Skip to contentSkip to conversation

mm/memory-failure: prioritize prctl(PR_MCE_KILL) over vm.memory_failure_early_kill

PATCH New Andrew Morton 1 participant 5 messages Jun 12, 2020 1 message not in the archive

Subsystems (8)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Andrew Morton

author maintainer
#1 lore raw

PATCH 1/5 mm/memory-failure: prioritize prctl(PR_MCE_KILL) over vm.memory_failure_early_kill

From: Naoya Horiguchi <[email protected]>
Subject: mm/memory-failure: prioritize prctl(PR_MCE_KILL) over vm.memory_failure_early_kill

Patch series "hwpoison: fixes signaling on memory error"

This is a small patchset to solve issues in memory error handler to send SIGBUS to proper process/thread as expected in configuration. Please see descriptions in individual patches for more details.

This patch (of 2):

Early-kill policy is controlled from two types of settings, one is per-process setting prctl(PR_MCE_KILL) and the other is system-wide setting vm.memory_failure_early_kill. Users expect per-process setting to override system-wide setting as many other settings do, but early-kill setting doesn't work as such. For example, if a system configures vm.memory_failure_early_kill to 1 (enabled), a process receives SIGBUS even if it's configured to explicitly disable PF_MCE_KILL by prctl(). That's not desirable for applications with their own policies.

This patch is suggesting to change the priority of these two types of settings, by checking sysctl_memory_failure_early_kill only when a given process has the default kill policy.

Note that this patch is solving a thread choice issue too. Originally, collect_procs() always chooses the main thread when vm.memory_failure_early_kill is 1, even if the process has a dedicated thread for memory error handling. SIGBUS should be sent to the dedicated thread if early-kill is enabled via vm.memory_failure_early_kill as we are doing for PR_MCE_KILL_EARLY processes.

---

 mm/memory-failure.c |   20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)
mm/memory-failure.c renamed from mm/memory-failure.c~mm-memory-failure-prioritize-prctlpr_mce_kill-over-vmmemory_failure_early_kill +10 -10
@@ -402,9 +402,15 @@ static struct task_struct *find_early_ki
 {
 	struct task_struct *t;
 
-	for_each_thread(tsk, t)
-		if ((t->flags & PF_MCE_PROCESS) && (t->flags & PF_MCE_EARLY))
-			return t;
+	for_each_thread(tsk, t) {
+		if (t->flags & PF_MCE_PROCESS) {
+			if (t->flags & PF_MCE_EARLY)
+				return t;
+		} else {
+			if (sysctl_memory_failure_early_kill)
+				return t;
+		}
+	}
 	return NULL;
 }
 
@@ -417,17 +423,11 @@ static struct task_struct *find_early_ki
 static struct task_struct *task_early_kill(struct task_struct *tsk,
 					   int force_early)
 {
-	struct task_struct *t;
 	if (!tsk->mm)
 		return NULL;
 	if (force_early)
 		return tsk;
-	t = find_early_kill_thread(tsk);
-	if (t)
-		return t;
-	if (sysctl_memory_failure_early_kill)
-		return tsk;
-	return NULL;
+	return find_early_kill_thread(tsk);
 }
 
 /*

_

author maintainer
#2 lore raw

PATCH 2/5 mm/memory-failure: send SIGBUS(BUS_MCEERR_AR) only to current thread

In reply to a message not in the archive

From: Naoya Horiguchi <[email protected]>
Subject: mm/memory-failure: send SIGBUS(BUS_MCEERR_AR) only to current thread

Action Required memory error should happen only when a processor is about to access to a corrupted memory, so it's synchronous and only affects current process/thread. Recently commit 872e9a205c84 ("mm,
memory_failure: don't send BUS_MCEERR_AO for action required error") fixed the issue that Action Required memory could unnecessarily send SIGBUS to the processes which share the error memory. But we still have another issue that we could send SIGBUS to a wrong thread.

This is because collect_procs() and task_early_kill() fails to add the current process to "to-kill" list. So this patch is suggesting to fix it. With this fix, SIGBUS(BUS_MCEERR_AR) is never sent to non-current process/thread.

---

 mm/memory-failure.c |   23 ++++++++++++++++-------
 1 file changed, 16 insertions(+), 7 deletions(-)
mm/memory-failure.c renamed from mm/memory-failure.c~mm-memory-failure-send-sigbusbus_mceerr_ar-only-to-current-thread +16 -7
@@ -212,15 +212,13 @@ static int kill_proc(struct to_kill *tk,
 	short addr_lsb = tk->size_shift;
 	int ret = 0;
 
-	if ((t->mm == current->mm) || !(flags & MF_ACTION_REQUIRED))
-		pr_err("Memory failure: %#lx: Sending SIGBUS to %s:%d due to hardware memory corruption\n",
+	pr_err("Memory failure: %#lx: Sending SIGBUS to %s:%d due to hardware memory corruption\n",
 			pfn, t->comm, t->pid);
 
 	if (flags & MF_ACTION_REQUIRED) {
-		if (t->mm == current->mm)
-			ret = force_sig_mceerr(BUS_MCEERR_AR,
+		WARN_ON_ONCE(t != current);
+		ret = force_sig_mceerr(BUS_MCEERR_AR,
 					 (void __user *)tk->addr, addr_lsb);
-		/* send no signal to non-current processes */
 	} else {
 		/*
 		 * Don't use force here, it's convenient if the signal
@@ -419,14 +417,25 @@ static struct task_struct *find_early_ki
  * to be signaled when some page under the process is hwpoisoned.
  * Return task_struct of the dedicated thread (main thread unless explicitly
  * specified) if the process is "early kill," and otherwise returns NULL.
+ *
+ * Note that the above is true for Action Optional case, but not for Action
+ * Required case where SIGBUS should sent only to the current thread.
  */
 static struct task_struct *task_early_kill(struct task_struct *tsk,
 					   int force_early)
 {
 	if (!tsk->mm)
 		return NULL;
-	if (force_early)
-		return tsk;
+	if (force_early) {
+		/*
+		 * Comparing ->mm here because current task might represent
+		 * a subthread, while tsk always points to the main thread.
+		 */
+		if (tsk->mm == current->mm)
+			return current;
+		else
+			return NULL;
+	}
 	return find_early_kill_thread(tsk);
 }
 

_

author maintainer
#3 lore raw

PATCH 3/5 ocfs2: fix build failure when TCP/IP is disabled

In reply to a message not in the archive

From: Tom Seewald <[email protected]>
Subject: ocfs2: fix build failure when TCP/IP is disabled

After commit 12abc5ee7873 ("tcp: add tcp_sock_set_nodelay") and commit c488aeadcbd0 ("tcp: add tcp_sock_set_user_timeout"), building the kernel with OCFS2_FS=y but without INET=y causes it to fail with:

ld: fs/ocfs2/cluster/tcp.o: in function `o2net_accept_many':
tcp.c:(.text+0x21b1): undefined reference to `tcp_sock_set_nodelay'
ld: tcp.c:(.text+0x21c1): undefined reference to `tcp_sock_set_user_timeout '
ld: fs/ocfs2/cluster/tcp.o: in function `o2net_start_connect':
tcp.c:(.text+0x2633): undefined reference to `tcp_sock_set_nodelay'
ld: tcp.c:(.text+0x2643): undefined reference to `tcp_sock_set_user_timeout '

This is due to tcp_sock_set_nodelay() and tcp_sock_set_user_timeout() being declared in linux/tcp.h and defined in net/ipv4/tcp.c, which depend on TCP/IP being enabled.

To fix this, make OCFS2_FS depend on INET=y which already requires NET=y.

12abc5ee7873 ("tcp: add tcp_sock_set_nodelay") and commit

Cc Sagi Grimberg <[email protected]>
Cc Jason Gunthorpe <[email protected]>
Cc David S. Miller <[email protected]>
Cc Mark Fasheh <[email protected]>
Cc Joel Becker <[email protected]>
Cc Junxiao Bi <[email protected]>
Cc Changwei Ge <[email protected]>
Cc Gang He <[email protected]>
Cc Jun Piao <[email protected]>

---

 fs/ocfs2/Kconfig |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/ocfs2/Kconfig renamed from fs/ocfs2/Kconfig~fix-build-failure-of-ocfs2-when-tcp-ip-is-disabled +1 -1
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0-only
 config OCFS2_FS
 	tristate "OCFS2 file system support"
-	depends on NET && SYSFS && CONFIGFS_FS
+	depends on INET && SYSFS && CONFIGFS_FS
 	select JBD2
 	select CRC32
 	select QUOTA

_

author maintainer
#4 lore raw

PATCH 4/5 lib/lzo: fix ambiguous encoding bug in lzo-rle

In reply to a message not in the archive

From: Dave Rodgman <[email protected]>
Subject: lib/lzo: fix ambiguous encoding bug in lzo-rle

In some rare cases, for input data over 32 KB, lzo-rle could encode two different inputs to the same compressed representation, so that decompression is then ambiguous (i.e. data may be corrupted - although zram is not affected because it operates over 4 KB pages).

This modifies the compressor without changing the decompressor or the bitstream format, such that:

- there is no change to how data produced by the old compressor is decompressed

- an old decompressor will correctly decode data from the updated compressor

- performance and compression ratio are not affected

- we avoid introducing a new bitstream format

In testing over 12.8M real-world files totalling 903 GB, three files were affected by this bug. I also constructed 37M semi-random 64 KB files totalling 2.27 TB, and saw no affected files. Finally I tested over files constructed to contain each of the ~1024 possible bad input sequences; for all of these cases, updated lzo-rle worked correctly.

There is no significant impact to performance or compression ratio.

Cc Mark Rutland <[email protected]>
Cc Dave Rodgman <[email protected]>
Cc Willy Tarreau <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Chao Yu <[email protected]>

---

 Documentation/lzo.txt    |    8 ++++++--
 lib/lzo/lzo1x_compress.c |   13 +++++++++++++
 2 files changed, 19 insertions(+), 2 deletions(-)
Documentation/lzo.txt renamed from Documentation/lzo.txt~lib-lzo-fix-ambiguous-encoding-bug-in-lzo-rle +6 -2
@@ -159,11 +159,15 @@ Byte sequences
            distance = 16384 + (H << 14) + D
            state = S (copy S literals after this block)
            End of stream is reached if distance == 16384
+           In version 1 only, to prevent ambiguity with the RLE case when
+           ((distance & 0x803f) == 0x803f) && (261 <= length <= 264), the
+           compressor must not emit block copies where distance and length
+           meet these conditions.
 
         In version 1 only, this instruction is also used to encode a run of
-        zeros if distance = 0xbfff, i.e. H = 1 and the D bits are all 1.
+           zeros if distance = 0xbfff, i.e. H = 1 and the D bits are all 1.
            In this case, it is followed by a fourth byte, X.
-           run length = ((X << 3) | (0 0 0 0 0 L L L)) + 4.
+           run length = ((X << 3) | (0 0 0 0 0 L L L)) + 4
 
       0 0 1 L L L L L  (32..63)
            Copy of small block within 16kB distance (preferably less than 34B)
lib/lzo/lzo1x_compress.c renamed from lib/lzo/lzo1x_compress.c~lib-lzo-fix-ambiguous-encoding-bug-in-lzo-rle +13 -0
@@ -268,6 +268,19 @@ m_len_done:
 				*op++ = (M4_MARKER | ((m_off >> 11) & 8)
 						| (m_len - 2));
 			else {
+				if (unlikely(((m_off & 0x403f) == 0x403f)
+						&& (m_len >= 261)
+						&& (m_len <= 264))
+						&& likely(bitstream_version)) {
+					// Under lzo-rle, block copies
+					// for 261 <= length <= 264 and
+					// (distance & 0x80f3) == 0x80f3
+					// can result in ambiguous
+					// output. Adjust length
+					// to 260 to prevent ambiguity.
+					ip -= m_len - 260;
+					m_len = 260;
+				}
 				m_len -= M4_MAX_LEN;
 				*op++ = (M4_MARKER | ((m_off >> 11) & 8));
 				while (unlikely(m_len > 255)) {

_

author maintainer
#5 lore raw

PATCH 5/5 amdgpu: a NULL ->mm does not mean a thread is a kthread

In reply to a message not in the archive

From: Christoph Hellwig <[email protected]>
Subject: amdgpu: a NULL ->mm does not mean a thread is a kthread

Use the proper API instead.

Fixes 70539bd795002 ("drm/amd: Update MEC HQD loading code for KFD")
Cc Al Viro <[email protected]>
Cc Alex Deucher <[email protected]>
Cc Zhenyu Wang <[email protected]>
Cc Zhi Wang <[email protected]>
Cc Felipe Balbi <[email protected]>
Cc "Michael S. Tsirkin" <[email protected]>
Cc Jason Wang <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>

---

 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h renamed from drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h~amdgpu-a-null-mm-does-not-mean-a-thread-is-a-kthread +1 -1
@@ -196,7 +196,7 @@ uint8_t amdgpu_amdkfd_get_xgmi_hops_coun
 			pagefault_disable();				\
 			if ((mmptr) == current->mm) {			\
 				valid = !get_user((dst), (wptr));	\
-			} else if (current->mm == NULL) {		\
+			} else if (current->flags & PF_KTHREAD) {	\
 				kthread_use_mm(mmptr);			\
 				valid = !get_user((dst), (wptr));	\
 				kthread_unuse_mm(mmptr);		\

_