ASoC: Fixes for v5.5
This is mostly driver specific fixes, plus an error handling fix
in the core. There is a rather large diffstat for the stm32 SAI
driver, this is a very large but mostly mechanical update which
wraps every register access in the driver to allow a fix to the
locking which avoids circular locks, the active change is much
smaller and more reasonably sized.
In case system has multiple HDA controllers, it can happen that
same HDA codec driver is used for codecs of multiple controllers.
In this case, SOF may fail to probe the HDA driver and SOF
initialization fails.
SOF HDA code currently relies that a call to request_module() will
also run device matching logic to attach driver to the codec instance.
However if driver for another HDA controller was already loaded and it
already loaded the HDA codec driver, this breaks current logic in SOF.
In this case the request_module() SOF does becomes a no-op and HDA
Codec driver is not attached to the codec instance sitting on the HDA
bus SOF is controlling. Typical scenario would be a system with both
external and internal GPUs, with driver of the external GPU loaded
first.
Fix this by adding similar logic as is used in legacy HDA driver
where an explicit device_attach() call is done after request_module().
Also add logic to propagate errors reported by device_attach() back
to caller. This also works in the case where drivers are not built
as modules.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20200110235751.3404-8-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Any app using ALSA OSS emulation on top of SOF will fail
to error from OSS SNDCTL_DSP_SETFMT ioctl. Reported initially
as an issue with xournalpp (application using PortAudio with
an OSS backend), but applies more generally to other apps
using OSS as well.
Problem is caused by SOF PCM not supporting repeated calls
to hw_params(), without matching calls to pcm_free(). This
is however exactly what the ALSA OSS PCM code is doing when
it is handling the OSS ioctls.
The problem will lead to leaking of DSP resources and eventual
failure of DSP PCM_PARAMS IPC.
BugLink: https://github.com/thesofproject/linux/issues/1510
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20200110235751.3404-7-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Since ApolloLake, Intel platforms require signed firmware. On all
Windows platforms the default is to require the Intel production key
be used. But some platforms allow for a community key to be used,
which allows developers to compile/build their own firmware.
In the linux-firmware tree, the default intel/sof path is used for
firmwares signed for the production key, and files signed with the
community key are located in intel/sof/community.
Since we don't have an API to query which key is used on what
platforms, we have to rely on DMI-based quirks.
Developers can bypass this mechanism by setting a kernel 'fw_path'
module parameter. Additional dynamic debug traces are provided to help
debug cases where the wrong file might be used.
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20200107160840.1524-2-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
ASoC: Fixes for v5.5
More fixes that have been collected, nothing super remarkable here - the
few core fixes are mainly error handling related as are many of the
driver fixes.
dsp_box is used to keep DSP initiated messages. The value of dsp_offset
is set by the DSP with the first message, so we need a way to bootstrap
it in order to get the first message.
We do this by setting the correct default dsp_box offset which on i.MX8
is not zero.
Very interesting is why it has worked until now.
On i.MX8, DSP communicates with ARM core using a shared SDRAM memory
area. Actually, there are two shared areas:
* SDRAM0 - starting at 0x92400000, size 0x800000
* SDRAM1 - starting at 0x92C00000, size 0x800000
SDRAM0 keeps the data sections, starting with .rodata. By chance
fw_ready structure was placed at the beginning of .rodata.
dsp_box_base is defined as SDRAM0 + dsp_box_offset and it is placed
at the beginning of SDRAM1 (dsp_box_offset should be 0x800000). But
because it is zero initialized by default it points to SDRAM0 where
by chance the fw_ready was placed in the SOF firmware.
Anyhow, SOF commit 7466bee378dd811b ("clk: make freq arrays constant")
fw_ready is no longer at the beginning of SDRAM0 and everything shows
how lucky we were until now.
Fix this by properly setting the default dsp_box offset.
Fixes: 202acc565a ("ASoC: SOF: imx: Add i.MX8 HW support")
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191220170531.10423-1-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In case a HDA codec probe fails, do not raise error immediately,
but instead remove the codec from bus->codec_mask and continue
probe for other codecs.
This allows for more robust behaviour in cases where one codec
in the system is faulty. SOF driver load can still proceed with
the codecs that can be probed successfully. Probe may still
fail if suitable machine driver is not found, but in many
cases the generic HDA machine driver can operate with a subset
of codecs.
Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191218002616.7652-6-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Add a state machine for FW boot to track the
different stages of FW boot and replace the boot_complete
field with fw_state field in struct snd_sof_dev.
This will be used to determine the actions to be performed
during system suspend.
One of the main motivations for adding this change is the
fact that errors during the top-level SOF device probe cannot
be propagated and therefore suspending the SOF device normally
during system suspend could potentially run into errors.
For example, with the current flow, if the FW boot failed
for some reason and the system suspends, the SOF device
suspend could fail because the CTX_SAVE IPC would be attempted
even though the FW never really booted successfully causing it
to time out. Another scenario that the state machine fixes
is when the runtime suspend for the SOF device fails and
the DSP is powered down nevertheless, the CTX_SAVE IPC during
system suspend would timeout because the DSP is already
powered down.
Reviewed-by: Curtis Malainey <cujomalainey@chromium.org>
Reviewed-by: Daniel Baluta <daniel.baluta@nxp.com>
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191218002616.7652-2-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
If sof_machine_check() fails during driver probe, the IPC
state is not initialized and this will lead to a NULL
dereference at driver unload. Example log is as follows:
[ 1535.980630] sof-audio-pci 0000:00:1f.3: error: no matching ASoC machine driver found - aborting probe
[ 1535.980631] sof-audio-pci 0000:00:1f.3: error: failed to get machine info -19
[ 1535.980632] sof-audio-pci 0000:00:1f.3: error: sof_probe_work failed err: -19
[ 1550.798373] BUG: kernel NULL pointer dereference, address: 0000000000000008
...
[ 1550.798393] Call Trace:
[ 1550.798397] snd_sof_ipc_free+0x15/0x30 [snd_sof]
[ 1550.798399] snd_sof_device_remove+0x29/0xa0 [snd_sof]
[ 1550.798400] sof_pci_remove+0x10/0x30 [snd_sof_pci]
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191218000518.5830-2-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
ASoC: Fixes for v5.5
A collection of fixes since the merge window, mostly driver specific but
there's a few in the core that clean up fallout from the refactorings
done in the last cycle.
Currently, SOF probes machine drivers by creating a platform device
and passing the machine description as private data.
This is driven by the ACPI restrictions. Ideally, ACPI tables
should contain the description for the machine driver. This is
not possible because ACPI tables are frozen and used on multiple
OS-es (e.g Windows).
In the case of Device Tree we don't have this restriction, so we
choose to probe the machine drivers by creating a DT node as is
the standard ALSA way.
This patch makes the probing of machine drivers from SOF
core optional allowing for Device Tree platforms to decouple
the SOF core from machine driver probing.
Along with this, it also consolidates the machine driver selection
for Intel platforms by defining optional ops for selecting the machine
driver based on the ACPI match for HDA and non-HDA platforms and
setting the mach params.
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191204211556.12671-11-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Move all the audio-specific code in the core,
audio-specific logic in the top-level PM callbacks
and the core header files into a separate file
(sof-audio.*) in preparation for adding an
audio client device.
In the process of moving all structure definitions
for widget, routes, pcm's etc, the snd_sof_dev
member in all these structs is replaced with
the snd_soc_component member. Also, use the component
device instead of the snd_sof_dev device wherever
possible in the PCM component driver,
control IO functions and the topology parser as the
component device will be moved over to the client
device later on.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191204211556.12671-9-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Currently the FW filename is obtained from the ACPI matching
table when determining which machine driver to use. In
preparation for making the machine driver ACPI match optional
for Device Tree platforms and moving the machine driver selection
out of the SOF core, this patch introduces the default_fw_filename
member in struct sof_dev_desc.
Once the machine driver selection is moved out of SOF core,
the nocodec_fw_filename will become obsolete and will be removed.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20191204211556.12671-8-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
