Skip to contentSkip to conversation

2.6.30-rc1 regression? -- epoll: BUG: sleeping function called from invalid context

DISCUSSION Stefan Richter 2 participants 4 messages Jun 15, 2009 – Jun 16, 2009 VFS

lore
Share & tools
lore
#0 lore raw

Looks like a regression after 2.6.29, before 2.6.30-rc1, caused by commit 5071f97ec6d74f006072de0ce89b67c8792fe5a1, "epoll: fix epoll's own poll" (since this introduced ep_scan_ready_list), but I haven't fully investigated yet whether this is really the cause.

Test case: Run any libraw1394 or libdc1394 based program on firewire-core on a kernel with the usual selection of debugging options configured in. I didn't have these options enabled for a while, hence noticed only now.

BUG: sleeping function called from invalid context at kernel/mutex.c:278
in_atomic(): 1, irqs_disabled(): 0, pid: 8301, name: dvgrab
no locks held by dvgrab/8301.
Pid: 8301, comm: dvgrab Tainted: G	  W  2.6.30 #2
Call Trace:
[<ffffffff80250bb2>] ? __debug_show_held_locks+0x22/0x24
[<ffffffff8022a91f>] __might_sleep+0x120/0x122
[<ffffffff8045192b>] mutex_lock_nested+0x25/0x2eb
[<ffffffff80253c25>] ? __lock_acquire+0x705/0x793
[<ffffffff802b8f5a>] ep_scan_ready_list+0x3c/0x185
[<ffffffff802b997f>] ? ep_read_events_proc+0x0/0x6c
[<ffffffff802b90b5>] ep_poll_readyevents_proc+0x12/0x14
[<ffffffff802b8b4f>] ep_call_nested+0x9f/0xfa
[<ffffffff802b90a3>] ? ep_poll_readyevents_proc+0x0/0x14
[<ffffffff802b8bf7>] ep_eventpoll_poll+0x4d/0x5b
[<ffffffff8029d525>] do_sys_poll+0x1b4/0x3b5
[<ffffffff8029e17a>] ? __pollwait+0x0/0xce
[<ffffffff8029e248>] ? pollwake+0x0/0x52
[<ffffffff8025162b>] ? mark_held_locks+0x4d/0x6a
[<ffffffff8020b900>] ? restore_args+0x0/0x30
[<ffffffff80251753>] ? trace_hardirqs_on_caller+0x10b/0x12f
[<ffffffff8025162b>] ? mark_held_locks+0x4d/0x6a
[<ffffffff8020b900>] ? restore_args+0x0/0x30
[<ffffffff80253c25>] ? __lock_acquire+0x705/0x793
[<ffffffff80251753>] ? trace_hardirqs_on_caller+0x10b/0x12f
[<ffffffff80251784>] ? trace_hardirqs_on+0xd/0xf
[<ffffffff80235a69>] ? timespec_add_safe+0x34/0x61
[<ffffffff802b9070>] ? ep_scan_ready_list+0x152/0x185
[<ffffffff802483f9>] ? ktime_get_ts+0x49/0x4e
[<ffffffff8029d26b>] ? poll_select_set_timeout+0x5c/0x7f
[<ffffffff8029d778>] sys_poll+0x52/0xb2
[<ffffffff8020aeab>] system_call_fastpath+0x16/0x1b

Any idea how to approach this?

Signature · 3 lines
Stefan Richter
-=====-==--= -==- =----
http://arcgraph.de/sr/
#1 lore raw
Quoted 43 lines · Stefan Richter

Looks like a regression after 2.6.29, before 2.6.30-rc1, caused by commit 5071f97ec6d74f006072de0ce89b67c8792fe5a1, "epoll: fix epoll's own poll" (since this introduced ep_scan_ready_list), but I haven't fully investigated yet whether this is really the cause.

Test case: Run any libraw1394 or libdc1394 based program on firewire-core on a kernel with the usual selection of debugging options configured in. I didn't have these options enabled for a while, hence noticed only now.

BUG: sleeping function called from invalid context at kernel/mutex.c:278
in_atomic(): 1, irqs_disabled(): 0, pid: 8301, name: dvgrab
no locks held by dvgrab/8301.
Pid: 8301, comm: dvgrab Tainted: G	  W  2.6.30 #2
Call Trace:
[<ffffffff80250bb2>] ? __debug_show_held_locks+0x22/0x24
[<ffffffff8022a91f>] __might_sleep+0x120/0x122
[<ffffffff8045192b>] mutex_lock_nested+0x25/0x2eb
[<ffffffff80253c25>] ? __lock_acquire+0x705/0x793
[<ffffffff802b8f5a>] ep_scan_ready_list+0x3c/0x185
[<ffffffff802b997f>] ? ep_read_events_proc+0x0/0x6c
[<ffffffff802b90b5>] ep_poll_readyevents_proc+0x12/0x14
[<ffffffff802b8b4f>] ep_call_nested+0x9f/0xfa
[<ffffffff802b90a3>] ? ep_poll_readyevents_proc+0x0/0x14
[<ffffffff802b8bf7>] ep_eventpoll_poll+0x4d/0x5b
[<ffffffff8029d525>] do_sys_poll+0x1b4/0x3b5
[<ffffffff8029e17a>] ? __pollwait+0x0/0xce
[<ffffffff8029e248>] ? pollwake+0x0/0x52
[<ffffffff8025162b>] ? mark_held_locks+0x4d/0x6a
[<ffffffff8020b900>] ? restore_args+0x0/0x30
[<ffffffff80251753>] ? trace_hardirqs_on_caller+0x10b/0x12f
[<ffffffff8025162b>] ? mark_held_locks+0x4d/0x6a
[<ffffffff8020b900>] ? restore_args+0x0/0x30
[<ffffffff80253c25>] ? __lock_acquire+0x705/0x793
[<ffffffff80251753>] ? trace_hardirqs_on_caller+0x10b/0x12f
[<ffffffff80251784>] ? trace_hardirqs_on+0xd/0xf
[<ffffffff80235a69>] ? timespec_add_safe+0x34/0x61
[<ffffffff802b9070>] ? ep_scan_ready_list+0x152/0x185
[<ffffffff802483f9>] ? ktime_get_ts+0x49/0x4e
[<ffffffff8029d26b>] ? poll_select_set_timeout+0x5c/0x7f
[<ffffffff8029d778>] sys_poll+0x52/0xb2
[<ffffffff8020aeab>] system_call_fastpath+0x16/0x1b
Quoted 1 line

Any idea how to approach this?

That's not the problem. The problem is the patch changing the cookie from "current" to the current CPU (hence bumping preempt count with get_cpu()). Need a fix. Working on it ...

- Davide

#2 lore raw
In reply to Stefan Richter
Quoted 43 lines · Stefan Richter

Looks like a regression after 2.6.29, before 2.6.30-rc1, caused by commit 5071f97ec6d74f006072de0ce89b67c8792fe5a1, "epoll: fix epoll's own poll" (since this introduced ep_scan_ready_list), but I haven't fully investigated yet whether this is really the cause.

Test case: Run any libraw1394 or libdc1394 based program on firewire-core on a kernel with the usual selection of debugging options configured in. I didn't have these options enabled for a while, hence noticed only now.

BUG: sleeping function called from invalid context at kernel/mutex.c:278
in_atomic(): 1, irqs_disabled(): 0, pid: 8301, name: dvgrab
no locks held by dvgrab/8301.
Pid: 8301, comm: dvgrab Tainted: G	  W  2.6.30 #2
Call Trace:
[<ffffffff80250bb2>] ? __debug_show_held_locks+0x22/0x24
[<ffffffff8022a91f>] __might_sleep+0x120/0x122
[<ffffffff8045192b>] mutex_lock_nested+0x25/0x2eb
[<ffffffff80253c25>] ? __lock_acquire+0x705/0x793
[<ffffffff802b8f5a>] ep_scan_ready_list+0x3c/0x185
[<ffffffff802b997f>] ? ep_read_events_proc+0x0/0x6c
[<ffffffff802b90b5>] ep_poll_readyevents_proc+0x12/0x14
[<ffffffff802b8b4f>] ep_call_nested+0x9f/0xfa
[<ffffffff802b90a3>] ? ep_poll_readyevents_proc+0x0/0x14
[<ffffffff802b8bf7>] ep_eventpoll_poll+0x4d/0x5b
[<ffffffff8029d525>] do_sys_poll+0x1b4/0x3b5
[<ffffffff8029e17a>] ? __pollwait+0x0/0xce
[<ffffffff8029e248>] ? pollwake+0x0/0x52
[<ffffffff8025162b>] ? mark_held_locks+0x4d/0x6a
[<ffffffff8020b900>] ? restore_args+0x0/0x30
[<ffffffff80251753>] ? trace_hardirqs_on_caller+0x10b/0x12f
[<ffffffff8025162b>] ? mark_held_locks+0x4d/0x6a
[<ffffffff8020b900>] ? restore_args+0x0/0x30
[<ffffffff80253c25>] ? __lock_acquire+0x705/0x793
[<ffffffff80251753>] ? trace_hardirqs_on_caller+0x10b/0x12f
[<ffffffff80251784>] ? trace_hardirqs_on+0xd/0xf
[<ffffffff80235a69>] ? timespec_add_safe+0x34/0x61
[<ffffffff802b9070>] ? ep_scan_ready_list+0x152/0x185
[<ffffffff802483f9>] ? ktime_get_ts+0x49/0x4e
[<ffffffff8029d26b>] ? poll_select_set_timeout+0x5c/0x7f
[<ffffffff8029d778>] sys_poll+0x52/0xb2
[<ffffffff8020aeab>] system_call_fastpath+0x16/0x1b
Quoted 1 line

Any idea how to approach this?

Do you have a chance to give the patch below a spin, in your context?

- Davide

---
 fs/eventpoll.c |   22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)
fs/eventpoll.c +13 -9
Index: linux-2.6.mod/fs/eventpoll.c
===================================================================
@@ -13,6 +13,7 @@
 
 #include <linux/init.h>
 #include <linux/kernel.h>
+#include <linux/hardirq.h>
 #include <linux/sched.h>
 #include <linux/fs.h>
 #include <linux/file.h>
@@ -98,7 +99,7 @@ struct epoll_filefd {
 struct nested_call_node {
 	struct list_head llink;
 	void *cookie;
-	int cpu;
+	void *ctx;
 };
 
 /*
@@ -317,17 +318,17 @@ static void ep_nested_calls_init(struct 
  * @nproc: Nested call core function pointer.
  * @priv: Opaque data to be passed to the @nproc callback.
  * @cookie: Cookie to be used to identify this nested call.
+ * @ctx: This instance context.
  *
  * Returns: Returns the code returned by the @nproc callback, or -1 if
  *          the maximum recursion limit has been exceeded.
  */
 static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
 			  int (*nproc)(void *, void *, int), void *priv,
-			  void *cookie)
+			  void *cookie, void *ctx)
 {
 	int error, call_nests = 0;
 	unsigned long flags;
-	int this_cpu = get_cpu();
 	struct list_head *lsthead = &ncalls->tasks_call_list;
 	struct nested_call_node *tncur;
 	struct nested_call_node tnode;
@@ -340,7 +341,7 @@ static int ep_call_nested(struct nested_
 	 * very much limited.
 	 */
 	list_for_each_entry(tncur, lsthead, llink) {
-		if (tncur->cpu == this_cpu &&
+		if (tncur->ctx == ctx &&
 		    (tncur->cookie == cookie || ++call_nests > max_nests)) {
 			/*
 			 * Ops ... loop detected or maximum nest level reached.
@@ -352,7 +353,7 @@ static int ep_call_nested(struct nested_
 	}
 
 	/* Add the current task and cookie to the list */
-	tnode.cpu = this_cpu;
+	tnode.ctx = ctx;
 	tnode.cookie = cookie;
 	list_add(&tnode.llink, lsthead);
 
@@ -364,10 +365,9 @@ static int ep_call_nested(struct nested_
 	/* Remove the current task from the list */
 	spin_lock_irqsave(&ncalls->lock, flags);
 	list_del(&tnode.llink);
- out_unlock:
+out_unlock:
 	spin_unlock_irqrestore(&ncalls->lock, flags);
 
-	put_cpu();
 	return error;
 }
 
@@ -408,8 +408,12 @@ static int ep_poll_wakeup_proc(void *pri
  */
 static void ep_poll_safewake(wait_queue_head_t *wq)
 {
+	int this_cpu = get_cpu();
+
 	ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
-		       ep_poll_wakeup_proc, NULL, wq);
+		       ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
+
+	put_cpu();
 }
 
 /*
@@ -663,7 +667,7 @@ static unsigned int ep_eventpoll_poll(st
 	 * could re-enter here.
 	 */
 	pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
-				   ep_poll_readyevents_proc, ep, ep);
+				   ep_poll_readyevents_proc, ep, ep, current);
 
 	return pollflags != -1 ? pollflags : 0;
 }
#3 lore raw
Quoted 5 lines · Davide Libenzi

Do you have a chance to give the patch below a spin, in your context?

- Davide

This works for me, i.e. the previously 100% reliably triggered BUGs went away, and testing did not reveal any issues anymore.

Quoted 90 lines
---
 fs/eventpoll.c |   22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)

Index: linux-2.6.mod/fs/eventpoll.c
===================================================================

fs/eventpoll.c +12 -8
@@ -13,6 +13,7 @@
 
 #include <linux/init.h>
 #include <linux/kernel.h>
+#include <linux/hardirq.h>
 #include <linux/sched.h>
 #include <linux/fs.h>
 #include <linux/file.h>
@@ -98,7 +99,7 @@ struct epoll_filefd {
 struct nested_call_node {
 	struct list_head llink;
 	void *cookie;
-	int cpu;
+	void *ctx;
 };
 
 /*
@@ -317,17 +318,17 @@ static void ep_nested_calls_init(struct 
  * @nproc: Nested call core function pointer.
  * @priv: Opaque data to be passed to the @nproc callback.
  * @cookie: Cookie to be used to identify this nested call.
+ * @ctx: This instance context.
  *
  * Returns: Returns the code returned by the @nproc callback, or -1 if
  *          the maximum recursion limit has been exceeded.
  */
 static int ep_call_nested(struct nested_calls *ncalls, int max_nests,
 			  int (*nproc)(void *, void *, int), void *priv,
-			  void *cookie)
+			  void *cookie, void *ctx)
 {
 	int error, call_nests = 0;
 	unsigned long flags;
-	int this_cpu = get_cpu();
 	struct list_head *lsthead = &ncalls->tasks_call_list;
 	struct nested_call_node *tncur;
 	struct nested_call_node tnode;
@@ -340,7 +341,7 @@ static int ep_call_nested(struct nested_
 	 * very much limited.
 	 */
 	list_for_each_entry(tncur, lsthead, llink) {
-		if (tncur->cpu == this_cpu &&
+		if (tncur->ctx == ctx &&
 		    (tncur->cookie == cookie || ++call_nests > max_nests)) {
 			/*
 			 * Ops ... loop detected or maximum nest level reached.
@@ -352,7 +353,7 @@ static int ep_call_nested(struct nested_
 	}
 
 	/* Add the current task and cookie to the list */
-	tnode.cpu = this_cpu;
+	tnode.ctx = ctx;
 	tnode.cookie = cookie;
 	list_add(&tnode.llink, lsthead);
 
@@ -364,10 +365,9 @@ static int ep_call_nested(struct nested_
 	/* Remove the current task from the list */
 	spin_lock_irqsave(&ncalls->lock, flags);
 	list_del(&tnode.llink);
- out_unlock:
+out_unlock:
 	spin_unlock_irqrestore(&ncalls->lock, flags);
 
-	put_cpu();
 	return error;
 }
 
@@ -408,8 +408,12 @@ static int ep_poll_wakeup_proc(void *pri
  */
 static void ep_poll_safewake(wait_queue_head_t *wq)
 {
+	int this_cpu = get_cpu();
+
 	ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS,
-		       ep_poll_wakeup_proc, NULL, wq);
+		       ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu);
+
+	put_cpu();
 }
 
 /*

fs/eventpoll.c static unsigned int ep_eventpoll_poll(st · line 673

Quoted 8 lines
fs/eventpoll.c +1 -1
@@ -663,7 +667,7 @@ static unsigned int ep_eventpoll_poll(st
 	 * could re-enter here.
 	 */
 	pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS,
-				   ep_poll_readyevents_proc, ep, ep);
+				   ep_poll_readyevents_proc, ep, ep, current);
 
 	return pollflags != -1 ? pollflags : 0;
 }

Thanks for the quick response,

Signature · 3 lines
Stefan Richter
-=====-==--= -==- =----
http://arcgraph.de/sr/