Re: [PATCH] mm: fix incorrect vbq reference in purge_fragmented_block

[Chuanhua Han]

Zhaoyang Huang <huangzhaoyang@xxxxxxxxx> 于2024年5月30日周四 17:25写道：
>
> On Thu, May 30, 2024 at 5:16 PM Chuanhua Han <chuanhuahan@xxxxxxxxx> wrote:
> >
> > zhaoyang.huang <zhaoyang.huang@xxxxxxxxxx> 于2024年5月30日周四 10:52写道：
> > >
> > > From: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
> > >
> > > Broken vbq->free reported on a v6.6 based system which is caused
> > > by invalid vbq->lock protect over vbq->free in purge_fragmented_block.
> > > This should be introduced by the Fixes below which ignored vbq->lock
> > > matter.
> > >
> > > Fixes: fc1e0d980037 ("mm/vmalloc: prevent stale TLBs in fully utilized blocks")
> > >
> > > Signed-off-by: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
> > > ---
> > >  mm/vmalloc.c | 11 +++++++----
> > >  1 file changed, 7 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > > index 22aa63f4ef63..112b50431725 100644
> > > --- a/mm/vmalloc.c
> > > +++ b/mm/vmalloc.c
> > > @@ -2614,9 +2614,10 @@ static void free_vmap_block(struct vmap_block *vb)
> > >  }
> > >
> > >  static bool purge_fragmented_block(struct vmap_block *vb,
> > > -               struct vmap_block_queue *vbq, struct list_head *purge_list,
> > > -               bool force_purge)
> > > +               struct list_head *purge_list, bool force_purge)
> > >  {
> > > +       struct vmap_block_queue *vbq;
> > > +
> > >         if (vb->free + vb->dirty != VMAP_BBMAP_BITS ||
> > >             vb->dirty == VMAP_BBMAP_BITS)
> > >                 return false;
> > > @@ -2625,6 +2626,8 @@ static bool purge_fragmented_block(struct vmap_block *vb,
> > >         if (!(force_purge || vb->free < VMAP_PURGE_THRESHOLD))
> > >                 return false;
> > >
> > > +       vbq = container_of(addr_to_vb_xa(vb->va->va_start),
> > > +               struct vmap_block_queue, vmap_blocks);
> > This seems to be the same as before fix :),  the vbq found by
> > addr_to_vb_xa is still added to the xarray vbq, not necessarily to the
> > free_list vbq,
> Yes, my fault. Should we expand the vmap_block_queue by introducing a
> cpu_id which I actually do in my local regression.
You may need to embed a cpu_id in vb, and then use cpu_id to get the
vbq where the free_list is located
>
> > These two vbqs may not be the same, we need to find the vbq when added
> > to free_list.
> >
> > For example:
> > We add vb to vbq1's xarray and vbq2's free_list, and we need to find
> > vbq2 instead of vbq1.
> > So I feel like this place isn't really fixed？
> > >         /* prevent further allocs after releasing lock */
> > >         WRITE_ONCE(vb->free, 0);
> > >         /* prevent purging it again */
> > > @@ -2664,7 +2667,7 @@ static void purge_fragmented_blocks(int cpu)
> > >                         continue;
> > >
> > >                 spin_lock(&vb->lock);
> > > -               purge_fragmented_block(vb, vbq, &purge, true);
> > > +               purge_fragmented_block(vb, &purge, true);
> > >                 spin_unlock(&vb->lock);
> > >         }
> > >         rcu_read_unlock();
> > > @@ -2801,7 +2804,7 @@ static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush)
> > >                          * not purgeable, check whether there is dirty
> > >                          * space to be flushed.
> > >                          */
> > > -                       if (!purge_fragmented_block(vb, vbq, &purge_list, false) &&
> > > +                       if (!purge_fragmented_block(vb, &purge_list, false) &&
> > >                             vb->dirty_max && vb->dirty != VMAP_BBMAP_BITS) {
> > >                                 unsigned long va_start = vb->va->va_start;
> > >                                 unsigned long s, e;
> > > --
> > > 2.25.1
> > >
> > >
> >
> >
> > --
> > Thanks,
> > Chuanhua



-- 
Thanks,
Chuanhua