Wei, Gang
2009-May-14 13:41 UTC
[Xen-devel] [PATCH]cpuidle: Fix possible false judge caused by type casting
Keir, this is a fix along with an added tuning option. It should not have any
negative impact. Could you take it in 3.4?
Jimmy
cpuidle: Fix possible false judge caused by type casting
For timer_deadline == 0 or timer_deadline - now > largest u32 case,
the expected_us (in u32 type) may become wrong.
Add a tunable option to ease conditional adjustment.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r fb3e1e84d8f4 xen/arch/x86/acpi/cpu_idle.c
--- a/xen/arch/x86/acpi/cpu_idle.c Wed May 06 12:40:51 2009 +0100
+++ b/xen/arch/x86/acpi/cpu_idle.c Thu May 14 17:36:11 2009 +0800
@@ -624,6 +624,7 @@ static int check_cx(struct acpi_processo
}
static unsigned int latency_factor = 2;
+integer_param("idle_latency_factor", latency_factor);
static void set_cx(
struct acpi_processor_power *acpi_power,
diff -r fb3e1e84d8f4 xen/arch/x86/acpi/cpuidle_menu.c
--- a/xen/arch/x86/acpi/cpuidle_menu.c Wed May 06 12:40:51 2009 +0100
+++ b/xen/arch/x86/acpi/cpuidle_menu.c Thu May 14 17:36:11 2009 +0800
@@ -45,9 +45,15 @@ struct menu_device
static DEFINE_PER_CPU(struct menu_device, menu_devices);
-static s_time_t get_sleep_length_ns(void)
+static unsigned int get_sleep_length_us(void)
{
- return per_cpu(timer_deadline, smp_processor_id()) - NOW();
+ s_time_t us = (per_cpu(timer_deadline, smp_processor_id()) - NOW()) / 1000;
+ /*
+ * while us < 0 or us > (u32)-1, return a large u32,
+ * choose (unsigned int)-2000 to avoid wrapping while added with exit
+ * latency because the latency should not larger than 2ms
+ */
+ return (us >> 32) ? (unsigned int)-2000 : (unsigned int)us;
}
static int menu_select(struct acpi_processor_power *power)
@@ -56,7 +62,7 @@ static int menu_select(struct acpi_proce
int i;
/* determine the expected residency time */
- data->expected_us = (u32) get_sleep_length_ns() / 1000;
+ data->expected_us = get_sleep_length_us();
/* find the deepest idle state that satisfies our constraints */
for ( i = 2; i < power->count; i++ )
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel