From f6d87f4bd259cf33e092cd1a8fde05f291c47af1 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 7 Nov 2008 13:18:30 +0100 Subject: genirq: keep affinities set from userspace across free/request_irq() Impact: preserve user-modified affinities on interrupts Kumar Galak noticed that commit 18404756765c713a0be4eb1082920c04822ce588 (genirq: Expose default irq affinity mask (take 3)) overrides an already set affinity setting across a free / request_irq(). Happens e.g. with ifdown/ifup of a network device. Change the logic to mark the affinities as set and keep them intact. This also fixes the unlocked access to irq_desc in irq_select_affinity() when called from irq_affinity_proc_write() Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- include/linux/irq.h | 8 ++----- kernel/irq/internals.h | 2 ++ kernel/irq/manage.c | 58 +++++++++++++++++++++++++++++++++++++++++--------- kernel/irq/migration.c | 11 ---------- kernel/irq/proc.c | 2 +- 5 files changed, 53 insertions(+), 28 deletions(-) diff --git a/include/linux/irq.h b/include/linux/irq.h index d058c57be02..36b186eb318 100644 --- a/include/linux/irq.h +++ b/include/linux/irq.h @@ -63,7 +63,8 @@ typedef void (*irq_flow_handler_t)(unsigned int irq, #define IRQ_MOVE_PENDING 0x00200000 /* need to re-target IRQ destination */ #define IRQ_NO_BALANCING 0x00400000 /* IRQ is excluded from balancing */ #define IRQ_SPURIOUS_DISABLED 0x00800000 /* IRQ was disabled by the spurious trap */ -#define IRQ_MOVE_PCNTXT 0x01000000 /* IRQ migration from process context */ +#define IRQ_MOVE_PCNTXT 0x01000000 /* IRQ migration from process context */ +#define IRQ_AFFINITY_SET 0x02000000 /* IRQ affinity was set from userspace*/ #ifdef CONFIG_IRQ_PER_CPU # define CHECK_IRQ_PER_CPU(var) ((var) & IRQ_PER_CPU) @@ -210,7 +211,6 @@ extern int setup_irq(unsigned int irq, struct irqaction *new); #ifdef CONFIG_GENERIC_PENDING_IRQ -void set_pending_irq(unsigned int irq, cpumask_t mask); void move_native_irq(int irq); void move_masked_irq(int irq); @@ -228,10 +228,6 @@ static inline void move_masked_irq(int irq) { } -static inline void set_pending_irq(unsigned int irq, cpumask_t mask) -{ -} - #endif /* CONFIG_GENERIC_PENDING_IRQ */ #else /* CONFIG_SMP */ diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h index c9767e64198..64c1c7253da 100644 --- a/kernel/irq/internals.h +++ b/kernel/irq/internals.h @@ -25,6 +25,8 @@ static inline void unregister_handler_proc(unsigned int irq, struct irqaction *action) { } #endif +extern int irq_select_affinity_usr(unsigned int irq); + /* * Debugging printout: */ diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index c498a1b8c62..634a2a95510 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -82,24 +82,27 @@ int irq_can_set_affinity(unsigned int irq) int irq_set_affinity(unsigned int irq, cpumask_t cpumask) { struct irq_desc *desc = irq_to_desc(irq); + unsigned long flags; if (!desc->chip->set_affinity) return -EINVAL; + spin_lock_irqsave(&desc->lock, flags); + #ifdef CONFIG_GENERIC_PENDING_IRQ if (desc->status & IRQ_MOVE_PCNTXT || desc->status & IRQ_DISABLED) { - unsigned long flags; - - spin_lock_irqsave(&desc->lock, flags); desc->affinity = cpumask; desc->chip->set_affinity(irq, cpumask); - spin_unlock_irqrestore(&desc->lock, flags); - } else - set_pending_irq(irq, cpumask); + } else { + desc->status |= IRQ_MOVE_PENDING; + desc->pending_mask = cpumask; + } #else desc->affinity = cpumask; desc->chip->set_affinity(irq, cpumask); #endif + desc->status |= IRQ_AFFINITY_SET; + spin_unlock_irqrestore(&desc->lock, flags); return 0; } @@ -107,24 +110,59 @@ int irq_set_affinity(unsigned int irq, cpumask_t cpumask) /* * Generic version of the affinity autoselector. */ -int irq_select_affinity(unsigned int irq) +int do_irq_select_affinity(unsigned int irq, struct irq_desc *desc) { cpumask_t mask; - struct irq_desc *desc; if (!irq_can_set_affinity(irq)) return 0; cpus_and(mask, cpu_online_map, irq_default_affinity); - desc = irq_to_desc(irq); + /* + * Preserve an userspace affinity setup, but make sure that + * one of the targets is online. + */ + if (desc->status & IRQ_AFFINITY_SET) { + if (cpus_intersects(desc->affinity, cpu_online_map)) + mask = desc->affinity; + else + desc->status &= ~IRQ_AFFINITY_SET; + } + desc->affinity = mask; desc->chip->set_affinity(irq, mask); return 0; } +#else +static inline int do_irq_select_affinity(unsigned int irq, struct irq_desc *d) +{ + return irq_select_affinity(irq); +} #endif +/* + * Called when affinity is set via /proc/irq + */ +int irq_select_affinity_usr(unsigned int irq) +{ + struct irq_desc *desc = irq_to_desc(irq); + unsigned long flags; + int ret; + + spin_lock_irqsave(&desc->lock, flags); + ret = do_irq_select_affinity(irq, desc); + spin_unlock_irqrestore(&desc->lock, flags); + + return ret; +} + +#else +static inline int do_select_irq_affinity(int irq, struct irq_desc *desc) +{ + return 0; +} #endif /** @@ -446,7 +484,7 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new) desc->depth = 1; /* Set default affinity mask once everything is setup */ - irq_select_affinity(irq); + do_irq_select_affinity(irq, desc); } else if ((new->flags & IRQF_TRIGGER_MASK) && (new->flags & IRQF_TRIGGER_MASK) diff --git a/kernel/irq/migration.c b/kernel/irq/migration.c index 90b920d3f52..9db681d9581 100644 --- a/kernel/irq/migration.c +++ b/kernel/irq/migration.c @@ -1,17 +1,6 @@ #include -void set_pending_irq(unsigned int irq, cpumask_t mask) -{ - struct irq_desc *desc = irq_to_desc(irq); - unsigned long flags; - - spin_lock_irqsave(&desc->lock, flags); - desc->status |= IRQ_MOVE_PENDING; - desc->pending_mask = mask; - spin_unlock_irqrestore(&desc->lock, flags); -} - void move_masked_irq(int irq) { struct irq_desc *desc = irq_to_desc(irq); diff --git a/kernel/irq/proc.c b/kernel/irq/proc.c index 4d161c70ba5..d257e7d6a8a 100644 --- a/kernel/irq/proc.c +++ b/kernel/irq/proc.c @@ -62,7 +62,7 @@ static ssize_t irq_affinity_proc_write(struct file *file, if (!cpus_intersects(new_value, cpu_online_map)) /* Special case for empty set - allow the architecture code to set default SMP affinity. */ - return irq_select_affinity(irq) ? -EINVAL : count; + return irq_select_affinity_usr(irq) ? -EINVAL : count; irq_set_affinity(irq, new_value); -- cgit v1.2.3 From 612e3684c1b7752d2890510e4f90115fd1eb2afb Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 7 Nov 2008 13:58:46 +0100 Subject: genirq: fix the affinity setting in setup_irq The affinity setting in setup irq is called before the NO_BALANCING flag is checked and might therefore override affinity settings from the calling code with the default setting. Move the NO_BALANCING flag check before the call to the affinity setting. Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- kernel/irq/manage.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index 634a2a95510..948a22a2c01 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -123,7 +123,7 @@ int do_irq_select_affinity(unsigned int irq, struct irq_desc *desc) * Preserve an userspace affinity setup, but make sure that * one of the targets is online. */ - if (desc->status & IRQ_AFFINITY_SET) { + if (desc->status & (IRQ_AFFINITY_SET | IRQ_NO_BALANCING)) { if (cpus_intersects(desc->affinity, cpu_online_map)) mask = desc->affinity; else @@ -483,6 +483,10 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new) /* Undo nested disables: */ desc->depth = 1; + /* Exclude IRQ from balancing if requested */ + if (new->flags & IRQF_NOBALANCING) + desc->status |= IRQ_NO_BALANCING; + /* Set default affinity mask once everything is setup */ do_irq_select_affinity(irq, desc); @@ -497,10 +501,6 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new) *p = new; - /* Exclude IRQ from balancing */ - if (new->flags & IRQF_NOBALANCING) - desc->status |= IRQ_NO_BALANCING; - /* Reset broken irq detection when installing new handler */ desc->irq_count = 0; desc->irqs_unhandled = 0; -- cgit v1.2.3 From 6c2e94033df5ca11149e52dd179b8dde3172e9bf Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 7 Nov 2008 12:33:49 +0100 Subject: x86: apic honour irq affinity which was set in early boot setup_ioapic_dest() is called after the non boot cpus have been brought up. It sets the irq affinity of all already configured interrupts to all cpus and ignores affinity settings which were done by the early bootup code. If the IRQ_NO_BALANCING or IRQ_AFFINITY_SET flags are set then use the affinity mask from the irq descriptor and not TARGET_CPUS. Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- arch/x86/kernel/io_apic.c | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-) diff --git a/arch/x86/kernel/io_apic.c b/arch/x86/kernel/io_apic.c index 7a3f2028e2e..988ee89467d 100644 --- a/arch/x86/kernel/io_apic.c +++ b/arch/x86/kernel/io_apic.c @@ -3761,7 +3761,9 @@ int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity) void __init setup_ioapic_dest(void) { int pin, ioapic, irq, irq_entry; + struct irq_desc *desc; struct irq_cfg *cfg; + cpumask_t mask; if (skip_ioapic_setup == 1) return; @@ -3778,16 +3780,30 @@ void __init setup_ioapic_dest(void) * cpu is online. */ cfg = irq_cfg(irq); - if (!cfg->vector) + if (!cfg->vector) { setup_IO_APIC_irq(ioapic, pin, irq, irq_trigger(irq_entry), irq_polarity(irq_entry)); + continue; + + } + + /* + * Honour affinities which have been set in early boot + */ + desc = irq_to_desc(irq); + if (desc->status & + (IRQ_NO_BALANCING | IRQ_AFFINITY_SET)) + mask = desc->affinity; + else + mask = TARGET_CPUS; + #ifdef CONFIG_INTR_REMAP - else if (intr_remapping_enabled) - set_ir_ioapic_affinity_irq(irq, TARGET_CPUS); -#endif + if (intr_remapping_enabled) + set_ir_ioapic_affinity_irq(irq, mask); else - set_ioapic_affinity_irq(irq, TARGET_CPUS); +#endif + set_ioapic_affinity_irq(irq, mask); } } -- cgit v1.2.3 From f131e2436ddbac2527bb2d6297a823aae4b024f8 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sat, 8 Nov 2008 09:57:40 +0100 Subject: irq: fix typo Impact: build fix fix build failure on UP. Signed-off-by: Ingo Molnar --- kernel/irq/manage.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index 948a22a2c01..435861284e4 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -159,7 +159,7 @@ int irq_select_affinity_usr(unsigned int irq) } #else -static inline int do_select_irq_affinity(int irq, struct irq_desc *desc) +static inline int do_irq_select_affinity(int irq, struct irq_desc *desc) { return 0; } -- cgit v1.2.3 From 3ff68a6a106c362a6811d3e51bced58e6fc87de7 Mon Sep 17 00:00:00 2001 From: Mark Nelson Date: Thu, 13 Nov 2008 21:37:41 +1100 Subject: genirq: __irq_set_trigger: change pr_warning to pr_debug Commit 0c5d1eb77a8be917b638344a22afe1398236482b (genirq: record trigger type) caused powerpc platforms that had no set_type() function in their struct irq_chip to spew out warnings about "No set_type function for IRQ...". This warning isn't necessarily justified though because the generic powerpc platform code calls set_irq_type() (which in turn calls __irq_set_trigger) with information from the device tree to establish the interrupt mappings, regardless of whether the PIC can actually set a type. A platform's irq_chip might not have a set_type function for a variety of reasons, for example: the platform may have the type essentially hard-coded, or as in the case for Cell interrupts are just messages past around that have no real concept of type, or the platform could even have a virtual PIC as on the PS3. Signed-off-by: Mark Nelson Signed-off-by: Ingo Molnar --- kernel/irq/manage.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index 435861284e4..801addda3c4 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -365,7 +365,7 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq, * IRQF_TRIGGER_* but the PIC does not support multiple * flow-types? */ - pr_warning("No set_type function for IRQ %d (%s)\n", irq, + pr_debug("No set_type function for IRQ %d (%s)\n", irq, chip ? (chip->name ? : "unknown") : "unknown"); return 0; } -- cgit v1.2.3 From 2ed1cdcf9a83205d1343f29b630abff232eaa72c Mon Sep 17 00:00:00 2001 From: Randy Dunlap Date: Fri, 21 Nov 2008 16:59:57 -0800 Subject: irq.h: fix missing/extra kernel-doc Impact: fix kernel-doc build Fix missing & excess irq.h kernel-doc: Warning(include/linux/irq.h:182): No description found for parameter 'irq' Warning(include/linux/irq.h:182): Excess struct/union/enum/typedef member 'affinity_entry' description in 'irq_desc' Signed-off-by: Randy Dunlap Cc: Andrew Morton Signed-off-by: Ingo Molnar --- include/linux/irq.h | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/include/linux/irq.h b/include/linux/irq.h index 36b186eb318..3dddfa703eb 100644 --- a/include/linux/irq.h +++ b/include/linux/irq.h @@ -131,7 +131,7 @@ struct irq_chip { /** * struct irq_desc - interrupt descriptor - * + * @irq: interrupt number for this descriptor * @handle_irq: highlevel irq-events handler [if NULL, __do_IRQ()] * @chip: low level interrupt hardware access * @msi_desc: MSI descriptor @@ -150,7 +150,6 @@ struct irq_chip { * @cpu: cpu index useful for balancing * @pending_mask: pending rebalanced interrupts * @dir: /proc/irq/ procfs entry - * @affinity_entry: /proc/irq/smp_affinity procfs entry on SMP * @name: flow handler name for /proc/interrupts output */ struct irq_desc { -- cgit v1.2.3