[PATCH bpf-next] selftests/bpf: Fix timer_start_deadlock failure due to hrtimer change

[PATCH bpf-next] selftests/bpf: Fix timer_start_deadlock failure due to hrtimer change

[Shung-Hsi Yu]

Since commit f2e388a019e4 ("hrtimer: Reduce trace noise in hrtimer_start()"),
hrtimer_cancel tracepoint is no longer called when a hrtimer is re-armed. So
instead of a hrtimer_cancel followed by hrtimer_start tracepoint events, there
is now only a since hrtimer_start tracepoint event with the new was_armed field
set to 1, to indicated that the hrtimer was previously armed.

Update timer_start_deadlock accordingly so it traces hrtimer_start tracepoint
instead, with was_armed used as guard.

Signed-off-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
---
Currently only fails on Linus' tree since hrtimer changes haven't made
its way into bpf-next yet.
---
 tools/testing/selftests/bpf/progs/timer_start_deadlock.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/tools/testing/selftests/bpf/progs/timer_start_deadlock.c b/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
index 019518ee18cd..afabd15bdac4 100644
--- a/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
+++ b/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
@@ -27,13 +27,13 @@ static int timer_cb(void *map, int *key, struct elem *value)
 	return 0;
 }
 
-SEC("tp_btf/hrtimer_cancel")
-int BPF_PROG(tp_hrtimer_cancel, struct hrtimer *hrtimer)
+SEC("tp_btf/hrtimer_start")
+int BPF_PROG(tp_hrtimer_start, struct hrtimer *hrtimer, enum hrtimer_mode mode, bool was_armed)
 {
 	struct bpf_timer *timer;
 	int key = 0;
 
-	if (!in_timer_start)
+	if (!in_timer_start || !was_armed)
 		return 0;
 
 	tp_called = 1;
@@ -60,7 +60,7 @@ int start_timer(void *ctx)
 
 	/*
 	 * call hrtimer_start() twice, so that 2nd call does
-	 * remove_hrtimer() and trace_hrtimer_cancel() tracepoint.
+	 * trace_hrtimer_start(was_armed=1) tracepoint.
 	 */
 	in_timer_start = 1;
 	bpf_timer_start(timer, 1000000000, 0);
-- 
2.53.0

Re: [PATCH bpf-next] selftests/bpf: Fix timer_start_deadlock failure due to hrtimer change

[Mykyta Yatsenko]

On 4/15/26 1:03 PM, Shung-Hsi Yu wrote:
> Since commit f2e388a019e4 ("hrtimer: Reduce trace noise in hrtimer_start()"),
> hrtimer_cancel tracepoint is no longer called when a hrtimer is re-armed. So
> instead of a hrtimer_cancel followed by hrtimer_start tracepoint events, there
> is now only a since hrtimer_start tracepoint event with the new was_armed field
> set to 1, to indicated that the hrtimer was previously armed.
> 
> Update timer_start_deadlock accordingly so it traces hrtimer_start tracepoint
> instead, with was_armed used as guard.
> 
> Signed-off-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
> ---
> Currently only fails on Linus' tree since hrtimer changes haven't made
> its way into bpf-next yet.
> ---

This patch fixes the failing test.

Tested-by: Mykyta Yatsenko <yatsenko@meta.com>

The logic is also looks correct, tracepoint used to run on timer_cancel, 
which was called due to the second timer_start, now it's running on 
timer_start with was_armed=true, so the logic is kept the same.

Acked-by: Mykyta Yatsenko <yatsenko@meta.com>

>   tools/testing/selftests/bpf/progs/timer_start_deadlock.c | 8 ++++----
>   1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/testing/selftests/bpf/progs/timer_start_deadlock.c b/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
> index 019518ee18cd..afabd15bdac4 100644
> --- a/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
> +++ b/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
> @@ -27,13 +27,13 @@ static int timer_cb(void *map, int *key, struct elem *value)
>   	return 0;
>   }
>   
> -SEC("tp_btf/hrtimer_cancel")
> -int BPF_PROG(tp_hrtimer_cancel, struct hrtimer *hrtimer)
> +SEC("tp_btf/hrtimer_start")
> +int BPF_PROG(tp_hrtimer_start, struct hrtimer *hrtimer, enum hrtimer_mode mode, bool was_armed)
>   {
>   	struct bpf_timer *timer;
>   	int key = 0;
>   
> -	if (!in_timer_start)
> +	if (!in_timer_start || !was_armed)
>   		return 0;
>   
>   	tp_called = 1;
> @@ -60,7 +60,7 @@ int start_timer(void *ctx)
>   
>   	/*
>   	 * call hrtimer_start() twice, so that 2nd call does
> -	 * remove_hrtimer() and trace_hrtimer_cancel() tracepoint.
> +	 * trace_hrtimer_start(was_armed=1) tracepoint.
>   	 */
>   	in_timer_start = 1;
>   	bpf_timer_start(timer, 1000000000, 0);

Re: [PATCH bpf-next] selftests/bpf: Fix timer_start_deadlock failure due to hrtimer change

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to bpf/bpf.git (master)
by Alexei Starovoitov <ast@kernel.org>:

On Wed, 15 Apr 2026 20:03:28 +0800 you wrote:
> Since commit f2e388a019e4 ("hrtimer: Reduce trace noise in hrtimer_start()"),
> hrtimer_cancel tracepoint is no longer called when a hrtimer is re-armed. So
> instead of a hrtimer_cancel followed by hrtimer_start tracepoint events, there
> is now only a since hrtimer_start tracepoint event with the new was_armed field
> set to 1, to indicated that the hrtimer was previously armed.
> 
> Update timer_start_deadlock accordingly so it traces hrtimer_start tracepoint
> instead, with was_armed used as guard.
> 
> [...]

Here is the summary with links:
  - [bpf-next] selftests/bpf: Fix timer_start_deadlock failure due to hrtimer change
    https://git.kernel.org/bpf/bpf/c/813f336269e6

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html