On Thu, Apr 17, 2014 at 08:15:46PM +0100, Ashwin Chaugule wrote:
PSCIv0.2 adds a new function called AFFINITY_INFO, which can be used to query if a specified CPU has actually gone offline. Calling this function via cpu_kill ensures that a CPU has quiesced after a call to cpu_die.
Signed-off-by: Ashwin Chaugule ashwin.chaugule@linaro.org Reviewed-by: Rob Herring robh@kernel.org
arch/arm/kernel/psci_smp.c | 21 +++++++++++++++++++++ include/uapi/linux/psci.h | 5 +++++ 2 files changed, 26 insertions(+)
diff --git a/arch/arm/kernel/psci_smp.c b/arch/arm/kernel/psci_smp.c index 570a48c..c6f1420 100644 --- a/arch/arm/kernel/psci_smp.c +++ b/arch/arm/kernel/psci_smp.c @@ -16,6 +16,7 @@ #include <linux/init.h> #include <linux/smp.h> #include <linux/of.h> +#include <uapi/linux/psci.h> #include <asm/psci.h> #include <asm/smp_plat.h> @@ -66,6 +67,25 @@ void __ref psci_cpu_die(unsigned int cpu) /* We should never return */ panic("psci: cpu %d failed to shutdown\n", cpu); }
+int __ref psci_cpu_kill(unsigned int cpu) +{
- int err;
- if (!psci_ops.affinity_info)
return 1;- err = psci_ops.affinity_info(cpu_logical_map(cpu), 0);
- if (err != PSCI_AFFINITY_INFO_RET_OFF) {
pr_err("psci: Cannot kill CPU:%d, psci ret val: %d\n",cpu, err);/* Make platform_cpu_kill() fail. */return 0;- }
We can race with the dying CPU here -- if we call AFFINITY_INFO before the dying cpu is sufficiently far through its CPU_OFF call it won't register as OFF.
Could we poll here instead (with a reasonable limit on the number of iterations)? That would enable us to not spuriously declare a CPU to be dead when it happened to take slightly longer than we expect to turn off.
Otherwise, this looks good. Thanks for implementing this :)
Cheers, Mark.