Skip to contentSkip to conversation

lockdep splat from fbmem

DISCUSSION Johannes Weiner 3 participants 7 messages Oct 23, 2008 – Oct 28, 2008 Framebuffer layer

lore
Share & tools
lore
#0 lore raw

Hi,

fb_mmap() is called under mmap_sem and acquires fb_info->lock.

Since 3e680aae4e53ab "fb: convert lock/unlock_kernel() into local fb mutex", fb_ioctl() takes fb_info->lock and does copy_to/from_user() under it, which in turn might acquire mmap_sem.

Just noticed the trace and wanted to let you know.

Hannes

#1 lore raw
Quoted 11 lines · Johannes Weiner

Hi,

fb_mmap() is called under mmap_sem and acquires fb_info->lock.

Since 3e680aae4e53ab "fb: convert lock/unlock_kernel() into local fb mutex", fb_ioctl() takes fb_info->lock and does copy_to/from_user() under it, which in turn might acquire mmap_sem.

Just noticed the trace and wanted to let you know.

Hannes

Good catch. We could try to push down the mutex and call all the copy_to/from_user() outside the lock. I can try look at it tonight, if someone doesn't fix it before.

Thanks for reporting,
-Andrea

#2 lore raw

lockdep splat from ioctl and mmap fops sharing lock

In reply to Johannes Weiner
Quoted 7 lines · Johannes Weiner

Hi,

fb_mmap() is called under mmap_sem and acquires fb_info->lock.

Since 3e680aae4e53ab "fb: convert lock/unlock_kernel() into local fb mutex", fb_ioctl() takes fb_info->lock and does copy_to/from_user() under it, which in turn might acquire mmap_sem.

More affected drivers:

agp_ioctl() doing usercopy under agp_fe.agp_mutex agp_mmap() taking agp_fe.agp_mutex under mmap_sem

dv1394_write() doing usercopy under video->mtx dv1394_mmap() taking video->mtx under mmap_sem

raw1394_ioctl() doing usercopy under fi->state_mutex raw1394_mmap() taking fi->state_mutex under mmap_sem

Hannes

#3 lore raw

Re: lockdep splat from ioctl and mmap fops sharing lock

Quoted 10 lines · Johannes Weiner
Quoted 7 lines · Johannes Weiner

Hi,

fb_mmap() is called under mmap_sem and acquires fb_info->lock.

Since 3e680aae4e53ab "fb: convert lock/unlock_kernel() into local fb mutex", fb_ioctl() takes fb_info->lock and does copy_to/from_user() under it, which in turn might acquire mmap_sem.

Quoted 7 lines

More affected drivers:

agp_ioctl() doing usercopy under agp_fe.agp_mutex agp_mmap() taking agp_fe.agp_mutex under mmap_sem

dv1394_write() doing usercopy under video->mtx dv1394_mmap() taking video->mtx under mmap_sem

OK, it doesn't surprise me that I didn't notice this one myself: The video->mtx usage is old, but dv1394 is not frequently used anymore; I for one don't test anything of it beyond loading and unloading.

Quoted 2 lines

raw1394_ioctl() doing usercopy under fi->state_mutex raw1394_mmap() taking fi->state_mutex under mmap_sem

The state_mutex in raw1394 however was introduced by me in patches written against 2.6.26 and 2.6.27-rcs. And I tested the ioctls and I'd like to think that I also tested mmaps. But maybe I didn't. Of course I ahve all sorts of lockdep options enabled.

So, was the usage of mmap_sem changed after 2.6.27 or were my tests insufficient?

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

Re: lockdep splat from ioctl and mmap fops sharing lock

Quoted 11 lines · Stefan Richter
Quoted 2 lines · Johannes Weiner

raw1394_ioctl() doing usercopy under fi->state_mutex raw1394_mmap() taking fi->state_mutex under mmap_sem

The state_mutex in raw1394 however was introduced by me in patches written against 2.6.26 and 2.6.27-rcs. And I tested the ioctls and I'd like to think that I also tested mmaps. But maybe I didn't. Of course I ahve all sorts of lockdep options enabled.

So, was the usage of mmap_sem changed after 2.6.27 or were my tests insufficient?

In linux-next/-mm, copy_to/from_user have lockdep annotations telling that they might fault and therefor acquire the mmap_sem in #PF.

But since faults on ioctl parameters are so rare, without these annotations you would probably never see a warning.

Hannes

#5 lore raw

Re: lockdep splat from ioctl and mmap fops sharing lock

Quoted 5 lines · Johannes Weiner
Quoted 3 lines · Stefan Richter
Quoted 2 lines · Johannes Weiner

raw1394_ioctl() doing usercopy under fi->state_mutex raw1394_mmap() taking fi->state_mutex under mmap_sem

...

Quoted 8 lines
Quoted 2 lines

was the usage of mmap_sem changed after 2.6.27 or were my tests insufficient?

In linux-next/-mm, copy_to/from_user have lockdep annotations telling that they might fault and therefor acquire the mmap_sem in #PF.

But since faults on ioctl parameters are so rare, without these annotations you would probably never see a warning.

Thanks for the report and explanation. I logged this as http://bugzilla.kernel.org/show_bug.cgi?id=11823 (dv1394, won't fix) and http://bugzilla.kernel.org/show_bug.cgi?id=11824 (raw1394, to be fixed in the 2.6.28-rc timeframe since it is a post 2.6.27 regression).

Signature · 3 lines
Stefan Richter
-=====-==--- =-=- ==--=
http://arcgraph.de/sr/
#6 lore raw
In reply to Johannes Weiner
Quoted 11 lines · Johannes Weiner

Hi,

fb_mmap() is called under mmap_sem and acquires fb_info->lock.

Since 3e680aae4e53ab "fb: convert lock/unlock_kernel() into local fb mutex", fb_ioctl() takes fb_info->lock and does copy_to/from_user() under it, which in turn might acquire mmap_sem.

Just noticed the trace and wanted to let you know.

Hannes

Johannes,

could you check the following fix for fbmem? maybe some parts are still deadlockable, but I'm not able to raise any lockdep trace with it using the latest Linus' git.

Thanks,
-Andrea

---
fb: avoid to call copy_from/to_user() with fb_info mutex held in fbmem ioctl()

---
 drivers/video/fbcmap.c |   24 ++++++--
 drivers/video/fbmem.c  |  150 ++++++++++++++++++++++++++++++------------------
 include/linux/fb.h     |    5 +-
 3 files changed, 114 insertions(+), 65 deletions(-)
drivers/video/fbcmap.c +17 -7
index 91b78e6..8f46589 100644
@@ -245,14 +245,11 @@ int fb_set_cmap(struct fb_cmap *cmap, struct fb_info *info)
 	return rc;
 }
 
-int fb_set_user_cmap(struct fb_cmap_user *cmap, struct fb_info *info)
+int fb_set_user_cmap(struct fb_cmap_user *cmap, int fbidx)
 {
 	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;
+	struct fb_info *info;
 
 	memset(&umap, 0, sizeof(struct fb_cmap));
 	rc = fb_alloc_cmap(&umap, cmap->len, cmap->transp != NULL);
@@ -262,11 +259,24 @@ 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(fbidx);
+	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 -57

219 lines not on this page · raw message

include/linux/fb.h +4 -1
index 75a81ea..9f4e70f 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(int fbidx);
+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, int fbidx);
 extern const struct fb_cmap *fb_default_cmap(int len);
 extern void fb_invert_cmaps(void);