Have rt_mutex use the rt_mutex specific scheduler helpers to avoid
recursion vs rtlock on the PI state.
[[ peterz: adapted to new names ]]
Reported-by: Crystal Wood <swood@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20230908162254.999499-6-bigeasy@linutronix.de
With PREEMPT_RT there is a rt_mutex recursion problem where
sched_submit_work() can use an rtlock (aka spinlock_t). More
specifically what happens is:
mutex_lock() /* really rt_mutex */
...
__rt_mutex_slowlock_locked()
task_blocks_on_rt_mutex()
// enqueue current task as waiter
// do PI chain walk
rt_mutex_slowlock_block()
schedule()
sched_submit_work()
...
spin_lock() /* really rtlock */
...
__rt_mutex_slowlock_locked()
task_blocks_on_rt_mutex()
// enqueue current task as waiter *AGAIN*
// *CONFUSION*
Fix this by making rt_mutex do the sched_submit_work() early, before
it enqueues itself as a waiter -- before it even knows *if* it will
wait.
[[ basically Thomas' patch but with different naming and a few asserts
added ]]
Originally-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20230908162254.999499-5-bigeasy@linutronix.de
There are currently two implementations of this basic __schedule()
loop, and there is soon to be a third.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20230908162254.999499-4-bigeasy@linutronix.de
With DEBUG_RT_MUTEXES enabled the fast-path rt_mutex_cmpxchg_acquire()
always fails and all lock operations take the slow path.
Provide a new helper inline rt_mutex_try_acquire() which maps to
rt_mutex_cmpxchg_acquire() in the non-debug case. For the debug case
it invokes rt_mutex_slowtrylock() which can acquire a non-contended
rtmutex under full debug coverage.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20230908162254.999499-3-bigeasy@linutronix.de
Even though sched_submit_work() is ran from preemptible context,
it is discouraged to have it use blocking locks due to the recursion
potential.
Enforce this.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20230908162254.999499-2-bigeasy@linutronix.de
Use FIELD_GET() to extract PCIe Negotiated Link Width field instead of
custom masking and shifting, and remove extract_width() which only
wraps that FIELD_GET().
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Link: https://lore.kernel.org/r/20230919125648.1920-2-ilpo.jarvinen@linux.intel.com
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: Dean Luick <dean.luick@cornelisnetworks.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
enable_mpesw() assumed only 2 ports are available, fix this by removing
that assumption and looping through the existing lag ports to enable multi-port
E-switch for cards with more than 2 ports.
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Add new health error syndrome to indicate that pci data poisoned error
has been received while fetching device ICM data.
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Whenever we have few destinations from Flow-table type we need to put
the one that goes to the wire to be the last one.
We are using FW in order to get iterator, the FW uses RX for the first
destinations and TX for the last destination, if we want the packet to
be directed to the wire it should be done in the TX path and not in the
RX.
The code now checks if the FT is directed to the wire and if so puts it
as the last destination.
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The driver should not allow rule that forward to more than one FT in TX
flow unless there is a specific support from the FW.
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
In order to have mcast offloads the driver needs the following:
It should know if that mcast comes from wire port, in addition the flow
should not be marked as any specific source, that way it will give the
flexibility for the driver not to be depended on the way iterator
implemented in the FW.
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
As matchall filter uses TSAR (Transmit Scheduling Arbiter) for rate
limit, the rate of police action should not be over the port's max
link speed, or the maximum aggregated speed of both ports if LAG is
configured.
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When LAG is configured, functions (PF,VF,SF) can utilize the maximum
aggregated link speed for transmission. Currently the aggregated link
speed is not considered.
Hence, improve it to use the aggregated link speed by referring to the
physical port's upper bonding device when LAG is configured.
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
table->max_sfs is initialized in mlx5_sf_dev_table_create() and only
used to check for 0 in mlx5_sf_dev_add(). mlx5_sf_dev_add() is called
either from mlx5_sf_dev_state_change_handler() or
mlx5_sf_dev_add_active_work(). Both ensure max SF count is not 0,
using mlx5_sf_max_functions() helper before calling mlx5_sf_dev_add().
So remove the redundant check and no longer used max_sfs field.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Historically, the SF table reference counting was present in order to
protect parallel executions of devlink ops. However, since currently
this is protected with devlink instance lock, the SF table reference
counting is no longer needed. Remove it entirely.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Don't call the same functions for SF deletion on multiple places.
Instead, introduce a helper mlx5_sf_del() and move the code there.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The function does not do deactivation, but it deletes all SFs instead.
Rename accordingly.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Instead of taking lock and calling mlx5_sf_dealloc(), move the lock
taking into mlx5_sf_dealloc(). The other caller of mlx5_sf_dealloc()
does not need it now, but will need it after a follow-up patch removing
the table reference counting.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
No need to lookup for sf by a port index. Convert the xarray to have
function id as an index and optimize the remaining function id
based lookup.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Benefit from the fact that struct devlink_port is eventually embedded
in struct mlx5_sf and use container_of() macro to get it instead of the
xarray lookup in every devlink port op.
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Before every call of mlx5_sf_dealloc(), there is a call to
mlx5_sf_id_erase(). So move it to the beginning of mlx5_sf_dealloc().
Also remove redundant mlx5_sf_id_erase() call from mlx5_sf_free()
as it is called only from mlx5_sf_dealloc().
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The current links of ABI can not be found for some time, let us fix
the broken links.
By the way, the latest and official ABI documentation releases are
available at https://github.com/loongson/la-abi-specs, but there are
no Chinese and pdf versions for now, so just do the minimal changes
to update the links so that they can be found, hope there are stable
links in the future.
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
The current links of ABI can not be found for some time, let us fix
the broken links.
By the way, the latest and official ABI documentation releases are
available at https://github.com/loongson/la-abi-specs, but there are
no Chinese and pdf versions for now, so just do the minimal changes
to update the links so that they can be found, hope there are stable
links in the future.
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
As Linus suggested, kasan_mem_to_shadow()/kasan_shadow_to_mem() are not
performance-critical and too big to inline. This is simply wrong so just
define them out-of-line.
If they really need to be inlined in future, such as the objtool / SMAP
issue for X86, we should mark them __always_inline.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
As Linus suggested, __HAVE_ARCH_XYZ is "stupid" and "having historical
uses of it doesn't make it good". So migrate __HAVE_ARCH_SHADOW_MAP to
separate macros named after the respective functions.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: WANG Xuerui <git@xen0n.name>
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
The initial aim is to silence the following objtool warning:
arch/loongarch/kernel/relocate_kernel.o: warning: objtool: relocate_new_kernel+0x74: unreachable instruction
There are two adjacent "b" instructions, the second one is unreachable,
it is dead code, just remove it.
Co-developed-by: Jinyang He <hejinyang@loongson.cn>
Signed-off-by: Jinyang He <hejinyang@loongson.cn>
Co-developed-by: Youling Tang <tangyouling@loongson.cn>
Signed-off-by: Youling Tang <tangyouling@loongson.cn>
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
Use _UL() and _ULL() that are provided by const.h.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
Since commit 0a6b58c5cd ("lockdep: fix static memory detection even
more") the lockdep code uses is_kernel_core_data(), is_kernel_rodata()
and init_section_contains() to verify if a lock is located inside a
kernel static data section.
This change triggers a failure on LoongArch, for which the vmlinux.lds.S
script misses to put the locks (as part of in the .data.rel symbols)
into the Linux data section.
This patch fixes the lockdep problem by moving *(.data.rel*) symbols
into the kernel data section (from _sdata to _edata).
Additionally, move other wrongly assigned symbols too:
- altinstructions into the _initdata section,
- PLT symbols behind the read-only section, and
- *(.la_abs) into the data section.
Cc: stable <stable@kernel.org> # v6.4+
Fixes: 0a6b58c5cd ("lockdep: fix static memory detection even more")
Reported-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
x86 optimized crypto modules are built as modules rather than build-in and
they are not loaded when the crypto API is initialized, resulting in the
generic builtin module (sha1-generic) being used instead.
It was discovered when creating a sha1/sha256 checksum of a 2Gb file by
using kcapi-tools because it would take significantly longer than creating
a sha512 checksum of the same file. trace-cmd showed that for sha1/256 the
generic module was used, whereas for sha512 the optimized module was used
instead.
Add module aliases() for these x86 optimized crypto modules based on CPU
feature bits so udev gets a chance to load them later in the boot
process. This resulted in ~3x decrease in the real-time execution of
kcapi-dsg.
Fix is inspired from commit
aa031b8f70 ("crypto: x86/sha512 - load based on CPU features")
where a similar fix was done for sha512.
Cc: stable@vger.kernel.org # 5.15+
Suggested-by: Dimitri John Ledkov <dimitri.ledkov@canonical.com>
Suggested-by: Julian Andres Klode <julian.klode@canonical.com>
Signed-off-by: Roxana Nicolescu <roxana.nicolescu@canonical.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When sec_aead_mac_init returns an error code, sec_cipher_map
will exit abnormally, the hardware sgl should be unmmaped.
Signed-off-by: Wenkai Lin <linwenkai6@hisilicon.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
QAT GEN4 devices support chained compression operations. These
allow, with a single request to firmware, to hash then compress
data.
Extend the configuration to enable such mode. The cfg_services
operations in sysfs are extended to allow the string "dcc". When
selected, the driver downloads to the device both the symmetric
crypto and the compression firmware images and sends an admin message
to firmware which enables `chained` operations.
In addition, it sets the device's capabilities as the combination
of compression and symmetric crypto capabilities, while excluding
the ICP_ACCEL_CAPABILITIES_CRYPTO_SYMMETRIC bit to indicate
that in this mode, symmetric crypto instances are not supported.
When "dcc" is enabled, the device will handle compression requests
as if the "dc" configuration is loaded ("dcc" is a variation of "dc")
and the driver will register the acomp algorithms.
As for the other extended configurations, "dcc" is only available for
qat_4xxx devices and the chaining service will be only accessible from
user space.
Signed-off-by: Adam Guerin <adam.guerin@intel.com>
Reviewed-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The data structure that associates a service id with its name is
replicated across the driver.
Remove duplication by moving this data structure to a new include file,
adf_cfg_services.h in order to have consistency across the drivers.
Note that the data structure is re-instantiated every time the new
include is added to a compilation unit.
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The function adf_dev_init(), through the subsystem qat_compression,
populates the list of list of compression instances
accel_dev->compression_list. If the list of instances is not empty,
the function adf_dev_start() will then call qat_compression_registers()
register the compression algorithms into the crypto framework.
If any of the functions in adf_dev_start() fail, the caller of such
function, in the error path calls adf_dev_down() which in turn call
adf_dev_stop() and adf_dev_shutdown(), see for example the function
state_store in adf_sriov.c.
However, if the registration of compression algorithms is not done,
adf_dev_stop() will try to unregister the algorithms regardless.
This might cause the counter active_devs in qat_compression.c to get
to a negative value.
Add a new state, ADF_STATUS_COMPRESSION_ALGS_REGISTERED, which tracks
if the compression algorithms are registered into the crypto framework.
Then use this to unregister the algorithms if such flag is set. This
ensures that the compression algorithms are only unregistered if
previously registered.
Fixes: 1198ae56c9 ("crypto: qat - expose deflate through acomp api for QAT GEN2")
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Adam Guerin <adam.guerin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The function adf_dev_init(), through the subsystem qat_crypto, populates
the list of list of crypto instances accel_dev->crypto_list.
If the list of instances is not empty, the function adf_dev_start() will
then call qat_algs_registers() and qat_asym_algs_register() to register
the crypto algorithms into the crypto framework.
If any of the functions in adf_dev_start() fail, the caller of such
function, in the error path calls adf_dev_down() which in turn call
adf_dev_stop() and adf_dev_shutdown(), see for example the function
state_store in adf_sriov.c.
However, if the registration of crypto algorithms is not done,
adf_dev_stop() will try to unregister the algorithms regardless.
This might cause the counter active_devs in qat_algs.c and
qat_asym_algs.c to get to a negative value.
Add a new state, ADF_STATUS_CRYPTO_ALGS_REGISTERED, which tracks if the
crypto algorithms are registered into the crypto framework. Then use
this to unregister the algorithms if such flag is set. This ensures that
the crypto algorithms are only unregistered if previously registered.
Fixes: d8cba25d2c ("crypto: qat - Intel(R) QAT driver framework")
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Adam Guerin <adam.guerin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
If the device is already in the up state, a subsequent write of `up` to
the sysfs attribute /sys/bus/pci/devices/<BDF>/qat/state brings the
device down.
Fix this behaviour by ignoring subsequent `up` commands if the device is
already in the up state.
Fixes: 1bdc85550a ("crypto: qat - fix concurrency issue when device state changes")
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Adam Guerin <adam.guerin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Do not shadow the return code from adf_dev_down() in the error path of
the DEV_DOWN command.
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Adam Guerin <adam.guerin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Commit 1bdc85550a ("crypto: qat - fix concurrency issue when device
state changes") introduced the function adf_dev_down() which wraps the
functions adf_dev_stop() and adf_dev_shutdown().
In a subsequent change, the sequence adf_dev_stop() followed by
adf_dev_shutdown() was then replaced across the driver with just a call
to the function adf_dev_down().
The functions adf_dev_stop() and adf_dev_shutdown() are called in error
paths to stop the accelerator and free up resources and can be called
even if the counterparts adf_dev_init() and adf_dev_start() did not
complete successfully.
However, the implementation of adf_dev_down() prevents the stop/shutdown
sequence if the device is found already down.
For example, if adf_dev_init() fails, the device status is not set as
started and therefore a call to adf_dev_down() won't be calling
adf_dev_shutdown() to undo what adf_dev_init() did.
Do not check if a device is started in adf_dev_down() but do the
equivalent check in adf_sysfs.c when handling a DEV_DOWN command from
the user.
Fixes: 2b60f79c7b ("crypto: qat - replace state machine calls")
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Adam Guerin <adam.guerin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Remove the support of zlib-deflate and gzip.
Signed-off-by: Yang Shen <shenyang39@huawei.com>
Reviewed-by: Longfang Liu <liulongfang@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add the deflate algorithm support for hisilicon zip hardware.
Signed-off-by: Yang Shen <shenyang39@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch adds two different implementations of ECB. First of
all an lskcipher wrapper around existing ciphers is introduced as
a temporary transition aid.
Secondly a permanent lskcipher template is also added. It's simply
a wrapper around the underlying lskcipher algorithm.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
As an aid to the transition from cipher algorithm implementations
to lskcipher, add a temporary wrapper when creating simple lskcipher
templates by using ecb(X) instead of X if an lskcipher implementation
of X cannot be found.
This can be reverted once all cipher implementations have switched
over to lskcipher.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add a new API type lskcipher designed for taking straight kernel
pointers instead of SG lists. Its relationship to skcipher will
be analogous to that between shash and ahash.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Move the macro CRYPTO_ALG_TYPE_AHASH_MASK out of linux/crypto.h
and into crypto/ahash.c so that it's not visible to users of the
Crypto API.
Also remove the unused CRYPTO_ALG_TYPE_HASH_MASK macro.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop using the obsolete crypto_has_alg helper that is type-agnostic.
Instead use the type-specific helpers such as the newly added
crypto_has_aead.
This means that changes in the underlying type/mask values won't
affect IPsec.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add the helper crypto_has_aead. This is meant to replace the
existing use of crypto_has_alg to locate AEAD algorithms.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>