Date:   Mon, 23 Jan 2023 18:18:03 +0300
From:   "Kirill A. Shutemov" <kirill@...temov.name>
To:     Vlastimil Babka <vbabka@...e.cz>
Cc:     "Huang, Kai" <kai.huang@...el.com>,
        "chao.p.peng@...ux.intel.com" <chao.p.peng@...ux.intel.com>,
        "tglx@...utronix.de" <tglx@...utronix.de>,
        "linux-arch@...r.kernel.org" <linux-arch@...r.kernel.org>,
        "kvm@...r.kernel.org" <kvm@...r.kernel.org>,
        "jmattson@...gle.com" <jmattson@...gle.com>,
        "Hocko, Michal" <mhocko@...e.com>,
        "pbonzini@...hat.com" <pbonzini@...hat.com>,
        "ak@...ux.intel.com" <ak@...ux.intel.com>,
        "Lutomirski, Andy" <luto@...nel.org>,
        "linux-fsdevel@...r.kernel.org" <linux-fsdevel@...r.kernel.org>,
        "tabba@...gle.com" <tabba@...gle.com>,
        "david@...hat.com" <david@...hat.com>,
        "michael.roth@....com" <michael.roth@....com>,
        "kirill.shutemov@...ux.intel.com" <kirill.shutemov@...ux.intel.com>,
        "corbet@....net" <corbet@....net>,
        "qemu-devel@...gnu.org" <qemu-devel@...gnu.org>,
        "dhildenb@...hat.com" <dhildenb@...hat.com>,
        "bfields@...ldses.org" <bfields@...ldses.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "x86@...nel.org" <x86@...nel.org>, "bp@...en8.de" <bp@...en8.de>,
        "ddutile@...hat.com" <ddutile@...hat.com>,
        "rppt@...nel.org" <rppt@...nel.org>,
        "shuah@...nel.org" <shuah@...nel.org>,
        "vkuznets@...hat.com" <vkuznets@...hat.com>,
        "mail@...iej.szmigiero.name" <mail@...iej.szmigiero.name>,
        "naoya.horiguchi@....com" <naoya.horiguchi@....com>,
        "qperret@...gle.com" <qperret@...gle.com>,
        "arnd@...db.de" <arnd@...db.de>,
        "linux-api@...r.kernel.org" <linux-api@...r.kernel.org>,
        "yu.c.zhang@...ux.intel.com" <yu.c.zhang@...ux.intel.com>,
        "Christopherson,, Sean" <seanjc@...gle.com>,
        "wanpengli@...cent.com" <wanpengli@...cent.com>,
        "vannapurve@...gle.com" <vannapurve@...gle.com>,
        "hughd@...gle.com" <hughd@...gle.com>,
        "aarcange@...hat.com" <aarcange@...hat.com>,
        "mingo@...hat.com" <mingo@...hat.com>,
        "hpa@...or.com" <hpa@...or.com>,
        "Nakajima, Jun" <jun.nakajima@...el.com>,
        "jlayton@...nel.org" <jlayton@...nel.org>,
        "joro@...tes.org" <joro@...tes.org>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>,
        "Wang, Wei W" <wei.w.wang@...el.com>,
        "steven.price@....com" <steven.price@....com>,
        "linux-doc@...r.kernel.org" <linux-doc@...r.kernel.org>,
        "Hansen, Dave" <dave.hansen@...el.com>,
        "akpm@...ux-foundation.org" <akpm@...ux-foundation.org>,
        "linmiaohe@...wei.com" <linmiaohe@...wei.com>
Subject: Re: [PATCH v10 1/9] mm: Introduce memfd_restricted system call to
 create restricted user memory

On Mon, Jan 23, 2023 at 03:03:45PM +0100, Vlastimil Babka wrote:
> On 12/22/22 01:37, Huang, Kai wrote:
> >>> I argue that this page pinning (or page migration prevention) is not
> >>> tied to where the page comes from, instead related to how the page will
> >>> be used. Whether the page is restrictedmem backed or GUP() backed, once
> >>> it's used by current version of TDX then the page pinning is needed. So
> >>> such page migration prevention is really TDX thing, even not KVM generic
> >>> thing (that's why I think we don't need change the existing logic of
> >>> kvm_release_pfn_clean()). 
> >>>
> > This essentially boils down to who "owns" page migration handling, and sadly,
> > page migration is kinda "owned" by the core-kernel, i.e. KVM cannot handle page
> > migration by itself -- it's just a passive receiver.
> > 
> > For normal pages, page migration is totally done by the core-kernel (i.e. it
> > unmaps page from VMA, allocates a new page, and uses migrate_pape() or a_ops-
> >> migrate_page() to actually migrate the page).
> > In the sense of TDX, conceptually it should be done in the same way. The more
> > important thing is: yes KVM can use get_page() to prevent page migration, but
> > when KVM wants to support it, KVM cannot just remove get_page(), as the core-
> > kernel will still just do migrate_page() which won't work for TDX (given
> > restricted_memfd doesn't have a_ops->migrate_page() implemented).
> > 
> > So I think the restricted_memfd filesystem should own page migration handling,
> > (i.e. by implementing a_ops->migrate_page() to either just reject page migration
> > or somehow support it).
> 
> While this thread seems to be settled on refcounts already, just wanted
> to point out that it wouldn't be ideal to prevent migrations by
> a_ops->migrate_page() rejecting them. It would mean cputime wasted (i.e.
> by memory compaction) by isolating the pages for migration and then
> releasing them after the callback rejects it (at least we wouldn't waste
> time creating and undoing migration entries in the userspace page tables
> as there's no mmap). Elevated refcount on the other hand is detected
> very early in compaction so no isolation is attempted, so from that
> aspect it's optimal.

Hm. Do we need a new hook in a_ops to check if the page is migratable
before going with longer path to migrate_page().

Or maybe add AS_UNMOVABLE?

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Powered by blists - more mailing lists