Skip to contentSkip to conversation

spin_lock_irqwhat ?

DISCUSSION Giuliano Pochini 3 participants 10 messages May 8, 2004 – May 16, 2004

lore
Share & tools
lore
#0 lore raw

Some drivers use spin_lock_irq() a lot, while others always use spin_lock_irqsave(). I can't see the difference. When it's safe using the _irq() version ?

--
Giuliano.

-------------------------------------------------------
This SF.Net email is sponsored by Sleepycat Software
Learn developer strategies Cisco, Motorola, Ericsson & Lucent use to deliver higher performing products faster, at low TCO.
http://www.sleepycat.com/telcomwpreg.php?From=osdnemail3

#1 lore raw
Quoted 3 lines · Giuliano Pochini

Some drivers use spin_lock_irq() a lot, while others always use spin_lock_irqsave(). I can't see the difference. When it's safe using the _irq() version ?

Only when you know that you are not in interrupt context.

Regards,
Clemens

-------------------------------------------------------
This SF.Net email is sponsored by Sleepycat Software
Learn developer strategies Cisco, Motorola, Ericsson & Lucent use to deliver higher performing products faster, at low TCO.
http://www.sleepycat.com/telcomwpreg.php?From=osdnemail3

#2 lore raw

At Mon, 10 May 2004 10:31:07 +0200 (METDST),

Quoted 7 lines · Clemens Ladisch
Quoted 3 lines · Giuliano Pochini

Some drivers use spin_lock_irq() a lot, while others always use spin_lock_irqsave(). I can't see the difference. When it's safe using the _irq() version ?

Only when you know that you are not in interrupt context.

also, not in the context which already disabled irq.

Takashi

-------------------------------------------------------
This SF.Net email is sponsored by Sleepycat Software
Learn developer strategies Cisco, Motorola, Ericsson & Lucent use to deliver higher performing products faster, at low TCO.
http://www.sleepycat.com/telcomwpreg.php?From=osdnemail3

#3 lore raw
Quoted 7 lines · On 10-May-2004 Takashi Iwai wrote:
Quoted 5 lines
Quoted 3 lines

Some drivers use spin_lock_irq() a lot, while others always use spin_lock_irqsave(). I can't see the difference. When it's safe using the _irq() version ?

Only when you know that you are not in interrupt context.

also, not in the context which already disabled irq.

Yes, I read the famous Rusty's unreliable guide. My question was ALSA-specific. The point is that I don't know when those conditions are true. Can and_pcm_period_elapsed() call anything other than the pcm callbacks ? Does the ALSA middle layer ever disable interrupts before calling a driver function ? If it doesn't, I can replace spin_lock_irqsave() with spin_lock_irq() almost everything. Some callbacks are atomic according to the tutorial, but it doesn't say if interrupts are disabled too. Are irq (and preempt) disabled before calling .prepare and .trigger callbacks ?

--
Giuliano.

-------------------------------------------------------
This SF.Net email is sponsored by Sleepycat Software
Learn developer strategies Cisco, Motorola, Ericsson & Lucent use to deliver higher performing products faster, at low TCO.
http://www.sleepycat.com/telcomwpreg.php?From=osdnemail3

#4 lore raw

At Mon, 10 May 2004 16:51:24 +0200 (CEST),

Quoted 12 lines · Giuliano Pochini
Quoted 7 lines · On 10-May-2004 Takashi Iwai wrote:
Quoted 5 lines
Quoted 3 lines

Some drivers use spin_lock_irq() a lot, while others always use spin_lock_irqsave(). I can't see the difference. When it's safe using the _irq() version ?

Only when you know that you are not in interrupt context.

also, not in the context which already disabled irq.

Quoted 10 lines

Yes, I read the famous Rusty's unreliable guide. My question was ALSA-specific. The point is that I don't know when those conditions are true. Can and_pcm_period_elapsed() call anything other than the pcm callbacks ? Does the ALSA middle layer ever disable interrupts before calling a driver function ? If it doesn't, I can replace spin_lock_irqsave() with spin_lock_irq() almost everything. Some callbacks are atomic according to the tutorial, but it doesn't say if interrupts are disabled too. Are irq (and preempt) disabled before calling .prepare and .trigger callbacks ?

prepare and trigger callbacks are already in irq-disabled. i.e. you need only spin_lock() in them.
also, ioctl(SNDRV_PCM_IOCTL1_RESET) is in the irq-disabled context.

others are not.

some points to be noted:

- prepare can be non-atomic by setting SNDRV_PCM_INFO_NONATOMIC_OPS to pcm->info_flags. (e.g. usbaudio.c)

- pointer callback is not protected. but it's called also in the interrupt context, so you can't use spin_lock_irq() in general.

Takashi

-------------------------------------------------------
This SF.Net email is sponsored by Sleepycat Software
Learn developer strategies Cisco, Motorola, Ericsson & Lucent use to deliver higher performing products faster, at low TCO.
http://www.sleepycat.com/telcomwpreg.php?From=osdnemail3

#5 lore raw
Quoted 2 lines · On 10-May-2004 Takashi Iwai wrote:

prepare and trigger callbacks are already in irq-disabled. i.e. you need only spin_lock() in them. [...]

Fine. I moved SetSampleRate() from prepare() to the hw_params() callback because it may need to reload the firmware.

--
Giuliano.

-------------------------------------------------------
This SF.Net email is sponsored by: SourceForge.net Broadband Sign-up now for SourceForge Broadband and get the fastest 6.0/768 connection for only $19.95/mo for the first 3 months!
http://ads.osdn.com/?ad_id=2562&alloc_id=6184&op=click

#6 lore raw
In reply to Takashi Iwai
Quoted 2 lines · On Mon, 10 May 2004 17:08:49 +0200 Takashi Iwai <[email protected]> wrote:

prepare and trigger callbacks are already in irq-disabled. i.e. you need only spin_lock() in them.

Does it mean that ALSA acquires the lock only when it calls PCM callbacks, that is trigger(), etc., are atomic only wrt other PCM functions ?

hdsp.c and rme9652.x use spin_*lock_irq() inside prepare().

--
Giuliano.

-------------------------------------------------------
This SF.Net email is sponsored by: SourceForge.net Broadband Sign-up now for SourceForge Broadband and get the fastest 6.0/768 connection for only $19.95/mo for the first 3 months!
http://ads.osdn.com/?ad_id=2562&alloc_id=6184&op=click

#7 lore raw

At Thu, 13 May 2004 20:57:51 +0200,

Quoted 10 lines · Giuliano Pochini
Quoted 2 lines · On Mon, 10 May 2004 17:08:49 +0200 Takashi Iwai <[email protected]> wrote:

prepare and trigger callbacks are already in irq-disabled. i.e. you need only spin_lock() in them.

Does it mean that ALSA acquires the lock only when it calls PCM callbacks, that is trigger(), etc., are atomic only wrt other PCM functions ?

sorry i don't understand your question.

the prepare and the trigger callbacks acquire several locks. first, the global rw_lock for the pcm linking (snd_pcm_link_rwlock), the group lock the substream belongs to, and the lock for the substream itself. and the first lock/unlock is done with *_irq().

Quoted 1 line

hdsp.c and rme9652.x use spin_*lock_irq() inside prepare().

oh, it's wrong. fixed on cvs now.

thanks,

Takashi

-------------------------------------------------------
This SF.Net email is sponsored by: SourceForge.net Broadband Sign-up now for SourceForge Broadband and get the fastest 6.0/768 connection for only $19.95/mo for the first 3 months!
http://ads.osdn.com/?ad_id=2562&alloc_id=6184&op=click

#8 lore raw
Quoted 6 lines · On Fri, 14 May 2004 12:37:46 +0200 Takashi Iwai <[email protected]> wrote:
Quoted 5 lines
Quoted 2 lines

prepare and trigger callbacks are already in irq-disabled. i.e. you need only spin_lock() in them.

Does it mean that ALSA acquires the lock only when it calls PCM callbacks, that is trigger(), etc., are atomic only wrt other PCM functions ?

Quoted 6 lines

sorry i don't understand your question.

the prepare and the trigger callbacks acquire several locks. first, the global rw_lock for the pcm linking (snd_pcm_link_rwlock), the group lock the substream belongs to, and the lock for the substream itself. and the first lock/unlock is done with *_irq().

Ok, are those locks acquired also before calling any other interface callback (control, midi...) ? If so, it shouldn't be necessary protecting with a spin_lock() the code which touches the hw inside prepare().

--
Giuliano.

-------------------------------------------------------
This SF.Net email is sponsored by: SourceForge.net Broadband Sign-up now for SourceForge Broadband and get the fastest 6.0/768 connection for only $19.95/mo for the first 3 months!
http://ads.osdn.com/?ad_id=2562&alloc_id=6184&op=click

#9 lore raw

At Sat, 15 May 2004 23:37:47 +0200,

Quoted 12 lines · Giuliano Pochini
Quoted 8 lines · On Fri, 14 May 2004 12:37:46 +0200 Takashi Iwai <[email protected]> wrote:
Quoted 5 lines
Quoted 2 lines

prepare and trigger callbacks are already in irq-disabled. i.e. you need only spin_lock() in them.

Does it mean that ALSA acquires the lock only when it calls PCM callbacks, that is trigger(), etc., are atomic only wrt other PCM functions ?

sorry i don't understand your question.

Quoted 8 lines
Quoted 4 lines

the prepare and the trigger callbacks acquire several locks. first, the global rw_lock for the pcm linking (snd_pcm_link_rwlock), the group lock the substream belongs to, and the lock for the substream itself. and the first lock/unlock is done with *_irq().

Ok, are those locks acquired also before calling any other interface callback (control, midi...) ? If so, it shouldn't be necessary protecting with a spin_lock() the code which touches the hw inside prepare().

no, all of these locks are specific to PCM.
you still likely need another spinlock for protecting the h/w registers reading/writing.

Takashi

-------------------------------------------------------
This SF.Net email is sponsored by: SourceForge.net Broadband Sign-up now for SourceForge Broadband and get the fastest 6.0/768 connection for only $19.95/mo for the first 3 months!
http://ads.osdn.com/?ad_id=2562&alloc_id=6184&op=click