Skip to contentSkip to conversation

ARM: smp: Allow real broadcast device selection instead of always dummy

PATCH New Santosh Shilimkar 4 participants 16 messages Mar 13, 2013 – Mar 14, 2013 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

 arch/arm/kernel/smp.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/smp.c +1 -1
index 31644f1..79078ed 100644
@@ -480,7 +480,7 @@ static void __cpuinit broadcast_timer_setup(struct clock_event_device *evt)
 	evt->features	= CLOCK_EVT_FEAT_ONESHOT |
 			  CLOCK_EVT_FEAT_PERIODIC |
 			  CLOCK_EVT_FEAT_DUMMY;
-	evt->rating	= 400;
+	evt->rating	= 100;
 	evt->mult	= 1;
 	evt->set_mode	= broadcast_timer_set_mode;
 
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

(Forgot to CC Thomas)

Quoted 12 lines · Santosh Shilimkar

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Quoted 9 lines
Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

Some one pointed me to a fix made by Mark which was discussed under '[BUG] ARM Architected timers appear broken in 3.9-rc1' subject. That patch seems to be more of work around since the root of the problem is incorrect dummy timer rating. Either way, both patches fix the issue.

Quoted 17 lines
 arch/arm/kernel/smp.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/smp.c +1 -1
index 31644f1..79078ed 100644
@@ -480,7 +480,7 @@ static void __cpuinit broadcast_timer_setup(struct clock_event_device *evt)
 	evt->features	= CLOCK_EVT_FEAT_ONESHOT |
 			  CLOCK_EVT_FEAT_PERIODIC |
 			  CLOCK_EVT_FEAT_DUMMY;
-	evt->rating	= 400;
+	evt->rating	= 100;
 	evt->mult	= 1;
 	evt->set_mode	= broadcast_timer_set_mode;
 
#2 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Hi Santosh,

Quoted 24 lines · Santosh Shilimkar

(Forgot to CC Thomas)

Quoted 12 lines · Santosh Shilimkar

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Quoted 9 lines
Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

Quoted 5 lines

Some one pointed me to a fix made by Mark which was discussed under '[BUG] ARM Architected timers appear broken in 3.9-rc1' subject. That patch seems to be more of work around since the root of the problem is incorrect dummy timer rating. Either way, both patches fix the issue.

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

If it is the former, Then I believe my patch solve the issue more generally - if you happen to register a dummy timer before other timers, it will become the broadcast source. Regardless of how temporary this is, it should never happen, and lowering the rating of the dummy won't fix this.

If it is the latter, then this patch would ensure that a real timer with a rating over 100 is selected in preference to the dummy, which is certainly what we want. The proposed generic dummy timer in Stephen Boyd's local timer API removal series [1] similarly uses a low rating to ensure that real timers are selected in preference to an always-registered dummy. I note that the architected timer has a higher rating (450) than the dummy (400), so this shouldn't currently be a problem.

Have I missed something?

[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-March/153208.html

Thanks,
Mark.

#3 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 27 lines · Mark Rutland

Hi Santosh,

Quoted 24 lines · Santosh Shilimkar

(Forgot to CC Thomas)

Quoted 21 lines · Santosh Shilimkar

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

Quoted 9 lines
Quoted 5 lines

Some one pointed me to a fix made by Mark which was discussed under '[BUG] ARM Architected timers appear broken in 3.9-rc1' subject. That patch seems to be more of work around since the root of the problem is incorrect dummy timer rating. Either way, both patches fix the issue.

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Quoted 5 lines

If it is the former, Then I believe my patch solve the issue more generally - if you happen to register a dummy timer before other timers, it will become the broadcast source. Regardless of how temporary this is, it should never happen, and lowering the rating of the dummy won't fix this.

Well by the time we need active broadcast functionality, clock-events are already chosen if the ratings are appropriate.

Quoted 8 lines

If it is the latter, then this patch would ensure that a real timer with a rating over 100 is selected in preference to the dummy, which is certainly what we want. The proposed generic dummy timer in Stephen Boyd's local timer API removal series [1] similarly uses a low rating to ensure that real timers are selected in preference to an always-registered dummy. I note that the architected timer has a higher rating (450) than the dummy (400), so this shouldn't currently be a problem.

Because we always register dummy broadcast on ARM now
and with higher rating, it is picked as broadcast source. We definitely don't want such a behavior when we have real broadcast device is available.

With your patch, we are trying to avoid the registration which goes against the the whole idea of registering it always and picking the right clock-event based on rating by clock-event core. The clock-event core except the proper ratings to be provided based on the capability of the timer source and its resolution and in that case dummy should have the lowest rating which is what I tried to patch.

Quoted 2 lines

Have I missed something?

Not much. I was just looking at x86 code as well after your email to see how the LAPIC issue is handled. They seems to also have correct rating to take care of the selection.

Probably we can merge both the fixes but from clock-event core code perspective, the ratings fix is more than enough.

Thomas might be able to comment better.

Regards,
Santosh

#4 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 34 lines · Santosh Shilimkar
Quoted 33 lines · Mark Rutland

Hi Santosh,

Quoted 29 lines · Santosh Shilimkar

(Forgot to CC Thomas)

Quoted 12 lines · Santosh Shilimkar

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Quoted 9 lines
Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

Some one pointed me to a fix made by Mark which was discussed under '[BUG] ARM Architected timers appear broken in 3.9-rc1' subject. That patch seems to be more of work around since the root of the problem is incorrect dummy timer rating. Either way, both patches fix the issue.

Quoted 4 lines
Quoted 3 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

Quoted 8 lines
Quoted 6 lines

If it is the former, Then I believe my patch solve the issue more generally - if you happen to register a dummy timer before other timers, it will become the broadcast source. Regardless of how temporary this is, it should never happen, and lowering the rating of the dummy won't fix this.

Well by the time we need active broadcast functionality, clock-events are already chosen if the ratings are appropriate.

That's a fair point. I still think it's worth having the check in the core broadcast code - rejecting dummy timers is always a sensible thing to do, and it prevents future crashes if new timers are added without sensible rating values.

Quoted 9 lines
Quoted 9 lines

If it is the latter, then this patch would ensure that a real timer with a rating over 100 is selected in preference to the dummy, which is certainly what we want. The proposed generic dummy timer in Stephen Boyd's local timer API removal series [1] similarly uses a low rating to ensure that real timers are selected in preference to an always-registered dummy. I note that the architected timer has a higher rating (450) than the dummy (400), so this shouldn't currently be a problem.

Quoted 3 lines

Because we always register dummy broadcast on ARM now
and with higher rating, it is picked as broadcast source. We definitely don't want such a behavior when we have real broadcast device is available.

Agreed. We *never* want to pick a dummy timer as a broadcast source, as this never makes sense.

Quoted 7 lines

With your patch, we are trying to avoid the registration which goes against the the whole idea of registering it always and picking the right clock-event based on rating by clock-event core. The clock-event core except the proper ratings to be provided based on the capability of the timer source and its resolution and in that case dummy should have the lowest rating which is what I tried to patch.

Quoted 8 lines
Quoted 2 lines

Have I missed something?

Not much. I was just looking at x86 code as well after your email to see how the LAPIC issue is handled. They seems to also have correct rating to take care of the selection.

Probably we can merge both the fixes but from clock-event core code perspective, the ratings fix is more than enough.

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Thomas has already pulled my fix into tip timers/urgent [1].

[1] https://git.kernel.org/cgit/linux/kernel/git/tip/tip.git/commit/?h=timers/urgent&id=a7dc19b8652c862d5b7c4d2339bd3c428bd29c4a

Thanks,
Mark.

#5 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 3 lines · Mark Rutland
Quoted 2 lines · Santosh Shilimkar
Quoted 1 line · Mark Rutland

Hi Santosh,

[..]

Quoted 51 lines
Quoted 5 lines
Quoted 4 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

Quoted 8 lines
Quoted 6 lines

If it is the former, Then I believe my patch solve the issue more generally - if you happen to register a dummy timer before other timers, it will become the broadcast source. Regardless of how temporary this is, it should never happen, and lowering the rating of the dummy won't fix this.

Well by the time we need active broadcast functionality, clock-events are already chosen if the ratings are appropriate.

That's a fair point. I still think it's worth having the check in the core broadcast code - rejecting dummy timers is always a sensible thing to do, and it prevents future crashes if new timers are added without sensible rating values.

Quoted 9 lines
Quoted 9 lines

If it is the latter, then this patch would ensure that a real timer with a rating over 100 is selected in preference to the dummy, which is certainly what we want. The proposed generic dummy timer in Stephen Boyd's local timer API removal series [1] similarly uses a low rating to ensure that real timers are selected in preference to an always-registered dummy. I note that the architected timer has a higher rating (450) than the dummy (400), so this shouldn't currently be a problem.

Quoted 3 lines

Because we always register dummy broadcast on ARM now
and with higher rating, it is picked as broadcast source. We definitely don't want such a behavior when we have real broadcast device is available.

Agreed. We *never* want to pick a dummy timer as a broadcast source, as this never makes sense.

Quoted 13 lines

With your patch, we are trying to avoid the registration which goes against the the whole idea of registering it always and picking the right clock-event based on rating by clock-event core. The clock-event core except the proper ratings to be provided based on the capability of the timer source and its resolution and in that case dummy should have the lowest rating which is what I tried to patch.

Quoted 2 lines

Have I missed something?

Not much. I was just looking at x86 code as well after your email to see how the LAPIC issue is handled. They seems to also have correct rating to take care of the selection.

Quoted 6 lines
Quoted 2 lines

Probably we can merge both the fixes but from clock-event core code perspective, the ratings fix is more than enough.

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Regards
Santosh

#6 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 14 lines · Santosh Shilimkar
Quoted 3 lines · Mark Rutland
Quoted 2 lines · Santosh Shilimkar
Quoted 1 line · Mark Rutland

Hi Santosh,

[..]

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

[...]

Quoted 5 lines
Quoted 3 lines

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Sure. Though could you reword the commit message? The patch solves the more general issue of a dummy being preferred over real hardware even outside of choosing the broadcast device.

#7 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 21 lines · Mark Rutland
Quoted 14 lines · Santosh Shilimkar
Quoted 3 lines · Mark Rutland
Quoted 2 lines · Santosh Shilimkar
Quoted 1 line · Mark Rutland

Hi Santosh,

[..]

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

[...]

Quoted 3 lines
Quoted 3 lines

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Quoted 8 lines
Quoted 2 lines

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Sure. Though could you reword the commit message? The patch solves the more general issue of a dummy being preferred over real hardware even outside of choosing the broadcast device.

Thanks. For record, patch is in end of the email which I plan to put into patch system.

Regards,
Santosh

Quoted 1 line

From 57c501bcdc88c7ff26a5c63956be07e94a5083c5 Mon Sep 17 00:00:00 2001

From: Santosh Shilimkar <[email protected]>
Date: Wed, 13 Mar 2013 12:33:16 +0530
Subject: [PATCH 1/2] ARM: smp: Avoid dummy clockevent being preferred over real hardware clock-event

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it may be preferred over a real clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real clockevent device is selected when available.

---
 arch/arm/kernel/smp.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/smp.c +1 -1
index 31644f1..79078ed 100644
@@ -480,7 +480,7 @@ static void __cpuinit broadcast_timer_setup(struct clock_event_device *evt)
 	evt->features	= CLOCK_EVT_FEAT_ONESHOT |
 			  CLOCK_EVT_FEAT_PERIODIC |
 			  CLOCK_EVT_FEAT_DUMMY;
-	evt->rating	= 400;
+	evt->rating	= 100;
 	evt->mult	= 1;
 	evt->set_mode	= broadcast_timer_set_mode;
 
Signature · 1 line
1.7.9.5
#8 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 47 lines · Santosh Shilimkar
Quoted 21 lines · Mark Rutland
Quoted 14 lines · Santosh Shilimkar
Quoted 3 lines · Mark Rutland
Quoted 2 lines · Santosh Shilimkar
Quoted 1 line · Mark Rutland

Hi Santosh,

[..]

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

[...]

Quoted 3 lines
Quoted 3 lines

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Quoted 8 lines
Quoted 2 lines

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Sure. Though could you reword the commit message? The patch solves the more general issue of a dummy being preferred over real hardware even outside of choosing the broadcast device.

Thanks. For record, patch is in end of the email which I plan to put into patch system.

Regards,
Santosh

Quoted 1 line

From 57c501bcdc88c7ff26a5c63956be07e94a5083c5 Mon Sep 17 00:00:00 2001

From: Santosh Shilimkar <[email protected]>
Date: Wed, 13 Mar 2013 12:33:16 +0530
Subject: [PATCH 1/2] ARM: smp: Avoid dummy clockevent being preferred over real hardware clock-event

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it may be preferred over a real clock event.

Quoted 5 lines

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real clockevent device is selected when available.

Quoted 23 lines
---
 arch/arm/kernel/smp.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/smp.c +1 -1
index 31644f1..79078ed 100644
@@ -480,7 +480,7 @@ static void __cpuinit broadcast_timer_setup(struct clock_event_device *evt)
 	evt->features	= CLOCK_EVT_FEAT_ONESHOT |
 			  CLOCK_EVT_FEAT_PERIODIC |
 			  CLOCK_EVT_FEAT_DUMMY;
-	evt->rating	= 400;
+	evt->rating	= 100;
 	evt->mult	= 1;
 	evt->set_mode	= broadcast_timer_set_mode;
 
Signature · 1 line
1.7.9.5
#9 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

In reply to Santosh Shilimkar
Quoted 29 lines · Santosh Shilimkar
Quoted 28 lines · Mark Rutland
Quoted 14 lines · Santosh Shilimkar
Quoted 3 lines · Mark Rutland
Quoted 2 lines · Santosh Shilimkar
Quoted 1 line · Mark Rutland

Hi Santosh,

[..]

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

[...]

Quoted 5 lines
Quoted 3 lines

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Sure. Though could you reword the commit message? The patch solves the more general issue of a dummy being preferred over real hardware even outside of choosing the broadcast device.

Quoted 7 lines
Quoted 1 line

Thanks. For record, patch is in end of the email which I plan to put into patch system.

Regards,
Santosh

The below patch seems fine. Are you intending for this to go in as a fix for 3.9-rc*, or as a cleanup for 3.10?

If you're aiming for the latter, it's going to clash with Stephen Boyd's local timer API removal [1], in which the generic dummy timer driver [2] (replacing the arm-specific dummy [3]) also has a rating of 100.

It would be nice if we could reduce the possibility of a conflict.

[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-March/154724.html
[2] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-March/154725.html
[3] http://lists.infradead.org/pipermail/linux-arm-kernel/2013-March/154726.html

Thanks,
Mark.

Quoted 39 lines

From 57c501bcdc88c7ff26a5c63956be07e94a5083c5 Mon Sep 17 00:00:00 2001
From: Santosh Shilimkar <[email protected]>
Date: Wed, 13 Mar 2013 12:33:16 +0530
Subject: [PATCH 1/2] ARM: smp: Avoid dummy clockevent being preferred over real hardware clock-event

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it may be preferred over a real clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real clockevent device is selected when available.

---
 arch/arm/kernel/smp.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/smp.c +1 -1
index 31644f1..79078ed 100644
@@ -480,7 +480,7 @@ static void __cpuinit broadcast_timer_setup(struct clock_event_device *evt)
 	evt->features	= CLOCK_EVT_FEAT_ONESHOT |
 			  CLOCK_EVT_FEAT_PERIODIC |
 			  CLOCK_EVT_FEAT_DUMMY;
-	evt->rating	= 400;
+	evt->rating	= 100;
 	evt->mult	= 1;
 	evt->set_mode	= broadcast_timer_set_mode;
 
Signature · 1 line
1.7.9.5
#10 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 32 lines · Mark Rutland
Quoted 31 lines · Santosh Shilimkar
Quoted 29 lines · Mark Rutland
Quoted 14 lines · Santosh Shilimkar
Quoted 3 lines · Mark Rutland
Quoted 2 lines · Santosh Shilimkar
Quoted 1 line · Mark Rutland

Hi Santosh,

[..]

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

Is the problem that the dummy timer is being registered as the broadcast source, or that it is selected as a local timer in preference of real timers?

Dummy timer is preferred over real broadcast timer.

Ok.

[...]

Quoted 5 lines
Quoted 3 lines

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Sure. Though could you reword the commit message? The patch solves the more general issue of a dummy being preferred over real hardware even outside of choosing the broadcast device.

Quoted 9 lines
Quoted 5 lines

Thanks. For record, patch is in end of the email which I plan to put into patch system.

Regards,
Santosh

The below patch seems fine. Are you intending for this to go in as a fix for 3.9-rc*, or as a cleanup for 3.10?

Fix for 3.9-rc*. I have put already it in patch system.

Regards,
Santosh

#11 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

In reply to Mark Rutland
Quoted 17 lines · Mark Rutland
Quoted 16 lines · Santosh Shilimkar
Quoted 5 lines · Mark Rutland
Quoted 4 lines · Santosh Shilimkar
Quoted 3 lines · Mark Rutland

I do agree it'd be worth lowering the dummy timer's rating to ensure it doesn't override a real timer elsewhere.

Quoted 7 lines
Quoted 2 lines

Yep. Can I add you acked-by tag then for $subject patch ? Would be good to get this one merged as well.

Sure. Though could you reword the commit message? The patch solves the more general issue of a dummy being preferred over real hardware even outside of choosing the broadcast device.

Thanks. For record, patch is in end of the email which I plan to put into patch system.

Quoted 8 lines
Quoted 2 lines

Regards,
Santosh

The below patch seems fine. Are you intending for this to go in as a fix for 3.9-rc*, or as a cleanup for 3.10?

If you're aiming for the latter, it's going to clash with Stephen Boyd's local timer API removal [1], in which the generic dummy timer driver [2] (replacing the arm-specific dummy [3]) also has a rating of 100.

Thanks for the heads up. Looks like the conflict will be trivial, but I wonder why need to put the patch at all? Per my understanding the regression has been fixed by your patch in Thomas' tree and then in 3.10 we can merge the local timer removal patches and fix up the rating at the same time.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
#12 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

In reply to Santosh Shilimkar
Quoted 22 lines · Santosh Shilimkar
Quoted 12 lines · Santosh Shilimkar

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Quoted 9 lines
Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

Quoted 5 lines

Some one pointed me to a fix made by Mark which was discussed under '[BUG] ARM Architected timers appear broken in 3.9-rc1' subject. That patch seems to be more of work around since the root of the problem is incorrect dummy timer rating. Either way, both patches fix the issue.

Well, using a dummy timer as the broadcast event device is a bug, no matter what the rating is. The fix is in Linus tree already.

Though making the rating of the dummy lower is definitely a good thing, so a real hardware device which is detected later can replace the dummy device. So yes, the rating should be low for the dummy timer.

Thanks,

tglx

#13 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 23 lines · Thomas Gleixner
Quoted 22 lines · Santosh Shilimkar
Quoted 21 lines · Santosh Shilimkar

With recent arm broadcast time clean-up from Mark Rutland, the dummy broadcast device is always registered with timer subsystem. And since the rating of the dummy clock event is very high, it is preferred over a real broad-cast clock event.

This is a change in behavior from past and not an intended one. So reduce the rating of the dummy clockevent so that real broadcast device is selected when available.

Without this all the C states with C3STOP won't work since the broad cast notifier will take an abort.

Cc Mark Rutland <[email protected]>
Cc Russell King <[email protected]>

---
Its a regression so hopefully can get into the 3.9-rcx. Noticed this one on A15 platform. A9 platform the issue may not be seen since the local timer check avoids dummy timer registration.

Quoted 9 lines
Quoted 5 lines

Some one pointed me to a fix made by Mark which was discussed under '[BUG] ARM Architected timers appear broken in 3.9-rc1' subject. That patch seems to be more of work around since the root of the problem is incorrect dummy timer rating. Either way, both patches fix the issue.

Well, using a dummy timer as the broadcast event device is a bug, no matter what the rating is. The fix is in Linus tree already.

Agree.

Quoted 5 lines

Though making the rating of the dummy lower is definitely a good thing, so a real hardware device which is detected later can replace the dummy device. So yes, the rating should be low for the dummy timer.

Exactly. As Mark pointed out, you have already applied Mark's patch. I was just wondering whether you can take the $subject patch as well with ack from Russell, Mark.

Regards,
Santosh

#14 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 9 lines · Santosh Shilimkar
Quoted 5 lines · Thomas Gleixner

Though making the rating of the dummy lower is definitely a good thing, so a real hardware device which is detected later can replace the dummy device. So yes, the rating should be low for the dummy timer.

Exactly. As Mark pointed out, you have already applied Mark's patch. I was just wondering whether you can take the $subject patch as well with ack from Russell, Mark.

It can go through the ARM tree as well, no dependencies on my stuff.

Thanks,

tglx

#15 lore raw

PATCH 1/1 ARM: smp: Allow real broadcast device selection instead of always dummy

Quoted 7 lines · Thomas Gleixner
Quoted 6 lines · Santosh Shilimkar
Quoted 5 lines · Thomas Gleixner

Though making the rating of the dummy lower is definitely a good thing, so a real hardware device which is detected later can replace the dummy device. So yes, the rating should be low for the dummy timer.

Quoted 6 lines
Quoted 3 lines

Exactly. As Mark pointed out, you have already applied Mark's patch. I was just wondering whether you can take the $subject patch as well with ack from Russell, Mark.

It can go through the ARM tree as well, no dependencies on my stuff.

Sure. I will push that through ARM tree.

Regards,
Santosh