Hi
There's a race between registration routine and interrupt handler in aw2 driver
which manifested as oops on some box [1].
When snd_aw2_create calls request_irq chip data are not yet fully initialized.
If interrupt raises before snd_aw2_create call snd_aw2_saa7146_setup, driver may
oops or read/write other port than its own. (chip->base_addr is initialized in
snd_aw2_saa7146_setup, but is used in snd_aw2_saa7146_interrupt via READREG/WRITEREG)
The simple fix would be to move call to snd_aw2_saa7146_setup before request_irq,
but snd_aw2_saa7146_setup enables interrupts...
PS: I don't have this hardware.
Marcin
[1] http://kerneloops.org/guilty.php?guilty=snd_aw2_saa7146_interrupt&version=2.6.26-rc&start=1703936&end=1736703&class=oops
At Sun, 22 Jun 2008 18:47:17 +0200,
Marcin Slusarz wrote:
>
> Hi
> There's a race between registration routine and interrupt handler in aw2 driver
> which manifested as oops on some box [1].
>
> When snd_aw2_create calls request_irq chip data are not yet fully initialized.
> If interrupt raises before snd_aw2_create call snd_aw2_saa7146_setup, driver may
> oops or read/write other port than its own. (chip->base_addr is initialized in
> snd_aw2_saa7146_setup, but is used in snd_aw2_saa7146_interrupt via READREG/WRITEREG)
>
> The simple fix would be to move call to snd_aw2_saa7146_setup before request_irq,
> but snd_aw2_saa7146_setup enables interrupts...
>
> PS: I don't have this hardware.
>
> Marcin
>
> [1] http://kerneloops.org/guilty.php?guilty=snd_aw2_saa7146_interrupt&version=2.6.26-rc&start=1703936&end=1736703&class=oops
Thanks for spotting the bug.
I fixed this on my git tree now and will post a pull request soon
later.
Takashi