The s390 CPU Measurement sampling facility has an overflow condition
which fires when all entries in a SBD are used.
The measurement alert interrupt is triggered and reads out all samples
in this SDB. It then tests the successor SDB, if this SBD is not full,
the interrupt handler does not read any samples at all from this SDB
The design waits for the hardware to fill this SBD and then trigger
another meassurement alert interrupt.
This scheme works nicely until
an perf_event_overflow() function call discards the sample due to
a too high sampling rate.
The interrupt handler has logic to read out a partially filled SDB
when the perf event overflow condition in linux common code is met.
This causes the CPUM sampling measurement hardware and the PMU
device driver to operate on the same SBD's trailer entry.
This should not happen.
This can be seen here using this trace:
cpumsf_pmu_add: tear:0xb5286000
hw_perf_event_update: sdbt 0xb5286000 full 1 over 0 flush_all:0
hw_perf_event_update: sdbt 0xb5286008 full 0 over 0 flush_all:0
above shows 1. interrupt
hw_perf_event_update: sdbt 0xb5286008 full 1 over 0 flush_all:0
hw_perf_event_update: sdbt 0xb5286008 full 0 over 0 flush_all:0
above shows 2. interrupt
... this goes on fine until...
hw_perf_event_update: sdbt 0xb5286068 full 1 over 0 flush_all:0
perf_push_sample1: overflow
one or more samples read from the IRQ handler are rejected by
perf_event_overflow() and the IRQ handler advances to the next SDB
and modifies the trailer entry of a partially filled SDB.
hw_perf_event_update: sdbt 0xb5286070 full 0 over 0 flush_all:1
timestamp: 14:32:52.519953
Next time the IRQ handler is called for this SDB the trailer entry shows
an overflow count of 19 missed entries.
hw_perf_event_update: sdbt 0xb5286070 full 1 over 19 flush_all:1
timestamp: 14:32:52.970058
Remove access to a follow on SDB when event overflow happened.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Function perf_event_ever_overflow() and perf_event_account_interrupt()
are called every time samples are processed by the interrupt handler.
However function perf_event_account_interrupt() has checks to avoid being
flooded with interrupts (more then 1000 samples are received per
task_tick). Samples are then dropped and a PERF_RECORD_THROTTLED is
added to the perf data. The perf subsystem limit calculation is:
maximum sample frequency := 100000 --> 1 samples per 10 us
task_tick = 10ms = 10000us --> 1000 samples per task_tick
The work flow is
measurement_alert() uses SDBT head and each SBDT points to 511
SDB pages, each with 126 sample entries. After processing 8 SBDs
and for each valid sample calling:
perf_event_overflow()
perf_event_account_interrupts()
there is a considerable amount of samples being dropped, especially when
the sample frequency is very high and near the 100000 limit.
To avoid the high amount of samples being dropped near the end of a
task_tick time frame, increment the sampling interval in case of
dropped events. The CPU Measurement sampling facility on the s390
supports only intervals, specifiing how many CPU cycles have to be
executed before a sample is generated. Increase the interval when the
samples being generated hit the task_tick limit.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Each SBDT is located at a 4KB page and contains 512 entries.
Each entry of a SDBT points to a SDB, a 4KB page containing
sampled data. The last entry is a link to another SDBT page.
When an event is created the function sequence executed is:
__hw_perf_event_init()
+--> allocate_buffers()
+--> realloc_sampling_buffers()
+---> alloc_sample_data_block()
Both functions realloc_sampling_buffers() and
alloc_sample_data_block() allocate pages and the allocation
can fail. This is handled correctly and all allocated
pages are freed and error -ENOMEM is returned to the
top calling function. Finally the event is not created.
Once the event has been created, the amount of initially
allocated SDBT and SDB can be too low. This is detected
during measurement interrupt handling, where the amount
of lost samples is calculated. If the number of lost samples
is too high considering sampling frequency and already allocated
SBDs, the number of SDBs is enlarged during the next execution
of cpumsf_pmu_enable().
If more SBDs need to be allocated, functions
realloc_sampling_buffers()
+---> alloc-sample_data_block()
are called to allocate more pages. Page allocation may fail
and the returned error is ignored. A SDBT and SDB setup
already exists.
However the modified SDBTs and SDBs might end up in a situation
where the first entry of an SDBT does not point to an SDB,
but another SDBT, basicly an SBDT without payload.
This can not be handled by the interrupt handler, where an SDBT
must have at least one entry pointing to an SBD.
Add a check to avoid SDBTs with out payload (SDBs) when enlarging
the buffer setup.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The macro TEAR_REG() saves the last used SDBT address
in the perf_hw_event structure. This is also done
by function hw_reset_registers() which is a one-liner
and simply uses macro TEAR_REG(). Remove function
hw_reset_registers(), which is only used one time and use
macro TEAR_REG() instead. This macro is used throughout
the code anyway.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
In interrupt handling the function extend_sampling_buffer()
is called after checking for a possibly extension.
This check is not necessary as the called function itself
performs this check again.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Replace hard coded function names in debug statements
by the "%s ...", __func__ construct suggested by checkpatch.pl
script. Use consistent debug print format of the form variable
blank value. Also add leading 0x for all hex values.
Print allocated page addresses consistantly as hex numbers
with leading 0x.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
When starting the CPU Measurement sampling facility using
qsi() function, this function may return an error value.
This error value is referenced in the else part of the
if statement to dump its value in a debug statement.
Right now this value is always zero because it has not been
assigned a value.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Replace hard coded function names in debug statements
by the "%s ...", __func__ construct suggested by checkpatch.pl
script.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Use consistant debug print format of the form variable
blank value. Also add leading 0x for all hex values.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Fix indentation in the s390 CPU Measuement Facility
sampling device dirver.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
s390 IBM z15 introduces a check if the CPU Mesurement Facility
sampling is temporarily unavailable. If this is the case return -EBUSY
and abort the setup of CPU Measuement facility sampling.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
A perf_event can be set up to deliver overflow notifications
via SIGIO signal. The setup of the event is:
1. create event with perf_event_open()
2. assign it a signal for I/O notification with fcntl()
3. Install signal handler and consume samples
The initial setup of perf_event_open() determines the
period/frequency time span needed to elapse before each signal
is delivered to the user process.
While the event is active, system call
ioctl(.., PERF_EVENT_IOC_PERIOD, value) can be used the change
the frequency/period time span of the active event.
The remaining signal handler invocations honour the new value.
This does not work on s390. In fact the time span does not change
regardless of ioctl's third argument 'value'. The call succeeds
but the time span does not change.
Support this behavior and make it common with other platforms.
This is achieved by changing the interval value of the sampling
control block accordingly and feed this new value every time
the event is enabled using pmu_event_enable().
Before this change the interval value was set only once at
pmu_event_add() and never changed.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Rewrite some lines to match line length and replace
format string 0x%x to %#x. Add and remove blank line.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
When pmu::setup_aux() is called the coresight PMU needs to know which
sink to use for the session by looking up the information in the
event's attr::config2 field.
As such simply replace the cpu information by the complete perf_event
structure and change all affected customers.
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Reviewed-by: Suzuki Poulouse <suzuki.poulose@arm.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: linux-s390@vger.kernel.org
Link: http://lkml.kernel.org/r/20190131184714.20388-2-mathieu.poirier@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Previously, the attribute entry for diagnostic sampling was added
if authorized. Otherwise, the array of struct attribute contains
two NULL values.
Change this logic and reserve space for the attribute for diagnostic
sampling. If diagnostic sampling is authorized, add an entry in the
respective position in the array of struct attribute.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Suggested-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Return an error when the function debug_register() fails allocating
the debug handle.
Also remove the registered debug handle when the initialization fails
later on.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Pull core signal handling updates from Eric Biederman:
"It was observed that a periodic timer in combination with a
sufficiently expensive fork could prevent fork from every completing.
This contains the changes to remove the need for that restart.
This set of changes is split into several parts:
- The first part makes PIDTYPE_TGID a proper pid type instead
something only for very special cases. The part starts using
PIDTYPE_TGID enough so that in __send_signal where signals are
actually delivered we know if the signal is being sent to a a group
of processes or just a single process.
- With that prep work out of the way the logic in fork is modified so
that fork logically makes signals received while it is running
appear to be received after the fork completes"
* 'siginfo-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (22 commits)
signal: Don't send signals to tasks that don't exist
signal: Don't restart fork when signals come in.
fork: Have new threads join on-going signal group stops
fork: Skip setting TIF_SIGPENDING in ptrace_init_task
signal: Add calculate_sigpending()
fork: Unconditionally exit if a fatal signal is pending
fork: Move and describe why the code examines PIDNS_ADDING
signal: Push pid type down into complete_signal.
signal: Push pid type down into __send_signal
signal: Push pid type down into send_signal
signal: Pass pid type into do_send_sig_info
signal: Pass pid type into send_sigio_to_task & send_sigurg_to_task
signal: Pass pid type into group_send_sig_info
signal: Pass pid and pid type into send_sigqueue
posix-timers: Noralize good_sigevent
signal: Use PIDTYPE_TGID to clearly store where file signals will be sent
pid: Implement PIDTYPE_TGID
pids: Move the pgrp and session pid pointers from task_struct to signal_struct
kvm: Don't open code task_pid in kvm_vcpu_ioctl
pids: Compute task_tgid using signal->leader_pid
...
Processing the samples in the AUX-area by perf requires the computation
of respective time stamps. The time stamps used by perf are based on
the monotonic clock. To convert the TOD clock value contained in an
SDB to a monotonic clock value, the TOD clock base is required. Hence,
also save the TOD clock base in the SDB.
Suggested-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Everywhere except in the pid array we distinguish between a tasks pid and
a tasks tgid (thread group id). Even in the enumeration we want that
distinction sometimes so we have added __PIDTYPE_TGID. With leader_pid
we almost have an implementation of PIDTYPE_TGID in struct signal_struct.
Add PIDTYPE_TGID as a first class member of the pid_type enumeration and
into the pids array. Then remove the __PIDTYPE_TGID special case and the
leader_pid in signal_struct.
The net size increase is just an extra pointer added to struct pid and
an extra pair of pointers of an hlist_node added to task_struct.
The effect on code maintenance is the removal of a number of special
cases today and the potential to remove many more special cases as
PIDTYPE_TGID gets used to it's fullest. The long term potential
is allowing zombie thread group leaders to exit, which will remove
a lot more special cases in the code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Correct a trinity finding for the perf_event_open() system call with
a perf event attribute structure that uses a frequency but has the
sampling frequency set to zero. This causes a FP divide exception during
the sample rate initialization for the hardware sampling facility.
Fixes: 8c069ff4bd ("s390/perf: add support for the CPU-Measurement Sampling Facility")
Cc: stable@vger.kernel.org # 3.14+
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Now that the SPDX tag is in all arch/s390/kernel/ files, that identifies
the license in a specific and legally-defined manner. So the extra GPL
text wording can be removed as it is no longer needed at all.
This is done on a quest to remove the 700+ different ways that files in
the kernel describe the GPL license text. And there's unneeded stuff
like the address (sometimes incorrect) for the FSF which is never
needed.
No copyright headers or other non-license-description text was removed.
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
It's good to have SPDX identifiers in all files to make it easier to
audit the kernel tree for correct licenses.
Update the arch/s390/kernel/ files with the correct SPDX license
identifier based on the license text in the file itself. The SPDX
identifier is a legally binding shorthand, which can be used instead of
the full boiler plate text.
This work is based on a script and data from Thomas Gleixner, Philippe
Ombredanne, and Kate Stewart.
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Kate Stewart <kstewart@linuxfoundation.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The hardware sampler creates samples that are processed at a later
point in time. The PID and TID values of the perf samples that are
created for hardware samples are initialized with values from the
current task. Hence, the PID and TID values are not correct and
perf samples are associated with wrong processes.
The PID and TID values are obtained from the Host Program Parameter
(HPP) field in the basic-sampling data entries. These PIDs are
valid in the init PID namespace. Ensure that the PIDs in the perf
samples are resolved considering the PID namespace in which the
perf event was created.
To correct the PID and TID values in the created perf samples,
a special overflow handler is installed. It replaces the default
overflow handler and does not become effective if any other
overflow handler is used. With the special overflow handler most
of the perf samples are associated with the right processes.
For processes, that are no longer exist, the association might
still be wrong.
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The lpp instruction is used to place the PID of the current
task in the program-parameter (PP) register. The register
contents is then included in the sampling data entries.
The lpp instruction loads the PP register only when at least
one sampling function is enabled. Otherwise it is executed
as a no-op.
Linux calls lpp at context switch. If the context switch
happens before the sampler is enabled, the PP register is
empty. That means, the PID of the task that is sampled is
not stored in sampling data until the next context switch.
Hence, always call lpp when enabling the sampler.
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Previously, the cpum_sf PMU was registered even if there is no
sampling mode authorized. Add a check and register cpum_sf only
at least one sampling mode is authorized.
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Raw sample was implemented to export the diagnostic samples.
With having this achieved with AUX buffers, there is no requirement
for basic samples to export raw data. In particular, most basic
sampling information are consumed for creating the perf event sample.
Signed-off-by: Pu Hou <bjhoupu@linux.vnet.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Modify PMU callback to use AUX buffer for diagnostic mode sampling.
Basic-mode sampling still use orignal way.
Signed-off-by: Pu Hou <bjhoupu@linux.vnet.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Current implementation uses a private buffer for cpumf to dump samples.
Samples first go to this buffer. Then copy to ring buffer allocated
by perf core. With AUX buffer, this copy is not needed. AUX buffer is
shared and zero-copy mapped to user space. The trailer information at
the end of each SDB(sample data block) is also exported to user space.
AUX buffer is used when diagnostic sampling mode is enabled.
This patch contains functions to setup/free AUX buffer or to begin/end
sampling per-cpu. Also include function called in interrupt to
collect samples.
Signed-off-by: Pu Hou <bjhoupu@linux.vnet.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Paul Burton reported that the nr_cpumask_bits check
within cpumsf_pmu_event_init() is not necessary.
Actually there is already a prior check within
perf_event_alloc(). Therefore remove the check.
Reported-by: Paul Burton <paul.burton@imgtec.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
A per-thread event could not be created correctly like below:
perf record --per-thread -e rB0000 -- sleep 1
Error:
The sys_perf_event_open() syscall returned with 19 (No such device) for event (rB0000).
/bin/dmesg may provide additional information.
No CONFIG_PERF_EVENTS=y kernel support configured?
This bug was introduced by:
commit c311c79799
Author: Alexey Dobriyan <adobriyan@gmail.com>
Date: Mon May 8 15:56:15 2017 -0700
cpumask: make "nr_cpumask_bits" unsigned
If a per-thread event is not attached to any CPU, the cpu field
in struct perf_event is -1. The above commit converts the CPU number
to unsigned int, which result in an illegal CPU number.
Fixes: c311c79799 ("cpumask: make "nr_cpumask_bits" unsigned")
Cc: <stable@vger.kernel.org> # v4.12+
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Pu Hou <bjhoupu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The P sample bit indicates problem state and not PER.
Fixes: commit a752598254 ("s390: rename struct psw_bits members")
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Rename a couple of the struct psw_bits members so it is more obvious
for what they are good. Initially I thought using the single character
names from the PoP would be sufficient and obvious, but admittedly
that is not true.
The current implementation is not easy to use, if one has to look into
the source file to figure out which member represents the 'per' bit
(which is the 'r' member).
Therefore rename the members to sane names that are identical to the
uapi psw mask defines:
r -> per
i -> io
e -> ext
t -> dat
m -> mcheck
w -> wait
p -> pstate
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Bit searching functions accept "unsigned long" indices but
"nr_cpumask_bits" is "int" which is signed, so inevitable sign
extensions occur on x86_64. Those MOVSX are #1 MOVSX bloat by number of
uses across whole kernel.
Change "nr_cpumask_bits" to unsigned, this number can't be negative
after all. It allows to do implicit zero-extension on x86_64 without
MOVSX.
Change signed comparisons into unsigned comparisons where necessary.
Other uses looks fine because it is either argument passed to a function
or comparison is already unsigned.
Net win on allyesconfig type of kernel: ~2.8 KB (!)
add/remove: 0/0 grow/shrink: 8/725 up/down: 93/-2926 (-2833)
function old new delta
xen_exit_mmap 691 735 +44
qstat_read 426 440 +14
__cpufreq_cooling_register 1678 1687 +9
trace_rb_cpu_prepare 447 455 +8
vermagic 54 60 +6
nfp_driver_version 54 60 +6
rcu_torture_stats_print 1147 1151 +4
find_next_push_cpu 267 269 +2
xen_irq_resume 961 960 -1
...
init_vp_index 946 906 -40
od_set_powersave_bias 328 281 -47
power_cpu_exit 193 139 -54
arch_show_interrupts 3538 3484 -54
select_idle_sibling 1558 1471 -87
Total: Before=158358910, After=158356077, chg -0.00%
Same arguments apply to "nr_cpu_ids" but I haven't yet found enough
courage to delve into this issue (and proper fix may require new type
"cpu_t" which is whole separate story).
Link: http://lkml.kernel.org/r/20170309205322.GA1728@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are three different code levels in regard to the identification
of guest samples. They differ in the way the LPP instruction is used.
1) Old kernels without the LPP instruction. The guest program parameter
is always zero.
2) Newer kernels load the process pid into the program parameter with LPP.
The guest program parameter is non-zero if the guest executes in a
process != idle.
3) The latest kernels load ((1UL << 31) | pid) with LPP to make the value
non-zero even for the idle task. The guest program parameter is non-zero
if the guest is running.
All kernels load the process pid to CR4 on context switch. The CPU sampling
code uses the value in CR4 to decide between guest and host samples in case
the guest program parameter is zero. The three cases:
1) CR4==pid, gpp==0
2) CR4==pid, gpp==pid
3) CR4==pid, gpp==((1UL << 31) | pid)
The load-control instruction to load the pid into CR4 is expensive and the
goal is to remove it. To distinguish the host CR4 from the guest pid for
the idle process the maximum value 0xffff for the PASN is used.
This adds a fourth case for a guest OS with an updated kernel:
4) CR4==0xffff, gpp=((1UL << 31) | pid)
The host kernel will have CR4==0xffff and will use (gpp!=0 || CR4!==0xffff)
to identify guest samples. This works nicely with all 4 cases, the only
possible issue would be a guest with an old kernel (gpp==0) and a process
pid of 0xffff. Well, don't do that..
Suggested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
When the state names got added a script was used to add the extra argument
to the calls. The script basically converted the state constant to a
string, but the cleanup to convert these strings into meaningful ones did
not happen.
Replace all the useless strings with 'subsys/xxx/yyy:state' strings which
are used in all the other places already.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Link: http://lkml.kernel.org/r/20161221192112.085444152@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Newer hardware provides the level of virtualization that a particular
sample belongs to. Use that information and fall back to the old
heuristics if the sample does not contain that information.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Use the psw_bits macro and simplify the code. The generated code is
also better since it doesn't contain any conditional branches anymore.
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Pull smp hotplug updates from Thomas Gleixner:
"This is the next part of the hotplug rework.
- Convert all notifiers with a priority assigned
- Convert all CPU_STARTING/DYING notifiers
The final removal of the STARTING/DYING infrastructure will happen
when the merge window closes.
Another 700 hundred line of unpenetrable maze gone :)"
* 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (70 commits)
timers/core: Correct callback order during CPU hot plug
leds/trigger/cpu: Move from CPU_STARTING to ONLINE level
powerpc/numa: Convert to hotplug state machine
arm/perf: Fix hotplug state machine conversion
irqchip/armada: Avoid unused function warnings
ARC/time: Convert to hotplug state machine
clocksource/atlas7: Convert to hotplug state machine
clocksource/armada-370-xp: Convert to hotplug state machine
clocksource/exynos_mct: Convert to hotplug state machine
clocksource/arm_global_timer: Convert to hotplug state machine
rcu: Convert rcutree to hotplug state machine
KVM/arm/arm64/vgic-new: Convert to hotplug state machine
smp/cfd: Convert core to hotplug state machine
x86/x2apic: Convert to CPU hotplug state machine
profile: Convert to hotplug state machine
timers/core: Convert to hotplug state machine
hrtimer: Convert to hotplug state machine
x86/tboot: Convert to hotplug state machine
arm64/armv8 deprecated: Convert to hotplug state machine
hwtracing/coresight-etm4x: Convert to hotplug state machine
...
Pull networking updates from David Miller:
1) Unified UDP encapsulation offload methods for drivers, from
Alexander Duyck.
2) Make DSA binding more sane, from Andrew Lunn.
3) Support QCA9888 chips in ath10k, from Anilkumar Kolli.
4) Several workqueue usage cleanups, from Bhaktipriya Shridhar.
5) Add XDP (eXpress Data Path), essentially running BPF programs on RX
packets as soon as the device sees them, with the option to mirror
the packet on TX via the same interface. From Brenden Blanco and
others.
6) Allow qdisc/class stats dumps to run lockless, from Eric Dumazet.
7) Add VLAN support to b53 and bcm_sf2, from Florian Fainelli.
8) Simplify netlink conntrack entry layout, from Florian Westphal.
9) Add ipv4 forwarding support to mlxsw spectrum driver, from Ido
Schimmel, Yotam Gigi, and Jiri Pirko.
10) Add SKB array infrastructure and convert tun and macvtap over to it.
From Michael S Tsirkin and Jason Wang.
11) Support qdisc packet injection in pktgen, from John Fastabend.
12) Add neighbour monitoring framework to TIPC, from Jon Paul Maloy.
13) Add NV congestion control support to TCP, from Lawrence Brakmo.
14) Add GSO support to SCTP, from Marcelo Ricardo Leitner.
15) Allow GRO and RPS to function on macsec devices, from Paolo Abeni.
16) Support MPLS over IPV4, from Simon Horman.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1622 commits)
xgene: Fix build warning with ACPI disabled.
be2net: perform temperature query in adapter regardless of its interface state
l2tp: Correctly return -EBADF from pppol2tp_getname.
net/mlx5_core/health: Remove deprecated create_singlethread_workqueue
net: ipmr/ip6mr: update lastuse on entry change
macsec: ensure rx_sa is set when validation is disabled
tipc: dump monitor attributes
tipc: add a function to get the bearer name
tipc: get monitor threshold for the cluster
tipc: make cluster size threshold for monitoring configurable
tipc: introduce constants for tipc address validation
net: neigh: disallow transition to NUD_STALE if lladdr is unchanged in neigh_update()
MAINTAINERS: xgene: Add driver and documentation path
Documentation: dtb: xgene: Add MDIO node
dtb: xgene: Add MDIO node
drivers: net: xgene: ethtool: Use phy_ethtool_gset and sset
drivers: net: xgene: Use exported functions
drivers: net: xgene: Enable MDIO driver
drivers: net: xgene: Add backward compatibility
drivers: net: phy: xgene: Add MDIO driver
...
This patch adds support for non-linear data on raw records. It
extends raw records to have one or multiple fragments that will
be written linearly into the ring slot, where each fragment can
optionally have a custom callback handler to walk and extract
complex, possibly non-linear data.
If a callback handler is provided for a fragment, then the new
__output_custom() will be used instead of __output_copy() for
the perf_output_sample() part. perf_prepare_sample() does all
the size calculation only once, so perf_output_sample() doesn't
need to redo the same work anymore, meaning real_size and padding
will be cached in the raw record. The raw record becomes 32 bytes
in size without holes; to not increase it further and to avoid
doing unnecessary recalculations in fast-path, we can reuse
next pointer of the last fragment, idea here is borrowed from
ZERO_OR_NULL_PTR(), which should keep the perf_output_sample()
path for PERF_SAMPLE_RAW minimal.
This facility is needed for BPF's event output helper as a first
user that will, in a follow-up, add an additional perf_raw_frag
to its perf_raw_record in order to be able to more efficiently
dump skb context after a linear head meta data related to it.
skbs can be non-linear and thus need a custom output function to
dump buffers. Currently, the skb data needs to be copied twice;
with the help of __output_custom() this work only needs to be
done once. Future users could be things like XDP/BPF programs
that work on different context though and would thus also have
a different callback function.
The few users of raw records are adapted to initialize their frag
data from the raw record itself, no change in behavior for them.
The code is based upon a PoC diff provided by Peter Zijlstra [1].
[1] http://thread.gmane.org/gmane.linux.network/421294
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Install the callbacks via the state machine and let the core invoke the
callbacks on the already online CPUs.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-s390@vger.kernel.org
Cc: rt@linutronix.de
Link: http://lkml.kernel.org/r/20160713153334.518084858@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Now that the oprofile sampling code is gone there is only one user of
the sampling facility left. Therefore the reserve and release
functions can be removed.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Since commit 3b9d6da67e ("cpu/hotplug: Fix rollback during error-out
in __cpu_disable()") it is ensured that callbacks of CPU_ONLINE and
CPU_DOWN_PREPARE are processed on the hotplugged CPU. Due to this SMP
function calls are no longer required.
Replace smp_call_function_single() with a direct call of
setup_pmc_cpu(). To keep the calling convention, interrupts are
explicitly disabled around the call.
Cc: linux-s390@vger.kernel.org
Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The cpumf_pmu_notfier() hotplug callback lacks handling of the
CPU_DOWN_FAILED case. That means, if CPU_DOWN_PREPARE failes, the PMC
of the CPU is not setup again. Furthermore the CPU_ONLINE_FROZEN case
will never be processed because of masking the switch expression with
CPU_TASKS_FROZEN.
Add handling for CPU_DOWN_FAILED transition to setup the PMC of the
CPU. Remove CPU_ONLINE_FROZEN case.
Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
Acked-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
commit e22cf8ca6f ("s390/cpumf: rework program parameter setting
to detect guest samples") requires guest changes to get proper
guest/host. We can do better: We can use the primary asn value,
which is set on all Linux variants to compare this with the host
pp value.
We now have the following cases:
1. Guest using PP
host sample: gpp == 0, asn == hpp --> host
guest sample: gpp != 0 --> guest
2. Guest not using PP
host sample: gpp == 0, asn == hpp --> host
guest sample: gpp == 0, asn != hpp --> guest
As soon as the host no longer sets CR4, we must back out
this heuristics - let's add a comment in switch_to.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The program parameter can be used to mark hardware samples with
some token. Previously, it was used to mark guest samples only.
Improve the program parameter doubleword by combining two parts,
the leftmost LPP part and the rightmost PID part. Set the PID
part for processes by using the task PID.
To distinguish host and guest samples for the kernel (PID part
is zero), the guest must always set the program paramater to a
non-zero value. Use the leftmost bit in the LPP part of the
program parameter to be able to detect guest kernel samples.
[brueckner@linux.vnet.ibm.com]: Split __LC_CURRENT and introduced
__LC_LPP. Corrected __LC_CURRENT users and adjusted assembler parts.
And updated the commit message accordingly.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Right now we use the address of the sie control block as tag for
the sampling data. This is hard to get for users. Let's just use
the PID of the cpu thread to mark the hardware samples.
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Most code already uses consts for the struct kernel_param_ops,
sweep the kernel for the last offending stragglers. Other than
include/linux/moduleparam.h and kernel/params.c all other changes
were generated with the following Coccinelle SmPL patch. Merge
conflicts between trees can be handled with Coccinelle.
In the future git could get Coccinelle merge support to deal with
patch --> fail --> grammar --> Coccinelle --> new patch conflicts
automatically for us on patches where the grammar is available and
the patch is of high confidence. Consider this a feature request.
Test compiled on x86_64 against:
* allnoconfig
* allmodconfig
* allyesconfig
@ const_found @
identifier ops;
@@
const struct kernel_param_ops ops = {
};
@ const_not_found depends on !const_found @
identifier ops;
@@
-struct kernel_param_ops ops = {
+const struct kernel_param_ops ops = {
};
Generated-by: Coccinelle SmPL
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Junio C Hamano <gitster@pobox.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: cocci@systeme.lip6.fr
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The SF_CYCLES_BASIC_DIAG is always registered even if it is turned of in the
current hardware configuration. Because diagnostic-sampling is typically not
turned on in the hardware configuration, do not register this perf event by
default. Enable it only if the diagnostic-sampling function is authorized.
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Pull s390 updates from Martin Schwidefsky:
"The most notable change for this pull request is the ftrace rework
from Heiko. It brings a small performance improvement and the ground
work to support a new gcc option to replace the mcount blocks with a
single nop.
Two new s390 specific system calls are added to emulate user space
mmio for PCI, an artifact of the how PCI memory is accessed.
Two patches for the memory management with changes to common code.
For KVM mm_forbids_zeropage is added which disables the empty zero
page for an mm that is used by a KVM process. And an optimization,
pmdp_get_and_clear_full is added analog to ptep_get_and_clear_full.
Some micro optimization for the cmpxchg and the spinlock code.
And as usual bug fixes and cleanups"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (46 commits)
s390/cputime: fix 31-bit compile
s390/scm_block: make the number of reqs per HW req configurable
s390/scm_block: handle multiple requests in one HW request
s390/scm_block: allocate aidaw pages only when necessary
s390/scm_block: use mempool to manage aidaw requests
s390/eadm: change timeout value
s390/mm: fix memory leak of ptlock in pmd_free_tlb
s390: use local symbol names in entry[64].S
s390/ptrace: always include vector registers in core files
s390/simd: clear vector register pointer on fork/clone
s390: translate cputime magic constants to macros
s390/idle: convert open coded idle time seqcount
s390/idle: add missing irq off lockdep annotation
s390/debug: avoid function call for debug_sprintf_*
s390/kprobes: fix instruction copy for out of line execution
s390: remove diag 44 calls from cpu_relax()
s390/dasd: retry partition detection
s390/dasd: fix list corruption for sleep_on requests
s390/dasd: fix infinite term I/O loop
s390/dasd: remove unused code
...
Thomas Richter