Skip to contentSkip to conversation

regression in TSC unstable?

PATCH In review Steven Rostedt 6 participants 20 messages May 8, 2009 – May 15, 2009 ACPI

lore
Share & tools
lore
#0 lore raw

The ring-buffer-benchmark module creates a producer and consumer and loops on do_gettimeofday until it hits about 10 seconds. Then it calculates the number of events recorded / time running.

After Ingo merged Linus's latest my tests when from 330ns per entry to 880ns. After bisecting it I came down to this change:

commit a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2
Author: Len Brown <[email protected]>
Date: Tue Apr 21 00:50:11 2009 -0400

ACPI: idle: mark_tsc_unstable() at init-time, not run-time

I've never had a problem with the TSC with this box before. It passes the synchronization phase, and until this commit, the TSC never was marked as unstable.

Should it now be unstable? Or is this a false positive?

# cat /proc/cpuinfo 
processor	: 0
vendor_id	: GenuineIntel
cpu family	: 15
model		: 4
model name	: Intel(R) Xeon(TM) CPU 2.80GHz
stepping	: 1
cpu MHz		: 2793.272
cache size	: 1024 KB
physical id	: 0
siblings	: 2
core id		: 0
cpu cores	: 1
apicid		: 0
initial apicid	: 0
fpu		: yes
fpu_exception	: yes
cpuid level	: 5
wp		: yes
flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca 
cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx 
lm constant_tsc pebs bts pni dtes64 monitor ds_cpl cid cx16 xtpr
bogomips	: 5586.54
clflush size	: 64
cache_alignment	: 128
address sizes	: 36 bits physical, 48 bits virtual
power management:

Anything else I might need to know?

-- Steve

reviewer
#1 lore raw
Quoted 5 lines · Steven Rostedt

The ring-buffer-benchmark module creates a producer and consumer and loops on do_gettimeofday until it hits about 10 seconds. Then it calculates the number of events recorded / time running.

After Ingo merged Linus's latest

merged what into what?
are you referring to the upstream kernel or something else?

Quoted 7 lines

my tests when from 330ns per entry to
880ns. After bisecting it I came down to this change:

commit a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2
Author: Len Brown <[email protected]>
Date: Tue Apr 21 00:50:11 2009 -0400

Quoted 8 lines

ACPI: idle: mark_tsc_unstable() at init-time, not run-time

I've never had a problem with the TSC with this box before. It passes the synchronization phase, and until this commit, the TSC never was marked as unstable.

Should it now be unstable? Or is this a false positive?

If you revert that patch the issue goes away?

Does your Xeon even export any C-states?
(cat /proc/acpi/processor/*/power)

does your dmesg before or afer the patch actually print "TSC halts in idle"?

thanks,
-Len

Quoted 38 lines
# cat /proc/cpuinfo 
processor	: 0
vendor_id	: GenuineIntel
cpu family	: 15
model		: 4
model name	: Intel(R) Xeon(TM) CPU 2.80GHz
stepping	: 1
cpu MHz		: 2793.272
cache size	: 1024 KB
physical id	: 0
siblings	: 2
core id		: 0
cpu cores	: 1
apicid		: 0
initial apicid	: 0
fpu		: yes
fpu_exception	: yes
cpuid level	: 5
wp		: yes
flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca 
cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx 
lm constant_tsc pebs bts pni dtes64 monitor ds_cpl cid cx16 xtpr
bogomips	: 5586.54
clflush size	: 64
cache_alignment	: 128
address sizes	: 36 bits physical, 48 bits virtual
power management:

Anything else I might need to know?

-- Steve

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#2 lore raw
Quoted 10 lines · Len Brown
Quoted 5 lines · Steven Rostedt

The ring-buffer-benchmark module creates a producer and consumer and loops on do_gettimeofday until it hits about 10 seconds. Then it calculates the number of events recorded / time running.

After Ingo merged Linus's latest

merged what into what?
are you referring to the upstream kernel or something else?

he's describing how he found it: a not-yet-upstream kernel-benchmark module showed a performance regression that came in via upstream. The regression is upstream too, just harder to notice.

Ingo

#3 lore raw

Re: Regression in TSC unstable -- confirmed

In reply to Steven Rostedt

I'm seeing the same issue with 2.6.30-rc5 (not tested previous RCs on that box) on a system for which TSC has always been usable.

$ dmesg | grep TSC
Fast TSC calibration using PIT
checking TSC synchronization [CPU#0 -> CPU#1]: passed.
Marking TSC unstable due to TSC halts in idle

The last line is new with 2.6.30. After reverting a71e4917dc0 the TSC works fine again.

The system has a Intel D945GCZ mainboard with dual core Pentium D (3.20GHz) processor running x86_64 and Debian Lenny.

Cheers,
FJP

(Thanks to Steven for doing the bisection.)

#4 lore raw

PATCH 1/1 ACPI: do not mark TSC unstable for invalid C-states

commit a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2 (ACPI: idle:
mark_tsc_unstable() at init-time, not run-time) marks TSC unstable even on non affected systems.

The reason is that the state enumeration does not check the cx->valid flag before calling tsc_halts_in_c(), so it marks TSC unstable for any enumerated albeit invalid C state which is known to turn off TSC.

Check cx->valid before checking whether the TSC might become unstable.

---
 drivers/acpi/processor_idle.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/acpi/processor_idle.c +1 -1
Index: linux-2.6/drivers/acpi/processor_idle.c
===================================================================
@@ -583,7 +583,7 @@ static int acpi_processor_power_verify(s
 
 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 		/* TSC could halt in idle, so notify users */
-		if (tsc_halts_in_c(cx->type))
+		if (cx->valid && tsc_halts_in_c(cx->type))
 			mark_tsc_unstable("TSC halts in idle");;
 #endif
 		switch (cx->type) {
author Tested-by on v1 1/1
#5 lore raw

Re: [PATCH] ACPI: do not mark TSC unstable for invalid C-states

Quoted 11 lines · Thomas Gleixner

commit a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2 (ACPI: idle:
mark_tsc_unstable() at init-time, not run-time) marks TSC unstable even on non affected systems.

The reason is that the state enumeration does not check the cx->valid flag before calling tsc_halts_in_c(), so it marks TSC unstable for any enumerated albeit invalid C state which is known to turn off TSC.

Check cx->valid before checking whether the TSC might become unstable.

-- Steve

Quoted 20 lines
---
 drivers/acpi/processor_idle.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: linux-2.6/drivers/acpi/processor_idle.c
===================================================================

drivers/acpi/processor_idle.c +1 -1
@@ -583,7 +583,7 @@ static int acpi_processor_power_verify(s
 
 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 		/* TSC could halt in idle, so notify users */
-		if (tsc_halts_in_c(cx->type))
+		if (cx->valid && tsc_halts_in_c(cx->type))
 			mark_tsc_unstable("TSC halts in idle");;
 #endif
 		switch (cx->type) {
Tested-by on v1 1/1
#6 lore raw

Re: [PATCH] ACPI: do not mark TSC unstable for invalid C-states

In reply to Thomas Gleixner
Quoted 11 lines · Thomas Gleixner

commit a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2 (ACPI: idle:
mark_tsc_unstable() at init-time, not run-time) marks TSC unstable even on non affected systems.

The reason is that the state enumeration does not check the cx->valid flag before calling tsc_halts_in_c(), so it marks TSC unstable for any enumerated albeit invalid C state which is known to turn off TSC.

Check cx->valid before checking whether the TSC might become unstable.

Thanks Thomas.

#7 lore raw

Re: [PATCH] ACPI: do not mark TSC unstable for invalid C-states

In reply to Thomas Gleixner
Quoted 20 lines · Thomas Gleixner

commit a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2 (ACPI: idle:
mark_tsc_unstable() at init-time, not run-time) marks TSC unstable even on non affected systems.

The reason is that the state enumeration does not check the cx->valid flag before calling tsc_halts_in_c(), so it marks TSC unstable for any enumerated albeit invalid C state which is known to turn off TSC.

Check cx->valid before checking whether the TSC might become unstable.

---
 drivers/acpi/processor_idle.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: linux-2.6/drivers/acpi/processor_idle.c
===================================================================

drivers/acpi/processor_idle.c static int acpi_processor_power_verify(s · line 587

Quoted 6 lines
drivers/acpi/processor_idle.c +1 -1
@@ -583,7 +583,7 @@ static int acpi_processor_power_verify(s
 
 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 		/* TSC could halt in idle, so notify users */
-		if (tsc_halts_in_c(cx->type))
+		if (cx->valid && tsc_halts_in_c(cx->type))
 			mark_tsc_unstable("TSC halts in idle");;

Small suggestion: can we fix the double ';' here at the same time?

drivers/acpi/processor_idle.c static int acpi_processor_power_verify(s · line 589

Quoted 4 lines
drivers/acpi/processor_idle.c +0 -0
 #endif
 		switch (cx->type) {

Janne

reviewer
#8 lore raw

Re: [PATCH] ACPI: do not mark TSC unstable for invalid C-states

applied (with ;; fixed:-)

sorry for causing the regression,
and thanks for quickly isolating it!

Len Brown, Intel Open Source Technology Center

reviewer
#9 lore raw

Re: [PATCH] ACPI: do not mark TSC unstable for invalid C-states

hang on, that patch will fail for the case where valid
is (about to be) set to 1 and we'll fail to disable the TSC for systems which _do_ have c3. I'll send an updated patch shortly.

thanks,
Len Brown, Intel Open Source Technology Center

#10 lore raw

Re: [PATCH] ACPI: do not mark TSC unstable for invalid C-states

Quoted 3 lines · Len Brown

hang on, that patch will fail for the case where valid
is (about to be) set to 1 and we'll fail to disable the TSC for systems which _do_ have c3. I'll send an updated patch shortly.

Yep, you are right. I missed the cx->valid update in apic_processor_power_verify_cX(). So my patch would break the laptops instead of breaking desktops and servers :)

Moving it to acpi_timer_check_state() is the best we can do and that saves us ifdeffery as well as acpi_timer_check_state() is already x86 only.

Thanks,

tglx

reviewer
#11 lore raw

PATCH 1/1 ACPI: idle: fix init-time TSC check regression

From: Len Brown <[email protected]>

A previous 2.6.30 patch, a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2, (ACPI: idle: mark_tsc_unstable() at init-time, not run-time) erroneously disabled the TSC on systems that did not actually have valid deep C-states.

Move the check after the deep-C-states are validated, via new helper, tsc_check_state(), hich replaces tsc_halts_in_c().

---
Thomas,
I decided to use a tsc-specific helper
rather than piggyback on acpi_timer_check_state(), since that is not general, but rather LAPIC timer specific. (and I'll re-name that to reflect its specific function
in a cosmetic follow-up patch)

 drivers/acpi/processor_idle.c |   17 +++++++++--------
 1 files changed, 9 insertions(+), 8 deletions(-)
drivers/acpi/processor_idle.c +9 -8
index f7ca8c5..421cc5c 100644
@@ -216,7 +216,7 @@ int acpi_processor_resume(struct acpi_device * device)
 }
 
 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
-static int tsc_halts_in_c(int state)
+static void tsc_check_state(int state)
 {
 	switch (boot_cpu_data.x86_vendor) {
 	case X86_VENDOR_AMD:
@@ -226,13 +226,17 @@ static int tsc_halts_in_c(int state)
 		 * C/P/S0/S1 states when this bit is set.
 		 */
 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
-			return 0;
+			return;
 
 		/*FALL THROUGH*/
 	default:
-		return state > ACPI_STATE_C1;
+		/* TSC could halt in idle, so notify users */
+		if (state > ACPI_STATE_C1)
+			mark_tsc_unstable("TSC halts in idle");
 	}
 }
+#else
+static void tsc_check_state(int state) { return; }
 #endif
 
 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
@@ -581,11 +585,6 @@ static int acpi_processor_power_verify(struct acpi_processor *pr)
 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 		struct acpi_processor_cx *cx = &pr->power.states[i];
 
-#if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
-		/* TSC could halt in idle, so notify users */
-		if (tsc_halts_in_c(cx->type))
-			mark_tsc_unstable("TSC halts in idle");;
-#endif
 		switch (cx->type) {
 		case ACPI_STATE_C1:
 			cx->valid = 1;
@@ -603,6 +602,8 @@ static int acpi_processor_power_verify(struct acpi_processor *pr)
 				acpi_timer_check_state(i, pr, cx);
 			break;
 		}
+		if (cx->valid)
+			tsc_check_state(cx->type);
 
 		if (cx->valid)
 			working++;
Signature · 1 line
1.6.3.1.9.g95405b
#12 lore raw

Re: [PATCH] ACPI: idle: fix init-time TSC check regression

Quoted 12 lines · Len Brown

From: Len Brown <[email protected]>

A previous 2.6.30 patch, a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2, (ACPI: idle: mark_tsc_unstable() at init-time, not run-time) erroneously disabled the TSC on systems that did not actually have valid deep C-states.

Move the check after the deep-C-states are validated, via new helper, tsc_check_state(), hich replaces tsc_halts_in_c().

Quoted 7 lines

---
Thomas,
I decided to use a tsc-specific helper
rather than piggyback on acpi_timer_check_state(), since that is not general, but rather LAPIC timer specific. (and I'll re-name that to reflect its specific function
in a cosmetic follow-up patch)

I have no strong opinion on that :)

drivers/acpi/processor_idle.c static int acpi_processor_power_verify(struct acpi_processor *pr) · line 609

Quoted 5 lines
drivers/acpi/processor_idle.c +2 -0
+		if (cx->valid)
+			tsc_check_state(cx->type);
 
 		if (cx->valid)
 			working++;

Could do with

      	 if (cx->valid) {
            	tsc_check_state(cx->type);
                working++;
      	 }

Otherwise Acked-by: Thomas Gleixner <[email protected]>

reviewer
#13 lore raw

Re: [PATCH] ACPI: idle: fix init-time TSC check regression

Quoted 8 lines

drivers/acpi/processor_idle.c static int acpi_processor_power_verify(struct acpi_processor *pr) · line 609

Quoted 5 lines
drivers/acpi/processor_idle.c +2 -0
+		if (cx->valid)
+			tsc_check_state(cx->type);
 
 		if (cx->valid)
 			working++;

Could do with

Quoted 4 lines
      	 if (cx->valid) {
            	tsc_check_state(cx->type);
                working++;
      	 }

right, i deferred that to a cosmetic-only patch that will also rename and move acpi_timer_check_state.

Quoted 1 line

Otherwise Acked-by: Thomas Gleixner <[email protected]>

thanks,
Len Brown, Intel Open Source Technology Center

#14 lore raw

Re: [PATCH] ACPI: idle: fix init-time TSC check regression

In reply to Len Brown
Quoted 9 lines · Len Brown

From: Len Brown <[email protected]>

A previous 2.6.30 patch, a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2, (ACPI: idle: mark_tsc_unstable() at init-time, not run-time) erroneously disabled the TSC on systems that did not actually have valid deep C-states.

Move the check after the deep-C-states are validated, via new helper, tsc_check_state(), hich replaces tsc_halts_in_c().

[...]

Quoted 10 lines
drivers/acpi/processor_idle.c +2 -0
@@ -603,6 +602,8 @@ static int acpi_processor_power_verify(struct
 acpi_processor *pr) 
                               acpi_timer_check_state(i, pr, cx);
                       break;
               }
+               if (cx->valid)
+                       tsc_check_state(cx->type);
 
               if (cx->valid)
                       working++;

Shouldn't those last two ifs not simply be combined?
Hmm. I guess not if tsc_check_state can change cx->valid, but that might be worth a comment.

But even with that I'll give the patch a try tomorrow.

Cheers,
FJP

reviewer
#15 lore raw

PATCH 1/1 ACPI: Idle C-states disabled by max_cstate should not disable the TSC

In reply to Len Brown

From: Len Brown <[email protected]>

Processor idle power states C2 and C3 stop the TSC on many machines. Linux recognizes this situation and marks the TSC as unstable:

Marking TSC unstable due to TSC halts in idle

But if those same machines are booted with "processor.max_cstate=1", then there is no need to validate C2 and C3, and no need to disable the TSC, which can be reliably used as a clocksource.

---
 drivers/acpi/processor_idle.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/acpi/processor_idle.c +1 -1
index e39a40a..e65476f 100644
@@ -582,7 +582,7 @@ static int acpi_processor_power_verify(struct acpi_processor *pr)
 
 	pr->power.timer_broadcast_on_state = INT_MAX;
 
-	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
+	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 		struct acpi_processor_cx *cx = &pr->power.states[i];
 
 		switch (cx->type) {
Signature · 1 line
1.6.3.1.9.g95405b
reviewer
#16 lore raw

PATCH 1/1 ACPI: idle: rename lapic timer workaround routines

From: Len Brown <[email protected]>

cosmetic only. The lapic_timer workaround routines are specific to the lapic_timer, and are not acpi-generic.

old:

acpi_timer_check_state()
acpi_propagate_timer_broadcast()
acpi_state_timer_broadcast()

new:

lapic_timer_check_state()
lapic_timer_propagate_broadcast()
lapic_timer_state_broadcast()

also, simplify the code in acpi_processor_power_verify() so that lapic_timer_check_state() is simply called from one place for all valid C-states, including C1.

---
 drivers/acpi/processor_idle.c |   35 ++++++++++++++++-------------------
 1 files changed, 16 insertions(+), 19 deletions(-)
drivers/acpi/processor_idle.c +16 -19
index e65476f..60c3bcd 100644
@@ -139,7 +139,7 @@ static void acpi_safe_halt(void)
  * are affected too. We pick the most conservative approach: we assume
  * that the local APIC stops in both C2 and C3.
  */
-static void acpi_timer_check_state(int state, struct acpi_processor *pr,
+static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 				   struct acpi_processor_cx *cx)
 {
 	struct acpi_processor_power *pwr = &pr->power;
@@ -159,7 +159,7 @@ static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 		pr->power.timer_broadcast_on_state = state;
 }
 
-static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
+static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
 {
 	unsigned long reason;
 
@@ -170,7 +170,7 @@ static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
 }
 
 /* Power(C) State timer broadcast control */
-static void acpi_state_timer_broadcast(struct acpi_processor *pr,
+static void lapic_timer_state_broadcast(struct acpi_processor *pr,
 				       struct acpi_processor_cx *cx,
 				       int broadcast)
 {
@@ -187,10 +187,10 @@ static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 
 #else
 
-static void acpi_timer_check_state(int state, struct acpi_processor *pr,
+static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 				   struct acpi_processor_cx *cstate) { }
-static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
-static void acpi_state_timer_broadcast(struct acpi_processor *pr,
+static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
+static void lapic_timer_state_broadcast(struct acpi_processor *pr,
 				       struct acpi_processor_cx *cx,
 				       int broadcast)
 {
@@ -592,24 +592,21 @@ static int acpi_processor_power_verify(struct acpi_processor *pr)
 
 		case ACPI_STATE_C2:
 			acpi_processor_power_verify_c2(cx);
-			if (cx->valid)
-				acpi_timer_check_state(i, pr, cx);
 			break;
 
 		case ACPI_STATE_C3:
 			acpi_processor_power_verify_c3(pr, cx);
-			if (cx->valid)
-				acpi_timer_check_state(i, pr, cx);
 			break;
 		}
-		if (cx->valid)
-			tsc_check_state(cx->type);
+		if (!cx->valid)
+			continue;
 
-		if (cx->valid)
-			working++;
+		lapic_timer_check_state(i, pr, cx);
+		tsc_check_state(cx->type);
+		working++;
 	}
 
-	acpi_propagate_timer_broadcast(pr);
+	lapic_timer_propagate_broadcast(pr);
 
 	return (working);
 }
@@ -863,7 +860,7 @@ static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 	 * Must be done before busmaster disable as we might need to
 	 * access HPET !
 	 */
-	acpi_state_timer_broadcast(pr, cx, 1);
+	lapic_timer_state_broadcast(pr, cx, 1);
 
 	if (cx->type == ACPI_STATE_C3)
 		ACPI_FLUSH_CPU_CACHE();
@@ -885,7 +882,7 @@ static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 
 	cx->usage++;
 
-	acpi_state_timer_broadcast(pr, cx, 0);
+	lapic_timer_state_broadcast(pr, cx, 0);
 	cx->time += sleep_ticks;
 	return idle_time;
 }
@@ -952,7 +949,7 @@ static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 	 * Must be done before busmaster disable as we might need to
 	 * access HPET !
 	 */
-	acpi_state_timer_broadcast(pr, cx, 1);
+	lapic_timer_state_broadcast(pr, cx, 1);
 
 	kt1 = ktime_get_real();
 	/*
@@ -997,7 +994,7 @@ static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 
 	cx->usage++;
 
-	acpi_state_timer_broadcast(pr, cx, 0);
+	lapic_timer_state_broadcast(pr, cx, 0);
 	cx->time += sleep_ticks;
 	return idle_time;
 }
Signature · 1 line
1.6.3.1.9.g95405b
Acked-by on v1 1/1
#17 lore raw

Re: [PATCH] ACPI: idle: rename lapic timer workaround routines

Quoted 17 lines · Len Brown

From: Len Brown <[email protected]>

cosmetic only. The lapic_timer workaround routines are specific to the lapic_timer, and are not acpi-generic.

old:

acpi_timer_check_state()
acpi_propagate_timer_broadcast()
acpi_state_timer_broadcast()

new:

lapic_timer_check_state()
lapic_timer_propagate_broadcast()
lapic_timer_state_broadcast()

Quoted 5 lines

also, simplify the code in acpi_processor_power_verify() so that lapic_timer_check_state() is simply called from one place for all valid C-states, including C1.

Acked-by on v1 1/1
#18 lore raw

Re: [PATCH] ACPI: Idle C-states disabled by max_cstate should not disable the TSC

In reply to Len Brown
Quoted 5 lines · Len Brown

From: Len Brown <[email protected]>

Processor idle power states C2 and C3 stop the TSC on many machines. Linux recognizes this situation and marks the TSC as unstable:

Quoted 7 lines

Marking TSC unstable due to TSC halts in idle

But if those same machines are booted with "processor.max_cstate=1", then there is no need to validate C2 and C3, and no need to disable the TSC, which can be reliably used as a clocksource.

Quoted 20 lines
---
 drivers/acpi/processor_idle.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/acpi/processor_idle.c +1 -1
index e39a40a..e65476f 100644
@@ -582,7 +582,7 @@ static int acpi_processor_power_verify(struct acpi_processor *pr)
 
 	pr->power.timer_broadcast_on_state = INT_MAX;
 
-	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
+	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 		struct acpi_processor_cx *cx = &pr->power.states[i];
 
 		switch (cx->type) {
Signature · 1 line
1.6.3.1.9.g95405b
Tested-by on v1 1/1
#19 lore raw

Re: [PATCH] ACPI: idle: fix init-time TSC check regression

In reply to Len Brown
Quoted 7 lines · Len Brown

From: Len Brown <[email protected]>

A previous 2.6.30 patch, a71e4917dc0ebbcb5a0ecb7ca3486643c1c9a6e2, (ACPI: idle: mark_tsc_unstable() at init-time, not run-time) erroneously disabled the TSC on systems that did not actually have valid deep C-states.

Quoted 5 lines

Move the check after the deep-C-states are validated, via new helper, tsc_check_state(), hich replaces tsc_halts_in_c().

OK on both my desktop (which had the regression) and laptop.

(Tested together with the 2 patches you sent later.)

Cheers,
FJP