Pull perf tools fixes from Namhyung Kim:
"Two simple fixes for this cycle:
- Do not use separate debug files for Intel PT decoding
- Fix size of raw data in the PowerPC VPA DTL samples"
* tag 'perf-tools-fixes-for-v7.3-2026-09-07' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools:
perf powerpc-vpadtl: Fix raw_size of DTL samples
perf symbol: Do not use debug file as the binary type
In powerpc_vpadtl_sample(), raw_data of the synthetic sample points to a
struct powerpc_vpadtl_entry (48 bytes), but raw_size is set to
sizeof(record). record is a struct powerpc_vpadtl_entry pointer, so
sizeof(record) is the size of the pointer (8 bytes on 64-bit) rather
than the size of the record itself.
As a result, consumers that bound their access to raw_data by raw_size
only see or copy the first 8 bytes of each DTL entry instead of the full
record.
Use sizeof(*record) so that raw_size reflects the actual length of the
raw data.
Fixes: 8644834a48 ("perf powerpc: Process the DTL entries in queue and deliver samples")
Signed-off-by: Wang Yan <wangyan01@kylinos.cn>
Reviewed-by: Athira Rajeev <atrajeev@linux.ibm.com>
Reviewed-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
dso__load() sets the binary type of a DSO to the type of the first symbol
source found. For a DSO with a separate debug file linked via
.gnu-debuglink, that is DSO_BINARY_TYPE__DEBUGLINK, which makes
dso__get_filename() return the name of the debug file instead of the file
that was actually executed.
Consumers that need to read instruction bytes, such as Intel PT decoding
in 'perf script', then read from the debug file and produce wrong
instructions.
Prefer DSO_BINARY_TYPE__BUILD_ID_CACHE, and otherwise
DSO_BINARY_TYPE__SYSTEM_PATH_DSO, over debug-only types, which restores
the behaviour of using a file that contains the executed instructions.
This is a workaround. Properly separating the binary file used for
instructions from the file used for debug symbols is left for later.
Example:
Create a shared object with a separate .gnu_debuglink debug file. Note
that 'objcopy --only-keep-debug' leaves .text as NOBITS, so instructions
read from the debug file are zeros:
# cat > foo.c << EOF
unsigned long foo_work(unsigned long n)
{
unsigned long s = 0;
for (unsigned long i = 0; i < n; i++)
s = s * 31 + i;
return s;
}
EOF
# cat > main.c << EOF
#include <stdio.h>
unsigned long foo_work(unsigned long n);
int main(void)
{
printf("%lu\n", foo_work(1000));
return 0;
}
EOF
# gcc -g -O2 -shared -fPIC -o libfoo.so foo.c
# gcc -g -O2 -o main main.c -L. -lfoo -Wl,-rpath,'$ORIGIN'
# objcopy --only-keep-debug libfoo.so libfoo.so.debug
# objcopy --strip-debug libfoo.so
# objcopy --add-gnu-debuglink=libfoo.so.debug libfoo.so
# perf record -e intel_pt//u ./main
Note that branch samples must be requested, because it is the resolving
of the branch target symbol that causes dso__load() to be called, and
hence the binary type to be set, before the decoder walks the code.
With '--itrace=e' alone, nothing loads symbols for libfoo.so, the binary
type is left as DSO_BINARY_TYPE__NOT_FOUND, the correct file is read
anyway, and no errors are reported either way.
Before:
# perf.before script --itrace=be 2>&1 | grep "instruction trace error"
instruction trace error type 1 time 2350.467489498 cpu 9 pid 75634 tid 75634 ip 0x77d48480718f code 6: Trace doesn't match instruction
instruction trace error type 1 time 2350.467489832 cpu 9 pid 75634 tid 75634 ip 0x77d484807341 code 6: Trace doesn't match instruction
instruction trace error type 1 time 2350.467496412 cpu 9 pid 75634 tid 75634 ip 0x5b4de37a8074 code 6: Trace doesn't match instruction
instruction trace error type 1 time 2350.467593393 cpu 9 pid 75634 tid 75634 ip 0x77d4848070d0 code 6: Trace doesn't match instruction
instruction trace error type 1 time 2350.467593954 cpu 9 pid 75634 tid 75634 ip 0x77d4848075a8 code 6: Trace doesn't match instruction
instruction trace error type 1 time 2350.467595728 cpu 9 pid 75634 tid 75634 ip 0x77d4848324de code 6: Trace doesn't match instruction
6 instruction trace errors
After:
# perf script --itrace=be 2>&1 | grep "instruction trace error"
#
Fixes: 5363c30678 ("perf symbol: Set binary_type of dso when loading")
Reported-by: Todd Lipcon <tlipcon@google.com>
Closes: https://lore.kernel.org/all/CAGH6UiG=RJLqBU3kLu9XJciPyPO1HZkbAPERguVUMRuWQgqf=A@mail.gmail.com/
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Pull perf tools updates from Namhyung Kim:
"perf c2c:
- Add 'function view' in perf c2c report TUI (switched by pressing
'TAB' in the cacheline view) to organize samples around functions
rather than cachelines in 3-level hierarchy:
Level 1: Read-side function (sorted by estimated Cycles %)
Level 2: Contending writer functions (sorted by Store count)
Level 3: Shared cacheline addresses
Users can navigate the entries and fold/unfold using 'e' key. An
example output would look like below:
Shared Data Functions Table (19 entries, sorted on Cycles %)
Cycles Store
% count Function / Contending function / Cacheline
----------------------------------------------------------------------
+ 35.67% 876 + [k] cpupri_set
+ 24.31% 424 + [k] pull_rt_task
- 16.53% 555 - [k] dequeue_pushable_task
145 - [k] pull_rt_task
145 0xff2d0082809da080
139 - [k] enqueue_pushable_task
70 0xff2d00a2071f9640
69 0xff2d0082809da000
python module support:
- Extend "perf" python module so that it can be fully functional. The
goal is to run scripts directly, not by 'perf script' command. This
would give better performance as well as more control to build
standalone programs with UI.
- Add LiveSession helper (perf_live.py) to enable live event
collection directly from Python using perf.evlist and
perf.parse_events.
perf stat:
- Add --hide-zero-events option to suppress zero-count events
- Reject conflicting --field-separator and --json-output options
- Fix duplicate event output with --for-each-cgroup
perf sched latency:
- Add -H/--histogram and --hist-mode (log|linear) options to show
scheduler wait latency histograms
- Add --time option to filter analysis by time span in 'perf sched
latency'
ARM CoreSight:
- Synthesize callchains for instruction samples from CoreSight trace
using thread stack ('--itrace=g...')
- Support call indentation ('perf script -F +callindent') to display
call depth hierarchy on branch samples
- Decode ETE (Embedded Trace Extension) exception packets
Build system:
- Add 'make install-build-deps' target to install required packages
- Parallelize JSON and metric pre-computation in jevents.py for
faster builds
Vendor event/metric updates:
- Add Intel Nova Lake events and update tables for existing models
- Update AMD Zen 5 and Zen 6 core events
- Update Arm64 Tegra410 metrics and PowerPC hcalls
Internal changes and fixes:
- Harden trace-event and synthetic event parsing against corrupted
data
- Fix unwinding of multi-threaded processes in libdw unwinder
- Fix memory leaks in various commands and python bindings
- Speed up 'perf test' shell tests"
* tag 'perf-tools-for-v7.3-2026-08-21' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools: (232 commits)
perf vendor events arm64: Fix Tegra410 Olympus event 0x0197
perf vendor events arm64: fix swapped MetricGroup for Tegra410 L1 prefetcher metrics
perf evlist: Warn when 'sleep' workload is used without system-wide (-a) option
perf c2c: document function view in perf-c2c man page
perf c2c: add function view browser UI and cacheline detail
perf c2c: build and finalize the function view hierarchy
perf c2c: add function view hierarchy entry creation
perf c2c: add function view stats merge and memory management
perf c2c: add HPP list parsing for function view columns
perf c2c: add column rendering for function view
perf c2c: add function view model skeleton
perf c2c: extract shared data structures into util/c2c.h
perf test sample-parsing: Validate PERF_FORMAT_GROUP values without LOST
perf dso: Replace assert with runtime check in dso__read_symbol()
perf dso: Guard against cache underflow on short reads in dso_cache__memcpy()
perf dso: Use stored fd error instead of stale errno in file_read() and file_size()
perf dso: Guard close() against invalid fd in dso__decompress_kmodule_path()
perf dso: Guard against errno==0 when dso__get_filename() returns NULL
perf build: install-build-deps: add RHEL family devel package mapping
perf build: Remove leftover feature tests for removed cxx and clang support
...
The name and description for NVIDIA Tegra410 Olympus implementation-defined
event 0x0197 incorrectly indicate a TXREQ limit between one quarter and
one half. Correct them to describe a TXREQ limit below one half.
Fixes: 86ff690f45 ("perf vendor events arm64: Add Tegra410 Olympus PMU events")
Signed-off-by: Besar Wicaksono <bwicaksono@nvidia.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The L1D prefetcher metrics (l1d_cache_rw_miss_ratio, l1d_demand_accesses,
etc.) were incorrectly assigned to MetricGroup "L1I_Prefetcher_Effectiveness"
and vice versa. This causes 'perf stat --metric-group L1D_Prefetcher_Effectiveness'
to display L1I metrics, which is misleading.
Swap the MetricGroup assignments so L1D metrics are in the L1D group and
L1I metrics are in the L1I group.
Fixes: 86ff690f45 ("perf vendor events arm64: Add Tegra410 Olympus PMU events")
Signed-off-by: Saurav Sachidanand <sauravsc@amazon.com>
Reviewed-by: Besar Wicaksono <bwicaksono@nvidia.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
A common mistake when trying to record system-wide profiles for a given
duration is running commands like 'perf record sleep 1' or 'perf stat
sleep 1' without passing '-a' / '--all-cpus'. When '-a' is omitted, perf
defaults to per-process monitoring of the sleep process itself, which
does not collect system-wide activity and records very few events.
Add a warning in evlist__prepare_workload() when the workload executable
is 'sleep' and system-wide mode is not enabled.
Assisted-by: Antigravity:gemini-3.6-flash
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Describe the function view hierarchy (read-side function -> contending
writer function -> shared cachelines), the per-level indentation, and the
keys, with a worked example.
Document that reliable function attribution requires `iaddr` in
`--coalesce`, that the reader and writer may be the same function, and why
the coalesced function view cannot distinguish same-thread from
different-thread accesses in that case. Also document that verbose mode
includes code addresses in function rows.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add the browser front end: create/run/delete the hist_browser and add the
title. The d shortcut opens the existing per-cacheline detail view for the
selected level-3 cacheline. Level-3 entries retain the source cacheline
index, so the shortcut can locate the original entry without relying on a
potentially ambiguous virtual address.
Report a warning when the common model rejects a cacheline coalescing field
list without `iaddr`. Without it, the detail histograms may already have
merged samples from different functions and cannot support reliable
function attribution.
Keep visible-row accounting local to the function view by wrapping the
generic browser refresh callback and recounting the currently reachable
hierarchy before each redraw. This keeps navigation correct when a level-1
row is collapsed while level-3 descendants remain expanded, without adding
C2C-specific hooks to the shared hist_browser. Also handle Ctrl-C like the
other function-view exit keys.
Keep callchains hidden while the function browser runs, restoring the
user's setting while opening the cacheline detail view.
Wire the builder into perf_c2c__browse_function_view().
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add the builder that walks the top-level cacheline entries and, for each
read-side function, correlates the functions that write the same lines
(level 2) and the specific cachelines they contend over (level 3) within
each retained detail histogram. Aggregate the write traffic per contending
function, resort by store count, and prune writers/functions with no
contention. The finalize pass then computes the Cycles % denominator from
the surviving level-1 entries after pruning, so the column shows each
function's share of the functions retained in the table rather than of the
whole recording -- the semantics documented for Cycles % in perf-c2c.txt.
Expose c2c_function__build() and c2c_function__reset() for the TUI front
end added by the next patch. The builder requires iaddr in the cacheline
coalescing fields and returns the completed hists through an output
argument. Validate the inputs before replacing an existing model.
Function-view entries do not carry callchains. Suppress callchain handling
while building and tearing down the model so the common API does not depend
on the caller's current callchain setting.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add the entry-creation layer: owned-reference child allocation and
insertion, and the level-1/2/3 lookup-or-create functions keyed by
function symbol (level 1 read-side, level 2 writer) and by the source
cacheline's existing index (level 3).
Give synthetic children normal entry operations and acquire their thread
and map-symbol references. This lets the hierarchy teardown use
hist_entry__delete() for the common fields while the function-view free
callback handles the private child tree and containing allocation.
Reuse cacheline_idx to preserve the source entry identity without adding
function-view-only state. Add c2c_function__find_cacheline() to locate the
original cacheline entry by the same index.
These are driven by the hierarchy builder in the next patch and are
__maybe_unused until then.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add the per-entry stats/cstats aggregation helpers and hierarchy teardown.
Child common fields are released through hist_entry__delete(), while the
function-view free callback handles the private child tree and containing
allocation. Also add a helper for pruning writer entries with no stores or
cacheline children.
These are used by the entry-creation and builder patches that follow and
are __maybe_unused until then.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add the parser that builds the function view's local HPP output and
sort lists from field strings. This includes dimension lookup, comparator
wrappers, c2c_fmt allocation, and the initialization entry points used by
the hierarchy builder.
The generic perf_hpp__setup_output_field() registers formats on the global
perf_hpp_list. Using it here would leave the function view's local list
without output columns and modify the cacheline view's list instead. Add
c2c_function_hists__setup_output_field() to append sort keys to the local
output list.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add renderers for the function view's Cycles %, Store count, and
hierarchy identity columns. The identity column renders the read-side
function, contending writer, or cacheline, with indentation for the
hierarchy level. Also add width and header helpers, estimated-cycle
calculation, comparators, and the dimension table that ties them together.
Clamp the identity renderer's returned length to its local buffer before
using it for pointer and padding calculations. This handles snprintf-style
would-have-been lengths without changing normal output.
The next patch connects these dimensions to the view's HPP lists, so the
symbols used only there are temporarily marked __maybe_unused.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The function browser belongs in libperf-ui.a, but that archive is also
linked into python/perf.so, where builtin command objects are unavailable.
The browser therefore cannot depend on types or callbacks owned by
builtin-c2c.c.
Move c2c_hists, compute_stats, c2c_hist_entry, and the shared column
formatting definitions from builtin-c2c.c to a new util/c2c.h. Move
c2c_fmt_free() and c2c_fmt_equal() to a new util/c2c.c.
Keep struct perf_c2c, the command instance, and
perf_c2c__browse_cacheline() private to builtin-c2c.c.
No functional change.
Signed-off-by: Jiebin Sun <jiebin.sun@intel.com>
Reviewed-by: Tianyou Li <tianyou.li@intel.com>
Reviewed-by: Wangyang Guo <wangyang.guo@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Pull failfs filesystem from Christian Brauner:
"Add failfs and expose a FD_FAILFS_ROOT sentinel.
This allows userspace to shed their filesystem state completely. A
process with its root or working directory in failfs must anchor every
path lookup at an explicit file descriptor. Absolute paths, absolute
symlinks and AT_FDCWD-relative lookups simply fail.
Failfs is the counterpart to nullfs. nullfs says adds a permanently
empty, immutable directory whose lookups fail with ENOENT but which
can be opened, read, stat'd and mounted upon. Failfs on the other hand
fails every operation. The root cannot be opened at all. A single
instance is mounted during early boot via kern_mount(), which makes it
logically distinct from every mount namespace.
This is accompanied by a new fchroot() system call which makes
chrooting via a file descriptor a first class concept. It's possible
to chroot into failfs as an unprivileged user provided the task has no
new privileges set"
* tag 'vfs-7.3-rc1.failfs' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
Documentation: add failfs documentation
selftests/filesystems: add failfs selftests
arch: hookup fchroot() system call
fs: support FD_FAILFS_ROOT in fchroot()
fs: add fchroot()
fs: support FD_FAILFS_ROOT in fchdir()
fs: add failfs
The sample parsing test only validates grouped read values when
PERF_FORMAT_LOST is present.
For PERF_FORMAT_GROUP without PERF_FORMAT_LOST, the contents of
read.group.values[] are not validated, allowing corruption of the parsed
value and id fields to go undetected.
The values are also handed to the synthesis as a plain array of struct
sample_read_value, which always has a 24-byte stride, while
read.group.values is expected to be packed according to read_format --
evsel__parse_sample() points it into the event data. Without
PERF_FORMAT_LOST the stride is 16, so both the synthesis and the
comparison walk overlapping bytes and the test passes regardless of the
contents.
Validate value and id for grouped reads and continue to validate lost
when PERF_FORMAT_LOST is present, walking the entries with
next_sample_read_value(). Also build the input packed using
sample_read_value_size() so the compared fields are the real ones.
Verified with a deliberate stride bug in copy_read_group_values(): the
test still passes without this change and fails at read_format 0xc with
it applied.
Signed-off-by: PVS Narasimha Rao <venkatasuryapala@gmail.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
dso__read_symbol() asserts that len <= jited_prog_len, where len comes
from sym->end - sym->start (parsed from PERF_RECORD_KSYMBOL in
perf.data). Both values originate from untrusted file input.
With NDEBUG (production builds), the assert is compiled out, allowing
an out-of-bounds heap read when the BPF program buffer is accessed.
Without NDEBUG, a crafted perf.data crashes perf with an assertion
failure.
Replace the assert with a runtime bounds check that returns NULL with
an appropriate error code, matching the existing error handling
pattern in this function.
Fixes: aa04707f50 ("perf dso: Support BPF programs in dso__read_symbol()")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Song Liu <song@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
dso_cache__memcpy() computes cache_offset = offset - cache->offset,
then cache_size = min(cache->size - cache_offset, size). The RB tree
lookup in __dso_cache__find() matches using the full
DSO__DATA_CACHE_SIZE window, but cache->size reflects the actual pread
return value from dso_cache__populate().
A short pread (e.g. near end-of-file) makes cache->size smaller than
DSO__DATA_CACHE_SIZE. If a subsequent access targets an offset past
cache->offset + cache->size but within the DSO__DATA_CACHE_SIZE
window, the cache entry is found but cache_offset exceeds cache->size.
Since both are u64, the subtraction cache->size - cache_offset wraps
to a large value, min() selects the caller's size, and memcpy reads
out of bounds.
Return 0 for an offset past the valid cached data. For a regular
file a short pread only happens at end-of-file, so 0 is what a direct
pread() at that offset would return: cached_io() stops its read loop
as on EOF. Re-reading from the backing file would not help — a
second pread at the same offset returns the same short count.
Fixes: 366df72657 ("perf dso: Refactor dso_cache__read()")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
file_read() and file_size() use ret = -errno when
dso__data(dso)->fd is negative after try_to_open_dso() fails. By this
point errno has been through mutex_lock(), nsinfo__mountns_enter(), and
multiple open() attempts inside try_to_open_dso() — it no longer
reflects the actual open failure. If errno happens to be 0, ret = 0
looks like EOF rather than an error, and file_size() callers like
dso__data_size() would then report a zero-sized file instead of
failing.
dso__data(dso)->fd is always negative on failure — -errno from
__open_dso() when no filename could be built (e.g. -EINVAL, -ENOENT),
or -1 when do_open() itself failed — and never 0, so use it directly
instead of reading the stale global errno.
No assert() or comment is needed after the assignment: the enclosing
if (dso__data(dso)->fd < 0) already guarantees ret < 0
[Namhyung Kim review].
Fixes: 33bdedcea2 ("perf tools: Protect dso cache fd with a mutex")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
dso__decompress_kmodule_path() unconditionally calls close(fd) on the
return value of decompress_kmodule(). When decompression fails or the
DSO is not compressed, decompress_kmodule() returns -1. close(-1)
fails with EBADF and clobbers errno, which callers up the chain
(dso__get_filename → __open_dso) depend on for error propagation.
Guard the close() call with fd >= 0 so only valid file descriptors are
closed.
Fixes: 42b3fa6708 ("perf tools: Introduce dso__decompress_kmodule_{fd,path}")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
__open_dso() computes fd = -errno when dso__get_filename() returns NULL.
Some failure paths in dso__get_filename() (e.g. binary type mismatch)
return NULL without making a syscall, leaving errno at 0 from a prior
successful call. fd = -0 = 0, which is stdin — subsequent code treats
it as a valid file descriptor.
Fall back to ENOENT when errno is 0, ensuring fd is always negative on
failure.
The forced ENOENT stays in errno for the callers that check it after a
negative fd. It must not misdirect the try_to_open_dso() fallback
loop, though: dso__get_filename()'s chroot fallback used to accept a
stale ENOENT even when stat() succeeded on a non-regular file (e.g. a
directory). Re-stat() there and only take the chroot path when
stat() actually failed with ENOENT [sashiko-bot review of PATCH 1/5].
Fixes: eba5102d2f ("perf tools: Add global list of opened dso objects")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
With the Fedora mapping in place, this patch extends it to the RHEL
family (RHEL, CentOS Stream, Rocky Linux, AlmaLinux, Oracle Linux),
which shares most Fedora package names and runs dnf (RHEL 8 and
later). The names that differ are handled by probing the enabled
repos:
- zlib.h comes from zlib-ng-compat-devel on the RHEL 10 family,
zlib-devel on RHEL 9 and earlier;
- there is no java-latest-openjdk-devel: the JDK devel package is
versioned per release, java-21-openjdk-devel on the RHEL 10
family, java-17-openjdk-devel on RHEL 9, java-11-openjdk-devel
on RHEL 8;
- libbpf-devel and capstone-devel live in the CRB repo on RHEL and
CentOS Stream 10, in EPEL on RHEL 9 and earlier;
- libbabeltrace2-devel is not packaged on the RHEL 10 family.
Packages not available on the enabled repos are skipped instead of
aborting the dnf transaction, and are listed at the end of the run, with
the repo that provides them pointed out in the header comment and help
text: a distro with CRB/EPEL enabled gets the full set, one without them
still installs what it can.
This also holds for the base set: e.g. 'rust' exists only as the
rust-toolset AppStream module on RHEL 8 and 9, where it is not
installable as a plain package, so it is skipped and noted there instead
of failing the whole dnf transaction.
The base set lists pkgconf-pkg-config instead of pkgconfig: both
families have been on pkgconf since Fedora 26 / RHEL 8, where
'pkgconfig' lives only as a virtual Provides of that subpackage, and
a minimal RHEL-family container may not have it preinstalled.
Validated on a fresh CentOS Stream 10 distrobox container, with the
CRB repo enabled, so the host system is not modified:
distrobox create --image quay.io/centos/centos:stream10
distrobox enter centos-stream10
dnf config-manager --set-enabled crb
make -C tools/perf install-build-deps
which installed the 28 available mapped packages; libbabeltrace2-devel, the
only mapped package with no RHEL 10 package, is reported at the end
of the run.
A subsequent 'make -C tools/perf feature-dump' enabled every feature
with an external dependency the RHEL 10 family provides, including
libbpf and libcapstone from the CRB repo, with only
babeltrace2-ctf-writer left out along with the deliberately unmapped
opt-in features.
Re-running the target is a no-op (dnf reports "Nothing to do"); with the
CRB repo disabled, the skipped packages are instead listed in the
end-of-run note, whose header comment and help text point out which repo
provides them.
Members of the family without dnf (RHEL 7 and earlier, e.g. Oracle
Linux 7, a yum-only distro) are rejected with an explicit error while
the dnf-based members get the full mapping.
Example of its --list:
$ grep PRETTY_NAME /etc/os-release
PRETTY_NAME="Fedora Linux 44 (Toolbx Container Image)"
$ tools/perf/scripts/install-build-deps.sh --list --distro rhel
bison
capstone-devel
clang-devel
elfutils-debuginfod-client-devel
elfutils-devel
elfutils-libelf-devel
flex
gcc
gcc-c++
glibc-devel
java-latest-openjdk-devel
kernel-headers
libbabeltrace2-devel
libbpf-devel
libpfm-devel
libstdc++-devel
libtraceevent-devel
libzstd-devel
llvm-devel
make
numactl-devel
openssl-devel
pkgconf-pkg-config
python3-devel
python3-setuptools
rust
slang-devel
systemtap-sdt-devel
xz-devel
zlib-ng-compat-devel
$
Assisted-by: opencode:deepseek-v4-flash-free
Assisted-by: claude:claude-opus-4-7
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
With the Fedora and Ubuntu mappings in place, this patch adds Debian
support: Debian installs the same devel packages, under the same
names, as the Ubuntu mapping, so it reuses debian_pkg_for() and
debian_base_pkgs as-is, with only auto-detection in detect_distro()
(and the shared apt-get install path) added, keeping the script's
per-distro dispatch ready for distros with their own package names.
Validated on a fresh Debian 13 (trixie) container so the host system
is not modified:
distrobox create --image debian:trixie
distrobox enter debian-trixie
make -C tools/perf install-build-deps
which installed the 29 mapped packages; re-running the target is a
no-op (apt-get reports "0 newly installed"). A subsequent clean build
enabled the same feature set as Ubuntu: 'perf version --build-options'
shows every feature with an external dependency Debian has a package
for [on], including the BPF skeletons compiled with clang/llvm
(libLLVM), the python binding and the C++-based features, with only
the deliberately unmapped libbfd family, libperl, libunwind and the
CoreSight (libopencsd) packages [OFF].
RHEL, whose package mapping is largely similar to Fedora's, is the
remaining planned distro, to be enabled once that mapping is
validated on it.
Example of its --list:
$ grep PRETTY_NAME /etc/os-release
PRETTY_NAME="Fedora Linux 44 (Toolbx Container Image)"
$ tools/perf/scripts/install-build-deps.sh --list --distro debian
bison
clang
default-jdk
flex
g++
gcc
libbabeltrace2-dev
libbpf-dev
libc6-dev
libcapstone-dev
libdebuginfod-dev
libdw-dev
libelf-dev
liblzma-dev
libnuma-dev
libpfm4-dev
libslang2-dev
libssl-dev
libtraceevent-dev
libzstd-dev
linux-libc-dev
llvm-dev
make
pkg-config
python3-dev
python3-setuptools
rustc
systemtap-sdt-dev
zlib1g-dev
$
Assisted-by: opencode:deepseek-v4-flash-free
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
With the framework and Fedora mapping in place, this patch adds the
Ubuntu (apt) mapping: same feature-to-package correspondence as the
Fedora one, adapted to Debian packaging conventions (libfoo-dev), on
a per-distro dispatch so future distros can pick their own mapping or
reuse one of these (Debian shares the Ubuntu mapping).
Notable differences from Fedora:
- base set: g++ (ships libstdc++-*-dev, covering cxa-demangle),
pkg-config (installed implicitly by Fedora's default toolchain
metapackage, but not by Ubuntu's), linux-libc-dev and libc6-dev
instead of kernel-headers and glibc-devel, and rustc for rust;
- cxa-demangle maps to nothing, covered by g++'s libstdc++;
- the clang-bpf-co-re test needs the clang compiler binary (Fedora's
clang-devel provides it transitively), and llvm-dev, which also
brings llvm-config (deps on the llvm package), used by the
llvm/llvm-perf tests;
- libslang maps to libslang2-dev and jvmti to default-jdk;
- the install command runs 'apt-get update' first since a fresh
container has no package indexes, unlike dnf.
Validated on a fresh Ubuntu 26.04 distrobox container so the host
system is not modified:
distrobox create --image ubuntu:26.04
distrobox enter ubuntu-26-04
make -C tools/perf install-build-deps
which installed the 29 mapped packages; a subsequent clean O= build
enabled every feature with an external dependency Ubuntu has a
package for: perf's build-options then showed all of them [on],
including the BPF skeletons requiring clang/llvm, the python binding
and the C++-based features, with only the deliberately unmapped
(deprecated) libbfd family, libperl and libunwind [OFF], and the
build linked libpfm, libbabeltrace2-ctf-writer, libcapstone,
libtraceevent, libslang, libnuma, libdw and libssl. Re-running the
target is a no-op (apt-get reports "0 newly installed").
Debian (trixie) is the next planned distro: it shares this Ubuntu
mapping, so enabling it reuses it as-is, once it gets validated on a
Debian release.
Example of its --list:
$ grep PRETTY_NAME /etc/os-release
PRETTY_NAME="Fedora Linux 44 (Toolbx Container Image)"
$ tools/perf/scripts/install-build-deps.sh --list --distro ubuntu
bison
clang
default-jdk
flex
g++
gcc
libbabeltrace2-dev
libbpf-dev
libc6-dev
libcapstone-dev
libdebuginfod-dev
libdw-dev
libelf-dev
liblzma-dev
libnuma-dev
libpfm4-dev
libslang2-dev
libssl-dev
libtraceevent-dev
libzstd-dev
linux-libc-dev
llvm-dev
make
pkg-config
python3-dev
python3-setuptools
rustc
systemtap-sdt-dev
zlib1g-dev
$
Assisted-by: opencode:deepseek-v4-flash-free
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
With the framework from the previous commit in place, this patch adds
the per-feature mapping for Fedora/dnf: for each feature test in
tools/build/feature/, the Fedora devel package providing the headers
or library the test compiles against, kept explicit in the script
next to the test that requires it.
Special cases:
- test-libdebuginfod.c includes <elfutils/debuginfod.h>, provided
by elfutils-debuginfod-client-devel, not elfutils-devel;
- the cxa-demangle test links against libstdc++'s builtin demangler,
pulling in libstdc++-devel;
- the BPF-oriented features (bpf, clang-bpf-co-re) get their headers
from the base packages and clang-devel.
Tests with no Fedora equivalent (bionic, compile-32, compile-x32) and
the opt-in/deprecated ones (libbfd disassembler family, GTK2, LIBPERL,
LIBUNWIND, CoreSight, and the tests perf itself doesn't check, like
libcpupower) are deliberately not mapped.
Validated on a fresh Fedora 44 toolbx container, so the host OS is not
modified:
toolbox create fedora:44
toolbox enter fedora:44
make -C tools/perf install-build-deps
which installed the 29 mapped packages; a subsequent clean O= build
enabled every feature with an external dependency Fedora provides
(feature tests went to 1, except bionic/compile-32/compile-x32, which
have no Fedora equivalent, and the libunwind-debug-frame tests, whose
symbols Fedora's libunwind does not export), linking libpfm,
libbabeltrace2-ctf-writer, libcapstone, libtraceevent, libslang and
libnuma, as well as building the BPF skeletons requiring clang/llvm.
Re-running the target is a no-op (dnf reports "Nothing to do").
RHEL and its derivatives share most Fedora package names but are
refused by the script until this mapping is validated on them.
Example of its --list option:
$ grep PRETTY_NAME /etc/os-release
PRETTY_NAME="Fedora Linux 44 (Toolbx Container Image)"
$ tools/perf/scripts/install-build-deps.sh --list
bison
capstone-devel
clang-devel
elfutils-debuginfod-client-devel
elfutils-devel
elfutils-libelf-devel
flex
gcc
gcc-c++
glibc-devel
java-latest-openjdk-devel
kernel-headers
libbabeltrace2-devel
libbpf-devel
libpfm-devel
libstdc++-devel
libtraceevent-devel
libzstd-devel
llvm-devel
make
numactl-devel
openssl-devel
python3-devel
python3-setuptools
rust
slang-devel
systemtap-sdt-devel
xz-devel
zlib-devel
$
Assisted-by: opencode:deepseek-v4-flash-free
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Installing the development packages needed to build perf is
error-prone on a fresh distro install: the packages are scattered
across the feature tests in tools/build/feature/, each checking for a
specific header/library, and the build only tells you what's missing
after failing a check.
This series adds a 'make -C tools/perf install-build-deps' target to
install them in one go, deriving the package list from the feature tests
themselves.
This commit adds the framework, on top of the parse-time compiler
probe guard from the previous commit:
- the install-build-deps target in tools/perf/Makefile.perf, exempted
from the config/feature detection pass, since it must run in a fresh
container, before gcc or pkg-config exist, to install them;
- the install-build-deps.sh script, with --list, --dry-run and
--distro options, distro detection (Fedora and Ubuntu), dnf and
apt-get drivers, root/passwordless-sudo handling, and the base
packages common to any build: compiler, C++ compiler, make, flex,
bison, libc and kernel headers, python3-setuptools (needed by the
python binding) and rust (checked by the rust feature test);
- the parse-time probes for optional tools, like pkg-config, use
'command -v' with stderr discarded, so a fresh container without
them gets no 'which: no pkg-config in (...)' spew from make;
- the script does not rely on 'set -e': its error paths are explicit,
since the make target runs it via $(SHELL), where a shebang option
would be ignored anyway, so direct and make-driven runs behave the
same.
The per-feature mappings, from each feature test to the devel package
providing its headers on a given distro, are added by the follow-up
patches, one per distro, together with the validation of each mapping
in a fresh container: until then the target installs just the base
toolchain.
Assisted-by: opencode:deepseek-v4-flash-free
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
machine__resolve() indirectly acquires a thread reference via
machine__findnew_thread(). Release it, as suggested by the documentation
of the former.
Signed-off-by: Michalis Niarchos <michael.niarchos@gmail.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
If scandir() finds no matching tasks in /proc, n is 0. If thread_nr is > 1,
we bypass the single-thread fast path and then clamp thread_nr to n, making
it 0. This results in a divide by zero when calculating num_per_thread.
Handle n <= 1 early to use the single-thread fast path and prevent the
crash.
Fixes: 340b47f510 ("perf top: Implement multithreading for perf_event__synthesize_threads")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
In pyrf__metrics_cb, PyDict_SetItem does not steal the reference of the
key and value, so they need to be decref'ed after successful insertion
to avoid memory leaks.
Fixes: 47b3e95728 ("perf python: Add metrics function")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
In pyrf_evsel__read, if PyArg_ParseTuple fails, the allocated count_values
is leaked. Move the allocation of count_values after the PyArg_ParseTuple
call to prevent the memory leak.
Fixes: 739621f657 ("perf python: Add evsel read method")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The metrics() function can return a dictionary where the value is either
a string or a list of strings, so the type signature in the stub file
should be Union[str, List[str]].
Fixes: 430da3cd03b4 ("perf python: Add perf.pyi stubs file")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add CHECK_INITIALIZED checks to the thread attribute getters
(get_pid, get_tid, get_ppid) to prevent crashes if they are accessed
before being properly initialized.
Fixes: 3b96bf7af60d ("perf python: Add python session abstraction wrapping perf's session")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Replace path clearing with memset so the entire struct is zeroed,
preventing uninitialized fields from causing errors later.
Fixes: 4cd0142f7dec ("perf python: Add wrapper for perf_data file abstraction")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
If val is NULL when setting an attribute, PyErr_SetString should be
called as deleting the attribute isn't supported. In addition, ensure
PyErr_Occurred is checked before setting the attribute to avoid setting
a garbage value.
Fixes: 877108e42b ("perf tools: Initial python binding")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add explicit Py_TYPE checks to ensure the arguments passed are
actually of the correct pyrf_thread_map and pyrf_cpu_map types.
Fixes: 877108e42b ("perf tools: Initial python binding")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The set_values function incorrectly assumed the list contained exactly
5 elements. Add a check to prevent out-of-bounds access.
Fixes: 877108e42b ("perf tools: Initial python binding")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Fix a memory leak in cmd_test() where skiplist was not freed on
exit paths.
Assisted-by: Antigravity:gemini-3.1-pro
Fixes: 2ae828786c ("perf test: Allow skipping tests")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
In perf_event__synthesize_threads(), fix an uninitialized pthread_join()
call when thread creation fails by only joining the successfully
created threads.
Assisted-by: Antigravity:gemini-3.1-pro
Fixes: 340b47f510 ("perf top: Implement multithreading for perf_event__synthesize_threads")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Fix a Python list object leak in pyrf_evlist__get_pollfd() by adding
a missing Py_DECREF on the error exit path.
Assisted-by: Antigravity:gemini-3.1-pro
Fixes: 877108e42b ("perf tools: Initial python binding")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Fix a memory leak in cmd_top() where top.sb_evlist was not freed if
evlist__add_bpf_sb_event() fails. Note that evlist__start_sb_thread() and
evlist__stop_sb_thread() take ownership of the evlist and free it, so
we must only free it if we fail before starting the thread. Also set
top.sb_evlist to NULL to prevent use-after-free bugs.
Apply the same fix to builtin-record.c to avoid leaking rec->sb_evlist
and calling pthread_join on uninitialized threads in the error path.
Assisted-by: Antigravity:gemini-3.1-pro
Fixes: b38d85ef49 ("perf bpf: Decouple creating the evlist from adding the SB event")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Fix a memory leak in cmd_stat() where evsel_list is leaked if an error
occurs while opening the output file.
Assisted-by: Antigravity:gemini-3.1-pro
Fixes: 361c99a661 ("perf evsel: Introduce perf_evlist")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Fix a memory leak in script_find_metrics() where metric_evlist is leaked
when returning early on error paths.
Assisted-by: Antigravity:gemini-3.1-pro
Fixes: 3622990efa ("perf script: Change metric format to use json metrics")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
While 'perf sched latency' reports task runtime and delay statistics
(average and maximum delay), it does not provide a visual representation
of how task wait times are distributed across latency ranges between
snapshots (start and finish of the analysis window).
The --histogram option collects CPU wait latencies (time between when
a task becomes runnable and when it gets scheduled onto a CPU) into 22
latency buckets, displaying an ASCII bar chart distribution.
The --hist-mode option configures the bucketing scheme:
- log (default). Logarithmic latency buckets ranging from
sub-microsecond (< 1 us) up to >= 1.05 seconds
- linear. Equal-width linear latency buckets
(i.e., 100 us steps up to >= 2.1 ms)
The --time option allows filtering trace event processing to a
specific time interval [start,stop].
Example histogram output excerpt:
❯ sudo perf sched latency --histogram --CPU 0
CPU Wait Latency Distribution Histogram (between snapshots) (total samples: 36114)
-------------------------------------------------------------------
Latency Range | Count | Pct | Histogram Graph
-------------------------------------------------------------------
< 1 us | 17 | 0.0% | #
2 - 4 us | 673 | 1.9% | #
4 - 8 us | 6237 | 17.3% | ######
8 - 16 us | 3224 | 8.9% | ###
16 - 32 us | 1388 | 3.8% | #
32 - 64 us | 709 | 2.0% | #
64 - 128 us | 690 | 1.9% | #
128 - 256 us | 789 | 2.2% | #
256 - 512 us | 541 | 1.5% | #
512 - 1024 us | 2256 | 6.2% | ##
1 - 2 ms | 3577 | 9.9% | ###
2 - 4 ms | 13259 | 36.7% | ##############
4 - 8 ms | 2523 | 7.0% | ##
8 - 16 ms | 222 | 0.6% | #
16 - 32 ms | 10 | 0.0% | #
>= 1.05 s | 3 | 0.0% | #
-------------------------------------------------------------------
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Aaron Tomlin <atomlin@atomlin.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Currently, 'perf sched latency' displays task runtime and delay values
exclusively in milliseconds (ms). This can be hard to read when
latencies are very small (in the microsecond or nanosecond range) or
unusually large (seconds).
Introduce auto-scaling for latency and runtime display columns. Values
are dynamically scaled and output with the most appropriate unit:
nanoseconds (ns), microseconds (us), milliseconds (ms), or seconds (s).
Additionally, rename column headers from "Runtime ms", "Avg delay ms",
and "Max delay ms" to "Runtime", "Avg delay", and "Max delay"
respectively, adjust spacing to maintain column alignment and stripe
redundant prefix strings from each row's format string to produce a
clean, tabular output.
For illustrative purposes, a comparison of the latency table header
before and after this change is shown below:
Before:
-------------------------------------------------------------------------------------------------------------------------------------------
Task | Runtime ms | Count | Avg delay ms | Max delay ms | Max delay start | Max delay end |
-------------------------------------------------------------------------------------------------------------------------------------------
kworker/2:2-mm_:154757 | 0.033 ms | 1 | avg: 0.829 ms | max: 0.829 ms | max start: 169486.543205 s | max end: 169486.544034 s
After:
------------------------------------------------------------------------------------------------------------------------------------------
Task | Runtime | Count | Avg delay | Max delay | Max delay start | Max delay end |
------------------------------------------------------------------------------------------------------------------------------------------
kworker/2:2-mm_:154757 | 32.873 us | 1 | 829.347 us | 829.347 us | 169486.543205 s | 169486.544034 s |
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Aaron Tomlin <atomlin@atomlin.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
For pipe mode streams, event attributes are received dynamically during
event processing, meaning session->evlist is not populated prior to
perf_session__process_events(). To handle pipe input correctly:
- Register the missing .attr, .tracing_data, .build_id, and .feature
callbacks in cmd_sched()
- Promote the handlers array to file-scope (latency_handlers[]) and
dynamically assign matching tracepoint handlers
(or process_sched_ignore) inside
perf_sched__process_tracepoint_sample() when evsel->handler is NULL;
replace process_sched_wakeup_ignore() with process_sched_ignore()
- Perform the trace check post-processing when handling pipe data
Fixes: 27295592c2 ("perf session: Share the common trace sample_check routine as perf_session__has_traces")
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Aaron Tomlin <atomlin@atomlin.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When 'perf sched latency' is executed on a perf.data file that lacks
tracepoint samples (i.e., a file recorded without the -R flag or
containing only non-tracepoint events), perf_session__has_traces()
correctly outputs an error message. However, perf_sched__read_events()
subsequently falls through and returns 0 (success).
Consequently, caller functions such as perf_sched__lat() assume event
processing succeeded and proceed to render empty latency header tables
and total summary statistics.
Fix this behaviour by ensuring perf_sched__read_events() aborts early and
returns a suitable error code when perf_session__has_traces() evaluates
to false.
Additionally, validate thread__get_runtime() against NULL in
map_switch_event() to prevent potential null-pointer dereferences.
Fixes: 27295592c2 ("perf session: Share the common trace sample_check routine as perf_session__has_traces")
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Aaron Tomlin <atomlin@atomlin.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Linus Torvalds