Re: [PATCH] dax: Fix missed PMD wakeups

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, Jul 3, 2019 at 12:53 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> On Wed, Jul 03, 2019 at 10:01:37AM -0700, Dan Williams wrote:
> > On Wed, Jul 3, 2019 at 5:17 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> > >
> > > On Wed, Jul 03, 2019 at 12:24:54AM -0700, Dan Williams wrote:
> > > > This fix may increase waitqueue contention, but a fix for that is saved
> > > > for a larger rework. In the meantime this fix is suitable for -stable
> > > > backports.
> > >
> > > I think this is too big for what it is; just the two-line patch to stop
> > > incorporating the low bits of the PTE would be more appropriate.
> >
> > Sufficient, yes, "appropriate", not so sure. All those comments about
> > pmd entry size are stale after this change.
>
> But then they'll have to be put back in again.  This seems to be working
> for me, although I doubt I'm actually hitting the edge case that rocksdb
> hits:

Seems to be holding up under testing here, a couple comments...

>
> diff --git a/fs/dax.c b/fs/dax.c
> index 2e48c7ebb973..e77bd6aef10c 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -198,6 +198,10 @@ static void dax_wake_entry(struct xa_state *xas, void *entry, bool wake_all)
>   * if it did.
>   *
>   * Must be called with the i_pages lock held.
> + *
> + * If the xa_state refers to a larger entry, then it may return a locked
> + * smaller entry (eg a PTE entry) without waiting for the smaller entry
> + * to be unlocked.
>   */
>  static void *get_unlocked_entry(struct xa_state *xas)
>  {
> @@ -211,7 +215,8 @@ static void *get_unlocked_entry(struct xa_state *xas)
>         for (;;) {
>                 entry = xas_find_conflict(xas);
>                 if (!entry || WARN_ON_ONCE(!xa_is_value(entry)) ||
> -                               !dax_is_locked(entry))
> +                               !dax_is_locked(entry) ||
> +                               dax_entry_order(entry) < xas_get_order(xas))

Doesn't this potentially allow a locked entry to be returned for a
caller that expects all value entries are unlocked?

>                         return entry;
>
>                 wq = dax_entry_waitqueue(xas, entry, &ewait.key);
> @@ -253,8 +258,12 @@ static void wait_entry_unlocked(struct xa_state *xas, void *entry)
>
>  static void put_unlocked_entry(struct xa_state *xas, void *entry)
>  {
> -       /* If we were the only waiter woken, wake the next one */
> -       if (entry)
> +       /*
> +        * If we were the only waiter woken, wake the next one.
> +        * Do not wake anybody if the entry is locked; that indicates
> +        * we weren't woken.
> +        */
> +       if (entry && !dax_is_locked(entry))
>                 dax_wake_entry(xas, entry, false);
>  }
>
> diff --git a/include/linux/xarray.h b/include/linux/xarray.h
> index 052e06ff4c36..b17289d92af4 100644
> --- a/include/linux/xarray.h
> +++ b/include/linux/xarray.h
> @@ -1529,6 +1529,27 @@ static inline void xas_set_order(struct xa_state *xas, unsigned long index,
>  #endif
>  }
>
> +/**
> + * xas_get_order() - Get the order of the entry being operated on.
> + * @xas: XArray operation state.
> + *
> + * Return: The order of the entry.
> + */
> +static inline unsigned int xas_get_order(const struct xa_state *xas)
> +{
> +       unsigned int order = xas->xa_shift;
> +
> +#ifdef CONFIG_XARRAY_MULTI
> +       unsigned int sibs = xas->xa_sibs;
> +
> +       while (sibs) {
> +               order++;
> +               sibs /= 2;
> +       }

Use ilog2() here?



[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [Samba]     [Device Mapper]     [CEPH Development]

  Powered by Linux