Re: [PATCH RFC v2] mm: use per_vma lock for MADV_DONTNEED

[SeongJae Park]

On Mon,  2 Jun 2025 10:35:19 -0700 SeongJae Park <sj@xxxxxxxxxx> wrote:

> On Sat, 31 May 2025 04:17:51 +0800 Barry Song <21cnbao@xxxxxxxxx> wrote:
> [...]
> > BTW, I found vector_madvise doesn't check the ret value of madvise_lock(),
> > it seems also a bug?
> > 
> > static ssize_t vector_madvise(struct mm_struct *mm, struct iov_iter *iter,
> > {
> >                         /* Drop and reacquire lock to unwind race. */
> >                         madvise_finish_tlb(&madv_behavior);
> >                        madvise_unlock(mm, behavior);
> >                        madvise_lock(mm, behavior);  /* missing the ret check */
> >                         madvise_init_tlb(&madv_behavior, mm)
> > }
> 
> Thank you for finding this, I will post a fix soon.

The fix is now available at
https://lore.kernel.org/20250602174926.1074-1-sj@xxxxxxxxxx


Thanks,
SJ

[...]