[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] x86/shadow: don't use map_domain_page_global() on paths that may not fail

>>> On 22.02.19 at 13:33, <tim@xxxxxxx> wrote:
> At 08:15 -0700 on 20 Feb (1550650529), Jan Beulich wrote:
>> The assumption (according to one comment) and hope (according to
>> another) that map_domain_page_global() can't fail are both wrong on
>> large enough systems. Do away with the guest_vtable field altogether,
>> and establish / tear down the desired mapping as necessary.
>> The alternatives, discarded as being undesirable, would have been to
>> either crash the guest in sh_update_cr3() when the mapping fails, or to
>> bubble up an error indicator, which upper layers would have a hard time
>> to deal with (other than again by crashing the guest).
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> I follow your argument, so Acked-by: Tim Deegan <tim@xxxxxxx>


> I would expect this to have a measurable cost on page fault times (on
> configurations where global map isn't just a directmap).  It would be
> good to know if that's the case.

I have absolutely no doubt that this will cost performance on huge
systems (and debug builds). That's basically why I've explicitly listed
the (discarded) options that would have had less overhead.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.