cvs commit: src/sys/alpha/alpha busdma_machdep.csrc/sys/i386/i386 src/sys/powerpc/powerpc busdma_machdep.c

[Scott Long]

Justin T. Gibbs wrote:
>>>>@@ -931,8 +932,12 @@
>>>>	while ((map = STAILQ_FIRST(&bounce_map_callbacklist)) != NULL) {
>>>>		STAILQ_REMOVE_HEAD(&bounce_map_callbacklist, links);
>>>>		mtx_unlock(&bounce_lock);
>>>>+		if (map->callback_mtx != NULL)
>>>>+			mtx_lock(map->callback_mtx);
>>>>		bus_dmamap_load(map->dmat, map, map->buf, map->buflen,
>>>>				map->callback, map->callback_arg, /*flags*/0);
>>>>+		if (map->callback_mtx != NULL)
>>>>+			mtx_unlock(map->callback_mtx);
>>>>		mtx_lock(&bounce_lock);
>>>>	}
>>>>	mtx_unlock(&bounce_lock);
>>>
>>>
>>>Good to see you working on this area.  Why would callback_mtx be null?  If
>>>it is null, is it safe to call the function without the lock held?
>>>
>>>-Nate
>>>
>>
>>Yes, calling with a NULL mutex would be allowed and would behave as
>>expected.
> 
> 
> Does this code run without giant?  If not, shouldn't it be dropping giant
> upon calling what should be an MP safe client?  If so, doesn't it need
> to pick up giant in for the non-MP safe clients?
> 
> --
> Justin
> 
> 

Ah, good point.  For all but ia64, some attempt has been made to make
the swi be MPSAFE.  I'll look into it.

Scott