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().
  ...

1 files changed, 1137 insertions, 0 deletions

diff --git a/drivers/misc/habanalabs/common/hw_queue.c b/drivers/misc/habanalabs/common/hw_queue.c
new file mode 100644
index 000000000..d0087c0ec
--- /dev/null
+++ b/drivers/misc/habanalabs/common/hw_queue.c
@@ -0,0 +1,1137 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Copyright 2016-2019 HabanaLabs, Ltd.
+ * All Rights Reserved.
+ */
+
+#include "habanalabs.h"
+
+#include <linux/slab.h>
+
+/*
+ * hl_queue_add_ptr - add to pi or ci and checks if it wraps around
+ *
+ * @ptr: the current pi/ci value
+ * @val: the amount to add
+ *
+ * Add val to ptr. It can go until twice the queue length.
+ */
+inline u32 hl_hw_queue_add_ptr(u32 ptr, u16 val)
+{
+	ptr += val;
+	ptr &= ((HL_QUEUE_LENGTH << 1) - 1);
+	return ptr;
+}
+static inline int queue_ci_get(atomic_t *ci, u32 queue_len)
+{
+	return atomic_read(ci) & ((queue_len << 1) - 1);
+}
+
+static inline int queue_free_slots(struct hl_hw_queue *q, u32 queue_len)
+{
+	int delta = (q->pi - queue_ci_get(&q->ci, queue_len));
+
+	if (delta >= 0)
+		return (queue_len - delta);
+	else
+		return (abs(delta) - queue_len);
+}
+
+void hl_hw_queue_update_ci(struct hl_cs *cs)
+{
+	struct hl_device *hdev = cs->ctx->hdev;
+	struct hl_hw_queue *q;
+	int i;
+
+	if (hdev->disabled)
+		return;
+
+	q = &hdev->kernel_queues[0];
+
+	/* There are no internal queues if H/W queues are being used */
+	if (!hdev->asic_prop.max_queues || q->queue_type == QUEUE_TYPE_HW)
+		return;
+
+	/* We must increment CI for every queue that will never get a
+	 * completion, there are 2 scenarios this can happen:
+	 * 1. All queues of a non completion CS will never get a completion.
+	 * 2. Internal queues never gets completion.
+	 */
+	for (i = 0 ; i < hdev->asic_prop.max_queues ; i++, q++) {
+		if (!cs_needs_completion(cs) || q->queue_type == QUEUE_TYPE_INT)
+			atomic_add(cs->jobs_in_queue_cnt[i], &q->ci);
+	}
+}
+
+/*
+ * hl_hw_queue_submit_bd() - Submit a buffer descriptor to an external or a
+ *                                H/W queue.
+ * @hdev: pointer to habanalabs device structure
+ * @q: pointer to habanalabs queue structure
+ * @ctl: BD's control word
+ * @len: BD's length
+ * @ptr: BD's pointer
+ *
+ * This function assumes there is enough space on the queue to submit a new
+ * BD to it. It initializes the next BD and calls the device specific
+ * function to set the pi (and doorbell)
+ *
+ * This function must be called when the scheduler mutex is taken
+ *
+ */
+void hl_hw_queue_submit_bd(struct hl_device *hdev, struct hl_hw_queue *q,
+		u32 ctl, u32 len, u64 ptr)
+{
+	struct hl_bd *bd;
+
+	bd = q->kernel_address;
+	bd += hl_pi_2_offset(q->pi);
+	bd->ctl = cpu_to_le32(ctl);
+	bd->len = cpu_to_le32(len);
+	bd->ptr = cpu_to_le64(ptr);
+
+	q->pi = hl_queue_inc_ptr(q->pi);
+	hdev->asic_funcs->ring_doorbell(hdev, q->hw_queue_id, q->pi);
+}
+
+/*
+ * ext_queue_sanity_checks - perform some sanity checks on external queue
+ *
+ * @hdev              : pointer to hl_device structure
+ * @q                 :	pointer to hl_hw_queue structure
+ * @num_of_entries    : how many entries to check for space
+ * @reserve_cq_entry  :	whether to reserve an entry in the cq
+ *
+ * H/W queues spinlock should be taken before calling this function
+ *
+ * Perform the following:
+ * - Make sure we have enough space in the h/w queue
+ * - Make sure we have enough space in the completion queue
+ * - Reserve space in the completion queue (needs to be reversed if there
+ *   is a failure down the road before the actual submission of work). Only
+ *   do this action if reserve_cq_entry is true
+ *
+ */
+static int ext_queue_sanity_checks(struct hl_device *hdev,
+				struct hl_hw_queue *q, int num_of_entries,
+				bool reserve_cq_entry)
+{
+	atomic_t *free_slots =
+			&hdev->completion_queue[q->cq_id].free_slots_cnt;
+	int free_slots_cnt;
+
+	/* Check we have enough space in the queue */
+	free_slots_cnt = queue_free_slots(q, HL_QUEUE_LENGTH);
+
+	if (free_slots_cnt < num_of_entries) {
+		dev_dbg(hdev->dev, "Queue %d doesn't have room for %d CBs\n",
+			q->hw_queue_id, num_of_entries);
+		return -EAGAIN;
+	}
+
+	if (reserve_cq_entry) {
+		/*
+		 * Check we have enough space in the completion queue
+		 * Add -1 to counter (decrement) unless counter was already 0
+		 * In that case, CQ is full so we can't submit a new CB because
+		 * we won't get ack on its completion
+		 * atomic_add_unless will return 0 if counter was already 0
+		 */
+		if (atomic_add_negative(num_of_entries * -1, free_slots)) {
+			dev_dbg(hdev->dev, "No space for %d on CQ %d\n",
+				num_of_entries, q->hw_queue_id);
+			atomic_add(num_of_entries, free_slots);
+			return -EAGAIN;
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * int_queue_sanity_checks - perform some sanity checks on internal queue
+ *
+ * @hdev              : pointer to hl_device structure
+ * @q                 :	pointer to hl_hw_queue structure
+ * @num_of_entries    : how many entries to check for space
+ *
+ * H/W queues spinlock should be taken before calling this function
+ *
+ * Perform the following:
+ * - Make sure we have enough space in the h/w queue
+ *
+ */
+static int int_queue_sanity_checks(struct hl_device *hdev,
+					struct hl_hw_queue *q,
+					int num_of_entries)
+{
+	int free_slots_cnt;
+
+	if (num_of_entries > q->int_queue_len) {
+		dev_err(hdev->dev,
+			"Cannot populate queue %u with %u jobs\n",
+			q->hw_queue_id, num_of_entries);
+		return -ENOMEM;
+	}
+
+	/* Check we have enough space in the queue */
+	free_slots_cnt = queue_free_slots(q, q->int_queue_len);
+
+	if (free_slots_cnt < num_of_entries) {
+		dev_dbg(hdev->dev, "Queue %d doesn't have room for %d CBs\n",
+			q->hw_queue_id, num_of_entries);
+		return -EAGAIN;
+	}
+
+	return 0;
+}
+
+/*
+ * hw_queue_sanity_checks() - Make sure we have enough space in the h/w queue
+ * @hdev: Pointer to hl_device structure.
+ * @q: Pointer to hl_hw_queue structure.
+ * @num_of_entries: How many entries to check for space.
+ *
+ * Notice: We do not reserve queue entries so this function mustn't be called
+ *         more than once per CS for the same queue
+ *
+ */
+static int hw_queue_sanity_checks(struct hl_device *hdev, struct hl_hw_queue *q,
+					int num_of_entries)
+{
+	int free_slots_cnt;
+
+	/* Check we have enough space in the queue */
+	free_slots_cnt = queue_free_slots(q, HL_QUEUE_LENGTH);
+
+	if (free_slots_cnt < num_of_entries) {
+		dev_dbg(hdev->dev, "Queue %d doesn't have room for %d CBs\n",
+			q->hw_queue_id, num_of_entries);
+		return -EAGAIN;
+	}
+
+	return 0;
+}
+
+/*
+ * hl_hw_queue_send_cb_no_cmpl - send a single CB (not a JOB) without completion
+ *
+ * @hdev: pointer to hl_device structure
+ * @hw_queue_id: Queue's type
+ * @cb_size: size of CB
+ * @cb_ptr: pointer to CB location
+ *
+ * This function sends a single CB, that must NOT generate a completion entry.
+ * Sending CPU messages can be done instead via 'hl_hw_queue_submit_bd()'
+ */
+int hl_hw_queue_send_cb_no_cmpl(struct hl_device *hdev, u32 hw_queue_id,
+				u32 cb_size, u64 cb_ptr)
+{
+	struct hl_hw_queue *q = &hdev->kernel_queues[hw_queue_id];
+	int rc = 0;
+
+	hdev->asic_funcs->hw_queues_lock(hdev);
+
+	if (hdev->disabled) {
+		rc = -EPERM;
+		goto out;
+	}
+
+	/*
+	 * hl_hw_queue_send_cb_no_cmpl() is called for queues of a H/W queue
+	 * type only on init phase, when the queues are empty and being tested,
+	 * so there is no need for sanity checks.
+	 */
+	if (q->queue_type != QUEUE_TYPE_HW) {
+		rc = ext_queue_sanity_checks(hdev, q, 1, false);
+		if (rc)
+			goto out;
+	}
+
+	hl_hw_queue_submit_bd(hdev, q, 0, cb_size, cb_ptr);
+
+out:
+	hdev->asic_funcs->hw_queues_unlock(hdev);
+
+	return rc;
+}
+
+/*
+ * ext_queue_schedule_job - submit a JOB to an external queue
+ *
+ * @job: pointer to the job that needs to be submitted to the queue
+ *
+ * This function must be called when the scheduler mutex is taken
+ *
+ */
+static void ext_queue_schedule_job(struct hl_cs_job *job)
+{
+	struct hl_device *hdev = job->cs->ctx->hdev;
+	struct hl_hw_queue *q = &hdev->kernel_queues[job->hw_queue_id];
+	struct hl_cq_entry cq_pkt;
+	struct hl_cq *cq;
+	u64 cq_addr;
+	struct hl_cb *cb;
+	u32 ctl;
+	u32 len;
+	u64 ptr;
+
+	/*
+	 * Update the JOB ID inside the BD CTL so the device would know what
+	 * to write in the completion queue
+	 */
+	ctl = ((q->pi << BD_CTL_SHADOW_INDEX_SHIFT) & BD_CTL_SHADOW_INDEX_MASK);
+
+	cb = job->patched_cb;
+	len = job->job_cb_size;
+	ptr = cb->bus_address;
+
+	/* Skip completion flow in case this is a non completion CS */
+	if (!cs_needs_completion(job->cs))
+		goto submit_bd;
+
+	cq_pkt.data = cpu_to_le32(
+			((q->pi << CQ_ENTRY_SHADOW_INDEX_SHIFT)
+				& CQ_ENTRY_SHADOW_INDEX_MASK) |
+			FIELD_PREP(CQ_ENTRY_SHADOW_INDEX_VALID_MASK, 1) |
+			FIELD_PREP(CQ_ENTRY_READY_MASK, 1));
+
+	/*
+	 * No need to protect pi_offset because scheduling to the
+	 * H/W queues is done under the scheduler mutex
+	 *
+	 * No need to check if CQ is full because it was already
+	 * checked in ext_queue_sanity_checks
+	 */
+	cq = &hdev->completion_queue[q->cq_id];
+	cq_addr = cq->bus_address + cq->pi * sizeof(struct hl_cq_entry);
+
+	hdev->asic_funcs->add_end_of_cb_packets(hdev, cb->kernel_address, len,
+						job->user_cb_size,
+						cq_addr,
+						le32_to_cpu(cq_pkt.data),
+						q->msi_vec,
+						job->contains_dma_pkt);
+
+	q->shadow_queue[hl_pi_2_offset(q->pi)] = job;
+
+	cq->pi = hl_cq_inc_ptr(cq->pi);
+
+submit_bd:
+	hl_hw_queue_submit_bd(hdev, q, ctl, len, ptr);
+}
+
+/*
+ * int_queue_schedule_job - submit a JOB to an internal queue
+ *
+ * @job: pointer to the job that needs to be submitted to the queue
+ *
+ * This function must be called when the scheduler mutex is taken
+ *
+ */
+static void int_queue_schedule_job(struct hl_cs_job *job)
+{
+	struct hl_device *hdev = job->cs->ctx->hdev;
+	struct hl_hw_queue *q = &hdev->kernel_queues[job->hw_queue_id];
+	struct hl_bd bd;
+	__le64 *pi;
+
+	bd.ctl = 0;
+	bd.len = cpu_to_le32(job->job_cb_size);
+
+	if (job->is_kernel_allocated_cb)
+		/* bus_address is actually a mmu mapped address
+		 * allocated from an internal pool
+		 */
+		bd.ptr = cpu_to_le64(job->user_cb->bus_address);
+	else
+		bd.ptr = cpu_to_le64((u64) (uintptr_t) job->user_cb);
+
+	pi = q->kernel_address + (q->pi & (q->int_queue_len - 1)) * sizeof(bd);
+
+	q->pi++;
+	q->pi &= ((q->int_queue_len << 1) - 1);
+
+	hdev->asic_funcs->pqe_write(hdev, pi, &bd);
+
+	hdev->asic_funcs->ring_doorbell(hdev, q->hw_queue_id, q->pi);
+}
+
+/*
+ * hw_queue_schedule_job - submit a JOB to a H/W queue
+ *
+ * @job: pointer to the job that needs to be submitted to the queue
+ *
+ * This function must be called when the scheduler mutex is taken
+ *
+ */
+static void hw_queue_schedule_job(struct hl_cs_job *job)
+{
+	struct hl_device *hdev = job->cs->ctx->hdev;
+	struct hl_hw_queue *q = &hdev->kernel_queues[job->hw_queue_id];
+	u64 ptr;
+	u32 offset, ctl, len;
+
+	/*
+	 * Upon PQE completion, COMP_DATA is used as the write data to the
+	 * completion queue (QMAN HBW message), and COMP_OFFSET is used as the
+	 * write address offset in the SM block (QMAN LBW message).
+	 * The write address offset is calculated as "COMP_OFFSET << 2".
+	 */
+	offset = job->cs->sequence & (hdev->asic_prop.max_pending_cs - 1);
+	ctl = ((offset << BD_CTL_COMP_OFFSET_SHIFT) & BD_CTL_COMP_OFFSET_MASK) |
+		((q->pi << BD_CTL_COMP_DATA_SHIFT) & BD_CTL_COMP_DATA_MASK);
+
+	len = job->job_cb_size;
+
+	/*
+	 * A patched CB is created only if a user CB was allocated by driver and
+	 * MMU is disabled. If MMU is enabled, the user CB should be used
+	 * instead. If the user CB wasn't allocated by driver, assume that it
+	 * holds an address.
+	 */
+	if (job->patched_cb)
+		ptr = job->patched_cb->bus_address;
+	else if (job->is_kernel_allocated_cb)
+		ptr = job->user_cb->bus_address;
+	else
+		ptr = (u64) (uintptr_t) job->user_cb;
+
+	hl_hw_queue_submit_bd(hdev, q, ctl, len, ptr);
+}
+
+static int init_signal_cs(struct hl_device *hdev,
+		struct hl_cs_job *job, struct hl_cs_compl *cs_cmpl)
+{
+	struct hl_sync_stream_properties *prop;
+	struct hl_hw_sob *hw_sob;
+	u32 q_idx;
+	int rc = 0;
+
+	q_idx = job->hw_queue_id;
+	prop = &hdev->kernel_queues[q_idx].sync_stream_prop;
+	hw_sob = &prop->hw_sob[prop->curr_sob_offset];
+
+	cs_cmpl->hw_sob = hw_sob;
+	cs_cmpl->sob_val = prop->next_sob_val;
+
+	dev_dbg(hdev->dev,
+		"generate signal CB, sob_id: %d, sob val: %u, q_idx: %d, seq: %llu\n",
+		cs_cmpl->hw_sob->sob_id, cs_cmpl->sob_val, q_idx,
+		cs_cmpl->cs_seq);
+
+	/* we set an EB since we must make sure all oeprations are done
+	 * when sending the signal
+	 */
+	hdev->asic_funcs->gen_signal_cb(hdev, job->patched_cb,
+				cs_cmpl->hw_sob->sob_id, 0, true);
+
+	rc = hl_cs_signal_sob_wraparound_handler(hdev, q_idx, &hw_sob, 1,
+								false);
+
+	job->cs->sob_addr_offset = hw_sob->sob_addr;
+	job->cs->initial_sob_count = prop->next_sob_val - 1;
+
+	return rc;
+}
+
+void hl_hw_queue_encaps_sig_set_sob_info(struct hl_device *hdev,
+			struct hl_cs *cs, struct hl_cs_job *job,
+			struct hl_cs_compl *cs_cmpl)
+{
+	struct hl_cs_encaps_sig_handle *handle = cs->encaps_sig_hdl;
+	u32 offset = 0;
+
+	cs_cmpl->hw_sob = handle->hw_sob;
+
+	/* Note that encaps_sig_wait_offset was validated earlier in the flow
+	 * for offset value which exceeds the max reserved signal count.
+	 * always decrement 1 of the offset since when the user
+	 * set offset 1 for example he mean to wait only for the first
+	 * signal only, which will be pre_sob_val, and if he set offset 2
+	 * then the value required is (pre_sob_val + 1) and so on...
+	 * if user set wait offset to 0, then treat it as legacy wait cs,
+	 * wait for the next signal.
+	 */
+	if (job->encaps_sig_wait_offset)
+		offset = job->encaps_sig_wait_offset - 1;
+
+	cs_cmpl->sob_val = handle->pre_sob_val + offset;
+}
+
+static int init_wait_cs(struct hl_device *hdev, struct hl_cs *cs,
+		struct hl_cs_job *job, struct hl_cs_compl *cs_cmpl)
+{
+	struct hl_gen_wait_properties wait_prop;
+	struct hl_sync_stream_properties *prop;
+	struct hl_cs_compl *signal_cs_cmpl;
+	u32 q_idx;
+
+	q_idx = job->hw_queue_id;
+	prop = &hdev->kernel_queues[q_idx].sync_stream_prop;
+
+	signal_cs_cmpl = container_of(cs->signal_fence,
+					struct hl_cs_compl,
+					base_fence);
+
+	if (cs->encaps_signals) {
+		/* use the encaps signal handle stored earlier in the flow
+		 * and set the SOB information from the encaps
+		 * signals handle
+		 */
+		hl_hw_queue_encaps_sig_set_sob_info(hdev, cs, job, cs_cmpl);
+
+		dev_dbg(hdev->dev, "Wait for encaps signals handle, qidx(%u), CS sequence(%llu), sob val: 0x%x, offset: %u\n",
+				cs->encaps_sig_hdl->q_idx,
+				cs->encaps_sig_hdl->cs_seq,
+				cs_cmpl->sob_val,
+				job->encaps_sig_wait_offset);
+	} else {
+		/* Copy the SOB id and value of the signal CS */
+		cs_cmpl->hw_sob = signal_cs_cmpl->hw_sob;
+		cs_cmpl->sob_val = signal_cs_cmpl->sob_val;
+	}
+
+	/* check again if the signal cs already completed.
+	 * if yes then don't send any wait cs since the hw_sob
+	 * could be in reset already. if signal is not completed
+	 * then get refcount to hw_sob to prevent resetting the sob
+	 * while wait cs is not submitted.
+	 * note that this check is protected by two locks,
+	 * hw queue lock and completion object lock,
+	 * and the same completion object lock also protects
+	 * the hw_sob reset handler function.
+	 * The hw_queue lock prevent out of sync of hw_sob
+	 * refcount value, changed by signal/wait flows.
+	 */
+	spin_lock(&signal_cs_cmpl->lock);
+
+	if (completion_done(&cs->signal_fence->completion)) {
+		spin_unlock(&signal_cs_cmpl->lock);
+		return -EINVAL;
+	}
+
+	kref_get(&cs_cmpl->hw_sob->kref);
+
+	spin_unlock(&signal_cs_cmpl->lock);
+
+	dev_dbg(hdev->dev,
+		"generate wait CB, sob_id: %d, sob_val: 0x%x, mon_id: %d, q_idx: %d, seq: %llu\n",
+		cs_cmpl->hw_sob->sob_id, cs_cmpl->sob_val,
+		prop->base_mon_id, q_idx, cs->sequence);
+
+	wait_prop.data = (void *) job->patched_cb;
+	wait_prop.sob_base = cs_cmpl->hw_sob->sob_id;
+	wait_prop.sob_mask = 0x1;
+	wait_prop.sob_val = cs_cmpl->sob_val;
+	wait_prop.mon_id = prop->base_mon_id;
+	wait_prop.q_idx = q_idx;
+	wait_prop.size = 0;
+
+	hdev->asic_funcs->gen_wait_cb(hdev, &wait_prop);
+
+	mb();
+	hl_fence_put(cs->signal_fence);
+	cs->signal_fence = NULL;
+
+	return 0;
+}
+
+/*
+ * init_signal_wait_cs - initialize a signal/wait CS
+ * @cs: pointer to the signal/wait CS
+ *
+ * H/W queues spinlock should be taken before calling this function
+ */
+static int init_signal_wait_cs(struct hl_cs *cs)
+{
+	struct hl_ctx *ctx = cs->ctx;
+	struct hl_device *hdev = ctx->hdev;
+	struct hl_cs_job *job;
+	struct hl_cs_compl *cs_cmpl =
+			container_of(cs->fence, struct hl_cs_compl, base_fence);
+	int rc = 0;
+
+	/* There is only one job in a signal/wait CS */
+	job = list_first_entry(&cs->job_list, struct hl_cs_job,
+				cs_node);
+
+	if (cs->type & CS_TYPE_SIGNAL)
+		rc = init_signal_cs(hdev, job, cs_cmpl);
+	else if (cs->type & CS_TYPE_WAIT)
+		rc = init_wait_cs(hdev, cs, job, cs_cmpl);
+
+	return rc;
+}
+
+static int encaps_sig_first_staged_cs_handler
+			(struct hl_device *hdev, struct hl_cs *cs)
+{
+	struct hl_cs_compl *cs_cmpl =
+			container_of(cs->fence,
+					struct hl_cs_compl, base_fence);
+	struct hl_cs_encaps_sig_handle *encaps_sig_hdl;
+	struct hl_encaps_signals_mgr *mgr;
+	int rc = 0;
+
+	mgr = &cs->ctx->sig_mgr;
+
+	spin_lock(&mgr->lock);
+	encaps_sig_hdl = idr_find(&mgr->handles, cs->encaps_sig_hdl_id);
+	if (encaps_sig_hdl) {
+		/*
+		 * Set handler CS sequence,
+		 * the CS which contains the encapsulated signals.
+		 */
+		encaps_sig_hdl->cs_seq = cs->sequence;
+		/* store the handle and set encaps signal indication,
+		 * to be used later in cs_do_release to put the last
+		 * reference to encaps signals handlers.
+		 */
+		cs_cmpl->encaps_signals = true;
+		cs_cmpl->encaps_sig_hdl = encaps_sig_hdl;
+
+		/* set hw_sob pointer in completion object
+		 * since it's used in cs_do_release flow to put
+		 * refcount to sob
+		 */
+		cs_cmpl->hw_sob = encaps_sig_hdl->hw_sob;
+		cs_cmpl->sob_val = encaps_sig_hdl->pre_sob_val +
+						encaps_sig_hdl->count;
+
+		dev_dbg(hdev->dev, "CS seq (%llu) added to encaps signal handler id (%u), count(%u), qidx(%u), sob(%u), val(%u)\n",
+				cs->sequence, encaps_sig_hdl->id,
+				encaps_sig_hdl->count,
+				encaps_sig_hdl->q_idx,
+				cs_cmpl->hw_sob->sob_id,
+				cs_cmpl->sob_val);
+
+	} else {
+		dev_err(hdev->dev, "encaps handle id(%u) wasn't found!\n",
+				cs->encaps_sig_hdl_id);
+		rc = -EINVAL;
+	}
+
+	spin_unlock(&mgr->lock);
+
+	return rc;
+}
+
+/*
+ * hl_hw_queue_schedule_cs - schedule a command submission
+ * @cs: pointer to the CS
+ */
+int hl_hw_queue_schedule_cs(struct hl_cs *cs)
+{
+	enum hl_device_status status;
+	struct hl_cs_counters_atomic *cntr;
+	struct hl_ctx *ctx = cs->ctx;
+	struct hl_device *hdev = ctx->hdev;
+	struct hl_cs_job *job, *tmp;
+	struct hl_hw_queue *q;
+	int rc = 0, i, cq_cnt;
+	bool first_entry;
+	u32 max_queues;
+
+	cntr = &hdev->aggregated_cs_counters;
+
+	hdev->asic_funcs->hw_queues_lock(hdev);
+
+	if (!hl_device_operational(hdev, &status)) {
+		atomic64_inc(&cntr->device_in_reset_drop_cnt);
+		atomic64_inc(&ctx->cs_counters.device_in_reset_drop_cnt);
+		dev_err(hdev->dev,
+			"device is %s, CS rejected!\n", hdev->status[status]);
+		rc = -EPERM;
+		goto out;
+	}
+
+	max_queues = hdev->asic_prop.max_queues;
+
+	q = &hdev->kernel_queues[0];
+	for (i = 0, cq_cnt = 0 ; i < max_queues ; i++, q++) {
+		if (cs->jobs_in_queue_cnt[i]) {
+			switch (q->queue_type) {
+			case QUEUE_TYPE_EXT:
+				rc = ext_queue_sanity_checks(hdev, q,
+						cs->jobs_in_queue_cnt[i],
+						cs_needs_completion(cs) ?
+								true : false);
+				break;
+			case QUEUE_TYPE_INT:
+				rc = int_queue_sanity_checks(hdev, q,
+						cs->jobs_in_queue_cnt[i]);
+				break;
+			case QUEUE_TYPE_HW:
+				rc = hw_queue_sanity_checks(hdev, q,
+						cs->jobs_in_queue_cnt[i]);
+				break;
+			default:
+				dev_err(hdev->dev, "Queue type %d is invalid\n",
+					q->queue_type);
+				rc = -EINVAL;
+				break;
+			}
+
+			if (rc) {
+				atomic64_inc(
+					&ctx->cs_counters.queue_full_drop_cnt);
+				atomic64_inc(&cntr->queue_full_drop_cnt);
+				goto unroll_cq_resv;
+			}
+
+			if (q->queue_type == QUEUE_TYPE_EXT)
+				cq_cnt++;
+		}
+	}
+
+	if ((cs->type == CS_TYPE_SIGNAL) || (cs->type == CS_TYPE_WAIT)) {
+		rc = init_signal_wait_cs(cs);
+		if (rc)
+			goto unroll_cq_resv;
+	} else if (cs->type == CS_TYPE_COLLECTIVE_WAIT) {
+		rc = hdev->asic_funcs->collective_wait_init_cs(cs);
+		if (rc)
+			goto unroll_cq_resv;
+	}
+
+	rc = hdev->asic_funcs->pre_schedule_cs(cs);
+	if (rc) {
+		dev_err(hdev->dev,
+			"Failed in pre-submission operations of CS %d.%llu\n",
+			ctx->asid, cs->sequence);
+		goto unroll_cq_resv;
+	}
+
+	hdev->shadow_cs_queue[cs->sequence &
+				(hdev->asic_prop.max_pending_cs - 1)] = cs;
+
+	if (cs->encaps_signals && cs->staged_first) {
+		rc = encaps_sig_first_staged_cs_handler(hdev, cs);
+		if (rc)
+			goto unroll_cq_resv;
+	}
+
+	spin_lock(&hdev->cs_mirror_lock);
+
+	/* Verify staged CS exists and add to the staged list */
+	if (cs->staged_cs && !cs->staged_first) {
+		struct hl_cs *staged_cs;
+
+		staged_cs = hl_staged_cs_find_first(hdev, cs->staged_sequence);
+		if (!staged_cs) {
+			dev_err(hdev->dev,
+				"Cannot find staged submission sequence %llu",
+				cs->staged_sequence);
+			rc = -EINVAL;
+			goto unlock_cs_mirror;
+		}
+
+		if (is_staged_cs_last_exists(hdev, staged_cs)) {
+			dev_err(hdev->dev,
+				"Staged submission sequence %llu already submitted",
+				cs->staged_sequence);
+			rc = -EINVAL;
+			goto unlock_cs_mirror;
+		}
+
+		list_add_tail(&cs->staged_cs_node, &staged_cs->staged_cs_node);
+
+		/* update stream map of the first CS */
+		if (hdev->supports_wait_for_multi_cs)
+			staged_cs->fence->stream_master_qid_map |=
+					cs->fence->stream_master_qid_map;
+	}
+
+	list_add_tail(&cs->mirror_node, &hdev->cs_mirror_list);
+
+	/* Queue TDR if the CS is the first entry and if timeout is wanted */
+	first_entry = list_first_entry(&hdev->cs_mirror_list,
+					struct hl_cs, mirror_node) == cs;
+	if ((hdev->timeout_jiffies != MAX_SCHEDULE_TIMEOUT) &&
+				first_entry && cs_needs_timeout(cs)) {
+		cs->tdr_active = true;
+		schedule_delayed_work(&cs->work_tdr, cs->timeout_jiffies);
+
+	}
+
+	spin_unlock(&hdev->cs_mirror_lock);
+
+	list_for_each_entry_safe(job, tmp, &cs->job_list, cs_node)
+		switch (job->queue_type) {
+		case QUEUE_TYPE_EXT:
+			ext_queue_schedule_job(job);
+			break;
+		case QUEUE_TYPE_INT:
+			int_queue_schedule_job(job);
+			break;
+		case QUEUE_TYPE_HW:
+			hw_queue_schedule_job(job);
+			break;
+		default:
+			break;
+		}
+
+	cs->submitted = true;
+
+	goto out;
+
+unlock_cs_mirror:
+	spin_unlock(&hdev->cs_mirror_lock);
+unroll_cq_resv:
+	q = &hdev->kernel_queues[0];
+	for (i = 0 ; (i < max_queues) && (cq_cnt > 0) ; i++, q++) {
+		if ((q->queue_type == QUEUE_TYPE_EXT) &&
+						(cs->jobs_in_queue_cnt[i])) {
+			atomic_t *free_slots =
+				&hdev->completion_queue[i].free_slots_cnt;
+			atomic_add(cs->jobs_in_queue_cnt[i], free_slots);
+			cq_cnt--;
+		}
+	}
+
+out:
+	hdev->asic_funcs->hw_queues_unlock(hdev);
+
+	return rc;
+}
+
+/*
+ * hl_hw_queue_inc_ci_kernel - increment ci for kernel's queue
+ *
+ * @hdev: pointer to hl_device structure
+ * @hw_queue_id: which queue to increment its ci
+ */
+void hl_hw_queue_inc_ci_kernel(struct hl_device *hdev, u32 hw_queue_id)
+{
+	struct hl_hw_queue *q = &hdev->kernel_queues[hw_queue_id];
+
+	atomic_inc(&q->ci);
+}
+
+static int ext_and_cpu_queue_init(struct hl_device *hdev, struct hl_hw_queue *q,
+					bool is_cpu_queue)
+{
+	void *p;
+	int rc;
+
+	if (is_cpu_queue)
+		p = hl_cpu_accessible_dma_pool_alloc(hdev, HL_QUEUE_SIZE_IN_BYTES, &q->bus_address);
+	else
+		p = hl_asic_dma_alloc_coherent(hdev, HL_QUEUE_SIZE_IN_BYTES, &q->bus_address,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!p)
+		return -ENOMEM;
+
+	q->kernel_address = p;
+
+	q->shadow_queue = kmalloc_array(HL_QUEUE_LENGTH, sizeof(struct hl_cs_job *), GFP_KERNEL);
+	if (!q->shadow_queue) {
+		dev_err(hdev->dev,
+			"Failed to allocate shadow queue for H/W queue %d\n",
+			q->hw_queue_id);
+		rc = -ENOMEM;
+		goto free_queue;
+	}
+
+	/* Make sure read/write pointers are initialized to start of queue */
+	atomic_set(&q->ci, 0);
+	q->pi = 0;
+
+	return 0;
+
+free_queue:
+	if (is_cpu_queue)
+		hl_cpu_accessible_dma_pool_free(hdev, HL_QUEUE_SIZE_IN_BYTES, q->kernel_address);
+	else
+		hl_asic_dma_free_coherent(hdev, HL_QUEUE_SIZE_IN_BYTES, q->kernel_address,
+						q->bus_address);
+
+	return rc;
+}
+
+static int int_queue_init(struct hl_device *hdev, struct hl_hw_queue *q)
+{
+	void *p;
+
+	p = hdev->asic_funcs->get_int_queue_base(hdev, q->hw_queue_id,
+					&q->bus_address, &q->int_queue_len);
+	if (!p) {
+		dev_err(hdev->dev,
+			"Failed to get base address for internal queue %d\n",
+			q->hw_queue_id);
+		return -EFAULT;
+	}
+
+	q->kernel_address = p;
+	q->pi = 0;
+	atomic_set(&q->ci, 0);
+
+	return 0;
+}
+
+static int cpu_queue_init(struct hl_device *hdev, struct hl_hw_queue *q)
+{
+	return ext_and_cpu_queue_init(hdev, q, true);
+}
+
+static int ext_queue_init(struct hl_device *hdev, struct hl_hw_queue *q)
+{
+	return ext_and_cpu_queue_init(hdev, q, false);
+}
+
+static int hw_queue_init(struct hl_device *hdev, struct hl_hw_queue *q)
+{
+	void *p;
+
+	p = hl_asic_dma_alloc_coherent(hdev, HL_QUEUE_SIZE_IN_BYTES, &q->bus_address,
+					GFP_KERNEL | __GFP_ZERO);
+	if (!p)
+		return -ENOMEM;
+
+	q->kernel_address = p;
+
+	/* Make sure read/write pointers are initialized to start of queue */
+	atomic_set(&q->ci, 0);
+	q->pi = 0;
+
+	return 0;
+}
+
+static void sync_stream_queue_init(struct hl_device *hdev, u32 q_idx)
+{
+	struct hl_sync_stream_properties *sync_stream_prop;
+	struct asic_fixed_properties *prop = &hdev->asic_prop;
+	struct hl_hw_sob *hw_sob;
+	int sob, reserved_mon_idx, queue_idx;
+
+	sync_stream_prop = &hdev->kernel_queues[q_idx].sync_stream_prop;
+
+	/* We use 'collective_mon_idx' as a running index in order to reserve
+	 * monitors for collective master/slave queues.
+	 * collective master queue gets 2 reserved monitors
+	 * collective slave queue gets 1 reserved monitor
+	 */
+	if (hdev->kernel_queues[q_idx].collective_mode ==
+			HL_COLLECTIVE_MASTER) {
+		reserved_mon_idx = hdev->collective_mon_idx;
+
+		/* reserve the first monitor for collective master queue */
+		sync_stream_prop->collective_mstr_mon_id[0] =
+			prop->collective_first_mon + reserved_mon_idx;
+
+		/* reserve the second monitor for collective master queue */
+		sync_stream_prop->collective_mstr_mon_id[1] =
+			prop->collective_first_mon + reserved_mon_idx + 1;
+
+		hdev->collective_mon_idx += HL_COLLECTIVE_RSVD_MSTR_MONS;
+	} else if (hdev->kernel_queues[q_idx].collective_mode ==
+			HL_COLLECTIVE_SLAVE) {
+		reserved_mon_idx = hdev->collective_mon_idx++;
+
+		/* reserve a monitor for collective slave queue */
+		sync_stream_prop->collective_slave_mon_id =
+			prop->collective_first_mon + reserved_mon_idx;
+	}
+
+	if (!hdev->kernel_queues[q_idx].supports_sync_stream)
+		return;
+
+	queue_idx = hdev->sync_stream_queue_idx++;
+
+	sync_stream_prop->base_sob_id = prop->sync_stream_first_sob +
+			(queue_idx * HL_RSVD_SOBS);
+	sync_stream_prop->base_mon_id = prop->sync_stream_first_mon +
+			(queue_idx * HL_RSVD_MONS);
+	sync_stream_prop->next_sob_val = 1;
+	sync_stream_prop->curr_sob_offset = 0;
+
+	for (sob = 0 ; sob < HL_RSVD_SOBS ; sob++) {
+		hw_sob = &sync_stream_prop->hw_sob[sob];
+		hw_sob->hdev = hdev;
+		hw_sob->sob_id = sync_stream_prop->base_sob_id + sob;
+		hw_sob->sob_addr =
+			hdev->asic_funcs->get_sob_addr(hdev, hw_sob->sob_id);
+		hw_sob->q_idx = q_idx;
+		kref_init(&hw_sob->kref);
+	}
+}
+
+static void sync_stream_queue_reset(struct hl_device *hdev, u32 q_idx)
+{
+	struct hl_sync_stream_properties *prop =
+			&hdev->kernel_queues[q_idx].sync_stream_prop;
+
+	/*
+	 * In case we got here due to a stuck CS, the refcnt might be bigger
+	 * than 1 and therefore we reset it.
+	 */
+	kref_init(&prop->hw_sob[prop->curr_sob_offset].kref);
+	prop->curr_sob_offset = 0;
+	prop->next_sob_val = 1;
+}
+
+/*
+ * queue_init - main initialization function for H/W queue object
+ *
+ * @hdev: pointer to hl_device device structure
+ * @q: pointer to hl_hw_queue queue structure
+ * @hw_queue_id: The id of the H/W queue
+ *
+ * Allocate dma-able memory for the queue and initialize fields
+ * Returns 0 on success
+ */
+static int queue_init(struct hl_device *hdev, struct hl_hw_queue *q,
+			u32 hw_queue_id)
+{
+	int rc;
+
+	q->hw_queue_id = hw_queue_id;
+
+	switch (q->queue_type) {
+	case QUEUE_TYPE_EXT:
+		rc = ext_queue_init(hdev, q);
+		break;
+	case QUEUE_TYPE_INT:
+		rc = int_queue_init(hdev, q);
+		break;
+	case QUEUE_TYPE_CPU:
+		rc = cpu_queue_init(hdev, q);
+		break;
+	case QUEUE_TYPE_HW:
+		rc = hw_queue_init(hdev, q);
+		break;
+	case QUEUE_TYPE_NA:
+		q->valid = 0;
+		return 0;
+	default:
+		dev_crit(hdev->dev, "wrong queue type %d during init\n",
+			q->queue_type);
+		rc = -EINVAL;
+		break;
+	}
+
+	sync_stream_queue_init(hdev, q->hw_queue_id);
+
+	if (rc)
+		return rc;
+
+	q->valid = 1;
+
+	return 0;
+}
+
+/*
+ * hw_queue_fini - destroy queue
+ *
+ * @hdev: pointer to hl_device device structure
+ * @q: pointer to hl_hw_queue queue structure
+ *
+ * Free the queue memory
+ */
+static void queue_fini(struct hl_device *hdev, struct hl_hw_queue *q)
+{
+	if (!q->valid)
+		return;
+
+	/*
+	 * If we arrived here, there are no jobs waiting on this queue
+	 * so we can safely remove it.
+	 * This is because this function can only called when:
+	 * 1. Either a context is deleted, which only can occur if all its
+	 *    jobs were finished
+	 * 2. A context wasn't able to be created due to failure or timeout,
+	 *    which means there are no jobs on the queue yet
+	 *
+	 * The only exception are the queues of the kernel context, but
+	 * if they are being destroyed, it means that the entire module is
+	 * being removed. If the module is removed, it means there is no open
+	 * user context. It also means that if a job was submitted by
+	 * the kernel driver (e.g. context creation), the job itself was
+	 * released by the kernel driver when a timeout occurred on its
+	 * Completion. Thus, we don't need to release it again.
+	 */
+
+	if (q->queue_type == QUEUE_TYPE_INT)
+		return;
+
+	kfree(q->shadow_queue);
+
+	if (q->queue_type == QUEUE_TYPE_CPU)
+		hl_cpu_accessible_dma_pool_free(hdev, HL_QUEUE_SIZE_IN_BYTES, q->kernel_address);
+	else
+		hl_asic_dma_free_coherent(hdev, HL_QUEUE_SIZE_IN_BYTES, q->kernel_address,
+						q->bus_address);
+}
+
+int hl_hw_queues_create(struct hl_device *hdev)
+{
+	struct asic_fixed_properties *asic = &hdev->asic_prop;
+	struct hl_hw_queue *q;
+	int i, rc, q_ready_cnt;
+
+	hdev->kernel_queues = kcalloc(asic->max_queues,
+				sizeof(*hdev->kernel_queues), GFP_KERNEL);
+
+	if (!hdev->kernel_queues) {
+		dev_err(hdev->dev, "Not enough memory for H/W queues\n");
+		return -ENOMEM;
+	}
+
+	/* Initialize the H/W queues */
+	for (i = 0, q_ready_cnt = 0, q = hdev->kernel_queues;
+			i < asic->max_queues ; i++, q_ready_cnt++, q++) {
+
+		q->queue_type = asic->hw_queues_props[i].type;
+		q->supports_sync_stream =
+				asic->hw_queues_props[i].supports_sync_stream;
+		q->collective_mode = asic->hw_queues_props[i].collective_mode;
+		rc = queue_init(hdev, q, i);
+		if (rc) {
+			dev_err(hdev->dev,
+				"failed to initialize queue %d\n", i);
+			goto release_queues;
+		}
+	}
+
+	return 0;
+
+release_queues:
+	for (i = 0, q = hdev->kernel_queues ; i < q_ready_cnt ; i++, q++)
+		queue_fini(hdev, q);
+
+	kfree(hdev->kernel_queues);
+
+	return rc;
+}
+
+void hl_hw_queues_destroy(struct hl_device *hdev)
+{
+	struct hl_hw_queue *q;
+	u32 max_queues = hdev->asic_prop.max_queues;
+	int i;
+
+	for (i = 0, q = hdev->kernel_queues ; i < max_queues ; i++, q++)
+		queue_fini(hdev, q);
+
+	kfree(hdev->kernel_queues);
+}
+
+void hl_hw_queue_reset(struct hl_device *hdev, bool hard_reset)
+{
+	struct hl_hw_queue *q;
+	u32 max_queues = hdev->asic_prop.max_queues;
+	int i;
+
+	for (i = 0, q = hdev->kernel_queues ; i < max_queues ; i++, q++) {
+		if ((!q->valid) ||
+			((!hard_reset) && (q->queue_type == QUEUE_TYPE_CPU)))
+			continue;
+		q->pi = 0;
+		atomic_set(&q->ci, 0);
+
+		if (q->supports_sync_stream)
+			sync_stream_queue_reset(hdev, q->hw_queue_id);
+	}
+}