Hi Tim:
I noticed your recent patch turned off PFEC_reserved_bit in case
another CPU had already modified a magic PTE. It seems that it
still leaves PFEC_page_present enabled. This could confuse the
guest if the PTE modification is such that the PTE is now present
and valid. In fact the guest may treat it as a protection fault.
You could turn the present bit off, but that is not strictly correct
if say an MMIO va was reused for normal memory.
The safest solution is probably to just return EXCRET_fault_fixed
and let the CPU figure it out.
Cheers,
--
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
At 15:42 +1100 on 15 Dec (1166197339), Herbert Xu wrote:> I noticed your recent patch turned off PFEC_reserved_bit in case > another CPU had already modified a magic PTE. It seems that it > still leaves PFEC_page_present enabled. This could confuse the > guest if the PTE modification is such that the PTE is now present > and valid. In fact the guest may treat it as a protection fault.It''s just a spurious page fault -- the hardware is happy to throw these at operating systems and so am I. :)> The safest solution is probably to just return EXCRET_fault_fixed > and let the CPU figure it out.That does seem cleaner. Is this a problem you were seeing on a particular system, or just a general improvement? Cheers, Tim. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Fri, Dec 15, 2006 at 09:27:45AM +0000, Tim Deegan wrote:> At 15:42 +1100 on 15 Dec (1166197339), Herbert Xu wrote: > > I noticed your recent patch turned off PFEC_reserved_bit in case > > another CPU had already modified a magic PTE. It seems that it > > still leaves PFEC_page_present enabled. This could confuse the > > guest if the PTE modification is such that the PTE is now present > > and valid. In fact the guest may treat it as a protection fault. > > It''s just a spurious page fault -- the hardware is happy to throw these > at operating systems and so am I. :)It''s only a spurious page fault if PFEC_page_present is clear. Otherwise the OS (Linux in particular) may treat this as a protection error which may: 1) Kill a user-space process if in user-mode. 2) Oops the kernel otherwise.> That does seem cleaner. Is this a problem you were seeing on a > particular system, or just a general improvement?I just happened to be looking at the code :) Cheers, -- Visit Openswan at http://www.openswan.org/ Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel