Skip to contentSkip to conversation

'sleeping function called from invalid context' bug when mounting an IDE device

PATCH Vitaly Wool 3 participants 4 messages Jul 8, 2005

lore
Share & tools
lore
#0 lore raw

Hi Ingo,

I've come across the following problem during the debugging of IDE driver for Philips PNX0105 ARM9 platform in RT mode (CONFIG_PREEMPT_RT).
When I mount/unmount a device, the following error is printed out to a terminal:

Quoted 3 lines
# mount /dev/discs/disc0/part1 /mnt/
BUG: sleeping function called from invalid context IRQ 27(655) at
kernel/rt.c:1449in_atomic():0 [00000000], irqs_disabled():128

And gdb shows the following:

Quoted 19 lines

(gdb) b kernel/sched.c:5430
Breakpoint 1 at 0xc0034880: file kernel/sched.c, line 5430. (gdb) c
Continuing.

Breakpoint 1, __might_sleep (file=0x56 <Address 0x56 out of bounds>, line=1449)

	 at kernel/sched.c:5434

5434 in_atomic(), preempt_count(),
irqs_disabled());
(gdb) bt
#0 __might_sleep (file=0x56 <Address 0x56 out of bounds>, line=1449)

	 at kernel/sched.c:5434

#1 0xc004f700 in __spin_lock (lock=0xc01ee10c, eip=3222413300) at kernel/rt.c:1449
#2 0xc004f778 in _spin_lock_irqsave (spin=0x56) at kernel/rt.c:1474 #3 0xc0121ff4 in ide_intr (irq=27, dev_id=0x0, regs=0x0) at drivers/ide/ide-io.c:1416
#4 0xc00201c0 in __do_irq (irq=27, action=0xc03f8df4, regs=0x0)

Quoted 9 lines
	 at arch/arm/kernel/irq.c:530

#5 0xc001fec0 in do_hardirq (desc=0xc021f2bc) at arch/arm/kernel/irq.c:394
#6 0xc0020008 in do_irqd (__desc=0xc021f2bc) at arch/arm/kernel/irq.c:447
#7 0xc004c310 in kthread (_create=0xc02abe20) at kernel/kthread.c:95 #8 0xc0039af4 in exit_notify (tsk=0x0) at kernel/exit.c:785 Previous frame inner to this frame (corrupt stack?)

So, the problem is in the generic IDE code, namely, in ide_intr() taking ide_lock.

The suggested patch is the following:

drivers/ide/ide-probe.c +1 -4
Index: linux/drivers/ide/ide-probe.c
===================================================================
@@ -1050,16 +1050,13 @@
 	 * Allocate the irq, if not already obtained for another hwif
 	 */
 	if (!match || match->irq != hwif->irq) {
-		int sa = SA_INTERRUPT;
+		int sa = 0;
 #if defined(__mc68000__) || defined(CONFIG_APUS)
 		sa = SA_SHIRQ;
 #endif /* __mc68000__ || CONFIG_APUS */
 
 		if (IDE_CHIPSET_IS_PCI(hwif->chipset)) {
 			sa = SA_SHIRQ;
-#ifndef CONFIG_IDEPCI_SHARE_IRQ
-			sa |= SA_INTERRUPT;
-#endif /* CONFIG_IDEPCI_SHARE_IRQ */
 		}
 
 		if (hwif->io_ports[IDE_CONTROL_OFFSET])

Can you please comment on it?

Thanks,
Vitaly

#1 lore raw
Quoted 6 lines · Vitaly Wool

Hi Ingo,

I've come across the following problem during the debugging of IDE driver for Philips PNX0105 ARM9 platform in RT mode (CONFIG_PREEMPT_RT). When I mount/unmount a device, the following error is printed out to a terminal:

could you send me the full backtrace?

Quoted 2 lines

So, the problem is in the generic IDE code, namely, in ide_intr() taking ide_lock.

which version did you try, and does this happen with the latest patch too?

Ingo

#2 lore raw
Quoted 9 lines · Ingo Molnar
Quoted 6 lines · Vitaly Wool

Hi Ingo,

I've come across the following problem during the debugging of IDE driver for Philips PNX0105 ARM9 platform in RT mode (CONFIG_PREEMPT_RT). When I mount/unmount a device, the following error is printed out to a terminal:

Quoted 7 lines

could you send me the full backtrace?

Quoted 2 lines

So, the problem is in the generic IDE code, namely, in ide_intr() taking ide_lock.

which version did you try, and does this happen with the latest patch too?

Interrupts should be enabled unconditionally for threaded interrupt handlers. Or at least the generics work that way.

Daniel

#3 lore raw
Quoted 8 lines · Daniel Walker
Quoted 5 lines
Quoted 2 lines

So, the problem is in the generic IDE code, namely, in ide_intr() taking ide_lock.

which version did you try, and does this happen with the latest patch too?

Interrupts should be enabled unconditionally for threaded interrupt handlers. Or at least the generics work that way.

yes, that's why i'm asking about the version and the full backtrace. Threaded irq handlers run with irqs enabled:

        /*
         * Unconditionally enable interrupts for threaded
         * IRQ handlers:
         */
        if (!hardirq_count() || !(action->flags & SA_INTERRUPT))
                raw_local_irq_enable();

    Ingo