On Friday 02 July 2010 12:05:05 Jan Beulich wrote:
> Christoph,
>
> that patch of yours, even though it doesn't say so, appears to be a
> plain revert of 3.2's c/s 16898. Consequently it re-introduces the
> problems fixed by that earlier c/s, in particular the bad behavior of
> the SVM decoder near page boundaries. Presumably that patch,
> never having been part of an official release, has seen very little
> testing, but as it was taken into our post-release SLE10 SP3 code
> base is now apparently causing regressions there.
>
> In order to find a way to better fix the problems you tried to
> address, could you provide as much detail as you can recall/
> reconstruct on what lead you to that patch of yours (the patch
> comment unfortunately was rather terse), and namely what the
> precise problem was that you encountered? In particular I'm
> having trouble connecting the patch to the error messages noted
> in its description - afaics the SVM decoder shouldn't be used for
> the (generic) HVM MMIO path.
>
> Thanks, Jan
>
> (original mail was Cc-ed to xen-devel, please add this back on your reply)
Following setup did not work on one of our PC-Ware boxes with
a Rev.F Athlon X2 and xen-3.2-testing:
Guest: SLES10 64bit
Memory: 2GB
VCPUs: 2
The guest becomes unresponsive and is using 100% CPU. I tried to assign only 1
VCPU, but the result was the same.
The exact same test did work on a Family10h box.
xm dmesg gets flooded with this repeating two lines:
(XEN) platform.c:1049:d6 handle_mmio: failed to get instruction
(XEN) instrlen.c:252:d6 Cannot read from address 802eb001 (eip 802eb001, mode
2)
The specific changeset has been nailed down to:
16984:9750ac9b7e8b bad
16932:d31315b23f28 bad
16907:0016f5a1dd5a bad
16907 with 16898 reverted good
16898:784805591fa2 ? that's the cause
16888:9733f681d5c5 good
16881:4073b3ded545 good (3.2.1 release)
16718:f4a57e0474af good (3.2.0 release)
Checked following xen-unstable changesets on the same box:
18510:694b7daa353c (current tip)
17575:01aa7c088e98 (SVM: clean up __get_instruction_length_from_list())
Both are working. So it's only a problem within xen-3.2-testing.
The testing results shown so far mean that c/s 16898 has a dependency
to at least one c/s in xen-unstable prior to 17575 that has *not* been
backported to Xen-3.2-testing.
The issue is described in c/s 17202 (xen-unstable tree):
-----------------------------------------------------------------
description: SVM: handle page faults in emulated instruction fetches
Deal with failures in hvm_copy_from_guest_virt when fetching
instructions in the various SVM emulation paths. Since we know that
the instruction was fetchable by the hardware, we can usually just
return from the VMEXIT and try again; whatever caused us to fail will
cause the hardware to fail next time and we'll get the correct exit
code.
-----------------------------------------------------------------
Unfortunately, the SVM emulation code has been rewritten in
c/s 17090 (xen-unstable tree) and not been backported to Xen 3.2.x.
Therefore, the real fix is way more complicated than the diff
from c/s 17202.
Hope, this helps Jan.
Christoph
--
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach b. Muenchen
Geschaeftsfuehrer: Alberto Bozzo, Andrew Bowd
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home