Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
In this case, use fifo_low, because it only cares about being above
SCHED_NORMAL. Effectively no change in behaviour.
Cc: alexander.deucher@amd.com
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
Use sched_set_fifo(); Effectively changes prio from 16 to 50.
Cc: airlied@redhat.com
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
Effectively changes prio from 99 to 50.
XXX this thing is horrific, it basically open-codes a stop-machine and
idle.
Cc: daniel.lezcano@linaro.org
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
In this case, use fifo_low, because it only cares about being above
SCHED_NORMAL. Effectively changes prio from 2 to 1.
Cc: axboe@kernel.dk
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
In this case, use fifo_low, because it only cares about being above
SCHED_NORMAL. Effectively no change in behaviour.
XXX: this driver is still complete crap; why isn't it using proper
idle injection or at the very least play_idle() ?
Cc: rafael.j.wysocki@intel.com
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
Use sched_set_fifo() to request SCHED_FIFO and delegate
actual priority selection to userspace. Effectively no change in
behaviour.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Because SCHED_FIFO is a broken scheduler model (see previous patches)
take away the priority field, the kernel can't possibly make an
informed decision.
In this case, use fifo_low, because it only cares about being above
SCHED_NORMAL. Effectively no change in behaviour.
Cc: rmk+kernel@arm.linux.org.uk
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Nicolas Pitre <nico@fluxnic.net>
SCHED_FIFO (or any static priority scheduler) is a broken scheduler
model; it is fundamentally incapable of resource management, the one
thing an OS is actually supposed to do.
It is impossible to compose static priority workloads. One cannot take
two well designed and functional static priority workloads and mash
them together and still expect them to work.
Therefore it doesn't make sense to expose the priority field; the
kernel is fundamentally incapable of setting a sensible value, it
needs systems knowledge that it doesn't have.
Take away sched_setschedule() / sched_setattr() from modules and
replace them with:
- sched_set_fifo(p); create a FIFO task (at prio 50)
- sched_set_fifo_low(p); create a task higher than NORMAL,
which ends up being a FIFO task at prio 1.
- sched_set_normal(p, nice); (re)set the task to normal
This stops the proliferation of randomly chosen, and irrelevant, FIFO
priorities that dont't really mean anything anyway.
The system administrator/integrator, whoever has insight into the
actual system design and requirements (userspace) can set-up
appropriate priorities if and when needed.
Cc: airlied@redhat.com
Cc: alexander.deucher@amd.com
Cc: awalls@md.metrocast.net
Cc: axboe@kernel.dk
Cc: broonie@kernel.org
Cc: daniel.lezcano@linaro.org
Cc: gregkh@linuxfoundation.org
Cc: hannes@cmpxchg.org
Cc: herbert@gondor.apana.org.au
Cc: hverkuil@xs4all.nl
Cc: john.stultz@linaro.org
Cc: nico@fluxnic.net
Cc: paulmck@kernel.org
Cc: rafael.j.wysocki@intel.com
Cc: rmk+kernel@arm.linux.org.uk
Cc: sudeep.holla@arm.com
Cc: tglx@linutronix.de
Cc: ulf.hansson@linaro.org
Cc: wim@linux-watchdog.org
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Paul E. McKenney <paulmck@kernel.org>
Explain the rationale for annotating WARN(), even though, strictly
speaking printk() and friends are very much not safe in many of the
places we put them.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
The UBSAN instrumentation only inserts external CALLs when things go
'BAD', much like WARN(). So treat them similar to WARN()s for noinstr,
that is: allow them, at the risk of taking the machine down, to get
their message out.
Suggested-by: Marco Elver <elver@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Marco Elver <elver@google.com>
Adds the portable definitions for __no_sanitize_address, and
__no_sanitize_undefined, and subsequently changes noinstr to use the
attributes to disable instrumentation via KASAN or UBSAN.
Reported-by: syzbot+dc1fa714cb070b184db5@syzkaller.appspotmail.com
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
Link: https://lore.kernel.org/lkml/000000000000d2474c05a6c938fe@google.com/
Adds config variable CC_HAS_WORKING_NOSANITIZE_ADDRESS, which will be
true if we have a compiler that does not fail builds due to
no_sanitize_address functions. This does not yet mean they work as
intended, but for automated build-tests, this is the minimum
requirement.
For example, we require that __always_inline functions used from
no_sanitize_address functions do not generate instrumentation. On GCC <=
7 this fails to build entirely, therefore we make the minimum version
GCC 8.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Acked-by: Andrey Konovalov <andreyknvl@google.com>
Link: https://lkml.kernel.org/r/20200602175859.GC2604@hirez.programming.kicks-ass.net
The 'noinstr' function attribute means no-instrumentation, this should
very much include *SAN. Because lots of that is broken at present,
only include KCSAN for now, as that is limited to clang11, which has
sane function attribute behaviour.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
There are no more user of this function attribute, also, with us now
actively supporting '__no_kcsan inline' it doesn't make sense to have
in any case.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Now that KCSAN relies on -tsan-distinguish-volatile we no longer need
the annotation for constant_test_bit(). Remove it.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Factorize in a single place the calculation of the divider to be used to
to compute *_avg from *_sum value
Suggested-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200612154703.23555-1-vincent.guittot@linaro.org
When a task has a runtime that cannot be served within the scheduling
deadline by any of the idle CPU (later_mask) the task is doomed to miss
its deadline.
This can happen since the SCHED_DEADLINE admission control guarantees
only bounded tardiness and not the hard respect of all deadlines.
In this case try to select the idle CPU with the largest CPU capacity
to minimize tardiness.
Favor task_cpu(p) if it has max capacity of !fitting CPUs so that
find_later_rq() can potentially still return it (most likely cache-hot)
early.
Signed-off-by: Luca Abeni <luca.abeni@santannapisa.it>
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/20200520134243.19352-6-dietmar.eggemann@arm.com
The current SCHED_DEADLINE (DL) scheduler uses a global EDF scheduling
algorithm w/o considering CPU capacity or task utilization.
This works well on homogeneous systems where DL tasks are guaranteed
to have a bounded tardiness but presents issues on heterogeneous
systems.
A DL task can migrate to a CPU which does not have enough CPU capacity
to correctly serve the task (e.g. a task w/ 70ms runtime and 100ms
period on a CPU w/ 512 capacity).
Add the DL fitness function dl_task_fits_capacity() for DL admission
control on heterogeneous systems. A task fits onto a CPU if:
CPU original capacity / 1024 >= task runtime / task deadline
Use this function on heterogeneous systems to try to find a CPU which
meets this criterion during task wakeup, push and offline migration.
On homogeneous systems the original behavior of the DL admission
control should be retained.
Signed-off-by: Luca Abeni <luca.abeni@santannapisa.it>
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/20200520134243.19352-5-dietmar.eggemann@arm.com
The current SCHED_DEADLINE (DL) admission control ensures that
sum of reserved CPU bandwidth < x * M
where
x = /proc/sys/kernel/sched_rt_{runtime,period}_us
M = # CPUs in root domain.
DL admission control works well for homogeneous systems where the
capacity of all CPUs are equal (1024). I.e. bounded tardiness for DL
and non-starvation of non-DL tasks is guaranteed.
But on heterogeneous systems where capacity of CPUs are different it
could fail by over-allocating CPU time on smaller capacity CPUs.
On an Arm big.LITTLE/DynamIQ system DL tasks can easily starve other
tasks making it unusable.
Fix this by explicitly considering the CPU capacity in the DL admission
test by replacing M with the root domain CPU capacity sum.
Signed-off-by: Luca Abeni <luca.abeni@santannapisa.it>
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/20200520134243.19352-4-dietmar.eggemann@arm.com
Capacity-aware SCHED_DEADLINE Admission Control (AC) needs root domain
(rd) CPU capacity sum.
Introduce dl_bw_capacity() which for a symmetric rd w/ a CPU capacity
of SCHED_CAPACITY_SCALE simply relies on dl_bw_cpus() to return #CPUs
multiplied by SCHED_CAPACITY_SCALE.
For an asymmetric rd or a CPU capacity < SCHED_CAPACITY_SCALE it
computes the CPU capacity sum over rd span and cpu_active_mask.
A 'XXX Fix:' comment was added to highlight that if 'rq->rd ==
def_root_domain' AC should be performed against the capacity of the
CPU the task is running on rather the rd CPU capacity sum. This
issue already exists w/o capacity awareness.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/20200520134243.19352-3-dietmar.eggemann@arm.com
Return the weight of the root domain (rd) span in case it is a subset
of the cpu_active_mask.
Continue to compute the number of CPUs over rd span and cpu_active_mask
when in hotplug.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/20200520134243.19352-2-dietmar.eggemann@arm.com
During sched domain init, we check whether non-topological SD_flags are
returned by tl->sd_flags(), if found, fire a waning and correct the
violation, but the code failed to correct the violation. Correct this.
Fixes: 143e1e28cb ("sched: Rework sched_domain topology definition")
Signed-off-by: Peng Liu <iwtbavbm@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/20200609150936.GA13060@iZj6chx1xj0e0buvshuecpZ
With commit:
'b7031a02ec ("sched/fair: Add NOHZ_STATS_KICK")'
rebalance_domains of the local cfs_rq happens before others idle cpus have
updated nohz.next_balance and its value is overwritten.
Move the update of nohz.next_balance for other idles cpus before balancing
and updating the next_balance of local cfs_rq.
Also, the nohz.next_balance is now updated only if all idle cpus got a
chance to rebalance their domains and the idle balance has not been aborted
because of new activities on the CPU. In case of need_resched, the idle
load balance will be kick the next jiffie in order to address remaining
ilb.
Fixes: b7031a02ec ("sched/fair: Add NOHZ_STATS_KICK")
Reported-by: Peng Liu <iwtbavbm@gmail.com>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Link: https://lkml.kernel.org/r/20200609123748.18636-1-vincent.guittot@linaro.org
This is a kernel enhancement that configures the cpu affinity of kernel
threads via kernel boot option nohz_full=.
When this option is specified, the cpumask is immediately applied upon
kthread launch. This does not affect kernel threads that specify cpu
and node.
This allows CPU isolation (that is not allowing certain threads
to execute on certain CPUs) without using the isolcpus=domain parameter,
making it possible to enable load balancing on such CPUs
during runtime (see kernel-parameters.txt).
Note-1: this is based off on Wind River's patch at
https://github.com/starlingx-staging/stx-integ/blob/master/kernel/kernel-std/centos/patches/affine-compute-kernel-threads.patch
Difference being that this patch is limited to modifying kernel thread
cpumask. Behaviour of other threads can be controlled via cgroups or
sched_setaffinity.
Note-2: Wind River's patch was based off Christoph Lameter's patch at
https://lwn.net/Articles/565932/ with the only difference being
the kernel parameter changed from kthread to kthread_cpus.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200527142909.23372-3-frederic@kernel.org
Next patch will switch unbound kernel threads mask to
housekeeping_cpumask(), a subset of cpu_possible_mask. So in order to
ease bisection, lets first switch kthreads default affinity from
cpu_all_mask to cpu_possible_mask.
It looks safe to do so as cpu_possible_mask seem to be initialized
at setup_arch() time, way before kthreadd is created.
Suggested-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200527142909.23372-2-frederic@kernel.org
Each psi group requires a dedicated kthread_delayed_work and
kthread_worker. Since no other work can be performed using psi_group's
kthread_worker, the same result can be obtained using a task_struct and
a timer directly. This makes psi triggering simpler by removing lists
and locks involved with kthread_worker usage and eliminates the need for
poll_scheduled atomic use in the hot path.
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200528195442.190116-1-surenb@google.com
Be defensive against the case where the processor reports a base_freq
larger than turbo_freq (the ratio would be zero).
Fixes: 1567c3e346 ("x86, sched: Add support for frequency invariance")
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200531182453.15254-4-ggherdovich@suse.cz
There may be CPUs that support turbo boost but don't declare any turbo
ratio, i.e. their MSR_TURBO_RATIO_LIMIT is all zeroes. In that condition
scale-invariant calculations can't be performed.
Fixes: 1567c3e346 ("x86, sched: Add support for frequency invariance")
Suggested-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Link: https://lkml.kernel.org/r/20200531182453.15254-3-ggherdovich@suse.cz
The product mcnt * arch_max_freq_ratio can overflows u64.
For context, a large value for arch_max_freq_ratio would be 5000,
corresponding to a turbo_freq/base_freq ratio of 5 (normally it's more like
1500-2000). A large increment frequency for the MPERF counter would be 5GHz
(the base clock of all CPUs on the market today is less than that). With
these figures, a CPU would need to go without a scheduler tick for around 8
days for the u64 overflow to happen. It is unlikely, but the check is
warranted.
Under similar conditions, the difference acnt of two consecutive APERF
readings can overflow as well.
In these circumstances is appropriate to disable frequency invariant
accounting: the feature relies on measures of the clock frequency done at
every scheduler tick, which need to be "fresh" to be at all meaningful.
A note on i386: prior to version 5.1, the GCC compiler didn't have the
builtin function __builtin_mul_overflow. In these GCC versions the macro
check_mul_overflow needs __udivdi3() to do (u64)a/b, which the kernel
doesn't provide. For this reason this change fails to build on i386 if
GCC<5.1, and we protect the entire frequency invariant code behind
CONFIG_X86_64 (special thanks to "kbuild test robot" <lkp@intel.com>).
Fixes: 1567c3e346 ("x86, sched: Add support for frequency invariance")
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200531182453.15254-2-ggherdovich@suse.cz
The util_est signals are key elements for EAS task placement and
frequency selection. Having tracepoints to track these signals enables
load-tracking and schedutil testing and/or debugging by a toolkit.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/1590597554-370150-1-git-send-email-vincent.donnefort@arm.com
Since commit 8ec59c0f5f ("sched/topology: Remove unused 'sd'
parameter from arch_scale_cpu_capacity()") it is no longer needed.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20200603080304.16548-5-dietmar.eggemann@arm.com
The idle task and stop task sched_classes return 0 in this function.
The single call site in sched_rr_get_interval() calls
p->sched_class->get_rr_interval() only conditional in case it is
defined. Otherwise time_slice=0 will be used.
The deadline sched class does not define it. Commit a57beec5d4
("sched: Make sched_class::get_rr_interval() optional") introduced
the default time-slice=0 for sched classes which do not provide this
function.
So .get_rr_interval for idle and stop sched_class can be removed to
shrink the code a little.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200603080304.16548-4-dietmar.eggemann@arm.com
Commit 6d1cafd8b5 ("sched: Resched proper CPU on yield_to()") moved
the code to resched the CPU from yield_to_task_fair() to yield_to()
making the preempt parameter in sched_class->yield_to_task()
unnecessary. Remove it. No other sched_class implements yield_to_task().
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200603080304.16548-3-dietmar.eggemann@arm.com
Besides in PELT cap_scale() is used in the Deadline scheduler class for
scale-invariant bandwidth enforcement.
Remove the cap_scale() definition in kernel/sched/pelt.c and keep the
one in kernel/sched/sched.h.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20200603080304.16548-2-dietmar.eggemann@arm.com
People report that utime and stime from /proc/<pid>/stat become very
wrong when the numbers are big enough, especially if you watch these
counters incrementally.
Specifically, the current implementation of: stime*rtime/total,
results in a saw-tooth function on top of the desired line, where the
teeth grow in size the larger the values become. IOW, it has a
relative error.
The result is that, when watching incrementally as time progresses
(for large values), we'll see periods of pure stime or utime increase,
irrespective of the actual ratio we're striving for.
Replace scale_stime() with a math64.h helper: mul_u64_u64_div_u64()
that is far more accurate. This also allows architectures to override
the implementation -- for instance they can opt for the old algorithm
if this new one turns out to be too expensive for them.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200519172506.GA317395@hirez.programming.kicks-ass.net
Current version supports a server line starting Intel® Xeon® Processor
Scalable Family and introduces mapping for IIO Uncore units only.
Other units can be added on demand.
IIO stack to PMON mapping is exposed through:
/sys/devices/uncore_iio_<pmu_idx>/dieX
where dieX is file which holds "Segment:Root Bus" for PCIe root port,
which can be monitored by that IIO PMON block.
Details are explained in Documentation/ABI/testing/sysfs-devices-mapping
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Alexander Antonov <alexander.antonov@linux.intel.com>
Signed-off-by: Roman Sudarikov <roman.sudarikov@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Reviewed-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Link: https://lkml.kernel.org/r/20200601083543.30011-4-alexander.antonov@linux.intel.com
The accessor to return number of dies on the platform.
Signed-off-by: Alexander Antonov <alexander.antonov@linux.intel.com>
Signed-off-by: Roman Sudarikov <roman.sudarikov@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Reviewed-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Link: https://lkml.kernel.org/r/20200601083543.30011-3-alexander.antonov@linux.intel.com
Each Uncore unit type, by its nature, can be mapped to its own context -
which platform component each PMON block of that type is supposed to
monitor.
Intel® Xeon® Scalable processor family (code name Skylake-SP) makes
significant changes in the integrated I/O (IIO) architecture. The new
solution introduces IIO stacks which are responsible for managing traffic
between the PCIe domain and the Mesh domain. Each IIO stack has its own
PMON block and can handle either DMI port, x16 PCIe root port, MCP-Link
or various built-in accelerators. IIO PMON blocks allow concurrent
monitoring of I/O flows up to 4 x4 bifurcation within each IIO stack.
Software is supposed to program required perf counters within each IIO
stack and gather performance data. The tricky thing here is that IIO PMON
reports data per IIO stack but users have no idea what IIO stacks are -
they only know devices which are connected to the platform.
Understanding IIO stack concept to find which IIO stack that particular
IO device is connected to, or to identify an IIO PMON block to program
for monitoring specific IIO stack assumes a lot of implicit knowledge
about given Intel server platform architecture.
Usage example:
ls /sys/devices/uncore_<type>_<pmu_idx>/die*
Signed-off-by: Alexander Antonov <alexander.antonov@linux.intel.com>
Signed-off-by: Roman Sudarikov <roman.sudarikov@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Reviewed-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Link: https://lkml.kernel.org/r/20200601083543.30011-2-alexander.antonov@linux.intel.com
An oops will be triggered, if perf tries to access an invalid address
which exceeds the mapped area.
Check the address before the actual access to MMIO sapce of an uncore
unit.
Suggested-by: David Laight <David.Laight@ACULAB.COM>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/1590679169-61823-3-git-send-email-kan.liang@linux.intel.com
Perf cannot validate an address before the actual access to MMIO space
of some uncore units, e.g. IMC on TGL. Accessing an invalid address,
which exceeds mapped area, can trigger oops.
Perf never records the size of mapped area. Generic functions, e.g.
uncore_mmio_read_counter(), cannot get the correct size for address
validation.
Add mmio_map_size in intel_uncore_type to record the size of mapped
area. Print warning message if ioremap fails.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/1590679169-61823-2-git-send-email-kan.liang@linux.intel.com
When counting IMC uncore events on some TGL machines, an oops will be
triggered.
[ 393.101262] BUG: unable to handle page fault for address:
ffffb45200e15858
[ 393.101269] #PF: supervisor read access in kernel mode
[ 393.101271] #PF: error_code(0x0000) - not-present page
Current perf uncore driver still use the IMC MAP SIZE inherited from
SNB, which is 0x6000.
However, the offset of IMC uncore counters is larger than 0x6000,
e.g. 0xd8a0.
Enlarge the IMC MAP SIZE for TGL to 0xe000.
Fixes: fdb6482244 ("perf/x86: Add Intel Tiger Lake uncore support")
Reported-by: Ammy Yi <ammy.yi@intel.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Ammy Yi <ammy.yi@intel.com>
Tested-by: Chao Qin <chao.qin@intel.com>
Link: https://lkml.kernel.org/r/1590679169-61823-1-git-send-email-kan.liang@linux.intel.com
Add perf text poke events for ftrace trampolines when created and when
freed.
There can be 3 text_poke events for ftrace trampolines:
1. NULL -> trampoline
By ftrace_update_trampoline() when !ops->trampoline
Trampoline created
2. [e.g. on x86] CALL rel32 -> CALL rel32
By arch_ftrace_update_trampoline() when ops->trampoline and
ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP
[e.g. on x86] via text_poke_bp() which generates text poke events
Trampoline-called function target updated
3. trampoline -> NULL
By ftrace_trampoline_free() when ops->trampoline and
ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP
Trampoline freed
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200512121922.8997-9-adrian.hunter@intel.com
Symbols are needed for tools to describe instruction addresses. Pages
allocated for ftrace's purposes need symbols to be created for them.
Add such symbols to be visible via perf ksymbol events.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200512121922.8997-8-adrian.hunter@intel.com
Symbols are needed for tools to describe instruction addresses. Pages
allocated for ftrace's purposes need symbols to be created for them.
Add such symbols to be visible via /proc/kallsyms.
Example on x86 with CONFIG_DYNAMIC_FTRACE=y
# echo function > /sys/kernel/debug/tracing/current_tracer
# cat /proc/kallsyms | grep '\[__builtin__ftrace\]'
ffffffffc0238000 t ftrace_trampoline [__builtin__ftrace]
Note: This patch adds "__builtin__ftrace" as a module name in /proc/kallsyms for
symbols for pages allocated for ftrace's purposes, even though "__builtin__ftrace"
is not a module.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200512121922.8997-7-adrian.hunter@intel.com
Add perf text poke events for kprobes. That includes:
- the replaced instruction(s) which are executed out-of-line
i.e. arch_copy_kprobe() and arch_remove_kprobe()
- the INT3 that activates the kprobe
i.e. arch_arm_kprobe() and arch_disarm_kprobe()
- optimised kprobe function
i.e. arch_prepare_optimized_kprobe() and
__arch_remove_optimized_kprobe()
- optimised kprobe
i.e. arch_optimize_kprobes() and arch_unoptimize_kprobe()
Resulting in 8 possible text_poke events:
0: NULL -> probe.ainsn.insn (if ainsn.boostable && !kp.post_handler)
arch_copy_kprobe()
1: old0 -> INT3 arch_arm_kprobe()
// boosted kprobe active
2: NULL -> optprobe_trampoline arch_prepare_optimized_kprobe()
3: INT3,old1,old2,old3,old4 -> JMP32 arch_optimize_kprobes()
// optprobe active
4: JMP32 -> INT3,old1,old2,old3,old4
// optprobe disabled and kprobe active (this sometimes goes back to 3)
arch_unoptimize_kprobe()
5: optprobe_trampoline -> NULL arch_remove_optimized_kprobe()
// boosted kprobe active
6: INT3 -> old0 arch_disarm_kprobe()
7: probe.ainsn.insn -> NULL (if ainsn.boostable && !kp.post_handler)
arch_remove_kprobe()
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lkml.kernel.org/r/20200512121922.8997-6-adrian.hunter@intel.com
Symbols are needed for tools to describe instruction addresses. Pages
allocated for kprobe's purposes need symbols to be created for them.
Add such symbols to be visible via perf ksymbol events.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lkml.kernel.org/r/20200512121922.8997-5-adrian.hunter@intel.com
Symbols are needed for tools to describe instruction addresses. Pages
allocated for kprobe's purposes need symbols to be created for them.
Add such symbols to be visible via /proc/kallsyms.
Note: kprobe insn pages are not used if ftrace is configured. To see the
effect of this patch, the kernel must be configured with:
# CONFIG_FUNCTION_TRACER is not set
CONFIG_KPROBES=y
and for optimised kprobes:
CONFIG_OPTPROBES=y
Example on x86:
# perf probe __schedule
Added new event:
probe:__schedule (on __schedule)
# cat /proc/kallsyms | grep '\[__builtin__kprobes\]'
ffffffffc00d4000 t kprobe_insn_page [__builtin__kprobes]
ffffffffc00d6000 t kprobe_optinsn_page [__builtin__kprobes]
Note: This patch adds "__builtin__kprobes" as a module name in
/proc/kallsyms for symbols for pages allocated for kprobes' purposes, even
though "__builtin__kprobes" is not a module.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lkml.kernel.org/r/20200528080058.20230-1-adrian.hunter@intel.com
