When compiling with CONFIG_DEBUG_ATOMIC_SLEEP=y the mxs-dcp driver
prints warnings such as:
WARNING: CPU: 0 PID: 120 at kernel/sched/core.c:7736 __might_sleep+0x98/0x9c
do not call blocking ops when !TASK_RUNNING; state=1 set at [<8081978c>] dcp_chan_thread_sha+0x3c/0x2ec
The problem is that blocking ops will manipulate current->state
themselves so it is not allowed to call them between
set_current_state(TASK_INTERRUPTIBLE) and schedule().
Fix this by converting the per-chan mutex to a spinlock (it only
protects tiny list ops anyway) and rearranging the wait logic so that
callbacks are called current->state as TASK_RUNNING. Those callbacks
will indeed call blocking ops themselves so this is required.
Cc: <stable@vger.kernel.org>
Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Crypto framework requires export/import in async hash. If driver do not
implement them, wrapper functions in framework will be used, and it will
cause error during ahash alg registration (unless one disables crypto
internal tests). To make change in framework and expose this requirement,
I will remove wrappers from crypto/ahash.c , but this can broke code which
depends on them. Add empty hash export and import, with the same behaviour
as in framework and expose this directly in driver. This can also prevent
OOPS when config option in Cryptographic API 'Disable run-time self tests'
will be enabled.
Signed-off-by: Kamil Konieczny <k.konieczny@partner.samsung.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Print error message on platform_get_irq failure before return.
Signed-off-by: Gustavo A. R. Silva <garsilva@embeddedor.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
return type of wait_for_completion_timeout is unsigned long not int, this
patch changes the type of ret from int to unsigned long.
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Optimize the hashing operation in the MXS-DCP by doing two adjustments:
1) Given that the output buffer for the hash is now always correctly aligned,
we can just use the buffer for the DCP DMA to store the resulting hash.
We thus get rid of one copying of data. Moreover, we remove an entry from
dcp_coherent_block{} and thus lower the memory footprint of the driver.
2) We map the output buffer for the hash for DMA only in case we will output
the hash, not always, as it was now.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Shawn Guo <shawn.guo@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The DCP needs the bounce buffers, DMA descriptors and result buffers aligned
to 64 bytes (yet another hardware limitation). Make sure they are aligned by
properly aligning the structure which contains them during allocation.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Shawn Guo <shawn.guo@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
We should test the error case for each platform_get_irq() assignment and
propagate the error accordingly.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Marek Vasut <marex@denx.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
stmp_reset_block() may fail, so check its return value and propagate it in the
case of error.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Marek Vasut <marex@denx.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Using devm_kzalloc() can make the code cleaner.
While at it, remove the devm_kzalloc error message as there is standard OOM
message done by the core.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Acked-by: Marek Vasut <marex@denx.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Move the AES operation type and mode from async crypto context to
crypto request context. This allows for recycling of the async crypto
context for different kinds of operations.
I found this problem when I used dm-crypt, which uses the same async
crypto context (actx) for both encryption and decryption requests.
Since the requests are enqueued into the processing queue, immediatelly
storing the type of operation into async crypto context (actx) caused
corruption of this information when encryption and decryption operations
followed imediatelly one after the other. When the first operation was
dequeued, the second operation was already enqueued and overwritten the
type of operation in actx, thus causing incorrect result of the first
operation.
Fix this problem by storing the type of operation into the crypto request
context.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Shawn Guo <shawn.guo@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fixes the following sparse warning:
drivers/crypto/mxs-dcp.c:103:1: warning:
symbol 'global_mutex' was not declared. Should it be static?
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: Marek Vasut <marex@denx.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add support for the MXS DCP block. The driver currently supports
SHA-1/SHA-256 hashing and AES-128 CBC/ECB modes. The non-standard
CRC32 is not yet supported.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Shawn Guo <shawn.guo@linaro.org>
Cc: devicetree@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
