[PATCH] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[PATCH] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[David Wang]

commit a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting
bug at task exit") made changes to __perf_remove_from_context() to
coordinate its changes with perf_event_exit_event(), but the change are
unconditional, it impacts callpaths to __perf_remove_from_context()
other than from perf_event_exit_event(). One of the impact is to cgroup,
which is not properly handled and would cause kernel panic with high
probalibity during reboot on some system[1].

To confine the side effects, make the changes to
__perf_remove_from_context() conditional, restore to its previous state
except when DETACH_EXIT is set.

Closes: https://lore.kernel.org/lkml/20250601173603.3920-1-00107082@163.com/ [1]
Fixes: a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting bug at task exit")
Signed-off-by: David Wang <00107082@163.com>
---
 kernel/events/core.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 95e703891b24..6a7e3f5c5af5 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2466,7 +2466,7 @@ __perf_remove_from_context(struct perf_event *event,
 			   void *info)
 {
 	struct perf_event_pmu_context *pmu_ctx = event->pmu_ctx;
-	enum perf_event_state state = PERF_EVENT_STATE_OFF;
+	enum perf_event_state exit_state = PERF_EVENT_STATE_EXIT;
 	unsigned long flags = (unsigned long)info;
 
 	ctx_time_update(cpuctx, ctx);
@@ -2475,19 +2475,20 @@ __perf_remove_from_context(struct perf_event *event,
 	 * Ensure event_sched_out() switches to OFF, at the very least
 	 * this avoids raising perf_pending_task() at this time.
 	 */
-	if (flags & DETACH_EXIT)
-		state = PERF_EVENT_STATE_EXIT;
 	if (flags & DETACH_DEAD) {
 		event->pending_disable = 1;
-		state = PERF_EVENT_STATE_DEAD;
+		exit_state = PERF_EVENT_STATE_DEAD;
 	}
 	event_sched_out(event, ctx);
-	perf_event_set_state(event, min(event->state, state));
+	if (flags & DETACH_EXIT)
+		perf_event_set_state(event, min(event->state, exit_state));
 	if (flags & DETACH_GROUP)
 		perf_group_detach(event);
 	if (flags & DETACH_CHILD)
 		perf_child_detach(event);
 	list_del_event(event, ctx);
+	if (flags & DETACH_DEAD)
+		event->state = PERF_EVENT_STATE_DEAD;
 
 	if (!pmu_ctx->nr_events) {
 		pmu_ctx->rotate_necessary = 0;
-- 
2.39.2

[PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[David Wang]

commit a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting
bug at task exit") made changes to __perf_remove_from_context() to
coordinate its changes with perf_event_exit_event(), but the change are
unconditional, it impacts callpaths to __perf_remove_from_context()
other than from perf_event_exit_event(). One of the impact is to cgroup,
which is not properly handled and would cause kernel panic with high
probalibity during reboot on some system[1].

To confine the side effects, make the changes to
__perf_remove_from_context() conditional, restore to its previous state
except when DETACH_EXIT is set.

Closes: https://lore.kernel.org/lkml/20250601173603.3920-1-00107082@163.com/ [1]
Fixes: a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting bug at task exit")
Signed-off-by: David Wang <00107082@163.com>
---
Changes:
Address yeoreum.yun@arm.com's concern about missing cgroup event.
---
 kernel/events/core.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 95e703891b24..e2c0f34b0789 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2466,7 +2466,7 @@ __perf_remove_from_context(struct perf_event *event,
 			   void *info)
 {
 	struct perf_event_pmu_context *pmu_ctx = event->pmu_ctx;
-	enum perf_event_state state = PERF_EVENT_STATE_OFF;
+	enum perf_event_state exit_state = PERF_EVENT_STATE_EXIT;
 	unsigned long flags = (unsigned long)info;
 
 	ctx_time_update(cpuctx, ctx);
@@ -2475,19 +2475,20 @@ __perf_remove_from_context(struct perf_event *event,
 	 * Ensure event_sched_out() switches to OFF, at the very least
 	 * this avoids raising perf_pending_task() at this time.
 	 */
-	if (flags & DETACH_EXIT)
-		state = PERF_EVENT_STATE_EXIT;
 	if (flags & DETACH_DEAD) {
 		event->pending_disable = 1;
-		state = PERF_EVENT_STATE_DEAD;
+		exit_state = PERF_EVENT_STATE_DEAD;
 	}
 	event_sched_out(event, ctx);
-	perf_event_set_state(event, min(event->state, state));
 	if (flags & DETACH_GROUP)
 		perf_group_detach(event);
 	if (flags & DETACH_CHILD)
 		perf_child_detach(event);
 	list_del_event(event, ctx);
+	if (flags & DETACH_EXIT)
+		perf_event_set_state(event, min(event->state, exit_state));
+	if (flags & DETACH_DEAD)
+		event->state = PERF_EVENT_STATE_DEAD;
 
 	if (!pmu_ctx->nr_events) {
 		pmu_ctx->rotate_necessary = 0;
-- 
2.39.2

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[Peter Zijlstra]

On Tue, Jun 03, 2025 at 04:33:04PM +0800, David Wang wrote:
> commit a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting
> bug at task exit") made changes to __perf_remove_from_context() to
> coordinate its changes with perf_event_exit_event(), but the change are
> unconditional, it impacts callpaths to __perf_remove_from_context()
> other than from perf_event_exit_event(). One of the impact is to cgroup,
> which is not properly handled and would cause kernel panic with high
> probalibity during reboot on some system[1].

Sorry, but no. This does not describe the problem adequately. I would
have to go read your [1] to figure out what is actually broken.

That is, having read the above, I'm still clueless as to what the actual
problem is.

> To confine the side effects, make the changes to
> __perf_remove_from_context() conditional, restore to its previous state
> except when DETACH_EXIT is set.
> 
> Closes: https://lore.kernel.org/lkml/20250601173603.3920-1-00107082@163.com/ [1]
> Fixes: a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting bug at task exit")
> Signed-off-by: David Wang <00107082@163.com>
> ---
> Changes:
> Address yeoreum.yun@arm.com's concern about missing cgroup event.
> ---
>  kernel/events/core.c | 11 ++++++-----
>  1 file changed, 6 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 95e703891b24..e2c0f34b0789 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -2466,7 +2466,7 @@ __perf_remove_from_context(struct perf_event *event,
>  			   void *info)
>  {
>  	struct perf_event_pmu_context *pmu_ctx = event->pmu_ctx;
> -	enum perf_event_state state = PERF_EVENT_STATE_OFF;
> +	enum perf_event_state exit_state = PERF_EVENT_STATE_EXIT;
>  	unsigned long flags = (unsigned long)info;
>  
>  	ctx_time_update(cpuctx, ctx);
> @@ -2475,19 +2475,20 @@ __perf_remove_from_context(struct perf_event *event,
>  	 * Ensure event_sched_out() switches to OFF, at the very least
>  	 * this avoids raising perf_pending_task() at this time.
>  	 */
> -	if (flags & DETACH_EXIT)
> -		state = PERF_EVENT_STATE_EXIT;
>  	if (flags & DETACH_DEAD) {
>  		event->pending_disable = 1;
> -		state = PERF_EVENT_STATE_DEAD;
> +		exit_state = PERF_EVENT_STATE_DEAD;
>  	}
>  	event_sched_out(event, ctx);
> -	perf_event_set_state(event, min(event->state, state));
>  	if (flags & DETACH_GROUP)
>  		perf_group_detach(event);
>  	if (flags & DETACH_CHILD)
>  		perf_child_detach(event);
>  	list_del_event(event, ctx);
> +	if (flags & DETACH_EXIT)
> +		perf_event_set_state(event, min(event->state, exit_state));
> +	if (flags & DETACH_DEAD)
> +		event->state = PERF_EVENT_STATE_DEAD;

Urgh, no. Trying to reverse engineer the above, the intent appears to be
to not set OFF.

This can be achieved by doing:

-       enum perf_event_state state = PERF_EVENT_STATE_OFF;
+       enum perf_event_state state = event->state;

No other changes required. You also move the location of
perf_event_set_state(), but it is entirely unclear to me if that is
actually needed.

Worse, you split the means of setting state -- that is entirely uncalled
for.

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[David Wang]

At 2025-06-03 17:13:52, "Peter Zijlstra" <peterz@infradead.org> wrote:
>On Tue, Jun 03, 2025 at 04:33:04PM +0800, David Wang wrote:
>> commit a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting
>> bug at task exit") made changes to __perf_remove_from_context() to
>> coordinate its changes with perf_event_exit_event(), but the change are
>> unconditional, it impacts callpaths to __perf_remove_from_context()
>> other than from perf_event_exit_event(). One of the impact is to cgroup,
>> which is not properly handled and would cause kernel panic with high
>> probalibity during reboot on some system[1].
>
>Sorry, but no. This does not describe the problem adequately. I would
>have to go read your [1] to figure out what is actually broken.
>
>That is, having read the above, I'm still clueless as to what the actual
>problem is.

well, short story is commit a3c3c66670ce introduce a kernel panic when reboot the system
after perf_event_open with cgroup.
My understanding is commit a3c3c66670ce make changes to call path
perf_event_exit_event() --> __perf_remove_from_context(), but this changes affect other
call path as well, for example
perf_event_release_kernel() --> perf_remove_from_context()
(As yeoreum.yun@arm.com pointed out,  the change in perf_remove_from_context() made
perf_event_set_state() happened before list_del_event(), resulting in perf_cgroup_event_disable()
not called.)

My suggestion here is to confine the effect of commit a3c3c66670ce only to call chain
perf_event_exit_event() --> __perf_remove_from_context()


(But this v2 version is totally wrong, should be ignored; it breaks commit a3c3c66670ce)



>
>> To confine the side effects, make the changes to
>> __perf_remove_from_context() conditional, restore to its previous state
>> except when DETACH_EXIT is set.
>> 
>> Closes: https://lore.kernel.org/lkml/20250601173603.3920-1-00107082@163.com/ [1]
>> Fixes: a3c3c66670ce ("perf/core: Fix child_total_time_enabled accounting bug at task exit")
>> Signed-off-by: David Wang <00107082@163.com>
>> ---
>> Changes:
>> Address yeoreum.yun@arm.com's concern about missing cgroup event.
>> ---
>>  kernel/events/core.c | 11 ++++++-----
>>  1 file changed, 6 insertions(+), 5 deletions(-)
>> 
>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>> index 95e703891b24..e2c0f34b0789 100644
>> --- a/kernel/events/core.c
>> +++ b/kernel/events/core.c
>> @@ -2466,7 +2466,7 @@ __perf_remove_from_context(struct perf_event *event,
>>  			   void *info)
>>  {
>>  	struct perf_event_pmu_context *pmu_ctx = event->pmu_ctx;
>> -	enum perf_event_state state = PERF_EVENT_STATE_OFF;
>> +	enum perf_event_state exit_state = PERF_EVENT_STATE_EXIT;
>>  	unsigned long flags = (unsigned long)info;
>>  
>>  	ctx_time_update(cpuctx, ctx);
>> @@ -2475,19 +2475,20 @@ __perf_remove_from_context(struct perf_event *event,
>>  	 * Ensure event_sched_out() switches to OFF, at the very least
>>  	 * this avoids raising perf_pending_task() at this time.
>>  	 */
>> -	if (flags & DETACH_EXIT)
>> -		state = PERF_EVENT_STATE_EXIT;
>>  	if (flags & DETACH_DEAD) {
>>  		event->pending_disable = 1;
>> -		state = PERF_EVENT_STATE_DEAD;
>> +		exit_state = PERF_EVENT_STATE_DEAD;
>>  	}
>>  	event_sched_out(event, ctx);
>> -	perf_event_set_state(event, min(event->state, state));
>>  	if (flags & DETACH_GROUP)
>>  		perf_group_detach(event);
>>  	if (flags & DETACH_CHILD)
>>  		perf_child_detach(event);
>>  	list_del_event(event, ctx);
>> +	if (flags & DETACH_EXIT)
>> +		perf_event_set_state(event, min(event->state, exit_state));
>> +	if (flags & DETACH_DEAD)
>> +		event->state = PERF_EVENT_STATE_DEAD;
>
>Urgh, no. Trying to reverse engineer the above, the intent appears to be
>to not set OFF.
>
>This can be achieved by doing:
>
>-       enum perf_event_state state = PERF_EVENT_STATE_OFF;
>+       enum perf_event_state state = event->state;
>
>No other changes required. You also move the location of
>perf_event_set_state(), but it is entirely unclear to me if that is
>actually needed.
>
>Worse, you split the means of setting state -- that is entirely uncalled
>for. 
Yes, that is very wired to me too..... commit  a3c3c66670ce wants to use perf_event_set_state to update time,
but the original code use just event->state = ...

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[Peter Zijlstra]

On Tue, Jun 03, 2025 at 06:44:58PM +0800, David Wang wrote:


> (As yeoreum.yun@arm.com pointed out,  the change in perf_remove_from_context() made
> perf_event_set_state() happened before list_del_event(), resulting in perf_cgroup_event_disable()
> not called.)

Aah, d'0h. Let me see what we should do there.

> My suggestion here is to confine the effect of commit a3c3c66670ce only to call chain
> perf_event_exit_event() --> __perf_remove_from_context()
> 
> 
> (But this v2 version is totally wrong, should be ignored; it breaks commit a3c3c66670ce)

Right. Because we moved that state update earlier because
perf_child_detach() wants up-to-date timestamps.

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[Peter Zijlstra]

On Tue, Jun 03, 2025 at 02:50:56PM +0200, Peter Zijlstra wrote:
> On Tue, Jun 03, 2025 at 06:44:58PM +0800, David Wang wrote:
> 
> 
> > (As yeoreum.yun@arm.com pointed out,  the change in perf_remove_from_context() made
> > perf_event_set_state() happened before list_del_event(), resulting in perf_cgroup_event_disable()
> > not called.)
> 
> Aah, d'0h. Let me see what we should do there.

Does this help? This way event_sched_out() will call
perf_cgroup_event_disable().


diff --git a/kernel/events/core.c b/kernel/events/core.c
index f34c99f8ce8f..adbb0372825f 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2494,9 +2494,9 @@ __perf_remove_from_context(struct perf_event *event,
 	if (flags & DETACH_REVOKE)
 		state = PERF_EVENT_STATE_REVOKED;
 	if (flags & DETACH_DEAD) {
-		event->pending_disable = 1;
 		state = PERF_EVENT_STATE_DEAD;
 	}
+	event->pending_disable = 1;
 	event_sched_out(event, ctx);
 	perf_event_set_state(event, min(event->state, state));
 

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[David Wang]

At 2025-06-03 20:54:40, "Peter Zijlstra" <peterz@infradead.org> wrote:
>On Tue, Jun 03, 2025 at 02:50:56PM +0200, Peter Zijlstra wrote:
>> On Tue, Jun 03, 2025 at 06:44:58PM +0800, David Wang wrote:
>> 
>> 
>> > (As yeoreum.yun@arm.com pointed out,  the change in perf_remove_from_context() made
>> > perf_event_set_state() happened before list_del_event(), resulting in perf_cgroup_event_disable()
>> > not called.)
>> 
>> Aah, d'0h. Let me see what we should do there.
>
>Does this help? This way event_sched_out() will call
>perf_cgroup_event_disable().
>
>
>diff --git a/kernel/events/core.c b/kernel/events/core.c
>index f34c99f8ce8f..adbb0372825f 100644
>--- a/kernel/events/core.c
>+++ b/kernel/events/core.c
>@@ -2494,9 +2494,9 @@ __perf_remove_from_context(struct perf_event *event,
> 	if (flags & DETACH_REVOKE)
> 		state = PERF_EVENT_STATE_REVOKED;
> 	if (flags & DETACH_DEAD) {
>-		event->pending_disable = 1;
> 		state = PERF_EVENT_STATE_DEAD;
> 	}
>+	event->pending_disable = 1;
> 	event_sched_out(event, ctx);
> 	perf_event_set_state(event, min(event->state, state));
> 

Ok, I will give it a try and update later.

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[David Wang]

At 2025-06-03 21:03:55, "David Wang" <00107082@163.com> wrote:
>
>At 2025-06-03 20:54:40, "Peter Zijlstra" <peterz@infradead.org> wrote:
>>On Tue, Jun 03, 2025 at 02:50:56PM +0200, Peter Zijlstra wrote:
>>> On Tue, Jun 03, 2025 at 06:44:58PM +0800, David Wang wrote:
>>> 
>>> 
>>> > (As yeoreum.yun@arm.com pointed out,  the change in perf_remove_from_context() made
>>> > perf_event_set_state() happened before list_del_event(), resulting in perf_cgroup_event_disable()
>>> > not called.)
>>> 
>>> Aah, d'0h. Let me see what we should do there.
>>
>>Does this help? This way event_sched_out() will call
>>perf_cgroup_event_disable().
>>
>>
>>diff --git a/kernel/events/core.c b/kernel/events/core.c
>>index f34c99f8ce8f..adbb0372825f 100644
>>--- a/kernel/events/core.c
>>+++ b/kernel/events/core.c
>>@@ -2494,9 +2494,9 @@ __perf_remove_from_context(struct perf_event *event,
>> 	if (flags & DETACH_REVOKE)
>> 		state = PERF_EVENT_STATE_REVOKED;
>> 	if (flags & DETACH_DEAD) {
>>-		event->pending_disable = 1;
>> 		state = PERF_EVENT_STATE_DEAD;
>> 	}
>>+	event->pending_disable = 1;
>> 	event_sched_out(event, ctx);
>> 	perf_event_set_state(event, min(event->state, state));
>> 
>
>Ok, I will give it a try and update later.

Sadly no, caught a kernel panic at the first round....

I tried to use perf to reproduce this, but no luck so far. Following is the code I used to reproduce.

(The code is silly, but valid I think....)
To reproduce, I use following steps:
Open two terminals:
1. In terminal A
mkdir /sys/fs/cgroup/mytest
echo $$ > /sys/fs/cgroup/mytest/cgroup.procs
2. In terminal B
[g++ following code if not done yet g++ -o profiler xx.cpp]
./profiler mytest
3. Do something in terminal A, usually I would run following command under kernel source tree
for i in {1..200}; do find ./ -name nottobefound > /dev/null; done
4. wait for 5~10mintes
5. In terminal B, ctrl-C stop the profiler
6. reboot
(On my system, with 6.15 at most 4 rounds of test would catch a kernel panic.)

I could not reproduce it with my KVM, maybe I need more trials.
Not sure whether anyone else could reproduce this. 


---
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <asm/unistd.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <poll.h>
#include <signal.h>
#include <fcntl.h>
#include <elf.h>
#include <string.h>

#include <vector>
#include <string>
#include <map>
#include <unordered_map>
#include <unordered_set>
#include <algorithm>
using namespace std;


#define MAXN  512
#define MAXCPU 128
#define error(msg) do { perror(msg); exit(1); } while(0)

static long perf_event_open(struct perf_event_attr *perf_event,
		pid_t pid, int cpu, int group_fd, unsigned long flags) {
    return syscall(__NR_perf_event_open, perf_event,
		    pid, cpu, group_fd, flags);
}

struct pollfd polls[MAXCPU];
// res for cleanup
static long long psize;
map<int, pair<void*, long long>> res;
static long long eventc = 0;

void int_exit(int _) {
    for (auto x: res) {
        auto y = x.second;
        void* addr = y.first;
        munmap(addr, (1+MAXN)*psize);
        close(x.first);
    }
    res.clear();
    printf("total %lld events collect\n", eventc);
    exit(0);
}
int process_event(char *base, unsigned long long size, unsigned long long offset) {
	struct perf_event_header* p = NULL;
	offset%=size;
	p = (struct perf_event_header*) (base+offset);
	eventc++;
	return p->size;
}

int main(int argc, char *argv[]) {
	if (argc<2) { printf("Need cgroup name\n"); return 1; }
	char xb[256];
	snprintf(xb, sizeof(xb), "/sys/fs/cgroup/%s", argv[1]);
	int cgroup_id = open(xb, O_CLOEXEC);
	if (cgroup_id <= 0) error("error open cgroup dir");
	int cpu_num = sysconf(_SC_NPROCESSORS_ONLN);
	psize = sysconf(_SC_PAGE_SIZE); // getpagesize();
	struct perf_event_attr attr;
	memset(&attr, 0, sizeof(attr));
	attr.type = PERF_TYPE_SOFTWARE;
	attr.size = sizeof(attr);
	attr.config = PERF_COUNT_SW_CPU_CLOCK;
	attr.sample_freq = 9999;//777; // adjust it
	attr.freq = 1;
	attr.wakeup_events = 16;
	attr.sample_type = PERF_SAMPLE_CALLCHAIN;
	attr.sample_max_stack = 32;
	attr.exclude_callchain_user = 1;
	// start perf event
	int i, k, fd;
	void* addr;
	for (i=0, k=0; i<cpu_num&&i<MAXCPU; i++) {
		printf("attaching cpu %d\n", i);
		fd = perf_event_open(&attr, cgroup_id, i, -1, PERF_FLAG_FD_CLOEXEC|PERF_FLAG_PID_CGROUP);
		if (fd<0) error("fail to open perf event");
		addr = mmap(NULL, (1+MAXN)*psize, PROT_READ, MAP_SHARED, fd, 0);
		if (addr == MAP_FAILED) error("mmap failed");
		res[fd] = {addr, 0};
		polls[k].fd = fd;
		polls[k].events = POLLIN;
		polls[k].revents = 0;
		k++;
	}
	signal(SIGINT, int_exit);
	signal(SIGTERM, int_exit);

	unsigned long long head;
	int event_size;
	struct perf_event_mmap_page *mp;
	while (poll(polls, k, -1)>0) {
		for (i=0; i<k; i++) {
			if ((polls[i].revents&POLLIN)==0) continue;
			fd = polls[i].fd;
			addr = res[fd].first;
			mp = (struct perf_event_mmap_page *)addr;
			head = res[fd].second;
			ioctl(fd, PERF_EVENT_IOC_PAUSE_OUTPUT, 1);
			if (head>mp->data_head) head=mp->data_head;
			head = mp->data_head-((mp->data_head-head)%mp->data_size);
			while(head<mp->data_head) {
				head += process_event((char*)addr+mp->data_offset, mp->data_size, head);
			}
			res[fd].second = mp->data_head;
			ioctl(fd, PERF_EVENT_IOC_PAUSE_OUTPUT, 0);
		}
	}
	int_exit(0);
	return 0;
}

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[Yeoreum Yun]

Hi Peter,

> On Tue, Jun 03, 2025 at 02:50:56PM +0200, Peter Zijlstra wrote:
> > On Tue, Jun 03, 2025 at 06:44:58PM +0800, David Wang wrote:
> >
> >
> > > (As yeoreum.yun@arm.com pointed out,  the change in perf_remove_from_context() made
> > > perf_event_set_state() happened before list_del_event(), resulting in perf_cgroup_event_disable()
> > > not called.)
> >
> > Aah, d'0h. Let me see what we should do there.
>
> Does this help? This way event_sched_out() will call
> perf_cgroup_event_disable().
>
>
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index f34c99f8ce8f..adbb0372825f 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -2494,9 +2494,9 @@ __perf_remove_from_context(struct perf_event *event,
>  	if (flags & DETACH_REVOKE)
>  		state = PERF_EVENT_STATE_REVOKED;
>  	if (flags & DETACH_DEAD) {
> -		event->pending_disable = 1;
>  		state = PERF_EVENT_STATE_DEAD;
>  	}
> +	event->pending_disable = 1;

I think it would break if event->state is "PERF_EVENT_STATE_ERROR".

TBH, there is the patch to solve this problem:
  https://lore.kernel.org/all/20250602184049.4010919-1-yeoreum.yun@arm.com/

Does it have a problem?

Thanks.

--
Sincerely,
Yeoreum Yun

Re: [PATCH v2] perf/core: restore __perf_remove_from_context when DETACH_EXIT not set

[Peter Zijlstra]

On Tue, Jun 03, 2025 at 02:22:37PM +0100, Yeoreum Yun wrote:

> I think it would break if event->state is "PERF_EVENT_STATE_ERROR".

Indeed.

> TBH, there is the patch to solve this problem:
>   https://lore.kernel.org/all/20250602184049.4010919-1-yeoreum.yun@arm.com/
> 
> Does it have a problem?

It does mean we can remove that pending_disable thing from DETACH_DEAD I
think.

Also let me edit the Changelog, the actual splat doesn't really have
much useful information.