lkml.org 
[lkml]   [2013]   [Nov]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH] mm/zswap: don't allow entry eviction if in use by load
On Wed, Nov 20, 2013 at 8:59 PM, Weijie Yang <weijie.yang.kh@gmail.com> wrote:
> Hello Dan
>
> On Thu, Nov 21, 2013 at 3:48 AM, Dan Streetman <ddstreet@ieee.org> wrote:
>> The changes in commit 0ab0abcf511545d1fddbe72a36b3ca73388ac937
>> introduce a bug in writeback, if an entry is in use by load
>> it will be evicted anyway, which isn't correct (technically,
>> the code currently in zbud doesn't actually care much what the
>> zswap evict function returns, but that could change).
>
> Thanks for your work. Howerver it is not a bug.
>
> I have thought about this situation, and it will never happen.
> If entry is being loaded, its corresponding page must be in swapcache
> so zswap_get_swap_cache_page() will return ZSWAP_SWAPCACHE_EXIST

ah, ok.

While you do imply that with the fail: comment, I personally think it
should also be stated in the refcount check comment; a comment
indicating failure can happen due to concurrent load does not make
clear that it will *always* fail in cases of concurrent load and so
that case doesn't need to be checked for in the success path.
Additionally, the lack of a check here is assuming that zswap won't be
updated to ever inc the refcount anywhere besides the load function,
which might cause unexpected breakage later; i.e., this is coding to
the current implementation, not to the entry->refcount api.

Can I also ask why you do a rb_search instead of just checking the
entry->refcount? Doing the search is going to take longer than just
checking the refcount; is there some case where the entry will not be
in the rb but will have a nonzero refcount?


\
 
 \ /
  Last update: 2013-11-21 23:01    [W:0.326 / U:1.940 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site