https://bugzilla.novell.com/show_bug.cgi?id=413842
User jbeulich@novell.com added comment
https://bugzilla.novell.com/show_bug.cgi?id=413842#c27
--- Comment #27 from Jan Beulich 2008-08-15 06:39:26 MDT ---
This looks pretty certainly like a hardware issue (wbinvd works while clflush
doesn't, albeit there's a perhaps signficant difference [also to native] in
that clflush executes directly in the dom0 kernel [as it's a non-privileged
instruction], i.e. at ring 3, while wbinvd gets trapped and executed in the
hypervisor, i.e. at ring 0).
The only way software could be involved here is if there was ever a cachable
mapping established for the bus address range in question, which there is no
indication of (and hence clflush should actually be a no-op on that region).
Neverthless, could you provide the MTRR settings under both native and Xen
kernel (here with the mptspi driver loading suppressed), as reported through
/proc/mtrr?
The other somewhat reasonable alternative would be if this happens (due to a
bug - generally the condition right at the beginning of the function should be
true and hence prevent the body of the function from being executed, but 2.6.27
fixes the condition to use >= rather than > as the relation) in the context of
or after executing cpa_process_alias() - the address you provided
(FFFFC200100C0000) indicates we aren't in the context of that function, but we
don't know whether the previous two pages got us in there (perhaps due to that
wrong relation).
--
Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.