summaryrefslogtreecommitdiffstats
path: root/drivers/crypto
diff options
context:
space:
mode:
authorDmitry Kasatkin <dmitry.kasatkin@nokia.com>2010-11-19 16:04:29 +0200
committerHerbert Xu <herbert@gondor.apana.org.au>2010-11-27 16:37:19 +0800
commit887c883eea9867535059f3c8414c8cfc952ccff1 (patch)
tree2a35344c0b1de81b5b4a70bfe416c6605c5809c7 /drivers/crypto
parenta55b290b0e41e02d1969589c5d77d966ac2b7ec8 (diff)
downloadtalos-op-linux-887c883eea9867535059f3c8414c8cfc952ccff1.tar.gz
talos-op-linux-887c883eea9867535059f3c8414c8cfc952ccff1.zip
crypto: omap-sham - zero-copy scatterlist handling
If scatterlist have more than one entry, current driver uses aligned buffer to copy data to to accelerator to tackle possible issues with DMA and SHA buffer alignment. This commit adds more intelligence to verify SG alignment and possibility to use DMA directly on the data without using copy buffer. Signed-off-by: Dmitry Kasatkin <dmitry.kasatkin@nokia.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Diffstat (limited to 'drivers/crypto')
-rw-r--r--drivers/crypto/omap-sham.c87
1 files changed, 61 insertions, 26 deletions
diff --git a/drivers/crypto/omap-sham.c b/drivers/crypto/omap-sham.c
index d88d7ebfffa7..eb988e7a2fd9 100644
--- a/drivers/crypto/omap-sham.c
+++ b/drivers/crypto/omap-sham.c
@@ -74,7 +74,7 @@
#define FLAGS_FINUP 0x0002
#define FLAGS_FINAL 0x0004
-#define FLAGS_FAST 0x0008
+#define FLAGS_SG 0x0008
#define FLAGS_SHA1 0x0010
#define FLAGS_DMA_ACTIVE 0x0020
#define FLAGS_OUTPUT_READY 0x0040
@@ -393,6 +393,8 @@ static int omap_sham_xmit_dma_map(struct omap_sham_dev *dd,
return -EINVAL;
}
+ ctx->flags &= ~FLAGS_SG;
+
/* next call does not fail... so no unmap in the case of error */
return omap_sham_xmit_dma(dd, ctx->dma_addr, length, final);
}
@@ -403,9 +405,6 @@ static int omap_sham_update_dma_slow(struct omap_sham_dev *dd)
unsigned int final;
size_t count;
- if (!ctx->total)
- return 0;
-
omap_sham_append_sg(ctx);
final = (ctx->flags & FLAGS_FINUP) && !ctx->total;
@@ -422,25 +421,62 @@ static int omap_sham_update_dma_slow(struct omap_sham_dev *dd)
return 0;
}
-static int omap_sham_update_dma_fast(struct omap_sham_dev *dd)
+/* Start address alignment */
+#define SG_AA(sg) (IS_ALIGNED(sg->offset, sizeof(u32)))
+/* SHA1 block size alignment */
+#define SG_SA(sg) (IS_ALIGNED(sg->length, SHA1_MD5_BLOCK_SIZE))
+
+static int omap_sham_update_dma_start(struct omap_sham_dev *dd)
{
struct omap_sham_reqctx *ctx = ahash_request_ctx(dd->req);
- unsigned int length;
+ unsigned int length, final, tail;
+ struct scatterlist *sg;
+
+ if (!ctx->total)
+ return 0;
+
+ if (ctx->bufcnt || ctx->offset)
+ return omap_sham_update_dma_slow(dd);
+
+ dev_dbg(dd->dev, "fast: digcnt: %d, bufcnt: %u, total: %u\n",
+ ctx->digcnt, ctx->bufcnt, ctx->total);
+
+ sg = ctx->sg;
- ctx->flags |= FLAGS_FAST;
+ if (!SG_AA(sg))
+ return omap_sham_update_dma_slow(dd);
- length = min(ctx->total, sg_dma_len(ctx->sg));
- ctx->total = length;
+ if (!sg_is_last(sg) && !SG_SA(sg))
+ /* size is not SHA1_BLOCK_SIZE aligned */
+ return omap_sham_update_dma_slow(dd);
+
+ length = min(ctx->total, sg->length);
+
+ if (sg_is_last(sg)) {
+ if (!(ctx->flags & FLAGS_FINUP)) {
+ /* not last sg must be SHA1_MD5_BLOCK_SIZE aligned */
+ tail = length & (SHA1_MD5_BLOCK_SIZE - 1);
+ /* without finup() we need one block to close hash */
+ if (!tail)
+ tail = SHA1_MD5_BLOCK_SIZE;
+ length -= tail;
+ }
+ }
if (!dma_map_sg(dd->dev, ctx->sg, 1, DMA_TO_DEVICE)) {
dev_err(dd->dev, "dma_map_sg error\n");
return -EINVAL;
}
+ ctx->flags |= FLAGS_SG;
+
ctx->total -= length;
+ ctx->offset = length; /* offset where to start slow */
+
+ final = (ctx->flags & FLAGS_FINUP) && !ctx->total;
/* next call does not fail... so no unmap in the case of error */
- return omap_sham_xmit_dma(dd, sg_dma_address(ctx->sg), length, 1);
+ return omap_sham_xmit_dma(dd, sg_dma_address(ctx->sg), length, final);
}
static int omap_sham_update_cpu(struct omap_sham_dev *dd)
@@ -460,11 +496,17 @@ static int omap_sham_update_dma_stop(struct omap_sham_dev *dd)
struct omap_sham_reqctx *ctx = ahash_request_ctx(dd->req);
omap_stop_dma(dd->dma_lch);
- if (ctx->flags & FLAGS_FAST)
+ if (ctx->flags & FLAGS_SG) {
dma_unmap_sg(dd->dev, ctx->sg, 1, DMA_TO_DEVICE);
- else
+ if (ctx->sg->length == ctx->offset) {
+ ctx->sg = sg_next(ctx->sg);
+ if (ctx->sg)
+ ctx->offset = 0;
+ }
+ } else {
dma_unmap_single(dd->dev, ctx->dma_addr, ctx->buflen,
DMA_TO_DEVICE);
+ }
return 0;
}
@@ -545,10 +587,8 @@ static int omap_sham_update_req(struct omap_sham_dev *dd)
if (ctx->flags & FLAGS_CPU)
err = omap_sham_update_cpu(dd);
- else if (ctx->flags & FLAGS_FAST)
- err = omap_sham_update_dma_fast(dd);
else
- err = omap_sham_update_dma_slow(dd);
+ err = omap_sham_update_dma_start(dd);
/* wait for dma completion before can take more data */
dev_dbg(dd->dev, "update: err: %d, digcnt: %d\n", err, ctx->digcnt);
@@ -730,18 +770,13 @@ static int omap_sham_update(struct ahash_request *req)
*/
omap_sham_append_sg(ctx);
return 0;
- } else if (ctx->bufcnt + ctx->total <= 64) {
+ } else if (ctx->bufcnt + ctx->total <= SHA1_MD5_BLOCK_SIZE) {
+ /*
+ * faster to use CPU for short transfers
+ */
ctx->flags |= FLAGS_CPU;
- } else if (!ctx->bufcnt && sg_is_last(ctx->sg)) {
- /* may be can use faster functions */
- int aligned = IS_ALIGNED((u32)ctx->sg->offset,
- sizeof(u32));
- if (aligned)
- /* digest: first and final */
- ctx->flags |= FLAGS_FAST;
}
- } else if (ctx->bufcnt + ctx->total <= ctx->buflen) {
- /* if not finaup -> not fast */
+ } else if (ctx->bufcnt + ctx->total < ctx->buflen) {
omap_sham_append_sg(ctx);
return 0;
}
@@ -1026,7 +1061,7 @@ static void omap_sham_done_task(unsigned long data)
dd->flags &= ~FLAGS_DMA_ACTIVE;
omap_sham_update_dma_stop(dd);
if (!dd->err)
- err = omap_sham_update_dma_slow(dd);
+ err = omap_sham_update_dma_start(dd);
}
err = dd->err ? : err;
OpenPOWER on IntegriCloud