Skip to contentSkip to conversation

x86,mm: fix pte_free()

PATCH In review Peter Zijlstra 5 participants 8 messages Jan 23, 2009 – Jan 26, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86,mm: fix pte_free()

On -rt we were seeing spurious bad page states like:

Bad page state in process 'firefox'
page:c1bc2380 flags:0x40000000 mapping:c1bc2390 mapcount:0 count:0 Trying to fix it up, but a reboot is needed

Backtrace:
Pid: 503, comm: firefox Not tainted 2.6.26.8-rt13 #3
[<c043d0f3>] ? printk+0x14/0x19
[<c0272d4e>] bad_page+0x4e/0x79
[<c0273831>] free_hot_cold_page+0x5b/0x1d3
[<c02739f6>] free_hot_page+0xf/0x11
[<c0273a18>] __free_pages+0x20/0x2b
[<c027d170>] __pte_alloc+0x87/0x91
[<c027d25e>] handle_mm_fault+0xe4/0x733
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c0218875>] do_page_fault+0x36f/0x88a

This is the case where a concurrent fault already installed the PTE and we get to free the newly allocated one.

This is due to pgtable_page_ctor() doing the spin_lock_init(&page->ptl) which is overlaid with the {private, mapping} struct.

union {
    struct {
        unsigned long private;
        struct address_space *mapping;
    };
#if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS
    spinlock_t ptl;
#endif
    struct kmem_cache *slab;
    struct page *first_page;
};

Normally the spinlock is small enough to not stomp on page->mapping, but PREEMPT_RT=y has huge 'spin'locks.

But lockdep kernels should also be able to trigger this splat, as the lock tracking code grows the spinlock to cover page->mapping.

The obvious fix is calling pgtable_page_dtor() like the regular pte free path __pte_free_tlb() does.

It seems all architectures except x86 and nm10300 already do this, and nm10300 doesn't seem to use pgtable_page_ctor(), which suggests it doesn't do SMP or simply doesnt do MMU at all or something.

---
 arch/x86/include/asm/pgalloc.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/x86/include/asm/pgalloc.h +1 -0
index cb7c151..b99023c 100644
@@ -42,6 +42,7 @@ static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 
 static inline void pte_free(struct mm_struct *mm, struct page *pte)
 {
+	pgtable_page_dtor();
 	__free_page(pte);
 }
 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#1 lore raw
Quoted 49 lines · Peter Zijlstra

On -rt we were seeing spurious bad page states like:

Bad page state in process 'firefox'
page:c1bc2380 flags:0x40000000 mapping:c1bc2390 mapcount:0 count:0 Trying to fix it up, but a reboot is needed

Backtrace:
Pid: 503, comm: firefox Not tainted 2.6.26.8-rt13 #3
[<c043d0f3>] ? printk+0x14/0x19
[<c0272d4e>] bad_page+0x4e/0x79
[<c0273831>] free_hot_cold_page+0x5b/0x1d3
[<c02739f6>] free_hot_page+0xf/0x11
[<c0273a18>] __free_pages+0x20/0x2b
[<c027d170>] __pte_alloc+0x87/0x91
[<c027d25e>] handle_mm_fault+0xe4/0x733
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c0218875>] do_page_fault+0x36f/0x88a

This is the case where a concurrent fault already installed the PTE and
we get to free the newly allocated one.

This is due to pgtable_page_ctor() doing the spin_lock_init(&page->ptl)
which is overlaid with the {private, mapping} struct.

union {
    struct {
        unsigned long private;
        struct address_space *mapping;
    };
#if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS
    spinlock_t ptl;
#endif
    struct kmem_cache *slab;
    struct page *first_page;
};

Normally the spinlock is small enough to not stomp on page->mapping, but
PREEMPT_RT=y has huge 'spin'locks.

But lockdep kernels should also be able to trigger this splat, as the lock tracking code grows the spinlock to cover page->mapping.

The obvious fix is calling pgtable_page_dtor() like the regular pte free
path __pte_free_tlb() does.

Quoted 6 lines

It seems all architectures except x86 and nm10300 already do this, and nm10300 doesn't seem to use pgtable_page_ctor(), which suggests it doesn't do SMP or simply doesnt do MMU at all or something.

Now one that's not obviously borken,..

---
 arch/x86/include/asm/pgalloc.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/x86/include/asm/pgalloc.h +1 -0
index cb7c151..dd14c54 100644
@@ -42,6 +42,7 @@ static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 
 static inline void pte_free(struct mm_struct *mm, struct page *pte)
 {
+	pgtable_page_dtor(pte);
 	__free_page(pte);
 }
 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Acked-by
#2 lore raw
Quoted 50 lines · Peter Zijlstra
Quoted 49 lines · Peter Zijlstra

On -rt we were seeing spurious bad page states like:

Bad page state in process 'firefox'
page:c1bc2380 flags:0x40000000 mapping:c1bc2390 mapcount:0 count:0 Trying to fix it up, but a reboot is needed

Backtrace:
Pid: 503, comm: firefox Not tainted 2.6.26.8-rt13 #3
[<c043d0f3>] ? printk+0x14/0x19
[<c0272d4e>] bad_page+0x4e/0x79
[<c0273831>] free_hot_cold_page+0x5b/0x1d3
[<c02739f6>] free_hot_page+0xf/0x11
[<c0273a18>] __free_pages+0x20/0x2b
[<c027d170>] __pte_alloc+0x87/0x91
[<c027d25e>] handle_mm_fault+0xe4/0x733
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c0218875>] do_page_fault+0x36f/0x88a

This is the case where a concurrent fault already installed the PTE and
we get to free the newly allocated one.

This is due to pgtable_page_ctor() doing the spin_lock_init(&page->ptl)
which is overlaid with the {private, mapping} struct.

union {
    struct {
        unsigned long private;
        struct address_space *mapping;
    };
#if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS
    spinlock_t ptl;
#endif
    struct kmem_cache *slab;
    struct page *first_page;
};

Normally the spinlock is small enough to not stomp on page->mapping, but
PREEMPT_RT=y has huge 'spin'locks.

But lockdep kernels should also be able to trigger this splat, as the lock tracking code grows the spinlock to cover page->mapping.

The obvious fix is calling pgtable_page_dtor() like the regular pte free
path __pte_free_tlb() does.

Quoted 6 lines
Quoted 6 lines

It seems all architectures except x86 and nm10300 already do this, and nm10300 doesn't seem to use pgtable_page_ctor(), which suggests it doesn't do SMP or simply doesnt do MMU at all or something.

Thanks, Peter: good catch. That pgtable_page_dtor() had long been there in pte_free(), then somehow got lost in one of 2.6.26's rearrangements.

Quoted 2 lines

Now one that's not obviously borken,..

And I can quite see why you voided the first version:
your mind rightly stalled on that foul "struct page *pte". Oh well, clean that up some other time.

Hugh

Quoted 10 lines
---
 arch/x86/include/asm/pgalloc.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/x86/include/asm/pgalloc.h +0 -0
index cb7c151..dd14c54 100644

arch/x86/include/asm/pgalloc.h static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte) · line 47

Quoted 6 lines
arch/x86/include/asm/pgalloc.h +1 -0
@@ -42,6 +42,7 @@ static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 
 static inline void pte_free(struct mm_struct *mm, struct page *pte)
 {
+	pgtable_page_dtor(pte);
 	__free_page(pte);
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw
In reply to Peter Zijlstra
Quoted 60 lines · Peter Zijlstra

On -rt we were seeing spurious bad page states like:

Bad page state in process 'firefox'
page:c1bc2380 flags:0x40000000 mapping:c1bc2390 mapcount:0 count:0 Trying to fix it up, but a reboot is needed

Backtrace:
Pid: 503, comm: firefox Not tainted 2.6.26.8-rt13 #3
[<c043d0f3>] ? printk+0x14/0x19
[<c0272d4e>] bad_page+0x4e/0x79
[<c0273831>] free_hot_cold_page+0x5b/0x1d3
[<c02739f6>] free_hot_page+0xf/0x11
[<c0273a18>] __free_pages+0x20/0x2b
[<c027d170>] __pte_alloc+0x87/0x91
[<c027d25e>] handle_mm_fault+0xe4/0x733
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c0218875>] do_page_fault+0x36f/0x88a

This is the case where a concurrent fault already installed the PTE and we get to free the newly allocated one.

This is due to pgtable_page_ctor() doing the spin_lock_init(&page->ptl) which is overlaid with the {private, mapping} struct.

union {
    struct {
        unsigned long private;
        struct address_space *mapping;
    };
#if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS
    spinlock_t ptl;
#endif
    struct kmem_cache *slab;
    struct page *first_page;
};

Normally the spinlock is small enough to not stomp on page->mapping, but PREEMPT_RT=y has huge 'spin'locks.

But lockdep kernels should also be able to trigger this splat, as the lock tracking code grows the spinlock to cover page->mapping.

The obvious fix is calling pgtable_page_dtor() like the regular pte free path __pte_free_tlb() does.

It seems all architectures except x86 and nm10300 already do this, and nm10300 doesn't seem to use pgtable_page_ctor(), which suggests it doesn't do SMP or simply doesnt do MMU at all or something.

---
 arch/x86/include/asm/pgalloc.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/x86/include/asm/pgalloc.h +0 -0
index cb7c151..b99023c 100644

arch/x86/include/asm/pgalloc.h static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte) · line 45

Quoted 4 lines
arch/x86/include/asm/pgalloc.h +1 -0
@@ -42,6 +42,7 @@ static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
 
 static inline void pte_free(struct mm_struct *mm, struct page *pte)
 {
+	pgtable_page_dtor();

i suspect on lockdep we dont see this in practice because it initializes things to NULL, which hides the issue. On -rt we initialize list heads there which brings up the wrong warning in the page free logic.

So i agree with the fix, but the patch does not look right: shouldnt that be pgtable_page_dtor(pte), so that we get ->mapping cleared via pte_lock_deinit()? (which i guess your intention was here - this probably wont even build)

Ingo

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#4 lore raw
Quoted 4 lines · Ingo Molnar

So i agree with the fix, but the patch does not look right: shouldnt that be pgtable_page_dtor(pte), so that we get ->mapping cleared via pte_lock_deinit()? (which i guess your intention was here - this probably wont even build)

Yeah, I somehow fudged it, already send out a better one. -- One of them days I guess :-(

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#5 lore raw
Quoted 9 lines · Peter Zijlstra
Quoted 4 lines · Ingo Molnar

So i agree with the fix, but the patch does not look right: shouldnt that be pgtable_page_dtor(pte), so that we get ->mapping cleared via pte_lock_deinit()? (which i guess your intention was here - this probably wont even build)

Yeah, I somehow fudged it, already send out a better one. -- One of them days I guess :-(

no problem - applied to tip/x86/urgent, thanks Peter!

Ingo

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#6 lore raw
Quoted 10 lines · Peter Zijlstra
Quoted 8 lines · Ingo Molnar
Quoted 5 lines

So i agree with the fix, but the patch does not look right: shouldnt that be pgtable_page_dtor(pte), so that we get ->mapping cleared via pte_lock_deinit()? (which i guess your intention was here - this probably wont even build)

Quoted 4 lines
Quoted 2 lines

Yeah, I somehow fudged it, already send out a better one. -- One of them days I guess :-(

no problem - applied to tip/x86/urgent, thanks Peter!

please fix typo. s/nm10300/MN10300/ :)
at first look, I don't understand his intention.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#7 lore raw
In reply to Peter Zijlstra
Quoted 3 lines · Peter Zijlstra

It seems all architectures except x86 and nm10300 already do this, and nm10300 doesn't seem to use pgtable_page_ctor(), which suggests it doesn't do SMP or simply doesnt do MMU at all or something.

MN10300 does not, as yet, do SMP.

David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>