Skip to contentSkip to conversation

2.6.-test10: bad "__might_sleep" in ALSA sound drivers

DISCUSSION Torrey Hoffman 3 participants 3 messages Nov 25, 2003 Sound

lore
Share & tools
lore
#0 lore raw

I get this every time I boot 2.6.0-test10. I think this problem has been around for a while.

Debug: sleeping function called from invalid context at include/asm/semaphore.h:119
in_atomic():1, irqs_disabled():0

Call Trace:
 [<c01229d0>] __might_sleep+0xab/0xc9
 [<fcff57bd>] ap_cs8427_sendbytes+0x38/0xd2 [snd_ice1712]
 [<fcfa825a>] snd_i2c_sendbytes+0x22/0x26 [snd_i2c]
 [<fcfaf15e>] snd_cs8427_reg_read+0x2c/0x8c [snd_cs8427]
 [<fcfaf414>] snd_cs8427_create+0xa4/0x347 [snd_cs8427]
 [<c02dc4b0>] snd_device_new+0x20/0x6a
 [<fcff151a>] snd_ice1712_init_cs8427+0x2f/0x6f [snd_ice1712]
 [<fcff6153>] snd_ice1712_delta_init+0x230/0x2b9 [snd_ice1712]
 [<fcff5227>] snd_ice1712_probe+0x323/0x34c [snd_ice1712]
 [<c017360f>] dput+0x24/0x25f
 [<c020c833>] pci_device_probe_static+0x4d/0x5e
 [<c020c87c>] __pci_device_probe+0x38/0x4b
 [<c020c8bb>] pci_device_probe+0x2c/0x48
 [<c0272efd>] bus_match+0x3d/0x65
 [<c0273016>] driver_attach+0x59/0x83
 [<c02732df>] bus_add_driver+0x9e/0xb1
 [<c020ca82>] pci_register_driver+0x6b/0x90
 [<fcfa2015>] alsa_card_ice1712_init+0x15/0x4e [snd_ice1712]
 [<c013cbc5>] sys_init_module+0x14f/0x233
 [<c010b3e5>] sysenter_past_esp+0x52/0x71

The system is a P4 with HT, kernel has SMP & preempt enabled. There are two sound cards. Here are all the snd_ drivers mentioned in the output of lsmod:

snd_ice1712            55428  -
snd_ice17xx_ak4xxx      2880  -
snd_cs8427              8224  -
snd_i2c                 4512  -
snd_ak4xxx_adda         4832  -
snd_intel8x0           28612  -
snd_ac97_codec         50660  -
snd_pcm                88352  -
snd_timer              22052  -
snd_page_alloc          8772  -
snd_mpu401_uart         6016  -
snd_rawmidi            20192  -
snd_seq_device          6440  -
Signature · 1 line
Torrey Hoffman <[email protected]>
#1 lore raw
Quoted 7 lines · Torrey Hoffman

I get this every time I boot 2.6.0-test10. I think this problem has been around for a while.

Debug: sleeping function called from invalid context at include/asm/semaphore.h:119
in_atomic():1, irqs_disabled():0

Quoted 10 lines
Call Trace:
 [<c01229d0>] __might_sleep+0xab/0xc9
 [<fcff57bd>] ap_cs8427_sendbytes+0x38/0xd2 [snd_ice1712]
 [<fcfa825a>] snd_i2c_sendbytes+0x22/0x26 [snd_i2c]
 [<fcfaf15e>] snd_cs8427_reg_read+0x2c/0x8c [snd_cs8427]
 [<fcfaf414>] snd_cs8427_create+0xa4/0x347 [snd_cs8427]
 [<c02dc4b0>] snd_device_new+0x20/0x6a
 [<fcff151a>] snd_ice1712_init_cs8427+0x2f/0x6f [snd_ice1712]
 [<fcff6153>] snd_ice1712_delta_init+0x230/0x2b9 [snd_ice1712]
 [<fcff5227>] snd_ice1712_probe+0x323/0x34c [snd_ice1712]

snd_cs8427_create() does:

	snd_i2c_lock(bus);
	if ((err = snd_cs8427_reg_read(device, CS8427_REG_ID_AND_VER)) != CS8427_VER8427A) {

the first statement takes a spinlock. snd_cs8427_reg_read() then calls ap_cs8427_sendbytes() which downs a semaphore.

It might be more appropriate to use a semaphore for ice1712_t.gpio_mutex?

#2 lore raw

Hi,

At Mon, 24 Nov 2003 18:00:17 -0800,

Quoted 22 lines · Andrew Morton
Quoted 7 lines · Torrey Hoffman

I get this every time I boot 2.6.0-test10. I think this problem has been around for a while.

Debug: sleeping function called from invalid context at include/asm/semaphore.h:119
in_atomic():1, irqs_disabled():0

Quoted 10 lines
Call Trace:
 [<c01229d0>] __might_sleep+0xab/0xc9
 [<fcff57bd>] ap_cs8427_sendbytes+0x38/0xd2 [snd_ice1712]
 [<fcfa825a>] snd_i2c_sendbytes+0x22/0x26 [snd_i2c]
 [<fcfaf15e>] snd_cs8427_reg_read+0x2c/0x8c [snd_cs8427]
 [<fcfaf414>] snd_cs8427_create+0xa4/0x347 [snd_cs8427]
 [<c02dc4b0>] snd_device_new+0x20/0x6a
 [<fcff151a>] snd_ice1712_init_cs8427+0x2f/0x6f [snd_ice1712]
 [<fcff6153>] snd_ice1712_delta_init+0x230/0x2b9 [snd_ice1712]
 [<fcff5227>] snd_ice1712_probe+0x323/0x34c [snd_ice1712]

snd_cs8427_create() does:

Quoted 8 lines
	snd_i2c_lock(bus);
	if ((err = snd_cs8427_reg_read(device, CS8427_REG_ID_AND_VER)) != CS8427_VER8427A) {

the first statement takes a spinlock. snd_cs8427_reg_read() then calls ap_cs8427_sendbytes() which downs a semaphore.

It might be more appropriate to use a semaphore for ice1712_t.gpio_mutex?

the attached patch fixes this, simply replacing spinlock with mutex. it's already in ALSA 1.0.0pre release.

--
Takashi Iwai <[email protected]> ALSA Developer - www.alsa-project.org

include/sound/i2c.h +5 -5
diff -ru linux/include/sound linux/include/sound
@@ -58,7 +58,7 @@
 	snd_card_t *card;	/* card which I2C belongs to */
 	char name[32];		/* some useful label */
 
-	spinlock_t lock;
+	struct semaphore lock_mutex;
 
 	snd_i2c_bus_t *master;	/* master bus when SCK/SCL is shared */
 	struct list_head buses;	/* master: slave buses sharing SCK/SCL, slave: link list */
@@ -84,15 +84,15 @@
 
 static inline void snd_i2c_lock(snd_i2c_bus_t *bus) {
 	if (bus->master)
-		spin_lock(&bus->master->lock);
+		down(&bus->master->lock_mutex);
 	else
-		spin_lock(&bus->lock);
+		down(&bus->lock_mutex);
 }
 static inline void snd_i2c_unlock(snd_i2c_bus_t *bus) {
 	if (bus->master)
-		spin_unlock(&bus->master->lock);
+		up(&bus->master->lock_mutex);
 	else
-		spin_unlock(&bus->lock);
+		up(&bus->lock_mutex);
 }
 
 int snd_i2c_sendbytes(snd_i2c_device_t *device, unsigned char *bytes, int count);
sound/i2c/i2c.c +1 -1
diff -ru linux/sound linux/sound
@@ -84,7 +84,7 @@
 	bus = (snd_i2c_bus_t *)snd_magic_kcalloc(snd_i2c_bus_t, 0, GFP_KERNEL);
 	if (bus == NULL)
 		return -ENOMEM;
-	spin_lock_init(&bus->lock);
+	init_MUTEX(&bus->lock_mutex);
 	INIT_LIST_HEAD(&bus->devices);
 	INIT_LIST_HEAD(&bus->buses);
 	bus->card = card;