Hello Minchan,
On (03/18/16 10:25), Minchan Kim wrote:
[..]
> > aha, ok.
> >
> > > (ie, simple code, removing
> > > max_comp_streams knob, no need to this your stat, guarantee parallel
> > > level, guarantee consumed memory space).
> >
> > I'll take a look and prepare some numbers (most likely next week).
>
> Sounds great to me!
so I have schematically this thing now. streams are per-cpu and contain
scratch buffer and work mem.
zram_bvec_write()
{
*get_cpu_ptr(comp->stream);
zcomp_compress();
zs_malloc()
put_cpu_ptr(comp->stream);
}
this, however, makes zsmalloc unhapy. pool has GFP_NOIO | __GFP_HIGHMEM
gfp, and GFP_NOIO is ___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM. this
__GFP_DIRECT_RECLAIM is in the conflict with per-cpu streams, because
per-cpu streams require disabled preemption (up until we copy stream
buffer to zspage). so what options do we have here... from the top of
my head (w/o a lot of thinking)...
-- remove __GFP_DIRECT_RECLAIM from pool gfp mask, which a bit is risky...
IOW, make pool gfp '___GFP_KSWAPD_RECLAIM | __GFP_HIGHMEM'
-- kmalloc/kfree temp buffer for every RW op, which is ugly... because
it sort of voids the whole purpose of per-cpu streams.
-- ... hm.
-ss
