v1 -> v2:
* ignore empty zero page for madvise_hwpoison directly
[ 57.579580] Injecting memory failure for page 0x19d0 at 0xb77d2000
[ 57.579824] MCE 0x19d0: non LRU page recovery: Ignored
[ 91.290453] MCE: Software-unpoisoned page 0x19d0
[ 91.290456] BUG: Bad page state in process bash pfn:019d0
[ 91.290466] page:f3461a00 count:0 mapcount:0 mapping: (null) index:0x0
[ 91.290467] page flags: 0x40000404(referenced|reserved)
[ 91.290469] Modules linked in: nfsd auth_rpcgss i915 nfs_acl nfs lockd video drm_kms_helper drm bnep rfcomm sunrpc bluetooth psmouse parport_pc ppdev lp serio_raw fscache parport gpio_ich lpc_ich mac_hid i2c_algo_bit tpm_tis wmi usb_storage hid_generic usbhid hid e1000e firewire_ohci firewire_core ahci ptp libahci pps_core crc_itu_t
[ 91.290486] CPU: 3 PID: 2123 Comm: bash Not tainted 3.11.0-rc6+ #12
[ 91.290487] Hardware name: LENOVO 7034DD7/ , BIOS 9HKT47AUS 01//2012
[ 91.290488] 00000000 00000000 e9625ea0 c15ec49b f3461a00 e9625eb8 c15ea119 c17cbf18
[ 91.290491] ef084314 000019d0 f3461a00 e9625ed8 c110dc8a f3461a00 00000001 00000000
[ 91.290494] f3461a00 40000404 00000000 e9625ef8 c110dcc1 f3461a00 f3461a00 000019d0
[ 91.290497] Call Trace:
[ 91.290501] [<c15ec49b>] dump_stack+0x41/0x52
[ 91.290504] [<c15ea119>] bad_page+0xcf/0xeb
[ 91.290515] [<c110dc8a>] free_pages_prepare+0x12a/0x140
[ 91.290517] [<c110dcc1>] free_hot_cold_page+0x21/0x110
[ 91.290519] [<c11123c1>] __put_single_page+0x21/0x30
[ 91.290521] [<c1112815>] put_page+0x25/0x40
[ 91.290524] [<c11544e7>] unpoison_memory+0x107/0x200
[ 91.290526] [<c104a537>] ? ns_capable+0x27/0x60
[ 91.290528] [<c1155720>] hwpoison_unpoison+0x20/0x30
[ 91.290530] [<c1178266>] simple_attr_write+0xb6/0xd0
[ 91.290532] [<c11781b0>] ? generic_fh_to_dentry+0x50/0x50
[ 91.290535] [<c1158c60>] vfs_write+0xa0/0x1b0
[ 91.290537] [<c11781b0>] ? generic_fh_to_dentry+0x50/0x50
[ 91.290539] [<c11590df>] SyS_write+0x4f/0x90
[ 91.290549] [<c15f9a81>] sysenter_do_call+0x12/0x22
[ 91.290550] Disabling lock debugging due to kernel taint
Testcase:
#define _GNU_SOURCE
#include <stdlib.h>
#include <stdio.h>
#include <sys/mman.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/types.h>
#include <errno.h>
#define PAGES_TO_TEST 1
#define PAGE_SIZE 4096
int main(void)
{
char *mem;
mem = mmap(NULL, PAGES_TO_TEST * PAGE_SIZE,
PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (madvise(mem, PAGES_TO_TEST * PAGE_SIZE, MADV_HWPOISON) == -1)
return -1;
munmap(mem, PAGES_TO_TEST * PAGE_SIZE);
return 0;
}
There is one page reference count for default empty zero page, madvise_hwpoison
add another one by get_user_pages_fast. memory_hwpoison reduce one page reference
count since it's a non LRU page. unpoison_memory release the last page reference
count and free empty zero page to buddy system which is not correct since empty
zero page has PG_reserved flag. This patch fix it by ignore empty zero page for
madvise_hwpoison directly.
Suggested-by: Naoya Horiguchi <[email protected]>
Signed-off-by: Wanpeng Li <[email protected]>
---
mm/madvise.c | 2 ++
mm/memory-failure.c | 3 +++
2 files changed, 5 insertions(+)
diff --git a/mm/madvise.c b/mm/madvise.c
index 212f5f1..a20764c 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -352,6 +352,8 @@ static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
int ret = get_user_pages_fast(start, 1, 0, &p);
if (ret != 1)
return ret;
+ if (page_to_pfn(p) == my_zero_pfn(0))
+ continue;
if (bhv == MADV_SOFT_OFFLINE) {
pr_info("Soft offlining page %#lx at %#lx\n",
page_to_pfn(p), start);
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 7cdabc0..68cbca0 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -1331,6 +1331,9 @@ int unpoison_memory(unsigned long pfn)
if (!pfn_valid(pfn))
return -ENXIO;
+ if (pfn == my_zero_pfn(0))
+ return 0;
+
p = pfn_to_page(pfn);
page = compound_head(p);
--
1.8.1.2
Hwpoison inject doesn't implement read method for corrupt-pfn/unpoison-pfn
attributes:
# cat /sys/kernel/debug/hwpoison/corrupt-pfn
cat: /sys/kernel/debug/hwpoison/corrupt-pfn: Permission denied
# cat /sys/kernel/debug/hwpoison/unpoison-pfn
cat: /sys/kernel/debug/hwpoison/unpoison-pfn: Permission denied
This patch change the permission of corrupt-pfn/unpoison-pfn to 0200.
Reviewed-by: Naoya Horiguchi <[email protected]>
Signed-off-by: Wanpeng Li <[email protected]>
---
mm/hwpoison-inject.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/mm/hwpoison-inject.c b/mm/hwpoison-inject.c
index 3a61efc..afc2daa 100644
--- a/mm/hwpoison-inject.c
+++ b/mm/hwpoison-inject.c
@@ -88,12 +88,12 @@ static int pfn_inject_init(void)
* hardware status change, hence do not require hardware support.
* They are mainly for testing hwpoison in software level.
*/
- dentry = debugfs_create_file("corrupt-pfn", 0600, hwpoison_dir,
+ dentry = debugfs_create_file("corrupt-pfn", 0200, hwpoison_dir,
NULL, &hwpoison_fops);
if (!dentry)
goto fail;
- dentry = debugfs_create_file("unpoison-pfn", 0600, hwpoison_dir,
+ dentry = debugfs_create_file("unpoison-pfn", 0200, hwpoison_dir,
NULL, &unpoison_fops);
if (!dentry)
goto fail;
--
1.8.1.2
The return value outside for loop is always zero which means madvise_hwpoison
return success, however, this is not truth for soft_offline_page w/ failure
return value.
Signed-off-by: Wanpeng Li <[email protected]>
---
mm/madvise.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mm/madvise.c b/mm/madvise.c
index a20764c..19b71e4 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -359,7 +359,7 @@ static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
page_to_pfn(p), start);
ret = soft_offline_page(p, MF_COUNT_INCREASED);
if (ret)
- break;
+ return ret;
continue;
}
pr_info("Injecting memory failure for page %#lx at %#lx\n",
--
1.8.1.2
On Tue, Aug 27, 2013 at 10:39:31AM +0800, Wanpeng Li wrote:
> The return value outside for loop is always zero which means madvise_hwpoison
> return success, however, this is not truth for soft_offline_page w/ failure
> return value.
I don't understand what you want to do for what reason. Could you clarify
those?
>
> Signed-off-by: Wanpeng Li <[email protected]>
> ---
> mm/madvise.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/madvise.c b/mm/madvise.c
> index a20764c..19b71e4 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -359,7 +359,7 @@ static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
> page_to_pfn(p), start);
> ret = soft_offline_page(p, MF_COUNT_INCREASED);
> if (ret)
> - break;
> + return ret;
> continue;
> }
> pr_info("Injecting memory failure for page %#lx at %#lx\n",
This seems to introduce no behavioral change.
Thanks,
Naoya Horiguchi
On Tue, Aug 27, 2013 at 10:39:29AM +0800, Wanpeng Li wrote:
> v1 -> v2:
> * ignore empty zero page for madvise_hwpoison directly
>
> [ 57.579580] Injecting memory failure for page 0x19d0 at 0xb77d2000
> [ 57.579824] MCE 0x19d0: non LRU page recovery: Ignored
> [ 91.290453] MCE: Software-unpoisoned page 0x19d0
> [ 91.290456] BUG: Bad page state in process bash pfn:019d0
> [ 91.290466] page:f3461a00 count:0 mapcount:0 mapping: (null) index:0x0
> [ 91.290467] page flags: 0x40000404(referenced|reserved)
> [ 91.290469] Modules linked in: nfsd auth_rpcgss i915 nfs_acl nfs lockd video drm_kms_helper drm bnep rfcomm sunrpc bluetooth psmouse parport_pc ppdev lp serio_raw fscache parport gpio_ich lpc_ich mac_hid i2c_algo_bit tpm_tis wmi usb_storage hid_generic usbhid hid e1000e firewire_ohci firewire_core ahci ptp libahci pps_core crc_itu_t
> [ 91.290486] CPU: 3 PID: 2123 Comm: bash Not tainted 3.11.0-rc6+ #12
> [ 91.290487] Hardware name: LENOVO 7034DD7/ , BIOS 9HKT47AUS 01//2012
> [ 91.290488] 00000000 00000000 e9625ea0 c15ec49b f3461a00 e9625eb8 c15ea119 c17cbf18
> [ 91.290491] ef084314 000019d0 f3461a00 e9625ed8 c110dc8a f3461a00 00000001 00000000
> [ 91.290494] f3461a00 40000404 00000000 e9625ef8 c110dcc1 f3461a00 f3461a00 000019d0
> [ 91.290497] Call Trace:
> [ 91.290501] [<c15ec49b>] dump_stack+0x41/0x52
> [ 91.290504] [<c15ea119>] bad_page+0xcf/0xeb
> [ 91.290515] [<c110dc8a>] free_pages_prepare+0x12a/0x140
> [ 91.290517] [<c110dcc1>] free_hot_cold_page+0x21/0x110
> [ 91.290519] [<c11123c1>] __put_single_page+0x21/0x30
> [ 91.290521] [<c1112815>] put_page+0x25/0x40
> [ 91.290524] [<c11544e7>] unpoison_memory+0x107/0x200
> [ 91.290526] [<c104a537>] ? ns_capable+0x27/0x60
> [ 91.290528] [<c1155720>] hwpoison_unpoison+0x20/0x30
> [ 91.290530] [<c1178266>] simple_attr_write+0xb6/0xd0
> [ 91.290532] [<c11781b0>] ? generic_fh_to_dentry+0x50/0x50
> [ 91.290535] [<c1158c60>] vfs_write+0xa0/0x1b0
> [ 91.290537] [<c11781b0>] ? generic_fh_to_dentry+0x50/0x50
> [ 91.290539] [<c11590df>] SyS_write+0x4f/0x90
> [ 91.290549] [<c15f9a81>] sysenter_do_call+0x12/0x22
> [ 91.290550] Disabling lock debugging due to kernel taint
>
> Testcase:
>
> #define _GNU_SOURCE
> #include <stdlib.h>
> #include <stdio.h>
> #include <sys/mman.h>
> #include <unistd.h>
> #include <fcntl.h>
> #include <sys/types.h>
> #include <errno.h>
>
> #define PAGES_TO_TEST 1
> #define PAGE_SIZE 4096
>
> int main(void)
> {
> char *mem;
>
> mem = mmap(NULL, PAGES_TO_TEST * PAGE_SIZE,
> PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
>
> if (madvise(mem, PAGES_TO_TEST * PAGE_SIZE, MADV_HWPOISON) == -1)
> return -1;
>
> munmap(mem, PAGES_TO_TEST * PAGE_SIZE);
>
> return 0;
> }
>
> There is one page reference count for default empty zero page, madvise_hwpoison
> add another one by get_user_pages_fast. memory_hwpoison reduce one page reference
> count since it's a non LRU page. unpoison_memory release the last page reference
> count and free empty zero page to buddy system which is not correct since empty
> zero page has PG_reserved flag. This patch fix it by ignore empty zero page for
> madvise_hwpoison directly.
>
> Suggested-by: Naoya Horiguchi <[email protected]>
> Signed-off-by: Wanpeng Li <[email protected]>
Thank you.
Reviewed-by: Naoya Horiguchi <[email protected]>
> ---
> mm/madvise.c | 2 ++
> mm/memory-failure.c | 3 +++
> 2 files changed, 5 insertions(+)
>
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 212f5f1..a20764c 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -352,6 +352,8 @@ static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
> int ret = get_user_pages_fast(start, 1, 0, &p);
> if (ret != 1)
> return ret;
> + if (page_to_pfn(p) == my_zero_pfn(0))
> + continue;
> if (bhv == MADV_SOFT_OFFLINE) {
> pr_info("Soft offlining page %#lx at %#lx\n",
> page_to_pfn(p), start);
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 7cdabc0..68cbca0 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1331,6 +1331,9 @@ int unpoison_memory(unsigned long pfn)
> if (!pfn_valid(pfn))
> return -ENXIO;
>
> + if (pfn == my_zero_pfn(0))
> + return 0;
> +
> p = pfn_to_page(pfn);
> page = compound_head(p);
On Tue, Aug 27, 2013 at 11:38:27AM +0800, Wanpeng Li wrote:
> Hi Naoya,
> On Mon, Aug 26, 2013 at 11:28:16PM -0400, Naoya Horiguchi wrote:
> >On Tue, Aug 27, 2013 at 10:39:31AM +0800, Wanpeng Li wrote:
> >> The return value outside for loop is always zero which means madvise_hwpoison
> >> return success, however, this is not truth for soft_offline_page w/ failure
> >> return value.
> >
> >I don't understand what you want to do for what reason. Could you clarify
> >those?
>
> int ret is defined in two place in madvise_hwpoison. One is out of for
> loop and its value is always zero(zero means success for madvise), the
> other one is in for loop. The soft_offline_page function maybe return
> -EBUSY and break, however, the ret out of for loop is return which means
> madvise_hwpoison success.
Oh, I see. Thanks.
Reviewed-by: Naoya Horiguchi <[email protected]>
> Regards,
> Wanpeng Li
>
> >
> >>
> >> Signed-off-by: Wanpeng Li <[email protected]>
> >> ---
> >> mm/madvise.c | 2 +-
> >> 1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/mm/madvise.c b/mm/madvise.c
> >> index a20764c..19b71e4 100644
> >> --- a/mm/madvise.c
> >> +++ b/mm/madvise.c
> >> @@ -359,7 +359,7 @@ static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
> >> page_to_pfn(p), start);
> >> ret = soft_offline_page(p, MF_COUNT_INCREASED);
> >> if (ret)
> >> - break;
> >> + return ret;
> >> continue;
> >> }
> >> pr_info("Injecting memory failure for page %#lx at %#lx\n",
> >
> >This seems to introduce no behavioral change.
> >
> >Thanks,
> >Naoya Horiguchi
>
> --
> 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>
>
On Mon, Aug 26, 2013 at 11:41:36PM -0400, Naoya Horiguchi wrote:
> Date: Mon, 26 Aug 2013 23:41:36 -0400
> From: Naoya Horiguchi <[email protected]>
> To: Wanpeng Li <[email protected]>
> Cc: Andrew Morton <[email protected]>, Andi Kleen
> <[email protected]>, Fengguang Wu <[email protected]>, Tony Luck
> <[email protected]>, [email protected], [email protected],
> [email protected]
> Subject: Re: [PATCH v2 3/3] mm/hwpoison: fix return value of
> madvise_hwpoison
> User-Agent: Mutt 1.5.21 (2010-09-15)
>
> On Tue, Aug 27, 2013 at 11:38:27AM +0800, Wanpeng Li wrote:
> > Hi Naoya,
> > On Mon, Aug 26, 2013 at 11:28:16PM -0400, Naoya Horiguchi wrote:
> > >On Tue, Aug 27, 2013 at 10:39:31AM +0800, Wanpeng Li wrote:
> > >> The return value outside for loop is always zero which means madvise_hwpoison
> > >> return success, however, this is not truth for soft_offline_page w/ failure
> > >> return value.
> > >
> > >I don't understand what you want to do for what reason. Could you clarify
> > >those?
> >
> > int ret is defined in two place in madvise_hwpoison. One is out of for
> > loop and its value is always zero(zero means success for madvise), the
> > other one is in for loop. The soft_offline_page function maybe return
> > -EBUSY and break, however, the ret out of for loop is return which means
> > madvise_hwpoison success.
>
> Oh, I see. Thanks.
>
I don't think such change is a good idea. The original code is obviously
easy to confuse people. Why not removing redundant local variable?
On Tue, Aug 27, 2013 at 04:05:23PM +0800, Wanpeng Li wrote:
> Date: Tue, 27 Aug 2013 16:05:23 +0800
> From: Wanpeng Li <[email protected]>
> To: Chen Gong <[email protected]>
> Cc: Naoya Horiguchi <[email protected]>, Andrew Morton
> <[email protected]>, Andi Kleen <[email protected]>, Fengguang
> Wu <[email protected]>, Tony Luck <[email protected]>,
> [email protected], [email protected]
> Subject: Re: [PATCH v2 3/3] mm/hwpoison: fix return value of
> madvise_hwpoison
> User-Agent: Mutt/1.5.21 (2010-09-15)
>
> Hi Chen,
> On Tue, Aug 27, 2013 at 03:37:01AM -0400, Chen Gong wrote:
> >On Mon, Aug 26, 2013 at 11:41:36PM -0400, Naoya Horiguchi wrote:
> >> Date: Mon, 26 Aug 2013 23:41:36 -0400
> >> From: Naoya Horiguchi <[email protected]>
> >> To: Wanpeng Li <[email protected]>
> >> Cc: Andrew Morton <[email protected]>, Andi Kleen
> >> <[email protected]>, Fengguang Wu <[email protected]>, Tony Luck
> >> <[email protected]>, [email protected], [email protected],
> >> [email protected]
> >> Subject: Re: [PATCH v2 3/3] mm/hwpoison: fix return value of
> >> madvise_hwpoison
> >> User-Agent: Mutt 1.5.21 (2010-09-15)
> >>
> >> On Tue, Aug 27, 2013 at 11:38:27AM +0800, Wanpeng Li wrote:
> >> > Hi Naoya,
> >> > On Mon, Aug 26, 2013 at 11:28:16PM -0400, Naoya Horiguchi wrote:
> >> > >On Tue, Aug 27, 2013 at 10:39:31AM +0800, Wanpeng Li wrote:
> >> > >> The return value outside for loop is always zero which means madvise_hwpoison
> >> > >> return success, however, this is not truth for soft_offline_page w/ failure
> >> > >> return value.
> >> > >
> >> > >I don't understand what you want to do for what reason. Could you clarify
> >> > >those?
> >> >
> >> > int ret is defined in two place in madvise_hwpoison. One is out of for
> >> > loop and its value is always zero(zero means success for madvise), the
> >> > other one is in for loop. The soft_offline_page function maybe return
> >> > -EBUSY and break, however, the ret out of for loop is return which means
> >> > madvise_hwpoison success.
> >>
> >> Oh, I see. Thanks.
> >>
> >I don't think such change is a good idea. The original code is obviously
> >easy to confuse people. Why not removing redundant local variable?
> >
>
> I think the trick here is get_user_pages_fast will return the number of
> pages pinned. It is always 1 in madvise_hwpoison, the return value of
> memory_failure is ignored. Therefore we still need to reset ret to 0
> before return madvise_hwpoison.
>
> Regards,
> Wanpeng Li
>
It looks like the original author wrote in that way deliberately but
botching it. FWIW, I just think its harmness is more than good.
On Tue, 27 Aug 2013 16:02:29 +0800 Wanpeng Li <[email protected]> wrote:
> Hi Chen,
> On Tue, Aug 27, 2013 at 03:37:01AM -0400, Chen Gong wrote:
> >> On Tue, Aug 27, 2013 at 11:38:27AM +0800, Wanpeng Li wrote:
> >> > Hi Naoya,
> >> > On Mon, Aug 26, 2013 at 11:28:16PM -0400, Naoya Horiguchi wrote:
> >> > >On Tue, Aug 27, 2013 at 10:39:31AM +0800, Wanpeng Li wrote:
> >> > >> The return value outside for loop is always zero which means madvise_hwpoison
> >> > >> return success, however, this is not truth for soft_offline_page w/ failure
> >> > >> return value.
> >> > >
> >> > >I don't understand what you want to do for what reason. Could you clarify
> >> > >those?
> >> >
> >> > int ret is defined in two place in madvise_hwpoison. One is out of for
> >> > loop and its value is always zero(zero means success for madvise), the
> >> > other one is in for loop. The soft_offline_page function maybe return
> >> > -EBUSY and break, however, the ret out of for loop is return which means
> >> > madvise_hwpoison success.
> >>
> >> Oh, I see. Thanks.
> >>
> >I don't think such change is a good idea. The original code is obviously
> >easy to confuse people. Why not removing redundant local variable?
> >
>
> I think the trick here is get_user_pages_fast will return the number of
> pages pinned. It is always 1 in madvise_hwpoison, the return value of
> memory_failure is ignored. Therefore we still need to reset ret to 0
> before return madvise_hwpoison.
erk, madvise_hwpoison() has two locals with the same name. Bad.
From: Andrew Morton <[email protected]>
Subject: mm/madvise.c:madvise_hwpoison(): remove local `ret'
madvise_hwpoison() has two locals called "ret". Fix it all up.
Cc: Wanpeng Li <[email protected]>
Cc: Naoya Horiguchi <[email protected]>
Cc: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
---
mm/madvise.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff -puN mm/madvise.c~a mm/madvise.c
--- a/mm/madvise.c~a
+++ a/mm/madvise.c
@@ -343,15 +343,16 @@ static long madvise_remove(struct vm_are
*/
static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
{
- int ret = 0;
-
if (!capable(CAP_SYS_ADMIN))
return -EPERM;
for (; start < end; start += PAGE_SIZE) {
struct page *p;
- int ret = get_user_pages_fast(start, 1, 0, &p);
+ int ret;
+
+ ret = get_user_pages_fast(start, 1, 0, &p);
if (ret != 1)
return ret;
+
if (PageHWPoison(p)) {
put_page(p);
continue;
@@ -369,7 +370,7 @@ static int madvise_hwpoison(int bhv, uns
/* Ignore return value for now */
memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED);
}
- return ret;
+ return 0;
}
#endif
_