Re: [PATCH v1 1/9] mm: Separate free page chunk merging into its own routine

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 03/28/2017 03:44 PM, Wei Liu wrote:
> On Tue, Mar 28, 2017 at 03:41:10PM -0400, Boris Ostrovsky wrote:
>> On 03/28/2017 03:20 PM, Wei Liu wrote:
>>> On Fri, Mar 24, 2017 at 01:04:56PM -0400, Boris Ostrovsky wrote:
>>>> This is needed for subsequent changes to memory scrubbing. No
>>>> logic change, only code re-factoring.
>>> Actually there is a slight change in logic: pg and order could be
>>> updated in the original merge code. With this patch you still pass the
>>> original pg to reserve_offlined_page if tainted is true. I don't think
>>> this matters in terms of correctness, but it is worth pointing out.
>> Actually, for pg it does matter since reserve_offlined_page() is passed
>> buddy head. So I should make merge_chunks() return new head.
>>
> OK. I think the downside is reserve_offlined_page doesn't merge as many
> pages as it should have, so it is still correct, just the result is
> suboptimal, but I could be wrong -- it's a bit late here. :-)

You are wrong that my patch was right. ;-)

I was passing old head which, if it was merged back in merge_chunks(),
would have lost its PGC_need_scrub bit. And so when
reserve_offlined_page() tries to re-merge buddy's pieces all need_scrub
info would be lost.

Again, this would have been OK with previous version of the series where
every page in the buddy was marked with PGC_need_scrub. But this is no
longer the case.

-boris


>
> I think returning the updated pg is probably best.
>
> Wei.
>
>> ('order' is not used after merge_chunks(), so that's OK).
>>
>> -boris


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