Re: [PATCH v1] LoongArch: Fix __smp_mb__{before,after}_atomic()

["Kumar Kartikeya Dwivedi" <memxor@gmail.com>]

On Tue Aug 18, 2026 at 3:50 PM CEST, Tiezhu Yang wrote:
> On 2026/8/18 下午9:35, Kumar Kartikeya Dwivedi wrote:
>> On Tue Aug 18, 2026 at 3:28 PM CEST, Tiezhu Yang wrote:
>>> On 2026/8/17 上午11:59, Tiezhu Yang wrote:
>>>> When testing the BPF selftest "sudo ./test_progs -t timer_lockup", there
>>>> is a kernel lockup and panic:
>>>
>>> ...
>>>
>>>> With this patch, the lockless "store-before-load" ordering is enforced by
>>>> the DBAR instruction. The BPF timer_lockup selftest was stressed for 5000
>>>> consecutive loops on a physical LoongArch machine without encountering any
>>>> further lockups or warnings:
>>>>
>>>>     for i in {1..5000}; do sudo ./test_progs -t timer_lockup; done
>>>
>>> ...
>>>
>>>> diff --git a/arch/loongarch/include/asm/barrier.h b/arch/loongarch/include/asm/barrier.h
>>>> index 4b663f197706..adfe343dfa65 100644
>>>> --- a/arch/loongarch/include/asm/barrier.h
>>>> +++ b/arch/loongarch/include/asm/barrier.h
>>>> @@ -57,8 +57,8 @@
>>>>    #define __WEAK_LLSC_MB		"			\n"
>>>>    #endif
>>>>
>>>> -#define __smp_mb__before_atomic()	barrier()
>>>> -#define __smp_mb__after_atomic()	barrier()
>>>> +#define __smp_mb__before_atomic()	__smp_mb()
>>>> +#define __smp_mb__after_atomic()	__smp_mb()
>>>>
>>>>    /**
>>>>     * array_index_mask_nospec() - generate a ~0 mask when index < size, 0 otherwise
>>>
>>> Hi all,
>>>
>>> To address the soft lockup while avoiding the performance overhead of
>>> executing two separate instructions, I think there is a more elegant
>>> way to fix this directly in the BPF core helper:
>>>
>>> We can replace atomic_inc() and smp_mb__after_atomic() with a single
>>> atomic_fetch_add() in bpf_timer_cancel(). This consolidates the logic
>>> into a single native atomic operation with full ordering, which not
>>> only guarantees the store-load order to eliminate the deadlock but
>>> also improves the performance for weak memory model architectures.
>>>
>>> Any thoughts on this approach?
>>>
>>> ```
>>> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
>>> index c18f1e16edee..ee2b3a4dcc05 100644
>>> --- a/kernel/bpf/helpers.c
>>> +++ b/kernel/bpf/helpers.c
>>> @@ -1591,9 +1591,7 @@ BPF_CALL_1(bpf_timer_cancel, struct bpf_async_kern
>>> *, async)
>>>            */
>>>           if (!cur_t)
>>>                   goto drop;
>>> -       atomic_inc(&t->cancelling);
>>> -       /* Need full barrier after relaxed atomic_inc */
>>> -       smp_mb__after_atomic();
>>> +       atomic_fetch_add(1, &t->cancelling);
>>>           inc = true;
>>>           if (atomic_read(&cur_t->cancelling)) {
>>>                   /* We're cancelling timer t, while some other timer
>>> callback is
>>> ```
>>>
>>> I tested the above diff, the BPF timer_lockup selftest was stressed
>>> for 5000 consecutive loops on a physical LoongArch machine without
>>> encountering any further lockups or warnings.
>>
>> There is a full barrier in both cases. I don't know what performance improvement
>> will be achieved by changing this.
>
> We performed benchmarking using UnixBench on a physical LoongArch
> machine, the UnixBench score is different (amadd.w + dbar < amadd_db.w).
>

I mean, this function is already quite heavy. We have multiple fully ordered
atomics (refcount bumps/drops, xchg(), etc.) spread across the operation. Do you
observe any measurable speedup in the throughput of this function?  The second
question is whether timer cancellation is really frequent. In practice, I don't
think that is the case.  The actual hrtimer_cancel() in itself is heavy and
waits synchronously for the callback to finish.

I'm not opposed to it or anything, I just don't think it's worth it in this
case. If the latency of cancel is a problem there are other bigger opportunities
to pursue than this.

>> Don't you need to fix the lowering for
>> smp_mb__after_atomic() anyway? It's used in several other places in the kernel.
>
> The LoongArch architecture-level fix for smp_mb__after_atomic() is
> a fundamental bug fix that will be pushed separately to the LoongArch
> tree. This BPF-layer change is intended as a generic, cross-architecture
> performance optimization.
>
> Thanks,
> Tiezhu