Peter,
What's your opinion about the patch? We hit it when enabling Medfield Android mobile.
This patch would put AP to a loop.
Another method to fix it is just to enlarge the wait time, for example, from 2HZ to 10HZ.
Yanmin
-------- Forwarded Message --------
> From: Chen, LinX Z <[email protected]>
> To: [email protected]
> Cc: [email protected], [email protected], [email protected],
> [email protected]
> Subject: [PATCH] x86/smp: Fix cpuN startup panic
> Date: Tue, 07 Aug 2012 18:50:40 +0900
>
> From: Lin Chen <[email protected]>
>
> We hit a panic while doing cpu hotplug test.
> <0>[ 627.982857] Kernel panic - not syncing: smp_callin: CPU1 started up but did not get a callout!
> <0>[ 627.982864]
> <4>[ 627.982876] Pid: 0, comm: kworker/0:1 Tainted: G ...
> <4>[ 627.982883] Call Trace:
> <4>[ 627.982903] [<c18f2977>] panic+0x66/0x16c
> <4>[ 627.982918] [<c12234cc>] ? default_get_apic_id+0x1c/0x40
> <4>[ 627.982931] [<c18ef96d>] start_secondary+0xda/0x252
>
> During BSP bootup AP, it is possible that BSP be preempted before
> finishing STARTUP sequence of AP(set cpu_callout_mask) which maybe cause
> AP busy wait for it. At present, AP will wait for 2 seconds then panic.
>
> This patch let AP waits until BSP finish the startup sequence and gives
> WARNING when BSP is preempted more than 2 seconds.
>
> Signed-off-by: Yanmin Zhang <[email protected]>
> Signed-off-by: Lin Chen <[email protected]>
> ---
> arch/x86/kernel/smpboot.c | 11 ++++++-----
> 1 files changed, 6 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
> index 7c5a8c3..a9e3379 100644
> --- a/arch/x86/kernel/smpboot.c
> +++ b/arch/x86/kernel/smpboot.c
> @@ -165,19 +165,20 @@ static void __cpuinit smp_callin(void)
> * Waiting 2s total for startup (udelay is not yet working)
> */
> timeout = jiffies + 2*HZ;
> - while (time_before(jiffies, timeout)) {
> + while (1) {
> /*
> * Has the boot CPU finished it's STARTUP sequence?
> */
> if (cpumask_test_cpu(cpuid, cpu_callout_mask))
> break;
> cpu_relax();
> + if (!time_before(jiffies, timeout)) {
> + WARN(1, "%s: CPU%d started up but did not get a callout!\n",
> + __func__, cpuid);
> + timeout = jiffies + 2*HZ;
> + }
> }
>
> - if (!time_before(jiffies, timeout)) {
> - panic("%s: CPU%d started up but did not get a callout!\n",
> - __func__, cpuid);
> - }
>
> /*
> * the boot CPU has finished the init stage and is spinning
> --
> 1.7.1
On 08/09/2012 05:35 PM, Yanmin Zhang wrote:
> Peter,
>
> What's your opinion about the patch? We hit it when enabling Medfield Android mobile.
> This patch would put AP to a loop.
>
> Another method to fix it is just to enlarge the wait time, for example, from 2HZ to 10HZ.
>
Hmm... and I presume running with preemption disabled doesn't work for
hotplug (not that this is a hotplug situation)...
2 seconds is a *long* time, but I guess there isn't really nay harm in
making it 10 seconds in the sense that the system is basically dead at
that point...
tglx - you have been looking at the CPU startup lately, any opinions?
-hpa
--
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel. I don't speak on their behalf.