powerpc/cell: disable ptcal in case of kdump reboot

PATCH In review Arnd Bergmann 3 participants 7 messages Jun 11, 2008 – Jun 12, 2008

Subsystems (2)

PowerPC Cell broadband engine architecture

lore

Share & tools

lore

Arnd Bergmann

author

2008-06-11 16:28 UTC #0 lore raw

PATCH 1/1 powerpc/cell: disable ptcal in case of kdump reboot

The ibm,cbe-start-ptcal rtas call transfers ownership of a memory range to the hardware in order to do periodic memory interface recalibration.
When we do a kexec, we need to get this page back, so we don't step on the new kernel's toes.
We used to do this correctly in case of a user triggered kexec, but not for kdump.

This patch disables ptcal from the cell specific machine_crash_shutdown() callback. I also needed to fix the pseries code that tries to install its default_machine_crash_shutdown() code independent of what machine we're running on.

The fix has been verified by Akira Tsukamoto, the original reporter of the problem.

Acked-by Akira Tsukamoto <[email protected]>

Signed-off-by Arnd Bergmann <[email protected]>

---
 arch/powerpc/platforms/cell/ras.c      |    6 ++++++
 arch/powerpc/platforms/cell/ras.h      |    1 +
 arch/powerpc/platforms/cell/setup.c    |    2 +-
 arch/powerpc/platforms/pseries/kexec.c |    2 +-
 4 files changed, 9 insertions(+), 2 deletions(-)

arch/powerpc/platforms/cell/ras.c +6 -0

index 655704a..21b313d 100644
@@ -226,6 +226,12 @@ static int cbe_ptcal_notify_reboot(struct notifier_block *nb,
 	return cbe_ptcal_disable();
 }
 
+void cbe_machine_crash_shutdown(struct pt_regs *regs)
+{
+	cbe_ptcal_disable();
+	return default_machine_crash_shutdown(regs);
+}
+
 static struct notifier_block cbe_ptcal_reboot_notifier = {
 	.notifier_call = cbe_ptcal_notify_reboot
 };

arch/powerpc/platforms/cell/ras.h +1 -0

index eb7ee54..d1401f9 100644
@@ -5,5 +5,6 @@ extern void cbe_system_error_exception(struct pt_regs *regs);
 extern void cbe_maintenance_exception(struct pt_regs *regs);
 extern void cbe_thermal_exception(struct pt_regs *regs);
 extern void cbe_ras_init(void);
+extern void cbe_machine_crash_shutdown(struct pt_regs *regs);
 
 #endif /* RAS_H */

arch/powerpc/platforms/cell/setup.c +1 -1

index ab721b5..b555030 100644
@@ -292,6 +292,6 @@ define_machine(cell) {
 #ifdef CONFIG_KEXEC
 	.machine_kexec		= default_machine_kexec,
 	.machine_kexec_prepare	= default_machine_kexec_prepare,
-	.machine_crash_shutdown	= default_machine_crash_shutdown,
+	.machine_crash_shutdown	= cbe_machine_crash_shutdown,
 #endif
 };

arch/powerpc/platforms/pseries/kexec.c +1 -1

index e9dd5fe..53cbd53 100644
@@ -70,4 +70,4 @@ static int __init pseries_kexec_setup(void)
 
 	return 0;
 }
-__initcall(pseries_kexec_setup);
+machine_device_initcall(pseries, pseries_kexec_setup);

Signature · 1 line

1.5.4.3

Arnd Bergmann

author

2008-06-11 16:34 UTC #1 lore raw

please pull merge branch of cell-2.6.git

Hi Paul,

Please pull the fix I just sent from

master.kernel.org/pub/scm/linux/kernel/git/arnd/cell-2.6.git merge

Jeremy might want to look at it as well, since he wrote the code originally.

Thanks,

	Arnd <><

---
Arnd Bergmann <[email protected]>:
powerpc/cell: disable ptcal in case of kdump reboot

 arch/powerpc/platforms/cell/ras.c      |    6 ++++++
 arch/powerpc/platforms/cell/ras.h      |    1 +
 arch/powerpc/platforms/cell/setup.c    |    2 +-
 arch/powerpc/platforms/pseries/kexec.c |    2 +-
 4 files changed, 9 insertions(+), 2 deletions(-)

Michael Ellerman

2008-06-11 16:49 UTC #2 lore raw

In reply to Arnd Bergmann

Quoted 7 lines · Arnd Bergmann

The ibm,cbe-start-ptcal rtas call transfers ownership of a memory range to the hardware in order to do periodic memory interface recalibration.
When we do a kexec, we need to get this page back, so we don't step on the new kernel's toes.
We used to do this correctly in case of a user triggered kexec, but not for kdump.

Used to?

Quoted 2 lines

This patch disables ptcal from the cell specific machine_crash_shutdown() callback.

I'd rather you used the crash_shutdown_register() infrastructure Mikey added for EHEA - it's designed for just this sort of thing.

Quoted 3 lines

I also needed to fix the pseries code that tries to install its default_machine_crash_shutdown() code independent
of what machine we're running on.

That's a buglet.

cheers

Signature · 8 lines

Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person

Arnd Bergmann

author

2008-06-12 07:39 UTC #3 lore raw

Quoted 5 lines · Michael Ellerman

Quoted 2 lines · Arnd Bergmann
We used to do this correctly in case of a user triggered kexec, but not for kdump.
Used to?

Sorry, wrong wording. I meant without this patch, it's correct for kexec.

Quoted 5 lines

Quoted 2 lines
This patch disables ptcal from the cell specific
machine_crash_shutdown() callback.
I'd rather you used the crash_shutdown_register() infrastructure Mikey added for EHEA - it's designed for just this sort of thing.

Thanks for pointing this out, I wasn't aware of that call. New patch coming up.

	Arnd <><

Michael Neuling

2008-06-12 08:06 UTC #4 lore raw

Quoted 10 lines · In message <[email protected]> you wrote:

Quoted 5 lines · Michael Ellerman
Quoted 2 lines · Arnd Bergmann
We used to do this correctly in case of a user triggered kexec, but not for kdump.
Used to?
Sorry, wrong wording. I meant without this patch, it's correct for kexec.

Quoted 8 lines

Quoted 5 lines
Quoted 2 lines
This patch disables ptcal from the cell specific
machine_crash_shutdown() callback.
I'd rather you used the crash_shutdown_register() infrastructure Mikey added for EHEA - it's designed for just this sort of thing.
Thanks for pointing this out, I wasn't aware of that call. New patch coming up.

In theory we can register lots of call backs but currently the maximum is set to 1 through CRASH_HANDLER_MAX. This is to discourage people from using this infrastructure at all.

Unless there's some crazy arse company that's going to mix a CBE with an eHEA, we are probably ok leaving this at 1. Right?

Mikey

Michael Ellerman

2008-06-12 09:04 UTC #5 lore raw

Quoted 17 lines · Michael Neuling

Quoted 16 lines · In message <[email protected]> you wrote:
Quoted 5 lines · Michael Ellerman
Quoted 2 lines · Arnd Bergmann
We used to do this correctly in case of a user triggered kexec, but not for kdump.
Used to?
Sorry, wrong wording. I meant without this patch, it's correct for kexec.
Quoted 5 lines
Quoted 2 lines
This patch disables ptcal from the cell specific
machine_crash_shutdown() callback.
I'd rather you used the crash_shutdown_register() infrastructure Mikey added for EHEA - it's designed for just this sort of thing.

Quoted 6 lines

Quoted 2 lines
Thanks for pointing this out, I wasn't aware of that call. New patch coming up.
In theory we can register lots of call backs but currently the maximum is set to 1 through CRASH_HANDLER_MAX. This is to discourage people from using this infrastructure at all.

Yeah true. In this case I think it's the less horrible solution to something that's fundamentally kdump-unfriendly.

Quoted 2 lines

Unless there's some crazy arse company that's going to mix a CBE with an eHEA, we are probably ok leaving this at 1. Right?

No comment *cough*. I guess I'd say just increase it to 2 on the off chance, can't hurt much.

cheers

Signature · 8 lines

Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person

Michael Neuling

2008-06-12 11:22 UTC #6 lore raw

Quoted 27 lines · In message <1213261480.8652.5.camel@localhost> you wrote:

Quoted 7 lines · Michael Neuling
Quoted 6 lines · In message <[email protected]> you wrote:
Quoted 5 lines · Michael Ellerman
Quoted 2 lines · Arnd Bergmann
We used to do this correctly in case of a user triggered kexec, but not for kdump.
=20
Used to?
Quoted 8 lines
Quoted 8 lines
=20
Sorry, wrong wording. I meant without this patch, it's correct for kexec.
=20
Quoted 4 lines
Quoted 2 lines
This patch disables ptcal from the cell specific
machine_crash_shutdown() callback.=20
=20
I'd rather you used the crash_shutdown_register() infrastructure Mike=
y
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
added for EHEA - it's designed for just this sort of thing.
=20
Thanks for pointing this out, I wasn't aware of that call. New patch coming up.
=20
In theory we can register lots of call backs but currently the maximum is set to 1 through CRASH_HANDLER_MAX. This is to discourage people from using this infrastructure at all.

Quoted 8 lines

Yeah true. In this case I think it's the less horrible solution to something that's fundamentally kdump-unfriendly.
Quoted 2 lines
Unless there's some crazy arse company that's going to mix a CBE with an eHEA, we are probably ok leaving this at 1. Right?
No comment *cough*. I guess I'd say just increase it to 2 on the off chance, can't hurt much.

Ok.. but this one really really is the last one!! :-)

Mikey