[alsa-devel] ALSA sound core device deinit crash

Takashi Iwai tiwai at suse.de
Fri Jul 8 07:53:21 CEST 2016


(Please don't drop Cc to ML)

On Fri, 08 Jul 2016 07:35:37 +0200,
b_lkasam at codeaurora.org wrote:
> 
> On 2016-07-08 10:58, Takashi Iwai wrote:
> > On Fri, 08 Jul 2016 07:19:05 +0200,
> > b_lkasam at codeaurora.org wrote:
> >>
> >> Hi Alsa team,
> >> There is kernel crash observed when soundcard register failure case as
> >> below ->
> >>
> >> diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c old mode
> >> 100644 new mode 100755 index 0495890..60a1eb0
> >> --- a/sound/soc/soc-core.c
> >> +++ b/sound/soc/soc-core.c
> >> @@ -1382,6 +1382,7 @@ static int soc_probe_link_dais(struct
> >> snd_soc_card *card, int num, int order)
> >> /* do machine specific initialization */ if (dai_link->init) { ret =
> >> dai_link->init(rtd);
> >> + ret = -ENODEV; //  -> we can force error here to reproduce crash
> >> easily.
> >> if (ret < 0) {
> >> dev_err(card->dev, "ASoC: failed to init %s: %d\n", dai_link->name,
> >> ret);
> >>
> >> If sound card fails at initialize at above, it is crashing in
> >> pcm_chmap_ctl_private_free().
> >>
> >> <1>[   40.646642] [01-01-2016 00:07:16 CPU:0x3] Unable to handle
> >> kernel
> >> paging request at virtual address ffffffc0da644b68
> >> <1>[   40.646664] [01-01-2016 00:07:16 CPU:0x3] pgd = ffffffc001d28000
> >> <1>[   40.646673] [01-01-2016 00:07:16 CPU:0x3] [ffffffc0da644b68]
> >> *pgd=00000000857fc003, *pud=00000000857fc003, *pmd=000000017ddd8003,
> >> *pte=00c000015a644793
> >> <0>[   40.646697] [01-01-2016 00:07:16 CPU:0x3] Internal error: Oops:
> >> 9600004f [#1] PREEMPT SMP
> >> <6>[   40.646708] [01-01-2016 00:07:16 CPU:0x3] Modules linked in:
> >> brcm_bt_drv fm_drv brcm_hci_ldisc texfat(PO)
> >> <6>[   40.646735] [01-01-2016 00:07:16 CPU:0x3] CPU: 3 PID: 299 Comm:
> >> kworker/u8:8 Tainted: P           O   3.18.24-g41b2dda2-00002-gbe25a74
> >> #1
> >> <6>[   40.646744] [01-01-2016 00:07:16 CPU:0x3] Hardware name:
> >> Qualcomm
> >> Technologies, Inc. MSM 8996 v3.x + PMI8996 MTP (DT)
> >> <6>[   40.646763] [01-01-2016 00:07:16 CPU:0x3] Workqueue: deferwq
> >> deferred_probe_work_func
> >> <6>[   40.646773] [01-01-2016 00:07:16 CPU:0x3] task: ffffffc0e951c880
> >> ti: ffffffc03368c000 task.ti: ffffffc03368c000
> >> <6>[   40.646787] [01-01-2016 00:07:16 CPU:0x3] PC is at
> >> pcm_chmap_ctl_private_free+0x1c/0x2c
> >> <6>[   40.646798] [01-01-2016 00:07:16 CPU:0x3] LR is at
> >> snd_ctl_free_one+0x20/0x34
> >>
> >>
> >> FIX:
> >>
> >> Can you look at the change below and share your comments on this?
> >> diff --git a/sound/core/device.c b/sound/core/device.c old mode 100644
> >> new mode 100755 index 41bec30..eaffde1
> >> --- a/sound/core/device.c
> >> +++ b/sound/core/device.c
> >> @@ -219,6 +219,7 @@ void snd_device_free_all(struct snd_card *card)
> >>
> >>           if (snd_BUG_ON(!card))
> >>                   return;
> >> -       list_for_each_entry_safe_reverse(dev, next, &card->devices,
> >> list)
> >> +               list_for_each_entry_safe(dev, next, &card->devices,
> >> list)
> >>                   __snd_device_free(dev); }
> >>
> >>
> >> Since control sound device has the lowest type value
> >> (SNDRV_DEV_CONTROL), it will be the first entry linked in the
> >> card->devices linked list head and will be the last one to be freed.
> >>
> >> This issue seems to be resolved by modifying the sequence the sound
> >> devices in the card->devices list are freed as shown below (from
> >> “prev”
> >> direction to “next” direction) but I’m not sure if this is a right
> >> approach from ALSA perspective.
> >
> > This doesn't look correct.  The strange thing is that this error
> > shouldn't happen no matter which free loop direction is.  The chmap
> > ctl should have been already removed by the disconnection before
> > freeing.  It means that either the disconnection isn't done properly
> > or something else is missing.
> >
> > Could you give the full stack trace?  It's important to know which
> > code path triggers it.
> >
> >
> > thanks,
> >
> > Takashi
> 
> 
> Hi Takashi,
> 
> Here is full stack trace -->
> 
> 
> [Callstack]
> : (struct snd_pcm_chmap *)info = 0xFFFFFFC0DA6A5200
> : (struct snd_pcm *)pcm = 0xFFFFFFC0DA644A80 //part of buddy page,
> read-only
> : info->stream = 0
> : snd_card_free() was called
> 
> -012|pcm_chmap_ctl_private_free()
> //info->pcm->streams[info->stream].chmap_kctl = NULL
> -013|snd_ctl_free_one()
> -014|snd_ctl_remove()
> -015|snd_ctl_dev_free()
> -016|__snd_device_free()
> -017|snd_device_free_all()
> -018|snd_card_do_free(inline)
> -018|release_card_device()
> -019|device_release()
> -020|kobject_cleanup(inline)
> -020|kobject_release()
> -021|kobject_put()
> -022|put_device()
> -023|snd_card_free_when_closed()
> -024|snd_card_free()
> -025|snd_soc_instantiate_card(inline) -> //Here instantiate card
> failed for some reason, then triggers snd_card_free
> -025|snd_soc_register_card()
> -026|msm8996_asoc_machine_probe()
> -027|platform_drv_probe()
> -028|really_probe(inline)
> -028|driver_probe_device()
> -029|__device_attach()
> -030|bus_for_each_drv()
> -031|device_attach()
> -032|bus_probe_device()
> -033|deferred_probe_work_func()
> -034|static_key_count(inline)
> -034|static_key_false(inline)
> -034|trace_workqueue_execute_end(inline)
> -034|process_one_work()
> -035|worker_thread()
> -036|kthread()
> -037|ret_from_fork(asm)
> ---|end of frame
> 
> 
> -----------------------
> trigger point in soc-core.c
> in API snd_soc_instantiate_card(),
> 
>         card_probe_error:
>                  if (card->remove)
>                         card->remove(card);
> 
>                  snd_card_free(card->snd_card); -> this is
> snd_card_free which internally leads to above function call.
> ------------------------------------------

OK, thanks.  So this is the case where it frees before registering,
and indeed there is a bug in the PCM chmap code.  It's freed at
disconnection but the disconnect is called only when it was
registered.

Below is a quick fix.  Give it a try.


thanks,

Takashi

---
diff --git a/sound/core/control.c b/sound/core/control.c
index 9ff081cd03f4..fb096cb20a80 100644
--- a/sound/core/control.c
+++ b/sound/core/control.c
@@ -160,6 +160,8 @@ void snd_ctl_notify(struct snd_card *card, unsigned int mask,
 	
 	if (snd_BUG_ON(!card || !id))
 		return;
+	if (card->shutdown)
+		return;
 	read_lock(&card->ctl_files_rwlock);
 #if IS_ENABLED(CONFIG_SND_MIXER_OSS)
 	card->mixer_oss_change_count++;
diff --git a/sound/core/pcm.c b/sound/core/pcm.c
index 308c9ecf73db..8e980aa678d0 100644
--- a/sound/core/pcm.c
+++ b/sound/core/pcm.c
@@ -849,6 +849,14 @@ int snd_pcm_new_internal(struct snd_card *card, const char *id, int device,
 }
 EXPORT_SYMBOL(snd_pcm_new_internal);
 
+static void free_chmap(struct snd_pcm_str *pstr)
+{
+	if (pstr->chmap_kctl) {
+		snd_ctl_remove(pstr->pcm->card, pstr->chmap_kctl);
+		pstr->chmap_kctl = NULL;
+	}
+}
+
 static void snd_pcm_free_stream(struct snd_pcm_str * pstr)
 {
 	struct snd_pcm_substream *substream, *substream_next;
@@ -871,6 +879,7 @@ static void snd_pcm_free_stream(struct snd_pcm_str * pstr)
 		kfree(setup);
 	}
 #endif
+	free_chmap(pstr);
 	if (pstr->substream_count)
 		put_device(&pstr->dev);
 }
@@ -1135,10 +1144,7 @@ static int snd_pcm_dev_disconnect(struct snd_device *device)
 	for (cidx = 0; cidx < 2; cidx++) {
 		if (!pcm->internal)
 			snd_unregister_device(&pcm->streams[cidx].dev);
-		if (pcm->streams[cidx].chmap_kctl) {
-			snd_ctl_remove(pcm->card, pcm->streams[cidx].chmap_kctl);
-			pcm->streams[cidx].chmap_kctl = NULL;
-		}
+		free_chmap(&pcm->streams[cidx]);
 	}
 	mutex_unlock(&pcm->open_mutex);
 	mutex_unlock(&register_mutex);


More information about the Alsa-devel mailing list