Re: X server in dom0: Bad VBT signature

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Mon, Sep 03, 2012 at 06:16:07AM -0400, Konrad Rzeszutek Wilk wrote:
> In Linux upstream we fixed this by adding an extra bit to the PFN in the
> P2M lookup table - an IDENTITY_FRAME_BIT (1<<63) that would identity
> said PFN->MFN as a 1:1. During bootup when we scan the E820 we would
> mark all everything that is not E820_RAM (and E820_UNUSUABLE) as such
> (including the gaps). 
> 
> Granted there was also some underlaying changes in the technology - we
> changed the P2M from an array to be an tree - this way we could replace
> swatches of empty space in the tree to point to an 'p2m_identity' leaf.
> 
> Anyhow, the point was that with this - we could omit some of the changes
> of having to identity various drivers that assume that PFN == MFN and
> when they request a PTE we would insert the proper mfn (in the case of
> MMIO space that would be the pfn).
> 
> Could that solve this problem?

I don't think: the problem here is not with the p2m table.
The problem is to identify, when the pmap module gets a request to
map a an address, if this is a PA (which has to be translated to a MA), or
if it's already a MA.
Device drivers requests are already identified as being MA (thanks to the
bus_space interface, driver requests have their own path separate from
VM requests). The problem is when a mapping request comes from userland
via /dev/mem: we don't know if it is a PA or a MA (it depends on who is
doing the request) and as both spaces uses the same addresses there
are conflicts.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--