Messages in this thread | | | Date | Mon, 23 Nov 2015 16:47:40 -0800 | From | Andrew Morton <> | Subject | Re: [PATCH] zram/zcomp: use GFP_NOIO to allocate streams |
| |
On Tue, 24 Nov 2015 09:30:27 +0900 Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com> wrote:
> On (11/23/15 15:18), Andrew Morton wrote: > [..] > > > --- a/drivers/block/zram/zcomp_lz4.c > > > +++ b/drivers/block/zram/zcomp_lz4.c > > > @@ -20,10 +20,13 @@ static void *zcomp_lz4_create(void) > > > void *ret; > > > > > > ret = kzalloc(LZ4_MEM_COMPRESS, > > > - __GFP_NORETRY|__GFP_NOWARN|__GFP_NOMEMALLOC); > > > - if (!ret) > > > - ret = vzalloc(LZ4_MEM_COMPRESS); > > > - return ret; > > > + __GFP_NORETRY | __GFP_NOWARN | __GFP_NOMEMALLOC); > > > > But here we've still lost __GFP_RECLAIM, unnecessarily. And it's quite > > unclear why __GFP_NORETRY and __GFP_NOMEMALLOC are being used. > > __GFP_NORETRY > > we are guaranteed to have at least one compression stream, so sooner or > later every IO operation will be served. any IO that has failed in > zcomp_lz4_create() or zcomp_lzo_create() will simply wait for already > available compression stream to become idle. so this allocation is not > so dramatically important - we just increase the level of parallelism > (N idle streams let N IO operations to execute concurrently). apart from > that we are in a low memory condition (or whatever was the reason the > kernel failed to allocate LZ4_MEM_COMPRESS or LZO1X_MEM_COMPRESS) and > we can avoid pressuring the kernel furher. > > for the same reason __GFP_NOMEMALLOC is used -- we don't want to waste > an emergency memory for compression streams. >
Doesn't make a lot of sense to me. We use a weakened gfp for the kmalloc and if that fails, fall into vmalloc() using the stronger gfp anyway.
Perhaps it makes sense for higher-order allocations: we don't want to thrash around trying to create an order-2 page - we'd prefer to give up and fall into vmalloc to do a bunch of order-0 allocations.
But this argument holds for 1000 other kmalloc->vmalloc allocation attempts - what's special about this one?
And whatever is the reason for this peculiar setup,
a) where's the proof that the change is actually beneficial?
b) let's get a good code comment in place so that future readers are not similarly puzzled.
| |