[ Upstream commit ad789f84c9a145f8a18744c0387cec22ec51651e ]
The handling of sysrq key can be activated by echoing the key to
/proc/sysrq-trigger or via the magic key sequence typed into a terminal
that is connected to the system in some way (serial, USB or other mean).
In the former case, the handling is done in a user context. In the
latter case, it is likely to be in an interrupt context.
Currently in print_cpu() of kernel/sched/debug.c, sched_debug_lock is
taken with interrupt disabled for the whole duration of the calls to
print_*_stats() and print_rq() which could last for the quite some time
if the information dump happens on the serial console.
If the system has many cpus and the sched_debug_lock is somehow busy
(e.g. parallel sysrq-t), the system may hit a hard lockup panic
depending on the actually serial console implementation of the
system.
The purpose of sched_debug_lock is to serialize the use of the global
cgroup_path[] buffer in print_cpu(). The rests of the printk calls don't
need serialization from sched_debug_lock.
Calling printk() with interrupt disabled can still be problematic if
multiple instances are running. Allocating a stack buffer of PATH_MAX
bytes is not feasible because of the limited size of the kernel stack.
The solution implemented in this patch is to allow only one caller at a
time to use the full size group_path[], while other simultaneous callers
will have to use shorter stack buffers with the possibility of path
name truncation. A "..." suffix will be printed if truncation may have
happened. The cgroup path name is provided for informational purpose
only, so occasional path name truncation should not be a big problem.
Fixes: efe25c2c7b ("sched: Reinstate group names in /proc/sched_debug")
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20210415195426.6677-1-longman@redhat.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 39a2a6eb5c9b66ea7c8055026303b3aa681b49a5 ]
Syzbot reported a handful of occurrences where an sd->nr_balance_failed can
grow to much higher values than one would expect.
A successful load_balance() resets it to 0; a failed one increments
it. Once it gets to sd->cache_nice_tries + 3, this *should* trigger an
active balance, which will either set it to sd->cache_nice_tries+1 or reset
it to 0. However, in case the to-be-active-balanced task is not allowed to
run on env->dst_cpu, then the increment is done without any further
modification.
This could then be repeated ad nauseam, and would explain the absurdly high
values reported by syzbot (86, 149). VincentG noted there is value in
letting sd->cache_nice_tries grow, so the shift itself should be
fixed. That means preventing:
"""
If the value of the right operand is negative or is greater than or equal
to the width of the promoted left operand, the behavior is undefined.
"""
Thus we need to cap the shift exponent to
BITS_PER_TYPE(typeof(lefthand)) - 1.
I had a look around for other similar cases via coccinelle:
@expr@
position pos;
expression E1;
expression E2;
@@
(
E1 >> E2@pos
|
E1 >> E2@pos
)
@cst depends on expr@
position pos;
expression expr.E1;
constant cst;
@@
(
E1 >> cst@pos
|
E1 << cst@pos
)
@script:python depends on !cst@
pos << expr.pos;
exp << expr.E2;
@@
# Dirty hack to ignore constexpr
if exp.upper() != exp:
coccilib.report.print_report(pos[0], "Possible UB shift here")
The only other match in kernel/sched is rq_clock_thermal() which employs
sched_thermal_decay_shift, and that exponent is already capped to 10, so
that one is fine.
Fixes: 5a7f555904 ("sched/fair: Relax constraint on task's load during load balance")
Reported-by: syzbot+d7581744d5fd27c9fbe1@syzkaller.appspotmail.com
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: http://lore.kernel.org/r/000000000000ffac1205b9a2112f@google.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 9d10a13d1e4c349b76f1c675a874a7f981d6d3b4 ]
psi_group_cpu->tasks, represented by the unsigned int, stores the
number of tasks that could be stalled on a psi resource(io/mem/cpu).
Decrementing these counters at zero leads to wrapping which further
leads to the psi_group_cpu->state_mask is being set with the
respective pressure state. This could result into the unnecessary time
sampling for the pressure state thus cause the spurious psi events.
This can further lead to wrong actions being taken at the user land
based on these psi events.
Though psi_bug is set under these conditions but that just for debug
purpose. Fix it by decrementing the ->tasks count only when it is
non-zero.
Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: https://lkml.kernel.org/r/1618585336-37219-1-git-send-email-charante@codeaurora.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0c2de3f054a59f15e01804b75a04355c48de628c ]
The current sched_slice() seems to have issues; there's two possible
things that could be improved:
- the 'nr_running' used for __sched_period() is daft when cgroups are
considered. Using the RQ wide h_nr_running seems like a much more
consistent number.
- (esp) cgroups can slice it real fine, which makes for easy
over-scheduling, ensure min_gran is what the name says.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/20210412102001.611897312@infradead.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 9bcb959d05eeb564dfc9cac13a59843a4fb2edf2 ]
During load balance, LBF_SOME_PINNED will be set if any candidate task
cannot be detached due to CPU affinity constraints. This can result in
setting env->sd->parent->sgc->group_imbalance, which can lead to a group
being classified as group_imbalanced (rather than any of the other, lower
group_type) when balancing at a higher level.
In workloads involving a single task per CPU, LBF_SOME_PINNED can often be
set due to per-CPU kthreads being the only other runnable tasks on any
given rq. This results in changing the group classification during
load-balance at higher levels when in reality there is nothing that can be
done for this affinity constraint: per-CPU kthreads, as the name implies,
don't get to move around (modulo hotplug shenanigans).
It's not as clear for userspace tasks - a task could be in an N-CPU cpuset
with N-1 offline CPUs, making it an "accidental" per-CPU task rather than
an intended one. KTHREAD_IS_PER_CPU gives us an indisputable signal which
we can leverage here to not set LBF_SOME_PINNED.
Note that the aforementioned classification to group_imbalance (when
nothing can be done) is especially problematic on big.LITTLE systems, which
have a topology the likes of:
DIE [ ]
MC [ ][ ]
0 1 2 3
L L B B
arch_scale_cpu_capacity(L) < arch_scale_cpu_capacity(B)
Here, setting LBF_SOME_PINNED due to a per-CPU kthread when balancing at MC
level on CPUs [0-1] will subsequently prevent CPUs [2-3] from classifying
the [0-1] group as group_misfit_task when balancing at DIE level. Thus, if
CPUs [0-1] are running CPU-bound (misfit) tasks, ill-timed per-CPU kthreads
can significantly delay the upgmigration of said misfit tasks. Systems
relying on ASYM_PACKING are likely to face similar issues.
Signed-off-by: Lingutla Chandrasekhar <clingutla@codeaurora.org>
[Use kthread_is_per_cpu() rather than p->nr_cpus_allowed]
[Reword changelog]
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20210407220628.3798191-2-valentin.schneider@arm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b89997aa88f0b07d8a6414c908af75062103b8c9 ]
Being called for each dequeue, util_est reduces the number of its updates
by filtering out when the EWMA signal is different from the task util_avg
by less than 1%. It is a problem for a sudden util_avg ramp-up. Due to the
decay from a previous high util_avg, EWMA might now be close enough to
the new util_avg. No update would then happen while it would leave
ue.enqueued with an out-of-date value.
Taking into consideration the two util_est members, EWMA and enqueued for
the filtering, ensures, for both, an up-to-date value.
This is for now an issue only for the trace probe that might return the
stale value. Functional-wise, it isn't a problem, as the value is always
accessed through max(enqueued, ewma).
This problem has been observed using LISA's UtilConvergence:test_means on
the sd845c board.
No regression observed with Hackbench on sd845c and Perf-bench sched pipe
on hikey/hikey960.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20210225165820.1377125-1-vincent.donnefort@arm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 156ec6f42b8d300dbbf382738ff35c8bad8f4c3a ]
Hung tasks and RCU stall cases were reported on systems which were not
100% busy. Investigation of such unexpected cases (no sign of potential
starvation caused by tasks hogging the system) pointed out that the
periodic sched tick timer wasn't serviced anymore after a certain point
and that caused all machinery that depends on it (timers, RCU, etc.) to
stop working as well. This issues was however only reproducible if
HRTICK was enabled.
Looking at core dumps it was found that the rbtree of the hrtimer base
used also for the hrtick was corrupted (i.e. next as seen from the base
root and actual leftmost obtained by traversing the tree are different).
Same base is also used for periodic tick hrtimer, which might get "lost"
if the rbtree gets corrupted.
Much alike what described in commit 1f71addd34 ("tick/sched: Do not
mess with an enqueued hrtimer") there is a race window between
hrtimer_set_expires() in hrtick_start and hrtimer_start_expires() in
__hrtick_restart() in which the former might be operating on an already
queued hrtick hrtimer, which might lead to corruption of the base.
Use hrtick_start() (which removes the timer before enqueuing it back) to
ensure hrtick hrtimer reprogramming is entirely guarded by the base
lock, so that no race conditions can occur.
Signed-off-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>
Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20210208073554.14629-2-juri.lelli@redhat.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 43789ef3f7d61aa7bed0cb2764e588fc990c30ef upstream.
Entering RCU idle mode may cause a deferred wake up of an RCU NOCB_GP
kthread (rcuog) to be serviced.
Usually a local wake up happening while running the idle task is handled
in one of the need_resched() checks carefully placed within the idle
loop that can break to the scheduler.
Unfortunately the call to rcu_idle_enter() is already beyond the last
generic need_resched() check and we may halt the CPU with a resched
request unhandled, leaving the task hanging.
Fix this with splitting the rcuog wakeup handling from rcu_idle_enter()
and place it before the last generic need_resched() check in the idle
loop. It is then assumed that no call to call_rcu() will be performed
after that in the idle loop until the CPU is put in low power mode.
Fixes: 96d3fd0d31 (rcu: Break call_rcu() deadlock involving scheduler and perf)
Reported-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20210131230548.32970-3-frederic@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 0ae78eec8aa64e645866e75005162603a77a0f49 ]
If the task is pinned to a cpu, setting the misfit status means that
we'll unnecessarily continuously attempt to migrate the task but fail.
This continuous failure will cause the balance_interval to increase to
a high value, and eventually cause unnecessary significant delays in
balancing the system when real imbalance happens.
Caught while testing uclamp where rt-app calibration loop was pinned to
cpu 0, shortly after which we spawn another task with high util_clamp
value. The task was failing to migrate after over 40ms of runtime due to
balance_interval unnecessary expanded to a very high value from the
calibration loop.
Not done here, but it could be useful to extend the check for pinning to
verify that the affinity of the task has a cpu that fits. We could end
up in a similar situation otherwise.
Fixes: 3b1baa6496 ("sched/fair: Add 'group_misfit_task' load-balance type")
Signed-off-by: Qais Yousef <qais.yousef@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Quentin Perret <qperret@google.com>
Acked-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/20210119120755.2425264-1-qais.yousef@arm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 8c1f560c1ea3f19e22ba356f62680d9d449c9ec2 ]
CPU (root cfs_rq) estimated utilization (util_est) is currently used in
dequeue_task_fair() to drive frequency selection before it is updated.
with:
CPU_util : rq->cfs.avg.util_avg
CPU_util_est : rq->cfs.avg.util_est
CPU_utilization : max(CPU_util, CPU_util_est)
task_util : p->se.avg.util_avg
task_util_est : p->se.avg.util_est
dequeue_task_fair():
/* (1) CPU_util and task_util update + inform schedutil about
CPU_utilization changes */
for_each_sched_entity() /* 2 loops */
(dequeue_entity() ->) update_load_avg() -> cfs_rq_util_change()
-> cpufreq_update_util() ->...-> sugov_update_[shared\|single]
-> sugov_get_util() -> cpu_util_cfs()
/* (2) CPU_util_est and task_util_est update */
util_est_dequeue()
cpu_util_cfs() uses CPU_utilization which could lead to a false (too
high) utilization value for schedutil in task ramp-down or ramp-up
scenarios during task dequeue.
To mitigate the issue split the util_est update (2) into:
(A) CPU_util_est update in util_est_dequeue()
(B) task_util_est update in util_est_update()
Place (A) before (1) and keep (B) where (2) is. The latter is necessary
since (B) relies on task_util update in (1).
Fixes: 7f65ea42eb ("sched/fair: Add util_est on top of PELT")
Signed-off-by: Xuewen Yan <xuewen.yan@unisoc.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/1608283672-18240-1-git-send-email-xuewen.yan94@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 345a957fcc95630bf5535d7668a59ed983eb49a7 ]
do_sched_yield() invokes schedule() with interrupts disabled which is
not allowed. This goes back to the pre git era to commit a6efb709806c
("[PATCH] irqlock patch 2.5.27-H6") in the history tree.
Reenable interrupts and remove the misleading comment which "explains" it.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/87r1pt7y5c.fsf@nanos.tec.linutronix.de
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a57415f5d1e43c3a5c5d412cd85e2792d7ed9b11 ]
When change sched_rt_{runtime, period}_us, we validate that the new
settings should at least accommodate the currently allocated -dl
bandwidth:
sched_rt_handler()
--> sched_dl_bandwidth_validate()
{
new_bw = global_rt_runtime()/global_rt_period();
for_each_possible_cpu(cpu) {
dl_b = dl_bw_of(cpu);
if (new_bw < dl_b->total_bw) <-------
ret = -EBUSY;
}
}
But under CONFIG_SMP, dl_bw is per root domain , but not per CPU,
dl_b->total_bw is the allocated bandwidth of the whole root domain.
Instead, we should compare dl_b->total_bw against "cpus*new_bw",
where 'cpus' is the number of CPUs of the root domain.
Also, below annotation(in kernel/sched/sched.h) implied implementation
only appeared in SCHED_DEADLINE v2[1], then deadline scheduler kept
evolving till got merged(v9), but the annotation remains unchanged,
meaningless and misleading, update it.
* With respect to SMP, the bandwidth is given on a per-CPU basis,
* meaning that:
* - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
* - dl_total_bw array contains, in the i-eth element, the currently
* allocated bandwidth on the i-eth CPU.
[1]: https://lore.kernel.org/lkml/1267385230.13676.101.camel@Palantir/
Fixes: 332ac17ef5 ("sched/deadline: Add bandwidth management for SCHED_DEADLINE tasks")
Signed-off-by: Peng Liu <iwtbavbm@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/db6bbda316048cda7a1bbc9571defde193a8d67e.1602171061.git.iwtbavbm@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
membarrier()'s MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE is documented as
syncing the core on all sibling threads but not necessarily the calling
thread. This behavior is fundamentally buggy and cannot be used safely.
Suppose a user program has two threads. Thread A is on CPU 0 and thread B
is on CPU 1. Thread A modifies some text and calls
membarrier(MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE).
Then thread B executes the modified code. If, at any point after
membarrier() decides which CPUs to target, thread A could be preempted and
replaced by thread B on CPU 0. This could even happen on exit from the
membarrier() syscall. If this happens, thread B will end up running on CPU
0 without having synced.
In principle, this could be fixed by arranging for the scheduler to issue
sync_core_before_usermode() whenever switching between two threads in the
same mm if there is any possibility of a concurrent membarrier() call, but
this would have considerable overhead. Instead, make membarrier() sync the
calling CPU as well.
As an optimization, this avoids an extra smp_mb() in the default
barrier-only mode and an extra rseq preempt on the caller.
Fixes: 70216e18e5 ("membarrier: Provide core serializing command, *_SYNC_CORE")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Link: https://lore.kernel.org/r/250ded637696d490c69bef1877148db86066881c.1607058304.git.luto@kernel.org
membarrier() does not explicitly sync_core() remote CPUs; instead, it
relies on the assumption that an IPI will result in a core sync. On x86,
this may be true in practice, but it's not architecturally reliable. In
particular, the SDM and APM do not appear to guarantee that interrupt
delivery is serializing. While IRET does serialize, IPI return can
schedule, thereby switching to another task in the same mm that was
sleeping in a syscall. The new task could then SYSRET back to usermode
without ever executing IRET.
Make this more robust by explicitly calling sync_core_before_usermode()
on remote cores. (This also helps people who search the kernel tree for
instances of sync_core() and sync_core_before_usermode() -- one might be
surprised that the core membarrier code doesn't currently show up in a
such a search.)
Fixes: 70216e18e5 ("membarrier: Provide core serializing command, *_SYNC_CORE")
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/776b448d5f7bd6b12690707f5ed67bcda7f1d427.1607058304.git.luto@kernel.org
Pull locking fixes from Thomas Gleixner:
"Two more places which invoke tracing from RCU disabled regions in the
idle path.
Similar to the entry path the low level idle functions have to be
non-instrumentable"
* tag 'locking-urgent-2020-11-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
intel_idle: Fix intel_idle() vs tracing
sched/idle: Fix arch_cpu_idle() vs tracing
We call arch_cpu_idle() with RCU disabled, but then use
local_irq_{en,dis}able(), which invokes tracing, which relies on RCU.
Switch all arch_cpu_idle() implementations to use
raw_local_irq_{en,dis}able() and carefully manage the
lockdep,rcu,tracing state like we do in entry.
(XXX: we really should change arch_cpu_idle() to not return with
interrupts enabled)
Reported-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Link: https://lkml.kernel.org/r/20201120114925.594122626@infradead.org
Pull scheduler fixes from Thomas Gleixner:
"A couple of scheduler fixes:
- Make the conditional update of the overutilized state work
correctly by caching the relevant flags state before overwriting
them and checking them afterwards.
- Fix a data race in the wakeup path which caused loadavg on ARM64
platforms to become a random number generator.
- Fix the ordering of the iowaiter accounting operations so it can't
be decremented before it is incremented.
- Fix a bug in the deadline scheduler vs. priority inheritance when a
non-deadline task A has inherited the parameters of a deadline task
B and then blocks on a non-deadline task C.
The second inheritance step used the static deadline parameters of
task A, which are usually 0, instead of further propagating task
B's parameters. The zero initialized parameters trigger a bug in
the deadline scheduler"
* tag 'sched-urgent-2020-11-22' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/deadline: Fix priority inheritance with multiple scheduling classes
sched: Fix rq->nr_iowait ordering
sched: Fix data-race in wakeup
sched/fair: Fix overutilized update in enqueue_task_fair()
Glenn reported that "an application [he developed produces] a BUG in
deadline.c when a SCHED_DEADLINE task contends with CFS tasks on nested
PTHREAD_PRIO_INHERIT mutexes. I believe the bug is triggered when a CFS
task that was boosted by a SCHED_DEADLINE task boosts another CFS task
(nested priority inheritance).
------------[ cut here ]------------
kernel BUG at kernel/sched/deadline.c:1462!
invalid opcode: 0000 [#1] PREEMPT SMP
CPU: 12 PID: 19171 Comm: dl_boost_bug Tainted: ...
Hardware name: ...
RIP: 0010:enqueue_task_dl+0x335/0x910
Code: ...
RSP: 0018:ffffc9000c2bbc68 EFLAGS: 00010002
RAX: 0000000000000009 RBX: ffff888c0af94c00 RCX: ffffffff81e12500
RDX: 000000000000002e RSI: ffff888c0af94c00 RDI: ffff888c10b22600
RBP: ffffc9000c2bbd08 R08: 0000000000000009 R09: 0000000000000078
R10: ffffffff81e12440 R11: ffffffff81e1236c R12: ffff888bc8932600
R13: ffff888c0af94eb8 R14: ffff888c10b22600 R15: ffff888bc8932600
FS: 00007fa58ac55700(0000) GS:ffff888c10b00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fa58b523230 CR3: 0000000bf44ab003 CR4: 00000000007606e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
? intel_pstate_update_util_hwp+0x13/0x170
rt_mutex_setprio+0x1cc/0x4b0
task_blocks_on_rt_mutex+0x225/0x260
rt_spin_lock_slowlock_locked+0xab/0x2d0
rt_spin_lock_slowlock+0x50/0x80
hrtimer_grab_expiry_lock+0x20/0x30
hrtimer_cancel+0x13/0x30
do_nanosleep+0xa0/0x150
hrtimer_nanosleep+0xe1/0x230
? __hrtimer_init_sleeper+0x60/0x60
__x64_sys_nanosleep+0x8d/0xa0
do_syscall_64+0x4a/0x100
entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x7fa58b52330d
...
---[ end trace 0000000000000002 ]—
He also provided a simple reproducer creating the situation below:
So the execution order of locking steps are the following
(N1 and N2 are non-deadline tasks. D1 is a deadline task. M1 and M2
are mutexes that are enabled * with priority inheritance.)
Time moves forward as this timeline goes down:
N1 N2 D1
| | |
| | |
Lock(M1) | |
| | |
| Lock(M2) |
| | |
| | Lock(M2)
| | |
| Lock(M1) |
| (!!bug triggered!) |
Daniel reported a similar situation as well, by just letting ksoftirqd
run with DEADLINE (and eventually block on a mutex).
Problem is that boosted entities (Priority Inheritance) use static
DEADLINE parameters of the top priority waiter. However, there might be
cases where top waiter could be a non-DEADLINE entity that is currently
boosted by a DEADLINE entity from a different lock chain (i.e., nested
priority chains involving entities of non-DEADLINE classes). In this
case, top waiter static DEADLINE parameters could be null (initialized
to 0 at fork()) and replenish_dl_entity() would hit a BUG().
Fix this by keeping track of the original donor and using its parameters
when a task is boosted.
Reported-by: Glenn Elliott <glenn@aurora.tech>
Reported-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Link: https://lkml.kernel.org/r/20201117061432.517340-1-juri.lelli@redhat.com
schedule() ttwu()
deactivate_task(); if (p->on_rq && ...) // false
atomic_dec(&task_rq(p)->nr_iowait);
if (prev->in_iowait)
atomic_inc(&rq->nr_iowait);
Allows nr_iowait to be decremented before it gets incremented,
resulting in more dodgy IO-wait numbers than usual.
Note that because we can now do ttwu_queue_wakelist() before
p->on_cpu==0, we lose the natural ordering and have to further delay
the decrement.
Fixes: c6e7bd7afa ("sched/core: Optimize ttwu() spinning on p->on_cpu")
Reported-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Link: https://lkml.kernel.org/r/20201117093829.GD3121429@hirez.programming.kicks-ass.net
Pull scheduler fixes from Thomas Gleixner:
"A set of scheduler fixes:
- Address a load balancer regression by making the load balancer use
the same logic as the wakeup path to spread tasks in the LLC domain
- Prefer the CPU on which a task run last over the local CPU in the
fast wakeup path for asymmetric CPU capacity systems to align with
the symmetric case. This ensures more locality and prevents massive
migration overhead on those asymetric systems
- Fix a memory corruption bug in the scheduler debug code caused by
handing a modified buffer pointer to kfree()"
* tag 'sched-urgent-2020-11-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/debug: Fix memory corruption caused by multiple small reads of flags
sched/fair: Prefer prev cpu in asymmetric wakeup path
sched/fair: Ensure tasks spreading in LLC during LB
Reading /proc/sys/kernel/sched_domain/cpu*/domain0/flags mutliple times
with small reads causes oopses with slub corruption issues because the kfree is
free'ing an offset from a previous allocation. Fix this by adding in a new
pointer 'buf' for the allocation and kfree and use the temporary pointer tmp
to handle memory copies of the buf offsets.
Fixes: 5b9f8ff7b3 ("sched/debug: Output SD flag names rather than their values")
Reported-by: Jeff Bastian <jbastian@redhat.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/20201029151103.373410-1-colin.king@canonical.com
During fast wakeup path, scheduler always check whether local or prev
cpus are good candidates for the task before looking for other cpus in
the domain. With commit b7a331615d ("sched/fair: Add asymmetric CPU
capacity wakeup scan") the heterogenous system gains a dedicated path
but doesn't try to reuse prev cpu whenever possible. If the previous
cpu is idle and belong to the LLC domain, we should check it 1st
before looking for another cpu because it stays one of the best
candidate and this also stabilizes task placement on the system.
This change aligns asymmetric path behavior with symmetric one and reduces
cases where the task migrates across all cpus of the sd_asym_cpucapacity
domains at wakeup.
This change does not impact normal EAS mode but only the overloaded case or
when EAS is not used.
- On hikey960 with performance governor (EAS disable)
./perf bench sched pipe -T -l 50000
mainline w/ patch
# migrations 999364 0
ops/sec 149313(+/-0.28%) 182587(+/- 0.40) +22%
- On hikey with performance governor
./perf bench sched pipe -T -l 50000
mainline w/ patch
# migrations 0 0
ops/sec 47721(+/-0.76%) 47899(+/- 0.56) +0.4%
According to test on hikey, the patch doesn't impact symmetric system
compared to current implementation (only tested on arm64)
Also read the uclamped value of task's utilization at most twice instead
instead each time we compare task's utilization with cpu's capacity.
Fixes: b7a331615d ("sched/fair: Add asymmetric CPU capacity wakeup scan")
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/20201029161824.26389-1-vincent.guittot@linaro.org
schbench shows latency increase for 95 percentile above since:
commit 0b0695f2b3 ("sched/fair: Rework load_balance()")
Align the behavior of the load balancer with the wake up path, which tries
to select an idle CPU which belongs to the LLC for a waking task.
calculate_imbalance() will use nr_running instead of the spare
capacity when CPUs share resources (ie cache) at the domain level. This
will ensure a better spread of tasks on idle CPUs.
Running schbench on a hikey (8cores arm64) shows the problem:
tip/sched/core :
schbench -m 2 -t 4 -s 10000 -c 1000000 -r 10
Latency percentiles (usec)
50.0th: 33
75.0th: 45
90.0th: 51
95.0th: 4152
*99.0th: 14288
99.5th: 14288
99.9th: 14288
min=0, max=14276
tip/sched/core + patch :
schbench -m 2 -t 4 -s 10000 -c 1000000 -r 10
Latency percentiles (usec)
50.0th: 34
75.0th: 47
90.0th: 52
95.0th: 78
*99.0th: 94
99.5th: 94
99.9th: 94
min=0, max=94
Fixes: 0b0695f2b3 ("sched/fair: Rework load_balance()")
Reported-by: Chris Mason <clm@fb.com>
Suggested-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rik van Riel <riel@surriel.com>
Tested-by: Rik van Riel <riel@surriel.com>
Link: https://lkml.kernel.org/r/20201102102457.28808-1-vincent.guittot@linaro.org
A new cpufreq governor flag will be added subsequently, so replace
the bool dynamic_switching fleid in struct cpufreq_governor with a
flags field and introduce CPUFREQ_GOV_DYNAMIC_SWITCHING to set for
the "dynamic switching" governors instead of it.
No intentional functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The cpufreq policy's frequency limits (min/max) can get changed at any
point of time, while schedutil is trying to update the next frequency.
Though the schedutil governor has necessary locking and support in place
to make sure we don't miss any of those updates, there is a corner case
where the governor will find that the CPU is already running at the
desired frequency and so may skip an update.
For example, consider that the CPU can run at 1 GHz, 1.2 GHz and 1.4 GHz
and is running at 1 GHz currently. Schedutil tries to update the
frequency to 1.2 GHz, during this time the policy limits get changed as
policy->min = 1.4 GHz. As schedutil (and cpufreq core) does clamp the
frequency at various instances, we will eventually set the frequency to
1.4 GHz, while we will save 1.2 GHz in sg_policy->next_freq.
Now lets say the policy limits get changed back at this time with
policy->min as 1 GHz. The next time schedutil is invoked by the
scheduler, we will reevaluate the next frequency (because
need_freq_update will get set due to limits change event) and lets say
we want to set the frequency to 1.2 GHz again. At this point
sugov_update_next_freq() will find the next_freq == current_freq and
will abort the update, while the CPU actually runs at 1.4 GHz.
Until now need_freq_update was used as a flag to indicate that the
policy's frequency limits have changed, and that we should consider the
new limits while reevaluating the next frequency.
This patch fixes the above mentioned issue by extending the purpose of
the need_freq_update flag. If this flag is set now, the schedutil
governor will not try to abort a frequency change even if next_freq ==
current_freq.
As similar behavior is required in the case of
CPUFREQ_NEED_UPDATE_LIMITS flag as well, need_freq_update will never be
set to false if that flag is set for the driver.
We also don't need to consider the need_freq_update flag in
sugov_update_single() anymore to handle the special case of busy CPU, as
we won't abort a frequency update anymore.
Reported-by: zhuguangqing <zhuguangqing@xiaomi.com>
Suggested-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Rearrange code to avoid a branch ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Because sugov_update_next_freq() may skip a frequency update even if
the need_freq_update flag has been set for the policy at hand, policy
limits updates may not take effect as expected.
For example, if the intel_pstate driver operates in the passive mode
with HWP enabled, it needs to update the HWP min and max limits when
the policy min and max limits change, respectively, but that may not
happen if the target frequency does not change along with the limit
at hand. In particular, if the policy min is changed first, causing
the target frequency to be adjusted to it, and the policy max limit
is changed later to the same value, the HWP max limit will not be
updated to follow it as expected, because the target frequency is
still equal to the policy min limit and it will not change until
that limit is updated.
To address this issue, modify get_next_freq() to let the driver
callback run if the CPUFREQ_NEED_UPDATE_LIMITS cpufreq driver flag
is set regardless of whether or not the new frequency to set is
equal to the previous one.
Fixes: f6ebbcf08f ("cpufreq: intel_pstate: Implement passive mode with HWP enabled")
Reported-by: Zhang Rui <rui.zhang@intel.com>
Tested-by: Zhang Rui <rui.zhang@intel.com>
Cc: 5.9+ <stable@vger.kernel.org> # 5.9+: 1c534352f4 cpufreq: Introduce CPUFREQ_NEED_UPDATE_LIMITS ...
Cc: 5.9+ <stable@vger.kernel.org> # 5.9+: a62f68f5ca cpufreq: Introduce cpufreq_driver_test_flags()
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pull scheduler fixes from Thomas Gleixner:
"Two scheduler fixes:
- A trivial build fix for sched_feat() to compile correctly with
CONFIG_JUMP_LABEL=n
- Replace a zero lenght array with a flexible array"
* tag 'sched-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/features: Fix !CONFIG_JUMP_LABEL case
sched: Replace zero-length array with flexible-array
Pull more power management updates from Rafael Wysocki:
"First of all, the adaptive voltage scaling (AVS) drivers go to new
platform-specific locations as planned (this part was reported to have
merge conflicts against the new arm-soc updates in linux-next).
In addition to that, there are some fixes (intel_idle, intel_pstate,
RAPL, acpi_cpufreq), the addition of on/off notifiers and idle state
accounting support to the generic power domains (genpd) code and some
janitorial changes all over.
Specifics:
- Move the AVS drivers to new platform-specific locations and get rid
of the drivers/power/avs directory (Ulf Hansson).
- Add on/off notifiers and idle state accounting support to the
generic power domains (genpd) framework (Ulf Hansson, Lina Iyer).
- Ulf will maintain the PM domain part of cpuidle-psci (Ulf Hansson).
- Make intel_idle disregard ACPI _CST if it cannot use the data
returned by that method (Mel Gorman).
- Modify intel_pstate to avoid leaving useless sysfs directory
structure behind if it cannot be registered (Chen Yu).
- Fix domain detection in the RAPL power capping driver and prevent
it from failing to enumerate the Psys RAPL domain (Zhang Rui).
- Allow acpi-cpufreq to use ACPI _PSD information with Family 19 and
later AMD chips (Wei Huang).
- Update the driver assumptions comment in intel_idle and fix a
kerneldoc comment in the runtime PM framework (Alexander Monakov,
Bean Huo).
- Avoid unnecessary resets of the cached frequency in the schedutil
cpufreq governor to reduce overhead (Wei Wang).
- Clean up the cpufreq core a bit (Viresh Kumar).
- Make assorted minor janitorial changes (Daniel Lezcano, Geert
Uytterhoeven, Hubert Jasudowicz, Tom Rix).
- Clean up and optimize the cpupower utility somewhat (Colin Ian
King, Martin Kaistra)"
* tag 'pm-5.10-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (23 commits)
PM: sleep: remove unreachable break
PM: AVS: Drop the avs directory and the corresponding Kconfig
PM: AVS: qcom-cpr: Move the driver to the qcom specific drivers
PM: runtime: Fix typo in pm_runtime_set_active() helper comment
PM: domains: Fix build error for genpd notifiers
powercap: Fix typo in Kconfig "Plance" -> "Plane"
cpufreq: schedutil: restore cached freq when next_f is not changed
acpi-cpufreq: Honor _PSD table setting on new AMD CPUs
PM: AVS: smartreflex Move driver to soc specific drivers
PM: AVS: rockchip-io: Move the driver to the rockchip specific drivers
PM: domains: enable domain idle state accounting
PM: domains: Add curly braces to delimit comment + statement block
PM: domains: Add support for PM domain on/off notifiers for genpd
powercap/intel_rapl: enumerate Psys RAPL domain together with package RAPL domain
powercap/intel_rapl: Fix domain detection
intel_idle: Ignore _CST if control cannot be taken from the platform
cpuidle: Remove pointless stub
intel_idle: mention assumption that WBINVD is not needed
MAINTAINERS: Add section for cpuidle-psci PM domain
cpufreq: intel_pstate: Delete intel_pstate sysfs if failed to register the driver
...
We have the raw cached freq to reduce the chance in calling cpufreq
driver where it could be costly in some arch/SoC.
Currently, the raw cached freq is reset in sugov_update_single() when
it avoids frequency reduction (which is not desirable sometimes), but
it is better to restore the previous value of it in that case,
because it may not change in the next cycle and it is not necessary
to change the CPU frequency then.
Adapted from https://android-review.googlesource.com/1352810/
Signed-off-by: Wei Wang <wvw@google.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Subject edit and changelog rewrite ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
A previous commit changed the notification mode from true/false to an
int, allowing notify-no, notify-yes, or signal-notify. This was
backwards compatible in the sense that any existing true/false user
would translate to either 0 (on notification sent) or 1, the latter
which mapped to TWA_RESUME. TWA_SIGNAL was assigned a value of 2.
Clean this up properly, and define a proper enum for the notification
mode. Now we have:
- TWA_NONE. This is 0, same as before the original change, meaning no
notification requested.
- TWA_RESUME. This is 1, same as before the original change, meaning
that we use TIF_NOTIFY_RESUME.
- TWA_SIGNAL. This uses TIF_SIGPENDING/JOBCTL_TASK_WORK for the
notification.
Clean up all the callers, switching their 0/1/false/true to using the
appropriate TWA_* mode for notifications.
Fixes: e91b481623 ("task_work: teach task_work_add() to do signal_wake_up()")
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit:
765cc3a4b2 ("sched/core: Optimize sched_feat() for !CONFIG_SCHED_DEBUG builds")
made sched features static for !CONFIG_SCHED_DEBUG configurations, but
overlooked the CONFIG_SCHED_DEBUG=y and !CONFIG_JUMP_LABEL cases.
For the latter echoing changes to /sys/kernel/debug/sched_features has
the nasty effect of effectively changing what sched_features reports,
but without actually changing the scheduler behaviour (since different
translation units get different sysctl_sched_features).
Fix CONFIG_SCHED_DEBUG=y and !CONFIG_JUMP_LABEL configurations by properly
restructuring ifdefs.
Fixes: 765cc3a4b2 ("sched/core: Optimize sched_feat() for !CONFIG_SCHED_DEBUG builds")
Co-developed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Patrick Bellasi <patrick.bellasi@matbug.net>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lore.kernel.org/r/20201013053114.160628-1-juri.lelli@redhat.com
In the following commit:
04f5c362ec: ("sched/fair: Replace zero-length array with flexible-array")
a zero-length array cpumask[0] has been replaced with cpumask[].
But there is still a cpumask[0] in 'struct sched_group_capacity'
which was missed.
The point of using [] instead of [0] is that with [] the compiler will
generate a build warning if it isn't the last member of a struct.
[ mingo: Rewrote the changelog. ]
Signed-off-by: zhuguangqing <zhuguangqing@xiaomi.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20201014140220.11384-1-zhuguangqing83@gmail.com
Pull power management updates from Rafael Wysocki:
"These rework the collection of cpufreq statistics to allow it to take
place if fast frequency switching is enabled in the governor, rework
the frequency invariance handling in the cpufreq core and drivers, add
new hardware support to a couple of cpufreq drivers, fix a number of
assorted issues and clean up the code all over.
Specifics:
- Rework cpufreq statistics collection to allow it to take place when
fast frequency switching is enabled in the governor (Viresh Kumar).
- Make the cpufreq core set the frequency scale on behalf of the
driver and update several cpufreq drivers accordingly (Ionela
Voinescu, Valentin Schneider).
- Add new hardware support to the STI and qcom cpufreq drivers and
improve them (Alain Volmat, Manivannan Sadhasivam).
- Fix multiple assorted issues in cpufreq drivers (Jon Hunter,
Krzysztof Kozlowski, Matthias Kaehlcke, Pali Rohár, Stephan
Gerhold, Viresh Kumar).
- Fix several assorted issues in the operating performance points
(OPP) framework (Stephan Gerhold, Viresh Kumar).
- Allow devfreq drivers to fetch devfreq instances by DT enumeration
instead of using explicit phandles and modify the devfreq core code
to support driver-specific devfreq DT bindings (Leonard Crestez,
Chanwoo Choi).
- Improve initial hardware resetting in the tegra30 devfreq driver
and clean up the tegra cpuidle driver (Dmitry Osipenko).
- Update the cpuidle core to collect state entry rejection statistics
and expose them via sysfs (Lina Iyer).
- Improve the ACPI _CST code handling diagnostics (Chen Yu).
- Update the PSCI cpuidle driver to allow the PM domain
initialization to occur in the OSI mode as well as in the PC mode
(Ulf Hansson).
- Rework the generic power domains (genpd) core code to allow domain
power off transition to be aborted in the absence of the "power
off" domain callback (Ulf Hansson).
- Fix two suspend-to-idle issues in the ACPI EC driver (Rafael
Wysocki).
- Fix the handling of timer_expires in the PM-runtime framework on
32-bit systems and the handling of device links in it (Grygorii
Strashko, Xiang Chen).
- Add IO requests batching support to the hibernate image saving and
reading code and drop a bogus get_gendisk() from there (Xiaoyi
Chen, Christoph Hellwig).
- Allow PCIe ports to be put into the D3cold power state if they are
power-manageable via ACPI (Lukas Wunner).
- Add missing header file include to a power capping driver (Pujin
Shi).
- Clean up the qcom-cpr AVS driver a bit (Liu Shixin).
- Kevin Hilman steps down as designated reviwer of adaptive voltage
scaling (AVS) drivers (Kevin Hilman)"
* tag 'pm-5.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (65 commits)
cpufreq: stats: Fix string format specifier mismatch
arm: disable frequency invariance for CONFIG_BL_SWITCHER
cpufreq,arm,arm64: restructure definitions of arch_set_freq_scale()
cpufreq: stats: Add memory barrier to store_reset()
cpufreq: schedutil: Simplify sugov_fast_switch()
ACPI: EC: PM: Drop ec_no_wakeup check from acpi_ec_dispatch_gpe()
ACPI: EC: PM: Flush EC work unconditionally after wakeup
PCI/ACPI: Whitelist hotplug ports for D3 if power managed by ACPI
PM: hibernate: remove the bogus call to get_gendisk() in software_resume()
cpufreq: Move traces and update to policy->cur to cpufreq core
cpufreq: stats: Enable stats for fast-switch as well
cpufreq: stats: Mark few conditionals with unlikely()
cpufreq: stats: Remove locking
cpufreq: stats: Defer stats update to cpufreq_stats_record_transition()
PM: domains: Allow to abort power off when no ->power_off() callback
PM: domains: Rename power state enums for genpd
PM / devfreq: tegra30: Improve initial hardware resetting
PM / devfreq: event: Change prototype of devfreq_event_get_edev_by_phandle function
PM / devfreq: Change prototype of devfreq_get_devfreq_by_phandle function
PM / devfreq: Add devfreq_get_devfreq_by_node function
...
Pull scheduler updates from Ingo Molnar:
- reorganize & clean up the SD* flags definitions and add a bunch of
sanity checks. These new checks caught quite a few bugs or at least
inconsistencies, resulting in another set of patches.
- rseq updates, add MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ
- add a new tracepoint to improve CPU capacity tracking
- improve overloaded SMP system load-balancing behavior
- tweak SMT balancing
- energy-aware scheduling updates
- NUMA balancing improvements
- deadline scheduler fixes and improvements
- CPU isolation fixes
- misc cleanups, simplifications and smaller optimizations
* tag 'sched-core-2020-10-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (42 commits)
sched/deadline: Unthrottle PI boosted threads while enqueuing
sched/debug: Add new tracepoint to track cpu_capacity
sched/fair: Tweak pick_next_entity()
rseq/selftests: Test MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ
rseq/selftests,x86_64: Add rseq_offset_deref_addv()
rseq/membarrier: Add MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ
sched/fair: Use dst group while checking imbalance for NUMA balancer
sched/fair: Reduce busy load balance interval
sched/fair: Minimize concurrent LBs between domain level
sched/fair: Reduce minimal imbalance threshold
sched/fair: Relax constraint on task's load during load balance
sched/fair: Remove the force parameter of update_tg_load_avg()
sched/fair: Fix wrong cpu selecting from isolated domain
sched: Remove unused inline function uclamp_bucket_base_value()
sched/rt: Disable RT_RUNTIME_SHARE by default
sched/deadline: Fix stale throttling on de-/boosted tasks
sched/numa: Use runnable_avg to classify node
sched/topology: Move sd_flag_debug out of #ifdef CONFIG_SYSCTL
MAINTAINERS: Add myself as SCHED_DEADLINE reviewer
sched/topology: Move SD_DEGENERATE_GROUPS_MASK out of linux/sched/topology.h
...
Drop a redundant local variable definition from sugov_fast_switch()
and rearrange the code in there to avoid the redundant logical
negation.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The cpufreq core handles the updates to policy->cur and recording of
cpufreq trace events for all the governors except schedutil's fast
switch case.
Move that as well to cpufreq core for consistency and readability.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
stress-ng has a test (stress-ng --cyclic) that creates a set of threads
under SCHED_DEADLINE with the following parameters:
dl_runtime = 10000 (10 us)
dl_deadline = 100000 (100 us)
dl_period = 100000 (100 us)
These parameters are very aggressive. When using a system without HRTICK
set, these threads can easily execute longer than the dl_runtime because
the throttling happens with 1/HZ resolution.
During the main part of the test, the system works just fine because
the workload does not try to run over the 10 us. The problem happens at
the end of the test, on the exit() path. During exit(), the threads need
to do some cleanups that require real-time mutex locks, mainly those
related to memory management, resulting in this scenario:
Note: locks are rt_mutexes...
------------------------------------------------------------------------
TASK A: TASK B: TASK C:
activation
activation
activation
lock(a): OK! lock(b): OK!
<overrun runtime>
lock(a)
-> block (task A owns it)
-> self notice/set throttled
+--< -> arm replenished timer
| switch-out
| lock(b)
| -> <C prio > B prio>
| -> boost TASK B
| unlock(a) switch-out
| -> handle lock a to B
| -> wakeup(B)
| -> B is throttled:
| -> do not enqueue
| switch-out
|
|
+---------------------> replenishment timer
-> TASK B is boosted:
-> do not enqueue
------------------------------------------------------------------------
BOOM: TASK B is runnable but !enqueued, holding TASK C: the system
crashes with hung task C.
This problem is avoided by removing the throttle state from the boosted
thread while boosting it (by TASK A in the example above), allowing it to
be queued and run boosted.
The next replenishment will take care of the runtime overrun, pushing
the deadline further away. See the "while (dl_se->runtime <= 0)" on
replenish_dl_entity() for more information.
Reported-by: Mark Simmons <msimmons@redhat.com>
Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Juri Lelli <juri.lelli@redhat.com>
Tested-by: Mark Simmons <msimmons@redhat.com>
Link: https://lkml.kernel.org/r/5076e003450835ec74e6fa5917d02c4fa41687e6.1600170294.git.bristot@redhat.com
Currently, pick_next_entity(...) has the following structure
(simplified):
[...]
if (last_buddy_ok())
result = last_buddy;
if (next_buddy_ok())
result = next_buddy;
[...]
The intended behavior is to prefer next buddy over last buddy;
the current code somewhat obfuscates this, and also wastes
cycles checking the last buddy when eventually the next buddy is
picked up.
So this patch refactors two 'ifs' above into
[...]
if (next_buddy_ok())
result = next_buddy;
else if (last_buddy_ok())
result = last_buddy;
[...]
Signed-off-by: Peter Oskolkov <posk@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guitttot@linaro.org>
Link: https://lkml.kernel.org/r/20200930173532.1069092-1-posk@google.com
This patchset is based on Google-internal RSEQ work done by Paul
Turner and Andrew Hunter.
When working with per-CPU RSEQ-based memory allocations, it is
sometimes important to make sure that a global memory location is no
longer accessed from RSEQ critical sections. For example, there can be
two per-CPU lists, one is "active" and accessed per-CPU, while another
one is inactive and worked on asynchronously "off CPU" (e.g. garbage
collection is performed). Then at some point the two lists are
swapped, and a fast RCU-like mechanism is required to make sure that
the previously active list is no longer accessed.
This patch introduces such a mechanism: in short, membarrier() syscall
issues an IPI to a CPU, restarting a potentially active RSEQ critical
section on the CPU.
Signed-off-by: Peter Oskolkov <posk@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Link: https://lkml.kernel.org/r/20200923233618.2572849-1-posk@google.com
The busy_factor, which increases load balance interval when a cpu is busy,
is set to 32 by default. This value generates some huge LB interval on
large system like the THX2 made of 2 node x 28 cores x 4 threads.
For such system, the interval increases from 112ms to 3584ms at MC level.
And from 228ms to 7168ms at NUMA level.
Even on smaller system, a lower busy factor has shown improvement on the
fair distribution of the running time so let reduce it for all.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Phil Auld <pauld@redhat.com>
Link: https://lkml.kernel.org/r/20200921072424.14813-5-vincent.guittot@linaro.org
sched domains tend to trigger simultaneously the load balance loop but
the larger domains often need more time to collect statistics. This
slowness makes the larger domain trying to detach tasks from a rq whereas
tasks already migrated somewhere else at a sub-domain level. This is not
a real problem for idle LB because the period of smaller domains will
increase with its CPUs being busy and this will let time for higher ones
to pulled tasks. But this becomes a problem when all CPUs are already busy
because all domains stay synced when they trigger their LB.
A simple way to minimize simultaneous LB of all domains is to decrement the
the busy interval by 1 jiffies. Because of the busy_factor, the interval of
larger domain will not be a multiple of smaller ones anymore.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Phil Auld <pauld@redhat.com>
Link: https://lkml.kernel.org/r/20200921072424.14813-4-vincent.guittot@linaro.org
The 25% default imbalance threshold for DIE and NUMA domain is large
enough to generate significant unfairness between threads. A typical
example is the case of 11 threads running on 2x4 CPUs. The imbalance of
20% between the 2 groups of 4 cores is just low enough to not trigger
the load balance between the 2 groups. We will have always the same 6
threads on one group of 4 CPUs and the other 5 threads on the other
group of CPUS. With a fair time sharing in each group, we ends up with
+20% running time for the group of 5 threads.
Consider decreasing the imbalance threshold for overloaded case where we
use the load to balance task and to ensure fair time sharing.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Phil Auld <pauld@redhat.com>
Acked-by: Hillf Danton <hdanton@sina.com>
Link: https://lkml.kernel.org/r/20200921072424.14813-3-vincent.guittot@linaro.org
Some UCs like 9 always running tasks on 8 CPUs can't be balanced and the
load balancer currently migrates the waiting task between the CPUs in an
almost random manner. The success of a rq pulling a task depends of the
value of nr_balance_failed of its domains and its ability to be faster
than others to detach it. This behavior results in an unfair distribution
of the running time between tasks because some CPUs will run most of the
time, if not always, the same task whereas others will share their time
between several tasks.
Instead of using nr_balance_failed as a boolean to relax the condition
for detaching task, the LB will use nr_balanced_failed to relax the
threshold between the tasks'load and the imbalance. This mecanism
prevents the same rq or domain to always win the load balance fight.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Phil Auld <pauld@redhat.com>
Link: https://lkml.kernel.org/r/20200921072424.14813-2-vincent.guittot@linaro.org
