Skip to contentSkip to conversation

ALSA: hda: Abort capability probe at invalid register read

PATCH In review Takashi Iwai 3 participants 9 messages Oct 17, 2017 – Oct 18, 2017 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ALSA: hda: Abort capability probe at invalid register read

The loop in snd_hdac_bus_parse_capabilities() may go to nirvana when it hits an invalid register value read:

 BUG: unable to handle kernel paging request at ffffad5dc41f3fff
 IP: pci_azx_readl+0x5/0x10 [snd_hda_intel]
 Call Trace:
  snd_hdac_bus_parse_capabilities+0x3c/0x1f0 [snd_hda_core]
  azx_probe_continue+0x7d5/0x940 [snd_hda_intel]
  .....

This happened on a new Intel machine, and we need to check the value and abort the loop accordingly.

[Note: the fixes tag below indicates only the commit where this patch
can be applied; the original problem was introduced even before that commit]

Fixes 6720b38420a0 ("ALSA: hda - move bus_parse_capabilities to core")
---
 sound/hda/hdac_controller.c | 5 +++++
 1 file changed, 5 insertions(+)
sound/hda/hdac_controller.c +5 -0
index 978dc1801b3a..f6d2985b2520 100644
@@ -284,6 +284,11 @@ int snd_hdac_bus_parse_capabilities(struct hdac_bus *bus)
 		dev_dbg(bus->dev, "HDA capability ID: 0x%x\n",
 			(cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF);
 
+		if (cur_cap == -1) {
+			dev_dbg(bus->dev, "Invalid capability reg read\n");
+			break;
+		}
+
 		switch ((cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF) {
 		case AZX_ML_CAP_ID:
 			dev_dbg(bus->dev, "Found ML capability\n");
Signature · 1 line
2.14.2
#1 lore raw

Review

Quoted 4 lines
(quoted diff) +4 -0
+		if (cur_cap == -1) {
+			dev_dbg(bus->dev, "Invalid capability reg read\n");
+			break;
+		}

Since we already have a default case in the current code, can we use it ? I am sending you a untested patch, as proposal.

From 1690624051a75b95592fb1c497a0a8844cbd1c64 Mon Sep 17 00:00:00 2001
From: Rakesh Ughreja <[email protected]>
Date: Wed, 18 Oct 2017 04:46:46 +0530
Subject: [PATCH] ALSA: hda: Abort capability probe at invalid register read

In the case of unknown capability break the loop by setting the current capability to zero

---
 sound/hda/hdac_controller.c | 1 +
 1 file changed, 1 insertion(+)
sound/hda/hdac_controller.c +1 -0
index 978dc18..86c2b4c 100644
@@ -315,6 +315,7 @@ int snd_hdac_bus_parse_capabilities(struct hdac_bus *bus)
 
 		default:
 			dev_dbg(bus->dev, "Unknown capability %d\n", cur_cap);
+			cur_cap = 0;
 			break;
 		}
 
Signature · 1 line
2.7.4
#2 lore raw
Quoted 8 lines · Rakesh A

Review

Quoted 5 lines
(quoted diff) +4 -0
+		if (cur_cap == -1) {
+			dev_dbg(bus->dev, "Invalid capability reg read\n");
+			break;
+		}

Since we already have a default case in the current code, can we use it ? I am sending you a untested patch, as proposal.

In general, the value -1 indicates a clear error, and it has to be handled alone, not checking a part of the bits.

I'm fine to add your change in addition, of course. And, I believe it'd be better to make dev_dbg() to dev_err() or dev_warn() in the default case, since the situation is really unexpected.

That being said, feel free to send the patch to change that. I'll apply it to for-next branch, while my fix would go into for-linus as a quick fix.

thanks,

Takashi

#3 lore raw
Quoted 4 lines

That being said, feel free to send the patch to change that. I'll apply it to for-next branch, while my fix would go into for-linus as a quick fix.

Sent a patch to you.

Regards,
Rakesh

#4 lore raw
In reply to Takashi Iwai
Quoted 10 lines · Takashi Iwai

The loop in snd_hdac_bus_parse_capabilities() may go to nirvana when it hits an invalid register value read:

 BUG: unable to handle kernel paging request at ffffad5dc41f3fff
 IP: pci_azx_readl+0x5/0x10 [snd_hda_intel]
 Call Trace:
  snd_hdac_bus_parse_capabilities+0x3c/0x1f0 [snd_hda_core]
  azx_probe_continue+0x7d5/0x940 [snd_hda_intel]
  .....
Quoted 2 lines

This happened on a new Intel machine, and we need to check the value and abort the loop accordingly.

okay and what machine is the problem here. I have had a similar bug report from Gfx CI guys on CFL machine. Turns out the BIOS was buggy and we fixed that up by upgrading the BIOS.

Yes it is a good idea to keep this guard but -1 would mean that HW read is failing which points to some other issue here

Quoted 31 lines

[Note: the fixes tag below indicates only the commit where this patch
can be applied; the original problem was introduced even before that commit]

Fixes 6720b38420a0 ("ALSA: hda - move bus_parse_capabilities to core")
---
 sound/hda/hdac_controller.c | 5 +++++
 1 file changed, 5 insertions(+)
sound/hda/hdac_controller.c +5 -0
index 978dc1801b3a..f6d2985b2520 100644
@@ -284,6 +284,11 @@ int snd_hdac_bus_parse_capabilities(struct hdac_bus *bus)
 		dev_dbg(bus->dev, "HDA capability ID: 0x%x\n",
 			(cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF);
 
+		if (cur_cap == -1) {
+			dev_dbg(bus->dev, "Invalid capability reg read\n");
+			break;
+		}
+
 		switch ((cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF) {
 		case AZX_ML_CAP_ID:
 			dev_dbg(bus->dev, "Found ML capability\n");
Signature · 1 line
2.14.2
Signature · 1 line
~Vinod
#5 lore raw
Quoted 12 lines · Vinod Koul
Quoted 10 lines · Takashi Iwai

The loop in snd_hdac_bus_parse_capabilities() may go to nirvana when it hits an invalid register value read:

 BUG: unable to handle kernel paging request at ffffad5dc41f3fff
 IP: pci_azx_readl+0x5/0x10 [snd_hda_intel]
 Call Trace:
  snd_hdac_bus_parse_capabilities+0x3c/0x1f0 [snd_hda_core]
  azx_probe_continue+0x7d5/0x940 [snd_hda_intel]
  .....
Quoted 6 lines
Quoted 2 lines

This happened on a new Intel machine, and we need to check the value and abort the loop accordingly.

okay and what machine is the problem here. I have had a similar bug report from Gfx CI guys on CFL machine. Turns out the BIOS was buggy and we fixed that up by upgrading the BIOS.

Yes, it's a CFL-H. Possibly a buggy BIOS, but the driver still shouldn't crash.

Quoted 2 lines

Yes it is a good idea to keep this guard but -1 would mean that HW read is failing which points to some other issue here

Right.

Takashi

Quoted 34 lines
Quoted 30 lines

[Note: the fixes tag below indicates only the commit where this patch
can be applied; the original problem was introduced even before that commit]

Fixes 6720b38420a0 ("ALSA: hda - move bus_parse_capabilities to core")
---
 sound/hda/hdac_controller.c | 5 +++++
 1 file changed, 5 insertions(+)
sound/hda/hdac_controller.c +5 -0
index 978dc1801b3a..f6d2985b2520 100644
@@ -284,6 +284,11 @@ int snd_hdac_bus_parse_capabilities(struct hdac_bus *bus)
 		dev_dbg(bus->dev, "HDA capability ID: 0x%x\n",
 			(cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF);
 
+		if (cur_cap == -1) {
+			dev_dbg(bus->dev, "Invalid capability reg read\n");
+			break;
+		}
+
 		switch ((cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF) {
 		case AZX_ML_CAP_ID:
 			dev_dbg(bus->dev, "Found ML capability\n");
Signature · 1 line
2.14.2
Signature · 1 line
~Vinod
#6 lore raw
Quoted 17 lines · Takashi Iwai
Quoted 15 lines · Vinod Koul
Quoted 12 lines · Takashi Iwai

The loop in snd_hdac_bus_parse_capabilities() may go to nirvana when it hits an invalid register value read:

 BUG: unable to handle kernel paging request at ffffad5dc41f3fff
 IP: pci_azx_readl+0x5/0x10 [snd_hda_intel]
 Call Trace:
  snd_hdac_bus_parse_capabilities+0x3c/0x1f0 [snd_hda_core]
  azx_probe_continue+0x7d5/0x940 [snd_hda_intel]
  .....

This happened on a new Intel machine, and we need to check the value and abort the loop accordingly.

Quoted 6 lines
Quoted 3 lines

okay and what machine is the problem here. I have had a similar bug report from Gfx CI guys on CFL machine. Turns out the BIOS was buggy and we fixed that up by upgrading the BIOS.

Yes, it's a CFL-H. Possibly a buggy BIOS, but the driver still shouldn't crash.

Okay so can you ask them to update BIOS and check.

Quoted 4 lines
Quoted 2 lines

Yes it is a good idea to keep this guard but -1 would mean that HW read is failing which points to some other issue here

Right.

In this case should we send this to stable? I have not seen this crashing till now except bad BIOS issue

Quoted 38 lines

Takashi

Quoted 34 lines
Quoted 30 lines

[Note: the fixes tag below indicates only the commit where this patch
can be applied; the original problem was introduced even before that commit]

Fixes 6720b38420a0 ("ALSA: hda - move bus_parse_capabilities to core")
---
 sound/hda/hdac_controller.c | 5 +++++
 1 file changed, 5 insertions(+)
sound/hda/hdac_controller.c +5 -0
index 978dc1801b3a..f6d2985b2520 100644
@@ -284,6 +284,11 @@ int snd_hdac_bus_parse_capabilities(struct hdac_bus *bus)
 		dev_dbg(bus->dev, "HDA capability ID: 0x%x\n",
 			(cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF);
 
+		if (cur_cap == -1) {
+			dev_dbg(bus->dev, "Invalid capability reg read\n");
+			break;
+		}
+
 		switch ((cur_cap & AZX_CAP_HDR_ID_MASK) >> AZX_CAP_HDR_ID_OFF) {
 		case AZX_ML_CAP_ID:
 			dev_dbg(bus->dev, "Found ML capability\n");
Signature · 1 line
2.14.2
Signature · 1 line
~Vinod
Signature · 1 line
~Vinod
#7 lore raw
Quoted 28 lines · Vinod Koul
Quoted 17 lines · Takashi Iwai
Quoted 15 lines · Vinod Koul
Quoted 12 lines · Takashi Iwai

The loop in snd_hdac_bus_parse_capabilities() may go to nirvana when it hits an invalid register value read:

 BUG: unable to handle kernel paging request at ffffad5dc41f3fff
 IP: pci_azx_readl+0x5/0x10 [snd_hda_intel]
 Call Trace:
  snd_hdac_bus_parse_capabilities+0x3c/0x1f0 [snd_hda_core]
  azx_probe_continue+0x7d5/0x940 [snd_hda_intel]
  .....

This happened on a new Intel machine, and we need to check the value and abort the loop accordingly.

Quoted 6 lines
Quoted 3 lines

okay and what machine is the problem here. I have had a similar bug report from Gfx CI guys on CFL machine. Turns out the BIOS was buggy and we fixed that up by upgrading the BIOS.

Yes, it's a CFL-H. Possibly a buggy BIOS, but the driver still shouldn't crash.

Okay so can you ask them to update BIOS and check.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

Yes it is a good idea to keep this guard but -1 would mean that HW read is failing which points to some other issue here

Right.

In this case should we send this to stable? I have not seen this crashing till now except bad BIOS issue

People will be getting test hardware now and see the Oops. We can't guarantee the sane BIOS, and obviously the current code does crash easily, and yet the code fix is trivial -- a perfect situation for stable :)

Takashi

Acked-by
#8 lore raw
Quoted 35 lines · Takashi Iwai
Quoted 33 lines · Vinod Koul
Quoted 17 lines · Takashi Iwai
Quoted 15 lines · Vinod Koul
Quoted 12 lines · Takashi Iwai

The loop in snd_hdac_bus_parse_capabilities() may go to nirvana when it hits an invalid register value read:

 BUG: unable to handle kernel paging request at ffffad5dc41f3fff
 IP: pci_azx_readl+0x5/0x10 [snd_hda_intel]
 Call Trace:
  snd_hdac_bus_parse_capabilities+0x3c/0x1f0 [snd_hda_core]
  azx_probe_continue+0x7d5/0x940 [snd_hda_intel]
  .....

This happened on a new Intel machine, and we need to check the value and abort the loop accordingly.

Quoted 6 lines
Quoted 3 lines

okay and what machine is the problem here. I have had a similar bug report from Gfx CI guys on CFL machine. Turns out the BIOS was buggy and we fixed that up by upgrading the BIOS.

Yes, it's a CFL-H. Possibly a buggy BIOS, but the driver still shouldn't crash.

Okay so can you ask them to update BIOS and check.

Quoted 4 lines
Quoted 2 lines

Yes it is a good idea to keep this guard but -1 would mean that HW read is failing which points to some other issue here

Right.

Quoted 7 lines
Quoted 2 lines

In this case should we send this to stable? I have not seen this crashing till now except bad BIOS issue

People will be getting test hardware now and see the Oops. We can't guarantee the sane BIOS, and obviously the current code does crash easily, and yet the code fix is trivial -- a perfect situation for stable :)

Right :), But do make sure to ask for BIOS update on that board.

Signature · 1 line
~Vinod