Merge tag 'net-next-6.3' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-nextgrafted

Pull networking updates from Jakub Kicinski:
 "Core:

   - Add dedicated kmem_cache for typical/small skb->head, avoid having
     to access struct page at kfree time, and improve memory use.

   - Introduce sysctl to set default RPS configuration for new netdevs.

   - Define Netlink protocol specification format which can be used to
     describe messages used by each family and auto-generate parsers.
     Add tools for generating kernel data structures and uAPI headers.

   - Expose all net/core sysctls inside netns.

   - Remove 4s sleep in netpoll if carrier is instantly detected on
     boot.

   - Add configurable limit of MDB entries per port, and port-vlan.

   - Continue populating drop reasons throughout the stack.

   - Retire a handful of legacy Qdiscs and classifiers.

  Protocols:

   - Support IPv4 big TCP (TSO frames larger than 64kB).

   - Add IP_LOCAL_PORT_RANGE socket option, to control local port range
     on socket by socket basis.

   - Track and report in procfs number of MPTCP sockets used.

   - Support mixing IPv4 and IPv6 flows in the in-kernel MPTCP path
     manager.

   - IPv6: don't check net.ipv6.route.max_size and rely on garbage
     collection to free memory (similarly to IPv4).

   - Support Penultimate Segment Pop (PSP) flavor in SRv6 (RFC8986).

   - ICMP: add per-rate limit counters.

   - Add support for user scanning requests in ieee802154.

   - Remove static WEP support.

   - Support minimal Wi-Fi 7 Extremely High Throughput (EHT) rate
     reporting.

   - WiFi 7 EHT channel puncturing support (client & AP).

  BPF:

   - Add a rbtree data structure following the "next-gen data structure"
     precedent set by recently added linked list, that is, by using
     kfunc + kptr instead of adding a new BPF map type.

   - Expose XDP hints via kfuncs with initial support for RX hash and
     timestamp metadata.

   - Add BPF_F_NO_TUNNEL_KEY extension to bpf_skb_set_tunnel_key to
     better support decap on GRE tunnel devices not operating in collect
     metadata.

   - Improve x86 JIT's codegen for PROBE_MEM runtime error checks.

   - Remove the need for trace_printk_lock for bpf_trace_printk and
     bpf_trace_vprintk helpers.

   - Extend libbpf's bpf_tracing.h support for tracing arguments of
     kprobes/uprobes and syscall as a special case.

   - Significantly reduce the search time for module symbols by
     livepatch and BPF.

   - Enable cpumasks to be used as kptrs, which is useful for tracing
     programs tracking which tasks end up running on which CPUs in
     different time intervals.

   - Add support for BPF trampoline on s390x and riscv64.

   - Add capability to export the XDP features supported by the NIC.

   - Add __bpf_kfunc tag for marking kernel functions as kfuncs.

   - Add cgroup.memory=nobpf kernel parameter option to disable BPF
     memory accounting for container environments.

  Netfilter:

   - Remove the CLUSTERIP target. It has been marked as obsolete for
     years, and we still have WARN splats wrt races of the out-of-band
     /proc interface installed by this target.

   - Add 'destroy' commands to nf_tables. They are identical to the
     existing 'delete' commands, but do not return an error if the
     referenced object (set, chain, rule...) did not exist.

  Driver API:

   - Improve cpumask_local_spread() locality to help NICs set the right
     IRQ affinity on AMD platforms.

   - Separate C22 and C45 MDIO bus transactions more clearly.

   - Introduce new DCB table to control DSCP rewrite on egress.

   - Support configuration of Physical Layer Collision Avoidance (PLCA)
     Reconciliation Sublayer (RS) (802.3cg-2019). Modern version of
     shared medium Ethernet.

   - Support for MAC Merge layer (IEEE 802.3-2018 clause 99). Allowing
     preemption of low priority frames by high priority frames.

   - Add support for controlling MACSec offload using netlink SET.

   - Rework devlink instance refcounts to allow registration and
     de-registration under the instance lock. Split the code into
     multiple files, drop some of the unnecessarily granular locks and
     factor out common parts of netlink operation handling.

   - Add TX frame aggregation parameters (for USB drivers).

   - Add a new attr TCA_EXT_WARN_MSG to report TC (offload) warning
     messages with notifications for debug.

   - Allow offloading of UDP NEW connections via act_ct.

   - Add support for per action HW stats in TC.

   - Support hardware miss to TC action (continue processing in SW from
     a specific point in the action chain).

   - Warn if old Wireless Extension user space interface is used with
     modern cfg80211/mac80211 drivers. Do not support Wireless
     Extensions for Wi-Fi 7 devices at all. Everyone should switch to
     using nl80211 interface instead.

   - Improve the CAN bit timing configuration. Use extack to return
     error messages directly to user space, update the SJW handling,
     including the definition of a new default value that will benefit
     CAN-FD controllers, by increasing their oscillator tolerance.

  New hardware / drivers:

   - Ethernet:
      - nVidia BlueField-3 support (control traffic driver)
      - Ethernet support for imx93 SoCs
      - Motorcomm yt8531 gigabit Ethernet PHY
      - onsemi NCN26000 10BASE-T1S PHY (with support for PLCA)
      - Microchip LAN8841 PHY (incl. cable diagnostics and PTP)
      - Amlogic gxl MDIO mux

   - WiFi:
      - RealTek RTL8188EU (rtl8xxxu)
      - Qualcomm Wi-Fi 7 devices (ath12k)

   - CAN:
      - Renesas R-Car V4H

  Drivers:

   - Bluetooth:
      - Set Per Platform Antenna Gain (PPAG) for Intel controllers.

   - Ethernet NICs:
      - Intel (1G, igc):
         - support TSN / Qbv / packet scheduling features of i226 model
      - Intel (100G, ice):
         - use GNSS subsystem instead of TTY
         - multi-buffer XDP support
         - extend support for GPIO pins to E823 devices
      - nVidia/Mellanox:
         - update the shared buffer configuration on PFC commands
         - implement PTP adjphase function for HW offset control
         - TC support for Geneve and GRE with VF tunnel offload
         - more efficient crypto key management method
         - multi-port eswitch support
      - Netronome/Corigine:
         - add DCB IEEE support
         - support IPsec offloading for NFP3800
      - Freescale/NXP (enetc):
         - support XDP_REDIRECT for XDP non-linear buffers
         - improve reconfig, avoid link flap and waiting for idle
         - support MAC Merge layer
      - Other NICs:
         - sfc/ef100: add basic devlink support for ef100
         - ionic: rx_push mode operation (writing descriptors via MMIO)
         - bnxt: use the auxiliary bus abstraction for RDMA
         - r8169: disable ASPM and reset bus in case of tx timeout
         - cpsw: support QSGMII mode for J721e CPSW9G
         - cpts: support pulse-per-second output
         - ngbe: add an mdio bus driver
         - usbnet: optimize usbnet_bh() by avoiding unnecessary queuing
         - r8152: handle devices with FW with NCM support
         - amd-xgbe: support 10Mbps, 2.5GbE speeds and rx-adaptation
         - virtio-net: support multi buffer XDP
         - virtio/vsock: replace virtio_vsock_pkt with sk_buff
         - tsnep: XDP support

   - Ethernet high-speed switches:
      - nVidia/Mellanox (mlxsw):
         - add support for latency TLV (in FW control messages)
      - Microchip (sparx5):
         - separate explicit and implicit traffic forwarding rules, make
           the implicit rules always active
         - add support for egress DSCP rewrite
         - IS0 VCAP support (Ingress Classification)
         - IS2 VCAP filters (protos, L3 addrs, L4 ports, flags, ToS
           etc.)
         - ES2 VCAP support (Egress Access Control)
         - support for Per-Stream Filtering and Policing (802.1Q,
           8.6.5.1)

   - Ethernet embedded switches:
      - Marvell (mv88e6xxx):
         - add MAB (port auth) offload support
         - enable PTP receive for mv88e6390
      - NXP (ocelot):
         - support MAC Merge layer
         - support for the the vsc7512 internal copper phys
      - Microchip:
         - lan9303: convert to PHYLINK
         - lan966x: support TC flower filter statistics
         - lan937x: PTP support for KSZ9563/KSZ8563 and LAN937x
         - lan937x: support Credit Based Shaper configuration
         - ksz9477: support Energy Efficient Ethernet
      - other:
         - qca8k: convert to regmap read/write API, use bulk operations
         - rswitch: Improve TX timestamp accuracy

   - Intel WiFi (iwlwifi):
      - EHT (Wi-Fi 7) rate reporting
      - STEP equalizer support: transfer some STEP (connection to radio
        on platforms with integrated wifi) related parameters from the
        BIOS to the firmware.

   - Qualcomm 802.11ax WiFi (ath11k):
      - IPQ5018 support
      - Fine Timing Measurement (FTM) responder role support
      - channel 177 support

   - MediaTek WiFi (mt76):
      - per-PHY LED support
      - mt7996: EHT (Wi-Fi 7) support
      - Wireless Ethernet Dispatch (WED) reset support
      - switch to using page pool allocator

   - RealTek WiFi (rtw89):
      - support new version of Bluetooth co-existance

   - Mobile:
      - rmnet: support TX aggregation"

* tag 'net-next-6.3' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next: (1872 commits)
  page_pool: add a comment explaining the fragment counter usage
  net: ethtool: fix __ethtool_dev_mm_supported() implementation
  ethtool: pse-pd: Fix double word in comments
  xsk: add linux/vmalloc.h to xsk.c
  sefltests: netdevsim: wait for devlink instance after netns removal
  selftest: fib_tests: Always cleanup before exit
  net/mlx5e: Align IPsec ASO result memory to be as required by hardware
  net/mlx5e: TC, Set CT miss to the specific ct action instance
  net/mlx5e: Rename CHAIN_TO_REG to MAPPED_OBJ_TO_REG
  net/mlx5: Refactor tc miss handling to a single function
  net/mlx5: Kconfig: Make tc offload depend on tc skb extension
  net/sched: flower: Support hardware miss to tc action
  net/sched: flower: Move filter handle initialization earlier
  net/sched: cls_api: Support hardware miss to tc action
  net/sched: Rename user cookie and act cookie
  sfc: fix builds without CONFIG_RTC_LIB
  sfc: clean up some inconsistent indentings
  net/mlx4_en: Introduce flexible array to silence overflow warning
  net: lan966x: Fix possible deadlock inside PTP
  net/ulp: Remove redundant ->clone() test in inet_clone_ulp().
  ...

35 files changed, 8493 insertions, 0 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
new file mode 100644
index 000000000..5a074b997
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
@@ -0,0 +1,30 @@
+# SPDX-License-Identifier: MIT
+nvkm-y += nvkm/engine/fifo/base.o
+nvkm-y += nvkm/engine/fifo/cgrp.o
+nvkm-y += nvkm/engine/fifo/chan.o
+nvkm-y += nvkm/engine/fifo/chid.o
+nvkm-y += nvkm/engine/fifo/runl.o
+nvkm-y += nvkm/engine/fifo/runq.o
+
+nvkm-y += nvkm/engine/fifo/nv04.o
+nvkm-y += nvkm/engine/fifo/nv10.o
+nvkm-y += nvkm/engine/fifo/nv17.o
+nvkm-y += nvkm/engine/fifo/nv40.o
+nvkm-y += nvkm/engine/fifo/nv50.o
+nvkm-y += nvkm/engine/fifo/g84.o
+nvkm-y += nvkm/engine/fifo/g98.o
+nvkm-y += nvkm/engine/fifo/gf100.o
+nvkm-y += nvkm/engine/fifo/gk104.o
+nvkm-y += nvkm/engine/fifo/gk110.o
+nvkm-y += nvkm/engine/fifo/gk208.o
+nvkm-y += nvkm/engine/fifo/gk20a.o
+nvkm-y += nvkm/engine/fifo/gm107.o
+nvkm-y += nvkm/engine/fifo/gm200.o
+nvkm-y += nvkm/engine/fifo/gp100.o
+nvkm-y += nvkm/engine/fifo/gv100.o
+nvkm-y += nvkm/engine/fifo/tu102.o
+nvkm-y += nvkm/engine/fifo/ga100.o
+nvkm-y += nvkm/engine/fifo/ga102.o
+
+nvkm-y += nvkm/engine/fifo/ucgrp.o
+nvkm-y += nvkm/engine/fifo/uchan.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
new file mode 100644
index 000000000..5ea9a2ff0
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
@@ -0,0 +1,389 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/bar.h>
+#include <subdev/mc.h>
+#include <subdev/mmu.h>
+
+#include <nvif/cl0080.h>
+#include <nvif/unpack.h>
+
+bool
+nvkm_fifo_ctxsw_in_progress(struct nvkm_engine *engine)
+{
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+
+	nvkm_runl_foreach(runl, engine->subdev.device->fifo) {
+		nvkm_runl_foreach_engn(engn, runl) {
+			if (engn->engine == engine)
+				return engn->func->chsw ? engn->func->chsw(engn) : false;
+		}
+	}
+
+	return false;
+}
+
+void
+nvkm_fifo_pause(struct nvkm_fifo *fifo, unsigned long *flags)
+{
+	return fifo->func->pause(fifo, flags);
+}
+
+void
+nvkm_fifo_start(struct nvkm_fifo *fifo, unsigned long *flags)
+{
+	return fifo->func->start(fifo, flags);
+}
+
+void
+nvkm_fifo_fault(struct nvkm_fifo *fifo, struct nvkm_fault_data *info)
+{
+	return fifo->func->mmu_fault->recover(fifo, info);
+}
+
+static int
+nvkm_fifo_class_new(struct nvkm_device *device, const struct nvkm_oclass *oclass,
+		    void *argv, u32 argc, struct nvkm_object **pobject)
+{
+	struct nvkm_fifo *fifo = nvkm_fifo(oclass->engine);
+
+	if (oclass->engn == &fifo->func->cgrp.user)
+		return nvkm_ucgrp_new(fifo, oclass, argv, argc, pobject);
+
+	if (oclass->engn == &fifo->func->chan.user)
+		return nvkm_uchan_new(fifo, NULL, oclass, argv, argc, pobject);
+
+	WARN_ON(1);
+	return -ENOSYS;
+}
+
+static const struct nvkm_device_oclass
+nvkm_fifo_class = {
+	.ctor = nvkm_fifo_class_new,
+};
+
+static int
+nvkm_fifo_class_get(struct nvkm_oclass *oclass, int index, const struct nvkm_device_oclass **class)
+{
+	struct nvkm_fifo *fifo = nvkm_fifo(oclass->engine);
+	const struct nvkm_fifo_func_cgrp *cgrp = &fifo->func->cgrp;
+	const struct nvkm_fifo_func_chan *chan = &fifo->func->chan;
+	int c = 0;
+
+	/* *_CHANNEL_GROUP_* */
+	if (cgrp->user.oclass) {
+		if (c++ == index) {
+			oclass->base = cgrp->user;
+			oclass->engn = &fifo->func->cgrp.user;
+			*class = &nvkm_fifo_class;
+			return 0;
+		}
+	}
+
+	/* *_CHANNEL_DMA, *_CHANNEL_GPFIFO_* */
+	if (chan->user.oclass) {
+		if (c++ == index) {
+			oclass->base = chan->user;
+			oclass->engn = &fifo->func->chan.user;
+			*class = &nvkm_fifo_class;
+			return 0;
+		}
+	}
+
+	return c;
+}
+
+static int
+nvkm_fifo_fini(struct nvkm_engine *engine, bool suspend)
+{
+	struct nvkm_fifo *fifo = nvkm_fifo(engine);
+	struct nvkm_runl *runl;
+
+	nvkm_inth_block(&fifo->engine.subdev.inth);
+
+	nvkm_runl_foreach(runl, fifo)
+		nvkm_runl_fini(runl);
+
+	return 0;
+}
+
+static int
+nvkm_fifo_init(struct nvkm_engine *engine)
+{
+	struct nvkm_fifo *fifo = nvkm_fifo(engine);
+	struct nvkm_runq *runq;
+	struct nvkm_runl *runl;
+	u32 mask = 0;
+
+	if (fifo->func->init_pbdmas) {
+		nvkm_runq_foreach(runq, fifo)
+			mask |= BIT(runq->id);
+
+		fifo->func->init_pbdmas(fifo, mask);
+
+		nvkm_runq_foreach(runq, fifo)
+			runq->func->init(runq);
+	}
+
+	nvkm_runl_foreach(runl, fifo) {
+		if (runl->func->init)
+			runl->func->init(runl);
+	}
+
+	if (fifo->func->init)
+		fifo->func->init(fifo);
+
+	nvkm_inth_allow(&fifo->engine.subdev.inth);
+	return 0;
+}
+
+static int
+nvkm_fifo_info(struct nvkm_engine *engine, u64 mthd, u64 *data)
+{
+	struct nvkm_fifo *fifo = nvkm_fifo(engine);
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+	int ret;
+
+	ret = nvkm_subdev_oneinit(&fifo->engine.subdev);
+	if (ret)
+		return ret;
+
+	switch (mthd) {
+	case NV_DEVICE_HOST_CHANNELS: *data = fifo->chid ? fifo->chid->nr : 0; return 0;
+	case NV_DEVICE_HOST_RUNLISTS:
+		*data = 0;
+		nvkm_runl_foreach(runl, fifo)
+			*data |= BIT(runl->id);
+		return 0;
+	case NV_DEVICE_HOST_RUNLIST_ENGINES:
+		runl = nvkm_runl_get(fifo, *data, 0);
+		if (runl) {
+			*data = 0;
+			nvkm_runl_foreach_engn(engn, runl) {
+#define CASE(n) case NVKM_ENGINE_##n: *data |= NV_DEVICE_HOST_RUNLIST_ENGINES_##n; break
+				switch (engn->engine->subdev.type) {
+				case NVKM_ENGINE_DMAOBJ:
+					break;
+				CASE(SW    );
+				CASE(GR    );
+				CASE(MPEG  );
+				CASE(ME    );
+				CASE(CIPHER);
+				CASE(BSP   );
+				CASE(VP    );
+				CASE(CE    );
+				CASE(SEC   );
+				CASE(MSVLD );
+				CASE(MSPDEC);
+				CASE(MSPPP );
+				CASE(MSENC );
+				CASE(VIC   );
+				CASE(SEC2  );
+				CASE(NVDEC );
+				CASE(NVENC );
+				default:
+					WARN_ON(1);
+					break;
+				}
+#undef CASE
+			}
+			return 0;
+		}
+		return -EINVAL;
+	case NV_DEVICE_HOST_RUNLIST_CHANNELS:
+		if (!fifo->chid) {
+			runl = nvkm_runl_get(fifo, *data, 0);
+			if (runl) {
+				*data = runl->chid->nr;
+				return 0;
+			}
+		}
+		return -EINVAL;
+	default:
+		break;
+	}
+
+	return -ENOSYS;
+}
+
+static int
+nvkm_fifo_oneinit(struct nvkm_engine *engine)
+{
+	struct nvkm_subdev *subdev = &engine->subdev;
+	struct nvkm_device *device = subdev->device;
+	struct nvkm_fifo *fifo = nvkm_fifo(engine);
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+	int ret, nr, i;
+
+	/* Initialise CHID/CGID allocator(s) on GPUs where they aren't per-runlist. */
+	if (fifo->func->chid_nr) {
+		ret = fifo->func->chid_ctor(fifo, fifo->func->chid_nr(fifo));
+		if (ret)
+			return ret;
+	}
+
+	/* Create runqueues for each PBDMA. */
+	if (fifo->func->runq_nr) {
+		for (nr = fifo->func->runq_nr(fifo), i = 0; i < nr; i++) {
+			if (!nvkm_runq_new(fifo, i))
+				return -ENOMEM;
+		}
+	}
+
+	/* Create runlists. */
+	ret = fifo->func->runl_ctor(fifo);
+	if (ret)
+		return ret;
+
+	nvkm_runl_foreach(runl, fifo) {
+		RUNL_DEBUG(runl, "chan:%06x", runl->chan);
+		nvkm_runl_foreach_engn(engn, runl) {
+			ENGN_DEBUG(engn, "");
+		}
+	}
+
+	/* Register interrupt handler. */
+	if (fifo->func->intr) {
+		ret = nvkm_inth_add(&device->mc->intr, NVKM_INTR_SUBDEV, NVKM_INTR_PRIO_NORMAL,
+				    subdev, fifo->func->intr, &subdev->inth);
+		if (ret) {
+			nvkm_error(subdev, "intr %d\n", ret);
+			return ret;
+		}
+	}
+
+	/* Initialise non-stall intr handling. */
+	if (fifo->func->nonstall_ctor) {
+		ret = fifo->func->nonstall_ctor(fifo);
+		if (ret) {
+			nvkm_error(subdev, "nonstall %d\n", ret);
+		}
+	}
+
+	/* Allocate USERD + BAR1 polling area. */
+	if (fifo->func->chan.func->userd->bar == 1) {
+		struct nvkm_vmm *bar1 = nvkm_bar_bar1_vmm(device);
+
+		ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, fifo->chid->nr *
+				      fifo->func->chan.func->userd->size, 0, true,
+				      &fifo->userd.mem);
+		if (ret)
+			return ret;
+
+		ret = nvkm_vmm_get(bar1, 12, nvkm_memory_size(fifo->userd.mem), &fifo->userd.bar1);
+		if (ret)
+			return ret;
+
+		ret = nvkm_memory_map(fifo->userd.mem, 0, bar1, fifo->userd.bar1, NULL, 0);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static void
+nvkm_fifo_preinit(struct nvkm_engine *engine)
+{
+	nvkm_mc_reset(engine->subdev.device, NVKM_ENGINE_FIFO, 0);
+}
+
+static void *
+nvkm_fifo_dtor(struct nvkm_engine *engine)
+{
+	struct nvkm_fifo *fifo = nvkm_fifo(engine);
+	struct nvkm_runl *runl, *runt;
+	struct nvkm_runq *runq, *rtmp;
+
+	if (fifo->userd.bar1)
+		nvkm_vmm_put(nvkm_bar_bar1_vmm(engine->subdev.device), &fifo->userd.bar1);
+	nvkm_memory_unref(&fifo->userd.mem);
+
+	list_for_each_entry_safe(runl, runt, &fifo->runls, head)
+		nvkm_runl_del(runl);
+	list_for_each_entry_safe(runq, rtmp, &fifo->runqs, head)
+		nvkm_runq_del(runq);
+
+	nvkm_chid_unref(&fifo->cgid);
+	nvkm_chid_unref(&fifo->chid);
+
+	nvkm_event_fini(&fifo->nonstall.event);
+	mutex_destroy(&fifo->mutex);
+	return fifo;
+}
+
+static const struct nvkm_engine_func
+nvkm_fifo = {
+	.dtor = nvkm_fifo_dtor,
+	.preinit = nvkm_fifo_preinit,
+	.oneinit = nvkm_fifo_oneinit,
+	.info = nvkm_fifo_info,
+	.init = nvkm_fifo_init,
+	.fini = nvkm_fifo_fini,
+	.base.sclass = nvkm_fifo_class_get,
+};
+
+int
+nvkm_fifo_new_(const struct nvkm_fifo_func *func, struct nvkm_device *device,
+	       enum nvkm_subdev_type type, int inst, struct nvkm_fifo **pfifo)
+{
+	struct nvkm_fifo *fifo;
+	int ret;
+
+	if (!(fifo = *pfifo = kzalloc(sizeof(*fifo), GFP_KERNEL)))
+		return -ENOMEM;
+
+	fifo->func = func;
+	INIT_LIST_HEAD(&fifo->runqs);
+	INIT_LIST_HEAD(&fifo->runls);
+	/*TODO: Needs to be >CTXSW_TIMEOUT, so RC can recover before this is hit.
+	 *      CTXSW_TIMEOUT HW default seems to differ between GPUs, so just a
+	 *      large number for now until we support changing it.
+	 */
+	fifo->timeout.chan_msec = 10000;
+	spin_lock_init(&fifo->lock);
+	mutex_init(&fifo->mutex);
+
+	ret = nvkm_engine_ctor(&nvkm_fifo, device, type, inst, true, &fifo->engine);
+	if (ret)
+		return ret;
+
+	if (func->nonstall) {
+		ret = nvkm_event_init(func->nonstall, &fifo->engine.subdev, 1, 1,
+				      &fifo->nonstall.event);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c
new file mode 100644
index 000000000..ea53fb3d5
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c
@@ -0,0 +1,252 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+#include "priv.h"
+
+#include <core/gpuobj.h>
+#include <subdev/mmu.h>
+
+static void
+nvkm_cgrp_ectx_put(struct nvkm_cgrp *cgrp, struct nvkm_ectx **pectx)
+{
+	struct nvkm_ectx *ectx = *pectx;
+
+	if (ectx) {
+		struct nvkm_engn *engn = ectx->engn;
+
+		if (refcount_dec_and_test(&ectx->refs)) {
+			CGRP_TRACE(cgrp, "dtor ectx %d[%s]", engn->id, engn->engine->subdev.name);
+			nvkm_object_del(&ectx->object);
+			list_del(&ectx->head);
+			kfree(ectx);
+		}
+
+		*pectx = NULL;
+	}
+}
+
+static int
+nvkm_cgrp_ectx_get(struct nvkm_cgrp *cgrp, struct nvkm_engn *engn, struct nvkm_ectx **pectx,
+		   struct nvkm_chan *chan, struct nvkm_client *client)
+{
+	struct nvkm_engine *engine = engn->engine;
+	struct nvkm_oclass cclass = {
+		.client = client,
+		.engine = engine,
+	};
+	struct nvkm_ectx *ectx;
+	int ret = 0;
+
+	/* Look for an existing context for this engine in the channel group. */
+	ectx = nvkm_list_find(ectx, &cgrp->ectxs, head, ectx->engn == engn);
+	if (ectx) {
+		refcount_inc(&ectx->refs);
+		*pectx = ectx;
+		return 0;
+	}
+
+	/* Nope - create a fresh one. */
+	CGRP_TRACE(cgrp, "ctor ectx %d[%s]", engn->id, engn->engine->subdev.name);
+	if (!(ectx = *pectx = kzalloc(sizeof(*ectx), GFP_KERNEL)))
+		return -ENOMEM;
+
+	ectx->engn = engn;
+	refcount_set(&ectx->refs, 1);
+	refcount_set(&ectx->uses, 0);
+	list_add_tail(&ectx->head, &cgrp->ectxs);
+
+	/* Allocate the HW structures. */
+	if (engine->func->fifo.cclass)
+		ret = engine->func->fifo.cclass(chan, &cclass, &ectx->object);
+	else if (engine->func->cclass)
+		ret = nvkm_object_new_(engine->func->cclass, &cclass, NULL, 0, &ectx->object);
+
+	if (ret)
+		nvkm_cgrp_ectx_put(cgrp, pectx);
+
+	return ret;
+}
+
+void
+nvkm_cgrp_vctx_put(struct nvkm_cgrp *cgrp, struct nvkm_vctx **pvctx)
+{
+	struct nvkm_vctx *vctx = *pvctx;
+
+	if (vctx) {
+		struct nvkm_engn *engn = vctx->ectx->engn;
+
+		if (refcount_dec_and_test(&vctx->refs)) {
+			CGRP_TRACE(cgrp, "dtor vctx %d[%s]", engn->id, engn->engine->subdev.name);
+			nvkm_vmm_put(vctx->vmm, &vctx->vma);
+			nvkm_gpuobj_del(&vctx->inst);
+
+			nvkm_cgrp_ectx_put(cgrp, &vctx->ectx);
+			if (vctx->vmm) {
+				atomic_dec(&vctx->vmm->engref[engn->engine->subdev.type]);
+				nvkm_vmm_unref(&vctx->vmm);
+			}
+			list_del(&vctx->head);
+			kfree(vctx);
+		}
+
+		*pvctx = NULL;
+	}
+}
+
+int
+nvkm_cgrp_vctx_get(struct nvkm_cgrp *cgrp, struct nvkm_engn *engn, struct nvkm_chan *chan,
+		   struct nvkm_vctx **pvctx, struct nvkm_client *client)
+{
+	struct nvkm_ectx *ectx;
+	struct nvkm_vctx *vctx;
+	int ret;
+
+	/* Look for an existing sub-context for this engine+VEID in the channel group. */
+	vctx = nvkm_list_find(vctx, &cgrp->vctxs, head,
+			      vctx->ectx->engn == engn && vctx->vmm == chan->vmm);
+	if (vctx) {
+		refcount_inc(&vctx->refs);
+		*pvctx = vctx;
+		return 0;
+	}
+
+	/* Nope - create a fresh one.  But, context first. */
+	ret = nvkm_cgrp_ectx_get(cgrp, engn, &ectx, chan, client);
+	if (ret) {
+		CGRP_ERROR(cgrp, "ectx %d[%s]: %d", engn->id, engn->engine->subdev.name, ret);
+		return ret;
+	}
+
+	/* Now, create the sub-context. */
+	CGRP_TRACE(cgrp, "ctor vctx %d[%s]", engn->id, engn->engine->subdev.name);
+	if (!(vctx = *pvctx = kzalloc(sizeof(*vctx), GFP_KERNEL))) {
+		nvkm_cgrp_ectx_put(cgrp, &ectx);
+		return -ENOMEM;
+	}
+
+	vctx->ectx = ectx;
+	vctx->vmm = nvkm_vmm_ref(chan->vmm);
+	refcount_set(&vctx->refs, 1);
+	list_add_tail(&vctx->head, &cgrp->vctxs);
+
+	/* MMU on some GPUs needs to know engine usage for TLB invalidation. */
+	if (vctx->vmm)
+		atomic_inc(&vctx->vmm->engref[engn->engine->subdev.type]);
+
+	/* Allocate the HW structures. */
+	if (engn->func->bind) {
+		ret = nvkm_object_bind(vctx->ectx->object, NULL, 0, &vctx->inst);
+		if (ret == 0 && engn->func->ctor)
+			ret = engn->func->ctor(engn, vctx);
+	}
+
+	if (ret)
+		nvkm_cgrp_vctx_put(cgrp, pvctx);
+
+	return ret;
+}
+
+static void
+nvkm_cgrp_del(struct kref *kref)
+{
+	struct nvkm_cgrp *cgrp = container_of(kref, typeof(*cgrp), kref);
+	struct nvkm_runl *runl = cgrp->runl;
+
+	if (runl->cgid)
+		nvkm_chid_put(runl->cgid, cgrp->id, &cgrp->lock);
+
+	mutex_destroy(&cgrp->mutex);
+	nvkm_vmm_unref(&cgrp->vmm);
+	kfree(cgrp);
+}
+
+void
+nvkm_cgrp_unref(struct nvkm_cgrp **pcgrp)
+{
+	struct nvkm_cgrp *cgrp = *pcgrp;
+
+	if (!cgrp)
+		return;
+
+	kref_put(&cgrp->kref, nvkm_cgrp_del);
+	*pcgrp = NULL;
+}
+
+struct nvkm_cgrp *
+nvkm_cgrp_ref(struct nvkm_cgrp *cgrp)
+{
+	if (cgrp)
+		kref_get(&cgrp->kref);
+
+	return cgrp;
+}
+
+void
+nvkm_cgrp_put(struct nvkm_cgrp **pcgrp, unsigned long irqflags)
+{
+	struct nvkm_cgrp *cgrp = *pcgrp;
+
+	if (!cgrp)
+		return;
+
+	*pcgrp = NULL;
+	spin_unlock_irqrestore(&cgrp->lock, irqflags);
+}
+
+int
+nvkm_cgrp_new(struct nvkm_runl *runl, const char *name, struct nvkm_vmm *vmm, bool hw,
+	      struct nvkm_cgrp **pcgrp)
+{
+	struct nvkm_cgrp *cgrp;
+
+	if (!(cgrp = *pcgrp = kmalloc(sizeof(*cgrp), GFP_KERNEL)))
+		return -ENOMEM;
+
+	cgrp->func = runl->fifo->func->cgrp.func;
+	strscpy(cgrp->name, name, sizeof(cgrp->name));
+	cgrp->runl = runl;
+	cgrp->vmm = nvkm_vmm_ref(vmm);
+	cgrp->hw = hw;
+	cgrp->id = -1;
+	kref_init(&cgrp->kref);
+	INIT_LIST_HEAD(&cgrp->chans);
+	cgrp->chan_nr = 0;
+	spin_lock_init(&cgrp->lock);
+	INIT_LIST_HEAD(&cgrp->ectxs);
+	INIT_LIST_HEAD(&cgrp->vctxs);
+	mutex_init(&cgrp->mutex);
+	atomic_set(&cgrp->rc, NVKM_CGRP_RC_NONE);
+
+	if (runl->cgid) {
+		cgrp->id = nvkm_chid_get(runl->cgid, cgrp);
+		if (cgrp->id < 0) {
+			RUNL_ERROR(runl, "!cgids");
+			nvkm_cgrp_unref(pcgrp);
+			return -ENOSPC;
+		}
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h
new file mode 100644
index 000000000..5f6abd59a
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h
@@ -0,0 +1,75 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVKM_CGRP_H__
+#define __NVKM_CGRP_H__
+#include <core/os.h>
+struct nvkm_chan;
+struct nvkm_client;
+
+struct nvkm_vctx {
+	struct nvkm_ectx *ectx;
+	struct nvkm_vmm *vmm;
+	refcount_t refs;
+
+	struct nvkm_gpuobj *inst;
+	struct nvkm_vma *vma;
+
+	struct list_head head;
+};
+
+struct nvkm_ectx {
+	struct nvkm_engn *engn;
+	refcount_t refs;
+	refcount_t uses;
+
+	struct nvkm_object *object;
+
+	struct list_head head;
+};
+
+struct nvkm_cgrp {
+	const struct nvkm_cgrp_func {
+		void (*preempt)(struct nvkm_cgrp *);
+	} *func;
+	char name[64];
+	struct nvkm_runl *runl;
+	struct nvkm_vmm *vmm;
+	bool hw;
+	int id;
+	struct kref kref;
+
+	struct list_head chans;
+	int chan_nr;
+
+	spinlock_t lock; /* protects irq handler channel (group) lookup */
+
+	struct list_head ectxs;
+	struct list_head vctxs;
+	struct mutex mutex;
+
+#define NVKM_CGRP_RC_NONE    0
+#define NVKM_CGRP_RC_PENDING 1
+#define NVKM_CGRP_RC_RUNNING 2
+	atomic_t rc;
+
+	struct list_head head;
+};
+
+int nvkm_cgrp_new(struct nvkm_runl *, const char *name, struct nvkm_vmm *, bool hw,
+		  struct nvkm_cgrp **);
+struct nvkm_cgrp *nvkm_cgrp_ref(struct nvkm_cgrp *);
+void nvkm_cgrp_unref(struct nvkm_cgrp **);
+int nvkm_cgrp_vctx_get(struct nvkm_cgrp *, struct nvkm_engn *, struct nvkm_chan *,
+		       struct nvkm_vctx **, struct nvkm_client *);
+void nvkm_cgrp_vctx_put(struct nvkm_cgrp *, struct nvkm_vctx **);
+
+void nvkm_cgrp_put(struct nvkm_cgrp **, unsigned long irqflags);
+
+#define nvkm_cgrp_foreach_chan(chan,cgrp) list_for_each_entry((chan), &(cgrp)->chans, head)
+#define nvkm_cgrp_foreach_chan_safe(chan,ctmp,cgrp) \
+	list_for_each_entry_safe((chan), (ctmp), &(cgrp)->chans, head)
+
+#define CGRP_PRCLI(c,l,p,f,a...) RUNL_PRINT((c)->runl, l, p, "%04x:[%s]"f, (c)->id, (c)->name, ##a)
+#define CGRP_PRINT(c,l,p,f,a...) RUNL_PRINT((c)->runl, l, p, "%04x:"f, (c)->id, ##a)
+#define CGRP_ERROR(c,f,a...) CGRP_PRCLI((c), ERROR,    err, " "f"\n", ##a)
+#define CGRP_TRACE(c,f,a...) CGRP_PRINT((c), TRACE,   info, " "f"\n", ##a)
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
new file mode 100644
index 000000000..b7c9d6115
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
@@ -0,0 +1,480 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "chan.h"
+#include "chid.h"
+#include "cgrp.h"
+#include "chid.h"
+#include "runl.h"
+#include "priv.h"
+
+#include <core/ramht.h>
+#include <subdev/mmu.h>
+#include <engine/dma.h>
+
+#include <nvif/if0020.h>
+
+const struct nvkm_event_func
+nvkm_chan_event = {
+};
+
+void
+nvkm_chan_cctx_bind(struct nvkm_chan *chan, struct nvkm_engn *engn, struct nvkm_cctx *cctx)
+{
+	struct nvkm_cgrp *cgrp = chan->cgrp;
+	struct nvkm_runl *runl = cgrp->runl;
+	struct nvkm_engine *engine = engn->engine;
+
+	if (!engn->func->bind)
+		return;
+
+	CHAN_TRACE(chan, "%sbind cctx %d[%s]", cctx ? "" : "un", engn->id, engine->subdev.name);
+
+	/* Prevent any channel in channel group from being rescheduled, kick them
+	 * off host and any engine(s) they're loaded on.
+	 */
+	if (cgrp->hw)
+		nvkm_runl_block(runl);
+	else
+		nvkm_chan_block(chan);
+	nvkm_chan_preempt(chan, true);
+
+	/* Update context pointer. */
+	engn->func->bind(engn, cctx, chan);
+
+	/* Resume normal operation. */
+	if (cgrp->hw)
+		nvkm_runl_allow(runl);
+	else
+		nvkm_chan_allow(chan);
+}
+
+void
+nvkm_chan_cctx_put(struct nvkm_chan *chan, struct nvkm_cctx **pcctx)
+{
+	struct nvkm_cctx *cctx = *pcctx;
+
+	if (cctx) {
+		struct nvkm_engn *engn = cctx->vctx->ectx->engn;
+
+		if (refcount_dec_and_mutex_lock(&cctx->refs, &chan->cgrp->mutex)) {
+			CHAN_TRACE(chan, "dtor cctx %d[%s]", engn->id, engn->engine->subdev.name);
+			nvkm_cgrp_vctx_put(chan->cgrp, &cctx->vctx);
+			list_del(&cctx->head);
+			kfree(cctx);
+			mutex_unlock(&chan->cgrp->mutex);
+		}
+
+		*pcctx = NULL;
+	}
+}
+
+int
+nvkm_chan_cctx_get(struct nvkm_chan *chan, struct nvkm_engn *engn, struct nvkm_cctx **pcctx,
+		   struct nvkm_client *client)
+{
+	struct nvkm_cgrp *cgrp = chan->cgrp;
+	struct nvkm_vctx *vctx;
+	struct nvkm_cctx *cctx;
+	int ret;
+
+	/* Look for an existing channel context for this engine+VEID. */
+	mutex_lock(&cgrp->mutex);
+	cctx = nvkm_list_find(cctx, &chan->cctxs, head,
+			      cctx->vctx->ectx->engn == engn && cctx->vctx->vmm == chan->vmm);
+	if (cctx) {
+		refcount_inc(&cctx->refs);
+		*pcctx = cctx;
+		mutex_unlock(&chan->cgrp->mutex);
+		return 0;
+	}
+
+	/* Nope - create a fresh one.  But, sub-context first. */
+	ret = nvkm_cgrp_vctx_get(cgrp, engn, chan, &vctx, client);
+	if (ret) {
+		CHAN_ERROR(chan, "vctx %d[%s]: %d", engn->id, engn->engine->subdev.name, ret);
+		goto done;
+	}
+
+	/* Now, create the channel context - to track engine binding. */
+	CHAN_TRACE(chan, "ctor cctx %d[%s]", engn->id, engn->engine->subdev.name);
+	if (!(cctx = *pcctx = kzalloc(sizeof(*cctx), GFP_KERNEL))) {
+		nvkm_cgrp_vctx_put(cgrp, &vctx);
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	cctx->vctx = vctx;
+	refcount_set(&cctx->refs, 1);
+	refcount_set(&cctx->uses, 0);
+	list_add_tail(&cctx->head, &chan->cctxs);
+done:
+	mutex_unlock(&cgrp->mutex);
+	return ret;
+}
+
+int
+nvkm_chan_preempt_locked(struct nvkm_chan *chan, bool wait)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+
+	CHAN_TRACE(chan, "preempt");
+	chan->func->preempt(chan);
+	if (!wait)
+		return 0;
+
+	return nvkm_runl_preempt_wait(runl);
+}
+
+int
+nvkm_chan_preempt(struct nvkm_chan *chan, bool wait)
+{
+	int ret;
+
+	if (!chan->func->preempt)
+		return 0;
+
+	mutex_lock(&chan->cgrp->runl->mutex);
+	ret = nvkm_chan_preempt_locked(chan, wait);
+	mutex_unlock(&chan->cgrp->runl->mutex);
+	return ret;
+}
+
+void
+nvkm_chan_remove_locked(struct nvkm_chan *chan)
+{
+	struct nvkm_cgrp *cgrp = chan->cgrp;
+	struct nvkm_runl *runl = cgrp->runl;
+
+	if (list_empty(&chan->head))
+		return;
+
+	CHAN_TRACE(chan, "remove");
+	if (!--cgrp->chan_nr) {
+		runl->cgrp_nr--;
+		list_del(&cgrp->head);
+	}
+	runl->chan_nr--;
+	list_del_init(&chan->head);
+	atomic_set(&runl->changed, 1);
+}
+
+void
+nvkm_chan_remove(struct nvkm_chan *chan, bool preempt)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+
+	mutex_lock(&runl->mutex);
+	if (preempt && chan->func->preempt)
+		nvkm_chan_preempt_locked(chan, true);
+	nvkm_chan_remove_locked(chan);
+	nvkm_runl_update_locked(runl, true);
+	mutex_unlock(&runl->mutex);
+}
+
+void
+nvkm_chan_insert(struct nvkm_chan *chan)
+{
+	struct nvkm_cgrp *cgrp = chan->cgrp;
+	struct nvkm_runl *runl = cgrp->runl;
+
+	mutex_lock(&runl->mutex);
+	if (WARN_ON(!list_empty(&chan->head))) {
+		mutex_unlock(&runl->mutex);
+		return;
+	}
+
+	CHAN_TRACE(chan, "insert");
+	list_add_tail(&chan->head, &cgrp->chans);
+	runl->chan_nr++;
+	if (!cgrp->chan_nr++) {
+		list_add_tail(&cgrp->head, &cgrp->runl->cgrps);
+		runl->cgrp_nr++;
+	}
+	atomic_set(&runl->changed, 1);
+	nvkm_runl_update_locked(runl, true);
+	mutex_unlock(&runl->mutex);
+}
+
+static void
+nvkm_chan_block_locked(struct nvkm_chan *chan)
+{
+	CHAN_TRACE(chan, "block %d", atomic_read(&chan->blocked));
+	if (atomic_inc_return(&chan->blocked) == 1)
+		chan->func->stop(chan);
+}
+
+void
+nvkm_chan_error(struct nvkm_chan *chan, bool preempt)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&chan->lock, flags);
+	if (atomic_inc_return(&chan->errored) == 1) {
+		CHAN_ERROR(chan, "errored - disabling channel");
+		nvkm_chan_block_locked(chan);
+		if (preempt)
+			chan->func->preempt(chan);
+		nvkm_event_ntfy(&chan->cgrp->runl->chid->event, chan->id, NVKM_CHAN_EVENT_ERRORED);
+	}
+	spin_unlock_irqrestore(&chan->lock, flags);
+}
+
+void
+nvkm_chan_block(struct nvkm_chan *chan)
+{
+	spin_lock_irq(&chan->lock);
+	nvkm_chan_block_locked(chan);
+	spin_unlock_irq(&chan->lock);
+}
+
+void
+nvkm_chan_allow(struct nvkm_chan *chan)
+{
+	spin_lock_irq(&chan->lock);
+	CHAN_TRACE(chan, "allow %d", atomic_read(&chan->blocked));
+	if (atomic_dec_and_test(&chan->blocked))
+		chan->func->start(chan);
+	spin_unlock_irq(&chan->lock);
+}
+
+void
+nvkm_chan_del(struct nvkm_chan **pchan)
+{
+	struct nvkm_chan *chan = *pchan;
+
+	if (!chan)
+		return;
+
+	if (chan->func->ramfc->clear)
+		chan->func->ramfc->clear(chan);
+
+	nvkm_ramht_del(&chan->ramht);
+	nvkm_gpuobj_del(&chan->pgd);
+	nvkm_gpuobj_del(&chan->eng);
+	nvkm_gpuobj_del(&chan->cache);
+	nvkm_gpuobj_del(&chan->ramfc);
+
+	nvkm_memory_unref(&chan->userd.mem);
+
+	if (chan->cgrp) {
+		nvkm_chid_put(chan->cgrp->runl->chid, chan->id, &chan->cgrp->lock);
+		nvkm_cgrp_unref(&chan->cgrp);
+	}
+
+	if (chan->vmm) {
+		nvkm_vmm_part(chan->vmm, chan->inst->memory);
+		nvkm_vmm_unref(&chan->vmm);
+	}
+
+	nvkm_gpuobj_del(&chan->push);
+	nvkm_gpuobj_del(&chan->inst);
+	kfree(chan);
+}
+
+void
+nvkm_chan_put(struct nvkm_chan **pchan, unsigned long irqflags)
+{
+	struct nvkm_chan *chan = *pchan;
+
+	if (!chan)
+		return;
+
+	*pchan = NULL;
+	spin_unlock_irqrestore(&chan->cgrp->lock, irqflags);
+}
+
+struct nvkm_chan *
+nvkm_chan_get_inst(struct nvkm_engine *engine, u64 inst, unsigned long *pirqflags)
+{
+	struct nvkm_fifo *fifo = engine->subdev.device->fifo;
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+	struct nvkm_chan *chan;
+
+	nvkm_runl_foreach(runl, fifo) {
+		nvkm_runl_foreach_engn(engn, runl) {
+			if (engine == &fifo->engine || engn->engine == engine) {
+				chan = nvkm_runl_chan_get_inst(runl, inst, pirqflags);
+				if (chan || engn->engine == engine)
+					return chan;
+			}
+		}
+	}
+
+	return NULL;
+}
+
+struct nvkm_chan *
+nvkm_chan_get_chid(struct nvkm_engine *engine, int id, unsigned long *pirqflags)
+{
+	struct nvkm_fifo *fifo = engine->subdev.device->fifo;
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+
+	nvkm_runl_foreach(runl, fifo) {
+		nvkm_runl_foreach_engn(engn, runl) {
+			if (fifo->chid || engn->engine == engine)
+				return nvkm_runl_chan_get_chid(runl, id, pirqflags);
+		}
+	}
+
+	return NULL;
+}
+
+int
+nvkm_chan_new_(const struct nvkm_chan_func *func, struct nvkm_runl *runl, int runq,
+	       struct nvkm_cgrp *cgrp, const char *name, bool priv, u32 devm, struct nvkm_vmm *vmm,
+	       struct nvkm_dmaobj *dmaobj, u64 offset, u64 length,
+	       struct nvkm_memory *userd, u64 ouserd, struct nvkm_chan **pchan)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_chan *chan;
+	int ret;
+
+	/* Validate arguments against class requirements. */
+	if ((runq && runq >= runl->func->runqs) ||
+	    (!func->inst->vmm != !vmm) ||
+	    ((func->userd->bar < 0) == !userd) ||
+	    (!func->ramfc->ctxdma != !dmaobj) ||
+	    ((func->ramfc->devm < devm) && devm != BIT(0)) ||
+	    (!func->ramfc->priv && priv)) {
+		RUNL_DEBUG(runl, "args runq:%d:%d vmm:%d:%p userd:%d:%p "
+				 "push:%d:%p devm:%08x:%08x priv:%d:%d",
+			   runl->func->runqs, runq, func->inst->vmm, vmm,
+			   func->userd->bar < 0, userd, func->ramfc->ctxdma, dmaobj,
+			   func->ramfc->devm, devm, func->ramfc->priv, priv);
+		return -EINVAL;
+	}
+
+	if (!(chan = *pchan = kzalloc(sizeof(*chan), GFP_KERNEL)))
+		return -ENOMEM;
+
+	chan->func = func;
+	strscpy(chan->name, name, sizeof(chan->name));
+	chan->runq = runq;
+	chan->id = -1;
+	spin_lock_init(&chan->lock);
+	atomic_set(&chan->blocked, 1);
+	atomic_set(&chan->errored, 0);
+	INIT_LIST_HEAD(&chan->cctxs);
+	INIT_LIST_HEAD(&chan->head);
+
+	/* Join channel group.
+	 *
+	 * GK110 and newer support channel groups (aka TSGs), where individual channels
+	 * share a timeslice, and, engine context(s).
+	 *
+	 * As such, engine contexts are tracked in nvkm_cgrp and we need them even when
+	 * channels aren't in an API channel group, and on HW that doesn't support TSGs.
+	 */
+	if (!cgrp) {
+		ret = nvkm_cgrp_new(runl, chan->name, vmm, fifo->func->cgrp.force, &chan->cgrp);
+		if (ret) {
+			RUNL_DEBUG(runl, "cgrp %d", ret);
+			return ret;
+		}
+
+		cgrp = chan->cgrp;
+	} else {
+		if (cgrp->runl != runl || cgrp->vmm != vmm) {
+			RUNL_DEBUG(runl, "cgrp %d %d", cgrp->runl != runl, cgrp->vmm != vmm);
+			return -EINVAL;
+		}
+
+		chan->cgrp = nvkm_cgrp_ref(cgrp);
+	}
+
+	/* Allocate instance block. */
+	ret = nvkm_gpuobj_new(device, func->inst->size, 0x1000, func->inst->zero, NULL,
+			      &chan->inst);
+	if (ret) {
+		RUNL_DEBUG(runl, "inst %d", ret);
+		return ret;
+	}
+
+	/* Initialise virtual address-space. */
+	if (func->inst->vmm) {
+		if (WARN_ON(vmm->mmu != device->mmu))
+			return -EINVAL;
+
+		ret = nvkm_vmm_join(vmm, chan->inst->memory);
+		if (ret) {
+			RUNL_DEBUG(runl, "vmm %d", ret);
+			return ret;
+		}
+
+		chan->vmm = nvkm_vmm_ref(vmm);
+	}
+
+	/* Allocate HW ctxdma for push buffer. */
+	if (func->ramfc->ctxdma) {
+		ret = nvkm_object_bind(&dmaobj->object, chan->inst, -16, &chan->push);
+		if (ret) {
+			RUNL_DEBUG(runl, "bind %d", ret);
+			return ret;
+		}
+	}
+
+	/* Allocate channel ID. */
+	chan->id = nvkm_chid_get(runl->chid, chan);
+	if (chan->id < 0) {
+		RUNL_ERROR(runl, "!chids");
+		return -ENOSPC;
+	}
+
+	if (cgrp->id < 0)
+		cgrp->id = chan->id;
+
+	/* Initialise USERD. */
+	if (func->userd->bar < 0) {
+		if (ouserd + chan->func->userd->size >= nvkm_memory_size(userd)) {
+			RUNL_DEBUG(runl, "ouserd %llx", ouserd);
+			return -EINVAL;
+		}
+
+		ret = nvkm_memory_kmap(userd, &chan->userd.mem);
+		if (ret) {
+			RUNL_DEBUG(runl, "userd %d", ret);
+			return ret;
+		}
+
+		chan->userd.base = ouserd;
+	} else {
+		chan->userd.mem = nvkm_memory_ref(fifo->userd.mem);
+		chan->userd.base = chan->id * chan->func->userd->size;
+	}
+
+	if (chan->func->userd->clear)
+		chan->func->userd->clear(chan);
+
+	/* Initialise RAMFC. */
+	ret = chan->func->ramfc->write(chan, offset, length, devm, priv);
+	if (ret) {
+		RUNL_DEBUG(runl, "ramfc %d", ret);
+		return ret;
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
new file mode 100644
index 000000000..85b94f699
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
@@ -0,0 +1,78 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVKM_CHAN_H__
+#define __NVKM_CHAN_H__
+#include <engine/fifo.h>
+struct nvkm_dmaobj;
+struct nvkm_engn;
+struct nvkm_runl;
+
+extern const struct nvkm_event_func nvkm_chan_event;
+
+struct nvkm_cctx {
+	struct nvkm_vctx *vctx;
+	refcount_t refs;
+	refcount_t uses;
+
+	struct list_head head;
+};
+
+struct nvkm_chan_func {
+	const struct nvkm_chan_func_inst {
+		u32 size;
+		bool zero;
+		bool vmm;
+	} *inst;
+
+	const struct nvkm_chan_func_userd {
+		int bar;
+		u32 base;
+		u32 size;
+		void (*clear)(struct nvkm_chan *);
+	} *userd;
+
+	const struct nvkm_chan_func_ramfc {
+		const struct nvkm_ramfc_layout {
+			unsigned bits:6;
+			unsigned ctxs:5;
+			unsigned ctxp:8;
+			unsigned regs:5;
+			unsigned regp;
+		} *layout;
+		int (*write)(struct nvkm_chan *, u64 offset, u64 length, u32 devm, bool priv);
+		void (*clear)(struct nvkm_chan *);
+		bool ctxdma;
+		u32 devm;
+		bool priv;
+	} *ramfc;
+
+	void (*bind)(struct nvkm_chan *);
+	void (*unbind)(struct nvkm_chan *);
+	void (*start)(struct nvkm_chan *);
+	void (*stop)(struct nvkm_chan *);
+	void (*preempt)(struct nvkm_chan *);
+	u32 (*doorbell_handle)(struct nvkm_chan *);
+};
+
+int nvkm_chan_new_(const struct nvkm_chan_func *, struct nvkm_runl *, int runq, struct nvkm_cgrp *,
+		   const char *name, bool priv, u32 devm, struct nvkm_vmm *, struct nvkm_dmaobj *,
+		   u64 offset, u64 length, struct nvkm_memory *userd, u64 userd_bar1,
+		   struct nvkm_chan **);
+void nvkm_chan_del(struct nvkm_chan **);
+void nvkm_chan_allow(struct nvkm_chan *);
+void nvkm_chan_block(struct nvkm_chan *);
+void nvkm_chan_error(struct nvkm_chan *, bool preempt);
+void nvkm_chan_insert(struct nvkm_chan *);
+void nvkm_chan_remove(struct nvkm_chan *, bool preempt);
+void nvkm_chan_remove_locked(struct nvkm_chan *);
+int nvkm_chan_preempt(struct nvkm_chan *, bool wait);
+int nvkm_chan_preempt_locked(struct nvkm_chan *, bool wait);
+int nvkm_chan_cctx_get(struct nvkm_chan *, struct nvkm_engn *, struct nvkm_cctx **,
+		       struct nvkm_client * /*TODO: remove need for this */);
+void nvkm_chan_cctx_put(struct nvkm_chan *, struct nvkm_cctx **);
+void nvkm_chan_cctx_bind(struct nvkm_chan *, struct nvkm_engn *, struct nvkm_cctx *);
+
+#define CHAN_PRCLI(c,l,p,f,a...) CGRP_PRINT((c)->cgrp, l, p, "%04x:[%s]"f, (c)->id, (c)->name, ##a)
+#define CHAN_PRINT(c,l,p,f,a...) CGRP_PRINT((c)->cgrp, l, p, "%04x:"f, (c)->id, ##a)
+#define CHAN_ERROR(c,f,a...) CHAN_PRCLI((c), ERROR,    err, " "f"\n", ##a)
+#define CHAN_TRACE(c,f,a...) CHAN_PRINT((c), TRACE,   info, " "f"\n", ##a)
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chid.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chid.c
new file mode 100644
index 000000000..23944d95e
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chid.c
@@ -0,0 +1,111 @@
+/*
+ * Copyright 2020 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "chid.h"
+
+void
+nvkm_chid_put(struct nvkm_chid *chid, int id, spinlock_t *data_lock)
+{
+	if (id >= 0) {
+		spin_lock_irq(&chid->lock);
+		spin_lock(data_lock);
+		chid->data[id] = NULL;
+		spin_unlock(data_lock);
+		clear_bit(id, chid->used);
+		spin_unlock_irq(&chid->lock);
+	}
+}
+
+int
+nvkm_chid_get(struct nvkm_chid *chid, void *data)
+{
+	int id = -1, cid;
+
+	spin_lock_irq(&chid->lock);
+	cid = find_first_zero_bit(chid->used, chid->nr);
+	if (cid < chid->nr) {
+		set_bit(cid, chid->used);
+		chid->data[cid] = data;
+		id = cid;
+	}
+	spin_unlock_irq(&chid->lock);
+	return id;
+}
+
+static void
+nvkm_chid_del(struct kref *kref)
+{
+	struct nvkm_chid *chid = container_of(kref, typeof(*chid), kref);
+
+	nvkm_event_fini(&chid->event);
+
+	kvfree(chid->data);
+	kfree(chid);
+}
+
+void
+nvkm_chid_unref(struct nvkm_chid **pchid)
+{
+	struct nvkm_chid *chid = *pchid;
+
+	if (!chid)
+		return;
+
+	kref_put(&chid->kref, nvkm_chid_del);
+	*pchid = NULL;
+}
+
+struct nvkm_chid *
+nvkm_chid_ref(struct nvkm_chid *chid)
+{
+	if (chid)
+		kref_get(&chid->kref);
+
+	return chid;
+}
+
+int
+nvkm_chid_new(const struct nvkm_event_func *func, struct nvkm_subdev *subdev,
+	      int nr, int first, int count, struct nvkm_chid **pchid)
+{
+	struct nvkm_chid *chid;
+	int id;
+
+	if (!(chid = *pchid = kzalloc(struct_size(chid, used, nr), GFP_KERNEL)))
+		return -ENOMEM;
+
+	kref_init(&chid->kref);
+	chid->nr = nr;
+	chid->mask = chid->nr - 1;
+	spin_lock_init(&chid->lock);
+
+	if (!(chid->data = kvzalloc(sizeof(*chid->data) * nr, GFP_KERNEL))) {
+		nvkm_chid_unref(pchid);
+		return -ENOMEM;
+	}
+
+	for (id = 0; id < first; id++)
+		__set_bit(id, chid->used);
+	for (id = first + count; id < nr; id++)
+		__set_bit(id, chid->used);
+
+	return nvkm_event_init(func, subdev, 1, nr, &chid->event);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chid.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chid.h
new file mode 100644
index 000000000..2a42efb18
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chid.h
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVKM_CHID_H__
+#define __NVKM_CHID_H__
+#include <core/event.h>
+
+struct nvkm_chid {
+	struct kref kref;
+	int nr;
+	u32 mask;
+
+	struct nvkm_event event;
+
+	void **data;
+
+	spinlock_t lock;
+	unsigned long used[];
+};
+
+int nvkm_chid_new(const struct nvkm_event_func *, struct nvkm_subdev *,
+		  int nr, int first, int count, struct nvkm_chid **pchid);
+struct nvkm_chid *nvkm_chid_ref(struct nvkm_chid *);
+void nvkm_chid_unref(struct nvkm_chid **);
+int nvkm_chid_get(struct nvkm_chid *, void *data);
+void nvkm_chid_put(struct nvkm_chid *, int id, spinlock_t *data_lock);
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
new file mode 100644
index 000000000..6b229a3fb
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
@@ -0,0 +1,231 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "runl.h"
+
+#include <core/ramht.h>
+#include <subdev/timer.h>
+
+#include <nvif/class.h>
+
+static void
+g84_chan_bind(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x002600 + (chan->id * 4), chan->ramfc->addr >> 8);
+}
+
+static int
+g84_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+	const u32 limit2 = ilog2(length / 8);
+	int ret;
+
+	ret = nvkm_gpuobj_new(device, 0x0200, 0, true, chan->inst, &chan->eng);
+	if (ret)
+		return ret;
+
+	ret = nvkm_gpuobj_new(device, 0x4000, 0, false, chan->inst, &chan->pgd);
+	if (ret)
+		return ret;
+
+	ret = nvkm_gpuobj_new(device, 0x1000, 0x400, true, chan->inst, &chan->cache);
+	if (ret)
+		return ret;
+
+	ret = nvkm_gpuobj_new(device, 0x100, 0x100, true, chan->inst, &chan->ramfc);
+	if (ret)
+		return ret;
+
+	ret = nvkm_ramht_new(device, 0x8000, 16, chan->inst, &chan->ramht);
+	if (ret)
+		return ret;
+
+	nvkm_kmap(chan->ramfc);
+	nvkm_wo32(chan->ramfc, 0x3c, 0x403f6078);
+	nvkm_wo32(chan->ramfc, 0x44, 0x01003fff);
+	nvkm_wo32(chan->ramfc, 0x48, chan->push->node->offset >> 4);
+	nvkm_wo32(chan->ramfc, 0x50, lower_32_bits(offset));
+	nvkm_wo32(chan->ramfc, 0x54, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->ramfc, 0x60, 0x7fffffff);
+	nvkm_wo32(chan->ramfc, 0x78, 0x00000000);
+	nvkm_wo32(chan->ramfc, 0x7c, 0x30000000 | devm);
+	nvkm_wo32(chan->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				     (4 << 24) /* SEARCH_FULL */ |
+				     (chan->ramht->gpuobj->node->offset >> 4));
+	nvkm_wo32(chan->ramfc, 0x88, chan->cache->addr >> 10);
+	nvkm_wo32(chan->ramfc, 0x98, chan->inst->addr >> 12);
+	nvkm_done(chan->ramfc);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+g84_chan_ramfc = {
+	.write = g84_chan_ramfc_write,
+	.ctxdma = true,
+	.devm = 0xfff,
+};
+
+const struct nvkm_chan_func
+g84_chan = {
+	.inst = &nv50_chan_inst,
+	.userd = &nv50_chan_userd,
+	.ramfc = &g84_chan_ramfc,
+	.bind = g84_chan_bind,
+	.unbind = nv50_chan_unbind,
+	.start = nv50_chan_start,
+	.stop = nv50_chan_stop,
+};
+
+static void
+g84_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	struct nvkm_subdev *subdev = &chan->cgrp->runl->fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u64 start = 0, limit = 0;
+	u32 flags = 0, ptr0, save;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_GR    : ptr0 = 0x0020; break;
+	case NVKM_ENGINE_VP    :
+	case NVKM_ENGINE_MSPDEC: ptr0 = 0x0040; break;
+	case NVKM_ENGINE_MPEG  :
+	case NVKM_ENGINE_MSPPP : ptr0 = 0x0060; break;
+	case NVKM_ENGINE_BSP   :
+	case NVKM_ENGINE_MSVLD : ptr0 = 0x0080; break;
+	case NVKM_ENGINE_CIPHER:
+	case NVKM_ENGINE_SEC   : ptr0 = 0x00a0; break;
+	case NVKM_ENGINE_CE    : ptr0 = 0x00c0; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (!cctx) {
+		save = nvkm_mask(device, 0x002520, 0x0000003f, BIT(engn->id - 1));
+		nvkm_wr32(device, 0x0032fc, chan->inst->addr >> 12);
+		nvkm_msec(device, 2000,
+			if (nvkm_rd32(device, 0x0032fc) != 0xffffffff)
+				break;
+		);
+		nvkm_wr32(device, 0x002520, save);
+	} else {
+		flags = 0x00190000;
+		start = cctx->vctx->inst->addr;
+		limit = start + cctx->vctx->inst->size - 1;
+	}
+
+	nvkm_kmap(chan->eng);
+	nvkm_wo32(chan->eng, ptr0 + 0x00, flags);
+	nvkm_wo32(chan->eng, ptr0 + 0x04, lower_32_bits(limit));
+	nvkm_wo32(chan->eng, ptr0 + 0x08, lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x0c, upper_32_bits(limit) << 24 |
+					  lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x10, 0x00000000);
+	nvkm_wo32(chan->eng, ptr0 + 0x14, 0x00000000);
+	nvkm_done(chan->eng);
+}
+
+const struct nvkm_engn_func
+g84_engn = {
+	.bind = g84_ectx_bind,
+	.ramht_add = nv50_eobj_ramht_add,
+	.ramht_del = nv50_eobj_ramht_del,
+};
+
+static void
+g84_fifo_nonstall_block(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	nvkm_mask(fifo->engine.subdev.device, 0x002140, 0x40000000, 0x00000000);
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+static void
+g84_fifo_nonstall_allow(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	nvkm_mask(fifo->engine.subdev.device, 0x002140, 0x40000000, 0x40000000);
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+const struct nvkm_event_func
+g84_fifo_nonstall = {
+	.init = g84_fifo_nonstall_allow,
+	.fini = g84_fifo_nonstall_block,
+};
+
+static int
+g84_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_runl *runl;
+
+	runl = nvkm_runl_new(fifo, 0, 0, 0);
+	if (IS_ERR(runl))
+		return PTR_ERR(runl);
+
+	nvkm_runl_add(runl, 0, fifo->func->engn_sw, NVKM_ENGINE_SW, 0);
+	nvkm_runl_add(runl, 0, fifo->func->engn_sw, NVKM_ENGINE_DMAOBJ, 0);
+	nvkm_runl_add(runl, 1, fifo->func->engn, NVKM_ENGINE_GR, 0);
+	nvkm_runl_add(runl, 2, fifo->func->engn, NVKM_ENGINE_MPEG, 0);
+	nvkm_runl_add(runl, 3, fifo->func->engn, NVKM_ENGINE_ME, 0);
+	nvkm_runl_add(runl, 4, fifo->func->engn, NVKM_ENGINE_VP, 0);
+	nvkm_runl_add(runl, 5, fifo->func->engn, NVKM_ENGINE_CIPHER, 0);
+	nvkm_runl_add(runl, 6, fifo->func->engn, NVKM_ENGINE_BSP, 0);
+	return 0;
+}
+
+static const struct nvkm_fifo_func
+g84_fifo = {
+	.chid_nr = nv50_fifo_chid_nr,
+	.chid_ctor = nv50_fifo_chid_ctor,
+	.runl_ctor = g84_fifo_runl_ctor,
+	.init = nv50_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.nonstall = &g84_fifo_nonstall,
+	.runl = &nv50_runl,
+	.engn = &g84_engn,
+	.engn_sw = &nv50_engn_sw,
+	.cgrp = {{                          }, &nv04_cgrp },
+	.chan = {{ 0, 0, G82_CHANNEL_GPFIFO }, &g84_chan },
+};
+
+int
+g84_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	     struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&g84_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c
new file mode 100644
index 000000000..c6ca050c3
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c
@@ -0,0 +1,70 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+#include "chan.h"
+#include "runl.h"
+
+#include <nvif/class.h>
+
+static int
+g98_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_runl *runl;
+
+	runl = nvkm_runl_new(fifo, 0, 0, 0);
+	if (IS_ERR(runl))
+		return PTR_ERR(runl);
+
+	nvkm_runl_add(runl, 0, fifo->func->engn_sw, NVKM_ENGINE_SW, 0);
+	nvkm_runl_add(runl, 0, fifo->func->engn_sw, NVKM_ENGINE_DMAOBJ, 0);
+	nvkm_runl_add(runl, 1, fifo->func->engn, NVKM_ENGINE_GR, 0);
+	nvkm_runl_add(runl, 2, fifo->func->engn, NVKM_ENGINE_MSPPP, 0);
+	nvkm_runl_add(runl, 3, fifo->func->engn, NVKM_ENGINE_CE, 0);
+	nvkm_runl_add(runl, 4, fifo->func->engn, NVKM_ENGINE_MSPDEC, 0);
+	nvkm_runl_add(runl, 5, fifo->func->engn, NVKM_ENGINE_SEC, 0);
+	nvkm_runl_add(runl, 6, fifo->func->engn, NVKM_ENGINE_MSVLD, 0);
+	return 0;
+}
+
+static const struct nvkm_fifo_func
+g98_fifo = {
+	.chid_nr = nv50_fifo_chid_nr,
+	.chid_ctor = nv50_fifo_chid_ctor,
+	.runl_ctor = g98_fifo_runl_ctor,
+	.init = nv50_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.nonstall = &g84_fifo_nonstall,
+	.runl = &nv50_runl,
+	.engn = &g84_engn,
+	.engn_sw = &nv50_engn_sw,
+	.cgrp = {{                          }, &nv04_cgrp },
+	.chan = {{ 0, 0, G82_CHANNEL_GPFIFO }, &g84_chan },
+};
+
+int
+g98_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&g98_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c
new file mode 100644
index 000000000..12a5d99d5
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c
@@ -0,0 +1,550 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/top.h>
+#include <subdev/vfn.h>
+
+#include <nvif/class.h>
+
+/*TODO: allocate? */
+#define GA100_FIFO_NONSTALL_VECTOR 0
+
+static u32
+ga100_chan_doorbell_handle(struct nvkm_chan *chan)
+{
+	return (chan->cgrp->runl->doorbell << 16) | chan->id;
+}
+
+static void
+ga100_chan_stop(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->chan + (chan->id * 4), 0x00000003);
+}
+
+static void
+ga100_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	const int gfid = 0;
+
+	nvkm_wr32(device, runl->chan + (chan->id * 4), 0x00000002);
+	nvkm_wr32(device, runl->addr + 0x0090, (gfid << 16) | chan->id); /* INTERNAL_DOORBELL. */
+}
+
+static void
+ga100_chan_unbind(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->chan + (chan->id * 4), 0xffffffff);
+}
+
+static int
+ga100_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	const u32 limit2 = ilog2(length / 8);
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x010, 0x0000face);
+	nvkm_wo32(chan->inst, 0x030, 0x7ffff902);
+	nvkm_wo32(chan->inst, 0x048, lower_32_bits(offset));
+	nvkm_wo32(chan->inst, 0x04c, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->inst, 0x084, 0x20400000);
+	nvkm_wo32(chan->inst, 0x094, 0x30000000 | devm);
+	nvkm_wo32(chan->inst, 0x0e4, priv ? 0x00000020 : 0x00000000);
+	nvkm_wo32(chan->inst, 0x0e8, chan->id);
+	nvkm_wo32(chan->inst, 0x0f4, 0x00001000 | (priv ? 0x00000100 : 0x00000000));
+	nvkm_wo32(chan->inst, 0x0f8, 0x80000000 | GA100_FIFO_NONSTALL_VECTOR);
+	nvkm_mo32(chan->inst, 0x218, 0x00000000, 0x00000000);
+	nvkm_done(chan->inst);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+ga100_chan_ramfc = {
+	.write = ga100_chan_ramfc_write,
+	.devm = 0xfff,
+	.priv = true,
+};
+
+const struct nvkm_chan_func
+ga100_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gv100_chan_userd,
+	.ramfc = &ga100_chan_ramfc,
+	.unbind = ga100_chan_unbind,
+	.start = ga100_chan_start,
+	.stop = ga100_chan_stop,
+	.preempt = gk110_chan_preempt,
+	.doorbell_handle = ga100_chan_doorbell_handle,
+};
+
+static void
+ga100_cgrp_preempt(struct nvkm_cgrp *cgrp)
+{
+	struct nvkm_runl *runl = cgrp->runl;
+
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->addr + 0x098, 0x01000000 | cgrp->id);
+}
+
+const struct nvkm_cgrp_func
+ga100_cgrp = {
+	.preempt = ga100_cgrp_preempt,
+};
+
+static int
+ga100_engn_cxid(struct nvkm_engn *engn, bool *cgid)
+{
+	struct nvkm_runl *runl = engn->runl;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u32 stat = nvkm_rd32(device, runl->addr + 0x200 + engn->id * 0x40);
+
+	ENGN_DEBUG(engn, "status %08x", stat);
+	*cgid = true;
+
+	switch ((stat & 0x0000e000) >> 13) {
+	case 0 /* INVALID */: return -ENODEV;
+	case 1 /*   VALID */:
+	case 5 /*    SAVE */: return (stat & 0x00000fff);
+	case 6 /*    LOAD */: return (stat & 0x0fff0000) >> 16;
+	case 7 /*  SWITCH */:
+		if (nvkm_engine_chsw_load(engn->engine))
+			return (stat & 0x0fff0000) >> 16;
+		return (stat & 0x00000fff);
+	default:
+		WARN_ON(1);
+		break;
+	}
+
+	return -ENODEV;
+}
+
+const struct nvkm_engn_func
+ga100_engn = {
+	.cxid = ga100_engn_cxid,
+	.ctor = gk104_ectx_ctor,
+	.bind = gv100_ectx_bind,
+};
+
+const struct nvkm_engn_func
+ga100_engn_ce = {
+	.cxid = ga100_engn_cxid,
+	.ctor = gv100_ectx_ce_ctor,
+	.bind = gv100_ectx_ce_bind,
+};
+
+static bool
+ga100_runq_idle(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	return !(nvkm_rd32(device, 0x04015c + (runq->id * 0x800)) & 0x0000e000);
+}
+
+static bool
+ga100_runq_intr_1(struct nvkm_runq *runq, struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+	u32 inte = nvkm_rd32(device, 0x040180 + (runq->id * 0x800));
+	u32 intr = nvkm_rd32(device, 0x040148 + (runq->id * 0x800));
+	u32 stat = intr & inte;
+
+	if (!stat) {
+		RUNQ_DEBUG(runq, "inte1 %08x %08x", intr, inte);
+		return false;
+	}
+
+	if (stat & 0x80000000) {
+		u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x0800)) & runl->chid->mask;
+		struct nvkm_chan *chan;
+		unsigned long flags;
+
+		RUNQ_ERROR(runq, "CTXNOTVALID chid:%d", chid);
+		chan = nvkm_runl_chan_get_chid(runl, chid, &flags);
+		if (chan) {
+			nvkm_chan_error(chan, true);
+			nvkm_chan_put(&chan, flags);
+		}
+
+		nvkm_mask(device, 0x0400ac + (runq->id * 0x800), 0x00030000, 0x00030000);
+		stat &= ~0x80000000;
+	}
+
+	if (stat) {
+		RUNQ_ERROR(runq, "intr1 %08x", stat);
+		nvkm_wr32(device, 0x0401a0 + (runq->id * 0x800), stat);
+	}
+
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x800), intr);
+	return true;
+}
+
+static bool
+ga100_runq_intr_0(struct nvkm_runq *runq, struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+	u32 inte = nvkm_rd32(device, 0x040170 + (runq->id * 0x800));
+	u32 intr = nvkm_rd32(device, 0x040108 + (runq->id * 0x800));
+	u32 stat = intr & inte;
+
+	if (!stat) {
+		RUNQ_DEBUG(runq, "inte0 %08x %08x", intr, inte);
+		return false;
+	}
+
+	/*TODO: expand on this when fixing up gf100's version. */
+	if (stat & 0xc6afe000) {
+		u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x0800)) & runl->chid->mask;
+		struct nvkm_chan *chan;
+		unsigned long flags;
+
+		RUNQ_ERROR(runq, "intr0 %08x", stat);
+		chan = nvkm_runl_chan_get_chid(runl, chid, &flags);
+		if (chan) {
+			nvkm_chan_error(chan, true);
+			nvkm_chan_put(&chan, flags);
+		}
+
+		stat &= ~0xc6afe000;
+	}
+
+	if (stat) {
+		RUNQ_ERROR(runq, "intr0 %08x", stat);
+		nvkm_wr32(device, 0x040190 + (runq->id * 0x800), stat);
+	}
+
+	nvkm_wr32(device, 0x040108 + (runq->id * 0x800), intr);
+	return true;
+}
+
+static bool
+ga100_runq_intr(struct nvkm_runq *runq, struct nvkm_runl *runl)
+{
+	bool intr0 = ga100_runq_intr_0(runq, runl);
+	bool intr1 = ga100_runq_intr_1(runq, runl);
+
+	return intr0 || intr1;
+}
+
+static void
+ga100_runq_init(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x040108 + (runq->id * 0x800), 0xffffffff); /* INTR_0 */
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x800), 0xffffffff); /* INTR_1 */
+	nvkm_wr32(device, 0x040170 + (runq->id * 0x800), 0xffffffff); /* INTR_0_EN_SET_TREE */
+	nvkm_wr32(device, 0x040180 + (runq->id * 0x800), 0xffffffff); /* INTR_1_EN_SET_TREE */
+}
+
+const struct nvkm_runq_func
+ga100_runq = {
+	.init = ga100_runq_init,
+	.intr = ga100_runq_intr,
+	.idle = ga100_runq_idle,
+};
+
+static bool
+ga100_runl_preempt_pending(struct nvkm_runl *runl)
+{
+	return nvkm_rd32(runl->fifo->engine.subdev.device, runl->addr + 0x098) & 0x00100000;
+}
+
+static void
+ga100_runl_preempt(struct nvkm_runl *runl)
+{
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->addr + 0x098, 0x00000000);
+}
+
+static void
+ga100_runl_allow(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, runl->addr + 0x094, 0x00000001, 0x00000000);
+}
+
+static void
+ga100_runl_block(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, runl->addr + 0x094, 0x00000001, 0x00000001);
+}
+
+static bool
+ga100_runl_pending(struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+
+	return nvkm_rd32(device, runl->addr + 0x08c) & 0x00008000;
+}
+
+static void
+ga100_runl_commit(struct nvkm_runl *runl, struct nvkm_memory *memory, u32 start, int count)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u64 addr = nvkm_memory_addr(memory) + start;
+
+	nvkm_wr32(device, runl->addr + 0x080, lower_32_bits(addr));
+	nvkm_wr32(device, runl->addr + 0x084, upper_32_bits(addr));
+	nvkm_wr32(device, runl->addr + 0x088, count);
+}
+
+static irqreturn_t
+ga100_runl_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_runl *runl = container_of(inth, typeof(*runl), inth);
+	struct nvkm_engn *engn;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u32 inte = nvkm_rd32(device, runl->addr + 0x120);
+	u32 intr = nvkm_rd32(device, runl->addr + 0x100);
+	u32 stat = intr & inte;
+	u32 info;
+
+	if (!stat) {
+		RUNL_DEBUG(runl, "inte %08x %08x", intr, inte);
+		return IRQ_NONE;
+	}
+
+	if (stat & 0x00000007) {
+		nvkm_runl_foreach_engn_cond(engn, runl, stat & BIT(engn->id)) {
+			info = nvkm_rd32(device, runl->addr + 0x224 + (engn->id * 0x40));
+
+			tu102_fifo_intr_ctxsw_timeout_info(engn, info);
+
+			nvkm_wr32(device, runl->addr + 0x100, BIT(engn->id));
+			stat &= ~BIT(engn->id);
+		}
+	}
+
+	if (stat & 0x00000300) {
+		nvkm_wr32(device, runl->addr + 0x100, stat & 0x00000300);
+		stat &= ~0x00000300;
+	}
+
+	if (stat & 0x00010000) {
+		if (runl->runq[0]) {
+			if (runl->runq[0]->func->intr(runl->runq[0], runl))
+				stat &= ~0x00010000;
+		}
+	}
+
+	if (stat & 0x00020000) {
+		if (runl->runq[1]) {
+			if (runl->runq[1]->func->intr(runl->runq[1], runl))
+				stat &= ~0x00020000;
+		}
+	}
+
+	if (stat) {
+		RUNL_ERROR(runl, "intr %08x", stat);
+		nvkm_wr32(device, runl->addr + 0x140, stat);
+	}
+
+	nvkm_wr32(device, runl->addr + 0x180, 0x00000001);
+	return IRQ_HANDLED;
+}
+
+static void
+ga100_runl_fini(struct nvkm_runl *runl)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, runl->addr + 0x300, 0x80000000, 0x00000000);
+	nvkm_inth_block(&runl->inth);
+}
+
+static void
+ga100_runl_init(struct nvkm_runl *runl)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_runq *runq;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	int i;
+
+	/* Submit NULL runlist and preempt. */
+	nvkm_wr32(device, runl->addr + 0x088, 0x00000000);
+	runl->func->preempt(runl);
+
+	/* Enable doorbell. */
+	nvkm_mask(device, runl->addr + 0x300, 0x80000000, 0x80000000);
+
+	nvkm_wr32(device, runl->addr + 0x100, 0xffffffff); /* INTR_0 */
+	nvkm_wr32(device, runl->addr + 0x140, 0xffffffff); /* INTR_0_EN_CLEAR_TREE(0) */
+	nvkm_wr32(device, runl->addr + 0x120, 0x000f1307); /* INTR_0_EN_SET_TREE(0) */
+	nvkm_wr32(device, runl->addr + 0x148, 0xffffffff); /* INTR_0_EN_CLEAR_TREE(1) */
+	nvkm_wr32(device, runl->addr + 0x128, 0x00000000); /* INTR_0_EN_SET_TREE(1) */
+
+	/* Init PBDMA(s). */
+	for (i = 0; i < runl->runq_nr; i++) {
+		runq = runl->runq[i];
+		runq->func->init(runq);
+	}
+
+	nvkm_inth_allow(&runl->inth);
+}
+
+const struct nvkm_runl_func
+ga100_runl = {
+	.init = ga100_runl_init,
+	.fini = ga100_runl_fini,
+	.size = 16,
+	.update = nv50_runl_update,
+	.insert_cgrp = gv100_runl_insert_cgrp,
+	.insert_chan = gv100_runl_insert_chan,
+	.commit = ga100_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = ga100_runl_pending,
+	.block = ga100_runl_block,
+	.allow = ga100_runl_allow,
+	.preempt = ga100_runl_preempt,
+	.preempt_pending = ga100_runl_preempt_pending,
+};
+
+static int
+ga100_runl_new(struct nvkm_fifo *fifo, int id, u32 addr, struct nvkm_runl **prunl)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_runl *runl;
+	u32 chcfg  = nvkm_rd32(device, addr + 0x004);
+	u32 chnum  = 1 << (chcfg & 0x0000000f);
+	u32 chaddr = (chcfg & 0xfffffff0);
+	u32 dbcfg  = nvkm_rd32(device, addr + 0x008);
+	u32 vector = nvkm_rd32(device, addr + 0x160);
+	int i, ret;
+
+	runl = *prunl = nvkm_runl_new(fifo, id, addr, chnum);
+	if (IS_ERR(runl))
+		return PTR_ERR(runl);
+
+	for (i = 0; i < 2; i++) {
+		u32 pbcfg = nvkm_rd32(device, addr + 0x010 + (i * 0x04));
+		if (pbcfg & 0x80000000) {
+			runl->runq[runl->runq_nr] =
+				nvkm_runq_new(fifo, ((pbcfg & 0x03fffc00) - 0x040000) / 0x800);
+			if (!runl->runq[runl->runq_nr])
+				return -ENOMEM;
+
+			runl->runq_nr++;
+		}
+	}
+
+	ret = nvkm_inth_add(&device->vfn->intr, vector & 0x00000fff, NVKM_INTR_PRIO_NORMAL,
+			    &fifo->engine.subdev, ga100_runl_intr, &runl->inth);
+	if (ret)
+		return ret;
+
+	runl->chan = chaddr;
+	runl->doorbell = dbcfg >> 16;
+	return 0;
+}
+
+static irqreturn_t
+ga100_fifo_nonstall_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_fifo *fifo = container_of(inth, typeof(*fifo), nonstall.intr);
+
+	nvkm_event_ntfy(&fifo->nonstall.event, 0, NVKM_FIFO_NONSTALL_EVENT);
+	return IRQ_HANDLED;
+}
+
+static void
+ga100_fifo_nonstall_block(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+
+	nvkm_inth_block(&fifo->nonstall.intr);
+}
+
+static void
+ga100_fifo_nonstall_allow(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+
+	nvkm_inth_allow(&fifo->nonstall.intr);
+}
+
+const struct nvkm_event_func
+ga100_fifo_nonstall = {
+	.init = ga100_fifo_nonstall_allow,
+	.fini = ga100_fifo_nonstall_block,
+};
+
+int
+ga100_fifo_nonstall_ctor(struct nvkm_fifo *fifo)
+{
+	return nvkm_inth_add(&fifo->engine.subdev.device->vfn->intr, GA100_FIFO_NONSTALL_VECTOR,
+			     NVKM_INTR_PRIO_NORMAL, &fifo->engine.subdev, ga100_fifo_nonstall_intr,
+			     &fifo->nonstall.intr);
+}
+
+int
+ga100_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_top_device *tdev;
+	struct nvkm_runl *runl;
+	int id = 0, ret;
+
+	nvkm_list_foreach(tdev, &device->top->device, head, tdev->runlist >= 0) {
+		runl = nvkm_runl_get(fifo, -1, tdev->runlist);
+		if (!runl) {
+			ret = ga100_runl_new(fifo, id++, tdev->runlist, &runl);
+			if (ret)
+				return ret;
+		}
+
+		if (tdev->engine < 0)
+			continue;
+
+		nvkm_runl_add(runl, tdev->engine, (tdev->type == NVKM_ENGINE_CE) ?
+			      fifo->func->engn_ce : fifo->func->engn, tdev->type, tdev->inst);
+	}
+
+	return 0;
+}
+
+static const struct nvkm_fifo_func
+ga100_fifo = {
+	.runl_ctor = ga100_fifo_runl_ctor,
+	.mmu_fault = &tu102_fifo_mmu_fault,
+	.nonstall_ctor = ga100_fifo_nonstall_ctor,
+	.nonstall = &ga100_fifo_nonstall,
+	.runl = &ga100_runl,
+	.runq = &ga100_runq,
+	.engn = &ga100_engn,
+	.engn_ce = &ga100_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &ga100_cgrp, .force = true },
+	.chan = {{ 0, 0, AMPERE_CHANNEL_GPFIFO_A }, &ga100_chan },
+};
+
+int
+ga100_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&ga100_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c
new file mode 100644
index 000000000..2cdf5da33
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c
@@ -0,0 +1,45 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+
+#include <nvif/class.h>
+
+static const struct nvkm_fifo_func
+ga102_fifo = {
+	.runl_ctor = ga100_fifo_runl_ctor,
+	.mmu_fault = &tu102_fifo_mmu_fault,
+	.nonstall_ctor = ga100_fifo_nonstall_ctor,
+	.nonstall = &ga100_fifo_nonstall,
+	.runl = &ga100_runl,
+	.runq = &ga100_runq,
+	.engn = &ga100_engn,
+	.engn_ce = &ga100_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &ga100_cgrp, .force = true },
+	.chan = {{ 0, 0, AMPERE_CHANNEL_GPFIFO_B }, &ga100_chan },
+};
+
+int
+ga102_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&ga102_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
new file mode 100644
index 000000000..5bb65258c
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
@@ -0,0 +1,969 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/bar.h>
+#include <subdev/fault.h>
+#include <subdev/mc.h>
+#include <subdev/mmu.h>
+#include <engine/sw.h>
+
+#include <nvif/class.h>
+
+void
+gf100_chan_preempt(struct nvkm_chan *chan)
+{
+	nvkm_wr32(chan->cgrp->runl->fifo->engine.subdev.device, 0x002634, chan->id);
+}
+
+static void
+gf100_chan_stop(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x003004 + (chan->id * 8), 0x00000001, 0x00000000);
+}
+
+static void
+gf100_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x003004 + (chan->id * 8), 0x001f0001);
+}
+
+static void gf100_fifo_intr_engine(struct nvkm_fifo *);
+
+static void
+gf100_chan_unbind(struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+
+	/*TODO: Is this cargo-culted, or necessary? RM does *something* here... Why? */
+	gf100_fifo_intr_engine(fifo);
+
+	nvkm_wr32(device, 0x003000 + (chan->id * 8), 0x00000000);
+}
+
+static void
+gf100_chan_bind(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x003000 + (chan->id * 8), 0xc0000000 | chan->inst->addr >> 12);
+}
+
+static int
+gf100_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	const u64 userd = nvkm_memory_addr(chan->userd.mem) + chan->userd.base;
+	const u32 limit2 = ilog2(length / 8);
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x08, lower_32_bits(userd));
+	nvkm_wo32(chan->inst, 0x0c, upper_32_bits(userd));
+	nvkm_wo32(chan->inst, 0x10, 0x0000face);
+	nvkm_wo32(chan->inst, 0x30, 0xfffff902);
+	nvkm_wo32(chan->inst, 0x48, lower_32_bits(offset));
+	nvkm_wo32(chan->inst, 0x4c, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->inst, 0x54, 0x00000002);
+	nvkm_wo32(chan->inst, 0x84, 0x20400000);
+	nvkm_wo32(chan->inst, 0x94, 0x30000000 | devm);
+	nvkm_wo32(chan->inst, 0x9c, 0x00000100);
+	nvkm_wo32(chan->inst, 0xa4, 0x1f1f1f1f);
+	nvkm_wo32(chan->inst, 0xa8, 0x1f1f1f1f);
+	nvkm_wo32(chan->inst, 0xac, 0x0000001f);
+	nvkm_wo32(chan->inst, 0xb8, 0xf8000000);
+	nvkm_wo32(chan->inst, 0xf8, 0x10003080); /* 0x002310 */
+	nvkm_wo32(chan->inst, 0xfc, 0x10000010); /* 0x002350 */
+	nvkm_done(chan->inst);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+gf100_chan_ramfc = {
+	.write = gf100_chan_ramfc_write,
+	.devm = 0xfff,
+};
+
+void
+gf100_chan_userd_clear(struct nvkm_chan *chan)
+{
+	nvkm_kmap(chan->userd.mem);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x040, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x044, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x048, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x04c, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x050, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x058, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x05c, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x060, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x088, 0x00000000);
+	nvkm_wo32(chan->userd.mem, chan->userd.base + 0x08c, 0x00000000);
+	nvkm_done(chan->userd.mem);
+}
+
+static const struct nvkm_chan_func_userd
+gf100_chan_userd = {
+	.bar = 1,
+	.size = 0x1000,
+	.clear = gf100_chan_userd_clear,
+};
+
+const struct nvkm_chan_func_inst
+gf100_chan_inst = {
+	.size = 0x1000,
+	.zero = true,
+	.vmm = true,
+};
+
+static const struct nvkm_chan_func
+gf100_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gf100_chan_userd,
+	.ramfc = &gf100_chan_ramfc,
+	.bind = gf100_chan_bind,
+	.unbind = gf100_chan_unbind,
+	.start = gf100_chan_start,
+	.stop = gf100_chan_stop,
+	.preempt = gf100_chan_preempt,
+};
+
+static void
+gf100_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	u64 addr = 0ULL;
+	u32 ptr0;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_SW    : return;
+	case NVKM_ENGINE_GR    : ptr0 = 0x0210; break;
+	case NVKM_ENGINE_CE    : ptr0 = 0x0230 + (engn->engine->subdev.inst * 0x10); break;
+	case NVKM_ENGINE_MSPDEC: ptr0 = 0x0250; break;
+	case NVKM_ENGINE_MSPPP : ptr0 = 0x0260; break;
+	case NVKM_ENGINE_MSVLD : ptr0 = 0x0270; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (cctx) {
+		addr  = cctx->vctx->vma->addr;
+		addr |= 4ULL;
+	}
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, ptr0 + 0, lower_32_bits(addr));
+	nvkm_wo32(chan->inst, ptr0 + 4, upper_32_bits(addr));
+	nvkm_done(chan->inst);
+}
+
+static int
+gf100_ectx_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx)
+{
+	int ret;
+
+	ret = nvkm_vmm_get(vctx->vmm, 12, vctx->inst->size, &vctx->vma);
+	if (ret)
+		return ret;
+
+	return nvkm_memory_map(vctx->inst, 0, vctx->vmm, vctx->vma, NULL, 0);
+}
+
+bool
+gf100_engn_mmu_fault_triggered(struct nvkm_engn *engn)
+{
+	struct nvkm_runl *runl = engn->runl;
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u32 data = nvkm_rd32(device, 0x002a30 + (engn->id * 4));
+
+	ENGN_DEBUG(engn, "%08x: mmu fault triggered", data);
+	if (!(data & 0x00000100))
+		return false;
+
+	spin_lock(&fifo->lock);
+	nvkm_mask(device, 0x002a30 + (engn->id * 4), 0x00000100, 0x00000000);
+	if (atomic_dec_and_test(&runl->rc_triggered))
+		nvkm_mask(device, 0x002140, 0x00000100, 0x00000100);
+	spin_unlock(&fifo->lock);
+	return true;
+}
+
+void
+gf100_engn_mmu_fault_trigger(struct nvkm_engn *engn)
+{
+	struct nvkm_runl *runl = engn->runl;
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+
+	ENGN_DEBUG(engn, "triggering mmu fault on 0x%02x", engn->fault);
+	spin_lock(&fifo->lock);
+	if (atomic_inc_return(&runl->rc_triggered) == 1)
+		nvkm_mask(device, 0x002140, 0x00000100, 0x00000000);
+	nvkm_wr32(device, 0x002100, 0x00000100);
+	nvkm_wr32(device, 0x002a30 + (engn->id * 4), 0x00000100 | engn->fault);
+	spin_unlock(&fifo->lock);
+}
+
+/*TODO: clean all this up. */
+struct gf100_engn_status {
+	bool busy;
+	bool save;
+	bool unk0;
+	bool unk1;
+	u8   chid;
+};
+
+static void
+gf100_engn_status(struct nvkm_engn *engn, struct gf100_engn_status *status)
+{
+	u32 stat = nvkm_rd32(engn->engine->subdev.device, 0x002640 + (engn->id * 4));
+
+	status->busy = (stat & 0x10000000);
+	status->save = (stat & 0x00100000);
+	status->unk0 = (stat & 0x00004000);
+	status->unk1 = (stat & 0x00001000);
+	status->chid = (stat & 0x0000007f);
+
+	ENGN_DEBUG(engn, "%08x: busy %d save %d unk0 %d unk1 %d chid %d",
+		   stat, status->busy, status->save, status->unk0, status->unk1, status->chid);
+}
+
+static int
+gf100_engn_cxid(struct nvkm_engn *engn, bool *cgid)
+{
+	struct gf100_engn_status status;
+
+	gf100_engn_status(engn, &status);
+	if (status.busy) {
+		*cgid = false;
+		return status.chid;
+	}
+
+	return -ENODEV;
+}
+
+static bool
+gf100_engn_chsw(struct nvkm_engn *engn)
+{
+	struct gf100_engn_status status;
+
+	gf100_engn_status(engn, &status);
+	if (status.busy && (status.unk0 || status.unk1))
+		return true;
+
+	return false;
+}
+
+static const struct nvkm_engn_func
+gf100_engn = {
+	.chsw = gf100_engn_chsw,
+	.cxid = gf100_engn_cxid,
+	.mmu_fault_trigger = gf100_engn_mmu_fault_trigger,
+	.mmu_fault_triggered = gf100_engn_mmu_fault_triggered,
+	.ctor = gf100_ectx_ctor,
+	.bind = gf100_ectx_bind,
+};
+
+const struct nvkm_engn_func
+gf100_engn_sw = {
+};
+
+static const struct nvkm_bitfield
+gf100_runq_intr_0_names[] = {
+/*	{ 0x00008000, "" }	seen with null ib push */
+	{ 0x00200000, "ILLEGAL_MTHD" },
+	{ 0x00800000, "EMPTY_SUBC" },
+	{}
+};
+
+bool
+gf100_runq_intr(struct nvkm_runq *runq, struct nvkm_runl *null)
+{
+	struct nvkm_subdev *subdev = &runq->fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, 0x04010c + (runq->id * 0x2000));
+	u32 stat = nvkm_rd32(device, 0x040108 + (runq->id * 0x2000)) & mask;
+	u32 addr = nvkm_rd32(device, 0x0400c0 + (runq->id * 0x2000));
+	u32 data = nvkm_rd32(device, 0x0400c4 + (runq->id * 0x2000));
+	u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x2000)) & runq->fifo->chid->mask;
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00003ffc);
+	u32 show = stat;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+	char msg[128];
+
+	if (stat & 0x00800000) {
+		if (device->sw) {
+			if (nvkm_sw_mthd(device->sw, chid, subc, mthd, data))
+				show &= ~0x00800000;
+		}
+	}
+
+	if (show) {
+		nvkm_snprintbf(msg, sizeof(msg), runq->func->intr_0_names, show);
+		chan = nvkm_chan_get_chid(&runq->fifo->engine, chid, &flags);
+		nvkm_error(subdev, "PBDMA%d: %08x [%s] ch %d [%010llx %s] "
+				   "subc %d mthd %04x data %08x\n",
+			   runq->id, show, msg, chid, chan ? chan->inst->addr : 0,
+			   chan ? chan->name : "unknown", subc, mthd, data);
+
+		/*TODO: use proper procedure for clearing each exception / debug output */
+		if ((stat & 0xc67fe000) && chan)
+			nvkm_chan_error(chan, true);
+		nvkm_chan_put(&chan, flags);
+	}
+
+	nvkm_wr32(device, 0x0400c0 + (runq->id * 0x2000), 0x80600008);
+	nvkm_wr32(device, 0x040108 + (runq->id * 0x2000), stat);
+	return true;
+}
+
+void
+gf100_runq_init(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x04013c + (runq->id * 0x2000), 0x10000100, 0x00000000);
+	nvkm_wr32(device, 0x040108 + (runq->id * 0x2000), 0xffffffff); /* INTR */
+	nvkm_wr32(device, 0x04010c + (runq->id * 0x2000), 0xfffffeff); /* INTREN */
+}
+
+static const struct nvkm_runq_func
+gf100_runq = {
+	.init = gf100_runq_init,
+	.intr = gf100_runq_intr,
+	.intr_0_names = gf100_runq_intr_0_names,
+};
+
+bool
+gf100_runl_preempt_pending(struct nvkm_runl *runl)
+{
+	return nvkm_rd32(runl->fifo->engine.subdev.device, 0x002634) & 0x00100000;
+}
+
+static void
+gf100_runl_fault_clear(struct nvkm_runl *runl)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, 0x00262c, 0x00000000, 0x00000000);
+}
+
+static void
+gf100_runl_allow(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, 0x002630, engm, 0x00000000);
+}
+
+static void
+gf100_runl_block(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, 0x002630, engm, engm);
+}
+
+static bool
+gf100_runl_pending(struct nvkm_runl *runl)
+{
+	return nvkm_rd32(runl->fifo->engine.subdev.device, 0x00227c) & 0x00100000;
+}
+
+static void
+gf100_runl_commit(struct nvkm_runl *runl, struct nvkm_memory *memory, u32 start, int count)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u64 addr = nvkm_memory_addr(memory) + start;
+	int target;
+
+	switch (nvkm_memory_target(memory)) {
+	case NVKM_MEM_TARGET_VRAM: target = 0; break;
+	case NVKM_MEM_TARGET_NCOH: target = 3; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	nvkm_wr32(device, 0x002270, (target << 28) | (addr >> 12));
+	nvkm_wr32(device, 0x002274, 0x01f00000 | count);
+}
+
+static void
+gf100_runl_insert_chan(struct nvkm_chan *chan, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset + 0, chan->id);
+	nvkm_wo32(memory, offset + 4, 0x00000004);
+}
+
+static const struct nvkm_runl_func
+gf100_runl = {
+	.size = 8,
+	.update = nv50_runl_update,
+	.insert_chan = gf100_runl_insert_chan,
+	.commit = gf100_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = gf100_runl_pending,
+	.block = gf100_runl_block,
+	.allow = gf100_runl_allow,
+	.fault_clear = gf100_runl_fault_clear,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+static void
+gf100_fifo_nonstall_allow(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	nvkm_mask(fifo->engine.subdev.device, 0x002140, 0x80000000, 0x80000000);
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+void
+gf100_fifo_nonstall_block(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	nvkm_mask(fifo->engine.subdev.device, 0x002140, 0x80000000, 0x00000000);
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+const struct nvkm_event_func
+gf100_fifo_nonstall = {
+	.init = gf100_fifo_nonstall_allow,
+	.fini = gf100_fifo_nonstall_block,
+};
+
+static const struct nvkm_enum
+gf100_fifo_mmu_fault_engine[] = {
+	{ 0x00, "PGRAPH", NULL, NVKM_ENGINE_GR },
+	{ 0x03, "PEEPHOLE", NULL, NVKM_ENGINE_IFB },
+	{ 0x04, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0x05, "BAR3", NULL, NVKM_SUBDEV_INSTMEM },
+	{ 0x07, "PFIFO" },
+	{ 0x10, "PMSVLD", NULL, NVKM_ENGINE_MSVLD },
+	{ 0x11, "PMSPPP", NULL, NVKM_ENGINE_MSPPP },
+	{ 0x13, "PCOUNTER" },
+	{ 0x14, "PMSPDEC", NULL, NVKM_ENGINE_MSPDEC },
+	{ 0x15, "PCE0", NULL, NVKM_ENGINE_CE, 0 },
+	{ 0x16, "PCE1", NULL, NVKM_ENGINE_CE, 1 },
+	{ 0x17, "PMU" },
+	{}
+};
+
+static const struct nvkm_enum
+gf100_fifo_mmu_fault_reason[] = {
+	{ 0x00, "PT_NOT_PRESENT" },
+	{ 0x01, "PT_TOO_SHORT" },
+	{ 0x02, "PAGE_NOT_PRESENT" },
+	{ 0x03, "VM_LIMIT_EXCEEDED" },
+	{ 0x04, "NO_CHANNEL" },
+	{ 0x05, "PAGE_SYSTEM_ONLY" },
+	{ 0x06, "PAGE_READ_ONLY" },
+	{ 0x0a, "COMPRESSED_SYSRAM" },
+	{ 0x0c, "INVALID_STORAGE_TYPE" },
+	{}
+};
+
+static const struct nvkm_enum
+gf100_fifo_mmu_fault_hubclient[] = {
+	{ 0x01, "PCOPY0" },
+	{ 0x02, "PCOPY1" },
+	{ 0x04, "DISPATCH" },
+	{ 0x05, "CTXCTL" },
+	{ 0x06, "PFIFO" },
+	{ 0x07, "BAR_READ" },
+	{ 0x08, "BAR_WRITE" },
+	{ 0x0b, "PVP" },
+	{ 0x0c, "PMSPPP" },
+	{ 0x0d, "PMSVLD" },
+	{ 0x11, "PCOUNTER" },
+	{ 0x12, "PMU" },
+	{ 0x14, "CCACHE" },
+	{ 0x15, "CCACHE_POST" },
+	{}
+};
+
+static const struct nvkm_enum
+gf100_fifo_mmu_fault_gpcclient[] = {
+	{ 0x01, "TEX" },
+	{ 0x0c, "ESETUP" },
+	{ 0x0e, "CTXCTL" },
+	{ 0x0f, "PROP" },
+	{}
+};
+
+const struct nvkm_enum
+gf100_fifo_mmu_fault_access[] = {
+	{ 0x00, "READ" },
+	{ 0x01, "WRITE" },
+	{}
+};
+
+void
+gf100_fifo_mmu_fault_recover(struct nvkm_fifo *fifo, struct nvkm_fault_data *info)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	const struct nvkm_enum *er, *ee, *ec, *ea;
+	struct nvkm_engine *engine = NULL;
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+	char ct[8] = "HUB/";
+
+	/* Lookup engine by MMU fault ID. */
+	nvkm_runl_foreach(runl, fifo) {
+		engn = nvkm_runl_find_engn(engn, runl, engn->fault == info->engine);
+		if (engn) {
+			/* Fault triggered by CTXSW_TIMEOUT recovery procedure. */
+			if (engn->func->mmu_fault_triggered &&
+			    engn->func->mmu_fault_triggered(engn)) {
+				nvkm_runl_rc_engn(runl, engn);
+				return;
+			}
+
+			engine = engn->engine;
+			break;
+		}
+	}
+
+	er = nvkm_enum_find(fifo->func->mmu_fault->reason, info->reason);
+	ee = nvkm_enum_find(fifo->func->mmu_fault->engine, info->engine);
+	if (info->hub) {
+		ec = nvkm_enum_find(fifo->func->mmu_fault->hubclient, info->client);
+	} else {
+		ec = nvkm_enum_find(fifo->func->mmu_fault->gpcclient, info->client);
+		snprintf(ct, sizeof(ct), "GPC%d/", info->gpc);
+	}
+	ea = nvkm_enum_find(fifo->func->mmu_fault->access, info->access);
+
+	/* Handle BAR faults. */
+	if (ee && ee->data2) {
+		switch (ee->data2) {
+		case NVKM_SUBDEV_BAR:
+			nvkm_bar_bar1_reset(device);
+			break;
+		case NVKM_SUBDEV_INSTMEM:
+			nvkm_bar_bar2_reset(device);
+			break;
+		case NVKM_ENGINE_IFB:
+			nvkm_mask(device, 0x001718, 0x00000000, 0x00000000);
+			break;
+		default:
+			break;
+		}
+	}
+
+	chan = nvkm_chan_get_inst(&fifo->engine, info->inst, &flags);
+
+	nvkm_error(subdev,
+		   "fault %02x [%s] at %016llx engine %02x [%s] client %02x "
+		   "[%s%s] reason %02x [%s] on channel %d [%010llx %s]\n",
+		   info->access, ea ? ea->name : "", info->addr,
+		   info->engine, ee ? ee->name : engine ? engine->subdev.name : "",
+		   info->client, ct, ec ? ec->name : "",
+		   info->reason, er ? er->name : "",
+		   chan ? chan->id : -1, info->inst, chan ? chan->name : "unknown");
+
+	/* Handle host/engine faults. */
+	if (chan)
+		nvkm_runl_rc_cgrp(chan->cgrp);
+
+	nvkm_chan_put(&chan, flags);
+}
+
+static const struct nvkm_fifo_func_mmu_fault
+gf100_fifo_mmu_fault = {
+	.recover = gf100_fifo_mmu_fault_recover,
+	.access = gf100_fifo_mmu_fault_access,
+	.engine = gf100_fifo_mmu_fault_engine,
+	.reason = gf100_fifo_mmu_fault_reason,
+	.hubclient = gf100_fifo_mmu_fault_hubclient,
+	.gpcclient = gf100_fifo_mmu_fault_gpcclient,
+};
+
+void
+gf100_fifo_intr_ctxsw_timeout(struct nvkm_fifo *fifo, u32 engm)
+{
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn, *engn2;
+	bool cgid, cgid2;
+	int id, id2;
+
+	nvkm_runl_foreach(runl, fifo) {
+		/* Stop the runlist, and go through all engines serving it. */
+		nvkm_runl_block(runl);
+		nvkm_runl_foreach_engn_cond(engn, runl, engm & BIT(engn->id)) {
+			/* Determine what channel (group) the engine is on. */
+			id = engn->func->cxid(engn, &cgid);
+			if (id >= 0) {
+				/* Trigger MMU fault on any engine(s) on that channel (group). */
+				nvkm_runl_foreach_engn_cond(engn2, runl, engn2->func->cxid) {
+					id2 = engn2->func->cxid(engn2, &cgid2);
+					if (cgid2 == cgid && id2 == id)
+						engn2->func->mmu_fault_trigger(engn2);
+				}
+			}
+		}
+		nvkm_runl_allow(runl); /* HW will keep runlist blocked via ERROR_SCHED_DISABLE. */
+	}
+}
+
+static void
+gf100_fifo_intr_sched_ctxsw(struct nvkm_fifo *fifo)
+{
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+	u32 engm = 0;
+
+	/* Look for any engines that are busy, and awaiting chsw ack. */
+	nvkm_runl_foreach(runl, fifo) {
+		nvkm_runl_foreach_engn_cond(engn, runl, engn->func->chsw) {
+			if (WARN_ON(engn->fault < 0) || !engn->func->chsw(engn))
+				continue;
+
+			engm |= BIT(engn->id);
+		}
+	}
+
+	if (!engm)
+		return;
+
+	fifo->func->intr_ctxsw_timeout(fifo, engm);
+}
+
+static const struct nvkm_enum
+gf100_fifo_intr_sched_names[] = {
+	{ 0x0a, "CTXSW_TIMEOUT" },
+	{}
+};
+
+void
+gf100_fifo_intr_sched(struct nvkm_fifo *fifo)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 intr = nvkm_rd32(device, 0x00254c);
+	u32 code = intr & 0x000000ff;
+	const struct nvkm_enum *en;
+
+	en = nvkm_enum_find(gf100_fifo_intr_sched_names, code);
+
+	nvkm_error(subdev, "SCHED_ERROR %02x [%s]\n", code, en ? en->name : "");
+
+	switch (code) {
+	case 0x0a:
+		gf100_fifo_intr_sched_ctxsw(fifo);
+		break;
+	default:
+		break;
+	}
+}
+
+void
+gf100_fifo_intr_mmu_fault_unit(struct nvkm_fifo *fifo, int unit)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u32 inst = nvkm_rd32(device, 0x002800 + (unit * 0x10));
+	u32 valo = nvkm_rd32(device, 0x002804 + (unit * 0x10));
+	u32 vahi = nvkm_rd32(device, 0x002808 + (unit * 0x10));
+	u32 type = nvkm_rd32(device, 0x00280c + (unit * 0x10));
+	struct nvkm_fault_data info;
+
+	info.inst   =  (u64)inst << 12;
+	info.addr   = ((u64)vahi << 32) | valo;
+	info.time   = 0;
+	info.engine = unit;
+	info.valid  = 1;
+	info.gpc    = (type & 0x1f000000) >> 24;
+	info.client = (type & 0x00001f00) >> 8;
+	info.access = (type & 0x00000080) >> 7;
+	info.hub    = (type & 0x00000040) >> 6;
+	info.reason = (type & 0x0000000f);
+
+	nvkm_fifo_fault(fifo, &info);
+}
+
+void
+gf100_fifo_intr_mmu_fault(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	unsigned long mask = nvkm_rd32(device, 0x00259c);
+	int unit;
+
+	for_each_set_bit(unit, &mask, 32) {
+		fifo->func->intr_mmu_fault_unit(fifo, unit);
+		nvkm_wr32(device, 0x00259c, BIT(unit));
+	}
+}
+
+bool
+gf100_fifo_intr_pbdma(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_runq *runq;
+	u32 mask = nvkm_rd32(device, 0x0025a0);
+	bool handled = false;
+
+	nvkm_runq_foreach_cond(runq, fifo, mask & BIT(runq->id)) {
+		if (runq->func->intr(runq, NULL))
+			handled = true;
+
+		nvkm_wr32(device, 0x0025a0, BIT(runq->id));
+	}
+
+	return handled;
+}
+
+static void
+gf100_fifo_intr_runlist(struct nvkm_fifo *fifo)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 intr = nvkm_rd32(device, 0x002a00);
+
+	if (intr & 0x10000000) {
+		nvkm_wr32(device, 0x002a00, 0x10000000);
+		intr &= ~0x10000000;
+	}
+
+	if (intr) {
+		nvkm_error(subdev, "RUNLIST %08x\n", intr);
+		nvkm_wr32(device, 0x002a00, intr);
+	}
+}
+
+static void
+gf100_fifo_intr_engine_unit(struct nvkm_fifo *fifo, int engn)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 intr = nvkm_rd32(device, 0x0025a8 + (engn * 0x04));
+	u32 inte = nvkm_rd32(device, 0x002628);
+	u32 unkn;
+
+	nvkm_wr32(device, 0x0025a8 + (engn * 0x04), intr);
+
+	for (unkn = 0; unkn < 8; unkn++) {
+		u32 ints = (intr >> (unkn * 0x04)) & inte;
+		if (ints & 0x1) {
+			nvkm_event_ntfy(&fifo->nonstall.event, 0, NVKM_FIFO_NONSTALL_EVENT);
+			ints &= ~1;
+		}
+		if (ints) {
+			nvkm_error(subdev, "ENGINE %d %d %01x", engn, unkn, ints);
+			nvkm_mask(device, 0x002628, ints, 0);
+		}
+	}
+}
+
+static void
+gf100_fifo_intr_engine(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u32 mask = nvkm_rd32(device, 0x0025a4);
+
+	while (mask) {
+		u32 unit = __ffs(mask);
+		gf100_fifo_intr_engine_unit(fifo, unit);
+		mask &= ~(1 << unit);
+	}
+}
+
+static irqreturn_t
+gf100_fifo_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_fifo *fifo = container_of(inth, typeof(*fifo), engine.subdev.inth);
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, 0x002140);
+	u32 stat = nvkm_rd32(device, 0x002100) & mask;
+
+	if (stat & 0x00000001) {
+		u32 intr = nvkm_rd32(device, 0x00252c);
+		nvkm_warn(subdev, "INTR 00000001: %08x\n", intr);
+		nvkm_wr32(device, 0x002100, 0x00000001);
+		stat &= ~0x00000001;
+	}
+
+	if (stat & 0x00000100) {
+		gf100_fifo_intr_sched(fifo);
+		nvkm_wr32(device, 0x002100, 0x00000100);
+		stat &= ~0x00000100;
+	}
+
+	if (stat & 0x00010000) {
+		u32 intr = nvkm_rd32(device, 0x00256c);
+		nvkm_warn(subdev, "INTR 00010000: %08x\n", intr);
+		nvkm_wr32(device, 0x002100, 0x00010000);
+		stat &= ~0x00010000;
+	}
+
+	if (stat & 0x01000000) {
+		u32 intr = nvkm_rd32(device, 0x00258c);
+		nvkm_warn(subdev, "INTR 01000000: %08x\n", intr);
+		nvkm_wr32(device, 0x002100, 0x01000000);
+		stat &= ~0x01000000;
+	}
+
+	if (stat & 0x10000000) {
+		gf100_fifo_intr_mmu_fault(fifo);
+		stat &= ~0x10000000;
+	}
+
+	if (stat & 0x20000000) {
+		if (gf100_fifo_intr_pbdma(fifo))
+			stat &= ~0x20000000;
+	}
+
+	if (stat & 0x40000000) {
+		gf100_fifo_intr_runlist(fifo);
+		stat &= ~0x40000000;
+	}
+
+	if (stat & 0x80000000) {
+		gf100_fifo_intr_engine(fifo);
+		stat &= ~0x80000000;
+	}
+
+	if (stat) {
+		nvkm_error(subdev, "INTR %08x\n", stat);
+		spin_lock(&fifo->lock);
+		nvkm_mask(device, 0x002140, stat, 0x00000000);
+		spin_unlock(&fifo->lock);
+		nvkm_wr32(device, 0x002100, stat);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static void
+gf100_fifo_init_pbdmas(struct nvkm_fifo *fifo, u32 mask)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+
+	/* Enable PBDMAs. */
+	nvkm_wr32(device, 0x000204, mask);
+	nvkm_wr32(device, 0x002204, mask);
+
+	/* Assign engines to PBDMAs. */
+	if ((mask & 7) == 7) {
+		nvkm_wr32(device, 0x002208, ~(1 << 0)); /* PGRAPH */
+		nvkm_wr32(device, 0x00220c, ~(1 << 1)); /* PVP */
+		nvkm_wr32(device, 0x002210, ~(1 << 1)); /* PMSPP */
+		nvkm_wr32(device, 0x002214, ~(1 << 1)); /* PMSVLD */
+		nvkm_wr32(device, 0x002218, ~(1 << 2)); /* PCE0 */
+		nvkm_wr32(device, 0x00221c, ~(1 << 1)); /* PCE1 */
+	}
+
+	nvkm_mask(device, 0x002a04, 0xbfffffff, 0xbfffffff);
+}
+
+static void
+gf100_fifo_init(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x002200, 0x00000001, 0x00000001);
+	nvkm_wr32(device, 0x002254, 0x10000000 | fifo->userd.bar1->addr >> 12);
+
+	nvkm_wr32(device, 0x002100, 0xffffffff);
+	nvkm_wr32(device, 0x002140, 0x7fffffff);
+	nvkm_wr32(device, 0x002628, 0x00000001); /* ENGINE_INTR_EN */
+}
+
+static int
+gf100_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_runl *runl;
+
+	runl = nvkm_runl_new(fifo, 0, 0, 0);
+	if (IS_ERR(runl))
+		return PTR_ERR(runl);
+
+	nvkm_runl_add(runl,  0, fifo->func->engn, NVKM_ENGINE_GR, 0);
+	nvkm_runl_add(runl,  1, fifo->func->engn, NVKM_ENGINE_MSPDEC, 0);
+	nvkm_runl_add(runl,  2, fifo->func->engn, NVKM_ENGINE_MSPPP, 0);
+	nvkm_runl_add(runl,  3, fifo->func->engn, NVKM_ENGINE_MSVLD, 0);
+	nvkm_runl_add(runl,  4, fifo->func->engn, NVKM_ENGINE_CE, 0);
+	nvkm_runl_add(runl,  5, fifo->func->engn, NVKM_ENGINE_CE, 1);
+	nvkm_runl_add(runl, 15,   &gf100_engn_sw, NVKM_ENGINE_SW, 0);
+	return 0;
+}
+
+int
+gf100_fifo_runq_nr(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u32 save;
+
+	/* Determine number of PBDMAs by checking valid enable bits. */
+	save = nvkm_mask(device, 0x000204, 0xffffffff, 0xffffffff);
+	save = nvkm_mask(device, 0x000204, 0xffffffff, save);
+	return hweight32(save);
+}
+
+int
+gf100_fifo_chid_ctor(struct nvkm_fifo *fifo, int nr)
+{
+	return nvkm_chid_new(&nvkm_chan_event, &fifo->engine.subdev, nr, 0, nr, &fifo->chid);
+}
+
+static const struct nvkm_fifo_func
+gf100_fifo = {
+	.chid_nr = nv50_fifo_chid_nr,
+	.chid_ctor = gf100_fifo_chid_ctor,
+	.runq_nr = gf100_fifo_runq_nr,
+	.runl_ctor = gf100_fifo_runl_ctor,
+	.init = gf100_fifo_init,
+	.init_pbdmas = gf100_fifo_init_pbdmas,
+	.intr = gf100_fifo_intr,
+	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gf100_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gf100_runl,
+	.runq = &gf100_runq,
+	.engn = &gf100_engn,
+	.cgrp = {{                            }, &nv04_cgrp },
+	.chan = {{ 0, 0, FERMI_CHANNEL_GPFIFO }, &gf100_chan },
+};
+
+int
+gf100_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gf100_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
new file mode 100644
index 000000000..d8a4d773a
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -0,0 +1,833 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/mc.h>
+#include <subdev/mmu.h>
+#include <subdev/top.h>
+
+#include <nvif/class.h>
+#include <nvif/if900d.h>
+
+void
+gk104_chan_stop(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x800004 + (chan->id * 8), 0x00000800, 0x00000800);
+}
+
+void
+gk104_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x800004 + (chan->id * 8), 0x00000400, 0x00000400);
+}
+
+void
+gk104_chan_unbind(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x800000 + (chan->id * 8), 0x00000000);
+}
+
+void
+gk104_chan_bind_inst(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x800000 + (chan->id * 8), 0x80000000 | chan->inst->addr >> 12);
+}
+
+void
+gk104_chan_bind(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x800004 + (chan->id * 8), 0x000f0000, runl->id << 16);
+	gk104_chan_bind_inst(chan);
+}
+
+static int
+gk104_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	const u64 userd = nvkm_memory_addr(chan->userd.mem) + chan->userd.base;
+	const u32 limit2 = ilog2(length / 8);
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x08, lower_32_bits(userd));
+	nvkm_wo32(chan->inst, 0x0c, upper_32_bits(userd));
+	nvkm_wo32(chan->inst, 0x10, 0x0000face);
+	nvkm_wo32(chan->inst, 0x30, 0xfffff902);
+	nvkm_wo32(chan->inst, 0x48, lower_32_bits(offset));
+	nvkm_wo32(chan->inst, 0x4c, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->inst, 0x84, 0x20400000);
+	nvkm_wo32(chan->inst, 0x94, 0x30000000 | devm);
+	nvkm_wo32(chan->inst, 0x9c, 0x00000100);
+	nvkm_wo32(chan->inst, 0xac, 0x0000001f);
+	nvkm_wo32(chan->inst, 0xe4, priv ? 0x00000020 : 0x00000000);
+	nvkm_wo32(chan->inst, 0xe8, chan->id);
+	nvkm_wo32(chan->inst, 0xb8, 0xf8000000);
+	nvkm_wo32(chan->inst, 0xf8, 0x10003080); /* 0x002310 */
+	nvkm_wo32(chan->inst, 0xfc, 0x10000010); /* 0x002350 */
+	nvkm_done(chan->inst);
+	return 0;
+}
+
+const struct nvkm_chan_func_ramfc
+gk104_chan_ramfc = {
+	.write = gk104_chan_ramfc_write,
+	.devm = 0xfff,
+	.priv = true,
+};
+
+const struct nvkm_chan_func_userd
+gk104_chan_userd = {
+	.bar = 1,
+	.size = 0x200,
+	.clear = gf100_chan_userd_clear,
+};
+
+static const struct nvkm_chan_func
+gk104_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gk104_chan_userd,
+	.ramfc = &gk104_chan_ramfc,
+	.bind = gk104_chan_bind,
+	.unbind = gk104_chan_unbind,
+	.start = gk104_chan_start,
+	.stop = gk104_chan_stop,
+	.preempt = gf100_chan_preempt,
+};
+
+static void
+gk104_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	u32 ptr0, ptr1 = 0;
+	u64 addr = 0ULL;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_SW    : return;
+	case NVKM_ENGINE_GR    : ptr0 = 0x0210; break;
+	case NVKM_ENGINE_SEC   : ptr0 = 0x0220; break;
+	case NVKM_ENGINE_MSPDEC: ptr0 = 0x0250; break;
+	case NVKM_ENGINE_MSPPP : ptr0 = 0x0260; break;
+	case NVKM_ENGINE_MSVLD : ptr0 = 0x0270; break;
+	case NVKM_ENGINE_VIC   : ptr0 = 0x0280; break;
+	case NVKM_ENGINE_MSENC : ptr0 = 0x0290; break;
+	case NVKM_ENGINE_NVDEC :
+		ptr1 = 0x0270;
+		ptr0 = 0x0210;
+		break;
+	case NVKM_ENGINE_NVENC :
+		if (!engn->engine->subdev.inst)
+			ptr1 = 0x0290;
+		ptr0 = 0x0210;
+		break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (cctx) {
+		addr  = cctx->vctx->vma->addr;
+		addr |= 4ULL;
+	}
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, ptr0 + 0, lower_32_bits(addr));
+	nvkm_wo32(chan->inst, ptr0 + 4, upper_32_bits(addr));
+	if (ptr1) {
+		nvkm_wo32(chan->inst, ptr1 + 0, lower_32_bits(addr));
+		nvkm_wo32(chan->inst, ptr1 + 4, upper_32_bits(addr));
+	}
+	nvkm_done(chan->inst);
+}
+
+int
+gk104_ectx_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx)
+{
+	struct gf100_vmm_map_v0 args = { .priv = 1 };
+	int ret;
+
+	ret = nvkm_vmm_get(vctx->vmm, 12, vctx->inst->size, &vctx->vma);
+	if (ret)
+		return ret;
+
+	return nvkm_memory_map(vctx->inst, 0, vctx->vmm, vctx->vma, &args, sizeof(args));
+}
+
+/*TODO: clean this up */
+struct gk104_engn_status {
+	bool busy;
+	bool faulted;
+	bool chsw;
+	bool save;
+	bool load;
+	struct {
+		bool tsg;
+		u32 id;
+	} prev, next, *chan;
+};
+
+static void
+gk104_engn_status(struct nvkm_engn *engn, struct gk104_engn_status *status)
+{
+	u32 stat = nvkm_rd32(engn->runl->fifo->engine.subdev.device, 0x002640 + (engn->id * 0x08));
+
+	status->busy     = !!(stat & 0x80000000);
+	status->faulted  = !!(stat & 0x40000000);
+	status->next.tsg = !!(stat & 0x10000000);
+	status->next.id  =   (stat & 0x0fff0000) >> 16;
+	status->chsw     = !!(stat & 0x00008000);
+	status->save     = !!(stat & 0x00004000);
+	status->load     = !!(stat & 0x00002000);
+	status->prev.tsg = !!(stat & 0x00001000);
+	status->prev.id  =   (stat & 0x00000fff);
+	status->chan     = NULL;
+
+	if (status->busy && status->chsw) {
+		if (status->load && status->save) {
+			if (nvkm_engine_chsw_load(engn->engine))
+				status->chan = &status->next;
+			else
+				status->chan = &status->prev;
+		} else
+		if (status->load) {
+			status->chan = &status->next;
+		} else {
+			status->chan = &status->prev;
+		}
+	} else
+	if (status->load) {
+		status->chan = &status->prev;
+	}
+
+	ENGN_DEBUG(engn, "%08x: busy %d faulted %d chsw %d save %d load %d %sid %d%s-> %sid %d%s",
+		   stat, status->busy, status->faulted, status->chsw, status->save, status->load,
+		   status->prev.tsg ? "tsg" : "ch", status->prev.id,
+		   status->chan == &status->prev ? "*" : " ",
+		   status->next.tsg ? "tsg" : "ch", status->next.id,
+		   status->chan == &status->next ? "*" : " ");
+}
+
+int
+gk104_engn_cxid(struct nvkm_engn *engn, bool *cgid)
+{
+	struct gk104_engn_status status;
+
+	gk104_engn_status(engn, &status);
+	if (status.chan) {
+		*cgid = status.chan->tsg;
+		return status.chan->id;
+	}
+
+	return -ENODEV;
+}
+
+bool
+gk104_engn_chsw(struct nvkm_engn *engn)
+{
+	struct gk104_engn_status status;
+
+	gk104_engn_status(engn, &status);
+	if (status.busy && status.chsw)
+		return true;
+
+	return false;
+}
+
+const struct nvkm_engn_func
+gk104_engn = {
+	.chsw = gk104_engn_chsw,
+	.cxid = gk104_engn_cxid,
+	.mmu_fault_trigger = gf100_engn_mmu_fault_trigger,
+	.mmu_fault_triggered = gf100_engn_mmu_fault_triggered,
+	.ctor = gk104_ectx_ctor,
+	.bind = gk104_ectx_bind,
+};
+
+const struct nvkm_engn_func
+gk104_engn_ce = {
+	.chsw = gk104_engn_chsw,
+	.cxid = gk104_engn_cxid,
+	.mmu_fault_trigger = gf100_engn_mmu_fault_trigger,
+	.mmu_fault_triggered = gf100_engn_mmu_fault_triggered,
+};
+
+bool
+gk104_runq_idle(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	return !(nvkm_rd32(device, 0x003080 + (runq->id * 4)) & 0x0000e000);
+}
+
+static const struct nvkm_bitfield
+gk104_runq_intr_1_names[] = {
+	{ 0x00000001, "HCE_RE_ILLEGAL_OP" },
+	{ 0x00000002, "HCE_RE_ALIGNB" },
+	{ 0x00000004, "HCE_PRIV" },
+	{ 0x00000008, "HCE_ILLEGAL_MTHD" },
+	{ 0x00000010, "HCE_ILLEGAL_CLASS" },
+	{}
+};
+
+static bool
+gk104_runq_intr_1(struct nvkm_runq *runq)
+{
+	struct nvkm_subdev *subdev = &runq->fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, 0x04014c + (runq->id * 0x2000));
+	u32 stat = nvkm_rd32(device, 0x040148 + (runq->id * 0x2000)) & mask;
+	u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x2000)) & 0xfff;
+	char msg[128];
+
+	if (stat & 0x80000000) {
+		if (runq->func->intr_1_ctxnotvalid &&
+		    runq->func->intr_1_ctxnotvalid(runq, chid))
+			stat &= ~0x80000000;
+	}
+
+	if (stat) {
+		nvkm_snprintbf(msg, sizeof(msg), gk104_runq_intr_1_names, stat);
+		nvkm_error(subdev, "PBDMA%d: %08x [%s] ch %d %08x %08x\n",
+			   runq->id, stat, msg, chid,
+			   nvkm_rd32(device, 0x040150 + (runq->id * 0x2000)),
+			   nvkm_rd32(device, 0x040154 + (runq->id * 0x2000)));
+	}
+
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x2000), stat);
+	return true;
+}
+
+const struct nvkm_bitfield
+gk104_runq_intr_0_names[] = {
+	{ 0x00000001, "MEMREQ" },
+	{ 0x00000002, "MEMACK_TIMEOUT" },
+	{ 0x00000004, "MEMACK_EXTRA" },
+	{ 0x00000008, "MEMDAT_TIMEOUT" },
+	{ 0x00000010, "MEMDAT_EXTRA" },
+	{ 0x00000020, "MEMFLUSH" },
+	{ 0x00000040, "MEMOP" },
+	{ 0x00000080, "LBCONNECT" },
+	{ 0x00000100, "LBREQ" },
+	{ 0x00000200, "LBACK_TIMEOUT" },
+	{ 0x00000400, "LBACK_EXTRA" },
+	{ 0x00000800, "LBDAT_TIMEOUT" },
+	{ 0x00001000, "LBDAT_EXTRA" },
+	{ 0x00002000, "GPFIFO" },
+	{ 0x00004000, "GPPTR" },
+	{ 0x00008000, "GPENTRY" },
+	{ 0x00010000, "GPCRC" },
+	{ 0x00020000, "PBPTR" },
+	{ 0x00040000, "PBENTRY" },
+	{ 0x00080000, "PBCRC" },
+	{ 0x00100000, "XBARCONNECT" },
+	{ 0x00200000, "METHOD" },
+	{ 0x00400000, "METHODCRC" },
+	{ 0x00800000, "DEVICE" },
+	{ 0x02000000, "SEMAPHORE" },
+	{ 0x04000000, "ACQUIRE" },
+	{ 0x08000000, "PRI" },
+	{ 0x20000000, "NO_CTXSW_SEG" },
+	{ 0x40000000, "PBSEG" },
+	{ 0x80000000, "SIGNATURE" },
+	{}
+};
+
+bool
+gk104_runq_intr(struct nvkm_runq *runq, struct nvkm_runl *null)
+{
+	bool intr0 = gf100_runq_intr(runq, NULL);
+	bool intr1 = gk104_runq_intr_1(runq);
+
+	return intr0 || intr1;
+}
+
+void
+gk104_runq_init(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	gf100_runq_init(runq);
+
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x2000), 0xffffffff); /* HCE.INTR */
+	nvkm_wr32(device, 0x04014c + (runq->id * 0x2000), 0xffffffff); /* HCE.INTREN */
+}
+
+static u32
+gk104_runq_runm(struct nvkm_runq *runq)
+{
+	return nvkm_rd32(runq->fifo->engine.subdev.device, 0x002390 + (runq->id * 0x04));
+}
+
+const struct nvkm_runq_func
+gk104_runq = {
+	.init = gk104_runq_init,
+	.intr = gk104_runq_intr,
+	.intr_0_names = gk104_runq_intr_0_names,
+	.idle = gk104_runq_idle,
+};
+
+void
+gk104_runl_fault_clear(struct nvkm_runl *runl)
+{
+	nvkm_wr32(runl->fifo->engine.subdev.device, 0x00262c, BIT(runl->id));
+}
+
+void
+gk104_runl_allow(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, 0x002630, BIT(runl->id), 0x00000000);
+}
+
+void
+gk104_runl_block(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, 0x002630, BIT(runl->id), BIT(runl->id));
+}
+
+bool
+gk104_runl_pending(struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+
+	return nvkm_rd32(device, 0x002284 + (runl->id * 0x08)) & 0x00100000;
+}
+
+void
+gk104_runl_commit(struct nvkm_runl *runl, struct nvkm_memory *memory, u32 start, int count)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u64 addr = nvkm_memory_addr(memory) + start;
+	int target;
+
+	switch (nvkm_memory_target(memory)) {
+	case NVKM_MEM_TARGET_VRAM: target = 0; break;
+	case NVKM_MEM_TARGET_NCOH: target = 3; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	spin_lock_irq(&fifo->lock);
+	nvkm_wr32(device, 0x002270, (target << 28) | (addr >> 12));
+	nvkm_wr32(device, 0x002274, (runl->id << 20) | count);
+	spin_unlock_irq(&fifo->lock);
+}
+
+void
+gk104_runl_insert_chan(struct nvkm_chan *chan, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset + 0, chan->id);
+	nvkm_wo32(memory, offset + 4, 0x00000000);
+}
+
+static const struct nvkm_runl_func
+gk104_runl = {
+	.size = 8,
+	.update = nv50_runl_update,
+	.insert_chan = gk104_runl_insert_chan,
+	.commit = gk104_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = gk104_runl_pending,
+	.block = gk104_runl_block,
+	.allow = gk104_runl_allow,
+	.fault_clear = gk104_runl_fault_clear,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+static const struct nvkm_enum
+gk104_fifo_mmu_fault_engine[] = {
+	{ 0x00, "GR", NULL, NVKM_ENGINE_GR },
+	{ 0x01, "DISPLAY" },
+	{ 0x02, "CAPTURE" },
+	{ 0x03, "IFB", NULL, NVKM_ENGINE_IFB },
+	{ 0x04, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0x05, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
+	{ 0x06, "SCHED" },
+	{ 0x07, "HOST0" },
+	{ 0x08, "HOST1" },
+	{ 0x09, "HOST2" },
+	{ 0x0a, "HOST3" },
+	{ 0x0b, "HOST4" },
+	{ 0x0c, "HOST5" },
+	{ 0x0d, "HOST6" },
+	{ 0x0e, "HOST7" },
+	{ 0x0f, "HOSTSR" },
+	{ 0x10, "MSVLD", NULL, NVKM_ENGINE_MSVLD },
+	{ 0x11, "MSPPP", NULL, NVKM_ENGINE_MSPPP },
+	{ 0x13, "PERF" },
+	{ 0x14, "MSPDEC", NULL, NVKM_ENGINE_MSPDEC },
+	{ 0x15, "CE0", NULL, NVKM_ENGINE_CE, 0 },
+	{ 0x16, "CE1", NULL, NVKM_ENGINE_CE, 1 },
+	{ 0x17, "PMU" },
+	{ 0x18, "PTP" },
+	{ 0x19, "MSENC", NULL, NVKM_ENGINE_MSENC },
+	{ 0x1b, "CE2", NULL, NVKM_ENGINE_CE, 2 },
+	{}
+};
+
+const struct nvkm_enum
+gk104_fifo_mmu_fault_reason[] = {
+	{ 0x00, "PDE" },
+	{ 0x01, "PDE_SIZE" },
+	{ 0x02, "PTE" },
+	{ 0x03, "VA_LIMIT_VIOLATION" },
+	{ 0x04, "UNBOUND_INST_BLOCK" },
+	{ 0x05, "PRIV_VIOLATION" },
+	{ 0x06, "RO_VIOLATION" },
+	{ 0x07, "WO_VIOLATION" },
+	{ 0x08, "PITCH_MASK_VIOLATION" },
+	{ 0x09, "WORK_CREATION" },
+	{ 0x0a, "UNSUPPORTED_APERTURE" },
+	{ 0x0b, "COMPRESSION_FAILURE" },
+	{ 0x0c, "UNSUPPORTED_KIND" },
+	{ 0x0d, "REGION_VIOLATION" },
+	{ 0x0e, "BOTH_PTES_VALID" },
+	{ 0x0f, "INFO_TYPE_POISONED" },
+	{}
+};
+
+const struct nvkm_enum
+gk104_fifo_mmu_fault_hubclient[] = {
+	{ 0x00, "VIP" },
+	{ 0x01, "CE0" },
+	{ 0x02, "CE1" },
+	{ 0x03, "DNISO" },
+	{ 0x04, "FE" },
+	{ 0x05, "FECS" },
+	{ 0x06, "HOST" },
+	{ 0x07, "HOST_CPU" },
+	{ 0x08, "HOST_CPU_NB" },
+	{ 0x09, "ISO" },
+	{ 0x0a, "MMU" },
+	{ 0x0b, "MSPDEC" },
+	{ 0x0c, "MSPPP" },
+	{ 0x0d, "MSVLD" },
+	{ 0x0e, "NISO" },
+	{ 0x0f, "P2P" },
+	{ 0x10, "PD" },
+	{ 0x11, "PERF" },
+	{ 0x12, "PMU" },
+	{ 0x13, "RASTERTWOD" },
+	{ 0x14, "SCC" },
+	{ 0x15, "SCC_NB" },
+	{ 0x16, "SEC" },
+	{ 0x17, "SSYNC" },
+	{ 0x18, "GR_CE" },
+	{ 0x19, "CE2" },
+	{ 0x1a, "XV" },
+	{ 0x1b, "MMU_NB" },
+	{ 0x1c, "MSENC" },
+	{ 0x1d, "DFALCON" },
+	{ 0x1e, "SKED" },
+	{ 0x1f, "AFALCON" },
+	{}
+};
+
+const struct nvkm_enum
+gk104_fifo_mmu_fault_gpcclient[] = {
+	{ 0x00, "L1_0" }, { 0x01, "T1_0" }, { 0x02, "PE_0" },
+	{ 0x03, "L1_1" }, { 0x04, "T1_1" }, { 0x05, "PE_1" },
+	{ 0x06, "L1_2" }, { 0x07, "T1_2" }, { 0x08, "PE_2" },
+	{ 0x09, "L1_3" }, { 0x0a, "T1_3" }, { 0x0b, "PE_3" },
+	{ 0x0c, "RAST" },
+	{ 0x0d, "GCC" },
+	{ 0x0e, "GPCCS" },
+	{ 0x0f, "PROP_0" },
+	{ 0x10, "PROP_1" },
+	{ 0x11, "PROP_2" },
+	{ 0x12, "PROP_3" },
+	{ 0x13, "L1_4" }, { 0x14, "T1_4" }, { 0x15, "PE_4" },
+	{ 0x16, "L1_5" }, { 0x17, "T1_5" }, { 0x18, "PE_5" },
+	{ 0x19, "L1_6" }, { 0x1a, "T1_6" }, { 0x1b, "PE_6" },
+	{ 0x1c, "L1_7" }, { 0x1d, "T1_7" }, { 0x1e, "PE_7" },
+	{ 0x1f, "GPM" },
+	{ 0x20, "LTP_UTLB_0" },
+	{ 0x21, "LTP_UTLB_1" },
+	{ 0x22, "LTP_UTLB_2" },
+	{ 0x23, "LTP_UTLB_3" },
+	{ 0x24, "GPC_RGG_UTLB" },
+	{}
+};
+
+const struct nvkm_fifo_func_mmu_fault
+gk104_fifo_mmu_fault = {
+	.recover = gf100_fifo_mmu_fault_recover,
+	.access = gf100_fifo_mmu_fault_access,
+	.engine = gk104_fifo_mmu_fault_engine,
+	.reason = gk104_fifo_mmu_fault_reason,
+	.hubclient = gk104_fifo_mmu_fault_hubclient,
+	.gpcclient = gk104_fifo_mmu_fault_gpcclient,
+};
+
+static const struct nvkm_enum
+gk104_fifo_intr_bind_reason[] = {
+	{ 0x01, "BIND_NOT_UNBOUND" },
+	{ 0x02, "SNOOP_WITHOUT_BAR1" },
+	{ 0x03, "UNBIND_WHILE_RUNNING" },
+	{ 0x05, "INVALID_RUNLIST" },
+	{ 0x06, "INVALID_CTX_TGT" },
+	{ 0x0b, "UNBIND_WHILE_PARKED" },
+	{}
+};
+
+void
+gk104_fifo_intr_bind(struct nvkm_fifo *fifo)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	u32 intr = nvkm_rd32(subdev->device, 0x00252c);
+	u32 code = intr & 0x000000ff;
+	const struct nvkm_enum *en = nvkm_enum_find(gk104_fifo_intr_bind_reason, code);
+
+	nvkm_error(subdev, "BIND_ERROR %02x [%s]\n", code, en ? en->name : "");
+}
+
+void
+gk104_fifo_intr_chsw(struct nvkm_fifo *fifo)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 stat = nvkm_rd32(device, 0x00256c);
+
+	nvkm_error(subdev, "CHSW_ERROR %08x\n", stat);
+	nvkm_wr32(device, 0x00256c, stat);
+}
+
+static void
+gk104_fifo_intr_dropped_fault(struct nvkm_fifo *fifo)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	u32 stat = nvkm_rd32(subdev->device, 0x00259c);
+
+	nvkm_error(subdev, "DROPPED_MMU_FAULT %08x\n", stat);
+}
+
+void
+gk104_fifo_intr_runlist(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_runl *runl;
+	u32 mask = nvkm_rd32(device, 0x002a00);
+
+	nvkm_runl_foreach_cond(runl, fifo, mask & BIT(runl->id)) {
+		nvkm_wr32(device, 0x002a00, BIT(runl->id));
+	}
+}
+
+irqreturn_t
+gk104_fifo_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_fifo *fifo = container_of(inth, typeof(*fifo), engine.subdev.inth);
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, 0x002140);
+	u32 stat = nvkm_rd32(device, 0x002100) & mask;
+
+	if (stat & 0x00000001) {
+		gk104_fifo_intr_bind(fifo);
+		nvkm_wr32(device, 0x002100, 0x00000001);
+		stat &= ~0x00000001;
+	}
+
+	if (stat & 0x00000010) {
+		nvkm_error(subdev, "PIO_ERROR\n");
+		nvkm_wr32(device, 0x002100, 0x00000010);
+		stat &= ~0x00000010;
+	}
+
+	if (stat & 0x00000100) {
+		gf100_fifo_intr_sched(fifo);
+		nvkm_wr32(device, 0x002100, 0x00000100);
+		stat &= ~0x00000100;
+	}
+
+	if (stat & 0x00010000) {
+		gk104_fifo_intr_chsw(fifo);
+		nvkm_wr32(device, 0x002100, 0x00010000);
+		stat &= ~0x00010000;
+	}
+
+	if (stat & 0x00800000) {
+		nvkm_error(subdev, "FB_FLUSH_TIMEOUT\n");
+		nvkm_wr32(device, 0x002100, 0x00800000);
+		stat &= ~0x00800000;
+	}
+
+	if (stat & 0x01000000) {
+		nvkm_error(subdev, "LB_ERROR\n");
+		nvkm_wr32(device, 0x002100, 0x01000000);
+		stat &= ~0x01000000;
+	}
+
+	if (stat & 0x08000000) {
+		gk104_fifo_intr_dropped_fault(fifo);
+		nvkm_wr32(device, 0x002100, 0x08000000);
+		stat &= ~0x08000000;
+	}
+
+	if (stat & 0x10000000) {
+		gf100_fifo_intr_mmu_fault(fifo);
+		stat &= ~0x10000000;
+	}
+
+	if (stat & 0x20000000) {
+		if (gf100_fifo_intr_pbdma(fifo))
+			stat &= ~0x20000000;
+	}
+
+	if (stat & 0x40000000) {
+		gk104_fifo_intr_runlist(fifo);
+		stat &= ~0x40000000;
+	}
+
+	if (stat & 0x80000000) {
+		nvkm_wr32(device, 0x002100, 0x80000000);
+		nvkm_event_ntfy(&fifo->nonstall.event, 0, NVKM_FIFO_NONSTALL_EVENT);
+		stat &= ~0x80000000;
+	}
+
+	if (stat) {
+		nvkm_error(subdev, "INTR %08x\n", stat);
+		spin_lock(&fifo->lock);
+		nvkm_mask(device, 0x002140, stat, 0x00000000);
+		spin_unlock(&fifo->lock);
+		nvkm_wr32(device, 0x002100, stat);
+	}
+
+	return IRQ_HANDLED;
+}
+
+void
+gk104_fifo_init_pbdmas(struct nvkm_fifo *fifo, u32 mask)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x000204, mask);
+	nvkm_mask(device, 0x002a04, 0xbfffffff, 0xbfffffff);
+}
+
+void
+gk104_fifo_init(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+
+	if (fifo->func->chan.func->userd->bar == 1)
+		nvkm_wr32(device, 0x002254, 0x10000000 | fifo->userd.bar1->addr >> 12);
+
+	nvkm_wr32(device, 0x002100, 0xffffffff);
+	nvkm_wr32(device, 0x002140, 0x7fffffff);
+}
+
+int
+gk104_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_top_device *tdev;
+	struct nvkm_runl *runl;
+	struct nvkm_runq *runq;
+	const struct nvkm_engn_func *func;
+
+	nvkm_list_foreach(tdev, &device->top->device, head, tdev->runlist >= 0) {
+		runl = nvkm_runl_get(fifo, tdev->runlist, tdev->runlist);
+		if (!runl) {
+			runl = nvkm_runl_new(fifo, tdev->runlist, tdev->runlist, 0);
+			if (IS_ERR(runl))
+				return PTR_ERR(runl);
+
+			nvkm_runq_foreach_cond(runq, fifo, gk104_runq_runm(runq) & BIT(runl->id)) {
+				if (WARN_ON(runl->runq_nr == ARRAY_SIZE(runl->runq)))
+					return -ENOMEM;
+
+				runl->runq[runl->runq_nr++] = runq;
+			}
+
+		}
+
+		if (tdev->engine < 0)
+			continue;
+
+		switch (tdev->type) {
+		case NVKM_ENGINE_CE:
+			func = fifo->func->engn_ce;
+			break;
+		case NVKM_ENGINE_GR:
+			nvkm_runl_add(runl, 15, &gf100_engn_sw, NVKM_ENGINE_SW, 0);
+			fallthrough;
+		default:
+			func = fifo->func->engn;
+			break;
+		}
+
+		nvkm_runl_add(runl, tdev->engine, func, tdev->type, tdev->inst);
+	}
+
+	return 0;
+}
+
+int
+gk104_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return 4096;
+}
+
+static const struct nvkm_fifo_func
+gk104_fifo = {
+	.chid_nr = gk104_fifo_chid_nr,
+	.chid_ctor = gf100_fifo_chid_ctor,
+	.runq_nr = gf100_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gk104_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gk104_runl,
+	.runq = &gk104_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{                               }, &nv04_cgrp },
+	.chan = {{ 0, 0, KEPLER_CHANNEL_GPFIFO_A }, &gk104_chan },
+};
+
+int
+gk104_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gk104_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c
new file mode 100644
index 000000000..a8ff21cf7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c
@@ -0,0 +1,132 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+
+#include <core/memory.h>
+#include <subdev/timer.h>
+
+#include <nvif/class.h>
+
+void
+gk110_chan_preempt(struct nvkm_chan *chan)
+{
+	struct nvkm_cgrp *cgrp = chan->cgrp;
+
+	if (cgrp->hw) {
+		cgrp->func->preempt(cgrp);
+		return;
+	}
+
+	gf100_chan_preempt(chan);
+}
+
+const struct nvkm_chan_func
+gk110_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gk104_chan_userd,
+	.ramfc = &gk104_chan_ramfc,
+	.bind = gk104_chan_bind,
+	.unbind = gk104_chan_unbind,
+	.start = gk104_chan_start,
+	.stop = gk104_chan_stop,
+	.preempt = gk110_chan_preempt,
+};
+
+static void
+gk110_cgrp_preempt(struct nvkm_cgrp *cgrp)
+{
+	nvkm_wr32(cgrp->runl->fifo->engine.subdev.device, 0x002634, 0x01000000 | cgrp->id);
+}
+
+const struct nvkm_cgrp_func
+gk110_cgrp = {
+	.preempt = gk110_cgrp_preempt,
+};
+
+void
+gk110_runl_insert_cgrp(struct nvkm_cgrp *cgrp, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset + 0, (cgrp->chan_nr << 26) | (128 << 18) |
+				      (3 << 14) | 0x00002000 | cgrp->id);
+	nvkm_wo32(memory, offset + 4, 0x00000000);
+}
+
+const struct nvkm_runl_func
+gk110_runl = {
+	.size = 8,
+	.update = nv50_runl_update,
+	.insert_cgrp = gk110_runl_insert_cgrp,
+	.insert_chan = gk104_runl_insert_chan,
+	.commit = gk104_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = gk104_runl_pending,
+	.block = gk104_runl_block,
+	.allow = gk104_runl_allow,
+	.fault_clear = gk104_runl_fault_clear,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+int
+gk110_fifo_chid_ctor(struct nvkm_fifo *fifo, int nr)
+{
+	int ret;
+
+	ret = nvkm_chid_new(&nvkm_chan_event, &fifo->engine.subdev, nr, 0, nr, &fifo->cgid);
+	if (ret)
+		return ret;
+
+	return gf100_fifo_chid_ctor(fifo, nr);
+}
+
+static const struct nvkm_fifo_func
+gk110_fifo = {
+	.chid_nr = gk104_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gf100_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gk104_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gk110_runl,
+	.runq = &gk104_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp },
+	.chan = {{ 0, 0, KEPLER_CHANNEL_GPFIFO_B }, &gk110_chan },
+};
+
+int
+gk110_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gk110_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c
new file mode 100644
index 000000000..8fa2b0be1
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c
@@ -0,0 +1,77 @@
+/*
+ * Copyright 2013 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "runq.h"
+
+#include <nvif/class.h>
+
+void
+gk208_runq_init(struct nvkm_runq *runq)
+{
+	gk104_runq_init(runq);
+
+	nvkm_wr32(runq->fifo->engine.subdev.device, 0x04012c + (runq->id * 0x2000), 0x000f4240);
+}
+
+const struct nvkm_runq_func
+gk208_runq = {
+	.init = gk208_runq_init,
+	.intr = gk104_runq_intr,
+	.intr_0_names = gk104_runq_intr_0_names,
+	.idle = gk104_runq_idle,
+};
+
+static int
+gk208_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return 1024;
+}
+
+static const struct nvkm_fifo_func
+gk208_fifo = {
+	.chid_nr = gk208_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gf100_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gk104_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gk110_runl,
+	.runq = &gk208_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp },
+	.chan = {{ 0, 0, KEPLER_CHANNEL_GPFIFO_A }, &gk110_chan },
+};
+
+int
+gk208_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gk208_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c
new file mode 100644
index 000000000..b63ca8361
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c
@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2014, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+
+#include <nvif/class.h>
+
+static const struct nvkm_fifo_func
+gk20a_fifo = {
+	.chid_nr = nv50_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gf100_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gk104_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gk110_runl,
+	.runq = &gk208_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{                               }, &gk110_cgrp },
+	.chan = {{ 0, 0, KEPLER_CHANNEL_GPFIFO_A }, &gk110_chan },
+};
+
+int
+gk20a_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gk20a_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c
new file mode 100644
index 000000000..5ba60021b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c
@@ -0,0 +1,156 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "chan.h"
+#include "runl.h"
+
+#include <core/gpuobj.h>
+#include <subdev/fault.h>
+
+#include <nvif/class.h>
+
+const struct nvkm_chan_func
+gm107_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gk104_chan_userd,
+	.ramfc = &gk104_chan_ramfc,
+	.bind = gk104_chan_bind_inst,
+	.unbind = gk104_chan_unbind,
+	.start = gk104_chan_start,
+	.stop = gk104_chan_stop,
+	.preempt = gk110_chan_preempt,
+};
+
+static void
+gm107_runl_insert_chan(struct nvkm_chan *chan, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset + 0, chan->id);
+	nvkm_wo32(memory, offset + 4, chan->inst->addr >> 12);
+}
+
+const struct nvkm_runl_func
+gm107_runl = {
+	.size = 8,
+	.update = nv50_runl_update,
+	.insert_cgrp = gk110_runl_insert_cgrp,
+	.insert_chan = gm107_runl_insert_chan,
+	.commit = gk104_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = gk104_runl_pending,
+	.block = gk104_runl_block,
+	.allow = gk104_runl_allow,
+	.fault_clear = gk104_runl_fault_clear,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+static const struct nvkm_enum
+gm107_fifo_mmu_fault_engine[] = {
+	{ 0x01, "DISPLAY" },
+	{ 0x02, "CAPTURE" },
+	{ 0x03, "IFB", NULL, NVKM_ENGINE_IFB },
+	{ 0x04, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0x05, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
+	{ 0x06, "SCHED" },
+	{ 0x07, "HOST0" },
+	{ 0x08, "HOST1" },
+	{ 0x09, "HOST2" },
+	{ 0x0a, "HOST3" },
+	{ 0x0b, "HOST4" },
+	{ 0x0c, "HOST5" },
+	{ 0x0d, "HOST6" },
+	{ 0x0e, "HOST7" },
+	{ 0x0f, "HOSTSR" },
+	{ 0x13, "PERF" },
+	{ 0x17, "PMU" },
+	{ 0x18, "PTP" },
+	{}
+};
+
+const struct nvkm_fifo_func_mmu_fault
+gm107_fifo_mmu_fault = {
+	.recover = gf100_fifo_mmu_fault_recover,
+	.access = gf100_fifo_mmu_fault_access,
+	.engine = gm107_fifo_mmu_fault_engine,
+	.reason = gk104_fifo_mmu_fault_reason,
+	.hubclient = gk104_fifo_mmu_fault_hubclient,
+	.gpcclient = gk104_fifo_mmu_fault_gpcclient,
+};
+
+void
+gm107_fifo_intr_mmu_fault_unit(struct nvkm_fifo *fifo, int unit)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u32 inst = nvkm_rd32(device, 0x002800 + (unit * 0x10));
+	u32 valo = nvkm_rd32(device, 0x002804 + (unit * 0x10));
+	u32 vahi = nvkm_rd32(device, 0x002808 + (unit * 0x10));
+	u32 type = nvkm_rd32(device, 0x00280c + (unit * 0x10));
+	struct nvkm_fault_data info;
+
+	info.inst   =  (u64)inst << 12;
+	info.addr   = ((u64)vahi << 32) | valo;
+	info.time   = 0;
+	info.engine = unit;
+	info.valid  = 1;
+	info.gpc    = (type & 0x1f000000) >> 24;
+	info.client = (type & 0x00003f00) >> 8;
+	info.access = (type & 0x00000080) >> 7;
+	info.hub    = (type & 0x00000040) >> 6;
+	info.reason = (type & 0x0000000f);
+
+	nvkm_fifo_fault(fifo, &info);
+}
+
+static int
+gm107_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return 2048;
+}
+
+static const struct nvkm_fifo_func
+gm107_fifo = {
+	.chid_nr = gm107_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gf100_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gm107_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gm107_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gm107_runl,
+	.runq = &gk208_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp },
+	.chan = {{ 0, 0, KEPLER_CHANNEL_GPFIFO_B }, &gm107_chan },
+};
+
+int
+gm107_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gm107_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c
new file mode 100644
index 000000000..d92d1ac39
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c
@@ -0,0 +1,66 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+
+#include <nvif/class.h>
+
+int
+gm200_fifo_runq_nr(struct nvkm_fifo *fifo)
+{
+	return nvkm_rd32(fifo->engine.subdev.device, 0x002004) & 0x000000ff;
+}
+
+int
+gm200_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return nvkm_rd32(fifo->engine.subdev.device, 0x002008);
+}
+
+static const struct nvkm_fifo_func
+gm200_fifo = {
+	.chid_nr = gm200_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gm200_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gm107_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gm107_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gm107_runl,
+	.runq = &gk208_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{ 0, 0,  KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp },
+	.chan = {{ 0, 0, MAXWELL_CHANNEL_GPFIFO_A }, &gm107_chan },
+};
+
+int
+gm200_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gm200_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
new file mode 100644
index 000000000..65bdb6a7d
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
@@ -0,0 +1,139 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "runl.h"
+
+#include <core/gpuobj.h>
+#include <subdev/fault.h>
+
+#include <nvif/class.h>
+
+static void
+gp100_runl_insert_chan(struct nvkm_chan *chan, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset + 0, chan->id | chan->runq << 14);
+	nvkm_wo32(memory, offset + 4, chan->inst->addr >> 12);
+}
+
+static const struct nvkm_runl_func
+gp100_runl = {
+	.runqs = 2,
+	.size = 8,
+	.update = nv50_runl_update,
+	.insert_cgrp = gk110_runl_insert_cgrp,
+	.insert_chan = gp100_runl_insert_chan,
+	.commit = gk104_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = gk104_runl_pending,
+	.block = gk104_runl_block,
+	.allow = gk104_runl_allow,
+	.fault_clear = gk104_runl_fault_clear,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+static const struct nvkm_enum
+gp100_fifo_mmu_fault_engine[] = {
+	{ 0x01, "DISPLAY" },
+	{ 0x03, "IFB", NULL, NVKM_ENGINE_IFB },
+	{ 0x04, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0x05, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
+	{ 0x06, "HOST0" },
+	{ 0x07, "HOST1" },
+	{ 0x08, "HOST2" },
+	{ 0x09, "HOST3" },
+	{ 0x0a, "HOST4" },
+	{ 0x0b, "HOST5" },
+	{ 0x0c, "HOST6" },
+	{ 0x0d, "HOST7" },
+	{ 0x0e, "HOST8" },
+	{ 0x0f, "HOST9" },
+	{ 0x10, "HOST10" },
+	{ 0x13, "PERF" },
+	{ 0x17, "PMU" },
+	{ 0x18, "PTP" },
+	{ 0x1f, "PHYSICAL" },
+	{}
+};
+
+static const struct nvkm_fifo_func_mmu_fault
+gp100_fifo_mmu_fault = {
+	.recover = gf100_fifo_mmu_fault_recover,
+	.access = gf100_fifo_mmu_fault_access,
+	.engine = gp100_fifo_mmu_fault_engine,
+	.reason = gk104_fifo_mmu_fault_reason,
+	.hubclient = gk104_fifo_mmu_fault_hubclient,
+	.gpcclient = gk104_fifo_mmu_fault_gpcclient,
+};
+
+static void
+gp100_fifo_intr_mmu_fault_unit(struct nvkm_fifo *fifo, int unit)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	u32 inst = nvkm_rd32(device, 0x002800 + (unit * 0x10));
+	u32 valo = nvkm_rd32(device, 0x002804 + (unit * 0x10));
+	u32 vahi = nvkm_rd32(device, 0x002808 + (unit * 0x10));
+	u32 type = nvkm_rd32(device, 0x00280c + (unit * 0x10));
+	struct nvkm_fault_data info;
+
+	info.inst   =  (u64)inst << 12;
+	info.addr   = ((u64)vahi << 32) | valo;
+	info.time   = 0;
+	info.engine = unit;
+	info.valid  = 1;
+	info.gpc    = (type & 0x1f000000) >> 24;
+	info.hub    = (type & 0x00100000) >> 20;
+	info.access = (type & 0x00070000) >> 16;
+	info.client = (type & 0x00007f00) >> 8;
+	info.reason = (type & 0x0000001f);
+
+	nvkm_fifo_fault(fifo, &info);
+}
+
+static const struct nvkm_fifo_func
+gp100_fifo = {
+	.chid_nr = gm200_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gm200_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_mmu_fault_unit = gp100_fifo_intr_mmu_fault_unit,
+	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gp100_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gp100_runl,
+	.runq = &gk208_runq,
+	.engn = &gk104_engn,
+	.engn_ce = &gk104_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp, .force = true },
+	.chan = {{ 0, 0, PASCAL_CHANNEL_GPFIFO_A }, &gm107_chan },
+};
+
+int
+gp100_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gp100_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c
new file mode 100644
index 000000000..33066c8cd
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c
@@ -0,0 +1,492 @@
+/*
+ * Copyright 2018 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+#include "chan.h"
+#include "chid.h"
+#include "cgrp.h"
+#include "runl.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/mmu.h>
+
+#include <nvif/class.h>
+
+static u32
+gv100_chan_doorbell_handle(struct nvkm_chan *chan)
+{
+	return chan->id;
+}
+
+static int
+gv100_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	const u64 userd = nvkm_memory_addr(chan->userd.mem) + chan->userd.base;
+	const u32 limit2 = ilog2(length / 8);
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x008, lower_32_bits(userd));
+	nvkm_wo32(chan->inst, 0x00c, upper_32_bits(userd));
+	nvkm_wo32(chan->inst, 0x010, 0x0000face);
+	nvkm_wo32(chan->inst, 0x030, 0x7ffff902);
+	nvkm_wo32(chan->inst, 0x048, lower_32_bits(offset));
+	nvkm_wo32(chan->inst, 0x04c, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->inst, 0x084, 0x20400000);
+	nvkm_wo32(chan->inst, 0x094, 0x30000000 | devm);
+	nvkm_wo32(chan->inst, 0x0e4, priv ? 0x00000020 : 0x00000000);
+	nvkm_wo32(chan->inst, 0x0e8, chan->id);
+	nvkm_wo32(chan->inst, 0x0f4, 0x00001000 | (priv ? 0x00000100 : 0x00000000));
+	nvkm_wo32(chan->inst, 0x0f8, 0x10003080);
+	nvkm_mo32(chan->inst, 0x218, 0x00000000, 0x00000000);
+	nvkm_done(chan->inst);
+	return 0;
+}
+
+const struct nvkm_chan_func_ramfc
+gv100_chan_ramfc = {
+	.write = gv100_chan_ramfc_write,
+	.devm = 0xfff,
+	.priv = true,
+};
+
+const struct nvkm_chan_func_userd
+gv100_chan_userd = {
+	.bar = -1,
+	.size = 0x200,
+	.clear = gf100_chan_userd_clear,
+};
+
+static const struct nvkm_chan_func
+gv100_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gv100_chan_userd,
+	.ramfc = &gv100_chan_ramfc,
+	.bind = gk104_chan_bind_inst,
+	.unbind = gk104_chan_unbind,
+	.start = gk104_chan_start,
+	.stop = gk104_chan_stop,
+	.preempt = gk110_chan_preempt,
+	.doorbell_handle = gv100_chan_doorbell_handle,
+};
+
+void
+gv100_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	u64 addr = 0ULL;
+
+	if (cctx) {
+		addr  = cctx->vctx->vma->addr;
+		addr |= 4ULL;
+	}
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x210, lower_32_bits(addr));
+	nvkm_wo32(chan->inst, 0x214, upper_32_bits(addr));
+	nvkm_mo32(chan->inst, 0x0ac, 0x00010000, cctx ? 0x00010000 : 0x00000000);
+	nvkm_done(chan->inst);
+}
+
+const struct nvkm_engn_func
+gv100_engn = {
+	.chsw = gk104_engn_chsw,
+	.cxid = gk104_engn_cxid,
+	.ctor = gk104_ectx_ctor,
+	.bind = gv100_ectx_bind,
+};
+
+void
+gv100_ectx_ce_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	const u64 bar2 = cctx ? nvkm_memory_bar2(cctx->vctx->inst->memory) : 0ULL;
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x220, lower_32_bits(bar2));
+	nvkm_wo32(chan->inst, 0x224, upper_32_bits(bar2));
+	nvkm_mo32(chan->inst, 0x0ac, 0x00020000, cctx ? 0x00020000 : 0x00000000);
+	nvkm_done(chan->inst);
+}
+
+int
+gv100_ectx_ce_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx)
+{
+	if (nvkm_memory_bar2(vctx->inst->memory) == ~0ULL)
+		return -EFAULT;
+
+	return 0;
+}
+
+const struct nvkm_engn_func
+gv100_engn_ce = {
+	.chsw = gk104_engn_chsw,
+	.cxid = gk104_engn_cxid,
+	.ctor = gv100_ectx_ce_ctor,
+	.bind = gv100_ectx_ce_bind,
+};
+
+static bool
+gv100_runq_intr_1_ctxnotvalid(struct nvkm_runq *runq, int chid)
+{
+	struct nvkm_fifo *fifo = runq->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+
+	RUNQ_ERROR(runq, "CTXNOTVALID chid:%d", chid);
+
+	chan = nvkm_chan_get_chid(&fifo->engine, chid, &flags);
+	if (WARN_ON_ONCE(!chan))
+		return false;
+
+	nvkm_chan_error(chan, true);
+	nvkm_chan_put(&chan, flags);
+
+	nvkm_mask(device, 0x0400ac + (runq->id * 0x2000), 0x00030000, 0x00030000);
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x2000), 0x80000000);
+	return true;
+}
+
+const struct nvkm_runq_func
+gv100_runq = {
+	.init = gk208_runq_init,
+	.intr = gk104_runq_intr,
+	.intr_0_names = gk104_runq_intr_0_names,
+	.intr_1_ctxnotvalid = gv100_runq_intr_1_ctxnotvalid,
+	.idle = gk104_runq_idle,
+};
+
+void
+gv100_runl_preempt(struct nvkm_runl *runl)
+{
+	nvkm_wr32(runl->fifo->engine.subdev.device, 0x002638, BIT(runl->id));
+}
+
+void
+gv100_runl_insert_chan(struct nvkm_chan *chan, struct nvkm_memory *memory, u64 offset)
+{
+	const u64 user = nvkm_memory_addr(chan->userd.mem) + chan->userd.base;
+	const u64 inst = chan->inst->addr;
+
+	nvkm_wo32(memory, offset + 0x0, lower_32_bits(user) | chan->runq << 1);
+	nvkm_wo32(memory, offset + 0x4, upper_32_bits(user));
+	nvkm_wo32(memory, offset + 0x8, lower_32_bits(inst) | chan->id);
+	nvkm_wo32(memory, offset + 0xc, upper_32_bits(inst));
+}
+
+void
+gv100_runl_insert_cgrp(struct nvkm_cgrp *cgrp, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset + 0x0, (128 << 24) | (3 << 16) | 0x00000001);
+	nvkm_wo32(memory, offset + 0x4, cgrp->chan_nr);
+	nvkm_wo32(memory, offset + 0x8, cgrp->id);
+	nvkm_wo32(memory, offset + 0xc, 0x00000000);
+}
+
+static const struct nvkm_runl_func
+gv100_runl = {
+	.runqs = 2,
+	.size = 16,
+	.update = nv50_runl_update,
+	.insert_cgrp = gv100_runl_insert_cgrp,
+	.insert_chan = gv100_runl_insert_chan,
+	.commit = gk104_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = gk104_runl_pending,
+	.block = gk104_runl_block,
+	.allow = gk104_runl_allow,
+	.preempt = gv100_runl_preempt,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+const struct nvkm_enum
+gv100_fifo_mmu_fault_gpcclient[] = {
+	{ 0x00, "T1_0" },
+	{ 0x01, "T1_1" },
+	{ 0x02, "T1_2" },
+	{ 0x03, "T1_3" },
+	{ 0x04, "T1_4" },
+	{ 0x05, "T1_5" },
+	{ 0x06, "T1_6" },
+	{ 0x07, "T1_7" },
+	{ 0x08, "PE_0" },
+	{ 0x09, "PE_1" },
+	{ 0x0a, "PE_2" },
+	{ 0x0b, "PE_3" },
+	{ 0x0c, "PE_4" },
+	{ 0x0d, "PE_5" },
+	{ 0x0e, "PE_6" },
+	{ 0x0f, "PE_7" },
+	{ 0x10, "RAST" },
+	{ 0x11, "GCC" },
+	{ 0x12, "GPCCS" },
+	{ 0x13, "PROP_0" },
+	{ 0x14, "PROP_1" },
+	{ 0x15, "PROP_2" },
+	{ 0x16, "PROP_3" },
+	{ 0x17, "GPM" },
+	{ 0x18, "LTP_UTLB_0" },
+	{ 0x19, "LTP_UTLB_1" },
+	{ 0x1a, "LTP_UTLB_2" },
+	{ 0x1b, "LTP_UTLB_3" },
+	{ 0x1c, "LTP_UTLB_4" },
+	{ 0x1d, "LTP_UTLB_5" },
+	{ 0x1e, "LTP_UTLB_6" },
+	{ 0x1f, "LTP_UTLB_7" },
+	{ 0x20, "RGG_UTLB" },
+	{ 0x21, "T1_8" },
+	{ 0x22, "T1_9" },
+	{ 0x23, "T1_10" },
+	{ 0x24, "T1_11" },
+	{ 0x25, "T1_12" },
+	{ 0x26, "T1_13" },
+	{ 0x27, "T1_14" },
+	{ 0x28, "T1_15" },
+	{ 0x29, "TPCCS_0" },
+	{ 0x2a, "TPCCS_1" },
+	{ 0x2b, "TPCCS_2" },
+	{ 0x2c, "TPCCS_3" },
+	{ 0x2d, "TPCCS_4" },
+	{ 0x2e, "TPCCS_5" },
+	{ 0x2f, "TPCCS_6" },
+	{ 0x30, "TPCCS_7" },
+	{ 0x31, "PE_8" },
+	{ 0x32, "PE_9" },
+	{ 0x33, "TPCCS_8" },
+	{ 0x34, "TPCCS_9" },
+	{ 0x35, "T1_16" },
+	{ 0x36, "T1_17" },
+	{ 0x37, "T1_18" },
+	{ 0x38, "T1_19" },
+	{ 0x39, "PE_10" },
+	{ 0x3a, "PE_11" },
+	{ 0x3b, "TPCCS_10" },
+	{ 0x3c, "TPCCS_11" },
+	{ 0x3d, "T1_20" },
+	{ 0x3e, "T1_21" },
+	{ 0x3f, "T1_22" },
+	{ 0x40, "T1_23" },
+	{ 0x41, "PE_12" },
+	{ 0x42, "PE_13" },
+	{ 0x43, "TPCCS_12" },
+	{ 0x44, "TPCCS_13" },
+	{ 0x45, "T1_24" },
+	{ 0x46, "T1_25" },
+	{ 0x47, "T1_26" },
+	{ 0x48, "T1_27" },
+	{ 0x49, "PE_14" },
+	{ 0x4a, "PE_15" },
+	{ 0x4b, "TPCCS_14" },
+	{ 0x4c, "TPCCS_15" },
+	{ 0x4d, "T1_28" },
+	{ 0x4e, "T1_29" },
+	{ 0x4f, "T1_30" },
+	{ 0x50, "T1_31" },
+	{ 0x51, "PE_16" },
+	{ 0x52, "PE_17" },
+	{ 0x53, "TPCCS_16" },
+	{ 0x54, "TPCCS_17" },
+	{ 0x55, "T1_32" },
+	{ 0x56, "T1_33" },
+	{ 0x57, "T1_34" },
+	{ 0x58, "T1_35" },
+	{ 0x59, "PE_18" },
+	{ 0x5a, "PE_19" },
+	{ 0x5b, "TPCCS_18" },
+	{ 0x5c, "TPCCS_19" },
+	{ 0x5d, "T1_36" },
+	{ 0x5e, "T1_37" },
+	{ 0x5f, "T1_38" },
+	{ 0x60, "T1_39" },
+	{}
+};
+
+const struct nvkm_enum
+gv100_fifo_mmu_fault_hubclient[] = {
+	{ 0x00, "VIP" },
+	{ 0x01, "CE0" },
+	{ 0x02, "CE1" },
+	{ 0x03, "DNISO" },
+	{ 0x04, "FE" },
+	{ 0x05, "FECS" },
+	{ 0x06, "HOST" },
+	{ 0x07, "HOST_CPU" },
+	{ 0x08, "HOST_CPU_NB" },
+	{ 0x09, "ISO" },
+	{ 0x0a, "MMU" },
+	{ 0x0b, "NVDEC" },
+	{ 0x0d, "NVENC1" },
+	{ 0x0e, "NISO" },
+	{ 0x0f, "P2P" },
+	{ 0x10, "PD" },
+	{ 0x11, "PERF" },
+	{ 0x12, "PMU" },
+	{ 0x13, "RASTERTWOD" },
+	{ 0x14, "SCC" },
+	{ 0x15, "SCC_NB" },
+	{ 0x16, "SEC" },
+	{ 0x17, "SSYNC" },
+	{ 0x18, "CE2" },
+	{ 0x19, "XV" },
+	{ 0x1a, "MMU_NB" },
+	{ 0x1b, "NVENC0" },
+	{ 0x1c, "DFALCON" },
+	{ 0x1d, "SKED" },
+	{ 0x1e, "AFALCON" },
+	{ 0x1f, "DONT_CARE" },
+	{ 0x20, "HSCE0" },
+	{ 0x21, "HSCE1" },
+	{ 0x22, "HSCE2" },
+	{ 0x23, "HSCE3" },
+	{ 0x24, "HSCE4" },
+	{ 0x25, "HSCE5" },
+	{ 0x26, "HSCE6" },
+	{ 0x27, "HSCE7" },
+	{ 0x28, "HSCE8" },
+	{ 0x29, "HSCE9" },
+	{ 0x2a, "HSHUB" },
+	{ 0x2b, "PTP_X0" },
+	{ 0x2c, "PTP_X1" },
+	{ 0x2d, "PTP_X2" },
+	{ 0x2e, "PTP_X3" },
+	{ 0x2f, "PTP_X4" },
+	{ 0x30, "PTP_X5" },
+	{ 0x31, "PTP_X6" },
+	{ 0x32, "PTP_X7" },
+	{ 0x33, "NVENC2" },
+	{ 0x34, "VPR_SCRUBBER0" },
+	{ 0x35, "VPR_SCRUBBER1" },
+	{ 0x36, "DWBIF" },
+	{ 0x37, "FBFALCON" },
+	{ 0x38, "CE_SHIM" },
+	{ 0x39, "GSP" },
+	{}
+};
+
+const struct nvkm_enum
+gv100_fifo_mmu_fault_reason[] = {
+	{ 0x00, "PDE" },
+	{ 0x01, "PDE_SIZE" },
+	{ 0x02, "PTE" },
+	{ 0x03, "VA_LIMIT_VIOLATION" },
+	{ 0x04, "UNBOUND_INST_BLOCK" },
+	{ 0x05, "PRIV_VIOLATION" },
+	{ 0x06, "RO_VIOLATION" },
+	{ 0x07, "WO_VIOLATION" },
+	{ 0x08, "PITCH_MASK_VIOLATION" },
+	{ 0x09, "WORK_CREATION" },
+	{ 0x0a, "UNSUPPORTED_APERTURE" },
+	{ 0x0b, "COMPRESSION_FAILURE" },
+	{ 0x0c, "UNSUPPORTED_KIND" },
+	{ 0x0d, "REGION_VIOLATION" },
+	{ 0x0e, "POISONED" },
+	{ 0x0f, "ATOMIC_VIOLATION" },
+	{}
+};
+
+static const struct nvkm_enum
+gv100_fifo_mmu_fault_engine[] = {
+	{ 0x01, "DISPLAY" },
+	{ 0x03, "PTP" },
+	{ 0x04, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0x05, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
+	{ 0x06, "PWR_PMU" },
+	{ 0x08, "IFB", NULL, NVKM_ENGINE_IFB },
+	{ 0x09, "PERF" },
+	{ 0x1f, "PHYSICAL" },
+	{ 0x20, "HOST0" },
+	{ 0x21, "HOST1" },
+	{ 0x22, "HOST2" },
+	{ 0x23, "HOST3" },
+	{ 0x24, "HOST4" },
+	{ 0x25, "HOST5" },
+	{ 0x26, "HOST6" },
+	{ 0x27, "HOST7" },
+	{ 0x28, "HOST8" },
+	{ 0x29, "HOST9" },
+	{ 0x2a, "HOST10" },
+	{ 0x2b, "HOST11" },
+	{ 0x2c, "HOST12" },
+	{ 0x2d, "HOST13" },
+	{}
+};
+
+const struct nvkm_enum
+gv100_fifo_mmu_fault_access[] = {
+	{ 0x0, "VIRT_READ" },
+	{ 0x1, "VIRT_WRITE" },
+	{ 0x2, "VIRT_ATOMIC" },
+	{ 0x3, "VIRT_PREFETCH" },
+	{ 0x4, "VIRT_ATOMIC_WEAK" },
+	{ 0x8, "PHYS_READ" },
+	{ 0x9, "PHYS_WRITE" },
+	{ 0xa, "PHYS_ATOMIC" },
+	{ 0xb, "PHYS_PREFETCH" },
+	{}
+};
+
+static const struct nvkm_fifo_func_mmu_fault
+gv100_fifo_mmu_fault = {
+	.recover = gf100_fifo_mmu_fault_recover,
+	.access = gv100_fifo_mmu_fault_access,
+	.engine = gv100_fifo_mmu_fault_engine,
+	.reason = gv100_fifo_mmu_fault_reason,
+	.hubclient = gv100_fifo_mmu_fault_hubclient,
+	.gpcclient = gv100_fifo_mmu_fault_gpcclient,
+};
+
+static void
+gv100_fifo_intr_ctxsw_timeout(struct nvkm_fifo *fifo, u32 engm)
+{
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+
+	nvkm_runl_foreach(runl, fifo) {
+		nvkm_runl_foreach_engn_cond(engn, runl, engm & BIT(engn->id))
+			nvkm_runl_rc_engn(runl, engn);
+	}
+}
+
+static const struct nvkm_fifo_func
+gv100_fifo = {
+	.chid_nr = gm200_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gm200_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = gk104_fifo_init_pbdmas,
+	.intr = gk104_fifo_intr,
+	.intr_ctxsw_timeout = gv100_fifo_intr_ctxsw_timeout,
+	.mmu_fault = &gv100_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &gv100_runl,
+	.runq = &gv100_runq,
+	.engn = &gv100_engn,
+	.engn_ce = &gv100_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp, .force = true },
+	.chan = {{ 0, 0,  VOLTA_CHANNEL_GPFIFO_A }, &gv100_chan },
+};
+
+int
+gv100_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&gv100_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
new file mode 100644
index 000000000..674faf002
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
@@ -0,0 +1,543 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+
+#include "regsnv04.h"
+
+#include <core/ramht.h>
+#include <subdev/instmem.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <engine/sw.h>
+
+#include <nvif/class.h>
+
+void
+nv04_chan_stop(struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_memory *fctx = device->imem->ramfc;
+	const struct nvkm_ramfc_layout *c;
+	unsigned long flags;
+	u32 data = chan->ramfc_offset;
+	u32 chid;
+
+	/* prevent fifo context switches */
+	spin_lock_irqsave(&fifo->lock, flags);
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 0);
+
+	/* if this channel is active, replace it with a null context */
+	chid = nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH1) & fifo->chid->mask;
+	if (chid == chan->id) {
+		nvkm_mask(device, NV04_PFIFO_CACHE1_DMA_PUSH, 0x00000001, 0);
+		nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0, 0);
+		nvkm_mask(device, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0);
+
+		c = chan->func->ramfc->layout;
+		nvkm_kmap(fctx);
+		do {
+			u32 rm = ((1ULL << c->bits) - 1) << c->regs;
+			u32 cm = ((1ULL << c->bits) - 1) << c->ctxs;
+			u32 rv = (nvkm_rd32(device, c->regp) &  rm) >> c->regs;
+			u32 cv = (nvkm_ro32(fctx, c->ctxp + data) & ~cm);
+			nvkm_wo32(fctx, c->ctxp + data, cv | (rv << c->ctxs));
+		} while ((++c)->bits);
+		nvkm_done(fctx);
+
+		c = chan->func->ramfc->layout;
+		do {
+			nvkm_wr32(device, c->regp, 0x00000000);
+		} while ((++c)->bits);
+
+		nvkm_wr32(device, NV03_PFIFO_CACHE1_GET, 0);
+		nvkm_wr32(device, NV03_PFIFO_CACHE1_PUT, 0);
+		nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH1, fifo->chid->mask);
+		nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0, 1);
+		nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
+	}
+
+	/* restore normal operation, after disabling dma mode */
+	nvkm_mask(device, NV04_PFIFO_MODE, BIT(chan->id), 0);
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 1);
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+void
+nv04_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	nvkm_mask(fifo->engine.subdev.device, NV04_PFIFO_MODE, BIT(chan->id), BIT(chan->id));
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+void
+nv04_chan_ramfc_clear(struct nvkm_chan *chan)
+{
+	struct nvkm_memory *ramfc = chan->cgrp->runl->fifo->engine.subdev.device->imem->ramfc;
+	const struct nvkm_ramfc_layout *c = chan->func->ramfc->layout;
+
+	nvkm_kmap(ramfc);
+	do {
+		nvkm_wo32(ramfc, chan->ramfc_offset + c->ctxp, 0x00000000);
+	} while ((++c)->bits);
+	nvkm_done(ramfc);
+}
+
+static int
+nv04_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	struct nvkm_memory *ramfc = chan->cgrp->runl->fifo->engine.subdev.device->imem->ramfc;
+	const u32 base = chan->id * 32;
+
+	chan->ramfc_offset = base;
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, base + 0x00, offset);
+	nvkm_wo32(ramfc, base + 0x04, offset);
+	nvkm_wo32(ramfc, base + 0x08, chan->push->addr >> 4);
+	nvkm_wo32(ramfc, base + 0x10, NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+				      NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+				      NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+				      NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_done(ramfc);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+nv04_chan_ramfc = {
+	.layout = (const struct nvkm_ramfc_layout[]) {
+		{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+		{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+		{ 16,  0, 0x08,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+		{ 16, 16, 0x08,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+		{ 32,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+		{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+		{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_ENGINE },
+		{ 32,  0, 0x18,  0, NV04_PFIFO_CACHE1_PULL1 },
+		{}
+	},
+	.write = nv04_chan_ramfc_write,
+	.clear = nv04_chan_ramfc_clear,
+	.ctxdma = true,
+};
+
+const struct nvkm_chan_func_userd
+nv04_chan_userd = {
+	.bar = 0,
+	.base = 0x800000,
+	.size = 0x010000,
+};
+
+const struct nvkm_chan_func_inst
+nv04_chan_inst = {
+	.size = 0x1000,
+};
+
+static const struct nvkm_chan_func
+nv04_chan = {
+	.inst = &nv04_chan_inst,
+	.userd = &nv04_chan_userd,
+	.ramfc = &nv04_chan_ramfc,
+	.start = nv04_chan_start,
+	.stop = nv04_chan_stop,
+};
+
+const struct nvkm_cgrp_func
+nv04_cgrp = {
+};
+
+void
+nv04_eobj_ramht_del(struct nvkm_chan *chan, int hash)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_instmem *imem = fifo->engine.subdev.device->imem;
+
+	mutex_lock(&fifo->mutex);
+	nvkm_ramht_remove(imem->ramht, hash);
+	mutex_unlock(&fifo->mutex);
+}
+
+static int
+nv04_eobj_ramht_add(struct nvkm_engn *engn, struct nvkm_object *eobj, struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_instmem *imem = fifo->engine.subdev.device->imem;
+	u32 context = 0x80000000 | chan->id << 24 | engn->id << 16;
+	int hash;
+
+	mutex_lock(&fifo->mutex);
+	hash = nvkm_ramht_insert(imem->ramht, eobj, chan->id, 4, eobj->handle, context);
+	mutex_unlock(&fifo->mutex);
+	return hash;
+}
+
+const struct nvkm_engn_func
+nv04_engn = {
+	.ramht_add = nv04_eobj_ramht_add,
+	.ramht_del = nv04_eobj_ramht_del,
+};
+
+void
+nv04_fifo_pause(struct nvkm_fifo *fifo, unsigned long *pflags)
+__acquires(fifo->lock)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	*pflags = flags;
+
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 0x00000000);
+	nvkm_mask(device, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0x00000000);
+
+	/* in some cases the puller may be left in an inconsistent state
+	 * if you try to stop it while it's busy translating handles.
+	 * sometimes you get a CACHE_ERROR, sometimes it just fails
+	 * silently; sending incorrect instance offsets to PGRAPH after
+	 * it's started up again.
+	 *
+	 * to avoid this, we invalidate the most recently calculated
+	 * instance.
+	 */
+	nvkm_msec(device, 2000,
+		u32 tmp = nvkm_rd32(device, NV04_PFIFO_CACHE1_PULL0);
+		if (!(tmp & NV04_PFIFO_CACHE1_PULL0_HASH_BUSY))
+			break;
+	);
+
+	if (nvkm_rd32(device, NV04_PFIFO_CACHE1_PULL0) &
+			  NV04_PFIFO_CACHE1_PULL0_HASH_FAILED)
+		nvkm_wr32(device, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_CACHE_ERROR);
+
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_HASH, 0x00000000);
+}
+
+void
+nv04_fifo_start(struct nvkm_fifo *fifo, unsigned long *pflags)
+__releases(fifo->lock)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	unsigned long flags = *pflags;
+
+	nvkm_mask(device, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0x00000001);
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 0x00000001);
+
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+const struct nvkm_runl_func
+nv04_runl = {
+};
+
+static const char *
+nv_dma_state_err(u32 state)
+{
+	static const char * const desc[] = {
+		"NONE", "CALL_SUBR_ACTIVE", "INVALID_MTHD", "RET_SUBR_INACTIVE",
+		"INVALID_CMD", "IB_EMPTY"/* NV50+ */, "MEM_FAULT", "UNK"
+	};
+	return desc[(state >> 29) & 0x7];
+}
+
+static bool
+nv04_fifo_swmthd(struct nvkm_device *device, u32 chid, u32 addr, u32 data)
+{
+	struct nvkm_sw *sw = device->sw;
+	const int subc = (addr & 0x0000e000) >> 13;
+	const int mthd = (addr & 0x00001ffc);
+	const u32 mask = 0x0000000f << (subc * 4);
+	u32 engine = nvkm_rd32(device, 0x003280);
+	bool handled = false;
+
+	switch (mthd) {
+	case 0x0000 ... 0x0000: /* subchannel's engine -> software */
+		nvkm_wr32(device, 0x003280, (engine &= ~mask));
+		fallthrough;
+	case 0x0180 ... 0x01fc: /* handle -> instance */
+		data = nvkm_rd32(device, 0x003258) & 0x0000ffff;
+		fallthrough;
+	case 0x0100 ... 0x017c:
+	case 0x0200 ... 0x1ffc: /* pass method down to sw */
+		if (!(engine & mask) && sw)
+			handled = nvkm_sw_mthd(sw, chid, subc, mthd, data);
+		break;
+	default:
+		break;
+	}
+
+	return handled;
+}
+
+static void
+nv04_fifo_intr_cache_error(struct nvkm_fifo *fifo, u32 chid, u32 get)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+	u32 pull0 = nvkm_rd32(device, 0x003250);
+	u32 mthd, data;
+	int ptr;
+
+	/* NV_PFIFO_CACHE1_GET actually goes to 0xffc before wrapping on my
+	 * G80 chips, but CACHE1 isn't big enough for this much data.. Tests
+	 * show that it wraps around to the start at GET=0x800.. No clue as to
+	 * why..
+	 */
+	ptr = (get & 0x7ff) >> 2;
+
+	if (device->card_type < NV_40) {
+		mthd = nvkm_rd32(device, NV04_PFIFO_CACHE1_METHOD(ptr));
+		data = nvkm_rd32(device, NV04_PFIFO_CACHE1_DATA(ptr));
+	} else {
+		mthd = nvkm_rd32(device, NV40_PFIFO_CACHE1_METHOD(ptr));
+		data = nvkm_rd32(device, NV40_PFIFO_CACHE1_DATA(ptr));
+	}
+
+	if (!(pull0 & 0x00000100) ||
+	    !nv04_fifo_swmthd(device, chid, mthd, data)) {
+		chan = nvkm_chan_get_chid(&fifo->engine, chid, &flags);
+		nvkm_error(subdev, "CACHE_ERROR - "
+			   "ch %d [%s] subc %d mthd %04x data %08x\n",
+			   chid, chan ? chan->name : "unknown",
+			   (mthd >> 13) & 7, mthd & 0x1ffc, data);
+		nvkm_chan_put(&chan, flags);
+	}
+
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_DMA_PUSH, 0);
+	nvkm_wr32(device, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_CACHE_ERROR);
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0,
+		nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH0) & ~1);
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_GET, get + 4);
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0,
+		nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH0) | 1);
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_HASH, 0);
+
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_DMA_PUSH,
+		nvkm_rd32(device, NV04_PFIFO_CACHE1_DMA_PUSH) | 1);
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
+}
+
+static void
+nv04_fifo_intr_dma_pusher(struct nvkm_fifo *fifo, u32 chid)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 dma_get = nvkm_rd32(device, 0x003244);
+	u32 dma_put = nvkm_rd32(device, 0x003240);
+	u32 push = nvkm_rd32(device, 0x003220);
+	u32 state = nvkm_rd32(device, 0x003228);
+	struct nvkm_chan *chan;
+	unsigned long flags;
+	const char *name;
+
+	chan = nvkm_chan_get_chid(&fifo->engine, chid, &flags);
+	name = chan ? chan->name : "unknown";
+	if (device->card_type == NV_50) {
+		u32 ho_get = nvkm_rd32(device, 0x003328);
+		u32 ho_put = nvkm_rd32(device, 0x003320);
+		u32 ib_get = nvkm_rd32(device, 0x003334);
+		u32 ib_put = nvkm_rd32(device, 0x003330);
+
+		nvkm_error(subdev, "DMA_PUSHER - "
+			   "ch %d [%s] get %02x%08x put %02x%08x ib_get %08x "
+			   "ib_put %08x state %08x (err: %s) push %08x\n",
+			   chid, name, ho_get, dma_get, ho_put, dma_put,
+			   ib_get, ib_put, state, nv_dma_state_err(state),
+			   push);
+
+		/* METHOD_COUNT, in DMA_STATE on earlier chipsets */
+		nvkm_wr32(device, 0x003364, 0x00000000);
+		if (dma_get != dma_put || ho_get != ho_put) {
+			nvkm_wr32(device, 0x003244, dma_put);
+			nvkm_wr32(device, 0x003328, ho_put);
+		} else
+		if (ib_get != ib_put)
+			nvkm_wr32(device, 0x003334, ib_put);
+	} else {
+		nvkm_error(subdev, "DMA_PUSHER - ch %d [%s] get %08x put %08x "
+				   "state %08x (err: %s) push %08x\n",
+			   chid, name, dma_get, dma_put, state,
+			   nv_dma_state_err(state), push);
+
+		if (dma_get != dma_put)
+			nvkm_wr32(device, 0x003244, dma_put);
+	}
+	nvkm_chan_put(&chan, flags);
+
+	nvkm_wr32(device, 0x003228, 0x00000000);
+	nvkm_wr32(device, 0x003220, 0x00000001);
+	nvkm_wr32(device, 0x002100, NV_PFIFO_INTR_DMA_PUSHER);
+}
+
+irqreturn_t
+nv04_fifo_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_fifo *fifo = container_of(inth, typeof(*fifo), engine.subdev.inth);
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, NV03_PFIFO_INTR_EN_0);
+	u32 stat = nvkm_rd32(device, NV03_PFIFO_INTR_0) & mask;
+	u32 reassign, chid, get, sem;
+
+	reassign = nvkm_rd32(device, NV03_PFIFO_CACHES) & 1;
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 0);
+
+	chid = nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH1) & fifo->chid->mask;
+	get  = nvkm_rd32(device, NV03_PFIFO_CACHE1_GET);
+
+	if (stat & NV_PFIFO_INTR_CACHE_ERROR) {
+		nv04_fifo_intr_cache_error(fifo, chid, get);
+		stat &= ~NV_PFIFO_INTR_CACHE_ERROR;
+	}
+
+	if (stat & NV_PFIFO_INTR_DMA_PUSHER) {
+		nv04_fifo_intr_dma_pusher(fifo, chid);
+		stat &= ~NV_PFIFO_INTR_DMA_PUSHER;
+	}
+
+	if (stat & NV_PFIFO_INTR_SEMAPHORE) {
+		stat &= ~NV_PFIFO_INTR_SEMAPHORE;
+		nvkm_wr32(device, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_SEMAPHORE);
+
+		sem = nvkm_rd32(device, NV10_PFIFO_CACHE1_SEMAPHORE);
+		nvkm_wr32(device, NV10_PFIFO_CACHE1_SEMAPHORE, sem | 0x1);
+
+		nvkm_wr32(device, NV03_PFIFO_CACHE1_GET, get + 4);
+		nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
+	}
+
+	if (device->card_type == NV_50) {
+		if (stat & 0x00000010) {
+			stat &= ~0x00000010;
+			nvkm_wr32(device, 0x002100, 0x00000010);
+		}
+
+		if (stat & 0x40000000) {
+			nvkm_wr32(device, 0x002100, 0x40000000);
+			nvkm_event_ntfy(&fifo->nonstall.event, 0, NVKM_FIFO_NONSTALL_EVENT);
+			stat &= ~0x40000000;
+		}
+	}
+
+	if (stat) {
+		nvkm_warn(subdev, "intr %08x\n", stat);
+		nvkm_mask(device, NV03_PFIFO_INTR_EN_0, stat, 0x00000000);
+		nvkm_wr32(device, NV03_PFIFO_INTR_0, stat);
+	}
+
+	nvkm_wr32(device, NV03_PFIFO_CACHES, reassign);
+	return IRQ_HANDLED;
+}
+
+void
+nv04_fifo_init(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_instmem *imem = device->imem;
+	struct nvkm_ramht *ramht = imem->ramht;
+	struct nvkm_memory *ramro = imem->ramro;
+	struct nvkm_memory *ramfc = imem->ramfc;
+
+	nvkm_wr32(device, NV04_PFIFO_DELAY_0, 0x000000ff);
+	nvkm_wr32(device, NV04_PFIFO_DMA_TIMESLICE, 0x0101ffff);
+
+	nvkm_wr32(device, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
+					    ((ramht->bits - 9) << 16) |
+					    (ramht->gpuobj->addr >> 8));
+	nvkm_wr32(device, NV03_PFIFO_RAMRO, nvkm_memory_addr(ramro) >> 8);
+	nvkm_wr32(device, NV03_PFIFO_RAMFC, nvkm_memory_addr(ramfc) >> 8);
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH1, fifo->chid->mask);
+
+	nvkm_wr32(device, NV03_PFIFO_INTR_0, 0xffffffff);
+	nvkm_wr32(device, NV03_PFIFO_INTR_EN_0, 0xffffffff);
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0, 1);
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 1);
+}
+
+int
+nv04_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_runl *runl;
+
+	runl = nvkm_runl_new(fifo, 0, 0, 0);
+	if (IS_ERR(runl))
+		return PTR_ERR(runl);
+
+	nvkm_runl_add(runl, 0, fifo->func->engn_sw, NVKM_ENGINE_SW, 0);
+	nvkm_runl_add(runl, 0, fifo->func->engn_sw, NVKM_ENGINE_DMAOBJ, 0);
+	nvkm_runl_add(runl, 1, fifo->func->engn   , NVKM_ENGINE_GR, 0);
+	nvkm_runl_add(runl, 2, fifo->func->engn   , NVKM_ENGINE_MPEG, 0); /* NV31- */
+	return 0;
+}
+
+int
+nv04_fifo_chid_ctor(struct nvkm_fifo *fifo, int nr)
+{
+	/* The last CHID is reserved by HW as a "channel invalid" marker. */
+	return nvkm_chid_new(&nvkm_chan_event, &fifo->engine.subdev, nr, 0, nr - 1, &fifo->chid);
+}
+
+static int
+nv04_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return 16;
+}
+
+static const struct nvkm_fifo_func
+nv04_fifo = {
+	.chid_nr = nv04_fifo_chid_nr,
+	.chid_ctor = nv04_fifo_chid_ctor,
+	.runl_ctor = nv04_fifo_runl_ctor,
+	.init = nv04_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.runl = &nv04_runl,
+	.engn = &nv04_engn,
+	.engn_sw = &nv04_engn,
+	.cgrp = {{                        }, &nv04_cgrp },
+	.chan = {{ 0, 0, NV03_CHANNEL_DMA }, &nv04_chan },
+};
+
+int
+nv04_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	      struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&nv04_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
new file mode 100644
index 000000000..a4bcf6b0a
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
@@ -0,0 +1,113 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "runl.h"
+
+#include <core/gpuobj.h>
+#include <subdev/instmem.h>
+
+#include "regsnv04.h"
+
+#include <nvif/class.h>
+
+static int
+nv10_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	struct nvkm_memory *ramfc = chan->cgrp->runl->fifo->engine.subdev.device->imem->ramfc;
+	const u32 base = chan->id * 32;
+
+	chan->ramfc_offset = base;
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, base + 0x00, offset);
+	nvkm_wo32(ramfc, base + 0x04, offset);
+	nvkm_wo32(ramfc, base + 0x0c, chan->push->addr >> 4);
+	nvkm_wo32(ramfc, base + 0x14, NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+				      NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+				      NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+				      NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_done(ramfc);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+nv10_chan_ramfc = {
+	.layout = (const struct nvkm_ramfc_layout[]) {
+		{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+		{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+		{ 32,  0, 0x08,  0, NV10_PFIFO_CACHE1_REF_CNT },
+		{ 16,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+		{ 16, 16, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+		{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+		{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+		{ 32,  0, 0x18,  0, NV04_PFIFO_CACHE1_ENGINE },
+		{ 32,  0, 0x1c,  0, NV04_PFIFO_CACHE1_PULL1 },
+		{}
+	},
+	.write = nv10_chan_ramfc_write,
+	.clear = nv04_chan_ramfc_clear,
+	.ctxdma = true,
+};
+
+static const struct nvkm_chan_func
+nv10_chan = {
+	.inst = &nv04_chan_inst,
+	.userd = &nv04_chan_userd,
+	.ramfc = &nv10_chan_ramfc,
+	.start = nv04_chan_start,
+	.stop = nv04_chan_stop,
+};
+
+int
+nv10_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return 32;
+}
+
+static const struct nvkm_fifo_func
+nv10_fifo = {
+	.chid_nr = nv10_fifo_chid_nr,
+	.chid_ctor = nv04_fifo_chid_ctor,
+	.runl_ctor = nv04_fifo_runl_ctor,
+	.init = nv04_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.runl = &nv04_runl,
+	.engn = &nv04_engn,
+	.engn_sw = &nv04_engn,
+	.cgrp = {{                        }, &nv04_cgrp },
+	.chan = {{ 0, 0, NV10_CHANNEL_DMA }, &nv10_chan },
+};
+
+int
+nv10_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	      struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&nv10_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
new file mode 100644
index 000000000..c70f44fd4
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
@@ -0,0 +1,142 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+
+#include "regsnv04.h"
+
+#include <core/ramht.h>
+#include <subdev/instmem.h>
+
+#include <nvif/class.h>
+
+static int
+nv17_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	struct nvkm_memory *ramfc = chan->cgrp->runl->fifo->engine.subdev.device->imem->ramfc;
+	const u32 base = chan->id * 64;
+
+	chan->ramfc_offset = base;
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, base + 0x00, offset);
+	nvkm_wo32(ramfc, base + 0x04, offset);
+	nvkm_wo32(ramfc, base + 0x0c, chan->push->addr >> 4);
+	nvkm_wo32(ramfc, base + 0x14, NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+				      NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+				      NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+				      NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_done(ramfc);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+nv17_chan_ramfc = {
+	.layout = (const struct nvkm_ramfc_layout[]) {
+		{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+		{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+		{ 32,  0, 0x08,  0, NV10_PFIFO_CACHE1_REF_CNT },
+		{ 16,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+		{ 16, 16, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+		{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+		{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+		{ 32,  0, 0x18,  0, NV04_PFIFO_CACHE1_ENGINE },
+		{ 32,  0, 0x1c,  0, NV04_PFIFO_CACHE1_PULL1 },
+		{ 32,  0, 0x20,  0, NV10_PFIFO_CACHE1_ACQUIRE_VALUE },
+		{ 32,  0, 0x24,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP },
+		{ 32,  0, 0x28,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT },
+		{ 32,  0, 0x2c,  0, NV10_PFIFO_CACHE1_SEMAPHORE },
+		{ 32,  0, 0x30,  0, NV10_PFIFO_CACHE1_DMA_SUBROUTINE },
+		{}
+	},
+	.write = nv17_chan_ramfc_write,
+	.clear = nv04_chan_ramfc_clear,
+	.ctxdma = true,
+};
+
+static const struct nvkm_chan_func
+nv17_chan = {
+	.inst = &nv04_chan_inst,
+	.userd = &nv04_chan_userd,
+	.ramfc = &nv17_chan_ramfc,
+	.start = nv04_chan_start,
+	.stop = nv04_chan_stop,
+};
+
+static void
+nv17_fifo_init(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_instmem *imem = device->imem;
+	struct nvkm_ramht *ramht = imem->ramht;
+	struct nvkm_memory *ramro = imem->ramro;
+	struct nvkm_memory *ramfc = imem->ramfc;
+
+	nvkm_wr32(device, NV04_PFIFO_DELAY_0, 0x000000ff);
+	nvkm_wr32(device, NV04_PFIFO_DMA_TIMESLICE, 0x0101ffff);
+
+	nvkm_wr32(device, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
+					    ((ramht->bits - 9) << 16) |
+					    (ramht->gpuobj->addr >> 8));
+	nvkm_wr32(device, NV03_PFIFO_RAMRO, nvkm_memory_addr(ramro) >> 8);
+	nvkm_wr32(device, NV03_PFIFO_RAMFC, nvkm_memory_addr(ramfc) >> 8 |
+					    0x00010000);
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH1, fifo->chid->mask);
+
+	nvkm_wr32(device, NV03_PFIFO_INTR_0, 0xffffffff);
+	nvkm_wr32(device, NV03_PFIFO_INTR_EN_0, 0xffffffff);
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0, 1);
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 1);
+}
+
+static const struct nvkm_fifo_func
+nv17_fifo = {
+	.chid_nr = nv10_fifo_chid_nr,
+	.chid_ctor = nv04_fifo_chid_ctor,
+	.runl_ctor = nv04_fifo_runl_ctor,
+	.init = nv17_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.runl = &nv04_runl,
+	.engn = &nv04_engn,
+	.engn_sw = &nv04_engn,
+	.cgrp = {{                        }, &nv04_cgrp },
+	.chan = {{ 0, 0, NV17_CHANNEL_DMA }, &nv17_chan },
+};
+
+int
+nv17_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	      struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&nv17_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
new file mode 100644
index 000000000..e50a94b6d
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
@@ -0,0 +1,252 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+
+#include "regsnv04.h"
+
+#include <core/ramht.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+
+#include <nvif/class.h>
+
+static int
+nv40_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	struct nvkm_memory *ramfc = chan->cgrp->runl->fifo->engine.subdev.device->imem->ramfc;
+	const u32 base = chan->id * 128;
+
+	chan->ramfc_offset = base;
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, base + 0x00, offset);
+	nvkm_wo32(ramfc, base + 0x04, offset);
+	nvkm_wo32(ramfc, base + 0x0c, chan->push->addr >> 4);
+	nvkm_wo32(ramfc, base + 0x18, 0x30000000 |
+				      NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+				      NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+				      NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+				      NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_wo32(ramfc, base + 0x3c, 0x0001ffff);
+	nvkm_done(ramfc);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+nv40_chan_ramfc = {
+	.layout = (const struct nvkm_ramfc_layout[]) {
+		{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+		{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+		{ 32,  0, 0x08,  0, NV10_PFIFO_CACHE1_REF_CNT },
+		{ 32,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+		{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+		{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+		{ 28,  0, 0x18,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+		{  2, 28, 0x18, 28, 0x002058 },
+		{ 32,  0, 0x1c,  0, NV04_PFIFO_CACHE1_ENGINE },
+		{ 32,  0, 0x20,  0, NV04_PFIFO_CACHE1_PULL1 },
+		{ 32,  0, 0x24,  0, NV10_PFIFO_CACHE1_ACQUIRE_VALUE },
+		{ 32,  0, 0x28,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP },
+		{ 32,  0, 0x2c,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT },
+		{ 32,  0, 0x30,  0, NV10_PFIFO_CACHE1_SEMAPHORE },
+		{ 32,  0, 0x34,  0, NV10_PFIFO_CACHE1_DMA_SUBROUTINE },
+		{ 32,  0, 0x38,  0, NV40_PFIFO_GRCTX_INSTANCE },
+		{ 17,  0, 0x3c,  0, NV04_PFIFO_DMA_TIMESLICE },
+		{ 32,  0, 0x40,  0, 0x0032e4 },
+		{ 32,  0, 0x44,  0, 0x0032e8 },
+		{ 32,  0, 0x4c,  0, 0x002088 },
+		{ 32,  0, 0x50,  0, 0x003300 },
+		{ 32,  0, 0x54,  0, 0x00330c },
+		{}
+	},
+	.write = nv40_chan_ramfc_write,
+	.clear = nv04_chan_ramfc_clear,
+	.ctxdma = true,
+};
+
+static const struct nvkm_chan_func_userd
+nv40_chan_userd = {
+	.bar = 0,
+	.base = 0xc00000,
+	.size = 0x001000,
+};
+
+static const struct nvkm_chan_func
+nv40_chan = {
+	.inst = &nv04_chan_inst,
+	.userd = &nv40_chan_userd,
+	.ramfc = &nv40_chan_ramfc,
+	.start = nv04_chan_start,
+	.stop = nv04_chan_stop,
+};
+
+static int
+nv40_eobj_ramht_add(struct nvkm_engn *engn, struct nvkm_object *eobj, struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_instmem *imem = fifo->engine.subdev.device->imem;
+	u32 context = chan->id << 23 | engn->id << 20;
+	int hash;
+
+	mutex_lock(&fifo->mutex);
+	hash = nvkm_ramht_insert(imem->ramht, eobj, chan->id, 4, eobj->handle, context);
+	mutex_unlock(&fifo->mutex);
+	return hash;
+}
+
+static void
+nv40_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_memory *ramfc = device->imem->ramfc;
+	u32 inst = 0x00000000, reg, ctx;
+	int chid;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_GR:
+		reg = 0x0032e0;
+		ctx = 0x38;
+		break;
+	case NVKM_ENGINE_MPEG:
+		if (WARN_ON(device->chipset < 0x44))
+			return;
+		reg = 0x00330c;
+		ctx = 0x54;
+		break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (cctx)
+		inst = cctx->vctx->inst->addr >> 4;
+
+	spin_lock_irq(&fifo->lock);
+	nvkm_mask(device, 0x002500, 0x00000001, 0x00000000);
+
+	chid = nvkm_rd32(device, 0x003204) & (fifo->chid->nr - 1);
+	if (chid == chan->id)
+		nvkm_wr32(device, reg, inst);
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, chan->ramfc_offset + ctx, inst);
+	nvkm_done(ramfc);
+
+	nvkm_mask(device, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irq(&fifo->lock);
+}
+
+static const struct nvkm_engn_func
+nv40_engn = {
+	.bind = nv40_ectx_bind,
+	.ramht_add = nv40_eobj_ramht_add,
+	.ramht_del = nv04_eobj_ramht_del,
+};
+
+static const struct nvkm_engn_func
+nv40_engn_sw = {
+	.ramht_add = nv40_eobj_ramht_add,
+	.ramht_del = nv04_eobj_ramht_del,
+};
+
+static void
+nv40_fifo_init(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_fb *fb = device->fb;
+	struct nvkm_instmem *imem = device->imem;
+	struct nvkm_ramht *ramht = imem->ramht;
+	struct nvkm_memory *ramro = imem->ramro;
+	struct nvkm_memory *ramfc = imem->ramfc;
+
+	nvkm_wr32(device, 0x002040, 0x000000ff);
+	nvkm_wr32(device, 0x002044, 0x2101ffff);
+	nvkm_wr32(device, 0x002058, 0x00000001);
+
+	nvkm_wr32(device, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
+					    ((ramht->bits - 9) << 16) |
+					    (ramht->gpuobj->addr >> 8));
+	nvkm_wr32(device, NV03_PFIFO_RAMRO, nvkm_memory_addr(ramro) >> 8);
+
+	switch (device->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		nvkm_wr32(device, 0x002230, 0x00000001);
+		fallthrough;
+	case 0x40:
+	case 0x41:
+	case 0x42:
+	case 0x43:
+	case 0x45:
+	case 0x48:
+		nvkm_wr32(device, 0x002220, 0x00030002);
+		break;
+	default:
+		nvkm_wr32(device, 0x002230, 0x00000000);
+		nvkm_wr32(device, 0x002220, ((fb->ram->size - 512 * 1024 +
+					      nvkm_memory_addr(ramfc)) >> 16) |
+					    0x00030000);
+		break;
+	}
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH1, fifo->chid->mask);
+
+	nvkm_wr32(device, NV03_PFIFO_INTR_0, 0xffffffff);
+	nvkm_wr32(device, NV03_PFIFO_INTR_EN_0, 0xffffffff);
+
+	nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0, 1);
+	nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
+	nvkm_wr32(device, NV03_PFIFO_CACHES, 1);
+}
+
+static const struct nvkm_fifo_func
+nv40_fifo = {
+	.chid_nr = nv10_fifo_chid_nr,
+	.chid_ctor = nv04_fifo_chid_ctor,
+	.runl_ctor = nv04_fifo_runl_ctor,
+	.init = nv40_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.runl = &nv04_runl,
+	.engn = &nv40_engn,
+	.engn_sw = &nv40_engn_sw,
+	.cgrp = {{                        }, &nv04_cgrp },
+	.chan = {{ 0, 0, NV40_CHANNEL_DMA }, &nv40_chan },
+};
+
+int
+nv40_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	      struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&nv40_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
new file mode 100644
index 000000000..954b5f3a7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
@@ -0,0 +1,398 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+
+#include <core/ramht.h>
+#include <subdev/timer.h>
+
+#include <nvif/class.h>
+
+void
+nv50_eobj_ramht_del(struct nvkm_chan *chan, int hash)
+{
+	nvkm_ramht_remove(chan->ramht, hash);
+}
+
+int
+nv50_eobj_ramht_add(struct nvkm_engn *engn, struct nvkm_object *eobj, struct nvkm_chan *chan)
+{
+	return nvkm_ramht_insert(chan->ramht, eobj, 0, 4, eobj->handle, engn->id << 20);
+}
+
+void
+nv50_chan_stop(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x002600 + (chan->id * 4), 0x80000000, 0x00000000);
+}
+
+void
+nv50_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_mask(device, 0x002600 + (chan->id * 4), 0x80000000, 0x80000000);
+}
+
+void
+nv50_chan_unbind(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x002600 + (chan->id * 4), 0x00000000);
+}
+
+static void
+nv50_chan_bind(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x002600 + (chan->id * 4), chan->ramfc->addr >> 12);
+}
+
+static int
+nv50_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+	const u32 limit2 = ilog2(length / 8);
+	int ret;
+
+	ret = nvkm_gpuobj_new(device, 0x0200, 0x1000, true, chan->inst, &chan->ramfc);
+	if (ret)
+		return ret;
+
+	ret = nvkm_gpuobj_new(device, 0x1200, 0, true, chan->inst, &chan->eng);
+	if (ret)
+		return ret;
+
+	ret = nvkm_gpuobj_new(device, 0x4000, 0, false, chan->inst, &chan->pgd);
+	if (ret)
+		return ret;
+
+	ret = nvkm_ramht_new(device, 0x8000, 16, chan->inst, &chan->ramht);
+	if (ret)
+		return ret;
+
+	nvkm_kmap(chan->ramfc);
+	nvkm_wo32(chan->ramfc, 0x3c, 0x403f6078);
+	nvkm_wo32(chan->ramfc, 0x44, 0x01003fff);
+	nvkm_wo32(chan->ramfc, 0x48, chan->push->node->offset >> 4);
+	nvkm_wo32(chan->ramfc, 0x50, lower_32_bits(offset));
+	nvkm_wo32(chan->ramfc, 0x54, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->ramfc, 0x60, 0x7fffffff);
+	nvkm_wo32(chan->ramfc, 0x78, 0x00000000);
+	nvkm_wo32(chan->ramfc, 0x7c, 0x30000000 | devm);
+	nvkm_wo32(chan->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				     (4 << 24) /* SEARCH_FULL */ |
+				     (chan->ramht->gpuobj->node->offset >> 4));
+	nvkm_done(chan->ramfc);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+nv50_chan_ramfc = {
+	.write = nv50_chan_ramfc_write,
+	.ctxdma = true,
+	.devm = 0xfff,
+};
+
+const struct nvkm_chan_func_userd
+nv50_chan_userd = {
+	.bar = 0,
+	.base = 0xc00000,
+	.size = 0x002000,
+};
+
+const struct nvkm_chan_func_inst
+nv50_chan_inst = {
+	.size = 0x10000,
+	.vmm = true,
+};
+
+static const struct nvkm_chan_func
+nv50_chan = {
+	.inst = &nv50_chan_inst,
+	.userd = &nv50_chan_userd,
+	.ramfc = &nv50_chan_ramfc,
+	.bind = nv50_chan_bind,
+	.unbind = nv50_chan_unbind,
+	.start = nv50_chan_start,
+	.stop = nv50_chan_stop,
+};
+
+static void
+nv50_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	struct nvkm_subdev *subdev = &chan->cgrp->runl->fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u64 start = 0, limit = 0;
+	u32 flags = 0, ptr0, save;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_GR    : ptr0 = 0x0000; break;
+	case NVKM_ENGINE_MPEG  : ptr0 = 0x0060; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (!cctx) {
+		/* HW bug workaround:
+		 *
+		 * PFIFO will hang forever if the connected engines don't report
+		 * that they've processed the context switch request.
+		 *
+		 * In order for the kickoff to work, we need to ensure all the
+		 * connected engines are in a state where they can answer.
+		 *
+		 * Newer chipsets don't seem to suffer from this issue, and well,
+		 * there's also a "ignore these engines" bitmask reg we can use
+		 * if we hit the issue there..
+		 */
+		save = nvkm_mask(device, 0x00b860, 0x00000001, 0x00000001);
+
+		/* Tell engines to save out contexts. */
+		nvkm_wr32(device, 0x0032fc, chan->inst->addr >> 12);
+		nvkm_msec(device, 2000,
+			if (nvkm_rd32(device, 0x0032fc) != 0xffffffff)
+				break;
+		);
+		nvkm_wr32(device, 0x00b860, save);
+	} else {
+		flags = 0x00190000;
+		start = cctx->vctx->inst->addr;
+		limit = start + cctx->vctx->inst->size - 1;
+	}
+
+	nvkm_kmap(chan->eng);
+	nvkm_wo32(chan->eng, ptr0 + 0x00, flags);
+	nvkm_wo32(chan->eng, ptr0 + 0x04, lower_32_bits(limit));
+	nvkm_wo32(chan->eng, ptr0 + 0x08, lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x0c, upper_32_bits(limit) << 24 |
+					  lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x10, 0x00000000);
+	nvkm_wo32(chan->eng, ptr0 + 0x14, 0x00000000);
+	nvkm_done(chan->eng);
+}
+
+static const struct nvkm_engn_func
+nv50_engn = {
+	.bind = nv50_ectx_bind,
+	.ramht_add = nv50_eobj_ramht_add,
+	.ramht_del = nv50_eobj_ramht_del,
+};
+
+const struct nvkm_engn_func
+nv50_engn_sw = {
+	.ramht_add = nv50_eobj_ramht_add,
+	.ramht_del = nv50_eobj_ramht_del,
+};
+
+static bool
+nv50_runl_pending(struct nvkm_runl *runl)
+{
+	return nvkm_rd32(runl->fifo->engine.subdev.device, 0x0032ec) & 0x00000100;
+}
+
+int
+nv50_runl_wait(struct nvkm_runl *runl)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+
+	nvkm_msec(fifo->engine.subdev.device, fifo->timeout.chan_msec,
+		if (!nvkm_runl_update_pending(runl))
+			return 0;
+		usleep_range(1, 2);
+	);
+
+	return -ETIMEDOUT;
+}
+
+static void
+nv50_runl_commit(struct nvkm_runl *runl, struct nvkm_memory *memory, u32 start, int count)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u64 addr = nvkm_memory_addr(memory) + start;
+
+	nvkm_wr32(device, 0x0032f4, addr >> 12);
+	nvkm_wr32(device, 0x0032ec, count);
+}
+
+static void
+nv50_runl_insert_chan(struct nvkm_chan *chan, struct nvkm_memory *memory, u64 offset)
+{
+	nvkm_wo32(memory, offset, chan->id);
+}
+
+static struct nvkm_memory *
+nv50_runl_alloc(struct nvkm_runl *runl, u32 *offset)
+{
+	const u32 segment = ALIGN((runl->cgrp_nr + runl->chan_nr) * runl->func->size, 0x1000);
+	const u32 maxsize = (runl->cgid ? runl->cgid->nr : 0) + runl->chid->nr;
+	int ret;
+
+	if (unlikely(!runl->mem)) {
+		ret = nvkm_memory_new(runl->fifo->engine.subdev.device, NVKM_MEM_TARGET_INST,
+				      maxsize * 2 * runl->func->size, 0, false, &runl->mem);
+		if (ret) {
+			RUNL_ERROR(runl, "alloc %d\n", ret);
+			return ERR_PTR(ret);
+		}
+	} else {
+		if (runl->offset + segment >= nvkm_memory_size(runl->mem)) {
+			ret = runl->func->wait(runl);
+			if (ret) {
+				RUNL_DEBUG(runl, "rewind timeout");
+				return ERR_PTR(ret);
+			}
+
+			runl->offset = 0;
+		}
+	}
+
+	*offset = runl->offset;
+	runl->offset += segment;
+	return runl->mem;
+}
+
+int
+nv50_runl_update(struct nvkm_runl *runl)
+{
+	struct nvkm_memory *memory;
+	struct nvkm_cgrp *cgrp;
+	struct nvkm_chan *chan;
+	u32 start, offset, count;
+
+	/*TODO: prio, interleaving. */
+
+	RUNL_TRACE(runl, "RAMRL: update cgrps:%d chans:%d", runl->cgrp_nr, runl->chan_nr);
+	memory = nv50_runl_alloc(runl, &start);
+	if (IS_ERR(memory))
+		return PTR_ERR(memory);
+
+	RUNL_TRACE(runl, "RAMRL: update start:%08x", start);
+	offset = start;
+
+	nvkm_kmap(memory);
+	nvkm_runl_foreach_cgrp(cgrp, runl) {
+		if (cgrp->hw) {
+			CGRP_TRACE(cgrp, "     RAMRL+%08x: chans:%d", offset, cgrp->chan_nr);
+			runl->func->insert_cgrp(cgrp, memory, offset);
+			offset += runl->func->size;
+		}
+
+		nvkm_cgrp_foreach_chan(chan, cgrp) {
+			CHAN_TRACE(chan, "RAMRL+%08x: [%s]", offset, chan->name);
+			runl->func->insert_chan(chan, memory, offset);
+			offset += runl->func->size;
+		}
+	}
+	nvkm_done(memory);
+
+	/*TODO: look into using features on newer HW to guarantee forward progress. */
+	list_rotate_left(&runl->cgrps);
+
+	count = (offset - start) / runl->func->size;
+	RUNL_TRACE(runl, "RAMRL: commit start:%08x count:%d", start, count);
+
+	runl->func->commit(runl, memory, start, count);
+	return 0;
+}
+
+const struct nvkm_runl_func
+nv50_runl = {
+	.size = 4,
+	.update = nv50_runl_update,
+	.insert_chan = nv50_runl_insert_chan,
+	.commit = nv50_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = nv50_runl_pending,
+};
+
+void
+nv50_fifo_init(struct nvkm_fifo *fifo)
+{
+	struct nvkm_runl *runl = nvkm_runl_first(fifo);
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	int i;
+
+	nvkm_mask(device, 0x000200, 0x00000100, 0x00000000);
+	nvkm_mask(device, 0x000200, 0x00000100, 0x00000100);
+	nvkm_wr32(device, 0x00250c, 0x6f3cfc34);
+	nvkm_wr32(device, 0x002044, 0x01003fff);
+
+	nvkm_wr32(device, 0x002100, 0xffffffff);
+	nvkm_wr32(device, 0x002140, 0xbfffffff);
+
+	for (i = 0; i < 128; i++)
+		nvkm_wr32(device, 0x002600 + (i * 4), 0x00000000);
+
+	atomic_set(&runl->changed, 1);
+	runl->func->update(runl);
+
+	nvkm_wr32(device, 0x003200, 0x00000001);
+	nvkm_wr32(device, 0x003250, 0x00000001);
+	nvkm_wr32(device, 0x002500, 0x00000001);
+}
+
+int
+nv50_fifo_chid_ctor(struct nvkm_fifo *fifo, int nr)
+{
+	/* CHID 0 is unusable (some kind of PIO channel?), 127 is "channel invalid". */
+	return nvkm_chid_new(&nvkm_chan_event, &fifo->engine.subdev, nr, 1, nr - 2, &fifo->chid);
+}
+
+int
+nv50_fifo_chid_nr(struct nvkm_fifo *fifo)
+{
+	return 128;
+}
+
+static const struct nvkm_fifo_func
+nv50_fifo = {
+	.chid_nr = nv50_fifo_chid_nr,
+	.chid_ctor = nv50_fifo_chid_ctor,
+	.runl_ctor = nv04_fifo_runl_ctor,
+	.init = nv50_fifo_init,
+	.intr = nv04_fifo_intr,
+	.pause = nv04_fifo_pause,
+	.start = nv04_fifo_start,
+	.runl = &nv50_runl,
+	.engn = &nv50_engn,
+	.engn_sw = &nv50_engn_sw,
+	.cgrp = {{                           }, &nv04_cgrp },
+	.chan = {{ 0, 0, NV50_CHANNEL_GPFIFO }, &nv50_chan },
+};
+
+int
+nv50_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	      struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&nv50_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
new file mode 100644
index 000000000..4d448be19
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
@@ -0,0 +1,209 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVKM_FIFO_PRIV_H__
+#define __NVKM_FIFO_PRIV_H__
+#define nvkm_fifo(p) container_of((p), struct nvkm_fifo, engine)
+#include <engine/fifo.h>
+#include <core/enum.h>
+struct nvkm_cctx;
+struct nvkm_cgrp;
+struct nvkm_engn;
+struct nvkm_memory;
+struct nvkm_runl;
+struct nvkm_runq;
+struct nvkm_vctx;
+
+struct nvkm_fifo_func {
+	int (*chid_nr)(struct nvkm_fifo *);
+	int (*chid_ctor)(struct nvkm_fifo *, int nr);
+	int (*runq_nr)(struct nvkm_fifo *);
+	int (*runl_ctor)(struct nvkm_fifo *);
+
+	void (*init)(struct nvkm_fifo *);
+	void (*init_pbdmas)(struct nvkm_fifo *, u32 mask);
+
+	irqreturn_t (*intr)(struct nvkm_inth *);
+	void (*intr_mmu_fault_unit)(struct nvkm_fifo *, int unit);
+	void (*intr_ctxsw_timeout)(struct nvkm_fifo *, u32 engm);
+
+	const struct nvkm_fifo_func_mmu_fault {
+		void (*recover)(struct nvkm_fifo *, struct nvkm_fault_data *);
+		const struct nvkm_enum *access;
+		const struct nvkm_enum *engine;
+		const struct nvkm_enum *reason;
+		const struct nvkm_enum *hubclient;
+		const struct nvkm_enum *gpcclient;
+	} *mmu_fault;
+
+	void (*pause)(struct nvkm_fifo *, unsigned long *);
+	void (*start)(struct nvkm_fifo *, unsigned long *);
+
+	int (*nonstall_ctor)(struct nvkm_fifo *);
+	const struct nvkm_event_func *nonstall;
+
+	const struct nvkm_runl_func *runl;
+	const struct nvkm_runq_func *runq;
+	const struct nvkm_engn_func *engn;
+	const struct nvkm_engn_func *engn_sw;
+	const struct nvkm_engn_func *engn_ce;
+
+	struct nvkm_fifo_func_cgrp {
+		struct nvkm_sclass user;
+		const struct nvkm_cgrp_func *func;
+		bool force;
+	} cgrp;
+
+	struct nvkm_fifo_func_chan {
+		struct nvkm_sclass user;
+		const struct nvkm_chan_func *func;
+	} chan;
+};
+
+int nvkm_fifo_new_(const struct nvkm_fifo_func *, struct nvkm_device *, enum nvkm_subdev_type, int,
+		   struct nvkm_fifo **);
+
+int nv04_fifo_chid_ctor(struct nvkm_fifo *, int);
+int nv04_fifo_runl_ctor(struct nvkm_fifo *);
+void nv04_fifo_init(struct nvkm_fifo *);
+irqreturn_t nv04_fifo_intr(struct nvkm_inth *);
+void nv04_fifo_pause(struct nvkm_fifo *, unsigned long *);
+void nv04_fifo_start(struct nvkm_fifo *, unsigned long *);
+extern const struct nvkm_runl_func nv04_runl;
+extern const struct nvkm_engn_func nv04_engn;
+extern const struct nvkm_cgrp_func nv04_cgrp;
+extern const struct nvkm_chan_func_inst nv04_chan_inst;
+extern const struct nvkm_chan_func_userd nv04_chan_userd;
+void nv04_chan_ramfc_clear(struct nvkm_chan *);
+void nv04_chan_start(struct nvkm_chan *);
+void nv04_chan_stop(struct nvkm_chan *);
+void nv04_eobj_ramht_del(struct nvkm_chan *, int);
+
+int nv10_fifo_chid_nr(struct nvkm_fifo *);
+
+int nv50_fifo_chid_nr(struct nvkm_fifo *);
+int nv50_fifo_chid_ctor(struct nvkm_fifo *, int);
+void nv50_fifo_init(struct nvkm_fifo *);
+extern const struct nvkm_runl_func nv50_runl;
+int nv50_runl_update(struct nvkm_runl *);
+int nv50_runl_wait(struct nvkm_runl *);
+extern const struct nvkm_engn_func nv50_engn_sw;
+extern const struct nvkm_chan_func_inst nv50_chan_inst;
+extern const struct nvkm_chan_func_userd nv50_chan_userd;
+void nv50_chan_unbind(struct nvkm_chan *);
+void nv50_chan_start(struct nvkm_chan *);
+void nv50_chan_stop(struct nvkm_chan *);
+void nv50_chan_preempt(struct nvkm_chan *);
+int nv50_eobj_ramht_add(struct nvkm_engn *, struct nvkm_object *, struct nvkm_chan *);
+void nv50_eobj_ramht_del(struct nvkm_chan *, int);
+
+extern const struct nvkm_event_func g84_fifo_nonstall;
+extern const struct nvkm_engn_func g84_engn;
+extern const struct nvkm_chan_func g84_chan;
+
+int gf100_fifo_chid_ctor(struct nvkm_fifo *, int);
+int gf100_fifo_runq_nr(struct nvkm_fifo *);
+bool gf100_fifo_intr_pbdma(struct nvkm_fifo *);
+void gf100_fifo_intr_mmu_fault(struct nvkm_fifo *);
+void gf100_fifo_intr_mmu_fault_unit(struct nvkm_fifo *, int);
+void gf100_fifo_intr_sched(struct nvkm_fifo *);
+void gf100_fifo_intr_ctxsw_timeout(struct nvkm_fifo *, u32);
+void gf100_fifo_mmu_fault_recover(struct nvkm_fifo *, struct nvkm_fault_data *);
+extern const struct nvkm_enum gf100_fifo_mmu_fault_access[];
+extern const struct nvkm_event_func gf100_fifo_nonstall;
+bool gf100_runl_preempt_pending(struct nvkm_runl *);
+void gf100_runq_init(struct nvkm_runq *);
+bool gf100_runq_intr(struct nvkm_runq *, struct nvkm_runl *);
+void gf100_engn_mmu_fault_trigger(struct nvkm_engn *);
+bool gf100_engn_mmu_fault_triggered(struct nvkm_engn *);
+extern const struct nvkm_engn_func gf100_engn_sw;
+extern const struct nvkm_chan_func_inst gf100_chan_inst;
+void gf100_chan_userd_clear(struct nvkm_chan *);
+void gf100_chan_preempt(struct nvkm_chan *);
+
+int gk104_fifo_chid_nr(struct nvkm_fifo *);
+int gk104_fifo_runl_ctor(struct nvkm_fifo *);
+void gk104_fifo_init(struct nvkm_fifo *);
+void gk104_fifo_init_pbdmas(struct nvkm_fifo *, u32);
+irqreturn_t gk104_fifo_intr(struct nvkm_inth *);
+void gk104_fifo_intr_runlist(struct nvkm_fifo *);
+void gk104_fifo_intr_chsw(struct nvkm_fifo *);
+void gk104_fifo_intr_bind(struct nvkm_fifo *);
+extern const struct nvkm_fifo_func_mmu_fault gk104_fifo_mmu_fault;
+extern const struct nvkm_enum gk104_fifo_mmu_fault_reason[];
+extern const struct nvkm_enum gk104_fifo_mmu_fault_hubclient[];
+extern const struct nvkm_enum gk104_fifo_mmu_fault_gpcclient[];
+void gk104_runl_insert_chan(struct nvkm_chan *, struct nvkm_memory *, u64);
+void gk104_runl_commit(struct nvkm_runl *, struct nvkm_memory *, u32, int);
+bool gk104_runl_pending(struct nvkm_runl *);
+void gk104_runl_block(struct nvkm_runl *, u32);
+void gk104_runl_allow(struct nvkm_runl *, u32);
+void gk104_runl_fault_clear(struct nvkm_runl *);
+extern const struct nvkm_runq_func gk104_runq;
+void gk104_runq_init(struct nvkm_runq *);
+bool gk104_runq_intr(struct nvkm_runq *, struct nvkm_runl *);
+extern const struct nvkm_bitfield gk104_runq_intr_0_names[];
+bool gk104_runq_idle(struct nvkm_runq *);
+extern const struct nvkm_engn_func gk104_engn;
+bool gk104_engn_chsw(struct nvkm_engn *);
+int gk104_engn_cxid(struct nvkm_engn *, bool *cgid);
+int gk104_ectx_ctor(struct nvkm_engn *, struct nvkm_vctx *);
+extern const struct nvkm_engn_func gk104_engn_ce;
+extern const struct nvkm_chan_func_userd gk104_chan_userd;
+extern const struct nvkm_chan_func_ramfc gk104_chan_ramfc;
+void gk104_chan_bind(struct nvkm_chan *);
+void gk104_chan_bind_inst(struct nvkm_chan *);
+void gk104_chan_unbind(struct nvkm_chan *);
+void gk104_chan_start(struct nvkm_chan *);
+void gk104_chan_stop(struct nvkm_chan *);
+
+int gk110_fifo_chid_ctor(struct nvkm_fifo *, int);
+extern const struct nvkm_runl_func gk110_runl;
+extern const struct nvkm_cgrp_func gk110_cgrp;
+void gk110_runl_insert_cgrp(struct nvkm_cgrp *, struct nvkm_memory *, u64);
+extern const struct nvkm_chan_func gk110_chan;
+void gk110_chan_preempt(struct nvkm_chan *);
+
+extern const struct nvkm_runq_func gk208_runq;
+void gk208_runq_init(struct nvkm_runq *);
+
+void gm107_fifo_intr_mmu_fault_unit(struct nvkm_fifo *, int);
+extern const struct nvkm_fifo_func_mmu_fault gm107_fifo_mmu_fault;
+extern const struct nvkm_runl_func gm107_runl;
+extern const struct nvkm_chan_func gm107_chan;
+
+int gm200_fifo_chid_nr(struct nvkm_fifo *);
+int gm200_fifo_runq_nr(struct nvkm_fifo *);
+
+extern const struct nvkm_enum gv100_fifo_mmu_fault_access[];
+extern const struct nvkm_enum gv100_fifo_mmu_fault_reason[];
+extern const struct nvkm_enum gv100_fifo_mmu_fault_hubclient[];
+extern const struct nvkm_enum gv100_fifo_mmu_fault_gpcclient[];
+void gv100_runl_insert_cgrp(struct nvkm_cgrp *, struct nvkm_memory *, u64);
+void gv100_runl_insert_chan(struct nvkm_chan *, struct nvkm_memory *, u64);
+void gv100_runl_preempt(struct nvkm_runl *);
+extern const struct nvkm_runq_func gv100_runq;
+extern const struct nvkm_engn_func gv100_engn;
+void gv100_ectx_bind(struct nvkm_engn *, struct nvkm_cctx *, struct nvkm_chan *);
+extern const struct nvkm_engn_func gv100_engn_ce;
+int gv100_ectx_ce_ctor(struct nvkm_engn *, struct nvkm_vctx *);
+void gv100_ectx_ce_bind(struct nvkm_engn *, struct nvkm_cctx *, struct nvkm_chan *);
+extern const struct nvkm_chan_func_userd gv100_chan_userd;
+extern const struct nvkm_chan_func_ramfc gv100_chan_ramfc;
+
+void tu102_fifo_intr_ctxsw_timeout_info(struct nvkm_engn *, u32 info);
+extern const struct nvkm_fifo_func_mmu_fault tu102_fifo_mmu_fault;
+
+int ga100_fifo_runl_ctor(struct nvkm_fifo *);
+int ga100_fifo_nonstall_ctor(struct nvkm_fifo *);
+extern const struct nvkm_event_func ga100_fifo_nonstall;
+extern const struct nvkm_runl_func ga100_runl;
+extern const struct nvkm_runq_func ga100_runq;
+extern const struct nvkm_engn_func ga100_engn;
+extern const struct nvkm_engn_func ga100_engn_ce;
+extern const struct nvkm_cgrp_func ga100_cgrp;
+extern const struct nvkm_chan_func ga100_chan;
+
+int nvkm_uchan_new(struct nvkm_fifo *, struct nvkm_cgrp *, const struct nvkm_oclass *,
+		   void *argv, u32 argc, struct nvkm_object **);
+int nvkm_ucgrp_new(struct nvkm_fifo *, const struct nvkm_oclass *, void *argv, u32 argc,
+		   struct nvkm_object **);
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/regsnv04.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/regsnv04.h
new file mode 100644
index 000000000..4445a12b9
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/regsnv04.h
@@ -0,0 +1,133 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NV04_FIFO_REGS_H__
+#define __NV04_FIFO_REGS_H__
+
+#define NV04_PFIFO_DELAY_0                                 0x00002040
+#define NV04_PFIFO_DMA_TIMESLICE                           0x00002044
+#define NV04_PFIFO_NEXT_CHANNEL                            0x00002050
+#define NV03_PFIFO_INTR_0                                  0x00002100
+#define NV03_PFIFO_INTR_EN_0                               0x00002140
+#    define NV_PFIFO_INTR_CACHE_ERROR                          (1<<0)
+#    define NV_PFIFO_INTR_RUNOUT                               (1<<4)
+#    define NV_PFIFO_INTR_RUNOUT_OVERFLOW                      (1<<8)
+#    define NV_PFIFO_INTR_DMA_PUSHER                          (1<<12)
+#    define NV_PFIFO_INTR_DMA_PT                              (1<<16)
+#    define NV_PFIFO_INTR_SEMAPHORE                           (1<<20)
+#    define NV_PFIFO_INTR_ACQUIRE_TIMEOUT                     (1<<24)
+#define NV03_PFIFO_RAMHT                                   0x00002210
+#define NV03_PFIFO_RAMFC                                   0x00002214
+#define NV03_PFIFO_RAMRO                                   0x00002218
+#define NV40_PFIFO_RAMFC                                   0x00002220
+#define NV03_PFIFO_CACHES                                  0x00002500
+#define NV04_PFIFO_MODE                                    0x00002504
+#define NV04_PFIFO_DMA                                     0x00002508
+#define NV04_PFIFO_SIZE                                    0x0000250c
+#define NV50_PFIFO_CTX_TABLE(c)                        (0x2600+(c)*4)
+#define NV50_PFIFO_CTX_TABLE__SIZE                                128
+#define NV50_PFIFO_CTX_TABLE_CHANNEL_ENABLED                  (1<<31)
+#define NV50_PFIFO_CTX_TABLE_UNK30_BAD                        (1<<30)
+#define NV50_PFIFO_CTX_TABLE_INSTANCE_MASK_G80             0x0FFFFFFF
+#define NV50_PFIFO_CTX_TABLE_INSTANCE_MASK_G84             0x00FFFFFF
+#define NV03_PFIFO_CACHE0_PUSH0                            0x00003000
+#define NV03_PFIFO_CACHE0_PULL0                            0x00003040
+#define NV04_PFIFO_CACHE0_PULL0                            0x00003050
+#define NV04_PFIFO_CACHE0_PULL1                            0x00003054
+#define NV03_PFIFO_CACHE1_PUSH0                            0x00003200
+#define NV03_PFIFO_CACHE1_PUSH1                            0x00003204
+#define NV03_PFIFO_CACHE1_PUSH1_DMA                            (1<<8)
+#define NV40_PFIFO_CACHE1_PUSH1_DMA                           (1<<16)
+#define NV03_PFIFO_CACHE1_PUSH1_CHID_MASK                  0x0000000f
+#define NV10_PFIFO_CACHE1_PUSH1_CHID_MASK                  0x0000001f
+#define NV50_PFIFO_CACHE1_PUSH1_CHID_MASK                  0x0000007f
+#define NV03_PFIFO_CACHE1_PUT                              0x00003210
+#define NV04_PFIFO_CACHE1_DMA_PUSH                         0x00003220
+#define NV04_PFIFO_CACHE1_DMA_FETCH                        0x00003224
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_8_BYTES         0x00000000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_16_BYTES        0x00000008
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_24_BYTES        0x00000010
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_32_BYTES        0x00000018
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_40_BYTES        0x00000020
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_48_BYTES        0x00000028
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_56_BYTES        0x00000030
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_64_BYTES        0x00000038
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_72_BYTES        0x00000040
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_80_BYTES        0x00000048
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_88_BYTES        0x00000050
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_96_BYTES        0x00000058
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_104_BYTES       0x00000060
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_112_BYTES       0x00000068
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_120_BYTES       0x00000070
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES       0x00000078
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_136_BYTES       0x00000080
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_144_BYTES       0x00000088
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_152_BYTES       0x00000090
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_160_BYTES       0x00000098
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_168_BYTES       0x000000A0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_176_BYTES       0x000000A8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_184_BYTES       0x000000B0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_192_BYTES       0x000000B8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_200_BYTES       0x000000C0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_208_BYTES       0x000000C8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_216_BYTES       0x000000D0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_224_BYTES       0x000000D8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_232_BYTES       0x000000E0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_240_BYTES       0x000000E8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_248_BYTES       0x000000F0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_256_BYTES       0x000000F8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE                 0x0000E000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_32_BYTES        0x00000000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_64_BYTES        0x00002000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_96_BYTES        0x00004000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES       0x00006000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_160_BYTES       0x00008000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_192_BYTES       0x0000A000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_224_BYTES       0x0000C000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_256_BYTES       0x0000E000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS             0x001F0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_0           0x00000000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_1           0x00010000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_2           0x00020000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_3           0x00030000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_4           0x00040000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_5           0x00050000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_6           0x00060000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_7           0x00070000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8           0x00080000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_9           0x00090000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_10          0x000A0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_11          0x000B0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_12          0x000C0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_13          0x000D0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_14          0x000E0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_15          0x000F0000
+#    define NV_PFIFO_CACHE1_ENDIAN                         0x80000000
+#    define NV_PFIFO_CACHE1_LITTLE_ENDIAN                  0x7FFFFFFF
+#    define NV_PFIFO_CACHE1_BIG_ENDIAN                     0x80000000
+#define NV04_PFIFO_CACHE1_DMA_STATE                        0x00003228
+#define NV04_PFIFO_CACHE1_DMA_INSTANCE                     0x0000322c
+#define NV04_PFIFO_CACHE1_DMA_CTL                          0x00003230
+#define NV04_PFIFO_CACHE1_DMA_PUT                          0x00003240
+#define NV04_PFIFO_CACHE1_DMA_GET                          0x00003244
+#define NV10_PFIFO_CACHE1_REF_CNT                          0x00003248
+#define NV10_PFIFO_CACHE1_DMA_SUBROUTINE                   0x0000324C
+#define NV03_PFIFO_CACHE1_PULL0                            0x00003240
+#define NV04_PFIFO_CACHE1_PULL0                            0x00003250
+#    define NV04_PFIFO_CACHE1_PULL0_HASH_FAILED            0x00000010
+#    define NV04_PFIFO_CACHE1_PULL0_HASH_BUSY              0x00001000
+#define NV03_PFIFO_CACHE1_PULL1                            0x00003250
+#define NV04_PFIFO_CACHE1_PULL1                            0x00003254
+#define NV04_PFIFO_CACHE1_HASH                             0x00003258
+#define NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT                  0x00003260
+#define NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP                0x00003264
+#define NV10_PFIFO_CACHE1_ACQUIRE_VALUE                    0x00003268
+#define NV10_PFIFO_CACHE1_SEMAPHORE                        0x0000326C
+#define NV03_PFIFO_CACHE1_GET                              0x00003270
+#define NV04_PFIFO_CACHE1_ENGINE                           0x00003280
+#define NV04_PFIFO_CACHE1_DMA_DCOUNT                       0x000032A0
+#define NV40_PFIFO_GRCTX_INSTANCE                          0x000032E0
+#define NV40_PFIFO_UNK32E4                                 0x000032E4
+#define NV04_PFIFO_CACHE1_METHOD(i)                (0x00003800+(i*8))
+#define NV04_PFIFO_CACHE1_DATA(i)                  (0x00003804+(i*8))
+#define NV40_PFIFO_CACHE1_METHOD(i)                (0x00090000+(i*8))
+#define NV40_PFIFO_CACHE1_DATA(i)                  (0x00090004+(i*8))
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c
new file mode 100644
index 000000000..b5836cbc2
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c
@@ -0,0 +1,430 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "runl.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "priv.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/timer.h>
+#include <subdev/top.h>
+
+struct nvkm_cgrp *
+nvkm_engn_cgrp_get(struct nvkm_engn *engn, unsigned long *pirqflags)
+{
+	struct nvkm_cgrp *cgrp = NULL;
+	struct nvkm_chan *chan;
+	bool cgid;
+	int id;
+
+	id = engn->func->cxid(engn, &cgid);
+	if (id < 0)
+		return NULL;
+
+	if (!cgid) {
+		chan = nvkm_runl_chan_get_chid(engn->runl, id, pirqflags);
+		if (chan)
+			cgrp = chan->cgrp;
+	} else {
+		cgrp = nvkm_runl_cgrp_get_cgid(engn->runl, id, pirqflags);
+	}
+
+	WARN_ON(!cgrp);
+	return cgrp;
+}
+
+static void
+nvkm_runl_rc(struct nvkm_runl *runl)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_cgrp *cgrp, *gtmp;
+	struct nvkm_chan *chan, *ctmp;
+	struct nvkm_engn *engn;
+	unsigned long flags;
+	int rc, state, i;
+	bool reset;
+
+	/* Runlist is blocked before scheduling recovery - fetch count. */
+	BUG_ON(!mutex_is_locked(&runl->mutex));
+	rc = atomic_xchg(&runl->rc_pending, 0);
+	if (!rc)
+		return;
+
+	/* Look for channel groups flagged for RC. */
+	nvkm_runl_foreach_cgrp_safe(cgrp, gtmp, runl) {
+		state = atomic_cmpxchg(&cgrp->rc, NVKM_CGRP_RC_PENDING, NVKM_CGRP_RC_RUNNING);
+		if (state == NVKM_CGRP_RC_PENDING) {
+			/* Disable all channels in them, and remove from runlist. */
+			nvkm_cgrp_foreach_chan_safe(chan, ctmp, cgrp) {
+				nvkm_chan_error(chan, false);
+				nvkm_chan_remove_locked(chan);
+			}
+		}
+	}
+
+	/* On GPUs with runlist preempt, wait for PBDMA(s) servicing runlist to go idle. */
+	if (runl->func->preempt) {
+		for (i = 0; i < runl->runq_nr; i++) {
+			struct nvkm_runq *runq = runl->runq[i];
+
+			if (runq) {
+				nvkm_msec(fifo->engine.subdev.device, 2000,
+					if (runq->func->idle(runq))
+						break;
+				);
+			}
+		}
+	}
+
+	/* Look for engines that are still on flagged channel groups - reset them. */
+	nvkm_runl_foreach_engn_cond(engn, runl, engn->func->cxid) {
+		cgrp = nvkm_engn_cgrp_get(engn, &flags);
+		if (!cgrp) {
+			ENGN_DEBUG(engn, "cxid not valid");
+			continue;
+		}
+
+		reset = atomic_read(&cgrp->rc) == NVKM_CGRP_RC_RUNNING;
+		nvkm_cgrp_put(&cgrp, flags);
+		if (!reset) {
+			ENGN_DEBUG(engn, "cxid not in recovery");
+			continue;
+		}
+
+		ENGN_DEBUG(engn, "resetting...");
+		/*TODO: can we do something less of a potential catastrophe on failure? */
+		WARN_ON(nvkm_engine_reset(engn->engine));
+	}
+
+	/* Submit runlist update, and clear any remaining exception state. */
+	runl->func->update(runl);
+	if (runl->func->fault_clear)
+		runl->func->fault_clear(runl);
+
+	/* Unblock runlist processing. */
+	while (rc--)
+		nvkm_runl_allow(runl);
+	runl->func->wait(runl);
+}
+
+static void
+nvkm_runl_rc_runl(struct nvkm_runl *runl)
+{
+	RUNL_ERROR(runl, "rc scheduled");
+
+	nvkm_runl_block(runl);
+	if (runl->func->preempt)
+		runl->func->preempt(runl);
+
+	atomic_inc(&runl->rc_pending);
+	schedule_work(&runl->work);
+}
+
+void
+nvkm_runl_rc_cgrp(struct nvkm_cgrp *cgrp)
+{
+	if (atomic_cmpxchg(&cgrp->rc, NVKM_CGRP_RC_NONE, NVKM_CGRP_RC_PENDING) != NVKM_CGRP_RC_NONE)
+		return;
+
+	CGRP_ERROR(cgrp, "rc scheduled");
+	nvkm_runl_rc_runl(cgrp->runl);
+}
+
+void
+nvkm_runl_rc_engn(struct nvkm_runl *runl, struct nvkm_engn *engn)
+{
+	struct nvkm_cgrp *cgrp;
+	unsigned long flags;
+
+	/* Lookup channel group currently on engine. */
+	cgrp = nvkm_engn_cgrp_get(engn, &flags);
+	if (!cgrp) {
+		ENGN_DEBUG(engn, "rc skipped, not on channel");
+		return;
+	}
+
+	nvkm_runl_rc_cgrp(cgrp);
+	nvkm_cgrp_put(&cgrp, flags);
+}
+
+static void
+nvkm_runl_work(struct work_struct *work)
+{
+	struct nvkm_runl *runl = container_of(work, typeof(*runl), work);
+
+	mutex_lock(&runl->mutex);
+	nvkm_runl_rc(runl);
+	mutex_unlock(&runl->mutex);
+
+}
+
+struct nvkm_chan *
+nvkm_runl_chan_get_inst(struct nvkm_runl *runl, u64 inst, unsigned long *pirqflags)
+{
+	struct nvkm_chid *chid = runl->chid;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+	int id;
+
+	spin_lock_irqsave(&chid->lock, flags);
+	for_each_set_bit(id, chid->used, chid->nr) {
+		chan = chid->data[id];
+		if (likely(chan)) {
+			if (chan->inst->addr == inst) {
+				spin_lock(&chan->cgrp->lock);
+				*pirqflags = flags;
+				spin_unlock(&chid->lock);
+				return chan;
+			}
+		}
+	}
+	spin_unlock_irqrestore(&chid->lock, flags);
+	return NULL;
+}
+
+struct nvkm_chan *
+nvkm_runl_chan_get_chid(struct nvkm_runl *runl, int id, unsigned long *pirqflags)
+{
+	struct nvkm_chid *chid = runl->chid;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+
+	spin_lock_irqsave(&chid->lock, flags);
+	if (!WARN_ON(id >= chid->nr)) {
+		chan = chid->data[id];
+		if (likely(chan)) {
+			spin_lock(&chan->cgrp->lock);
+			*pirqflags = flags;
+			spin_unlock(&chid->lock);
+			return chan;
+		}
+	}
+	spin_unlock_irqrestore(&chid->lock, flags);
+	return NULL;
+}
+
+struct nvkm_cgrp *
+nvkm_runl_cgrp_get_cgid(struct nvkm_runl *runl, int id, unsigned long *pirqflags)
+{
+	struct nvkm_chid *cgid = runl->cgid;
+	struct nvkm_cgrp *cgrp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&cgid->lock, flags);
+	if (!WARN_ON(id >= cgid->nr)) {
+		cgrp = cgid->data[id];
+		if (likely(cgrp)) {
+			spin_lock(&cgrp->lock);
+			*pirqflags = flags;
+			spin_unlock(&cgid->lock);
+			return cgrp;
+		}
+	}
+	spin_unlock_irqrestore(&cgid->lock, flags);
+	return NULL;
+}
+
+int
+nvkm_runl_preempt_wait(struct nvkm_runl *runl)
+{
+	return nvkm_msec(runl->fifo->engine.subdev.device, runl->fifo->timeout.chan_msec,
+		if (!runl->func->preempt_pending(runl))
+			break;
+
+		nvkm_runl_rc(runl);
+		usleep_range(1, 2);
+	) < 0 ? -ETIMEDOUT : 0;
+}
+
+bool
+nvkm_runl_update_pending(struct nvkm_runl *runl)
+{
+	if (!runl->func->pending(runl))
+		return false;
+
+	nvkm_runl_rc(runl);
+	return true;
+}
+
+void
+nvkm_runl_update_locked(struct nvkm_runl *runl, bool wait)
+{
+	if (atomic_xchg(&runl->changed, 0) && runl->func->update) {
+		runl->func->update(runl);
+		if (wait)
+			runl->func->wait(runl);
+	}
+}
+
+void
+nvkm_runl_allow(struct nvkm_runl *runl)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	if (!--runl->blocked) {
+		RUNL_TRACE(runl, "running");
+		runl->func->allow(runl, ~0);
+	}
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+void
+nvkm_runl_block(struct nvkm_runl *runl)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	unsigned long flags;
+
+	spin_lock_irqsave(&fifo->lock, flags);
+	if (!runl->blocked++) {
+		RUNL_TRACE(runl, "stopped");
+		runl->func->block(runl, ~0);
+	}
+	spin_unlock_irqrestore(&fifo->lock, flags);
+}
+
+void
+nvkm_runl_fini(struct nvkm_runl *runl)
+{
+	if (runl->func->fini)
+		runl->func->fini(runl);
+
+	flush_work(&runl->work);
+}
+
+void
+nvkm_runl_del(struct nvkm_runl *runl)
+{
+	struct nvkm_engn *engn, *engt;
+
+	nvkm_memory_unref(&runl->mem);
+
+	list_for_each_entry_safe(engn, engt, &runl->engns, head) {
+		list_del(&engn->head);
+		kfree(engn);
+	}
+
+	nvkm_chid_unref(&runl->chid);
+	nvkm_chid_unref(&runl->cgid);
+
+	list_del(&runl->head);
+	mutex_destroy(&runl->mutex);
+	kfree(runl);
+}
+
+struct nvkm_engn *
+nvkm_runl_add(struct nvkm_runl *runl, int engi, const struct nvkm_engn_func *func,
+	      enum nvkm_subdev_type type, int inst)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_engine *engine;
+	struct nvkm_engn *engn;
+
+	engine = nvkm_device_engine(device, type, inst);
+	if (!engine) {
+		RUNL_DEBUG(runl, "engn %d.%d[%s] not found", engi, inst, nvkm_subdev_type[type]);
+		return NULL;
+	}
+
+	if (!(engn = kzalloc(sizeof(*engn), GFP_KERNEL)))
+		return NULL;
+
+	engn->func = func;
+	engn->runl = runl;
+	engn->id = engi;
+	engn->engine = engine;
+	engn->fault = -1;
+	list_add_tail(&engn->head, &runl->engns);
+
+	/* Lookup MMU engine ID for fault handling. */
+	if (device->top)
+		engn->fault = nvkm_top_fault_id(device, engine->subdev.type, engine->subdev.inst);
+
+	if (engn->fault < 0 && fifo->func->mmu_fault) {
+		const struct nvkm_enum *map = fifo->func->mmu_fault->engine;
+
+		while (map->name) {
+			if (map->data2 == engine->subdev.type && map->inst == engine->subdev.inst) {
+				engn->fault = map->value;
+				break;
+			}
+			map++;
+		}
+	}
+
+	return engn;
+}
+
+struct nvkm_runl *
+nvkm_runl_get(struct nvkm_fifo *fifo, int runi, u32 addr)
+{
+	struct nvkm_runl *runl;
+
+	nvkm_runl_foreach(runl, fifo) {
+		if ((runi >= 0 && runl->id == runi) || (runi < 0 && runl->addr == addr))
+			return runl;
+	}
+
+	return NULL;
+}
+
+struct nvkm_runl *
+nvkm_runl_new(struct nvkm_fifo *fifo, int runi, u32 addr, int id_nr)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_runl *runl;
+	int ret;
+
+	if (!(runl = kzalloc(sizeof(*runl), GFP_KERNEL)))
+		return NULL;
+
+	runl->func = fifo->func->runl;
+	runl->fifo = fifo;
+	runl->id = runi;
+	runl->addr = addr;
+	INIT_LIST_HEAD(&runl->engns);
+	INIT_LIST_HEAD(&runl->cgrps);
+	atomic_set(&runl->changed, 0);
+	mutex_init(&runl->mutex);
+	INIT_WORK(&runl->work, nvkm_runl_work);
+	atomic_set(&runl->rc_triggered, 0);
+	atomic_set(&runl->rc_pending, 0);
+	list_add_tail(&runl->head, &fifo->runls);
+
+	if (!fifo->chid) {
+		if ((ret = nvkm_chid_new(&nvkm_chan_event, subdev, id_nr, 0, id_nr, &runl->cgid)) ||
+		    (ret = nvkm_chid_new(&nvkm_chan_event, subdev, id_nr, 0, id_nr, &runl->chid))) {
+			RUNL_ERROR(runl, "cgid/chid: %d", ret);
+			nvkm_runl_del(runl);
+			return NULL;
+		}
+	} else {
+		runl->cgid = nvkm_chid_ref(fifo->cgid);
+		runl->chid = nvkm_chid_ref(fifo->chid);
+	}
+
+	return runl;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
new file mode 100644
index 000000000..c93d21bb7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
@@ -0,0 +1,125 @@
+#ifndef __NVKM_RUNL_H__
+#define __NVKM_RUNL_H__
+#include <core/intr.h>
+struct nvkm_cctx;
+struct nvkm_cgrp;
+struct nvkm_chan;
+struct nvkm_memory;
+struct nvkm_object;
+struct nvkm_vctx;
+enum nvkm_subdev_type;
+
+struct nvkm_engn {
+	const struct nvkm_engn_func {
+		bool (*chsw)(struct nvkm_engn *);
+		int (*cxid)(struct nvkm_engn *, bool *cgid);
+		void (*mmu_fault_trigger)(struct nvkm_engn *);
+		bool (*mmu_fault_triggered)(struct nvkm_engn *);
+		int (*ctor)(struct nvkm_engn *, struct nvkm_vctx *);
+		void (*bind)(struct nvkm_engn *, struct nvkm_cctx *, struct nvkm_chan *);
+		int (*ramht_add)(struct nvkm_engn *, struct nvkm_object *, struct nvkm_chan *);
+		void (*ramht_del)(struct nvkm_chan *, int hash);
+	} *func;
+	struct nvkm_runl *runl;
+	int id;
+
+	struct nvkm_engine *engine;
+
+	int fault;
+
+	struct list_head head;
+};
+
+#define ENGN_PRINT(e,l,p,f,a...)                                                           \
+	RUNL_PRINT((e)->runl, l, p, "%02d[%8s]:"f, (e)->id, (e)->engine->subdev.name, ##a)
+#define ENGN_DEBUG(e,f,a...) ENGN_PRINT((e), DEBUG,   info, " "f"\n", ##a)
+
+struct nvkm_runl {
+	const struct nvkm_runl_func {
+		void (*init)(struct nvkm_runl *);
+		void (*fini)(struct nvkm_runl *);
+		int runqs;
+		u8 size;
+		int (*update)(struct nvkm_runl *);
+		void (*insert_cgrp)(struct nvkm_cgrp *, struct nvkm_memory *, u64 offset);
+		void (*insert_chan)(struct nvkm_chan *, struct nvkm_memory *, u64 offset);
+		void (*commit)(struct nvkm_runl *, struct nvkm_memory *, u32 start, int count);
+		int (*wait)(struct nvkm_runl *);
+		bool (*pending)(struct nvkm_runl *);
+		void (*block)(struct nvkm_runl *, u32 engm);
+		void (*allow)(struct nvkm_runl *, u32 engm);
+		void (*fault_clear)(struct nvkm_runl *);
+		void (*preempt)(struct nvkm_runl *);
+		bool (*preempt_pending)(struct nvkm_runl *);
+	} *func;
+	struct nvkm_fifo *fifo;
+	int id;
+	u32 addr;
+	u32 chan;
+	u16 doorbell;
+
+	struct nvkm_chid *cgid;
+#define NVKM_CHAN_EVENT_ERRORED BIT(0)
+	struct nvkm_chid *chid;
+
+	struct list_head engns;
+
+	struct nvkm_runq *runq[2];
+	int runq_nr;
+
+	struct nvkm_inth inth;
+
+	struct list_head cgrps;
+	int cgrp_nr;
+	int chan_nr;
+	atomic_t changed;
+	struct nvkm_memory *mem;
+	u32 offset;
+	struct mutex mutex;
+
+	int blocked;
+
+	struct work_struct work;
+	atomic_t rc_triggered;
+	atomic_t rc_pending;
+
+	struct list_head head;
+};
+
+struct nvkm_runl *nvkm_runl_new(struct nvkm_fifo *, int runi, u32 addr, int id_nr);
+struct nvkm_runl *nvkm_runl_get(struct nvkm_fifo *, int runi, u32 addr);
+struct nvkm_engn *nvkm_runl_add(struct nvkm_runl *, int engi, const struct nvkm_engn_func *,
+				enum nvkm_subdev_type, int inst);
+void nvkm_runl_del(struct nvkm_runl *);
+void nvkm_runl_fini(struct nvkm_runl *);
+void nvkm_runl_block(struct nvkm_runl *);
+void nvkm_runl_allow(struct nvkm_runl *);
+void nvkm_runl_update_locked(struct nvkm_runl *, bool wait);
+bool nvkm_runl_update_pending(struct nvkm_runl *);
+int nvkm_runl_preempt_wait(struct nvkm_runl *);
+
+void nvkm_runl_rc_engn(struct nvkm_runl *, struct nvkm_engn *);
+void nvkm_runl_rc_cgrp(struct nvkm_cgrp *);
+
+struct nvkm_cgrp *nvkm_runl_cgrp_get_cgid(struct nvkm_runl *, int cgid, unsigned long *irqflags);
+struct nvkm_chan *nvkm_runl_chan_get_chid(struct nvkm_runl *, int chid, unsigned long *irqflags);
+struct nvkm_chan *nvkm_runl_chan_get_inst(struct nvkm_runl *, u64 inst, unsigned long *irqflags);
+
+#define nvkm_runl_find_engn(engn,runl,cond) nvkm_list_find(engn, &(runl)->engns, head, (cond))
+
+#define nvkm_runl_first(fifo) list_first_entry(&(fifo)->runls, struct nvkm_runl, head)
+#define nvkm_runl_foreach(runl,fifo) list_for_each_entry((runl), &(fifo)->runls, head)
+#define nvkm_runl_foreach_cond(runl,fifo,cond) nvkm_list_foreach(runl, &(fifo)->runls, head, (cond))
+#define nvkm_runl_foreach_engn(engn,runl) list_for_each_entry((engn), &(runl)->engns, head)
+#define nvkm_runl_foreach_engn_cond(engn,runl,cond) \
+	nvkm_list_foreach(engn, &(runl)->engns, head, (cond))
+#define nvkm_runl_foreach_cgrp(cgrp,runl) list_for_each_entry((cgrp), &(runl)->cgrps, head)
+#define nvkm_runl_foreach_cgrp_safe(cgrp,gtmp,runl) \
+	list_for_each_entry_safe((cgrp), (gtmp), &(runl)->cgrps, head)
+
+#define RUNL_PRINT(r,l,p,f,a...)                                                          \
+	nvkm_printk__(&(r)->fifo->engine.subdev, NV_DBG_##l, p, "%06x:"f, (r)->addr, ##a)
+#define RUNL_ERROR(r,f,a...) RUNL_PRINT((r), ERROR,    err, " "f"\n", ##a)
+#define RUNL_DEBUG(r,f,a...) RUNL_PRINT((r), DEBUG,   info, " "f"\n", ##a)
+#define RUNL_TRACE(r,f,a...) RUNL_PRINT((r), TRACE,   info, " "f"\n", ##a)
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.c
new file mode 100644
index 000000000..33bcf5fb3
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.c
@@ -0,0 +1,45 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "runq.h"
+#include "priv.h"
+
+void
+nvkm_runq_del(struct nvkm_runq *runq)
+{
+	list_del(&runq->head);
+	kfree(runq);
+}
+
+struct nvkm_runq *
+nvkm_runq_new(struct nvkm_fifo *fifo, int pbid)
+{
+	struct nvkm_runq *runq;
+
+	if (!(runq = kzalloc(sizeof(*runq), GFP_KERNEL)))
+		return NULL;
+
+	runq->func = fifo->func->runq;
+	runq->fifo = fifo;
+	runq->id = pbid;
+	list_add_tail(&runq->head, &fifo->runqs);
+	return runq;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.h
new file mode 100644
index 000000000..2cb4836e8
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.h
@@ -0,0 +1,31 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVKM_RUNQ_H__
+#define __NVKM_RUNQ_H__
+#include <core/os.h>
+struct nvkm_runl;
+
+struct nvkm_runq {
+	const struct nvkm_runq_func {
+		void (*init)(struct nvkm_runq *);
+		bool (*intr)(struct nvkm_runq *, struct nvkm_runl *);
+		const struct nvkm_bitfield *intr_0_names;
+		bool (*intr_1_ctxnotvalid)(struct nvkm_runq *, int chid);
+		bool (*idle)(struct nvkm_runq *);
+	} *func;
+	struct nvkm_fifo *fifo;
+	int id;
+
+	struct list_head head;
+};
+
+struct nvkm_runq *nvkm_runq_new(struct nvkm_fifo *, int pbid);
+void nvkm_runq_del(struct nvkm_runq *);
+
+#define nvkm_runq_foreach(runq,fifo) list_for_each_entry((runq), &(fifo)->runqs, head)
+#define nvkm_runq_foreach_cond(runq,fifo,cond) nvkm_list_foreach(runq, &(fifo)->runqs, head, (cond))
+
+#define RUNQ_PRINT(r,l,p,f,a...)							   \
+	nvkm_printk__(&(r)->fifo->engine.subdev, NV_DBG_##l, p, "PBDMA%d:"f, (r)->id, ##a)
+#define RUNQ_ERROR(r,f,a...) RUNQ_PRINT((r), ERROR,    err, " "f"\n", ##a)
+#define RUNQ_DEBUG(r,f,a...) RUNQ_PRINT((r), DEBUG,   info, " "f"\n", ##a)
+#endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c
new file mode 100644
index 000000000..ea9e151db
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c
@@ -0,0 +1,286 @@
+/*
+ * Copyright 2018 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "runl.h"
+
+#include <core/memory.h>
+#include <subdev/mc.h>
+#include <subdev/vfn.h>
+
+#include <nvif/class.h>
+
+static u32
+tu102_chan_doorbell_handle(struct nvkm_chan *chan)
+{
+	return (chan->cgrp->runl->id << 16) | chan->id;
+}
+
+static void
+tu102_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	gk104_chan_start(chan);
+	nvkm_wr32(device, device->vfn->addr.user + 0x0090, chan->func->doorbell_handle(chan));
+}
+
+static const struct nvkm_chan_func
+tu102_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gv100_chan_userd,
+	.ramfc = &gv100_chan_ramfc,
+	.bind = gk104_chan_bind_inst,
+	.unbind = gk104_chan_unbind,
+	.start = tu102_chan_start,
+	.stop = gk104_chan_stop,
+	.preempt = gk110_chan_preempt,
+	.doorbell_handle = tu102_chan_doorbell_handle,
+};
+
+static bool
+tu102_runl_pending(struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+
+	return nvkm_rd32(device, 0x002b0c + (runl->id * 0x10)) & 0x00008000;
+}
+
+static void
+tu102_runl_commit(struct nvkm_runl *runl, struct nvkm_memory *memory, u32 start, int count)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u64 addr = nvkm_memory_addr(memory) + start;
+	/*XXX: target? */
+
+	nvkm_wr32(device, 0x002b00 + (runl->id * 0x10), lower_32_bits(addr));
+	nvkm_wr32(device, 0x002b04 + (runl->id * 0x10), upper_32_bits(addr));
+	nvkm_wr32(device, 0x002b08 + (runl->id * 0x10), count);
+}
+
+static const struct nvkm_runl_func
+tu102_runl = {
+	.runqs = 2,
+	.size = 16,
+	.update = nv50_runl_update,
+	.insert_cgrp = gv100_runl_insert_cgrp,
+	.insert_chan = gv100_runl_insert_chan,
+	.commit = tu102_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = tu102_runl_pending,
+	.block = gk104_runl_block,
+	.allow = gk104_runl_allow,
+	.preempt = gv100_runl_preempt,
+	.preempt_pending = gf100_runl_preempt_pending,
+};
+
+static const struct nvkm_enum
+tu102_fifo_mmu_fault_engine[] = {
+	{ 0x01, "DISPLAY" },
+	{ 0x03, "PTP" },
+	{ 0x06, "PWR_PMU" },
+	{ 0x08, "IFB", NULL, NVKM_ENGINE_IFB },
+	{ 0x09, "PERF" },
+	{ 0x1f, "PHYSICAL" },
+	{ 0x20, "HOST0" },
+	{ 0x21, "HOST1" },
+	{ 0x22, "HOST2" },
+	{ 0x23, "HOST3" },
+	{ 0x24, "HOST4" },
+	{ 0x25, "HOST5" },
+	{ 0x26, "HOST6" },
+	{ 0x27, "HOST7" },
+	{ 0x28, "HOST8" },
+	{ 0x29, "HOST9" },
+	{ 0x2a, "HOST10" },
+	{ 0x2b, "HOST11" },
+	{ 0x2c, "HOST12" },
+	{ 0x2d, "HOST13" },
+	{ 0x2e, "HOST14" },
+	{ 0x80, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0xc0, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
+	{}
+};
+
+const struct nvkm_fifo_func_mmu_fault
+tu102_fifo_mmu_fault = {
+	.recover = gf100_fifo_mmu_fault_recover,
+	.access = gv100_fifo_mmu_fault_access,
+	.engine = tu102_fifo_mmu_fault_engine,
+	.reason = gv100_fifo_mmu_fault_reason,
+	.hubclient = gv100_fifo_mmu_fault_hubclient,
+	.gpcclient = gv100_fifo_mmu_fault_gpcclient,
+};
+
+void
+tu102_fifo_intr_ctxsw_timeout_info(struct nvkm_engn *engn, u32 info)
+{
+	struct nvkm_runl *runl = engn->runl;
+	struct nvkm_cgrp *cgrp;
+	unsigned long flags;
+
+	/* Check that engine hasn't become unstuck since timeout raised. */
+	ENGN_DEBUG(engn, "CTXSW_TIMEOUT %08x", info);
+	if (info & 0xc0000000)
+		return;
+
+	/* Determine channel group the engine is stuck on, and schedule recovery. */
+	switch (info & 0x0000c000) {
+	case 0x00004000: /* LOAD */
+		cgrp = nvkm_runl_cgrp_get_cgid(runl, info & 0x3fff0000, &flags);
+		break;
+	case 0x00008000: /* SAVE */
+	case 0x0000c000: /* SWITCH */
+		cgrp = nvkm_runl_cgrp_get_cgid(runl, info & 0x00003fff, &flags);
+		break;
+	default:
+		cgrp = NULL;
+		break;
+	}
+
+	if (!WARN_ON(!cgrp)) {
+		nvkm_runl_rc_cgrp(cgrp);
+		nvkm_cgrp_put(&cgrp, flags);
+	}
+}
+
+static void
+tu102_fifo_intr_ctxsw_timeout(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_runl *runl;
+	struct nvkm_engn *engn;
+	u32 engm = nvkm_rd32(device, 0x002a30);
+	u32 info;
+
+	nvkm_runl_foreach(runl, fifo) {
+		nvkm_runl_foreach_engn_cond(engn, runl, engm & BIT(engn->id)) {
+			info = nvkm_rd32(device, 0x003200 + (engn->id * 4));
+			tu102_fifo_intr_ctxsw_timeout_info(engn, info);
+		}
+	}
+
+	nvkm_wr32(device, 0x002a30, engm);
+}
+
+static void
+tu102_fifo_intr_sched(struct nvkm_fifo *fifo)
+{
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	u32 intr = nvkm_rd32(subdev->device, 0x00254c);
+	u32 code = intr & 0x000000ff;
+
+	nvkm_error(subdev, "SCHED_ERROR %02x\n", code);
+}
+
+static irqreturn_t
+tu102_fifo_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_fifo *fifo = container_of(inth, typeof(*fifo), engine.subdev.inth);
+	struct nvkm_subdev *subdev = &fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, 0x002140);
+	u32 stat = nvkm_rd32(device, 0x002100) & mask;
+
+	if (stat & 0x00000001) {
+		gk104_fifo_intr_bind(fifo);
+		nvkm_wr32(device, 0x002100, 0x00000001);
+		stat &= ~0x00000001;
+	}
+
+	if (stat & 0x00000002) {
+		tu102_fifo_intr_ctxsw_timeout(fifo);
+		stat &= ~0x00000002;
+	}
+
+	if (stat & 0x00000100) {
+		tu102_fifo_intr_sched(fifo);
+		nvkm_wr32(device, 0x002100, 0x00000100);
+		stat &= ~0x00000100;
+	}
+
+	if (stat & 0x00010000) {
+		gk104_fifo_intr_chsw(fifo);
+		nvkm_wr32(device, 0x002100, 0x00010000);
+		stat &= ~0x00010000;
+	}
+
+	if (stat & 0x20000000) {
+		if (gf100_fifo_intr_pbdma(fifo))
+			stat &= ~0x20000000;
+	}
+
+	if (stat & 0x40000000) {
+		gk104_fifo_intr_runlist(fifo);
+		stat &= ~0x40000000;
+	}
+
+	if (stat & 0x80000000) {
+		nvkm_wr32(device, 0x002100, 0x80000000);
+		nvkm_event_ntfy(&fifo->nonstall.event, 0, NVKM_FIFO_NONSTALL_EVENT);
+		stat &= ~0x80000000;
+	}
+
+	if (stat) {
+		nvkm_error(subdev, "INTR %08x\n", stat);
+		spin_lock(&fifo->lock);
+		nvkm_mask(device, 0x002140, stat, 0x00000000);
+		spin_unlock(&fifo->lock);
+		nvkm_wr32(device, 0x002100, stat);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static void
+tu102_fifo_init_pbdmas(struct nvkm_fifo *fifo, u32 mask)
+{
+	/* Not directly related to PBDMAs, but, enables doorbell to function. */
+	nvkm_mask(fifo->engine.subdev.device, 0xb65000, 0x80000000, 0x80000000);
+}
+
+static const struct nvkm_fifo_func
+tu102_fifo = {
+	.chid_nr = gm200_fifo_chid_nr,
+	.chid_ctor = gk110_fifo_chid_ctor,
+	.runq_nr = gm200_fifo_runq_nr,
+	.runl_ctor = gk104_fifo_runl_ctor,
+	.init = gk104_fifo_init,
+	.init_pbdmas = tu102_fifo_init_pbdmas,
+	.intr = tu102_fifo_intr,
+	.mmu_fault = &tu102_fifo_mmu_fault,
+	.nonstall = &gf100_fifo_nonstall,
+	.runl = &tu102_runl,
+	.runq = &gv100_runq,
+	.engn = &gv100_engn,
+	.engn_ce = &gv100_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &gk110_cgrp, .force = true },
+	.chan = {{ 0, 0, TURING_CHANNEL_GPFIFO_A }, &tu102_chan },
+};
+
+int
+tu102_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&tu102_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ucgrp.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ucgrp.c
new file mode 100644
index 000000000..52c594dfb
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ucgrp.c
@@ -0,0 +1,125 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#define nvkm_ucgrp(p) container_of((p), struct nvkm_ucgrp, object)
+#include "priv.h"
+#include "cgrp.h"
+#include "runl.h"
+
+#include <subdev/mmu.h>
+
+#include <nvif/if0021.h>
+
+struct nvkm_ucgrp {
+	struct nvkm_object object;
+	struct nvkm_cgrp *cgrp;
+};
+
+static int
+nvkm_ucgrp_chan_new(const struct nvkm_oclass *oclass, void *argv, u32 argc,
+		    struct nvkm_object **pobject)
+{
+	struct nvkm_cgrp *cgrp = nvkm_ucgrp(oclass->parent)->cgrp;
+
+	return nvkm_uchan_new(cgrp->runl->fifo, cgrp, oclass, argv, argc, pobject);
+}
+
+static int
+nvkm_ucgrp_sclass(struct nvkm_object *object, int index, struct nvkm_oclass *oclass)
+{
+	struct nvkm_cgrp *cgrp = nvkm_ucgrp(object)->cgrp;
+	struct nvkm_fifo *fifo = cgrp->runl->fifo;
+	const struct nvkm_fifo_func_chan *chan = &fifo->func->chan;
+	int c = 0;
+
+	/* *_CHANNEL_GPFIFO_* */
+	if (chan->user.oclass) {
+		if (c++ == index) {
+			oclass->base = chan->user;
+			oclass->ctor = nvkm_ucgrp_chan_new;
+			return 0;
+		}
+	}
+
+	return -EINVAL;
+}
+
+static void *
+nvkm_ucgrp_dtor(struct nvkm_object *object)
+{
+	struct nvkm_ucgrp *ucgrp = nvkm_ucgrp(object);
+
+	nvkm_cgrp_unref(&ucgrp->cgrp);
+	return ucgrp;
+}
+
+static const struct nvkm_object_func
+nvkm_ucgrp = {
+	.dtor = nvkm_ucgrp_dtor,
+	.sclass = nvkm_ucgrp_sclass,
+};
+
+int
+nvkm_ucgrp_new(struct nvkm_fifo *fifo, const struct nvkm_oclass *oclass, void *argv, u32 argc,
+	       struct nvkm_object **pobject)
+{
+	union nvif_cgrp_args *args = argv;
+	struct nvkm_runl *runl;
+	struct nvkm_vmm *vmm;
+	struct nvkm_ucgrp *ucgrp;
+	int ret;
+
+	if (argc < sizeof(args->v0) || args->v0.version != 0)
+		return -ENOSYS;
+	argc -= sizeof(args->v0);
+
+	if (args->v0.namelen != argc)
+		return -EINVAL;
+
+	/* Lookup objects referenced in args. */
+	runl = nvkm_runl_get(fifo, args->v0.runlist, 0);
+	if (!runl)
+		return -EINVAL;
+
+	vmm = nvkm_uvmm_search(oclass->client, args->v0.vmm);
+	if (IS_ERR(vmm))
+		return PTR_ERR(vmm);
+
+	/* Allocate channel group. */
+	if (!(ucgrp = kzalloc(sizeof(*ucgrp), GFP_KERNEL))) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	nvkm_object_ctor(&nvkm_ucgrp, oclass, &ucgrp->object);
+	*pobject = &ucgrp->object;
+
+	ret = nvkm_cgrp_new(runl, args->v0.name, vmm, true, &ucgrp->cgrp);
+	if (ret)
+		goto done;
+
+	/* Return channel group info to caller. */
+	args->v0.cgid = ucgrp->cgrp->id;
+
+done:
+	nvkm_vmm_unref(&vmm);
+	return ret;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
new file mode 100644
index 000000000..1dac95ae7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
@@ -0,0 +1,409 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#define nvkm_uchan(p) container_of((p), struct nvkm_uchan, object)
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+
+#include <core/gpuobj.h>
+#include <core/oproxy.h>
+#include <subdev/mmu.h>
+#include <engine/dma.h>
+
+#include <nvif/if0020.h>
+
+struct nvkm_uchan {
+	struct nvkm_object object;
+	struct nvkm_chan *chan;
+};
+
+static int
+nvkm_uchan_uevent(struct nvkm_object *object, void *argv, u32 argc, struct nvkm_uevent *uevent)
+{
+	struct nvkm_chan *chan = nvkm_uchan(object)->chan;
+	struct nvkm_runl *runl = chan->cgrp->runl;
+	union nvif_chan_event_args *args = argv;
+
+	if (!uevent)
+		return 0;
+	if (argc != sizeof(args->v0) || args->v0.version != 0)
+		return -ENOSYS;
+
+	switch (args->v0.type) {
+	case NVIF_CHAN_EVENT_V0_NON_STALL_INTR:
+		return nvkm_uevent_add(uevent, &runl->fifo->nonstall.event, 0,
+				       NVKM_FIFO_NONSTALL_EVENT, NULL);
+	case NVIF_CHAN_EVENT_V0_KILLED:
+		return nvkm_uevent_add(uevent, &runl->chid->event, chan->id,
+				       NVKM_CHAN_EVENT_ERRORED, NULL);
+	default:
+		break;
+	}
+
+	return -ENOSYS;
+}
+
+struct nvkm_uobj {
+	struct nvkm_oproxy oproxy;
+	struct nvkm_chan *chan;
+	struct nvkm_cctx *cctx;
+	int hash;
+};
+
+static int
+nvkm_uchan_object_fini_1(struct nvkm_oproxy *oproxy, bool suspend)
+{
+	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
+	struct nvkm_chan *chan = uobj->chan;
+	struct nvkm_cctx *cctx = uobj->cctx;
+	struct nvkm_ectx *ectx = cctx->vctx->ectx;
+
+	if (!ectx->object)
+		return 0;
+
+	/* Unbind engine context from channel, if no longer required. */
+	if (refcount_dec_and_mutex_lock(&cctx->uses, &chan->cgrp->mutex)) {
+		nvkm_chan_cctx_bind(chan, ectx->engn, NULL);
+
+		if (refcount_dec_and_test(&ectx->uses))
+			nvkm_object_fini(ectx->object, false);
+		mutex_unlock(&chan->cgrp->mutex);
+	}
+
+	return 0;
+}
+
+static int
+nvkm_uchan_object_init_0(struct nvkm_oproxy *oproxy)
+{
+	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
+	struct nvkm_chan *chan = uobj->chan;
+	struct nvkm_cctx *cctx = uobj->cctx;
+	struct nvkm_ectx *ectx = cctx->vctx->ectx;
+	int ret = 0;
+
+	if (!ectx->object)
+		return 0;
+
+	/* Bind engine context to channel, if it hasn't been already. */
+	if (!refcount_inc_not_zero(&cctx->uses)) {
+		mutex_lock(&chan->cgrp->mutex);
+		if (!refcount_inc_not_zero(&cctx->uses)) {
+			if (!refcount_inc_not_zero(&ectx->uses)) {
+				ret = nvkm_object_init(ectx->object);
+				if (ret == 0)
+					refcount_set(&ectx->uses, 1);
+			}
+
+			if (ret == 0) {
+				nvkm_chan_cctx_bind(chan, ectx->engn, cctx);
+				refcount_set(&cctx->uses, 1);
+			}
+		}
+		mutex_unlock(&chan->cgrp->mutex);
+	}
+
+	return ret;
+}
+
+static void
+nvkm_uchan_object_dtor(struct nvkm_oproxy *oproxy)
+{
+	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
+	struct nvkm_engn *engn;
+
+	if (!uobj->cctx)
+		return;
+
+	engn = uobj->cctx->vctx->ectx->engn;
+	if (engn->func->ramht_del)
+		engn->func->ramht_del(uobj->chan, uobj->hash);
+
+	nvkm_chan_cctx_put(uobj->chan, &uobj->cctx);
+}
+
+static const struct nvkm_oproxy_func
+nvkm_uchan_object = {
+	.dtor[1] = nvkm_uchan_object_dtor,
+	.init[0] = nvkm_uchan_object_init_0,
+	.fini[1] = nvkm_uchan_object_fini_1,
+};
+
+static int
+nvkm_uchan_object_new(const struct nvkm_oclass *oclass, void *argv, u32 argc,
+		      struct nvkm_object **pobject)
+{
+	struct nvkm_chan *chan = nvkm_uchan(oclass->parent)->chan;
+	struct nvkm_cgrp *cgrp = chan->cgrp;
+	struct nvkm_engn *engn;
+	struct nvkm_uobj *uobj;
+	int ret;
+
+	/* Lookup host engine state for target engine. */
+	engn = nvkm_runl_find_engn(engn, cgrp->runl, engn->engine == oclass->engine);
+	if (WARN_ON(!engn))
+		return -EINVAL;
+
+	/* Allocate SW object. */
+	if (!(uobj = kzalloc(sizeof(*uobj), GFP_KERNEL)))
+		return -ENOMEM;
+
+	nvkm_oproxy_ctor(&nvkm_uchan_object, oclass, &uobj->oproxy);
+	uobj->chan = chan;
+	*pobject = &uobj->oproxy.base;
+
+	/* Ref. channel context for target engine.*/
+	ret = nvkm_chan_cctx_get(chan, engn, &uobj->cctx, oclass->client);
+	if (ret)
+		return ret;
+
+	/* Allocate HW object. */
+	ret = oclass->base.ctor(&(const struct nvkm_oclass) {
+					.base = oclass->base,
+					.engn = oclass->engn,
+					.handle = oclass->handle,
+					.object = oclass->object,
+					.client = oclass->client,
+					.parent = uobj->cctx->vctx->ectx->object ?: oclass->parent,
+					.engine = engn->engine,
+				 }, argv, argc, &uobj->oproxy.object);
+	if (ret)
+		return ret;
+
+	if (engn->func->ramht_add) {
+		uobj->hash = engn->func->ramht_add(engn, uobj->oproxy.object, uobj->chan);
+		if (uobj->hash < 0)
+			return uobj->hash;
+	}
+
+	return 0;
+}
+
+static int
+nvkm_uchan_sclass(struct nvkm_object *object, int index, struct nvkm_oclass *oclass)
+{
+	struct nvkm_chan *chan = nvkm_uchan(object)->chan;
+	struct nvkm_engn *engn;
+	int ret, runq = 0;
+
+	nvkm_runl_foreach_engn(engn, chan->cgrp->runl) {
+		struct nvkm_engine *engine = engn->engine;
+		int c = 0;
+
+		/* Each runqueue, on runlists with multiple, has its own LCE. */
+		if (engn->runl->func->runqs) {
+			if (engine->subdev.type == NVKM_ENGINE_CE) {
+				if (chan->runq != runq++)
+					continue;
+			}
+		}
+
+		oclass->engine = engine;
+		oclass->base.oclass = 0;
+
+		if (engine->func->fifo.sclass) {
+			ret = engine->func->fifo.sclass(oclass, index);
+			if (oclass->base.oclass) {
+				if (!oclass->base.ctor)
+					oclass->base.ctor = nvkm_object_new;
+				oclass->ctor = nvkm_uchan_object_new;
+				return 0;
+			}
+
+			index -= ret;
+			continue;
+		}
+
+		while (engine->func->sclass[c].oclass) {
+			if (c++ == index) {
+				oclass->base = engine->func->sclass[index];
+				if (!oclass->base.ctor)
+					oclass->base.ctor = nvkm_object_new;
+				oclass->ctor = nvkm_uchan_object_new;
+				return 0;
+			}
+		}
+
+		index -= c;
+	}
+
+	return -EINVAL;
+}
+
+static int
+nvkm_uchan_map(struct nvkm_object *object, void *argv, u32 argc,
+	       enum nvkm_object_map *type, u64 *addr, u64 *size)
+{
+	struct nvkm_chan *chan = nvkm_uchan(object)->chan;
+	struct nvkm_device *device = chan->cgrp->runl->fifo->engine.subdev.device;
+
+	if (chan->func->userd->bar < 0)
+		return -ENOSYS;
+
+	*type = NVKM_OBJECT_MAP_IO;
+	*addr = device->func->resource_addr(device, chan->func->userd->bar) +
+		chan->func->userd->base + chan->userd.base;
+	*size = chan->func->userd->size;
+	return 0;
+}
+
+static int
+nvkm_uchan_fini(struct nvkm_object *object, bool suspend)
+{
+	struct nvkm_chan *chan = nvkm_uchan(object)->chan;
+
+	nvkm_chan_block(chan);
+	nvkm_chan_remove(chan, true);
+
+	if (chan->func->unbind)
+		chan->func->unbind(chan);
+
+	return 0;
+}
+
+static int
+nvkm_uchan_init(struct nvkm_object *object)
+{
+	struct nvkm_chan *chan = nvkm_uchan(object)->chan;
+
+	if (atomic_read(&chan->errored))
+		return 0;
+
+	if (chan->func->bind)
+		chan->func->bind(chan);
+
+	nvkm_chan_allow(chan);
+	nvkm_chan_insert(chan);
+	return 0;
+}
+
+static void *
+nvkm_uchan_dtor(struct nvkm_object *object)
+{
+	struct nvkm_uchan *uchan = nvkm_uchan(object);
+
+	nvkm_chan_del(&uchan->chan);
+	return uchan;
+}
+
+static const struct nvkm_object_func
+nvkm_uchan = {
+	.dtor = nvkm_uchan_dtor,
+	.init = nvkm_uchan_init,
+	.fini = nvkm_uchan_fini,
+	.map = nvkm_uchan_map,
+	.sclass = nvkm_uchan_sclass,
+	.uevent = nvkm_uchan_uevent,
+};
+
+int
+nvkm_uchan_new(struct nvkm_fifo *fifo, struct nvkm_cgrp *cgrp, const struct nvkm_oclass *oclass,
+	       void *argv, u32 argc, struct nvkm_object **pobject)
+{
+	union nvif_chan_args *args = argv;
+	struct nvkm_runl *runl;
+	struct nvkm_vmm *vmm = NULL;
+	struct nvkm_dmaobj *ctxdma = NULL;
+	struct nvkm_memory *userd = NULL;
+	struct nvkm_uchan *uchan;
+	struct nvkm_chan *chan;
+	int ret;
+
+	if (argc < sizeof(args->v0) || args->v0.version != 0)
+		return -ENOSYS;
+	argc -= sizeof(args->v0);
+
+	if (args->v0.namelen != argc)
+		return -EINVAL;
+
+	/* Lookup objects referenced in args. */
+	runl = nvkm_runl_get(fifo, args->v0.runlist, 0);
+	if (!runl)
+		return -EINVAL;
+
+	if (args->v0.vmm) {
+		vmm = nvkm_uvmm_search(oclass->client, args->v0.vmm);
+		if (IS_ERR(vmm))
+			return PTR_ERR(vmm);
+	}
+
+	if (args->v0.ctxdma) {
+		ctxdma = nvkm_dmaobj_search(oclass->client, args->v0.ctxdma);
+		if (IS_ERR(ctxdma)) {
+			ret = PTR_ERR(ctxdma);
+			goto done;
+		}
+	}
+
+	if (args->v0.huserd) {
+		userd = nvkm_umem_search(oclass->client, args->v0.huserd);
+		if (IS_ERR(userd)) {
+			ret = PTR_ERR(userd);
+			userd = NULL;
+			goto done;
+		}
+	}
+
+	/* Allocate channel. */
+	if (!(uchan = kzalloc(sizeof(*uchan), GFP_KERNEL))) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	nvkm_object_ctor(&nvkm_uchan, oclass, &uchan->object);
+	*pobject = &uchan->object;
+
+	ret = nvkm_chan_new_(fifo->func->chan.func, runl, args->v0.runq, cgrp, args->v0.name,
+			     args->v0.priv != 0, args->v0.devm, vmm, ctxdma, args->v0.offset,
+			     args->v0.length, userd, args->v0.ouserd, &uchan->chan);
+	if (ret)
+		goto done;
+
+	chan = uchan->chan;
+
+	/* Return channel info to caller. */
+	if (chan->func->doorbell_handle)
+		args->v0.token = chan->func->doorbell_handle(chan);
+	else
+		args->v0.token = ~0;
+
+	args->v0.chid = chan->id;
+
+	switch (nvkm_memory_target(chan->inst->memory)) {
+	case NVKM_MEM_TARGET_INST: args->v0.aper = NVIF_CHAN_V0_INST_APER_INST; break;
+	case NVKM_MEM_TARGET_VRAM: args->v0.aper = NVIF_CHAN_V0_INST_APER_VRAM; break;
+	case NVKM_MEM_TARGET_HOST: args->v0.aper = NVIF_CHAN_V0_INST_APER_HOST; break;
+	case NVKM_MEM_TARGET_NCOH: args->v0.aper = NVIF_CHAN_V0_INST_APER_NCOH; break;
+	default:
+		WARN_ON(1);
+		ret = -EFAULT;
+		break;
+	}
+
+	args->v0.inst = nvkm_memory_addr(chan->inst->memory);
+done:
+	nvkm_memory_unref(&userd);
+	nvkm_vmm_unref(&vmm);
+	return ret;
+}