Received-SPF: pass (google.com: domain of linux-crypto+bounces-2451-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Barry Song <21cnbao@gmail.com>
To: herbert@gondor.apana.org.au,
	davem@davemloft.net,
	linux-crypto@vger.kernel.org
Cc: akpm@linux-foundation.org,
	chrisl@kernel.org,
	sjenning@redhat.com,
	vitaly.wool@konsulko.com,
	linux-kernel@vger.kernel.org,
	Barry Song <v-songbaohua@oppo.com>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Nhat Pham <nphamcs@gmail.com>,
	Yosry Ahmed <yosryahmed@google.com>,
	Chengming Zhou <zhouchengming@bytedance.com>
Subject: [PATCH v7] crypto: scompress: remove memcpy if sg_nents is 1 and pages are lowmem
Date: Sat,  2 Mar 2024 08:27:45 +1300
Message-Id: <20240301192745.14987-1-21cnbao@gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

From: Barry Song <v-songbaohua@oppo.com>

while sg_nents is 1, which is always true for the current kernel
as the only user - zswap is this case, we might have a chance to
remove memcpy, thus improve the performance.
Though sg_nents is 1, its buffer might cross two pages. If those
pages are highmem, we have no cheap way to map them to contiguous
virtual address because kmap doesn't support more than one page
(kmap single higmem page could be still expensive for tlb) and
vmap is expensive.
So we also test and enure page is not highmem in order to safely
use page_to_virt before removing the memcpy. The good news is
that in the most majority of cases, we are lowmem, and we are
always lowmem in those modern and popular hardware.

Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Tested-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 -v7:
 * fix the problem pointed out by Herbert - flush all pages if dst
   is longer than one page.

 crypto/scompress.c | 36 +++++++++++++++++++++++++++++-------
 1 file changed, 29 insertions(+), 7 deletions(-)

diff --git a/crypto/scompress.c b/crypto/scompress.c
index b108a30a7600..60bbb7ea4060 100644
--- a/crypto/scompress.c
+++ b/crypto/scompress.c
@@ -117,6 +117,7 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)
 	struct crypto_scomp *scomp = *tfm_ctx;
 	void **ctx = acomp_request_ctx(req);
 	struct scomp_scratch *scratch;
+	void *src, *dst;
 	unsigned int dlen;
 	int ret;
 
@@ -134,13 +135,25 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)
 	scratch = raw_cpu_ptr(&scomp_scratch);
 	spin_lock(&scratch->lock);
 
-	scatterwalk_map_and_copy(scratch->src, req->src, 0, req->slen, 0);
+	if (sg_nents(req->src) == 1 && !PageHighMem(sg_page(req->src))) {
+		src = page_to_virt(sg_page(req->src)) + req->src->offset;
+	} else {
+		scatterwalk_map_and_copy(scratch->src, req->src, 0,
+					 req->slen, 0);
+		src = scratch->src;
+	}
+
+	if (req->dst && sg_nents(req->dst) == 1 && !PageHighMem(sg_page(req->dst)))
+		dst = page_to_virt(sg_page(req->dst)) + req->dst->offset;
+	else
+		dst = scratch->dst;
+
 	if (dir)
-		ret = crypto_scomp_compress(scomp, scratch->src, req->slen,
-					    scratch->dst, &req->dlen, *ctx);
+		ret = crypto_scomp_compress(scomp, src, req->slen,
+					    dst, &req->dlen, *ctx);
 	else
-		ret = crypto_scomp_decompress(scomp, scratch->src, req->slen,
-					      scratch->dst, &req->dlen, *ctx);
+		ret = crypto_scomp_decompress(scomp, src, req->slen,
+					      dst, &req->dlen, *ctx);
 	if (!ret) {
 		if (!req->dst) {
 			req->dst = sgl_alloc(req->dlen, GFP_ATOMIC, NULL);
@@ -152,8 +165,17 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)
 			ret = -ENOSPC;
 			goto out;
 		}
-		scatterwalk_map_and_copy(scratch->dst, req->dst, 0, req->dlen,
-					 1);
+		if (dst == scratch->dst) {
+			scatterwalk_map_and_copy(scratch->dst, req->dst, 0,
+						 req->dlen, 1);
+		} else {
+			int nr_pages = DIV_ROUND_UP(req->dst->offset + req->dlen, PAGE_SIZE);
+			int i;
+			struct page *dst_page = sg_page(req->dst);
+
+			for (i = 0; i < nr_pages; i++)
+				flush_dcache_page(dst_page + i);
+		}
 	}
 out:
 	spin_unlock(&scratch->lock);
-- 
2.34.1