Messages in this thread | | | Date | Sun, 29 Mar 2015 11:02:18 +0300 | From | Boaz Harrosh <> | Subject | Re: Should implementations of ->direct_access be allowed to sleep? |
| |
On 03/26/2015 09:32 PM, Dave Chinner wrote: <> >> I'm leaning towards the latter. But I'm not sure what GFP flags to >> recommend that brd use ... GFP_NOWAIT | __GFP_ZERO, perhaps? > > What, so we get random IO failures under memory pressure? > > I really think we should allow .direct_access to sleep. It means we > can use existing drivers and it also allows future implementations > that might require, say, RDMA to be performed to update a page > before access is granted. i.e. .direct_access is the first hook into > the persistent device at page fault time.... >
I agree with Dave. Last I tried (couple years ago) doing any allocation GFP_NOWAIT on FS IO paths fails really badly in all kind of surprising ways. The Kernel is built in to that allocation pressure.
I think that ->direct_access should not be any different then any other block-device access, ie allow to sleep.
With brd a user can make sure not to sleep if he pre-allocates ie call ->direct_access at least once on a given offset-length. But I would not like to even do that guaranty. ->direct_access should be allowed to sleep. Well written code has many ways to allow sleep yet be very low latency. (So I do not see what we are missing)
> Cheers, > Dave.
Thanks Boaz
| |