linux-stable.git/kernel/trace, branch linux-5.15.y

tracing: Fix potential deadlock in cpu hotplug with osnoise

2026-04-18T08:33:38+00:00

[ Upstream commit 1f9885732248d22f788e4992c739a98c88ab8a55 ]

The following sequence may leads deadlock in cpu hotplug:

    task1        task2        task3
    -----        -----        -----

 mutex_lock(&interface_lock)

            [CPU GOING OFFLINE]

            cpus_write_lock();
            osnoise_cpu_die();
              kthread_stop(task3);
                wait_for_completion();

                      osnoise_sleep();
                        mutex_lock(&interface_lock);

 cpus_read_lock();

 [DEAD LOCK]

Fix by swap the order of cpus_read_lock() and mutex_lock(&interface_lock).

Cc: stable@vger.kernel.org
Cc: 
Cc: 
Cc: 
Cc: 
Fixes: bce29ac9ce0bb ("trace: Add osnoise tracer")
Link: https://patch.msgid.link/20260326141953414bVSj33dAYktqp9Oiyizq8@zte.com.cn
Reviewed-by: Masami Hiramatsu (Google) 
Signed-off-by: Luo Haiyang 
Signed-off-by: Steven Rostedt (Google) 
[ adapted guard() macros to lock/unlock calls ]
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

tracing: Fix trace_buf_size= cmdline parameter with sizes >= 2G

2026-04-18T08:33:15+00:00

commit d008ba8be8984760e36d7dcd4adbd5a41a645708 upstream.

Some of the sizing logic through tracer_alloc_buffers() uses int
internally, causing unexpected behavior if the user passes a value that
does not fit in an int (on my x86 machine, the result is uselessly tiny
buffers).

Fix by plumbing the parameter's real type (unsigned long) through to the
ring buffer allocation functions, which already use unsigned long.

It has always been possible to create larger ring buffers via the sysfs
interface: this only affects the cmdline parameter.

Cc: stable@vger.kernel.org
Cc: Masami Hiramatsu 
Cc: Mathieu Desnoyers 
Link: https://patch.msgid.link/bff42a4288aada08bdf74da3f5b67a2c28b761f8.1772852067.git.calvin@wbinvd.org
Fixes: 73c5162aa362 ("tracing: keep ring buffer to minimum size till used")
Signed-off-by: Calvin Owens 
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Greg Kroah-Hartman

tracing: Fix to set write permission to per-cpu buffer_size_kb

2026-03-04T12:20:35+00:00

[ Upstream commit f844282deed7481cf2f813933229261e27306551 ]

Since the per-cpu buffer_size_kb file is writable for changing
per-cpu ring buffer size, the file should have the write access
permission.

Cc: stable@vger.kernel.org
Cc: Mathieu Desnoyers 
Link: https://patch.msgid.link/177071301597.2293046.11683339475076917920.stgit@mhiramat.tok.corp.google.com
Fixes: 21ccc9cd7211 ("tracing: Disable "other" permission bits in the tracefs files")
Signed-off-by: Masami Hiramatsu (Google) 
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Sasha Levin

tracing: Fix false sharing in hwlat get_sample()

2026-03-04T12:20:14+00:00

[ Upstream commit f743435f988cb0cf1f521035aee857851b25e06d ]

The get_sample() function in the hwlat tracer assumes the caller holds
hwlat_data.lock, but this is not actually happening. The result is
unprotected data access to hwlat_data, and in per-cpu mode can result in
false sharing which may show up as false positive latency events.

The specific case of false sharing observed was primarily between
hwlat_data.sample_width and hwlat_data.count. These are separated by
just 8B and are therefore likely to share a cache line. When one thread
modifies count, the cache line is in a modified state so when other
threads read sample_width in the main latency detection loop, they fetch
the modified cache line. On some systems, the fetch itself may be slow
enough to count as a latency event, which could set up a self
reinforcing cycle of latency events as each event increments count which
then causes more latency events, continuing the cycle.

The other result of the unprotected data access is that hwlat_data.count
can end up with duplicate or missed values, which was observed on some
systems in testing.

Convert hwlat_data.count to atomic64_t so it can be safely modified
without locking, and prevent false sharing by pulling sample_width into
a local variable.

One system this was tested on was a dual socket server with 32 CPUs on
each numa node. With settings of 1us threshold, 1000us width, and
2000us window, this change reduced the number of latency events from
500 per second down to approximately 1 event per minute. Some machines
tested did not exhibit measurable latency from the false sharing.

Cc: Masami Hiramatsu 
Cc: Mathieu Desnoyers 
Link: https://patch.msgid.link/20260210074810.6328-1-clord@mykolab.com
Signed-off-by: Colin Lord 
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Sasha Levin

tracing: Remove duplicate ENABLE_EVENT_STR and DISABLE_EVENT_STR macros

2026-03-04T12:19:44+00:00

[ Upstream commit 9df0e49c5b9b8d051529be9994e4f92f2d20be6f ]

The macros ENABLE_EVENT_STR and DISABLE_EVENT_STR were added to trace.h so
that more than one file can have access to them, but was never removed
from their original location. Remove the duplicates.

Cc: Masami Hiramatsu 
Cc: Mathieu Desnoyers 
Cc: Tom Zanussi 
Link: https://patch.msgid.link/20260126130037.4ba201f9@gandalf.local.home
Fixes: d0bad49bb0a09 ("tracing: Add enable_hist/disable_hist triggers")
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Sasha Levin

tracing: Fix ftrace event field alignments

2026-02-11T12:35:33+00:00

[ Upstream commit 033c55fe2e326bea022c3cc5178ecf3e0e459b82 ]

The fields of ftrace specific events (events used to save ftrace internal
events like function traces and trace_printk) are generated similarly to
how normal trace event fields are generated. That is, the fields are added
to a trace_events_fields array that saves the name, offset, size,
alignment and signness of the field. It is used to produce the output in
the format file in tracefs so that tooling knows how to parse the binary
data of the trace events.

The issue is that some of the ftrace event structures are packed. The
function graph exit event structures are one of them. The 64 bit calltime
and rettime fields end up 4 byte aligned, but the algorithm to show to
userspace shows them as 8 byte aligned.

The macros that create the ftrace events has one for embedded structure
fields. There's two macros for theses fields:

  __field_desc() and __field_packed()

The difference of the latter macro is that it treats the field as packed.

Rename that field to __field_desc_packed() and create replace the
__field_packed() to be a normal field that is packed and have the calltime
and rettime use those.

This showed up on 32bit architectures for function graph time fields. It
had:

 ~# cat /sys/kernel/tracing/events/ftrace/funcgraph_exit/format
[..]
        field:unsigned long func;       offset:8;       size:4; signed:0;
        field:unsigned int depth;       offset:12;      size:4; signed:0;
        field:unsigned int overrun;     offset:16;      size:4; signed:0;
        field:unsigned long long calltime;      offset:24;      size:8; signed:0;
        field:unsigned long long rettime;       offset:32;      size:8; signed:0;

Notice that overrun is at offset 16 with size 4, where in the structure
calltime is at offset 20 (16 + 4), but it shows the offset at 24. That's
because it used the alignment of unsigned long long when used as a
declaration and not as a member of a structure where it would be aligned
by word size (in this case 4).

By using the proper structure alignment, the format has it at the correct
offset:

 ~# cat /sys/kernel/tracing/events/ftrace/funcgraph_exit/format
[..]
        field:unsigned long func;       offset:8;       size:4; signed:0;
        field:unsigned int depth;       offset:12;      size:4; signed:0;
        field:unsigned int overrun;     offset:16;      size:4; signed:0;
        field:unsigned long long calltime;      offset:20;      size:8; signed:0;
        field:unsigned long long rettime;       offset:28;      size:8; signed:0;

Cc: stable@vger.kernel.org
Cc: Mathieu Desnoyers 
Cc: Mark Rutland 
Acked-by: Masami Hiramatsu (Google) 
Reported-by: "jempty.liang" 
Link: https://patch.msgid.link/20260204113628.53faec78@gandalf.local.home
Fixes: 04ae87a52074e ("ftrace: Rework event_create_dir()")
Closes: https://lore.kernel.org/all/20260130015740.212343-1-imntjempty@163.com/
Closes: https://lore.kernel.org/all/20260202123342.2544795-1-imntjempty@163.com/
Signed-off-by: Steven Rostedt (Google) 
[ Context / renames ]
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

ring-buffer: Avoid softlockup in ring_buffer_resize() during memory free

2026-02-11T12:35:30+00:00

[ Upstream commit 6435ffd6c7fcba330dfa91c58dc30aed2df3d0bf ]

When user resize all trace ring buffer through file 'buffer_size_kb',
then in ring_buffer_resize(), kernel allocates buffer pages for each
cpu in a loop.

If the kernel preemption model is PREEMPT_NONE and there are many cpus
and there are many buffer pages to be freed, it may not give up cpu
for a long time and finally cause a softlockup.

To avoid it, call cond_resched() after each cpu buffer free as Commit
f6bd2c92488c ("ring-buffer: Avoid softlockup in ring_buffer_resize()")
does.

Detailed call trace as follow:

  rcu: INFO: rcu_sched self-detected stall on CPU
  rcu: 	24-....: (14837 ticks this GP) idle=521c/1/0x4000000000000000 softirq=230597/230597 fqs=5329
  rcu: 	(t=15004 jiffies g=26003221 q=211022 ncpus=96)
  CPU: 24 UID: 0 PID: 11253 Comm: bash Kdump: loaded Tainted: G            EL      6.18.2+ #278 NONE
  pc : arch_local_irq_restore+0x8/0x20
   arch_local_irq_restore+0x8/0x20 (P)
   free_frozen_page_commit+0x28c/0x3b0
   __free_frozen_pages+0x1c0/0x678
   ___free_pages+0xc0/0xe0
   free_pages+0x3c/0x50
   ring_buffer_resize.part.0+0x6a8/0x880
   ring_buffer_resize+0x3c/0x58
   __tracing_resize_ring_buffer.part.0+0x34/0xd8
   tracing_resize_ring_buffer+0x8c/0xd0
   tracing_entries_write+0x74/0xd8
   vfs_write+0xcc/0x288
   ksys_write+0x74/0x118
   __arm64_sys_write+0x24/0x38

Cc: 
Link: https://patch.msgid.link/20251228065008.2396573-1-mawupeng1@huawei.com
Signed-off-by: Wupeng Ma 
Acked-by: Masami Hiramatsu (Google) 
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Sasha Levin

tracing: Do not register unsupported perf events

2026-01-19T12:09:53+00:00

commit ef7f38df890f5dcd2ae62f8dbde191d72f3bebae upstream.

Synthetic events currently do not have a function to register perf events.
This leads to calling the tracepoint register functions with a NULL
function pointer which triggers:

 ------------[ cut here ]------------
 WARNING: kernel/tracepoint.c:175 at tracepoint_add_func+0x357/0x370, CPU#2: perf/2272
 Modules linked in: kvm_intel kvm irqbypass
 CPU: 2 UID: 0 PID: 2272 Comm: perf Not tainted 6.18.0-ftest-11964-ge022764176fc-dirty #323 PREEMPTLAZY
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.17.0-debian-1.17.0-1 04/01/2014
 RIP: 0010:tracepoint_add_func+0x357/0x370
 Code: 28 9c e8 4c 0b f5 ff eb 0f 4c 89 f7 48 c7 c6 80 4d 28 9c e8 ab 89 f4 ff 31 c0 5b 41 5c 41 5d 41 5e 41 5f 5d c3 cc cc cc cc cc <0f> 0b 49 c7 c6 ea ff ff ff e9 ee fe ff ff 0f 0b e9 f9 fe ff ff 0f
 RSP: 0018:ffffabc0c44d3c40 EFLAGS: 00010246
 RAX: 0000000000000001 RBX: ffff9380aa9e4060 RCX: 0000000000000000
 RDX: 000000000000000a RSI: ffffffff9e1d4a98 RDI: ffff937fcf5fd6c8
 RBP: 0000000000000001 R08: 0000000000000007 R09: ffff937fcf5fc780
 R10: 0000000000000003 R11: ffffffff9c193910 R12: 000000000000000a
 R13: ffffffff9e1e5888 R14: 0000000000000000 R15: ffffabc0c44d3c78
 FS:  00007f6202f5f340(0000) GS:ffff93819f00f000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000055d3162281a8 CR3: 0000000106a56003 CR4: 0000000000172ef0
 Call Trace:
  
  tracepoint_probe_register+0x5d/0x90
  synth_event_reg+0x3c/0x60
  perf_trace_event_init+0x204/0x340
  perf_trace_init+0x85/0xd0
  perf_tp_event_init+0x2e/0x50
  perf_try_init_event+0x6f/0x230
  ? perf_event_alloc+0x4bb/0xdc0
  perf_event_alloc+0x65a/0xdc0
  __se_sys_perf_event_open+0x290/0x9f0
  do_syscall_64+0x93/0x7b0
  ? entry_SYSCALL_64_after_hwframe+0x76/0x7e
  ? trace_hardirqs_off+0x53/0xc0
  entry_SYSCALL_64_after_hwframe+0x76/0x7e

Instead, have the code return -ENODEV, which doesn't warn and has perf
error out with:

 # perf record -e synthetic:futex_wait
Error:
The sys_perf_event_open() syscall returned with 19 (No such device) for event (synthetic:futex_wait).
"dmesg | grep -i perf" may provide additional information.

Ideally perf should support synthetic events, but for now just fix the
warning. The support can come later.

Cc: stable@vger.kernel.org
Cc: Masami Hiramatsu 
Cc: Mathieu Desnoyers 
Cc: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://patch.msgid.link/20251216182440.147e4453@gandalf.local.home
Fixes: 4b147936fa509 ("tracing: Add support for 'synthetic' events")
Reported-by: Ian Rogers 
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Greg Kroah-Hartman

tracing: Fix memory leaks in create_field_var()

2025-12-06T21:09:26+00:00

[ Upstream commit 80f0d631dcc76ee1b7755bfca1d8417d91d71414 ]

The function create_field_var() allocates memory for 'val' through
create_hist_field() inside parse_atom(), and for 'var' through
create_var(), which in turn allocates var->type and var->var.name
internally. Simply calling kfree() to release these structures will
result in memory leaks.

Use destroy_hist_field() to properly free 'val', and explicitly release
the memory of var->type and var->var.name before freeing 'var' itself.

Link: https://patch.msgid.link/20251106120132.3639920-1-zilin@seu.edu.cn
Fixes: 02205a6752f22 ("tracing: Add support for 'field variables'")
Signed-off-by: Zilin Guan 
Signed-off-by: Steven Rostedt (Google) 
Signed-off-by: Sasha Levin

minmax: make generic MIN() and MAX() macros available everywhere

2025-10-19T14:21:52+00:00

[ Upstream commit 1a251f52cfdc417c84411a056bc142cbd77baef4 ]

This just standardizes the use of MIN() and MAX() macros, with the very
traditional semantics.  The goal is to use these for C constant
expressions and for top-level / static initializers, and so be able to
simplify the min()/max() macros.

These macro names were used by various kernel code - they are very
traditional, after all - and all such users have been fixed up, with a
few different approaches:

 - trivial duplicated macro definitions have been removed

   Note that 'trivial' here means that it's obviously kernel code that
   already included all the major kernel headers, and thus gets the new
   generic MIN/MAX macros automatically.

 - non-trivial duplicated macro definitions are guarded with #ifndef

   This is the "yes, they define their own versions, but no, the include
   situation is not entirely obvious, and maybe they don't get the
   generic version automatically" case.

 - strange use case #1

   A couple of drivers decided that the way they want to describe their
   versioning is with

	#define MAJ 1
	#define MIN 2
	#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN)

   which adds zero value and I just did my Alexander the Great
   impersonation, and rewrote that pointless Gordian knot as

	#define DRV_VERSION "1.2"

   instead.

 - strange use case #2

   A couple of drivers thought that it's a good idea to have a random
   'MIN' or 'MAX' define for a value or index into a table, rather than
   the traditional macro that takes arguments.

   These values were re-written as C enum's instead. The new
   function-line macros only expand when followed by an open
   parenthesis, and thus don't clash with enum use.

Happily, there weren't really all that many of these cases, and a lot of
users already had the pattern of using '#ifndef' guarding (or in one
case just using '#undef MIN') before defining their own private version
that does the same thing. I left such cases alone.

Cc: David Laight 
Cc: Lorenzo Stoakes 
Signed-off-by: Linus Torvalds 
Signed-off-by: Eliav Farber 
Signed-off-by: Greg Kroah-Hartman