The HAVE_ prefix means that the code could be enabled. Add another
variable for HAVE_HARDLOCKUP_DETECTOR_SPARC64 without this prefix.
It will be set when it should be built. It will make it compatible
with the other hardlockup detectors.
Before, it is far from obvious that the SPARC64 variant is actually used:
$> make ARCH=sparc64 defconfig
$> grep HARDLOCKUP_DETECTOR .config
CONFIG_HAVE_HARDLOCKUP_DETECTOR_BUDDY=y
CONFIG_HAVE_HARDLOCKUP_DETECTOR_SPARC64=y
After, it is more clear:
$> make ARCH=sparc64 defconfig
$> grep HARDLOCKUP_DETECTOR .config
CONFIG_HAVE_HARDLOCKUP_DETECTOR_BUDDY=y
CONFIG_HAVE_HARDLOCKUP_DETECTOR_SPARC64=y
CONFIG_HARDLOCKUP_DETECTOR_SPARC64=y
Link: https://lkml.kernel.org/r/20230616150618.6073-6-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
There are several hardlockup detector implementations and several Kconfig
values which allow selection and build of the preferred one.
CONFIG_HARDLOCKUP_DETECTOR was introduced by the commit 23637d477c
("lockup_detector: Introduce CONFIG_HARDLOCKUP_DETECTOR") in v2.6.36.
It was a preparation step for introducing the new generic perf hardlockup
detector.
The existing arch-specific variants did not support the to-be-created
generic build configurations, sysctl interface, etc. This distinction
was made explicit by the commit 4a7863cc2e ("x86, nmi_watchdog:
Remove ARCH_HAS_NMI_WATCHDOG and rely on CONFIG_HARDLOCKUP_DETECTOR")
in v2.6.38.
CONFIG_HAVE_NMI_WATCHDOG was introduced by the commit d314d74c69
("nmi watchdog: do not use cpp symbol in Kconfig") in v3.4-rc1. It replaced
the above mentioned ARCH_HAS_NMI_WATCHDOG. At that time, it was still used
by three architectures, namely blackfin, mn10300, and sparc.
The support for blackfin and mn10300 architectures has been completely
dropped some time ago. And sparc is the only architecture with the historic
NMI watchdog at the moment.
And the old sparc implementation is really special. It is always built on
sparc64. It used to be always enabled until the commit 7a5c8b57ce
("sparc: implement watchdog_nmi_enable and watchdog_nmi_disable") added
in v4.10-rc1.
There are only few locations where the sparc64 NMI watchdog interacts
with the generic hardlockup detectors code:
+ implements arch_touch_nmi_watchdog() which is called from the generic
touch_nmi_watchdog()
+ implements watchdog_hardlockup_enable()/disable() to support
/proc/sys/kernel/nmi_watchdog
+ is always preferred over other generic watchdogs, see
CONFIG_HARDLOCKUP_DETECTOR
+ includes asm/nmi.h into linux/nmi.h because some sparc-specific
functions are needed in sparc-specific code which includes
only linux/nmi.h.
The situation became more complicated after the commit 05a4a95279
("kernel/watchdog: split up config options") and commit 2104180a53
("powerpc/64s: implement arch-specific hardlockup watchdog") in v4.13-rc1.
They introduced HAVE_HARDLOCKUP_DETECTOR_ARCH. It was used for powerpc
specific hardlockup detector. It was compatible with the perf one
regarding the general boot, sysctl, and programming interfaces.
HAVE_HARDLOCKUP_DETECTOR_ARCH was defined as a superset of
HAVE_NMI_WATCHDOG. It made some sense because all arch-specific
detectors had some common requirements, namely:
+ implemented arch_touch_nmi_watchdog()
+ included asm/nmi.h into linux/nmi.h
+ defined the default value for /proc/sys/kernel/nmi_watchdog
But it actually has made things pretty complicated when the generic
buddy hardlockup detector was added. Before the generic perf detector
was newer supported together with an arch-specific one. But the buddy
detector could work on any SMP system. It means that an architecture
could support both the arch-specific and buddy detector.
As a result, there are few tricky dependencies. For example,
CONFIG_HARDLOCKUP_DETECTOR depends on:
((HAVE_HARDLOCKUP_DETECTOR_PERF || HAVE_HARDLOCKUP_DETECTOR_BUDDY) && !HAVE_NMI_WATCHDOG) || HAVE_HARDLOCKUP_DETECTOR_ARCH
The problem is that the very special sparc implementation is defined as:
HAVE_NMI_WATCHDOG && !HAVE_HARDLOCKUP_DETECTOR_ARCH
Another problem is that the meaning of HAVE_NMI_WATCHDOG is far from clear
without reading understanding the history.
Make the logic less tricky and more self-explanatory by making
HAVE_NMI_WATCHDOG specific for the sparc64 implementation. And rename it to
HAVE_HARDLOCKUP_DETECTOR_SPARC64.
Note that HARDLOCKUP_DETECTOR_PREFER_BUDDY, HARDLOCKUP_DETECTOR_PERF,
and HARDLOCKUP_DETECTOR_BUDDY may conflict only with
HAVE_HARDLOCKUP_DETECTOR_ARCH. They depend on HARDLOCKUP_DETECTOR
and it is not longer enabled when HAVE_NMI_WATCHDOG is set.
Link: https://lkml.kernel.org/r/20230616150618.6073-5-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
In the patch ("watchdog/hardlockup: detect hard lockups using secondary
(buddy) CPUs"), we added a call from the common watchdog.c file into the
buddy. That call could be done more cleanly. Specifically:
1. If we move the call into watchdog_hardlockup_kick() then it keeps
watchdog_timer_fn() simpler.
2. We don't need to pass an "unsigned long" to the buddy for the timer
count. In the patch ("watchdog/hardlockup: add a "cpu" param to
watchdog_hardlockup_check()") the count was changed to "atomic_t"
which is backed by an int, so we should match types.
Link: https://lkml.kernel.org/r/20230526184139.6.I006c7d958a1ea5c4e1e4dc44a25596d9bb5fd3ba@changeid
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Suggested-by: Petr Mladek <pmladek@suse.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Right now there is one arch (sparc64) that selects HAVE_NMI_WATCHDOG
without selecting HAVE_HARDLOCKUP_DETECTOR_ARCH. Because of that one
architecture, we have some special case code in the watchdog core to
handle the fact that watchdog_hardlockup_probe() isn't implemented.
Let's implement watchdog_hardlockup_probe() for sparc64 and get rid of the
special case.
As a side effect of doing this, code inspection tells us that we could fix
a minor bug where the system won't properly realize that NMI watchdogs are
disabled. Specifically, on powerpc if CONFIG_PPC_WATCHDOG is turned off
the arch might still select CONFIG_HAVE_HARDLOCKUP_DETECTOR_ARCH which
selects CONFIG_HAVE_NMI_WATCHDOG. Since CONFIG_PPC_WATCHDOG was off then
nothing will override the "weak" watchdog_hardlockup_probe() and we'll
fallback to looking at CONFIG_HAVE_NMI_WATCHDOG.
Link: https://lkml.kernel.org/r/20230526184139.2.Ic6ebbf307ca0efe91f08ce2c1eb4a037ba6b0700@changeid
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Suggested-by: Petr Mladek <pmladek@suse.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "watchdog: Cleanup / fixes after buddy series v5 reviews".
This patch series attempts to finish resolving the feedback received
from Petr Mladek on the v5 series I posted.
Probably the only thing that wasn't fully as clean as Petr requested was
the Kconfig stuff. I couldn't find a better way to express it without a
more major overhaul. In the very least, I renamed "NON_ARCH" to
"PERF_OR_BUDDY" in the hopes that will make it marginally better.
Nothing in this series is terribly critical and even the bugfixes are
small. However, it does cleanup a few things that were pointed out in
review.
This patch (of 10):
The permissions for the kernel.nmi_watchdog sysctl have always been set at
compile time despite the fact that a watchdog can fail to probe. Let's
fix this and set the permissions based on whether the hardlockup detector
actually probed.
Link: https://lkml.kernel.org/r/20230527014153.2793931-1-dianders@chromium.org
Link: https://lkml.kernel.org/r/20230526184139.1.I0d75971cc52a7283f495aac0bd5c3041aadc734e@changeid
Fixes: a994a3147e ("watchdog/hardlockup/perf: Implement init time detection of perf")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reported-by: Petr Mladek <pmladek@suse.com>
Closes: https://lore.kernel.org/r/ZHCn4hNxFpY5-9Ki@alley
Reviewed-by: Petr Mladek <pmladek@suse.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Move out flags validation and license checks out of the permission
checks. They were intermingled, which makes subsequent changes harder.
Clean this up: perform straightforward flag validation upfront, and
fetch and check license later, right where we use it. Also consolidate
capabilities check in one block, right after basic attribute sanity
checks.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/bpf/20230613223533.3689589-5-andrii@kernel.org
This allows to do more centralized decisions later on, and generally
makes it very explicit which maps are privileged and which are not
(e.g., LRU_HASH and LRU_PERCPU_HASH, which are privileged HASH variants,
as opposed to unprivileged HASH and HASH_PERCPU; now this is explicit
and easy to verify).
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/bpf/20230613223533.3689589-4-andrii@kernel.org
Currently find_and_alloc_map() performs two separate functions: some
argument sanity checking and partial map creation workflow hanling.
Neither of those functions are self-sufficient and are augmented by
further checks and initialization logic in the caller (map_create()
function). So unify all the sanity checks, permission checks, and
creation and initialization logic in one linear piece of code in
map_create() instead. This also make it easier to further enhance
permission checks and keep them located in one place.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/bpf/20230613223533.3689589-3-andrii@kernel.org
Make each bpf() syscall command a bit more self-contained, making it
easier to further enhance it. We move sysctl_unprivileged_bpf_disabled
handling down to map_create() and bpf_prog_load(), two special commands
in this regard.
Also swap the order of checks, calling bpf_capable() only if
sysctl_unprivileged_bpf_disabled is true, avoiding unnecessary audit
messages.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/bpf/20230613223533.3689589-2-andrii@kernel.org
Commit c78f261e5dcb ("posix-timers: Clarify posix_timer_fn() comments")
turns an ifdef CONFIG_HIGH_RES_TIMERS into an conditional on
"IS_ENABLED(CONFIG_HIGHRES_TIMERS)"; note that the new conditional refers
to "HIGHRES_TIMERS" not "HIGH_RES_TIMERS" as before.
Fix this typo introduced in that refactoring.
Fixes: c78f261e5dcb ("posix-timers: Clarify posix_timer_fn() comments")
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20230609094643.26253-1-lukas.bulwahn@gmail.com
release_posix_timers() is called for cleaning up both hashed and unhashed
timers. The cases are differentiated by an argument and the usage is
hideous.
Seperate the actual free path out and use it for unhashed timers. Provide a
function for hashed timers.
No functional change.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/r/20230425183313.301432503@linutronix.de
Technically it's not required to set k_itimer::it_signal to NULL on exit()
because there is no other thread anymore which could lookup the timer
concurrently.
Set it to NULL for consistency sake and add a comment to that effect.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/r/20230425183313.196462644@linutronix.de
posix_timer_add() tries to allocate a posix timer ID by starting from the
cached ID which was stored by the last successful allocation.
This is done in a loop searching the ID space for a free slot one by
one. The loop has to terminate when the search wrapped around to the
starting point.
But that's racy vs. establishing the starting point. That is read out
lockless, which leads to the following problem:
CPU0 CPU1
posix_timer_add()
start = sig->posix_timer_id;
lock(hash_lock);
... posix_timer_add()
if (++sig->posix_timer_id < 0)
start = sig->posix_timer_id;
sig->posix_timer_id = 0;
So CPU1 can observe a negative start value, i.e. -1, and the loop break
never happens because the condition can never be true:
if (sig->posix_timer_id == start)
break;
While this is unlikely to ever turn into an endless loop as the ID space is
huge (INT_MAX), the racy read of the start value caught the attention of
KCSAN and Dmitry unearthed that incorrectness.
Rewrite it so that all id operations are under the hash lock.
Reported-by: syzbot+5c54bd3eb218bb595aa9@syzkaller.appspotmail.com
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/r/87bkhzdn6g.ffs@tglx
itimer_delete() has a retry loop when the timer is concurrently expired. On
non-RT kernels this just spin-waits until the timer callback has completed,
except for posix CPU timers which have HAVE_POSIX_CPU_TIMERS_TASK_WORK
enabled.
In that case and on RT kernels the existing task could live lock when
preempting the task which does the timer delivery.
Replace spin_unlock() with an invocation of timer_wait_running() to handle
it the same way as the other retry loops in the posix timer code.
Fixes: ec8f954a40 ("posix-timers: Use a callback for cancel synchronization on PREEMPT_RT")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/r/87v8g7c50d.ffs@tglx
irq_domain_debugfs_init() is defined in irqdomain.c, but the
declaration is in a header that is not included here:
kernel/irq/irqdomain.c:1965:13: error: no previous prototype for 'irq_domain_debugfs_init' [-Werror=missing-prototypes]
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20230516200432.554240-1-arnd@kernel.org
After commit 8ad075c2eb ("sched: Async unthrottling for cfs
bandwidth"), we may update the rq clock multiple times in the loop of
__cfsb_csd_unthrottle().
A prior (although less common) instance of this problem exists in
unthrottle_offline_cfs_rqs().
Cure both by ensuring update_rq_clock() is called before the loop and
setting RQCF_ACT_SKIP during the loop, to supress further updates.
The alternative would be pulling update_rq_clock() out of
unthrottle_cfs_rq(), but that gives an even bigger mess.
Fixes: 8ad075c2eb ("sched: Async unthrottling for cfs bandwidth")
Reviewed-By: Ben Segall <bsegall@google.com>
Suggested-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20230613082012.49615-4-jiahao.os@bytedance.com
There is a double update_rq_clock() invocation:
__balance_push_cpu_stop()
update_rq_clock()
__migrate_task()
update_rq_clock()
Sadly select_fallback_rq() also needs update_rq_clock() for
__do_set_cpus_allowed(), it is not possible to remove the update from
__balance_push_cpu_stop(). So remove it from __migrate_task() and
ensure all callers of this function call update_rq_clock() prior to
calling it.
Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20230613082012.49615-3-jiahao.os@bytedance.com
When using a cpufreq governor that uses
cpufreq_add_update_util_hook(), it is possible to trigger a missing
update_rq_clock() warning for the CPU hotplug path:
rq_attach_root()
set_rq_offline()
rq_offline_rt()
__disable_runtime()
sched_rt_rq_enqueue()
enqueue_top_rt_rq()
cpufreq_update_util()
data->func(data, rq_clock(rq), flags)
Move update_rq_clock() from sched_cpu_deactivate() (one of it's
callers) into set_rq_offline() such that it covers all
set_rq_offline() usage.
Additionally change rq_attach_root() to use rq_lock_irqsave() so that
it will properly manage the runqueue clock flags.
Suggested-by: Ben Segall <bsegall@google.com>
Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20230613082012.49615-2-jiahao.os@bytedance.com
According to the GRUB[1] rule, the runtime is depreciated as:
"dq = -max{u, (1 - Uinact - Uextra)} dt" (1)
To guarantee that deadline tasks doesn't starve lower class tasks,
we do not allocate the full bandwidth of the cpu to deadline tasks.
Maximum bandwidth usable by deadline tasks is denoted by "Umax".
Considering Umax, equation (1) becomes:
"dq = -(max{u, (Umax - Uinact - Uextra)} / Umax) dt" (2)
Current implementation has a minor bug in equation (2), which this
patch fixes.
The reclamation logic is verified by a sample program which creates
multiple deadline threads and observing their utilization. The tests
were run on an isolated cpu(isolcpus=3) on a 4 cpu system.
Tests on 6.3.0
==============
RUN 1: runtime=7ms, deadline=period=10ms, RT capacity = 95%
TID[693]: RECLAIM=1, (r=7ms, d=10ms, p=10ms), Util: 93.33
TID[693]: RECLAIM=1, (r=7ms, d=10ms, p=10ms), Util: 93.35
RUN 2: runtime=1ms, deadline=period=100ms, RT capacity = 95%
TID[708]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 16.69
TID[708]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 16.69
RUN 3: 2 tasks
Task 1: runtime=1ms, deadline=period=10ms
Task 2: runtime=1ms, deadline=period=100ms
TID[631]: RECLAIM=1, (r=1ms, d=10ms, p=10ms), Util: 62.67
TID[632]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 6.37
TID[631]: RECLAIM=1, (r=1ms, d=10ms, p=10ms), Util: 62.38
TID[632]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 6.23
As seen above, the reclamation doesn't reclaim the maximum allowed
bandwidth and as the bandwidth of tasks gets smaller, the reclaimed
bandwidth also comes down.
Tests with this patch applied
=============================
RUN 1: runtime=7ms, deadline=period=10ms, RT capacity = 95%
TID[608]: RECLAIM=1, (r=7ms, d=10ms, p=10ms), Util: 95.19
TID[608]: RECLAIM=1, (r=7ms, d=10ms, p=10ms), Util: 95.16
RUN 2: runtime=1ms, deadline=period=100ms, RT capacity = 95%
TID[616]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 95.27
TID[616]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 95.21
RUN 3: 2 tasks
Task 1: runtime=1ms, deadline=period=10ms
Task 2: runtime=1ms, deadline=period=100ms
TID[620]: RECLAIM=1, (r=1ms, d=10ms, p=10ms), Util: 86.64
TID[621]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 8.66
TID[620]: RECLAIM=1, (r=1ms, d=10ms, p=10ms), Util: 86.45
TID[621]: RECLAIM=1, (r=1ms, d=100ms, p=100ms), Util: 8.73
Running tasks on all cpus allowing for migration also showed that
the utilization is reclaimed to the maximum. Running 10 tasks on
3 cpus SCHED_FLAG_RECLAIM - top shows:
%Cpu0 : 94.6 us, 0.0 sy, 0.0 ni, 5.4 id, 0.0 wa
%Cpu1 : 95.2 us, 0.0 sy, 0.0 ni, 4.8 id, 0.0 wa
%Cpu2 : 95.8 us, 0.0 sy, 0.0 ni, 4.2 id, 0.0 wa
[1]: Abeni, Luca & Lipari, Giuseppe & Parri, Andrea & Sun, Youcheng.
(2015). Parallel and sequential reclaiming in multicore
real-time global scheduling.
Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Daniel Bristot de Oliveira <bristot@kernel.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lore.kernel.org/r/20230530135526.2385378-1-vineeth@bitbyteword.org
Add a tracepoint for when a CSD is queued to a remote CPU's
call_single_queue. This allows finding exactly which CPU queued a given CSD
when looking at a csd_function_{entry,exit} event, and also enables us to
accurately measure IPI delivery time with e.g. a synthetic event:
$ echo 'hist:keys=cpu,csd.hex:ts=common_timestamp.usecs' >\
/sys/kernel/tracing/events/smp/csd_queue_cpu/trigger
$ echo 'csd_latency unsigned int dst_cpu; unsigned long csd; u64 time' >\
/sys/kernel/tracing/synthetic_events
$ echo \
'hist:keys=common_cpu,csd.hex:'\
'time=common_timestamp.usecs-$ts:'\
'onmatch(smp.csd_queue_cpu).trace(csd_latency,common_cpu,csd,$time)' >\
/sys/kernel/tracing/events/smp/csd_function_entry/trigger
$ trace-cmd record -e 'synthetic:csd_latency' hackbench
$ trace-cmd report
<...>-467 [001] 21.824263: csd_queue_cpu: cpu=0 callsite=try_to_wake_up+0x2ea func=sched_ttwu_pending csd=0xffff8880076148b8
<...>-467 [001] 21.824280: ipi_send_cpu: cpu=0 callsite=try_to_wake_up+0x2ea callback=generic_smp_call_function_single_interrupt+0x0
<...>-489 [000] 21.824299: csd_function_entry: func=sched_ttwu_pending csd=0xffff8880076148b8
<...>-489 [000] 21.824320: csd_latency: dst_cpu=0, csd=18446612682193848504, time=36
Suggested-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Leonardo Bras <leobras@redhat.com>
Tested-and-reviewed-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20230615065944.188876-7-leobras@redhat.com
The recently added ipi_send_{cpu,cpumask} tracepoints allow finding sources
of IPIs targeting CPUs running latency-sensitive applications.
For NOHZ_FULL CPUs, all IPIs are interference, and those tracepoints are
sufficient to find them and work on getting rid of them. In some setups
however, not *all* IPIs are to be suppressed, but long-running IPI
callbacks can still be problematic.
Add a pair of tracepoints to mark the start and end of processing a CSD IPI
callback, similar to what exists for softirq, workqueue or timer callbacks.
Signed-off-by: Leonardo Bras <leobras@redhat.com>
Tested-and-reviewed-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20230615065944.188876-5-leobras@redhat.com
