Repository navigation
Conversation
This patch uses rtd instead of pcm at snd_soc_pcm_component_new/free() parameter. This is prepare for dai_link remove bug fix on topology. Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
soc-topology adds extra dai_link by using snd_soc_add_dai_link(),
and removes it by snd_soc_romove_dai_link().
This snd_soc_add/remove_dai_link() and/or its related
functions are unbalanced before, and now, these are balance-uped.
But, it finds the random operation issue, and it is reported by
Pierre-Louis.
When card was released, topology will call snd_soc_remove_dai_link()
via (A).
static void soc_cleanup_card_resources(struct snd_soc_card *card)
{
struct snd_soc_dai_link *link, *_link;
/* This should be called before snd_card_free() */
(A) soc_remove_link_components(card);
/* free the ALSA card at first; this syncs with pending operations */
if (card->snd_card) {
(B) snd_card_free(card->snd_card);
card->snd_card = NULL;
}
/* remove and free each DAI */
(X) soc_remove_link_dais(card);
for_each_card_links_safe(card, link, _link)
(C) snd_soc_remove_dai_link(card, link);
...
}
At (A), topology calls snd_soc_remove_dai_link().
Then topology rtd, and its related all data are freed.
Next, (B) is called, and then, pcm->private_free = soc_pcm_private_free()
is called.
static void soc_pcm_private_free(struct snd_pcm *pcm)
{
struct snd_soc_pcm_runtime *rtd = pcm->private_data;
/* need to sync the delayed work before releasing resources */
flush_delayed_work(&rtd->delayed_work);
snd_soc_pcm_component_free(rtd);
}
Here, it gets rtd via pcm->private_data.
But, topology related rtd are already freed at (A).
Normal sound card has no damage, becase it frees rtd at (C).
These are finalizing rtd related data.
Thus, these should be called when rtd was freed, not sound card
was freed. It is very natural and understandable.
In other words, pcm->private_free = soc_pcm_private_free()
is no longer needed.
Extra issue is that there is zero chance to call
soc_remove_dai() for topology related dai at (X).
Because (A) removes rtd connection from card too, and,
(X) is based on card connected rtd.
This means, (X) need to be called before (C) (= for normal sound)
and (A) (= for topology).
Now, I want to focus this patch which is the reason why
snd_card_free() = (B) is located there.
commit 4efda5f
("ASoC: Fix use-after-free at card unregistration")
Original snd_card_free() was called last of this function.
But moved to top to avoid use-after-free issue.
The issue was happen at soc_pcm_free() which was pcm->private_free,
today it is updated/renamed to soc_pcm_private_free().
In other words, (B) need to be called before (C) (= for normal sound)
and (A) (= for topology), because it needs (not yet freed) rtd.
But, (A) need to be called before (B),
because it needs card->snd_card pointer.
If we call flush_delayed_work() and snd_soc_pcm_component_free()
(= same as soc_pcm_private_free()) when rtd was freed (= (C), (A)),
there is no reason to call snd_card_free() at top of this function.
It can be called end of this function, again.
But, in such case, it will likely break unbind again, as Takashi-san
reported. When unbind is performed in a busy state, the code may
release still-in-use resources.
At least we need to call snd_card_disconnect_sync() at the first place.
The final code will be...
static void soc_cleanup_card_resources(struct snd_soc_card *card)
{
struct snd_soc_dai_link *link, *_link;
if (card->snd_card)
(Z) snd_card_disconnect_sync(card->snd_card);
(X) soc_remove_link_dais(card);
(A) soc_remove_link_components(card);
for_each_card_links_safe(card, link, _link)
(C) snd_soc_remove_dai_link(card, link);
...
if (card->snd_card) {
(B) snd_card_free(card->snd_card);
card->snd_card = NULL;
}
}
To avoid release still-in-use resources,
call snd_card_disconnect_sync() at (Z).
(X) is needed for both non-topology and topology.
topology removes rtd via (A), and
non topology removes rtd via (C).
snd_card_free() is no longer related to use-after-free issue.
Thus, locating (B) is no problem.
Fixes: df95a16 ("ASoC: soc-core: fix RIP warning on card removal")
Fixes: bc7a909 ("ASoC: soc-core: add soc_unbind_dai_link()")
Reported-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Tested-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Collaborator
|
Sure, I will start stress-test on BYT/BSW first. |
Collaborator
Collaborator
|
I ran extra stress-test. For both BYT_MB_RT5682 and BYT_MB_NOCODEC, kmod load/reload tests are rock solid! BSW_CYN_MAX98090 has a separate issue. Updated: |
Member
Author
|
will be merged upstream |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Updated from PR #1504 (based on newer tip)
@fredoh9 can you take this for a spin on CI devices?
@morimoto FYI