On Sat, Aug 23, 2008 at 06:01:51PM +0200, Ingo Molnar wrote:
quoted text
> 
> * Mikael Pettersson <mikpe@it.uu.se> wrote:
> 
> > Since 2.6.27-rc1 my Core2Duo has been getting sporadic oopses
> > from hpet_rtc_interrupt, usually during shutdown or reboot,
> > but occasionally also early in init. Today I finally managed
> > to capture one via a serial cable:
> > 
> > INIT: version 2.86 booting
> > 		Welcome to Fedora Core
> > 		Press 'I' to enter interactive startup.
> > BUG: NMI Watchdog detected LOCKUP on CPU0, ip c0117092, registers:
> > Modules linked in: ehci_hcd uhci_hcd usbcore
> > 
> > Pid: 311, comm: nash-hotplug Not tainted (2.6.27-rc4 #1)
> > EIP: 0060:[<c0117092>] EFLAGS: 00000097 CPU: 0
> > EIP is at hpet_rtc_interrupt+0x2d2/0x310
> > EAX: 00000000 EBX: 00000002 ECX: 00000046 EDX: 00000002
> > ESI: 000000a6 EDI: ffff8e25 EBP: 00000008 ESP: f7bd7f28
> >  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
> > Process nash-hotplug (pid: 311, ti=f7bd6000 task=f7b70460 task.ti=f7bd6000)
> > Stack: f7bd7f6c c0139cc0 00000000 c035ba04 00000000 00000000 00000000 00000000 
> >        00000000 00000000 00000000 00000000 00000000 f7b845a0 00000000 00000000 
> >        00000008 c01478a8 c035bf80 f7b845a0 c035bfb0 00000008 c0148f71 00000400 
> > Call Trace:
> >  [<c0139cc0>] hrtimer_run_pending+0x20/0x90
> >  [<c01478a8>] handle_IRQ_event+0x28/0x50
> >  [<c0148f71>] handle_edge_irq+0xa1/0x120
> >  [<c010615b>] do_IRQ+0x3b/0x70
> >  [<c0113225>] smp_apic_timer_interrupt+0x55/0x80
> >  [<c0103c4f>] common_interrupt+0x23/0x28
> >  [<c02c0000>] unix_release_sock+0xc0/0x220
> >  =======================
> > Code: 89 44 24 18 0f b6 c2 e8 5d 74 0c 00 8b 0d d8 9c 3b c0 89 44 24 1c 8b 44 24 0c 48 89 44 24 20 e9 84 fd ff ff 90 8d 74 26 00 f3 90 <a1> 80 ba 35 c0 29 f8 83 f8 01 76 f2 e9 e1 fe ff ff 90 8d 74 26 
> > 
> > This points to the following loop in hpet_rtc_interrupt:
> > 
> > 0xc0117090 <hpet_rtc_interrupt+720>:    pause  
> > 0xc0117092 <hpet_rtc_interrupt+722>:    mov    0xc035ba80,%eax
> > 0xc0117097 <hpet_rtc_interrupt+727>:    sub    %edi,%eax
> > 0xc0117099 <hpet_rtc_interrupt+729>:    cmp    __PLACEHOLDER__0_x1,%eax
> > 0xc011709c <hpet_rtc_interrupt+732>:    jbe    0xc0117090 <hpet_rtc_interrupt+720>
> > 
> > Note: 0xc035ba80 == &jiffies
> > 
> > This loop originates from asm-generic/rtc.h:get_rtc_time()
> > 
> > 		while (jiffies - uip_watchdog < 2*HZ/100) {
> > 			barrier();
> > 			cpu_relax();
> > 		}
> > 
> > Note: HZ == CONFIG_HZ == 100
> > 
> > The bug may not originate from the 2.6.27-rc series as I only recently 
> > enabled HPET in this machine's kernels (not due to HPET problems, it 
> > inherited its .config way back from an older machine w/o HPET).
> 
> argh, that loop in asm-generic/rtc.h:get_rtc_time looks extremely 
> fragile, we'll lock up if it's ever called with hardirqs off!
> 
> Does the patch below do the trick?
> 
> 	Ingo
> 
> ----------------->
> >From 2273cc870b52a7ed09eb225142a6db97299e4f39 Mon Sep 17 00:00:00 2001
> From: Ingo Molnar <mingo@elte.hu>
> Date: Sat, 23 Aug 2008 17:59:07 +0200
> Subject: [PATCH] rtc: fix deadlock
> 
> if get_rtc_time() is _ever_ called with IRQs off, we deadlock badly
> in it, waiting for jiffies to increment.
> 
> So make the code more robust by doing an explicit mdelay(20).
> 
> This solves a very hard to reproduce/debug hard lockup reported
> by Mikael Pettersson.
> 
> Reported-by: Mikael Pettersson <mikpe@it.uu.se>
> Signed-off-by: Ingo Molnar <mingo@elte.hu>
> ---
>  include/asm-generic/rtc.h |   12 ++++--------
>  1 files changed, 4 insertions(+), 8 deletions(-)
> 
> diff --git a/include/asm-generic/rtc.h b/include/asm-generic/rtc.h
> index be4af00..71ef3f0 100644
> --- a/include/asm-generic/rtc.h
> +++ b/include/asm-generic/rtc.h
> @@ -15,6 +15,7 @@
>  #include <linux/mc146818rtc.h>
>  #include <linux/rtc.h>
>  #include <linux/bcd.h>
> +#include <linux/delay.h>
>  
>  #define RTC_PIE 0x40		/* periodic interrupt enable */
>  #define RTC_AIE 0x20		/* alarm interrupt enable */
> @@ -43,7 +44,6 @@ static inline unsigned char rtc_is_updating(void)
>  
>  static inline unsigned int get_rtc_time(struct rtc_time *time)
>  {
> -	unsigned long uip_watchdog = jiffies;
>  	unsigned char ctrl;
>  	unsigned long flags;
>  
> @@ -53,19 +53,15 @@ static inline unsigned int get_rtc_time(struct rtc_time *time)
>  
>  	/*
>  	 * read RTC once any update in progress is done. The update
> -	 * can take just over 2ms. We wait 10 to 20ms. There is no need to
> +	 * can take just over 2ms. We wait 20ms. There is no need to
>  	 * to poll-wait (up to 1s - eeccch) for the falling edge of RTC_UIP.
>  	 * If you need to know *exactly* when a second has started, enable
>  	 * periodic update complete interrupts, (via ioctl) and then 
>  	 * immediately read /dev/rtc which will block until you get the IRQ.
>  	 * Once the read clears, read the RTC time (again via ioctl). Easy.
>  	 */
> -
> -	if (rtc_is_updating() != 0)
> -		while (jiffies - uip_watchdog < 2*HZ/100) {
> -			barrier();
> -			cpu_relax();
> -		}
> +	if (rtc_is_updating())
> +		mdelay(20);
>  
>  	/*
>  	 * Only the values that we read from the RTC are set. We leave

This patch fixes a regression since 2.6.24: 2.6.25 and 2.6.26 occasionally
locked up hard here without a trace and even alt-sysrq did not work
anymore. It's easy to reproduce with

	while :; do hwclock; done

Others are experiencing this issue too:
- http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=494036
- http://kerneltrap.org/mailarchive/message-id/20080821163920.GA19140@gamma.logic.tuwien...
- people (me included) experienced booting problems because of
  this (lockup after initscripts says "Setting the system clock").

maybe this is 2.6.25.x and 2.6.26.x material too?


Anyway, get_rtc_time() is called by interrupt handler(s). I think 20ms
is a awful lot of time for an interrupt handler.

-- 
Frank
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/