Skip to contentSkip to conversation

- SN: validate smp_affinity mask on intr redirect

PATCH Superseded John Keller 2 participants 3 messages May 3, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 - SN: validate smp_affinity mask on intr redirect

On SN, only allow one bit to be set in the smp_affinty mask when redirecting an interrupt. Currently setting multiple bits is allowed, but only the first bit is used in determining the CPU to redirect to. This has caused confusion among some customers.

---

 arch/ia64/kernel/irq.c |   11 +++++++++++
 include/asm-ia64/irq.h |    5 +++++
 include/linux/irq.h    |    4 ++++
 kernel/irq/proc.c      |    3 +++
 4 files changed, 23 insertions(+)

Index: linux-2.6/kernel/irq/proc.c
=================================--- linux-2.6.orig/kernel/irq/proc.c 2007-05-01 15:47:01.656642577 -0500
+++ linux-2.6/kernel/irq/proc.c 2007-05-01 15:48:42.725254306 -0500
@@ -42,6 +42,9 @@ static int irq_affinity_write_proc(struc
if (err)
return err;

+ if (!is_affinity_mask_valid(new_value))
+ return -EINVAL;
+
/*
* Do not allow disabling IRQs completely - it's a too easy
* way to make the system unusable accidentally :-) At least
Index: linux-2.6/arch/ia64/kernel/irq.c
=================================--- linux-2.6.orig/arch/ia64/kernel/irq.c 2007-05-01 15:47:01.660643076 -0500
+++ linux-2.6/arch/ia64/kernel/irq.c 2007-05-01 15:48:42.729254805 -0500
@@ -104,6 +104,17 @@ void set_irq_affinity_info (unsigned int
irq_redir[irq] = (char) (redir & 0xff);
}
}
+
+int is_affinity_mask_valid(cpumask_t cpumask)
+{
+ if (ia64_platform_is("sn2")) {
+ /* Only allow one CPU to be specified in the smp_affinity mask */
+ if (cpus_weight(cpumask) != 1)
+ return 0;
+ }
+ return 1;
+}
+
#endif /* CONFIG_SMP */

#ifdef CONFIG_HOTPLUG_CPU
Index: linux-2.6/include/linux/irq.h
=================================--- linux-2.6.orig/include/linux/irq.h 2007-05-01 15:47:01.660643076 -0500
+++ linux-2.6/include/linux/irq.h 2007-05-01 15:48:42.729254805 -0500
@@ -228,6 +228,10 @@ static inline void set_pending_irq(unsig

#endif /* CONFIG_GENERIC_PENDING_IRQ */

+#ifndef ARCH_HAS_IRQ_AFFINITY_VALIDATION
+#define is_irq_affinity_mask_valid(cpumask) 1
+#endif
+
extern int irq_set_affinity(unsigned int irq, cpumask_t cpumask); extern int irq_can_set_affinity(unsigned int irq);

Index: linux-2.6/include/asm-ia64/irq.h
=================================--- linux-2.6.orig/include/asm-ia64/irq.h 2007-05-01 15:47:01.660643076 -0500
+++ linux-2.6/include/asm-ia64/irq.h 2007-05-01 15:48:42.729254805 -0500
@@ -11,6 +11,8 @@
* 02/29/00 D.Mosberger moved most things into hw_irq.h
*/

+#include <linux/cpumask.h>
+
#define NR_IRQS 256
#define NR_IRQ_VECTORS NR_IRQS

@@ -30,4 +32,7 @@ extern void disable_irq_nosync (unsigned
extern void enable_irq (unsigned int);
extern void set_irq_affinity_info (unsigned int irq, int dest, int redir);

+#define ARCH_HAS_IRQ_AFFINITY_VALIDATION +extern int is_affinity_mask_valid(cpumask_t cpumask); +
#endif /* _ASM_IA64_IRQ_H */

#1 lore raw
Quoted 8 lines · On Thu, 03 May 2007 07:56:02 -0500 John Keller <[email protected]> wrote:

On SN, only allow one bit to be set in the smp_affinty mask when redirecting an interrupt. Currently setting multiple bits is allowed, but only the first bit is used in determining the CPU to redirect to. This has caused confusion among some customers.

...

Review

Quoted 10 lines
(quoted diff) +10 -0
+
+int is_affinity_mask_valid(cpumask_t cpumask)
+{
+	if (ia64_platform_is("sn2")) {
+		/* Only allow one CPU to be specified in the smp_affinity mask */
+		if (cpus_weight(cpumask) != 1)
+			return 0;
+	}
+	return 1;
+}

A bool returning true or false would be appropriate, given this function's clearly-boolean name.

Review

Quoted 3 lines
(quoted diff) +3 -0
+#ifndef ARCH_HAS_IRQ_AFFINITY_VALIDATION
+#define is_irq_affinity_mask_valid(cpumask)         1
+#endif

The ARCH_HAS_FOO things have never been popular. There are a couple of alternatives going around:

a) use __attribute__((weak)) and provide a one-line default implementation in kernel/irq/proc.c. This has a small runtime cost.

b) In include/asm-ia64/irq.h, do

   #define is_affinity_mask_valid is_affinity_mask_valid

and in kernel/irq/proc.c do

   #ifndef is_affinity_mask_valid
   #define is_affinity_mask_valid() 1
   #endif

This has minimal runtime cost, but is a bit tricksy.

It has the advantage that it doesn't introduce some new identifier which needs to be maintained.

Review line 15

Quoted 12 lines

extern int irq_set_affinity(unsigned int irq, cpumask_t cpumask); extern int irq_can_set_affinity(unsigned int irq);

Index: linux-2.6/include/asm-ia64/irq.h
=================================> --- linux-2.6.orig/include/asm-ia64/irq.h 2007-05-01 15:47:01.660643076 -0500
+++ linux-2.6/include/asm-ia64/irq.h 2007-05-01 15:48:42.729254805 -0500

(quoted diff) +2 -0
@@ -11,6 +11,8 @@
  * 02/29/00     D.Mosberger	moved most things into hw_irq.h
  */
 
+#include <linux/cpumask.h>
+

This is the sort of inclusion which tends to make things blow up. I trust this was carefully compile-tested.

With a) above, the is_affinity_mask_valid() declaration would be in a different header. With b), no additional include will be needed.

Quoted 10 lines
(quoted diff) +3 -0
 #define NR_IRQS		256
 #define NR_IRQ_VECTORS	NR_IRQS
 
@@ -30,4 +32,7 @@ extern void disable_irq_nosync (unsigned
 extern void enable_irq (unsigned int);
 extern void set_irq_affinity_info (unsigned int irq, int dest, int redir);
 
+#define ARCH_HAS_IRQ_AFFINITY_VALIDATION
+extern int is_affinity_mask_valid(cpumask_t cpumask);
+
#2 lore raw

I'll rework as per your suggestions.

Thanks,
John

Quoted 82 lines
Quoted 8 lines · On Thu, 03 May 2007 07:56:02 -0500 John Keller <[email protected]> wrote:

On SN, only allow one bit to be set in the smp_affinty mask when redirecting an interrupt. Currently setting multiple bits is allowed, but only the first bit is used in determining the CPU to redirect to. This has caused confusion among some customers.

...

Review

Quoted 10 lines
(quoted diff) +10 -0
+
+int is_affinity_mask_valid(cpumask_t cpumask)
+{
+	if (ia64_platform_is("sn2")) {
+		/* Only allow one CPU to be specified in the smp_affinity mask */
+		if (cpus_weight(cpumask) != 1)
+			return 0;
+	}
+	return 1;
+}

A bool returning true or false would be appropriate, given this function's clearly-boolean name.

Review

Quoted 3 lines
(quoted diff) +3 -0
+#ifndef ARCH_HAS_IRQ_AFFINITY_VALIDATION
+#define is_irq_affinity_mask_valid(cpumask)         1
+#endif

The ARCH_HAS_FOO things have never been popular. There are a couple of alternatives going around:

a) use __attribute__((weak)) and provide a one-line default implementation in kernel/irq/proc.c. This has a small runtime cost.

b) In include/asm-ia64/irq.h, do

   #define is_affinity_mask_valid is_affinity_mask_valid

and in kernel/irq/proc.c do

   #ifndef is_affinity_mask_valid
   #define is_affinity_mask_valid() 1
   #endif

This has minimal runtime cost, but is a bit tricksy.

It has the advantage that it doesn't introduce some new identifier which needs to be maintained.

Review line 15

Quoted 12 lines

extern int irq_set_affinity(unsigned int irq, cpumask_t cpumask); extern int irq_can_set_affinity(unsigned int irq);

Index: linux-2.6/include/asm-ia64/irq.h
=================================> > --- linux-2.6.orig/include/asm-ia64/irq.h 2007-05-01 15:47:01.660643076 -0500
+++ linux-2.6/include/asm-ia64/irq.h 2007-05-01 15:48:42.729254805 -0500

(quoted diff) +2 -0
@@ -11,6 +11,8 @@
  * 02/29/00     D.Mosberger	moved most things into hw_irq.h
  */
 
+#include <linux/cpumask.h>
+

This is the sort of inclusion which tends to make things blow up. I trust this was carefully compile-tested.

With a) above, the is_affinity_mask_valid() declaration would be in a different header. With b), no additional include will be needed.

Quoted 10 lines
(quoted diff) +3 -0
 #define NR_IRQS		256
 #define NR_IRQ_VECTORS	NR_IRQS
 
@@ -30,4 +32,7 @@ extern void disable_irq_nosync (unsigned
 extern void enable_irq (unsigned int);
 extern void set_irq_affinity_info (unsigned int irq, int dest, int redir);
 
+#define ARCH_HAS_IRQ_AFFINITY_VALIDATION
+extern int is_affinity_mask_valid(cpumask_t cpumask);
+