On Fri, Feb 16, 2024 at 11:10:04PM +1300, Barry Song wrote:
> On Fri, Feb 16, 2024 at 9:30 PM Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
> >
> > On Fri, Feb 16, 2024 at 05:08:14PM +1300, Barry Song wrote:
> > > From: Barry Song <v-songbaohua@xxxxxxxx>
> > >
> > > Most compressors are actually CPU-based and won't sleep during
> > > compression and decompression. We should remove the redundant
> > > memcpy for them.
> > >
> > > Signed-off-by: Barry Song <v-songbaohua@xxxxxxxx>
> > > Tested-by: Chengming Zhou <zhouchengming@xxxxxxxxxxxxx>
> > > Reviewed-by: Nhat Pham <nphamcs@xxxxxxxxx>
> > > ---
> > > mm/zswap.c | 6 ++++--
> > > 1 file changed, 4 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/mm/zswap.c b/mm/zswap.c
> > > index 350dd2fc8159..6319d2281020 100644
> > > --- a/mm/zswap.c
> > > +++ b/mm/zswap.c
> > > @@ -168,6 +168,7 @@ struct crypto_acomp_ctx {
> > > struct crypto_wait wait;
> > > u8 *buffer;
> > > struct mutex mutex;
> > > + bool is_sleepable;
> > > };
> > >
> > > /*
> > > @@ -716,6 +717,7 @@ static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
> > > goto acomp_fail;
> > > }
> > > acomp_ctx->acomp = acomp;
> > > + acomp_ctx->is_sleepable = acomp_is_sleepable(acomp);
> >
> > Just one question here. In patch 1, sleepable seems to mean "not async".
> > IIUC, even a synchronous algorithm may sleep (e.g. if there is a
> > cond_resched or waiting for a mutex). Does sleepable in acomp terms the
> > same as "atomic" in scheduling/preemption terms?
>
> I think the answer is yes though async and sleepable are slightly
> different semantically
> generally speaking. but for comp cases, they are equal.
>
> We have two backends for compression/ decompression - scomp and acomp. if comp
> is using scomp backend, we can safely think they are not sleepable at
> least from the
> below three facts.
>
> 1. in zRAM, we are using scomp APIs only - crypto_comp_decompress()/
> crypto_comp_compress(), which are definitely scomp, we have never considered
> sleeping problem in zram drivers:
> static int zram_read_from_zspool(struct zram *zram, struct page *page,
> u32 index)
> {
> struct zcomp_strm *zstrm;
> unsigned long handle;
> unsigned int size;
> void *src, *dst;
> u32 prio;
> int ret;
>
> handle = zram_get_handle(zram, index);
> ...
> src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
> if (size == PAGE_SIZE) {
> dst = kmap_local_page(page);
> memcpy(dst, src, PAGE_SIZE);
> kunmap_local(dst);
> ret = 0;
> } else {
> dst = kmap_local_page(page);
> ret = zcomp_decompress(zstrm, src, size, dst);
> kunmap_local(dst);
> zcomp_stream_put(zram->comps[prio]);
> }
> zs_unmap_object(zram->mem_pool, handle);
> return ret;
> }
>
> 2. zswap used to only support scomp before we moved to use
> crypto_acomp_compress()
> and crypto_acomp_decompress() APIs whose backends can be either scomp
> or acomp, thus new hardware-based compression drivers can be used in zswap.
>
> But before we moved to these new APIs in commit 1ec3b5fe6eec782 ("mm/zswap:
> move to use crypto_acomp API for hardware acceleration") , zswap had
> never considered
> sleeping problems just like zRAM.
>
> 3. There is no sleeping in drivers using scomp backend.
>
> $ git grep crypto_register_scomp
> crypto/842.c: ret = crypto_register_scomp(&scomp);
> crypto/deflate.c: ret = crypto_register_scomp(&scomp);
> crypto/lz4.c: ret = crypto_register_scomp(&scomp);
> crypto/lz4hc.c: ret = crypto_register_scomp(&scomp);
> crypto/lzo-rle.c: ret = crypto_register_scomp(&scomp);
> crypto/lzo.c: ret = crypto_register_scomp(&scomp);
> crypto/zstd.c: ret = crypto_register_scomp(&scomp);
> drivers/crypto/cavium/zip/zip_main.c: ret =
> crypto_register_scomp(&zip_scomp_deflate);
> drivers/crypto/cavium/zip/zip_main.c: ret =
> crypto_register_scomp(&zip_scomp_lzs);
>
> which are the most common cases.
Thanks for explaining. Ideally we should be able to catch any violations
with proper debug options as you mentioned. Please include more info the
commit message about sleepability, a summarized version of what you
described above.
