Intel HD Audio: sound stops working in Xen PV dom0 in >=5.17

DISCUSSION Takashi Iwai 2 participants 25 messages Dec 9, 2022 – Jan 24, 2023 1 message not in the archive Sound

lore

Share & tools

lore

A message that is not in the archive (sent off-list or not yet received): Y5KPAs6f7S2dEoxR@mail-itl

Takashi Iwai

author

2022-12-09 08:11 UTC #1 lore raw

Quoted 58 lines · Marek Marczykowski-Górecki

Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
I'm not sure if that happens before sound stops working, after, of if it's related at all, but that's pretty much the only sound-related error I found in logs.
When the issue happens, on rare occasions it starts working again later for a short time, but generally the fix is to reboot. Reloading all snd_* modules (surprisingly) do not help. I don't know what exactly triggers the issue, sometimes is happen after short time like 15 minutes uptime, but usually after several hours. I guess it depends on usage pattern, but I haven't spotted any specific relation.
I managed to bisect it to this commit:
2c95b92ecd92e784785b1db8cccc4f0f2bfa850c is the first bad commit commit 2c95b92ecd92e784785b1db8cccc4f0f2bfa850c
Author: Takashi Iwai <[email protected]>
Date: Tue Nov 16 08:33:58 2021 +0100
ALSA: memalloc: Unify x86 SG-buffer handling (take#3)
This is a second attempt to unify the x86-specific SG-buffer handling code with the new standard non-contiguous page handler.
The first try (in commit 2d9ea39917a4) failed due to the wrong page and address calculations, hence reverted. (And the second try failed due to a copy&paste error.) Now it's corrected with the previous fix for noncontig pages, and the proper sg page iteration by this patch.
After the migration, SNDRV_DMA_TYPE_DMA_SG becomes identical with SNDRV_DMA_TYPE_NONCONTIG on x86, while others still fall back to SNDRV_DMA_TYPE_DEV.
Tested-by Alex Xu (Hello71) <[email protected]>
Tested-by Harald Arnesen <[email protected]>
Link https://lore.kernel.org/r/[email protected] open here
Link https://lore.kernel.org/r/[email protected] open here
Link https://lore.kernel.org/r/[email protected] open here
Signed-off-by Takashi Iwai <[email protected]>
include/sound/memalloc.h | 14 ++--
sound/core/Makefile | 1 -
sound/core/memalloc.c | 53 ++++++++++++-
sound/core/sgbuf.c | 201 -----------------------------------------------
4 files changed, 56 insertions(+), 213 deletions(-) delete mode 100644 sound/core/sgbuf.c
I've seen further follow ups to this commit, but I still observe this issue on Linux 6.0.8.

Quoted 8 lines

I have observed this issue on KBL-based system, but I've got reports also from users of other platforms (including as old as Sandy Bridge).
I tried to include all relevant information above, but some more details can be found at original report at
https://github.com/QubesOS/qubes-issues/issues/7465
Any ideas?

Hm, is it specific to Xen, i.e. if you run the normal kernel on the same machine, does it still work?

In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).

If the problem persists, another thing to check is the hack below works.

thanks,

Takashi

-- 8< --

sound/pci/hda/hda_intel.c +7 -0

@@ -1808,9 +1808,16 @@ static int azx_create(struct snd_card *card, struct pci_dev *pci,
 	if (err < 0)
 		return err;
 
+#if 0
 	/* use the non-cached pages in non-snoop mode */
 	if (!azx_snoop(chip))
 		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_WC_SG;
+#else
+	if (!azx_snoop(chip))
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_SG;
+	else
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV;
+#endif
 
 	if (chip->driver_type == AZX_DRIVER_NVIDIA) {
 		dev_dbg(chip->card->dev, "Enable delay in RIRB handling\n");

Marek Marczykowski-Górecki

2022-12-09 12:41 UTC #2 lore raw

Quoted 63 lines · Takashi Iwai

Quoted 61 lines · Marek Marczykowski-Górecki
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
I'm not sure if that happens before sound stops working, after, of if it's related at all, but that's pretty much the only sound-related error I found in logs.
When the issue happens, on rare occasions it starts working again later for a short time, but generally the fix is to reboot. Reloading all snd_* modules (surprisingly) do not help. I don't know what exactly triggers the issue, sometimes is happen after short time like 15 minutes uptime, but usually after several hours. I guess it depends on usage pattern, but I haven't spotted any specific relation.
I managed to bisect it to this commit:
2c95b92ecd92e784785b1db8cccc4f0f2bfa850c is the first bad commit commit 2c95b92ecd92e784785b1db8cccc4f0f2bfa850c
Author: Takashi Iwai <[email protected]>
Date: Tue Nov 16 08:33:58 2021 +0100
ALSA: memalloc: Unify x86 SG-buffer handling (take#3)
This is a second attempt to unify the x86-specific SG-buffer handling code with the new standard non-contiguous page handler.
The first try (in commit 2d9ea39917a4) failed due to the wrong page and address calculations, hence reverted. (And the second try failed due to a copy&paste error.) Now it's corrected with the previous fix for noncontig pages, and the proper sg page iteration by this patch.
After the migration, SNDRV_DMA_TYPE_DMA_SG becomes identical with SNDRV_DMA_TYPE_NONCONTIG on x86, while others still fall back to SNDRV_DMA_TYPE_DEV.
Tested-by Alex Xu (Hello71) <[email protected]>
Tested-by Harald Arnesen <[email protected]>
Link https://lore.kernel.org/r/[email protected] open here
Link https://lore.kernel.org/r/[email protected] open here
Link https://lore.kernel.org/r/[email protected] open here
Signed-off-by Takashi Iwai <[email protected]>
include/sound/memalloc.h | 14 ++--
sound/core/Makefile | 1 -
sound/core/memalloc.c | 53 ++++++++++++-
sound/core/sgbuf.c | 201 -----------------------------------------------
4 files changed, 56 insertions(+), 213 deletions(-) delete mode 100644 sound/core/sgbuf.c
I've seen further follow ups to this commit, but I still observe this issue on Linux 6.0.8.
I have observed this issue on KBL-based system, but I've got reports also from users of other platforms (including as old as Sandy Bridge).

Quoted 8 lines

Quoted 5 lines
I tried to include all relevant information above, but some more details can be found at original report at
https://github.com/QubesOS/qubes-issues/issues/7465
Any ideas?
Hm, is it specific to Xen, i.e. if you run the normal kernel on the same machine, does it still work?

I don't know if that's specific to Xen, but I assume if it wouldn't be, there would be a lot more bug reports. I can't think of any other relevant difference. Unfortunately, I can't run Linux without Xen on this system long enough to confirm.

Quoted 4 lines

In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).

Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/

Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.

Quoted 2 lines

If the problem persists, another thing to check is the hack below works.

Thanks, I'll check both and report back.

Quoted 24 lines

thanks,

Takashi

-- 8< --

sound/pci/hda/hda_intel.c +7 -0

@@ -1808,9 +1808,16 @@ static int azx_create(struct snd_card *card, struct pci_dev *pci,
 	if (err < 0)
 		return err;
 
+#if 0
 	/* use the non-cached pages in non-snoop mode */
 	if (!azx_snoop(chip))
 		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_WC_SG;
+#else
+	if (!azx_snoop(chip))
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_SG;
+	else
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV;
+#endif
 
 	if (chip->driver_type == AZX_DRIVER_NVIDIA) {
 		dev_dbg(chip->card->dev, "Enable delay in RIRB handling\n");

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Marek Marczykowski-Górecki

2022-12-10 01:01 UTC #3 lore raw

Quoted 4 lines · Marek Marczykowski-Górecki

Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki

Quoted 8 lines

Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting

Hit the issue again, this message did not appear in the log (or at least not yet).

(...)

Quoted 4 lines

Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).

This did not helped.

Quoted 8 lines

Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/

Quoted 5 lines

Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.

Trying this one now.

Quoted 20 lines

-- 8< --

sound/pci/hda/hda_intel.c +7 -0

@@ -1808,9 +1808,16 @@ static int azx_create(struct snd_card *card, struct pci_dev *pci,
 	if (err < 0)
 		return err;
 
+#if 0
 	/* use the non-cached pages in non-snoop mode */
 	if (!azx_snoop(chip))
 		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_WC_SG;
+#else
+	if (!azx_snoop(chip))
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_SG;
+	else
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV;
+#endif
 
 	if (chip->driver_type == AZX_DRIVER_NVIDIA) {
 		dev_dbg(chip->card->dev, "Enable delay in RIRB handling\n");

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Marek Marczykowski-Górecki

2022-12-10 16:18 UTC #4 lore raw

Quoted 34 lines · Marek Marczykowski-Górecki

Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 8 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/

Quoted 7 lines

Quoted 5 lines
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.

And this one didn't either :/

When it stopped working, I did two things:
1. switched audio profiles ("configuration" tab in pavucontrol) several times; this on its own did not helped
2. reloaded sound related modules, but did not loaded them all back (see attached list before and after).

After this, it worked again for a few minutes. Not sure if/which the above actions were relevant, tho...

Another observation: when it stops working, it's never during a playback. It's always that at some point starting an audio stream results in a silence.

Quoted 20 lines

-- 8< --

sound/pci/hda/hda_intel.c +7 -0

@@ -1808,9 +1808,16 @@ static int azx_create(struct snd_card *card, struct pci_dev *pci,
 	if (err < 0)
 		return err;
 
+#if 0
 	/* use the non-cached pages in non-snoop mode */
 	if (!azx_snoop(chip))
 		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_WC_SG;
+#else
+	if (!azx_snoop(chip))
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV_SG;
+	else
+		azx_bus(chip)->dma_type = SNDRV_DMA_TYPE_DEV;
+#endif
 
 	if (chip->driver_type == AZX_DRIVER_NVIDIA) {
 		dev_dbg(chip->card->dev, "Enable delay in RIRB handling\n");

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

snd_hda_codec_hdmi     86016  1
snd_ctl_led            24576  0
snd_hda_codec_conexant    32768  1
snd_hda_codec_generic    98304  1 snd_hda_codec_conexant
ledtrig_audio          16384  2 snd_ctl_led,snd_hda_codec_generic
snd_hda_intel          61440  5
snd_intel_dspcfg       36864  1 snd_hda_intel
snd_intel_sdw_acpi     20480  1 snd_intel_dspcfg
snd_hda_codec         184320  4 snd_hda_codec_generic,snd_hda_codec_conexant,snd_hda_codec_hdmi,snd_hda_intel
snd_hda_core          118784  5 snd_hda_codec_generic,snd_hda_codec_conexant,snd_hda_codec_hdmi,snd_hda_intel,snd_hda_codec
snd_hwdep              16384  1 snd_hda_codec
snd_seq                94208  0
snd_seq_device         16384  1 snd_seq
snd_pcm               151552  5 snd_hda_codec_hdmi,snd_hda_intel,snd_hda_codec,snd_hda_core
snd_timer              49152  3 snd_seq,snd_pcm
snd                   126976  19 snd_ctl_led,snd_hda_codec_generic,snd_seq,snd_hda_codec_conexant,snd_seq_device,snd_hda_codec_hdmi,snd_hwdep,snd_hda_intel,snd_hda_codec,snd_timer,snd_pcm
soundcore              16384  2 snd_ctl_led,snd
snd_sof_pci_intel_skl    16384  0
snd_sof_intel_hda_common   217088  1 snd_sof_pci_intel_skl
soundwire_intel        53248  1 snd_sof_intel_hda_common
snd_sof_intel_hda      20480  1 snd_sof_intel_hda_common
snd_sof_pci            24576  2 snd_sof_intel_hda_common,snd_sof_pci_intel_skl
snd_sof_xtensa_dsp     20480  1 snd_sof_intel_hda_common
snd_sof               339968  2 snd_sof_pci,snd_sof_intel_hda_common
snd_sof_utils          20480  1 snd_sof
snd_soc_avs           172032  0
snd_soc_hda_codec      28672  1 snd_soc_avs
snd_soc_skl           217088  0
snd_soc_hdac_hda       28672  2 snd_sof_intel_hda_common,snd_soc_skl
snd_hda_ext_core       36864  5 snd_soc_avs,snd_soc_hda_codec,snd_sof_intel_hda_common,snd_soc_hdac_hda,snd_soc_skl
snd_soc_sst_ipc        20480  1 snd_soc_skl
snd_soc_sst_dsp        36864  1 snd_soc_skl
snd_soc_acpi_intel_match    73728  3 snd_sof_intel_hda_common,snd_soc_skl,snd_sof_pci_intel_skl
snd_soc_acpi           16384  3 snd_soc_acpi_intel_match,snd_sof_intel_hda_common,snd_soc_skl
snd_hda_codec_hdmi     86016  1
snd_soc_core          393216  7 snd_soc_avs,snd_soc_hda_codec,soundwire_intel,snd_sof,snd_sof_intel_hda_common,snd_soc_hdac_hda,snd_soc_skl
snd_ctl_led            24576  0
snd_compress           28672  1 snd_soc_core
ac97_bus               16384  1 snd_soc_core
snd_hda_codec_conexant    32768  1
snd_pcm_dmaengine      16384  1 snd_soc_core
snd_hda_codec_generic    98304  1 snd_hda_codec_conexant
snd_hda_intel          61440  2
snd_intel_dspcfg       36864  5 snd_soc_avs,snd_hda_intel,snd_sof,snd_sof_intel_hda_common,snd_soc_skl
snd_intel_sdw_acpi     20480  2 snd_sof_intel_hda_common,snd_intel_dspcfg
snd_hda_codec         184320  9 snd_hda_codec_generic,snd_hda_codec_conexant,snd_soc_avs,snd_hda_codec_hdmi,snd_soc_hda_codec,snd_hda_intel,snd_soc_hdac_hda,snd_soc_skl,snd_sof_intel_hda
snd_hda_core          118784  12 snd_hda_codec_generic,snd_hda_codec_conexant,snd_soc_avs,snd_hda_codec_hdmi,snd_soc_hda_codec,snd_hda_intel,snd_hda_ext_core,snd_hda_codec,snd_sof_intel_hda_common,snd_soc_hdac_hda,snd_soc_skl,snd_sof_intel_hda
snd_hwdep              16384  1 snd_hda_codec
snd_seq                94208  0
snd_seq_device         16384  1 snd_seq
snd_pcm               151552  13 snd_soc_avs,snd_hda_codec_hdmi,snd_hda_intel,snd_hda_codec,soundwire_intel,snd_sof,snd_sof_intel_hda_common,snd_compress,snd_soc_core,snd_sof_utils,snd_soc_skl,snd_hda_core,snd_pcm_dmaengine
snd_timer              49152  2 snd_seq,snd_pcm
ledtrig_audio          16384  3 snd_ctl_led,snd_hda_codec_generic,thinkpad_acpi
snd                   126976  19 snd_ctl_led,snd_hda_codec_generic,snd_seq,snd_hda_codec_conexant,snd_seq_device,snd_hda_codec_hdmi,snd_hwdep,snd_hda_intel,snd_hda_codec,snd_sof,snd_timer,snd_compress,thinkpad_acpi,snd_soc_core,snd_pcm
soundcore              16384  2 snd_ctl_led,snd

Marek Marczykowski-Górecki

2022-12-20 04:44 UTC #5 lore raw

Quoted 51 lines · Marek Marczykowski-Górecki

Quoted 34 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 8 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Quoted 7 lines
Quoted 5 lines
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
When it stopped working, I did two things:
1. switched audio profiles ("configuration" tab in pavucontrol) several times; this on its own did not helped
2. reloaded sound related modules, but did not loaded them all back (see attached list before and after).

Quoted 6 lines

After this, it worked again for a few minutes. Not sure if/which the above actions were relevant, tho...
Another observation: when it stops working, it's never during a playback. It's always that at some point starting an audio stream results in a silence.

Any other ideas? Or maybe there is another patch I should try?

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2022-12-22 08:10 UTC #6 lore raw

In reply to Marek Marczykowski-Górecki

Quoted 39 lines · Marek Marczykowski-Górecki

Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.

Quoted 6 lines

Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/

(Sorry for the late reply, as I've been off in the last weeks.)

I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.

thanks,

Takashi

-- 8< --

sound/core/Kconfig +2 -2

@@ -225,8 +225,8 @@ config SND_VMASTER
 	bool
 
 config SND_DMA_SGBUF
-	def_bool y
-	depends on X86
+	def_bool n
+#	depends on X86
 
 config SND_CTL_LED
 	tristate

Marek Marczykowski-Górecki

2022-12-27 15:28 UTC #7 lore raw

Quoted 46 lines · Takashi Iwai

Quoted 44 lines · Marek Marczykowski-Górecki
Quoted 41 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 8 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Quoted 5 lines
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.

Quoted 7 lines

Quoted 1 line
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.

This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.

Quoted 14 lines

-- 8< --

sound/core/Kconfig +2 -2

@@ -225,8 +225,8 @@ config SND_VMASTER
 	bool
 
 config SND_DMA_SGBUF
-	def_bool y
-	depends on X86
+	def_bool n
+#	depends on X86
 
 config SND_CTL_LED
 	tristate

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-16 15:56 UTC #8 lore raw

Quoted 52 lines · Marek Marczykowski-Górecki

Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)

Quoted 8 lines

Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.

Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.

Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)

thanks,

Takashi

-- 8< --

sound/core/memalloc.c +34 ~~-10~~

@@ -719,17 +719,30 @@ static const struct snd_malloc_ops snd_dma_sg_wc_ops = {
 struct snd_dma_sg_fallback {
 	size_t count;
 	struct page **pages;
+	dma_addr_t *addrs;
 };
 
 static void __snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab,
 				       struct snd_dma_sg_fallback *sgbuf)
 {
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
-	size_t i;
-
-	for (i = 0; i < sgbuf->count && sgbuf->pages[i]; i++)
-		do_free_pages(page_address(sgbuf->pages[i]), PAGE_SIZE, wc);
+	size_t i, size;
+
+	if (sgbuf->pages && sgbuf->addrs) {
+		i = 0;
+		while (i < sgbuf->count) {
+			if (!sgbuf->pages[i] || !sgbuf->addrs[i])
+				break;
+			size = sgbuf->addrs[i] & ~PAGE_MASK;
+			if (!WARN_ON(size))
+				break;
+			dma_free_coherent(dmab->dev.dev, size,
+					  page_address(sgbuf->pages[i]),
+					  sgbuf->addrs[i] & PAGE_MASK);
+			i += size;
+		}
+	}
 	kvfree(sgbuf->pages);
+	kvfree(sgbuf->addrs);
 	kfree(sgbuf);
 }

@@ -738,9 +751,8 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	struct snd_dma_sg_fallback *sgbuf;
 	struct page **pagep, *curp;
 	size_t chunk, npages;
-	dma_addr_t addr;
+	dma_addr_t *addrp;
 	void *p;
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 
 	sgbuf = kzalloc(sizeof(*sgbuf), GFP_KERNEL);
 	if (!sgbuf)

@@ -748,14 +760,16 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	size = PAGE_ALIGN(size);
 	sgbuf->count = size >> PAGE_SHIFT;
 	sgbuf->pages = kvcalloc(sgbuf->count, sizeof(*sgbuf->pages), GFP_KERNEL);
-	if (!sgbuf->pages)
+	sgbuf->addrs = kvcalloc(sgbuf->count, sizeof(*sgbuf->addrs), GFP_KERNEL);
+	if (!sgbuf->pages || !sgbuf->addrs)
 		goto error;
 
 	pagep = sgbuf->pages;
-	chunk = size;
+	addrp = sgbuf->addrs;
+	chunk = PAGE_SIZE * (PAGE_SIZE - 1); /* to fit in low bits in addrs */
 	while (size > 0) {
 		chunk = min(size, chunk);
-		p = do_alloc_pages(dmab->dev.dev, chunk, &addr, wc);
+		p = dma_alloc_coherent(dmab->dev.dev, chunk, addrp, DEFAULT_GFP);
 		if (!p) {
 			if (chunk <= PAGE_SIZE)
 				goto error;

@@ -767,6 +781,8 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 		size -= chunk;
 		/* fill pages */
 		npages = chunk >> PAGE_SHIFT;
+		*addrp |= npages; /* store in lower bits */
+		addrp += npages;
 		curp = virt_to_page(p);
 		while (npages--)
 			*pagep++ = curp++;

@@ -775,6 +791,10 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	p = vmap(sgbuf->pages, sgbuf->count, VM_MAP, PAGE_KERNEL);
 	if (!p)
 		goto error;
+
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wc(sgbuf->pages, sgbuf->count);
+
 	dmab->private_data = sgbuf;
 	/* store the first page address for convenience */
 	dmab->addr = snd_sgbuf_get_addr(dmab, 0);

@@ -787,7 +807,11 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 
 static void snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab)
 {
+	struct snd_dma_sg_fallback *sgbuf = dmab->private_data;
+
 	vunmap(dmab->area);
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wb(sgbuf->pages, sgbuf->count);
 	__snd_dma_sg_fallback_free(dmab, dmab->private_data);
 }

Takashi Iwai

author

2023-01-17 08:00 UTC #9 lore raw

Quoted 64 lines · Takashi Iwai

Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.

Quoted 8 lines

Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)

Gah, there was an obvious typo, scratch that.

Below is a proper patch. Please try this one instead.

thanks,

Takashi

-- 8< --
From: Takashi Iwai <[email protected]>
Subject: [PATCH] ALSA: memalloc: Use coherent DMA allocation for fallback again

We switched the memory allocation for fallback cases in the noncontig type to use the standard alloc_pages*() at the commit a8d302a0b770 ("ALSA: memalloc: Revive x86-specific WC page allocations again"), while we used the dma_alloc_coherent() in the past. The reason was that the page address retrieved from the virtual pointer returned from dma_alloc_coherent() can't be used with IOMMU systems. Meanwhile, we explicitly disabled the fallback allocation for IOMMU systems at the commit 9736a325137b ("ALSA: memalloc: Don't fall back for SG-buffer with IOMMU") after the commit above; that is, the usage of dma_alloc_coherent() should be OK again.

Now, we've received reports that the current fallback page allocation caused a regression on Xen (and maybe other) systems; the sound disappear partially or completely. The further investigation showed that this can be worked around by the dma_alloc_coherent() pages. So, it's time to take it back.

This patch switches back to the dma_alloc_coherent() for the fallback allocations. Unlike the previous implementation, the allocation is implemented in a more optimized way to try larger chunks. The page count is stored in the lower bits of the addresses.

Fixes a8d302a0b770 ("ALSA: memalloc: Revive x86-specific WC page allocations again")

Fixes 9736a325137b ("ALSA: memalloc: Don't fall back for SG-buffer with IOMMU")

Link https://lore.kernel.org/r/[email protected] open here

Signed-off-by Takashi Iwai <[email protected]>

---
 sound/core/memalloc.c | 44 +++++++++++++++++++++++++++++++++----------
 1 file changed, 34 insertions(+), 10 deletions(-)

sound/core/memalloc.c +34 ~~-10~~

index 81025f50a542..dff07cd6f209 100644
@@ -719,17 +719,30 @@ static const struct snd_malloc_ops snd_dma_sg_wc_ops = {
 struct snd_dma_sg_fallback {
 	size_t count;
 	struct page **pages;
+	dma_addr_t *addrs;
 };
 
 static void __snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab,
 				       struct snd_dma_sg_fallback *sgbuf)
 {
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
-	size_t i;
-
-	for (i = 0; i < sgbuf->count && sgbuf->pages[i]; i++)
-		do_free_pages(page_address(sgbuf->pages[i]), PAGE_SIZE, wc);
+	size_t i, size;
+
+	if (sgbuf->pages && sgbuf->addrs) {
+		i = 0;
+		while (i < sgbuf->count) {
+			if (!sgbuf->pages[i] || !sgbuf->addrs[i])
+				break;
+			size = sgbuf->addrs[i] & ~PAGE_MASK;
+			if (WARN_ON(!size))
+				break;
+			dma_free_coherent(dmab->dev.dev, size,
+					  page_address(sgbuf->pages[i]),
+					  sgbuf->addrs[i] & PAGE_MASK);
+			i += size;
+		}
+	}
 	kvfree(sgbuf->pages);
+	kvfree(sgbuf->addrs);
 	kfree(sgbuf);
 }

@@ -738,9 +751,8 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	struct snd_dma_sg_fallback *sgbuf;
 	struct page **pagep, *curp;
 	size_t chunk, npages;
-	dma_addr_t addr;
+	dma_addr_t *addrp;
 	void *p;
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 
 	sgbuf = kzalloc(sizeof(*sgbuf), GFP_KERNEL);
 	if (!sgbuf)

@@ -748,14 +760,16 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	size = PAGE_ALIGN(size);
 	sgbuf->count = size >> PAGE_SHIFT;
 	sgbuf->pages = kvcalloc(sgbuf->count, sizeof(*sgbuf->pages), GFP_KERNEL);
-	if (!sgbuf->pages)
+	sgbuf->addrs = kvcalloc(sgbuf->count, sizeof(*sgbuf->addrs), GFP_KERNEL);
+	if (!sgbuf->pages || !sgbuf->addrs)
 		goto error;
 
 	pagep = sgbuf->pages;
-	chunk = size;
+	addrp = sgbuf->addrs;
+	chunk = (PAGE_SIZE - 1) << PAGE_SHIFT; /* to fit in low bits in addrs */
 	while (size > 0) {
 		chunk = min(size, chunk);
-		p = do_alloc_pages(dmab->dev.dev, chunk, &addr, wc);
+		p = dma_alloc_coherent(dmab->dev.dev, chunk, addrp, DEFAULT_GFP);
 		if (!p) {
 			if (chunk <= PAGE_SIZE)
 				goto error;

@@ -767,6 +781,8 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 		size -= chunk;
 		/* fill pages */
 		npages = chunk >> PAGE_SHIFT;
+		*addrp |= npages; /* store in lower bits */
+		addrp += npages;
 		curp = virt_to_page(p);
 		while (npages--)
 			*pagep++ = curp++;

@@ -775,6 +791,10 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	p = vmap(sgbuf->pages, sgbuf->count, VM_MAP, PAGE_KERNEL);
 	if (!p)
 		goto error;
+
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wc(sgbuf->pages, sgbuf->count);
+
 	dmab->private_data = sgbuf;
 	/* store the first page address for convenience */
 	dmab->addr = snd_sgbuf_get_addr(dmab, 0);

@@ -787,7 +807,11 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 
 static void snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab)
 {
+	struct snd_dma_sg_fallback *sgbuf = dmab->private_data;
+
 	vunmap(dmab->area);
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wb(sgbuf->pages, sgbuf->count);
 	__snd_dma_sg_fallback_free(dmab, dmab->private_data);
 }

Signature · 1 line

2.35.3

Marek Marczykowski-Górecki

2023-01-17 11:37 UTC #10 lore raw

Quoted 72 lines · Takashi Iwai

Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.

Quoted 6 lines

Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.

Thanks, I'll give it a try.

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Marek Marczykowski-Górecki

2023-01-17 14:22 UTC #11 lore raw

Quoted 73 lines · Marek Marczykowski-Górecki

Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.

Quoted 8 lines

Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.

Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-17 14:34 UTC #12 lore raw

Quoted 78 lines · Marek Marczykowski-Górecki

Quoted 76 lines · Marek Marczykowski-Górecki
Quoted 75 lines · Takashi Iwai
Quoted 72 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)

Quoted 8 lines

Quoted 5 lines
Quoted 3 lines
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.

Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.

Could you try the one below instead of the previous?

Takashi

-- 8< --

sound/core/memalloc.c +3 -3

@@ -538,11 +538,11 @@ static const struct snd_malloc_ops snd_dma_wc_ops = {
  */
 static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 {
-	struct sg_table *sgt;
+	struct sg_table *sgt = NULL;
 	void *p;
 
-	sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
-				      DEFAULT_GFP, 0);
+	// sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
+	//			      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
 	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)

Marek Marczykowski-Górecki

2023-01-17 16:50 UTC #13 lore raw

Quoted 86 lines · Takashi Iwai

Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few hours. pavucontrol still shows meter bars moving, but the speakers remain silent. At least on some occasions I see the following message in dmesg:
  [ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.

Quoted 7 lines

Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?

Unfortunately, this one doesn't fix it either :/

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-17 16:53 UTC #14 lore raw

Quoted 91 lines · Marek Marczykowski-Górecki

Quoted 89 lines · Takashi Iwai
Quoted 86 lines · Marek Marczykowski-Górecki
Quoted 73 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.

Quoted 6 lines

Quoted 4 lines
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/

Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.

Takashi

Marek Marczykowski-Górecki

2023-01-17 20:35 UTC #15 lore raw

Quoted 96 lines · Takashi Iwai

Quoted 94 lines · Marek Marczykowski-Górecki
Quoted 92 lines · Takashi Iwai
Quoted 78 lines · Marek Marczykowski-Górecki
Quoted 76 lines · Marek Marczykowski-Górecki
Quoted 75 lines · Takashi Iwai
Quoted 72 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 4 lines · Marek Marczykowski-Górecki
Quoted 3 lines · Takashi Iwai
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Quoted 8 lines
Quoted 8 lines
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV that may be relevant. It configures PAT differently than native Linux. Theoretically Linux adapts automatically and using proper API (like set_memory_wc()) should just work, but at least for i915 driver it causes issues (not fully tracked down yet). Details about that bug report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.

Quoted 8 lines

Quoted 3 lines
Quoted 1 line
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.

Oh, I noticed only now: the last patch made it fail to initialize. I don't see obvious errors in dmesg, but when trying aplay, I get:

ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory

# dmesg |grep snd
[   21.947940] snd_hda_intel 0000:00:1f.3: bound 0000:00:02.0 (ops i915_audio_component_bind_ops [i915])
[   21.985769] snd_hda_codec_conexant hdaudioC0D0: CX8200: BIOS auto-probing.
[   21.987956] snd_hda_codec_conexant hdaudioC0D0: vmaster hook already present before cdev!
[   21.990073] snd_hda_codec_conexant hdaudioC0D0: autoconfig for CX8200: line_outs=1 (0x17/0x0/0x0/0x0/0x0) type:speaker
[   21.991126] snd_hda_codec_conexant hdaudioC0D0:    speaker_outs=0 (0x0/0x0/0x0/0x0/0x0)
[   21.992188] snd_hda_codec_conexant hdaudioC0D0:    hp_outs=1 (0x16/0x0/0x0/0x0/0x0)
[   21.993234] snd_hda_codec_conexant hdaudioC0D0:    mono: mono_out=0x0
[   21.994274] snd_hda_codec_conexant hdaudioC0D0:    inputs:
[   22.000517] snd_hda_codec_conexant hdaudioC0D0:      Internal Mic=0x1a
[   22.001586] snd_hda_codec_conexant hdaudioC0D0:      Mic=0x19

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-18 09:00 UTC #16 lore raw

Quoted 100 lines · Marek Marczykowski-Górecki

Quoted 98 lines · Takashi Iwai
Quoted 96 lines · Marek Marczykowski-Górecki
Quoted 93 lines · Takashi Iwai
Quoted 78 lines · Marek Marczykowski-Górecki
Quoted 76 lines · Marek Marczykowski-Górecki
Quoted 75 lines · Takashi Iwai
Quoted 72 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
Quoted 12 lines · Marek Marczykowski-Górecki
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least not yet).
(...)
Quoted 4 lines
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Quoted 11 lines
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?

Quoted 8 lines

Quoted 6 lines
Quoted 1 line
Unfortunately, this one doesn't fix it either :/
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.

The "last patch" means the patch to enforce the fallback allocation?

Quoted 6 lines

I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory

It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.

If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.

Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.

thanks,

Takashi

Marek Marczykowski-Górecki

2023-01-18 10:40 UTC #17 lore raw

Quoted 104 lines · Takashi Iwai

Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 39 lines · Marek Marczykowski-Górecki
Quoted 37 lines · Marek Marczykowski-Górecki
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
Quoted 6 lines
Quoted 4 lines
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/

Quoted 8 lines

Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?

Yes, the one about dma_alloc_noncontiguous().

Quoted 10 lines

Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.

While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.

Quoted 5 lines

If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.

Will do.

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-18 12:40 UTC #18 lore raw

Quoted 131 lines · Marek Marczykowski-Górecki

Quoted 104 lines · Takashi Iwai
Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
Quoted 45 lines · Marek Marczykowski-Górecki
On Sat, Dec 10, 2022 at 02:00:06AM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only about BDL pages. Could you check the patch below instead? It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?
Yes, the one about dma_alloc_noncontiguous().
Quoted 10 lines
Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.
While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.

Quoted 7 lines

Quoted 5 lines
If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.
Will do.

If get_dma_ops() is non-NULL, it means we need some Xen-specific workaround not to use dma_alloc_noncontiguous().
What's the best way to see whether the driver is running on Xen PV?

Meanwhile, it's helpful if you can try the combo of my last two patches, too. It should work, and if it doesn't, it implies that we're looking at a wrong place.

thanks,

Takashi

Marek Marczykowski-Górecki

2023-01-20 01:12 UTC #19 lore raw

Quoted 136 lines · Takashi Iwai

Quoted 134 lines · Marek Marczykowski-Górecki
Quoted 104 lines · Takashi Iwai
Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 52 lines · Marek Marczykowski-Górecki
Quoted 50 lines · Takashi Iwai
On Sat, 10 Dec 2022 17:17:42 +0100,
Marek Marczykowski-Górecki wrote:
On Sat, Dec 10, 2022 at 02:00:06AM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
Quoted 8 lines
Quoted 3 lines
I think the hack doesn't influence on the PCM buffer pages, but only
about BDL pages. Could you check the patch below instead?
It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be nice. Let me know if I can collect any more info that would help with that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?
Yes, the one about dma_alloc_noncontiguous().
Quoted 10 lines
Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.
While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.
Quoted 3 lines
If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.

Quoted 6 lines

Quoted 4 lines
Quoted 2 lines
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.
Will do.
If get_dma_ops() is non-NULL,

Yes, it's non-NULL.

Quoted 3 lines

it means we need some Xen-specific
workaround not to use dma_alloc_noncontiguous().
What's the best way to see whether the driver is running on Xen PV?

Usually it's this: cpu_feature_enabled(X86_FEATURE_XENPV)

Quoted 3 lines

Meanwhile, it's helpful if you can try the combo of my last two patches, too. It should work, and if it doesn't, it implies that we're looking at a wrong place.

It doesn't because the last of them causes "Cannot allocate memory". I'm trying now with this on top:

---8<---

sound/core/memalloc.c +1 -1

index 97d7b8106869..e927d18d1ebb 100644
@@ -545,7 +545,7 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	// sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
 	//	      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (!sgt) { // && !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else

---8<---

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Marek Marczykowski-Górecki

2023-01-20 02:25 UTC #20 lore raw

Quoted 165 lines · Marek Marczykowski-Górecki

Quoted 136 lines · Takashi Iwai
Quoted 134 lines · Marek Marczykowski-Górecki
Quoted 104 lines · Takashi Iwai
Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
Quoted 60 lines · Marek Marczykowski-Górecki
On Thu, Dec 22, 2022 at 09:09:15AM +0100, Takashi Iwai wrote:
On Sat, 10 Dec 2022 17:17:42 +0100,
Marek Marczykowski-Górecki wrote:
On Sat, Dec 10, 2022 at 02:00:06AM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
I think the hack doesn't influence on the PCM buffer pages, but only
about BDL pages. Could you check the patch below instead?
It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be
nice. Let me know if I can collect any more info that would help with
that.
Then we seem to go back again with the coherent memory allocation for the fallback sg cases. It was changed because the use of dma_alloc_coherent() caused a problem with IOMMU case for retrieving the page addresses, but since the commit 9736a325137b, we essentially avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?
Yes, the one about dma_alloc_noncontiguous().
Quoted 10 lines
Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.
While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.
Quoted 3 lines
If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.
Will do.
If get_dma_ops() is non-NULL,
Yes, it's non-NULL.
Quoted 3 lines
it means we need some Xen-specific
workaround not to use dma_alloc_noncontiguous().
What's the best way to see whether the driver is running on Xen PV?
Usually it's this: cpu_feature_enabled(X86_FEATURE_XENPV)
Quoted 3 lines
Meanwhile, it's helpful if you can try the combo of my last two patches, too. It should work, and if it doesn't, it implies that we're looking at a wrong place.
It doesn't because the last of them causes "Cannot allocate memory". I'm trying now with this on top:
---8<---
sound/core/memalloc.c +0 -0
index 97d7b8106869..e927d18d1ebb 100644

sound/core/memalloc.c static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)

Quoted 9 lines

sound/core/memalloc.c +1 -2

@@ -545,7 +545,7 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	// sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
 	//	      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (!sgt) { // && !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else
---8<---

Unfortunately, the above doesn't help. I mean, I don't get an error anymore, but no sound output either (even though pavucontrol says I should hear it). So, it's like the original issue, but without any delay, just straight from the start.

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-20 07:27 UTC #21 lore raw

Quoted 173 lines · Marek Marczykowski-Górecki

Quoted 171 lines · Marek Marczykowski-Górecki
Quoted 136 lines · Takashi Iwai
Quoted 134 lines · Marek Marczykowski-Górecki
Quoted 104 lines · Takashi Iwai
Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 72 lines · Takashi Iwai
Quoted 70 lines · Takashi Iwai
On Tue, 27 Dec 2022 16:26:54 +0100,
Marek Marczykowski-Górecki wrote:
On Thu, Dec 22, 2022 at 09:09:15AM +0100, Takashi Iwai wrote:
On Sat, 10 Dec 2022 17:17:42 +0100,
Marek Marczykowski-Górecki wrote:
On Sat, Dec 10, 2022 at 02:00:06AM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
I think the hack doesn't influence on the PCM buffer pages, but only
about BDL pages. Could you check the patch below instead?
It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be
nice. Let me know if I can collect any more info that would help with
that.
Then we seem to go back again with the coherent memory allocation for
the fallback sg cases. It was changed because the use of
dma_alloc_coherent() caused a problem with IOMMU case for retrieving
the page addresses, but since the commit 9736a325137b, we essentially
avoid the fallback when IOMMU is used, so it should be fine again.
Quoted 6 lines
Quoted 2 lines
Let me know if the patch like below works for you instead of the
previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?
Yes, the one about dma_alloc_noncontiguous().
Quoted 10 lines
Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.
While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.
Quoted 3 lines
If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.
Will do.
If get_dma_ops() is non-NULL,
Yes, it's non-NULL.
Quoted 3 lines
it means we need some Xen-specific
workaround not to use dma_alloc_noncontiguous().
What's the best way to see whether the driver is running on Xen PV?
Usually it's this: cpu_feature_enabled(X86_FEATURE_XENPV)
Quoted 3 lines
Meanwhile, it's helpful if you can try the combo of my last two patches, too. It should work, and if it doesn't, it implies that we're looking at a wrong place.
It doesn't because the last of them causes "Cannot allocate memory". I'm trying now with this on top:
---8<---
sound/core/memalloc.c +1 -1
index 97d7b8106869..e927d18d1ebb 100644
@@ -545,7 +545,7 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	// sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
 	//	      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (!sgt) { // && !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)

Quoted 8 lines

sound/core/memalloc.c static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
Quoted 3 lines
sound/core/memalloc.c +0 -1
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else
---8<---
Unfortunately, the above doesn't help. I mean, I don't get an error anymore, but no sound output either (even though pavucontrol says I should hear it). So, it's like the original issue, but without any delay, just straight from the start.

Hmm, it's the result with the combination of both patches, right? What I meant as the combo is something like below.

Takashi

-- 8< --

sound/core/memalloc.c +38 ~~-13~~

@@ -15,6 +15,7 @@
 #include <linux/vmalloc.h>
 #ifdef CONFIG_X86
 #include <asm/set_memory.h>
+#include <xen/xen.h>
 #endif
 #include <sound/memalloc.h>
 #include "memalloc_local.h"

@@ -541,10 +542,8 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	struct sg_table *sgt;
 	void *p;
 
-	sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
-				      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (xen_domain() || !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else

@@ -552,6 +551,8 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 		return snd_dma_sg_fallback_alloc(dmab, size);
 	}
 #endif
+	sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
+				      DEFAULT_GFP, 0);
 	if (!sgt)
 		return NULL;

@@ -719,17 +720,30 @@ static const struct snd_malloc_ops snd_dma_sg_wc_ops = {
 struct snd_dma_sg_fallback {
 	size_t count;
 	struct page **pages;
+	dma_addr_t *addrs;
 };
 
 static void __snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab,
 				       struct snd_dma_sg_fallback *sgbuf)
 {
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
-	size_t i;
-
-	for (i = 0; i < sgbuf->count && sgbuf->pages[i]; i++)
-		do_free_pages(page_address(sgbuf->pages[i]), PAGE_SIZE, wc);
+	size_t i, size;
+
+	if (sgbuf->pages && sgbuf->addrs) {
+		i = 0;
+		while (i < sgbuf->count) {
+			if (!sgbuf->pages[i] || !sgbuf->addrs[i])
+				break;
+			size = sgbuf->addrs[i] & ~PAGE_MASK;
+			if (WARN_ON(!size))
+				break;
+			dma_free_coherent(dmab->dev.dev, size,
+					  page_address(sgbuf->pages[i]),
+					  sgbuf->addrs[i] & PAGE_MASK);
+			i += size;
+		}
+	}
 	kvfree(sgbuf->pages);
+	kvfree(sgbuf->addrs);
 	kfree(sgbuf);
 }

@@ -738,9 +752,8 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	struct snd_dma_sg_fallback *sgbuf;
 	struct page **pagep, *curp;
 	size_t chunk, npages;
-	dma_addr_t addr;
+	dma_addr_t *addrp;
 	void *p;
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 
 	sgbuf = kzalloc(sizeof(*sgbuf), GFP_KERNEL);
 	if (!sgbuf)

@@ -748,14 +761,16 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	size = PAGE_ALIGN(size);
 	sgbuf->count = size >> PAGE_SHIFT;
 	sgbuf->pages = kvcalloc(sgbuf->count, sizeof(*sgbuf->pages), GFP_KERNEL);
-	if (!sgbuf->pages)
+	sgbuf->addrs = kvcalloc(sgbuf->count, sizeof(*sgbuf->addrs), GFP_KERNEL);
+	if (!sgbuf->pages || !sgbuf->addrs)
 		goto error;
 
 	pagep = sgbuf->pages;
-	chunk = size;
+	addrp = sgbuf->addrs;
+	chunk = (PAGE_SIZE - 1) << PAGE_SHIFT; /* to fit in low bits in addrs */
 	while (size > 0) {
 		chunk = min(size, chunk);
-		p = do_alloc_pages(dmab->dev.dev, chunk, &addr, wc);
+		p = dma_alloc_coherent(dmab->dev.dev, chunk, addrp, DEFAULT_GFP);
 		if (!p) {
 			if (chunk <= PAGE_SIZE)
 				goto error;

@@ -767,6 +782,8 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 		size -= chunk;
 		/* fill pages */
 		npages = chunk >> PAGE_SHIFT;
+		*addrp |= npages; /* store in lower bits */
+		addrp += npages;
 		curp = virt_to_page(p);
 		while (npages--)
 			*pagep++ = curp++;

@@ -775,6 +792,10 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	p = vmap(sgbuf->pages, sgbuf->count, VM_MAP, PAGE_KERNEL);
 	if (!p)
 		goto error;
+
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wc(sgbuf->pages, sgbuf->count);
+
 	dmab->private_data = sgbuf;
 	/* store the first page address for convenience */
 	dmab->addr = snd_sgbuf_get_addr(dmab, 0);

@@ -787,7 +808,11 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 
 static void snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab)
 {
+	struct snd_dma_sg_fallback *sgbuf = dmab->private_data;
+
 	vunmap(dmab->area);
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wb(sgbuf->pages, sgbuf->count);
 	__snd_dma_sg_fallback_free(dmab, dmab->private_data);
 }

Marek Marczykowski-Górecki

2023-01-20 12:12 UTC #22 lore raw

Quoted 184 lines · Takashi Iwai

Quoted 173 lines · Marek Marczykowski-Górecki
Quoted 171 lines · Marek Marczykowski-Górecki
Quoted 136 lines · Takashi Iwai
Quoted 134 lines · Marek Marczykowski-Górecki
Quoted 104 lines · Takashi Iwai
Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
Quoted 78 lines · Takashi Iwai
On Mon, 16 Jan 2023 16:55:11 +0100,
Takashi Iwai wrote:
On Tue, 27 Dec 2022 16:26:54 +0100,
Marek Marczykowski-Górecki wrote:
On Thu, Dec 22, 2022 at 09:09:15AM +0100, Takashi Iwai wrote:
On Sat, 10 Dec 2022 17:17:42 +0100,
Marek Marczykowski-Górecki wrote:
On Sat, Dec 10, 2022 at 02:00:06AM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
I think the hack doesn't influence on the PCM buffer pages, but only
about BDL pages. Could you check the patch below instead?
It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be
nice. Let me know if I can collect any more info that would help with
that.
Then we seem to go back again with the coherent memory allocation for
the fallback sg cases. It was changed because the use of
dma_alloc_coherent() caused a problem with IOMMU case for retrieving
the page addresses, but since the commit 9736a325137b, we essentially
avoid the fallback when IOMMU is used, so it should be fine again.
Let me know if the patch like below works for you instead of the
previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?
Yes, the one about dma_alloc_noncontiguous().
Quoted 10 lines
Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.
While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.
Quoted 3 lines
If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.
Will do.
If get_dma_ops() is non-NULL,
Yes, it's non-NULL.
Quoted 3 lines
it means we need some Xen-specific
workaround not to use dma_alloc_noncontiguous().
What's the best way to see whether the driver is running on Xen PV?
Usually it's this: cpu_feature_enabled(X86_FEATURE_XENPV)
Quoted 3 lines
Meanwhile, it's helpful if you can try the combo of my last two patches, too. It should work, and if it doesn't, it implies that we're looking at a wrong place.
It doesn't because the last of them causes "Cannot allocate memory". I'm trying now with this on top:
---8<---
sound/core/memalloc.c +1 -1
index 97d7b8106869..e927d18d1ebb 100644
@@ -545,7 +545,7 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	// sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
 	//	      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (!sgt) { // && !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)
Quoted 8 lines
sound/core/memalloc.c static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
Quoted 3 lines
sound/core/memalloc.c +0 -1
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else
---8<---
Unfortunately, the above doesn't help. I mean, I don't get an error anymore, but no sound output either (even though pavucontrol says I should hear it). So, it's like the original issue, but without any delay, just straight from the start.

Quoted 1 line

Hmm, it's the result with the combination of both patches, right?

Yes.

Quoted 1 line

What I meant as the combo is something like below.

Something like this, yes.

BTW, xen_domain() will also return true on PVH/HVM domain, which should not need any of this special treatment. It's PV that is weird.

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-20 13:20 UTC #23 lore raw

Quoted 186 lines · Marek Marczykowski-Górecki

Quoted 184 lines · Takashi Iwai
Quoted 181 lines · Marek Marczykowski-Górecki
Quoted 165 lines · Marek Marczykowski-Górecki
Quoted 136 lines · Takashi Iwai
Quoted 134 lines · Marek Marczykowski-Górecki
Quoted 104 lines · Takashi Iwai
Quoted 102 lines · Marek Marczykowski-Górecki
Quoted 100 lines · Takashi Iwai
Quoted 97 lines · Marek Marczykowski-Górecki
Quoted 86 lines · Takashi Iwai
Quoted 84 lines · Marek Marczykowski-Górecki
Quoted 81 lines · Marek Marczykowski-Górecki
On Tue, Jan 17, 2023 at 08:58:57AM +0100, Takashi Iwai wrote:
On Mon, 16 Jan 2023 16:55:11 +0100,
Takashi Iwai wrote:
On Tue, 27 Dec 2022 16:26:54 +0100,
Marek Marczykowski-Górecki wrote:
On Thu, Dec 22, 2022 at 09:09:15AM +0100, Takashi Iwai wrote:
On Sat, 10 Dec 2022 17:17:42 +0100,
Marek Marczykowski-Górecki wrote:
On Sat, Dec 10, 2022 at 02:00:06AM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 01:40:15PM +0100, Marek Marczykowski-Górecki wrote:
On Fri, Dec 09, 2022 at 09:10:19AM +0100, Takashi Iwai wrote:
On Fri, 09 Dec 2022 02:27:30 +0100,
Marek Marczykowski-Górecki wrote:
Hi,
Under Xen PV dom0, with Linux >= 5.17, sound stops working after few
hours. pavucontrol still shows meter bars moving, but the speakers
remain silent. At least on some occasions I see the following message in
dmesg:
[ 2142.484553] snd_hda_intel 0000:00:1f.3: Unstable LPIB (18144 >= 6396); disabling LPIB delay counting
Hit the issue again, this message did not appear in the log (or at least
not yet).
(...)
In anyway, please check the behavior with 6.1-rc8 + the commit
cc26516374065a34e10c9a8bf3e940e42cd96e2a
ALSA: memalloc: Allocate more contiguous pages for fallback case
from for-next of my sound git tree (which will be in 6.2-rc1).
This did not helped.
Looking at the mentioned commits, there is one specific aspect of Xen PV
that may be relevant. It configures PAT differently than native Linux.
Theoretically Linux adapts automatically and using proper API (like
set_memory_wc()) should just work, but at least for i915 driver it
causes issues (not fully tracked down yet). Details about that bug
report include some more background:
https://lore.kernel.org/intel-gfx/Y5Hst0bCxQDTN7lK@mail-itl/
Anyway, I have tested it on a Xen modified to setup PAT the same way as
native Linux and the audio issue is still there.
If the problem persists, another thing to check is the hack below
works.
Trying this one now.
And this one didn't either :/
(Sorry for the late reply, as I've been off in the last weeks.)
I think the hack doesn't influence on the PCM buffer pages, but only
about BDL pages. Could you check the patch below instead?
It'll disable the SG-buffer handling on x86 completely.
This seems to "fix" the issue, thanks!
I guess I'll run it this way for now, but a proper solution would be
nice. Let me know if I can collect any more info that would help with
that.
Then we seem to go back again with the coherent memory allocation for
the fallback sg cases. It was changed because the use of
dma_alloc_coherent() caused a problem with IOMMU case for retrieving
the page addresses, but since the commit 9736a325137b, we essentially
avoid the fallback when IOMMU is used, so it should be fine again.
Let me know if the patch like below works for you instead of the
previous hack to disable SG-buffer (note: totally untested!)
Gah, there was an obvious typo, scratch that.
Below is a proper patch. Please try this one instead.
Thanks, I'll give it a try.
Quoted 7 lines
Quoted 2 lines
Unfortunately, it doesn't help, it stopped working again, after about 3h uptime.
Aha, then it might be rather other way round; dma_alloc_noncontiguous() doesn't work on Xen properly.
Could you try the one below instead of the previous?
Unfortunately, this one doesn't fix it either :/
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Hmm. Then how about applying both of the last two patches? The last one to enforce the fallback allocation and the previous one to use dma_alloc_coherent(). It should be essentially reverting to the old way.
Oh, I noticed only now: the last patch made it fail to initialize.
The "last patch" means the patch to enforce the fallback allocation?
Yes, the one about dma_alloc_noncontiguous().
Quoted 10 lines
Quoted 6 lines
I
don't see obvious errors in dmesg, but when trying aplay, I get:
ALSA lib pcm_direct.c:1284:(snd1_pcm_direct_initialize_slave) unable to install hw params ALSA lib pcm_dmix.c:1087:(snd_pcm_dmix_open) unable to initialize slave
aplay: main:830: audio open error: Cannot allocate memory
It's -ENOMEM, so it must be from there. Does it appear always? If yes, your system is with IOMMU, and the patch made return always NULL intentionally.
While the system do have IOMMU, it isn't configured by Linux, but by Xen. And it maps all the memory that Linux see.
Quoted 3 lines
If that's the case, the problem is that IOMMU doesn't handle the coherent memory on Xen.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Please check more explicitly, whether get_dma_ops(dmab->dev.dev) call in snd_dma_noncontig_alloc() returns NULL or not.
Will do.
If get_dma_ops() is non-NULL,
Yes, it's non-NULL.
Quoted 3 lines
it means we need some Xen-specific
workaround not to use dma_alloc_noncontiguous().
What's the best way to see whether the driver is running on Xen PV?
Usually it's this: cpu_feature_enabled(X86_FEATURE_XENPV)
Quoted 3 lines
Meanwhile, it's helpful if you can try the combo of my last two patches, too. It should work, and if it doesn't, it implies that we're looking at a wrong place.
It doesn't because the last of them causes "Cannot allocate memory". I'm trying now with this on top:
---8<---
sound/core/memalloc.c +0 -0
index 97d7b8106869..e927d18d1ebb 100644
sound/core/memalloc.c static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
Quoted 9 lines
sound/core/memalloc.c +1 -2
@@ -545,7 +545,7 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	// sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
 	//	      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (!sgt) { // && !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else
---8<---
Unfortunately, the above doesn't help. I mean, I don't get an error anymore, but no sound output either (even though pavucontrol says I should hear it). So, it's like the original issue, but without any delay, just straight from the start.

Quoted 7 lines

Quoted 1 line
Hmm, it's the result with the combination of both patches, right?
Yes.
Quoted 1 line
What I meant as the combo is something like below.
Something like this, yes.

It's puzzling, then. The patch changes the allocation with the dma_alloc_coherent(), and that's what does with the Kconfig hack you've tested. One possible significant difference is the use of the DMA address.

Quoted 2 lines

BTW, xen_domain() will also return true on PVH/HVM domain, which should not need any of this special treatment. It's PV that is weird.

OK, then it can be an overkill.

Below is another try: it changes the different use of the DMA buffer address. Let's cross fingers.

thanks,

Takashi

-- 8< --

sound/core/memalloc.c +52 ~~-15~~

@@ -541,10 +541,9 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 	struct sg_table *sgt;
 	void *p;
 
-	sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
-				      DEFAULT_GFP, 0);
 #ifdef CONFIG_SND_DMA_SGBUF
-	if (!sgt && !get_dma_ops(dmab->dev.dev)) {
+	if (cpu_feature_enabled(X86_FEATURE_XENPV) ||
+	    !get_dma_ops(dmab->dev.dev)) {
 		if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG)
 			dmab->dev.type = SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 		else

@@ -552,6 +551,8 @@ static void *snd_dma_noncontig_alloc(struct snd_dma_buffer *dmab, size_t size)
 		return snd_dma_sg_fallback_alloc(dmab, size);
 	}
 #endif
+	sgt = dma_alloc_noncontiguous(dmab->dev.dev, size, dmab->dev.dir,
+				      DEFAULT_GFP, 0);
 	if (!sgt)
 		return NULL;

@@ -719,17 +720,30 @@ static const struct snd_malloc_ops snd_dma_sg_wc_ops = {
 struct snd_dma_sg_fallback {
 	size_t count;
 	struct page **pages;
+	dma_addr_t *addrs;
 };
 
 static void __snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab,
 				       struct snd_dma_sg_fallback *sgbuf)
 {
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
-	size_t i;
-
-	for (i = 0; i < sgbuf->count && sgbuf->pages[i]; i++)
-		do_free_pages(page_address(sgbuf->pages[i]), PAGE_SIZE, wc);
+	size_t i, size;
+
+	if (sgbuf->pages && sgbuf->addrs) {
+		i = 0;
+		while (i < sgbuf->count) {
+			if (!sgbuf->pages[i] || !sgbuf->addrs[i])
+				break;
+			size = sgbuf->addrs[i] & ~PAGE_MASK;
+			if (WARN_ON(!size))
+				break;
+			dma_free_coherent(dmab->dev.dev, size,
+					  page_address(sgbuf->pages[i]),
+					  sgbuf->addrs[i] & PAGE_MASK);
+			i += size;
+		}
+	}
 	kvfree(sgbuf->pages);
+	kvfree(sgbuf->addrs);
 	kfree(sgbuf);
 }

@@ -738,9 +752,9 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	struct snd_dma_sg_fallback *sgbuf;
 	struct page **pagep, *curp;
 	size_t chunk, npages;
+	dma_addr_t *addrp;
 	dma_addr_t addr;
 	void *p;
-	bool wc = dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK;
 
 	sgbuf = kzalloc(sizeof(*sgbuf), GFP_KERNEL);
 	if (!sgbuf)

@@ -748,14 +762,16 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 	size = PAGE_ALIGN(size);
 	sgbuf->count = size >> PAGE_SHIFT;
 	sgbuf->pages = kvcalloc(sgbuf->count, sizeof(*sgbuf->pages), GFP_KERNEL);
-	if (!sgbuf->pages)
+	sgbuf->addrs = kvcalloc(sgbuf->count, sizeof(*sgbuf->addrs), GFP_KERNEL);
+	if (!sgbuf->pages || !sgbuf->addrs)
 		goto error;
 
 	pagep = sgbuf->pages;
-	chunk = size;
+	addrp = sgbuf->addrs;
+	chunk = (PAGE_SIZE - 1) << PAGE_SHIFT; /* to fit in low bits in addrs */
 	while (size > 0) {
 		chunk = min(size, chunk);
-		p = do_alloc_pages(dmab->dev.dev, chunk, &addr, wc);
+		p = dma_alloc_coherent(dmab->dev.dev, chunk, &addr, DEFAULT_GFP);
 		if (!p) {
 			if (chunk <= PAGE_SIZE)
 				goto error;

@@ -767,17 +783,25 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 		size -= chunk;
 		/* fill pages */
 		npages = chunk >> PAGE_SHIFT;
+		*addrp = npages; /* store in lower bits */
 		curp = virt_to_page(p);
-		while (npages--)
+		while (npages--) {
 			*pagep++ = curp++;
+			*addrp++ |= addr;
+			addr += PAGE_SIZE;
+		}
 	}
 
 	p = vmap(sgbuf->pages, sgbuf->count, VM_MAP, PAGE_KERNEL);
 	if (!p)
 		goto error;
+
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wc(sgbuf->pages, sgbuf->count);
+
 	dmab->private_data = sgbuf;
 	/* store the first page address for convenience */
-	dmab->addr = snd_sgbuf_get_addr(dmab, 0);
+	dmab->addr = sgbuf->addrs[0] & PAGE_MASK;
 	return p;
 
  error:

@@ -787,10 +811,23 @@ static void *snd_dma_sg_fallback_alloc(struct snd_dma_buffer *dmab, size_t size)
 
 static void snd_dma_sg_fallback_free(struct snd_dma_buffer *dmab)
 {
+	struct snd_dma_sg_fallback *sgbuf = dmab->private_data;
+
 	vunmap(dmab->area);
+	if (dmab->dev.type == SNDRV_DMA_TYPE_DEV_WC_SG_FALLBACK)
+		set_pages_array_wb(sgbuf->pages, sgbuf->count);
 	__snd_dma_sg_fallback_free(dmab, dmab->private_data);
 }
 
+static dma_addr_t snd_dma_sg_fallback_get_addr(struct snd_dma_buffer *dmab,
+					       size_t offset)
+{
+	struct snd_dma_sg_fallback *sgbuf = dmab->private_data;
+	size_t index = offset >> PAGE_SHIFT;
+
+	return (sgbuf->addrs[index] & PAGE_MASK) | (offset & ~PAGE_MASK);
+}
+
 static int snd_dma_sg_fallback_mmap(struct snd_dma_buffer *dmab,
 				    struct vm_area_struct *area)
 {

@@ -805,8 +842,8 @@ static const struct snd_malloc_ops snd_dma_sg_fallback_ops = {
 	.alloc = snd_dma_sg_fallback_alloc,
 	.free = snd_dma_sg_fallback_free,
 	.mmap = snd_dma_sg_fallback_mmap,
+	.get_addr = snd_dma_sg_fallback_get_addr,
 	/* reuse vmalloc helpers */
-	.get_addr = snd_dma_vmalloc_get_addr,
 	.get_page = snd_dma_vmalloc_get_page,
 	.get_chunk_size = snd_dma_vmalloc_get_chunk_size,
 };

Marek Marczykowski-Górecki

2023-01-23 21:32 UTC #24 lore raw

Quoted 8 lines · Takashi Iwai

Quoted 1 line · Marek Marczykowski-Górecki
It's puzzling, then. The patch changes the allocation with the dma_alloc_coherent(), and that's what does with the Kconfig hack you've tested. One possible significant difference is the use of the DMA address.

Quoted 7 lines

Quoted 2 lines
BTW, xen_domain() will also return true on PVH/HVM domain, which should not need any of this special treatment. It's PV that is weird.
OK, then it can be an overkill.
Below is another try: it changes the different use of the DMA buffer address. Let's cross fingers.

3 days update and it still works!

Signature · 3 lines

Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Takashi Iwai

author

2023-01-24 09:25 UTC #25 lore raw

Quoted 13 lines · Marek Marczykowski-Górecki

Quoted 11 lines · Takashi Iwai
Quoted 1 line · Marek Marczykowski-Górecki
It's puzzling, then. The patch changes the allocation with the dma_alloc_coherent(), and that's what does with the Kconfig hack you've tested. One possible significant difference is the use of the DMA address.
Quoted 2 lines
BTW, xen_domain() will also return true on PVH/HVM domain, which should not need any of this special treatment. It's PV that is weird.

Quoted 6 lines

Quoted 4 lines
OK, then it can be an overkill.
Below is another try: it changes the different use of the DMA buffer address. Let's cross fingers.
3 days update and it still works!

Great, I'm going to submit the proper patches, then. Thanks for your patient testing!

Takashi