Re: rme9652: possible deadlock

[Abramo Bagnara <abramo.bagnara@libero.it>]

Jaroslav Kysela ha scritto:
> 
> 
> Bingo, I got it. You mean this scenario (linked two streams):
> 
> CPU0: stream1 interrupt: stream1 locked -> snd_pcm_stop()
> CPU1: stream2 interrupt: stream2 locked -> snd_pcm_stop()
> CPU0: try to lock stream2 (in ACTION macro)
> CPU1: try to lock stream1 (in ACTION macro)
> 
> Yes, this is next deadlock issue.
> 
> I have probably only one solution in my brain: Separate snd_pcm_stop() to 
> two steps:
> 
> 1) stop only the locked stream and unlock it
> 2) stop all other linked streams

My proposal is to have a per "linked PCM group" lock and to use it for 
proper locking (with also some side benefits substituting some use of 
snd_pcm_link_lock).

This mean also we need to change the linked list struct from

struct _snd_pcm_substream {
	...
         /* -- linked substreams -- */
         snd_pcm_substream_t *link_next;
         snd_pcm_substream_t *link_prev;

to
	struct list_head link_list;
	snd_pcm_link_t *link;	/* 0 for non linked stream */

typedef struct _snd_pcm_link {
	spinlock_t lock;
	struct list_head substreams;
};

Of course creation and destruction of a snd_pcm_link_t need to be 
protected by snd_pcm_link_lock.


-- 
Abramo Bagnara                       mailto:abramo.bagnara@libero.it

Opera Unica                          Phone: +39.546.656023
Via Emilia Interna, 140
48014 Castel Bolognese (RA) - Italy



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf