ASoC: Fixes for v5.6

A few fixes sent in since the merge window, none of them with global
impact but all important for the users they affect.

Some code in HD-audio driver calls snprintf() in a loop and still
expects that the return value were actually written size, while
snprintf() returns the expected would-be length instead.  When the
given buffer limit were small, this leads to a buffer overflow.

Use scnprintf() for addressing those issues.  It returns the actually
written size unlike snprintf().

Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200218091409.27162-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>

snd_hdac_ext_bus_link_get() does not work correctly in case
there are multiple codecs on the bus. It unconditionally
resets the bus->codec_mask value. As per documentation in
hdaudio.h and existing use in client code, this field should
be used to store bit flag of detected codecs on the bus.

By overwriting value of the codec_mask, information on all
detected codecs is lost. No current user of hdac is impacted,
but use of bus->codec_mask is planned in future patches
for SOF.

Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Takashi Iwai <tiwai@suse.de>
Link: https://lore.kernel.org/r/20200206200223.7715-1-kai.vehmanen@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>

Tegra HDA has FIFO size which can hold upto 10 audio frames to support
DVFS. When HDA DMA RUN bit is set to 0 to stop the stream, the DMA RUN
bit will be cleared to 0 only after transferring all the remaining audio
frames queued up in the fifo. This is not in sync with spec which states
the controller will stop transmitting(output) in the beginning of the
next frame for the relevant stream.

The above behavior with Tegra HDA was resulting in machine check error
during the system suspend flow with active audio playback with below kernel
error logs.
[ 33.524583] mc-err: [mcerr] (hda) csr_hdar: EMEM address decode error
[ 33.531088] mc-err: [mcerr] status = 0x20000015; addr = 0x00000000
[ 33.537431] mc-err: [mcerr] secure: no, access-type: read, SMMU fault: none

This was due to the fifo has more than one audio frame when the DMA
RUN bit is set to 0 during system suspend flow and the timeout handling in
snd_hdac_stream_sync() was not designed to handle this scenario. So the
DMA will continue running even after timeout hit until all remaining
audio frames in the fifo are transferred, but the suspend flow will try
to reset the controller and turn off the hda clocks without the knowledge
of the DMA is still running and could result in mc-err.

The above issue can be resolved by doing stream reset with the help of
snd_hdac_stream_reset() which would ensure the DMA RUN bit is cleared
if the timeout was hit in snd_hdac_stream_sync().

Signed-off-by: Mohan Kumar <mkumard@nvidia.com>
Link: https://lore.kernel.org/r/20200128051508.26064-1-mkumard@nvidia.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>

Like many other drivers, HD-audio drivers also do PCM buffer
preallocation to assure the buffer pages allocated at the early boot
stage.  This step is useful for platforms that may fail to allocate
the PCM hardware buffers -- which is mostly for either large
continuous pages or with the specific DMA mask (like emu10k1).

OTOH, when a buffer is allocated as SG-buffer and the DMA mask is
either 32 or 64 bits, the allocation almost never fails unless it hits
the real OOM situation.  In such a case, we don't need the
preallocation inevitably unlike the cases above.

That said, we may drop the preallocation for HD-audio that does
allocate via SG-buffers, and the patch achieves it.

However, there is one caveat: the buffer allocation behavior depends
on CONFIG_SND_DMA_SGBUF, and it falls back to the continuous pages
when it's not set.  And, currently this SG buffer allocation is
enabled only on x86 platforms.  So, covering those fall-outs, the
patch adjusts CONFIG_SND_HDA_PREALLOC_SIZE depending on the condition,
and keeps the old behavior as-is for non-x86 platforms.

On x86, the kconfig item is no longer adjustable but always set to
zero for disabling the preallocation.  You can still enable the
preallocation via procfs interface at any time later, too.

Link: https://lore.kernel.org/r/20200120124423.11862-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>

Make W=1 throws a lot of warnings, with multiple misalignments between
function params and their descriptions.

Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20200113205638.27338-1-pierre-louis.bossart@linux.intel.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>

In the commit 8e85def5723e ("ALSA: hda: enable regmap internal
locking"), we re-enabled the regmap lock due to the reported
regression that showed the possible concurrent accesses.  It was a
temporary workaround, and there are still a few opened races even
after the revert.  In this patch, we cover those still opened windows
with a proper mutex lock and disable the regmap internal lock again.

First off, the patch introduces a new snd_hdac_device.regmap_lock
mutex that is applied for each snd_hdac_regmap_*() call, including
read, write and update helpers.  The mutex is applied carefully so
that it won't block the self-power-up procedure in the helper
function.  Also, this assures the protection for the accesses without
regmap, too.

The snd_hdac_regmap_update_raw() is refactored to use the standard
regmap_update_bits_check() function instead of the open-code.  The
non-regmap case is still open-coded but it's an easy part.  The all
read and write operations are in the single mutex protection, so it's
now race-free.

In addition, a couple of new helper functions are added:
snd_hdac_regmap_update_raw_once() and snd_hdac_regmap_sync().  Both
are called from HD-audio legacy driver.  The former is to initialize
the given verb bits but only once when it's not initialized yet.  Due
to this condition, the function invokes regcache_cache_only(), and
it's now performed inside the regmap_lock (formerly it was racy) too.
The latter function is for simply invoking regcache_sync() inside the
regmap_lock, which is called from the codec resume call path.
Along with that, the HD-audio codec driver code is slightly modified /
simplified to adapt those new functions.

And finally, snd_hdac_regmap_read_raw(), *_write_raw(), etc are
rewritten with the helper macro.  It's just for simplification because
the code logic is identical among all those functions.

Tested-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Link: https://lore.kernel.org/r/20200109090104.26073-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>

Sync 5.5-devel branch once again for applying the HD-audio fixes.

Signed-off-by: Takashi Iwai <tiwai@suse.de>

This reverts commit 42ec336f1f9d ("ALSA: hda: Disable regmap
internal locking").

Without regmap locking, there is a race between snd_hda_codec_amp_init()
and PM callbacks issuing regcache_sync(). This was caught by
following kernel warning trace:

<4> [358.080081] WARNING: CPU: 2 PID: 4157 at drivers/base/regmap/regcache.c:498 regcache_cache_only+0xf5/0x130
[...]
<4> [358.080148] Call Trace:
<4> [358.080158]  snd_hda_codec_amp_init+0x4e/0x100 [snd_hda_codec]
<4> [358.080169]  snd_hda_codec_amp_init_stereo+0x40/0x80 [snd_hda_codec]

Suggested-by: Takashi Iwai <tiwai@suse.de>
BugLink: https://gitlab.freedesktop.org/drm/intel/issues/592
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Link: https://lore.kernel.org/r/20200108180856.5194-1-kai.vehmanen@linux.intel.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>

Link: https://lore.kernel.org/r/20200108153430.31456-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>