| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20200210122441.308894706@linuxfoundation.org>
Date: Mon, 10 Feb 2020 04:31:32 -0800
From: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To: linux-kernel@...r.kernel.org
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
stable@...r.kernel.org,
Jonathan Cameron <Jonathan.Cameron@...wei.com>,
Zhou Wang <wangzhou1@...ilicon.com>,
Herbert Xu <herbert@...dor.apana.org.au>
Subject: [PATCH 5.5 179/367] crypto: hisilicon - Use the offset fields in sqe to avoid need to split scatterlists
From: Jonathan Cameron <Jonathan.Cameron@...wei.com>
commit 484a897ffa3005f16cd9a31efd747bcf8155826f upstream.
We can configure sgl offset fields in ZIP sqe to let ZIP engine read/write
sgl data with skipped data. Hence no need to splite the sgl.
Fixes: 62c455ca853e (crypto: hisilicon - add HiSilicon ZIP accelerator support)
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@...wei.com>
Signed-off-by: Zhou Wang <wangzhou1@...ilicon.com>
Signed-off-by: Herbert Xu <herbert@...dor.apana.org.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
---
drivers/crypto/hisilicon/Kconfig | 1
drivers/crypto/hisilicon/zip/zip.h | 4 +
drivers/crypto/hisilicon/zip/zip_crypto.c | 92 +++++++-----------------------
3 files changed, 27 insertions(+), 70 deletions(-)
--- a/drivers/crypto/hisilicon/Kconfig
+++ b/drivers/crypto/hisilicon/Kconfig
@@ -44,7 +44,6 @@ config CRYPTO_DEV_HISI_ZIP
depends on ARM64 || (COMPILE_TEST && 64BIT)
depends on !CPU_BIG_ENDIAN || COMPILE_TEST
select CRYPTO_DEV_HISI_QM
- select SG_SPLIT
help
Support for HiSilicon ZIP Driver
--- a/drivers/crypto/hisilicon/zip/zip.h
+++ b/drivers/crypto/hisilicon/zip/zip.h
@@ -11,6 +11,10 @@
/* hisi_zip_sqe dw3 */
#define HZIP_BD_STATUS_M GENMASK(7, 0)
+/* hisi_zip_sqe dw7 */
+#define HZIP_IN_SGE_DATA_OFFSET_M GENMASK(23, 0)
+/* hisi_zip_sqe dw8 */
+#define HZIP_OUT_SGE_DATA_OFFSET_M GENMASK(23, 0)
/* hisi_zip_sqe dw9 */
#define HZIP_REQ_TYPE_M GENMASK(7, 0)
#define HZIP_ALG_TYPE_ZLIB 0x02
--- a/drivers/crypto/hisilicon/zip/zip_crypto.c
+++ b/drivers/crypto/hisilicon/zip/zip_crypto.c
@@ -46,10 +46,8 @@ enum hisi_zip_alg_type {
struct hisi_zip_req {
struct acomp_req *req;
- struct scatterlist *src;
- struct scatterlist *dst;
- size_t slen;
- size_t dlen;
+ int sskip;
+ int dskip;
struct hisi_acc_hw_sgl *hw_src;
struct hisi_acc_hw_sgl *hw_dst;
dma_addr_t dma_src;
@@ -119,13 +117,15 @@ static void hisi_zip_config_tag(struct h
static void hisi_zip_fill_sqe(struct hisi_zip_sqe *sqe, u8 req_type,
dma_addr_t s_addr, dma_addr_t d_addr, u32 slen,
- u32 dlen)
+ u32 dlen, int sskip, int dskip)
{
memset(sqe, 0, sizeof(struct hisi_zip_sqe));
- sqe->input_data_length = slen;
+ sqe->input_data_length = slen - sskip;
+ sqe->dw7 = FIELD_PREP(HZIP_IN_SGE_DATA_OFFSET_M, sskip);
+ sqe->dw8 = FIELD_PREP(HZIP_OUT_SGE_DATA_OFFSET_M, dskip);
sqe->dw9 = FIELD_PREP(HZIP_REQ_TYPE_M, req_type);
- sqe->dest_avail_out = dlen;
+ sqe->dest_avail_out = dlen - dskip;
sqe->source_addr_l = lower_32_bits(s_addr);
sqe->source_addr_h = upper_32_bits(s_addr);
sqe->dest_addr_l = lower_32_bits(d_addr);
@@ -327,11 +327,6 @@ static void hisi_zip_remove_req(struct h
{
struct hisi_zip_req_q *req_q = &qp_ctx->req_q;
- if (qp_ctx->qp->alg_type == HZIP_ALG_TYPE_COMP)
- kfree(req->dst);
- else
- kfree(req->src);
-
write_lock(&req_q->req_lock);
clear_bit(req->req_id, req_q->req_bitmap);
memset(req, 0, sizeof(struct hisi_zip_req));
@@ -359,8 +354,8 @@ static void hisi_zip_acomp_cb(struct his
}
dlen = sqe->produced;
- hisi_acc_sg_buf_unmap(dev, req->src, req->hw_src);
- hisi_acc_sg_buf_unmap(dev, req->dst, req->hw_dst);
+ hisi_acc_sg_buf_unmap(dev, acomp_req->src, req->hw_src);
+ hisi_acc_sg_buf_unmap(dev, acomp_req->dst, req->hw_dst);
head_size = (qp->alg_type == 0) ? TO_HEAD_SIZE(qp->req_type) : 0;
acomp_req->dlen = dlen + head_size;
@@ -454,20 +449,6 @@ static size_t get_comp_head_size(struct
}
}
-static int get_sg_skip_bytes(struct scatterlist *sgl, size_t bytes,
- size_t remains, struct scatterlist **out)
-{
-#define SPLIT_NUM 2
- size_t split_sizes[SPLIT_NUM];
- int out_mapped_nents[SPLIT_NUM];
-
- split_sizes[0] = bytes;
- split_sizes[1] = remains;
-
- return sg_split(sgl, 0, 0, SPLIT_NUM, split_sizes, out,
- out_mapped_nents, GFP_KERNEL);
-}
-
static struct hisi_zip_req *hisi_zip_create_req(struct acomp_req *req,
struct hisi_zip_qp_ctx *qp_ctx,
size_t head_size, bool is_comp)
@@ -475,31 +456,7 @@ static struct hisi_zip_req *hisi_zip_cre
struct hisi_zip_req_q *req_q = &qp_ctx->req_q;
struct hisi_zip_req *q = req_q->q;
struct hisi_zip_req *req_cache;
- struct scatterlist *out[2];
- struct scatterlist *sgl;
- size_t len;
- int ret, req_id;
-
- /*
- * remove/add zlib/gzip head, as hardware operations do not include
- * comp head. so split req->src to get sgl without heads in acomp, or
- * add comp head to req->dst ahead of that hardware output compressed
- * data in sgl splited from req->dst without comp head.
- */
- if (is_comp) {
- sgl = req->dst;
- len = req->dlen - head_size;
- } else {
- sgl = req->src;
- len = req->slen - head_size;
- }
-
- ret = get_sg_skip_bytes(sgl, head_size, len, out);
- if (ret)
- return ERR_PTR(ret);
-
- /* sgl for comp head is useless, so free it now */
- kfree(out[0]);
+ int req_id;
write_lock(&req_q->req_lock);
@@ -507,7 +464,6 @@ static struct hisi_zip_req *hisi_zip_cre
if (req_id >= req_q->size) {
write_unlock(&req_q->req_lock);
dev_dbg(&qp_ctx->qp->qm->pdev->dev, "req cache is full!\n");
- kfree(out[1]);
return ERR_PTR(-EBUSY);
}
set_bit(req_id, req_q->req_bitmap);
@@ -515,16 +471,13 @@ static struct hisi_zip_req *hisi_zip_cre
req_cache = q + req_id;
req_cache->req_id = req_id;
req_cache->req = req;
+
if (is_comp) {
- req_cache->src = req->src;
- req_cache->dst = out[1];
- req_cache->slen = req->slen;
- req_cache->dlen = req->dlen - head_size;
+ req_cache->sskip = 0;
+ req_cache->dskip = head_size;
} else {
- req_cache->src = out[1];
- req_cache->dst = req->dst;
- req_cache->slen = req->slen - head_size;
- req_cache->dlen = req->dlen;
+ req_cache->sskip = head_size;
+ req_cache->dskip = 0;
}
write_unlock(&req_q->req_lock);
@@ -536,6 +489,7 @@ static int hisi_zip_do_work(struct hisi_
struct hisi_zip_qp_ctx *qp_ctx)
{
struct hisi_zip_sqe *zip_sqe = &qp_ctx->zip_sqe;
+ struct acomp_req *a_req = req->req;
struct hisi_qp *qp = qp_ctx->qp;
struct device *dev = &qp->qm->pdev->dev;
struct hisi_acc_sgl_pool *pool = qp_ctx->sgl_pool;
@@ -543,16 +497,16 @@ static int hisi_zip_do_work(struct hisi_
dma_addr_t output;
int ret;
- if (!req->src || !req->slen || !req->dst || !req->dlen)
+ if (!a_req->src || !a_req->slen || !a_req->dst || !a_req->dlen)
return -EINVAL;
- req->hw_src = hisi_acc_sg_buf_map_to_hw_sgl(dev, req->src, pool,
+ req->hw_src = hisi_acc_sg_buf_map_to_hw_sgl(dev, a_req->src, pool,
req->req_id << 1, &input);
if (IS_ERR(req->hw_src))
return PTR_ERR(req->hw_src);
req->dma_src = input;
- req->hw_dst = hisi_acc_sg_buf_map_to_hw_sgl(dev, req->dst, pool,
+ req->hw_dst = hisi_acc_sg_buf_map_to_hw_sgl(dev, a_req->dst, pool,
(req->req_id << 1) + 1,
&output);
if (IS_ERR(req->hw_dst)) {
@@ -561,8 +515,8 @@ static int hisi_zip_do_work(struct hisi_
}
req->dma_dst = output;
- hisi_zip_fill_sqe(zip_sqe, qp->req_type, input, output, req->slen,
- req->dlen);
+ hisi_zip_fill_sqe(zip_sqe, qp->req_type, input, output, a_req->slen,
+ a_req->dlen, req->sskip, req->dskip);
hisi_zip_config_buf_type(zip_sqe, HZIP_SGL);
hisi_zip_config_tag(zip_sqe, req->req_id);
@@ -574,9 +528,9 @@ static int hisi_zip_do_work(struct hisi_
return -EINPROGRESS;
err_unmap_output:
- hisi_acc_sg_buf_unmap(dev, req->dst, req->hw_dst);
+ hisi_acc_sg_buf_unmap(dev, a_req->dst, req->hw_dst);
err_unmap_input:
- hisi_acc_sg_buf_unmap(dev, req->src, req->hw_src);
+ hisi_acc_sg_buf_unmap(dev, a_req->src, req->hw_src);
return ret;
}
Powered by blists - more mailing lists