aboutsummaryrefslogtreecommitdiff
path: root/Documentation/scheduler/completion.rst
diff options
context:
space:
mode:
authorLibravatar Linus Torvalds <torvalds@linux-foundation.org>2023-02-21 18:24:12 -0800
committerLibravatar Linus Torvalds <torvalds@linux-foundation.org>2023-02-21 18:24:12 -0800
commit5b7c4cabbb65f5c469464da6c5f614cbd7f730f2 (patch)
treecc5c2d0a898769fd59549594fedb3ee6f84e59a0 /Documentation/scheduler/completion.rst
downloadlinux-5b7c4cabbb65f5c469464da6c5f614cbd7f730f2.tar.gz
linux-5b7c4cabbb65f5c469464da6c5f614cbd7f730f2.zip
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(). ...
Diffstat (limited to 'Documentation/scheduler/completion.rst')
-rw-r--r--Documentation/scheduler/completion.rst293
1 files changed, 293 insertions, 0 deletions
diff --git a/Documentation/scheduler/completion.rst b/Documentation/scheduler/completion.rst
new file mode 100644
index 000000000..9f039b4f4
--- /dev/null
+++ b/Documentation/scheduler/completion.rst
@@ -0,0 +1,293 @@
+================================================
+Completions - "wait for completion" barrier APIs
+================================================
+
+Introduction:
+-------------
+
+If you have one or more threads that must wait for some kernel activity
+to have reached a point or a specific state, completions can provide a
+race-free solution to this problem. Semantically they are somewhat like a
+pthread_barrier() and have similar use-cases.
+
+Completions are a code synchronization mechanism which is preferable to any
+misuse of locks/semaphores and busy-loops. Any time you think of using
+yield() or some quirky msleep(1) loop to allow something else to proceed,
+you probably want to look into using one of the wait_for_completion*()
+calls and complete() instead.
+
+The advantage of using completions is that they have a well defined, focused
+purpose which makes it very easy to see the intent of the code, but they
+also result in more efficient code as all threads can continue execution
+until the result is actually needed, and both the waiting and the signalling
+is highly efficient using low level scheduler sleep/wakeup facilities.
+
+Completions are built on top of the waitqueue and wakeup infrastructure of
+the Linux scheduler. The event the threads on the waitqueue are waiting for
+is reduced to a simple flag in 'struct completion', appropriately called "done".
+
+As completions are scheduling related, the code can be found in
+kernel/sched/completion.c.
+
+
+Usage:
+------
+
+There are three main parts to using completions:
+
+ - the initialization of the 'struct completion' synchronization object
+ - the waiting part through a call to one of the variants of wait_for_completion(),
+ - the signaling side through a call to complete() or complete_all().
+
+There are also some helper functions for checking the state of completions.
+Note that while initialization must happen first, the waiting and signaling
+part can happen in any order. I.e. it's entirely normal for a thread
+to have marked a completion as 'done' before another thread checks whether
+it has to wait for it.
+
+To use completions you need to #include <linux/completion.h> and
+create a static or dynamic variable of type 'struct completion',
+which has only two fields::
+
+ struct completion {
+ unsigned int done;
+ wait_queue_head_t wait;
+ };
+
+This provides the ->wait waitqueue to place tasks on for waiting (if any), and
+the ->done completion flag for indicating whether it's completed or not.
+
+Completions should be named to refer to the event that is being synchronized on.
+A good example is::
+
+ wait_for_completion(&early_console_added);
+
+ complete(&early_console_added);
+
+Good, intuitive naming (as always) helps code readability. Naming a completion
+'complete' is not helpful unless the purpose is super obvious...
+
+
+Initializing completions:
+-------------------------
+
+Dynamically allocated completion objects should preferably be embedded in data
+structures that are assured to be alive for the life-time of the function/driver,
+to prevent races with asynchronous complete() calls from occurring.
+
+Particular care should be taken when using the _timeout() or _killable()/_interruptible()
+variants of wait_for_completion(), as it must be assured that memory de-allocation
+does not happen until all related activities (complete() or reinit_completion())
+have taken place, even if these wait functions return prematurely due to a timeout
+or a signal triggering.
+
+Initializing of dynamically allocated completion objects is done via a call to
+init_completion()::
+
+ init_completion(&dynamic_object->done);
+
+In this call we initialize the waitqueue and set ->done to 0, i.e. "not completed"
+or "not done".
+
+The re-initialization function, reinit_completion(), simply resets the
+->done field to 0 ("not done"), without touching the waitqueue.
+Callers of this function must make sure that there are no racy
+wait_for_completion() calls going on in parallel.
+
+Calling init_completion() on the same completion object twice is
+most likely a bug as it re-initializes the queue to an empty queue and
+enqueued tasks could get "lost" - use reinit_completion() in that case,
+but be aware of other races.
+
+For static declaration and initialization, macros are available.
+
+For static (or global) declarations in file scope you can use
+DECLARE_COMPLETION()::
+
+ static DECLARE_COMPLETION(setup_done);
+ DECLARE_COMPLETION(setup_done);
+
+Note that in this case the completion is boot time (or module load time)
+initialized to 'not done' and doesn't require an init_completion() call.
+
+When a completion is declared as a local variable within a function,
+then the initialization should always use DECLARE_COMPLETION_ONSTACK()
+explicitly, not just to make lockdep happy, but also to make it clear
+that limited scope had been considered and is intentional::
+
+ DECLARE_COMPLETION_ONSTACK(setup_done)
+
+Note that when using completion objects as local variables you must be
+acutely aware of the short life time of the function stack: the function
+must not return to a calling context until all activities (such as waiting
+threads) have ceased and the completion object is completely unused.
+
+To emphasise this again: in particular when using some of the waiting API variants
+with more complex outcomes, such as the timeout or signalling (_timeout(),
+_killable() and _interruptible()) variants, the wait might complete
+prematurely while the object might still be in use by another thread - and a return
+from the wait_on_completion*() caller function will deallocate the function
+stack and cause subtle data corruption if a complete() is done in some
+other thread. Simple testing might not trigger these kinds of races.
+
+If unsure, use dynamically allocated completion objects, preferably embedded
+in some other long lived object that has a boringly long life time which
+exceeds the life time of any helper threads using the completion object,
+or has a lock or other synchronization mechanism to make sure complete()
+is not called on a freed object.
+
+A naive DECLARE_COMPLETION() on the stack triggers a lockdep warning.
+
+Waiting for completions:
+------------------------
+
+For a thread to wait for some concurrent activity to finish, it
+calls wait_for_completion() on the initialized completion structure::
+
+ void wait_for_completion(struct completion *done)
+
+A typical usage scenario is::
+
+ CPU#1 CPU#2
+
+ struct completion setup_done;
+
+ init_completion(&setup_done);
+ initialize_work(...,&setup_done,...);
+
+ /* run non-dependent code */ /* do setup */
+
+ wait_for_completion(&setup_done); complete(setup_done);
+
+This is not implying any particular order between wait_for_completion() and
+the call to complete() - if the call to complete() happened before the call
+to wait_for_completion() then the waiting side simply will continue
+immediately as all dependencies are satisfied; if not, it will block until
+completion is signaled by complete().
+
+Note that wait_for_completion() is calling spin_lock_irq()/spin_unlock_irq(),
+so it can only be called safely when you know that interrupts are enabled.
+Calling it from IRQs-off atomic contexts will result in hard-to-detect
+spurious enabling of interrupts.
+
+The default behavior is to wait without a timeout and to mark the task as
+uninterruptible. wait_for_completion() and its variants are only safe
+in process context (as they can sleep) but not in atomic context,
+interrupt context, with disabled IRQs, or preemption is disabled - see also
+try_wait_for_completion() below for handling completion in atomic/interrupt
+context.
+
+As all variants of wait_for_completion() can (obviously) block for a long
+time depending on the nature of the activity they are waiting for, so in
+most cases you probably don't want to call this with held mutexes.
+
+
+wait_for_completion*() variants available:
+------------------------------------------
+
+The below variants all return status and this status should be checked in
+most(/all) cases - in cases where the status is deliberately not checked you
+probably want to make a note explaining this (e.g. see
+arch/arm/kernel/smp.c:__cpu_up()).
+
+A common problem that occurs is to have unclean assignment of return types,
+so take care to assign return-values to variables of the proper type.
+
+Checking for the specific meaning of return values also has been found
+to be quite inaccurate, e.g. constructs like::
+
+ if (!wait_for_completion_interruptible_timeout(...))
+
+... would execute the same code path for successful completion and for the
+interrupted case - which is probably not what you want::
+
+ int wait_for_completion_interruptible(struct completion *done)
+
+This function marks the task TASK_INTERRUPTIBLE while it is waiting.
+If a signal was received while waiting it will return -ERESTARTSYS; 0 otherwise::
+
+ unsigned long wait_for_completion_timeout(struct completion *done, unsigned long timeout)
+
+The task is marked as TASK_UNINTERRUPTIBLE and will wait at most 'timeout'
+jiffies. If a timeout occurs it returns 0, else the remaining time in
+jiffies (but at least 1).
+
+Timeouts are preferably calculated with msecs_to_jiffies() or usecs_to_jiffies(),
+to make the code largely HZ-invariant.
+
+If the returned timeout value is deliberately ignored a comment should probably explain
+why (e.g. see drivers/mfd/wm8350-core.c wm8350_read_auxadc())::
+
+ long wait_for_completion_interruptible_timeout(struct completion *done, unsigned long timeout)
+
+This function passes a timeout in jiffies and marks the task as
+TASK_INTERRUPTIBLE. If a signal was received it will return -ERESTARTSYS;
+otherwise it returns 0 if the completion timed out, or the remaining time in
+jiffies if completion occurred.
+
+Further variants include _killable which uses TASK_KILLABLE as the
+designated tasks state and will return -ERESTARTSYS if it is interrupted,
+or 0 if completion was achieved. There is a _timeout variant as well::
+
+ long wait_for_completion_killable(struct completion *done)
+ long wait_for_completion_killable_timeout(struct completion *done, unsigned long timeout)
+
+The _io variants wait_for_completion_io() behave the same as the non-_io
+variants, except for accounting waiting time as 'waiting on IO', which has
+an impact on how the task is accounted in scheduling/IO stats::
+
+ void wait_for_completion_io(struct completion *done)
+ unsigned long wait_for_completion_io_timeout(struct completion *done, unsigned long timeout)
+
+
+Signaling completions:
+----------------------
+
+A thread that wants to signal that the conditions for continuation have been
+achieved calls complete() to signal exactly one of the waiters that it can
+continue::
+
+ void complete(struct completion *done)
+
+... or calls complete_all() to signal all current and future waiters::
+
+ void complete_all(struct completion *done)
+
+The signaling will work as expected even if completions are signaled before
+a thread starts waiting. This is achieved by the waiter "consuming"
+(decrementing) the done field of 'struct completion'. Waiting threads
+wakeup order is the same in which they were enqueued (FIFO order).
+
+If complete() is called multiple times then this will allow for that number
+of waiters to continue - each call to complete() will simply increment the
+done field. Calling complete_all() multiple times is a bug though. Both
+complete() and complete_all() can be called in IRQ/atomic context safely.
+
+There can only be one thread calling complete() or complete_all() on a
+particular 'struct completion' at any time - serialized through the wait
+queue spinlock. Any such concurrent calls to complete() or complete_all()
+probably are a design bug.
+
+Signaling completion from IRQ context is fine as it will appropriately
+lock with spin_lock_irqsave()/spin_unlock_irqrestore() and it will never
+sleep.
+
+
+try_wait_for_completion()/completion_done():
+--------------------------------------------
+
+The try_wait_for_completion() function will not put the thread on the wait
+queue but rather returns false if it would need to enqueue (block) the thread,
+else it consumes one posted completion and returns true::
+
+ bool try_wait_for_completion(struct completion *done)
+
+Finally, to check the state of a completion without changing it in any way,
+call completion_done(), which returns false if there are no posted
+completions that were not yet consumed by waiters (implying that there are
+waiters) and true otherwise::
+
+ bool completion_done(struct completion *done)
+
+Both try_wait_for_completion() and completion_done() are safe to be called in
+IRQ or atomic context.