Skip to contentSkip to conversation

io_apic fix spinlock in resume [Was: Re: swsusp regression (s2dsk) [Was: 2.6.18-rc2-mm1]]

PATCH In review Jiri Slaby 2 participants 2 messages Jul 31, 2006

lore
Share & tools
lore

Jiri Slaby

author
#0 lore raw

PATCH 1/1 io_apic fix spinlock in resume [Was: Re: swsusp regression (s2dsk) [Was: 2.6.18-rc2-mm1]]

Quoted 32 lines · Jiri Slaby

Rafael J. Wysocki napsal(a):

Quoted 23 lines
Quoted 23 lines · Pavel Machek
Quoted 22 lines
Quoted 22 lines

Hi!

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

I have problems with swsusp again. While suspending, the very
last thing kernel
writes is 'restoring higmem' and then hangs, hardly. No sysrq
response at all.
Here is a snapshot of the screen:
http://www.fi.muni.cz/~xslaby/sklad/swsusp_higmem.gif

It's SMP system (HT), higmem enabled (1 gig of ram).
Most probably it hangs in device_power_up(), so the problem
seems to be
with one of the devices that are resumed with IRQs off.

Quoted 6 lines
Quoted 6 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines
Quoted 2 lines
Quoted 1 line
Quoted 1 line

Does vanila .18-rc2 work?

Yup, it does.

Can you try up kernel, no highmem? (mem=512M)?

It writes then:
p16v: status 0xffffffff, mask 0x00001000, pvoice f7c04a20, use 0 in endless loop when resuming -- after reading from swap.

Okay, so we have two different problems here.

Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Quoted 1 line

One is "hang during suspend" with smp/highmem mode,

That one is "interesting". I've no idea why the restoration of highmem would
have caused the box to hang like that. Jiri, could you please post the output
of dmesg after a fresh boot?

Quoted 4 lines

higmem is ok. ioapic0 is the culprit -- its class resume dies:

        if (cls->resume)
                cls->resume(dev); <----

in __sysdev_resume

io_apic fix spinlock in resume

In io_apic class resume after Andi's cleanup was wiped out one unlock of spinlock. Get him back to allow suspending (and resuming) of machine.

---
commit 2d82ba5b564500cb29613ee9c24b1d0efa829518
tree 64551b2f45e6ffd3f220af85f25dae2199897652
parent 8e013921e94b248b33880b58b46e5eba4a931b51
author Jiri Slaby <[email protected]> Tue, 01 Aug 2006 01:16:13 +0159 committer Jiri Slaby <[email protected]> Tue, 01 Aug 2006 01:16:13 +0159

 arch/i386/kernel/io_apic.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/i386/kernel/io_apic.c +1 -0
index fa0eb9f..617037a 100644
@@ -2360,6 +2360,7 @@ static int ioapic_resume(struct sys_devi
 		reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
 		io_apic_write(dev->id, 0, reg_00.raw);
 	}
+	spin_unlock_irqrestore(&ioapic_lock, flags);
 	for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
 		ioapic_write_entry(dev->id, i, entry[i]);
 
#1 lore raw
Quoted 5 lines

io_apic fix spinlock in resume

In io_apic class resume after Andi's cleanup was wiped out one unlock of spinlock. Get him back to allow suspending (and resuming) of machine.

Oops. I fixed the original patch. Thanks.

-Andi