- Thoroughly rewrite the data structures that implement perf task context handling,
with the goal of fixing various quirks and unfeatures both in already merged,
and in upcoming proposed code.
The old data structure is the per task and per cpu perf_event_contexts:
task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context
^ | ^ | ^
`---------------------------------' | `--> pmu ---'
v ^
perf_event ------'
In this new design this is replaced with a single task context and
a single CPU context, plus intermediate data-structures:
task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context
^ | ^ ^
`---------------------------' | |
| | perf_cpu_pmu_context <--.
| `----. ^ |
| | | |
| v v |
| ,--> perf_event_pmu_context |
| | |
| | |
v v |
perf_event ---> pmu ----------------'
[ See commit bd27568117 for more details. ]
This rewrite was developed by Peter Zijlstra and Ravi Bangoria.
- Optimize perf_tp_event()
- Update the Intel uncore PMU driver, extending it with UPI topology discovery
on various hardware models.
- Misc fixes & cleanups
Signed-off-by: Ingo Molnar <mingo@kernel.org>
-----BEGIN PGP SIGNATURE-----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=KNaI
-----END PGP SIGNATURE-----
Merge tag 'perf-core-2022-12-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull perf events updates from Ingo Molnar:
- Thoroughly rewrite the data structures that implement perf task
context handling, with the goal of fixing various quirks and
unfeatures both in already merged, and in upcoming proposed code.
The old data structure is the per task and per cpu
perf_event_contexts:
task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context
^ | ^ | ^
`---------------------------------' | `--> pmu ---'
v ^
perf_event ------'
In this new design this is replaced with a single task context and a
single CPU context, plus intermediate data-structures:
task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context
^ | ^ ^
`---------------------------' | |
| | perf_cpu_pmu_context <--.
| `----. ^ |
| | | |
| v v |
| ,--> perf_event_pmu_context |
| | |
| | |
v v |
perf_event ---> pmu ----------------'
[ See commit bd27568117 for more details. ]
This rewrite was developed by Peter Zijlstra and Ravi Bangoria.
- Optimize perf_tp_event()
- Update the Intel uncore PMU driver, extending it with UPI topology
discovery on various hardware models.
- Misc fixes & cleanups
* tag 'perf-core-2022-12-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (25 commits)
perf/x86/intel/uncore: Fix reference count leak in __uncore_imc_init_box()
perf/x86/intel/uncore: Fix reference count leak in snr_uncore_mmio_map()
perf/x86/intel/uncore: Fix reference count leak in hswep_has_limit_sbox()
perf/x86/intel/uncore: Fix reference count leak in sad_cfg_iio_topology()
perf/x86/intel/uncore: Make set_mapping() procedure void
perf/x86/intel/uncore: Update sysfs-devices-mapping file
perf/x86/intel/uncore: Enable UPI topology discovery for Sapphire Rapids
perf/x86/intel/uncore: Enable UPI topology discovery for Icelake Server
perf/x86/intel/uncore: Get UPI NodeID and GroupID
perf/x86/intel/uncore: Enable UPI topology discovery for Skylake Server
perf/x86/intel/uncore: Generalize get_topology() for SKX PMUs
perf/x86/intel/uncore: Disable I/O stacks to PMU mapping on ICX-D
perf/x86/intel/uncore: Clear attr_update properly
perf/x86/intel/uncore: Introduce UPI topology type
perf/x86/intel/uncore: Generalize IIO topology support
perf/core: Don't allow grouping events from different hw pmus
perf/amd/ibs: Make IBS a core pmu
perf: Fix function pointer case
perf/x86/amd: Remove the repeated declaration
perf: Fix possible memleak in pmu_dev_alloc()
...
- Factor out handle_write() function and simplify 3215 console
write operation.
- When 3170 terminal emulator is connected to the 3215 console
driver the boot time could be very long due to limited buffer
space or missing operator input. Add con3215_drop command line
parameter and con3215_drop sysfs attribute file to instruct
the kernel drop console data when such conditions are met.
- Fix white space errors in 3215 console driver.
- Move enum paiext_mode definition to a header file and rename
it to paievt_mode to indicate this is now used for several
events. Rename PAI_MODE_COUNTER to PAI_MODE_COUNTING to make
consistent with PAI_MODE_SAMPLING.
- Simplify the logic of PMU pai_crypto mapped buffer reference
counter and make it consistent with PMU pai_ext.
- Rename PMU pai_crypto mapped buffer structure member users
to active_events to make it consistent with PMU pai_ext.
- Enable HUGETLB_PAGE_OPTIMIZE_VMEMMAP configuration option.
This results in saving of 12K per 1M hugetlb page (~1.2%)
and 32764K per 2G hugetlb page (~1.6%).
- Use generic serial.h, bugs.h, shmparam.h and vga.h header
files and scrap s390-specific versions.
- The generic percpu setup code does not expect the s390-like
implementation and emits a warning. To get rid of that warning
and provide sane CPU-to-node and CPU-to-CPU distance mappings
implementat a minimal version of setup_per_cpu_areas().
- Use kstrtobool() instead of strtobool() for re-IPL sysfs device
attributes.
- Avoid unnecessary lookup of a pointer to MSI descriptor when
setting IRQ affinity for a PCI device.
- Get rid of "an incompatible function type cast" warning by
changing debug_sprintf_format_fn() function prototype so it
matches the debug_format_proc_t function type.
- Remove unused info_blk_hdr__pcpus() and get_page_state()
functions.
- Get rid of clang "unused unused insn cache ops function"
warning by moving s390_insn definition to a private header.
- Get rid of clang "unused function" warning by making function
raw3270_state_final() only available if CONFIG_TN3270_CONSOLE
is enabled.
- Use kstrobool() to parse sclp_con_drop parameter to make it
identical to the con3215_drop parameter and allow passing
values like "yes" and "true".
- Use sysfs_emit() for all SCLP sysfs show functions, which is
the current standard way to generate output strings.
- Make SCLP con_drop sysfs attribute also writable and allow to
change its value during runtime. This makes SCLP console drop
handling consistent with the 3215 device driver.
- Virtual and physical addresses are indentical on s390. However,
there is still a confusion when pointers are directly casted to
physical addresses or vice versa. Use correct address converters
virt_to_phys() and phys_to_virt() for s390 channel IO drivers.
- Support for power managemant has been removed from s390 since
quite some time. Remove unused power managemant code from the
appldata device driver.
- Allow memory tools like KASAN see memory accesses from the
checksum code. Switch to GENERIC_CSUM if KASAN is enabled,
just like x86 does.
- Add support of ECKD DASDs disks so it could be used as boot
and dump devices.
- Follow checkpatch recommendations and use octal values instead
of S_IRUGO and S_IWUSR for dump device attributes in sysfs.
- Changes to vx-insn.h do not cause a recompile of C files that
use asm(".include \"asm/vx-insn.h\"\n") magic to access vector
instruction macros from inline assemblies. Add wrapper include
header file to avoid this problem.
- Use vector instruction macros instead of byte patterns to
increase register validation routine readability.
- The current machine check register validation handling does not
take into account various scenarios and might lead to killing a
wrong user process or potentially ignore corrupted FPU registers.
Simplify logic of the machine check handler and stop the whole
machine if the previous context was kerenel mode. If the previous
context was user mode, kill the current task.
- Introduce sclp_emergency_printk() function which can be used to
emit a message in emergency cases. It is supposed to be used in
cases where regular console device drivers may not work anymore,
e.g. unrecoverable machine checks.
Keep the early Service-Call Control Block so it can also be used
after initdata has been freed to allow sclp_emergency_printk()
implementation.
- In case a system will be stopped because of an unrecoverable
machine check error print the machine check interruption code
to give a hint of what went wrong.
- Move storage error checking from the assembly entry code to C
in order to simplify machine check handling. Enter the handler
with DAT turned on, which simplifies the entry code even more.
- The machine check extended save areas are allocated using
a private "nmi_save_areas" slab cache which guarantees a
required power-of-two alignment. Get rid of that cache in
favour of kmalloc().
-----BEGIN PGP SIGNATURE-----
iI0EABYIADUWIQQrtrZiYVkVzKQcYivNdxKlNrRb8AUCY5ckrhccYWdvcmRlZXZA
bGludXguaWJtLmNvbQAKCRDNdxKlNrRb8NlrAQD8NCLeEAkhGCRnzdTyngExCrzV
Mw//cEnksUkIPqalJgEArbyFjGh05ecNaiDQduH8Gh94/qOhGE4obMdTgMWq7QY=
=3aou
-----END PGP SIGNATURE-----
Merge tag 's390-6.2-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Pull s390 updates from Alexander Gordeev:
- Factor out handle_write() function and simplify 3215 console write
operation
- When 3170 terminal emulator is connected to the 3215 console driver
the boot time could be very long due to limited buffer space or
missing operator input. Add con3215_drop command line parameter and
con3215_drop sysfs attribute file to instruct the kernel drop console
data when such conditions are met
- Fix white space errors in 3215 console driver
- Move enum paiext_mode definition to a header file and rename it to
paievt_mode to indicate this is now used for several events. Rename
PAI_MODE_COUNTER to PAI_MODE_COUNTING to make consistent with
PAI_MODE_SAMPLING
- Simplify the logic of PMU pai_crypto mapped buffer reference counter
and make it consistent with PMU pai_ext
- Rename PMU pai_crypto mapped buffer structure member users to
active_events to make it consistent with PMU pai_ext
- Enable HUGETLB_PAGE_OPTIMIZE_VMEMMAP configuration option. This
results in saving of 12K per 1M hugetlb page (~1.2%) and 32764K per
2G hugetlb page (~1.6%)
- Use generic serial.h, bugs.h, shmparam.h and vga.h header files and
scrap s390-specific versions
- The generic percpu setup code does not expect the s390-like
implementation and emits a warning. To get rid of that warning and
provide sane CPU-to-node and CPU-to-CPU distance mappings implementat
a minimal version of setup_per_cpu_areas()
- Use kstrtobool() instead of strtobool() for re-IPL sysfs device
attributes
- Avoid unnecessary lookup of a pointer to MSI descriptor when setting
IRQ affinity for a PCI device
- Get rid of "an incompatible function type cast" warning by changing
debug_sprintf_format_fn() function prototype so it matches the
debug_format_proc_t function type
- Remove unused info_blk_hdr__pcpus() and get_page_state() functions
- Get rid of clang "unused unused insn cache ops function" warning by
moving s390_insn definition to a private header
- Get rid of clang "unused function" warning by making function
raw3270_state_final() only available if CONFIG_TN3270_CONSOLE is
enabled
- Use kstrobool() to parse sclp_con_drop parameter to make it identical
to the con3215_drop parameter and allow passing values like "yes" and
"true"
- Use sysfs_emit() for all SCLP sysfs show functions, which is the
current standard way to generate output strings
- Make SCLP con_drop sysfs attribute also writable and allow to change
its value during runtime. This makes SCLP console drop handling
consistent with the 3215 device driver
- Virtual and physical addresses are indentical on s390. However, there
is still a confusion when pointers are directly casted to physical
addresses or vice versa. Use correct address converters
virt_to_phys() and phys_to_virt() for s390 channel IO drivers
- Support for power managemant has been removed from s390 since quite
some time. Remove unused power managemant code from the appldata
device driver
- Allow memory tools like KASAN see memory accesses from the checksum
code. Switch to GENERIC_CSUM if KASAN is enabled, just like x86 does
- Add support of ECKD DASDs disks so it could be used as boot and dump
devices
- Follow checkpatch recommendations and use octal values instead of
S_IRUGO and S_IWUSR for dump device attributes in sysfs
- Changes to vx-insn.h do not cause a recompile of C files that use
asm(".include \"asm/vx-insn.h\"\n") magic to access vector
instruction macros from inline assemblies. Add wrapper include header
file to avoid this problem
- Use vector instruction macros instead of byte patterns to increase
register validation routine readability
- The current machine check register validation handling does not take
into account various scenarios and might lead to killing a wrong user
process or potentially ignore corrupted FPU registers. Simplify logic
of the machine check handler and stop the whole machine if the
previous context was kerenel mode. If the previous context was user
mode, kill the current task
- Introduce sclp_emergency_printk() function which can be used to emit
a message in emergency cases. It is supposed to be used in cases
where regular console device drivers may not work anymore, e.g.
unrecoverable machine checks
Keep the early Service-Call Control Block so it can also be used
after initdata has been freed to allow sclp_emergency_printk()
implementation
- In case a system will be stopped because of an unrecoverable machine
check error print the machine check interruption code to give a hint
of what went wrong
- Move storage error checking from the assembly entry code to C in
order to simplify machine check handling. Enter the handler with DAT
turned on, which simplifies the entry code even more
- The machine check extended save areas are allocated using a private
"nmi_save_areas" slab cache which guarantees a required power-of-two
alignment. Get rid of that cache in favour of kmalloc()
* tag 's390-6.2-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (38 commits)
s390/nmi: get rid of private slab cache
s390/nmi: move storage error checking back to C, enter with DAT on
s390/nmi: print machine check interruption code before stopping system
s390/sclp: introduce sclp_emergency_printk()
s390/sclp: keep sclp_early_sccb
s390/nmi: rework register validation handling
s390/nmi: use vector instruction macros instead of byte patterns
s390/vx: add vx-insn.h wrapper include file
s390/ipl: use octal values instead of S_* macros
s390/ipl: add eckd dump support
s390/ipl: add eckd support
vfio/ccw: identify CCW data addresses as physical
vfio/ccw: sort out physical vs virtual pointers usage
s390/checksum: support GENERIC_CSUM, enable it for KASAN
s390/appldata: remove power management callbacks
s390/cio: sort out physical vs virtual pointers usage
s390/sclp: allow to change sclp_console_drop during runtime
s390/sclp: convert to use sysfs_emit()
s390/sclp: use kstrobool() to parse sclp_con_drop parameter
s390/3270: make raw3270_state_final() depend on CONFIG_TN3270_CONSOLE
...
There have been various issues and limitations with the way perf uses
(task) contexts to track events. Most notable is the single hardware
PMU task context, which has resulted in a number of yucky things (both
proposed and merged).
Notably:
- HW breakpoint PMU
- ARM big.little PMU / Intel ADL PMU
- Intel Branch Monitoring PMU
- AMD IBS PMU
- S390 cpum_cf PMU
- PowerPC trace_imc PMU
*Current design:*
Currently we have a per task and per cpu perf_event_contexts:
task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context
^ | ^ | ^
`---------------------------------' | `--> pmu ---'
v ^
perf_event ------'
Each task has an array of pointers to a perf_event_context. Each
perf_event_context has a direct relation to a PMU and a group of
events for that PMU. The task related perf_event_context's have a
pointer back to that task.
Each PMU has a per-cpu pointer to a per-cpu perf_cpu_context, which
includes a perf_event_context, which again has a direct relation to
that PMU, and a group of events for that PMU.
The perf_cpu_context also tracks which task context is currently
associated with that CPU and includes a few other things like the
hrtimer for rotation etc.
Each perf_event is then associated with its PMU and one
perf_event_context.
*Proposed design:*
New design proposed by this patch reduce to a single task context and
a single CPU context but adds some intermediate data-structures:
task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context
^ | ^ ^
`---------------------------' | |
| | perf_cpu_pmu_context <--.
| `----. ^ |
| | | |
| v v |
| ,--> perf_event_pmu_context |
| | |
| | |
v v |
perf_event ---> pmu ----------------'
With the new design, perf_event_context will hold all events for all
pmus in the (respective pinned/flexible) rbtrees. This can be achieved
by adding pmu to rbtree key:
{cpu, pmu, cgroup, group_index}
Each perf_event_context carries a list of perf_event_pmu_context which
is used to hold per-pmu-per-context state. For example, it keeps track
of currently active events for that pmu, a pmu specific task_ctx_data,
a flag to tell whether rotation is required or not etc.
Additionally, perf_cpu_pmu_context is used to hold per-pmu-per-cpu
state like hrtimer details to drive the event rotation, a pointer to
perf_event_pmu_context of currently running task and some other
ancillary information.
Each perf_event is associated to it's pmu, perf_event_context and
perf_event_pmu_context.
Further optimizations to current implementation are possible. For
example, ctx_resched() can be optimized to reschedule only single pmu
events.
Much thanks to Ravi for picking this up and pushing it towards
completion.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Co-developed-by: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20221008062424.313-1-ravi.bangoria@amd.com
Commit 838d9bb62d ("perf: Use sample_flags for raw_data")
changed the way the raw data of an event is collected.
Adjust the PMU pai_ext to the new scheme.
Fixes: 838d9bb62d ("perf: Use sample_flags for raw_data")
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Move enum definition to header file. This is done in preparation
for a follow on patch where this enum will be used in another source
file.
Also change the enum name from paiext_mode to paievt_mode
to indicate this enum is now used for several events.
Make naming consistent and rename PAI_MODE_COUNTER to PAI_MODE_COUNTING.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
PMU device driver perf_paiext supports Processor Activity
Instrumentation Extension (PAIE1), available with IBM z16:
- maps a 512 byte block to lowcore address 0x1508 called PAIE1 control
block.
- maps a 1024 byte block at PAIE1 control block entry with index 2.
- uses control register bit 14 to enable PAIE1 control block lookup.
- turn PAIE1 nnpa counting on and off by setting bit 63 in
PAIE1 control block entry with index 2.
- creates a sample with raw data on each context switch out when
at context switch some mapped counters have a value of nonzero.
This device driver only supports CPU wide context, no task context
is allowed.
Support for counting:
- one or more counters can be specified using
perf stat -e pai_ext/xxx/
where xxx stands for the counter event name. Multiple invocation
of this command is possible. The counter names are listed in
/sys/devices/pai_ext/events directory.
- one special counters can be specified using
perf stat -e pai_ext/NNPA_ALL/
which returns the sum of all incremented nnpa counters.
- multiple counting events can run in parallel.
Support for Sampling:
- one event pai_ext/NNPA_ALL/ is reserved for sampling.
The event collects data at context switch out and saves them in
the ring buffer.
- no multiple invocations are possible.
The PAIE1 nnpa counter events are system wide. No task context is
supported. Therefore some restrictions documented in function
paiext_busy() apply.
Extend qpaci assembly instruction to query supported memory mapped nnpa
counters. It returns the number of counters (no holes allowed in that
range).
PAIE1 nnpa counter events can not be created when a CPU hot plug
add is processed. This means a CPU hot plug add does not get
the necessary PAIE1 event to record PAIE1 nnpa counter increments
on the newly added CPU. CPU hot plug remove removes the event and
terminates the counting of PAIE1 counters immediately.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Reviewed-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Linus Torvalds
Peter Zijlstra
Thomas Richter