[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Xen-users] System seams slower with Xenified kernel and xend dies with SIG 15



 

> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Petersson, Mats
> Sent: 02 August 2006 11:09
> To: Gabriel Rossetti; xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-users] System seams slower with Xenified 
> kernel and xend dies with SIG 15
> 
> > -----Original Message-----
> > From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> > [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> > Gabriel Rossetti
> > Sent: 02 August 2006 10:55
> > To: xen-users@xxxxxxxxxxxxxxxxxxx
> > Subject: Re: [Xen-users] System seams slower with Xenified 
> > kernel and xend dies with SIG 15
> > 
> > Ok, but wouldn't the OS take care of that? I don't think that 
> > the driver
> > does any
> > address translations, it would make no sense, it would use virtual
> > addresses and the OS, in my case a Xenified
> > linux kernel, takes care of translations. I suspect that some of the
> > changes made for Xen is the memory management, no?
> 
> Yes, but some of the changes in the Xenified kernel is to 
> add/change the
> code in drivers... And since the binary for the ATI (or 
> nVidia) drivers
> aren't source-code, they don't get the new code by recompiling the
> kernel... 
> 
> > If so then I would make sense that it stays transparent so that
> > everything would still work and
> > not have to recompile everything like in Denali. At least 
> > that's what I
> > think, from what I saw in
> > an OS class, and that I had to implement an OS's memory 
> > managment, but I
> > may be wrong. Also,
> > my binary Intel ipw3945 wireless drivers work fine, so I don't think
> > that is it, but then again, I may
> > be wrong. From my light investigation I think it's a path 
> problem, it
> > can't find something apparently,
> > and I don't get why not. I'll have to look at it more when I 
> > find the time.
> 
> Maybe the wireless driver is more "well-behaved". 

Thinking another few seconds about it, I think one main difference is
that a wireless driver doesn't put physical addresses within the packet
for the PCI-device to use as addresses for other things, it just says
"Here's a packet, go send it" to the DMA-API, which works fine. 

Graphics drivers will have things like:

Here's a DMA-buffer:
        <BitBlt from(0,0,100,100) src=(some address),
to(100,100,200,200) dest=(some other address)>

The addresses for src and dest need to be translated correctly to
machine physical addresses. The natural operation here is to just look
up the virtual address and translate it to a physical one - but the
normal OS doesn't understand the fact that there are two types of
physical address - one that it uses internally, and another one for
external devices... 

Sorry for the post-on-post thing... 

--
Mats
> 
> --
> Mats
> > 
> > Gabriel
> > 
> > 
> > 
> > Petersson, Mats wrote:
> > >
> > > Can't comment on xend dieing, but any binary driver would 
> > be unlikely to
> > > work in Xenified kernel, since the Xenified kernel changes 
> > some of the
> > > things that a driver needs to do when it translates 
> memory addresses
> > > from virtual to physical - there are now two different 
> > types of physical
> > > address: pseudo-physical, which is what the OS sees, and machine
> > > physical, which is what the PROCESSOR sees. If you have a 
> > Dom0 that has
> > > 512MB of RAM, it will not necessarily have all it's memory from
> > > 0..512MB, but it may actually be located anywhere Xen likes 
> > it to be...
> > > [Dom0 is most likely starting somewhere low, since it's 
> > loaded early -
> > > but that's just luck, rather than planned for]. Machine 
> physical is
> > > needed for the PCI (PCIe == PCIx == AGP == PCI in this case 
> > - they are
> > > just different hardware implementations of the same 
> > protocol) so that
> > > the card can access memory directly, and for sure all 
> > modern graphics
> > > cards have direct memory access capabilities (and most 
> > likely doesn't do
> > > much other than a standard 15 year old VGA card if you 
> > don't use these
> > > capability). 
> > >
> > >
> > > --
> > > Mats
> > >
> > >
> > >   
> > 
> > 
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-users
> > 
> > 
> > 
> 
> 
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
> 
> 
> 



_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.