Normal 512-byte get/set of a TLV isn't supported but we were
registering the normal get/set anyway and relying on omitting
the SNDRV_CTL_ELEM_ACCESS_[READ|WRITE] flags to prevent them
being called.
Trouble is if this gets broken in the core ALSA code - as it has
been since at least 4.14 - the standard get/set can be called
unexpectedly and corrupt memory.
There's no point providing functions that won't be called and
it's a trivial change. The benefit is that if the ALSA core gets
broken again we get a big fat immediate NULL dereference instead
of a memory corruption timebomb.
Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
Clear the buffer data structure on each trigger start such that the
buffer is in a sensible state even if the DSP itself didn't restart.
This is necessary to support voice control streams which can trigger
multiple times without reloading the firmware.
Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
For controls marked write-only don't initialize the cache from the
content of the DSP memory. We stil need the cache for any new data
that is written to this control, and we need to return something
for a read of the ALSA control because most user-side code assumes
all ALSA controls are readable. The cache is already created zero-
filled so the only change needed is to skip populating it from
DSP memory if the control isn't readable.
Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The checks for whether another region/block header could be present
are subtracting the size from the current offset. Obviously we should
instead subtract the offset from the size.
The checks for whether the region/block data fit in the file are
adding the data size to the current offset and header size, without
checking for integer overflow. Rearrange these so that overflow is
impossible.
Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Acked-by: Charles Keepax <ckeepax@opensource.cirrus.com>
Tested-by: Charles Keepax <ckeepax@opensource.cirrus.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
The list terminator is 0xbedead but the message warning if it
wasn't found was showing that 0xbeadead was expected.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The xinfo member of struct wm_coeff_ctl_ops is never used.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The shift member of struct soc_mixer_control is unsigned int.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Newer ADSP2V2 codecs include a memory protection unit that can
be set to trap illegal accesses. When enabling an ADSPV2 core we
must configure the memory region traps so that the firmware can
access its own memory.
Signed-off-by: Mayuresh Kulkarni <mkulkarni@opensource.wolfsonmicro.com>
Signed-off-by: Nikesh Oswal <Nikesh.Oswal@wolfsonmicro.com>
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Adds support for ADSP2V2 cores. Primary differences are that
they use a 32-bit register map compared to the 16-bit register
map of ADSP2V1, and there are some changes to clocking control.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
We should not be writing acknowledge controls until the firmware is
running, as in the case of preloaded firmwares the DSP memory may be
unaccessible to whilst in the preloaded state. This means a write to the
control during this time could be lost.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Volatile controls should only be accessed when the firmware is active,
currently however writes to these controls will succeed, but the data
will be lost, if the firmware is powered down. Update this behaviour such
that an error is returned the same as it is for reads.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Currently we are not disabling MEM_ENA on the error path, we should
really do this to unwind the state back to how it was. This patch adds a
clear of MEM_ENA on the error path, again there is no major issues
caused by this minor fix.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The booted and running flags should really only be set once all the
steps at that power level have been complete. Currently operations can
fail after the flags have been set, which would leave us in an
inconsistent state where the flags are set but the things expected to
reach that level have not happened. Whilst there isn't really any major
impact from this it is best to clean it up.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The recent refactoring overlooked some places which should be covered by
the pwr_lock, all code that affects or depends on the power status of
the DSP should be covered, this patch adds the missing coverage.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
As requirements to bring up audio paths are continuous getting tighter
and the DSP download to most ADSP devices happens over an external bus
it can become an important factor in the path bring up time. As such
sometimes it is a reasonable trade off to download the firmware ahead of
when it will be required and take a small hit on power consumption for
keeping the core powered up.
This "preloading" adds an additional control for each DSP core "DSPx
Preload Switch" that when set to true will power up the DSP core and
download the firmware currently selected in the "DSPx Firmware" control.
Whilst the core is preloaded the current firmware can not be changed and
the CODEC will be kept powered up and SYSCLK held on. Although future
improvements may allow the SYSCLK to be powered down as well because
the hardware only requires SYSCLK whilst the download is actually taking
place, but this is not covered in this series.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Protect against corrupt firmware files by ensuring that the length we
get for the data in a region actually lies within the available firmware
file data buffer.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
buf was allocated by kzalloc() so it should be passed to kfree()
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
We are not checking the return value from a call to wm_adsp_buffer_init
it looks like this used to be returned at the bottom of the function but
probably got missed when more error paths were added. This patch adds
the appropriate error check.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The kcontrol pointer in wm_coeff_ctl is not used now.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
We were adding firmware controls as card controls (using
snd_soc_add_codec_controls). The DSP is part of a specific codec so
we should be adding them as codec controls. Adding as codec controls
also means that if the codec has a name_prefix it will be added to
the control name, which won't happen when adding as a card control.
As that was the only use of the card pointer in struct wm_adsp it can
be removed.
For ADSP2 codecs a wm_adsp2_codec_probe() was added since the original
control handling was written, and that's the logical place to store a
pointer to the codec rather than delaying it until the codec is
powered-up.
For ADSP1 we don't use a codec_probe() stage so the codec pointer
initialization replaces the original card pointer initialization in
wm_adsp1_event().
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The control list may contain shutdown controls for firmwares that are
not currently active, attempting to write this will at best fail. To
avoid this issue we skip any control that is not active.
Fixes: commit f4f0c4c60c ("ASoC: wm_adsp: Signal firmware shutdown
through event control")
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The lookup of the base register corresponding to a control is
duplicated in read and write so factor it out into a separate
function.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch handles publishing acknowledged controls through ALSA.
These controls allow user-side to send events to the firmware and
wait for the firmware to acknowledge it.
Note that although acked controls only operate in the direction
host->firmware, and therefore they are write-only as seen from user-
side code, we have to make them readable to account for all the code
out there that assumes that ALSA controls are always readable (amixer
for example.)
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Add support for firmware controls marked SYSTEM. These are
internal to the driver-firmware interface and do not have
a user-accessible ALSA control.
Signed-off-by: Stuart Henderson <stuarth@opensource.wolfsonmicro.com>
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch factors out converting a memory region type into
a name string, mainly so that it can be used in log commands.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
If the firmware has any system event signalling controls, signal
them during DSP PRE_PMD to tell the firmware it is about to be stopped.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The booting process for the DSP is clearly separated into two parts, the
preloader brings up the core and downloads code, then the main widget
starts the code actually executing. However the shutdown sequence is all
handled with the main widget.
To allow the preloading to be run independently of the main audio bring
up it makes sense, and is generally just cleaner, for the preloader
widget to shutdown those things it initialised. This patch moves the
appropriate parts of the shutdown process into the preloader widget.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Between when we load the DSP and when it actually starts running put the
core into a lower power state where the memory is retained but nothing
is clocked.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Replace the 1ms msleep in wm_adsp2_ena with a usleep_range, as per
normal guidance on delay functions. Also tighten up the delay a little
as 1ms was quite generous.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
As part of the work to download firmware before the audio path is brought
up the DSP will be put into a low power state between downloading firmware
to the core and starting it running. This will mean that the firmware ALSA
controls are not accessible in the hardware during this period of time.
To prepare for this change we gate access to the hardware in the ALSA
control handlers on the DSP being running rather than simply booted and
move the synchronisation of the control caches out of the preloader delayed
work and into the main DAPM thread after the DSP will have been brought out
of its low power state.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Currently the wm_adsp driver has a flag that indicates the DSP is
"running", this flag is used to gate access to the hardware. However this
flag is actually set in the firmware download thread after the firmware has
been downloaded, but this is before the core is actually started running,
so really it currently indicates that the core has been booted and is
perhaps running.
This patch clearly separates out the concepts of booted (firmware is
downloaded) and running (code is executing on the DSP) within the wm_adsp
driver.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch reorders the clearing of the DMA masks to avoid potential
artefacts being introduced.
Signed-off-by: Simon Trimmer <simont@opensource.wolfsonmicro.com>
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
If the DSP is powered down whilst a compressed stream is being processed
we should treat this as a fatal error, clearly the stream is no longer
valid.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
If we encounter a fatal error on the compressed stream call the new
snd_compr_stop_error to shutdown the stream and allow the core to
inform user-space that the stream is no longer valid.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Reviewed-by: Vinod Koul <vinod.koul@intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The DSP uses an IRQ to indicate data is available on the compressed
stream. For voice trigger use-cases the first such IRQ can be considered
an indication that the user has spoken the key phrase triggering the
firmware. Provide a means for the ADSP code to communicate back to the
calling driver whether an IRQ should be considered as trigger event or
not.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
If someone powers down the DSP core (through routing changes
say) whilst a compressed record is in progress we can end up
using a freed pointer to the buffer object. When a compressed
audio stream is triggered we attach it to a buffer on a physical
DSP. This patch adds a detach of the buffer from the stream when
the stream is freed or when the DSP is powered down which avoids
the situation where we use a buffer when it is no longer valid.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Move wm_adsp_compr_attach and wm_adsp_compr_attached functions so they
will stay logically grouped with similar functions after some additional
changes.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch adds support for the arbitrary length TLV based binary
controls. This allows users to properly access controls that are
more than 512 bytes in length.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Add a function to delete and free the contents of the alg_regions list.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The patch adds a wm_adsp2_remove() function to ensure that memory
is freed when the driver is unloaded or shut down.
Signed-off-by: Richard Fitzgerald <rf@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
If we encounter an error on the DSP side whilst user-space is
waiting on the poll we should call snd_compr_fragment_elapsed,
although data is not actually available we want to wake
user-space such that the error can be propagated out
quickly. Additionally some versions of the DSP firmware are
not super consistent about actually generating an IRQ if they
encounter an error, as such we will check the DSP error status
every time we run out of available data as well, to ensure we
catch it.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Factor out the reading of the DSP error flag into its own function to
support further improvements to the code.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
All other debug messages talk about data on the compressed stream in
bytes except avail which is shown in words. To avoid confusion show
avail in bytes as well.
Signed-off-by: Charles Keepax <ckeepax@opensource.wolfsonmicro.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
