mirrors/linux-stable
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git synced 2024-10-05 16:37:50 +00:00
Code Issues Releases Wiki Activity

crypto: caam - fix S/G table passing page boundary

Browse source 
According to CAAM RM:
-crypto engine reads 4 S/G entries (64 bytes) at a time,
even if the S/G table has fewer entries
-it's the responsibility of the user / programmer to make sure
this HW behaviour has no side effect

The drivers do not take care of this currently, leading to IOMMU faults
when the S/G table ends close to a page boundary - since only one page
is DMA mapped, while CAAM's DMA engine accesses two pages.

Fix this by rounding up the number of allocated S/G table entries
to a multiple of 4.
Note that in case of two *contiguous* S/G tables, only the last table
might needs extra entries.

Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This commit is contained in:
Horia Geantă 2019-05-03 17:17:38 +03:00 committed by Herbert Xu
parent dcd9c76e5a
commit a5e5c13398
6 changed files with 147 additions and 37 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
drivers/crypto/caam/caamalg.c
View file

@ -1381,8 +1381,16 @@ static struct aead_edesc *aead_edesc_alloc(struct aead_request *req,
}
}
/*
* HW reads 4 S/G entries at a time; make sure the reads don't go beyond
* the end of the table by allocating more S/G entries.
*/
sec4_sg_len = mapped_src_nents > 1 ? mapped_src_nents : 0;
sec4_sg_len += mapped_dst_nents > 1 ? mapped_dst_nents : 0;
if (mapped_dst_nents > 1)
sec4_sg_len += pad_sg_nents(mapped_dst_nents);
else
sec4_sg_len = pad_sg_nents(sec4_sg_len);
sec4_sg_bytes = sec4_sg_len * sizeof(struct sec4_sg_entry);
/* allocate space for base edesc and hw desc commands, link tables */
@ -1720,7 +1728,25 @@ static struct skcipher_edesc *skcipher_edesc_alloc(struct skcipher_request *req,
else
sec4_sg_ents = mapped_src_nents + !!ivsize;
dst_sg_idx = sec4_sg_ents;
sec4_sg_ents += mapped_dst_nents > 1 ? mapped_dst_nents : 0;
/*
* HW reads 4 S/G entries at a time; make sure the reads don't go beyond
* the end of the table by allocating more S/G entries. Logic:
* if (src != dst && output S/G)
* pad output S/G, if needed
* else if (src == dst && S/G)
* overlapping S/Gs; pad one of them
* else if (input S/G) ...
* pad input S/G, if needed
*/
if (mapped_dst_nents > 1)
sec4_sg_ents += pad_sg_nents(mapped_dst_nents);
else if ((req->src == req->dst) && (mapped_src_nents > 1))
sec4_sg_ents = max(pad_sg_nents(sec4_sg_ents),
!!ivsize + pad_sg_nents(mapped_src_nents));
else
sec4_sg_ents = pad_sg_nents(sec4_sg_ents);
sec4_sg_bytes = sec4_sg_ents * sizeof(struct sec4_sg_entry);
/*

40
drivers/crypto/caam/caamalg_qi.c
View file

@ -4,7 +4,7 @@
* Based on caamalg.c
*
* Copyright 2013-2016 Freescale Semiconductor, Inc.
* Copyright 2016-2018 NXP
* Copyright 2016-2019 NXP
*/
#include "compat.h"
@ -1019,9 +1019,24 @@ static struct aead_edesc *aead_edesc_alloc(struct aead_request *req,
/*
* Create S/G table: req->assoclen, [IV,] req->src [, req->dst].
* Input is not contiguous.
* HW reads 4 S/G entries at a time; make sure the reads don't go beyond
* the end of the table by allocating more S/G entries. Logic:
* if (src != dst && output S/G)
* pad output S/G, if needed
* else if (src == dst && S/G)
* overlapping S/Gs; pad one of them
* else if (input S/G) ...
* pad input S/G, if needed
*/
qm_sg_ents = 1 + !!ivsize + mapped_src_nents +
(mapped_dst_nents > 1 ? mapped_dst_nents : 0);
qm_sg_ents = 1 + !!ivsize + mapped_src_nents;
if (mapped_dst_nents > 1)
qm_sg_ents += pad_sg_nents(mapped_dst_nents);
else if ((req->src == req->dst) && (mapped_src_nents > 1))
qm_sg_ents = max(pad_sg_nents(qm_sg_ents),
1 + !!ivsize + pad_sg_nents(mapped_src_nents));
else
qm_sg_ents = pad_sg_nents(qm_sg_ents);
sg_table = &edesc->sgt[0];
qm_sg_bytes = qm_sg_ents * sizeof(*sg_table);
if (unlikely(offsetof(struct aead_edesc, sgt) + qm_sg_bytes + ivsize >
@ -1276,7 +1291,24 @@ static struct skcipher_edesc *skcipher_edesc_alloc(struct skcipher_request *req,
qm_sg_ents = 1 + mapped_src_nents;
dst_sg_idx = qm_sg_ents;
qm_sg_ents += mapped_dst_nents > 1 ? mapped_dst_nents : 0;
/*
* HW reads 4 S/G entries at a time; make sure the reads don't go beyond
* the end of the table by allocating more S/G entries. Logic:
* if (src != dst && output S/G)
* pad output S/G, if needed
* else if (src == dst && S/G)
* overlapping S/Gs; pad one of them
* else if (input S/G) ...
* pad input S/G, if needed
*/
if (mapped_dst_nents > 1)
qm_sg_ents += pad_sg_nents(mapped_dst_nents);
else if ((req->src == req->dst) && (mapped_src_nents > 1))
qm_sg_ents = max(pad_sg_nents(qm_sg_ents),
1 + pad_sg_nents(mapped_src_nents));
else
qm_sg_ents = pad_sg_nents(qm_sg_ents);
qm_sg_bytes = qm_sg_ents * sizeof(struct qm_sg_entry);
if (unlikely(offsetof(struct skcipher_edesc, sgt) + qm_sg_bytes +
ivsize > CAAM_QI_MEMCACHE_SIZE)) {

63
drivers/crypto/caam/caamalg_qi2.c
View file

@ -1,7 +1,7 @@
// SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
/*
* Copyright 2015-2016 Freescale Semiconductor Inc.
* Copyright 2017-2018 NXP
* Copyright 2017-2019 NXP
*/
#include "compat.h"
@ -460,9 +460,25 @@ static struct aead_edesc *aead_edesc_alloc(struct aead_request *req,
/*
* Create S/G table: req->assoclen, [IV,] req->src [, req->dst].
* Input is not contiguous.
* HW reads 4 S/G entries at a time; make sure the reads don't go beyond
* the end of the table by allocating more S/G entries. Logic:
* if (src != dst && output S/G)
* pad output S/G, if needed
* else if (src == dst && S/G)
* overlapping S/Gs; pad one of them
* else if (input S/G) ...
* pad input S/G, if needed
*/
qm_sg_nents = 1 + !!ivsize + mapped_src_nents +
(mapped_dst_nents > 1 ? mapped_dst_nents : 0);
qm_sg_nents = 1 + !!ivsize + mapped_src_nents;
if (mapped_dst_nents > 1)
qm_sg_nents += pad_sg_nents(mapped_dst_nents);
else if ((req->src == req->dst) && (mapped_src_nents > 1))
qm_sg_nents = max(pad_sg_nents(qm_sg_nents),
1 + !!ivsize +
pad_sg_nents(mapped_src_nents));
else
qm_sg_nents = pad_sg_nents(qm_sg_nents);
sg_table = &edesc->sgt[0];
qm_sg_bytes = qm_sg_nents * sizeof(*sg_table);
if (unlikely(offsetof(struct aead_edesc, sgt) + qm_sg_bytes + ivsize >
@ -1086,7 +1102,24 @@ static struct skcipher_edesc *skcipher_edesc_alloc(struct skcipher_request *req)
qm_sg_ents = 1 + mapped_src_nents;
dst_sg_idx = qm_sg_ents;
qm_sg_ents += mapped_dst_nents > 1 ? mapped_dst_nents : 0;
/*
* HW reads 4 S/G entries at a time; make sure the reads don't go beyond
* the end of the table by allocating more S/G entries. Logic:
* if (src != dst && output S/G)
* pad output S/G, if needed
* else if (src == dst && S/G)
* overlapping S/Gs; pad one of them
* else if (input S/G) ...
* pad input S/G, if needed
*/
if (mapped_dst_nents > 1)
qm_sg_ents += pad_sg_nents(mapped_dst_nents);
else if ((req->src == req->dst) && (mapped_src_nents > 1))
qm_sg_ents = max(pad_sg_nents(qm_sg_ents),
1 + pad_sg_nents(mapped_src_nents));
else
qm_sg_ents = pad_sg_nents(qm_sg_ents);
qm_sg_bytes = qm_sg_ents * sizeof(struct dpaa2_sg_entry);
if (unlikely(offsetof(struct skcipher_edesc, sgt) + qm_sg_bytes +
ivsize > CAAM_QI_MEMCACHE_SIZE)) {
@ -3418,7 +3451,7 @@ static int ahash_update_ctx(struct ahash_request *req)
edesc->src_nents = src_nents;
qm_sg_src_index = 1 + (*buflen ? 1 : 0);
qm_sg_bytes = (qm_sg_src_index + mapped_nents) *
qm_sg_bytes = pad_sg_nents(qm_sg_src_index + mapped_nents) *
sizeof(*sg_table);
sg_table = &edesc->sgt[0];
@ -3503,7 +3536,7 @@ static int ahash_final_ctx(struct ahash_request *req)
gfp_t flags = (req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP) ?
GFP_KERNEL : GFP_ATOMIC;
int buflen = *current_buflen(state);
int qm_sg_bytes, qm_sg_src_index;
int qm_sg_bytes;
int digestsize = crypto_ahash_digestsize(ahash);
struct ahash_edesc *edesc;
struct dpaa2_sg_entry *sg_table;
@ -3514,8 +3547,7 @@ static int ahash_final_ctx(struct ahash_request *req)
if (!edesc)
return -ENOMEM;
qm_sg_src_index = 1 + (buflen ? 1 : 0);
qm_sg_bytes = qm_sg_src_index * sizeof(*sg_table);
qm_sg_bytes = pad_sg_nents(1 + (buflen ? 1 : 0)) * sizeof(*sg_table);
sg_table = &edesc->sgt[0];
ret = ctx_map_to_qm_sg(ctx->dev, state, ctx->ctx_len, sg_table,
@ -3527,7 +3559,7 @@ static int ahash_final_ctx(struct ahash_request *req)
if (ret)
goto unmap_ctx;
dpaa2_sg_set_final(sg_table + qm_sg_src_index - 1, true);
dpaa2_sg_set_final(sg_table + (buflen ? 1 : 0), true);
edesc->qm_sg_dma = dma_map_single(ctx->dev, sg_table, qm_sg_bytes,
DMA_TO_DEVICE);
@ -3608,7 +3640,8 @@ static int ahash_finup_ctx(struct ahash_request *req)
edesc->src_nents = src_nents;
qm_sg_src_index = 1 + (buflen ? 1 : 0);
qm_sg_bytes = (qm_sg_src_index + mapped_nents) * sizeof(*sg_table);
qm_sg_bytes = pad_sg_nents(qm_sg_src_index + mapped_nents) *
sizeof(*sg_table);
sg_table = &edesc->sgt[0];
ret = ctx_map_to_qm_sg(ctx->dev, state, ctx->ctx_len, sg_table,
@ -3705,7 +3738,7 @@ static int ahash_digest(struct ahash_request *req)
int qm_sg_bytes;
struct dpaa2_sg_entry *sg_table = &edesc->sgt[0];
qm_sg_bytes = mapped_nents * sizeof(*sg_table);
qm_sg_bytes = pad_sg_nents(mapped_nents) * sizeof(*sg_table);
sg_to_qm_sg_last(req->src, mapped_nents, sg_table, 0);
edesc->qm_sg_dma = dma_map_single(ctx->dev, sg_table,
qm_sg_bytes, DMA_TO_DEVICE);
@ -3877,7 +3910,8 @@ static int ahash_update_no_ctx(struct ahash_request *req)
}
edesc->src_nents = src_nents;
qm_sg_bytes = (1 + mapped_nents) * sizeof(*sg_table);
qm_sg_bytes = pad_sg_nents(1 + mapped_nents) *
sizeof(*sg_table);
sg_table = &edesc->sgt[0];
ret = buf_map_to_qm_sg(ctx->dev, sg_table, state);
@ -3996,7 +4030,7 @@ static int ahash_finup_no_ctx(struct ahash_request *req)
}
edesc->src_nents = src_nents;
qm_sg_bytes = (2 + mapped_nents) * sizeof(*sg_table);
qm_sg_bytes = pad_sg_nents(2 + mapped_nents) * sizeof(*sg_table);
sg_table = &edesc->sgt[0];
ret = buf_map_to_qm_sg(ctx->dev, sg_table, state);
@ -4111,7 +4145,8 @@ static int ahash_update_first(struct ahash_request *req)
int qm_sg_bytes;
sg_to_qm_sg_last(req->src, mapped_nents, sg_table, 0);
qm_sg_bytes = mapped_nents * sizeof(*sg_table);
qm_sg_bytes = pad_sg_nents(mapped_nents) *
sizeof(*sg_table);
edesc->qm_sg_dma = dma_map_single(ctx->dev, sg_table,
qm_sg_bytes,
DMA_TO_DEVICE);

33
drivers/crypto/caam/caamhash.c
View file

@ -759,7 +759,8 @@ static int ahash_edesc_add_src(struct caam_hash_ctx *ctx,
if (nents > 1 || first_sg) {
struct sec4_sg_entry *sg = edesc->sec4_sg;
unsigned int sgsize = sizeof(*sg) * (first_sg + nents);
unsigned int sgsize = sizeof(*sg) *
pad_sg_nents(first_sg + nents);
sg_to_sec4_sg_last(req->src, nents, sg + first_sg, 0);
@ -819,6 +820,8 @@ static int ahash_update_ctx(struct ahash_request *req)
}
if (to_hash) {
int pad_nents;
src_nents = sg_nents_for_len(req->src,
req->nbytes - (*next_buflen));
if (src_nents < 0) {
@ -838,15 +841,14 @@ static int ahash_update_ctx(struct ahash_request *req)
}
sec4_sg_src_index = 1 + (*buflen ? 1 : 0);
sec4_sg_bytes = (sec4_sg_src_index + mapped_nents) *
sizeof(struct sec4_sg_entry);
pad_nents = pad_sg_nents(sec4_sg_src_index + mapped_nents);
sec4_sg_bytes = pad_nents * sizeof(struct sec4_sg_entry);
/*
* allocate space for base edesc and hw desc commands,
* link tables
*/
edesc = ahash_edesc_alloc(ctx, sec4_sg_src_index + mapped_nents,
ctx->sh_desc_update,
edesc = ahash_edesc_alloc(ctx, pad_nents, ctx->sh_desc_update,
ctx->sh_desc_update_dma, flags);
if (!edesc) {
dma_unmap_sg(jrdev, req->src, src_nents, DMA_TO_DEVICE);
@ -935,18 +937,17 @@ static int ahash_final_ctx(struct ahash_request *req)
GFP_KERNEL : GFP_ATOMIC;
int buflen = *current_buflen(state);
u32 *desc;
int sec4_sg_bytes, sec4_sg_src_index;
int sec4_sg_bytes;
int digestsize = crypto_ahash_digestsize(ahash);
struct ahash_edesc *edesc;
int ret;
sec4_sg_src_index = 1 + (buflen ? 1 : 0);
sec4_sg_bytes = sec4_sg_src_index * sizeof(struct sec4_sg_entry);
sec4_sg_bytes = pad_sg_nents(1 + (buflen ? 1 : 0)) *
sizeof(struct sec4_sg_entry);
/* allocate space for base edesc and hw desc commands, link tables */
edesc = ahash_edesc_alloc(ctx, sec4_sg_src_index,
ctx->sh_desc_fin, ctx->sh_desc_fin_dma,
flags);
edesc = ahash_edesc_alloc(ctx, 4, ctx->sh_desc_fin,
ctx->sh_desc_fin_dma, flags);
if (!edesc)
return -ENOMEM;
@ -963,7 +964,7 @@ static int ahash_final_ctx(struct ahash_request *req)
if (ret)
goto unmap_ctx;
sg_to_sec4_set_last(edesc->sec4_sg + sec4_sg_src_index - 1);
sg_to_sec4_set_last(edesc->sec4_sg + (buflen ? 1 : 0));
edesc->sec4_sg_dma = dma_map_single(jrdev, edesc->sec4_sg,
sec4_sg_bytes, DMA_TO_DEVICE);
@ -1246,6 +1247,8 @@ static int ahash_update_no_ctx(struct ahash_request *req)
}
if (to_hash) {
int pad_nents;
src_nents = sg_nents_for_len(req->src,
req->nbytes - *next_buflen);
if (src_nents < 0) {
@ -1264,14 +1267,14 @@ static int ahash_update_no_ctx(struct ahash_request *req)
mapped_nents = 0;
}
sec4_sg_bytes = (1 + mapped_nents) *
sizeof(struct sec4_sg_entry);
pad_nents = pad_sg_nents(1 + mapped_nents);
sec4_sg_bytes = pad_nents * sizeof(struct sec4_sg_entry);
/*
* allocate space for base edesc and hw desc commands,
* link tables
*/
edesc = ahash_edesc_alloc(ctx, 1 + mapped_nents,
edesc = ahash_edesc_alloc(ctx, pad_nents,
ctx->sh_desc_update_first,
ctx->sh_desc_update_first_dma,
flags);

7
drivers/crypto/caam/caampkc.c
View file

@ -3,7 +3,7 @@
* caam - Freescale FSL CAAM support for Public Key Cryptography
*
* Copyright 2016 Freescale Semiconductor, Inc.
* Copyright 2018 NXP
* Copyright 2018-2019 NXP
*
* There is no Shared Descriptor for PKC so that the Job Descriptor must carry
* all the desired key parameters, input and output pointers.
@ -239,8 +239,11 @@ static struct rsa_edesc *rsa_edesc_alloc(struct akcipher_request *req,
if (src_nents > 1)
sec4_sg_len = src_nents;
if (dst_nents > 1)
sec4_sg_len += dst_nents;
sec4_sg_len += pad_sg_nents(dst_nents);
else
sec4_sg_len = pad_sg_nents(sec4_sg_len);
sec4_sg_bytes = sec4_sg_len * sizeof(struct sec4_sg_entry);

11
drivers/crypto/caam/desc_constr.h
View file

@ -3,6 +3,7 @@
* caam descriptor construction helper functions
*
* Copyright 2008-2012 Freescale Semiconductor, Inc.
* Copyright 2019 NXP
*/
#ifndef DESC_CONSTR_H
@ -37,6 +38,16 @@
extern bool caam_little_end;
/*
* HW fetches 4 S/G table entries at a time, irrespective of how many entries
* are in the table. It's SW's responsibility to make sure these accesses
* do not have side effects.
*/
static inline int pad_sg_nents(int sg_nents)
{
return ALIGN(sg_nents, 4);
}
static inline int desc_len(u32 * const desc)
{
return caam32_to_cpu(*desc) & HDR_DESCLEN_MASK;