[PATCH v4 0/4] ASoC: SOF: Intel: power optimizations with HDaudio SPIB register
This patchset was initially provided in a larger series that was split in two [1]. This part only provides support for the SPIB register support, added on Intel platforms since Skylake (2015).
The use of the SPIB register helps reduce power consumption - though to a smaller degree than DMI_L1. This hardware capability is however incompatible with userspace-initiated rewinds typically used by PulseAudio.
In the past (2015..2017) Intel suggested an API extension to let applications disable rewinds. At the time the feedback was that such a capability was too Intel-specific and SPIB remained unused except for loading DSP code. We now see devices with smaller batteries being released, and it's time to revisit Linux support for SPIB to extend battery life.
In this update the rewinds are disabled via an opt-in kernel parameter. In the previous reviews, there was consensus that a Kconfig option was too complicated for distributions to set, and we are missing a TBD API to expose such capabilities to user-space.
The debate on whether or not to use rewinds, and the impact of disabling rewinds, will likely be closed when Intel releases the 'deep-buffer' support, currently under development [2][3]. With this solution, rewinds will not be needed, ever. When an application deals with content that is not latency-sensitive (e.g. music playback), it will be able to reduce power consumption by selecting a different PCM device with increased buffering capabilities. Low-latency streams will be handled by the 'regular' path. In other words, the impossible compromise between power and latency will be handled with different PCM devices/profiles for the same endpoint, and we can push the design of capability negotiation to a later time when all the building blocks (firmware topology, kernel, userspace) are ready - we still have firmware xruns, DPCM race conditions to solve, and a need to describe these alternate PCM devices with UCM using 'modifiers'.
[1] https://lore.kernel.org/alsa-devel/20210610205326.1176400-1-pierre-louis.bos... [2] https://github.com/thesofproject/linux/pull/3146 [3] https://github.com/thesofproject/sof/pull/4611
Changes since v3: Added Takashi's Reviewed-by tags on patches 1 and 2 Used SYNC_APPLPTR instead of EXPLICIT_SYNC as suggested by Takashi Added a missing .ack callback for IceLake
Changes since v2: Added check on appl_ptr (Takashi) Moved rewind detection to deal with SYNC_PTR (Takashi)
Pierre-Louis Bossart (2): ALSA: pcm: unconditionally check if appl_ptr is in 0..boundary range ALSA: pcm: introduce INFO_NO_REWINDS flag
Ranjani Sridharan (2): ASoC: SOF: pcm: add .ack callback support ASoC: SOF: Intel: add .ack support for HDaudio platforms
include/uapi/sound/asound.h | 2 +- sound/core/pcm_lib.c | 17 +++++++++++++ sound/soc/sof/intel/apl.c | 1 + sound/soc/sof/intel/cnl.c | 1 + sound/soc/sof/intel/hda-pcm.c | 41 ++++++++++++++++++++++++++++++-- sound/soc/sof/intel/hda-stream.c | 2 ++ sound/soc/sof/intel/hda.h | 1 + sound/soc/sof/intel/icl.c | 1 + sound/soc/sof/intel/tgl.c | 1 + sound/soc/sof/ops.h | 10 ++++++++ sound/soc/sof/pcm.c | 9 +++++++ sound/soc/sof/sof-priv.h | 3 +++ 12 files changed, 86 insertions(+), 3 deletions(-)
In some cases, the appl_ptr passed by userspace is not checked before being used. This patch adds an unconditional check and returns an error code should the appl_ptr exceed the ALSA 'boundary'.
Suggested-by: Takashi Iwai tiwai@suse.de Reviewed-by: Takashi Iwai tiwai@suse.de Reviewed-by: Ranjani Sridharan ranjani.sridharan@linux.intel.com Reviewed-by: Kai Vehmanen kai.vehmanen@linux.intel.com Signed-off-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com --- sound/core/pcm_lib.c | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/sound/core/pcm_lib.c b/sound/core/pcm_lib.c index 4f4b4739f987..fdd992772b20 100644 --- a/sound/core/pcm_lib.c +++ b/sound/core/pcm_lib.c @@ -2133,6 +2133,9 @@ int pcm_lib_apply_appl_ptr(struct snd_pcm_substream *substream, if (old_appl_ptr == appl_ptr) return 0;
+ if (appl_ptr >= runtime->boundary) + return -EINVAL; + runtime->control->appl_ptr = appl_ptr; if (substream->ops->ack) { ret = substream->ops->ack(substream);
When the hardware can only deal with a monotonically increasing appl_ptr, this flag can be set.
In case the application requests a rewind, be it with a snd_pcm_rewind() or with a direct change of a mmap'ed pointer followed by a SNDRV_PCM_IOCTL_SYNC_PTR, this patch checks if a rewind occurred and returns an error.
Credits to Takashi Iwai for identifying the path with SYNC_PTR and suggesting the pointer checks.
Suggested-by: Takashi Iwai tiwai@suse.de Reviewed-by: Takashi Iwai tiwai@suse.de Reviewed-by: Ranjani Sridharan ranjani.sridharan@linux.intel.com Reviewed-by: Kai Vehmanen kai.vehmanen@linux.intel.com Signed-off-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com --- include/uapi/sound/asound.h | 2 +- sound/core/pcm_lib.c | 14 ++++++++++++++ 2 files changed, 15 insertions(+), 1 deletion(-)
diff --git a/include/uapi/sound/asound.h b/include/uapi/sound/asound.h index 5fbb79e30819..ff7e638221c5 100644 --- a/include/uapi/sound/asound.h +++ b/include/uapi/sound/asound.h @@ -300,7 +300,7 @@ typedef int __bitwise snd_pcm_subformat_t; #define SNDRV_PCM_INFO_HAS_LINK_ESTIMATED_ATIME 0x04000000 /* report estimated link audio time */ #define SNDRV_PCM_INFO_HAS_LINK_SYNCHRONIZED_ATIME 0x08000000 /* report synchronized audio/system time */ #define SNDRV_PCM_INFO_EXPLICIT_SYNC 0x10000000 /* needs explicit sync of pointers and data */ - +#define SNDRV_PCM_INFO_NO_REWINDS 0x20000000 /* hardware can only support monotonic changes of appl_ptr */ #define SNDRV_PCM_INFO_DRAIN_TRIGGER 0x40000000 /* internal kernel flag - trigger in drain */ #define SNDRV_PCM_INFO_FIFO_IN_FRAMES 0x80000000 /* internal kernel flag - FIFO size is in frames */
diff --git a/sound/core/pcm_lib.c b/sound/core/pcm_lib.c index fdd992772b20..f2090025236b 100644 --- a/sound/core/pcm_lib.c +++ b/sound/core/pcm_lib.c @@ -2128,6 +2128,7 @@ int pcm_lib_apply_appl_ptr(struct snd_pcm_substream *substream, { struct snd_pcm_runtime *runtime = substream->runtime; snd_pcm_uframes_t old_appl_ptr = runtime->control->appl_ptr; + snd_pcm_sframes_t diff; int ret;
if (old_appl_ptr == appl_ptr) @@ -2135,6 +2136,19 @@ int pcm_lib_apply_appl_ptr(struct snd_pcm_substream *substream,
if (appl_ptr >= runtime->boundary) return -EINVAL; + /* + * check if a rewind is requested by the application + */ + if (substream->runtime->info & SNDRV_PCM_INFO_NO_REWINDS) { + diff = appl_ptr - old_appl_ptr; + if (diff >= 0) { + if (diff > runtime->buffer_size) + return -EINVAL; + } else { + if (runtime->boundary + diff > runtime->buffer_size) + return -EINVAL; + } + }
runtime->control->appl_ptr = appl_ptr; if (substream->ops->ack) {
From: Ranjani Sridharan ranjani.sridharan@linux.intel.com
Add the indirections required at the core level for platform-specific operations on ack.
Note that on errors in the .ack the ALSA core will restore the previous appl_ptr.
Reviewed-by: Péter Ujfalusi peter.ujfalusi@linux.intel.com Reviewed-by: Kai Vehmanen kai.vehmanen@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 --- sound/soc/sof/ops.h | 10 ++++++++++ sound/soc/sof/pcm.c | 9 +++++++++ sound/soc/sof/sof-priv.h | 3 +++ 3 files changed, 22 insertions(+)
diff --git a/sound/soc/sof/ops.h b/sound/soc/sof/ops.h index 09bf38fdfb8a..8c58741176a0 100644 --- a/sound/soc/sof/ops.h +++ b/sound/soc/sof/ops.h @@ -454,6 +454,16 @@ snd_sof_pcm_platform_pointer(struct snd_sof_dev *sdev, return 0; }
+/* pcm ack */ +static inline int snd_sof_pcm_platform_ack(struct snd_sof_dev *sdev, + struct snd_pcm_substream *substream) +{ + if (sof_ops(sdev) && sof_ops(sdev)->pcm_ack) + return sof_ops(sdev)->pcm_ack(sdev, substream); + + return 0; +} + #if IS_ENABLED(CONFIG_SND_SOC_SOF_DEBUG_PROBES) static inline int snd_sof_probe_compr_assign(struct snd_sof_dev *sdev, diff --git a/sound/soc/sof/pcm.c b/sound/soc/sof/pcm.c index 31dd79b794f1..98aa5a6579e3 100644 --- a/sound/soc/sof/pcm.c +++ b/sound/soc/sof/pcm.c @@ -917,6 +917,14 @@ static void sof_pcm_remove(struct snd_soc_component *component) snd_soc_tplg_component_remove(component); }
+static int sof_pcm_ack(struct snd_soc_component *component, + struct snd_pcm_substream *substream) +{ + struct snd_sof_dev *sdev = snd_soc_component_get_drvdata(component); + + return snd_sof_pcm_platform_ack(sdev, substream); +} + void snd_sof_new_platform_drv(struct snd_sof_dev *sdev) { struct snd_soc_component_driver *pd = &sdev->plat_drv; @@ -935,6 +943,7 @@ void snd_sof_new_platform_drv(struct snd_sof_dev *sdev) pd->hw_free = sof_pcm_hw_free; pd->trigger = sof_pcm_trigger; pd->pointer = sof_pcm_pointer; + pd->ack = sof_pcm_ack;
#if IS_ENABLED(CONFIG_SND_SOC_SOF_DEBUG_PROBES) pd->compress_ops = &sof_probe_compressed_ops; diff --git a/sound/soc/sof/sof-priv.h b/sound/soc/sof/sof-priv.h index 9a8af76b2f8b..1e924d849731 100644 --- a/sound/soc/sof/sof-priv.h +++ b/sound/soc/sof/sof-priv.h @@ -206,6 +206,9 @@ struct snd_sof_dsp_ops { snd_pcm_uframes_t (*pcm_pointer)(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream); /* optional */
+ /* pcm ack */ + int (*pcm_ack)(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream); /* optional */ + #if IS_ENABLED(CONFIG_SND_SOC_SOF_DEBUG_PROBES) /* Except for probe_pointer, all probe ops are mandatory */ int (*probe_assign)(struct snd_sof_dev *sdev,
From: Ranjani Sridharan ranjani.sridharan@linux.intel.com
When we disable rewinds, then the .ack can be used to program SPIB with the application pointer, which allows the HDaudio DMA to save power by opportunistically bursting data transfers when the path to memory is enabled (and conversely to shut it down when there are no transfer requests).
The SPIB register can only be programmed with incremental values with wrap-around after the DMA RUN bits are set. For simplicity, we set the INFO_NO_REWINDS flag in the .open callback when we already need to program the SNDRV_PCM_INFO_SYNC_APPLPTR flag.
Rewinds are not used by many applications. One notable application using rewinds is PulseAudio. Practical experiments with Ubuntu/PulseAudio default settings did not show any audible issues, but the user may hear volume changes and notification with a delay, depending on the size of the ring buffer and latency constraints.
The choice of disabling rewinds is exposed as a kernel parameter and not a Kconfig option to avoid any undesirable side-effects.
Reviewed-by: Péter Ujfalusi peter.ujfalusi@linux.intel.com Reviewed-by: Kai Vehmanen kai.vehmanen@linux.intel.com Co-developed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com Signed-off-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com Signed-off-by: Ranjani Sridharan ranjani.sridharan@linux.intel.com --- sound/soc/sof/intel/apl.c | 1 + sound/soc/sof/intel/cnl.c | 1 + sound/soc/sof/intel/hda-pcm.c | 41 ++++++++++++++++++++++++++++++-- sound/soc/sof/intel/hda-stream.c | 2 ++ sound/soc/sof/intel/hda.h | 1 + sound/soc/sof/intel/icl.c | 1 + sound/soc/sof/intel/tgl.c | 1 + 7 files changed, 46 insertions(+), 2 deletions(-)
diff --git a/sound/soc/sof/intel/apl.c b/sound/soc/sof/intel/apl.c index 917f78cf6daf..689679014ade 100644 --- a/sound/soc/sof/intel/apl.c +++ b/sound/soc/sof/intel/apl.c @@ -78,6 +78,7 @@ const struct snd_sof_dsp_ops sof_apl_ops = { .pcm_hw_free = hda_dsp_stream_hw_free, .pcm_trigger = hda_dsp_pcm_trigger, .pcm_pointer = hda_dsp_pcm_pointer, + .pcm_ack = hda_dsp_pcm_ack,
#if IS_ENABLED(CONFIG_SND_SOC_SOF_HDA_PROBES) /* probe callbacks */ diff --git a/sound/soc/sof/intel/cnl.c b/sound/soc/sof/intel/cnl.c index 3957e2b3db32..d128c08ba726 100644 --- a/sound/soc/sof/intel/cnl.c +++ b/sound/soc/sof/intel/cnl.c @@ -283,6 +283,7 @@ const struct snd_sof_dsp_ops sof_cnl_ops = { .pcm_hw_free = hda_dsp_stream_hw_free, .pcm_trigger = hda_dsp_pcm_trigger, .pcm_pointer = hda_dsp_pcm_pointer, + .pcm_ack = hda_dsp_pcm_ack,
#if IS_ENABLED(CONFIG_SND_SOC_SOF_HDA_PROBES) /* probe callbacks */ diff --git a/sound/soc/sof/intel/hda-pcm.c b/sound/soc/sof/intel/hda-pcm.c index cc8ddef37f37..974383cd0440 100644 --- a/sound/soc/sof/intel/hda-pcm.c +++ b/sound/soc/sof/intel/hda-pcm.c @@ -32,6 +32,10 @@ static bool hda_always_enable_dmi_l1; module_param_named(always_enable_dmi_l1, hda_always_enable_dmi_l1, bool, 0444); MODULE_PARM_DESC(always_enable_dmi_l1, "SOF HDA always enable DMI l1");
+static bool hda_disable_rewinds = IS_ENABLED(CONFIG_SND_SOC_SOF_HDA_DISABLE_REWINDS); +module_param_named(disable_rewinds, hda_disable_rewinds, bool, 0444); +MODULE_PARM_DESC(disable_rewinds, "SOF HDA disable rewinds"); + u32 hda_dsp_get_mult_div(struct snd_sof_dev *sdev, int rate) { switch (rate) { @@ -120,8 +124,11 @@ int hda_dsp_pcm_hw_params(struct snd_sof_dev *sdev, return ret; }
- /* disable SPIB, to enable buffer wrap for stream */ - hda_dsp_stream_spib_config(sdev, stream, HDA_DSP_SPIB_DISABLE, 0); + /* enable SPIB when rewinds are disabled */ + if (hda_disable_rewinds) + hda_dsp_stream_spib_config(sdev, stream, HDA_DSP_SPIB_ENABLE, 0); + else + hda_dsp_stream_spib_config(sdev, stream, HDA_DSP_SPIB_DISABLE, 0);
/* update no_stream_position flag for ipc params */ if (hda && hda->no_ipc_position) { @@ -140,6 +147,29 @@ int hda_dsp_pcm_hw_params(struct snd_sof_dev *sdev, return 0; }
+/* update SPIB register with appl position */ +int hda_dsp_pcm_ack(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream) +{ + struct hdac_stream *hstream = substream->runtime->private_data; + struct hdac_ext_stream *hext_stream = stream_to_hdac_ext_stream(hstream); + struct snd_pcm_runtime *runtime = substream->runtime; + ssize_t appl_pos, buf_size; + u32 spib; + + appl_pos = frames_to_bytes(runtime, runtime->control->appl_ptr); + buf_size = frames_to_bytes(runtime, runtime->buffer_size); + + spib = appl_pos % buf_size; + + /* Allowable value for SPIB is 1 byte to max buffer size */ + if (!spib) + spib = buf_size; + + sof_io_write(sdev, hext_stream->spib_addr, spib); + + return 0; +} + int hda_dsp_pcm_trigger(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream, int cmd) { @@ -234,6 +264,13 @@ int hda_dsp_pcm_open(struct snd_sof_dev *sdev, return -EINVAL; }
+ /* + * if we want the .ack to work, we need to prevent the control from being mapped. + * The status can still be mapped. + */ + if (hda_disable_rewinds) + runtime->hw.info |= SNDRV_PCM_INFO_NO_REWINDS | SNDRV_PCM_INFO_SYNC_APPLPTR; + /* * All playback streams are DMI L1 capable, capture streams need * pause push/release to be disabled diff --git a/sound/soc/sof/intel/hda-stream.c b/sound/soc/sof/intel/hda-stream.c index 1d845c2cbc33..b6f037815344 100644 --- a/sound/soc/sof/intel/hda-stream.c +++ b/sound/soc/sof/intel/hda-stream.c @@ -655,6 +655,8 @@ int hda_dsp_stream_hw_free(struct snd_sof_dev *sdev, SOF_HDA_REG_PP_PPCTL, mask, 0); spin_unlock_irq(&bus->reg_lock);
+ hda_dsp_stream_spib_config(sdev, link_dev, HDA_DSP_SPIB_DISABLE, 0); + stream->substream = NULL;
return 0; diff --git a/sound/soc/sof/intel/hda.h b/sound/soc/sof/intel/hda.h index 1195018a1f4f..6829d36fbfe9 100644 --- a/sound/soc/sof/intel/hda.h +++ b/sound/soc/sof/intel/hda.h @@ -533,6 +533,7 @@ int hda_dsp_pcm_trigger(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream, int cmd); snd_pcm_uframes_t hda_dsp_pcm_pointer(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream); +int hda_dsp_pcm_ack(struct snd_sof_dev *sdev, struct snd_pcm_substream *substream);
/* * DSP Stream Operations. diff --git a/sound/soc/sof/intel/icl.c b/sound/soc/sof/intel/icl.c index 0b2cc331d55b..1e3bc555b807 100644 --- a/sound/soc/sof/intel/icl.c +++ b/sound/soc/sof/intel/icl.c @@ -77,6 +77,7 @@ const struct snd_sof_dsp_ops sof_icl_ops = { .pcm_hw_free = hda_dsp_stream_hw_free, .pcm_trigger = hda_dsp_pcm_trigger, .pcm_pointer = hda_dsp_pcm_pointer, + .pcm_ack = hda_dsp_pcm_ack,
#if IS_ENABLED(CONFIG_SND_SOC_SOF_HDA_PROBES) /* probe callbacks */ diff --git a/sound/soc/sof/intel/tgl.c b/sound/soc/sof/intel/tgl.c index 48da8e7a67bc..2a29058e0c20 100644 --- a/sound/soc/sof/intel/tgl.c +++ b/sound/soc/sof/intel/tgl.c @@ -73,6 +73,7 @@ const struct snd_sof_dsp_ops sof_tgl_ops = { .pcm_hw_free = hda_dsp_stream_hw_free, .pcm_trigger = hda_dsp_pcm_trigger, .pcm_pointer = hda_dsp_pcm_pointer, + .pcm_ack = hda_dsp_pcm_ack,
#if IS_ENABLED(CONFIG_SND_SOC_SOF_HDA_PROBES) /* probe callbacks */
On Fri, 19 Nov 2021 17:08:48 -0600, Pierre-Louis Bossart wrote:
This patchset was initially provided in a larger series that was split in two [1]. This part only provides support for the SPIB register support, added on Intel platforms since Skylake (2015).
The use of the SPIB register helps reduce power consumption - though to a smaller degree than DMI_L1. This hardware capability is however incompatible with userspace-initiated rewinds typically used by PulseAudio.
[...]
Applied to
https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git for-next
Thanks!
[1/4] ALSA: pcm: unconditionally check if appl_ptr is in 0..boundary range commit: 0e888a74e52db369e19aec908131cf171079b306 [2/4] ALSA: pcm: introduce INFO_NO_REWINDS flag commit: b456abe63f60ad93c83a526d33b71574bc32656c [3/4] ASoC: SOF: pcm: add .ack callback support commit: 4a39ea3f07f14f21a6b97e78c972f71fc5761d3a [4/4] ASoC: SOF: Intel: add .ack support for HDaudio platforms commit: 6c26b5054ce2b822856e32f1840d13f777c6f295
All being well this means that it will be integrated into the linux-next tree (usually sometime in the next 24 hours) and sent to Linus during the next merge window (or sooner if it is a bug fix), however if problems are discovered then the patch may be dropped or reverted.
You may get further e-mails resulting from automated or manual testing and review of the tree, please engage with people reporting problems and send followup patches addressing any issues that are reported if needed.
If any updates are required or you are submitting further changes they should be sent as incremental updates against current git, existing patches will not be replaced.
Please add any relevant lists and maintainers to the CCs when replying to this mail.
Thanks, Mark
participants (2)
-
Mark Brown -
Pierre-Louis Bossart