On 01/10/2011 09:21 PM, Benjamin Herrenschmidt wrote:
> On Mon, 2011-01-10 at 17:20 -0800, Yinghai Lu wrote:
>>
>> for early access, can not use udelay yet, it will take some one.
>> Also msleep will cause crash, because it needs scheduler there.
>
> Right, and that's for such special cases (hopefully rare) that we have
> system_state... Much better than your function pointers I reckon.
system_state does not work.
it only have BOOTING and RUNNING ...
RUNNING is set in init/main.c::init_post().
so early_quirk and pci_quirk all with BOOTING stage...
slab_is_available() could be used, but looks alike abuse.
>
> We could even wrap it into a safe_delay() function or whatever (in fact
> why not make msleep() itself safe ? It's not like it was timing critical
> code :-)
like
void safe_udelay(unsigned long usecs)
{
if (slab_is_available())
udelay(usecs)
else
early_udelay(usecs);
}
or wonder if you are happy with
void __weak safe_udelay(unsigned long usecs)
{
udelay(usecs);
}
and will have x86 have it's own safe_udelay...
Yinghai
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
