linux-kernel - Re: [External] Re: [PATCH v9 03/11] mm/hugetlb: Free the vmemmap pages associated with each HugeTLB page

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAMZfGtU4QRXp_ufJX_XZzbANbhLO1sdGvDEe+ugTg7WfVc-VYA@mail.gmail.com>
Date:   Thu, 17 Dec 2020 17:05:07 +0800
From:   Muchun Song <songmuchun@...edance.com>
To:     Mike Kravetz <mike.kravetz@...cle.com>,
        Oscar Salvador <osalvador@...e.de>
Cc:     Jonathan Corbet <corbet@....net>,
        Thomas Gleixner <tglx@...utronix.de>, mingo@...hat.com,
        bp@...en8.de, x86@...nel.org, hpa@...or.com,
        dave.hansen@...ux.intel.com, luto@...nel.org,
        Peter Zijlstra <peterz@...radead.org>, viro@...iv.linux.org.uk,
        Andrew Morton <akpm@...ux-foundation.org>, paulmck@...nel.org,
        mchehab+huawei@...nel.org, pawan.kumar.gupta@...ux.intel.com,
        Randy Dunlap <rdunlap@...radead.org>, oneukum@...e.com,
        anshuman.khandual@....com, jroedel@...e.de,
        Mina Almasry <almasrymina@...gle.com>,
        David Rientjes <rientjes@...gle.com>,
        Matthew Wilcox <willy@...radead.org>,
        Michal Hocko <mhocko@...e.com>,
        "Song Bao Hua (Barry Song)" <song.bao.hua@...ilicon.com>,
        David Hildenbrand <david@...hat.com>,
        Xiongchun duan <duanxiongchun@...edance.com>,
        linux-doc@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>,
        Linux Memory Management List <linux-mm@...ck.org>,
        linux-fsdevel <linux-fsdevel@...r.kernel.org>
Subject: Re: [External] Re: [PATCH v9 03/11] mm/hugetlb: Free the vmemmap
 pages associated with each HugeTLB page

On Thu, Dec 17, 2020 at 2:54 PM Muchun Song <songmuchun@...edance.com> wrote:
>
> On Thu, Dec 17, 2020 at 6:52 AM Mike Kravetz <mike.kravetz@...cle.com> wrote:
> >
> > On 12/16/20 2:25 PM, Oscar Salvador wrote:
> > > On Wed, Dec 16, 2020 at 02:08:30PM -0800, Mike Kravetz wrote:
> > >>> + * vmemmap_rmap_walk - walk vmemmap page table
> > >>> +
> > >>> +static void vmemmap_pte_range(pmd_t *pmd, unsigned long addr,
> > >>> +                         unsigned long end, struct vmemmap_rmap_walk *walk)
> > >>> +{
> > >>> +   pte_t *pte;
> > >>> +
> > >>> +   pte = pte_offset_kernel(pmd, addr);
> > >>> +   do {
> > >>> +           BUG_ON(pte_none(*pte));
> > >>> +
> > >>> +           if (!walk->reuse)
> > >>> +                   walk->reuse = pte_page(pte[VMEMMAP_TAIL_PAGE_REUSE]);
> > >>
> > >> It may be just me, but I don't like the pte[-1] here.  It certainly does work
> > >> as designed because we want to remap all pages in the range to the page before
> > >> the range (at offset -1).  But, we do not really validate this 'reuse' page.
> > >> There is the BUG_ON(pte_none(*pte)) as a sanity check, but we do nothing similar
> > >> for pte[-1].  Based on the usage for HugeTLB pages, we can be confident that
> > >> pte[-1] is actually a pte.  In discussions with Oscar, you mentioned another
> > >> possible use for these routines.
> > >
> > > Without giving it much of a thought, I guess we could duplicate the
> > > BUG_ON for the pte outside the loop, and add a new one for pte[-1].
> > > Also, since walk->reuse seems to not change once it is set, we can take
> > > it outside the loop? e.g:
> > >
> > >       pte *pte;
> > >
> > >       pte = pte_offset_kernel(pmd, addr);
> > >       BUG_ON(pte_none(*pte));
> > >       BUG_ON(pte_none(pte[VMEMMAP_TAIL_PAGE_REUSE]));
> > >       walk->reuse = pte_page(pte[VMEMMAP_TAIL_PAGE_REUSE]);
> > >       do {
> > >               ....
> > >       } while...
> > >
> > > Or I am not sure whether we want to keep it inside the loop in case
> > > future cases change walk->reuse during the operation.
> > > But to be honest, I do not think it is realistic of all future possible
> > > uses of this, so I would rather keep it simple for now.
> >
> > I was thinking about possibly passing the 'reuse' address as another parameter
> > to vmemmap_remap_reuse().  We could add this addr to the vmemmap_rmap_walk
> > struct and set walk->reuse when we get to the pte for that address.  Of
> > course this would imply that the addr would need to be part of the range.
>
> Maybe adding another one parameter is unnecessary.  How about doing
> this in the vmemmap_remap_reuse?
>
> The 'reuse' address just is start + PAGE_SIZE.
>
> void vmemmap_remap_free(unsigned long start, unsigned long size)
> {
>          unsigned long end = start + size;
>          unsigned long reuse_addr = start + PAGE_SIZE;
                                           ^^^
                                        Here is "-"
Sorry.

>          LIST_HEAD(vmemmap_pages);
>
>          struct vmemmap_remap_walk walk = {
>                   .remap_pte = vmemmap_remap_pte,
>                   .vmemmap_pages = &vmemmap_pages,
>                   .reuse_addr = reuse_addr.
>          };
>
> }
>
> >
> > Ideally, we would walk the page table to get to the reuse page.  My concern
> > was not explicitly about adding the BUG_ON.  In more general use, *pte could
> > be the first entry on a pte page.  And, then pte[-1] may not even be a pte.
> >
> > Again, I don't think this matters for the current HugeTLB use case.  Just a
> > little concerned if code is put to use for other purposes.
> > --
> > Mike Kravetz
>
>
>
> --
> Yours,
> Muchun



-- 
Yours,
Muchun