illegal context for sleeping ... rmmod ide-cd + ide-scsi

illegal context for sleeping ... rmmod ide-cd + ide-scsi

[Douglas Gilbert]

In lk 2.5.69 (and in 68) both the ide-cd and ide-scsi
modules generate a "sleeping function called from illegal
context" stack trace when removed.

After "rmmod ide-cd" this appears:
  Debug: sleeping function called from illegal context
         at include/asm/semaphore.h:119
  Call Trace:
   [<c011dcec>] __might_sleep+0x5c/0x70
   [<c025b634>] auto_remove_settings+0x24/0x80
   [<c025db8a>] ide_unregister_subdriver+0x15a/0x3d0
   [<e10d48df>] ide_cdrom_cleanup+0x2f/0x110 [ide_cd]
   [<c02346de>] driver_unregister+0x2e/0x42
   [<e10d90e0>] ide_cdrom_driver+0x0/0xd8 [ide_cd]
   [<e10d91a8>] ide_cdrom_driver+0xc8/0xd8 [ide_cd]
   [<c025e270>] ide_unregister_driver+0x100/0x1df
   [<c0155f1f>] unmap_vma_list+0x1f/0x30
   [<e10d9200>] +0x0/0x140 [ide_cd]
   [<e10d4df2>] +0x12/0x20 [ide_cd]
   [<e10d90e0>] ide_cdrom_driver+0x0/0xd8 [ide_cd]
   [<c013d336>] sys_delete_module+0x1d6/0x240
   [<c0156457>] sys_munmap+0x57/0x80
   [<c0109eaf>] syscall_call+0x7/0xb

After "rmmod ide-scsi" this appears:
  Debug: sleeping function called from illegal context
        at include/asm/semaphore.h:119
  Call Trace:
   [<e10c8068>] idescsi_driver+0xc8/0xd8 [ide_scsi]
   [<c011dcec>] __might_sleep+0x5c/0x70
   [<e10c8068>] idescsi_driver+0xc8/0xd8 [ide_scsi]
   [<c025b634>] auto_remove_settings+0x24/0x80
   [<c025db8a>] ide_unregister_subdriver+0x15a/0x3d0
   [<e10c7ff8>] idescsi_driver+0x58/0xd8 [ide_scsi]
   [<e10c7ff8>] idescsi_driver+0x58/0xd8 [ide_scsi]
   [<e10c542d>] idescsi_cleanup+0x1d/0x60 [ide_scsi]
   [<e10c7fa0>] idescsi_driver+0x0/0xd8 [ide_scsi]
   [<c025e270>] ide_unregister_driver+0x100/0x1df
   [<e10c6e87>] +0xfa/0x10f [ide_scsi]
   [<e10c8300>] +0x0/0x140 [ide_scsi]
   [<e10c663f>] +0x2f/0x50 [ide_scsi]
   [<e10c7fa0>] idescsi_driver+0x0/0xd8 [ide_scsi]
   [<c013d336>] sys_delete_module+0x1d6/0x240
   [<c0156457>] sys_munmap+0x57/0x80
   [<c0109eaf>] syscall_call+0x7/0xb

My .config has these kernel debugging options set:
#
# Kernel hacking
#
CONFIG_DEBUG_KERNEL=y
CONFIG_DEBUG_STACKOVERFLOW=y
CONFIG_DEBUG_SLAB=y
CONFIG_DEBUG_IOVIRT=y
CONFIG_MAGIC_SYSRQ=y
CONFIG_DEBUG_SPINLOCK=y
CONFIG_KALLSYMS=y
CONFIG_DEBUG_SPINLOCK_SLEEP=y
CONFIG_FRAME_POINTER=y

Doug Gilbert

Re: illegal context for sleeping ... rmmod ide-cd + ide-scsi

[Andrew Morton]

Douglas Gilbert <dougg@torque.net> wrote:
>
> In lk 2.5.69 (and in 68) both the ide-cd and ide-scsi
> modules generate a "sleeping function called from illegal
> context" stack trace when removed.
> 
> After "rmmod ide-cd" this appears:
>   Debug: sleeping function called from illegal context
>          at include/asm/semaphore.h:119
>   Call Trace:
>    [<c011dcec>] __might_sleep+0x5c/0x70

ide_unregister_subdriver() does spin_lock_irqsave(&ide_lock), then
calls auto_remove_settings(), which does down(&ide_setting_sem);

A simple fix might be:

diff -puN drivers/ide/ide.c~ide_setting_sem-fix drivers/ide/ide.c
--- 25/drivers/ide/ide.c~ide_setting_sem-fix	2003-05-05 01:59:01.000000000 -0700
+++ 25-akpm/drivers/ide/ide.c	2003-05-05 02:00:21.000000000 -0700
@@ -1131,13 +1131,12 @@ ide_settings_t *ide_find_setting_by_name
  *
  *	Automatically remove all the driver specific settings for this
  *	drive. This function may sleep and must not be called from IRQ
- *	context. Takes the settings_lock
+ *	context. The caller must hold ide_setting_sem.
  */
  
 static void auto_remove_settings (ide_drive_t *drive)
 {
 	ide_settings_t *setting;
-	down(&ide_setting_sem);
 repeat:
 	setting = drive->settings;
 	while (setting) {
@@ -1147,7 +1146,6 @@ repeat:
 		}
 		setting = setting->next;
 	}
-	up(&ide_setting_sem);
 }
 
 /**
@@ -2350,9 +2348,11 @@ int ide_unregister_subdriver (ide_drive_
 {
 	unsigned long flags;
 	
+	down(&ide_setting_sem);
 	spin_lock_irqsave(&ide_lock, flags);
 	if (drive->usage || drive->driver == &idedefault_driver || DRIVER(drive)->busy) {
 		spin_unlock_irqrestore(&ide_lock, flags);
+		up(&ide_setting_sem);
 		return 1;
 	}
 #if defined(CONFIG_BLK_DEV_IDEPNP) && defined(CONFIG_PNP) && defined(MODULE)
@@ -2363,6 +2363,7 @@ int ide_unregister_subdriver (ide_drive_
 	ide_remove_proc_entries(drive->proc, generic_subdriver_entries);
 #endif
 	auto_remove_settings(drive);
+	up(&ide_setting_sem);
 	drive->driver = &idedefault_driver;
 	setup_driver_defaults(drive);
 	spin_unlock_irqrestore(&ide_lock, flags);

_

Re: illegal context for sleeping ... rmmod ide-cd + ide-scsi

[Alan Cox]

On Llu, 2003-05-05 at 12:51, Douglas Gilbert wrote:
> > ide_unregister_subdriver() does spin_lock_irqsave(&ide_lock), then
> > calls auto_remove_settings(), which does down(&ide_setting_sem);
> > 
> > A simple fix might be:
> 
> Andrew,
> Thanks. That patch clears the reported problem.

This is already fixed in 2.4.x btw. Just hadn't got pushed into 2.5 yet

The 2.5.x code has another problem as well there is a basically unfixable
deadlock in the proc and config stuff when flipping a device in and out of
scsi mode

Re: illegal context for sleeping ... rmmod ide-cd + ide-scsi

[Douglas Gilbert]

Andrew Morton wrote:
> Douglas Gilbert <dougg@torque.net> wrote:
> 
>>In lk 2.5.69 (and in 68) both the ide-cd and ide-scsi
>>modules generate a "sleeping function called from illegal
>>context" stack trace when removed.
>>
>>After "rmmod ide-cd" this appears:
>>  Debug: sleeping function called from illegal context
>>         at include/asm/semaphore.h:119
>>  Call Trace:
>>   [<c011dcec>] __might_sleep+0x5c/0x70
> 
> 
> ide_unregister_subdriver() does spin_lock_irqsave(&ide_lock), then
> calls auto_remove_settings(), which does down(&ide_setting_sem);
> 
> A simple fix might be:

Andrew,
Thanks. That patch clears the reported problem.

Doug Gilbert