linux-dev/arch/s390/kernel/perf_cpum_cf.c, branch linus/master

s390: raise minimum supported machine generation to z10

2022-03-10T14:58:17Z

Machine generations up to z9 (released in May 2006) have been officially out of service for several years now (z9 end of service - January 31, 2019). No distributions build kernels supporting those old machine generations anymore, except Debian, which seems to pick the oldest supported generation. The team supporting Debian on s390 has been notified about the change. Raising minimum supported machine generation to z10 helps to reduce maintenance cost and effectively remove code, which is not getting enough testing coverage due to lack of older hardware and distributions support. Besides that this unblocks some optimization opportunities and allows to use wider instruction set in asm files for future features implementation. Due to this change spectre mitigation and usercopy implementations could be drastically simplified and many newer instructions could be converted from ".insn" encoding to instruction names. Acked-by: Ilya Leoshkevich Reviewed-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390/cpumf: cpum_cf PMU displays invalid value after hotplug remove

2021-11-08T13:17:49Z

When a CPU is hotplugged while the perf stat -e cycles command is running, a wrong (very large) value is displayed immediately after the CPU removal: Check the values, shouldn't be too high as in time counts unit events 1.001101919 29261846 cycles 2.002454499 17523405 cycles 3.003659292 24361161 cycles 4.004816983 18446744073638406144 cycles 5.005671647 cycles ... The CPU hotplug off took place after 3 seconds. The issue is the read of the event count value after 4 seconds when the CPU is not available and the read of the counter returns an error. This is treated as a counter value of zero. This results in a very large value (0 - previous_value). Fix this by detecting the hotplugged off CPU and report 0 instead of a very large number. Cc: stable@vger.kernel.org Fixes: a029a4eab39e ("s390/cpumf: Allow concurrent access for CPU Measurement Counter Facility") Reported-by: Sumanth Korikkar Signed-off-by: Thomas Richter Reviewed-by: Sumanth Korikkar Signed-off-by: Vasily Gorbik

s390/cpumf: Allow multiple processes to access /dev/hwc

2021-10-26T13:21:29Z

Commit a029a4eab39e ("s390/cpumf: Allow concurrent access for CPU Measurement Counter Facility") added CPU Measurement counter facility access to multiple consumers. It allows concurrent access to the CPU Measurement counter facility via several perf_event_open() system call invocations and via ioctl() system call of device /dev/hwc. However the access via device /dev/hwc was exclusive, only one process was able to open this device. The patch removes this restriction. Now multiple invocations of lshwc can execute in parallel. They can access different CPUs and counter sets or CPUs and counter set can overlap. Signed-off-by: Thomas Richter Suggested-by: Heiko Carstens Reviewed-by: Sumanth Korikkar Acked-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390/cpum_cf: move array from header to C file

2021-09-07T11:38:41Z

Move array from header to C file to avoid that it gets defined in every C file where the header is included: In file included from arch/s390/kernel/perf_cpum_cf_common.c:19: ./arch/s390/include/asm/cpu_mcf.h:27:18: warning: ‘cpumf_ctr_ctl’ defined but not used [-Wunused-const-variable=] 27 | static const u64 cpumf_ctr_ctl[CPUMF_CTR_SET_MAX] = { | ^~~~~~~~~~~~~ Signed-off-by: Heiko Carstens

s390: replace deprecated CPU-hotplug functions

2021-08-05T12:10:53Z

The functions get_online_cpus() and put_online_cpus() have been deprecated during the CPU hotplug rework. They map directly to cpus_read_lock() and cpus_read_unlock(). Replace deprecated CPU-hotplug functions with the official version. The behavior remains unchanged. Signed-off-by: Sebastian Andrzej Siewior Link: https://lore.kernel.org/r/20210803141621.780504-5-bigeasy@linutronix.de Signed-off-by: Heiko Carstens

s390/cpumf: fix semicolon.cocci warnings

2021-07-20T15:59:40Z

arch/s390/kernel/perf_cpum_cf.c:748:2-3: Unneeded semicolon Remove unneeded semicolon. Generated by: scripts/coccinelle/misc/semicolon.cocci Fixes: a029a4eab39e ("s390/cpumf: Allow concurrent access for CPU Measurement Counter Facility") CC: Thomas Richter Reported-by: kernel test robot Signed-off-by: kernel test robot Signed-off-by: Heiko Carstens

s390/cpumf: Allow concurrent access for CPU Measurement Counter Facility

2021-07-05T10:44:23Z

Commit cf6acb8bdb1d ("s390/cpumf: Add support for complete counter set extraction") allows access to the CPU Measurement Counter Facility via character device /dev/hwctr. The access was exclusive via this device or via perf_event_open() system call. Only one path at a time was permitted. The CPU Measurement Counter Facility device driver blocked access to other processes. This patch removes this restriction and allows concurrent access to the CPU Measurement Counter Facility from multiple processes at the same time via perf_event_open() SVC and via /dev/hwctr device. The access via /dev/hwctr device is still exclusive, only one process is allowed to access this device. This patch - moves the /dev/hwctr device access from file perf_cpum_cf_diag.c. to file perf_cpum_cf.c. - use only one trace buffer .../s390dbf/cpum_cf. - remove cfset_csd structure and includes its members it into the structure cpu_cf_events. This results in one data structure and simplifies the access. - rework function familiy ctr_set_enable, ctr_set_disable, ctr_set_start and ctr_set_stop which operate on a counter set number. Now they operate on a counter set bit mask. - move CF_DIAG event functionality to file perf_cpum_cf.c. It now contains the complete functionality of the CPU Measurement Counter Facility: - Performance measurement support for counters using perf stat. - Support for complete counter set extraction with device /dev/hwctr. - Support for counter set extraction event CF_DIAG attached to samples using perf record. - removes file perf_cpum_cf_diag.c Signed-off-by: Thomas Richter Reviewed-by: Sumanth Korikkar Signed-off-by: Vasily Gorbik

s390/cpumf: remove WARN_ON_ONCE in counter start handler

2021-06-07T15:06:58Z

Remove some WARN_ON_ONCE() warnings when a counter is started. Each counter is installed function calls event_sched_in() --> cpumf_pmu_add(..., PERF_EF_START). This is done after the event has been created using perf_pmu_event_init() which verifies the counter is valid. Member hwc->config must be valid at this point. Function cpumf_pmu_start(..., PERF_EF_RELOAD) is called from function cpumf_pmu_add() for counter events. All other invocations of cpumf_pmu_start(..., PERF_EF_RELOAD) are from the performance subsystem for sampling events. Signed-off-by: Thomas Richter Acked-by: Sumanth Korikkar Signed-off-by: Vasily Gorbik

s390/cpumf: remove counter transaction call backs

2021-06-07T15:06:58Z

The command 'perf stat -e cycles ...' triggers the following function sequence in the CPU Measurement Facility counter device driver: perf_pmu_event_init() __hw_perf_event_init() validate_ctr_auth() validate_ctr_version() During event creation, the counter number is checked in functions validate_ctr_auth() and validate_ctr_version() to verify it is a valid counter and supported by the hardware. If this is not the case, both functions return an error and the event is not created. System call perf_event_open() returns an error in this case. Later on the event is installed in the kernel event subsystem and the driver functions cpumf_pmu_add() and cpumf_pmu_commit_txn() are called to install the counter event by the hardware. Since both events have been verified at event creation, there is no need to re-evaluate the authorization state. This can not change since on * LPARs the authorization change requires a restart of the LPAR (and thus a reboot of the kernel) * DPMs can not take resources away, just add them. Also the sequence of CPU Measurement facility counter device driver calls is cpumf_pmu_start_txn cpumf_pmu_add cpumf_pmu_start cpumf_pmu_commit_txn for every single event. Which means the condition in cpumf_pmu_add() is never met and validate_ctr_auth() is never called. This leaves the counter device driver transaction functions with just one task: start_txn: Verify a transaction is not in flight and call perf_pmu_disable() cancel_txn, commit_txn: Verify a transaction is in flight and call perf_pmu_enable() The same functionality is provided by the default transaction handling functions in kernel/events/core.c. Use those by removing the counter device driver private call back functions. Suggested-by: Sumanth Korikkar Signed-off-by: Thomas Richter Reviewed-by: Sumanth Korikkar Signed-off-by: Vasily Gorbik

s390/cpumf: remove call to perf_event_update_userpage

2021-04-30T15:17:01Z

The function cpumf_pmu_add and cpumf_pmu_del call function perf_event_update_userpage(). This calls is obsolete, the calls add and delete a counter event. Counter events do not sample data and the event->rb member to access the sampling ring buffer is always NULL. The function perf_event_update_userpage() simply returns in this case. Signed-off-by: Thomas Richter Acked-by : Sumanth Korikkar Signed-off-by: Heiko Carstens