This reverts commit 99b04f4c ("ASoC: add Component level
pcm_new/pcm_free"), which started calling the pcm_new callback for every
component in a *card* when creating a new pcm, something which does not
seem to make any sense.

This specifically led to memory leaks in systems with more than one
platform component and where DMA memory is allocated in the
platform-driver callback. For example, when both mcasp devices are being
used on an am335x board, DMA memory would be allocated twice for every
DAI link during probe.

When CONFIG_SND_VERBOSE_PROCFS was set this fortunately also led to
warnings such as:

WARNING: CPU: 0 PID: 565 at ../fs/proc/generic.c:346 proc_register+0x110/0x154
proc_dir_entry 'sub0/prealloc' already registered

Since there seems to be no users of the new component callbacks, and the
current implementation introduced a regression, let's revert the
offending commit for now.

Fixes: 99b04f4c ("ASoC: add Component level pcm_new/pcm_free")
Signed-off-by: NJohan Hovold <johan@kernel.org>
Reviewed-by: NLinus Walleij <linus.walleij@linaro.org>
Tested-by: NLinus Walleij <linus.walleij@linaro.org>
Signed-off-by: NMark Brown <broonie@kernel.org>
Cc: stable <stable@vger.kernel.org>	# 4.10

Now that all users of old copy and silence ops have been converted to
the new PCM ops, the old stuff can be retired and go away.
Reviewed-by: NTakashi Sakamoto <o-takashi@sakamocchi.jp>
Signed-off-by: NTakashi Iwai <tiwai@suse.de>

For supporting the explicit in-kernel copy of PCM buffer data, and
also for further code refactoring, three new PCM ops, copy_user,
copy_kernel and fill_silence, are introduced.  The old copy and
silence ops will be deprecated and removed later once when all callers
are converted.

The copy_kernel ops is the new one, and it's supposed to transfer the
PCM data from the given kernel buffer to the hardware ring-buffer (or
vice-versa depending on the stream direction), while the copy_user ops
is equivalent with the former copy ops, to transfer the data from the
user-space buffer.

The major difference of the new copy_* and fill_silence ops from the
previous ops is that the new ops take bytes instead of frames for size
and position arguments.  It has two merits: first, it allows the
callback implementation often simpler (just call directly memcpy() &
co), and second, it may unify the implementations of both interleaved
and non-interleaved cases, as we'll see in the later patch.

As of this stage, copy_kernel ops isn't referred yet, but only
copy_user is used.
Reviewed-by: NTakashi Sakamoto <o-takashi@sakamocchi.jp>
Acked-by: NMark Brown <broonie@kernel.org>
Signed-off-by: NTakashi Iwai <tiwai@suse.de>

No one is using snd_soc_platform_trigger().
Let's remove it.
Signed-off-by: NKuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

No existing platform is using .bespoke_trigger.
Let's remove it.
Signed-off-by: NKuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

No existing platform is using .delay.
Let's remove it.
Signed-off-by: NKuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

When multiple front-ends are using the same back-end, putting state of a
front-end to STOP state upon receiving pause command will result in backend
stream getting released by DPCM framework unintentionally. In order to
avoid backend to be released when another active front-end stream is
present, put the stream state to PAUSED state instead of STOP state.
Signed-off-by: NPatrick Lai <plai@codeaurora.org>
Signed-off-by: NMark Brown <broonie@kernel.org>

In current ALSA SoC, Platform only has pcm_new/pcm_free feature,
but it should be supported on Component level. This patch adds it.
Signed-off-by: NKuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

dpcm_state_string() returns a pointer to a string literal. Modifying a
string literal causes undefined behaviour. So make the return type of the
function const char * to make it explicit that the returned value should
not be modified.

This patch is purely cosmetic, none of the users of dpcm_state_string()
attempt to modify the returned content.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@kernel.org>

When operating the BE, we should print out the dai_link name of BE other
than FE. This is useful when analyzing the kernel log.
Signed-off-by: NDonglin Peng <pengdonglin@smartisan.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

If be_hw_param_fixup is defined for BE then it will
force the BE to a specific configuration supported
by HW. In this case don't apply symmetry.
Signed-off-by: NJeeja KP <jeeja.kp@intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Currently in situations where a normal CODEC to CODEC link follows a
DPCM DAI, an error in the following form will be logged:

ASoC: can't get [playback|capture] BE for <widget name>
ASoC: no BE found for <widget name>

This happens because all widgets in a path containing a DPCM DAI will
be passed to dpcm_add_paths, which will try to interpret the CODEC<->CODEC
as if it were a DPCM DAI, in turn causing the error.

This patch aims to resolve the described issue by stopping the DPCM graph
walk, initiated from dpcm_path_get, at the first widget associated with
a DPCM BE.
Signed-off-by: NPiotr Stankiewicz <piotrs@opensource.wolfsonmicro.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Certain situations may warrant examining DAPM paths only to a certain
arbitrary point, as opposed to always following them to the end. For
instance, when establishing a connection between a front-end DAI link
and a back-end DAI link in a DPCM path, it does not make sense to walk
the DAPM graph beyond the first widget associated with a back-end link.

This patch introduces a mechanism which lets a user of
dai_get_connected_widgets supply a function which will be called for
every node during the graph walk. When invoked, this function can
execute arbitrary logic to decide whether the walk, given a DAPM widget
and walk direction, should be terminated at that point or continued
as normal.
Signed-off-by: NPiotr Stankiewicz <piotrs@opensource.wolfsonmicro.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

While performing hw_free, DPCM checks the BE state but leaves out
the suspend state. The suspend state needs to be checked as well,
as we might be suspended and then usermode closes rather than
resuming the audio stream.

This was found by a stress testing of system with playback in
loop and killed after few seconds running in background and second
script running suspend-resume test in loop
Signed-off-by: NVinod Koul <vinod.koul@intel.com>
Acked-by: NLiam Girdwood <liam.r.girdwood@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org

Current DPCM doesn't copy dpcm->hw_params and doesn't call be_hw_params
if some FE are connected. But 2nd or later FE might want to know BE hw_params.
This patch solves this issue.
Signed-off-by: NKuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

If a device would like to use delayed suspending then PM
recommendation is to set ‘power.use_autosuspend’ flag. To allow
users to do so we need to change runtime calls in core to use
autosuspend counterparts.

For user who do not wish to use delayed suspend not setting the
device's ‘power.use_autosuspend’ flag will result in non-delayed
suspend even with these APIs which incidentally is also the default
behaviour, so only users will be impacted who opt in for this.
Signed-off-by: NSanyog Kale <sanyog.r.kale@intel.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

DPCM does not fully support symmetry attributes. soc_pcm_apply_symmetry()
is skipped in soc_pcm_open() for DPCM, without being applied elsewhere.
So HW parameters cannot be correctly limited, and user space can do
playback/capture at different rates while HW actually does not support it.
soc_pcm_params_symmetry() will return error and the second stream stops.

This patch adds soc_pcm_apply_symmetry() for FE, BE, and codec DAIs
in DPCM path that was skipped in soc_pcm_open().
Signed-off-by: NPC Liao <pc.liao@mediatek.com>
Signed-off-by: NKoro Chen <koro.chen@mediatek.com>
Acked-by: NLiam Girdwood <liam.r.girdwood@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Currently the number of DAI links is statically defined by the machine
driver at build time using an array. This makes it difficult to shrink/
grow the number of DAI links at runtime in order to reflect any changes
in topology.

We can change the DAI link array in the core to a list so that PCMs and
FE DAI links can be added and deleted at runtime to reflect changes in
use case and DSP topology. The machine driver can still register DAI links
as an array.

As the 1st step, this patch change the PCM runtime array to a list. A new
PCM runtime is added to the list when a DAI link is bound successfully.

Later patches will further implement the DAI link list.

More:
- define snd_soc_new/free_pcm_runtime() to create/free a runtime.
- define soc_add_pcm_runtime() to add a runtime to the rtd list.
- define soc_remove_pcm_runtimes() to clean up the runtime list.

- traverse the rtd list to probe the link components and dais.

- Add a field "num" to PCM runtime struct, used to specify the device
  number when creating the pcm device, and for a soc card to access
  its dai_props array.

- The following 3rd party machine/platform drivers iterate the rtd list
  to check the runtimes:
  sound/soc/intel/atom/sst-mfld-platform-pcm.c
  sound/soc/intel/boards/cht_bsw_rt5645.c
  sound/soc/intel/boards/cht_bsw_rt5672.c
  sound/soc/intel/boards/cht_bsw_max98090_ti.c
Signed-off-by: NMengdong Lin <mengdong.lin@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

During suspend/resume, there is a flow that if a driver does not support
SNDRV_PCM_INFO_RESUME, it will fail at snd_pcm_resume(), and user space
can then issue SNDRV_PCM_IOCTL_PREPARE to let audio continue to play.

However, in dpcm_be_dai_prepare() it only allows BEs to be prepared
in state SND_SOC_DPCM_STATE_HW_PARAMS or SND_SOC_DPCM_STATE_STOP.
The BE state will then stay in SND_SOC_DPCM_STATE_SUSPEND, consequently
dpcm_be_dai_shutdown() is skipped in the end of playback and
be_substream->runtime is not cleared while this runtime is actually freed
by snd_pcm_detach_substream(). If another suspend comes, a NULL pointer
dereference will happen in snd_pcm_suspend() when accessing
BE substream's runtime.
Signed-off-by: NKoro Chen <koro.chen@mediatek.com>
Acked-by: NLiam Girdwood <liam.r.girdwood@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Use the new snd_pcm_hw_constraint_single() helper function instead of
calling snd_pcm_hw_constraint_minmax() with the same value for min and max
to install a constraint that limits the possible configuration values to a
single value. Using snd_pcm_hw_constraint_single() makes the indented
result clearer and is slightly shorter.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Acked-by: NMark Brown <broonie@kernel.org>
Signed-off-by: NTakashi Iwai <tiwai@suse.de>

Add the capability to use multiple codecs on the same DAI linke where
one codec is used for playback and another one is used for capture.

Tested on a setup using an SSM2518 for playback and an ICS43432 I2S MEMS
microphone for capture.

No verification is made that the playback and capture codec setups are
compatible in terms of number of TDM slots (where applicable).
Signed-off-by: NRicard Wanderlof <ricardw@axis.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

When running dapm_dai_get_connected_widgets() currently in
is_connected_{input,output}_ep() for each widget that gets added the array
is resized and the code also loops over all existing entries to avoid
adding a widget multiple times.

The former can be avoided by collecting the widgets in a linked list and
only once we have all widgets allocate the array.

The later can be avoided by changing when the widget is added. Currently it
is added when walking the neighbor lists of a widget. Since a widget can be
neighbors with multiple other widgets it could get added twice and hence
the check is necessary. But the main body of is_connected_{input,output}_ep
is guaranteed to be only executed at most once per widget. So adding the
widget to the list at the beginning of the function automatically makes
sure that each widget gets only added once. The only difference is that
using this method the starting point itself will also end up on the list,
but it can easily be skipped when creating the array.

Overall this reduces the code size and speeds things slightly up.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@kernel.org>

In dpcm_get_be(), it looks for a BE rtd that has the DAI widget
according to current stream type. Only playback_widgets are searched
in the case of playback stream and vice versa. However, the DAI widget
itself can be playback or capture.

If the DAI widget is capture, but current stream type is playback,
dpcm_get_be() will always fail to find a rtd, print error messages,
and continue to the next DAI widget in list. We can just skip this
DAI widget to further suppress error messages. This happens in a
special case when 2 codecs are inter-connected, and the 1st codec's
"capture" widget is used to send data to the 2nd codec during "playback":

mtk-rt5650-rt5676 sound: ASoC: can't get playback BE for Sub AIF2 Capture
rt5650_rt5676 Playback: ASoC: no BE found for Sub AIF2 Capture

Add checks to continue to next DAI widget if current DAI widget's
direction does not match the stream type.
Signed-off-by: NKoro Chen <koro.chen@mediatek.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Current DPCM is caring only FE format. but it will be no sound
if FE/BE was below style, and user selects S24_LE format.

        FE: S16_LE/S24_LE
        BE: S16_LE

DPCM can rewrite the format, so basically we don't want to
constrain with the BE constraints. But sometimes it will be trouble.
This patch adds new .dpcm_merged_format on struct snd_soc_dai_link.
DPCM will use FE / BE merged format if .struct snd_soc_dai_link
has it. We can have other .dpcm_merged_xxx in the future

    .dpcm_merged_foramt
    .dpcm_merged_rate
    .dpcm_merged_chan
Signed-off-by: NKuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Tested-by: NKeita Kobayashi <keita.kobayashi.ym@renesas.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

If the registration of a debugfs directory fails this is treated as a
non-fatal error in ASoC and operation continues as normal. This means we
need to be careful and check if the parent debugfs directory exists if we
try to register a debugfs file or sub-directory. Otherwise we might end up
passing NULL for the parent and the file or directory will be registered in
the top-level debugfs directory.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@kernel.org>

Failing to register the debugfs entries is not fatal and will not affect
normal operation of the sound card. Don't abort the card registration if
soc_dpcm_debugfs_add() fails.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@kernel.org>

ALSA core with commit 257f8cce - "ALSA: pcm: Allow nonatomic trigger
operations" allows trigger ops to implemented as nonatomic. For ASoC, we can
specify this in dailinks and is updated while snd_pcm is created
Signed-off-by: NSubhransu S. Prusty <subhransu.s.prusty@intel.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>
Cc: Takashi Iwai <tiwai@suse.de>
Signed-off-by: NMark Brown <broonie@kernel.org>

sound/soc/soc-pcm.c: In function ‘soc_pcm_set_msb’:
sound/soc/soc-pcm.c:307:11: warning: unused variable ‘i’ [-Wunused-variable]

Fixes: 0e2a3751 ('ASoC: pcm: Use wildcard msbits constraints')
Signed-off-by: NTakashi Iwai <tiwai@suse.de>

Use the new wildcard msbits constraints instead of installing a constraint
for each available sample format width.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Acked-by: NMark Brown <broonie@kernel.org>
Signed-off-by: NTakashi Iwai <tiwai@suse.de>

Both codec and cpu DAI prepare print the same error message making it a bit
more difficult to grep quickly from sources. Fix this by telling it
explicitly.
Signed-off-by: NJarkko Nikula <jarkko.nikula@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Do not free BE hw if it's still used by other FE during dpcm runtime
shutdown. Otherwise the BE runtime state will be STATE_HW_FREE and
won't be updated to STATE_CLOSE when shutdown ends, because BE dai
shutdown function won't close pcm when detecting BE is still under
use. With STATE_HW_FREE, BE can't be triggered start again.

This corner case can easily appear when one BE is used by two FE,
without this patch "ASoC: dpcm: Fix race between FE/BE updates and
trigger"(ea9d0d77). One FE tries to
shutdown but it's raced against xrun on another FE. It improves the
be dai hw_free logic.
Signed-off-by: NQiao Zhou <zhouqiao@marvell.com>
Acked-by: NLiam Girdwood <liam.r.girdwood@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Skip dpcm path checking for playback or capture, if corresponding FE
doesn't support playback or capture, or currently is not ready. It
can reduce the unnecessary cost to search connected widgets.

[Tweaked comments for clarity -- broonie]
Signed-off-by: NQiao Zhou <zhouqiao@marvell.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

DPCM can update the FE/BE connection states totally asynchronously
from the FE's PCM state.  Most of FE/BE state changes are protected by
mutex, so that they won't race, but there are still some actions that
are uncovered.  For example, suppose to switch a BE while a FE's
stream is running.  This would call soc_dpcm_runtime_update(), which
sets FE's runtime_update flag, then sets up and starts BEs, and clears
FE's runtime_update flag again.

When a device emits XRUN during this operation, the PCM core triggers
snd_pcm_stop(XRUN).  Since the trigger action is an atomic ops, this
isn't blocked by the mutex, thus it kicks off DPCM's trigger action.
It eventually updates and clears FE's runtime_update flag while
soc_dpcm_runtime_update() is running concurrently, and it results in
confusion.

Usually, for avoiding such a race, we take a lock.  There is a PCM
stream lock for that purpose.  However, as already mentioned, the
trigger action is atomic, and we can't take the lock for the whole
soc_dpcm_runtime_update() or other operations that include the lengthy
jobs like hw_params or prepare.

This patch provides an alternative solution.  This adds a way to defer
the conflicting trigger callback to be executed at the end of FE/BE
state changes.  For doing it, two things are introduced:

- Each runtime_update state change of FEs is protected via PCM stream
  lock.
- The FE's trigger callback checks the runtime_update flag.  If it's
  not set, the trigger action is executed there.  If set, mark the
  pending trigger action and returns immediately.
- At the exit of runtime_update state change, it checks whether the
  pending trigger is present.  If yes, it executes the trigger action
  at this point.
Reported-and-tested-by: NQiao Zhou <zhouqiao@marvell.com>
Signed-off-by: NTakashi Iwai <tiwai@suse.de>
Acked-by: NLiam Girdwood <liam.r.girdwood@linux.intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org

We call mute for codec dai only, we should call this for cpu dai as well to
allow cpu dais (FEs) in DSPs to be muted/unmuted on shutdown/startup
Signed-off-by: NRamesh Babu <ramesh.babu@intel.com>
Signed-off-by: NSubhransu S. Prusty <subhransu.s.prusty@intel.com>
Signed-off-by: NVinod Koul <vinod.koul@intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

In the SNDRV_PCM_STREAM_CAPTURE branch in soc_pcm_apply_msb(), look at
sig_bits of the capture stream, not the playback one.

Spotted by coverity.
Signed-off-by: NDaniel Mack <daniel@zonque.org>
Signed-off-by: NMark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org

dpcm_path_get may return -ENOMEM when allocating memory for list
fails. We should not keep processing path or start up dpcm dai in
this case.
Signed-off-by: NQiao Zhou <zhouqiao@marvell.com>
Acked-by: NVinod Koul <vinod.koul@intel.com>
Signed-off-by: NMark Brown <broonie@kernel.org>

Add a function helper to factorize the hw_params code.

Suggested by Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NBenoit Cousson <bcousson@baylibre.com>
Tested-by: NLars-Peter Clausen <lars@metafoo.de>
Reviewed-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@linaro.org>

Add multicodec support in soc-pcm.c
Signed-off-by: NBenoit Cousson <bcousson@baylibre.com>
Signed-off-by: NMisael Lopez Cruz <misael.lopez@ti.com>
Signed-off-by: NFabien Parent <fparent@baylibre.com>
Tested-by: NLars-Peter Clausen <lars@metafoo.de>
Reviewed-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@linaro.org>

Refactor the function to facilitate the migration to
multiple codecs.

Fix a trailing space in the header as well.

No functional change.
Signed-off-by: NBenoit Cousson <bcousson@baylibre.com>
Signed-off-by: NMark Brown <broonie@linaro.org>

The component struct already has a name and id field which are initialized to
the same values as the same fields in the CODEC and platform structs. So remove
them from the CODEC and platform structs and used the ones from the component
struct instead.
Signed-off-by: NLars-Peter Clausen <lars@metafoo.de>
Signed-off-by: NMark Brown <broonie@linaro.org>