From: Song Liu <songliubraving@fb.com>
To: Oleg Nesterov <oleg@redhat.com>
Cc: lkml <linux-kernel@vger.kernel.org>,
"linux-mm@kvack.org" <linux-mm@kvack.org>,
"akpm@linux-foundation.org" <akpm@linux-foundation.org>,
"matthew.wilcox@oracle.com" <matthew.wilcox@oracle.com>,
"kirill.shutemov@linux.intel.com"
<kirill.shutemov@linux.intel.com>,
Kernel Team <Kernel-team@fb.com>,
"william.kucharski@oracle.com" <william.kucharski@oracle.com>,
"srikar@linux.vnet.ibm.com" <srikar@linux.vnet.ibm.com>
Subject: Re: [PATCH v2 1/2] khugepaged: enable collapse pmd for pte-mapped THP
Date: Fri, 2 Aug 2019 20:59:13 +0000 [thread overview]
Message-ID: <0893F3AE-CAC5-4241-BA69-585F268E482C@fb.com> (raw)
In-Reply-To: <20190802103112.GA20111@redhat.com>
> On Aug 2, 2019, at 3:31 AM, Oleg Nesterov <oleg@redhat.com> wrote:
>
> On 08/01, Song Liu wrote:
>>
>>
>>> On Aug 1, 2019, at 7:50 AM, Oleg Nesterov <oleg@redhat.com> wrote:
>>>
>>> On 07/31, Song Liu wrote:
>>>>
>>>> +static int khugepaged_add_pte_mapped_thp(struct mm_struct *mm,
>>>> + unsigned long addr)
>>>> +{
>>>> + struct mm_slot *mm_slot;
>>>> + int ret = 0;
>>>> +
>>>> + /* hold mmap_sem for khugepaged_test_exit() */
>>>> + VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
>>>> + VM_BUG_ON(addr & ~HPAGE_PMD_MASK);
>>>> +
>>>> + if (unlikely(khugepaged_test_exit(mm)))
>>>> + return 0;
>>>> +
>>>> + if (!test_bit(MMF_VM_HUGEPAGE, &mm->flags) &&
>>>> + !test_bit(MMF_DISABLE_THP, &mm->flags)) {
>>>> + ret = __khugepaged_enter(mm);
>>>> + if (ret)
>>>> + return ret;
>>>> + }
>>>
>>> could you explain why do we need mm->mmap_sem, khugepaged_test_exit() check
>>> and __khugepaged_enter() ?
>>
>> If the mm doesn't have a mm_slot, we would like to create one here (by
>> calling __khugepaged_enter()).
>
> I can be easily wrong, I never read this code before, but this doesn't
> look correct.
>
> Firstly, mm->mmap_sem cam ONLY help if a) the task already has mm_slot
> and b) this mm_slot is khugepaged_scan.mm_slot. Otherwise khugepaged_exit()
> won't take mmap_sem for writing and thus we can't rely on test_exit().
>
> and this means that down_read(mmap_sem) before khugepaged_add_pte_mapped_thp()
> is pointless and can't help; this mm was found by vma_interval_tree_foreach().
>
> so __khugepaged_enter() can race with khugepaged_exit() and this is wrong
> in any case.
>
>> This happens when the THP is created by another mm, or by tmpfs with
>> "huge=always"; and then page table of this mm got split by split_huge_pmd().
>> With current kernel, this happens when we attach/detach uprobe to a file
>> in tmpfs with huge=always.
>
> Well. In this particular case khugepaged_enter() was likely already called
> by shmem_mmap() or khugepaged_enter_vma_merge(), or madvise.
>
> (in fact I think do_set_pmd() or shmem_fault() should call _enter() too,
> like do_huge_pmd_anonymous_page() does, but this is another story).
Hmm.. I didn't notice the one in shmem_mmap(). And yes, you are right, we
don't really need __khugepaged_enter() in khugepaged_add_pte_mapped_thp(),
especially when uprobes.c doesn't call it.
This should simplify this patch.
>
>
> And I forgot to mention... I don't understand why
> khugepaged_collapse_pte_mapped_thps() has to be called with khugepaged_mm_lock.
You are right that khugepaged_collapse_pte_mapped_thps() doesn't need
khugepaged_mm_lock in this version. For v1, when uprobes.c calls
khugepaged_add_pte_mapped_thp(), this is necessary.
Let me clean this up.
Thanks,
Song
next prev parent reply other threads:[~2019-08-02 20:59 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-07-31 18:33 [PATCH v2 0/2] khugepaged: collapse pmd for pte-mapped THP Song Liu
2019-07-31 18:33 ` [PATCH v2 1/2] khugepaged: enable " Song Liu
2019-08-01 12:43 ` Oleg Nesterov
2019-08-01 17:11 ` Song Liu
2019-08-01 14:50 ` Oleg Nesterov
2019-08-01 17:37 ` Song Liu
2019-08-02 10:31 ` Oleg Nesterov
2019-08-02 20:59 ` Song Liu [this message]
2019-07-31 18:33 ` [PATCH v2 2/2] uprobe: collapse THP pmd after removing all uprobes Song Liu
2019-08-01 11:19 ` [PATCH v2 0/2] khugepaged: collapse pmd for pte-mapped THP Kirill A. Shutemov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=0893F3AE-CAC5-4241-BA69-585F268E482C@fb.com \
--to=songliubraving@fb.com \
--cc=Kernel-team@fb.com \
--cc=akpm@linux-foundation.org \
--cc=kirill.shutemov@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=matthew.wilcox@oracle.com \
--cc=oleg@redhat.com \
--cc=srikar@linux.vnet.ibm.com \
--cc=william.kucharski@oracle.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).