Re: VT-d flush timeout

["Jan Beulich" <JBeulich@suse.com>]

>>> On 26.09.14 at 01:23, <eddie.dong@intel.com> wrote:
>> >>
>> >
>> > Giving this path of long time wait-loop only happens at the case when
>> > the hardware fails, I don't care if it enters panic in 1 seconds or in
>> > 10ms seconds. But the software compatibility (for all existing
>> > platforms and potential future platforms) is much more important.
>> 
>> I agree for the paths leading to a panic(). But there's one such case where 
> it
>> doesn't: snb_vtd_ops_preamble().
>> 
> 
> For some reason, it is using a same MAGIC TIMEOUT value with VTd flush 
> timeout.
> They are actually different and this one is used to prevent the IGD to enter 
> RC6 state :)
> 
> Whether the IGD enters RC6 state in 1second is another story, it doesn't 
> indicate the 
> hardware is bad, like the VTD flush timeout indicates. So it is not 
> necessary to enter panic.
> 
> Maybe we should cook a patch to use different TIMEOUT MACRO, though they are 
> 1second in both cases.

If it needs to stay 1 second, more than just introducing a separate
macro should be done. If the value can be reduced to a couple of
milliseconds, keeping the code as is with just the new macro put in
place would be fine.

Jan