[Xen-devel] 2nd opinion on backportability of c35eefded2

["Jan Beulich" <JBeulich@xxxxxxxx>]

Hi George,

after quite a bit of debugging on 4.6.1 I learned that said commit
("x86/P2M: consolidate handling of types not requiring a valid MFN")
is more than just cleanup: Since p2m_set_entry() happily performs
arithmetic on the passed in MFN, shadow mode guests (verified) as
well as HAP ones when 1Gb / 2Mb mappings are unavailable (not
verified), if any of the MFNs below 1Gb are invalid (reported with
e.g. "crashkernel=512M@16M"), p2m_pt_set_entry() would (in
the context of guest_physmap_mark_populate_on_demand())
produce invalid instead of PoD entries, resulting in subsequent
attempts by the guest to use (e.g. balloon out) these pages to
fail (the balloon failure results in a crash during boot).

Now, while the backport to 4.6 is straightforward, I'm having the
vague feeling that this change might depend on some earlier one,
but I can't pinpoint anything. Hence I would appreciate if you
could take a look and provide your judgment.

Thanks, Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel