Re: Re: pv_ops dom0 USB fixed

[Jeremy Fitzhardinge <jeremy@goop.org>]

Gerd Hoffmann wrote:
> Jeremy Fitzhardinge wrote:
>   
>>> Intel 64bit box, 6GB RAM (dom0_mem=2g), running rawhide.
>>>       
>> OK, that's interesting.  The symptom is the same (drive stops talking),
>> but after a while rather than immediate.
>>     
>
> Yep.
>
>   
>> That suggests there might be
>> some issue with what memory gets allocated or something?
>>     
>
> Or translated.  Maybe it grooks due to being fed with an invalid bus
> address where it can't DMA do.
>   

Hm.  It just uses dmam_alloc_coherent() to allocate its control 
structures, which ends up calling xen_alloc_coherent like everyone else 
does.  And presumably the actual IO goes via sg bios like everyone else.

>> Does putting the controller in IDE/legacy mode work OK?
>>     
>
> Didn't try, and can't easily because it is a headless box sitting in the
> corner.  Speaking of memory:  IIRC ahci can do 64bit DMA, which legacy
> IDE can't.  Maybe that makes AHCI trigger the bug while it doesn't
> happen with IDE?
>   

Yeah, that's worth investigating.  Though my test machine only has 2G, 
so it shouldn't make a difference.  And my other test machine has a 
Fusion which is also 64-bit capable.

    J