From: Michel Lespinasse <walken@google.com>
To: Sasha Levin <levinsasha928@gmail.com>
Cc: linux-mm@kvack.org, riel@redhat.com, peterz@infradead.org,
aarcange@redhat.com, hughd@google.com, daniel.santos@pobox.com,
linux-kernel@vger.kernel.org, akpm@linux-foundation.org,
Dave Jones <davej@redhat.com>, Jiri Slaby <jslaby@suse.cz>
Subject: Re: [PATCH 6/7] mm: add CONFIG_DEBUG_VM_RB build option
Date: Fri, 14 Sep 2012 17:00:29 -0700 [thread overview]
Message-ID: <20120915000029.GA29426@google.com> (raw)
In-Reply-To: <CANN689Ff3W4z=+3J8aGO-2GrPHGJ=ote_f5q9jzRQRAP+b0T4Q@mail.gmail.com>
On Fri, Sep 14, 2012 at 3:46 PM, Michel Lespinasse <walken@google.com> wrote:
> On Fri, Sep 14, 2012 at 3:14 PM, Sasha Levin <levinsasha928@gmail.com> wrote:
>> On 09/04/2012 11:20 AM, Michel Lespinasse wrote:
>>> Add a CONFIG_DEBUG_VM_RB build option for the previously existing
>>> DEBUG_MM_RB code. Now that Andi Kleen modified it to avoid using
>>> recursive algorithms, we can expose it a bit more.
>>>
>>> Also extend this code to validate_mm() after stack expansion, and to
>>> check that the vma's start and last pgoffs have not changed since the
>>> nodes were inserted on the anon vma interval tree (as it is important
>>> that the nodes be reindexed after each such update).
>>
>> This patch exposes the following warning:
>>
>> [ 24.977502] ------------[ cut here ]------------
>> [ 24.979089] WARNING: at mm/interval_tree.c:110
>> anon_vma_interval_tree_verify+0x81/0xa0()
>> [ 24.981765] Pid: 5928, comm: trinity-child37 Tainted: G W
>> 3.6.0-rc5-next-20120914-sasha-00003-g7deb7fa-dirty #333
>> [ 24.985501] Call Trace:
>> [ 24.986345] [<ffffffff81224c91>] ? anon_vma_interval_tree_verify+0x81/0xa0
>> [ 24.988535] [<ffffffff81106766>] warn_slowpath_common+0x86/0xb0
>> [ 24.990636] [<ffffffff81106855>] warn_slowpath_null+0x15/0x20
>> [ 24.992658] [<ffffffff81224c91>] anon_vma_interval_tree_verify+0x81/0xa0
>> [ 24.994980] [<ffffffff8122e6e8>] validate_mm+0x58/0x1e0
>> [ 24.996772] [<ffffffff8122e934>] vma_link+0x94/0xe0
>> [ 24.997719] [<ffffffff812315e9>] copy_vma+0x279/0x2e0
>> [ 24.998522] [<ffffffff8117a7fd>] ? trace_hardirqs_off+0xd/0x10
>> [ 25.000772] [<ffffffff81232e89>] move_vma+0xa9/0x260
>> [ 25.002499] [<ffffffff812334b5>] sys_mremap+0x475/0x540
>> [ 25.004364] [<ffffffff8374b6e8>] tracesys+0xe1/0xe6
>> [ 25.006108] ---[ end trace 7c901670963aa6e2 ]---
>>
>> The code line is
>>
>> WARN_ON_ONCE(node->cached_vma_last != avc_last_pgoff(node));
>
> That's very interesting (and potentially relevant to another bug
> that's been reported too).
>
> I'd like to know, what workload did you use that triggered this ?
> (I find it hard to test mremap as I don't know of enough users of it)
All right. Hugh managed to reproduce the issue on his suse laptop, and
I came up with a fix.
The problem was that in mremap, the new vma's vm_{start,end,pgoff}
fields need to be updated before calling anon_vma_clone() so that the
new vma will be properly indexed.
Patch attached. I expect this should also explain Jiri's reported
failure involving splitting THP pages during mremap(), even though we
did not manage to reproduce that one.
---------------------------------8<-------------------------------
From: Michel Lespinasse <walken@google.com>
Date: Fri, 14 Sep 2012 16:43:49 -0700
Subject: [PATCH] mm anon rmap: in mremap, set the new vma's position before
anon_vma_clone()
anon_vma_clone() expects new_vma->vm_{start,end,pgoff} to be correctly set
so that the new vma can be indexed on the anon interval tree.
copy_vma() was failing to do that, which broke mremap().
Signed-off-by: Michel Lespinasse <walken@google.com>
---
mm/mmap.c | 8 ++++----
1 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/mm/mmap.c b/mm/mmap.c
index cc8c64077a42..7e672800b5d4 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2446,16 +2446,16 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
if (new_vma) {
*new_vma = *vma;
+ new_vma->vm_start = addr;
+ new_vma->vm_end = addr + len;
+ new_vma->vm_pgoff = pgoff;
pol = mpol_dup(vma_policy(vma));
if (IS_ERR(pol))
goto out_free_vma;
+ vma_set_policy(new_vma, pol);
INIT_LIST_HEAD(&new_vma->anon_vma_chain);
if (anon_vma_clone(new_vma, vma))
goto out_free_mempol;
- vma_set_policy(new_vma, pol);
- new_vma->vm_start = addr;
- new_vma->vm_end = addr + len;
- new_vma->vm_pgoff = pgoff;
if (new_vma->vm_file) {
get_file(new_vma->vm_file);
--
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.
next prev parent reply other threads:[~2012-09-15 0:00 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-09-04 9:20 [PATCH 0/7] use interval trees for anon rmap Michel Lespinasse
2012-09-04 9:20 ` [PATCH 1/7] mm: interval tree updates Michel Lespinasse
2012-09-07 22:13 ` Andrew Morton
2012-09-07 22:29 ` Michel Lespinasse
2012-09-07 22:55 ` Andrew Morton
2012-09-07 23:26 ` Michel Lespinasse
2012-09-08 4:45 ` Hillf Danton
2012-09-07 23:26 ` Michel Lespinasse
2012-09-04 9:20 ` [PATCH 2/7] mm: fix potential anon_vma locking issue in mprotect() Michel Lespinasse
2012-09-04 14:27 ` Andrea Arcangeli
2012-09-04 21:53 ` Michel Lespinasse
2012-09-04 22:16 ` Andrea Arcangeli
2012-09-05 0:45 ` Michel Lespinasse
2012-09-04 9:20 ` [PATCH 3/7] mm anon rmap: remove anon_vma_moveto_tail Michel Lespinasse
2012-09-04 9:20 ` [PATCH 4/7] mm anon rmap: replace same_anon_vma linked list with an interval tree Michel Lespinasse
2012-09-05 0:51 ` Michel Lespinasse
2012-09-04 9:20 ` [PATCH 5/7] mm rmap: remove vma_address check for address inside vma Michel Lespinasse
2012-09-04 9:20 ` [PATCH 6/7] mm: add CONFIG_DEBUG_VM_RB build option Michel Lespinasse
2012-09-14 22:14 ` Sasha Levin
2012-09-14 22:40 ` Sasha Levin
2012-09-14 22:46 ` Michel Lespinasse
2012-09-15 0:00 ` Michel Lespinasse [this message]
2012-09-15 7:52 ` Jiri Slaby
2012-09-16 19:07 ` Hugh Dickins
2012-09-22 7:19 ` Jiri Slaby
2012-09-15 9:26 ` Sasha Levin
2012-09-20 21:39 ` Fengguang Wu
2012-09-20 22:27 ` Hugh Dickins
2012-09-20 22:31 ` Fengguang Wu
2012-09-04 9:20 ` [PATCH 7/7] mm: avoid taking rmap locks in move_ptes() Michel Lespinasse
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=20120915000029.GA29426@google.com \
--to=walken@google.com \
--cc=aarcange@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=daniel.santos@pobox.com \
--cc=davej@redhat.com \
--cc=hughd@google.com \
--cc=jslaby@suse.cz \
--cc=levinsasha928@gmail.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=peterz@infradead.org \
--cc=riel@redhat.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).