Skip to contentSkip to conversation

ASoC: core: add snd_BUG_ON() check in devm_snd_soc_register_card()

PATCH In review [email protected] 2 participants 7 messages Aug 3 – Aug 4

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ASoC: core: add snd_BUG_ON() check in devm_snd_soc_register_card()

From: bui duc phuc <[email protected]>

card is dereferenced immediately, so passing NULL results in a NULL pointer dereference. A NULL dev is silently accepted, defeating the documented automatic-unregistration semantics of this devm variant.

Add snd_BUG_ON() checks on both arguments to catch API misuse early.

---
 sound/soc/soc-devres.c | 3 +++
 1 file changed, 3 insertions(+)
sound/soc/soc-devres.c +3 -0
index 718165ba84ac..3e946196022a 100644
@@ -59,6 +59,9 @@ EXPORT_SYMBOL_GPL(devm_snd_soc_register_component);
  */
 int devm_snd_soc_register_card(struct device *dev, struct snd_soc_card *card)
 {
+	if (snd_BUG_ON(!dev || !card))
+		return -EINVAL;
+
 	card->devres_dev = dev;
 	return snd_soc_register_card(card);
 }
Signature · 1 line
2.43.0
#1 lore raw
Quoted 18 lines · On 8/3/2026 12:27 PM, [email protected] wrote:

From: bui duc phuc <[email protected]>

card is dereferenced immediately, so passing NULL results in a NULL pointer dereference. A NULL dev is silently accepted, defeating the documented automatic-unregistration semantics of this devm variant.

Add snd_BUG_ON() checks on both arguments to catch API misuse early.

---
  sound/soc/soc-devres.c | 3 +++
  1 file changed, 3 insertions(+)
sound/soc/soc-devres.c +0 -0
index 718165ba84ac..3e946196022a 100644

sound/soc/soc-devres.c EXPORT_SYMBOL_GPL(devm_snd_soc_register_component); · line 67

Quoted 9 lines
sound/soc/soc-devres.c +3 -0
@@ -59,6 +59,9 @@ EXPORT_SYMBOL_GPL(devm_snd_soc_register_component);
   */
  int devm_snd_soc_register_card(struct device *dev, struct snd_soc_card *card)
  {
+	if (snd_BUG_ON(!dev || !card))
+		return -EINVAL;
+
  	card->devres_dev = dev;
  	return snd_soc_register_card(card);
  }

Not sure whether we need such defenses. If we are to follow such approach, many ASoC-driver APIs require BUG_ON/WARN_ON() update.

register_card() is typically called as a last step, once 'card' is already allocated. At the same time, the register_card() typically occurs in probe() context of given device so we're always certain the 'dev' is not NULL.

TLDR: this is not a UAPI, (obvious) NULL-checks can be skipped.

#2 lore raw

Hi Cezary,

Thank you for your review.

Quoted 10 lines

Not sure whether we need such defenses. If we are to follow such approach, many ASoC-driver APIs require BUG_ON/WARN_ON() update.

register_card() is typically called as a last step, once 'card' is already allocated. At the same time, the register_card() typically occurs in probe() context of given device so we're always certain the 'dev' is not NULL.

TLDR: this is not a UAPI, (obvious) NULL-checks can be skipped.

I completely understand where you're coming from.
My thinking was that this function is referenced quite widely (currently in 112 files).
Given how frequently it is used, I felt that adding a simple defensive check here
could make the API a little more robust and help avoid potential NULL pointer dereferences for both existing and future callers.

I wasn't suggesting that every ASoC API should be updated this way. I was only thinking of APIs that are used very frequently, especially those commonly called from probe paths.

Also, it's not always obvious to callers what assumptions a core API relies on or
the context in which it is expected to be used.

That was simply the motivation behind this patch.

Best regards,
Phuc

#3 lore raw
Quoted 22 lines · Bui Duc Phuc
Quoted 9 lines

Not sure whether we need such defenses. If we are to follow such approach, many ASoC-driver APIs require BUG_ON/WARN_ON() update.

register_card() is typically called as a last step, once 'card' is already allocated. At the same time, the register_card() typically occurs in probe() context of given device so we're always certain the 'dev' is not NULL.

TLDR: this is not a UAPI, (obvious) NULL-checks can be skipped.

I completely understand where you're coming from.
My thinking was that this function is referenced quite widely (currently in 112 files).
Given how frequently it is used, I felt that adding a simple defensive check here
could make the API a little more robust and help avoid potential NULL pointer dereferences for both existing and future callers.

I wasn't suggesting that every ASoC API should be updated this way. I was only thinking of APIs that are used very frequently, especially those commonly called from probe paths.

Quoted 5 lines

Also, it's not always obvious to callers what assumptions a core API relies on or
the context in which it is expected to be used.

That was simply the motivation behind this patch.

snd_BUG_ON() translates to WARN_ON() in debug conditions. After reading the macro documentation, pr_xxx() or equivalents are recommended when dealing with invalid arguments. Same results after grepping for WARNs in sound/soc or in general references such as driver/base/core.c - it's clear to me the macro is not used as plain null-arg-check.

Updating just one function with the check raises the question when reading the file - Why just one devm_snd_xxx() has the check and the rest do not?

Kind regards,
Czarek

#4 lore raw

Hi Cezary,

Thank you for your feedback.

Quoted 6 lines

snd_BUG_ON() translates to WARN_ON() in debug conditions. After reading the macro documentation, pr_xxx() or equivalents are recommended when dealing with invalid arguments. Same results after grepping for WARNs in sound/soc or in general references such as driver/base/core.c - it's clear to me the macro is not used as plain null-arg-check.

I couldn't find anything in the snd_BUG_ON() documentation that explicitly says the
macro should not be used for plain NULL argument checks.

Also, there are quite a few places in the sound subsystem where WARN_ON() is used to
validate NULL arguments, for example:

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/usb/endpoint.c#L857

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/soc/ti/davinci-mcasp.c#L2476

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/soc/meson/meson-codec-glue.c#L59

and even in the Intel code that you maintain:

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/soc/intel/atom/sst-mfld-platform-pcm.c#L31

That said, I'm also fine with not using snd_BUG_ON() here. Replacing it with a simple argument check such as:

if (!dev || !card) {
      pr_err("invalid arguments\n");
      return -EINVAL;
}

is perfectly reasonable to me. It's not a big issue either way.

Quoted 4 lines

Updating just one function with the check raises the question when reading the file - Why just one devm_snd_xxx() has the check and the rest do not?

I think the remaining helper functions can be updated gradually over time.

To me, this discussion itself is a good example. You and I interpreted the intended use of snd_BUG_ON() differently.
I don't think the macro is wrong; rather, its intended usage is not clearly documented or consistently reflected by existing code. Therefore, I think it would be beneficial to either improve the API documentation to clarify the intended usage,
or add explicit argument validation, such as the check I proposed for devm_snd_soc_register_card().
Either approach would help avoid misunderstandings and make the API easier to use correctly.

Of course, that's just my opinion. If the SOUND subsystem does not intend to make these APIs more explicit or easier to use, that's perfectly fine as well. I'll follow the preferred direction.

Best regards,
Phuc

#5 lore raw
Quoted 10 lines · Bui Duc Phuc
Quoted 6 lines

snd_BUG_ON() translates to WARN_ON() in debug conditions. After reading the macro documentation, pr_xxx() or equivalents are recommended when dealing with invalid arguments. Same results after grepping for WARNs in sound/soc or in general references such as driver/base/core.c - it's clear to me the macro is not used as plain null-arg-check.

I couldn't find anything in the snd_BUG_ON() documentation that explicitly says the
macro should not be used for plain NULL argument checks.

My message reads: translates to WARN_ON() (..)
And WARN_ON's doc is what I meant.

Quoted 9 lines

Also, there are quite a few places in the sound subsystem where WARN_ON() is used to
validate NULL arguments, for example:

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/usb/endpoint.c#L857

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/soc/ti/davinci-mcasp.c#L2476

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/soc/meson/meson-codec-glue.c#L59

Nah, these are not plain null checks and if I missed something and any of them are, that's a patch candidate.

Quoted 3 lines

and even in the Intel code that you maintain:

https://elixir.bootlin.com/linux/v7.2-rc5/source/sound/soc/intel/atom/sst-mfld-platform-pcm.c#L31

That's a clear example of when _not to_ use WARN_ON(), good catch. :)
Maintain != reviewed/acked. Many maintainers maintain code that they inherited and were not there when it was merged.

Quoted 9 lines

That said, I'm also fine with not using snd_BUG_ON() here. Replacing it with a simple argument check such as:

if (!dev || !card) {
       pr_err("invalid arguments\n");
       return -EINVAL;
}

is perfectly reasonable to me. It's not a big issue either way.

While not a fan, still better than WARN_ON().

Quoted 11 lines
Quoted 4 lines

Updating just one function with the check raises the question when reading the file - Why just one devm_snd_xxx() has the check and the rest do not?

I think the remaining helper functions can be updated gradually over time.

To me, this discussion itself is a good example. You and I interpreted the intended use of snd_BUG_ON() differently.
I don't think the macro is wrong; rather, its intended usage is not clearly documented or consistently reflected by existing code.

Not a fan of snd_BUG_ON() in general, I'd rather have it removed and WARN_ON(s) updated is one believes something is missing or wrong with them.

Anyhow, it's up to Mark. I'd avoid selecting just one function and filling it with null-arg-checks.

#6 lore raw

Hi Cezary,

Quoted 4 lines

My message reads: translates to WARN_ON() (..)
And WARN_ON's doc is what I meant.

Ah, I see. I had interpreted your comment differently. Thanks for the clarification.

Best regards,
Phuc