[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20130411131813.17FD7E0085@blue.fi.intel.com>
Date: Thu, 11 Apr 2013 16:18:13 +0300 (EEST)
From: "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>
To: Minchan Kim <minchan@...nel.org>
Cc: Sasha Levin <sasha.levin@...cle.com>,
"Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
Andrew Morton <akpm@...ux-foundation.org>,
David Rientjes <rientjes@...gle.com>,
Andrea Arcangeli <aarcange@...hat.com>,
"H. Peter Anvin" <hpa@...or.com>, Mel Gorman <mgorman@...e.de>,
Dave Jones <davej@...hat.com>, linux-mm <linux-mm@...ck.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Dave Hansen <dave@...1.net>
Subject: Re: mm: BUG in do_huge_pmd_wp_page
Minchan Kim wrote:
> On Fri, Mar 29, 2013 at 09:04:16AM -0400, Sasha Levin wrote:
> > Hi all,
> >
> > While fuzzing with trinity inside a KVM tools guest running latest -next kernel,
> > I've stumbled on the following.
> >
> > It seems that the code in do_huge_pmd_wp_page() was recently modified in
> > "thp: do_huge_pmd_wp_page(): handle huge zero page".
> >
> > Here's the trace:
> >
> > [ 246.244708] BUG: unable to handle kernel paging request at ffff88009c422000
> > [ 246.245743] IP: [<ffffffff81a0a795>] copy_page_rep+0x5/0x10
> > [ 246.250569] PGD 7232067 PUD 7235067 PMD bfefe067 PTE 800000009c422060
> > [ 246.251529] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
> > [ 246.252325] Dumping ftrace buffer:
> > [ 246.252791] (ftrace buffer empty)
> > [ 246.252869] Modules linked in:
> > [ 246.252869] CPU 3
> > [ 246.252869] Pid: 11985, comm: trinity-child12 Tainted: G W 3.9.0-rc4-next-20130328-sasha-00014-g91a3267 #319
> > [ 246.252869] RIP: 0010:[<ffffffff81a0a795>] [<ffffffff81a0a795>] copy_page_rep+0x5/0x10
> > [ 246.252869] RSP: 0018:ffff88000015bc40 EFLAGS: 00010286
> > [ 246.252869] RAX: ffff88000015bfd8 RBX: 0000000002710880 RCX: 0000000000000200
> > [ 246.252869] RDX: 0000000000000000 RSI: ffff88009c422000 RDI: ffff88009a422000
> > [ 246.252869] RBP: ffff88000015bc98 R08: 0000000002718000 R09: 0000000000000001
> > [ 246.252869] R10: 0000000000000001 R11: 0000000000000000 R12: ffff880000000000
> > [ 246.252869] R13: ffff88000015bfd8 R14: ffff88000015bfd8 R15: fffffffffff80000
> > [ 246.252869] FS: 00007f53db93f700(0000) GS:ffff8800bba00000(0000) knlGS:0000000000000000
> > [ 246.252869] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [ 246.252869] CR2: ffff88009c422000 CR3: 0000000000159000 CR4: 00000000000406e0
> > [ 246.252869] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > [ 246.252869] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> > [ 246.252869] Process trinity-child12 (pid: 11985, threadinfo ffff88000015a000, task ffff88009c60b000)
> > [ 246.252869] Stack:
> > [ 246.252869] ffffffff81234aae ffff88000015bc88 ffffffff81273639 0000000000a00000
> > [ 246.252869] 0000000002718000 ffff8800ab36d050 ffff880000153800 ffffea0002690000
> > [ 246.252869] 0000000000a00000 ffff8800ab36d000 ffffea0002710000 ffff88000015bd48
> > [ 246.252869] Call Trace:
> > [ 246.252869] [<ffffffff81234aae>] ? copy_user_huge_page+0x1de/0x240
> > [ 246.252869] [<ffffffff81273639>] ? mem_cgroup_charge_common+0xa9/0xc0
> > [ 246.252869] [<ffffffff8126b4d7>] do_huge_pmd_wp_page+0x9f7/0xc60
> > [ 246.252869] [<ffffffff81a0acd9>] ? __const_udelay+0x29/0x30
> > [ 246.252869] [<ffffffff8123364e>] handle_mm_fault+0x26e/0x650
> > [ 246.252869] [<ffffffff8117dc1a>] ? __lock_is_held+0x5a/0x80
> > [ 246.252869] [<ffffffff83db3814>] ? __do_page_fault+0x514/0x5e0
> > [ 246.252869] [<ffffffff83db3870>] __do_page_fault+0x570/0x5e0
> > [ 246.252869] [<ffffffff811c6500>] ? rcu_eqs_exit_common+0x60/0x260
> > [ 246.252869] [<ffffffff811c740e>] ? rcu_eqs_enter_common+0x33e/0x3b0
> > [ 246.252869] [<ffffffff811c679c>] ? rcu_eqs_exit+0x9c/0xb0
> > [ 246.252869] [<ffffffff83db3912>] do_page_fault+0x32/0x50
> > [ 246.252869] [<ffffffff83db2ef0>] do_async_page_fault+0x30/0xc0
> > [ 246.252869] [<ffffffff83db01e8>] async_page_fault+0x28/0x30
> > [ 246.252869] Code: 90 90 90 90 90 90 9c fa 65 48 3b 06 75 14 65 48 3b 56 08 75 0d 65 48 89 1e 65 48 89 4e 08 9d b0 01 c3 9d 30
> > c0 c3 b9 00 02 00 00 <f3> 48 a5 c3 0f 1f 80 00 00 00 00 eb ee 66 66 66 90 66 66 66 90
> > [ 246.252869] RIP [<ffffffff81a0a795>] copy_page_rep+0x5/0x10
> > [ 246.252869] RSP <ffff88000015bc40>
> > [ 246.252869] CR2: ffff88009c422000
> > [ 246.252869] ---[ end trace 09fbe37b108d5766 ]---
> >
> > And this is the code:
> >
> > if (is_huge_zero_pmd(orig_pmd))
> > clear_huge_page(new_page, haddr, HPAGE_PMD_NR);
> > else
> > copy_user_huge_page(new_page, page, haddr, vma, HPAGE_PMD_NR); <--- this
> >
> >
> > Thanks,
> > Sasha
>
> I don't know this issue was already resolved. If so, my reply become a just
> question to Kirill regardless of this BUG.
>
> When I am looking at the code, I was wonder about the logic of GHZP(aka,
> get_huge_zero_page) reference handling. The logic depends on that page
> allocator never alocate PFN 0.
>
> Who makes sure it? What happens if allocator allocates PFN 0?
> I don't know all of architecture makes sure it.
> You investigated it for all arches?
>
> If not,
> CPU 1 CPU 2 CPU 3
>
> shrink_huge_zero_page
> huge_zero_refcount = 0;
> GHZP
> pfn_0_zero_page = alloc_pages
> GHZP
> pfn_some_zero_page = alloc_page
> huge_zero_pfn = 0
> huge_zero_pfn = pfn_0
> huge_zero_refcount = 2
> huge_zero_pfn = pfn_some
> huge_zero_refcount = 2
>
> So, if you want to stick this logic, at least, don't we need BUG_ON to check
> pfn 0 allocation in get_huge_zero_page?
I don't think it's related to oops in the thread (I was not able to
reproduce it), but nice catch anyway.
What about the patch below?
=====
>From 4579aefd606b2dd82797af163ce6d08912894b3a Mon Sep 17 00:00:00 2001
From: "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>
Date: Thu, 11 Apr 2013 15:47:50 +0300
Subject: [PATCH] thp: fix huge zero page logic for page with pfn == 0
Current implementation of huge zero page uses pfn value 0 to indicate
that the page hasn't allocated yet. It assumes that buddy page allocator
can't return page with pfn == 0.
Let's rework the code to store 'struct page *' of huge zero page, not
its pfn. This way we can avoid the weak assumption.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@...ux.intel.com>
Reported-by: Minchan Kim <minchan@...nel.org>
---
mm/huge_memory.c | 43 +++++++++++++++++++++----------------------
1 file changed, 21 insertions(+), 22 deletions(-)
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 45eaae0..bc2a548 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -163,25 +163,24 @@ static int start_khugepaged(void)
}
static atomic_t huge_zero_refcount;
-static unsigned long huge_zero_pfn __read_mostly;
+static struct page *huge_zero_page __read_mostly;
-static inline bool is_huge_zero_pfn(unsigned long pfn)
+static inline bool is_huge_zero_page(struct page *page)
{
- unsigned long zero_pfn = ACCESS_ONCE(huge_zero_pfn);
- return zero_pfn && pfn == zero_pfn;
+ return ACCESS_ONCE(huge_zero_page) == page;
}
static inline bool is_huge_zero_pmd(pmd_t pmd)
{
- return is_huge_zero_pfn(pmd_pfn(pmd));
+ return is_huge_zero_page(pmd_page(pmd));
}
-static unsigned long get_huge_zero_page(void)
+static struct page *get_huge_zero_page(void)
{
struct page *zero_page;
retry:
if (likely(atomic_inc_not_zero(&huge_zero_refcount)))
- return ACCESS_ONCE(huge_zero_pfn);
+ return ACCESS_ONCE(huge_zero_page);
zero_page = alloc_pages((GFP_TRANSHUGE | __GFP_ZERO) & ~__GFP_MOVABLE,
HPAGE_PMD_ORDER);
@@ -191,7 +190,7 @@ retry:
}
count_vm_event(THP_ZERO_PAGE_ALLOC);
preempt_disable();
- if (cmpxchg(&huge_zero_pfn, 0, page_to_pfn(zero_page))) {
+ if (cmpxchg(&huge_zero_page, NULL, zero_page)) {
preempt_enable();
__free_page(zero_page);
goto retry;
@@ -200,7 +199,7 @@ retry:
/* We take additional reference here. It will be put back by shrinker */
atomic_set(&huge_zero_refcount, 2);
preempt_enable();
- return ACCESS_ONCE(huge_zero_pfn);
+ return ACCESS_ONCE(huge_zero_page);
}
static void put_huge_zero_page(void)
@@ -220,9 +219,9 @@ static int shrink_huge_zero_page(struct shrinker *shrink,
return atomic_read(&huge_zero_refcount) == 1 ? HPAGE_PMD_NR : 0;
if (atomic_cmpxchg(&huge_zero_refcount, 1, 0) == 1) {
- unsigned long zero_pfn = xchg(&huge_zero_pfn, 0);
- BUG_ON(zero_pfn == 0);
- __free_page(__pfn_to_page(zero_pfn));
+ struct page *zero_page = xchg(&huge_zero_page, NULL);
+ BUG_ON(zero_page == NULL);
+ __free_page(zero_page);
}
return 0;
@@ -764,12 +763,12 @@ static inline struct page *alloc_hugepage(int defrag)
static bool set_huge_zero_page(pgtable_t pgtable, struct mm_struct *mm,
struct vm_area_struct *vma, unsigned long haddr, pmd_t *pmd,
- unsigned long zero_pfn)
+ struct page *zero_page)
{
pmd_t entry;
if (!pmd_none(*pmd))
return false;
- entry = pfn_pmd(zero_pfn, vma->vm_page_prot);
+ entry = mk_pmd(zero_page, vma->vm_page_prot);
entry = pmd_wrprotect(entry);
entry = pmd_mkhuge(entry);
set_pmd_at(mm, haddr, pmd, entry);
@@ -794,20 +793,20 @@ int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
if (!(flags & FAULT_FLAG_WRITE) &&
transparent_hugepage_use_zero_page()) {
pgtable_t pgtable;
- unsigned long zero_pfn;
+ struct page *zero_page;
bool set;
pgtable = pte_alloc_one(mm, haddr);
if (unlikely(!pgtable))
return VM_FAULT_OOM;
- zero_pfn = get_huge_zero_page();
- if (unlikely(!zero_pfn)) {
+ zero_page = get_huge_zero_page();
+ if (unlikely(!zero_page)) {
pte_free(mm, pgtable);
count_vm_event(THP_FAULT_FALLBACK);
goto out;
}
spin_lock(&mm->page_table_lock);
set = set_huge_zero_page(pgtable, mm, vma, haddr, pmd,
- zero_pfn);
+ zero_page);
spin_unlock(&mm->page_table_lock);
if (!set) {
pte_free(mm, pgtable);
@@ -886,16 +885,16 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
* a page table.
*/
if (is_huge_zero_pmd(pmd)) {
- unsigned long zero_pfn;
+ struct page *zero_page;
bool set;
/*
* get_huge_zero_page() will never allocate a new page here,
* since we already have a zero page to copy. It just takes a
* reference.
*/
- zero_pfn = get_huge_zero_page();
+ zero_page = get_huge_zero_page();
set = set_huge_zero_page(pgtable, dst_mm, vma, addr, dst_pmd,
- zero_pfn);
+ zero_page);
BUG_ON(!set); /* unexpected !pmd_none(dst_pmd) */
ret = 0;
goto out_unlock;
@@ -1803,7 +1802,7 @@ int split_huge_page(struct page *page)
struct anon_vma *anon_vma;
int ret = 1;
- BUG_ON(is_huge_zero_pfn(page_to_pfn(page)));
+ BUG_ON(is_huge_zero_page(page));
BUG_ON(!PageAnon(page));
/*
--
Kirill A. Shutemov
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists