Message ID | 20231213-zswap-dstmem-v1-3-896763369d04@bytedance.com (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | mm/zswap: dstmem reuse optimizations and cleanups | expand |
On Tue, Dec 12, 2023 at 8:18 PM Chengming Zhou <zhouchengming@bytedance.com> wrote: > > The zswap_load() and zswap_writeback_entry() have the same part that > decompress the data from zswap_entry to page, so refactor out the > common part as __zswap_load(entry, page). > > Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> > Reviewed-by: Nhat Pham <nphamcs@gmail.com> Great cleanup, Reviewed-by: Yosry Ahmed <yosryahmed@google.com>
On Tue, Dec 12, 2023 at 8:18 PM Chengming Zhou <zhouchengming@bytedance.com> wrote: > > The zswap_load() and zswap_writeback_entry() have the same part that > decompress the data from zswap_entry to page, so refactor out the > common part as __zswap_load(entry, page). > > Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> > Reviewed-by: Nhat Pham <nphamcs@gmail.com> On a second look, there a few nits here. First I think it makes more sense to move this refactoring ahead of reusing destmem. Right now, we add the destmem reuse to zswap_load() only, then we do the refactor and zswap_writeback_entry() gets it automatically, so there is a slight change coming to zswap_writeback_entry() hidden in the refactoring patch. Let's refactor out __zswap_load() first, then reuse destmem in it. > --- > mm/zswap.c | 107 ++++++++++++++++++++++--------------------------------------- > 1 file changed, 38 insertions(+), 69 deletions(-) > > diff --git a/mm/zswap.c b/mm/zswap.c > index fa186945010d..2f095c919a5c 100644 > --- a/mm/zswap.c > +++ b/mm/zswap.c > @@ -1392,6 +1392,41 @@ static int zswap_enabled_param_set(const char *val, > return ret; > } > > +static void __zswap_load(struct zswap_entry *entry, struct page *page) > +{ > + struct scatterlist input, output; > + unsigned int dlen = PAGE_SIZE; > + struct crypto_acomp_ctx *acomp_ctx; > + struct zpool *zpool; > + u8 *src; > + int ret; > + > + acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx); > + mutex_lock(acomp_ctx->mutex); > + > + zpool = zswap_find_zpool(entry); > + src = zpool_map_handle(zpool, entry->handle, ZPOOL_MM_RO); > + if (!zpool_can_sleep_mapped(zpool)) { > + memcpy(acomp_ctx->dstmem, src, entry->length); > + src = acomp_ctx->dstmem; > + zpool_unmap_handle(zpool, entry->handle); > + } > + > + sg_init_one(&input, src, entry->length); > + sg_init_table(&output, 1); > + sg_set_page(&output, page, PAGE_SIZE, 0); > + acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen); We should pass PAGE_SIZE here directly, BUG_ON(acomp_ctx->req->dlen) below, and remove the dlen variable. > + ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait); We should just BUG_ON() here directly an remove the ret variable.
On 2023/12/14 08:52, Yosry Ahmed wrote: > On Tue, Dec 12, 2023 at 8:18 PM Chengming Zhou > <zhouchengming@bytedance.com> wrote: >> >> The zswap_load() and zswap_writeback_entry() have the same part that >> decompress the data from zswap_entry to page, so refactor out the >> common part as __zswap_load(entry, page). >> >> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> >> Reviewed-by: Nhat Pham <nphamcs@gmail.com> > > On a second look, there a few nits here. > > First I think it makes more sense to move this refactoring ahead of > reusing destmem. Right now, we add the destmem reuse to zswap_load() > only, then we do the refactor and zswap_writeback_entry() gets it > automatically, so there is a slight change coming to > zswap_writeback_entry() hidden in the refactoring patch. > > Let's refactor out __zswap_load() first, then reuse destmem in it. > Ok, will put it first. >> --- >> mm/zswap.c | 107 ++++++++++++++++++++++--------------------------------------- >> 1 file changed, 38 insertions(+), 69 deletions(-) >> >> diff --git a/mm/zswap.c b/mm/zswap.c >> index fa186945010d..2f095c919a5c 100644 >> --- a/mm/zswap.c >> +++ b/mm/zswap.c >> @@ -1392,6 +1392,41 @@ static int zswap_enabled_param_set(const char *val, >> return ret; >> } >> >> +static void __zswap_load(struct zswap_entry *entry, struct page *page) >> +{ >> + struct scatterlist input, output; >> + unsigned int dlen = PAGE_SIZE; >> + struct crypto_acomp_ctx *acomp_ctx; >> + struct zpool *zpool; >> + u8 *src; >> + int ret; >> + >> + acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx); >> + mutex_lock(acomp_ctx->mutex); >> + >> + zpool = zswap_find_zpool(entry); >> + src = zpool_map_handle(zpool, entry->handle, ZPOOL_MM_RO); >> + if (!zpool_can_sleep_mapped(zpool)) { >> + memcpy(acomp_ctx->dstmem, src, entry->length); >> + src = acomp_ctx->dstmem; >> + zpool_unmap_handle(zpool, entry->handle); >> + } >> + >> + sg_init_one(&input, src, entry->length); >> + sg_init_table(&output, 1); >> + sg_set_page(&output, page, PAGE_SIZE, 0); >> + acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen); > > We should pass PAGE_SIZE here directly, BUG_ON(acomp_ctx->req->dlen) > below, and remove the dlen variable. > >> + ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait); > > We should just BUG_ON() here directly an remove the ret variable. Ok, thanks!
On 2023/12/14 08:52, Yosry Ahmed wrote: > On Tue, Dec 12, 2023 at 8:18 PM Chengming Zhou > <zhouchengming@bytedance.com> wrote: >> >> The zswap_load() and zswap_writeback_entry() have the same part that >> decompress the data from zswap_entry to page, so refactor out the >> common part as __zswap_load(entry, page). >> >> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> >> Reviewed-by: Nhat Pham <nphamcs@gmail.com> > > On a second look, there a few nits here. > > First I think it makes more sense to move this refactoring ahead of > reusing destmem. Right now, we add the destmem reuse to zswap_load() > only, then we do the refactor and zswap_writeback_entry() gets it > automatically, so there is a slight change coming to > zswap_writeback_entry() hidden in the refactoring patch. > > Let's refactor out __zswap_load() first, then reuse destmem in it. I tried but found that putting the __zswap_load() first would introduce another failure case in zswap_writeback_entry(), since the temporary memory allocation may fail. So instead, I also move the dstmem reusing in zswap_writeback_entry() to the dstmem reusing patch. Then this patch becomes having only refactoring. Thanks.
On Mon, Dec 18, 2023 at 12:15 AM Chengming Zhou <zhouchengming@bytedance.com> wrote: > > On 2023/12/14 08:52, Yosry Ahmed wrote: > > On Tue, Dec 12, 2023 at 8:18 PM Chengming Zhou > > <zhouchengming@bytedance.com> wrote: > >> > >> The zswap_load() and zswap_writeback_entry() have the same part that > >> decompress the data from zswap_entry to page, so refactor out the > >> common part as __zswap_load(entry, page). > >> > >> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> > >> Reviewed-by: Nhat Pham <nphamcs@gmail.com> > > > > On a second look, there a few nits here. > > > > First I think it makes more sense to move this refactoring ahead of > > reusing destmem. Right now, we add the destmem reuse to zswap_load() > > only, then we do the refactor and zswap_writeback_entry() gets it > > automatically, so there is a slight change coming to > > zswap_writeback_entry() hidden in the refactoring patch. > > > > Let's refactor out __zswap_load() first, then reuse destmem in it. > > I tried but found that putting the __zswap_load() first would introduce > another failure case in zswap_writeback_entry(), since the temporary > memory allocation may fail. > > So instead, I also move the dstmem reusing in zswap_writeback_entry() to > the dstmem reusing patch. Then this patch becomes having only refactoring. We could have still refactored __zswap_load() first by making it return an int initially when split, then void later. Anyway, it's not a big deal. The new series looks fine.
diff --git a/mm/zswap.c b/mm/zswap.c index fa186945010d..2f095c919a5c 100644 --- a/mm/zswap.c +++ b/mm/zswap.c @@ -1392,6 +1392,41 @@ static int zswap_enabled_param_set(const char *val, return ret; } +static void __zswap_load(struct zswap_entry *entry, struct page *page) +{ + struct scatterlist input, output; + unsigned int dlen = PAGE_SIZE; + struct crypto_acomp_ctx *acomp_ctx; + struct zpool *zpool; + u8 *src; + int ret; + + acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx); + mutex_lock(acomp_ctx->mutex); + + zpool = zswap_find_zpool(entry); + src = zpool_map_handle(zpool, entry->handle, ZPOOL_MM_RO); + if (!zpool_can_sleep_mapped(zpool)) { + memcpy(acomp_ctx->dstmem, src, entry->length); + src = acomp_ctx->dstmem; + zpool_unmap_handle(zpool, entry->handle); + } + + sg_init_one(&input, src, entry->length); + sg_init_table(&output, 1); + sg_set_page(&output, page, PAGE_SIZE, 0); + acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen); + ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait); + dlen = acomp_ctx->req->dlen; + mutex_unlock(acomp_ctx->mutex); + + if (zpool_can_sleep_mapped(zpool)) + zpool_unmap_handle(zpool, entry->handle); + + BUG_ON(ret); + BUG_ON(dlen != PAGE_SIZE); +} + /********************************* * writeback code **********************************/ @@ -1413,23 +1448,12 @@ static int zswap_writeback_entry(struct zswap_entry *entry, swp_entry_t swpentry = entry->swpentry; struct page *page; struct mempolicy *mpol; - struct scatterlist input, output; - struct crypto_acomp_ctx *acomp_ctx; - struct zpool *pool = zswap_find_zpool(entry); bool page_was_allocated; - u8 *src, *tmp = NULL; - unsigned int dlen; int ret; struct writeback_control wbc = { .sync_mode = WB_SYNC_NONE, }; - if (!zpool_can_sleep_mapped(pool)) { - tmp = kmalloc(PAGE_SIZE, GFP_KERNEL); - if (!tmp) - return -ENOMEM; - } - /* try to allocate swap cache page */ mpol = get_task_policy(current); page = __read_swap_cache_async(swpentry, GFP_KERNEL, mpol, @@ -1462,33 +1486,7 @@ static int zswap_writeback_entry(struct zswap_entry *entry, } spin_unlock(&tree->lock); - /* decompress */ - acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx); - dlen = PAGE_SIZE; - - src = zpool_map_handle(pool, entry->handle, ZPOOL_MM_RO); - if (!zpool_can_sleep_mapped(pool)) { - memcpy(tmp, src, entry->length); - src = tmp; - zpool_unmap_handle(pool, entry->handle); - } - - mutex_lock(acomp_ctx->mutex); - sg_init_one(&input, src, entry->length); - sg_init_table(&output, 1); - sg_set_page(&output, page, PAGE_SIZE, 0); - acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen); - ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait); - dlen = acomp_ctx->req->dlen; - mutex_unlock(acomp_ctx->mutex); - - if (!zpool_can_sleep_mapped(pool)) - kfree(tmp); - else - zpool_unmap_handle(pool, entry->handle); - - BUG_ON(ret); - BUG_ON(dlen != PAGE_SIZE); + __zswap_load(entry, page); /* page is up to date */ SetPageUptodate(page); @@ -1508,9 +1506,6 @@ static int zswap_writeback_entry(struct zswap_entry *entry, return ret; fail: - if (!zpool_can_sleep_mapped(pool)) - kfree(tmp); - /* * If we get here because the page is already in swapcache, a * load may be happening concurrently. It is safe and okay to @@ -1769,11 +1764,7 @@ bool zswap_load(struct folio *folio) struct page *page = &folio->page; struct zswap_tree *tree = zswap_trees[type]; struct zswap_entry *entry; - struct scatterlist input, output; - struct crypto_acomp_ctx *acomp_ctx; - unsigned int dlen = PAGE_SIZE; - u8 *src, *dst; - struct zpool *zpool; + u8 *dst; bool ret; VM_WARN_ON_ONCE(!folio_test_locked(folio)); @@ -1795,29 +1786,7 @@ bool zswap_load(struct folio *folio) goto stats; } - /* decompress */ - acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx); - mutex_lock(acomp_ctx->mutex); - - zpool = zswap_find_zpool(entry); - src = zpool_map_handle(zpool, entry->handle, ZPOOL_MM_RO); - if (!zpool_can_sleep_mapped(zpool)) { - memcpy(acomp_ctx->dstmem, src, entry->length); - src = acomp_ctx->dstmem; - zpool_unmap_handle(zpool, entry->handle); - } - - sg_init_one(&input, src, entry->length); - sg_init_table(&output, 1); - sg_set_page(&output, page, PAGE_SIZE, 0); - acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen); - if (crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait)) - WARN_ON(1); - mutex_unlock(acomp_ctx->mutex); - - if (zpool_can_sleep_mapped(zpool)) - zpool_unmap_handle(zpool, entry->handle); - + __zswap_load(entry, page); ret = true; stats: count_vm_event(ZSWPIN);