Re: dm bufio: Reduce dm_bufio_lock contention

[Michal Hocko <mhocko@kernel.org>]

On Fri 22-06-18 11:01:51, Michal Hocko wrote:
> On Thu 21-06-18 21:17:24, Mikulas Patocka wrote:
[...]
> > What about this patch? If __GFP_NORETRY and __GFP_FS is not set (i.e. the 
> > request comes from a block device driver or a filesystem), we should not 
> > sleep.
> 
> Why? How are you going to audit all the callers that the behavior makes
> sense and moreover how are you going to ensure that future usage will
> still make sense. The more subtle side effects gfp flags have the harder
> they are to maintain.

So just as an excercise. Try to explain the above semantic to users. We
currently have the following.

 * __GFP_NORETRY: The VM implementation will try only very lightweight
 *   memory direct reclaim to get some memory under memory pressure (thus
 *   it can sleep). It will avoid disruptive actions like OOM killer. The
 *   caller must handle the failure which is quite likely to happen under
 *   heavy memory pressure. The flag is suitable when failure can easily be
 *   handled at small cost, such as reduced throughput

 * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the
 *   allocator recursing into the filesystem which might already be holding
 *   locks.

So how are you going to explain gfp & (__GFP_NORETRY | ~__GFP_FS)? What
is the actual semantic without explaining the whole reclaim or force
users to look into the code to understand that? What about GFP_NOIO |
__GFP_NORETRY? What does it mean to that "should not sleep". Do all
shrinkers have to follow that as well?
-- 
Michal Hocko
SUSE Labs