dev/sound/pcm/sound.h shadows INTR_MPSAFE declaration
Yuriy Tsibizov
Yuriy.Tsibizov at gfk.ru
Tue Apr 6 23:23:58 PDT 2004
> From: Doug White [mailto:dwhite at gumbysoft.com]
> On Tue, 6 Apr 2004, Yuriy Tsibizov wrote:
>
> > > > Today I've noticied that my driver is marked as Giant-locked:
> > >
> > > Is it actually MPSAFE? If so, feel free to mark it as such,
> > > although pcm
> > > isn't fully locked yet, so we take no responsibility for
> any damage
> > > caused :)
> >
> > Some sound drivers in -CURRENT allocate interrupts with INTR_MPSAFE.
> > The problem is that after including dev/sound/pcm/sound.h
> INTR_MPSAFE no longer equals to 512, but it is set to 0. And
> if sound driver interrupt was allocated as
> INTR_TYPE_AV|INTR_MPSAFE it will be marked only as INTR_TYPE_TTY.
>
> Ever think that might be intentional? :) I'd poke
> -multimedia to get the
> sound driver folks, though.
Exactly the same piece of code exist on RELENG_4 branch. I think it was used to allow INTR_MPSAFE|INTR_TYPE_AV inside sound
drivers in 4.x, where INTR_MPSAFE and INTR_TYPE_AV are not defined.
If PCM subsystem in 5.x requires that sound drivers should not mark interrupts as INTR_MPSAFE it can be done in less ambiguous way, like this:
-sound.h-
/*
* PCM subsystem in not fully locked.
* This should force use of Giant lock for sound drivers
* even when they claim to be MPSAFE.
*/
#ifndef I_THINK_THAT_PCM_IS_MPSAFE
#define INTR_MPSAFE 0
#endif
-sound.h-
and remove wrong #ifdef INTR_MPSAFE checks, because INTR_MPSAFE was never defined. On RELENG_4 it can be replaced with
FreeBSD version check.
Yuriy.
More information about the freebsd-current
mailing list