Re: [Crash-utility] [PATCH 2/3] crash: ARM: get online CPUs from cpu_active_mask
On 2014/4/26 3:20, Dave Anderson wrote:
----- Original Message -----
> For ARM32 platfrom, The system will "offline" all CPUs except the
> crashing one, by clear the cpu_online_mask. So we need to find
> another way to get online-CPUs number for crash utility.
>
> This patch uses cpu_active_mask to get that value.
>
> Signed-off-by: Liu Hua <sdu.liu(a)huawei.com>
> ---
> arm.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arm.c b/arm.c
> index 3c38cd5..c94d7bf 100644
> --- a/arm.c
> +++ b/arm.c
> @@ -1518,7 +1518,7 @@ arm_display_machine_stats(void)
> static int
> arm_get_smp_cpus(void)
> {
> - return get_cpus_online();
> + return get_cpus_active();
> }
>
> /*
When did this start happening? I ask because no other ARM users have
reported this until now. I've only got one sample SMP vmcore, and
clearly the non-crashing cpu was not offlined:
Hi Dave,
commit 36a2e5010fe87 "kexec: offline non panic CPUs on Kdump panic "
introduces this change. <v3.10.27-1-g36a2e50>
The following messges come from the ARM boards on my hand. The crash
utility has added my patchs.
(1) ARM A15 16 CPUs
crash> sys
KERNEL: vmlinux
DUMPFILE: vmcore_my [PARTIAL DUMP]
CPUS: 16
DATE: Fri Feb 28 10:07:55 2014
UPTIME: 00:03:57
LOAD AVERAGE: 0.03, 0.01, 0.01
TASKS: 140
NODENAME: arma15el
RELEASE: 3.10.37+
VERSION: #2 SMP Thu Apr 24 21:13:28 CST 2014
MACHINE: armv7l (unknown Mhz)
MEMORY: 3.1 GB
PANIC: "Internal error: Oops: a07 [#1] SMP ARM" (check log for details
crash> help -k
....
cpu_possible_map: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
cpu_present_map: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
cpu_online_map: 0
...
(2) ARM A9 4 CPUs
crash> sys
KERNEL: vmlinux
DUMPFILE: vmcore
CPUS: 2
DATE: Mon Apr 14 11:09:10 2014
UPTIME: 00:06:17
LOAD AVERAGE: 0.16, 0.05, 0.02
TASKS: 45
NODENAME: arma9el
RELEASE: 3.10.37+
VERSION: #6 SMP Tue Apr 22 13:23:30 CST 2014
MACHINE: armv7l (unknown Mhz)
MEMORY: 1.5 GB
PANIC: "Internal error: Oops: 817 [#1] SMP ARM" (check log for details)
crash> help -k
....
cpu_possible_map: 0 1
cpu_present_map: 0 1
cpu_online_map: 1
crash> p per_cpu__runqueues
PER-CPU DATA TYPE:
struct rq runqueues;
PER-CPU ADDRESSES:
[0]: c069a680
[1]: c06a2680
With my patches, "sys" command is ok. And I can get per_cpu variables
correctly.
But something should be done for "help -k". May be I should add fully support
for cpu_active_mask. If necessory, I will resend this patch series.
crash> sys
...
Also, backwards-compatibility must be maintained, and it appears that
the cpu_active mask was only first introduced in 2.6.27. So perhaps it
should be something like:
Absolutely, I will repost this one.
--- arm.c.orig 2014-04-25 14:44:23.557665220 -0400
+++ arm.c 2014-04-25 15:13:46.709594145 -0400
@@ -1518,7 +1518,7 @@
static int
arm_get_smp_cpus(void)
{
- return get_cpus_online();
+ return MAX(get_cpus_active(), get_cpus_online());
}
Dave