[PATCH 2/2] kdump shutdown hook support

STALE6784d

27 messages, 5 authors, 2008-01-18 · open the first message on its own page

[PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2007-12-12 05:45:12

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.

This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable.  

I've tested with bogus data and instruction exceptions.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/crash.c |   90 +++++++++++++++++++++++++++++++++++++++++---
 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 88 insertions(+), 5 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/crash.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/crash.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_SHUTDOWN_HANDLES_NUM 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_SHUTDOWN_HANDLES_NUM+1];
+static DEFINE_SPINLOCK(crash_handles_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,69 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/* 
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i;
+
+	spin_lock(&crash_handles_lock);
+	for(i = 0 ; i <= CRASH_SHUTDOWN_HANDLES_NUM; i++)
+		if (!crash_shutdown_handles[i])
+			break;
+
+	if (i == CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		spin_unlock(&crash_handles_lock);
+		return 1;
+	}
+
+	/* Insert handle at end */
+	crash_shutdown_handles[i] = handler;
+	spin_unlock(&crash_handles_lock);
+	return 0;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i;
+
+	spin_lock(&crash_handles_lock);
+	for(i = 0 ; i <= CRASH_SHUTDOWN_HANDLES_NUM; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		spin_unlock(&crash_handles_lock);
+		return 1;
+	}
+
+	/* Shift handles down */
+	while(crash_shutdown_handles[i]) {
+		crash_shutdown_handles[i] = crash_shutdown_handles[i+1];
+		i++;
+	}
+	spin_unlock(&crash_handles_lock);
+	return 0;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static long crash_shutdown_buf[SETJMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
 
 	/*
 	 * This function is only called after the system

@@ -301,14 +368,27 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/* Call registered shutdown routines */
+	__debugger_fault_handler = handle_fault;
+	i = 0;
+	while(crash_shutdown_handles[i]){
+		if (setjmp(crash_shutdown_buf) == 0) {
+			asm volatile("sync; isync");
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(200);
+		}
+		i++;
 	}
 
 	/*

Index: linux-2.6-ozlabs/include/asm-powerpc/kexec.h
===================================================================

--- linux-2.6-ozlabs.orig/include/asm-powerpc/kexec.h
+++ linux-2.6-ozlabs/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

Re: [PATCH 2/2] kdump shutdown hook support

From: Olof Johansson <hidden>
Date: 2007-12-12 06:56:41

Hi,

A couple of comments below.


-Olof

On Wed, Dec 12, 2007 at 04:45:12PM +1100, Michael Neuling wrote:

quoted hunk

Index: linux-2.6-ozlabs/arch/powerpc/kernel/crash.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/crash.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_SHUTDOWN_HANDLES_NUM 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_SHUTDOWN_HANDLES_NUM+1];
+static DEFINE_SPINLOCK(crash_handles_lock);

Not 'handlers'?

quoted hunk

+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,69 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/* 
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i;
+
+	spin_lock(&crash_handles_lock);
+	for(i = 0 ; i <= CRASH_SHUTDOWN_HANDLES_NUM; i++)

Missing space after for. There's a handful more of these through the
patch.

quoted hunk

+		if (!crash_shutdown_handles[i])
+			break;
+
+	if (i == CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		spin_unlock(&crash_handles_lock);
+		return 1;
+	}
+
+	/* Insert handle at end */
+	crash_shutdown_handles[i] = handler;
+	spin_unlock(&crash_handles_lock);
+	return 0;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i;
+
+	spin_lock(&crash_handles_lock);
+	for(i = 0 ; i <= CRASH_SHUTDOWN_HANDLES_NUM; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		spin_unlock(&crash_handles_lock);
+		return 1;
+	}
+
+	/* Shift handles down */
+	while(crash_shutdown_handles[i]) {
+		crash_shutdown_handles[i] = crash_shutdown_handles[i+1];
+		i++;
+	}
+	spin_unlock(&crash_handles_lock);
+	return 0;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static long crash_shutdown_buf[SETJMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
 
 	/*
 	 * This function is only called after the system

@@ -301,14 +368,27 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/* Call registered shutdown routines */
+	__debugger_fault_handler = handle_fault;
+	i = 0;
+	while(crash_shutdown_handles[i]){

This could do nicely as a for loop instead:

for (i = 0; crash_shutdown_handles[i]; i++) {

+		if (setjmp(crash_shutdown_buf) == 0) {
+			asm volatile("sync; isync");
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(200);

This looks a bit random. Why the handcoded barriers, and why the delay?
At least comment why the delay is needed (and why just 200 is
sufficient). I don't see a need for the barriers at all here?

quoted hunk

+		}
+		i++;
 	}
 
 	/*
Index: linux-2.6-ozlabs/include/asm-powerpc/kexec.h
===================================================================

--- linux-2.6-ozlabs.orig/include/asm-powerpc/kexec.h
+++ linux-2.6-ozlabs/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev

Re: [PATCH 2/2] kdump shutdown hook support

From: Michael Ellerman <hidden>
Date: 2007-12-12 23:07:19

On Wed, 2007-12-12 at 16:45 +1100, Michael Neuling wrote:

quoted hunk

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.

This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable.  

I've tested with bogus data and instruction exceptions.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/crash.c |   90 +++++++++++++++++++++++++++++++++++++++++---
 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 88 insertions(+), 5 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/crash.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/crash.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_SHUTDOWN_HANDLES_NUM 1

CRASH_HANDLER_MAX ?

quoted hunk

+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_SHUTDOWN_HANDLES_NUM+1];
+static DEFINE_SPINLOCK(crash_handles_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,69 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/* 
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handles_lock);

+	for(i = 0 ; i <= CRASH_SHUTDOWN_HANDLES_NUM; i++) {
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at end */
+			crash_shutdown_handles[i] = handler;
+			rc = 0;
+			break;
+		}
+	}

+	if (i == CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc = 1;
+	}
+
+	spin_unlock(&crash_handles_lock);

Perhaps?

+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i;
+
+	spin_lock(&crash_handles_lock);
+	for(i = 0 ; i <= CRASH_SHUTDOWN_HANDLES_NUM; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		spin_unlock(&crash_handles_lock);
+		return 1;
+	}
+
+	/* Shift handles down */
+	while(crash_shutdown_handles[i]) {
+		crash_shutdown_handles[i] = crash_shutdown_handles[i+1];
+		i++;
+	}
+	spin_unlock(&crash_handles_lock);
+	return 0;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static long crash_shutdown_buf[SETJMP_BUF_LEN];

unsigned long?

quoted hunk

+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
 
 	/*
 	 * This function is only called after the system

@@ -301,14 +368,27 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/* Call registered shutdown routines */
+	__debugger_fault_handler = handle_fault;
+	i = 0;
+	while(crash_shutdown_handles[i]){
+		if (setjmp(crash_shutdown_buf) == 0) {
+			asm volatile("sync; isync");
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(200);
+		}
+		i++;
 	}

You should probably reset __debugger_fault_handler, just to be safe.

cheers

-- 
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

Re: [PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2007-12-13 00:04:02


In message <1197500839.7695.19.camel@concordia> you wrote:

--=-Kza0KCx0MG8nsjfq7kOz
Content-Type: text/plain
Content-Transfer-Encoding: quoted-printable

On Wed, 2007-12-12 at 16:45 +1100, Michael Neuling wrote:

quoted

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.
=20
This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable. =20
=20
I've tested with bogus data and instruction exceptions.
=20
Signed-off-by: Michael Neuling <redacted>
---
=20
 arch/powerpc/kernel/crash.c |   90 +++++++++++++++++++++++++++++++++++++=

++++---

quoted

 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 88 insertions(+), 5 deletions(-)
=20
Index: linux-2.6-ozlabs/arch/powerpc/kernel/crash.c
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=

=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

quoted

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/crash.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>

=20
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu =3D -1;
 static cpumask_t cpus_in_crash =3D CPU_MASK_NONE;
 cpumask_t cpus_in_sr =3D CPU_MASK_NONE;

=20
+#define CRASH_SHUTDOWN_HANDLES_NUM 1

CRASH_HANDLER_MAX ?

yep, MAX is probably a more representative name

quoted

+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_SHUTDOWN_HANDLES_NU=

M+1];

quoted

+static DEFINE_SPINLOCK(crash_handles_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset =3D ATOMIC_INIT(0);
=20

@@ -285,9 +292,69 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */

=20
+/*=20
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handles_lock);

quoted

+	for(i =3D 0 ; i <=3D CRASH_SHUTDOWN_HANDLES_NUM; i++) {
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at end */
+			crash_shutdown_handles[i] =3D handler;
+			rc =3D 0;
+			break;
+		}
+	}
=20
+	if (i =3D=3D CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc =3D 1;
+	}
+
+	spin_unlock(&crash_handles_lock);

Perhaps?

Yep, looks nicer.  I'll fix the one below too.

quoted

+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i;
+
+	spin_lock(&crash_handles_lock);
+	for(i =3D 0 ; i <=3D CRASH_SHUTDOWN_HANDLES_NUM; i++)
+		if (crash_shutdown_handles[i] =3D=3D handler)
+			break;
+
+	if (i =3D=3D CRASH_SHUTDOWN_HANDLES_NUM){
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		spin_unlock(&crash_handles_lock);
+		return 1;
+	}
+
+	/* Shift handles down */
+	while(crash_shutdown_handles[i]) {
+		crash_shutdown_handles[i] =3D crash_shutdown_handles[i+1];
+		i++;
+	}
+	spin_unlock(&crash_handles_lock);
+	return 0;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static long crash_shutdown_buf[SETJMP_BUF_LEN];

unsigned long?

yep

quoted

+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
=20
 	/*
 	 * This function is only called after the system

@@ -301,14 +368,27 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();

=20
-	for_each_irq(irq) {
-		struct irq_desc *desc =3D irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc =3D irq_desc + i;
=20
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
=20
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/* Call registered shutdown routines */
+	__debugger_fault_handler =3D handle_fault;
+	i =3D 0;
+	while(crash_shutdown_handles[i]){
+		if (setjmp(crash_shutdown_buf) =3D=3D 0) {
+			asm volatile("sync; isync");
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(200);
+		}
+		i++;
 	}

You should probably reset __debugger_fault_handler, just to be safe.

Agreed.

Mikey

Re: [PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2007-12-13 00:53:55

quoted

+		if (setjmp(crash_shutdown_buf) == 0) {
+			asm volatile("sync; isync");
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(200);

This looks a bit random. Why the handcoded barriers, and why the delay?
I don't see a need for the barriers at all here?

I was following the bouncing ball here from the xmon code.  We don't
want the hardware executing any of potentially dangerous hooks before
we've setup the setjmp save buffer.  Similar, any code past the
dangerous hooks shouldn't be executed before we take a machine check.

It's definitely not a fast path so.... :-)

At least comment why the delay is needed (and why just 200 is
sufficient).

xmon.c says: 
   /* wait a little while to see if we get a machine check */

I'll update the "200" with a #define SETJMP_MACHINE_CHECK_DELAY and push
the same change into the xmon code.  This delay in the machine check is
apparently required on 601 machines.

All you other comments I agree with.

Mikey

[PATCH 0/2] Add crashdump shutdown hooks

From: Michael Neuling <hidden>
Date: 2007-12-13 03:16:17

The following patches add crashdump shutdown hooks for POWERPC.
Signed-off-by: Michael Neuling <redacted>
---

This is an updated series following comments from the first post.
Adds some documentation, better code flow, fixes 32 bit compiles and
other updates based on feedback

Again these are based on paulus' for 2.6.25 tree.

[PATCH 1/2] Make setjmp/longjmp code generic

From: Michael Neuling <hidden>
Date: 2007-12-13 03:16:18

This makes the setjmp/longjmp code used by xmon, generically available
to other code.  It also removes the requirement for debugger hooks to
be only called on 0x300 (data storage) exception and adds some
documentation.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/misc.S   |  128 ++++++++++++++++++++++++++++++++++++++++
 arch/powerpc/mm/fault.c      |    6 -
 arch/powerpc/xmon/Makefile   |    2 
 arch/powerpc/xmon/setjmp.S   |  135 -------------------------------------------
 arch/powerpc/xmon/xmon.c     |   38 ++++--------
 include/asm-powerpc/setjmp.h |   19 ++++++
 6 files changed, 164 insertions(+), 164 deletions(-)

Index: clone3/arch/powerpc/kernel/misc.S
===================================================================

--- clone3.orig/arch/powerpc/kernel/misc.S
+++ clone3/arch/powerpc/kernel/misc.S

@@ -8,6 +8,8 @@
  * Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
  * PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
  *
+ * setjmp/longjmp and xmon_save_regs code by Paul Mackerras.
+ *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version

@@ -15,6 +17,8 @@
  */
 #include <asm/ppc_asm.h>
 #include <asm/unistd.h>
+#include <asm/asm-compat.h>
+#include <asm/asm-offsets.h>
 
 	.text

@@ -51,3 +55,127 @@ _GLOBAL(kernel_execve)
 	bnslr
 	neg	r3,r3
 	blr
+
+_GLOBAL(setjmp)
+	mflr	r0
+	PPC_STL	r0,0(r3)
+	PPC_STL	r1,SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	mfcr	r0
+	PPC_STL	r0,3*SZL(r3)
+	PPC_STL	r13,4*SZL(r3)
+	PPC_STL	r14,5*SZL(r3)
+	PPC_STL	r15,6*SZL(r3)
+	PPC_STL	r16,7*SZL(r3)
+	PPC_STL	r17,8*SZL(r3)
+	PPC_STL	r18,9*SZL(r3)
+	PPC_STL	r19,10*SZL(r3)
+	PPC_STL	r20,11*SZL(r3)
+	PPC_STL	r21,12*SZL(r3)
+	PPC_STL	r22,13*SZL(r3)
+	PPC_STL	r23,14*SZL(r3)
+	PPC_STL	r24,15*SZL(r3)
+	PPC_STL	r25,16*SZL(r3)
+	PPC_STL	r26,17*SZL(r3)
+	PPC_STL	r27,18*SZL(r3)
+	PPC_STL	r28,19*SZL(r3)
+	PPC_STL	r29,20*SZL(r3)
+	PPC_STL	r30,21*SZL(r3)
+	PPC_STL	r31,22*SZL(r3)
+	li	r3,0
+	blr
+
+_GLOBAL(longjmp)
+	PPC_LCMPI r4,0
+	bne	1f
+	li	r4,1
+1:	PPC_LL	r13,4*SZL(r3)
+	PPC_LL	r14,5*SZL(r3)
+	PPC_LL	r15,6*SZL(r3)
+	PPC_LL	r16,7*SZL(r3)
+	PPC_LL	r17,8*SZL(r3)
+	PPC_LL	r18,9*SZL(r3)
+	PPC_LL	r19,10*SZL(r3)
+	PPC_LL	r20,11*SZL(r3)
+	PPC_LL	r21,12*SZL(r3)
+	PPC_LL	r22,13*SZL(r3)
+	PPC_LL	r23,14*SZL(r3)
+	PPC_LL	r24,15*SZL(r3)
+	PPC_LL	r25,16*SZL(r3)
+	PPC_LL	r26,17*SZL(r3)
+	PPC_LL	r27,18*SZL(r3)
+	PPC_LL	r28,19*SZL(r3)
+	PPC_LL	r29,20*SZL(r3)
+	PPC_LL	r30,21*SZL(r3)
+	PPC_LL	r31,22*SZL(r3)
+	PPC_LL	r0,3*SZL(r3)
+	mtcrf	0x38,r0
+	PPC_LL	r0,0(r3)
+	PPC_LL	r1,SZL(r3)
+	PPC_LL	r2,2*SZL(r3)
+	mtlr	r0
+	mr	r3,r4
+	blr
+
+#ifdef CONFIG_XMON
+/*
+ * Grab the register values as they are now.
+ * This won't do a particularily good job because we really
+ * want our caller's caller's registers, and our caller has
+ * already executed its prologue.
+ * ToDo: We could reach back into the caller's save area to do
+ * a better job of representing the caller's state (note that
+ * that will be different for 32-bit and 64-bit, because of the
+ * different ABIs, though).
+ */
+_GLOBAL(xmon_save_regs)
+	PPC_STL	r0,0*SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	PPC_STL	r3,3*SZL(r3)
+	PPC_STL	r4,4*SZL(r3)
+	PPC_STL	r5,5*SZL(r3)
+	PPC_STL	r6,6*SZL(r3)
+	PPC_STL	r7,7*SZL(r3)
+	PPC_STL	r8,8*SZL(r3)
+	PPC_STL	r9,9*SZL(r3)
+	PPC_STL	r10,10*SZL(r3)
+	PPC_STL	r11,11*SZL(r3)
+	PPC_STL	r12,12*SZL(r3)
+	PPC_STL	r13,13*SZL(r3)
+	PPC_STL	r14,14*SZL(r3)
+	PPC_STL	r15,15*SZL(r3)
+	PPC_STL	r16,16*SZL(r3)
+	PPC_STL	r17,17*SZL(r3)
+	PPC_STL	r18,18*SZL(r3)
+	PPC_STL	r19,19*SZL(r3)
+	PPC_STL	r20,20*SZL(r3)
+	PPC_STL	r21,21*SZL(r3)
+	PPC_STL	r22,22*SZL(r3)
+	PPC_STL	r23,23*SZL(r3)
+	PPC_STL	r24,24*SZL(r3)
+	PPC_STL	r25,25*SZL(r3)
+	PPC_STL	r26,26*SZL(r3)
+	PPC_STL	r27,27*SZL(r3)
+	PPC_STL	r28,28*SZL(r3)
+	PPC_STL	r29,29*SZL(r3)
+	PPC_STL	r30,30*SZL(r3)
+	PPC_STL	r31,31*SZL(r3)
+	/* go up one stack frame for SP */
+	PPC_LL	r4,0(r1)
+	PPC_STL	r4,1*SZL(r3)
+	/* get caller's LR */
+	PPC_LL	r0,LRSAVE(r4)
+	PPC_STL	r0,_NIP-STACK_FRAME_OVERHEAD(r3)
+	PPC_STL	r0,_LINK-STACK_FRAME_OVERHEAD(r3)
+	mfmsr	r0
+	PPC_STL	r0,_MSR-STACK_FRAME_OVERHEAD(r3)
+	mfctr	r0
+	PPC_STL	r0,_CTR-STACK_FRAME_OVERHEAD(r3)
+	mfxer	r0
+	PPC_STL	r0,_XER-STACK_FRAME_OVERHEAD(r3)
+	mfcr	r0
+	PPC_STL	r0,_CCR-STACK_FRAME_OVERHEAD(r3)
+	li	r0,0
+	PPC_STL	r0,_TRAP-STACK_FRAME_OVERHEAD(r3)
+	blr
+#endif

Index: clone3/arch/powerpc/mm/fault.c
===================================================================

--- clone3.orig/arch/powerpc/mm/fault.c
+++ clone3/arch/powerpc/mm/fault.c

@@ -167,10 +167,8 @@ int __kprobes do_page_fault(struct pt_re
 	if (notify_page_fault(regs))
 		return 0;
 
-	if (trap == 0x300) {
-		if (debugger_fault_handler(regs))
-			return 0;
-	}
+	if (unlikely(debugger_fault_handler(regs)))
+		return 0;
 
 	/* On a kernel SLB miss we can only check for a valid exception entry */
 	if (!user_mode(regs) && (address >= TASK_SIZE))

Index: clone3/arch/powerpc/xmon/Makefile
===================================================================

--- clone3.orig/arch/powerpc/xmon/Makefile
+++ clone3/arch/powerpc/xmon/Makefile

@@ -4,7 +4,7 @@ ifdef CONFIG_PPC64
 EXTRA_CFLAGS += -mno-minimal-toc
 endif
 
-obj-y			+= xmon.o setjmp.o start.o nonstdio.o
+obj-y			+= xmon.o start.o nonstdio.o
 
 ifdef CONFIG_XMON_DISASSEMBLY
 obj-y			+= ppc-dis.o ppc-opc.o

Index: clone3/arch/powerpc/xmon/setjmp.S
===================================================================

--- clone3.orig/arch/powerpc/xmon/setjmp.S
+++ /dev/null

@@ -1,135 +0,0 @@
-/*
- * Copyright (C) 1996 Paul Mackerras.
- *
- *      This program is free software; you can redistribute it and/or
- *      modify it under the terms of the GNU General Public License
- *      as published by the Free Software Foundation; either version
- *      2 of the License, or (at your option) any later version.
- *
- * NOTE: assert(sizeof(buf) > 23 * sizeof(long))
- */
-#include <asm/processor.h>
-#include <asm/ppc_asm.h>
-#include <asm/asm-offsets.h>
-
-_GLOBAL(xmon_setjmp)
-	mflr	r0
-	PPC_STL	r0,0(r3)
-	PPC_STL	r1,SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	mfcr	r0
-	PPC_STL	r0,3*SZL(r3)
-	PPC_STL	r13,4*SZL(r3)
-	PPC_STL	r14,5*SZL(r3)
-	PPC_STL	r15,6*SZL(r3)
-	PPC_STL	r16,7*SZL(r3)
-	PPC_STL	r17,8*SZL(r3)
-	PPC_STL	r18,9*SZL(r3)
-	PPC_STL	r19,10*SZL(r3)
-	PPC_STL	r20,11*SZL(r3)
-	PPC_STL	r21,12*SZL(r3)
-	PPC_STL	r22,13*SZL(r3)
-	PPC_STL	r23,14*SZL(r3)
-	PPC_STL	r24,15*SZL(r3)
-	PPC_STL	r25,16*SZL(r3)
-	PPC_STL	r26,17*SZL(r3)
-	PPC_STL	r27,18*SZL(r3)
-	PPC_STL	r28,19*SZL(r3)
-	PPC_STL	r29,20*SZL(r3)
-	PPC_STL	r30,21*SZL(r3)
-	PPC_STL	r31,22*SZL(r3)
-	li	r3,0
-	blr
-
-_GLOBAL(xmon_longjmp)
-	PPC_LCMPI r4,0
-	bne	1f
-	li	r4,1
-1:	PPC_LL	r13,4*SZL(r3)
-	PPC_LL	r14,5*SZL(r3)
-	PPC_LL	r15,6*SZL(r3)
-	PPC_LL	r16,7*SZL(r3)
-	PPC_LL	r17,8*SZL(r3)
-	PPC_LL	r18,9*SZL(r3)
-	PPC_LL	r19,10*SZL(r3)
-	PPC_LL	r20,11*SZL(r3)
-	PPC_LL	r21,12*SZL(r3)
-	PPC_LL	r22,13*SZL(r3)
-	PPC_LL	r23,14*SZL(r3)
-	PPC_LL	r24,15*SZL(r3)
-	PPC_LL	r25,16*SZL(r3)
-	PPC_LL	r26,17*SZL(r3)
-	PPC_LL	r27,18*SZL(r3)
-	PPC_LL	r28,19*SZL(r3)
-	PPC_LL	r29,20*SZL(r3)
-	PPC_LL	r30,21*SZL(r3)
-	PPC_LL	r31,22*SZL(r3)
-	PPC_LL	r0,3*SZL(r3)
-	mtcrf	0x38,r0
-	PPC_LL	r0,0(r3)
-	PPC_LL	r1,SZL(r3)
-	PPC_LL	r2,2*SZL(r3)
-	mtlr	r0
-	mr	r3,r4
-	blr
-
-/*
- * Grab the register values as they are now.
- * This won't do a particularily good job because we really
- * want our caller's caller's registers, and our caller has
- * already executed its prologue.
- * ToDo: We could reach back into the caller's save area to do
- * a better job of representing the caller's state (note that
- * that will be different for 32-bit and 64-bit, because of the
- * different ABIs, though).
- */
-_GLOBAL(xmon_save_regs)
-	PPC_STL	r0,0*SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	PPC_STL	r3,3*SZL(r3)
-	PPC_STL	r4,4*SZL(r3)
-	PPC_STL	r5,5*SZL(r3)
-	PPC_STL	r6,6*SZL(r3)
-	PPC_STL	r7,7*SZL(r3)
-	PPC_STL	r8,8*SZL(r3)
-	PPC_STL	r9,9*SZL(r3)
-	PPC_STL	r10,10*SZL(r3)
-	PPC_STL	r11,11*SZL(r3)
-	PPC_STL	r12,12*SZL(r3)
-	PPC_STL	r13,13*SZL(r3)
-	PPC_STL	r14,14*SZL(r3)
-	PPC_STL	r15,15*SZL(r3)
-	PPC_STL	r16,16*SZL(r3)
-	PPC_STL	r17,17*SZL(r3)
-	PPC_STL	r18,18*SZL(r3)
-	PPC_STL	r19,19*SZL(r3)
-	PPC_STL	r20,20*SZL(r3)
-	PPC_STL	r21,21*SZL(r3)
-	PPC_STL	r22,22*SZL(r3)
-	PPC_STL	r23,23*SZL(r3)
-	PPC_STL	r24,24*SZL(r3)
-	PPC_STL	r25,25*SZL(r3)
-	PPC_STL	r26,26*SZL(r3)
-	PPC_STL	r27,27*SZL(r3)
-	PPC_STL	r28,28*SZL(r3)
-	PPC_STL	r29,29*SZL(r3)
-	PPC_STL	r30,30*SZL(r3)
-	PPC_STL	r31,31*SZL(r3)
-	/* go up one stack frame for SP */
-	PPC_LL	r4,0(r1)
-	PPC_STL	r4,1*SZL(r3)
-	/* get caller's LR */
-	PPC_LL	r0,LRSAVE(r4)
-	PPC_STL	r0,_NIP-STACK_FRAME_OVERHEAD(r3)
-	PPC_STL	r0,_LINK-STACK_FRAME_OVERHEAD(r3)
-	mfmsr	r0
-	PPC_STL	r0,_MSR-STACK_FRAME_OVERHEAD(r3)
-	mfctr	r0
-	PPC_STL	r0,_CTR-STACK_FRAME_OVERHEAD(r3)
-	mfxer	r0
-	PPC_STL	r0,_XER-STACK_FRAME_OVERHEAD(r3)
-	mfcr	r0
-	PPC_STL	r0,_CCR-STACK_FRAME_OVERHEAD(r3)
-	li	r0,0
-	PPC_STL	r0,_TRAP-STACK_FRAME_OVERHEAD(r3)
-	blr

Index: clone3/arch/powerpc/xmon/xmon.c
===================================================================

--- clone3.orig/arch/powerpc/xmon/xmon.c
+++ clone3/arch/powerpc/xmon/xmon.c

@@ -40,6 +40,7 @@
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
 #include <asm/firmware.h>
+#include <asm/setjmp.h>
 
 #ifdef CONFIG_PPC64
 #include <asm/hvcall.h>

@@ -71,12 +72,9 @@ static unsigned long ncsum = 4096;
 static int termch;
 static char tmpstr[128];
 
-#define JMP_BUF_LEN	23
-static long bus_error_jmp[JMP_BUF_LEN];
+static long bus_error_jmp[SETJMP_BUF_LEN];
 static int catch_memory_errors;
 static long *xmon_fault_jmp[NR_CPUS];
-#define setjmp xmon_setjmp
-#define longjmp xmon_longjmp
 
 /* Breakpoint stuff */
 struct bpt {

@@ -162,8 +160,6 @@ int xmon_no_auto_backtrace;
 extern void xmon_enter(void);
 extern void xmon_leave(void);
 
-extern long setjmp(long *);
-extern void longjmp(long *, long);
 extern void xmon_save_regs(struct pt_regs *);
 
 #ifdef CONFIG_PPC64

@@ -338,7 +334,7 @@ static int xmon_core(struct pt_regs *reg
 {
 	int cmd = 0;
 	struct bpt *bp;
-	long recurse_jmp[JMP_BUF_LEN];
+	long recurse_jmp[SETJMP_BUF_LEN];
 	unsigned long offset;
 	unsigned long flags;
 #ifdef CONFIG_SMP

@@ -1428,7 +1424,7 @@ void prregs(struct pt_regs *fp)
 			sync();
 			regs = *(struct pt_regs *)base;
 			sync();
-			__delay(200);
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
 		} else {
 			catch_memory_errors = 0;
 			printf("*** Error reading registers from "REG"\n",

@@ -1497,8 +1493,7 @@ void cacheflush(void)
 				cinval((void *) adrs);
 		}
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 	}
 	catch_memory_errors = 0;
 }

@@ -1533,8 +1528,7 @@ read_spr(int n)
 		ret = code();
 
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	}

@@ -1569,8 +1563,7 @@ write_spr(int n, unsigned long val)
 		code(val);
 
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	}
 }

@@ -1676,8 +1669,7 @@ mread(unsigned long adrs, void *buf, int
 			}
 		}
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	}
 	catch_memory_errors = 0;

@@ -1713,8 +1705,7 @@ mwrite(unsigned long adrs, void *buf, in
 			}
 		}
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	} else {
 		printf("*** Error writing address %x\n", adrs + n);

@@ -2521,8 +2512,7 @@ static void xmon_print_symbol(unsigned l
 		name = kallsyms_lookup(address, &size, &offset, &modname,
 				       tmpstr);
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 	}
 
 	catch_memory_errors = 0;

@@ -2777,7 +2767,7 @@ static void stop_spus(void)
 			spu_mfc_sr1_set(spu, tmp);
 
 			sync();
-			__delay(200);
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
 
 			spu_info[i].stopped_ok = 1;

@@ -2817,7 +2807,7 @@ static void restart_spus(void)
 					spu_info[i].saved_spu_runcntl_RW);
 
 			sync();
-			__delay(200);
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
 
 			printf("Restarted spu %.2d\n", i);
 		} else {

@@ -2837,7 +2827,7 @@ do {									\
 		printf("  %-*s = "format"\n", DUMP_WIDTH,		\
 				#field, value);				\
 		sync();							\
-		__delay(200);						\
+		__delay(SETJMP_MACHINE_CHECK_DELAY);			\
 	} else {							\
 		catch_memory_errors = 0;				\
 		printf("  %-*s = *** Error reading field.\n",		\

@@ -2899,7 +2889,7 @@ static void dump_spu_ls(unsigned long nu
 		sync();
 		ls_addr = (unsigned long)spu_info[num].spu->local_store;
 		sync();
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 	} else {
 		catch_memory_errors = 0;
 		printf("*** Error: accessing spu info for spu %d\n", num);

Index: clone3/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ clone3/include/asm-powerpc/setjmp.h

@@ -0,0 +1,19 @@
+/*
+ * Copyright (C) 2007 Michael Neuling
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define SETJMP_BUF_LEN    23
+#define SETJMP_MACHINE_CHECK_DELAY    200
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

[PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2007-12-13 03:16:18

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.

This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable.  

Tested with bogus data and instruction exceptions.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/crash.c |  102 +++++++++++++++++++++++++++++++++++++++++---
 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 100 insertions(+), 5 deletions(-)

Index: clone3/arch/powerpc/kernel/crash.c
===================================================================

--- clone3.orig/arch/powerpc/kernel/crash.c
+++ clone3/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_HANDLER_MAX 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
+static DEFINE_SPINLOCK(crash_handers_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,72 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/*
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at first empty entry */
+			crash_shutdown_handles[i] = handler;
+			rc = 0;
+			break;
+		}
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc = 1;
+	}
+
+	spin_unlock(&crash_handers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		rc = 1;
+	} else {
+		/* Shift handles down */
+		for (; crash_shutdown_handles[i]; i++)
+			crash_shutdown_handles[i] =
+				crash_shutdown_handles[i+1];
+		rc = 0;
+	}
+
+	spin_unlock(&crash_handers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static unsigned long crash_shutdown_buf[SETJMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
+	int (*old_handler)(struct pt_regs *regs);
+
 
 	/*
 	 * This function is only called after the system

@@ -301,15 +371,37 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/*
+	 * Call registered shutdown routines savely.  Swap out
+	 * __debugger_fault_handler, and replace on exit.
+	 */
+	old_handler = __debugger_fault_handler;
+	__debugger_fault_handler = handle_fault;
+	for (i = 0; crash_shutdown_handles[i]; i++) {
+		if (setjmp(crash_shutdown_buf) == 0) {
+			/*
+			 * Insert syncs and delay to ensure
+			 * instructions in the dangerous region don't
+			 * leak away from this protected region.
+			 */
+			asm volatile("sync; isync");
+			/* dangerous region */
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
+		}
 	}
+	__debugger_fault_handler = old_handler;
 
 	/*
 	 * Make a note of crashing cpu. Will be used in machine_kexec

Index: clone3/include/asm-powerpc/kexec.h
===================================================================

--- clone3.orig/include/asm-powerpc/kexec.h
+++ clone3/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

Re: [PATCH 2/2] kdump shutdown hook support

From: Olof Johansson <hidden>
Date: 2007-12-13 05:07:07

Hi,

On Thu, Dec 13, 2007 at 02:16:18PM +1100, Michael Neuling wrote:

quoted hunk

Index: clone3/arch/powerpc/kernel/crash.c
===================================================================

--- clone3.orig/arch/powerpc/kernel/crash.c
+++ clone3/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_HANDLER_MAX 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
+static DEFINE_SPINLOCK(crash_handers_lock);

How about just using 'handlers' instead of 'handers' and 'handles'? :)

quoted hunk

 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,72 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/*
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at first empty entry */
+			crash_shutdown_handles[i] = handler;
+			rc = 0;
+			break;
+		}
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc = 1;
+	}
+
+	spin_unlock(&crash_handers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		rc = 1;
+	} else {
+		/* Shift handles down */
+		for (; crash_shutdown_handles[i]; i++)
+			crash_shutdown_handles[i] =
+				crash_shutdown_handles[i+1];
+		rc = 0;
+	}
+
+	spin_unlock(&crash_handers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static unsigned long crash_shutdown_buf[SETJMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
+	int (*old_handler)(struct pt_regs *regs);
+
 
 	/*
 	 * This function is only called after the system
@@ -301,15 +371,37 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/*
+	 * Call registered shutdown routines savely.  Swap out
+	 * __debugger_fault_handler, and replace on exit.
+	 */
+	old_handler = __debugger_fault_handler;
+	__debugger_fault_handler = handle_fault;
+	for (i = 0; crash_shutdown_handles[i]; i++) {
+		if (setjmp(crash_shutdown_buf) == 0) {
+			/*
+			 * Insert syncs and delay to ensure
+			 * instructions in the dangerous region don't
+			 * leak away from this protected region.
+			 */
+			asm volatile("sync; isync");
+			/* dangerous region */
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(SETJMP_MACHINE_CHECK_DELAY);

Where is this defined?

quoted hunk

+		}
 	}
+	__debugger_fault_handler = old_handler;
 
 	/*
 	 * Make a note of crashing cpu. Will be used in machine_kexec
Index: clone3/include/asm-powerpc/kexec.h
===================================================================

--- clone3.orig/include/asm-powerpc/kexec.h
+++ clone3/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev

Re: [PATCH 2/2] kdump shutdown hook support

From: Olof Johansson <hidden>
Date: 2007-12-13 05:08:00

On Wed, Dec 12, 2007 at 11:12:24PM -0600, Olof Johansson wrote:

Hi,

quoted

+			__delay(SETJMP_MACHINE_CHECK_DELAY);

Where is this defined?

Oops, nevermind. In 1/2, of course. :-)


-Olof

[PATCH 0/2] Add crashdump shutdown hooks

From: Michael Neuling <hidden>
Date: 2007-12-13 09:59:17

The following patches add crashdump shutdown hooks for POWERPC.
Signed-off-by: Michael Neuling <redacted>
---
Fixes stupid variable name noticed by Olof.

[PATCH 1/2] Make setjmp/longjmp code generic

From: Michael Neuling <hidden>
Date: 2007-12-13 09:59:17

This makes the setjmp/longjmp code used by xmon, generically available
to other code.  It also removes the requirement for debugger hooks to
be only called on 0x300 (data storage) exception and adds some
documentation.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/misc.S   |  128 ++++++++++++++++++++++++++++++++++++++++
 arch/powerpc/mm/fault.c      |    6 -
 arch/powerpc/xmon/Makefile   |    2 
 arch/powerpc/xmon/setjmp.S   |  135 -------------------------------------------
 arch/powerpc/xmon/xmon.c     |   38 ++++--------
 include/asm-powerpc/setjmp.h |   19 ++++++
 6 files changed, 164 insertions(+), 164 deletions(-)

Index: clone3/arch/powerpc/kernel/misc.S
===================================================================

--- clone3.orig/arch/powerpc/kernel/misc.S
+++ clone3/arch/powerpc/kernel/misc.S

@@ -8,6 +8,8 @@
  * Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
  * PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
  *
+ * setjmp/longjmp and xmon_save_regs code by Paul Mackerras.
+ *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version

@@ -15,6 +17,8 @@
  */
 #include <asm/ppc_asm.h>
 #include <asm/unistd.h>
+#include <asm/asm-compat.h>
+#include <asm/asm-offsets.h>
 
 	.text

@@ -51,3 +55,127 @@ _GLOBAL(kernel_execve)
 	bnslr
 	neg	r3,r3
 	blr
+
+_GLOBAL(setjmp)
+	mflr	r0
+	PPC_STL	r0,0(r3)
+	PPC_STL	r1,SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	mfcr	r0
+	PPC_STL	r0,3*SZL(r3)
+	PPC_STL	r13,4*SZL(r3)
+	PPC_STL	r14,5*SZL(r3)
+	PPC_STL	r15,6*SZL(r3)
+	PPC_STL	r16,7*SZL(r3)
+	PPC_STL	r17,8*SZL(r3)
+	PPC_STL	r18,9*SZL(r3)
+	PPC_STL	r19,10*SZL(r3)
+	PPC_STL	r20,11*SZL(r3)
+	PPC_STL	r21,12*SZL(r3)
+	PPC_STL	r22,13*SZL(r3)
+	PPC_STL	r23,14*SZL(r3)
+	PPC_STL	r24,15*SZL(r3)
+	PPC_STL	r25,16*SZL(r3)
+	PPC_STL	r26,17*SZL(r3)
+	PPC_STL	r27,18*SZL(r3)
+	PPC_STL	r28,19*SZL(r3)
+	PPC_STL	r29,20*SZL(r3)
+	PPC_STL	r30,21*SZL(r3)
+	PPC_STL	r31,22*SZL(r3)
+	li	r3,0
+	blr
+
+_GLOBAL(longjmp)
+	PPC_LCMPI r4,0
+	bne	1f
+	li	r4,1
+1:	PPC_LL	r13,4*SZL(r3)
+	PPC_LL	r14,5*SZL(r3)
+	PPC_LL	r15,6*SZL(r3)
+	PPC_LL	r16,7*SZL(r3)
+	PPC_LL	r17,8*SZL(r3)
+	PPC_LL	r18,9*SZL(r3)
+	PPC_LL	r19,10*SZL(r3)
+	PPC_LL	r20,11*SZL(r3)
+	PPC_LL	r21,12*SZL(r3)
+	PPC_LL	r22,13*SZL(r3)
+	PPC_LL	r23,14*SZL(r3)
+	PPC_LL	r24,15*SZL(r3)
+	PPC_LL	r25,16*SZL(r3)
+	PPC_LL	r26,17*SZL(r3)
+	PPC_LL	r27,18*SZL(r3)
+	PPC_LL	r28,19*SZL(r3)
+	PPC_LL	r29,20*SZL(r3)
+	PPC_LL	r30,21*SZL(r3)
+	PPC_LL	r31,22*SZL(r3)
+	PPC_LL	r0,3*SZL(r3)
+	mtcrf	0x38,r0
+	PPC_LL	r0,0(r3)
+	PPC_LL	r1,SZL(r3)
+	PPC_LL	r2,2*SZL(r3)
+	mtlr	r0
+	mr	r3,r4
+	blr
+
+#ifdef CONFIG_XMON
+/*
+ * Grab the register values as they are now.
+ * This won't do a particularily good job because we really
+ * want our caller's caller's registers, and our caller has
+ * already executed its prologue.
+ * ToDo: We could reach back into the caller's save area to do
+ * a better job of representing the caller's state (note that
+ * that will be different for 32-bit and 64-bit, because of the
+ * different ABIs, though).
+ */
+_GLOBAL(xmon_save_regs)
+	PPC_STL	r0,0*SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	PPC_STL	r3,3*SZL(r3)
+	PPC_STL	r4,4*SZL(r3)
+	PPC_STL	r5,5*SZL(r3)
+	PPC_STL	r6,6*SZL(r3)
+	PPC_STL	r7,7*SZL(r3)
+	PPC_STL	r8,8*SZL(r3)
+	PPC_STL	r9,9*SZL(r3)
+	PPC_STL	r10,10*SZL(r3)
+	PPC_STL	r11,11*SZL(r3)
+	PPC_STL	r12,12*SZL(r3)
+	PPC_STL	r13,13*SZL(r3)
+	PPC_STL	r14,14*SZL(r3)
+	PPC_STL	r15,15*SZL(r3)
+	PPC_STL	r16,16*SZL(r3)
+	PPC_STL	r17,17*SZL(r3)
+	PPC_STL	r18,18*SZL(r3)
+	PPC_STL	r19,19*SZL(r3)
+	PPC_STL	r20,20*SZL(r3)
+	PPC_STL	r21,21*SZL(r3)
+	PPC_STL	r22,22*SZL(r3)
+	PPC_STL	r23,23*SZL(r3)
+	PPC_STL	r24,24*SZL(r3)
+	PPC_STL	r25,25*SZL(r3)
+	PPC_STL	r26,26*SZL(r3)
+	PPC_STL	r27,27*SZL(r3)
+	PPC_STL	r28,28*SZL(r3)
+	PPC_STL	r29,29*SZL(r3)
+	PPC_STL	r30,30*SZL(r3)
+	PPC_STL	r31,31*SZL(r3)
+	/* go up one stack frame for SP */
+	PPC_LL	r4,0(r1)
+	PPC_STL	r4,1*SZL(r3)
+	/* get caller's LR */
+	PPC_LL	r0,LRSAVE(r4)
+	PPC_STL	r0,_NIP-STACK_FRAME_OVERHEAD(r3)
+	PPC_STL	r0,_LINK-STACK_FRAME_OVERHEAD(r3)
+	mfmsr	r0
+	PPC_STL	r0,_MSR-STACK_FRAME_OVERHEAD(r3)
+	mfctr	r0
+	PPC_STL	r0,_CTR-STACK_FRAME_OVERHEAD(r3)
+	mfxer	r0
+	PPC_STL	r0,_XER-STACK_FRAME_OVERHEAD(r3)
+	mfcr	r0
+	PPC_STL	r0,_CCR-STACK_FRAME_OVERHEAD(r3)
+	li	r0,0
+	PPC_STL	r0,_TRAP-STACK_FRAME_OVERHEAD(r3)
+	blr
+#endif

Index: clone3/arch/powerpc/mm/fault.c
===================================================================

--- clone3.orig/arch/powerpc/mm/fault.c
+++ clone3/arch/powerpc/mm/fault.c

@@ -167,10 +167,8 @@ int __kprobes do_page_fault(struct pt_re
 	if (notify_page_fault(regs))
 		return 0;
 
-	if (trap == 0x300) {
-		if (debugger_fault_handler(regs))
-			return 0;
-	}
+	if (unlikely(debugger_fault_handler(regs)))
+		return 0;
 
 	/* On a kernel SLB miss we can only check for a valid exception entry */
 	if (!user_mode(regs) && (address >= TASK_SIZE))

Index: clone3/arch/powerpc/xmon/Makefile
===================================================================

--- clone3.orig/arch/powerpc/xmon/Makefile
+++ clone3/arch/powerpc/xmon/Makefile

@@ -4,7 +4,7 @@ ifdef CONFIG_PPC64
 EXTRA_CFLAGS += -mno-minimal-toc
 endif
 
-obj-y			+= xmon.o setjmp.o start.o nonstdio.o
+obj-y			+= xmon.o start.o nonstdio.o
 
 ifdef CONFIG_XMON_DISASSEMBLY
 obj-y			+= ppc-dis.o ppc-opc.o

Index: clone3/arch/powerpc/xmon/setjmp.S
===================================================================

--- clone3.orig/arch/powerpc/xmon/setjmp.S
+++ /dev/null

@@ -1,135 +0,0 @@
-/*
- * Copyright (C) 1996 Paul Mackerras.
- *
- *      This program is free software; you can redistribute it and/or
- *      modify it under the terms of the GNU General Public License
- *      as published by the Free Software Foundation; either version
- *      2 of the License, or (at your option) any later version.
- *
- * NOTE: assert(sizeof(buf) > 23 * sizeof(long))
- */
-#include <asm/processor.h>
-#include <asm/ppc_asm.h>
-#include <asm/asm-offsets.h>
-
-_GLOBAL(xmon_setjmp)
-	mflr	r0
-	PPC_STL	r0,0(r3)
-	PPC_STL	r1,SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	mfcr	r0
-	PPC_STL	r0,3*SZL(r3)
-	PPC_STL	r13,4*SZL(r3)
-	PPC_STL	r14,5*SZL(r3)
-	PPC_STL	r15,6*SZL(r3)
-	PPC_STL	r16,7*SZL(r3)
-	PPC_STL	r17,8*SZL(r3)
-	PPC_STL	r18,9*SZL(r3)
-	PPC_STL	r19,10*SZL(r3)
-	PPC_STL	r20,11*SZL(r3)
-	PPC_STL	r21,12*SZL(r3)
-	PPC_STL	r22,13*SZL(r3)
-	PPC_STL	r23,14*SZL(r3)
-	PPC_STL	r24,15*SZL(r3)
-	PPC_STL	r25,16*SZL(r3)
-	PPC_STL	r26,17*SZL(r3)
-	PPC_STL	r27,18*SZL(r3)
-	PPC_STL	r28,19*SZL(r3)
-	PPC_STL	r29,20*SZL(r3)
-	PPC_STL	r30,21*SZL(r3)
-	PPC_STL	r31,22*SZL(r3)
-	li	r3,0
-	blr
-
-_GLOBAL(xmon_longjmp)
-	PPC_LCMPI r4,0
-	bne	1f
-	li	r4,1
-1:	PPC_LL	r13,4*SZL(r3)
-	PPC_LL	r14,5*SZL(r3)
-	PPC_LL	r15,6*SZL(r3)
-	PPC_LL	r16,7*SZL(r3)
-	PPC_LL	r17,8*SZL(r3)
-	PPC_LL	r18,9*SZL(r3)
-	PPC_LL	r19,10*SZL(r3)
-	PPC_LL	r20,11*SZL(r3)
-	PPC_LL	r21,12*SZL(r3)
-	PPC_LL	r22,13*SZL(r3)
-	PPC_LL	r23,14*SZL(r3)
-	PPC_LL	r24,15*SZL(r3)
-	PPC_LL	r25,16*SZL(r3)
-	PPC_LL	r26,17*SZL(r3)
-	PPC_LL	r27,18*SZL(r3)
-	PPC_LL	r28,19*SZL(r3)
-	PPC_LL	r29,20*SZL(r3)
-	PPC_LL	r30,21*SZL(r3)
-	PPC_LL	r31,22*SZL(r3)
-	PPC_LL	r0,3*SZL(r3)
-	mtcrf	0x38,r0
-	PPC_LL	r0,0(r3)
-	PPC_LL	r1,SZL(r3)
-	PPC_LL	r2,2*SZL(r3)
-	mtlr	r0
-	mr	r3,r4
-	blr
-
-/*
- * Grab the register values as they are now.
- * This won't do a particularily good job because we really
- * want our caller's caller's registers, and our caller has
- * already executed its prologue.
- * ToDo: We could reach back into the caller's save area to do
- * a better job of representing the caller's state (note that
- * that will be different for 32-bit and 64-bit, because of the
- * different ABIs, though).
- */
-_GLOBAL(xmon_save_regs)
-	PPC_STL	r0,0*SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	PPC_STL	r3,3*SZL(r3)
-	PPC_STL	r4,4*SZL(r3)
-	PPC_STL	r5,5*SZL(r3)
-	PPC_STL	r6,6*SZL(r3)
-	PPC_STL	r7,7*SZL(r3)
-	PPC_STL	r8,8*SZL(r3)
-	PPC_STL	r9,9*SZL(r3)
-	PPC_STL	r10,10*SZL(r3)
-	PPC_STL	r11,11*SZL(r3)
-	PPC_STL	r12,12*SZL(r3)
-	PPC_STL	r13,13*SZL(r3)
-	PPC_STL	r14,14*SZL(r3)
-	PPC_STL	r15,15*SZL(r3)
-	PPC_STL	r16,16*SZL(r3)
-	PPC_STL	r17,17*SZL(r3)
-	PPC_STL	r18,18*SZL(r3)
-	PPC_STL	r19,19*SZL(r3)
-	PPC_STL	r20,20*SZL(r3)
-	PPC_STL	r21,21*SZL(r3)
-	PPC_STL	r22,22*SZL(r3)
-	PPC_STL	r23,23*SZL(r3)
-	PPC_STL	r24,24*SZL(r3)
-	PPC_STL	r25,25*SZL(r3)
-	PPC_STL	r26,26*SZL(r3)
-	PPC_STL	r27,27*SZL(r3)
-	PPC_STL	r28,28*SZL(r3)
-	PPC_STL	r29,29*SZL(r3)
-	PPC_STL	r30,30*SZL(r3)
-	PPC_STL	r31,31*SZL(r3)
-	/* go up one stack frame for SP */
-	PPC_LL	r4,0(r1)
-	PPC_STL	r4,1*SZL(r3)
-	/* get caller's LR */
-	PPC_LL	r0,LRSAVE(r4)
-	PPC_STL	r0,_NIP-STACK_FRAME_OVERHEAD(r3)
-	PPC_STL	r0,_LINK-STACK_FRAME_OVERHEAD(r3)
-	mfmsr	r0
-	PPC_STL	r0,_MSR-STACK_FRAME_OVERHEAD(r3)
-	mfctr	r0
-	PPC_STL	r0,_CTR-STACK_FRAME_OVERHEAD(r3)
-	mfxer	r0
-	PPC_STL	r0,_XER-STACK_FRAME_OVERHEAD(r3)
-	mfcr	r0
-	PPC_STL	r0,_CCR-STACK_FRAME_OVERHEAD(r3)
-	li	r0,0
-	PPC_STL	r0,_TRAP-STACK_FRAME_OVERHEAD(r3)
-	blr

Index: clone3/arch/powerpc/xmon/xmon.c
===================================================================

--- clone3.orig/arch/powerpc/xmon/xmon.c
+++ clone3/arch/powerpc/xmon/xmon.c

@@ -40,6 +40,7 @@
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
 #include <asm/firmware.h>
+#include <asm/setjmp.h>
 
 #ifdef CONFIG_PPC64
 #include <asm/hvcall.h>

@@ -71,12 +72,9 @@ static unsigned long ncsum = 4096;
 static int termch;
 static char tmpstr[128];
 
-#define JMP_BUF_LEN	23
-static long bus_error_jmp[JMP_BUF_LEN];
+static long bus_error_jmp[SETJMP_BUF_LEN];
 static int catch_memory_errors;
 static long *xmon_fault_jmp[NR_CPUS];
-#define setjmp xmon_setjmp
-#define longjmp xmon_longjmp
 
 /* Breakpoint stuff */
 struct bpt {

@@ -162,8 +160,6 @@ int xmon_no_auto_backtrace;
 extern void xmon_enter(void);
 extern void xmon_leave(void);
 
-extern long setjmp(long *);
-extern void longjmp(long *, long);
 extern void xmon_save_regs(struct pt_regs *);
 
 #ifdef CONFIG_PPC64

@@ -338,7 +334,7 @@ static int xmon_core(struct pt_regs *reg
 {
 	int cmd = 0;
 	struct bpt *bp;
-	long recurse_jmp[JMP_BUF_LEN];
+	long recurse_jmp[SETJMP_BUF_LEN];
 	unsigned long offset;
 	unsigned long flags;
 #ifdef CONFIG_SMP

@@ -1428,7 +1424,7 @@ void prregs(struct pt_regs *fp)
 			sync();
 			regs = *(struct pt_regs *)base;
 			sync();
-			__delay(200);
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
 		} else {
 			catch_memory_errors = 0;
 			printf("*** Error reading registers from "REG"\n",

@@ -1497,8 +1493,7 @@ void cacheflush(void)
 				cinval((void *) adrs);
 		}
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 	}
 	catch_memory_errors = 0;
 }

@@ -1533,8 +1528,7 @@ read_spr(int n)
 		ret = code();
 
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	}

@@ -1569,8 +1563,7 @@ write_spr(int n, unsigned long val)
 		code(val);
 
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	}
 }

@@ -1676,8 +1669,7 @@ mread(unsigned long adrs, void *buf, int
 			}
 		}
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	}
 	catch_memory_errors = 0;

@@ -1713,8 +1705,7 @@ mwrite(unsigned long adrs, void *buf, in
 			}
 		}
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 		n = size;
 	} else {
 		printf("*** Error writing address %x\n", adrs + n);

@@ -2521,8 +2512,7 @@ static void xmon_print_symbol(unsigned l
 		name = kallsyms_lookup(address, &size, &offset, &modname,
 				       tmpstr);
 		sync();
-		/* wait a little while to see if we get a machine check */
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 	}
 
 	catch_memory_errors = 0;

@@ -2777,7 +2767,7 @@ static void stop_spus(void)
 			spu_mfc_sr1_set(spu, tmp);
 
 			sync();
-			__delay(200);
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
 
 			spu_info[i].stopped_ok = 1;

@@ -2817,7 +2807,7 @@ static void restart_spus(void)
 					spu_info[i].saved_spu_runcntl_RW);
 
 			sync();
-			__delay(200);
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
 
 			printf("Restarted spu %.2d\n", i);
 		} else {

@@ -2837,7 +2827,7 @@ do {									\
 		printf("  %-*s = "format"\n", DUMP_WIDTH,		\
 				#field, value);				\
 		sync();							\
-		__delay(200);						\
+		__delay(SETJMP_MACHINE_CHECK_DELAY);			\
 	} else {							\
 		catch_memory_errors = 0;				\
 		printf("  %-*s = *** Error reading field.\n",		\

@@ -2899,7 +2889,7 @@ static void dump_spu_ls(unsigned long nu
 		sync();
 		ls_addr = (unsigned long)spu_info[num].spu->local_store;
 		sync();
-		__delay(200);
+		__delay(SETJMP_MACHINE_CHECK_DELAY);
 	} else {
 		catch_memory_errors = 0;
 		printf("*** Error: accessing spu info for spu %d\n", num);

Index: clone3/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ clone3/include/asm-powerpc/setjmp.h

@@ -0,0 +1,19 @@
+/*
+ * Copyright (C) 2007 Michael Neuling
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define SETJMP_BUF_LEN    23
+#define SETJMP_MACHINE_CHECK_DELAY    200
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

[PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2007-12-13 09:59:17

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.

This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable.  

Tested with bogus data and instruction exceptions.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/crash.c |  102 +++++++++++++++++++++++++++++++++++++++++---
 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 100 insertions(+), 5 deletions(-)

Index: clone3/arch/powerpc/kernel/crash.c
===================================================================

--- clone3.orig/arch/powerpc/kernel/crash.c
+++ clone3/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_HANDLER_MAX 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
+static DEFINE_SPINLOCK(crash_handlers_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,72 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/*
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handlers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at first empty entry */
+			crash_shutdown_handles[i] = handler;
+			rc = 0;
+			break;
+		}
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc = 1;
+	}
+
+	spin_unlock(&crash_handlers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handlers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		rc = 1;
+	} else {
+		/* Shift handles down */
+		for (; crash_shutdown_handles[i]; i++)
+			crash_shutdown_handles[i] =
+				crash_shutdown_handles[i+1];
+		rc = 0;
+	}
+
+	spin_unlock(&crash_handlers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static unsigned long crash_shutdown_buf[SETJMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
+	int (*old_handler)(struct pt_regs *regs);
+
 
 	/*
 	 * This function is only called after the system

@@ -301,15 +371,37 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/*
+	 * Call registered shutdown routines savely.  Swap out
+	 * __debugger_fault_handler, and replace on exit.
+	 */
+	old_handler = __debugger_fault_handler;
+	__debugger_fault_handler = handle_fault;
+	for (i = 0; crash_shutdown_handles[i]; i++) {
+		if (setjmp(crash_shutdown_buf) == 0) {
+			/*
+			 * Insert syncs and delay to ensure
+			 * instructions in the dangerous region don't
+			 * leak away from this protected region.
+			 */
+			asm volatile("sync; isync");
+			/* dangerous region */
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			__delay(SETJMP_MACHINE_CHECK_DELAY);
+		}
 	}
+	__debugger_fault_handler = old_handler;
 
 	/*
 	 * Make a note of crashing cpu. Will be used in machine_kexec

Index: clone3/include/asm-powerpc/kexec.h
===================================================================

--- clone3.orig/include/asm-powerpc/kexec.h
+++ clone3/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

[PATCH 0/2] Add crashdump shutdown hooks

From: Michael Neuling <hidden>
Date: 2008-01-17 04:45:42

The following patches add crashdump shutdown hooks for POWERPC.

Signed-off-by: Michael Neuling <redacted>
---
Updated for suggests made by Paulus

[PATCH 1/2] Make setjmp/longjmp code generic

From: Michael Neuling <hidden>
Date: 2008-01-17 04:45:42

This makes the setjmp/longjmp code used by xmon, generically available
to other code.  It also removes the requirement for debugger hooks to
be only called on 0x300 (data storage) exception.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/misc.S   |   65 +++++++++++++++++++++++++++++++++++++++++++
 arch/powerpc/mm/fault.c      |    6 +--
 arch/powerpc/xmon/setjmp.S   |   61 ----------------------------------------
 arch/powerpc/xmon/xmon.c     |    6 ---
 include/asm-powerpc/setjmp.h |   18 +++++++++++
 5 files changed, 86 insertions(+), 70 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/misc.S
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/misc.S
+++ linux-2.6-ozlabs/arch/powerpc/kernel/misc.S

@@ -8,6 +8,8 @@
  * Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
  * PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
  *
+ * setjmp/longjmp and xmon_save_regs code by Paul Mackerras.
+ *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version

@@ -15,6 +17,8 @@
  */
 #include <asm/ppc_asm.h>
 #include <asm/unistd.h>
+#include <asm/asm-compat.h>
+#include <asm/asm-offsets.h>
 
 	.text

@@ -51,3 +55,64 @@ _GLOBAL(kernel_execve)
 	bnslr
 	neg	r3,r3
 	blr
+
+_GLOBAL(setjmp)
+	mflr	r0
+	PPC_STL	r0,0(r3)
+	PPC_STL	r1,SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	mfcr	r0
+	PPC_STL	r0,3*SZL(r3)
+	PPC_STL	r13,4*SZL(r3)
+	PPC_STL	r14,5*SZL(r3)
+	PPC_STL	r15,6*SZL(r3)
+	PPC_STL	r16,7*SZL(r3)
+	PPC_STL	r17,8*SZL(r3)
+	PPC_STL	r18,9*SZL(r3)
+	PPC_STL	r19,10*SZL(r3)
+	PPC_STL	r20,11*SZL(r3)
+	PPC_STL	r21,12*SZL(r3)
+	PPC_STL	r22,13*SZL(r3)
+	PPC_STL	r23,14*SZL(r3)
+	PPC_STL	r24,15*SZL(r3)
+	PPC_STL	r25,16*SZL(r3)
+	PPC_STL	r26,17*SZL(r3)
+	PPC_STL	r27,18*SZL(r3)
+	PPC_STL	r28,19*SZL(r3)
+	PPC_STL	r29,20*SZL(r3)
+	PPC_STL	r30,21*SZL(r3)
+	PPC_STL	r31,22*SZL(r3)
+	li	r3,0
+	blr
+
+_GLOBAL(longjmp)
+	PPC_LCMPI r4,0
+	bne	1f
+	li	r4,1
+1:	PPC_LL	r13,4*SZL(r3)
+	PPC_LL	r14,5*SZL(r3)
+	PPC_LL	r15,6*SZL(r3)
+	PPC_LL	r16,7*SZL(r3)
+	PPC_LL	r17,8*SZL(r3)
+	PPC_LL	r18,9*SZL(r3)
+	PPC_LL	r19,10*SZL(r3)
+	PPC_LL	r20,11*SZL(r3)
+	PPC_LL	r21,12*SZL(r3)
+	PPC_LL	r22,13*SZL(r3)
+	PPC_LL	r23,14*SZL(r3)
+	PPC_LL	r24,15*SZL(r3)
+	PPC_LL	r25,16*SZL(r3)
+	PPC_LL	r26,17*SZL(r3)
+	PPC_LL	r27,18*SZL(r3)
+	PPC_LL	r28,19*SZL(r3)
+	PPC_LL	r29,20*SZL(r3)
+	PPC_LL	r30,21*SZL(r3)
+	PPC_LL	r31,22*SZL(r3)
+	PPC_LL	r0,3*SZL(r3)
+	mtcrf	0x38,r0
+	PPC_LL	r0,0(r3)
+	PPC_LL	r1,SZL(r3)
+	PPC_LL	r2,2*SZL(r3)
+	mtlr	r0
+	mr	r3,r4
+	blr

Index: linux-2.6-ozlabs/arch/powerpc/mm/fault.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/mm/fault.c
+++ linux-2.6-ozlabs/arch/powerpc/mm/fault.c

@@ -167,10 +167,8 @@ int __kprobes do_page_fault(struct pt_re
 	if (notify_page_fault(regs))
 		return 0;
 
-	if (trap == 0x300) {
-		if (debugger_fault_handler(regs))
-			return 0;
-	}
+	if (unlikely(debugger_fault_handler(regs)))
+		return 0;
 
 	/* On a kernel SLB miss we can only check for a valid exception entry */
 	if (!user_mode(regs) && (address >= TASK_SIZE))

Index: linux-2.6-ozlabs/arch/powerpc/xmon/setjmp.S
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/xmon/setjmp.S
+++ linux-2.6-ozlabs/arch/powerpc/xmon/setjmp.S

@@ -12,67 +12,6 @@
 #include <asm/ppc_asm.h>
 #include <asm/asm-offsets.h>
 
-_GLOBAL(xmon_setjmp)
-	mflr	r0
-	PPC_STL	r0,0(r3)
-	PPC_STL	r1,SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	mfcr	r0
-	PPC_STL	r0,3*SZL(r3)
-	PPC_STL	r13,4*SZL(r3)
-	PPC_STL	r14,5*SZL(r3)
-	PPC_STL	r15,6*SZL(r3)
-	PPC_STL	r16,7*SZL(r3)
-	PPC_STL	r17,8*SZL(r3)
-	PPC_STL	r18,9*SZL(r3)
-	PPC_STL	r19,10*SZL(r3)
-	PPC_STL	r20,11*SZL(r3)
-	PPC_STL	r21,12*SZL(r3)
-	PPC_STL	r22,13*SZL(r3)
-	PPC_STL	r23,14*SZL(r3)
-	PPC_STL	r24,15*SZL(r3)
-	PPC_STL	r25,16*SZL(r3)
-	PPC_STL	r26,17*SZL(r3)
-	PPC_STL	r27,18*SZL(r3)
-	PPC_STL	r28,19*SZL(r3)
-	PPC_STL	r29,20*SZL(r3)
-	PPC_STL	r30,21*SZL(r3)
-	PPC_STL	r31,22*SZL(r3)
-	li	r3,0
-	blr
-
-_GLOBAL(xmon_longjmp)
-	PPC_LCMPI r4,0
-	bne	1f
-	li	r4,1
-1:	PPC_LL	r13,4*SZL(r3)
-	PPC_LL	r14,5*SZL(r3)
-	PPC_LL	r15,6*SZL(r3)
-	PPC_LL	r16,7*SZL(r3)
-	PPC_LL	r17,8*SZL(r3)
-	PPC_LL	r18,9*SZL(r3)
-	PPC_LL	r19,10*SZL(r3)
-	PPC_LL	r20,11*SZL(r3)
-	PPC_LL	r21,12*SZL(r3)
-	PPC_LL	r22,13*SZL(r3)
-	PPC_LL	r23,14*SZL(r3)
-	PPC_LL	r24,15*SZL(r3)
-	PPC_LL	r25,16*SZL(r3)
-	PPC_LL	r26,17*SZL(r3)
-	PPC_LL	r27,18*SZL(r3)
-	PPC_LL	r28,19*SZL(r3)
-	PPC_LL	r29,20*SZL(r3)
-	PPC_LL	r30,21*SZL(r3)
-	PPC_LL	r31,22*SZL(r3)
-	PPC_LL	r0,3*SZL(r3)
-	mtcrf	0x38,r0
-	PPC_LL	r0,0(r3)
-	PPC_LL	r1,SZL(r3)
-	PPC_LL	r2,2*SZL(r3)
-	mtlr	r0
-	mr	r3,r4
-	blr
-
 /*
  * Grab the register values as they are now.
  * This won't do a particularily good job because we really

Index: linux-2.6-ozlabs/arch/powerpc/xmon/xmon.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/xmon/xmon.c
+++ linux-2.6-ozlabs/arch/powerpc/xmon/xmon.c

@@ -40,6 +40,7 @@
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
 #include <asm/firmware.h>
+#include <asm/setjmp.h>
 
 #ifdef CONFIG_PPC64
 #include <asm/hvcall.h>

@@ -71,12 +72,9 @@ static unsigned long ncsum = 4096;
 static int termch;
 static char tmpstr[128];
 
-#define JMP_BUF_LEN	23
 static long bus_error_jmp[JMP_BUF_LEN];
 static int catch_memory_errors;
 static long *xmon_fault_jmp[NR_CPUS];
-#define setjmp xmon_setjmp
-#define longjmp xmon_longjmp
 
 /* Breakpoint stuff */
 struct bpt {

@@ -162,8 +160,6 @@ int xmon_no_auto_backtrace;
 extern void xmon_enter(void);
 extern void xmon_leave(void);
 
-extern long setjmp(long *);
-extern void longjmp(long *, long);
 extern void xmon_save_regs(struct pt_regs *);
 
 #ifdef CONFIG_PPC64

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2007 Michael Neuling
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

[PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2008-01-17 04:45:42

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.

This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable.  

Tested with bogus data and instruction exceptions.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/crash.c |  106 +++++++++++++++++++++++++++++++++++++++++---
 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 104 insertions(+), 5 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/crash.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/crash.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_HANDLER_MAX 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
+static DEFINE_SPINLOCK(crash_handlers_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,72 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/*
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handlers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at first empty entry */
+			crash_shutdown_handles[i] = handler;
+			rc = 0;
+			break;
+		}
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc = 1;
+	}
+
+	spin_unlock(&crash_handlers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handlers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		rc = 1;
+	} else {
+		/* Shift handles down */
+		for (; crash_shutdown_handles[i]; i++)
+			crash_shutdown_handles[i] =
+				crash_shutdown_handles[i+1];
+		rc = 0;
+	}
+
+	spin_unlock(&crash_handlers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
+	int (*old_handler)(struct pt_regs *regs);
+
 
 	/*
 	 * This function is only called after the system

@@ -301,15 +371,41 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/*
+	 * Call registered shutdown routines savely.  Swap out
+	 * __debugger_fault_handler, and replace on exit.
+	 */
+	old_handler = __debugger_fault_handler;
+	__debugger_fault_handler = handle_fault;
+	for (i = 0; crash_shutdown_handles[i]; i++) {
+		if (setjmp(crash_shutdown_buf) == 0) {
+			/*
+			 * Insert syncs and delay to ensure
+			 * instructions in the dangerous region don't
+			 * leak away from this protected region.
+			 */
+			asm volatile("sync; isync");
+			/* dangerous region */
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			/* 
+			 * wait a little while to see if we get a
+			 * machine check 
+			 */
+			__delay(200);
+		}
 	}
+	__debugger_fault_handler = old_handler;
 
 	/*
 	 * Make a note of crashing cpu. Will be used in machine_kexec

Index: linux-2.6-ozlabs/include/asm-powerpc/kexec.h
===================================================================

--- linux-2.6-ozlabs.orig/include/asm-powerpc/kexec.h
+++ linux-2.6-ozlabs/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

Re: [PATCH 1/2] Make setjmp/longjmp code generic

From: Olof Johansson <hidden>
Date: 2008-01-17 05:28:24

Hi,

On Thu, Jan 17, 2008 at 03:45:42PM +1100, Michael Neuling wrote:

quoted hunk

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2007 Michael Neuling
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

Should the above be inside #ifdef __KERNEL__?


-Olof

Re: [PATCH 1/2] Make setjmp/longjmp code generic

From: Michael Neuling <hidden>
Date: 2008-01-17 05:36:49


In message [off-list ref] you wrote:

Hi,

On Thu, Jan 17, 2008 at 03:45:42PM +1100, Michael Neuling wrote:

quoted

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2007 Michael Neuling

                     ^^^^^^

quoted

+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

Should the above be inside #ifdef __KERNEL__?

Yep... _and_ it's 2008 now!

I'll update.

Mikey

Re: [PATCH 1/2] Make setjmp/longjmp code generic

From: Stephen Rothwell <hidden>
Date: 2008-01-17 06:36:53

On Thu, 17 Jan 2008 16:36:49 +1100 Michael Neuling [off-list ref] wrote:

quoted

Should the above be inside #ifdef __KERNEL__?

Yep... _and_ it's 2008 now!

I think that if you don't list the file for exporting to user mode (in
the Kbuild file), then you shouldn't need the __KERNEL__ protectors.

But it is, indeed, 2008. :-)

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

Re: [PATCH 1/2] Make setjmp/longjmp code generic

From: Stephen Rothwell <hidden>
Date: 2008-01-17 06:39:38

On Thu, 17 Jan 2008 15:45:42 +1100 Michael Neuling [off-list ref] wrote:

quoted hunk

+++ linux-2.6-ozlabs/arch/powerpc/kernel/misc.S

@@ -8,6 +8,8 @@
  * Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
  * PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
  *
+ * setjmp/longjmp and xmon_save_regs code by Paul Mackerras.

                         ^^^^^^^^^^^^^^
This doesn't exist here.

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

Re: [PATCH 1/2] Make setjmp/longjmp code generic

From: Josh Boyer <hidden>
Date: 2008-01-17 12:47:16

On Thu, 17 Jan 2008 16:36:49 +1100
Michael Neuling [off-list ref] wrote:


In message [off-list ref] you wrote:

quoted

Hi,

On Thu, Jan 17, 2008 at 03:45:42PM +1100, Michael Neuling wrote:

quoted

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2007 Michael Neuling

                     ^^^^^^

quoted

+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

Should the above be inside #ifdef __KERNEL__?

Yep... _and_ it's 2008 now!

I'll update.

While you're off updating it, remove the (C).  It's bogus and shouldn't
be there.

josh

Re: [PATCH 1/2] Make setjmp/longjmp code generic

From: Michael Ellerman <hidden>
Date: 2008-01-17 21:49:22

On Thu, 2008-01-17 at 06:45 -0600, Josh Boyer wrote:

On Thu, 17 Jan 2008 16:36:49 +1100
Michael Neuling [off-list ref] wrote:

quoted

In message [off-list ref] you wrote:

quoted

Hi,

On Thu, Jan 17, 2008 at 03:45:42PM +1100, Michael Neuling wrote:

quoted

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2007 Michael Neuling

                     ^^^^^^

quoted

+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

Should the above be inside #ifdef __KERNEL__?

Yep... _and_ it's 2008 now!

I'll update.

While you're off updating it, remove the (C).  It's bogus and shouldn't
be there.

And it should say IBM corp too.

Oh and my car needs a wash when you're free ..

cheers

-- 
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

[PATCH] Add crashdump shutdown hooks

From: Michael Neuling <hidden>
Date: 2008-01-17 22:16:07

The following patches add crashdump shutdown hooks for POWERPC.

Signed-off-by: Michael Neuling <redacted>
---
Updated some comments in the code, as noticed by sfr, Josh Boyer and a
dirty car owner.

[PATCH] Make setjmp/longjmp code generic

From: Michael Neuling <hidden>
Date: 2008-01-17 22:16:07

This makes the setjmp/longjmp code used by xmon, generically available
to other code.  It also removes the requirement for debugger hooks to
be only called on 0x300 (data storage) exception.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/misc.S   |   65 +++++++++++++++++++++++++++++++++++++++++++
 arch/powerpc/mm/fault.c      |    6 +--
 arch/powerpc/xmon/setjmp.S   |   61 ----------------------------------------
 arch/powerpc/xmon/xmon.c     |    6 ---
 include/asm-powerpc/setjmp.h |   18 +++++++++++
 5 files changed, 86 insertions(+), 70 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/misc.S
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/misc.S
+++ linux-2.6-ozlabs/arch/powerpc/kernel/misc.S

@@ -8,6 +8,8 @@
  * Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
  * PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
  *
+ * setjmp/longjmp code by Paul Mackerras.
+ *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version

@@ -15,6 +17,8 @@
  */
 #include <asm/ppc_asm.h>
 #include <asm/unistd.h>
+#include <asm/asm-compat.h>
+#include <asm/asm-offsets.h>
 
 	.text

@@ -51,3 +55,64 @@ _GLOBAL(kernel_execve)
 	bnslr
 	neg	r3,r3
 	blr
+
+_GLOBAL(setjmp)
+	mflr	r0
+	PPC_STL	r0,0(r3)
+	PPC_STL	r1,SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	mfcr	r0
+	PPC_STL	r0,3*SZL(r3)
+	PPC_STL	r13,4*SZL(r3)
+	PPC_STL	r14,5*SZL(r3)
+	PPC_STL	r15,6*SZL(r3)
+	PPC_STL	r16,7*SZL(r3)
+	PPC_STL	r17,8*SZL(r3)
+	PPC_STL	r18,9*SZL(r3)
+	PPC_STL	r19,10*SZL(r3)
+	PPC_STL	r20,11*SZL(r3)
+	PPC_STL	r21,12*SZL(r3)
+	PPC_STL	r22,13*SZL(r3)
+	PPC_STL	r23,14*SZL(r3)
+	PPC_STL	r24,15*SZL(r3)
+	PPC_STL	r25,16*SZL(r3)
+	PPC_STL	r26,17*SZL(r3)
+	PPC_STL	r27,18*SZL(r3)
+	PPC_STL	r28,19*SZL(r3)
+	PPC_STL	r29,20*SZL(r3)
+	PPC_STL	r30,21*SZL(r3)
+	PPC_STL	r31,22*SZL(r3)
+	li	r3,0
+	blr
+
+_GLOBAL(longjmp)
+	PPC_LCMPI r4,0
+	bne	1f
+	li	r4,1
+1:	PPC_LL	r13,4*SZL(r3)
+	PPC_LL	r14,5*SZL(r3)
+	PPC_LL	r15,6*SZL(r3)
+	PPC_LL	r16,7*SZL(r3)
+	PPC_LL	r17,8*SZL(r3)
+	PPC_LL	r18,9*SZL(r3)
+	PPC_LL	r19,10*SZL(r3)
+	PPC_LL	r20,11*SZL(r3)
+	PPC_LL	r21,12*SZL(r3)
+	PPC_LL	r22,13*SZL(r3)
+	PPC_LL	r23,14*SZL(r3)
+	PPC_LL	r24,15*SZL(r3)
+	PPC_LL	r25,16*SZL(r3)
+	PPC_LL	r26,17*SZL(r3)
+	PPC_LL	r27,18*SZL(r3)
+	PPC_LL	r28,19*SZL(r3)
+	PPC_LL	r29,20*SZL(r3)
+	PPC_LL	r30,21*SZL(r3)
+	PPC_LL	r31,22*SZL(r3)
+	PPC_LL	r0,3*SZL(r3)
+	mtcrf	0x38,r0
+	PPC_LL	r0,0(r3)
+	PPC_LL	r1,SZL(r3)
+	PPC_LL	r2,2*SZL(r3)
+	mtlr	r0
+	mr	r3,r4
+	blr

Index: linux-2.6-ozlabs/arch/powerpc/mm/fault.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/mm/fault.c
+++ linux-2.6-ozlabs/arch/powerpc/mm/fault.c

@@ -167,10 +167,8 @@ int __kprobes do_page_fault(struct pt_re
 	if (notify_page_fault(regs))
 		return 0;
 
-	if (trap == 0x300) {
-		if (debugger_fault_handler(regs))
-			return 0;
-	}
+	if (unlikely(debugger_fault_handler(regs)))
+		return 0;
 
 	/* On a kernel SLB miss we can only check for a valid exception entry */
 	if (!user_mode(regs) && (address >= TASK_SIZE))

Index: linux-2.6-ozlabs/arch/powerpc/xmon/setjmp.S
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/xmon/setjmp.S
+++ linux-2.6-ozlabs/arch/powerpc/xmon/setjmp.S

@@ -12,67 +12,6 @@
 #include <asm/ppc_asm.h>
 #include <asm/asm-offsets.h>
 
-_GLOBAL(xmon_setjmp)
-	mflr	r0
-	PPC_STL	r0,0(r3)
-	PPC_STL	r1,SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	mfcr	r0
-	PPC_STL	r0,3*SZL(r3)
-	PPC_STL	r13,4*SZL(r3)
-	PPC_STL	r14,5*SZL(r3)
-	PPC_STL	r15,6*SZL(r3)
-	PPC_STL	r16,7*SZL(r3)
-	PPC_STL	r17,8*SZL(r3)
-	PPC_STL	r18,9*SZL(r3)
-	PPC_STL	r19,10*SZL(r3)
-	PPC_STL	r20,11*SZL(r3)
-	PPC_STL	r21,12*SZL(r3)
-	PPC_STL	r22,13*SZL(r3)
-	PPC_STL	r23,14*SZL(r3)
-	PPC_STL	r24,15*SZL(r3)
-	PPC_STL	r25,16*SZL(r3)
-	PPC_STL	r26,17*SZL(r3)
-	PPC_STL	r27,18*SZL(r3)
-	PPC_STL	r28,19*SZL(r3)
-	PPC_STL	r29,20*SZL(r3)
-	PPC_STL	r30,21*SZL(r3)
-	PPC_STL	r31,22*SZL(r3)
-	li	r3,0
-	blr
-
-_GLOBAL(xmon_longjmp)
-	PPC_LCMPI r4,0
-	bne	1f
-	li	r4,1
-1:	PPC_LL	r13,4*SZL(r3)
-	PPC_LL	r14,5*SZL(r3)
-	PPC_LL	r15,6*SZL(r3)
-	PPC_LL	r16,7*SZL(r3)
-	PPC_LL	r17,8*SZL(r3)
-	PPC_LL	r18,9*SZL(r3)
-	PPC_LL	r19,10*SZL(r3)
-	PPC_LL	r20,11*SZL(r3)
-	PPC_LL	r21,12*SZL(r3)
-	PPC_LL	r22,13*SZL(r3)
-	PPC_LL	r23,14*SZL(r3)
-	PPC_LL	r24,15*SZL(r3)
-	PPC_LL	r25,16*SZL(r3)
-	PPC_LL	r26,17*SZL(r3)
-	PPC_LL	r27,18*SZL(r3)
-	PPC_LL	r28,19*SZL(r3)
-	PPC_LL	r29,20*SZL(r3)
-	PPC_LL	r30,21*SZL(r3)
-	PPC_LL	r31,22*SZL(r3)
-	PPC_LL	r0,3*SZL(r3)
-	mtcrf	0x38,r0
-	PPC_LL	r0,0(r3)
-	PPC_LL	r1,SZL(r3)
-	PPC_LL	r2,2*SZL(r3)
-	mtlr	r0
-	mr	r3,r4
-	blr
-
 /*
  * Grab the register values as they are now.
  * This won't do a particularily good job because we really

Index: linux-2.6-ozlabs/arch/powerpc/xmon/xmon.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/xmon/xmon.c
+++ linux-2.6-ozlabs/arch/powerpc/xmon/xmon.c

@@ -40,6 +40,7 @@
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
 #include <asm/firmware.h>
+#include <asm/setjmp.h>
 
 #ifdef CONFIG_PPC64
 #include <asm/hvcall.h>

@@ -71,12 +72,9 @@ static unsigned long ncsum = 4096;
 static int termch;
 static char tmpstr[128];
 
-#define JMP_BUF_LEN	23
 static long bus_error_jmp[JMP_BUF_LEN];
 static int catch_memory_errors;
 static long *xmon_fault_jmp[NR_CPUS];
-#define setjmp xmon_setjmp
-#define longjmp xmon_longjmp
 
 /* Breakpoint stuff */
 struct bpt {

@@ -162,8 +160,6 @@ int xmon_no_auto_backtrace;
 extern void xmon_enter(void);
 extern void xmon_leave(void);
 
-extern long setjmp(long *);
-extern void longjmp(long *, long);
 extern void xmon_save_regs(struct pt_regs *);
 
 #ifdef CONFIG_PPC64

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright �� 2008 Michael Neuling IBM Corporation
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

[PATCH 0/2] Add crashdump shutdown hooks

From: Michael Neuling <hidden>
Date: 2008-01-18 04:50:30

The following patches add crashdump shutdown hooks for POWERPC.

Signed-off-by: Michael Neuling <redacted>
---
Sorry, last update didn't include both patches.

[PATCH 1/2] Make setjmp/longjmp code generic

From: Michael Neuling <hidden>
Date: 2008-01-18 04:50:30

This makes the setjmp/longjmp code used by xmon, generically available
to other code.  It also removes the requirement for debugger hooks to
be only called on 0x300 (data storage) exception.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/misc.S   |   65 +++++++++++++++++++++++++++++++++++++++++++
 arch/powerpc/mm/fault.c      |    6 +--
 arch/powerpc/xmon/setjmp.S   |   61 ----------------------------------------
 arch/powerpc/xmon/xmon.c     |    6 ---
 include/asm-powerpc/setjmp.h |   18 +++++++++++
 5 files changed, 86 insertions(+), 70 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/misc.S
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/misc.S
+++ linux-2.6-ozlabs/arch/powerpc/kernel/misc.S

@@ -8,6 +8,8 @@
  * Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
  * PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
  *
+ * setjmp/longjmp code by Paul Mackerras.
+ *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version

@@ -15,6 +17,8 @@
  */
 #include <asm/ppc_asm.h>
 #include <asm/unistd.h>
+#include <asm/asm-compat.h>
+#include <asm/asm-offsets.h>
 
 	.text

@@ -51,3 +55,64 @@ _GLOBAL(kernel_execve)
 	bnslr
 	neg	r3,r3
 	blr
+
+_GLOBAL(setjmp)
+	mflr	r0
+	PPC_STL	r0,0(r3)
+	PPC_STL	r1,SZL(r3)
+	PPC_STL	r2,2*SZL(r3)
+	mfcr	r0
+	PPC_STL	r0,3*SZL(r3)
+	PPC_STL	r13,4*SZL(r3)
+	PPC_STL	r14,5*SZL(r3)
+	PPC_STL	r15,6*SZL(r3)
+	PPC_STL	r16,7*SZL(r3)
+	PPC_STL	r17,8*SZL(r3)
+	PPC_STL	r18,9*SZL(r3)
+	PPC_STL	r19,10*SZL(r3)
+	PPC_STL	r20,11*SZL(r3)
+	PPC_STL	r21,12*SZL(r3)
+	PPC_STL	r22,13*SZL(r3)
+	PPC_STL	r23,14*SZL(r3)
+	PPC_STL	r24,15*SZL(r3)
+	PPC_STL	r25,16*SZL(r3)
+	PPC_STL	r26,17*SZL(r3)
+	PPC_STL	r27,18*SZL(r3)
+	PPC_STL	r28,19*SZL(r3)
+	PPC_STL	r29,20*SZL(r3)
+	PPC_STL	r30,21*SZL(r3)
+	PPC_STL	r31,22*SZL(r3)
+	li	r3,0
+	blr
+
+_GLOBAL(longjmp)
+	PPC_LCMPI r4,0
+	bne	1f
+	li	r4,1
+1:	PPC_LL	r13,4*SZL(r3)
+	PPC_LL	r14,5*SZL(r3)
+	PPC_LL	r15,6*SZL(r3)
+	PPC_LL	r16,7*SZL(r3)
+	PPC_LL	r17,8*SZL(r3)
+	PPC_LL	r18,9*SZL(r3)
+	PPC_LL	r19,10*SZL(r3)
+	PPC_LL	r20,11*SZL(r3)
+	PPC_LL	r21,12*SZL(r3)
+	PPC_LL	r22,13*SZL(r3)
+	PPC_LL	r23,14*SZL(r3)
+	PPC_LL	r24,15*SZL(r3)
+	PPC_LL	r25,16*SZL(r3)
+	PPC_LL	r26,17*SZL(r3)
+	PPC_LL	r27,18*SZL(r3)
+	PPC_LL	r28,19*SZL(r3)
+	PPC_LL	r29,20*SZL(r3)
+	PPC_LL	r30,21*SZL(r3)
+	PPC_LL	r31,22*SZL(r3)
+	PPC_LL	r0,3*SZL(r3)
+	mtcrf	0x38,r0
+	PPC_LL	r0,0(r3)
+	PPC_LL	r1,SZL(r3)
+	PPC_LL	r2,2*SZL(r3)
+	mtlr	r0
+	mr	r3,r4
+	blr

Index: linux-2.6-ozlabs/arch/powerpc/mm/fault.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/mm/fault.c
+++ linux-2.6-ozlabs/arch/powerpc/mm/fault.c

@@ -167,10 +167,8 @@ int __kprobes do_page_fault(struct pt_re
 	if (notify_page_fault(regs))
 		return 0;
 
-	if (trap == 0x300) {
-		if (debugger_fault_handler(regs))
-			return 0;
-	}
+	if (unlikely(debugger_fault_handler(regs)))
+		return 0;
 
 	/* On a kernel SLB miss we can only check for a valid exception entry */
 	if (!user_mode(regs) && (address >= TASK_SIZE))

Index: linux-2.6-ozlabs/arch/powerpc/xmon/setjmp.S
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/xmon/setjmp.S
+++ linux-2.6-ozlabs/arch/powerpc/xmon/setjmp.S

@@ -12,67 +12,6 @@
 #include <asm/ppc_asm.h>
 #include <asm/asm-offsets.h>
 
-_GLOBAL(xmon_setjmp)
-	mflr	r0
-	PPC_STL	r0,0(r3)
-	PPC_STL	r1,SZL(r3)
-	PPC_STL	r2,2*SZL(r3)
-	mfcr	r0
-	PPC_STL	r0,3*SZL(r3)
-	PPC_STL	r13,4*SZL(r3)
-	PPC_STL	r14,5*SZL(r3)
-	PPC_STL	r15,6*SZL(r3)
-	PPC_STL	r16,7*SZL(r3)
-	PPC_STL	r17,8*SZL(r3)
-	PPC_STL	r18,9*SZL(r3)
-	PPC_STL	r19,10*SZL(r3)
-	PPC_STL	r20,11*SZL(r3)
-	PPC_STL	r21,12*SZL(r3)
-	PPC_STL	r22,13*SZL(r3)
-	PPC_STL	r23,14*SZL(r3)
-	PPC_STL	r24,15*SZL(r3)
-	PPC_STL	r25,16*SZL(r3)
-	PPC_STL	r26,17*SZL(r3)
-	PPC_STL	r27,18*SZL(r3)
-	PPC_STL	r28,19*SZL(r3)
-	PPC_STL	r29,20*SZL(r3)
-	PPC_STL	r30,21*SZL(r3)
-	PPC_STL	r31,22*SZL(r3)
-	li	r3,0
-	blr
-
-_GLOBAL(xmon_longjmp)
-	PPC_LCMPI r4,0
-	bne	1f
-	li	r4,1
-1:	PPC_LL	r13,4*SZL(r3)
-	PPC_LL	r14,5*SZL(r3)
-	PPC_LL	r15,6*SZL(r3)
-	PPC_LL	r16,7*SZL(r3)
-	PPC_LL	r17,8*SZL(r3)
-	PPC_LL	r18,9*SZL(r3)
-	PPC_LL	r19,10*SZL(r3)
-	PPC_LL	r20,11*SZL(r3)
-	PPC_LL	r21,12*SZL(r3)
-	PPC_LL	r22,13*SZL(r3)
-	PPC_LL	r23,14*SZL(r3)
-	PPC_LL	r24,15*SZL(r3)
-	PPC_LL	r25,16*SZL(r3)
-	PPC_LL	r26,17*SZL(r3)
-	PPC_LL	r27,18*SZL(r3)
-	PPC_LL	r28,19*SZL(r3)
-	PPC_LL	r29,20*SZL(r3)
-	PPC_LL	r30,21*SZL(r3)
-	PPC_LL	r31,22*SZL(r3)
-	PPC_LL	r0,3*SZL(r3)
-	mtcrf	0x38,r0
-	PPC_LL	r0,0(r3)
-	PPC_LL	r1,SZL(r3)
-	PPC_LL	r2,2*SZL(r3)
-	mtlr	r0
-	mr	r3,r4
-	blr
-
 /*
  * Grab the register values as they are now.
  * This won't do a particularily good job because we really

Index: linux-2.6-ozlabs/arch/powerpc/xmon/xmon.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/xmon/xmon.c
+++ linux-2.6-ozlabs/arch/powerpc/xmon/xmon.c

@@ -40,6 +40,7 @@
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
 #include <asm/firmware.h>
+#include <asm/setjmp.h>
 
 #ifdef CONFIG_PPC64
 #include <asm/hvcall.h>

@@ -71,12 +72,9 @@ static unsigned long ncsum = 4096;
 static int termch;
 static char tmpstr[128];
 
-#define JMP_BUF_LEN	23
 static long bus_error_jmp[JMP_BUF_LEN];
 static int catch_memory_errors;
 static long *xmon_fault_jmp[NR_CPUS];
-#define setjmp xmon_setjmp
-#define longjmp xmon_longjmp
 
 /* Breakpoint stuff */
 struct bpt {

@@ -162,8 +160,6 @@ int xmon_no_auto_backtrace;
 extern void xmon_enter(void);
 extern void xmon_leave(void);
 
-extern long setjmp(long *);
-extern void longjmp(long *, long);
 extern void xmon_save_regs(struct pt_regs *);
 
 #ifdef CONFIG_PPC64

Index: linux-2.6-ozlabs/include/asm-powerpc/setjmp.h
===================================================================

--- /dev/null
+++ linux-2.6-ozlabs/include/asm-powerpc/setjmp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright �� 2008 Michael Neuling IBM Corporation
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ */
+#ifndef _ASM_POWERPC_SETJMP_H
+#define _ASM_POWERPC_SETJMP_H
+
+#define JMP_BUF_LEN    23
+
+extern long setjmp(long *);
+extern void longjmp(long *, long);
+
+#endif /* _ASM_POWERPC_SETJMP_H */

[PATCH 2/2] kdump shutdown hook support

From: Michael Neuling <hidden>
Date: 2008-01-18 04:50:30

This adds hooks into the default_machine_crash_shutdown so drivers can
register a function to be run in the first kernel before we hand off
to the second kernel.  This should only be used in exceptional
circumstances, like where the device can't be reset in the second
kernel alone (as is the case with eHEA).  To emphasize this, the
number of handles allowed to be registered is currently #def to 1.

This uses the setjmp/longjmp code to call out to the registered hooks,
so any bogus exceptions we encounter will hopefully be recoverable.  

Tested with bogus data and instruction exceptions.

Signed-off-by: Michael Neuling <redacted>
---

 arch/powerpc/kernel/crash.c |  106 +++++++++++++++++++++++++++++++++++++++++---
 include/asm-powerpc/kexec.h |    3 +
 2 files changed, 104 insertions(+), 5 deletions(-)

Index: linux-2.6-ozlabs/arch/powerpc/kernel/crash.c
===================================================================

--- linux-2.6-ozlabs.orig/arch/powerpc/kernel/crash.c
+++ linux-2.6-ozlabs/arch/powerpc/kernel/crash.c

@@ -32,6 +32,8 @@
 #include <asm/lmb.h>
 #include <asm/firmware.h>
 #include <asm/smp.h>
+#include <asm/system.h>
+#include <asm/setjmp.h>
 
 #ifdef DEBUG
 #include <asm/udbg.h>

@@ -45,6 +47,11 @@ int crashing_cpu = -1;
 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
 cpumask_t cpus_in_sr = CPU_MASK_NONE;
 
+#define CRASH_HANDLER_MAX 1
+/* NULL terminated list of shutdown handles */
+static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
+static DEFINE_SPINLOCK(crash_handlers_lock);
+
 #ifdef CONFIG_SMP
 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);

@@ -285,9 +292,72 @@ static inline void crash_kexec_stop_spus
 }
 #endif /* CONFIG_SPU_BASE */
 
+/*
+ * Register a function to be called on shutdown.  Only use this if you
+ * can't reset your device in the second kernel.
+ */
+int crash_shutdown_register(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handlers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (!crash_shutdown_handles[i]) {
+			/* Insert handle at first empty entry */
+			crash_shutdown_handles[i] = handler;
+			rc = 0;
+			break;
+		}
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handles full, "
+		       "not registered.\n");
+		rc = 1;
+	}
+
+	spin_unlock(&crash_handlers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_register);
+
+int crash_shutdown_unregister(crash_shutdown_t handler)
+{
+	unsigned int i, rc;
+
+	spin_lock(&crash_handlers_lock);
+	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
+		if (crash_shutdown_handles[i] == handler)
+			break;
+
+	if (i == CRASH_HANDLER_MAX) {
+		printk(KERN_ERR "Crash shutdown handle not found\n");
+		rc = 1;
+	} else {
+		/* Shift handles down */
+		for (; crash_shutdown_handles[i]; i++)
+			crash_shutdown_handles[i] =
+				crash_shutdown_handles[i+1];
+		rc = 0;
+	}
+
+	spin_unlock(&crash_handlers_lock);
+	return rc;
+}
+EXPORT_SYMBOL(crash_shutdown_unregister);
+
+static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
+
+static int handle_fault(struct pt_regs *regs)
+{
+	longjmp(crash_shutdown_buf, 1);
+	return 0;
+}
+
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
-	unsigned int irq;
+	unsigned int i;
+	int (*old_handler)(struct pt_regs *regs);
+
 
 	/*
 	 * This function is only called after the system

@@ -301,15 +371,41 @@ void default_machine_crash_shutdown(stru
 	 */
 	hard_irq_disable();
 
-	for_each_irq(irq) {
-		struct irq_desc *desc = irq_desc + irq;
+	for_each_irq(i) {
+		struct irq_desc *desc = irq_desc + i;
 
 		if (desc->status & IRQ_INPROGRESS)
-			desc->chip->eoi(irq);
+			desc->chip->eoi(i);
 
 		if (!(desc->status & IRQ_DISABLED))
-			desc->chip->disable(irq);
+			desc->chip->disable(i);
+	}
+
+	/*
+	 * Call registered shutdown routines savely.  Swap out
+	 * __debugger_fault_handler, and replace on exit.
+	 */
+	old_handler = __debugger_fault_handler;
+	__debugger_fault_handler = handle_fault;
+	for (i = 0; crash_shutdown_handles[i]; i++) {
+		if (setjmp(crash_shutdown_buf) == 0) {
+			/*
+			 * Insert syncs and delay to ensure
+			 * instructions in the dangerous region don't
+			 * leak away from this protected region.
+			 */
+			asm volatile("sync; isync");
+			/* dangerous region */
+			crash_shutdown_handles[i]();
+			asm volatile("sync; isync");
+			/* 
+			 * wait a little while to see if we get a
+			 * machine check 
+			 */
+			__delay(200);
+		}
 	}
+	__debugger_fault_handler = old_handler;
 
 	/*
 	 * Make a note of crashing cpu. Will be used in machine_kexec

Index: linux-2.6-ozlabs/include/asm-powerpc/kexec.h
===================================================================

--- linux-2.6-ozlabs.orig/include/asm-powerpc/kexec.h
+++ linux-2.6-ozlabs/include/asm-powerpc/kexec.h

@@ -123,6 +123,9 @@ struct pt_regs;
 extern void default_machine_kexec(struct kimage *image);
 extern int default_machine_kexec_prepare(struct kimage *image);
 extern void default_machine_crash_shutdown(struct pt_regs *regs);
+typedef void (*crash_shutdown_t)(void);
+extern int crash_shutdown_register(crash_shutdown_t handler);
+extern int crash_shutdown_unregister(crash_shutdown_t handler);
 
 extern void machine_kexec_simple(struct kimage *image);
 extern void crash_kexec_secondary(struct pt_regs *regs);

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help