HWPOISON: Intro

PATCH In review Andi Kleen 7 participants 53 messages Aug 5, 2009 – Aug 12, 2009

Subsystems (27)

Memory management Documentation x86 VFS +23 more subsystems

lore

Share & tools

lore

Andi Kleen

author

2009-08-05 09:36 UTC #0 lore raw

cover HWPOISON: Intro

[AK: This is a version with the correct file list
and some fixes. Please disregard the version I posted yesterday]

New version of the hwpoison patchkit. Various changes. Believed to address all earlier review comments.

Active error truncate is enabled per file system now, so it adds a new VFS operation "error_remove_page" for this.
This prevents any truncation on metadata pages, on those it just does invalidate.

Also various bug fixes, most of them from Fengguang.

Please see the individual patches for changelog.

Should be good to go now.

Passes the hwpoison specific parts of the mce-test test suite
(git://git.kernel.org/pub/scm/utils/cpu/mce/mce-test.git)

Also available as git tree from
git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-mce-2.6.git hwpoison

Andrew, Please consider for merging.

Thanks,
-Andi

Signed-off-by Andi Kleen <[email protected]>

---

Upcoming Intel CPUs have support for recovering from some memory errors (``MCA recovery''). This requires the OS to declare a page "poisoned", kill the processes associated with it and avoid using it in the future.

This patchkit implements the necessary infrastructure in the VM.

To quote the overview comment:

* High level machine check handler. Handles pages reported by the
* hardware as being corrupted usually due to a 2bit ECC memory or cache
* failure.
*
* This focusses on pages detected as corrupted in the background.
* When the current CPU tries to consume corruption the currently
* running process can just be killed directly instead. This implies
* that if the error cannot be handled for some reason it's safe to
* just ignore it because no corruption has been consumed yet. Instead
* when that happens another machine check will happen.
*
* Handles page cache pages in various states. The tricky part
* here is that we can access any page asynchronous to other VM
* users, because memory failures could happen anytime and anywhere,
* possibly violating some of their assumptions. This is why this code
* has to be extremely careful. Generally it tries to use normal locking
* rules, as in get the standard locks, even if that means the
* error handling takes potentially a long time.
*
* Some of the operations here are somewhat inefficient and have non
* linear algorithmic complexity, because the data structures have not
* been optimized for this case. This is in particular the case
* for the mapping from a vma to a process. Since this case is expected
* to be rare we hope we can get away with this.

The code consists of a the high level handler in mm/memory-failure.c, a new page poison bit and various checks in the VM to handle poisoned pages.

The main target right now is KVM guests, but it works for all kinds of applications.

For the KVM use there was need for a new signal type so that KVM can inject the machine check into the guest with the proper address. This in theory allows other applications to handle memory failures too. The expection is that near all applications won't do that, but some very specialized ones might.

This is not fully complete yet, in particular there are still ways to access poison through various ways (crash dump, /proc/kcore etc.) that need to be plugged too.

-Andi

Andi Kleen

author

2009-08-05 09:36 UTC #1 lore raw

PATCH 1/19 HWPOISON: Add page flag for poisoned pages

Hardware poisoned pages need special handling in the VM and shouldn't be touched again. This requires a new page flag. Define it here.

The page flags wars seem to be over, so it shouldn't be a problem to get a new one.

v2: Add TestSetHWPoison (suggested by Johannes Weiner)

Acked-by Christoph Lameter <[email protected]>

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/page-flags.h |   17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

include/linux/page-flags.h +16 -1

Index: linux/include/linux/page-flags.h
===================================================================
@@ -51,6 +51,9 @@
  * PG_buddy is set to indicate that the page is free and in the buddy system
  * (see mm/page_alloc.c).
  *
+ * PG_hwpoison indicates that a page got corrupted in hardware and contains
+ * data with incorrect ECC bits that triggered a machine check. Accessing is
+ * not safe since it may cause another machine check. Don't touch!
  */
 
 /*

@@ -102,6 +105,9 @@ enum pageflags {
 #ifdef CONFIG_IA64_UNCACHED_ALLOCATOR
 	PG_uncached,		/* Page has been mapped as uncached */
 #endif
+#ifdef CONFIG_MEMORY_FAILURE
+	PG_hwpoison,		/* hardware poisoned page. Don't touch */
+#endif
 	__NR_PAGEFLAGS,
 
 	/* Filesystems */

@@ -263,6 +269,15 @@ PAGEFLAG(Uncached, uncached)
 PAGEFLAG_FALSE(Uncached)
 #endif
 
+#ifdef CONFIG_MEMORY_FAILURE
+PAGEFLAG(HWPoison, hwpoison)
+TESTSETFLAG(HWPoison, hwpoison)
+#define __PG_HWPOISON (1UL << PG_hwpoison)
+#else
+PAGEFLAG_FALSE(HWPoison)
+#define __PG_HWPOISON 0
+#endif
+
 static inline int PageUptodate(struct page *page)
 {
 	int ret = test_bit(PG_uptodate, &(page)->flags);

@@ -387,7 +402,7 @@ static inline void __ClearPageTail(struc
 	 1 << PG_private | 1 << PG_private_2 | \
 	 1 << PG_buddy	 | 1 << PG_writeback | 1 << PG_reserved | \
 	 1 << PG_slab	 | 1 << PG_swapcache | 1 << PG_active | \
-	 1 << PG_unevictable | __PG_MLOCKED)
+	 1 << PG_unevictable | __PG_MLOCKED | __PG_HWPOISON)
 
 /*
  * Flags checked when a page is prepped for return by the page allocator.

--
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>

Andi Kleen

author

2009-08-05 09:36 UTC #2 lore raw

PATCH 2/19 HWPOISON: Export some rmap vma locking to outside world

In reply to Andi Kleen

Needed for later patch that walks rmap entries on its own.

This used to be very frowned upon, but memory-failure.c does some rather specialized rmap walking and rmap has been stable for quite some time, so I think it's ok now to export it.

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/rmap.h |    6 ++++++
 mm/rmap.c            |    4 ++--
 2 files changed, 8 insertions(+), 2 deletions(-)

include/linux/rmap.h +6 -0

Index: linux/include/linux/rmap.h
===================================================================
@@ -112,6 +112,12 @@ int page_mkclean(struct page *);
  */
 int try_to_munlock(struct page *);
 
+/*
+ * Called by memory-failure.c to kill processes.
+ */
+struct anon_vma *page_lock_anon_vma(struct page *page);
+void page_unlock_anon_vma(struct anon_vma *anon_vma);
+
 #else	/* !CONFIG_MMU */
 
 #define anon_vma_init()		do {} while (0)

mm/rmap.c +2 -2

Index: linux/mm/rmap.c
===================================================================
@@ -191,7 +191,7 @@ void __init anon_vma_init(void)
  * Getting a lock on a stable anon_vma from a page off the LRU is
  * tricky: page_lock_anon_vma rely on RCU to guard against the races.
  */
-static struct anon_vma *page_lock_anon_vma(struct page *page)
+struct anon_vma *page_lock_anon_vma(struct page *page)
 {
 	struct anon_vma *anon_vma;
 	unsigned long anon_mapping;

@@ -211,7 +211,7 @@ out:
 	return NULL;
 }
 
-static void page_unlock_anon_vma(struct anon_vma *anon_vma)
+void page_unlock_anon_vma(struct anon_vma *anon_vma)
 {
 	spin_unlock(&anon_vma->lock);
 	rcu_read_unlock();

Andi Kleen

author

2009-08-05 09:36 UTC #3 lore raw

PATCH 3/19 HWPOISON: Add support for poison swap entries v2

In reply to Andi Kleen

Memory migration uses special swap entry types to trigger special actions on page faults. Extend this mechanism to also support poisoned swap entries, to trigger poison handling on page faults. This allows follow-on patches to prevent processes from faulting in poisoned pages again.

v2: Fix overflow in MAX_SWAPFILES (Fengguang Wu)
v3: Better overflow fix (Hidehiro Kawai)

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/swap.h    |   34 ++++++++++++++++++++++++++++------
 include/linux/swapops.h |   38 ++++++++++++++++++++++++++++++++++++++
 mm/swapfile.c           |    4 ++--
 3 files changed, 68 insertions(+), 8 deletions(-)

include/linux/swap.h +28 -6

Index: linux/include/linux/swap.h
===================================================================
@@ -34,16 +34,38 @@ static inline int current_is_kswapd(void
  * the type/offset into the pte as 5/27 as well.
  */
 #define MAX_SWAPFILES_SHIFT	5
-#ifndef CONFIG_MIGRATION
-#define MAX_SWAPFILES		(1 << MAX_SWAPFILES_SHIFT)
+
+/*
+ * Use some of the swap files numbers for other purposes. This
+ * is a convenient way to hook into the VM to trigger special
+ * actions on faults.
+ */
+
+/*
+ * NUMA node memory migration support
+ */
+#ifdef CONFIG_MIGRATION
+#define SWP_MIGRATION_NUM 2
+#define SWP_MIGRATION_READ	(MAX_SWAPFILES + SWP_HWPOISON_NUM)
+#define SWP_MIGRATION_WRITE	(MAX_SWAPFILES + SWP_HWPOISON_NUM + 1)
 #else
-/* Use last two entries for page migration swap entries */
-#define MAX_SWAPFILES		((1 << MAX_SWAPFILES_SHIFT)-2)
-#define SWP_MIGRATION_READ	MAX_SWAPFILES
-#define SWP_MIGRATION_WRITE	(MAX_SWAPFILES + 1)
+#define SWP_MIGRATION_NUM 0
 #endif
 
 /*
+ * Handling of hardware poisoned pages with memory corruption.
+ */
+#ifdef CONFIG_MEMORY_FAILURE
+#define SWP_HWPOISON_NUM 1
+#define SWP_HWPOISON		MAX_SWAPFILES
+#else
+#define SWP_HWPOISON_NUM 0
+#endif
+
+#define MAX_SWAPFILES \
+	((1 << MAX_SWAPFILES_SHIFT) - SWP_MIGRATION_NUM - SWP_HWPOISON_NUM)
+
+/*
  * Magic header for a swap area. The first part of the union is
  * what the swap magic looks like for the old (limited to 128MB)
  * swap area format, the second part of the union adds - in the

include/linux/swapops.h +38 -0

Index: linux/include/linux/swapops.h
===================================================================
@@ -131,3 +131,41 @@ static inline int is_write_migration_ent
 
 #endif
 
+#ifdef CONFIG_MEMORY_FAILURE
+/*
+ * Support for hardware poisoned pages
+ */
+static inline swp_entry_t make_hwpoison_entry(struct page *page)
+{
+	BUG_ON(!PageLocked(page));
+	return swp_entry(SWP_HWPOISON, page_to_pfn(page));
+}
+
+static inline int is_hwpoison_entry(swp_entry_t entry)
+{
+	return swp_type(entry) == SWP_HWPOISON;
+}
+#else
+
+static inline swp_entry_t make_hwpoison_entry(struct page *page)
+{
+	return swp_entry(0, 0);
+}
+
+static inline int is_hwpoison_entry(swp_entry_t swp)
+{
+	return 0;
+}
+#endif
+
+#if defined(CONFIG_MEMORY_FAILURE) || defined(CONFIG_MIGRATION)
+static inline int non_swap_entry(swp_entry_t entry)
+{
+	return swp_type(entry) >= MAX_SWAPFILES;
+}
+#else
+static inline int non_swap_entry(swp_entry_t entry)
+{
+	return 0;
+}
+#endif

mm/swapfile.c +2 -2

Index: linux/mm/swapfile.c
===================================================================
@@ -697,7 +697,7 @@ int free_swap_and_cache(swp_entry_t entr
 	struct swap_info_struct *p;
 	struct page *page = NULL;
 
-	if (is_migration_entry(entry))
+	if (non_swap_entry(entry))
 		return 1;
 
 	p = swap_info_get(entry);

@@ -2083,7 +2083,7 @@ static int __swap_duplicate(swp_entry_t
 	int count;
 	bool has_cache;
 
-	if (is_migration_entry(entry))
+	if (non_swap_entry(entry))
 		return -EINVAL;
 
 	type = swp_type(entry);

Andi Kleen

author

2009-08-05 09:36 UTC #4 lore raw

PATCH 4/19 HWPOISON: Add new SIGBUS error codes for hardware poison signals

In reply to Andi Kleen

Add new SIGBUS codes for reporting machine checks as signals. When the hardware detects an uncorrected ECC error it can trigger these signals.

This is needed for telling KVM's qemu about machine checks that happen to guests, so that it can inject them, but might be also useful for other programs. I find it useful in my test programs.

This patch merely defines the new types.

- Define two new si_codes for SIGBUS. BUS_MCEERR_AO and BUS_MCEERR_AR
* BUS_MCEERR_AO is for "Action Optional" machine checks, which means that some corruption has been detected in the background, but nothing has been consumed so far. The program can ignore those if it wants (but most programs would already get killed)
* BUS_MCEERR_AR is for "Action Required" machine checks. This happens when corrupted data is consumed or the application ran into an area which has been known to be corrupted earlier. These require immediate action and cannot just returned to. Most programs would kill themselves.
- They report the address of the corruption in the user address space in si_addr.
- Define a new si_addr_lsb field that reports the extent of the corruption to user space. That's currently always a (small) page. The user application cannot tell where in this page the corruption happened.

AK: I plan to write a man page update before anyone asks.

Signed-off-by Andi Kleen <[email protected]>

---
 include/asm-generic/siginfo.h |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

include/asm-generic/siginfo.h +7 -1

Index: linux/include/asm-generic/siginfo.h
===================================================================
@@ -82,6 +82,7 @@ typedef struct siginfo {
 #ifdef __ARCH_SI_TRAPNO
 			int _trapno;	/* TRAP # which caused the signal */
 #endif
+			short _addr_lsb; /* LSB of the reported address */
 		} _sigfault;
 
 		/* SIGPOLL */

@@ -112,6 +113,7 @@ typedef struct siginfo {
 #ifdef __ARCH_SI_TRAPNO
 #define si_trapno	_sifields._sigfault._trapno
 #endif
+#define si_addr_lsb	_sifields._sigfault._addr_lsb
 #define si_band		_sifields._sigpoll._band
 #define si_fd		_sifields._sigpoll._fd

@@ -192,7 +194,11 @@ typedef struct siginfo {
 #define BUS_ADRALN	(__SI_FAULT|1)	/* invalid address alignment */
 #define BUS_ADRERR	(__SI_FAULT|2)	/* non-existant physical address */
 #define BUS_OBJERR	(__SI_FAULT|3)	/* object specific hardware error */
-#define NSIGBUS		3
+/* hardware memory error consumed on a machine check: action required */
+#define BUS_MCEERR_AR	(__SI_FAULT|4)
+/* hardware memory error detected in process but not consumed: action optional*/
+#define BUS_MCEERR_AO	(__SI_FAULT|5)
+#define NSIGBUS		5
 
 /*
  * SIGTRAP si_codes

Andi Kleen

author

2009-08-05 09:36 UTC #5 lore raw

PATCH 5/19 HWPOISON: Add basic support for poisoned pages in fault handler v3

In reply to Andi Kleen

- Add a new VM_FAULT_HWPOISON error code to handle_mm_fault. Right now architectures have to explicitely enable poison page support, so this is forward compatible to all architectures. They only need to add it when they enable poison page support.
- Add poison page handling in swap in fault code

v2: Add missing delayacct_clear_flag (Hidehiro Kawai)
v3: Really use delayacct_clear_flag (Hidehiro Kawai)

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/mm.h |    3 ++-
 mm/memory.c        |   18 +++++++++++++++---
 2 files changed, 17 insertions(+), 4 deletions(-)

mm/memory.c +15 -3

Index: linux/mm/memory.c
===================================================================
@@ -1319,7 +1319,8 @@ int __get_user_pages(struct task_struct
 				if (ret & VM_FAULT_ERROR) {
 					if (ret & VM_FAULT_OOM)
 						return i ? i : -ENOMEM;
-					else if (ret & VM_FAULT_SIGBUS)
+					if (ret &
+					    (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
 						return i ? i : -EFAULT;
 					BUG();
 				}

@@ -2511,8 +2512,15 @@ static int do_swap_page(struct mm_struct
 		goto out;
 
 	entry = pte_to_swp_entry(orig_pte);
-	if (is_migration_entry(entry)) {
-		migration_entry_wait(mm, pmd, address);
+	if (unlikely(non_swap_entry(entry))) {
+		if (is_migration_entry(entry)) {
+			migration_entry_wait(mm, pmd, address);
+		} else if (is_hwpoison_entry(entry)) {
+			ret = VM_FAULT_HWPOISON;
+		} else {
+			print_bad_pte(vma, address, pte, NULL);
+			ret = VM_FAULT_OOM;
+		}
 		goto out;
 	}
 	delayacct_set_flag(DELAYACCT_PF_SWAPIN);

@@ -2536,6 +2544,10 @@ static int do_swap_page(struct mm_struct
 		/* Had to read the page from swap area: Major fault */
 		ret = VM_FAULT_MAJOR;
 		count_vm_event(PGMAJFAULT);
+	} else if (PageHWPoison(page)) {
+		ret = VM_FAULT_HWPOISON;
+		delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+		goto out;
 	}
 
 	lock_page(page);

include/linux/mm.h +2 -1

Index: linux/include/linux/mm.h
===================================================================
@@ -700,11 +700,12 @@ static inline int page_mapped(struct pag
 #define VM_FAULT_SIGBUS	0x0002
 #define VM_FAULT_MAJOR	0x0004
 #define VM_FAULT_WRITE	0x0008	/* Special case for get_user_pages */
+#define VM_FAULT_HWPOISON 0x0010	/* Hit poisoned page */
 
 #define VM_FAULT_NOPAGE	0x0100	/* ->fault installed the pte, not return page */
 #define VM_FAULT_LOCKED	0x0200	/* ->fault locked the returned page */
 
-#define VM_FAULT_ERROR	(VM_FAULT_OOM | VM_FAULT_SIGBUS)
+#define VM_FAULT_ERROR	(VM_FAULT_OOM | VM_FAULT_SIGBUS | VM_FAULT_HWPOISON)
 
 /*
  * Can be called by the pagefault handler when it gets a VM_FAULT_OOM.

Andi Kleen

author

2009-08-05 09:36 UTC #6 lore raw

PATCH 6/19 HWPOISON: Add various poison checks in mm/memory.c v2

In reply to Andi Kleen

Bail out early when hardware poisoned pages are found in page fault handling. Since they are poisoned they should not be mapped freshly into processes, because that would cause another (potentially deadly) machine check

This is generally handled in the same way as OOM, just a different error code is returned to the architecture code.

v2: Do a page unlock if needed (Fengguang Wu)

Signed-off-by Andi Kleen <[email protected]>

---
 mm/memory.c |    6 ++++++
 1 file changed, 6 insertions(+)

mm/memory.c +6 -0

Index: linux/mm/memory.c
===================================================================
@@ -2711,6 +2711,12 @@ static int __do_fault(struct mm_struct *
 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
 		return ret;
 
+	if (unlikely(PageHWPoison(vmf.page))) {
+		if (ret & VM_FAULT_LOCKED)
+			unlock_page(vmf.page);
+		return VM_FAULT_HWPOISON;
+	}
+
 	/*
 	 * For consistency in subsequent calls, make the faulted page always
 	 * locked.

Andi Kleen

author

2009-08-05 09:36 UTC #7 lore raw

PATCH 7/19 HWPOISON: x86: Add VM_FAULT_HWPOISON handling to x86 page fault handler v2

In reply to Andi Kleen

Add VM_FAULT_HWPOISON handling to the x86 page fault handler. This is very similar to VM_FAULT_OOM, the only difference is that a different si_code is passed to user space and the new addr_lsb field is initialized.

v2: Make the printk more verbose/unique

Signed-off-by Andi Kleen <[email protected]>

---
 arch/x86/mm/fault.c |   19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)

arch/x86/mm/fault.c +15 -4

Index: linux/arch/x86/mm/fault.c
===================================================================
@@ -167,6 +167,7 @@ force_sig_info_fault(int si_signo, int s
 	info.si_errno	= 0;
 	info.si_code	= si_code;
 	info.si_addr	= (void __user *)address;
+	info.si_addr_lsb = si_code == BUS_MCEERR_AR ? PAGE_SHIFT : 0;
 
 	force_sig_info(si_signo, &info, tsk);
 }

@@ -799,10 +800,12 @@ out_of_memory(struct pt_regs *regs, unsi
 }
 
 static void
-do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
+do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
+	  unsigned int fault)
 {
 	struct task_struct *tsk = current;
 	struct mm_struct *mm = tsk->mm;
+	int code = BUS_ADRERR;
 
 	up_read(&mm->mmap_sem);

@@ -818,7 +821,15 @@ do_sigbus(struct pt_regs *regs, unsigned
 	tsk->thread.error_code	= error_code;
 	tsk->thread.trap_no	= 14;
 
-	force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
+#ifdef CONFIG_MEMORY_FAILURE
+	if (fault & VM_FAULT_HWPOISON) {
+		printk(KERN_ERR
+	"MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
+			tsk->comm, tsk->pid, address);
+		code = BUS_MCEERR_AR;
+	}
+#endif
+	force_sig_info_fault(SIGBUS, code, address, tsk);
 }
 
 static noinline void

@@ -828,8 +839,8 @@ mm_fault_error(struct pt_regs *regs, uns
 	if (fault & VM_FAULT_OOM) {
 		out_of_memory(regs, error_code, address);
 	} else {
-		if (fault & VM_FAULT_SIGBUS)
-			do_sigbus(regs, error_code, address);
+		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON))
+			do_sigbus(regs, error_code, address, fault);
 		else
 			BUG();
 	}

Andi Kleen

author

2009-08-05 09:36 UTC #9 lore raw

PATCH 9/19 HWPOISON: Handle hardware poisoned pages in try_to_unmap

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

When a page has the poison bit set replace the PTE with a poison entry. This causes the right error handling to be done later when a process runs into it.

v2: add a new flag to not do that (needed for the memory-failure handler later) (Fengguang)
v3: remove unnecessary is_migration_entry() test (Fengguang, Minchan)

Reviewed-by Minchan Kim <[email protected]>

Reviewed-by Wu Fengguang <[email protected]>

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/rmap.h |    1 +
 mm/rmap.c            |    9 ++++++++-
 2 files changed, 9 insertions(+), 1 deletion(-)

mm/rmap.c +8 -1

Index: linux/mm/rmap.c
===================================================================
@@ -819,7 +819,14 @@ static int try_to_unmap_one(struct page
 	/* Update high watermark before we lower rss */
 	update_hiwater_rss(mm);
 
-	if (PageAnon(page)) {
+	if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
+		if (PageAnon(page))
+			dec_mm_counter(mm, anon_rss);
+		else
+			dec_mm_counter(mm, file_rss);
+		set_pte_at(mm, address, pte,
+				swp_entry_to_pte(make_hwpoison_entry(page)));
+	} else if (PageAnon(page)) {
 		swp_entry_t entry = { .val = page_private(page) };
 
 		if (PageSwapCache(page)) {

include/linux/rmap.h +1 -0

Index: linux/include/linux/rmap.h
===================================================================
@@ -93,6 +93,7 @@ enum ttu_flags {
 
 	TTU_IGNORE_MLOCK = (1 << 8),	/* ignore mlock */
 	TTU_IGNORE_ACCESS = (1 << 9),	/* don't age */
+	TTU_IGNORE_HWPOISON = (1 << 10),/* corrupted page is recoverable */
 };
 #define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)

Andi Kleen

author

2009-08-05 09:36 UTC #8 lore raw

PATCH 8/19 HWPOISON: Use bitmask/action code for try_to_unmap behaviour

In reply to Andi Kleen

try_to_unmap currently has multiple modi (migration, munlock, normal unmap) which are selected by magic flag variables. The logic is not very straight forward, because each of these flag change multiple behaviours (e.g. migration turns off aging, not only sets up migration ptes etc.) Also the different flags interact in magic ways.

A later patch in this series adds another mode to try_to_unmap, so this becomes quickly unmanageable.

Replace the different flags with a action code (migration, munlock, munmap) and some additional flags as modifiers (ignore mlock, ignore aging). This makes the logic more straight forward and allows easier extension to new behaviours. Change all the caller to declare what they want to do.

This patch is supposed to be a nop in behaviour. If anyone can prove it is not that would be a bug.

Cc [email protected]

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/rmap.h |   13 ++++++++++++-
 mm/migrate.c         |    2 +-
 mm/rmap.c            |   40 ++++++++++++++++++++++------------------
 mm/vmscan.c          |    2 +-
 4 files changed, 36 insertions(+), 21 deletions(-)

include/linux/rmap.h +12 -1

Index: linux/include/linux/rmap.h
===================================================================
@@ -85,7 +85,18 @@ static inline void page_dup_rmap(struct
  */
 int page_referenced(struct page *, int is_locked,
 			struct mem_cgroup *cnt, unsigned long *vm_flags);
-int try_to_unmap(struct page *, int ignore_refs);
+enum ttu_flags {
+	TTU_UNMAP = 0,			/* unmap mode */
+	TTU_MIGRATION = 1,		/* migration mode */
+	TTU_MUNLOCK = 2,		/* munlock mode */
+	TTU_ACTION_MASK = 0xff,
+
+	TTU_IGNORE_MLOCK = (1 << 8),	/* ignore mlock */
+	TTU_IGNORE_ACCESS = (1 << 9),	/* don't age */
+};
+#define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
+
+int try_to_unmap(struct page *, enum ttu_flags flags);
 
 /*
  * Called from mm/filemap_xip.c to unmap empty zero page

mm/rmap.c +22 ~~-18~~

Index: linux/mm/rmap.c
===================================================================
@@ -773,7 +773,7 @@ void page_remove_rmap(struct page *page)
  * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
  */
 static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
-				int migration)
+				enum ttu_flags flags)
 {
 	struct mm_struct *mm = vma->vm_mm;
 	unsigned long address;

@@ -795,11 +795,13 @@ static int try_to_unmap_one(struct page
 	 * If it's recently referenced (perhaps page_referenced
 	 * skipped over this mm) then we should reactivate it.
 	 */
-	if (!migration) {
+	if (!(flags & TTU_IGNORE_MLOCK)) {
 		if (vma->vm_flags & VM_LOCKED) {
 			ret = SWAP_MLOCK;
 			goto out_unmap;
 		}
+	}
+	if (!(flags & TTU_IGNORE_ACCESS)) {
 		if (ptep_clear_flush_young_notify(vma, address, pte)) {
 			ret = SWAP_FAIL;
 			goto out_unmap;

@@ -839,12 +841,12 @@ static int try_to_unmap_one(struct page
 			 * pte. do_swap_page() will wait until the migration
 			 * pte is removed and then restart fault handling.
 			 */
-			BUG_ON(!migration);
+			BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
 			entry = make_migration_entry(page, pte_write(pteval));
 		}
 		set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
 		BUG_ON(pte_file(*pte));
-	} else if (PAGE_MIGRATION && migration) {
+	} else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
 		/* Establish migration entry for a file page */
 		swp_entry_t entry;
 		entry = make_migration_entry(page, pte_write(pteval));

@@ -1013,12 +1015,13 @@ static int try_to_mlock_page(struct page
  * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
  * 'LOCKED.
  */
-static int try_to_unmap_anon(struct page *page, int unlock, int migration)
+static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
 {
 	struct anon_vma *anon_vma;
 	struct vm_area_struct *vma;
 	unsigned int mlocked = 0;
 	int ret = SWAP_AGAIN;
+	int unlock = TTU_ACTION(flags) == TTU_MUNLOCK;
 
 	if (MLOCK_PAGES && unlikely(unlock))
 		ret = SWAP_SUCCESS;	/* default for try_to_munlock() */

@@ -1034,7 +1037,7 @@ static int try_to_unmap_anon(struct page
 				continue;  /* must visit all unlocked vmas */
 			ret = SWAP_MLOCK;  /* saw at least one mlocked vma */
 		} else {
-			ret = try_to_unmap_one(page, vma, migration);
+			ret = try_to_unmap_one(page, vma, flags);
 			if (ret == SWAP_FAIL || !page_mapped(page))
 				break;
 		}

@@ -1058,8 +1061,7 @@ static int try_to_unmap_anon(struct page
 /**
  * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
  * @page: the page to unmap/unlock
- * @unlock:  request for unlock rather than unmap [unlikely]
- * @migration:  unmapping for migration - ignored if @unlock
+ * @flags: action and flags
  *
  * Find all the mappings of a page using the mapping pointer and the vma chains
  * contained in the address_space struct it points to.

@@ -1071,7 +1073,7 @@ static int try_to_unmap_anon(struct page
  * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
  * 'LOCKED.
  */
-static int try_to_unmap_file(struct page *page, int unlock, int migration)
+static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
 {
 	struct address_space *mapping = page->mapping;
 	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);

@@ -1083,6 +1085,7 @@ static int try_to_unmap_file(struct page
 	unsigned long max_nl_size = 0;
 	unsigned int mapcount;
 	unsigned int mlocked = 0;
+	int unlock = TTU_ACTION(flags) == TTU_MUNLOCK;
 
 	if (MLOCK_PAGES && unlikely(unlock))
 		ret = SWAP_SUCCESS;	/* default for try_to_munlock() */

@@ -1095,7 +1098,7 @@ static int try_to_unmap_file(struct page
 				continue;	/* must visit all vmas */
 			ret = SWAP_MLOCK;
 		} else {
-			ret = try_to_unmap_one(page, vma, migration);
+			ret = try_to_unmap_one(page, vma, flags);
 			if (ret == SWAP_FAIL || !page_mapped(page))
 				goto out;
 		}

@@ -1120,7 +1123,8 @@ static int try_to_unmap_file(struct page
 			ret = SWAP_MLOCK;	/* leave mlocked == 0 */
 			goto out;		/* no need to look further */
 		}
-		if (!MLOCK_PAGES && !migration && (vma->vm_flags & VM_LOCKED))
+		if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) &&
+			(vma->vm_flags & VM_LOCKED))
 			continue;
 		cursor = (unsigned long) vma->vm_private_data;
 		if (cursor > max_nl_cursor)

@@ -1154,7 +1158,7 @@ static int try_to_unmap_file(struct page
 	do {
 		list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
 						shared.vm_set.list) {
-			if (!MLOCK_PAGES && !migration &&
+			if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) &&
 			    (vma->vm_flags & VM_LOCKED))
 				continue;
 			cursor = (unsigned long) vma->vm_private_data;

@@ -1194,7 +1198,7 @@ out:
 /**
  * try_to_unmap - try to remove all page table mappings to a page
  * @page: the page to get unmapped
- * @migration: migration flag
+ * @flags: action and flags
  *
  * Tries to remove all the page table entries which are mapping this
  * page, used in the pageout path.  Caller must hold the page lock.

@@ -1205,16 +1209,16 @@ out:
  * SWAP_FAIL	- the page is unswappable
  * SWAP_MLOCK	- page is mlocked.
  */
-int try_to_unmap(struct page *page, int migration)
+int try_to_unmap(struct page *page, enum ttu_flags flags)
 {
 	int ret;
 
 	BUG_ON(!PageLocked(page));
 
 	if (PageAnon(page))
-		ret = try_to_unmap_anon(page, 0, migration);
+		ret = try_to_unmap_anon(page, flags);
 	else
-		ret = try_to_unmap_file(page, 0, migration);
+		ret = try_to_unmap_file(page, flags);
 	if (ret != SWAP_MLOCK && !page_mapped(page))
 		ret = SWAP_SUCCESS;
 	return ret;

@@ -1239,8 +1243,8 @@ int try_to_munlock(struct page *page)
 	VM_BUG_ON(!PageLocked(page) || PageLRU(page));
 
 	if (PageAnon(page))
-		return try_to_unmap_anon(page, 1, 0);
+		return try_to_unmap_anon(page, TTU_MUNLOCK);
 	else
-		return try_to_unmap_file(page, 1, 0);
+		return try_to_unmap_file(page, TTU_MUNLOCK);
 }

mm/vmscan.c +1 -1

Index: linux/mm/vmscan.c
===================================================================
@@ -654,7 +654,7 @@ static unsigned long shrink_page_list(st
 		 * processes. Try to unmap it here.
 		 */
 		if (page_mapped(page) && mapping) {
-			switch (try_to_unmap(page, 0)) {
+			switch (try_to_unmap(page, TTU_UNMAP)) {
 			case SWAP_FAIL:
 				goto activate_locked;
 			case SWAP_AGAIN:

mm/migrate.c +1 -1

Index: linux/mm/migrate.c
===================================================================
@@ -669,7 +669,7 @@ static int unmap_and_move(new_page_t get
 	}
 
 	/* Establish migration ptes or remove ptes */
-	try_to_unmap(page, 1);
+	try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
 
 	if (!page_mapped(page))
 		rc = move_to_new_page(newpage, page);

Andi Kleen

author

2009-08-05 09:36 UTC #10 lore raw

PATCH 10/19 HWPOISON: check and isolate corrupted free pages v2

In reply to Andi Kleen

From: Wu Fengguang <[email protected]>

If memory corruption hits the free buddy pages, we can safely ignore them. No one will access them until page allocation time, then prep_new_page() will automatically check and isolate PG_hwpoison page for us (for 0-order allocation).

This patch expands prep_new_page() to check every component page in a high order page allocation, in order to completely stop PG_hwpoison pages from being recirculated.

Note that the common case -- only allocating a single page, doesn't do any more work than before. Allocating > order 0 does a bit more work, but that's relatively uncommon.

This simple implementation may drop some innocent neighbor pages, hopefully it is not a big problem because the event should be rare enough.

This patch adds some runtime costs to high order page users.

[AK: Improved description]

v2: Andi Kleen:
Port to -mm code
Move check into separate function.
Don't dump stack in bad_pages for hwpoisoned pages.

Signed-off-by Wu Fengguang <[email protected]>

Signed-off-by Andi Kleen <[email protected]>

---
 mm/page_alloc.c |   20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

mm/page_alloc.c +19 -1

Index: linux/mm/page_alloc.c
===================================================================
@@ -234,6 +234,12 @@ static void bad_page(struct page *page)
 	static unsigned long nr_shown;
 	static unsigned long nr_unshown;
 
+	/* Don't complain about poisoned pages */
+	if (PageHWPoison(page)) {
+		__ClearPageBuddy(page);
+		return;
+	}
+
 	/*
 	 * Allow a burst of 60 reports, then keep quiet for that minute;
 	 * or allow a steady drip of one report per second.

@@ -646,7 +652,7 @@ static inline void expand(struct zone *z
 /*
  * This page is about to be returned from the page allocator
  */
-static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
+static inline int check_new_page(struct page *page)
 {
 	if (unlikely(page_mapcount(page) |
 		(page->mapping != NULL)  |

@@ -655,6 +661,18 @@ static int prep_new_page(struct page *pa
 		bad_page(page);
 		return 1;
 	}
+	return 0;
+}
+
+static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
+{
+	int i;
+
+	for (i = 0; i < (1 << order); i++) {
+		struct page *p = page + i;
+		if (unlikely(check_new_page(p)))
+			return 1;
+	}
 
 	set_page_private(page, 0);
 	set_page_refcounted(page);

Andi Kleen

author

2009-08-05 09:36 UTC #11 lore raw

PATCH 11/19 HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Andi Kleen

From: Nick Piggin <[email protected]>

Extract out truncate_inode_page() out of the truncate path so that it can be used by memory-failure.c

[AK: description, headers, fix typos]
v2: Some white space changes from Fengguang Wu

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/mm.h |    2 ++
 mm/truncate.c      |   29 +++++++++++++++--------------
 2 files changed, 17 insertions(+), 14 deletions(-)

mm/truncate.c +15 ~~-14~~

Index: linux/mm/truncate.c
===================================================================
@@ -93,11 +93,11 @@ EXPORT_SYMBOL(cancel_dirty_page);
  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
  */
-static void
+static int
 truncate_complete_page(struct address_space *mapping, struct page *page)
 {
 	if (page->mapping != mapping)
-		return;
+		return -EIO;
 
 	if (page_has_private(page))
 		do_invalidatepage(page, 0);

@@ -108,6 +108,7 @@ truncate_complete_page(struct address_sp
 	remove_from_page_cache(page);
 	ClearPageMappedToDisk(page);
 	page_cache_release(page);	/* pagecache ref */
+	return 0;
 }
 
 /*

@@ -135,6 +136,16 @@ invalidate_complete_page(struct address_
 	return ret;
 }
 
+int truncate_inode_page(struct address_space *mapping, struct page *page)
+{
+	if (page_mapped(page)) {
+		unmap_mapping_range(mapping,
+				   (loff_t)page->index << PAGE_CACHE_SHIFT,
+				   PAGE_CACHE_SIZE, 0);
+	}
+	return truncate_complete_page(mapping, page);
+}
+
 /**
  * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
  * @mapping: mapping to truncate

@@ -196,12 +207,7 @@ void truncate_inode_pages_range(struct a
 				unlock_page(page);
 				continue;
 			}
-			if (page_mapped(page)) {
-				unmap_mapping_range(mapping,
-				  (loff_t)page_index<<PAGE_CACHE_SHIFT,
-				  PAGE_CACHE_SIZE, 0);
-			}
-			truncate_complete_page(mapping, page);
+			truncate_inode_page(mapping, page);
 			unlock_page(page);
 		}
 		pagevec_release(&pvec);

@@ -238,15 +244,10 @@ void truncate_inode_pages_range(struct a
 				break;
 			lock_page(page);
 			wait_on_page_writeback(page);
-			if (page_mapped(page)) {
-				unmap_mapping_range(mapping,
-				  (loff_t)page->index<<PAGE_CACHE_SHIFT,
-				  PAGE_CACHE_SIZE, 0);
-			}
+			truncate_inode_page(mapping, page);
 			if (page->index > next)
 				next = page->index;
 			next++;
-			truncate_complete_page(mapping, page);
 			unlock_page(page);
 		}
 		pagevec_release(&pvec);

include/linux/mm.h +2 -0

Index: linux/include/linux/mm.h
===================================================================
@@ -809,6 +809,8 @@ static inline void unmap_shared_mapping_
 extern int vmtruncate(struct inode * inode, loff_t offset);
 extern int vmtruncate_range(struct inode * inode, loff_t offset, loff_t end);
 
+int truncate_inode_page(struct address_space *mapping, struct page *page);
+
 #ifdef CONFIG_MMU
 extern int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, unsigned int flags);

Andi Kleen

author

2009-08-05 09:36 UTC #24 lore raw

PATCH 12/19 HWPOISON: Add invalidate_inode_page

In reply to Andi Kleen

From: Wu Fengguang <[email protected]>

Add a simple way to invalidate a single page This is just a refactoring of the truncate.c code. Originally from Fengguang, modified by Andi Kleen.

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/mm.h |    2 ++
 mm/truncate.c      |   26 ++++++++++++++++++++------
 2 files changed, 22 insertions(+), 6 deletions(-)

include/linux/mm.h +2 -0

Index: linux/include/linux/mm.h
===================================================================
@@ -811,6 +811,8 @@ extern int vmtruncate_range(struct inode
 
 int truncate_inode_page(struct address_space *mapping, struct page *page);
 
+int invalidate_inode_page(struct page *page);
+
 #ifdef CONFIG_MMU
 extern int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, unsigned int flags);

mm/truncate.c +20 -6

Index: linux/mm/truncate.c
===================================================================
@@ -146,6 +146,24 @@ int truncate_inode_page(struct address_s
 	return truncate_complete_page(mapping, page);
 }
 
+/*
+ * Safely invalidate one page from its pagecache mapping.
+ * It only drops clean, unused pages. The page must be locked.
+ *
+ * Returns 1 if the page is successfully invalidated, otherwise 0.
+ */
+int invalidate_inode_page(struct page *page)
+{
+	struct address_space *mapping = page_mapping(page);
+	if (!mapping)
+		return 0;
+	if (PageDirty(page) || PageWriteback(page))
+		return 0;
+	if (page_mapped(page))
+		return 0;
+	return invalidate_complete_page(mapping, page);
+}
+
 /**
  * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
  * @mapping: mapping to truncate

@@ -312,12 +330,8 @@ unsigned long invalidate_mapping_pages(s
 			if (lock_failed)
 				continue;
 
-			if (PageDirty(page) || PageWriteback(page))
-				goto unlock;
-			if (page_mapped(page))
-				goto unlock;
-			ret += invalidate_complete_page(mapping, page);
-unlock:
+			ret += invalidate_inode_page(page);
+
 			unlock_page(page);
 			if (next > end)
 				break;

Andi Kleen

author

2009-08-05 09:36 UTC #25 lore raw

PATCH 13/19 HWPOISON: Define a new error_remove_page address space op for async truncation

In reply to Andi Kleen

Truncating metadata pages is not safe right now before we haven't audited all file systems.

To enable truncation only for data address space define a new address_space callback error_remove_page.

This is used for memory_failure.c memory error handling.

This can be then set to truncate_inode_page()

This patch just defines the new operation and adds documentation.

Callers and users come in followon patches.

Signed-off-by Andi Kleen <[email protected]>

---
 Documentation/filesystems/vfs.txt |    7 +++++++
 include/linux/fs.h                |    1 +
 include/linux/mm.h                |    1 +
 mm/truncate.c                     |   17 +++++++++++++++++
 4 files changed, 26 insertions(+)

include/linux/fs.h +1 -0

Index: linux/include/linux/fs.h
===================================================================
@@ -595,6 +595,7 @@ struct address_space_operations {
 	int (*launder_page) (struct page *);
 	int (*is_partially_uptodate) (struct page *, read_descriptor_t *,
 					unsigned long);
+	int (*error_remove_page)(struct address_space *, struct page *);
 };
 
 /*

Documentation/filesystems/vfs.txt +7 -0

Index: linux/Documentation/filesystems/vfs.txt
===================================================================
@@ -536,6 +536,7 @@ struct address_space_operations {
 	/* migrate the contents of a page to the specified target */
 	int (*migratepage) (struct page *, struct page *);
 	int (*launder_page) (struct page *);
+	int (*error_remove_page) (struct mapping *mapping, struct page *page);
 };
 
   writepage: called by the VM to write a dirty page to backing store.

@@ -694,6 +695,12 @@ struct address_space_operations {
   	prevent redirtying the page, it is kept locked during the whole
 	operation.
 
+  error_remove_page: normally set to generic_error_remove_page if truncation
+	is ok for this address space. Used for memory failure handling.
+	Setting this implies you deal with pages going away under you,
+	unless you have them locked or reference counts increased.
+
+
 The File Object
 ===============

mm/truncate.c +17 -0

Index: linux/mm/truncate.c
===================================================================
@@ -147,6 +147,23 @@ int truncate_inode_page(struct address_s
 }
 
 /*
+ * Used to get rid of pages on hardware memory corruption.
+ */
+int generic_error_remove_page(struct address_space *mapping, struct page *page)
+{
+	if (!mapping)
+		return -EINVAL;
+	/*
+	 * Only punch for normal data pages for now.
+	 * Handling other types like directories would need more auditing.
+	 */
+	if (!S_ISREG(mapping->host->i_mode))
+		return -EIO;
+	return truncate_inode_page(mapping, page);
+}
+EXPORT_SYMBOL(generic_error_remove_page);
+
+/*
  * Safely invalidate one page from its pagecache mapping.
  * It only drops clean, unused pages. The page must be locked.
  *

include/linux/mm.h +1 -0

Index: linux/include/linux/mm.h
===================================================================
@@ -810,6 +810,7 @@ extern int vmtruncate(struct inode * ino
 extern int vmtruncate_range(struct inode * inode, loff_t offset, loff_t end);
 
 int truncate_inode_page(struct address_space *mapping, struct page *page);
+int generic_error_remove_page(struct address_space *mapping, struct page *page);
 
 int invalidate_inode_page(struct page *page);

Andi Kleen

author

2009-08-05 09:36 UTC #26 lore raw

PATCH 14/19 HWPOISON: Add PR_MCE_KILL prctl to control early kill behaviour per process

In reply to Andi Kleen

This allows processes to override their early/late kill behaviour on hardware memory errors.

Typically applications which are memory error aware is better of with early kill (see the error as soon as possible), all others with late kill (only
see the error when the error is really impacting execution)

There's a global sysctl, but this way an application can set its specific policy.

We're using two bits, one to signify that the process stated its intention and that

I also made the prctl future proof by enforcing the unused arguments are 0.

The state is inherited to children for now. I've been considering to reset it on exec, but not done for now (TBD).

Note this makes us officially run out of process flags on 32bit, but the next patch can easily add another field.

Manpage patch will be supplied separately.

Signed-off-by Andi Kleen <[email protected]>

---
 include/linux/prctl.h |    2 ++
 include/linux/sched.h |    2 ++
 kernel/sys.c          |   22 ++++++++++++++++++++++
 3 files changed, 26 insertions(+)

include/linux/sched.h +2 -0

Index: linux/include/linux/sched.h
===================================================================
@@ -1674,6 +1674,7 @@ extern cputime_t task_gtime(struct task_
 #define PF_EXITPIDONE	0x00000008	/* pi exit done on shut down */
 #define PF_VCPU		0x00000010	/* I'm a virtual CPU */
 #define PF_FORKNOEXEC	0x00000040	/* forked but didn't exec */
+#define PF_MCE_PROCESS  0x00000080      /* process policy on mce errors */
 #define PF_SUPERPRIV	0x00000100	/* used super-user privileges */
 #define PF_DUMPCORE	0x00000200	/* dumped core */
 #define PF_SIGNALED	0x00000400	/* killed by a signal */

@@ -1693,6 +1694,7 @@ extern cputime_t task_gtime(struct task_
 #define PF_SPREAD_PAGE	0x01000000	/* Spread page cache over cpuset */
 #define PF_SPREAD_SLAB	0x02000000	/* Spread some slab caches over cpuset */
 #define PF_THREAD_BOUND	0x04000000	/* Thread bound to specific cpu */
+#define PF_MCE_EARLY    0x08000000      /* Early kill for mce process policy */
 #define PF_MEMPOLICY	0x10000000	/* Non-default NUMA mempolicy */
 #define PF_MUTEX_TESTER	0x20000000	/* Thread belongs to the rt mutex tester */
 #define PF_FREEZER_SKIP	0x40000000	/* Freezer should not count it as freezeable */

kernel/sys.c +22 -0

Index: linux/kernel/sys.c
===================================================================
@@ -1528,6 +1528,28 @@ SYSCALL_DEFINE5(prctl, int, option, unsi
 				current->timer_slack_ns = arg2;
 			error = 0;
 			break;
+		case PR_MCE_KILL:
+			if (arg4 | arg5)
+				return -EINVAL;
+			switch (arg2) {
+			case 0:
+				if (arg3 != 0)
+					return -EINVAL;
+				current->flags &= ~PF_MCE_PROCESS;
+				break;
+			case 1:
+				current->flags |= PF_MCE_PROCESS;
+				if (arg3 != 0)
+					current->flags |= PF_MCE_EARLY;
+				else
+					current->flags &= ~PF_MCE_EARLY;
+				break;
+			default:
+				return -EINVAL;
+			}
+			error = 0;
+			break;
+
 		default:
 			error = -EINVAL;
 			break;

include/linux/prctl.h +2 -0

Index: linux/include/linux/prctl.h
===================================================================
@@ -88,4 +88,6 @@
 #define PR_TASK_PERF_COUNTERS_DISABLE		31
 #define PR_TASK_PERF_COUNTERS_ENABLE		32
 
+#define PR_MCE_KILL	33
+
 #endif /* _LINUX_PRCTL_H */

Andi Kleen

author

2009-08-05 09:36 UTC #28 lore raw

PATCH 16/19 HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Enable removing of corrupted pages through truncation for a bunch of file systems: ext*, xfs, gfs2, ocfs2, ntfs These should cover most server needs.

I chose the set of migration aware file systems for this for now, assuming they have been especially audited. But in general it should be safe for all file systems on the data area that support read/write and truncate.

Caveat: the hardware error handler does not take i_mutex for now before calling the truncate function. Is that ok?

Cc [email protected]

Signed-off-by Andi Kleen <[email protected]>

---
 fs/ext2/inode.c             |    2 ++
 fs/ext3/inode.c             |    3 +++
 fs/ext4/inode.c             |    4 ++++
 fs/gfs2/aops.c              |    3 +++
 fs/ntfs/aops.c              |    2 ++
 fs/ocfs2/aops.c             |    1 +
 fs/xfs/linux-2.6/xfs_aops.c |    1 +
 mm/shmem.c                  |    1 +
 8 files changed, 17 insertions(+)

fs/gfs2/aops.c +3 -0

Index: linux/fs/gfs2/aops.c
===================================================================
@@ -1135,6 +1135,7 @@ static const struct address_space_operat
 	.direct_IO = gfs2_direct_IO,
 	.migratepage = buffer_migrate_page,
 	.is_partially_uptodate = block_is_partially_uptodate,
+	.error_remove_page = generic_error_remove_page,
 };
 
 static const struct address_space_operations gfs2_ordered_aops = {

@@ -1151,6 +1152,7 @@ static const struct address_space_operat
 	.direct_IO = gfs2_direct_IO,
 	.migratepage = buffer_migrate_page,
 	.is_partially_uptodate = block_is_partially_uptodate,
+	.error_remove_page = generic_error_remove_page,
 };
 
 static const struct address_space_operations gfs2_jdata_aops = {

@@ -1166,6 +1168,7 @@ static const struct address_space_operat
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
 	.is_partially_uptodate = block_is_partially_uptodate,
+	.error_remove_page = generic_error_remove_page,
 };
 
 void gfs2_set_aops(struct inode *inode)

fs/ntfs/aops.c +2 -0

Index: linux/fs/ntfs/aops.c
===================================================================
@@ -1550,6 +1550,7 @@ const struct address_space_operations nt
 	.migratepage	= buffer_migrate_page,	/* Move a page cache page from
 						   one physical page to an
 						   other. */
+	.error_remove_page = generic_error_remove_page,
 };
 
 /**

@@ -1569,6 +1570,7 @@ const struct address_space_operations nt
 	.migratepage	= buffer_migrate_page,	/* Move a page cache page from
 						   one physical page to an
 						   other. */
+	.error_remove_page = generic_error_remove_page,
 };
 
 #ifdef NTFS_RW

fs/ocfs2/aops.c +1 -0

Index: linux/fs/ocfs2/aops.c
===================================================================
@@ -1968,4 +1968,5 @@ const struct address_space_operations oc
 	.releasepage		= ocfs2_releasepage,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate	= block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };

fs/xfs/linux-2.6/xfs_aops.c +1 -0

Index: linux/fs/xfs/linux-2.6/xfs_aops.c
===================================================================
@@ -1636,4 +1636,5 @@ const struct address_space_operations xf
 	.direct_IO		= xfs_vm_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };

mm/shmem.c +1 -0

Index: linux/mm/shmem.c
===================================================================
@@ -2421,6 +2421,7 @@ static const struct address_space_operat
 	.write_end	= shmem_write_end,
 #endif
 	.migratepage	= migrate_page,
+	.error_remove_page = generic_error_remove_page,
 };
 
 static const struct file_operations shmem_file_operations = {

fs/ext2/inode.c +2 -0

Index: linux/fs/ext2/inode.c
===================================================================
@@ -819,6 +819,7 @@ const struct address_space_operations ex
 	.writepages		= ext2_writepages,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate	= block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 const struct address_space_operations ext2_aops_xip = {

@@ -837,6 +838,7 @@ const struct address_space_operations ex
 	.direct_IO		= ext2_direct_IO,
 	.writepages		= ext2_writepages,
 	.migratepage		= buffer_migrate_page,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 /*

fs/ext3/inode.c +3 -0

Index: linux/fs/ext3/inode.c
===================================================================
@@ -1819,6 +1819,7 @@ static const struct address_space_operat
 	.direct_IO		= ext3_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 static const struct address_space_operations ext3_writeback_aops = {

@@ -1834,6 +1835,7 @@ static const struct address_space_operat
 	.direct_IO		= ext3_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 static const struct address_space_operations ext3_journalled_aops = {

@@ -1848,6 +1850,7 @@ static const struct address_space_operat
 	.invalidatepage		= ext3_invalidatepage,
 	.releasepage		= ext3_releasepage,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 void ext3_set_aops(struct inode *inode)

fs/ext4/inode.c +4 -0

Index: linux/fs/ext4/inode.c
===================================================================
@@ -3373,6 +3373,7 @@ static const struct address_space_operat
 	.direct_IO		= ext4_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 static const struct address_space_operations ext4_writeback_aops = {

@@ -3388,6 +3389,7 @@ static const struct address_space_operat
 	.direct_IO		= ext4_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 static const struct address_space_operations ext4_journalled_aops = {

@@ -3402,6 +3404,7 @@ static const struct address_space_operat
 	.invalidatepage		= ext4_invalidatepage,
 	.releasepage		= ext4_releasepage,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 static const struct address_space_operations ext4_da_aops = {

@@ -3418,6 +3421,7 @@ static const struct address_space_operat
 	.direct_IO		= ext4_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 void ext4_set_aops(struct inode *inode)

Andi Kleen

author

2009-08-05 09:36 UTC #50 lore raw

PATCH 17/19 HWPOISON: Enable error_remove_page for NFS

In reply to Andi Kleen

Enable hardware memory error handling for NFS

Truncation of data pages at runtime should be safe in NFS, even when it doesn't support migration so far.

Cc [email protected]

Signed-off-by Andi Kleen <[email protected]>

---
 fs/nfs/file.c |    1 +
 1 file changed, 1 insertion(+)

fs/nfs/file.c +1 -0

Index: linux/fs/nfs/file.c
===================================================================
@@ -480,6 +480,7 @@ const struct address_space_operations nf
 	.releasepage = nfs_release_page,
 	.direct_IO = nfs_direct_IO,
 	.launder_page = nfs_launder_page,
+	.error_remove_page = generic_error_remove_page,
 };
 
 /*

Andi Kleen

author

2009-08-05 09:36 UTC #51 lore raw

PATCH 18/19 HWPOISON: Add madvise() based injector for hardware poisoned pages v3

In reply to Andi Kleen

Impact: optional, useful for debugging

Add a new madvice sub command to inject poison for some pages in a process' address space. This is useful for testing the poison page handling.

Open issues:

- This patch allows root to tie up arbitary amounts of memory. Should this be disabled inside containers?
- There's a small race window between getting the page and injecting. The patch drops the ref count because otherwise memory_failure complains about dangling references. In theory with a multi threaded injector one could inject poison for a process foreign page this way. Not a serious issue right now.

v2: Use write flag for get_user_pages to make sure to always get a fresh page
v3: Don't request write mapping (Fengguang Wu)

Signed-off-by Andi Kleen <[email protected]>

---
 include/asm-generic/mman-common.h |    1 +
 mm/madvise.c                      |   34 ++++++++++++++++++++++++++++++++++
 2 files changed, 35 insertions(+)

mm/madvise.c +34 -0

Index: linux/mm/madvise.c
===================================================================
@@ -207,6 +207,36 @@ static long madvise_remove(struct vm_are
 	return error;
 }
 
+#ifdef CONFIG_MEMORY_FAILURE
+/*
+ * Error injection support for memory error handling.
+ */
+static int madvise_hwpoison(unsigned long start, unsigned long end)
+{
+	int ret = -EIO;
+	/*
+	 * RED-PEN
+	 * This allows to tie up arbitary amounts of memory.
+	 * Might be a good idea to disable it inside containers even for root.
+	 */
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+	for (; start < end; start += PAGE_SIZE) {
+		struct page *p;
+		int ret = get_user_pages(current, current->mm, start, 1,
+						0, 0, &p, NULL);
+		if (ret != 1)
+			return ret;
+		printk(KERN_INFO "Injecting memory failure for page %lx at %lx\n",
+		       page_to_pfn(p), start);
+		/* Ignore return value for now */
+		__memory_failure(page_to_pfn(p), 0, 1);
+		put_page(p);
+	}
+	return ret;
+}
+#endif
+
 static long
 madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
 		unsigned long start, unsigned long end, int behavior)

@@ -307,6 +337,10 @@ SYSCALL_DEFINE3(madvise, unsigned long,
 	int write;
 	size_t len;
 
+#ifdef CONFIG_MEMORY_FAILURE
+	if (behavior == MADV_HWPOISON)
+		return madvise_hwpoison(start, start+len_in);
+#endif
 	if (!madvise_behavior_valid(behavior))
 		return error;

include/asm-generic/mman-common.h +1 -0

Index: linux/include/asm-generic/mman-common.h
===================================================================
@@ -34,6 +34,7 @@
 #define MADV_REMOVE	9		/* remove these pages & resources */
 #define MADV_DONTFORK	10		/* don't inherit across fork */
 #define MADV_DOFORK	11		/* do inherit across fork */
+#define MADV_HWPOISON	12		/* poison a page for testing */
 
 /* compatibility flags */
 #define MAP_FILE	0

Andi Kleen

author

2009-08-05 09:36 UTC #27 lore raw

PATCH 15/19 HWPOISON: The high level memory error handler in the VM v7

In reply to Andi Kleen

Add the high level memory handler that poisons pages
that got corrupted by hardware (typically by a two bit flip in a DIMM or a cache) on the Linux level. The goal is to prevent everyone from accessing these pages in the future.

This done at the VM level by marking a page hwpoisoned and doing the appropriate action based on the type of page it is.

The code that does this is portable and lives in mm/memory-failure.c

To quote the overview comment:

* High level machine check handler. Handles pages reported by the
* hardware as being corrupted usually due to a 2bit ECC memory or cache
* failure.
*
* This focuses on pages detected as corrupted in the background.
* When the current CPU tries to consume corruption the currently
* running process can just be killed directly instead. This implies
* that if the error cannot be handled for some reason it's safe to
* just ignore it because no corruption has been consumed yet. Instead
* when that happens another machine check will happen.
*
* Handles page cache pages in various states. The tricky part
* here is that we can access any page asynchronous to other VM
* users, because memory failures could happen anytime and anywhere,
* possibly violating some of their assumptions. This is why this code
* has to be extremely careful. Generally it tries to use normal locking
* rules, as in get the standard locks, even if that means the
* error handling takes potentially a long time.
*
* Some of the operations here are somewhat inefficient and have non
* linear algorithmic complexity, because the data structures have not
* been optimized for this case. This is in particular the case
* for the mapping from a vma to a process. Since this case is expected
* to be rare we hope we can get away with this.

There are in principle two strategies to kill processes on poison:
- just unmap the data and wait for an actual reference before killing
- kill as soon as corruption is detected.
Both have advantages and disadvantages and should be used in different situations. Right now both are implemented and can be switched with a new sysctl vm.memory_failure_early_kill The default is early kill.

The patch does some rmap data structure walking on its own to collect processes to kill. This is unusual because normally all rmap data structure knowledge is in rmap.c only. I put it here for now to keep everything together and rmap knowledge has been seeping out anyways

v2: Fix anon vma unlock crash (noticed by Johannes Weiner <[email protected]>) Handle pages on free list correctly (also noticed by Johannes)
Fix inverted try_to_release_page check (found by Chris Mason)
Add documentation for the new sysctl.
Various other cleanups/comment fixes.
v3: Use blockable signal for AO SIGBUS for better qemu handling. Numerous fixes from Fengguang Wu:
New code layout for the table (redone by AK)
Move the hwpoison bit setting before the lock (Fengguang Wu)
Some code cleanups (Fengguang Wu, AK)
Add missing lru_drain (Fengguang Wu)
Do more checks for valid mappings (inspired by patch from Fengguang) Handle free pages and fixes for clean pages (Fengguang)
Removed swap cache handling for now, needs more work
Better mapping checks to avoid races (Fengguang)
Fix swapcache (Fengguang)
Handle private2 pages too (Fengguang)
v4: Various fixes based on review comments from Nick Piggin Document locking order.
Improved comments.
Slightly improved description
Remove bogus hunk.
Wait properly for writeback pages (Nick Piggin)
v5: Improve various comments
Handle page_address_in_vma() failure better by SIGKILL and also
make message debugging only
Clean up printks
Remove redundant PageWriteback check (Nick Piggin)
Add missing clear_page_mlock
Reformat state table to be <80 columns again
Use truncate helper instead of manual truncate in me_pagecache_* Check for metadata buffer pages and reject them.
A few cleanups.
v6:
Fix a printk broken in the last round of cleanups.
More minor cleanups and fixes based on comments from Fengguang Wu. Rename /proc/meminfo Header to "HardwareCorrupted"
Add a printk for the failed mapping case (Fengguang Wu)
Better clean page check (Fengguang Wu)
v7:
Use lru_add_drain_all
Use isolate_lru result in final status
[Merge with Fengguang's version. Only cherry picked changes.]
Fix bad pfn handling in action_result (Fengguang)
Reverse tasklist lock locking order (Nick Piggin)
Check if VMA has really page mapped (Fengguang)
Add a missing page->mapping test (Fengguang)
Account all bad pages (Fengguang)
Misc comment fixes (Fengguang, Andi)
Add sysctl to force panic on memory failure
Only truncate if address space has it enabled, otherwise invalidate Now selectable in Kconfig by the user.
Allow elevated reference count and return status
Misc changes

Cc [email protected]

Signed-off-by Andi Kleen <[email protected]>

Acked-by Rik van Riel <[email protected]>

Reviewed-by Hidehiro Kawai <[email protected]>

---
 Documentation/sysctl/vm.txt |   41 ++
 fs/proc/meminfo.c           |    9 
 include/linux/mm.h          |    7 
 include/linux/rmap.h        |    1 
 kernel/sysctl.c             |   25 +
 mm/Kconfig                  |    9 
 mm/Makefile                 |    1 
 mm/filemap.c                |    4 
 mm/memory-failure.c         |  828 ++++++++++++++++++++++++++++++++++++++++++++
 mm/rmap.c                   |    7 
 10 files changed, 929 insertions(+), 3 deletions(-)

mm/Makefile +1 -0

Index: linux/mm/Makefile
===================================================================
@@ -40,5 +40,6 @@ obj-$(CONFIG_SMP) += allocpercpu.o
 endif
 obj-$(CONFIG_QUICKLIST) += quicklist.o
 obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o
+obj-$(CONFIG_MEMORY_FAILURE) += memory-failure.o
 obj-$(CONFIG_DEBUG_KMEMLEAK) += kmemleak.o
 obj-$(CONFIG_DEBUG_KMEMLEAK_TEST) += kmemleak-test.o

mm/memory-failure.c new file +828 -0

828 lines not on this page · raw message

include/linux/mm.h +7 -0

Index: linux/include/linux/mm.h
===================================================================
@@ -1324,5 +1324,12 @@ void vmemmap_populate_print_last(void);
 extern int account_locked_memory(struct mm_struct *mm, struct rlimit *rlim,
 				 size_t size);
 extern void refund_locked_memory(struct mm_struct *mm, size_t size);
+
+extern void memory_failure(unsigned long pfn, int trapno);
+extern int __memory_failure(unsigned long pfn, int trapno, int ref);
+extern int sysctl_memory_failure_early_kill;
+extern int sysctl_memory_failure_recovery;
+extern atomic_long_t mce_bad_pages;
+
 #endif /* __KERNEL__ */
 #endif /* _LINUX_MM_H */

kernel/sysctl.c +25 -0

Index: linux/kernel/sysctl.c
===================================================================
@@ -1356,6 +1356,31 @@ static struct ctl_table vm_table[] = {
 		.mode		= 0644,
 		.proc_handler	= &scan_unevictable_handler,
 	},
+#ifdef CONFIG_MEMORY_FAILURE
+	{
+		.ctl_name	= CTL_UNNUMBERED,
+		.procname	= "memory_failure_early_kill",
+		.data		= &sysctl_memory_failure_early_kill,
+		.maxlen		= sizeof(sysctl_memory_failure_early_kill),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.strategy	= &sysctl_intvec,
+		.extra1		= &zero,
+		.extra2		= &one,
+	},
+	{
+		.ctl_name	= CTL_UNNUMBERED,
+		.procname	= "memory_failure_recovery",
+		.data		= &sysctl_memory_failure_recovery,
+		.maxlen		= sizeof(sysctl_memory_failure_recovery),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.strategy	= &sysctl_intvec,
+		.extra1		= &zero,
+		.extra2		= &one,
+	},
+#endif
+
 /*
  * NOTE: do not add new entries to this table unless you have read
  * Documentation/sysctl/ctl_unnumbered.txt

fs/proc/meminfo.c +8 -1

Index: linux/fs/proc/meminfo.c
===================================================================
@@ -95,7 +95,11 @@ static int meminfo_proc_show(struct seq_
 		"Committed_AS:   %8lu kB\n"
 		"VmallocTotal:   %8lu kB\n"
 		"VmallocUsed:    %8lu kB\n"
-		"VmallocChunk:   %8lu kB\n",
+		"VmallocChunk:   %8lu kB\n"
+#ifdef CONFIG_MEMORY_FAILURE
+		"HardwareCorrupted: %8lu kB\n"
+#endif
+		,
 		K(i.totalram),
 		K(i.freeram),
 		K(i.bufferram),

@@ -140,6 +144,9 @@ static int meminfo_proc_show(struct seq_
 		(unsigned long)VMALLOC_TOTAL >> 10,
 		vmi.used >> 10,
 		vmi.largest_chunk >> 10
+#ifdef CONFIG_MEMORY_FAILURE
+		,atomic_long_read(&mce_bad_pages) << (PAGE_SHIFT - 10)
+#endif
 		);
 
 	hugetlb_report_meminfo(m);

mm/Kconfig +9 -0

Index: linux/mm/Kconfig
===================================================================
@@ -233,6 +233,15 @@ config DEFAULT_MMAP_MIN_ADDR
 	  /proc/sys/vm/mmap_min_addr tunable.
 
 
+config MEMORY_FAILURE
+	depends on MMU
+	depends on X86_MCE
+	bool "Enable memory failure recovery"
+	help
+	  Enables code to recover from some memory failures on systems
+	  with MCA recovery. This allows a system to continue running
+	  even when some of its memory has uncorrected errors.
+
 config NOMMU_INITIAL_TRIM_EXCESS
 	int "Turn on mmap() excess space trimming before booting"
 	depends on !MMU

Documentation/sysctl/vm.txt +40 -1

59 lines not on this page · raw message

mm/filemap.c +4 -0

Index: linux/mm/filemap.c
===================================================================
@@ -105,6 +105,10 @@
  *
  *  ->task->proc_lock
  *    ->dcache_lock		(proc_pid_lookup)
+ *
+ *  (code doesn't rely on that order, so you could switch it around)
+ *  ->tasklist_lock             (memory_failure, collect_procs_ao)
+ *    ->i_mmap_lock
  */
 
 /*

mm/rmap.c +6 -1

Index: linux/mm/rmap.c
===================================================================
@@ -36,6 +36,11 @@
  *                 mapping->tree_lock (widely used, in set_page_dirty,
  *                           in arch-dependent flush_dcache_mmap_lock,
  *                           within inode_lock in __sync_single_inode)
+ *
+ * (code doesn't rely on that order so it could be switched around)
+ * ->tasklist_lock
+ *   anon_vma->lock      (memory_failure, collect_procs_anon)
+ *     pte map lock
  */
 
 #include <linux/mm.h>

@@ -311,7 +316,7 @@ pte_t *page_check_address(struct page *p
  * if the page is not mapped into the page tables of this VMA.  Only
  * valid for normal file or anonymous VMAs.
  */
-static int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
+int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
 {
 	unsigned long address;
 	pte_t *pte;

include/linux/rmap.h +1 -0

Index: linux/include/linux/rmap.h
===================================================================
@@ -129,6 +129,7 @@ int try_to_munlock(struct page *);
  */
 struct anon_vma *page_lock_anon_vma(struct page *page);
 void page_unlock_anon_vma(struct anon_vma *anon_vma);
+int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
 
 #else	/* !CONFIG_MMU */

Andi Kleen

author

2009-08-05 09:36 UTC #52 lore raw

PATCH 19/19 HWPOISON: Add simple debugfs interface to inject hwpoison on arbitary PFNs

In reply to Andi Kleen

Useful for some testing scenarios, although specific testing is often done better through MADV_POISON

This can be done with the x86 level MCE injector too, but this interface allows it to do independently from low level x86 changes.

Open issues:

Should be disabled for cgroups.

Signed-off-by Andi Kleen <[email protected]>

---
 mm/Kconfig           |    6 +++++-
 mm/Makefile          |    1 +
 mm/hwpoison-inject.c |   40 ++++++++++++++++++++++++++++++++++++++++
 mm/madvise.c         |    2 +-
 4 files changed, 47 insertions(+), 2 deletions(-)

mm/hwpoison-inject.c new file +40 -0

Index: linux/mm/hwpoison-inject.c
===================================================================
@@ -0,0 +1,40 @@
+/* Inject a hwpoison memory failure on a arbitary pfn */
+#include <linux/module.h>
+#include <linux/debugfs.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+
+static struct dentry *hwpoison_dir, *corrupt_pfn;
+
+static int hwpoison_inject(void *data, u64 val)
+{
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+	printk(KERN_INFO "Injecting memory failure at pfn %Lx\n", val);
+	return __memory_failure(val, 18, 0);
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(hwpoison_fops, NULL, hwpoison_inject, "%lli\n");
+
+static void pfn_inject_exit(void)
+{
+	if (hwpoison_dir)
+		debugfs_remove_recursive(hwpoison_dir);
+}
+
+static int pfn_inject_init(void)
+{
+	hwpoison_dir = debugfs_create_dir("hwpoison", NULL);
+	if (hwpoison_dir == NULL)
+		return -ENOMEM;
+	corrupt_pfn = debugfs_create_file("corrupt-pfn", 0600, hwpoison_dir,
+					  NULL, &hwpoison_fops);
+	if (corrupt_pfn == NULL) {
+		pfn_inject_exit();
+		return -ENOMEM;
+	}
+	return 0;
+}
+
+module_init(pfn_inject_init);
+module_exit(pfn_inject_exit);

mm/Kconfig +5 -1

Index: linux/mm/Kconfig
===================================================================
@@ -236,12 +236,16 @@ config DEFAULT_MMAP_MIN_ADDR
 config MEMORY_FAILURE
 	depends on MMU
 	depends on X86_MCE
-	bool "Enable memory failure recovery"
+	bool "Enable recovery from hardware memory errors"
 	help
 	  Enables code to recover from some memory failures on systems
 	  with MCA recovery. This allows a system to continue running
 	  even when some of its memory has uncorrected errors.
 
+config HWPOISON_INJECT
+	tristate "Poison pages injector"
+	depends on MEMORY_FAILURE && DEBUG_KERNEL
+
 config NOMMU_INITIAL_TRIM_EXCESS
 	int "Turn on mmap() excess space trimming before booting"
 	depends on !MMU

mm/Makefile +1 -0

Index: linux/mm/Makefile
===================================================================
@@ -41,5 +41,6 @@ endif
 obj-$(CONFIG_QUICKLIST) += quicklist.o
 obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o
 obj-$(CONFIG_MEMORY_FAILURE) += memory-failure.o
+obj-$(CONFIG_HWPOISON_INJECT) += hwpoison-inject.o
 obj-$(CONFIG_DEBUG_KMEMLEAK) += kmemleak.o
 obj-$(CONFIG_DEBUG_KMEMLEAK_TEST) += kmemleak-test.o

mm/madvise.c +1 -1

Index: linux/mm/madvise.c
===================================================================
@@ -213,7 +213,7 @@ static long madvise_remove(struct vm_are
  */
 static int madvise_hwpoison(unsigned long start, unsigned long end)
 {
-	int ret = -EIO;
+	int ret = 0;
 	/*
 	 * RED-PEN
 	 * This allows to tie up arbitary amounts of memory.

Nick Piggin

2009-08-05 10:20 UTC #12 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Andi Kleen

Quoted 21 lines · Andi Kleen

From: Nick Piggin <[email protected]>
Extract out truncate_inode_page() out of the truncate path so that it can be used by memory-failure.c
[AK: description, headers, fix typos]
v2: Some white space changes from Fengguang Wu
Signed-off-by Andi Kleen <[email protected]>
---
 include/linux/mm.h |    2 ++
 mm/truncate.c      |   29 +++++++++++++++--------------
 2 files changed, 17 insertions(+), 14 deletions(-)
Index: linux/mm/truncate.c
===================================================================

mm/truncate.c EXPORT_SYMBOL(cancel_dirty_page); · line 100

Quoted 10 lines

mm/truncate.c +2 -2

@@ -93,11 +93,11 @@ EXPORT_SYMBOL(cancel_dirty_page);
  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
  */
-static void
+static int
 truncate_complete_page(struct address_space *mapping, struct page *page)
 {
 	if (page->mapping != mapping)
-		return;
+		return -EIO;

Hmm, at this point, the page must have been removed from pagecache, so I don't know if you need to pass an error back?

Quoted 63 lines

mm/truncate.c +13 ~~-12~~

 	if (page_has_private(page))
 		do_invalidatepage(page, 0);

@@ -108,6 +108,7 @@ truncate_complete_page(struct address_sp
 	remove_from_page_cache(page);
 	ClearPageMappedToDisk(page);
 	page_cache_release(page);	/* pagecache ref */
+	return 0;
 }
 
 /*

@@ -135,6 +136,16 @@ invalidate_complete_page(struct address_
 	return ret;
 }
 
+int truncate_inode_page(struct address_space *mapping, struct page *page)
+{
+	if (page_mapped(page)) {
+		unmap_mapping_range(mapping,
+				   (loff_t)page->index << PAGE_CACHE_SHIFT,
+				   PAGE_CACHE_SIZE, 0);
+	}
+	return truncate_complete_page(mapping, page);
+}
+
 /**
  * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
  * @mapping: mapping to truncate

@@ -196,12 +207,7 @@ void truncate_inode_pages_range(struct a
 				unlock_page(page);
 				continue;
 			}
-			if (page_mapped(page)) {
-				unmap_mapping_range(mapping,
-				  (loff_t)page_index<<PAGE_CACHE_SHIFT,
-				  PAGE_CACHE_SIZE, 0);
-			}
-			truncate_complete_page(mapping, page);
+			truncate_inode_page(mapping, page);
 			unlock_page(page);
 		}
 		pagevec_release(&pvec);

@@ -238,15 +244,10 @@ void truncate_inode_pages_range(struct a
 				break;
 			lock_page(page);
 			wait_on_page_writeback(page);
-			if (page_mapped(page)) {
-				unmap_mapping_range(mapping,
-				  (loff_t)page->index<<PAGE_CACHE_SHIFT,
-				  PAGE_CACHE_SIZE, 0);
-			}
+			truncate_inode_page(mapping, page);
 			if (page->index > next)
 				next = page->index;
 			next++;
-			truncate_complete_page(mapping, page);
 			unlock_page(page);
 		}
 		pagevec_release(&pvec);

Index: linux/include/linux/mm.h
===================================================================

include/linux/mm.h static inline void unmap_shared_mapping_ · line 816

Quoted 8 lines

include/linux/mm.h +2 -0

@@ -809,6 +809,8 @@ static inline void unmap_shared_mapping_
 extern int vmtruncate(struct inode * inode, loff_t offset);
 extern int vmtruncate_range(struct inode * inode, loff_t offset, loff_t end);
 
+int truncate_inode_page(struct address_space *mapping, struct page *page);
+
 #ifdef CONFIG_MMU
 extern int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, unsigned int flags);

Christoph Hellwig

2009-08-05 11:12 UTC #29 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Quoted 5 lines · Andi Kleen

Enable removing of corrupted pages through truncation for a bunch of file systems: ext*, xfs, gfs2, ocfs2, ntfs These should cover most server needs.

Quoted 7 lines

I chose the set of migration aware file systems for this for now, assuming they have been especially audited. But in general it should be safe for all file systems on the data area that support read/write and truncate.
Caveat: the hardware error handler does not take i_mutex for now before calling the truncate function. Is that ok?

It will probably need locking, e.g. the iolock in XFS. I'll need to take a look at the actual implementation of generic_error_remove_page to make sense of this.

Is there any way for us to test this functionality without introducing real hardware problems?

Wu Fengguang

2009-08-05 12:17 UTC #30 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 11 lines · Christoph Hellwig

Quoted 10 lines · Andi Kleen
Enable removing of corrupted pages through truncation for a bunch of file systems: ext*, xfs, gfs2, ocfs2, ntfs These should cover most server needs.
I chose the set of migration aware file systems for this for now, assuming they have been especially audited. But in general it should be safe for all file systems on the data area that support read/write and truncate.

Quoted 6 lines

Quoted 2 lines
Caveat: the hardware error handler does not take i_mutex for now before calling the truncate function. Is that ok?
It will probably need locking, e.g. the iolock in XFS. I'll need to take a look at the actual implementation of generic_error_remove_page to make sense of this.

In patch 13, it simply calls truncate_inode_page() for S_ISREG inodes.

Nick suggests call truncate_inode_page() with i_mutex. Sure we can do mutex_trylock(i_mutex), but we'd appreciate it if some fs gurus can demonstrate some bad consequences of not doing so, thanks!

Quoted 2 lines

Is there any way for us to test this functionality without introducing real hardware problems?

We have some additional patches (ugly but works for now) that export interfaces for injecting hwpoison to selected types pages. It can guarantee only data/metadata pages of selected fs will be poisoned. Based on which we can do all kinds of stress testing in user space.

Thanks,
Fengguang

Wu Fengguang

2009-08-05 12:38 UTC #13 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Nick Piggin

Quoted 27 lines · Nick Piggin

Quoted 26 lines · Andi Kleen
From: Nick Piggin <[email protected]>
Extract out truncate_inode_page() out of the truncate path so that it can be used by memory-failure.c
[AK: description, headers, fix typos]
v2: Some white space changes from Fengguang Wu
Signed-off-by Andi Kleen <[email protected]>
---
 include/linux/mm.h |    2 ++
 mm/truncate.c      |   29 +++++++++++++++--------------
 2 files changed, 17 insertions(+), 14 deletions(-)
Index: linux/mm/truncate.c
===================================================================
mm/truncate.c +1 -1
@@ -93,11 +93,11 @@ EXPORT_SYMBOL(cancel_dirty_page);
  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
  */
-static void
+static int

Quoted 8 lines

mm/truncate.c EXPORT_SYMBOL(cancel_dirty_page); · line 100
Quoted 5 lines
mm/truncate.c +1 -1
 truncate_complete_page(struct address_space *mapping, struct page *page)
 {
 	if (page->mapping != mapping)
-		return;
+		return -EIO;
Hmm, at this point, the page must have been removed from pagecache, so I don't know if you need to pass an error back?

Me think so too. When called from hwpoison, the page count and lock have both be taken, so at least _in this case_,

        (page->mapping != mapping)

can be equally written as

        (page->mapping == NULL)

But anyway, the return value is now ignored in upper layer :)

Thanks,
Fengguang

Quoted 67 lines

mm/truncate.c +13 ~~-12~~

 	if (page_has_private(page))
 		do_invalidatepage(page, 0);

@@ -108,6 +108,7 @@ truncate_complete_page(struct address_sp
 	remove_from_page_cache(page);
 	ClearPageMappedToDisk(page);
 	page_cache_release(page);	/* pagecache ref */
+	return 0;
 }
 
 /*

@@ -135,6 +136,16 @@ invalidate_complete_page(struct address_
 	return ret;
 }
 
+int truncate_inode_page(struct address_space *mapping, struct page *page)
+{
+	if (page_mapped(page)) {
+		unmap_mapping_range(mapping,
+				   (loff_t)page->index << PAGE_CACHE_SHIFT,
+				   PAGE_CACHE_SIZE, 0);
+	}
+	return truncate_complete_page(mapping, page);
+}
+
 /**
  * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
  * @mapping: mapping to truncate

@@ -196,12 +207,7 @@ void truncate_inode_pages_range(struct a
 				unlock_page(page);
 				continue;
 			}
-			if (page_mapped(page)) {
-				unmap_mapping_range(mapping,
-				  (loff_t)page_index<<PAGE_CACHE_SHIFT,
-				  PAGE_CACHE_SIZE, 0);
-			}
-			truncate_complete_page(mapping, page);
+			truncate_inode_page(mapping, page);
 			unlock_page(page);
 		}
 		pagevec_release(&pvec);

@@ -238,15 +244,10 @@ void truncate_inode_pages_range(struct a
 				break;
 			lock_page(page);
 			wait_on_page_writeback(page);
-			if (page_mapped(page)) {
-				unmap_mapping_range(mapping,
-				  (loff_t)page->index<<PAGE_CACHE_SHIFT,
-				  PAGE_CACHE_SIZE, 0);
-			}
+			truncate_inode_page(mapping, page);
 			if (page->index > next)
 				next = page->index;
 			next++;
-			truncate_complete_page(mapping, page);
 			unlock_page(page);
 		}
 		pagevec_release(&pvec);

Index: linux/include/linux/mm.h
===================================================================

include/linux/mm.h +0 -0

@@ -809,6 +809,8 @@ static inline void unmap_shared_mapping_
 extern int vmtruncate(struct inode * inode, loff_t offset);
 extern int vmtruncate_range(struct inode * inode, loff_t offset, loff_t end);

Quoted 5 lines

include/linux/mm.h static inline void unmap_shared_mapping_ · line 816

Quoted 5 lines

include/linux/mm.h +2 -0

+int truncate_inode_page(struct address_space *mapping, struct page *page);
+
 #ifdef CONFIG_MMU
 extern int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, unsigned int flags);

Andi Kleen

author

2009-08-05 13:46 UTC #14 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Nick Piggin

Quoted 8 lines · Nick Piggin

mm/truncate.c EXPORT_SYMBOL(cancel_dirty_page); · line 100
Quoted 5 lines
truncate_complete_page(struct address_space *mapping, struct page *page) {
mm/truncate.c +1 -1
 	if (page->mapping != mapping)
-		return;
+		return -EIO;
Hmm, at this point, the page must have been removed from pagecache, so I don't know if you need to pass an error back?

It could be reused, which would be bad for us?

The final check is the page error count in the end anyways.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Andi Kleen

author

2009-08-05 13:50 UTC #31 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Christoph Hellwig

Quoted 11 lines · Christoph Hellwig

Quoted 10 lines · Andi Kleen
Enable removing of corrupted pages through truncation for a bunch of file systems: ext*, xfs, gfs2, ocfs2, ntfs These should cover most server needs.
I chose the set of migration aware file systems for this for now, assuming they have been especially audited. But in general it should be safe for all file systems on the data area that support read/write and truncate.

Quoted 6 lines

Quoted 2 lines
Caveat: the hardware error handler does not take i_mutex for now before calling the truncate function. Is that ok?
It will probably need locking, e.g. the iolock in XFS. I'll need to take a look at the actual implementation of generic_error_remove_page to make sense of this.

Thanks for the review.

It just calls the lower end of the truncate path.

http://git.kernel.org/?p=linux/kernel/git/ak/linux-mce-2.6.git;a=commitdiff;h=b5d3322870b0f6e369809674d2c73857fd6ef480

and

http://git.kernel.org/?p=linux/kernel/git/ak/linux-mce-2.6.git;a=commitdiff;h=4c2df4fa9e4687bde815883fabc602ccc5961272

Is that ok? It should be roughly the same as normal truncate does, except it doesn't take i_mutex currently.

We're careful to not do it for non regular inodes.

Quoted 3 lines

Is there any way for us to test this functionality without introducing real hardware problems?

Yes, there are three different injectors to chose from :)

The easiest one is usually the madvise(MADV_POISON) injector.

Just map a suitable page and

There's a test program in

http://git.kernel.org/?p=utils/cpu/mce/mce-test.git;a=blob;f=tsrc/tinjpage.c;h=954e1edab765d1c141f693ae9767ba9d5491c1aa;hb=HEAD

that can be extended for new tests.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Nick Piggin

2009-08-05 14:01 UTC #15 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Andi Kleen

Quoted 11 lines · Andi Kleen

Quoted 8 lines · Nick Piggin
mm/truncate.c EXPORT_SYMBOL(cancel_dirty_page); · line 100
Quoted 5 lines
truncate_complete_page(struct address_space *mapping, struct page *page) {
mm/truncate.c +1 -1
 	if (page->mapping != mapping)
-		return;
+		return -EIO;
Hmm, at this point, the page must have been removed from pagecache, so I don't know if you need to pass an error back?
It could be reused, which would be bad for us?

I haven't brought up the caller at this point, but IIRC you had the page locked and mapping confirmed at this point anyway so it would never be an error for your code.

Probably it would be nice to just force callers to verify the page. Normally IMO it is much nicer and clearer to do it at the time the page gets locked, unless there is good reason otherwise.

Andi Kleen

author

2009-08-05 14:10 UTC #16 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

Quoted 7 lines

I haven't brought up the caller at this point, but IIRC you had the page locked and mapping confirmed at this point anyway so it would never be an error for your code.
Probably it would be nice to just force callers to verify the page. Normally IMO it is much nicer and clearer to do it at the time the page gets locked, unless there is good reason otherwise.

Ok. I think I'll just keep it as it is for now.

The only reason I added the error code was to make truncate_inode_page fit into .error_remove_page, but then latter I did another wrapper so it could be removed again. But it won't hurt to have it either.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Nick Piggin

2009-08-05 14:16 UTC #17 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

Quoted 8 lines · Andi Kleen

Quoted 7 lines
I haven't brought up the caller at this point, but IIRC you had the page locked and mapping confirmed at this point anyway so it would never be an error for your code.
Probably it would be nice to just force callers to verify the page. Normally IMO it is much nicer and clearer to do it at the time the page gets locked, unless there is good reason otherwise.

Quoted 5 lines

Ok. I think I'll just keep it as it is for now.
The only reason I added the error code was to make truncate_inode_page fit into .error_remove_page, but then latter I did another wrapper so it could be removed again. But it won't hurt to have it either.

OK, it's more of a cleanup/nit.

One question I had for the others (Andrew? other mm guys?) what is the feelings of merging this feature? Leaving aside exact implementation and just considering the high level design and cost/benefit. Last time there were some people objecting, so I wonder the situation now? So does anybody need more convincing? :)

Also I will just cc linux-arch. It would be interesting to know whether powerpc, ia64, or s390 or others would be interested to use this feature?

Thanks,
Nick

Andi Kleen

author

2009-08-05 14:41 UTC #18 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

Quoted 5 lines

One question I had for the others (Andrew? other mm guys?) what is the feelings of merging this feature? Leaving aside exact implementation and just considering the high level design and cost/benefit. Last time there were some people objecting, so I wonder the situation now? So does anybody need more convincing? :)

The main objection last time was that it was a bit too late in the release schedule.

I can't remember anyone really questioning the basic feature itself.

Quoted 2 lines

Also I will just cc linux-arch. It would be interesting to know whether powerpc, ia64, or s390 or others would be interested to use this feature?

ia64 is interested (but no code so far) I talked to DaveM and he seems to be interested for sparc too. I would expect other server architectures to eventually use it as they get around to writing the necessary architecture specific glue.

-Andi

Signature · 1 line

[email protected] -- Speaking for myself only.

Nick Piggin

2009-08-05 14:44 UTC #19 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

Quoted 10 lines · Andi Kleen

Quoted 5 lines
One question I had for the others (Andrew? other mm guys?) what is the feelings of merging this feature? Leaving aside exact implementation and just considering the high level design and cost/benefit. Last time there were some people objecting, so I wonder the situation now? So does anybody need more convincing? :)
The main objection last time was that it was a bit too late in the release schedule.
I can't remember anyone really questioning the basic feature itself.

I can't exactly remember. Maybe it was in a thread with Alan and/or Arjan ;) I don't think the feature itself was questioned as much as cost/benefit. Maybe I was wrong...

I just want to see everyone is happy with the basic idea ;)

Quoted 7 lines

Quoted 2 lines
Also I will just cc linux-arch. It would be interesting to know whether powerpc, ia64, or s390 or others would be interested to use this feature?
ia64 is interested (but no code so far) I talked to DaveM and he seems to be interested for sparc too. I would expect other server architectures to eventually use it as they get around to writing the necessary architecture specific glue.

Matthew Wilcox

2009-08-05 15:00 UTC #20 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Andi Kleen

Quoted 11 lines · Andi Kleen

Quoted 5 lines
One question I had for the others (Andrew? other mm guys?) what is the feelings of merging this feature? Leaving aside exact implementation and just considering the high level design and cost/benefit. Last time there were some people objecting, so I wonder the situation now? So does anybody need more convincing? :)
The main objection last time was that it was a bit too late in the release schedule.
I can't remember anyone really questioning the basic feature itself.

Quoted 7 lines

Quoted 2 lines
Also I will just cc linux-arch. It would be interesting to know whether powerpc, ia64, or s390 or others would be interested to use this feature?
ia64 is interested (but no code so far) I talked to DaveM and he seems to be interested for sparc too. I would expect other server architectures to eventually use it as they get around to writing the necessary architecture specific glue.

parisc could certainly implement this. Don't know if there's interest.

Signature · 4 lines

Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."

Wu Fengguang

2009-08-05 15:12 UTC #23 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Nick Piggin

Quoted 11 lines · Nick Piggin

Quoted 10 lines · Andi Kleen
Quoted 8 lines · Nick Piggin
mm/truncate.c EXPORT_SYMBOL(cancel_dirty_page); · line 100
Quoted 5 lines
truncate_complete_page(struct address_space *mapping, struct page *page) {
mm/truncate.c +1 -1
 	if (page->mapping != mapping)
-		return;
+		return -EIO;
Hmm, at this point, the page must have been removed from pagecache, so I don't know if you need to pass an error back?

Quoted 5 lines

Quoted 1 line
It could be reused, which would be bad for us?
I haven't brought up the caller at this point, but IIRC you had the page locked and mapping confirmed at this point anyway so it would never be an error for your code.

Right, that 'if' will always evaluate to false for the hwpoison case. Because that 'mapping' was taken from 'page->mapping' inside page lock and they will just remain the same values.

Quoted 3 lines

Probably it would be nice to just force callers to verify the page. Normally IMO it is much nicer and clearer to do it at the time the page gets locked, unless there is good reason otherwise.

Yes we do checked page->mapping after taking page lock.

Martin Schwidefsky

2009-08-06 11:48 UTC #21 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

In reply to Nick Piggin

Quoted 17 lines · On Wed, 5 Aug 2009 16:16:42 +0200 Nick Piggin <[email protected]> wrote:

Quoted 8 lines · Andi Kleen
Quoted 7 lines
I haven't brought up the caller at this point, but IIRC you had the page locked and mapping confirmed at this point anyway so it would never be an error for your code.
Probably it would be nice to just force callers to verify the page. Normally IMO it is much nicer and clearer to do it at the time the page gets locked, unless there is good reason otherwise.
Quoted 5 lines
Ok. I think I'll just keep it as it is for now.
The only reason I added the error code was to make truncate_inode_page fit into .error_remove_page, but then latter I did another wrapper so it could be removed again. But it won't hurt to have it either.
OK, it's more of a cleanup/nit.

Quoted 8 lines

One question I had for the others (Andrew? other mm guys?) what is the feelings of merging this feature? Leaving aside exact implementation and just considering the high level design and cost/benefit. Last time there were some people objecting, so I wonder the situation now? So does anybody need more convincing? :)
Also I will just cc linux-arch. It would be interesting to know whether powerpc, ia64, or s390 or others would be interested to use this feature?

This is not relevant for s390, as current machines do transparent memory sparing if a memory module goes bad. Really old machines reported bad memory to the OS by means of a machine check (storage error uncorrected and storage error corrected). I have never seen this happen, the level below the OS deals with these errors for us.

Signature · 4 lines

blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.

Andi Kleen

author

2009-08-06 12:04 UTC #22 lore raw

Re: [PATCH] [11/19] HWPOISON: Refactor truncate to allow direct truncating of page v2

Quoted 5 lines

This is not relevant for s390, as current machines do transparent memory sparing if a memory module goes bad. Really old machines reported bad memory to the OS by means of a machine check (storage error uncorrected and storage error corrected). I have never seen this happen, the level below the OS deals with these errors for us.

Ok fine. It's for the poorer cousins then who can't afford memory mirroring.

-Andi

Signature · 1 line

[email protected] -- Speaking for myself only.

Hidehiro Kawai

2009-08-10 06:36 UTC #32 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Hi,

fs/ext3/inode.c static const struct address_space_operat · line 1823

Quoted 10 lines · Andi Kleen

Index: linux/fs/ext3/inode.c
===================================================================

fs/ext3/inode.c +1 -0

@@ -1819,6 +1819,7 @@ static const struct address_space_operat
 	.direct_IO		= ext3_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };

(I'm sorry if I'm missing the point.)

If my understanding is correct, the following scenario can happen:

1. An uncorrected error on a dirty page cache page is detected by memory scrubbing
2. Kernel unmaps and truncates the page to recover from the error
3. An application reads data from the file location corresponding to the truncated page
==> Old or garbage data will be read into a new page cache page
4. The application modifies the data and write back it to the disk
5. The file will corrurpt!

(Yes, the application is wrong to not do the right thing, i.e. fsync,
but it's not user's fault!)

A similar data corruption can be caused by a write I/O error, because dirty flag is cleared even if the page couldn't be written to the disk.

However, we have a way to avoid this kind of data corruption at least for ext3. If we mount an ext3 filesystem with data=ordered and data_err=abort, all I/O errors on file data block belonging to the committing transaction are checked. When I/O error is found, abort journaling and remount the filesystem with read-only to prevent further updates. This kind of feature is very important for mission critical systems.

If we merge this patch, we would face the data corruption problem again.

I think there are three options,

(1) drop this patch
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
(3) implement a more sophisticated error_remove_page function

fs/ext3/inode.c static const struct address_space_operat · line 1839

Quoted 7 lines

fs/ext3/inode.c +1 -0

 static const struct address_space_operations ext3_writeback_aops = {

@@ -1834,6 +1835,7 @@ static const struct address_space_operat
 	.direct_IO		= ext3_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };

The writeback case would be OK. It's not much different from the I/O error case.

fs/ext3/inode.c static const struct address_space_operat · line 1856

Quoted 9 lines

fs/ext3/inode.c +1 -0

 static const struct address_space_operations ext3_journalled_aops = {

@@ -1848,6 +1850,7 @@ static const struct address_space_operat
 	.invalidatepage		= ext3_invalidatepage,
 	.releasepage		= ext3_releasepage,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
 
 void ext3_set_aops(struct inode *inode)

I'm not sure about the journalled case. I'm going to take a look at it later.

Best regards,

Signature · 9 lines

Hidehiro Kawai
Hitachi, Systems Development Laboratory
Linux Technology Center

--
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>

Wu Fengguang

2009-08-10 07:07 UTC #33 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Hi Hidehiro,

Quoted 28 lines · Hidehiro Kawai

Hi,
fs/ext3/inode.c static const struct address_space_operat · line 1823
Quoted 10 lines · Andi Kleen
Index: linux/fs/ext3/inode.c
===================================================================
fs/ext3/inode.c +1 -0
@@ -1819,6 +1819,7 @@ static const struct address_space_operat
 	.direct_IO		= ext3_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };
(I'm sorry if I'm missing the point.)
If my understanding is correct, the following scenario can happen:
1. An uncorrected error on a dirty page cache page is detected by memory scrubbing
2. Kernel unmaps and truncates the page to recover from the error
3. An application reads data from the file location corresponding to the truncated page
==> Old or garbage data will be read into a new page cache page
4. The application modifies the data and write back it to the disk
5. The file will corrurpt!

Quoted 2 lines

(Yes, the application is wrong to not do the right thing, i.e. fsync,
but it's not user's fault!)

Right. Note that the data has already been corrupted and the above scenario can be called as re-corruption. We set AS_EIO to trigger some IO reporting mechanism so that it won't corrupt *silently*.

Quoted 3 lines

A similar data corruption can be caused by a write I/O error, because dirty flag is cleared even if the page couldn't be written to the disk.

Yes.

Quoted 7 lines

However, we have a way to avoid this kind of data corruption at least for ext3. If we mount an ext3 filesystem with data=ordered and data_err=abort, all I/O errors on file data block belonging to the committing transaction are checked. When I/O error is found, abort journaling and remount the filesystem with read-only to prevent further updates. This kind of feature is very important for mission critical systems.

Agreed. We also set PG_error, which should be enough to trigger such remount?

Quoted 9 lines

If we merge this patch, we would face the data corruption problem again.
I think there are three options,
(1) drop this patch
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
(3) implement a more sophisticated error_remove_page function

In fact we proposed a patch for preventing the re-corruption case, see

        http://lkml.org/lkml/2009/6/11/294

However it is hard to answer the (policy) question "How sticky should the EIO bit remain?".

Quoted 19 lines

fs/ext3/inode.c static const struct address_space_operat · line 1839

Quoted 7 lines

fs/ext3/inode.c +1 -0

 static const struct address_space_operations ext3_writeback_aops = {

@@ -1834,6 +1835,7 @@ static const struct address_space_operat
 	.direct_IO		= ext3_direct_IO,
 	.migratepage		= buffer_migrate_page,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };

The writeback case would be OK. It's not much different from the I/O error case.

Quoted 8 lines

fs/ext3/inode.c +1 -0

 static const struct address_space_operations ext3_journalled_aops = {

@@ -1848,6 +1850,7 @@ static const struct address_space_operat
 	.invalidatepage		= ext3_invalidatepage,
 	.releasepage		= ext3_releasepage,
 	.is_partially_uptodate  = block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
 };

Quoted 4 lines

fs/ext3/inode.c static const struct address_space_operat · line 1856
Quoted 1 line
fs/ext3/inode.c +0 -0
 void ext3_set_aops(struct inode *inode)
I'm not sure about the journalled case. I'm going to take a look at it later.

Thanks,
Fengguang

Andi Kleen

author

2009-08-10 07:44 UTC #37 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Hidehiro Kawai

Hi,

Quoted 1 line

If my understanding is correct, the following scenario can happen:

Yes it can happen.

Quoted 7 lines

1. An uncorrected error on a dirty page cache page is detected by memory scrubbing
2. Kernel unmaps and truncates the page to recover from the error
3. An application reads data from the file location corresponding to the truncated page
==> Old or garbage data will be read into a new page cache page

The problem currently is that the error is not sticky enough and doesn't stay around long enough. It gets reported once, but not in later IO operations.

However it's a generic problem not unique to hwpoison. Me
and Fengguang went through the error propagation as our test program triggered the problem and we looked like it was really a generic problem, not unique to hardware poison (e.g. the IO error handling
on metadata has exactly the same problem)

And redesigning VFS IO error reporting was a bit of of scope for hwpoison. So we decided to not be better than a normal IO error here for now.

An application that handles current IO errors correctly will also also handle hwpoison IO errors correctly. And application that doesn't handle current IO errors correctly will also
not necessarily handle hwpoison correctly (it's not better and not worse) So the hwpoison errors are pretty much the same as the normal IO errors.

The normal error path probably needs some improvements, in particular the address space EIO error error likely needs to be more sticky than it is today.

An application has to handle the error on the first strike.

That is something that could be improved in the VFS -- although I fear any improvements here could also break compatibility. I don't think it's a blocker on hwpoison for now. It needs more design effort and thinking (e.g. likely the address space IO error bit should be separated into multiple bits)

Perhaps you're interested in working on this?

Quoted 10 lines

4. The application modifies the data and write back it to the disk
5. The file will corrurpt!
(Yes, the application is wrong to not do the right thing, i.e. fsync,
but it's not user's fault!)
A similar data corruption can be caused by a write I/O error, because dirty flag is cleared even if the page couldn't be written to the disk.

Quoted 7 lines

However, we have a way to avoid this kind of data corruption at least for ext3. If we mount an ext3 filesystem with data=ordered and data_err=abort, all I/O errors on file data block belonging to the committing transaction are checked. When I/O error is found, abort journaling and remount the filesystem with read-only to prevent further updates. This kind of feature is very important for mission critical systems.

Well it sounds like a potentially useful enhancement to ext3 (or ext4).

One issue is that the default is not ordered anymore since Linus changed the default.

I'm sure other enhancements for IO errors could be done too. Some of the file systems also handle them still quite poorly (e.g. btrfs)

But again I don't think it's a blocker for hwpoison.

Quoted 6 lines

I think there are three options,
(1) drop this patch
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
(3) implement a more sophisticated error_remove_page function

(4) accept that hwpoison error handling is not better and not worse than normal IO error handling.

We opted for (4).

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Hidehiro Kawai

2009-08-11 03:48 UTC #34 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Wu Fengguang

Quoted 10 lines · Wu Fengguang

Quoted 7 lines
However, we have a way to avoid this kind of data corruption at least for ext3. If we mount an ext3 filesystem with data=ordered and data_err=abort, all I/O errors on file data block belonging to the committing transaction are checked. When I/O error is found, abort journaling and remount the filesystem with read-only to prevent further updates. This kind of feature is very important for mission critical systems.
Agreed. We also set PG_error, which should be enough to trigger such remount?

ext3 doesn't check PG_error. Maybe we need to do:

1. trylock_buffer()
2. if step 1. succeeds, then clear_buffer_dirty(), clear_buffer_uptodate(), and set_buffer_write_io_error()

Note that we can't truncate the corrupted page until the error check is done by kjournald.

Quoted 10 lines

Quoted 9 lines
If we merge this patch, we would face the data corruption problem again.
I think there are three options,
(1) drop this patch
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
(3) implement a more sophisticated error_remove_page function

Quoted 6 lines

In fact we proposed a patch for preventing the re-corruption case, see
        http://lkml.org/lkml/2009/6/11/294
However it is hard to answer the (policy) question "How sticky should the EIO bit remain?".

It's a good approach! This approach may also solve my concern, the re-corruption issue caused by transient IO errors.

But I also think it needs a bit more consideration. For example, if the application has the valid data in the user space buffer, it would try to re-write it after detecting an IO error from the previous write. In this case, we should clear the sticky error flag.

Thanks,

Signature · 9 lines

Hidehiro Kawai
Hitachi, Systems Development Laboratory
Linux Technology Center

--
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>

Hidehiro Kawai

2009-08-11 03:51 UTC #38 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Quoted 7 lines · Andi Kleen

Quoted 6 lines
1. An uncorrected error on a dirty page cache page is detected by memory scrubbing
2. Kernel unmaps and truncates the page to recover from the error
3. An application reads data from the file location corresponding to the truncated page
==> Old or garbage data will be read into a new page cache page

Quoted 5 lines

The problem currently is that the error is not sticky enough and doesn't stay around long enough. It gets reported once, but not in later IO operations.
However it's a generic problem not unique to hwpoison. Me

Yes, it's a generic problem, and introducing a sticky error flag is one of the approach to solve the problem. I think it is a good approach because it doesn't depend on individual filesystems.

Quoted 3 lines

And application
that doesn't handle current IO errors correctly will also
not necessarily handle hwpoison correctly (it's not better and not worse)

This is my main concern. I'd like to prevent re-corruption even if applications don't have good manners.

As for usual I/O error, ext3/4 can now do it by using data=ordered and data_err=abort mount options. Moreover, if you mount the ext3/4 filesystem with the additional errors=panic option, kernel gets panic on write error instead of read-only remount. Customers who regard data integrity is very important require these features.

But this patch (PATCH 16/19) introduce this problem again, because it doesn't provide a way to shut out further writes to the fs. Of course, we can do it by setting tolerant level to 0 or memory_failure_recovery to 0. But it would be overkill. That is why I suggested this:

Quoted 11 lines

Quoted 2 lines
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
That is something that could be improved in the VFS -- although I fear any improvements here could also break compatibility. I don't think it's a blocker on hwpoison for now. It needs more design effort and thinking (e.g. likely the address space IO error bit should be separated into multiple bits)
Perhaps you're interested in working on this?

Yes. Transient IO errors have a potential for causing re-corruption problem. Now ext3/4 provide ways to prevent it, but not the other filesystems. We would need a generic way.

Quoted 18 lines

Quoted 10 lines
4. The application modifies the data and write back it to the disk
5. The file will corrurpt!
(Yes, the application is wrong to not do the right thing, i.e. fsync, but it's not user's fault!)
A similar data corruption can be caused by a write I/O error, because dirty flag is cleared even if the page couldn't be written to the disk.
Quoted 7 lines
However, we have a way to avoid this kind of data corruption at least for ext3. If we mount an ext3 filesystem with data=ordered and data_err=abort, all I/O errors on file data block belonging to the committing transaction are checked. When I/O error is found, abort journaling and remount the filesystem with read-only to prevent further updates. This kind of feature is very important for mission critical systems.

Quoted 4 lines

Well it sounds like a potentially useful enhancement to ext3 (or ext4).
One issue is that the default is not ordered anymore since Linus changed the default.

Yes, but what is important is whether the system provides such feature or not.

Quoted 4 lines

I'm sure other enhancements for IO errors could be done too. Some of the file systems also handle them still quite poorly (e.g. btrfs)
But again I don't think it's a blocker for hwpoison.

Unfortunately, it can be a blocker. As I stated, we can block the possible re-corruption caused by transient IO errors on ext3/4 filesystems. But applying this patch (PATCH 16/19), re-corruption can happen even if we use data=ordered, data_err=abort and errors=panic mount options.

So...

Quoted 11 lines

Quoted 6 lines
I think there are three options,
(1) drop this patch
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
(3) implement a more sophisticated error_remove_page function
(4) accept that hwpoison error handling is not better and not worse than normal IO error handling.
We opted for (4).

Could you consider adopting (2) or (3)? Fengguang's sticky EIO approach (http://lkml.org/lkml/2009/6/11/294) is also OK. I hope HWPOISON patches are merged into 2.6.32. So (2) is the best answer for me, because it's simple and less intrusive.

Thanks,

Signature · 9 lines

Hidehiro Kawai
Hitachi, Systems Development Laboratory
Linux Technology Center

--
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>

Andi Kleen

author

2009-08-11 06:59 UTC #35 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Hidehiro Kawai

Quoted 10 lines · Hidehiro Kawai

Quoted 8 lines · Wu Fengguang
Quoted 7 lines
However, we have a way to avoid this kind of data corruption at least for ext3. If we mount an ext3 filesystem with data=ordered and data_err=abort, all I/O errors on file data block belonging to the committing transaction are checked. When I/O error is found, abort journaling and remount the filesystem with read-only to prevent further updates. This kind of feature is very important for mission critical systems.

Quoted 4 lines

Quoted 2 lines
Agreed. We also set PG_error, which should be enough to trigger such remount?
ext3 doesn't check PG_error. Maybe we need to do:

When we truncate the page it's gone so there's no page to set PG_error on.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Andi Kleen

author

2009-08-11 07:17 UTC #39 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Hidehiro Kawai

Quoted 6 lines · Hidehiro Kawai

Quoted 3 lines
And application
that doesn't handle current IO errors correctly will also
not necessarily handle hwpoison correctly (it's not better and not worse)
This is my main concern. I'd like to prevent re-corruption even if applications don't have good manners.

I don't think there's much we can do if the application doesn't check for IO errors properly. What would you do if it doesn't check for IO errors at all? If it checks for IO errors it simply has to check for them on all IO operations -- if they do they will detect hwpoison errors correctly too.

Quoted 5 lines

As for usual I/O error, ext3/4 can now do it by using data=ordered and data_err=abort mount options. Moreover, if you mount the ext3/4 filesystem with the additional errors=panic option, kernel gets panic on write error instead of read-only remount. Customers who regard data integrity is very important require these features.

Well they can also set vm.memory_failure_recovery = 0 then if they don't care about their uptime.

Quoted 2 lines

That is why I suggested this:
Quoted 1 line
Quoted 1 line
(2) merge this patch with new panic_on_dirty_page_cache_corruption

You probably mean panic_on_non_anonymous_dirty_page_cache Normally anonymous memory is dirty.

Quoted 1 line

Quoted 1 line
Quoted 1 line
sysctl

It's unclear to me this special mode is really desirable. Does it bring enough value to the user to justify the complexity of another exotic option? The case is relatively exotic, as in dirty write cache that is mapped to a file.

Try to explain it in documentation and you see how ridiculous it sounds; u it simply doesn't have clean semantics

("In case you have applications with broken error IO handling on your mission critical system ...")

Quoted 10 lines

Quoted 4 lines
I'm sure other enhancements for IO errors could be done too. Some of the file systems also handle them still quite poorly (e.g. btrfs)
But again I don't think it's a blocker for hwpoison.
Unfortunately, it can be a blocker. As I stated, we can block the possible re-corruption caused by transient IO errors on ext3/4 filesystems. But applying this patch (PATCH 16/19), re-corruption can happen even if we use data=ordered, data_err=abort and errors=panic mount options.

We don't corrupt data on disk. Applications
that don't check for IO errors correctly may see stale data from the same file on disk though.

This can happen in all the cases you listed above except for panic-on-error.

If you want panic-on-error behaviour simply set vm.memory_failure_recovery = 0

Quoted 7 lines

Quoted 4 lines
(4) accept that hwpoison error handling is not better and not worse than normal IO error handling.
We opted for (4).
Could you consider adopting (2) or (3)? Fengguang's sticky EIO approach (http://lkml.org/lkml/2009/6/11/294) is also OK.

I believe redesigned IO error handling does not belong in the core hwpoison patchkit. It's big enough as it is and I consider it frozen unless fatal bugs are found -- and frankly this is not a fatal error in my estimation.

If you want to have improved IO error handling feel free to submit it separately. I agree this area could use some work. But it probably needs more design work first.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Wu Fengguang

2009-08-11 13:00 UTC #36 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Hidehiro Kawai

Quoted 8 lines · Hidehiro Kawai

Quoted 6 lines · Wu Fengguang
In fact we proposed a patch for preventing the re-corruption case, see
        http://lkml.org/lkml/2009/6/11/294
However it is hard to answer the (policy) question "How sticky should the EIO bit remain?".

Quoted 7 lines

It's a good approach! This approach may also solve my concern, the re-corruption issue caused by transient IO errors.
But I also think it needs a bit more consideration. For example, if the application has the valid data in the user space buffer, it would try to re-write it after detecting an IO error from the previous write. In this case, we should clear the sticky error flag.

Yes, and maybe more than that. The IO error issue really deserves an independent work, which will inevitably involve lots of discussions with lots of people.

For the data re-corruption problem, "vm.memory_failure_recovery = 0" should be the most clean workaround for now. Can we settle with that? Our goal for this initial hwpoison implementation is to achieve good coverage (not necessarily every possible case :).

Thanks,
Fengguang

Hidehiro Kawai

Reviewed-by

2009-08-12 02:49 UTC #40 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Quoted 6 lines · Andi Kleen

Quoted 4 lines · Hidehiro Kawai
Quoted 3 lines
And application
that doesn't handle current IO errors correctly will also
not necessarily handle hwpoison correctly (it's not better and not worse)

Quoted 8 lines

Quoted 2 lines
This is my main concern. I'd like to prevent re-corruption even if applications don't have good manners.
I don't think there's much we can do if the application doesn't check for IO errors properly. What would you do if it doesn't check for IO errors at all? If it checks for IO errors it simply has to check for them on all IO operations -- if they do they will detect hwpoison errors correctly too.

I believe it's not uncommon for applications to do buffered write and then exit without fsync(). And I think it's difficult to preclude such applications and commands from the system perfectly.

Quoted 7 lines

Quoted 4 lines
That is why I suggested this:
Quoted 2 lines
Quoted 2 lines
(2) merge this patch with new panic_on_dirty_page_cache_corruption sysctl
You probably mean panic_on_non_anonymous_dirty_page_cache Normally anonymous memory is dirty.

Yes, and sorry for my ambiguous description. I used the word "cache" to intend to not include anonymous pages.

Quoted 10 lines

It's unclear to me this special mode is really desirable. Does it bring enough value to the user to justify the complexity of another exotic option? The case is relatively exotic, as in dirty write cache that is mapped to a file.
Try to explain it in documentation and you see how ridiculous it sounds; u it simply doesn't have clean semantics
("In case you have applications with broken error IO handling on your mission critical system ...")

Generally, dropping unwritten dirty page caches is considered to be risky. So the "panic on IO error" policy has been used as usual practice for some systems. I just suggested that we adopted this policy into machine check errors.

vm.memory_failure_recovery satisfies my minimal requirement. It's OK at this stage, but I'd like to improve this in the future.

Another option is to introduce "ignore all" policy instead of panicking at the beginig of memory_failure(). Perhaps it finally causes SRAR machine check, and then kernel will panic or a process will be killed. Anyway, this is a topic for the next stage.

Quoted 10 lines

Quoted 7 lines
Quoted 4 lines
(4) accept that hwpoison error handling is not better and not worse than normal IO error handling.
We opted for (4).
Could you consider adopting (2) or (3)? Fengguang's sticky EIO approach (http://lkml.org/lkml/2009/6/11/294) is also OK.
I believe redesigned IO error handling does not belong in the core hwpoison patchkit.

I agree.

Quoted 3 lines

If you want to have improved IO error handling feel free to submit it separately. I agree this area could use some work. But it probably needs more design work first.

Well, this patch set itself looks good to me.
I also looked into the other patches, I couldn't find any problems (although I'm not good judge of reviewing).

Reviewed-by Hidehiro Kawai <[email protected]>

Thanks,

Signature · 11 lines

Hidehiro Kawai
Hitachi, Systems Development Laboratory
Linux Technology Center



--
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>

Andi Kleen

author

2009-08-12 07:46 UTC #41 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 9 lines · Hidehiro Kawai

Quoted 5 lines
I don't think there's much we can do if the application doesn't check for IO errors properly. What would you do if it doesn't check for IO errors at all? If it checks for IO errors it simply has to check for them on all IO operations -- if they do they will detect hwpoison errors correctly too.
I believe it's not uncommon for applications to do buffered write and then exit without fsync(). And I think it's difficult to preclude such applications and commands from the system perfectly.

That's true, but for anything mission critical you would expect them to use some transactional mechanism, either with O_SYNC or fsync(). Otherwise they always risk data loss anyways.

Quoted 8 lines

Quoted 8 lines
It's unclear to me this special mode is really desirable. Does it bring enough value to the user to justify the complexity of another exotic option? The case is relatively exotic, as in dirty write cache that is mapped to a file.
Try to explain it in documentation and you see how ridiculous it sounds; u it simply doesn't have clean semantics

Quoted 7 lines

Quoted 2 lines
("In case you have applications with broken error IO handling on your mission critical system ...")
Generally, dropping unwritten dirty page caches is considered to be risky. So the "panic on IO error" policy has been used as usual practice for some systems. I just suggested that we adopted this policy into machine check errors.

Hmm, what we could possibly do -- as followon patches -- would be to let error_remove_page check the per file system panic-on-io-error super block setting for dirty pages and panic in this case too. Unfortunately this setting is currently per file system, not generic, so it would need to be a fs specific check (or the flag would need to be moved into a generic fs superblock field first)

I think that would be relatively clean semantics wise. Would you be interested in working on patches for that?

Quoted 4 lines

Another option is to introduce "ignore all" policy instead of panicking at the beginig of memory_failure(). Perhaps it finally causes SRAR machine check, and then kernel will panic or a process will be killed. Anyway, this is a topic for the next stage.

The problem is memory_failure() would then need to start distingushing between AR=1 and AR=0 which it doesn't today.

It could be done, but would need some more work.

Quoted 9 lines

Quoted 3 lines
If you want to have improved IO error handling feel free to submit it separately. I agree this area could use some work. But it probably needs more design work first.
Well, this patch set itself looks good to me.
I also looked into the other patches, I couldn't find any problems (although I'm not good judge of reviewing).
Reviewed-by Hidehiro Kawai <[email protected]>

Thanks for your review and your comments.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Nick Piggin

2009-08-12 08:05 UTC #44 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Quoted 5 lines · Andi Kleen

Quoted 4 lines · Hidehiro Kawai
Quoted 3 lines
And application
that doesn't handle current IO errors correctly will also
not necessarily handle hwpoison correctly (it's not better and not worse)

Quoted 8 lines

Quoted 2 lines
This is my main concern. I'd like to prevent re-corruption even if applications don't have good manners.
I don't think there's much we can do if the application doesn't check for IO errors properly. What would you do if it doesn't check for IO errors at all? If it checks for IO errors it simply has to check for them on all IO operations -- if they do they will detect hwpoison errors correctly too.

But will quite possibly do the wrong thing: ie. try to re-sync the same page again, or try to write the page to a new location, etc.

This is the whole problem with -EIO semantics I brought up.

Quoted 13 lines

Quoted 2 lines
That is why I suggested this:
Quoted 1 line
Quoted 1 line
(2) merge this patch with new panic_on_dirty_page_cache_corruption
You probably mean panic_on_non_anonymous_dirty_page_cache Normally anonymous memory is dirty.
Quoted 1 line
Quoted 1 line
Quoted 1 line
sysctl
It's unclear to me this special mode is really desirable. Does it bring enough value to the user to justify the complexity of another exotic option? The case is relatively exotic, as in dirty write cache that is mapped to a file.

Quoted 5 lines

Try to explain it in documentation and you see how ridiculous it sounds; u it simply doesn't have clean semantics
("In case you have applications with broken error IO handling on your mission critical system ...")

Not broken error handling. It is very simple: if the application is assuming EIO is an error with dirty data being sent to disk, rather than an error with the data itself (which I think may be a common assumption). Then it could have a problem.

If a database for example tries to write the data to another location in response to EIO and then record it in a list of failed IOs before halting the database. Then if it restarts it might try to again try writing out these failed IOs (eg. give the administrator a chance to fix IO devices). Completely made up scenario but it is not outlandish and it would cause bad data corruption.

A mission critical server will *definitely* want to panic on dirty page corruption, IMO, because by definition they should be able to tolerate panic. But if they do not know about this change to -EIO semantics, then it is quite possible to cause problems.

Andi Kleen

author

2009-08-12 08:23 UTC #45 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 3 lines

page corruption, IMO, because by definition they should be able to tolerate panic. But if they do not know about this change to -EIO semantics, then it is quite possible to cause problems.

There's no change really. You already have this problem with any metadata error, which can cause similar trouble. If the application handles those correctly it will also handle hwpoison correctly.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Nick Piggin

2009-08-12 08:46 UTC #46 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 8 lines · Andi Kleen

Quoted 3 lines
page corruption, IMO, because by definition they should be able to tolerate panic. But if they do not know about this change to -EIO semantics, then it is quite possible to cause problems.
There's no change really. You already have this problem with any metadata error, which can cause similar trouble. If the application handles those correctly it will also handle hwpoison correctly.

What do you mean metadata error?

To be clear, we're talking about safety given by a panic on any memory error, versus safety of this hwpoison code (specifically which removes dirty pagecache pages as -EIO).

Andi Kleen

author

2009-08-12 08:57 UTC #47 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 11 lines · Nick Piggin

Quoted 8 lines · Andi Kleen
Quoted 3 lines
page corruption, IMO, because by definition they should be able to tolerate panic. But if they do not know about this change to -EIO semantics, then it is quite possible to cause problems.
There's no change really. You already have this problem with any metadata error, which can cause similar trouble. If the application handles those correctly it will also handle hwpoison correctly.
What do you mean metadata error?

e.g. when there's an write error on the indirect block or any other fs metadata. This can also cause you to lose data. The error reporting also works through the address space like with hwpoison, so it only gets reported once.

I'm not really against fixing that (make the error more sticky as Fengguang puts it), but I don't think it needs to be mixed with hwpoison.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>

Nick Piggin

2009-08-12 09:05 UTC #48 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 11 lines · Andi Kleen

Quoted 10 lines · Nick Piggin
Quoted 8 lines · Andi Kleen
Quoted 3 lines
page corruption, IMO, because by definition they should be able to tolerate panic. But if they do not know about this change to -EIO semantics, then it is quite possible to cause problems.
There's no change really. You already have this problem with any metadata error, which can cause similar trouble. If the application handles those correctly it will also handle hwpoison correctly.

Quoted 6 lines

Quoted 1 line
What do you mean metadata error?
e.g. when there's an write error on the indirect block or any other fs metadata. This can also cause you to lose data. The error reporting also works through the address space like with hwpoison, so it only gets reported once.

Well, this is also a filesystem issue, but anyway the data typically does not get thrown out. So a subsequent fsync should be able to retry.

But if the filesystem can't handle such errors and loses the original data when there is an IO error in newly dirty metadata, then it's a problem in the filesystem really isn't it?

Quoted 3 lines

I'm not really against fixing that (make the error more sticky as Fengguang puts it), but I don't think it needs to be mixed with hwpoison.

I don't know if making it sticky realy "fixes" it. The problem is different semantics of what EIO means. My example illustrates this.

Wu Fengguang

2009-08-12 09:39 UTC #49 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 14 lines · Nick Piggin

Quoted 13 lines · Andi Kleen
Quoted 11 lines · Nick Piggin
Quoted 8 lines · Andi Kleen
Quoted 3 lines
page corruption, IMO, because by definition they should be able to tolerate panic. But if they do not know about this change to -EIO semantics, then it is quite possible to cause problems.
There's no change really. You already have this problem with any metadata error, which can cause similar trouble. If the application handles those correctly it will also handle hwpoison correctly.
What do you mean metadata error?

Quoted 8 lines

Quoted 4 lines
e.g. when there's an write error on the indirect block or any other fs metadata. This can also cause you to lose data. The error reporting also works through the address space like with hwpoison, so it only gets reported once.
Well, this is also a filesystem issue, but anyway the data typically does not get thrown out. So a subsequent fsync should be able to retry.

Right. In normal EIO, the data in page cache is still good and accessible.

Quoted 3 lines

But if the filesystem can't handle such errors and loses the original data when there is an IO error in newly dirty metadata, then it's a problem in the filesystem really isn't it?

Right, and the fs should report EIO on future sync attempts as long as the problem sticks.

Quoted 6 lines

Quoted 3 lines
I'm not really against fixing that (make the error more sticky as Fengguang puts it), but I don't think it needs to be mixed with hwpoison.
I don't know if making it sticky realy "fixes" it. The problem is different semantics of what EIO means. My example illustrates this.

Case 1: (re)sync on EIO: sticky EIO will help.

Case 2: read out the data from page cache and rewrite it somewhere. Sticky EIO is not enough, because here the application assumes the dirty page is still accessible. In this case, patch
http://lkml.org/lkml/2009/6/11/294 will help. It effectively freezes the radix tree, so that no new pages will be loaded to replace the corrupted data and fake a 'good' one.

Thanks,
Fengguang

Hidehiro Kawai

2009-08-12 09:52 UTC #42 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

In reply to Andi Kleen

Quoted 11 lines · Andi Kleen

Quoted 4 lines
Generally, dropping unwritten dirty page caches is considered to be risky. So the "panic on IO error" policy has been used as usual practice for some systems. I just suggested that we adopted this policy into machine check errors.
Hmm, what we could possibly do -- as followon patches -- would be to let error_remove_page check the per file system panic-on-io-error super block setting for dirty pages and panic in this case too. Unfortunately this setting is currently per file system, not generic, so it would need to be a fs specific check (or the flag would need to be moved into a generic fs superblock field first)

A generic setting would be better, so I suggested panic_on_dirty_page_cache_corruption flag which would be checked before invoking error_remove_page(). If we check per-filesystem settings, we might want to notify EIO to the filesystem.

Quoted 2 lines

I think that would be relatively clean semantics wise. Would you be interested in working on patches for that?

Yes. :-)
I will work on this as soon as I come back from summer vacation.

Quoted 9 lines

Quoted 4 lines
Another option is to introduce "ignore all" policy instead of panicking at the beginig of memory_failure(). Perhaps it finally causes SRAR machine check, and then kernel will panic or a process will be killed. Anyway, this is a topic for the next stage.
The problem is memory_failure() would then need to start distingushing between AR=1 and AR=0 which it doesn't today.
It could be done, but would need some more work.

It's my understanding that memory_failure() are never called in AR=1 case. Is it wrong?

Thanks,

Signature · 9 lines

Hidehiro Kawai
Hitachi, Systems Development Laboratory
Linux Technology Center

--
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>

Andi Kleen

author

2009-08-12 10:16 UTC #43 lore raw

Re: [PATCH] [16/19] HWPOISON: Enable .remove_error_page for migration aware file systems

Quoted 14 lines · Hidehiro Kawai

Quoted 11 lines · Andi Kleen
Quoted 4 lines
Generally, dropping unwritten dirty page caches is considered to be risky. So the "panic on IO error" policy has been used as usual practice for some systems. I just suggested that we adopted this policy into machine check errors.
Hmm, what we could possibly do -- as followon patches -- would be to let error_remove_page check the per file system panic-on-io-error super block setting for dirty pages and panic in this case too. Unfortunately this setting is currently per file system, not generic, so it would need to be a fs specific check (or the flag would need to be moved into a generic fs superblock field first)

Quoted 4 lines

A generic setting would be better, so I suggested panic_on_dirty_page_cache_corruption flag which would be checked before invoking error_remove_page(). If we check per-filesystem settings, we might want to notify EIO to the filesystem.

You mean remounting ro if that is set?
That makes sense, but I'm not sure how complicated it would be. I still would prefer to unify it with the file system settings.

Quoted 7 lines

Quoted 4 lines
The problem is memory_failure() would then need to start distingushing between AR=1 and AR=0 which it doesn't today.
It could be done, but would need some more work.
It's my understanding that memory_failure() are never called in AR=1 case. Is it wrong?

Today yes, but we don't want to hardcode that assumption. e.g. for IA64 they will definitely need the equivalent of AR=1 handling.

-Andi

Signature · 7 lines

[email protected] -- Speaking for myself only.

--
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>