Current Linus's tree has the following lockdep information. I've bisected it, but there are several unbisectable commits between "good" and "bad". And my machine is too slow to continue bisect. But by this unfinished git-bisect, I've known the latest good is "0191b625ca5a46206d2fb862bb08f36f2fcb3b31" and the most early bad is "5f34fe1cfc1bdd8b4711bbe37421fba4ed0d1ed4" dmesg is bellow and .config is attached. ======================================================= [ INFO: possible circular locking dependency detected ] 2.6.28-default #147 ------------------------------------------------------- X/4226 is trying to acquire lock: (&mm->mmap_sem){----}, at: [] might_fault+0x32/0x6d but task is already holding lock: (&dev->struct_mutex){--..}, at: [] i915_cmdbuffer+0xfb/0x441 [i915] which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (&dev->struct_mutex){--..}: [] add_lock_to_list+0x5e/0x8b [] validate_chain+0x8ee/0xbfb [] drm_vm_open+0x22/0x32 [drm] [] drm_vm_open+0x22/0x32 [drm] [] __lock_acquire+0x674/0x700 [] lock_acquire+0x58/0x71 [] drm_vm_open+0x22/0x32 [drm] [] mutex_lock_nested+0xdb/0x247 [] drm_vm_open+0x22/0x32 [drm] [] drm_vm_open+0x22/0x32 [drm] [] dup_mm+0x247/0x2ec [] copy_process+0x93b/0xf42 [] do_fork+0x115/0x29b [] might_fault+0x32/0x6d [] copy_to_user+0x25/0x3c [] sysenter_exit+0xf/0x16 [] sys_clone+0x28/0x2d [] sysenter_do_call+0x12/0x31 [] 0xffffffff -> #1 (&mm->mmap_sem/1){--..}: [] add_lock_to_list+0x5e/0x8b [] validate_chain+0x8ee/0xbfb [] dup_mm+0xa1/0x2ec [] dup_mm+0xa1/0x2ec [] __lock_acquire+0x674/0x700 [] lock_acquire+0x58/0x71 [] dup_mm+0xa1/0x2ec [] down_write_nested+0x30/0x49 [] dup_mm+0xa1/0x2ec [] dup_mm+0xa1/0x2ec [] trace_hardirqs_on_caller+0x10b/0x131 [] copy_process+0x93b/0xf42 [] do_fork+0x115/0x29b [] might_fault+0x32/0x6d [] copy_to_user+0x25/0x3c [] sysenter_exit+0xf/0x16 [] sys_clone+0x28/0x2d [] sysenter_do_call+0x12/0x31 [] 0xffffffff -> #0 (&mm->mmap_sem){----}: [] validate_chain+0x5e1/0xbfb [] __lock_acquire+0x674/0x700 [] lock_acquire+0x58/0x71 [] might_fault+0x32/0x6d [] might_fault+0x4f/0x6d [] might_fault+0x32/0x6d [] i915_emit_box+0x18/0x1fe [i915] [] i915_cmdbuffer+0x179/0x441 [i915] [] drm_ioctl+0x19f/0x211 [drm] [] i915_cmdbuffer+0x0/0x441 [i915] [] drm_ioctl+0x0/0x211 [drm] [] vfs_ioctl+0x3a/0x49 [] do_vfs_ioctl+0x3b6/0x409 [] _spin_unlock+0x14/0x1c [] handle_mm_fault+0x55e/0x571 [] down_read_trylock+0x36/0x3d [] sys_ioctl+0x41/0x58 [] sysenter_do_call+0x12/0x31 [] 0xffffffff other info that might help us debug this: 1 lock held by X/4226: #0: (&dev->struct_mutex){--..}, at: [] i915_cmdbuffer+0xfb/0x441 [i915] stack backtrace: Pid: 4226, comm: X Not tainted 2.6.28-default #147 Call Trace: [] print_circular_bug_tail+0xa6/0xb0 [] validate_chain+0x5e1/0xbfb [] __lock_acquire+0x674/0x700 [] lock_acquire+0x58/0x71 [] might_fault+0x32/0x6d [] might_fault+0x4f/0x6d [] might_fault+0x32/0x6d [] i915_emit_box+0x18/0x1fe [i915] [] i915_cmdbuffer+0x179/0x441 [i915] [] drm_ioctl+0x19f/0x211 [drm] [] i915_cmdbuffer+0x0/0x441 [i915] [] drm_ioctl+0x0/0x211 [drm] [] vfs_ioctl+0x3a/0x49 [] do_vfs_ioctl+0x3b6/0x409 [] _spin_unlock+0x14/0x1c [] handle_mm_fault+0x55e/0x571 [] down_read_trylock+0x36/0x3d [] sys_ioctl+0x41/0x58 [] sysenter_do_call+0x12/0x31