Skip to contentSkip to conversation

.29rc2 lockdep report. fb_mmap vs sys_mmap2

PATCH In review Dave Jones 5 participants 9 messages Jan 17, 2009 – Jan 19, 2009 Framebuffer layer

lore
Share & tools
lore

Dave Jones

author
#0 lore raw

Two mmaps enter! Fight!

=======================================================
[ INFO: possible circular locking dependency detected ] 2.6.29-0.41.rc2.fc11.i686 #1
-------------------------------------------------------
plymouthd/669 is trying to acquire lock:
(&fb_info->lock){--..}, at: [<c055355c>] fb_mmap+0x87/0x156

but task is already holding lock:
(&mm->mmap_sem){----}, at: [<c0406a5d>] sys_mmap2+0x44/0x7b

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&mm->mmap_sem){----}:

       [<c044fef3>] __lock_acquire+0x9af/0xb22
       [<c04500c1>] lock_acquire+0x5b/0x81
       [<c048e036>] might_fault+0x60/0x80
       [<c053dedc>] copy_to_user+0x2c/0xfc
       [<c05542cc>] fb_ioctl+0x247/0x338
       [<c04b0c9f>] vfs_ioctl+0x22/0x69
       [<c04b1214>] do_vfs_ioctl+0x46a/0x4a3
       [<c04b128d>] sys_ioctl+0x40/0x5a
       [<c0403b76>] syscall_call+0x7/0xb
       [<ffffffff>] 0xffffffff

-> #0 (&fb_info->lock){--..}:

       [<c044fdc8>] __lock_acquire+0x884/0xb22
       [<c04500c1>] lock_acquire+0x5b/0x81
       [<c06e1b98>] __mutex_lock_common+0xd5/0x329
       [<c06e1e84>] mutex_lock_nested+0x2e/0x36
       [<c055355c>] fb_mmap+0x87/0x156
       [<c0494620>] mmap_region+0x22b/0x43c
       [<c0494aa2>] do_mmap_pgoff+0x271/0x2d1
       [<c0406a73>] sys_mmap2+0x5a/0x7b
       [<c0403b76>] syscall_call+0x7/0xb
       [<ffffffff>] 0xffffffff

other info that might help us debug this:

1 lock held by plymouthd/669:
#0: (&mm->mmap_sem){----}, at: [<c0406a5d>] sys_mmap2+0x44/0x7b

 stack backtrace:
 Pid: 669, comm: plymouthd Not tainted 2.6.29-0.41.rc2.fc11.i686 #1
 Call Trace:
 [<c06e09b6>] ? printk+0xf/0x11
 [<c044f32d>] print_circular_bug_tail+0x5d/0x68
 [<c044fdc8>] __lock_acquire+0x884/0xb22
 [<c04500c1>] lock_acquire+0x5b/0x81
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c06e1b98>] __mutex_lock_common+0xd5/0x329
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c06e1e84>] mutex_lock_nested+0x2e/0x36
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c055355c>] fb_mmap+0x87/0x156
 [<c0494620>] mmap_region+0x22b/0x43c
 [<c0494aa2>] do_mmap_pgoff+0x271/0x2d1
 [<c0406a73>] sys_mmap2+0x5a/0x7b
 [<c0403b76>] syscall_call+0x7/0xb
Signature · 1 line
http://www.codemonkey.org.uk
#1 lore raw
Quoted 56 lines · Dave Jones

Two mmaps enter! Fight!

=======================================================
[ INFO: possible circular locking dependency detected ] 2.6.29-0.41.rc2.fc11.i686 #1
-------------------------------------------------------
plymouthd/669 is trying to acquire lock:
(&fb_info->lock){--..}, at: [<c055355c>] fb_mmap+0x87/0x156

but task is already holding lock:
(&mm->mmap_sem){----}, at: [<c0406a5d>] sys_mmap2+0x44/0x7b

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&mm->mmap_sem){----}:

       [<c044fef3>] __lock_acquire+0x9af/0xb22
       [<c04500c1>] lock_acquire+0x5b/0x81
       [<c048e036>] might_fault+0x60/0x80
       [<c053dedc>] copy_to_user+0x2c/0xfc
       [<c05542cc>] fb_ioctl+0x247/0x338
       [<c04b0c9f>] vfs_ioctl+0x22/0x69
       [<c04b1214>] do_vfs_ioctl+0x46a/0x4a3
       [<c04b128d>] sys_ioctl+0x40/0x5a
       [<c0403b76>] syscall_call+0x7/0xb
       [<ffffffff>] 0xffffffff

-> #0 (&fb_info->lock){--..}:

       [<c044fdc8>] __lock_acquire+0x884/0xb22
       [<c04500c1>] lock_acquire+0x5b/0x81
       [<c06e1b98>] __mutex_lock_common+0xd5/0x329
       [<c06e1e84>] mutex_lock_nested+0x2e/0x36
       [<c055355c>] fb_mmap+0x87/0x156
       [<c0494620>] mmap_region+0x22b/0x43c
       [<c0494aa2>] do_mmap_pgoff+0x271/0x2d1
       [<c0406a73>] sys_mmap2+0x5a/0x7b
       [<c0403b76>] syscall_call+0x7/0xb
       [<ffffffff>] 0xffffffff

other info that might help us debug this:

1 lock held by plymouthd/669:
#0: (&mm->mmap_sem){----}, at: [<c0406a5d>] sys_mmap2+0x44/0x7b

 stack backtrace:
 Pid: 669, comm: plymouthd Not tainted 2.6.29-0.41.rc2.fc11.i686 #1
 Call Trace:
 [<c06e09b6>] ? printk+0xf/0x11
 [<c044f32d>] print_circular_bug_tail+0x5d/0x68
 [<c044fdc8>] __lock_acquire+0x884/0xb22
 [<c04500c1>] lock_acquire+0x5b/0x81
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c06e1b98>] __mutex_lock_common+0xd5/0x329
Quoted 8 lines
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c06e1e84>] mutex_lock_nested+0x2e/0x36
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c055355c>] fb_mmap+0x87/0x156
 [<c0494620>] mmap_region+0x22b/0x43c
 [<c0494aa2>] do_mmap_pgoff+0x271/0x2d1
 [<c0406a73>] sys_mmap2+0x5a/0x7b
 [<c0403b76>] syscall_call+0x7/0xb

I reported this one already some time ago [1], Andrea Righi [cc'd] also had a fix [2] but it seemed to not have made it in. Andrea, perhaps you can resend it?

Hannes

[1] http://lkml.org/lkml/2008/10/22/759
[2] http://lkml.org/lkml/2008/10/28/377

#2 lore raw
Quoted 66 lines · Johannes Weiner
Quoted 56 lines · Dave Jones

Two mmaps enter! Fight!

=======================================================
[ INFO: possible circular locking dependency detected ] 2.6.29-0.41.rc2.fc11.i686 #1
-------------------------------------------------------
plymouthd/669 is trying to acquire lock:
(&fb_info->lock){--..}, at: [<c055355c>] fb_mmap+0x87/0x156

but task is already holding lock:
(&mm->mmap_sem){----}, at: [<c0406a5d>] sys_mmap2+0x44/0x7b

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&mm->mmap_sem){----}:

       [<c044fef3>] __lock_acquire+0x9af/0xb22
       [<c04500c1>] lock_acquire+0x5b/0x81
       [<c048e036>] might_fault+0x60/0x80
       [<c053dedc>] copy_to_user+0x2c/0xfc
       [<c05542cc>] fb_ioctl+0x247/0x338
       [<c04b0c9f>] vfs_ioctl+0x22/0x69
       [<c04b1214>] do_vfs_ioctl+0x46a/0x4a3
       [<c04b128d>] sys_ioctl+0x40/0x5a
       [<c0403b76>] syscall_call+0x7/0xb
       [<ffffffff>] 0xffffffff

-> #0 (&fb_info->lock){--..}:

       [<c044fdc8>] __lock_acquire+0x884/0xb22
       [<c04500c1>] lock_acquire+0x5b/0x81
       [<c06e1b98>] __mutex_lock_common+0xd5/0x329
       [<c06e1e84>] mutex_lock_nested+0x2e/0x36
       [<c055355c>] fb_mmap+0x87/0x156
       [<c0494620>] mmap_region+0x22b/0x43c
       [<c0494aa2>] do_mmap_pgoff+0x271/0x2d1
       [<c0406a73>] sys_mmap2+0x5a/0x7b
       [<c0403b76>] syscall_call+0x7/0xb
       [<ffffffff>] 0xffffffff

other info that might help us debug this:

1 lock held by plymouthd/669:
#0: (&mm->mmap_sem){----}, at: [<c0406a5d>] sys_mmap2+0x44/0x7b

 stack backtrace:
 Pid: 669, comm: plymouthd Not tainted 2.6.29-0.41.rc2.fc11.i686 #1
 Call Trace:
 [<c06e09b6>] ? printk+0xf/0x11
 [<c044f32d>] print_circular_bug_tail+0x5d/0x68
 [<c044fdc8>] __lock_acquire+0x884/0xb22
 [<c04500c1>] lock_acquire+0x5b/0x81
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c06e1b98>] __mutex_lock_common+0xd5/0x329
Quoted 8 lines
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c06e1e84>] mutex_lock_nested+0x2e/0x36
 [<c055355c>] ? fb_mmap+0x87/0x156
 [<c055355c>] fb_mmap+0x87/0x156
 [<c0494620>] mmap_region+0x22b/0x43c
 [<c0494aa2>] do_mmap_pgoff+0x271/0x2d1
 [<c0406a73>] sys_mmap2+0x5a/0x7b
 [<c0403b76>] syscall_call+0x7/0xb
Quoted 8 lines

I reported this one already some time ago [1], Andrea Righi [cc'd] also had a fix [2] but it seemed to not have made it in. Andrea, perhaps you can resend it?

Hannes

[1] http://lkml.org/lkml/2008/10/22/759
[2] http://lkml.org/lkml/2008/10/28/377

Sure, I can resend it soon, after a test.

-Andrea

#3 lore raw

PATCH 1/1 fbmem: fix copy_from/to_user() with mutex held

In reply to Johannes Weiner

Avoid to call copy_from/to_user() with fb_info->lock mutex held in fbmem fb_ioctl().

NOTE: it doesn't push down the fb_info->lock in each driver's fb_ioctl().

---
 drivers/video/fbcmap.c |   21 +++++--
 drivers/video/fbmem.c  |  158 ++++++++++++++++++++++++++++--------------------
 include/linux/fb.h     |    5 +-
 3 files changed, 112 insertions(+), 72 deletions(-)
drivers/video/fbcmap.c +15 -6
index 91b78e6..b19f12c 100644
@@ -250,10 +250,6 @@ int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *info)
 	int rc, size = cmap->len * sizeof(u16);
 	struct fb_cmap umap;
 
-	if (cmap->start < 0 || (!info->fbops->fb_setcolreg &&
-			        !info->fbops->fb_setcmap))
-		return -EINVAL;
-
 	memset(&umap, 0, sizeof(struct fb_cmap));
 	rc = fb_alloc_cmap(&umap, cmap->len, cmap->transp != NULL);
 	if (rc)
@@ -262,11 +258,24 @@ int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *info)
 	    copy_from_user(umap.green, cmap->green, size) ||
 	    copy_from_user(umap.blue, cmap->blue, size) ||
 	    (cmap->transp && copy_from_user(umap.transp, cmap->transp, size))) {
-		fb_dealloc_cmap(&umap);
-		return -EFAULT;
+		rc = -EFAULT;
+		goto out;
 	}
 	umap.start = cmap->start;
+	info = get_fb_info(info);
+	if (!info) {
+		rc = -ENODEV;
+		goto out;
+	}
+	if (cmap->start < 0 || (!info->fbops->fb_setcolreg &&
+				!info->fbops->fb_setcmap)) {
+		rc = -EINVAL;
+		goto out1;
+	}
 	rc = fb_set_cmap(&umap, info);
+out1:
+	put_fb_info(info);
+out:
 	fb_dealloc_cmap(&umap);
 	return rc;
 }
drivers/video/fbmem.c +93 -65

250 lines not on this page · raw message

include/linux/fb.h +4 -1
index 818fe21..4f02a63 100644
@@ -960,6 +960,9 @@ extern struct fb_info *registered_fb[FB_MAX];
 extern int num_registered_fb;
 extern struct class *fb_class;
 
+struct fb_info *get_fb_info(struct fb_info *info);
+void put_fb_info(struct fb_info *info);
+
 static inline void __fb_pad_aligned_buffer(u8 *dst, u32 d_pitch,
 					   u8 *src, u32 s_pitch, u32 height)
 {
@@ -1068,7 +1071,7 @@ extern void fb_dealloc_cmap(struct fb_cmap *cmap);
 extern int fb_copy_cmap(const struct fb_cmap *from, struct fb_cmap *to);
 extern int fb_cmap_to_user(const struct fb_cmap *from, struct fb_cmap_user *to);
 extern int fb_set_cmap(struct fb_cmap *cmap, struct fb_info *fb_info);
-extern int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *fb_info);
+extern int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *info);
 extern const struct fb_cmap *fb_default_cmap(int len);
 extern void fb_invert_cmaps(void);
 
#4 lore raw

Re: [PATCH] fbmem: fix copy_from/to_user() with mutex held

Hi,

Quoted 6 lines · Andrea Righi

Avoid to call copy_from/to_user() with fb_info->lock mutex held in fbmem fb_ioctl().

NOTE: it doesn't push down the fb_info->lock in each driver's fb_ioctl().

This should probably also include an explanation for WHY doing uaccess under fb_info->lock is verboten. Perhaps even a comment because I don't think it is obvious from the code.

Quoted 88 lines
---
 drivers/video/fbcmap.c |   21 +++++--
 drivers/video/fbmem.c  |  158 ++++++++++++++++++++++++++++--------------------
 include/linux/fb.h     |    5 +-
 3 files changed, 112 insertions(+), 72 deletions(-)
drivers/video/fbcmap.c +15 -6
index 91b78e6..b19f12c 100644
@@ -250,10 +250,6 @@ int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *info)
 	int rc, size = cmap->len * sizeof(u16);
 	struct fb_cmap umap;
 
-	if (cmap->start < 0 || (!info->fbops->fb_setcolreg &&
-			        !info->fbops->fb_setcmap))
-		return -EINVAL;
-
 	memset(&umap, 0, sizeof(struct fb_cmap));
 	rc = fb_alloc_cmap(&umap, cmap->len, cmap->transp != NULL);
 	if (rc)
@@ -262,11 +258,24 @@ int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *info)
 	    copy_from_user(umap.green, cmap->green, size) ||
 	    copy_from_user(umap.blue, cmap->blue, size) ||
 	    (cmap->transp && copy_from_user(umap.transp, cmap->transp, size))) {
-		fb_dealloc_cmap(&umap);
-		return -EFAULT;
+		rc = -EFAULT;
+		goto out;
 	}
 	umap.start = cmap->start;
+	info = get_fb_info(info);
+	if (!info) {
+		rc = -ENODEV;
+		goto out;
+	}
+	if (cmap->start < 0 || (!info->fbops->fb_setcolreg &&
+				!info->fbops->fb_setcmap)) {
+		rc = -EINVAL;
+		goto out1;
+	}
 	rc = fb_set_cmap(&umap, info);
+out1:
+	put_fb_info(info);
+out:
 	fb_dealloc_cmap(&umap);
 	return rc;
 }
drivers/video/fbmem.c +18 -2
index 756efeb..e51e9ba 100644
@@ -1006,6 +1006,23 @@ fb_blank(struct fb_info *info, int blank)
  	return ret;
 }
 
+struct fb_info *get_fb_info(struct fb_info *info)
+__acquires(&info->lock)
+{
+	mutex_lock(&info->lock);
+	if (!info->fbops) {
+		mutex_unlock(&info->lock);
+		return NULL;
+	}
+	return info;
+}
+
+void put_fb_info(struct fb_info *info)
+__releases(&info->lock)
+{
+	mutex_unlock(&info->lock);
+}
+
 static long do_fb_ioctl(struct fb_info *info, unsigned int cmd,
 			unsigned long arg)
 {
@@ -1013,25 +1030,28 @@ static long do_fb_ioctl(struct fb_info *info, unsigned int cmd,
 	struct fb_var_screeninfo var;
 	struct fb_fix_screeninfo fix;
 	struct fb_con2fbmap con2fb;
+	struct fb_cmap cmap_from;
 	struct fb_cmap_user cmap;
 	struct fb_event event;
 	void __user *argp = (void __user *)arg;
 	long ret = 0;
 
-	fb = info->fbops;
-	if (!fb)

drivers/video/fbmem.c static long do_fb_ioctl(struct fb_info *info, unsigned int cmd, · line 1044

Quoted 10 lines
drivers/video/fbmem.c +4 -4
-		return -ENODEV;
-
 	switch (cmd) {
 	case FBIOGET_VSCREENINFO:
-		ret = copy_to_user(argp, &info->var,
-				    sizeof(var)) ? -EFAULT : 0;
+		info = get_fb_info(info);
+		if (!info)
+			return -ENODEV;
+		memcpy(&var, &info->var, sizeof(var));

You don't need these memcpy()s:

		var = info->var

does the same much more readable.

Thanks,
hannes

#5 lore raw

Re: [PATCH] fbmem: fix copy_from/to_user() with mutex held

In reply to Andrea Righi
Quoted 7 lines · Andrea Righi
drivers/video/fbmem.c +7 -0
+struct fb_info *get_fb_info(struct fb_info *info)
+__acquires(&info->lock)
+{
+	mutex_lock(&info->lock);
+	if (!info->fbops) {
+		mutex_unlock(&info->lock);
+		return NULL;

drivers/video/fbmem.c fb_blank(struct fb_info *info, int blank) · line 1025

Quoted 10 lines
drivers/video/fbmem.c +10 -0
+	}
+	return info;
+}
+
+void put_fb_info(struct fb_info *info)
+__releases(&info->lock)
+{
+	mutex_unlock(&info->lock);
+}
+

These are IMO bad function names. get_... and put_... sound too much like functions that either read or write data, or increment and decrement the reference count of data. These names don't reflect that the sole purpose of these functions is to lock and unlock a mutex (i.e. to serialize some critical sections).

*If* you really need to hide mutex_(un)lock(&info->lock) inside wrappers, then the names of these wrappers should clearly state what they do, so that everybody knows immediately what is going on at the call sites.

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

Re: [PATCH] fbmem: fix copy_from/to_user() with mutex held

Quoted 12 lines · Stefan Richter
Quoted 11 lines · Andrea Righi
drivers/video/fbmem.c +11 -0
+struct fb_info *get_fb_info(struct fb_info *info)
+__acquires(&info->lock)
+{
+	mutex_lock(&info->lock);
+	if (!info->fbops) {
+		mutex_unlock(&info->lock);
+		return NULL;
+	}
+	return info;
+}
+
Quoted 8 lines

drivers/video/fbmem.c fb_blank(struct fb_info *info, int blank) · line 1025

Quoted 6 lines
drivers/video/fbmem.c +6 -0
+void put_fb_info(struct fb_info *info)
+__releases(&info->lock)
+{
+	mutex_unlock(&info->lock);
+}
+

These are IMO bad function names.

PS: The return value of the mutex_lock wrapper is also bad. A bool or int would IMO be clearer, similar to the return value of mutex_trylock.

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

Re: [PATCH] fbmem: fix copy_from/to_user() with mutex held

Quoted 22 lines · Stefan Richter
Quoted 12 lines · Stefan Richter
Quoted 11 lines · Andrea Righi
drivers/video/fbmem.c +11 -0
+struct fb_info *get_fb_info(struct fb_info *info)
+__acquires(&info->lock)
+{
+	mutex_lock(&info->lock);
+	if (!info->fbops) {
+		mutex_unlock(&info->lock);
+		return NULL;
+	}
+	return info;
+}
+
Quoted 8 lines

drivers/video/fbmem.c fb_blank(struct fb_info *info, int blank) · line 1025

Quoted 6 lines
drivers/video/fbmem.c +6 -0
+void put_fb_info(struct fb_info *info)
+__releases(&info->lock)
+{
+	mutex_unlock(&info->lock);
+}
+

These are IMO bad function names.

Quoted 2 lines

PS: The return value of the mutex_lock wrapper is also bad. A bool or int would IMO be clearer, similar to the return value of mutex_trylock.

That, and there is no possible way to get the sparse annotations right for that function, which means you'll get no help from sparse in lock checking.

So I'd suggest just opencoding these where needed instead of the wrappers.

Harvey

#8 lore raw

Re: [PATCH] fbmem: fix copy_from/to_user() with mutex held

Quoted 26 lines · Harvey Harrison
Quoted 22 lines · Stefan Richter
Quoted 12 lines · Stefan Richter
Quoted 11 lines · Andrea Righi
drivers/video/fbmem.c +11 -0
+struct fb_info *get_fb_info(struct fb_info *info)
+__acquires(&info->lock)
+{
+  mutex_lock(&info->lock);
+  if (!info->fbops) {
+          mutex_unlock(&info->lock);
+          return NULL;
+  }
+  return info;
+}
+
Quoted 8 lines

drivers/video/fbmem.c fb_blank(struct fb_info *info, int blank) · line 1025

Quoted 6 lines
drivers/video/fbmem.c +6 -0
+void put_fb_info(struct fb_info *info)
+__releases(&info->lock)
+{
+  mutex_unlock(&info->lock);
+}
+

These are IMO bad function names.

Quoted 2 lines

PS: The return value of the mutex_lock wrapper is also bad. A bool or int would IMO be clearer, similar to the return value of mutex_trylock.

Quoted 9 lines

That, and there is no possible way to get the sparse annotations right for that function, which means you'll get no help from sparse in lock checking.

So I'd suggest just opencoding these where needed instead of the wrappers.

Harvey

OK, thanks for all the suggestions. I'll fix everything and repost a new patch ASAP.

-Andrea