API:
- Move crypto engine callback from tfm ctx into algorithm object.
- Fix atomic sleep bug in crypto_destroy_instance.
- Move lib/mpi into lib/crypto.
Algorithms:
- Add chacha20 and poly1305 implementation for powerpc p10.
Drivers:
- Add AES skcipher and aead support to starfive.
- Add Dynamic Boost Control support to ccp.
- Add support for STM32P13 platform to stm32.
-----BEGIN PGP SIGNATURE-----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=LbcA
-----END PGP SIGNATURE-----
Merge tag 'v6.6-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Pull crypto updates from Herbert Xu:
"API:
- Move crypto engine callback from tfm ctx into algorithm object
- Fix atomic sleep bug in crypto_destroy_instance
- Move lib/mpi into lib/crypto
Algorithms:
- Add chacha20 and poly1305 implementation for powerpc p10
Drivers:
- Add AES skcipher and aead support to starfive
- Add Dynamic Boost Control support to ccp
- Add support for STM32P13 platform to stm32"
* tag 'v6.6-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (149 commits)
Revert "dt-bindings: crypto: qcom,prng: Add SM8450"
crypto: chelsio - Remove unused declarations
X.509: if signature is unsupported skip validation
crypto: qat - fix crypto capability detection for 4xxx
crypto: drivers - Explicitly include correct DT includes
crypto: engine - Remove crypto_engine_ctx
crypto: zynqmp - Use new crypto_engine_op interface
crypto: virtio - Use new crypto_engine_op interface
crypto: stm32 - Use new crypto_engine_op interface
crypto: jh7110 - Use new crypto_engine_op interface
crypto: rk3288 - Use new crypto_engine_op interface
crypto: omap - Use new crypto_engine_op interface
crypto: keembay - Use new crypto_engine_op interface
crypto: sl3516 - Use new crypto_engine_op interface
crypto: caam - Use new crypto_engine_op interface
crypto: aspeed - Remove non-standard sha512 algorithms
crypto: aspeed - Use new crypto_engine_op interface
crypto: amlogic - Use new crypto_engine_op interface
crypto: sun8i-ss - Use new crypto_engine_op interface
crypto: sun8i-ce - Use new crypto_engine_op interface
...
- Add vfio-ap support to pass-through crypto devices to secure execution
guests
- Add API ordinal 6 support to zcrypt_ep11misc device drive, which is
required to handle key generate and key derive (e.g. secure key to
protected key) correctly
- Add missing secure/has_secure sysfs files for the case where it is not
possible to figure where a system has been booted from. Existing user
space relies on that these files are always present
- Fix DCSS block device driver list corruption, caused by incorrect
error handling
- Convert virt_to_pfn() and pfn_to_virt() from defines to static inline
functions to enforce type checking
- Cleanups, improvements, and minor fixes to the kernel mapping setup
- Fix various virtual vs physical address confusions
- Move pfault code to separate file, since it has nothing to do with
regular fault handling
- Move s390 documentation to Documentation/arch/ like it has been done
for other architectures already
- Add HAVE_FUNCTION_GRAPH_RETVAL support
- Factor out the s390_hypfs filesystem and add a new config option for
it. The filesystem is deprecated and as soon as all users are gone it
can be removed some time in the not so near future
- Remove support for old CEX2 and CEX3 crypto cards from zcrypt device
driver
- Add support for user-defined certificates: receive user-defined
certificates with a diagnose call and provide them via 'cert_store'
keyring to user space
- Couple of other small fixes and improvements all over the place
-----BEGIN PGP SIGNATURE-----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=ewd9
-----END PGP SIGNATURE-----
Merge tag 's390-6.6-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Pull s390 updates from Heiko Carstens:
- Add vfio-ap support to pass-through crypto devices to secure
execution guests
- Add API ordinal 6 support to zcrypt_ep11misc device drive, which is
required to handle key generate and key derive (e.g. secure key to
protected key) correctly
- Add missing secure/has_secure sysfs files for the case where it is
not possible to figure where a system has been booted from. Existing
user space relies on that these files are always present
- Fix DCSS block device driver list corruption, caused by incorrect
error handling
- Convert virt_to_pfn() and pfn_to_virt() from defines to static inline
functions to enforce type checking
- Cleanups, improvements, and minor fixes to the kernel mapping setup
- Fix various virtual vs physical address confusions
- Move pfault code to separate file, since it has nothing to do with
regular fault handling
- Move s390 documentation to Documentation/arch/ like it has been done
for other architectures already
- Add HAVE_FUNCTION_GRAPH_RETVAL support
- Factor out the s390_hypfs filesystem and add a new config option for
it. The filesystem is deprecated and as soon as all users are gone it
can be removed some time in the not so near future
- Remove support for old CEX2 and CEX3 crypto cards from zcrypt device
driver
- Add support for user-defined certificates: receive user-defined
certificates with a diagnose call and provide them via 'cert_store'
keyring to user space
- Couple of other small fixes and improvements all over the place
* tag 's390-6.6-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (66 commits)
s390/pci: use builtin_misc_device macro to simplify the code
s390/vfio-ap: make sure nib is shared
KVM: s390: export kvm_s390_pv*_is_protected functions
s390/uv: export uv_pin_shared for direct usage
s390/vfio-ap: check for TAPQ response codes 0x35 and 0x36
s390/vfio-ap: handle queue state change in progress on reset
s390/vfio-ap: use work struct to verify queue reset
s390/vfio-ap: store entire AP queue status word with the queue object
s390/vfio-ap: remove upper limit on wait for queue reset to complete
s390/vfio-ap: allow deconfigured queue to be passed through to a guest
s390/vfio-ap: wait for response code 05 to clear on queue reset
s390/vfio-ap: clean up irq resources if possible
s390/vfio-ap: no need to check the 'E' and 'I' bits in APQSW after TAPQ
s390/ipl: refactor deprecated strncpy
s390/ipl: fix virtual vs physical address confusion
s390/zcrypt_ep11misc: support API ordinal 6 with empty pin-blob
s390/paes: fix PKEY_TYPE_EP11_AES handling for secure keyblobs
s390/pkey: fix PKEY_TYPE_EP11_AES handling for sysfs attributes
s390/pkey: fix PKEY_TYPE_EP11_AES handling in PKEY_VERIFYKEY2 IOCTL
s390/pkey: fix PKEY_TYPE_EP11_AES handling in PKEY_KBLOB2PROTK[23]
...
These declarations are not implemented now, remove them.
Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When extending the capability detection logic for 4xxx devices the
SMx algorithms were accidentally missed.
Enable these SMx capabilities by default for QAT GEN4 devices.
Check for device variants where the SMx algorithms are explicitly
disabled by the GEN4 hardware. This is indicated in fusectl1
register.
Mask out SM3 and SM4 based on a bit specific to those algorithms.
Mask out SM2 if the PKE slice is not present.
Fixes: 4b44d28c71 ("crypto: qat - extend crypto capability detection for 4xxx")
Signed-off-by: Adam Guerin <adam.guerin@intel.com>
Reviewed-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Fiona Trahe <fiona.trahe@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The DT of_device.h and of_platform.h date back to the separate
of_platform_bus_type before it as merged into the regular platform bus.
As part of that merge prepping Arm DT support 13 years ago, they
"temporarily" include each other. They also include platform_device.h
and of.h. As a result, there's a pretty much random mix of those include
files used throughout the tree. In order to detangle these headers and
replace the implicit includes with struct declarations, users need to
explicitly include the correct includes.
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Algorithms must never be added to a driver unless there is a generic
implementation. These truncated versions of sha512 slipped through.
Remove them as they are useless.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The header file jh7110-cryp uses ahash_request without including
crypto/hash.h. Fix that by adding the inclusion.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The callbacks for prepare and unprepare request in crypto_engine
is superfluous. They can be done directly from do_one_request.
Move the code into do_one_request and remove the unused callbacks.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Some newer SoCs (like SM8450) do not require a clock vote for the PRNG
to function. Make it entirely optional and rely on the bindings checker
to ensure platforms that need it, consume one.
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Reviewed-by: Bjorn Andersson <andersson@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Some configurations with gcc-12 or gcc-13 produce a warning for the source
and destination of a memcpy() in atmel_sha_hmac_compute_ipad_hash() potentially
overlapping:
In file included from include/linux/string.h:254,
from drivers/crypto/atmel-sha.c:15:
drivers/crypto/atmel-sha.c: In function 'atmel_sha_hmac_compute_ipad_hash':
include/linux/fortify-string.h:57:33: error: '__builtin_memcpy' accessing 129 or more bytes at offsets 408 and 280 overlaps 1 or more bytes at offset 408 [-Werror=restrict]
57 | #define __underlying_memcpy __builtin_memcpy
| ^
include/linux/fortify-string.h:648:9: note: in expansion of macro '__underlying_memcpy'
648 | __underlying_##op(p, q, __fortify_size); \
| ^~~~~~~~~~~~~
include/linux/fortify-string.h:693:26: note: in expansion of macro '__fortify_memcpy_chk'
693 | #define memcpy(p, q, s) __fortify_memcpy_chk(p, q, s, \
| ^~~~~~~~~~~~~~~~~~~~
drivers/crypto/atmel-sha.c:1773:9: note: in expansion of macro 'memcpy'
1773 | memcpy(hmac->opad, hmac->ipad, bs);
| ^~~~~~
The same thing happens in two more drivers that have the same logic:
drivers/crypto/chelsio/chcr_algo.c: In function 'chcr_ahash_setkey':
include/linux/fortify-string.h:57:33: error: '__builtin_memcpy' accessing 129 or more bytes at offsets 260 and 132 overlaps 1 or more bytes at offset 260 [-Werror=restrict]
drivers/crypto/bcm/cipher.c: In function 'ahash_hmac_setkey':
include/linux/fortify-string.h:57:33: error: '__builtin_memcpy' accessing between 129 and 4294967295 bytes at offsets 840 and 712 overlaps between 1 and 4294967167 bytes at offset 840 [-Werror=restrict]
I don't think it can actually happen because the size is strictly bounded
to the available block sizes, at most 128 bytes, though inlining decisions
could lead gcc to not see that.
Use the unsafe_memcpy() helper instead of memcpy(), with the only difference
being that this skips the hardening checks that produce the warning.
Suggested-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
'type' is an enum, thus cast of pointer on 64-bit compile test with W=1
causes:
exynos-rng.c:280:14: error: cast to smaller integer type 'enum exynos_prng_type' from 'const void *' [-Werror,-Wvoid-pointer-to-enum-cast]
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
There are cases when the interrupt status register (JRINTR) is non-zero,
even though:
1. An interrupt was generated, but it was masked OR
2. There was no interrupt generated at all
for the corresponding job ring.
1. The case when interrupt is masked (JRCFGR_LS[IMSK]=1b'1)
while other events have happened and are being accounted for, e.g.
-JRINTR[HALT]=2b'10 - input job ring underwent a flush of all on-going
jobs and processing of still-existing jobs (sitting in the ring) has been
halted
-JRINTR[HALT]=2b'01 - input job ring is currently undergoing a flush
-JRINTR[ENTER_FAIL]=1b'1 - SecMon / SNVS transitioned to FAIL MODE
It doesn't matter whether these events would assert the interrupt signal
or not, interrupt is anyhow masked.
2. The case when interrupt is not masked (JRCFGR_LS[IMSK]=1b'0), however
the events accounted for in JRINTR do not generate interrupts, e.g.:
-JRINTR[HALT]=2b'01
-JRINTR[ENTER_FAIL]=1b'1 and JRCFGR_MS[FAIL_MODE]=1b'0
Currently in these cases, when the JR interrupt handler is invoked (as a
consequence of JR sharing the interrupt line with other devices - e.g.
the two JRs on i.MX7ULP) it continues execution instead of returning
IRQ_NONE.
This could lead to situations like interrupt handler clearing JRINTR (and
thus also the JRINTR[HALT] field) while corresponding job ring is
suspended and then that job ring failing on resume path, due to expecting
JRINTR[HALT]=b'10 and reading instead JRINTR[HALT]=b'00.
Fix this by checking status of JRINTR[JRI] in the JR interrupt handler.
If JRINTR[JRI]=1b'0, there was no interrupt generated for this JR and
handler must return IRQ_NONE.
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Meenakshi Aggarwal <meenakshi.aggarwal@nxp.com>
Reviewed-by: Gaurav Jain <gaurav.jain@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In caam_jr_enqueue, under heavy DDR load, smp_wmb() or dma_wmb()
fail to make the input ring be updated before the CAAM starts
reading it. So, CAAM will process, again, an old descriptor address
and will put it in the output ring. This will make caam_jr_dequeue()
to fail, since this old descriptor is not in the software ring.
To fix this, use wmb() which works on the full system instead of
inner/outer shareable domains.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Signed-off-by: Meenakshi Aggarwal <meenakshi.aggarwal@nxp.com>
Reviewed-by: Gaurav Jain <gaurav.jain@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The newly added PM operations use the deprecated SIMPLE_DEV_PM_OPS() macro,
causing a warning in some configurations:
drivers/crypto/caam/ctrl.c:828:12: error: 'caam_ctrl_resume' defined but not used [-Werror=unused-function]
828 | static int caam_ctrl_resume(struct device *dev)
| ^~~~~~~~~~~~~~~~
drivers/crypto/caam/ctrl.c:818:12: error: 'caam_ctrl_suspend' defined but not used [-Werror=unused-function]
818 | static int caam_ctrl_suspend(struct device *dev)
| ^~~~~~~~~~~~~~~~~
drivers/crypto/caam/jr.c:732:12: error: 'caam_jr_resume' defined but not used [-Werror=unused-function]
732 | static int caam_jr_resume(struct device *dev)
| ^~~~~~~~~~~~~~
drivers/crypto/caam/jr.c:687:12: error: 'caam_jr_suspend' defined but not used [-Werror=unused-function]
687 | static int caam_jr_suspend(struct device *dev)
| ^~~~~~~~~~~~~~~
Use the normal DEFINE_SIMPLE_DEV_PM_OPS() variant now, and use pm_ptr() to
completely eliminate the structure in configs without CONFIG_PM.
Fixes: 322d74752c ("crypto: caam - add power management support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Meenakshi Aggarwal <meenakshi.aggarwal@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since commit ce753ad154 ("platform: finally disallow IRQ0 in
platform_get_irq() and its ilk"), there is no possible for
platform_get_irq() to return 0. Use the return value
from platform_get_irq().
Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use kfree_sensitive() instead of memset() and kfree().
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Acked-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The .remove() callback for a platform driver returns an int which makes
many driver authors wrongly assume it's possible to do error handling by
returning an error code. However the value returned is ignored (apart
from emitting a warning) and this typically results in resource leaks.
To improve here there is a quest to make the remove callback return
void. In the first step of this quest all drivers are converted to
.remove_new() which already returns void. Eventually after all drivers
are converted, .remove_new() is renamed to .remove().
Trivially convert this driver from always returning zero in the remove
callback to the void returning variant.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
stm32_hash_remove() is only called after stm32_hash_probe() succeeded. In
this case platform_set_drvdata() was called with a non-NULL data patameter.
The check for hdev being non-NULL can be dropped because hdev is never NULL
(or something bad like memory corruption happened and then the check
doesn't help any more either).
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
If pm_runtime_get() (disguised as pm_runtime_resume_and_get()) fails, this
means the clk wasn't prepared and enabled. Returning early in this case
however is wrong as then the following resource frees are skipped and this
is never catched up. So do all the cleanups but clk_disable_unprepare().
Also don't emit a warning, as stm32_hash_runtime_resume() already emitted
one.
Note that the return value of stm32_hash_remove() is mostly ignored by
the device core. The only effect of returning zero instead of an error
value is to suppress another warning in platform_remove(). So return 0
even if pm_runtime_resume_and_get() failed.
Fixes: 8b4d566de6 ("crypto: stm32/hash - Add power management support")
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
