Currently, accessing /proc/cpuinfo sends IPIs to idle CPUs in order to
learn their clock frequency. Which is a bit strange, given that waking
them from idle likely significantly changes their clock frequency.
This commit therefore avoids sending /proc/cpuinfo-induced IPIs to
idle CPUs.
[ paulmck: Also check for idle in arch_freq_prepare_all(). ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
The aperfmperf_snapshot_cpu() function is invoked upon access to
/proc/cpuinfo, and it does do an early exit if the specified CPU has
recently done a snapshot. Unfortunately, the indication that a snapshot
has been completed is set in an IPI handler, and the execution of this
handler can be delayed by any number of unfortunate events. This means
that a system that starts a number of applications, each of which
parses /proc/cpuinfo, can suffer from an smp_call_function_single()
storm, especially given that each access to /proc/cpuinfo invokes
smp_call_function_single() for all CPUs. Please note that this is not
theoretical speculation. Note also that one CPU's pending IPI serves
all requests, so there is no point in ever having more than one IPI
pending to a given CPU.
This commit therefore suppresses duplicate IPIs to a given CPU via a
new ->scfpending field in the aperfmperf_sample structure. This field
is set to the value one if an IPI is pending to the corresponding CPU
and to zero otherwise.
The aperfmperf_snapshot_cpu() function uses atomic_xchg() to set this
field to the value one and sample the old value. If this function's
"wait" parameter is zero, smp_call_function_single() is called only if
the old value of the ->scfpending field was zero. The IPI handler uses
atomic_set_release() to set this new field to zero just before returning,
so that the prior stores into the aperfmperf_sample structure are seen
by future requests that get to the atomic_xchg(). Future requests that
pass the elapsed-time check are ordered by the fact that on x86 loads act
as acquire loads, just as was the case prior to this change. The return
value is based off of the age of the prior snapshot, just as before.
Reported-by: Dave Jones <davej@codemonkey.org.uk>
[ paulmck: Allow /proc/cpuinfo to take advantage of arch_freq_get_on_cpu(). ]
[ paulmck: Add comment on memory barrier. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Since commit 7d5905dc14 ("x86 / CPU: Always show current CPU frequency
in /proc/cpuinfo") open and read of /proc/cpuinfo sends IPI to all CPUs.
Many applications read /proc/cpuinfo at the start for trivial reasons like
counting cores or detecting cpu features. While sensitive workloads like
DPDK network polling don't like any interrupts.
Integrates this feature with cpu isolation and do not send IPIs to CPUs
without housekeeping flag HK_FLAG_MISC (set by nohz_full).
Code that requests cpu frequency like show_cpuinfo() falls back to the last
frequency set by the cpufreq driver if this method returns 0.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Len Brown <len.brown@intel.com>
Cc: Frederic Weisbecker <frederic@kernel.org>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Link: https://lkml.kernel.org/r/155790354043.1104.15333317408370209.stgit@buzz
Based on 1 normalized pattern(s):
this file is licensed under gplv2
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-only
has been chosen to replace the boilerplate/reference in 22 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Armijn Hemel <armijn@tjaldur.nl>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190531190115.129548190@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Using static_cpu_has() is pointless on those paths, convert them to the
boot_cpu_has() variant.
No functional changes.
Reported-by: Nadav Amit <nadav.amit@gmail.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Rik van Riel <riel@surriel.com>
Reviewed-by: Juergen Gross <jgross@suse.com> # for paravirt
Cc: Aubrey Li <aubrey.li@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Thomas Lendacky <Thomas.Lendacky@amd.com>
Cc: linux-edac@vger.kernel.org
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: virtualization@lists.linux-foundation.org
Cc: x86@kernel.org
Link: https://lkml.kernel.org/r/20190330112022.28888-3-bp@alien8.de
... with the goal of eventually enabling -Wmissing-prototypes by
default. At least on x86.
Make functions static where possible, otherwise add prototypes or make
them visible through includes.
asm/trace/ changes courtesy of Steven Rostedt <rostedt@goodmis.org>.
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> # ACPI + cpufreq bits
Cc: Andrew Banman <andrew.banman@hpe.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Mike Travis <mike.travis@hpe.com>
Cc: "Steven Rostedt (VMware)" <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yi Wang <wang.yi59@zte.com.cn>
Cc: linux-acpi@vger.kernel.org
After commit 890da9cf09 (Revert "x86: do not use cpufreq_quick_get()
for /proc/cpuinfo "cpu MHz"") the "cpu MHz" number in /proc/cpuinfo
on x86 can be either the nominal CPU frequency (which is constant)
or the frequency most recently requested by a scaling governor in
cpufreq, depending on the cpufreq configuration. That is somewhat
inconsistent and is different from what it was before 4.13, so in
order to restore the previous behavior, make it report the current
CPU frequency like the scaling_cur_freq sysfs file in cpufreq.
To that end, modify the /proc/cpuinfo implementation on x86 to use
aperfmperf_snapshot_khz() to snapshot the APERF and MPERF feedback
registers, if available, and use their values to compute the CPU
frequency to be reported as "cpu MHz".
However, do that carefully enough to avoid accumulating delays that
lead to unacceptable access times for /proc/cpuinfo on systems with
many CPUs. Run aperfmperf_snapshot_khz() once on all CPUs
asynchronously at the /proc/cpuinfo open time, add a single delay
upfront (if necessary) at that point and simply compute the current
frequency while running show_cpuinfo() for each individual CPU.
Also, to avoid slowing down /proc/cpuinfo accesses too much, reduce
the default delay between consecutive APERF and MPERF reads to 10 ms,
which should be sufficient to get large enough numbers for the
frequency computation in all cases.
Fixes: 890da9cf09 (Revert "x86: do not use cpufreq_quick_get() for /proc/cpuinfo "cpu MHz"")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Ingo Molnar <mingo@kernel.org>
Even though aperfmperf_snapshot_khz() caches the samples.khz value to
return if called again in a sufficiently short time, its caller,
arch_freq_get_on_cpu(), still uses smp_call_function_single() to run it
which may allow user space to trigger an IPI storm by reading from the
scaling_cur_freq cpufreq sysfs file in a tight loop.
To avoid that, move the decision on whether or not to return the cached
samples.khz value to arch_freq_get_on_cpu().
This change was part of commit 941f5f0f6e ("x86: CPU: Fix up "cpu MHz"
in /proc/cpuinfo"), but it was not the reason for the revert and it
remains applicable.
Fixes: 4815d3c56d (cpufreq: x86: Make scaling_cur_freq behave more as expected)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: WANG Chao <chao.wang@ucloud.cn>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit 941f5f0f6e.
Sadly, it turns out that we really can't just do the cross-CPU IPI to
all CPU's to get their proper frequencies, because it's much too
expensive on systems with lots of cores.
So we'll have to revert this for now, and revisit it using a smarter
model (probably doing one system-wide IPI at open time, and doing all
the frequency calculations in parallel).
Reported-by: WANG Chao <chao.wang@ucloud.cn>
Reported-by: Ingo Molnar <mingo@kernel.org>
Cc: Rafael J Wysocki <rafael.j.wysocki@intel.com>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 890da9cf09 (Revert "x86: do not use cpufreq_quick_get() for
/proc/cpuinfo "cpu MHz"") is not sufficient to restore the previous
behavior of "cpu MHz" in /proc/cpuinfo on x86 due to some changes
made after the commit it has reverted.
To address this, make the code in question use arch_freq_get_on_cpu()
which also is used by cpufreq for reporting the current frequency of
CPUs and since that function doesn't really depend on cpufreq in any
way, drop the CONFIG_CPU_FREQ dependency for the object file
containing it.
Also refactor arch_freq_get_on_cpu() somewhat to avoid IPIs and
return cached values right away if it is called very often over a
short time (to prevent user space from triggering IPI storms through
it).
Fixes: 890da9cf09 (Revert "x86: do not use cpufreq_quick_get() for /proc/cpuinfo "cpu MHz"")
Cc: stable@kernel.org # 4.13 - together with 890da9cf09
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
According to Intel 64 and IA-32 Architectures SDM, Volume 3,
Chapter 14.2, "Software needs to exercise care to avoid delays
between the two RDMSRs (for example interrupts)".
So, disable interrupts during reading MSRs IA32_APERF and IA32_MPERF.
See also: commit 4ab60c3f32 (cpufreq: intel_pstate: Disable
interrupts during MSRs reading).
Signed-off-by: Doug Smythies <dsmythies@telus.net>
Reviewed-by: Len Brown <len.brown@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
After commit f8475cef90 "x86: use common aperfmperf_khz_on_cpu() to
calculate KHz using APERF/MPERF" the scaling_cur_freq policy attribute
in sysfs only behaves as expected on x86 with APERF/MPERF registers
available when it is read from at least twice in a row. The value
returned by the first read may not be meaningful, because the
computations in there use cached values from the previous iteration
of aperfmperf_snapshot_khz() which may be stale.
To prevent that from happening, modify arch_freq_get_on_cpu() to
call aperfmperf_snapshot_khz() twice, with a short delay between
these calls, if the previous invocation of aperfmperf_snapshot_khz()
was too far back in the past (specifically, more that 1s ago).
Also, as pointed out by Doug Smythies, aperf_delta is limited now
and the multiplication of it by cpu_khz won't overflow, so simplify
the s->khz computations too.
Fixes: f8475cef90 "x86: use common aperfmperf_khz_on_cpu() to calculate KHz using APERF/MPERF"
Reported-by: Doug Smythies <dsmythies@telus.net>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The goal of this change is to give users a uniform and meaningful
result when they read /sys/...cpufreq/scaling_cur_freq
on modern x86 hardware, as compared to what they get today.
Modern x86 processors include the hardware needed
to accurately calculate frequency over an interval --
APERF, MPERF, and the TSC.
Here we provide an x86 routine to make this calculation
on supported hardware, and use it in preference to any
driver driver-specific cpufreq_driver.get() routine.
MHz is computed like so:
MHz = base_MHz * delta_APERF / delta_MPERF
MHz is the average frequency of the busy processor
over a measurement interval. The interval is
defined to be the time between successive invocations
of aperfmperf_khz_on_cpu(), which are expected to to
happen on-demand when users read sysfs attribute
cpufreq/scaling_cur_freq.
As with previous methods of calculating MHz,
idle time is excluded.
base_MHz above is from TSC calibration global "cpu_khz".
This x86 native method to calculate MHz returns a meaningful result
no matter if P-states are controlled by hardware or firmware
and/or if the Linux cpufreq sub-system is or is-not installed.
When this routine is invoked more frequently, the measurement
interval becomes shorter. However, the code limits re-computation
to 10ms intervals so that average frequency remains meaningful.
Discerning users are encouraged to take advantage of
the turbostat(8) utility, which can gracefully handle
concurrent measurement intervals of arbitrary length.
Signed-off-by: Len Brown <len.brown@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
