Fix the VMI-Timer no-idle-hz code.
Do not setup a one shot alarm if we are keeping the periodic alarm
armed. Additionally, since the periodic alarm can be run at a lower
rate than HZ, let's fixup the guard to the no-idle-hz mode appropriately.
This fixes the bug where the no-idle-hz mode might have a higher interrupt
rate than the non-idle case.
Signed-off-by: Dan Hecht <[email protected]>
diff -r 9d107b81bb7d arch/i386/kernel/vmitime.c
--- a/arch/i386/kernel/vmitime.c Thu Feb 01 23:43:37 2007 -0800
+++ b/arch/i386/kernel/vmitime.c Thu Feb 01 23:52:59 2007 -0800
@@ -374,7 +374,6 @@ int vmi_stop_hz_timer(void)
unsigned long seq, next;
unsigned long long real_cycles_expiry;
int cpu = smp_processor_id();
- int idle;
BUG_ON(!irqs_disabled());
if (sysctl_hz_timer != 0)
@@ -382,13 +381,13 @@ int vmi_stop_hz_timer(void)
cpu_set(cpu, nohz_cpu_mask);
smp_mb();
+
if (rcu_needs_cpu(cpu) || local_softirq_pending() ||
- (next = next_timer_interrupt(), time_before_eq(next, jiffies))) {
+ (next = next_timer_interrupt(),
+ time_before_eq(next, jiffies + HZ/CONFIG_VMI_ALARM_HZ))) {
cpu_clear(cpu, nohz_cpu_mask);
- next = jiffies;
- idle = 0;
- } else
- idle = 1;
+ return 0;
+ }
/* Convert jiffies to the real cycle counter. */
do {
@@ -398,17 +397,13 @@ int vmi_stop_hz_timer(void)
} while (read_seqretry(&xtime_lock, seq));
/* This cpu is going idle. Disable the periodic alarm. */
- if (idle) {
- vmi_timer_ops.cancel_alarm(VMI_CYCLES_AVAILABLE);
- per_cpu(idle_start_jiffies, cpu) = jiffies;
- }
-
+ vmi_timer_ops.cancel_alarm(VMI_CYCLES_AVAILABLE);
+ per_cpu(idle_start_jiffies, cpu) = jiffies;
/* Set the real time alarm to expire at the next event. */
vmi_timer_ops.set_alarm(
- VMI_ALARM_WIRING | VMI_ALARM_IS_ONESHOT | VMI_CYCLES_REAL,
- real_cycles_expiry, 0);
-
- return idle;
+ VMI_ALARM_WIRING | VMI_ALARM_IS_ONESHOT | VMI_CYCLES_REAL,
+ real_cycles_expiry, 0);
+ return 1;
}
static void vmi_reenable_hz_timer(int cpu)
diff -r 9d107b81bb7d kernel/sysctl.c
--- a/kernel/sysctl.c Thu Feb 01 23:43:37 2007 -0800
+++ b/kernel/sysctl.c Thu Feb 01 23:52:59 2007 -0800
@@ -440,17 +440,7 @@ static ctl_table kern_table[] = {
.extra1 = &minolduid,
.extra2 = &maxolduid,
},
-#ifdef CONFIG_S390
-#ifdef CONFIG_MATHEMU
- {
- .ctl_name = KERN_IEEE_EMULATION_WARNINGS,
- .procname = "ieee_emulation_warnings",
- .data = &sysctl_ieee_emulation_warnings,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = &proc_dointvec,
- },
-#endif
+#if defined(CONFIG_S390) || defined(CONFIG_VMI)
#ifdef CONFIG_NO_IDLE_HZ
{
.ctl_name = KERN_HZ_TIMER,
@@ -459,6 +449,18 @@ static ctl_table kern_table[] = {
.maxlen = sizeof(int),
.mode = 0644,
.proc_handler = &proc_dointvec,
+ },
+#endif
+#endif
+#ifdef CONFIG_S390
+#ifdef CONFIG_MATHEMU
+ {
+ .ctl_name = KERN_IEEE_EMULATION_WARNINGS,
+ .procname = "ieee_emulation_warnings",
+ .data = &sysctl_ieee_emulation_warnings,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec,
},
#endif
{
> -#ifdef CONFIG_S390
> -#ifdef CONFIG_MATHEMU
> - {
> - .ctl_name = KERN_IEEE_EMULATION_WARNINGS,
> - .procname = "ieee_emulation_warnings",
> - .data = &sysctl_ieee_emulation_warnings,
> - .maxlen = sizeof(int),
> - .mode = 0644,
> - .proc_handler = &proc_dointvec,
> - },
I think there needs to be a way to have this syctl only
when it's really implemented for the paravirt ops.
I suppose you should move it to dynamic registration on demand.
-Andi
Andi Kleen wrote:
>> -#ifdef CONFIG_S390
>> -#ifdef CONFIG_MATHEMU
>> - {
>> - .ctl_name = KERN_IEEE_EMULATION_WARNINGS,
>> - .procname = "ieee_emulation_warnings",
>> - .data = &sysctl_ieee_emulation_warnings,
>> - .maxlen = sizeof(int),
>> - .mode = 0644,
>> - .proc_handler = &proc_dointvec,
>> - },
>>
>
> I think there needs to be a way to have this syctl only
> when it's really implemented for the paravirt ops.
>
> I suppose you should move it to dynamic registration on demand.
>
Hmm, is there a way to do that and share the sysctl with Xen? Yes, I
think if we move the data to paravirt.c and call
paravirt_start_no_idle_hz (or something_with_fewer_underbars), then we
can both share this call site instead of silly duplicating the .ctl_name ...
Will code it up.
Zach
Fix the VMI-Timer no-idle-hz code.
Do not setup a one shot alarm if we are keeping the periodic alarm
armed. Additionally, since the periodic alarm can be run at a lower
rate than HZ, let's fixup the guard to the no-idle-hz mode appropriately.
This fixes the bug where the no-idle-hz mode might have a higher interrupt
rate than the non-idle case.
Signed-off-by: Dan Hecht <[email protected]>
Signed-off-by: Zachary Amsden <[email protected]>
diff -r 9d107b81bb7d arch/i386/kernel/vmitime.c
--- a/arch/i386/kernel/vmitime.c Thu Feb 01 23:43:37 2007 -0800
+++ b/arch/i386/kernel/vmitime.c Thu Feb 01 23:52:59 2007 -0800
@@ -374,7 +374,6 @@ int vmi_stop_hz_timer(void)
unsigned long seq, next;
unsigned long long real_cycles_expiry;
int cpu = smp_processor_id();
- int idle;
BUG_ON(!irqs_disabled());
if (sysctl_hz_timer != 0)
@@ -382,13 +381,13 @@ int vmi_stop_hz_timer(void)
cpu_set(cpu, nohz_cpu_mask);
smp_mb();
+
if (rcu_needs_cpu(cpu) || local_softirq_pending() ||
- (next = next_timer_interrupt(), time_before_eq(next, jiffies))) {
+ (next = next_timer_interrupt(),
+ time_before_eq(next, jiffies + HZ/CONFIG_VMI_ALARM_HZ))) {
cpu_clear(cpu, nohz_cpu_mask);
- next = jiffies;
- idle = 0;
- } else
- idle = 1;
+ return 0;
+ }
/* Convert jiffies to the real cycle counter. */
do {
@@ -398,17 +397,13 @@ int vmi_stop_hz_timer(void)
} while (read_seqretry(&xtime_lock, seq));
/* This cpu is going idle. Disable the periodic alarm. */
- if (idle) {
- vmi_timer_ops.cancel_alarm(VMI_CYCLES_AVAILABLE);
- per_cpu(idle_start_jiffies, cpu) = jiffies;
- }
-
+ vmi_timer_ops.cancel_alarm(VMI_CYCLES_AVAILABLE);
+ per_cpu(idle_start_jiffies, cpu) = jiffies;
/* Set the real time alarm to expire at the next event. */
vmi_timer_ops.set_alarm(
- VMI_ALARM_WIRING | VMI_ALARM_IS_ONESHOT | VMI_CYCLES_REAL,
- real_cycles_expiry, 0);
-
- return idle;
+ VMI_ALARM_WIRING | VMI_ALARM_IS_ONESHOT | VMI_CYCLES_REAL,
+ real_cycles_expiry, 0);
+ return 1;
}
static void vmi_reenable_hz_timer(int cpu)