[PATCH 8/8] Replacing spinlocks by nx_copy_ctx on NX-SHA256 and NX-SHA512

From: Leonidas Da Silva Barbosa
Date: Thu Jul 24 2014 - 10:02:55 EST

Next message: Arnaldo Carvalho de Melo: "Re: [PATCH 13/52] perf tools: Add perf_pmu__scan_file()"
Previous message: Arnaldo Carvalho de Melo: "Re: [RFC PATCH 1/2] powerpc/perf: include util/util.h and remove stringify macros"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Replaces spinlock usage by a simple copy of the crypto context, avoiding
possible bottlenecks.

Signed-off-by: Leonidas Da Silva Barbosa <leosilva@xxxxxxxxxxxxxxxxxx>
---
drivers/crypto/nx/nx-sha256.c | 70 +++++++++++++++++++++++++----------------
drivers/crypto/nx/nx-sha512.c | 70 +++++++++++++++++++++++++----------------
2 files changed, 86 insertions(+), 54 deletions(-)

diff --git a/drivers/crypto/nx/nx-sha256.c b/drivers/crypto/nx/nx-sha256.c
index da0b24a..3182e9f 100644
--- a/drivers/crypto/nx/nx-sha256.c
+++ b/drivers/crypto/nx/nx-sha256.c
@@ -57,10 +57,14 @@ static int nx_sha256_update(struct shash_desc *desc, const u8 *data,
struct nx_sg *in_sg;
u64 to_process, leftover, total;
u32 max_sg_len;
- unsigned long irq_flags;
int rc = 0;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;
+
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;

/* 2 cases for total data len:
* 1: < SHA256_BLOCK_SIZE: copy into state, return 0
@@ -73,9 +77,9 @@ static int nx_sha256_update(struct shash_desc *desc, const u8 *data,
goto out;
}

- in_sg = nx_ctx->in_sg;
+ in_sg = curr_nx_ctx.in_sg;
max_sg_len = min_t(u32, nx_driver.of.max_sg_len/sizeof(struct nx_sg),
- nx_ctx->ap->sglen);
+ curr_nx_ctx.ap->sglen);

do {
/*
@@ -83,21 +87,21 @@ static int nx_sha256_update(struct shash_desc *desc, const u8 *data,
* this update. This value is also restricted by the sg list
* limits.
*/
- to_process = min_t(u64, total, nx_ctx->ap->databytelen);
+ to_process = min_t(u64, total, curr_nx_ctx.ap->databytelen);
to_process = min_t(u64, to_process,
NX_PAGE_SIZE * (max_sg_len - 1));
to_process = to_process & ~(SHA256_BLOCK_SIZE - 1);
leftover = total - to_process;

if (sctx->count) {
- in_sg = nx_build_sg_list(nx_ctx->in_sg,
+ in_sg = nx_build_sg_list(curr_nx_ctx.in_sg,
(u8 *) sctx->buf,
sctx->count, max_sg_len);
}
in_sg = nx_build_sg_list(in_sg, (u8 *) data,
to_process - sctx->count,
max_sg_len);
- nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) *
+ curr_nx_ctx.op.inlen = (curr_nx_ctx.in_sg - in_sg) *
sizeof(struct nx_sg);

if (NX_CPB_FDM(csbcpb) & NX_FDM_CONTINUATION) {
@@ -111,12 +115,12 @@ static int nx_sha256_update(struct shash_desc *desc, const u8 *data,
}

NX_CPB_FDM(csbcpb) |= NX_FDM_INTERMEDIATE;
- if (!nx_ctx->op.inlen || !nx_ctx->op.outlen) {
+ if (!curr_nx_ctx.op.inlen || !curr_nx_ctx.op.outlen) {
rc = -EINVAL;
goto out;
}

- rc = nx_hcall_sync(nx_ctx, &nx_ctx->op,
+ rc = nx_hcall_sync(&curr_nx_ctx, &curr_nx_ctx.op,
desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP);
if (rc)
goto out;
@@ -131,7 +135,7 @@ static int nx_sha256_update(struct shash_desc *desc, const u8 *data,
total -= to_process;
data += to_process - sctx->count;
sctx->count = 0;
- in_sg = nx_ctx->in_sg;
+ in_sg = curr_nx_ctx.in_sg;
} while (leftover >= SHA256_BLOCK_SIZE);

/* copy the leftover back into the state struct */
@@ -139,7 +143,7 @@ static int nx_sha256_update(struct shash_desc *desc, const u8 *data,
memcpy(sctx->buf, data, leftover);
sctx->count = leftover;
out:
- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return rc;
}

@@ -150,12 +154,16 @@ static int nx_sha256_final(struct shash_desc *desc, u8 *out)
struct nx_csbcpb *csbcpb = (struct nx_csbcpb *)nx_ctx->csbcpb;
struct nx_sg *in_sg, *out_sg;
u32 max_sg_len;
- unsigned long irq_flags;
int rc;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;
+
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;

- max_sg_len = min_t(u32, nx_driver.of.max_sg_len, nx_ctx->ap->sglen);
+ csbcpb = curr_nx_ctx.csbcpb;
+
+ max_sg_len = min_t(u32, nx_driver.of.max_sg_len, curr_nx_ctx.ap->sglen);

if (NX_CPB_FDM(csbcpb) & NX_FDM_CONTINUATION) {
/* we've hit the nx chip previously, now we're finalizing,
@@ -170,19 +178,19 @@ static int nx_sha256_final(struct shash_desc *desc, u8 *out)

csbcpb->cpb.sha256.message_bit_length += (u64)(sctx->count * 8);

- in_sg = nx_build_sg_list(nx_ctx->in_sg, (u8 *)sctx->buf,
+ in_sg = nx_build_sg_list(curr_nx_ctx.in_sg, (u8 *)sctx->buf,
sctx->count, max_sg_len);
- out_sg = nx_build_sg_list(nx_ctx->out_sg, out, SHA256_DIGEST_SIZE,
+ out_sg = nx_build_sg_list(curr_nx_ctx.out_sg, out, SHA256_DIGEST_SIZE,
max_sg_len);
- nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) * sizeof(struct nx_sg);
- nx_ctx->op.outlen = (nx_ctx->out_sg - out_sg) * sizeof(struct nx_sg);
+ curr_nx_ctx.op.inlen = (curr_nx_ctx.in_sg - in_sg) * sizeof(struct nx_sg);
+ curr_nx_ctx.op.outlen = (curr_nx_ctx.out_sg - out_sg) * sizeof(struct nx_sg);

- if (!nx_ctx->op.outlen) {
+ if (!curr_nx_ctx.op.outlen) {
rc = -EINVAL;
goto out;
}

- rc = nx_hcall_sync(nx_ctx, &nx_ctx->op,
+ rc = nx_hcall_sync(&curr_nx_ctx, &curr_nx_ctx.op,
desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP);
if (rc)
goto out;
@@ -193,7 +201,7 @@ static int nx_sha256_final(struct shash_desc *desc, u8 *out)
&(nx_ctx->stats->sha256_bytes));
memcpy(out, csbcpb->cpb.sha256.message_digest, SHA256_DIGEST_SIZE);
out:
- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return rc;
}

@@ -203,10 +211,14 @@ static int nx_sha256_export(struct shash_desc *desc, void *out)
struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
struct nx_csbcpb *csbcpb = (struct nx_csbcpb *)nx_ctx->csbcpb;
struct sha256_state *octx = out;
- unsigned long irq_flags;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;

+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;
+
octx->count = sctx->count +
(csbcpb->cpb.sha256.message_bit_length / 8);
memcpy(octx->buf, sctx->buf, sizeof(octx->buf));
@@ -228,7 +240,7 @@ static int nx_sha256_export(struct shash_desc *desc, void *out)
octx->state[7] = SHA256_H7;
}

- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return 0;
}

@@ -238,9 +250,13 @@ static int nx_sha256_import(struct shash_desc *desc, const void *in)
struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
struct nx_csbcpb *csbcpb = (struct nx_csbcpb *)nx_ctx->csbcpb;
const struct sha256_state *ictx = in;
- unsigned long irq_flags;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;
+
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;

memcpy(sctx->buf, ictx->buf, sizeof(ictx->buf));

@@ -255,7 +271,7 @@ static int nx_sha256_import(struct shash_desc *desc, const void *in)
NX_CPB_FDM(csbcpb) |= NX_FDM_INTERMEDIATE;
}

- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return 0;
}

diff --git a/drivers/crypto/nx/nx-sha512.c b/drivers/crypto/nx/nx-sha512.c
index 4ae5b0f..7fc2583 100644
--- a/drivers/crypto/nx/nx-sha512.c
+++ b/drivers/crypto/nx/nx-sha512.c
@@ -57,10 +57,14 @@ static int nx_sha512_update(struct shash_desc *desc, const u8 *data,
struct nx_sg *in_sg;
u64 to_process, leftover, total, spbc_bits;
u32 max_sg_len;
- unsigned long irq_flags;
int rc = 0;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;
+
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;

/* 2 cases for total data len:
* 1: < SHA512_BLOCK_SIZE: copy into state, return 0
@@ -73,9 +77,9 @@ static int nx_sha512_update(struct shash_desc *desc, const u8 *data,
goto out;
}

- in_sg = nx_ctx->in_sg;
+ in_sg = curr_nx_ctx.in_sg;
max_sg_len = min_t(u32, nx_driver.of.max_sg_len/sizeof(struct nx_sg),
- nx_ctx->ap->sglen);
+ curr_nx_ctx.ap->sglen);

do {
/*
@@ -83,21 +87,21 @@ static int nx_sha512_update(struct shash_desc *desc, const u8 *data,
* this update. This value is also restricted by the sg list
* limits.
*/
- to_process = min_t(u64, total, nx_ctx->ap->databytelen);
+ to_process = min_t(u64, total, curr_nx_ctx.ap->databytelen);
to_process = min_t(u64, to_process,
NX_PAGE_SIZE * (max_sg_len - 1));
to_process = to_process & ~(SHA512_BLOCK_SIZE - 1);
leftover = total - to_process;

if (sctx->count[0]) {
- in_sg = nx_build_sg_list(nx_ctx->in_sg,
+ in_sg = nx_build_sg_list(curr_nx_ctx.in_sg,
(u8 *) sctx->buf,
sctx->count[0], max_sg_len);
}
in_sg = nx_build_sg_list(in_sg, (u8 *) data,
to_process - sctx->count[0],
max_sg_len);
- nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) *
+ curr_nx_ctx.op.inlen = (curr_nx_ctx.in_sg - in_sg) *
sizeof(struct nx_sg);

if (NX_CPB_FDM(csbcpb) & NX_FDM_CONTINUATION) {
@@ -111,12 +115,12 @@ static int nx_sha512_update(struct shash_desc *desc, const u8 *data,
}

NX_CPB_FDM(csbcpb) |= NX_FDM_INTERMEDIATE;
- if (!nx_ctx->op.inlen || !nx_ctx->op.outlen) {
+ if (!curr_nx_ctx.op.inlen || !curr_nx_ctx.op.outlen) {
rc = -EINVAL;
goto out;
}

- rc = nx_hcall_sync(nx_ctx, &nx_ctx->op,
+ rc = nx_hcall_sync(&curr_nx_ctx, &curr_nx_ctx.op,
desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP);
if (rc)
goto out;
@@ -133,7 +137,7 @@ static int nx_sha512_update(struct shash_desc *desc, const u8 *data,
total -= to_process;
data += to_process - sctx->count[0];
sctx->count[0] = 0;
- in_sg = nx_ctx->in_sg;
+ in_sg = curr_nx_ctx.in_sg;
} while (leftover >= SHA512_BLOCK_SIZE);

/* copy the leftover back into the state struct */
@@ -141,7 +145,7 @@ static int nx_sha512_update(struct shash_desc *desc, const u8 *data,
memcpy(sctx->buf, data, leftover);
sctx->count[0] = leftover;
out:
- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return rc;
}

@@ -153,12 +157,16 @@ static int nx_sha512_final(struct shash_desc *desc, u8 *out)
struct nx_sg *in_sg, *out_sg;
u32 max_sg_len;
u64 count0;
- unsigned long irq_flags;
int rc;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;

- max_sg_len = min_t(u32, nx_driver.of.max_sg_len, nx_ctx->ap->sglen);
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;
+
+ max_sg_len = min_t(u32, nx_driver.of.max_sg_len, curr_nx_ctx.ap->sglen);

if (NX_CPB_FDM(csbcpb) & NX_FDM_CONTINUATION) {
/* we've hit the nx chip previously, now we're finalizing,
@@ -177,19 +185,19 @@ static int nx_sha512_final(struct shash_desc *desc, u8 *out)
if (csbcpb->cpb.sha512.message_bit_length_lo < count0)
csbcpb->cpb.sha512.message_bit_length_hi++;

- in_sg = nx_build_sg_list(nx_ctx->in_sg, sctx->buf, sctx->count[0],
+ in_sg = nx_build_sg_list(curr_nx_ctx.in_sg, sctx->buf, sctx->count[0],
max_sg_len);
- out_sg = nx_build_sg_list(nx_ctx->out_sg, out, SHA512_DIGEST_SIZE,
+ out_sg = nx_build_sg_list(curr_nx_ctx.out_sg, out, SHA512_DIGEST_SIZE,
max_sg_len);
- nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) * sizeof(struct nx_sg);
- nx_ctx->op.outlen = (nx_ctx->out_sg - out_sg) * sizeof(struct nx_sg);
+ curr_nx_ctx.op.inlen = (curr_nx_ctx.in_sg - in_sg) * sizeof(struct nx_sg);
+ curr_nx_ctx.op.outlen = (curr_nx_ctx.out_sg - out_sg) * sizeof(struct nx_sg);

- if (!nx_ctx->op.outlen) {
+ if (!curr_nx_ctx.op.outlen) {
rc = -EINVAL;
goto out;
}

- rc = nx_hcall_sync(nx_ctx, &nx_ctx->op,
+ rc = nx_hcall_sync(&curr_nx_ctx, &curr_nx_ctx.op,
desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP);
if (rc)
goto out;
@@ -200,7 +208,7 @@ static int nx_sha512_final(struct shash_desc *desc, u8 *out)

memcpy(out, csbcpb->cpb.sha512.message_digest, SHA512_DIGEST_SIZE);
out:
- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return rc;
}

@@ -210,9 +218,13 @@ static int nx_sha512_export(struct shash_desc *desc, void *out)
struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
struct nx_csbcpb *csbcpb = (struct nx_csbcpb *)nx_ctx->csbcpb;
struct sha512_state *octx = out;
- unsigned long irq_flags;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;
+
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;

/* move message_bit_length (128 bits) into count and convert its value
* to bytes */
@@ -244,7 +256,7 @@ static int nx_sha512_export(struct shash_desc *desc, void *out)
octx->state[7] = SHA512_H7;
}

- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return 0;
}

@@ -254,9 +266,13 @@ static int nx_sha512_import(struct shash_desc *desc, const void *in)
struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
struct nx_csbcpb *csbcpb = (struct nx_csbcpb *)nx_ctx->csbcpb;
const struct sha512_state *ictx = in;
- unsigned long irq_flags;

- spin_lock_irqsave(&nx_ctx->lock, irq_flags);
+ struct nx_crypto_ctx curr_nx_ctx;
+
+ if (nx_copy_ctx(&curr_nx_ctx, nx_ctx))
+ return -ENOMEM;
+
+ csbcpb = curr_nx_ctx.csbcpb;

memcpy(sctx->buf, ictx->buf, sizeof(ictx->buf));
sctx->count[0] = ictx->count[0] & 0x3f;
@@ -274,7 +290,7 @@ static int nx_sha512_import(struct shash_desc *desc, const void *in)
NX_CPB_FDM(csbcpb) |= NX_FDM_INTERMEDIATE;
}

- spin_unlock_irqrestore(&nx_ctx->lock, irq_flags);
+ kmem_cache_free(nx_driver.slab, curr_nx_ctx.kmem);
return 0;
}

--
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Next message: Arnaldo Carvalho de Melo: "Re: [PATCH 13/52] perf tools: Add perf_pmu__scan_file()"
Previous message: Arnaldo Carvalho de Melo: "Re: [RFC PATCH 1/2] powerpc/perf: include util/util.h and remove stringify macros"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]