Skip to contentSkip to conversation

Fix irq_affinity write from /proc for IPF

DISCUSSION Ashok Raj 2 participants 4 messages Mar 14, 2005 – Mar 15, 2005

lore
Share & tools
lore

Ashok Raj

author
#0 lore raw

Hi Andrew/Tony

This patch is required for IPF to perform deferred write to rte's when affinity is programmed via /proc. These entries can only be programmed when interrupt is pending.

We will eventually need the same method for x86 and x86_64 as well.

This patch is only for IPF though. (the others are comming, more testing and changes needed in my sandbox)

Could you please queue this up for next mm candidate? if it looks acceptable.

Since iam touching a common file for GENERIC_HARDIRQ, it would be best it its reviwed in the -mm releases for any potential conflicts.

[ sorry for the cross post to lia64 ]

Signature · 84 lines
Cheers,
Ashok Raj
- Open Source Technology Center



---
fix_ia64_smp_affinity - Make GENERIC_HARDIRQ work for IPF and CPU Hotplug

Signed-off-by: Ashok Raj <[email protected]>

Made GENERIC_HARDIRQ mechanism work for IPF and CPU hotplug. When write to 
/proc/irq is handled it is not appropriate to perform set_rte immediatly, 
since there is a race when the interrupt is asserted while the re-program 
is happening. Hence such programming is only safe when we do the re-program 
at the time of servicing an interrupt. This got broken when GENERIC_HARDIRQ 
got introduced for IPF.

- added CONFIG_PENDING_IRQ so default /proc/irq write handler can do the right 
  thing.

TBD: We currently dont handle redirectable hint either in the display, or when
we handle writes to /proc/irq/XX/smp_affinity. We need an arch specific way to 
account for the presence of "r" hint when we handle the proc write.
---

 release_dir-araj/arch/ia64/kernel/irq.c |   12 ++++++++++--
 release_dir-araj/kernel/irq/proc.c      |   10 ++++++++--
 2 files changed, 18 insertions(+), 4 deletions(-)

diff -puN arch/ia64/kernel/irq.c~fix_ia64_smp_affinity arch/ia64/kernel/irq.c
--- release_dir/arch/ia64/kernel/irq.c~fix_ia64_smp_affinity	2005-03-14 14:35:44.589293491 -0800
+++ release_dir-araj/arch/ia64/kernel/irq.c	2005-03-14 15:27:54.262106715 -0800
@@ -94,12 +94,20 @@ skip:
 /*
  * This is updated when the user sets irq affinity via /proc
  */
-cpumask_t __cacheline_aligned pending_irq_cpumask[NR_IRQS];
+static cpumask_t __cacheline_aligned pending_irq_cpumask[NR_IRQS];
 static unsigned long pending_irq_redir[BITS_TO_LONGS(NR_IRQS)];
 
-static cpumask_t irq_affinity [NR_IRQS] = { [0 ... NR_IRQS-1] = CPU_MASK_ALL };
 static char irq_redir [NR_IRQS]; // = { [0 ... NR_IRQS-1] = 1 };
 
+/*
+ * Arch specific routine for deferred write to iosapic rte to reprogram
+ * intr destination.
+ */
+void proc_set_irq_affinity(unsigned int irq, cpumask_t mask_val)
+{
+	pending_irq_cpumask[irq] = mask_val;
+}
+
 void set_irq_affinity_info (unsigned int irq, int hwid, int redir)
 {
 	cpumask_t mask = CPU_MASK_NONE;
diff -puN kernel/irq/proc.c~fix_ia64_smp_affinity kernel/irq/proc.c
--- release_dir/kernel/irq/proc.c~fix_ia64_smp_affinity	2005-03-14 14:41:05.475031747 -0800
+++ release_dir-araj/kernel/irq/proc.c	2005-03-14 15:27:59.436911339 -0800
@@ -19,6 +19,13 @@ static struct proc_dir_entry *root_irq_d
  */
 static struct proc_dir_entry *smp_affinity_entry[NR_IRQS];
 
+void __attribute__((weak))
+proc_set_irq_affinity(unsigned int irq, cpumask_t mask_val)
+{
+	irq_affinity[irq] = mask_val;
+	irq_desc[irq].handler->set_affinity(irq, mask_val);
+}
+
 static int irq_affinity_read_proc(char *page, char **start, off_t off,
 				  int count, int *eof, void *data)
 {
@@ -53,8 +60,7 @@ static int irq_affinity_write_proc(struc
 	if (cpus_empty(tmp))
 		return -EINVAL;
 
-	irq_affinity[irq] = new_value;
-	irq_desc[irq].handler->set_affinity(irq, new_value);
+	proc_set_irq_affinity(irq, new_value);
 
 	return full_count;
 }
_
#1 lore raw
Quoted 6 lines · Ashok Raj

Hi Andrew/Tony

This patch is required for IPF to perform deferred write to rte's when affinity is programmed via /proc. These entries can only be programmed when interrupt is pending.

"ia64" is preferred, please. Nobody knows what an IPF is.

Quoted 35 lines
 release_dir-araj/arch/ia64/kernel/irq.c |   12 ++++++++++--
 release_dir-araj/kernel/irq/proc.c      |   10 ++++++++--
 2 files changed, 18 insertions(+), 4 deletions(-)

diff -puN arch/ia64/kernel/irq.c~fix_ia64_smp_affinity arch/ia64/kernel/irq.c

arch/ia64/kernel/irq.c renamed from arch/ia64/kernel/irq.c~fix_ia64_smp_affinity +10 -2
@@ -94,12 +94,20 @@ skip:
 /*
  * This is updated when the user sets irq affinity via /proc
  */
-cpumask_t __cacheline_aligned pending_irq_cpumask[NR_IRQS];
+static cpumask_t __cacheline_aligned pending_irq_cpumask[NR_IRQS];
 static unsigned long pending_irq_redir[BITS_TO_LONGS(NR_IRQS)];
 
-static cpumask_t irq_affinity [NR_IRQS] = { [0 ... NR_IRQS-1] = CPU_MASK_ALL };
 static char irq_redir [NR_IRQS]; // = { [0 ... NR_IRQS-1] = 1 };
 
+/*
+ * Arch specific routine for deferred write to iosapic rte to reprogram
+ * intr destination.
+ */
+void proc_set_irq_affinity(unsigned int irq, cpumask_t mask_val)
+{
+	pending_irq_cpumask[irq] = mask_val;
+}
+
 void set_irq_affinity_info (unsigned int irq, int hwid, int redir)
 {
 	cpumask_t mask = CPU_MASK_NONE;

diff -puN kernel/irq/proc.c~fix_ia64_smp_affinity kernel/irq/proc.c

kernel/irq/proc.c static struct proc_dir_entry *root_irq_d · line 28

Quoted 10 lines
kernel/irq/proc.c renamed from kernel/irq/proc.c~fix_ia64_smp_affinity +7 -0
@@ -19,6 +19,13 @@ static struct proc_dir_entry *root_irq_d
  */
 static struct proc_dir_entry *smp_affinity_entry[NR_IRQS];
 
+void __attribute__((weak))
+proc_set_irq_affinity(unsigned int irq, cpumask_t mask_val)
+{
+	irq_affinity[irq] = mask_val;
+	irq_desc[irq].handler->set_affinity(irq, mask_val);
+}
+

Is it not possible for ia64's ->set_affinity() handler to do this deferring?

author
#2 lore raw
Quoted 4 lines · Andrew Morton
Quoted 1 line · Ashok Raj

"ia64" is preferred, please. Nobody knows what an IPF is.

Right!. Sorry about that.

Quoted 4 lines

Is it not possible for ia64's ->set_affinity() handler to do this deferring?

There are other places where we re-program, and its fine to call the current version of set_affinity directly, like when we are doing cpu offline and trying to force migrate irqs for ia64.

Changing the default set_affinity() for ia64 would result in many changes, this still keeps the same purpose of those access functions, and differentiates the proc write cases alone without changing the meaning of those handler functions. (and a smaller patch)

this would further complicate the force migrate irq's when we consider MSI interrupts as well. Since it would have its own set_affinity, and we need to hack into MSI's set affinity handler as well which would complicate things.

Signature · 3 lines
Cheers,
Ashok Raj
- Open Source Technology Center
#3 lore raw
Quoted 8 lines · Ashok Raj
Quoted 3 lines

Is it not possible for ia64's ->set_affinity() handler to do this deferring?

There are other places where we re-program, and its fine to call the current version of set_affinity directly, like when we are doing cpu offline and trying to force migrate irqs for ia64.

Quoted 8 lines

Changing the default set_affinity() for ia64 would result in many changes, this still keeps the same purpose of those access functions, and differentiates the proc write cases alone without changing the meaning of those handler functions. (and a smaller patch)

this would further complicate the force migrate irq's when we consider MSI interrupts as well. Since it would have its own set_affinity, and we need to hack into MSI's set affinity handler as well which would complicate things.

OK, just checking.

I'll include this change in the next batch, probably post-2.6.12-rc1, thanks.