In the case where multiple peripherals are attached on the same link,
it's possible that they are in different pm_runtime states.
The device_for_each_child() loop to resume all devices before a system
suspend would not work if one peripheral was active and others
suspended. pm_runtime_resume() returns 1 in the former case, which is
taken as a error. As a result, a pm_runtime suspended device might be
skipped if the first device was active.
This patch changes the behavior of the helper function to only return
zero or a negative error. A Fixes tag is not provided since there are
no existing configurations on Intel platforms with different types of
devices on the same link. Amplifiers may be used on the same link, but
they are used by the same dailink so their pm_runtime state is always
matching. This assumption may not be true in the future, so we should
improve the behavior and align with AMD.
Reported-by: Mukunda,Vijendar <vijendar.mukunda@amd.com>
Link: https://lore.kernel.org/lkml/4cbbff8a-c596-e9cc-a6cf-6f8b66607505@amd.com/
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Rander Wang <rander.wang@intel.com>
Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://lore.kernel.org/r/20230323025228.1537107-1-yung-chuan.liao@linux.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
This reverts commit
443a98e649 ("soundwire: bus: use pm_runtime_resume_and_get()")
Change calls to pm_runtime_resume_and_get() back to pm_runtime_get_sync().
This fixes a usage count underrun caused by doing a pm_runtime_put() even
though pm_runtime_resume_and_get() returned an error.
The three affected functions ignore -EACCES error from trying to get
pm_runtime, and carry on, including a put at the end of the function.
But pm_runtime_resume_and_get() does not increment the usage count if it
returns an error. So in the -EACCES case you must not call
pm_runtime_put().
The documentation for pm_runtime_get_sync() says:
"Consider using pm_runtime_resume_and_get() ... as this is likely to
result in cleaner code."
In this case I don't think it results in cleaner code because the
pm_runtime_put() at the end of the function would have to be conditional on
the return value from pm_runtime_resume_and_get() at the top of the
function.
pm_runtime_get_sync() doesn't have this problem because it always
increments the count, so always needs a put. The code can just flow through
and do the pm_runtime_put() unconditionally.
Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20230406134640.8582-1-rf@opensource.cirrus.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
The existing code copies the hw_params pointer and reuses it later in
.prepare, specifically to re-initialize the ALH DMA channel
information that's lost in suspend-resume cycles.
This is not needed, we can directly access the information from the
substream/rtd - as done for the HDAudio DAIs in
sound/soc/sof/intel/hda-dai.c
In addition, using the saved pointer causes the suspend-resume test
cases to fail on specific platforms, depending on which version of GCC
is used. Péter Ujfalusi and I have spent long hours to root-cause this
problem that was reported by the Intel CI first with 6.2-rc1 and again
v6.3-rc1. In the latter case we were lucky that the problem was 100%
reproducible on local test devices, and found out that adding a
dev_dbg() or adding a call to usleep_range() just before accessing the
saved pointer "fixed" the issue. With errors appearing just by
changing the compiler version or minor changes in the code generated,
clearly we have a memory management Heisenbug.
The root-cause seems to be that the hw_params pointer is not
persistent. The soc-pcm code allocates the hw_params structure on the
stack, and passes it to the BE dailink hw_params and DAIs
hw_params. Saving such a pointer and reusing it later during the
.prepare stage cannot possibly work reliably, it's broken-by-design
since v5.10. It's astonishing that the problem was not seen earlier.
This simple fix will have to be back-ported to -stable, due to changes
to avoid the use of the get/set_dmadata routines this patch will only
apply on kernels older than v6.1.
Fixes: a5a0239c27 ("soundwire: intel: reinitialize IP+DSP in .prepare(), but only when resuming")
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com>
Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://lore.kernel.org/r/20230321022642.1426611-1-yung-chuan.liao@linux.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Currently issuing a sdw_nread/nwrite_no_pm across a page boundary
will silently fail to write correctly as nothing updates the page
registers, meaning the same page of the chip will get rewritten
with each successive page of data.
As the sdw_msg structure contains page information it seems
reasonable that a single sdw_msg should always be within one
page. It is also mostly simpler to handle the paging at the
bus level rather than each master having to handle it in their
xfer_msg callback.
As such add handling to the bus code to split up a transfer into
multiple sdw_msg's when they go across page boundaries.
Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com>
Link: https://lore.kernel.org/r/20230322164948.566962-3-ckeepax@opensource.cirrus.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Replace the call to sdw_ch_mask_to_ch() with a call to hweight32().
sdw_ch_mask_to_ch() is counting the number of set bits. The hweight()
family of functions already do this, and they have an advantage of
using a bit-counting instruction if it is available on the target CPU.
This also fixes a potential infinite loop bug in the implementation of
sdw_ch_mask_to_ch().
Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com>
Link: https://lore.kernel.org/r/20230315145051.2299822-1-rf@opensource.cirrus.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
There are two issues related to the number of ports coming from
Devicetree when exceeding in total QCOM_SDW_MAX_PORTS. Both lead to
incorrect memory accesses:
1. With DTS having too big value of input or output ports, the driver,
when copying port parameters from local/stack arrays into 'pconfig'
array in 'struct qcom_swrm_ctrl', will iterate over their sizes.
2. If DTS also has too many parameters for these ports (e.g.
qcom,ports-sinterval-low), the driver will overflow buffers on the
stack when reading these properties from DTS.
Add a sanity check so incorrect DTS will not cause kernel memory
corruption.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Reviewed-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Link: https://lore.kernel.org/r/20230222144412.237832-2-krzysztof.kozlowski@linaro.org
Signed-off-by: Vinod Koul <vkoul@kernel.org>
The latest Cadence IP splits some of the existing registers into two,
separated by a fixed offset. The bitfields themselves remain at the
same position, so we can use new helpers to dynamically add the fixed
offset.
For example, the existing MCP_CONFIG is now split in two with
MCP_CONFIG and IP_MCP_CONFIG (the naming comes directly from the
design document).
This patch adds helpers to access registers with the IP_ prefix. The
addition of the 'ip' prefix for helpers, registers and bitfields is
intentional to help reviewers spot any mistake.
For existing solutions, the offset is exactly zero so there's no
functional change - the MCP_CONFIG and IP_MCP_CONFIG are aliased to
the same address.
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Rander Wang <rander.wang@intel.com>
Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://lore.kernel.org/r/20230314015410.487311-13-yung-chuan.liao@linux.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Now that the bus start/stop/clock_stop sequences use the ops, we can
move them to a different file to reuse them.
Note that we could in theory remove the abstraction for all those
sequences and directly call the functions in intel_auxdevice.c. To
allow for more flexibility and have means to special-case new
platforms, we decided to keep the abstraction. If in time it becomes
clear there is no benefit the abstraction will be simplified.
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Rander Wang <rander.wang@intel.com>
Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://lore.kernel.org/r/20230314015410.487311-9-yung-chuan.liao@linux.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Pull soundwire updates from Vinod Koul:
"This is a small update which features a bit of core changes and driver
updates in Intel and cadence driver.
Core:
- sdw_transfer_defer() API change to drop an argument
- Reset page address rework
- Export sdw_nwrite_no_pm and sdw_nread_no_pm APIs
Drivers:
- Cadence and related intel driver updates for FIFO handling and low
level msg transfers"
* tag 'soundwire-6.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vkoul/soundwire:
soundwire: cadence: further simplify low-level xfer_msg_defer() callback
soundwire: cadence: use directly bus sdw_defer structure
soundwire: bus: remove sdw_defer argument in sdw_transfer_defer()
soundwire: stream: use consistent pattern for freeing buffers
soundwire: bus: Remove unused reset_page_addr() callback
soundwire: bus: Don't zero page registers after every transaction
soundwire: bus_type: Avoid lockdep assert in sdw_drv_probe()
soundwire: stream: Move remaining register accesses over to no_pm
soundwire: debugfs: Switch to sdw_read_no_pm
soundwire: Provide build stubs for common functions
soundwire: bus: export sdw_nwrite_no_pm and sdw_nread_no_pm functions
soundwire: cadence: remove unused sdw_cdns_master_ops declaration
soundwire: enable optional clock registers for SoundWire 1.2 devices
ASoC/soundwire: remove is_sdca boolean property
soundwire: cadence: Drain the RX FIFO after an IO timeout
soundwire: cadence: Remove wasted space in response_buf
soundwire: cadence: Don't overflow the command FIFOs
soundwire: intel: remove DAI startup/shutdown
Pull driver core updates from Greg KH:
"Here is the large set of driver core changes for 6.3-rc1.
There's a lot of changes this development cycle, most of the work
falls into two different categories:
- fw_devlink fixes and updates. This has gone through numerous review
cycles and lots of review and testing by lots of different devices.
Hopefully all should be good now, and Saravana will be keeping a
watch for any potential regression on odd embedded systems.
- driver core changes to work to make struct bus_type able to be
moved into read-only memory (i.e. const) The recent work with Rust
has pointed out a number of areas in the driver core where we are
passing around and working with structures that really do not have
to be dynamic at all, and they should be able to be read-only
making things safer overall. This is the contuation of that work
(started last release with kobject changes) in moving struct
bus_type to be constant. We didn't quite make it for this release,
but the remaining patches will be finished up for the release after
this one, but the groundwork has been laid for this effort.
Other than that we have in here:
- debugfs memory leak fixes in some subsystems
- error path cleanups and fixes for some never-able-to-be-hit
codepaths.
- cacheinfo rework and fixes
- Other tiny fixes, full details are in the shortlog
All of these have been in linux-next for a while with no reported
problems"
[ Geert Uytterhoeven points out that that last sentence isn't true, and
that there's a pending report that has a fix that is queued up - Linus ]
* tag 'driver-core-6.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core: (124 commits)
debugfs: drop inline constant formatting for ERR_PTR(-ERROR)
OPP: fix error checking in opp_migrate_dentry()
debugfs: update comment of debugfs_rename()
i3c: fix device.h kernel-doc warnings
dma-mapping: no need to pass a bus_type into get_arch_dma_ops()
driver core: class: move EXPORT_SYMBOL_GPL() lines to the correct place
Revert "driver core: add error handling for devtmpfs_create_node()"
Revert "devtmpfs: add debug info to handle()"
Revert "devtmpfs: remove return value of devtmpfs_delete_node()"
driver core: cpu: don't hand-override the uevent bus_type callback.
devtmpfs: remove return value of devtmpfs_delete_node()
devtmpfs: add debug info to handle()
driver core: add error handling for devtmpfs_create_node()
driver core: bus: update my copyright notice
driver core: bus: add bus_get_dev_root() function
driver core: bus: constify bus_unregister()
driver core: bus: constify some internal functions
driver core: bus: constify bus_get_kset()
driver core: bus: constify bus_register/unregister_notifier()
driver core: remove private pointer from struct bus_type
...
Currently, port_prep callback only has commands for PRE_PREP, PREP,
and POST_PREP, which doesn't directly say whether this is for a
prepare or deprepare call. Extend the command list enum to say
whether the call is for prepare or deprepare aswell.
Also remove SDW_OPS_PORT_PREP from sdw_port_prep_ops as this is unused,
and update this enum to be simpler and more consistent with enum
sdw_clk_stop_type.
Note: Currently, the only users of SDW_OPS_PORT_POST_PREP are codec
drivers sound/soc/codecs/wsa881x.c and sound/soc/codecs/wsa883x.c, both
of which seem to assume that POST_PREP only occurs after a prepare,
even though it would also have occurred after a deprepare. Since it
doesn't make sense to mark the port prepared after a deprepare, changing
the enum to separate PORT_DEPREP from PORT_PREP should make the check
for PORT_PREP in those drivers be more logical.
Signed-off-by: Stefan Binding <sbinding@opensource.cirrus.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Acked-By: Vinod Koul <vkoul@kernel.org>
Link: https://lore.kernel.org/r/20230127165111.3010960-2-sbinding@opensource.cirrus.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Zeroing the page registers at the end of every paged transaction is just
overhead (40% overhead on a 1-register access, 25% on a 4-register
transaction). According to the spec a peripheral that supports paging
should only use the values in the page registers if the address is paged
(address bit 15 set). The core SoundWire code always writes the page
registers at the start of a paged transaction so there will never be a
transaction that uses the stale values from a previous paged transaction.
For peripherals that need large amounts of data to be transferred, for
example firmware or filter coefficients, the overhead of page register
zeroing can become quite significant.
Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20230123164949.245898-2-rf@opensource.cirrus.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
There is no need to play with the runtime reference everytime a register
is accessed. All the remaining "pm" style register accesses trace back
to 4 functions:
sdw_prepare_stream
sdw_deprepare_stream
sdw_enable_stream
sdw_disable_stream
Any sensible implementation will need to hold a runtime reference
across all those functions, it makes no sense to be allowing the
device/bus to suspend whilst streams are being prepared/enabled. And
certainly in the case of the all existing users, they all call these
functions from hw_params/prepare/trigger/hw_free callbacks in ALSA,
which will have already runtime resumed all the audio devices
associated during the open callback.
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com>
Link: https://lore.kernel.org/r/20221125142028.1118618-5-ckeepax@opensource.cirrus.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
