On 04/04/2018 08:52 PM, Matthew Wilcox wrote:
> On Wed, Apr 04, 2018 at 09:35:46AM -0700, Mike Kravetz wrote:
>> Running with this XArray series on top of next-20180329 consistently 'hangs'
>> on shutdown looping (?forever?) in tag_pages_for_writeback/xas_for_each_tag.
>> All I have to do is make sure there is some activity on the ext4 fs before
>> shutdown. Not sure if this is a 'next-20180329' issue or XArray issue.
>> But the fact that we are looping in xas_for_each_tag looks suspicious.
>
> Thanks for your help debugging this! Particularly collecting the xa_dump.
> I got bit by the undefined behaviour of shifting by BITS_PER_LONG,
> but of course it was subtle.
>
> The userspace testing framework wasn't catching this for a couple of
> reasons; I'll work on making sure it catches this kind of thing in
> the future.
>
> I'll fold this in and post a v11 later this week or early next week.
Nice! Added patch and it solves the hangs I observed.
--
Mike Kravetz
>
> diff --git a/include/linux/xarray.h b/include/linux/xarray.h
> index eac04922eba2..f5b7e507a86f 100644
> --- a/include/linux/xarray.h
> +++ b/include/linux/xarray.h
> @@ -904,9 +929,12 @@ static inline unsigned int xas_find_chunk(struct xa_state *xas, bool advance,
> if (advance)
> offset++;
> if (XA_CHUNK_SIZE == BITS_PER_LONG) {
> - unsigned long data = *addr & (~0UL << offset);
> - if (data)
> - return __ffs(data);
> + if (offset < XA_CHUNK_SIZE) {
> + unsigned long data = *addr & (~0UL << offset);
> +
> + if (data)
> + return __ffs(data);
> + }
> return XA_CHUNK_SIZE;
> }
>
>
