Re: [RFC PATCH 07/26] hugetlb: add hugetlb_pte to track HugeTLB page table entries

From: Mike Kravetz <mike.kravetz@oracle.com>
To: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: James Houghton <jthoughton@google.com>,
	Muchun Song <songmuchun@bytedance.com>,
	Peter Xu <peterx@redhat.com>,
	David Hildenbrand <david@redhat.com>,
	David Rientjes <rientjes@google.com>,
	Axel Rasmussen <axelrasmussen@google.com>,
	Mina Almasry <almasrymina@google.com>, Jue Wang <juew@google.com>,
	Manish Mishra <manish.mishra@nutanix.com>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [RFC PATCH 07/26] hugetlb: add hugetlb_pte to track HugeTLB page table entries
Date: Tue, 12 Jul 2022 10:51:06 -0700	[thread overview]
Message-ID: <Ys20iusvF1An6BLs@monkey> (raw)
In-Reply-To: <Ys1B+fXo2uSPd46B@work-vm>

On 07/12/22 10:42, Dr. David Alan Gilbert wrote:
> * Mike Kravetz (mike.kravetz@oracle.com) wrote:
> > On 06/24/22 17:36, James Houghton wrote:
> > > After high-granularity mapping, page table entries for HugeTLB pages can
> > > be of any size/type. (For example, we can have a 1G page mapped with a
> > > mix of PMDs and PTEs.) This struct is to help keep track of a HugeTLB
> > > PTE after we have done a page table walk.
> > 
> > This has been rolling around in my head.
> > 
> > Will this first use case (live migration) actually make use of this
> > 'mixed mapping' model where hugetlb pages could be mapped at the PUD,
> > PMD and PTE level all within the same vma?  I only understand the use
> > case from a high level.  But, it seems that we would want to only want
> > to migrate PTE (or PMD) sized pages and not necessarily a mix.
> 
> I suspect we would pick one size and use that size for all transfers
> when in postcopy; not sure if there are any side cases though.
> 
> > The only reason I ask is because the code might be much simpler if all
> > mappings within a vma were of the same size.  Of course, the
> > performance/latency of converting a large mapping may be prohibitively
> > expensive.
> 
> Imagine we're migrating a few TB VM, backed by 1GB hugepages, I'm guessing it
> would be nice to clean up the PTE/PMDs for split 1GB pages as they're
> completed rather than having thousands of them for the whole VM.
> (I'm not sure if that is already doable)

Seems that would be doable by calling MADV_COLLAPSE for 1GB pages as
they are completed.

Thanks for information on post copy.
-- 
Mike Kravetz