This commit causes all the readers to start running unmeasured load
until all readers have done at least one such run (thus having warmed
up), then run the measured load, and then run unmeasured load until all
readers have completed their measured load. This approach avoids any
thread running measured load while other readers are idle.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, readers are awakened individually. On most systems, this
results in significant wakeup delay from one reader to the next, which
can result in the first and last reader having sole access to the
synchronization primitive in question. If that synchronization primitive
involves shared memory, those readers will rack up a huge number of
operations in a very short time, causing large perturbations in the
results.
This commit therefore has the readers busy-wait after being awakened,
and uses a new n_started variable to synchronize their start times.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit converts the reader_task structure's "start" field to int
in order to demote a full barrier to an smp_load_acquire() and also to
simplify the code a bit. While in the area, and to enlist the compiler's
help in ensuring that nothing was missed, the field's name was changed
to start_reader.
Also while in the area, change the main_func() store to use
smp_store_release() to further fortify against wait/wake races.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit moves a printk() out of the measurement interval, converts
a atomic_dec()/atomic_read() pair to atomic_dec_and_test(), and adds
a smp_mb__before_atomic() to avoid potential wake/wait hangs. These
changes have the added benefit of reducing the number of loops required
for amortizing loop overhead for CONFIG_PREEMPT=n RCU measurements from
1,000,000 to 10,000. This reduction in turn shortens the test, reducing
the probability of interference.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Because the reset_readers() and process_durations() functions are used
only within kernel/rcu/refperf.c, this commit makes them static.
Reported-by: kbuild test robot <lkp@intel.com>
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, the buffer used to accumulate the thread-summary output is
fixed size, which will cause problems if someone decides to run on a large
number of PCUs. This commit therefore dynamically allocates this buffer.
[ paulmck: Fix memory allocation as suggested by KASAN. ]
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, the buffer used to accumulate the experiment-summary output
is fixed size, which will cause problems if someone decides to run
one hundred experiments. This commit therefore dynamically allocates
this buffer.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The current code uses the number of threads both to limit the number
of threads and to specify the number of experiments, but also varies
the number of threads as the experiments progress. This commit takes
a different approach by adding an refperf.nruns module parameter that
specifies the number of experiments, and furthermore uses the same
number of threads for each experiment.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit converts nreaders to a module parameter, with the default
of -1 specifying the old behavior of using 75% of the readers.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The CONFIG_PREEMPT=n rcu_read_lock()/rcu_read_unlock() pair's overhead,
even including loop overhead, is far less than one nanosecond.
Since logscale plots are not all that happy with zero values, provide
picoseconds as decimals.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Current runs show PREEMPT=n rcu_read_lock()/rcu_read_unlock() pairs
consuming between 20 and 30 nanoseconds, when in fact the actual value is
zero, give or take the barrier() asm's effect on compiler optimizations.
The additional overhead is caused by function calls through pointers
(especially in these days of Spectre mitigations) and perhaps also
needless argument passing, a non-const loop limit, and an upcounting loop.
This commit therefore combines the ->readlock() and ->readunlock()
function pointers into a single ->readsection() function pointer that
takes the loop count as a const parameter and keeps any data passed
from the read-lock to the read-unlock internal to this new function.
These changes reduce the measured overhead of the aforementioned
PREEMPT=n rcu_read_lock()/rcu_read_unlock() pairs from between 20 and
30 nanoseconds to somewhere south of 500 picoseconds.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds an rcuperf module parameter named "holdoff" that
defaults to 10 seconds if refperf is built in and to zero otherwise.
The assumption is that all the CPUs are online by the time that the
modprobe and insmod commands are going to do anything, and that normal
systems will have all the CPUs online within ten seconds.
Larger systems may take many tens of seconds or even minutes to get
to this point, hence this being a module parameter instead of being a
hard-coded constant.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit updates the rcutorture scripting to include the new refperf
torture-test module.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds comments explaining why the readers have otherwise insane
levels of measurement overhead, namely that they are intended as a test
load for update-side performance measurements, not as a straight-up
read-side performance test.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Add a test for comparing the performance of RCU with various read-side
synchronization mechanisms. The test has proved useful for collecting
data and performing these comparisons.
Currently RCU, SRCU, reader-writer lock, reader-writer semaphore and
reference counting can be measured using refperf.perf_type parameter.
Each invocation of the test runs measures performance of a specific
mechanism.
The maximum number of CPUs to concurrently run readers on is chosen by
the test itself and is 75% of the total number of CPUs. So if you had 24
CPUs, the test runs with a maximum of 18 parallel readers.
A number of experiments are conducted, and in each experiment, the
number of readers is increased by 1, upto the 75% of CPUs mark. During
each experiment, all readers execute an empty loop with refperf.loops
iterations and time the total loop duration. This is then averaged.
Example output:
Parameters "refperf.perf_type=srcu refperf.loops=2000000" looks like:
[ 3.347133] srcu-ref-perf:
[ 3.347133] Threads Time(ns)
[ 3.347133] 1 36
[ 3.347133] 2 34
[ 3.347133] 3 34
[ 3.347133] 4 34
[ 3.347133] 5 33
[ 3.347133] 6 33
[ 3.347133] 7 33
[ 3.347133] 8 33
[ 3.347133] 9 33
[ 3.347133] 10 33
[ 3.347133] 11 33
[ 3.347133] 12 33
[ 3.347133] 13 33
[ 3.347133] 14 33
[ 3.347133] 15 32
[ 3.347133] 16 33
[ 3.347133] 17 33
[ 3.347133] 18 34
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
wait_event() already retries if the condition for the wake up is not
satisifed after wake up. Remove them from the rcuperf test.
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Pull ARM OMAP fixes from Arnd Bergmann:
"The OMAP developers are particularly active at hunting down
regressions, so this is a separate branch with OMAP specific
fixes for v5.8:
As Tony explains
"The recent display subsystem (DSS) related platform data changes
caused display related regressions for suspend and resume. Looks
like I only tested suspend and resume before dropping the legacy
platform data, and forgot to test it after dropping it. Turns out
the main issue was that we no longer have platform code calling
pm_runtime_suspend for DSS like we did for the legacy platform data
case, and that fix is still being discussed on the dri-devel list
and will get merged separately. The DSS related testing exposed a
pile other other display related issues that also need fixing
though":
- Fix ti-sysc optional clock handling and reset status checks for
devices that reset automatically in idle like DSS
- Ignore ti-sysc clockactivity bit unless separately requested to
avoid unexpected performance issues
- Init ti-sysc framedonetv_irq to true and disable for am4
- Avoid duplicate DSS reset for legacy mode with dts data
- Remove LCD timings for am4 as they cause warnings now that we're
using generic panels
Other OMAP changes from Tony include:
- Fix omap_prm reset deassert as we still have drivers setting the
pm_runtime_irq_safe() flag
- Flush posted write for ti-sysc enable and disable
- Fix droid4 spi related errors with spi flags
- Fix am335x USB range and a typo for softreset
- Fix dra7 timer nodes for clocks for IPU and DSP
- Drop duplicate mailboxes after mismerge for dra7
- Prevent pocketgeagle header line signal from accidentally setting
micro-SD write protection signal by removing the default mux
- Fix NFSroot flakeyness after resume for duover by switching the
smsc911x gpio interrupt to back to level sensitive
- Fix regression for omap4 clockevent source after recent system
timer changes
- Yet another ethernet regression fix for the "rgmii" vs "rgmii-rxid"
phy-mode
- One patch to convert am3/am4 DT files to use the regular sdhci-omap
driver instead of the old hsmmc driver, this was meant for the
merge window but got lost in the process"
* tag 'arm-omap-fixes-5.8-1' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (21 commits)
ARM: dts: am5729: beaglebone-ai: fix rgmii phy-mode
ARM: dts: Fix omap4 system timer source clocks
ARM: dts: Fix duovero smsc interrupt for suspend
ARM: dts: am335x-pocketbeagle: Fix mmc0 Write Protect
Revert "bus: ti-sysc: Increase max softreset wait"
ARM: dts: am437x-epos-evm: remove lcd timings
ARM: dts: am437x-gp-evm: remove lcd timings
ARM: dts: am437x-sk-evm: remove lcd timings
ARM: dts: dra7-evm-common: Fix duplicate mailbox nodes
ARM: dts: dra7: Fix timer nodes properly for timer_sys_ck clocks
ARM: dts: Fix am33xx.dtsi ti,sysc-mask wrong softreset flag
ARM: dts: Fix am33xx.dtsi USB ranges length
bus: ti-sysc: Increase max softreset wait
ARM: OMAP2+: Fix legacy mode dss_reset
bus: ti-sysc: Fix uninitialized framedonetv_irq
bus: ti-sysc: Ignore clockactivity unless specified as a quirk
bus: ti-sysc: Use optional clocks on for enable and wait for softreset bit
ARM: dts: omap4-droid4: Fix spi configuration and increase rate
bus: ti-sysc: Flush posted write on enable and disable
soc: ti: omap-prm: use atomic iopoll instead of sleeping one
...
Pull ARM SoC fixes from Arnd Bergmann:
"Here are a couple of bug fixes, mostly for devicetree files
NXP i.MX:
- Use correct voltage on some i.MX8M board device trees to avoid
hardware damage
- Code fixes for a compiler warning and incorrect reference counting,
both harmless.
- Fix the i.MX8M SoC driver to correctly identify imx8mp
- Fix watchdog configuration in imx6ul-kontron device tree.
Broadcom:
- A small regression fix for the Raspberry-Pi firmware driver
- A Kconfig change to use the correct timer driver on Northstar
- A DT fix for the Luxul XWC-2000 machine
- Two more DT fixes for NSP SoCs
STmicroelectronics STI
- Revert one broken patch for L2 cache configuration
ARM Versatile Express:
- Fix a regression by reverting a broken DT cleanup
TEE drivers:
- MAINTAINERS: change tee mailing list"
* tag 'arm-fixes-5.8-1' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
Revert "ARM: sti: Implement dummy L2 cache's write_sec"
soc: imx8m: fix build warning
ARM: imx6: add missing put_device() call in imx6q_suspend_init()
ARM: imx5: add missing put_device() call in imx_suspend_alloc_ocram()
soc: imx8m: Correct i.MX8MP UID fuse offset
ARM: dts: imx6ul-kontron: Change WDOG_ANY signal from push-pull to open-drain
ARM: dts: imx6ul-kontron: Move watchdog from Kontron i.MX6UL/ULL board to SoM
arm64: dts: imx8mm-beacon: Fix voltages on LDO1 and LDO2
arm64: dts: imx8mn-ddr4-evk: correct ldo1/ldo2 voltage range
arm64: dts: imx8mm-evk: correct ldo1/ldo2 voltage range
ARM: dts: NSP: Correct FA2 mailbox node
ARM: bcm2835: Fix integer overflow in rpi_firmware_print_firmware_revision()
MAINTAINERS: change tee mailing list
ARM: dts: NSP: Disable PL330 by default, add dma-coherent property
ARM: bcm: Select ARM_TIMER_SP804 for ARCH_BCM_NSP
ARM: dts: BCM5301X: Add missing memory "device_type" for Luxul XWC-2000
arm: dts: vexpress: Move mcc node back into motherboard node
Pull timer fix from Ingo Molnar:
"A single DocBook fix"
* tag 'timers-urgent-2020-06-28' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
timekeeping: Fix kerneldoc system_device_crosststamp & al
Pull perf fix from Ingo Molnar:
"A single Kbuild dependency fix"
* tag 'perf-urgent-2020-06-28' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/x86/rapl: Fix RAPL config variable bug
Pull EFI fixes from Ingo Molnar:
- Fix build regression on v4.8 and older
- Robustness fix for TPM log parsing code
- kobject refcount fix for the ESRT parsing code
- Two efivarfs fixes to make it behave more like an ordinary file
system
- Style fixup for zero length arrays
- Fix a regression in path separator handling in the initrd loader
- Fix a missing prototype warning
- Add some kerneldoc headers for newly introduced stub routines
- Allow support for SSDT overrides via EFI variables to be disabled
- Report CPU mode and MMU state upon entry for 32-bit ARM
- Use the correct stack pointer alignment when entering from mixed mode
* tag 'efi-urgent-2020-06-28' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
efi/libstub: arm: Print CPU boot mode and MMU state at boot
efi/libstub: arm: Omit arch specific config table matching array on arm64
efi/x86: Setup stack correctly for efi_pe_entry
efi: Make it possible to disable efivar_ssdt entirely
efi/libstub: Descriptions for stub helper functions
efi/libstub: Fix path separator regression
efi/libstub: Fix missing-prototype warning for skip_spaces()
efi: Replace zero-length array and use struct_size() helper
efivarfs: Don't return -EINTR when rate-limiting reads
efivarfs: Update inode modification time for successful writes
efi/esrt: Fix reference count leak in esre_create_sysfs_entry.
efi/tpm: Verify event log header before parsing
efi/x86: Fix build with gcc 4
Pull scheduler fixes from Borislav Petkov:
"The most anticipated fix in this pull request is probably the horrible
build fix for the RANDSTRUCT fail that didn't make -rc2. Also included
is the cleanup that removes those BUILD_BUG_ON()s and replaces it with
ugly unions.
Also included is the try_to_wake_up() race fix that was first
triggered by Paul's RCU-torture runs, but was independently hit by
Dave Chinner's fstest runs as well"
* tag 'sched_urgent_for_5.8_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/cfs: change initial value of runnable_avg
smp, irq_work: Continue smp_call_function*() and irq_work*() integration
sched/core: s/WF_ON_RQ/WQ_ON_CPU/
sched/core: Fix ttwu() race
sched/core: Fix PI boosting between RT and DEADLINE tasks
sched/deadline: Initialize ->dl_boosted
sched/core: Check cpus_mask, not cpus_ptr in __set_cpus_allowed_ptr(), to fix mask corruption
sched/core: Fix CONFIG_GCC_PLUGIN_RANDSTRUCT build fail
Pull x86 fixes from Borislav Petkov:
- AMD Memory bandwidth counter width fix, by Babu Moger.
- Use the proper length type in the 32-bit truncate() syscall variant,
by Jiri Slaby.
- Reinit IA32_FEAT_CTL during wakeup to fix the case where after
resume, VMXON would #GP due to VMX not being properly enabled, by
Sean Christopherson.
- Fix a static checker warning in the resctrl code, by Dan Carpenter.
- Add a CR4 pinning mask for bits which cannot change after boot, by
Kees Cook.
- Align the start of the loop of __clear_user() to 16 bytes, to improve
performance on AMD zen1 and zen2 microarchitectures, by Matt Fleming.
* tag 'x86_urgent_for_5.8_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/asm/64: Align start of __clear_user() loop to 16-bytes
x86/cpu: Use pinning mask for CR4 bits needing to be 0
x86/resctrl: Fix a NULL vs IS_ERR() static checker warning in rdt_cdp_peer_get()
x86/cpu: Reinitialize IA32_FEAT_CTL MSR on BSP during wakeup
syscalls: Fix offset type of ksys_ftruncate()
x86/resctrl: Fix memory bandwidth counter width for AMD
Pull RCU-vs-KCSAN fixes from Borislav Petkov:
"A single commit that uses "arch_" atomic operations to avoid the
instrumentation that comes with the non-"arch_" versions.
In preparation for that commit, it also has another commit that makes
these "arch_" atomic operations available to generic code.
Without these commits, KCSAN uses can see pointless errors"
* tag 'rcu_urgent_for_5.8_rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
rcu: Fixup noinstr warnings
locking/atomics: Provide the arch_atomic_ interface to generic code
Paul E. McKenney
Joel Fernandes (Google)
Linus Torvalds