2014-10-15 14:54:03

by Boris Ostrovsky

Subject: [PATCH] xen/smp: Use die_complete completion when taking CPU down

Commit 2ed53c0d6cc9 ("x86/smpboot: Speed up suspend/resume by avoiding
100ms sleep for CPU offline during S3") introduced completions to CPU
offlining process. These completions are not initialized on Xen kernels
causing a panic in play_dead_common().

Add handling of die_complete on Xen (this requires moving die_complete
to global scope. While at it, move die_complete definition in
arch/x86/kernel/smpboot.c under ifdef CONFIG_HOTPLUG_CPU)

Signed-off-by: Boris Ostrovsky <[email protected]>
---
arch/x86/include/asm/cpu.h | 3 +++
arch/x86/kernel/smpboot.c | 4 +++-
arch/x86/xen/smp.c | 4 ++++
3 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/cpu.h b/arch/x86/include/asm/cpu.h
index d2b1298..b65d5c9 100644
--- a/arch/x86/include/asm/cpu.h
+++ b/arch/x86/include/asm/cpu.h
@@ -6,6 +6,7 @@
#include <linux/topology.h>
#include <linux/nodemask.h>
#include <linux/percpu.h>
+#include <linux/completion.h>

#ifdef CONFIG_SMP

@@ -26,6 +27,8 @@ struct x86_cpu {
};

#ifdef CONFIG_HOTPLUG_CPU
+DECLARE_PER_CPU(struct completion, die_complete);
+
extern int arch_register_cpu(int num);
extern void arch_unregister_cpu(int);
extern void start_cpu0(void);
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 2d5200e..abe9005 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -102,7 +102,9 @@ DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
EXPORT_PER_CPU_SYMBOL(cpu_info);

-static DEFINE_PER_CPU(struct completion, die_complete);
+#ifdef CONFIG_HOTPLUG_CPU
+DEFINE_PER_CPU(struct completion, die_complete);
+#endif

atomic_t init_deasserted;

diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
index 8650cdb..4da0008 100644
--- a/arch/x86/xen/smp.c
+++ b/arch/x86/xen/smp.c
@@ -498,6 +498,7 @@ static int xen_cpu_disable(void)
if (cpu == 0)
return -EBUSY;

+ init_completion(&per_cpu(die_complete, smp_processor_id()));
cpu_disable_common();

load_cr3(swapper_pg_dir);
@@ -510,6 +511,9 @@ static void xen_cpu_die(unsigned int cpu)
current->state = TASK_UNINTERRUPTIBLE;
schedule_timeout(HZ/10);
}
+
+ wait_for_completion_timeout(&per_cpu(die_complete, cpu), HZ);
+
xen_smp_intr_free(cpu);
xen_uninit_lock_cpu(cpu);
xen_teardown_timer(cpu);
--
1.8.1.4

2014-10-16 07:30:56

by Ingo Molnar

Subject: Re: [PATCH] xen/smp: Use die_complete completion when taking CPU down

* Boris Ostrovsky <[email protected]> wrote:

> diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
> index 8650cdb..4da0008 100644
> --- a/arch/x86/xen/smp.c
> +++ b/arch/x86/xen/smp.c
> @@ -498,6 +498,7 @@ static int xen_cpu_disable(void)
> if (cpu == 0)
> return -EBUSY;
>
> + init_completion(&per_cpu(die_complete, smp_processor_id()));
> cpu_disable_common();
>
> load_cr3(swapper_pg_dir);
> @@ -510,6 +511,9 @@ static void xen_cpu_die(unsigned int cpu)
> current->state = TASK_UNINTERRUPTIBLE;
> schedule_timeout(HZ/10);
> }
> +
> + wait_for_completion_timeout(&per_cpu(die_complete, cpu), HZ);
> +
> xen_smp_intr_free(cpu);
> xen_uninit_lock_cpu(cpu);
> xen_teardown_timer(cpu);

So why not put it into cpu_disable_common() and do the same
change for native_cpu_disable()?

Likewise, the waiting bit should be put into a common function as
well (newly introduced, if need to be), we don't want to expose
internal x86 core data structures to the Xen guest code.

Thanks,

Ingo