This reverts commit 83f44ae0f8.
Currently we store initial stacktrace entry twice for non-HW ot_regs, which
means callers that fail perf_hw_regs(regs) condition in perf_callchain_kernel.
It's easy to reproduce this bpftrace:
# bpftrace -e 'tracepoint:sched:sched_process_exec { print(kstack()); }'
Attaching 1 probe...
bprm_execve+1767
bprm_execve+1767
do_execveat_common.isra.0+425
__x64_sys_execve+56
do_syscall_64+133
entry_SYSCALL_64_after_hwframe+118
When perf_callchain_kernel calls unwind_start with first_frame, AFAICS
we do not skip regs->ip, but it's added as part of the unwind process.
Hence reverting the extra perf_callchain_store for non-hw regs leg.
I was not able to bisect this, so I'm not really sure why this was needed
in v5.2 and why it's not working anymore, but I could see double entries
as far as v5.10.
I did the test for both ORC and framepointer unwind with and without the
this fix and except for the initial entry the stacktraces are the same.
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20251104215405.168643-2-jolsa@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>
When running "perf mem record" command on CWF, the below KASAN
global-out-of-bounds warning is seen.
==================================================================
BUG: KASAN: global-out-of-bounds in cmt_latency_data+0x176/0x1b0
Read of size 4 at addr ffffffffb721d000 by task dtlb/9850
Call Trace:
kasan_report+0xb8/0xf0
cmt_latency_data+0x176/0x1b0
setup_arch_pebs_sample_data+0xf49/0x2560
intel_pmu_drain_arch_pebs+0x577/0xb00
handle_pmi_common+0x6c4/0xc80
The issue is caused by below code in __grt_latency_data(). The code
tries to access x86_hybrid_pmu structure which doesn't exist on
non-hybrid platform like CWF.
WARN_ON_ONCE(hybrid_pmu(event->pmu)->pmu_type == hybrid_big)
So add is_hybrid() check before calling this WARN_ON_ONCE to fix the
global-out-of-bounds access issue.
Fixes: 090262439f ("perf/x86/intel: Rename model-specific pebs_latency_data functions")
Reported-by: Xudong Hao <xudong.hao@intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Zide Chen <zide.chen@intel.com>
Cc: stable@vger.kernel.org
Link: https://patch.msgid.link/20251028064214.1451968-1-dapeng1.mi@linux.intel.com
Along with the introduction Perfmon v6, pmu counters could be
incontinuous, like fixed counters on CWF, only fixed counters 0-3 and
5-7 are supported, there is no fixed counter 4 on CWF. To accommodate
this change, archPerfmonExt CPUID (0x23) leaves are introduced to
enumerate the true-view of counters bitmap.
Current perf code already supports archPerfmonExt CPUID and uses
counters-bitmap to enumerate HW really supported counters, but
x86_pmu_show_pmu_cap() still only dumps the absolute counter number
instead of true-view bitmap, it's out-dated and may mislead readers.
So dump counters true-view bitmap in x86_pmu_show_pmu_cap() and
opportunistically change the dump sequence and words.
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Link: https://lore.kernel.org/r/20250820023032.17128-8-dapeng1.mi@linux.intel.com
When running perf_fuzzer on PTL, sometimes the below "unchecked MSR
access error" is seen when accessing IA32_PMC_x_CFG_B MSRs.
[ 55.611268] unchecked MSR access error: WRMSR to 0x1986 (tried to write 0x0000000200000001) at rIP: 0xffffffffac564b28 (native_write_msr+0x8/0x30)
[ 55.611280] Call Trace:
[ 55.611282] <TASK>
[ 55.611284] ? intel_pmu_config_acr+0x87/0x160
[ 55.611289] intel_pmu_enable_acr+0x6d/0x80
[ 55.611291] intel_pmu_enable_event+0xce/0x460
[ 55.611293] x86_pmu_start+0x78/0xb0
[ 55.611297] x86_pmu_enable+0x218/0x3a0
[ 55.611300] ? x86_pmu_enable+0x121/0x3a0
[ 55.611302] perf_pmu_enable+0x40/0x50
[ 55.611307] ctx_resched+0x19d/0x220
[ 55.611309] __perf_install_in_context+0x284/0x2f0
[ 55.611311] ? __pfx_remote_function+0x10/0x10
[ 55.611314] remote_function+0x52/0x70
[ 55.611317] ? __pfx_remote_function+0x10/0x10
[ 55.611319] generic_exec_single+0x84/0x150
[ 55.611323] smp_call_function_single+0xc5/0x1a0
[ 55.611326] ? __pfx_remote_function+0x10/0x10
[ 55.611329] perf_install_in_context+0xd1/0x1e0
[ 55.611331] ? __pfx___perf_install_in_context+0x10/0x10
[ 55.611333] __do_sys_perf_event_open+0xa76/0x1040
[ 55.611336] __x64_sys_perf_event_open+0x26/0x30
[ 55.611337] x64_sys_call+0x1d8e/0x20c0
[ 55.611339] do_syscall_64+0x4f/0x120
[ 55.611343] entry_SYSCALL_64_after_hwframe+0x76/0x7e
On PTL, GP counter 0 and 1 doesn't support auto counter reload feature,
thus it would trigger a #GP when trying to write 1 on bit 0 of CFG_B MSR
which requires to enable auto counter reload on GP counter 0.
The root cause of causing this issue is the check for auto counter
reload (ACR) counter mask from user space is incorrect in
intel_pmu_acr_late_setup() helper. It leads to an invalid ACR counter
mask from user space could be set into hw.config1 and then written into
CFG_B MSRs and trigger the MSR access warning.
e.g., User may create a perf event with ACR counter mask (config2=0xcb),
and there is only 1 event created, so "cpuc->n_events" is 1.
The correct check condition should be "i + idx >= cpuc->n_events"
instead of "i + idx > cpuc->n_events" (it looks a typo). Otherwise,
the counter mask would traverse twice and an invalid "cpuc->assign[1]"
bit (bit 0) is set into hw.config1 and cause MSR accessing error.
Besides, also check if the ACR counter mask corresponding events are
ACR events. If not, filter out these counter mask. If a event is not a
ACR event, it could be scheduled to an HW counter which doesn't support
ACR. It's invalid to add their counter index in ACR counter mask.
Furthermore, remove the WARN_ON_ONCE() since it's easily triggered as
user could set any invalid ACR counter mask and the warning message
could mislead users.
Fixes: ec980e4fac ("perf/x86/intel: Support auto counter reload")
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Link: https://lore.kernel.org/r/20250820023032.17128-3-dapeng1.mi@linux.intel.com
After the commit 'd971342d38bf ("perf/x86/intel: Decouple BTS
initialization from PEBS initialization")' is introduced, x86_pmu.bts
would initialized in bts_init() which is hooked by arch_initcall().
Whereas init_hw_perf_events() is hooked by early_initcall(). Once the
core PMU is initialized, nmi watchdog initialization is called
immediately before bts_init() is called. It leads to the BTS buffer is
not really initialized since bts_init() is not called and x86_pmu.bts is
still false at that time. Worse, BTS buffer would never be initialized
then unless all core PMU events are freed and reserve_ds_buffers()
is called again.
Thus aligning with init_hw_perf_events(), use early_initcall() to hook
bts_init() to ensure x86_pmu.bts is initialized before nmi watchdog
initialization.
Fixes: d971342d38 ("perf/x86/intel: Decouple BTS initialization from PEBS initialization")
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Link: https://lore.kernel.org/r/20250820023032.17128-2-dapeng1.mi@linux.intel.com
The Panther Lake supports CBOX, MC, sNCU, and HBO uncore PMON.
The CBOX is similar to Lunar Lake. The only difference is the number of
CBOX.
The other three uncore PMON can be retrieved from the discovery table.
The global control register resides in the sNCU. The global freeze bit
is set by default. It must be cleared before monitoring any uncore
counters.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Link: https://lore.kernel.org/r/20250707201750.616527-4-kan.liang@linux.intel.com
Starting from the Panther Lake, the discovery table mechanism is also
supported in client platforms. The difference is that the portal of the
global discovery table is retrieved from an MSR.
The layout of discovery tables are the same as the server platforms.
Factor out __parse_discovery_table() to parse discover tables.
The uncore PMON is Die scope. Need to parse the discovery tables for
each die.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Link: https://lore.kernel.org/r/20250707201750.616527-2-kan.liang@linux.intel.com
The perf_fuzzer found a hard-lockup crash on a RaptorLake machine:
Oops: general protection fault, maybe for address 0xffff89aeceab400: 0000
CPU: 23 UID: 0 PID: 0 Comm: swapper/23
Tainted: [W]=WARN
Hardware name: Dell Inc. Precision 9660/0VJ762
RIP: 0010:native_read_pmc+0x7/0x40
Code: cc e8 8d a9 01 00 48 89 03 5b cd cc cc cc cc 0f 1f ...
RSP: 000:fffb03100273de8 EFLAGS: 00010046
....
Call Trace:
<TASK>
icl_update_topdown_event+0x165/0x190
? ktime_get+0x38/0xd0
intel_pmu_read_event+0xf9/0x210
__perf_event_read+0xf9/0x210
CPUs 16-23 are E-core CPUs that don't support the perf metrics feature.
The icl_update_topdown_event() should not be invoked on these CPUs.
It's a regression of commit:
f9bdf1f953 ("perf/x86/intel: Avoid disable PMU if !cpuc->enabled in sample read")
The bug introduced by that commit is that the is_topdown_event() function
is mistakenly used to replace the is_topdown_count() call to check if the
topdown functions for the perf metrics feature should be invoked.
Fix it.
Fixes: f9bdf1f953 ("perf/x86/intel: Avoid disable PMU if !cpuc->enabled in sample read")
Closes: https://lore.kernel.org/lkml/352f0709-f026-cd45-e60c-60dfd97f73f3@maine.edu/
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Vince Weaver <vincent.weaver@maine.edu>
Cc: stable@vger.kernel.org # v6.15+
Link: https://lore.kernel.org/r/20250612143818.2889040-1-kan.liang@linux.intel.com
The MSR offset calculations in intel_pmu_config_acr() are buggy.
To calculate fixed counter MSR addresses in intel_pmu_config_acr(),
the HW counter index "idx" is subtracted by INTEL_PMC_IDX_FIXED.
This leads to the ACR mask value of fixed counters to be incorrectly
saved to the positions of GP counters in acr_cfg_b[], e.g.
For fixed counter 0, its ACR counter mask should be saved to
acr_cfg_b[32], but it's saved to acr_cfg_b[0] incorrectly.
Fix this issue.
[ mingo: Clarified & improved the changelog. ]
Fixes: ec980e4fac ("perf/x86/intel: Support auto counter reload")
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250529080236.2552247-2-dapeng1.mi@linux.intel.com
Pull core x86 updates from Ingo Molnar:
"Boot code changes:
- A large series of changes to reorganize the x86 boot code into a
better isolated and easier to maintain base of PIC early startup
code in arch/x86/boot/startup/, by Ard Biesheuvel.
Motivation & background:
| Since commit
|
| c88d71508e ("x86/boot/64: Rewrite startup_64() in C")
|
| dated Jun 6 2017, we have been using C code on the boot path in a way
| that is not supported by the toolchain, i.e., to execute non-PIC C
| code from a mapping of memory that is different from the one provided
| to the linker. It should have been obvious at the time that this was a
| bad idea, given the need to sprinkle fixup_pointer() calls left and
| right to manipulate global variables (including non-pointer variables)
| without crashing.
|
| This C startup code has been expanding, and in particular, the SEV-SNP
| startup code has been expanding over the past couple of years, and
| grown many of these warts, where the C code needs to use special
| annotations or helpers to access global objects.
This tree includes the first phase of this work-in-progress x86
boot code reorganization.
Scalability enhancements and micro-optimizations:
- Improve code-patching scalability (Eric Dumazet)
- Remove MFENCEs for X86_BUG_CLFLUSH_MONITOR (Andrew Cooper)
CPU features enumeration updates:
- Thorough reorganization and cleanup of CPUID parsing APIs (Ahmed S.
Darwish)
- Fix, refactor and clean up the cacheinfo code (Ahmed S. Darwish,
Thomas Gleixner)
- Update CPUID bitfields to x86-cpuid-db v2.3 (Ahmed S. Darwish)
Memory management changes:
- Allow temporary MMs when IRQs are on (Andy Lutomirski)
- Opt-in to IRQs-off activate_mm() (Andy Lutomirski)
- Simplify choose_new_asid() and generate better code (Borislav
Petkov)
- Simplify 32-bit PAE page table handling (Dave Hansen)
- Always use dynamic memory layout (Kirill A. Shutemov)
- Make SPARSEMEM_VMEMMAP the only memory model (Kirill A. Shutemov)
- Make 5-level paging support unconditional (Kirill A. Shutemov)
- Stop prefetching current->mm->mmap_lock on page faults (Mateusz
Guzik)
- Predict valid_user_address() returning true (Mateusz Guzik)
- Consolidate initmem_init() (Mike Rapoport)
FPU support and vector computing:
- Enable Intel APX support (Chang S. Bae)
- Reorgnize and clean up the xstate code (Chang S. Bae)
- Make task_struct::thread constant size (Ingo Molnar)
- Restore fpu_thread_struct_whitelist() to fix
CONFIG_HARDENED_USERCOPY=y (Kees Cook)
- Simplify the switch_fpu_prepare() + switch_fpu_finish() logic (Oleg
Nesterov)
- Always preserve non-user xfeatures/flags in __state_perm (Sean
Christopherson)
Microcode loader changes:
- Help users notice when running old Intel microcode (Dave Hansen)
- AMD: Do not return error when microcode update is not necessary
(Annie Li)
- AMD: Clean the cache if update did not load microcode (Boris
Ostrovsky)
Code patching (alternatives) changes:
- Simplify, reorganize and clean up the x86 text-patching code (Ingo
Molnar)
- Make smp_text_poke_batch_process() subsume
smp_text_poke_batch_finish() (Nikolay Borisov)
- Refactor the {,un}use_temporary_mm() code (Peter Zijlstra)
Debugging support:
- Add early IDT and GDT loading to debug relocate_kernel() bugs
(David Woodhouse)
- Print the reason for the last reset on modern AMD CPUs (Yazen
Ghannam)
- Add AMD Zen debugging document (Mario Limonciello)
- Fix opcode map (!REX2) superscript tags (Masami Hiramatsu)
- Stop decoding i64 instructions in x86-64 mode at opcode (Masami
Hiramatsu)
CPU bugs and bug mitigations:
- Remove X86_BUG_MMIO_UNKNOWN (Borislav Petkov)
- Fix SRSO reporting on Zen1/2 with SMT disabled (Borislav Petkov)
- Restructure and harmonize the various CPU bug mitigation methods
(David Kaplan)
- Fix spectre_v2 mitigation default on Intel (Pawan Gupta)
MSR API:
- Large MSR code and API cleanup (Xin Li)
- In-kernel MSR API type cleanups and renames (Ingo Molnar)
PKEYS:
- Simplify PKRU update in signal frame (Chang S. Bae)
NMI handling code:
- Clean up, refactor and simplify the NMI handling code (Sohil Mehta)
- Improve NMI duration console printouts (Sohil Mehta)
Paravirt guests interface:
- Restrict PARAVIRT_XXL to 64-bit only (Kirill A. Shutemov)
SEV support:
- Share the sev_secrets_pa value again (Tom Lendacky)
x86 platform changes:
- Introduce the <asm/amd/> header namespace (Ingo Molnar)
- i2c: piix4, x86/platform: Move the SB800 PIIX4 FCH definitions to
<asm/amd/fch.h> (Mario Limonciello)
Fixes and cleanups:
- x86 assembly code cleanups and fixes (Uros Bizjak)
- Misc fixes and cleanups (Andi Kleen, Andy Lutomirski, Andy
Shevchenko, Ard Biesheuvel, Bagas Sanjaya, Baoquan He, Borislav
Petkov, Chang S. Bae, Chao Gao, Dan Williams, Dave Hansen, David
Kaplan, David Woodhouse, Eric Biggers, Ingo Molnar, Josh Poimboeuf,
Juergen Gross, Malaya Kumar Rout, Mario Limonciello, Nathan
Chancellor, Oleg Nesterov, Pawan Gupta, Peter Zijlstra, Shivank
Garg, Sohil Mehta, Thomas Gleixner, Uros Bizjak, Xin Li)"
* tag 'x86-core-2025-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (331 commits)
x86/bugs: Fix spectre_v2 mitigation default on Intel
x86/bugs: Restructure ITS mitigation
x86/xen/msr: Fix uninitialized variable 'err'
x86/msr: Remove a superfluous inclusion of <asm/asm.h>
x86/paravirt: Restrict PARAVIRT_XXL to 64-bit only
x86/mm/64: Make 5-level paging support unconditional
x86/mm/64: Make SPARSEMEM_VMEMMAP the only memory model
x86/mm/64: Always use dynamic memory layout
x86/bugs: Fix indentation due to ITS merge
x86/cpuid: Rename hypervisor_cpuid_base()/for_each_possible_hypervisor_cpuid_base() to cpuid_base_hypervisor()/for_each_possible_cpuid_base_hypervisor()
x86/cpu/intel: Rename CPUID(0x2) descriptors iterator parameter
x86/cacheinfo: Rename CPUID(0x2) descriptors iterator parameter
x86/cpuid: Rename cpuid_get_leaf_0x2_regs() to cpuid_leaf_0x2()
x86/cpuid: Rename have_cpuid_p() to cpuid_feature()
x86/cpuid: Set <asm/cpuid/api.h> as the main CPUID header
x86/cpuid: Move CPUID(0x2) APIs into <cpuid/api.h>
x86/msr: Add rdmsrl_on_cpu() compatibility wrapper
x86/mm: Fix kernel-doc descriptions of various pgtable methods
x86/asm-offsets: Export certain 'struct cpuinfo_x86' fields for 64-bit asm use too
x86/boot: Defer initialization of VM space related global variables
...
Pull perf events updates from Ingo Molnar:
"Core & generic-arch updates:
- Add support for dynamic constraints and propagate it to the Intel
driver (Kan Liang)
- Fix & enhance driver-specific throttling support (Kan Liang)
- Record sample last_period before updating on the x86 and PowerPC
platforms (Mark Barnett)
- Make perf_pmu_unregister() usable (Peter Zijlstra)
- Unify perf_event_free_task() / perf_event_exit_task_context()
(Peter Zijlstra)
- Simplify perf_event_release_kernel() and perf_event_free_task()
(Peter Zijlstra)
- Allocate non-contiguous AUX pages by default (Yabin Cui)
Uprobes updates:
- Add support to emulate NOP instructions (Jiri Olsa)
- selftests/bpf: Add 5-byte NOP uprobe trigger benchmark (Jiri Olsa)
x86 Intel PMU enhancements:
- Support Intel Auto Counter Reload [ACR] (Kan Liang)
- Add PMU support for Clearwater Forest (Dapeng Mi)
- Arch-PEBS preparatory changes: (Dapeng Mi)
- Parse CPUID archPerfmonExt leaves for non-hybrid CPUs
- Decouple BTS initialization from PEBS initialization
- Introduce pairs of PEBS static calls
x86 AMD PMU enhancements:
- Use hrtimer for handling overflows in the AMD uncore driver
(Sandipan Das)
- Prevent UMC counters from saturating (Sandipan Das)
Fixes and cleanups:
- Fix put_ctx() ordering (Frederic Weisbecker)
- Fix irq work dereferencing garbage (Frederic Weisbecker)
- Misc fixes and cleanups (Changbin Du, Frederic Weisbecker, Ian
Rogers, Ingo Molnar, Kan Liang, Peter Zijlstra, Qing Wang, Sandipan
Das, Thorsten Blum)"
* tag 'perf-core-2025-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (60 commits)
perf/headers: Clean up <linux/perf_event.h> a bit
perf/uapi: Clean up <uapi/linux/perf_event.h> a bit
perf/uapi: Fix PERF_RECORD_SAMPLE comments in <uapi/linux/perf_event.h>
mips/perf: Remove driver-specific throttle support
xtensa/perf: Remove driver-specific throttle support
sparc/perf: Remove driver-specific throttle support
loongarch/perf: Remove driver-specific throttle support
csky/perf: Remove driver-specific throttle support
arc/perf: Remove driver-specific throttle support
alpha/perf: Remove driver-specific throttle support
perf/apple_m1: Remove driver-specific throttle support
perf/arm: Remove driver-specific throttle support
s390/perf: Remove driver-specific throttle support
powerpc/perf: Remove driver-specific throttle support
perf/x86/zhaoxin: Remove driver-specific throttle support
perf/x86/amd: Remove driver-specific throttle support
perf/x86/intel: Remove driver-specific throttle support
perf: Only dump the throttle log for the leader
perf: Fix the throttle logic for a group
perf/core: Add the is_event_in_freq_mode() helper to simplify the code
...
The main CPUID header <asm/cpuid.h> was originally a storefront for the
headers:
<asm/cpuid/api.h>
<asm/cpuid/leaf_0x2_api.h>
Now that the latter CPUID(0x2) header has been merged into the former,
there is no practical difference between <asm/cpuid.h> and
<asm/cpuid/api.h>.
Migrate all users to the <asm/cpuid/api.h> header, in preparation of
the removal of <asm/cpuid.h>.
Don't remove <asm/cpuid.h> just yet, in case some new code in -next
started using it.
Suggested-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Ahmed S. Darwish <darwi@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: John Ogness <john.ogness@linutronix.de>
Cc: x86-cpuid@lists.linux.dev
Link: https://lore.kernel.org/r/20250508150240.172915-3-darwi@linutronix.de
Currently, using PEBS-via-PT with a sample frequency instead of a sample
period, causes a segfault. For example:
BUG: kernel NULL pointer dereference, address: 0000000000000195
<NMI>
? __die_body.cold+0x19/0x27
? page_fault_oops+0xca/0x290
? exc_page_fault+0x7e/0x1b0
? asm_exc_page_fault+0x26/0x30
? intel_pmu_pebs_event_update_no_drain+0x40/0x60
? intel_pmu_pebs_event_update_no_drain+0x32/0x60
intel_pmu_drain_pebs_icl+0x333/0x350
handle_pmi_common+0x272/0x3c0
intel_pmu_handle_irq+0x10a/0x2e0
perf_event_nmi_handler+0x2a/0x50
That happens because intel_pmu_pebs_event_update_no_drain() assumes all the
pebs_enabled bits represent counter indexes, which is not always the case.
In this particular case, bits 60 and 61 are set for PEBS-via-PT purposes.
The behaviour of PEBS-via-PT with sample frequency is questionable because
although a PMI is generated (PEBS_PMI_AFTER_EACH_RECORD), the period is not
adjusted anyway.
Putting that aside, fix intel_pmu_pebs_event_update_no_drain() by passing
the mask of counter bits instead of 'size'. Note, prior to the Fixes
commit, 'size' would be limited to the maximum counter index, so the issue
was not hit.
Fixes: 722e42e45c ("perf/x86: Support counter mask")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Cc: linux-perf-users@vger.kernel.org
Link: https://lore.kernel.org/r/20250508134452.73960-1-adrian.hunter@intel.com
perf always allocates contiguous AUX pages based on aux_watermark.
However, this contiguous allocation doesn't benefit all PMUs. For
instance, ARM SPE and TRBE operate with virtual pages, and Coresight
ETR allocates a separate buffer. For these PMUs, allocating contiguous
AUX pages unnecessarily exacerbates memory fragmentation. This
fragmentation can prevent their use on long-running devices.
This patch modifies the perf driver to be memory-friendly by default,
by allocating non-contiguous AUX pages. For PMUs requiring contiguous
pages (Intel BTS and some Intel PT), the existing
PERF_PMU_CAP_AUX_NO_SG capability can be used. For PMUs that don't
require but can benefit from contiguous pages (some Intel PT), a new
capability, PERF_PMU_CAP_AUX_PREFER_LARGE, is added to maintain their
existing behavior.
Signed-off-by: Yabin Cui <yabinc@google.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: James Clark <james.clark@linaro.org>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20250508232642.148767-1-yabinc@google.com
Prepare to resolve conflicts with an upstream series of fixes that conflict
with pending x86 changes:
6f5bf947ba Merge tag 'its-for-linus-20250509' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Signed-off-by: Ingo Molnar <mingo@kernel.org>
__rdmsr() is the lowest level MSR write API, with native_rdmsr()
and native_rdmsrq() serving as higher-level wrappers around it.
#define native_rdmsr(msr, val1, val2) \
do { \
u64 __val = __rdmsr((msr)); \
(void)((val1) = (u32)__val); \
(void)((val2) = (u32)(__val >> 32)); \
} while (0)
static __always_inline u64 native_rdmsrq(u32 msr)
{
return __rdmsr(msr);
}
However, __rdmsr() continues to be utilized in various locations.
MSR APIs are designed for different scenarios, such as native or
pvops, with or without trace, and safe or non-safe. Unfortunately,
the current MSR API names do not adequately reflect these factors,
making it challenging to select the most appropriate API for
various situations.
To pave the way for improving MSR API names, convert __rdmsr()
uses to native_rdmsrq() to ensure consistent usage. Later, these
APIs can be renamed to better reflect their implications, such as
native or pvops, with or without trace, and safe or non-safe.
No functional change intended.
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <seanjc@google.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Uros Bizjak <ubizjak@gmail.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Link: https://lore.kernel.org/r/20250427092027.1598740-10-xin@zytor.com
__wrmsr() is the lowest level MSR write API, with native_wrmsr()
and native_wrmsrq() serving as higher-level wrappers around it:
#define native_wrmsr(msr, low, high) \
__wrmsr(msr, low, high)
#define native_wrmsrl(msr, val) \
__wrmsr((msr), (u32)((u64)(val)), \
(u32)((u64)(val) >> 32))
However, __wrmsr() continues to be utilized in various locations.
MSR APIs are designed for different scenarios, such as native or
pvops, with or without trace, and safe or non-safe. Unfortunately,
the current MSR API names do not adequately reflect these factors,
making it challenging to select the most appropriate API for
various situations.
To pave the way for improving MSR API names, convert __wrmsr()
uses to native_wrmsr{,q}() to ensure consistent usage. Later,
these APIs can be renamed to better reflect their implications,
such as native or pvops, with or without trace, and safe or
non-safe.
No functional change intended.
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <seanjc@google.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Uros Bizjak <ubizjak@gmail.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Link: https://lore.kernel.org/r/20250427092027.1598740-8-xin@zytor.com
When generating the MSR_IA32_PEBS_ENABLE value that will be loaded on
VM-Entry to a KVM guest, mask the value with the vCPU's desired PEBS_ENABLE
value. Consulting only the host kernel's host vs. guest masks results in
running the guest with PEBS enabled even when the guest doesn't want to use
PEBS. Because KVM uses perf events to proxy the guest virtual PMU, simply
looking at exclude_host can't differentiate between events created by host
userspace, and events created by KVM on behalf of the guest.
Running the guest with PEBS unexpectedly enabled typically manifests as
crashes due to a near-infinite stream of #PFs. E.g. if the guest hasn't
written MSR_IA32_DS_AREA, the CPU will hit page faults on address '0' when
trying to record PEBS events.
The issue is most easily reproduced by running `perf kvm top` from before
commit 7b100989b4 ("perf evlist: Remove __evlist__add_default") (after
which, `perf kvm top` effectively stopped using PEBS). The userspace side
of perf creates a guest-only PEBS event, which intel_guest_get_msrs()
misconstrues a guest-*owned* PEBS event.
Arguably, this is a userspace bug, as enabling PEBS on guest-only events
simply cannot work, and userspace can kill VMs in many other ways (there
is no danger to the host). However, even if this is considered to be bad
userspace behavior, there's zero downside to perf/KVM restricting PEBS to
guest-owned events.
Note, commit 854250329c ("KVM: x86/pmu: Disable guest PEBS temporarily
in two rare situations") fixed the case where host userspace is profiling
KVM *and* userspace, but missed the case where userspace is profiling only
KVM.
Fixes: c59a1f106f ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS")
Closes: https://lore.kernel.org/all/Z_VUswFkWiTYI0eD@do-x1carbon
Reported-by: Seth Forshee <sforshee@kernel.org>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Tested-by: "Seth Forshee (DigitalOcean)" <sforshee@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20250426001355.1026530-1-seanjc@google.com
The counter backwards may be observed in the PMI handler when
counters-snapshotting some non-precise events in the freq mode.
For the non-precise events, it's possible the counters-snapshotting
records a positive value for an overflowed PEBS event. Then the HW
auto-reload mechanism reset the counter to 0 immediately. Because the
pebs_event_reset is cleared in the freq mode, which doesn't set the
PERF_X86_EVENT_AUTO_RELOAD.
In the PMI handler, 0 will be read rather than the positive value
recorded in the counters-snapshotting record.
The counters-snapshotting case has to be specially handled. Since the
event value has been updated when processing the counters-snapshotting
record, only needs to set the new period for the counter via
x86_pmu_set_period().
Fixes: e02e9b0374 ("perf/x86/intel: Support PEBS counters snapshotting")
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250424134718.311934-6-kan.liang@linux.intel.com
A warning in intel_pmu_lbr_counters_reorder() may be triggered by below
perf command.
perf record -e "{cpu-clock,cycles/call-graph="lbr"/}" -- sleep 1
It's because the group is mistakenly treated as a branch counter group.
The hw.flags of the leader are used to determine whether a group is a
branch counters group. However, the hw.flags is only available for a
hardware event. The field to store the flags is a union type. For a
software event, it's a hrtimer. The corresponding bit may be set if the
leader is a software event.
For a branch counter group and other groups that have a group flag
(e.g., topdown, PEBS counters snapshotting, and ACR), the leader must
be a X86 event. Check the X86 event before checking the flag.
The patch only fixes the issue for the branch counter group.
The following patch will fix the other groups.
There may be an alternative way to fix the issue by moving the hw.flags
out of the union type. It should work for now. But it's still possible
that the flags will be used by other types of events later. As long as
that type of event is used as a leader, a similar issue will be
triggered. So the alternative way is dropped.
Fixes: 3374491619 ("perf/x86/intel: Support branch counters logging")
Closes: https://lore.kernel.org/lkml/20250412091423.1839809-1-luogengkun@huaweicloud.com/
Reported-by: Luo Gengkun <luogengkun@huaweicloud.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20250424134718.311934-2-kan.liang@linux.intel.com
Perf doesn't work at perf stat for hardware events on certain x86 platforms:
$perf stat -- sleep 1
Performance counter stats for 'sleep 1':
16.44 msec task-clock # 0.016 CPUs utilized
2 context-switches # 121.691 /sec
0 cpu-migrations # 0.000 /sec
54 page-faults # 3.286 K/sec
<not supported> cycles
<not supported> instructions
<not supported> branches
<not supported> branch-misses
The reason is that the check in x86_pmu_hw_config() for sampling events is
unexpectedly applied to counting events as well.
It should only impact x86 platforms with limit_period used for non-PEBS
events. For Intel platforms, it should only impact some older platforms,
e.g., HSW, BDW and NHM.
Fixes: 88ec7eedbb ("perf/x86: Fix low freqency setting issue")
Signed-off-by: Luo Gengkun <luogengkun@huaweicloud.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Link: https://lore.kernel.org/r/20250423064724.3716211-1-luogengkun@huaweicloud.com
Unlike L3 and DF counters, UMC counters (PERF_CTRs) set the Overflow bit
(bit 48) and saturate on overflow. A subsequent pmu->read() of the event
reports an incorrect accumulated count as there is no difference between
the previous and the current values of the counter.
To avoid this, inspect the current counter value and proactively reset
the corresponding PERF_CTR register on every pmu->read(). Combined with
the periodic reads initiated by the hrtimer, the counters never get a
chance saturate but the resolution reduces to 47 bits.
Fixes: 25e5684782 ("perf/x86/amd/uncore: Add memory controller support")
Signed-off-by: Sandipan Das <sandipan.das@amd.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Song Liu <song@kernel.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/dee9c8af2c6d66814cf4c6224529c144c620cf2c.1744906694.git.sandipan.das@amd.com
Uncore counters do not provide mechanisms like interrupts to report
overflows and the accumulated user-visible count is incorrect if there
is more than one overflow between two successive read requests for the
same event because the value of prev_count goes out-of-date for
calculating the correct delta.
To avoid this, start a hrtimer to periodically initiate a pmu->read() of
the active counters for keeping prev_count up-to-date. It should be
noted that the hrtimer duration should be lesser than the shortest time
it takes for a counter to overflow for this approach to be effective.
Signed-off-by: Sandipan Das <sandipan.das@amd.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/8ecf5fe20452da1cd19cf3ff4954d3e7c5137468.1744906694.git.sandipan.das@amd.com
Arch-PEBS retires IA32_PEBS_ENABLE and MSR_PEBS_DATA_CFG MSRs, so
intel_pmu_pebs_enable/disable() and intel_pmu_pebs_enable/disable_all()
are not needed to call for ach-PEBS.
To make the code cleaner, introduce static calls
x86_pmu_pebs_enable/disable() and x86_pmu_pebs_enable/disable_all()
instead of adding "x86_pmu.arch_pebs" check directly in these helpers.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20250415114428.341182-7-dapeng1.mi@linux.intel.com
