[PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[Mike Travis]

[-- Attachment #1: cpumask:update-irq_desc-to-use-cpumask_var_t.patch --]
[-- Type: text/plain, Size: 14894 bytes --]

Impact: reduce memory usage, use new cpumask API.

Replace the affinity and pending_masks with cpumask_var_t's.  This adds
to the significant size reduction done with the SPARSE_IRQS changes.

The added functions (init_alloc_desc_masks & init_copy_desc_masks) are
in the include file so they can be inlined (and optimized out for the
!CONFIG_CPUMASKS_OFFSTACK case.)  [Naming chosen to be consistent with
the other init*irq functions, as well as the backwards arg declaration
of "from, to" instead of the more common "to, from" standard.]

Includes a slight change to the declaration of struct irq_desc to embed
the pending_mask within ifdef(CONFIG_SMP) to be consistent with other
references, and some small changes to Xen.

Tested: sparse/non-sparse/cpumask_offstack/non-cpumask_offstack/nonuma/nosmp on x86_64

Signed-off-by: Mike Travis <travis@sgi.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: virtualization@lists.osdl.org
Cc: xen-devel@lists.xensource.com
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
---
 arch/x86/kernel/io_apic.c |   20 +++++------
 arch/x86/kernel/irq_32.c  |    2 -
 arch/x86/kernel/irq_64.c  |    2 -
 drivers/xen/events.c      |    4 +-
 include/linux/irq.h       |   81 ++++++++++++++++++++++++++++++++++++++++++++--
 kernel/irq/chip.c         |    5 ++
 kernel/irq/handle.c       |   26 +++++++-------
 kernel/irq/manage.c       |   12 +++---
 kernel/irq/migration.c    |   12 +++---
 kernel/irq/numa_migrate.c |   12 ++++++
 kernel/irq/proc.c         |    4 +-
 11 files changed, 135 insertions(+), 45 deletions(-)

--- linux-2.6-for-ingo.orig/arch/x86/kernel/io_apic.c
+++ linux-2.6-for-ingo/arch/x86/kernel/io_apic.c
@@ -358,7 +358,7 @@ set_extra_move_desc(struct irq_desc *des
 
 	if (!cfg->move_in_progress) {
 		/* it means that domain is not changed */
-		if (!cpumask_intersects(&desc->affinity, mask))
+		if (!cpumask_intersects(desc->affinity, mask))
 			cfg->move_desc_pending = 1;
 	}
 }
@@ -581,9 +581,9 @@ set_desc_affinity(struct irq_desc *desc,
 	if (assign_irq_vector(irq, cfg, mask))
 		return BAD_APICID;
 
-	cpumask_and(&desc->affinity, cfg->domain, mask);
+	cpumask_and(desc->affinity, cfg->domain, mask);
 	set_extra_move_desc(desc, mask);
-	return cpu_mask_to_apicid_and(&desc->affinity, cpu_online_mask);
+	return cpu_mask_to_apicid_and(desc->affinity, cpu_online_mask);
 }
 
 static void
@@ -2385,7 +2385,7 @@ migrate_ioapic_irq_desc(struct irq_desc 
 	if (cfg->move_in_progress)
 		send_cleanup_vector(cfg);
 
-	cpumask_copy(&desc->affinity, mask);
+	cpumask_copy(desc->affinity, mask);
 }
 
 static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
@@ -2407,11 +2407,11 @@ static int migrate_irq_remapped_level_de
 	}
 
 	/* everthing is clear. we have right of way */
-	migrate_ioapic_irq_desc(desc, &desc->pending_mask);
+	migrate_ioapic_irq_desc(desc, desc->pending_mask);
 
 	ret = 0;
 	desc->status &= ~IRQ_MOVE_PENDING;
-	cpumask_clear(&desc->pending_mask);
+	cpumask_clear(desc->pending_mask);
 
 unmask:
 	unmask_IO_APIC_irq_desc(desc);
@@ -2436,7 +2436,7 @@ static void ir_irq_migration(struct work
 				continue;
 			}
 
-			desc->chip->set_affinity(irq, &desc->pending_mask);
+			desc->chip->set_affinity(irq, desc->pending_mask);
 			spin_unlock_irqrestore(&desc->lock, flags);
 		}
 	}
@@ -2450,7 +2450,7 @@ static void set_ir_ioapic_affinity_irq_d
 {
 	if (desc->status & IRQ_LEVEL) {
 		desc->status |= IRQ_MOVE_PENDING;
-		cpumask_copy(&desc->pending_mask, mask);
+		cpumask_copy(desc->pending_mask, mask);
 		migrate_irq_remapped_level_desc(desc);
 		return;
 	}
@@ -2518,7 +2518,7 @@ static void irq_complete_move(struct irq
 
 		/* domain has not changed, but affinity did */
 		me = smp_processor_id();
-		if (cpu_isset(me, desc->affinity)) {
+		if (cpumask_test_cpu(me, desc->affinity)) {
 			*descp = desc = move_irq_desc(desc, me);
 			/* get the new one */
 			cfg = desc->chip_data;
@@ -4041,7 +4041,7 @@ void __init setup_ioapic_dest(void)
 			 */
 			if (desc->status &
 			    (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
-				mask = &desc->affinity;
+				mask = desc->affinity;
 			else
 				mask = TARGET_CPUS;
 
--- linux-2.6-for-ingo.orig/arch/x86/kernel/irq_32.c
+++ linux-2.6-for-ingo/arch/x86/kernel/irq_32.c
@@ -248,7 +248,7 @@ void fixup_irqs(void)
 		if (irq == 2)
 			continue;
 
-		affinity = &desc->affinity;
+		affinity = desc->affinity;
 		if (cpumask_any_and(affinity, cpu_online_mask) >= nr_cpu_ids) {
 			printk("Breaking affinity for irq %i\n", irq);
 			affinity = cpu_all_mask;
--- linux-2.6-for-ingo.orig/arch/x86/kernel/irq_64.c
+++ linux-2.6-for-ingo/arch/x86/kernel/irq_64.c
@@ -100,7 +100,7 @@ void fixup_irqs(void)
 		/* interrupt's are disabled at this point */
 		spin_lock(&desc->lock);
 
-		affinity = &desc->affinity;
+		affinity = desc->affinity;
 		if (!irq_has_action(irq) ||
 		    cpumask_equal(affinity, cpu_online_mask)) {
 			spin_unlock(&desc->lock);
--- linux-2.6-for-ingo.orig/drivers/xen/events.c
+++ linux-2.6-for-ingo/drivers/xen/events.c
@@ -125,7 +125,7 @@ static void bind_evtchn_to_cpu(unsigned 
 
 	BUG_ON(irq == -1);
 #ifdef CONFIG_SMP
-	irq_to_desc(irq)->affinity = cpumask_of_cpu(cpu);
+	cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
 #endif
 
 	__clear_bit(chn, cpu_evtchn_mask[cpu_evtchn[chn]]);
@@ -142,7 +142,7 @@ static void init_evtchn_cpu_bindings(voi
 
 	/* By default all event channels notify CPU#0. */
 	for_each_irq_desc(i, desc) {
-		desc->affinity = cpumask_of_cpu(0);
+		cpumask_copy(desc->affinity, cpumask_of(0));
 	}
 #endif
 
--- linux-2.6-for-ingo.orig/include/linux/irq.h
+++ linux-2.6-for-ingo/include/linux/irq.h
@@ -182,11 +182,11 @@ struct irq_desc {
 	unsigned int		irqs_unhandled;
 	spinlock_t		lock;
 #ifdef CONFIG_SMP
-	cpumask_t		affinity;
+	cpumask_var_t		affinity;
 	unsigned int		cpu;
-#endif
 #ifdef CONFIG_GENERIC_PENDING_IRQ
-	cpumask_t		pending_mask;
+	cpumask_var_t		pending_mask;
+#endif
 #endif
 #ifdef CONFIG_PROC_FS
 	struct proc_dir_entry	*dir;
@@ -422,4 +422,79 @@ extern int set_irq_msi(unsigned int irq,
 
 #endif /* !CONFIG_S390 */
 
+#ifdef CONFIG_SMP
+/**
+ * init_alloc_desc_masks - allocate cpumasks for irq_desc
+ * @desc:	pointer to irq_desc struct
+ * @boot:	true if need bootmem
+ *
+ * Allocates affinity and pending_mask cpumask if required.
+ * Returns true if successful (or not required).
+ * Side effect: affinity has all bits set, pending_mask has all bits clear.
+ */
+static inline bool init_alloc_desc_masks(struct irq_desc *desc, int node,
+								bool boot)
+{
+	if (boot) {
+		alloc_bootmem_cpumask_var(&desc->affinity);
+		cpumask_setall(desc->affinity);
+
+#ifdef CONFIG_GENERIC_PENDING_IRQ
+		alloc_bootmem_cpumask_var(&desc->pending_mask);
+		cpumask_clear(desc->pending_mask);
+#endif
+		return true;
+	}
+
+	if (!alloc_cpumask_var_node(&desc->affinity, GFP_ATOMIC, node))
+		return false;
+	cpumask_setall(desc->affinity);
+
+#ifdef CONFIG_GENERIC_PENDING_IRQ
+	if (!alloc_cpumask_var_node(&desc->pending_mask, GFP_ATOMIC, node)) {
+		free_cpumask_var(desc->affinity);
+		return false;
+	}
+	cpumask_clear(desc->pending_mask);
+#endif
+	return true;
+}
+
+/**
+ * init_copy_desc_masks - copy cpumasks for irq_desc
+ * @old_desc:	pointer to old irq_desc struct
+ * @new_desc:	pointer to new irq_desc struct
+ *
+ * Insures affinity and pending_masks are copied to new irq_desc.
+ * If !CONFIG_CPUMASKS_OFFSTACK the cpumasks are embedded in the
+ * irq_desc struct so the copy is redundant.
+ */
+
+static inline void init_copy_desc_masks(struct irq_desc *old_desc,
+					struct irq_desc *new_desc)
+{
+#ifdef CONFIG_CPUMASKS_OFFSTACK
+	cpumask_copy(new_desc->affinity, old_desc->affinity);
+
+#ifdef CONFIG_GENERIC_PENDING_IRQ
+	cpumask_copy(new_desc->pending_mask, old_desc->pending_mask);
+#endif
+#endif
+}
+
+#else /* !CONFIG_SMP */
+
+static inline bool init_alloc_desc_masks(struct irq_desc *desc, int node,
+								bool boot)
+{
+	return true;
+}
+
+static inline void init_copy_desc_masks(struct irq_desc *old_desc,
+					struct irq_desc *new_desc)
+{
+}
+
+#endif	/* CONFIG_SMP */
+
 #endif /* _LINUX_IRQ_H */
--- linux-2.6-for-ingo.orig/kernel/irq/chip.c
+++ linux-2.6-for-ingo/kernel/irq/chip.c
@@ -46,7 +46,10 @@ void dynamic_irq_init(unsigned int irq)
 	desc->irq_count = 0;
 	desc->irqs_unhandled = 0;
 #ifdef CONFIG_SMP
-	cpumask_setall(&desc->affinity);
+	cpumask_setall(desc->affinity);
+#ifdef CONFIG_GENERIC_PENDING_IRQ
+	cpumask_clear(desc->pending_mask);
+#endif
 #endif
 	spin_unlock_irqrestore(&desc->lock, flags);
 }
--- linux-2.6-for-ingo.orig/kernel/irq/handle.c
+++ linux-2.6-for-ingo/kernel/irq/handle.c
@@ -64,9 +64,6 @@ static struct irq_desc irq_desc_init = {
 	.handle_irq = handle_bad_irq,
 	.depth      = 1,
 	.lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
-#ifdef CONFIG_SMP
-	.affinity   = CPU_MASK_ALL
-#endif
 };
 
 void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr)
@@ -88,6 +85,8 @@ void init_kstat_irqs(struct irq_desc *de
 
 static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu)
 {
+	int node = cpu_to_node(cpu);
+
 	memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
 
 	spin_lock_init(&desc->lock);
@@ -101,6 +100,10 @@ static void init_one_irq_desc(int irq, s
 		printk(KERN_ERR "can not alloc kstat_irqs\n");
 		BUG_ON(1);
 	}
+	if (!init_alloc_desc_masks(desc, node, false)) {
+		printk(KERN_ERR "can not alloc irq_desc cpumasks\n");
+		BUG_ON(1);
+	}
 	arch_init_chip_data(desc, cpu);
 }
 
@@ -119,9 +122,6 @@ static struct irq_desc irq_desc_legacy[N
 		.handle_irq = handle_bad_irq,
 		.depth	    = 1,
 		.lock	    = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
-#ifdef CONFIG_SMP
-		.affinity   = CPU_MASK_ALL
-#endif
 	}
 };
 
@@ -141,7 +141,7 @@ int __init early_irq_init(void)
 		desc[i].irq = i;
 		desc[i].kstat_irqs = kstat_irqs_legacy[i];
 		lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
-
+		init_alloc_desc_masks(&desc[i], 0, true);
 		irq_desc_ptrs[i] = desc + i;
 	}
 
@@ -188,6 +188,10 @@ struct irq_desc *irq_to_desc_alloc_cpu(u
 		printk(KERN_ERR "can not alloc irq_desc\n");
 		BUG_ON(1);
 	}
+	if (!init_alloc_desc_masks(desc, node, false)) {
+		printk(KERN_ERR "can not alloc irq_desc cpumasks\n");
+		BUG_ON(1);
+	}
 	init_one_irq_desc(irq, desc, cpu);
 
 	irq_desc_ptrs[irq] = desc;
@@ -207,9 +211,6 @@ struct irq_desc irq_desc[NR_IRQS] __cach
 		.handle_irq = handle_bad_irq,
 		.depth = 1,
 		.lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
-#ifdef CONFIG_SMP
-		.affinity = CPU_MASK_ALL
-#endif
 	}
 };
 
@@ -222,9 +223,10 @@ int __init early_irq_init(void)
 	desc = irq_desc;
 	count = ARRAY_SIZE(irq_desc);
 
-	for (i = 0; i < count; i++)
+	for (i = 0; i < count; i++) {
 		desc[i].irq = i;
-
+		init_alloc_desc_masks(&desc[i], 0, true);
+	}
 	return arch_early_irq_init();
 }
 
--- linux-2.6-for-ingo.orig/kernel/irq/manage.c
+++ linux-2.6-for-ingo/kernel/irq/manage.c
@@ -98,14 +98,14 @@ int irq_set_affinity(unsigned int irq, c
 
 #ifdef CONFIG_GENERIC_PENDING_IRQ
 	if (desc->status & IRQ_MOVE_PCNTXT || desc->status & IRQ_DISABLED) {
-		cpumask_copy(&desc->affinity, cpumask);
+		cpumask_copy(desc->affinity, cpumask);
 		desc->chip->set_affinity(irq, cpumask);
 	} else {
 		desc->status |= IRQ_MOVE_PENDING;
-		cpumask_copy(&desc->pending_mask, cpumask);
+		cpumask_copy(desc->pending_mask, cpumask);
 	}
 #else
-	cpumask_copy(&desc->affinity, cpumask);
+	cpumask_copy(desc->affinity, cpumask);
 	desc->chip->set_affinity(irq, cpumask);
 #endif
 	desc->status |= IRQ_AFFINITY_SET;
@@ -127,16 +127,16 @@ int do_irq_select_affinity(unsigned int 
 	 * one of the targets is online.
 	 */
 	if (desc->status & (IRQ_AFFINITY_SET | IRQ_NO_BALANCING)) {
-		if (cpumask_any_and(&desc->affinity, cpu_online_mask)
+		if (cpumask_any_and(desc->affinity, cpu_online_mask)
 		    < nr_cpu_ids)
 			goto set_affinity;
 		else
 			desc->status &= ~IRQ_AFFINITY_SET;
 	}
 
-	cpumask_and(&desc->affinity, cpu_online_mask, irq_default_affinity);
+	cpumask_and(desc->affinity, cpu_online_mask, irq_default_affinity);
 set_affinity:
-	desc->chip->set_affinity(irq, &desc->affinity);
+	desc->chip->set_affinity(irq, desc->affinity);
 
 	return 0;
 }
--- linux-2.6-for-ingo.orig/kernel/irq/migration.c
+++ linux-2.6-for-ingo/kernel/irq/migration.c
@@ -18,7 +18,7 @@ void move_masked_irq(int irq)
 
 	desc->status &= ~IRQ_MOVE_PENDING;
 
-	if (unlikely(cpumask_empty(&desc->pending_mask)))
+	if (unlikely(cpumask_empty(desc->pending_mask)))
 		return;
 
 	if (!desc->chip->set_affinity)
@@ -38,13 +38,13 @@ void move_masked_irq(int irq)
 	 * For correct operation this depends on the caller
 	 * masking the irqs.
 	 */
-	if (likely(cpumask_any_and(&desc->pending_mask, cpu_online_mask)
+	if (likely(cpumask_any_and(desc->pending_mask, cpu_online_mask)
 		   < nr_cpu_ids)) {
-		cpumask_and(&desc->affinity,
-			    &desc->pending_mask, cpu_online_mask);
-		desc->chip->set_affinity(irq, &desc->affinity);
+		cpumask_and(desc->affinity,
+			    desc->pending_mask, cpu_online_mask);
+		desc->chip->set_affinity(irq, desc->affinity);
 	}
-	cpumask_clear(&desc->pending_mask);
+	cpumask_clear(desc->pending_mask);
 }
 
 void move_native_irq(int irq)
--- linux-2.6-for-ingo.orig/kernel/irq/numa_migrate.c
+++ linux-2.6-for-ingo/kernel/irq/numa_migrate.c
@@ -46,6 +46,7 @@ static void init_copy_one_irq_desc(int i
 	desc->cpu = cpu;
 	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
 	init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
+	init_copy_desc_masks(old_desc, desc);
 	arch_init_copy_chip_data(old_desc, desc, cpu);
 }
 
@@ -76,11 +77,20 @@ static struct irq_desc *__real_move_irq_
 	node = cpu_to_node(cpu);
 	desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
 	if (!desc) {
-		printk(KERN_ERR "irq %d: can not get new irq_desc for migration.\n", irq);
+		printk(KERN_ERR "irq %d: can not get new irq_desc "
+				"for migration.\n", irq);
 		/* still use old one */
 		desc = old_desc;
 		goto out_unlock;
 	}
+	if (!init_alloc_desc_masks(desc, node, false)) {
+		printk(KERN_ERR "irq %d: can not get new irq_desc cpumask "
+				"for migration.\n", irq);
+		/* still use old one */
+		kfree(desc);
+		desc = old_desc;
+		goto out_unlock;
+	}
 	init_copy_one_irq_desc(irq, old_desc, desc, cpu);
 
 	irq_desc_ptrs[irq] = desc;
--- linux-2.6-for-ingo.orig/kernel/irq/proc.c
+++ linux-2.6-for-ingo/kernel/irq/proc.c
@@ -20,11 +20,11 @@ static struct proc_dir_entry *root_irq_d
 static int irq_affinity_proc_show(struct seq_file *m, void *v)
 {
 	struct irq_desc *desc = irq_to_desc((long)m->private);
-	const struct cpumask *mask = &desc->affinity;
+	const struct cpumask *mask = desc->affinity;
 
 #ifdef CONFIG_GENERIC_PENDING_IRQ
 	if (desc->status & IRQ_MOVE_PENDING)
-		mask = &desc->pending_mask;
+		mask = desc->pending_mask;
 #endif
 	seq_cpumask(m, mask);
 	seq_putc(m, '\n');

-- 

Re: [PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[Yinghai Lu]

On Wed, Jan 7, 2009 at 11:58 AM, Mike Travis <travis@sgi.com> wrote:
| --- linux-2.6-for-ingo.orig/kernel/irq/numa_migrate.c
| +++ linux-2.6-for-ingo/kernel/irq/numa_migrate.c
| @@ -46,6 +46,7 @@ static void init_copy_one_irq_desc(int i
| 	desc->cpu = cpu;
| 	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
| 	init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
|+	init_copy_desc_masks(old_desc, desc);
| 	arch_init_copy_chip_data(old_desc, desc, cpu);
| }
|
|@@ -76,11 +77,20 @@ static struct irq_desc *__real_move_irq_
| 	node = cpu_to_node(cpu);
| 	desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
| 	if (!desc) {
|-		printk(KERN_ERR "irq %d: can not get new irq_desc for migration.\n", irq);
|+		printk(KERN_ERR "irq %d: can not get new irq_desc "
|+				"for migration.\n", irq);
| 		/* still use old one */
| 		desc = old_desc;
| 		goto out_unlock;
| 	}
|+	if (!init_alloc_desc_masks(desc, node, false)) {
|+		printk(KERN_ERR "irq %d: can not get new irq_desc cpumask "
|+				"for migration.\n", irq);
|+		/* still use old one */
|+		kfree(desc);
|+		desc = old_desc;
|+		goto out_unlock;
|+	}
| 	init_copy_one_irq_desc(irq, old_desc, desc, cpu);

desc new mask_var (allocated) aka the pointer is overwritten here...
you may need to calling move init_alloc_desc_masks() into
init_copy_one_irq_desc()

| 	irq_desc_ptrs[irq] = desc;

YH

Re: [PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[Mike Travis]

Yinghai Lu wrote:
> On Wed, Jan 7, 2009 at 11:58 AM, Mike Travis <travis@sgi.com> wrote:
> | --- linux-2.6-for-ingo.orig/kernel/irq/numa_migrate.c
> | +++ linux-2.6-for-ingo/kernel/irq/numa_migrate.c
> | @@ -46,6 +46,7 @@ static void init_copy_one_irq_desc(int i
> | 	desc->cpu = cpu;
> | 	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
> | 	init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
> |+	init_copy_desc_masks(old_desc, desc);
> | 	arch_init_copy_chip_data(old_desc, desc, cpu);
> | }
> |
> |@@ -76,11 +77,20 @@ static struct irq_desc *__real_move_irq_
> | 	node = cpu_to_node(cpu);
> | 	desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
> | 	if (!desc) {
> |-		printk(KERN_ERR "irq %d: can not get new irq_desc for migration.\n", irq);
> |+		printk(KERN_ERR "irq %d: can not get new irq_desc "
> |+				"for migration.\n", irq);
> | 		/* still use old one */
> | 		desc = old_desc;
> | 		goto out_unlock;
> | 	}
> |+	if (!init_alloc_desc_masks(desc, node, false)) {
> |+		printk(KERN_ERR "irq %d: can not get new irq_desc cpumask "
> |+				"for migration.\n", irq);
> |+		/* still use old one */
> |+		kfree(desc);
> |+		desc = old_desc;
> |+		goto out_unlock;
> |+	}
> | 	init_copy_one_irq_desc(irq, old_desc, desc, cpu);
> 
> desc new mask_var (allocated) aka the pointer is overwritten here...
> you may need to calling move init_alloc_desc_masks() into
> init_copy_one_irq_desc()

Wouldn't this in init_copy_one_irq_desc() take care  of that?

@@ -46,6 +46,7 @@ static void init_copy_one_irq_desc(int i
        desc->cpu = cpu;
        lockdep_set_class(&desc->lock, &irq_desc_lock_class);
        init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
+       init_copy_desc_masks(old_desc, desc);
        arch_init_copy_chip_data(old_desc, desc, cpu);

where:

static inline void init_copy_desc_masks(struct irq_desc *old_desc,
                                        struct irq_desc *new_desc)
{
#ifdef CONFIG_CPUMASKS_OFFSTACK
        cpumask_copy(new_desc->affinity, old_desc->affinity);

#ifdef CONFIG_GENERIC_PENDING_IRQ
        cpumask_copy(new_desc->pending_mask, old_desc->pending_mask);
#endif
#endif
}

In other words if the masks are not a cpumask[1] but instead a
*cpumask pointer, then the old masks are copied to the new desc/

Or am I missing your point?

Thanks!
Mike

> 
> | 	irq_desc_ptrs[irq] = desc;
> 
> YH
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: [PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[Yinghai Lu]

Mike Travis wrote:
> Yinghai Lu wrote:
>> On Wed, Jan 7, 2009 at 11:58 AM, Mike Travis <travis@sgi.com> wrote:
>> | --- linux-2.6-for-ingo.orig/kernel/irq/numa_migrate.c
>> | +++ linux-2.6-for-ingo/kernel/irq/numa_migrate.c
>> | @@ -46,6 +46,7 @@ static void init_copy_one_irq_desc(int i
>> | 	desc->cpu = cpu;
>> | 	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
>> | 	init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
>> |+	init_copy_desc_masks(old_desc, desc);
>> | 	arch_init_copy_chip_data(old_desc, desc, cpu);
>> | }
>> |
>> |@@ -76,11 +77,20 @@ static struct irq_desc *__real_move_irq_
>> | 	node = cpu_to_node(cpu);
>> | 	desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
>> | 	if (!desc) {
>> |-		printk(KERN_ERR "irq %d: can not get new irq_desc for migration.\n", irq);
>> |+		printk(KERN_ERR "irq %d: can not get new irq_desc "
>> |+				"for migration.\n", irq);
>> | 		/* still use old one */
>> | 		desc = old_desc;
>> | 		goto out_unlock;
>> | 	}
>> |+	if (!init_alloc_desc_masks(desc, node, false)) {
>> |+		printk(KERN_ERR "irq %d: can not get new irq_desc cpumask "
>> |+				"for migration.\n", irq);
>> |+		/* still use old one */
>> |+		kfree(desc);
>> |+		desc = old_desc;
>> |+		goto out_unlock;
>> |+	}
>> | 	init_copy_one_irq_desc(irq, old_desc, desc, cpu);
>>
>> desc new mask_var (allocated) aka the pointer is overwritten here...
>> you may need to calling move init_alloc_desc_masks() into
>> init_copy_one_irq_desc()
> 
> Wouldn't this in init_copy_one_irq_desc() take care  of that?
> 
> @@ -46,6 +46,7 @@ static void init_copy_one_irq_desc(int i
>         desc->cpu = cpu;
>         lockdep_set_class(&desc->lock, &irq_desc_lock_class);
>         init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids);
> +       init_copy_desc_masks(old_desc, desc);
>         arch_init_copy_chip_data(old_desc, desc, cpu);
> 
> where:
> 
> static inline void init_copy_desc_masks(struct irq_desc *old_desc,
>                                         struct irq_desc *new_desc)
> {
> #ifdef CONFIG_CPUMASKS_OFFSTACK
>         cpumask_copy(new_desc->affinity, old_desc->affinity);
> 
> #ifdef CONFIG_GENERIC_PENDING_IRQ
>         cpumask_copy(new_desc->pending_mask, old_desc->pending_mask);
> #endif
> #endif
> }
> 
> In other words if the masks are not a cpumask[1] but instead a
> *cpumask pointer, then the old masks are copied to the new desc/
> 
> Or am I missing your point?

static void init_copy_one_irq_desc(int irq, struct irq_desc *old_desc,
                 struct irq_desc *desc, int cpu)
{
        memcpy(desc, old_desc, sizeof(struct irq_desc));


will overwrite new_desc->affinity and pending_mask

YH

Re: [PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[Mike Travis]

Yinghai Lu wrote:
...
>> Or am I missing your point?
> 
> static void init_copy_one_irq_desc(int irq, struct irq_desc *old_desc,
>                  struct irq_desc *desc, int cpu)
> {
>         memcpy(desc, old_desc, sizeof(struct irq_desc));
> 
> 
> will overwrite new_desc->affinity and pending_mask
> 
> YH

Yup, apparently I was missing your point (banging on head once more).

Thanks for spotting this!

Ingo - what's the state of tip/cpus4096?  Shall I push this as an
append patch, or redo the original so bisectability works.

Also, Yinghai - would you know of a straight forward way to test
the irq migration?   I have taken cpus offline and back online
but wasn't sure if any irq's were being moved off of cpu 0 which
cannot be taken offline.

Thanks,
Mike

Re: [PATCH 1/5] cpumask: update irq_desc to use cpumask_var_t

[Yinghai Lu]

Mike Travis wrote:
> Yinghai Lu wrote:
> ...
>>> Or am I missing your point?
>> static void init_copy_one_irq_desc(int irq, struct irq_desc *old_desc,
>>                  struct irq_desc *desc, int cpu)
>> {
>>         memcpy(desc, old_desc, sizeof(struct irq_desc));
>>
>>
>> will overwrite new_desc->affinity and pending_mask
>>
>> YH
> 
> Yup, apparently I was missing your point (banging on head once more).
> 
> Thanks for spotting this!
> 
> Ingo - what's the state of tip/cpus4096?  Shall I push this as an
> append patch, or redo the original so bisectability works.
> 
> Also, Yinghai - would you know of a straight forward way to test
> the irq migration?   I have taken cpus offline and back online
> but wasn't sure if any irq's were being moved off of cpu 0 which
> cannot be taken offline.

use echo "mask" > /proc/irq/xxx/smp_affinity to change affinity

and check /proc/interrupts and in /sys/log/messages i will said allocate new irq desc on some other node/cpu.

YH

[PATCH 3/5] cpumask: convert misc driver functions

[Mike Travis]

[-- Attachment #1: cpumask:convert-drivers.patch --]
[-- Type: text/plain, Size: 7183 bytes --]

From: Rusty Russell <rusty@rustcorp.com.au>

Impact: use new cpumask API.

Convert misc driver functions to use struct cpumask.

To Do:
  - Convert iucv_buffer_cpumask to cpumask_var_t.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Dean Nelson <dcn@sgi.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: oprofile-list@lists.sf.net
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: virtualization@lists.osdl.org
Cc: xen-devel@lists.xensource.com
Cc: Ursula Braun <ursula.braun@de.ibm.com>
Cc: linux390@de.ibm.com
Cc: linux-s390@vger.kernel.org
---
 drivers/base/cpu.c             |    2 +-
 drivers/misc/sgi-xp/xpc_main.c |    2 +-
 drivers/oprofile/buffer_sync.c |   22 ++++++++++++++++++----
 drivers/oprofile/buffer_sync.h |    4 ++++
 drivers/oprofile/oprof.c       |    9 ++++++++-
 drivers/xen/manage.c           |    2 +-
 net/iucv/iucv.c                |   28 ++++++++++++++++++----------
 7 files changed, 51 insertions(+), 18 deletions(-)

--- linux-2.6-for-ingo.orig/drivers/base/cpu.c
+++ linux-2.6-for-ingo/drivers/base/cpu.c
@@ -107,7 +107,7 @@ static SYSDEV_ATTR(crash_notes, 0400, sh
 /*
  * Print cpu online, possible, present, and system maps
  */
-static ssize_t print_cpus_map(char *buf, cpumask_t *map)
+static ssize_t print_cpus_map(char *buf, const struct cpumask *map)
 {
 	int n = cpulist_scnprintf(buf, PAGE_SIZE-2, map);
 
--- linux-2.6-for-ingo.orig/drivers/misc/sgi-xp/xpc_main.c
+++ linux-2.6-for-ingo/drivers/misc/sgi-xp/xpc_main.c
@@ -318,7 +318,7 @@ xpc_hb_checker(void *ignore)
 
 	/* this thread was marked active by xpc_hb_init() */
 
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(XPC_HB_CHECK_CPU));
+	set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU));
 
 	/* set our heartbeating to other partitions into motion */
 	xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
--- linux-2.6-for-ingo.orig/drivers/oprofile/buffer_sync.c
+++ linux-2.6-for-ingo/drivers/oprofile/buffer_sync.c
@@ -37,7 +37,7 @@
 
 static LIST_HEAD(dying_tasks);
 static LIST_HEAD(dead_tasks);
-static cpumask_t marked_cpus = CPU_MASK_NONE;
+static cpumask_var_t marked_cpus;
 static DEFINE_SPINLOCK(task_mortuary);
 static void process_task_mortuary(void);
 
@@ -489,10 +489,10 @@ static void mark_done(int cpu)
 {
 	int i;
 
-	cpu_set(cpu, marked_cpus);
+	cpumask_set_cpu(cpu, marked_cpus);
 
 	for_each_online_cpu(i) {
-		if (!cpu_isset(i, marked_cpus))
+		if (!cpumask_test_cpu(i, marked_cpus))
 			return;
 	}
 
@@ -501,7 +501,7 @@ static void mark_done(int cpu)
 	 */
 	process_task_mortuary();
 
-	cpus_clear(marked_cpus);
+	cpumask_clear(marked_cpus);
 }
 
 
@@ -595,6 +595,20 @@ void sync_buffer(int cpu)
 	mutex_unlock(&buffer_mutex);
 }
 
+int __init buffer_sync_init(void)
+{
+	if (!alloc_cpumask_var(&marked_cpus, GFP_KERNEL))
+		return -ENOMEM;
+
+	cpumask_clear(marked_cpus);
+		return 0;
+}
+
+void __exit buffer_sync_cleanup(void)
+{
+	free_cpumask_var(marked_cpus);
+}
+
 /* The function can be used to add a buffer worth of data directly to
  * the kernel buffer. The buffer is assumed to be a circular buffer.
  * Take the entries from index start and end at index end, wrapping
--- linux-2.6-for-ingo.orig/drivers/oprofile/buffer_sync.h
+++ linux-2.6-for-ingo/drivers/oprofile/buffer_sync.h
@@ -19,4 +19,8 @@ void sync_stop(void);
 /* sync the given CPU's buffer */
 void sync_buffer(int cpu);
 
+/* initialize/destroy the buffer system. */
+int buffer_sync_init(void);
+void buffer_sync_cleanup(void);
+
 #endif /* OPROFILE_BUFFER_SYNC_H */
--- linux-2.6-for-ingo.orig/drivers/oprofile/oprof.c
+++ linux-2.6-for-ingo/drivers/oprofile/oprof.c
@@ -183,6 +183,10 @@ static int __init oprofile_init(void)
 {
 	int err;
 
+	err = buffer_sync_init();
+	if (err)
+		return err;
+
 	err = oprofile_arch_init(&oprofile_ops);
 
 	if (err < 0 || timer) {
@@ -191,8 +195,10 @@ static int __init oprofile_init(void)
 	}
 
 	err = oprofilefs_register();
-	if (err)
+	if (err) {
 		oprofile_arch_exit();
+		buffer_sync_cleanup();
+	}
 
 	return err;
 }
@@ -202,6 +208,7 @@ static void __exit oprofile_exit(void)
 {
 	oprofilefs_unregister();
 	oprofile_arch_exit();
+	buffer_sync_cleanup();
 }
 
 
--- linux-2.6-for-ingo.orig/drivers/xen/manage.c
+++ linux-2.6-for-ingo/drivers/xen/manage.c
@@ -100,7 +100,7 @@ static void do_suspend(void)
 	/* XXX use normal device tree? */
 	xenbus_suspend();
 
-	err = stop_machine(xen_suspend, &cancelled, &cpumask_of_cpu(0));
+	err = stop_machine(xen_suspend, &cancelled, cpumask_of(0));
 	if (err) {
 		printk(KERN_ERR "failed to start xen_suspend: %d\n", err);
 		goto out;
--- linux-2.6-for-ingo.orig/net/iucv/iucv.c
+++ linux-2.6-for-ingo/net/iucv/iucv.c
@@ -492,15 +492,14 @@ static void iucv_setmask_mp(void)
  *
  * Allow iucv interrupts on a single cpu.
  */
-static void iucv_setmask_up(void)
+static void iucv_setmask_up(struct cpumask *cpumask)
 {
-	cpumask_t cpumask;
 	int cpu;
 
 	/* Disable all cpu but the first in cpu_irq_cpumask. */
-	cpumask = iucv_irq_cpumask;
-	cpu_clear(first_cpu(iucv_irq_cpumask), cpumask);
-	for_each_cpu_mask_nr(cpu, cpumask)
+	cpumask_copy(cpumask, iucv_irq_cpumask);
+	cpumask_clear_cpu(cpumask_first(iucv_irq_cpumask), cpumask);
+	for_each_cpu(cpu, cpumask)
 		smp_call_function_single(cpu, iucv_block_cpu, NULL, 1);
 }
 
@@ -558,7 +557,7 @@ static void iucv_disable(void)
 static int __cpuinit iucv_cpu_notify(struct notifier_block *self,
 				     unsigned long action, void *hcpu)
 {
-	cpumask_t cpumask;
+	cpumask_var_t cpumask;
 	long cpu = (long) hcpu;
 
 	switch (action) {
@@ -593,15 +592,20 @@ static int __cpuinit iucv_cpu_notify(str
 		break;
 	case CPU_DOWN_PREPARE:
 	case CPU_DOWN_PREPARE_FROZEN:
-		cpumask = iucv_buffer_cpumask;
-		cpu_clear(cpu, cpumask);
-		if (cpus_empty(cpumask))
+		if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
+			return NOTIFY_BAD;
+		cpumask_copy(cpumask, &iucv_buffer_cpumask);
+		cpumask_clear_cpu(cpu, cpumask);
+		if (cpumask_empty(cpumask)) {
 			/* Can't offline last IUCV enabled cpu. */
+			free_cpumask_var(cpumask);
 			return NOTIFY_BAD;
+		}
 		smp_call_function_single(cpu, iucv_retrieve_cpu, NULL, 1);
 		if (cpus_empty(iucv_irq_cpumask))
 			smp_call_function_single(first_cpu(iucv_buffer_cpumask),
 						 iucv_allow_cpu, NULL, 1);
+		free_cpumask_var(cpumask);
 		break;
 	}
 	return NOTIFY_OK;
@@ -686,9 +690,12 @@ static void iucv_cleanup_queue(void)
 int iucv_register(struct iucv_handler *handler, int smp)
 {
 	int rc;
+	cpumask_var_t scratch;
 
 	if (!iucv_available)
 		return -ENOSYS;
+	if (!alloc_cpumask_var(&scratch, GFP_KERNEL))
+		return -ENOMEM;
 	mutex_lock(&iucv_register_mutex);
 	if (!smp)
 		iucv_nonsmp_handler++;
@@ -697,7 +704,7 @@ int iucv_register(struct iucv_handler *h
 		if (rc)
 			goto out_mutex;
 	} else if (!smp && iucv_nonsmp_handler == 1)
-		iucv_setmask_up();
+		iucv_setmask_up(scratch);
 	INIT_LIST_HEAD(&handler->paths);
 
 	spin_lock_bh(&iucv_table_lock);
@@ -706,6 +713,7 @@ int iucv_register(struct iucv_handler *h
 	rc = 0;
 out_mutex:
 	mutex_unlock(&iucv_register_mutex);
+	free_cpumask_var(scratch);
 	return rc;
 }
 EXPORT_SYMBOL(iucv_register);

-- 

Re: [PATCH 3/5] cpumask: convert misc driver functions

[Rusty Russell]

On Thursday 08 January 2009 06:28:35 Mike Travis wrote:
> From: Rusty Russell <rusty@rustcorp.com.au>
> 
> Impact: use new cpumask API.
> 
> Convert misc driver functions to use struct cpumask.
...
> @@ -593,15 +592,20 @@ static int __cpuinit iucv_cpu_notify(str
>  		break;
>  	case CPU_DOWN_PREPARE:
>  	case CPU_DOWN_PREPARE_FROZEN:
> -		cpumask = iucv_buffer_cpumask;
> -		cpu_clear(cpu, cpumask);
> -		if (cpus_empty(cpumask))
> +		if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
> +			return NOTIFY_BAD;
> +		cpumask_copy(cpumask, &iucv_buffer_cpumask);
> +		cpumask_clear_cpu(cpu, cpumask);
> +		if (cpumask_empty(cpumask)) {
>  			/* Can't offline last IUCV enabled cpu. */
> +			free_cpumask_var(cpumask);
>  			return NOTIFY_BAD;
> +		}
>  		smp_call_function_single(cpu, iucv_retrieve_cpu, NULL, 1);
>  		if (cpus_empty(iucv_irq_cpumask))
>  			smp_call_function_single(first_cpu(iucv_buffer_cpumask),
>  						 iucv_allow_cpu, NULL, 1);
> +		free_cpumask_var(cpumask);
>  		break;

Minor nit: this can be done better, actually:

@@ -596,9 +594,7 @@ static int __cpuinit iucv_cpu_notify(str
 	case CPU_DOWN_PREPARE_FROZEN:
 		if (!iucv_path_table)
 			break;
-		cpumask = iucv_buffer_cpumask;
-		cpu_clear(cpu, cpumask);
-		if (cpus_empty(cpumask))
+		if (cpumask_any_but(&iucv_buffer_cpumask, cpu) >= nr_cpu_ids)
 			/* Can't offline last IUCV enabled cpu. */
 			return NOTIFY_BAD;
 		smp_call_function_single(cpu, iucv_retrieve_cpu, NULL, 1);

Cheers,
Rusty.

Re: [PATCH 3/5] cpumask: convert misc driver functions

[Tony Luck]

+void __exit buffer_sync_cleanup(void)
+{
+	free_cpumask_var(marked_cpus);
+}

This breaks ia64 with CONFIG_OPROFILE=y

`buffer_sync_cleanup' referenced in section `.init.text' of
arch/ia64/oprofile/built-in.o: defined in discarded section
`.exit.text' of arch/ia64/oprofile/built-in.o
make: *** [.tmp_vmlinux1] Error 1

-Tony

Re: [PATCH 3/5] cpumask: convert misc driver functions

[Mike Travis]

Tony Luck wrote:
> +void __exit buffer_sync_cleanup(void)
> +{
> +	free_cpumask_var(marked_cpus);
> +}
> 
> This breaks ia64 with CONFIG_OPROFILE=y
> 
> `buffer_sync_cleanup' referenced in section `.init.text' of
> arch/ia64/oprofile/built-in.o: defined in discarded section
> `.exit.text' of arch/ia64/oprofile/built-in.o
> make: *** [.tmp_vmlinux1] Error 1
> 
> -Tony

Hi Tony,

Hmm, that is a tricky error.  I'll see what can be done to
eliminate it.

Thanks!
Mike

Re: [PATCH 3/5] cpumask: convert misc driver functions

[Robert Richter]

On 07.01.09 11:58:35, Mike Travis wrote:
> From: Rusty Russell <rusty@rustcorp.com.au>
> 
> Impact: use new cpumask API.
> 
> Convert misc driver functions to use struct cpumask.
> 
> To Do:
>   - Convert iucv_buffer_cpumask to cpumask_var_t.
> 
> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
> Signed-off-by: Mike Travis <travis@sgi.com>
> Acked-by: Dean Nelson <dcn@sgi.com>
> Cc: Robert Richter <robert.richter@amd.com>
> Cc: oprofile-list@lists.sf.net
> Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
> Cc: Chris Wright <chrisw@sous-sol.org>
> Cc: virtualization@lists.osdl.org
> Cc: xen-devel@lists.xensource.com
> Cc: Ursula Braun <ursula.braun@de.ibm.com>
> Cc: linux390@de.ibm.com
> Cc: linux-s390@vger.kernel.org
> ---
>  drivers/base/cpu.c             |    2 +-
>  drivers/misc/sgi-xp/xpc_main.c |    2 +-
>  drivers/oprofile/buffer_sync.c |   22 ++++++++++++++++++----
>  drivers/oprofile/buffer_sync.h |    4 ++++
>  drivers/oprofile/oprof.c       |    9 ++++++++-

Acked-by: Robert Richter <robert.richter@amd.com>

I will send out a delta patch that changes the initialization.

-Robert

>  drivers/xen/manage.c           |    2 +-
>  net/iucv/iucv.c                |   28 ++++++++++++++++++----------
>  7 files changed, 51 insertions(+), 18 deletions(-)

-- 
Advanced Micro Devices, Inc.
Operating System Research Center
email: robert.richter@amd.com