System takes a very long time to suspend after commit 215a22ed31
("ALSA: hda: Refactor codec PM to use direct-complete optimization"):
[ 90.065964] PM: suspend entry (s2idle)
[ 90.067337] Filesystems sync: 0.001 seconds
[ 90.185758] Freezing user space processes ... (elapsed 0.002 seconds) done.
[ 90.188713] OOM killer disabled.
[ 90.188714] Freezing remaining freezable tasks ... (elapsed 0.001 seconds) done.
[ 90.190024] printk: Suspending console(s) (use no_console_suspend to debug)
[ 90.904912] intel_pch_thermal 0000:00:12.0: CPU-PCH is cool [49C], continue to suspend
[ 321.262505] snd_hda_codec_realtek ehdaudio0D0: Unable to sync register 0x2b8000. -5
[ 328.426919] snd_hda_codec_realtek ehdaudio0D0: Unable to sync register 0x2b8000. -5
[ 329.490933] ACPI: EC: interrupt blocked
That commit keeps the codec suspended during the system suspend. However,
mute/micmute LED will clear codec's direct-complete flag by
dpm_clear_superiors_direct_complete().
This doesn't play well with SOF driver. When its runtime resume is
called for system suspend, hda_codec_jack_check() schedules
jackpoll_work which uses snd_hdac_is_power_on() to check whether codec
is suspended. Because the direct-complete path isn't taken,
pm_runtime_disable() isn't called so snd_hdac_is_power_on() returns
false and jackpoll continues to run, and snd_hda_power_up_pm() cannot
power up an already suspended codec in multiple attempts, causes the
long delay on system suspend:
if (dev->power.direct_complete) {
if (pm_runtime_status_suspended(dev)) {
pm_runtime_disable(dev);
if (pm_runtime_status_suspended(dev)) {
pm_dev_dbg(dev, state, "direct-complete ");
goto Complete;
}
pm_runtime_enable(dev);
}
dev->power.direct_complete = false;
}
When direct-complete path is taken, snd_hdac_is_power_on() returns true
and hda_jackpoll_work() is skipped by accident. So this is still not
correct.
If we were to use snd_hdac_is_power_on() in system PM path,
pm_runtime_status_suspended() should be used instead of
pm_runtime_suspended(), otherwise pm_runtime_{enable,disable}() may
change the outcome of snd_hdac_is_power_on().
Because devices suspend in reverse order (i.e. child first), it doesn't
make much sense to resume an already suspended codec from audio
controller. So avoid the issue by making sure jackpoll isn't used in
system PM process.
Fixes: 215a22ed31 ("ALSA: hda: Refactor codec PM to use direct-complete optimization")
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Reviewed-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Link: https://lore.kernel.org/r/20210112181128.1229827-3-kai.heng.feng@canonical.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch implements support for DSP D0i3 when the system
is in S0. The basic idea is to schedule a delayed work after
every successful IPC TX that checks if there are only
D0I3-compatible streams active and if so transition
the DSP to D0I3.
With the introduction of DSP D0I3 in S0, we need to
ensure that the DSP is in D0I0 before sending any new
IPCs. The exception for this would be the
compact IPCs that are used to set the DSP in
D0I3/D0I0 states.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.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/20200129220726.31792-9-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
The DSP device substates such as D0I0/D0I3
are platform-specific. Therefore, the d0_substate
field of struct snd_sof_dev is replaced
with the dsp_power_state field which represents the current
state of the DSP. This field holds both the device state
and the platform-specific substate values.
With the DSP device substates being platform-specific,
the DSP power state transitions need to be performed in
the platform-specific suspend/resume ops as well.
In order to achieve this, the ops signature has to be
modified to pass the target device state as an
argument. The target substate will be determined by
the platform-specific ops before performing the transition.
For example, in the case of the system suspending to S0IX,
the top-level SOF device suspend callback needs to
only determine if the DSP will be entering
D3 or remain in D0. The target substate in case the device
needs to remain in D0 (D0I0 or D0I3) will be determined
by the platform-specific suspend op.
With the addition of the extended set of power states for the DSP,
the set_power_state op for HDA platforms has to be extended
to handle only the appropriate state transitions. So, the
implementation for the Intel HDA platforms is also modified.
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/20200129220726.31792-6-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
When HDA controller is resumed from suspend, i915 HDMI/DP
codec requires that following order of actions is kept:
- i915 display power up and configuration of link params
- hda link reset and setup
Current SOF HDA code delegates display codec power control
to the codec driver. This works most of the time, but in
runtime PM sequences, the above constraint may be violated.
On platforms where BIOS values for HDA link parameters do
not match hardware reset defaults, this may lead to errors
in HDA verb transactions after resume.
Fix the issue by explicitly powering the display codec
in the HDA controller resume/suspend calls, thus ensuring
correct ordering. Special handling is needed for the D0i3
flow, where display power must be turned off even though
DSP is left powered.
Now that we have more invocations of the display power helper
functions, the conditional checks surrounding each call have
been moved inside hda_codec_i915_display_power(). The two
special cases of display powering at initial probe are handled
separately. The intent is to avoid powering the display whenever
no display codecs are used.
Note that early powering of display was removed in
commit 687ae9e287 ("ASoC: intel: skl: Fix display power regression").
This change was also copied to the SOF driver. No failures
have resulted as hardware default values for link parameters
have worked out of the box. However with recent i915 driver
changes like done in commit 87c1694533 ("drm/i915: save
AUD_FREQ_CNTRL state at audio domain suspend"), this does not
hold anymore and errors are hit.
Cc: Takashi Iwai <tiwai@suse.de>
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/20200206200223.7715-2-kai.vehmanen@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
The recent change to remove the bus->io_ops callbacks used an older
version of the SOF code base, and when merged into Mark's for-next it
invalidated changes, resulting in broken compilation identified by
kbuild and reproduced during the weekly SOF rebase.
Restore SOF code overridden by git merge and apply Takashi's intended
change in the 'right' location.
Fixes: c2f16a94a8 ("Merge branch 'topic/hda-bus-ops-cleanup'")
Reported-by: kbuild test robot <lkp@intel.com>
Cc: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20190812190502.30729-1-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
HD-audio drivers access to the mmio registers indirectly via the
corresponding bus->io_ops callbacks. This is because some platform
(notably Tegra SoC) requires the word-aligned access. But it's rather
a rare case, and other platforms suffer from the penalties by indirect
calls unnecessarily.
This patch is an attempt to optimize and cleanup for this situation.
Now the special aligned access is used only when a new kconfig
CONFIG_SND_HDA_ALIGNED_MMIO is set. And the HD-audio core itself
provides the aligned MMIO access helpers instead of the driver side.
If Kconfig isn't set (as default), the standard helpers like readl()
or writel() are used directly.
A couple of places in ASoC Intel drivers have the access via io_ops
reg_writel(), and they are replaced with the direct writel() calls.
And now with this patch, the whole bus->io_ops becomes empty, so it's
dropped completely. The bus initialization functions are changed
accordingly as well to drop the whole bus->io_ops.
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
For this bug, there are two capture pcm streams active, with one
stream and its related stream tag released before suspend. Later
when system suspend is done, the stream tag for the remaining
active stream is released by SOF driver. After system resume, hda
codec driver restores the stream tag for the active pcm stream,
but SOF goes to assign a new one, which now doesn't match with the
stream tag used by codec driver, and this causes DMA to fail
receiving data, leading to unrecoverable XRUN condition in FW.
For stream tag is stored in both hda codec and SOF driver, it
shouldn't be released only in SOF driver. This patch just keeps the
stream information in dma data and checks whether there is a stored
DMA data for stream resuming from S3 and restores it. And it also
removes DMA data when the stream is released.
Tested on Whiskey Lake platform.
GitHub issue: https://github.com/thesofproject/sof/issues/1594
Signed-off-by: Rander Wang <rander.wang@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20190722141402.7194-19-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In commit 7d4f606c50 ("ALSA: hda - WAKEEN feature enabling for
runtime pm"), legacy HD-A driver sets hda controller in reset mode after
entering runtime-suspend. And when resuming from suspend mode, it checks
hda controller & codec status to detect headphone hotplug event. Now
this patch does the same job in SOF runtime pm functions.
And we need to check all the non-hdmi codecs for some cases like playback
with HDMI or capture with DMIC connected to dsp. In these cases, only
controller is active and codecs are suspended, so codecs can't send
unsolicited event to controller. The jack polling operation will activate
codecs and unsolicited event can work even codecs become suspended later.
Tested on whiskylake with hda codecs.
Signed-off-by: Rander Wang <rander.wang@linux.intel.com>
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Link: https://lore.kernel.org/r/20190722141402.7194-13-pierre-louis.bossart@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Paused streams do not get suspended when the system enters S3.
So, clear and release link DMA channel for such streams in the
hda_dsp_set_hw_params_upon_resume() callback. Also, invalidate
the link DMA channel in the DAI config before restoring the
dai config upon resume. Also, modify the signature for the
set_hw_params_upon_resume() op to return an int.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
