Re: [PATCH] bpf: test_run: Fix timer mode initialization to NO_MIGRATE mode

[Yonghong Song <yonghong.song@linux.dev>]

On 10/5/25 10:43 PM, Sahil Chandna wrote:
> By default, the timer mode is being initialized to `NO_PREEMPT`.
> This disables preemption and forces execution in atomic context.
> This can cause issue with PREEMPT_RT when calling spin_lock_bh() due
> to sleeping nature of the lock.
> ...
> BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
> in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 6107, name: syz.0.17
> preempt_count: 1, expected: 0
> RCU nest depth: 1, expected: 1
> Preemption disabled at:
> [<ffffffff891fce58>] bpf_test_timer_enter+0xf8/0x140 net/bpf/test_run.c:42
> Call Trace:
>   <TASK>
>   dump_stack_lvl+0x189/0x250 lib/dump_stack.c:120
>   __might_resched+0x44b/0x5d0 kernel/sched/core.c:8957
>   __rt_spin_lock kernel/locking/spinlock_rt.c:48 [inline]
>   rt_spin_lock+0xc7/0x2c0 kernel/locking/spinlock_rt.c:57
>   spin_lock_bh include/linux/spinlock_rt.h:88 [inline]
>   __sock_map_delete net/core/sock_map.c:421 [inline]
>   sock_map_delete_elem+0xb7/0x170 net/core/sock_map.c:452
>   bpf_prog_2c29ac5cdc6b1842+0x43/0x4b
>   bpf_dispatcher_nop_func include/linux/bpf.h:1332 [inline]
> ...
> Change initialization to NO_MIGRATE mode to prevent this.
>
> Reported-by: syzbot+1f1fbecb9413cdbfbef8@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=1f1fbecb9413cdbfbef8
> Tested-by: syzbot+1f1fbecb9413cdbfbef8@syzkaller.appspotmail.com
> Signed-off-by: Sahil Chandna <chandna.linuxkernel@gmail.com>
> ---
>   net/bpf/test_run.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
> index 4a862d605386..daf966dfed69 100644
> --- a/net/bpf/test_run.c
> +++ b/net/bpf/test_run.c
> @@ -1368,7 +1368,7 @@ int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
>   				     const union bpf_attr *kattr,
>   				     union bpf_attr __user *uattr)
>   {
> -	struct bpf_test_timer t = { NO_PREEMPT };
> +	struct bpf_test_timer t = { NO_MIGRATE };

I checked the original reproducer. And changing from
NO_PREEMPT to NO_MIGRATE is needed only with
CONFIG_PREEMPT_RT enabled.

>   	u32 size = kattr->test.data_size_in;
>   	struct bpf_flow_dissector ctx = {};
>   	u32 repeat = kattr->test.repeat;
> @@ -1436,7 +1436,7 @@ int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
>   int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
>   				union bpf_attr __user *uattr)
>   {
> -	struct bpf_test_timer t = { NO_PREEMPT };
> +	struct bpf_test_timer t = { NO_MIGRATE };

This change is not needed for the above particular BUG.

>   	struct bpf_prog_array *progs = NULL;
>   	struct bpf_sk_lookup_kern ctx = {};
>   	u32 repeat = kattr->test.repeat;

Checking the git history. I found the earliest NO_PREEMPT usage can be traced
back to this commit:

commit a439184d515fbf4805f57d11fa5dfd4524d2c0eb
Author: Stanislav Fomichev <sdf@google.com>
Date:   Tue Feb 19 10:54:17 2019 -0800

     bpf/test_run: fix unkillable BPF_PROG_TEST_RUN for flow dissector

At that time, migrate_disable/enable() are not used yet.

So I suspect that we can remove NO_PREEMPT/NO_MIGRATE in test_run.c
and use migrate_disable()/migrate_enable() universally.