On Tue, 2011-01-04 at 18:38 +0000, Konrad Rzeszutek Wilk wrote: 
quoted text
> On Tue, Jan 04, 2011 at 05:18:58PM +0000, Ian Campbell wrote:
> > On Thu, 2010-12-30 at 19:48 +0000, Konrad Rzeszutek Wilk wrote:
> > > We walk the E820 region and start at 0 (for PV guests we start
> > > at ISA_END_ADDRESS)
> > 
> > I was trying to figure out what any of this had to do with HVM guests,
> > but you mean as opposed to dom0, which with my pedant hat on is also a
> > guest ;-).
> > 
> > >  and skip any E820 RAM regions. For all other
> > > regions and as well the gaps we set them to be identity mappings.
> > > 
> > > The reasons we do not want to set the identity mapping from 0->
> > > ISA_END_ADDRESS when running as PV is b/c that the kernel would
> > > try to read DMI information and fail (no permissions to read that).
> > 
> > The reason for this special case is that in domU we have already punched
> > a hole from 640k-1M into the e820 which the hypervisor gave us.
> 
> For the privileged guest - yes. But for the non-priviligied it does not have
> such range and would end up failing.

xen_memory_setup has: 
        e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
                        E820_RESERVED);
which is unconditional but is actually more for domU's benefit than
dom0's which already sees the host e820 presumably with the right hole
already in place, which we simply shadow, or maybe slightly extend,
here.

In a domU we do this because if you let these pages into the general
allocation pool then they will potentially get used as page table pages
(hence be R/O) but e.g. the DMI code tries to map them to probe for
signatures and tries to does so R/W which fails. We could try and find
everywhere in the kernel which does this or we can simply reserve the
region which stops it getting used for page tables or other special
things, and is somewhat less surprising for non-Xen code.

quoted text
> > Should we perhaps be doing this identity mapping before we punch that
> > extra hole? i.e. setup ID mappings based on the hypervisors idea of the
> > guest e820 not the munged one we subsequently magicked up? Only the
> 
> You mean the ISA_START_ADDRESS->ISA_END_ADDRESS we mark as reserved?

Yep. 

quoted text
> It sure would be easier

quoted text
> (and it would mean we can return that memory back to the hypervisor).

I don't think you can return it, since something like the DMI code which
wants to probe it expects to be able to map that PFN, if you've given
the MFN back then that will fail.

I suppose we could alias all such PFNs to the same scratch MFN but I'd
be concerned about some piece of code which expects to interact with
firmware scribbling over it and surprising some other piece of code
which interacts with the firmware...

Ian.

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/