When a dynamic probe is created without explicitly requested arguments
via perf probe --add, the Ftrace subsystem automatically appends
"__probe_ip" to the tracepoint format to record the instruction pointer.
Currently, perf trace prints this implicit field by default.
Furthermore, because the formatting logic defaults to a standard signed
integer representation, the kernel space memory address is erroneously
displayed as a meaningless negative integer.
❯ sudo ./perf trace --event probe:proc_sys_open --max-events 1
0.000 ps/1316543 probe:proc_sys_open(__probe_ip: -1406056956)
This patch addresses the user experience by combining two refinements:
1. "__probe_ip" is now hidden from the standard output, as its
presence adds no contextual value for a bare probe.
2. If the user explicitly requests verbose output (--verbose),
"__probe_ip" is intercepted and properly formatted as a hexadecimal
kernel address, restoring its utility for debugging inline
function hits.
❯ sudo ./perf trace --event probe:proc_sys_open --max-events 1
0.000 ps/1314074 probe:proc_sys_open()
❯ sudo ./perf trace --verbose --event probe:proc_sys_open --max-events 1
Using CPUID GenuineIntel-6-8E-C
mmap size 528384B
0.000 ps/1314366 probe:proc_sys_open(__probe_ip: 0xffffffffac314604)
Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Aaron Tomlin <atomlin@atomlin.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Daniel Vacek <neelx@suse.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sean Ashe <sean@ashe.io>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a new shell test to verify the feature. The test covers:
- Basic address remapping for user space samples.
- Pipe mode coverage for piped into.
- Callchain address remapping.
- Consistency of output before and after injection.
- Pipe mode report consistency.
- Dropping of samples that leak ASLR info (physical addresses).
- Kernel address remapping (utilizing a dedicated kernel-intensive VFS
dd workload to guarantee continuous timer interrupts sampling flow
inside kernel privilege states).
- Kernel report consistency with address normalization.
The test suite is hardened with global 'set -o pipefail' assertions
to catch pipeline failures, stream-consuming awk processors to handle
SIGPIPE signals, and a dedicated pipe output scenario validating raw
'perf inject -o -' stdout streams.
Note on kernel DSO normalization in the test script:
The test script deliberately normalizes all kernel DSOs to a generic
[kernel] tag before diffing, as obfuscating physical kernel addresses
forces perf report to occasionally shift samples between individual
modules and [kernel.kallsyms] due to the lack of valid host module
boundary maps.
Note on ARM:
Kernel-based ASLR test cases (test_kernel_aslr and test_kernel_report_aslr)
are skipped on ARM architectures (aarch64 and arm*) to bypass high latency
constraints (such as check_invariants() spending excessive execution time
in maps__split_kallsyms() on debug builds) and symbolization inconsistencies.
Assisted-by: Antigravity:gemini-3.1-pro
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Gabriel Marin <gmx@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Extend the ASLR tool stripping helpers to drop register dump payloads
by masking out the relevant perf_event_attr fields (sample_regs_user,
sample_regs_intr) when the delegated tool is handling the data.
struct aslr_evsel_priv maintains the original perf_event_attr values
and is looked up via the evsel_orig_attrs hashmap so that sample sizes
can be properly parsed even when bits are stripped from the pipeline.
This is critical for bounded array copying within aslr_tool__process_sample,
which relies on orig_sample_type to determine exactly which fields were
captured by the kernel before any stripping occurred.
This allows us to keep samples that would otherwise be dropped because
they contain registers, while still obfuscating the registers.
Committer notes:
Moved now used variables from the previous patch:
struct aslr_evsel_priv *priv = NULL;
u64 orig_sample_type;
u64 orig_regs_user;
u64 orig_regs_intr;
Assisted-by: Antigravity:gemini-3.1-pro
Co-developed-by: Gabriel Marin <gmx@google.com>
Signed-off-by: Gabriel Marin <gmx@google.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add the sample address remapping logic to the ASLR tool. This patch
implements aslr_tool__process_sample, which parses sample events,
remaps IPs, ADDRs, callchains, and branch stacks using the mappings
collected from metadata events, and drops potentially leaking raw,
register, stack, physical address, and aux samples.
Also adds the aslr_tool__remap_address helper function.
Note on cross-endian compatibility:
'perf inject' functions as an endianness converter. Input files are read,
and their events are byte-swapped to host endianness in memory. When the
tool emits its output, it writes a host-endian PERF_MAGIC in the file
header, thereby marking the output file as host-endian natively.
Because the output file is always written in host endianness, events and
payloads must be constructed entirely using host-endian layouts. For this
reason, this patch explicitly un-packs and repacks PERF_SAMPLE_TID (and
PERF_SAMPLE_CPU) using unions to ensure that the sequential 32-bit layout
is correctly aligned in host endianness. Similarly, branch stack flags
(which are modified in-place to host-endian bitfields by the parser) are
copied directly to the newly synthesized event. When re-parsing the newly
synthesized event, 'needs_swap=false' is explicitly used to prevent double
swapping the already host-endian fields.
Committer notes:
Removed several unused variables, they will be reintroduced in the
following patches where they are finally used:
struct aslr_evsel_priv *priv = NULL;
u64 orig_sample_type;
u64 orig_regs_user;
u64 orig_regs_intr;
Also used PRIx64 for two u64 args (addresses) and %zu for a size_t arg
(map__size()) to fix the build on 32-bit architectures.
Assisted-by: Antigravity:gemini-3.1-pro
Co-developed-by: Gabriel Marin <gmx@google.com>
Signed-off-by: Gabriel Marin <gmx@google.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If perf.data files are taken from one machine to another they may
leak virtual addresses and so weaken ASLR on the machine they are
coming from. Add an aslr option for perf inject that remaps all
virtual addresses, or drops data/events, so that the virtual address
information isn't leaked.
This patch introduces the core ASLR remapping tool infrastructure and
implements remapping/tracking for metadata events (MMAP, MMAP2, COMM,
FORK, EXIT, KSYMBOL, TEXT_POKE). Sample events are delegated without
remapping for now.
Assisted-by: Antigravity:gemini-3.1-pro
Signed-off-by: Ian Rogers <irogers@google.com>
Co-developed-by: Gabriel Marin <gmx@google.com>
Signed-off-by: Gabriel Marin <gmx@google.com>
Tested-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
During kernel ELF symbol parsing (dso__process_kernel_symbol), proc
kallsyms image loading (dso__load_kernel_sym,
dso__load_guest_kernel_sym), and dynamic kernel memory map alignment
updates (machine__update_kernel_mmap), the loader directly modifies live
virtual address boundary keys fields on map objects.
If these boundaries are mutated while the map pointer actively resides
inside the parent maps cache array list (kmaps) outside of any lock
closure, an unsafe concurrent window is exposed where parallel worker
lookup threads (e.g., inside perf top) can mistakenly assume the cache
remains sorted based on stale parameters, executing binary search
queries (bsearch) across an unsorted range and triggering lookup
failures.
Fix this by introducing maps__mutate_mapping() that explicitly acquires
the parent maps write semaphore lock, executes an incoming mutation
callback block to perform the field updates under lock protection, and
invalidates the sorted tracking flags prior to releasing the write lock.
This guarantees synchronization invariants, closing the concurrent
lookup race window. The adjacent module alignment pass inside
machine__create_kernel_maps() is safely preserved as a high-performance
lockless pass, as its invocation lifecycle bounds remain strictly
single-threaded by contract during session initialization construction.
To safely support this unconditional down_write write lock mutator
without recursive read-to-write self-deadlock upgrades during lazy
symbol loading, we introduce a public maps__load_maps() API.
It copies map pointers under a brief read lock and force-loads all
modules locklessly outside the lock. Callers (such as perf inject) must
pre-load all kernel symbol maps up front at startup using
maps__load_maps(), completely bypassing dynamic runtime mutations.
Fixes: 39b12f7812 ("perf tools: Make it possible to read object code from vmlinux")
Assisted-by: Antigravity:gemini-3.1-pro
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Gabriel Marin <gmx@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In aggregation modes (e.g. --per-socket, --per-die, etc.), a counter
might not be scheduled or counted on specific aggregate groups if it was
not assigned to the CPUs belonging to those groups. However, the
printout() check triggers the "print_free_counters_hint" logic
unconditionally for any supported counter with a missing count. This
results in a false "Some events weren't counted. Try disabling the NMI
watchdog" warning.
Furthermore, the NMI watchdog only reserves performance counters on core
PMUs. Uncore PMU events (e.g. CHA, IMC) are not affected by the NMI
watchdog, but their failures also falsely triggered this warning.
This warning was originally introduced in commit 02d492e5dc ("perf
stat: Issue a HW watchdog disable hint")
To fix this, restrict setting of print_free_counters_hint to only
trigger for core PMU events by checking counter->pmu and
counter->pmu->is_core.
Example before/after:
$ perf stat -M lpm_miss_lat --metric-only --per-socket -a -- sleep 1
Before:
Performance counter stats for 'system wide':
ns lpm_miss_lat_rem ns lpm_miss_lat_loc
S0 126 202.3 207.9
S1 126 231.9 259.3
1.006029831 seconds time elapsed
Some events weren't counted. Try disabling the NMI watchdog:
echo 0 > /proc/sys/kernel/nmi_watchdog
perf stat ...
echo 1 > /proc/sys/kernel/nmi_watchdog
After:
Performance counter stats for 'system wide':
ns lpm_miss_lat_rem ns lpm_miss_lat_loc
S0 126 202.3 207.9
S1 126 231.9 259.3
1.006029831 seconds time elapsed
Assisted-by: Gemini:gemini-next
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Chun-Tse Shao <ctshao@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When test descriptions are extremely long (e.g., the truncated perf.data
graceful handling test is 103 characters long), they wrap across terminal
boundaries.
Because the ANSI escape code to delete the line (PERF_COLOR_DELETE_LINE)
only clears a single terminal line, visual wrapping leaves orphan
wrapped lines on the screen, which results in the test description being
printed multiple times.
Resolve this by checking the terminal width (get_term_dimensions) and
dynamically truncating the printed test description to fit within the
available columns, leaving safety space for the prefix index and status
suffix.
Also, remove the width padding from the test suite headers which do not
display inline status messages. This prevents their trailing colons from
wrapping onto new lines on standard width terminals.
Finally, avoid GCC 16's -Wformat-truncation warnings by delegating the
description padding to pr_info's %-*s format specifier instead of padding
within a temporary buffer, and clamp the truncation limit to the temporary
buffer's size.
JUnit XML output and the failure summary report still print the full,
untruncated test descriptions.
Assisted-by: Gemini-CLI:Google Gemini 3.1 Pro
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
symbol-minimal.c's read_build_id() iterates ELF notes with the same
pattern as symbol-elf.c's elf_read_build_id(): pointer arithmetic
driven by n_namesz and n_descsz from 32-bit note header fields,
without validating that the name and desc fit within the note section
data. A malformed ELF file with oversized note sizes causes
out-of-bounds reads past the section data buffer.
Add the same bounds check as the libelf path: validate namesz and
descsz individually against remaining data before advancing the
pointer, avoiding size_t overflow on 32-bit.
Fixes: b691f64360 ("perf symbols: Implement poor man's ELF parser")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
elf_read_build_id() iterates ELF notes using pointer arithmetic
driven by n_namesz and n_descsz from the note headers. Neither
the note header read nor the subsequent name/desc advances are
checked against the section boundary. A malformed ELF file with
oversized note sizes causes out-of-bounds reads past the section
data buffer.
Add two bounds checks: verify the note header fits within the
remaining section data, and verify that namesz + descsz (after
alignment) fits before advancing the pointer.
Fixes: fd7a346ea2 ("perf symbols: Filename__read_build_id should look at .notes section too")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When perf_env__insert_bpf_prog_info() returns false (duplicate
program), the error path frees info_linear and info_node but not
info_node->metadata. If bpf_metadata_create() had succeeded, the
metadata allocation is permanently leaked.
Fix by calling bpf_metadata_free() on info_node->metadata before
freeing info_node. bpf_metadata_free() handles NULL, so this is
safe even when bpf_metadata_create() returned NULL.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Fixes: fdc3441f2d ("perf record: collect BPF metadata from new programs")
Cc: Blake Jones <blakejones@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
bpf_metadata_create() calls bpf_metadata_read_map_data() which
allocates map.btf and map.rodata. If the subsequent
bpf_metadata_alloc() fails, the code does 'continue' which skips
bpf_metadata_free_map_data(), permanently leaking both allocations.
Fix by calling bpf_metadata_free_map_data() before continue.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Fixes: ab38e84ba9 ("perf record: collect BPF metadata from existing BPF programs")
Cc: Blake Jones <blakejones@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
synthesize_bpf_prog_name() calls btf__type_by_id() and immediately
dereferences the result via t->name_off without checking for NULL.
btf__type_by_id() returns NULL when the type_id is invalid or out
of range. When processing perf.data files, finfo->type_id comes from
untrusted input, so an invalid ID causes a NULL pointer dereference.
Fix by checking t for NULL before dereferencing.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Fixes: fc462ac75b ("perf bpf: Extract logic to create program names from perf_event__synthesize_one_bpf_prog()")
Cc: Song Liu <songliubraving@fb.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
dso__read_running_kernel_build_id() uses sprintf() to format a sysfs
path from machine->root_dir into a PATH_MAX buffer. If root_dir is
close to PATH_MAX in length, appending "/sys/kernel/notes" (18 bytes)
overflows the stack buffer.
Switch to snprintf() with sizeof(path) to prevent the overflow.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Fixes: cdd059d731 ("perf tools: Move dso_* related functions into dso object")
Cc: Jiri Olsa <jolsa@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
hwmon_pmu__read_events() reads label files with read() into a stack
buffer, strips trailing newlines, then checks buf[0] == '\0'. When
read() returns 0 (empty file) or -1 (error), the buffer is never
written, so buf[0] reads uninitialized stack memory. If the garbage
byte is non-zero, the code falls through to strdup(buf) which copies
arbitrary stack data as the label string.
Fix by checking read_len <= 0 before accessing buf contents, closing
the fd and skipping the entry.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Fixes: 53cc0b351e ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs")
Cc: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
filename__decompress() has an early return path for files that are not
actually compressed. This path returns the fd from open() directly but
never writes to the pathname output parameter, leaving the caller with
an uninitialized buffer despite a successful return.
Callers like dso__decompress_kmodule_path() pass pathname to
decompress_kmodule() which uses it to set the decompressed file path.
If pathname is uninitialized, subsequent operations on the path produce
undefined behavior.
Fix by setting pathname to an empty string on the uncompressed path.
Callers already check for an empty pathname to distinguish temporary
decompressed files (which need unlink) from the original file.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Fixes: 7ac22b088a ("perf tools: Add filename__decompress function")
Cc: Jiri Olsa <jolsa@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When sysfs__read_build_id() matches NT_GNU_BUILD_ID with the right
namesz but the name content is not "GNU", it falls back to reading
descsz bytes into the stack buffer bf[BUFSIZ]:
} else if (read(fd, bf, descsz) != (ssize_t)descsz)
Unlike the else branch which validates namesz + descsz against
sizeof(bf), this path passes descsz directly to read() without any
bounds check. A crafted sysfs file with a large n_descsz overflows
the 8192-byte stack buffer.
Add a descsz > sizeof(bf) check before the read, breaking out of
the loop on oversized values.
Fixes: e5a1845fc0 ("perf symbols: Split out util/symbol-elf.c")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
parse_hwmon_filename() strips the "_alarm" suffix from event names
by copying into a 24-byte stack buffer:
strlcpy(fn_type, fn_item, fn_item_len - 5);
The third argument is the source length minus the suffix, not the
destination buffer capacity. A long event name ending in "_alarm"
can have fn_item_len - 5 > sizeof(fn_type), causing strlcpy() to
write past the 24-byte fn_type[] array. The assert() only validates
that the longest *valid* hwmon item fits, but does not protect
against crafted input.
Clamp the strlcpy size to min(fn_item_len - 5, sizeof(fn_type)).
Fixes: 4810b761f8 ("perf hwmon_pmu: Add hwmon filename parser")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
hwmon_pmu__for_each_event() formats description strings via:
len = snprintf(desc_buf, sizeof(desc_buf), "%s in unit %s named %s.", ...);
len += hwmon_pmu__describe_items(hwm, desc_buf + len, sizeof(desc_buf) - len, ...);
If value->label is long enough to cause snprintf() to truncate, it
returns the would-have-been-written count, making len exceed
sizeof(desc_buf). The subsequent sizeof(desc_buf) - len underflows
to a huge size_t value, disabling bounds checking in
hwmon_pmu__describe_items().
The alias_buf snprintf has the same issue. Switch both to scnprintf()
which returns actual bytes written.
Fixes: 53cc0b351e ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Three functions read sysfs files into fixed-size stack buffers using
the full buffer size, then null-terminate at buf[read_len]. If the
read fills the buffer exactly, read_len equals sizeof(buf) and the
null byte writes one past the array, corrupting an adjacent stack
variable.
Fix all three by reading sizeof(buf) - 1 bytes, reserving space for
the null terminator:
- hwmon_pmu__read_events(): buf[128]
- hwmon_pmu__describe_items(): buf[64]
- evsel__hwmon_pmu_read(): buf[32]
Fixes: 53cc0b351e ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
thread__set_comm_from_proc() calls procfs__read_str() then strips
the trailing newline via comm[sz - 1] = '\0'. procfs__read_str()
allocates the buffer before reading, so on an empty /proc/pid/comm
(reachable during late exit teardown) it returns success with sz = 0
and an unterminated heap buffer.
The sz - 1 underflow was the original sashiko finding: it writes a
null byte before the allocation. But even with a sz > 0 guard on
the newline strip, the unterminated buffer would still be passed to
thread__set_comm() which calls strlen() — an unbounded heap read.
Fix by treating sz == 0 as failure: free the buffer and return -1.
This is consistent with pmu.c's perf_pmu__parse_scale/unit which
already treat len == 0 from filename__read_str as an error.
Fixes: 2f3027ac28 ("perf thread: Introduce method to set comm from /proc/pid/self")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
dump_insn() tracks remaining buffer space with a 'left' variable,
but the loop subtracts the cumulative offset 'n' each iteration
instead of just the per-iteration delta:
n += snprintf(x->out + n, left, "%02x ", inbuf[i]);
left -= n; /* BUG: n is cumulative, not the delta */
After two iterations left goes massively negative, wrapping to a
huge value when passed as size_t to snprintf(), disabling all bounds
checking for the rest of the loop.
Switch to scnprintf() accumulation using sizeof(x->out) - n as the
remaining space, which is always correct and eliminates the separate
'left' variable entirely.
Fixes: 48d02a1d5c ("perf script: Add 'brstackinsn' for branch stacks")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
mkstemp() creates file descriptors without the close-on-exec flag.
These fds leak to child processes spawned during symbol resolution
(addr2line, objdump), wasting descriptors and potentially exposing
temporary file contents.
Replace mkstemp() with mkostemp(tmpbuf, O_CLOEXEC) at all three
call sites:
- filename__decompress() in dso.c
- read_gnu_debugdata() in symbol-elf.c
- kcore__init() in symbol-elf.c
Fixes: 42b3fa6708 ("perf tools: Introduce dso__decompress_kmodule_{fd,path}")
Fixes: b10f74308e ("perf symbol: Support .gnu_debugdata for symbols")
Fixes: afba19d9dc ("perf symbols: Workaround objdump difficulties with kcore")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephen Brennan <stephen.s.brennan@oracle.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
filename__read_debuglink() copies .gnu_debuglink section data into a
caller-provided buffer via:
strncpy(debuglink, data->d_buf, size);
where size is PATH_MAX. If the ELF section is smaller than size and
lacks a null terminator, strncpy reads past data->d_buf into adjacent
memory. A malformed ELF file can trigger this, potentially causing a
segfault or leaking heap data.
Additionally, strncpy does not guarantee null termination when the
source fills the buffer.
Replace with an explicit memcpy bounded by both the output buffer
size and the actual section data size (data->d_size), followed by
explicit null termination.
Fixes: e5a1845fc0 ("perf symbols: Split out util/symbol-elf.c")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
sysfs__read_build_id() reads ELF note headers from sysfs files. The
note's namesz and descsz fields are used to compute the skip size:
int n = namesz + descsz;
if (n > (int)sizeof(bf))
Both namesz and descsz are size_t from NOTE_ALIGN() of 32-bit note
header fields. Their sum can exceed INT_MAX, overflowing the signed
int n to a negative value. The check n > sizeof(bf) then evaluates
false (negative < positive in signed comparison), and read(fd, bf, n)
reinterprets the negative n as a huge size_t count — the kernel writes
up to MAX_RW_COUNT bytes into the 8192-byte stack buffer.
In practice the overflow is bounded by the sysfs file's actual size,
so a real sysfs notes file won't trigger it organically. But crafted
input (e.g. via a mounted debugfs/sysfs image) could.
Fix by validating namesz and descsz individually against the buffer
size before summing, and change n to size_t to avoid the signed
overflow entirely.
Fixes: f1617b4059 ("perf symbols: Record the build_ids of kernel modules too")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf_pmu__parse_scale() reads a PMU scale file then accesses
scale[sret - 1] to strip a trailing newline. Only sret < 0 is
guarded, so an empty file (sret == 0) causes scale[-1] — a stack
buffer underflow that reads and potentially writes out of bounds.
perf_pmu__parse_unit() has the same pattern: alias->unit[sret - 1]
with sret == 0 accesses the byte before the struct member, which
may corrupt the adjacent pmu_name pointer field.
Change both guards from sret < 0 to sret <= 0 so that empty files
are treated as read errors.
Fixes: 410136f5dd ("tools/perf/stat: Add event unit and scale support")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
pmu_id() calls filename__read_str() then strips the trailing newline
via str[len - 1] = 0. If the PMU identifier file is empty,
filename__read_str() succeeds with len = 0. len - 1 underflows
size_t to SIZE_MAX, writing a null byte before the heap allocation.
Add a len == 0 check before the newline stripping.
Fixes: 51d5484715 ("perf pmu: Add pmu_id()")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: John Garry <john.g.garry@oracle.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
I found this error when I pass EXTRA_CFLAGS=-fsanitize=address on Fedora
44 with GCC 16. Fix it by copying one less byte.
CC util/jitdump.o
util/jitdump.c: In function ‘jit_process’:
util/jitdump.c:237:9: error: ‘__builtin_strncpy’ specified bound 4096 equals destination size
[-Werror=stringop-truncation]
237 | strncpy(jd->dir, name, PATH_MAX);
| ^
cc1: all warnings being treated as errors
make[4]: *** [tools/build/Makefile.build:95: util/jitdump.o] Error 1
make[4]: *** Waiting for unfinished jobs....
make[3]: *** [tools/build/Makefile.build:158: util] Error 2
make[2]: *** [Makefile.perf:578: perf-util-in.o] Error 2
make[1]: *** [Makefile.perf:288: sub-make] Error 2
make: *** [Makefile:76: all] Error 2
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We already test branch output in perf script mode, but then retest it in
Perf report mode. This is more of a test of Perf itself than Coresight
because Perf uses the same samples to generate both outputs. Also we're
already testing instruction output in Perf report mode.
Remove this test for a speedup. On the systemwide test also remove the
Perf report test because systemwide mode records a lot more data so
running multiple tests on it has a big runtime impact.
Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Amir Ayupov <aaupov@meta.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Leach <mike.leach@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paschalis Mpeis <Paschalis.Mpeis@arm.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Like asm_pure_loop, this memcpy test only checks that 10 of each of a
few trace packet types occur after recording a lot of trace, which isn't
more specific than other existing Coresight tests.
Assume it was supposed to be a stress test for dumping and replace it
with one that doesn't require a custom binary and checks for a specific
amount of raw output. Don't bother checking for packets because the
other tests that test decoding will catch issues with malformed data.
This also adds coverage for exit snapshot mode which was missing.
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Leo Yan <leo.yan@arm.com>
Cc: Amir Ayupov <aaupov@meta.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Leach <mike.leach@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paschalis Mpeis <Paschalis.Mpeis@arm.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It's not obvious what this test is for so remove it. It's not a stress
test because it doesn't output lots of data and it's not a functional
test because it only looks for raw trace output. It seems to imply that
a program written in assembly influences whether trace would be
generated by the CPU or not, but the CPU doesn't know what language the
program is written in.
We already have lots of Coresight tests that test the full pipeline
including decoding, and in many more modes of operation than this one,
so if no trace was collected they will already fail leaving this one
redundant.
Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Amir Ayupov <aaupov@meta.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Leach <mike.leach@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paschalis Mpeis <Paschalis.Mpeis@arm.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Testing a long sequence without branches seems like it would be better
as a decoder unit test, and this test doesn't test decoding either, so
it's not clear what bugs this is trying to catch.
The new deterministic workload has somewhat long sequences when built
unoptimized, and we can always increase them later if we want to. But
now we test that decoding always gives the same result for the same
sequence of code which we've never had before.
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Leo Yan <leo.yan@arm.com>
Cc: Amir Ayupov <aaupov@meta.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Leach <mike.leach@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paschalis Mpeis <Paschalis.Mpeis@arm.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a --workload-ctl=fifo:ctl-fifo[,ack-fifo] option for 'perf test
-w'. When set, run_workload() opens the named FIFO, writes enable before
invoking the builtin workload, writes disable before returning, and
waits for ack responses when an ack FIFO is provided to ensure that the
workload doesn't run until the events are enabled.
This can be used to limit the scope of the recording to only the
workload execution and avoid recording Perf setup and teardown code if
Perf record is started with events disabled (-D 1).
Assisted-by: Codex:GPT-5.5
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Amir Ayupov <aaupov@meta.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Leo Yan <leo.yan@arm.com>
Cc: Mike Leach <mike.leach@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paschalis Mpeis <Paschalis.Mpeis@arm.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
PE_CONTEXT elements update the context ID and exception level, but the
decoder may still have prior packets cached for frontend processing.
Updating the context immediately in the decoder backend can make those
cached packets get consumed with the wrong thread or EL state.
Add a CS_ETM_CONTEXT packet carrying the TID and EL to the frontend,
this keeps context changes ordered with the rest of the packet stream
and avoids mismatches when synthesizing samples from cached packets.
Separate the memory access function into one for the frontend and one
for decoding. The frontend also needs memory access to attach the
instruction to samples. Because the frontend does memory access for
both previous and current packets, change all the frontend memory access
function signatures to take both a tidq and packet. But backend always
uses the current backend EL and thread from the tidq.
Treat context packets as a boundary for branch sample generation and
remove tidq->prev_packet_thread because it's not possible to branch to a
different thread, so only tracking the current thread is required for
sample generation.
Fixes: e573e978fb ("perf cs-etm: Inject capabilitity for CoreSight traces")
Reported-by: Amir Ayupov <aaupov@meta.com>
Closes: https://lore.kernel.org/linux-perf-users/20260515021135.1729028-1-aaupov@meta.com/
Co-authored-by: James Clark <james.clark@linaro.org>
Signed-off-by: Leo Yan <leo.yan@arm.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: Mike Leach <mike.leach@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paschalis Mpeis <Paschalis.Mpeis@arm.com>
Cc: Robert Walker <robert.walker@arm.com>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Suzuki Poulouse <suzuki.poulose@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Replace snprintf with scnprintf in buffer offset calculations to
ensure the 'used' count will not exceed the "len".
The current logic in perf_pmu__for_each_event uses an unconditional
+ 1 increment to buf_used to account for null terminators. This can
cause a stack buffer overflow in the subsequent scnprintf call.
When the local stack buffer buf (1024 bytes) is full, buf_used can
reach 1025. This causes the subsequent remaining space calculation
sizeof(buf) - buf_used to underflow.
Use sub_non_neg() to see if space actually existed, and only
increment the offset if remaining space is present.
Changes includes:
- Use sub_non_neg to check if space exists
- Replacing snprintf with scnprintf to ensure the return value
reflects the actual bytes written into the buffer.
- Only increment buf_used by 1 if space exists
- If a parameterized event uses a built-in perf keyword for its
parameter name (eg, config=?), the lexer parses it as a predefined
term token, which sets term->config to NULL. Add check to use
parse_events__term_type_str() if term->config is NULL.
Signed-off-by: Athira Rajeev <atrajeev@linux.ibm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Shivani Nittor <shivani@linux.ibm.com>
Cc: Tanushree.Shah@ibm.com
Cc: Tejas.Manhas1@ibm.com
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
