[PATCH] lockdep: fix ide/proc interaction
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Thu, 7 Dec 2006 04:36:13 +0000 (20:36 -0800)
committerLinus Torvalds <torvalds@woody.osdl.org>
Thu, 7 Dec 2006 16:39:32 +0000 (08:39 -0800)
  rmmod/3080 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
   (proc_subdir_lock){--..}, at: [<c04a33b0>] remove_proc_entry+0x40/0x191

  and this task is already holding:
   (ide_lock){++..}, at: [<c05651a2>] ide_unregister_subdriver+0x39/0xc8
  which would create a new lock dependency:
   (ide_lock){++..} -> (proc_subdir_lock){--..}

  but this new dependency connects a hard-irq-safe lock:
   (ide_lock){++..}
  ... which became hard-irq-safe at:
    [<c043c458>] lock_acquire+0x4b/0x6b
    [<c06129d7>] _spin_lock_irqsave+0x22/0x32
    [<c0567870>] ide_intr+0x17/0x1a9
    [<c044eb31>] handle_IRQ_event+0x20/0x4d
    [<c044ebf2>] __do_IRQ+0x94/0xef
    [<c0406771>] do_IRQ+0x9e/0xbd

  to a hard-irq-unsafe lock:
   (proc_subdir_lock){--..}
  ... which became hard-irq-unsafe at:
  ...  [<c043c458>] lock_acquire+0x4b/0x6b
    [<c06126ab>] _spin_lock+0x19/0x28
    [<c04a32f2>] xlate_proc_name+0x1b/0x99
    [<c04a3547>] proc_create+0x46/0xdf
    [<c04a3642>] create_proc_entry+0x62/0xa5
    [<c07c1972>] proc_misc_init+0x1c/0x1d2
    [<c07c1844>] proc_root_init+0x4c/0xe9
    [<c07ad703>] start_kernel+0x294/0x3b3

Move ide_remove_proc_entries() out from under ide_lock; there is nothing
that indicates that this is needed.

In specific, the call to ide_add_proc_entries() is unprotected, and there
is nothing else in the file using the respective ->proc fields. Also the
lock order around destroy_proc_ide_interface() suggests this.

Alan sayeth:

  proc_ide_write_settings walks the setting list under ide_setting_sem, read
  ditto.  remove_proc_entry is doing proc side housekeeping.

  Looks fine to me, although that old code is such a mess anything could be
  going on.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/ide/ide.c

index 287a66201150a0bd32432817a4357705b767329c..16890769dca686dff62ec62bc0c2f741b7ef190c 100644 (file)
@@ -973,8 +973,8 @@ ide_settings_t *ide_find_setting_by_name (ide_drive_t *drive, char *name)
  *     @drive: drive
  *
  *     Automatically remove all the driver specific settings for this
- *     drive. This function may sleep and must not be called from IRQ
- *     context. The caller must hold ide_setting_sem.
+ *     drive. This function may not be called from IRQ context. The
+ *     caller must hold ide_setting_sem.
  */
  
 static void auto_remove_settings (ide_drive_t *drive)
@@ -1874,11 +1874,22 @@ void ide_unregister_subdriver(ide_drive_t *drive, ide_driver_t *driver)
 {
        unsigned long flags;
        
-       down(&ide_setting_sem);
-       spin_lock_irqsave(&ide_lock, flags);
 #ifdef CONFIG_PROC_FS
        ide_remove_proc_entries(drive->proc, driver->proc);
 #endif
+       down(&ide_setting_sem);
+       spin_lock_irqsave(&ide_lock, flags);
+       /*
+        * ide_setting_sem protects the settings list
+        * ide_lock protects the use of settings
+        *
+        * so we need to hold both, ide_settings_sem because we want to
+        * modify the settings list, and ide_lock because we cannot take
+        * a setting out that is being used.
+        *
+        * OTOH both ide_{read,write}_setting are only ever used under
+        * ide_setting_sem.
+        */
        auto_remove_settings(drive);
        spin_unlock_irqrestore(&ide_lock, flags);
        up(&ide_setting_sem);