[PATCH 0/3] unify the allocation of irq vectors v2

[PATCH 0/3] unify the allocation of irq vectors v2

[Dean Nelson]

This patchset makes it so that all irq vectors are allocated via one function.

On Fri, Sep 19, 2008 at 10:48:56AM +0200, Ingo Molnar wrote:
> 
> Dean, just to make sure the useful bits are not lost now that the 
> direction has been changed: could you please repost the patchset but 
> without the driver API bits? It's still all a nice and useful 
> generalization and cleanup of the x86 vector allocation code, and we can 
> check it in -tip how well it works in practice. 

Ingo, hopefully this patchset reflects what you're looking for. Feel free
to change it anyway you like.

Thanks,
Dean

[PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Dean Nelson]

Change per_cpu variable vector_irq[] from holding an 'int' irq number to
holding a 'struct irq_desc' pointer.

Signed-off-by: Dean Nelson <dcn@sgi.com>

---

Note that this pre-allocates the irq_desc structure before we know whether
a vector will be successfully found. And should it not, the irq_desc structure
is left created with the irq still unallocated. Should someone in the future
attempt to allocate a vector with that same irq and succeed, they will get
the previously allocated irq_desc structure.

Also, I won't claim the changes to arch/x86/xen/irq.c were correctly done, and
I know they weren't tested.

 arch/x86/kernel/io_apic.c     |   32 ++++++++++++++++++--------------
 arch/x86/kernel/irq_32.c      |   14 ++++++--------
 arch/x86/kernel/irq_64.c      |   10 ++++------
 arch/x86/kernel/irqinit_32.c  |   29 +++++++++--------------------
 arch/x86/kernel/irqinit_64.c  |   29 +++++++++--------------------
 arch/x86/kernel/vmiclock_32.c |    2 +-
 arch/x86/xen/irq.c            |    3 ++-
 include/asm-x86/hw_irq.h      |    2 +-
 8 files changed, 50 insertions(+), 71 deletions(-)

Index: linux/arch/x86/kernel/io_apic.c
===================================================================
--- linux.orig/arch/x86/kernel/io_apic.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/io_apic.c	2008-09-19 14:13:09.000000000 -0500
@@ -1226,6 +1226,7 @@ static int __assign_irq_vector(int irq, 
 	unsigned int old_vector;
 	int cpu;
 	struct irq_cfg *cfg;
+	struct irq_desc *desc;
 
 	cfg = irq_cfg(irq);
 
@@ -1243,6 +1244,8 @@ static int __assign_irq_vector(int irq, 
 			return 0;
 	}
 
+	desc = irq_to_desc_alloc(irq);
+
 	for_each_cpu_mask_nr(cpu, mask) {
 		cpumask_t domain, new_mask;
 		int new_cpu;
@@ -1270,7 +1273,7 @@ next:
 			goto next;
 #endif
 		for_each_cpu_mask_nr(new_cpu, new_mask)
-			if (per_cpu(vector_irq, new_cpu)[vector] != -1)
+			if (per_cpu(vector_irq, new_cpu)[vector] != NULL)
 				goto next;
 		/* Found one! */
 		current_vector = vector;
@@ -1280,7 +1283,7 @@ next:
 			cfg->old_domain = cfg->domain;
 		}
 		for_each_cpu_mask_nr(new_cpu, new_mask)
-			per_cpu(vector_irq, new_cpu)[vector] = irq;
+			per_cpu(vector_irq, new_cpu)[vector] = desc;
 		cfg->vector = vector;
 		cfg->domain = domain;
 		return 0;
@@ -1311,7 +1314,7 @@ static void __clear_irq_vector(int irq)
 	vector = cfg->vector;
 	cpus_and(mask, cfg->domain, cpu_online_map);
 	for_each_cpu_mask_nr(cpu, mask)
-		per_cpu(vector_irq, cpu)[vector] = -1;
+		per_cpu(vector_irq, cpu)[vector] = NULL;
 
 	cfg->vector = 0;
 	cpus_clear(cfg->domain);
@@ -1323,23 +1326,26 @@ void __setup_vector_irq(int cpu)
 	/* This function must be called with vector_lock held */
 	int irq, vector;
 	struct irq_cfg *cfg;
+	struct irq_desc *desc;
 
 	/* Mark the inuse vectors */
 	for_each_irq_cfg(irq, cfg) {
 		if (!cpu_isset(cpu, cfg->domain))
 			continue;
 		vector = cfg->vector;
-		per_cpu(vector_irq, cpu)[vector] = irq;
+		desc = irq_to_desc(irq);
+		BUG_ON(desc == NULL);
+		per_cpu(vector_irq, cpu)[vector] = desc;
 	}
 	/* Mark the free vectors */
 	for (vector = 0; vector < NR_VECTORS; ++vector) {
-		irq = per_cpu(vector_irq, cpu)[vector];
-		if (irq < 0)
+		desc = per_cpu(vector_irq, cpu)[vector];
+		if (desc == NULL)
 			continue;
 
-		cfg = irq_cfg(irq);
+		cfg = irq_cfg(desc->irq);
 		if (!cpu_isset(cpu, cfg->domain))
-			per_cpu(vector_irq, cpu)[vector] = -1;
+			per_cpu(vector_irq, cpu)[vector] = NULL;
 	}
 }
 
@@ -2389,16 +2395,14 @@ asmlinkage void smp_irq_move_cleanup_int
 
 	me = smp_processor_id();
 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
-		unsigned int irq;
 		struct irq_desc *desc;
 		struct irq_cfg *cfg;
-		irq = __get_cpu_var(vector_irq)[vector];
 
-		desc = irq_to_desc(irq);
-		if (!desc)
+		desc = __get_cpu_var(vector_irq)[vector];
+		if (desc == NULL)
 			continue;
 
-		cfg = irq_cfg(irq);
+		cfg = irq_cfg(desc->irq);
 		spin_lock(&desc->lock);
 		if (!cfg->move_cleanup_count)
 			goto unlock;
@@ -2406,7 +2410,7 @@ asmlinkage void smp_irq_move_cleanup_int
 		if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
 			goto unlock;
 
-		__get_cpu_var(vector_irq)[vector] = -1;
+		__get_cpu_var(vector_irq)[vector] = NULL;
 		cfg->move_cleanup_count--;
 unlock:
 		spin_unlock(&desc->lock);
Index: linux/include/asm-x86/hw_irq.h
===================================================================
--- linux.orig/include/asm-x86/hw_irq.h	2008-09-19 14:12:25.000000000 -0500
+++ linux/include/asm-x86/hw_irq.h	2008-09-19 14:13:09.000000000 -0500
@@ -113,7 +113,7 @@ extern asmlinkage void smp_invalidate_in
 extern void (*const interrupt[NR_VECTORS])(void);
 #endif
 
-typedef int vector_irq_t[NR_VECTORS];
+typedef struct irq_desc *vector_irq_t[NR_VECTORS];
 DECLARE_PER_CPU(vector_irq_t, vector_irq);
 
 #ifdef CONFIG_X86_IO_APIC
Index: linux/arch/x86/kernel/irq_32.c
===================================================================
--- linux.orig/arch/x86/kernel/irq_32.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/irq_32.c	2008-09-19 14:13:09.000000000 -0500
@@ -226,26 +226,24 @@ unsigned int do_IRQ(struct pt_regs *regs
 	int overflow;
 	unsigned vector = ~regs->orig_ax;
 	struct irq_desc *desc;
-	unsigned irq;
 
 
 	old_regs = set_irq_regs(regs);
 	irq_enter();
-	irq = __get_cpu_var(vector_irq)[vector];
 
 	overflow = check_stack_overflow();
 
-	desc = irq_to_desc(irq);
-	if (unlikely(!desc)) {
-		printk(KERN_EMERG "%s: cannot handle IRQ %d vector %#x cpu %d\n",
-					__func__, irq, vector, smp_processor_id());
+	desc = __get_cpu_var(vector_irq)[vector];
+	if (unlikely(desc == NULL)) {
+		printk(KERN_EMERG "%s: cannot handle IRQ vector %#x cpu %d\n",
+					__func__, vector, smp_processor_id());
 		BUG();
 	}
 
-	if (!execute_on_irq_stack(overflow, desc, irq)) {
+	if (!execute_on_irq_stack(overflow, desc, desc->irq)) {
 		if (unlikely(overflow))
 			print_stack_overflow();
-		desc->handle_irq(irq, desc);
+		desc->handle_irq(desc->irq, desc);
 	}
 
 	irq_exit();
Index: linux/arch/x86/kernel/irq_64.c
===================================================================
--- linux.orig/arch/x86/kernel/irq_64.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/irq_64.c	2008-09-19 14:13:09.000000000 -0500
@@ -213,20 +213,18 @@ asmlinkage unsigned int do_IRQ(struct pt
 
 	/* high bit used in ret_from_ code  */
 	unsigned vector = ~regs->orig_ax;
-	unsigned irq;
 
 	exit_idle();
 	irq_enter();
-	irq = __get_cpu_var(vector_irq)[vector];
 
 #ifdef CONFIG_DEBUG_STACKOVERFLOW
 	stack_overflow_check(regs);
 #endif
 
-	desc = irq_to_desc(irq);
-	if (likely(desc))
-		generic_handle_irq_desc(irq, desc);
-	else {
+	desc = __get_cpu_var(vector_irq)[vector];
+	if (likely(desc != NULL)) {
+		generic_handle_irq_desc(desc->irq, desc);
+	} else {
 		if (!disable_apic)
 			ack_APIC_irq();
 
Index: linux/arch/x86/kernel/vmiclock_32.c
===================================================================
--- linux.orig/arch/x86/kernel/vmiclock_32.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/vmiclock_32.c	2008-09-19 14:13:09.000000000 -0500
@@ -242,7 +242,7 @@ void __init vmi_time_init(void)
 	vmi_time_init_clockevent();
 	setup_irq(0, &vmi_clock_action);
 	for_each_possible_cpu(cpu)
-		per_cpu(vector_irq, cpu)[vmi_get_timer_vector()] = 0;
+		per_cpu(vector_irq, cpu)[vmi_get_timer_vector()] = NULL;
 }
 
 #ifdef CONFIG_X86_LOCAL_APIC
Index: linux/arch/x86/xen/irq.c
===================================================================
--- linux.orig/arch/x86/xen/irq.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/xen/irq.c	2008-09-19 14:13:09.000000000 -0500
@@ -27,8 +27,9 @@ static void __init __xen_init_IRQ(void)
 	for(i = 0; i < NR_VECTORS; i++) {
 		int cpu;
 
+		desc = irq_to_desc_alloc(i);
 		for_each_possible_cpu(cpu)
-			per_cpu(vector_irq, cpu)[i] = i;
+			per_cpu(vector_irq, cpu)[i] = desc;
 	}
 
 	xen_init_IRQ();
Index: linux/arch/x86/kernel/irqinit_32.c
===================================================================
--- linux.orig/arch/x86/kernel/irqinit_32.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/irqinit_32.c	2008-09-19 14:13:09.000000000 -0500
@@ -59,6 +59,8 @@ static struct irqaction fpu_irq = {
 void __init init_ISA_irqs (void)
 {
 	int i;
+	int cpu;
+	unsigned int vector;
 
 #ifdef CONFIG_X86_LOCAL_APIC
 	init_bsp_APIC();
@@ -76,6 +78,12 @@ void __init init_ISA_irqs (void)
 		desc->action = NULL;
 		desc->depth = 1;
 
+		vector = IRQ0_VECTOR + i;
+		for_each_possible_cpu(cpu) {
+			BUG_ON(per_cpu(vector_irq, cpu)[vector] != NULL);
+			per_cpu(vector_irq, cpu)[vector] = desc;
+		}
+
 		set_irq_chip_and_handler_name(i, &i8259A_chip,
 					      handle_level_irq, "XT");
 	}
@@ -90,26 +98,7 @@ static struct irqaction irq2 = {
 	.name = "cascade",
 };
 
-DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
-	[0 ... IRQ0_VECTOR - 1] = -1,
-	[IRQ0_VECTOR] = 0,
-	[IRQ1_VECTOR] = 1,
-	[IRQ2_VECTOR] = 2,
-	[IRQ3_VECTOR] = 3,
-	[IRQ4_VECTOR] = 4,
-	[IRQ5_VECTOR] = 5,
-	[IRQ6_VECTOR] = 6,
-	[IRQ7_VECTOR] = 7,
-	[IRQ8_VECTOR] = 8,
-	[IRQ9_VECTOR] = 9,
-	[IRQ10_VECTOR] = 10,
-	[IRQ11_VECTOR] = 11,
-	[IRQ12_VECTOR] = 12,
-	[IRQ13_VECTOR] = 13,
-	[IRQ14_VECTOR] = 14,
-	[IRQ15_VECTOR] = 15,
-	[IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
-};
+DEFINE_PER_CPU(vector_irq_t, vector_irq);
 
 /* Overridden in paravirt.c */
 void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
Index: linux/arch/x86/kernel/irqinit_64.c
===================================================================
--- linux.orig/arch/x86/kernel/irqinit_64.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/irqinit_64.c	2008-09-19 14:13:09.000000000 -0500
@@ -114,30 +114,13 @@ static struct irqaction irq2 = {
 	.mask = CPU_MASK_NONE,
 	.name = "cascade",
 };
-DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
-	[0 ... IRQ0_VECTOR - 1] = -1,
-	[IRQ0_VECTOR] = 0,
-	[IRQ1_VECTOR] = 1,
-	[IRQ2_VECTOR] = 2,
-	[IRQ3_VECTOR] = 3,
-	[IRQ4_VECTOR] = 4,
-	[IRQ5_VECTOR] = 5,
-	[IRQ6_VECTOR] = 6,
-	[IRQ7_VECTOR] = 7,
-	[IRQ8_VECTOR] = 8,
-	[IRQ9_VECTOR] = 9,
-	[IRQ10_VECTOR] = 10,
-	[IRQ11_VECTOR] = 11,
-	[IRQ12_VECTOR] = 12,
-	[IRQ13_VECTOR] = 13,
-	[IRQ14_VECTOR] = 14,
-	[IRQ15_VECTOR] = 15,
-	[IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
-};
+DEFINE_PER_CPU(vector_irq_t, vector_irq);
 
 static void __init init_ISA_irqs (void)
 {
 	int i;
+	int cpu;
+	unsigned int vector;
 
 	init_bsp_APIC();
 	init_8259A(0);
@@ -150,6 +133,12 @@ static void __init init_ISA_irqs (void)
 		desc->action = NULL;
 		desc->depth = 1;
 
+		vector = IRQ0_VECTOR + i;
+		for_each_possible_cpu(cpu) {
+			BUG_ON(per_cpu(vector_irq, cpu)[vector] != NULL);
+			per_cpu(vector_irq, cpu)[vector] = desc;
+		}
+
 		/*
 		 * 16 old-style INTA-cycle interrupts:
 		 */

[PATCH 2/3] switch static system vector allocation to use vector_irq[] v2

[Dean Nelson]

Replace the current use of system_vectors[] for the allocation of static
system vectors by also using the per_cpu variable vector_irq[].

Signed-off-by: Dean Nelson <dcn@sgi.com>

---

 arch/x86/kernel/Makefile     |    2 
 arch/x86/kernel/apic.c       |    2 
 arch/x86/kernel/io_apic.c    |  112 ++++++++++++++++++++-------------------
 arch/x86/kernel/irq.c        |   22 +++++++
 arch/x86/kernel/irq_32.c     |    2 
 arch/x86/kernel/irq_64.c     |    2 
 arch/x86/kernel/irqinit_32.c |   11 ++-
 arch/x86/kernel/irqinit_64.c |   11 ++-
 include/asm-x86/desc.h       |   21 +++----
 include/asm-x86/hw_irq.h     |    4 -
 include/asm-x86/irq.h        |    8 ++
 11 files changed, 118 insertions(+), 79 deletions(-)

Index: linux/arch/x86/kernel/apic.c
===================================================================
--- linux.orig/arch/x86/kernel/apic.c	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/apic.c	2008-09-19 14:13:17.000000000 -0500
@@ -118,8 +118,6 @@ EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok
 
 int first_system_vector = 0xfe;
 
-char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE};
-
 /*
  * Debug level, exported for io_apic.c
  */
Index: linux/include/asm-x86/desc.h
===================================================================
--- linux.orig/include/asm-x86/desc.h	2008-09-19 14:12:25.000000000 -0500
+++ linux/include/asm-x86/desc.h	2008-09-19 14:13:17.000000000 -0500
@@ -5,6 +5,7 @@
 #include <asm/desc_defs.h>
 #include <asm/ldt.h>
 #include <asm/mmu.h>
+#include <asm/irq.h>
 #include <linux/smp.h>
 
 static inline void fill_ldt(struct desc_struct *desc,
@@ -320,20 +321,20 @@ static inline void set_intr_gate(unsigne
 	_set_gate(n, GATE_INTERRUPT, addr, 0, 0, __KERNEL_CS);
 }
 
-#define SYS_VECTOR_FREE		0
-#define SYS_VECTOR_ALLOCED	1
-
 extern int first_system_vector;
-extern char system_vectors[];
 
 static inline void alloc_system_vector(int vector)
 {
-	if (system_vectors[vector] == SYS_VECTOR_FREE) {
-		system_vectors[vector] = SYS_VECTOR_ALLOCED;
-		if (first_system_vector > vector)
-			first_system_vector = vector;
-	} else
-		BUG();
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&vector_lock, flags);
+	ret = grab_irq_vector(NON_IRQ_DESC, vector, &cpu_possible_map);
+	BUG_ON(ret != vector);
+
+	if (first_system_vector > vector)
+		first_system_vector = vector;
+	spin_unlock_irqrestore(&vector_lock, flags);
 }
 
 static inline void alloc_intr_gate(unsigned int n, void *addr)
Index: linux/arch/x86/kernel/io_apic.c
===================================================================
--- linux.orig/arch/x86/kernel/io_apic.c	2008-09-19 14:13:09.000000000 -0500
+++ linux/arch/x86/kernel/io_apic.c	2008-09-19 14:13:17.000000000 -0500
@@ -72,7 +72,6 @@
 int sis_apic_bug = -1;
 
 static DEFINE_SPINLOCK(ioapic_lock);
-static DEFINE_SPINLOCK(vector_lock);
 
 /*
  * # of IRQ routing registers
@@ -546,7 +545,7 @@ static void __target_IO_APIC_irq(unsigne
 	}
 }
 
-static int assign_irq_vector(int irq, cpumask_t mask);
+static int assign_irq_vector(int irq, cpumask_t *mask);
 
 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
 {
@@ -561,7 +560,7 @@ static void set_ioapic_affinity_irq(unsi
 		return;
 
 	cfg = irq_cfg(irq);
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cpus_and(tmp, cfg->domain, mask);
@@ -1196,20 +1195,33 @@ static int pin_2_irq(int idx, int apic, 
 	return irq;
 }
 
-void lock_vector_lock(void)
+static int ioapic_grab_irq_vector(struct irq_desc *desc, unsigned int vector,
+				  cpumask_t *domain, cpumask_t *domain_online)
 {
-	/* Used to the online set of cpus does not change
-	 * during assign_irq_vector.
-	 */
-	spin_lock(&vector_lock);
-}
+	struct irq_cfg *cfg;
+	int ret;
 
-void unlock_vector_lock(void)
-{
-	spin_unlock(&vector_lock);
+	BUG_ON(!spin_is_locked(&vector_lock));
+
+	ret = grab_irq_vector(desc, vector, domain_online);
+	if (ret == vector) {
+		cfg = irq_cfg(desc->irq);
+		if (cfg->vector) {
+			cfg->move_in_progress = 1;
+			cfg->old_domain = cfg->domain;
+		}
+		cfg->vector = vector;
+		cfg->domain = *domain;
+	}
+	return ret;
 }
 
-static int __assign_irq_vector(int irq, cpumask_t mask)
+/* The following scratch cpumask_t variables are protected by vector lock. */
+static cpumask_t eligible_cpus;
+static cpumask_t domain;
+static cpumask_t domain_online;
+
+static int __assign_irq_vector(int irq, cpumask_t *mask)
 {
 	/*
 	 * NOTE! The local APIC isn't very good at handling
@@ -1223,36 +1235,34 @@ static int __assign_irq_vector(int irq, 
 	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
 	 */
 	static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
-	unsigned int old_vector;
 	int cpu;
 	struct irq_cfg *cfg;
 	struct irq_desc *desc;
+	int ret;
 
-	cfg = irq_cfg(irq);
+	BUG_ON(!spin_is_locked(&vector_lock));
 
-	/* Only try and allocate irqs on cpus that are present */
-	cpus_and(mask, mask, cpu_online_map);
+	cfg = irq_cfg(irq);
 
 	if ((cfg->move_in_progress) || cfg->move_cleanup_count)
 		return -EBUSY;
 
-	old_vector = cfg->vector;
-	if (old_vector) {
-		cpumask_t tmp;
-		cpus_and(tmp, cfg->domain, mask);
-		if (!cpus_empty(tmp))
+	/* Only try and allocate irqs on cpus that are present */
+	cpus_and(eligible_cpus, *mask, cpu_online_map);
+
+	if (cfg->vector) {
+		cpus_and(domain, cfg->domain, eligible_cpus);
+		if (!cpus_empty(domain))
 			return 0;
 	}
 
 	desc = irq_to_desc_alloc(irq);
 
-	for_each_cpu_mask_nr(cpu, mask) {
-		cpumask_t domain, new_mask;
-		int new_cpu;
+	for_each_cpu_mask_nr(cpu, eligible_cpus) {
 		int vector, offset;
 
 		domain = vector_allocation_domain(cpu);
-		cpus_and(new_mask, domain, cpu_online_map);
+		cpus_and(domain_online, domain, cpu_online_map);
 
 		vector = current_vector;
 		offset = current_offset;
@@ -1272,26 +1282,20 @@ next:
 		if (vector == SYSCALL_VECTOR)
 			goto next;
 #endif
-		for_each_cpu_mask_nr(new_cpu, new_mask)
-			if (per_cpu(vector_irq, new_cpu)[vector] != NULL)
-				goto next;
+		ret = ioapic_grab_irq_vector(desc, vector, &domain,
+					     &domain_online);
+		if (ret != vector)
+			goto next;
+
 		/* Found one! */
 		current_vector = vector;
 		current_offset = offset;
-		if (old_vector) {
-			cfg->move_in_progress = 1;
-			cfg->old_domain = cfg->domain;
-		}
-		for_each_cpu_mask_nr(new_cpu, new_mask)
-			per_cpu(vector_irq, new_cpu)[vector] = desc;
-		cfg->vector = vector;
-		cfg->domain = domain;
 		return 0;
 	}
 	return -ENOSPC;
 }
 
-static int assign_irq_vector(int irq, cpumask_t mask)
+static int assign_irq_vector(int irq, cpumask_t *mask)
 {
 	int err;
 	unsigned long flags;
@@ -1334,13 +1338,13 @@ void __setup_vector_irq(int cpu)
 			continue;
 		vector = cfg->vector;
 		desc = irq_to_desc(irq);
-		BUG_ON(desc == NULL);
+		BUG_ON(desc == NULL || desc == NON_IRQ_DESC);
 		per_cpu(vector_irq, cpu)[vector] = desc;
 	}
 	/* Mark the free vectors */
 	for (vector = 0; vector < NR_VECTORS; ++vector) {
 		desc = per_cpu(vector_irq, cpu)[vector];
-		if (desc == NULL)
+		if (desc == NULL || desc == NON_IRQ_DESC)
 			continue;
 
 		cfg = irq_cfg(desc->irq);
@@ -1495,7 +1499,7 @@ static void setup_IO_APIC_irq(int apic, 
 	cfg = irq_cfg(irq);
 
 	mask = TARGET_CPUS;
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cpus_and(mask, cfg->domain, mask);
@@ -2280,7 +2284,7 @@ static void migrate_ioapic_irq(int irq, 
 	if (get_irte(irq, &irte))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cfg = irq_cfg(irq);
@@ -2399,7 +2403,7 @@ asmlinkage void smp_irq_move_cleanup_int
 		struct irq_cfg *cfg;
 
 		desc = __get_cpu_var(vector_irq)[vector];
-		if (desc == NULL)
+		if (desc == NULL || desc == NON_IRQ_DESC)
 			continue;
 
 		cfg = irq_cfg(desc->irq);
@@ -2764,6 +2768,7 @@ static inline void __init check_timer(vo
 	unsigned long flags;
 	unsigned int ver;
 	int no_pin1 = 0;
+	cpumask_t mask;
 
 	local_irq_save(flags);
 
@@ -2774,7 +2779,8 @@ static inline void __init check_timer(vo
 	 * get/set the timer IRQ vector:
 	 */
 	disable_8259A_irq(0);
-	assign_irq_vector(0, TARGET_CPUS);
+	mask = TARGET_CPUS;
+	assign_irq_vector(0, &mask);
 
 	/*
 	 * As IRQ0 is to be enabled in the 8259A, the virtual
@@ -3074,10 +3080,12 @@ unsigned int create_irq_nr(unsigned int 
 	unsigned int new;
 	unsigned long flags;
 	struct irq_cfg *cfg_new;
+	cpumask_t mask;
 
 #ifndef CONFIG_HAVE_SPARSE_IRQ
 	irq_want = nr_irqs - 1;
 #endif
+	mask = TARGET_CPUS;
 
 	irq = 0;
 	spin_lock_irqsave(&vector_lock, flags);
@@ -3090,7 +3098,7 @@ unsigned int create_irq_nr(unsigned int 
 		/* check if need to create one */
 		if (!cfg_new)
 			cfg_new = irq_cfg_alloc(new);
-		if (__assign_irq_vector(new, TARGET_CPUS) == 0)
+		if (__assign_irq_vector(new, &mask) == 0)
 			irq = new;
 		break;
 	}
@@ -3140,7 +3148,7 @@ static int msi_compose_msg(struct pci_de
 	cpumask_t tmp;
 
 	tmp = TARGET_CPUS;
-	err = assign_irq_vector(irq, tmp);
+	err = assign_irq_vector(irq, &tmp);
 	if (err)
 		return err;
 
@@ -3212,7 +3220,7 @@ static void set_msi_irq_affinity(unsigne
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cfg = irq_cfg(irq);
@@ -3251,7 +3259,7 @@ static void ir_set_msi_irq_affinity(unsi
 	if (get_irte(irq, &irte))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cfg = irq_cfg(irq);
@@ -3491,7 +3499,7 @@ static void dmar_msi_set_affinity(unsign
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cfg = irq_cfg(irq);
@@ -3552,7 +3560,7 @@ static void hpet_msi_set_affinity(unsign
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cfg = irq_cfg(irq);
@@ -3632,7 +3640,7 @@ static void set_ht_irq_affinity(unsigned
 	if (cpus_empty(tmp))
 		return;
 
-	if (assign_irq_vector(irq, mask))
+	if (assign_irq_vector(irq, &mask))
 		return;
 
 	cfg = irq_cfg(irq);
@@ -3663,7 +3671,7 @@ int arch_setup_ht_irq(unsigned int irq, 
 	cpumask_t tmp;
 
 	tmp = TARGET_CPUS;
-	err = assign_irq_vector(irq, tmp);
+	err = assign_irq_vector(irq, &tmp);
 	if (!err) {
 		struct ht_irq_msg msg;
 		unsigned dest;
Index: linux/arch/x86/kernel/irq_32.c
===================================================================
--- linux.orig/arch/x86/kernel/irq_32.c	2008-09-19 14:13:09.000000000 -0500
+++ linux/arch/x86/kernel/irq_32.c	2008-09-19 14:13:17.000000000 -0500
@@ -234,7 +234,7 @@ unsigned int do_IRQ(struct pt_regs *regs
 	overflow = check_stack_overflow();
 
 	desc = __get_cpu_var(vector_irq)[vector];
-	if (unlikely(desc == NULL)) {
+	if (unlikely(desc == NULL || desc == NON_IRQ_DESC)) {
 		printk(KERN_EMERG "%s: cannot handle IRQ vector %#x cpu %d\n",
 					__func__, vector, smp_processor_id());
 		BUG();
Index: linux/arch/x86/kernel/irq_64.c
===================================================================
--- linux.orig/arch/x86/kernel/irq_64.c	2008-09-19 14:13:09.000000000 -0500
+++ linux/arch/x86/kernel/irq_64.c	2008-09-19 14:13:17.000000000 -0500
@@ -222,7 +222,7 @@ asmlinkage unsigned int do_IRQ(struct pt
 #endif
 
 	desc = __get_cpu_var(vector_irq)[vector];
-	if (likely(desc != NULL)) {
+	if (likely(desc != NULL && desc != NON_IRQ_DESC)) {
 		generic_handle_irq_desc(desc->irq, desc);
 	} else {
 		if (!disable_apic)
Index: linux/include/asm-x86/irq.h
===================================================================
--- linux.orig/include/asm-x86/irq.h	2008-09-19 14:12:25.000000000 -0500
+++ linux/include/asm-x86/irq.h	2008-09-19 14:13:17.000000000 -0500
@@ -47,4 +47,12 @@ extern void native_init_IRQ(void);
 /* Interrupt vector management */
 extern DECLARE_BITMAP(used_vectors, NR_VECTORS);
 
+struct irq_desc;
+extern int grab_irq_vector(struct irq_desc *desc, unsigned int vector,
+			   cpumask_t *new_domain_mask);
+
+#define NON_IRQ_DESC	((struct irq_desc *)-1UL)
+
+extern spinlock_t vector_lock;
+
 #endif /* ASM_X86__IRQ_H */
Index: linux/arch/x86/kernel/Makefile
===================================================================
--- linux.orig/arch/x86/kernel/Makefile	2008-09-19 14:12:25.000000000 -0500
+++ linux/arch/x86/kernel/Makefile	2008-09-19 14:13:17.000000000 -0500
@@ -24,7 +24,7 @@ CFLAGS_tsc.o		:= $(nostackp)
 CFLAGS_paravirt.o	:= $(nostackp)
 
 obj-y			:= process_$(BITS).o signal_$(BITS).o entry_$(BITS).o
-obj-y			+= traps_$(BITS).o irq_$(BITS).o
+obj-y			+= traps_$(BITS).o irq.o irq_$(BITS).o
 obj-y			+= time_$(BITS).o ioport.o ldt.o
 obj-y			+= setup.o i8259.o irqinit_$(BITS).o setup_percpu.o
 obj-$(CONFIG_X86_VISWS)	+= visws_quirks.o
Index: linux/arch/x86/kernel/irq.c
===================================================================
--- /dev/null	1970-01-01 00:00:00.000000000 +0000
+++ linux/arch/x86/kernel/irq.c	2008-09-19 14:13:17.000000000 -0500
@@ -0,0 +1,22 @@
+#include <linux/irq.h>
+
+DEFINE_SPINLOCK(vector_lock);
+
+int grab_irq_vector(struct irq_desc *desc, unsigned int vector,
+		   cpumask_t *domain)
+{
+	int cpu;
+
+	BUG_ON(!spin_is_locked(&vector_lock));
+
+	for_each_cpu_mask_nr(cpu, *domain) {
+		if (per_cpu(vector_irq, cpu)[vector] != NULL)
+			return -EBUSY;
+	}
+
+	/* it's available, reserve it */
+	for_each_cpu_mask_nr(cpu, *domain)
+		per_cpu(vector_irq, cpu)[vector] = desc;
+
+	return vector;
+}
Index: linux/include/asm-x86/hw_irq.h
===================================================================
--- linux.orig/include/asm-x86/hw_irq.h	2008-09-19 14:13:09.000000000 -0500
+++ linux/include/asm-x86/hw_irq.h	2008-09-19 14:13:17.000000000 -0500
@@ -117,8 +117,8 @@ typedef struct irq_desc *vector_irq_t[NR
 DECLARE_PER_CPU(vector_irq_t, vector_irq);
 
 #ifdef CONFIG_X86_IO_APIC
-extern void lock_vector_lock(void);
-extern void unlock_vector_lock(void);
+static inline void lock_vector_lock(void) { spin_lock(&vector_lock); }
+static inline void unlock_vector_lock(void) { spin_unlock(&vector_lock); }
 extern void __setup_vector_irq(int cpu);
 #else
 static inline void lock_vector_lock(void) {}
Index: linux/arch/x86/kernel/irqinit_32.c
===================================================================
--- linux.orig/arch/x86/kernel/irqinit_32.c	2008-09-19 14:13:09.000000000 -0500
+++ linux/arch/x86/kernel/irqinit_32.c	2008-09-19 14:13:17.000000000 -0500
@@ -59,8 +59,9 @@ static struct irqaction fpu_irq = {
 void __init init_ISA_irqs (void)
 {
 	int i;
-	int cpu;
+	int ret;
 	unsigned int vector;
+	unsigned long flags;
 
 #ifdef CONFIG_X86_LOCAL_APIC
 	init_bsp_APIC();
@@ -79,10 +80,10 @@ void __init init_ISA_irqs (void)
 		desc->depth = 1;
 
 		vector = IRQ0_VECTOR + i;
-		for_each_possible_cpu(cpu) {
-			BUG_ON(per_cpu(vector_irq, cpu)[vector] != NULL);
-			per_cpu(vector_irq, cpu)[vector] = desc;
-		}
+		spin_lock_irqsave(&vector_lock, flags);
+		ret = grab_irq_vector(desc, vector, &cpu_possible_map);
+		BUG_ON(ret != vector);
+		spin_unlock_irqrestore(&vector_lock, flags);
 
 		set_irq_chip_and_handler_name(i, &i8259A_chip,
 					      handle_level_irq, "XT");
Index: linux/arch/x86/kernel/irqinit_64.c
===================================================================
--- linux.orig/arch/x86/kernel/irqinit_64.c	2008-09-19 14:13:09.000000000 -0500
+++ linux/arch/x86/kernel/irqinit_64.c	2008-09-19 14:13:17.000000000 -0500
@@ -119,8 +119,9 @@ DEFINE_PER_CPU(vector_irq_t, vector_irq)
 static void __init init_ISA_irqs (void)
 {
 	int i;
-	int cpu;
+	int ret;
 	unsigned int vector;
+	unsigned long flags;
 
 	init_bsp_APIC();
 	init_8259A(0);
@@ -134,10 +135,10 @@ static void __init init_ISA_irqs (void)
 		desc->depth = 1;
 
 		vector = IRQ0_VECTOR + i;
-		for_each_possible_cpu(cpu) {
-			BUG_ON(per_cpu(vector_irq, cpu)[vector] != NULL);
-			per_cpu(vector_irq, cpu)[vector] = desc;
-		}
+		spin_lock_irqsave(&vector_lock, flags);
+		ret = grab_irq_vector(desc, vector, &cpu_possible_map);
+		BUG_ON(ret != vector);
+		spin_unlock_irqrestore(&vector_lock, flags);
 
 		/*
 		 * 16 old-style INTA-cycle interrupts:

[PATCH 3/3] switch non-standard SYSCALL_VECTOR allocation to use vector_irq v2

[Dean Nelson]

Replace the current use of used_vectors[] for the allocation of a non-standard
SYSCALL_VECTOR by also using the per_cpu variable vector_irq[].

Signed-off-by: Dean Nelson <dcn@sgi.com>

---

 arch/x86/kernel/traps_32.c            |   16 ++++++++++------
 drivers/lguest/interrupts_and_traps.c |   28 ++++++++++++++++++++++------
 include/asm-x86/irq.h                 |    3 ---
 3 files changed, 32 insertions(+), 15 deletions(-)

Index: linux/arch/x86/kernel/traps_32.c
===================================================================
--- linux.orig/arch/x86/kernel/traps_32.c	2008-09-19 11:01:10.000000000 -0500
+++ linux/arch/x86/kernel/traps_32.c	2008-09-19 12:35:32.000000000 -0500
@@ -63,9 +63,6 @@
 
 #include "mach_traps.h"
 
-DECLARE_BITMAP(used_vectors, NR_VECTORS);
-EXPORT_SYMBOL_GPL(used_vectors);
-
 asmlinkage int system_call(void);
 
 /* Do we ignore FPU interrupts ? */
@@ -1185,6 +1182,8 @@ asmlinkage void math_emulate(long arg)
 void __init trap_init(void)
 {
 	int i;
+	int ret;
+	unsigned long flags;
 
 #ifdef CONFIG_EISA
 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
@@ -1232,10 +1231,15 @@ void __init trap_init(void)
 	set_system_gate(SYSCALL_VECTOR, &system_call);
 
 	/* Reserve all the builtin and the syscall vector: */
-	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
-		set_bit(i, used_vectors);
+	spin_lock_irqsave(&vector_lock, flags);
+	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) {
+		ret = grab_irq_vector(NON_IRQ_DESC, i, &cpu_possible_map);
+		BUG_ON(ret != i);
+	}
 
-	set_bit(SYSCALL_VECTOR, used_vectors);
+	ret = grab_irq_vector(NON_IRQ_DESC, SYSCALL_VECTOR, &cpu_possible_map);
+	BUG_ON(ret != SYSCALL_VECTOR);
+	spin_unlock_irqrestore(&vector_lock, flags);
 
 	/*
 	 * Should be a barrier for any external CPU state:
Index: linux/include/asm-x86/irq.h
===================================================================
--- linux.orig/include/asm-x86/irq.h	2008-09-19 11:09:32.000000000 -0500
+++ linux/include/asm-x86/irq.h	2008-09-19 12:35:32.000000000 -0500
@@ -44,9 +44,6 @@ extern unsigned int do_IRQ(struct pt_reg
 extern void init_IRQ(void);
 extern void native_init_IRQ(void);
 
-/* Interrupt vector management */
-extern DECLARE_BITMAP(used_vectors, NR_VECTORS);
-
 struct irq_desc;
 extern int grab_irq_vector(struct irq_desc *desc, unsigned int vector,
 			   cpumask_t *new_domain_mask);
Index: linux/drivers/lguest/interrupts_and_traps.c
===================================================================
--- linux.orig/drivers/lguest/interrupts_and_traps.c	2008-09-19 11:01:10.000000000 -0500
+++ linux/drivers/lguest/interrupts_and_traps.c	2008-09-19 12:41:02.000000000 -0500
@@ -221,19 +221,35 @@ bool check_syscall_vector(struct lguest 
 
 int init_interrupts(void)
 {
+	unsigned long flags;
+	int ret;
+
 	/* If they want some strange system call vector, reserve it now */
-	if (syscall_vector != SYSCALL_VECTOR
-	    && test_and_set_bit(syscall_vector, used_vectors)) {
-		printk("lg: couldn't reserve syscall %u\n", syscall_vector);
-		return -EBUSY;
+	if (syscall_vector != SYSCALL_VECTOR) {
+		spin_lock_irqsave(&vector_lock, flags);
+		ret = grab_irq_vector(NON_IRQ_DESC, syscall_vector,
+				      &cpu_possible_map);
+		spin_unlock_irqrestore(&vector_lock, flags);
+		if (ret != syscall_vector) {
+			printk(KERN_WARNING "lg: couldn't reserve syscall %u\n",
+			       syscall_vector);
+			return ret;
+		}
 	}
 	return 0;
 }
 
 void free_interrupts(void)
 {
-	if (syscall_vector != SYSCALL_VECTOR)
-		clear_bit(syscall_vector, used_vectors);
+	int cpu;
+
+	if (syscall_vector != SYSCALL_VECTOR) {
+		for_each_possible_cpu(cpu) {
+			BUG_ON(per_cpu(vector_irq, cpu)[syscall_vector] !=
+			       NON_IRQ_DESC);
+			per_cpu(vector_irq, cpu)[syscall_vector] = NULL;
+		}
+	}
 }
 
 /*H:220 Now we've got the routines to deliver interrupts, delivering traps like

Re: [PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Mike Travis]

Hi Ingo,

Note that my recent patch from a couple of weeks ago, stripped almost all of
the cpumask_t's from the io_apic functions.  Any chance we can apply them
before we add more?

I'd be willing to merge in Dean's patch over the top of mine (or vice-versa).

Thanks,
Mike

Dean Nelson wrote:
> Change per_cpu variable vector_irq[] from holding an 'int' irq number to
> holding a 'struct irq_desc' pointer.
> 
> Signed-off-by: Dean Nelson <dcn@sgi.com>
> 
> ---
> 
> Note that this pre-allocates the irq_desc structure before we know whether
> a vector will be successfully found. And should it not, the irq_desc structure
> is left created with the irq still unallocated. Should someone in the future
> attempt to allocate a vector with that same irq and succeed, they will get
> the previously allocated irq_desc structure.
> 
> Also, I won't claim the changes to arch/x86/xen/irq.c were correctly done, and
> I know they weren't tested.
> 
>  arch/x86/kernel/io_apic.c     |   32 ++++++++++++++++++--------------
>  arch/x86/kernel/irq_32.c      |   14 ++++++--------
>  arch/x86/kernel/irq_64.c      |   10 ++++------
>  arch/x86/kernel/irqinit_32.c  |   29 +++++++++--------------------
>  arch/x86/kernel/irqinit_64.c  |   29 +++++++++--------------------
>  arch/x86/kernel/vmiclock_32.c |    2 +-
>  arch/x86/xen/irq.c            |    3 ++-
>  include/asm-x86/hw_irq.h      |    2 +-
>  8 files changed, 50 insertions(+), 71 deletions(-)
> 
> Index: linux/arch/x86/kernel/io_apic.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/io_apic.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/kernel/io_apic.c	2008-09-19 14:13:09.000000000 -0500
> @@ -1226,6 +1226,7 @@ static int __assign_irq_vector(int irq, 
>  	unsigned int old_vector;
>  	int cpu;
>  	struct irq_cfg *cfg;
> +	struct irq_desc *desc;
>  
>  	cfg = irq_cfg(irq);
>  
> @@ -1243,6 +1244,8 @@ static int __assign_irq_vector(int irq, 
>  			return 0;
>  	}
>  
> +	desc = irq_to_desc_alloc(irq);
> +
>  	for_each_cpu_mask_nr(cpu, mask) {
>  		cpumask_t domain, new_mask;
>  		int new_cpu;
> @@ -1270,7 +1273,7 @@ next:
>  			goto next;
>  #endif
>  		for_each_cpu_mask_nr(new_cpu, new_mask)
> -			if (per_cpu(vector_irq, new_cpu)[vector] != -1)
> +			if (per_cpu(vector_irq, new_cpu)[vector] != NULL)
>  				goto next;
>  		/* Found one! */
>  		current_vector = vector;
> @@ -1280,7 +1283,7 @@ next:
>  			cfg->old_domain = cfg->domain;
>  		}
>  		for_each_cpu_mask_nr(new_cpu, new_mask)
> -			per_cpu(vector_irq, new_cpu)[vector] = irq;
> +			per_cpu(vector_irq, new_cpu)[vector] = desc;
>  		cfg->vector = vector;
>  		cfg->domain = domain;
>  		return 0;
> @@ -1311,7 +1314,7 @@ static void __clear_irq_vector(int irq)
>  	vector = cfg->vector;
>  	cpus_and(mask, cfg->domain, cpu_online_map);
>  	for_each_cpu_mask_nr(cpu, mask)
> -		per_cpu(vector_irq, cpu)[vector] = -1;
> +		per_cpu(vector_irq, cpu)[vector] = NULL;
>  
>  	cfg->vector = 0;
>  	cpus_clear(cfg->domain);
> @@ -1323,23 +1326,26 @@ void __setup_vector_irq(int cpu)
>  	/* This function must be called with vector_lock held */
>  	int irq, vector;
>  	struct irq_cfg *cfg;
> +	struct irq_desc *desc;
>  
>  	/* Mark the inuse vectors */
>  	for_each_irq_cfg(irq, cfg) {
>  		if (!cpu_isset(cpu, cfg->domain))
>  			continue;
>  		vector = cfg->vector;
> -		per_cpu(vector_irq, cpu)[vector] = irq;
> +		desc = irq_to_desc(irq);
> +		BUG_ON(desc == NULL);
> +		per_cpu(vector_irq, cpu)[vector] = desc;
>  	}
>  	/* Mark the free vectors */
>  	for (vector = 0; vector < NR_VECTORS; ++vector) {
> -		irq = per_cpu(vector_irq, cpu)[vector];
> -		if (irq < 0)
> +		desc = per_cpu(vector_irq, cpu)[vector];
> +		if (desc == NULL)
>  			continue;
>  
> -		cfg = irq_cfg(irq);
> +		cfg = irq_cfg(desc->irq);
>  		if (!cpu_isset(cpu, cfg->domain))
> -			per_cpu(vector_irq, cpu)[vector] = -1;
> +			per_cpu(vector_irq, cpu)[vector] = NULL;
>  	}
>  }
>  
> @@ -2389,16 +2395,14 @@ asmlinkage void smp_irq_move_cleanup_int
>  
>  	me = smp_processor_id();
>  	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
> -		unsigned int irq;
>  		struct irq_desc *desc;
>  		struct irq_cfg *cfg;
> -		irq = __get_cpu_var(vector_irq)[vector];
>  
> -		desc = irq_to_desc(irq);
> -		if (!desc)
> +		desc = __get_cpu_var(vector_irq)[vector];
> +		if (desc == NULL)
>  			continue;
>  
> -		cfg = irq_cfg(irq);
> +		cfg = irq_cfg(desc->irq);
>  		spin_lock(&desc->lock);
>  		if (!cfg->move_cleanup_count)
>  			goto unlock;
> @@ -2406,7 +2410,7 @@ asmlinkage void smp_irq_move_cleanup_int
>  		if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
>  			goto unlock;
>  
> -		__get_cpu_var(vector_irq)[vector] = -1;
> +		__get_cpu_var(vector_irq)[vector] = NULL;
>  		cfg->move_cleanup_count--;
>  unlock:
>  		spin_unlock(&desc->lock);
> Index: linux/include/asm-x86/hw_irq.h
> ===================================================================
> --- linux.orig/include/asm-x86/hw_irq.h	2008-09-19 14:12:25.000000000 -0500
> +++ linux/include/asm-x86/hw_irq.h	2008-09-19 14:13:09.000000000 -0500
> @@ -113,7 +113,7 @@ extern asmlinkage void smp_invalidate_in
>  extern void (*const interrupt[NR_VECTORS])(void);
>  #endif
>  
> -typedef int vector_irq_t[NR_VECTORS];
> +typedef struct irq_desc *vector_irq_t[NR_VECTORS];
>  DECLARE_PER_CPU(vector_irq_t, vector_irq);
>  
>  #ifdef CONFIG_X86_IO_APIC
> Index: linux/arch/x86/kernel/irq_32.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/irq_32.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/kernel/irq_32.c	2008-09-19 14:13:09.000000000 -0500
> @@ -226,26 +226,24 @@ unsigned int do_IRQ(struct pt_regs *regs
>  	int overflow;
>  	unsigned vector = ~regs->orig_ax;
>  	struct irq_desc *desc;
> -	unsigned irq;
>  
>  
>  	old_regs = set_irq_regs(regs);
>  	irq_enter();
> -	irq = __get_cpu_var(vector_irq)[vector];
>  
>  	overflow = check_stack_overflow();
>  
> -	desc = irq_to_desc(irq);
> -	if (unlikely(!desc)) {
> -		printk(KERN_EMERG "%s: cannot handle IRQ %d vector %#x cpu %d\n",
> -					__func__, irq, vector, smp_processor_id());
> +	desc = __get_cpu_var(vector_irq)[vector];
> +	if (unlikely(desc == NULL)) {
> +		printk(KERN_EMERG "%s: cannot handle IRQ vector %#x cpu %d\n",
> +					__func__, vector, smp_processor_id());
>  		BUG();
>  	}
>  
> -	if (!execute_on_irq_stack(overflow, desc, irq)) {
> +	if (!execute_on_irq_stack(overflow, desc, desc->irq)) {
>  		if (unlikely(overflow))
>  			print_stack_overflow();
> -		desc->handle_irq(irq, desc);
> +		desc->handle_irq(desc->irq, desc);
>  	}
>  
>  	irq_exit();
> Index: linux/arch/x86/kernel/irq_64.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/irq_64.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/kernel/irq_64.c	2008-09-19 14:13:09.000000000 -0500
> @@ -213,20 +213,18 @@ asmlinkage unsigned int do_IRQ(struct pt
>  
>  	/* high bit used in ret_from_ code  */
>  	unsigned vector = ~regs->orig_ax;
> -	unsigned irq;
>  
>  	exit_idle();
>  	irq_enter();
> -	irq = __get_cpu_var(vector_irq)[vector];
>  
>  #ifdef CONFIG_DEBUG_STACKOVERFLOW
>  	stack_overflow_check(regs);
>  #endif
>  
> -	desc = irq_to_desc(irq);
> -	if (likely(desc))
> -		generic_handle_irq_desc(irq, desc);
> -	else {
> +	desc = __get_cpu_var(vector_irq)[vector];
> +	if (likely(desc != NULL)) {
> +		generic_handle_irq_desc(desc->irq, desc);
> +	} else {
>  		if (!disable_apic)
>  			ack_APIC_irq();
>  
> Index: linux/arch/x86/kernel/vmiclock_32.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/vmiclock_32.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/kernel/vmiclock_32.c	2008-09-19 14:13:09.000000000 -0500
> @@ -242,7 +242,7 @@ void __init vmi_time_init(void)
>  	vmi_time_init_clockevent();
>  	setup_irq(0, &vmi_clock_action);
>  	for_each_possible_cpu(cpu)
> -		per_cpu(vector_irq, cpu)[vmi_get_timer_vector()] = 0;
> +		per_cpu(vector_irq, cpu)[vmi_get_timer_vector()] = NULL;
>  }
>  
>  #ifdef CONFIG_X86_LOCAL_APIC
> Index: linux/arch/x86/xen/irq.c
> ===================================================================
> --- linux.orig/arch/x86/xen/irq.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/xen/irq.c	2008-09-19 14:13:09.000000000 -0500
> @@ -27,8 +27,9 @@ static void __init __xen_init_IRQ(void)
>  	for(i = 0; i < NR_VECTORS; i++) {
>  		int cpu;
>  
> +		desc = irq_to_desc_alloc(i);
>  		for_each_possible_cpu(cpu)
> -			per_cpu(vector_irq, cpu)[i] = i;
> +			per_cpu(vector_irq, cpu)[i] = desc;
>  	}
>  
>  	xen_init_IRQ();
> Index: linux/arch/x86/kernel/irqinit_32.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/irqinit_32.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/kernel/irqinit_32.c	2008-09-19 14:13:09.000000000 -0500
> @@ -59,6 +59,8 @@ static struct irqaction fpu_irq = {
>  void __init init_ISA_irqs (void)
>  {
>  	int i;
> +	int cpu;
> +	unsigned int vector;
>  
>  #ifdef CONFIG_X86_LOCAL_APIC
>  	init_bsp_APIC();
> @@ -76,6 +78,12 @@ void __init init_ISA_irqs (void)
>  		desc->action = NULL;
>  		desc->depth = 1;
>  
> +		vector = IRQ0_VECTOR + i;
> +		for_each_possible_cpu(cpu) {
> +			BUG_ON(per_cpu(vector_irq, cpu)[vector] != NULL);
> +			per_cpu(vector_irq, cpu)[vector] = desc;
> +		}
> +
>  		set_irq_chip_and_handler_name(i, &i8259A_chip,
>  					      handle_level_irq, "XT");
>  	}
> @@ -90,26 +98,7 @@ static struct irqaction irq2 = {
>  	.name = "cascade",
>  };
>  
> -DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
> -	[0 ... IRQ0_VECTOR - 1] = -1,
> -	[IRQ0_VECTOR] = 0,
> -	[IRQ1_VECTOR] = 1,
> -	[IRQ2_VECTOR] = 2,
> -	[IRQ3_VECTOR] = 3,
> -	[IRQ4_VECTOR] = 4,
> -	[IRQ5_VECTOR] = 5,
> -	[IRQ6_VECTOR] = 6,
> -	[IRQ7_VECTOR] = 7,
> -	[IRQ8_VECTOR] = 8,
> -	[IRQ9_VECTOR] = 9,
> -	[IRQ10_VECTOR] = 10,
> -	[IRQ11_VECTOR] = 11,
> -	[IRQ12_VECTOR] = 12,
> -	[IRQ13_VECTOR] = 13,
> -	[IRQ14_VECTOR] = 14,
> -	[IRQ15_VECTOR] = 15,
> -	[IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
> -};
> +DEFINE_PER_CPU(vector_irq_t, vector_irq);
>  
>  /* Overridden in paravirt.c */
>  void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
> Index: linux/arch/x86/kernel/irqinit_64.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/irqinit_64.c	2008-09-19 14:12:25.000000000 -0500
> +++ linux/arch/x86/kernel/irqinit_64.c	2008-09-19 14:13:09.000000000 -0500
> @@ -114,30 +114,13 @@ static struct irqaction irq2 = {
>  	.mask = CPU_MASK_NONE,
>  	.name = "cascade",
>  };
> -DEFINE_PER_CPU(vector_irq_t, vector_irq) = {
> -	[0 ... IRQ0_VECTOR - 1] = -1,
> -	[IRQ0_VECTOR] = 0,
> -	[IRQ1_VECTOR] = 1,
> -	[IRQ2_VECTOR] = 2,
> -	[IRQ3_VECTOR] = 3,
> -	[IRQ4_VECTOR] = 4,
> -	[IRQ5_VECTOR] = 5,
> -	[IRQ6_VECTOR] = 6,
> -	[IRQ7_VECTOR] = 7,
> -	[IRQ8_VECTOR] = 8,
> -	[IRQ9_VECTOR] = 9,
> -	[IRQ10_VECTOR] = 10,
> -	[IRQ11_VECTOR] = 11,
> -	[IRQ12_VECTOR] = 12,
> -	[IRQ13_VECTOR] = 13,
> -	[IRQ14_VECTOR] = 14,
> -	[IRQ15_VECTOR] = 15,
> -	[IRQ15_VECTOR + 1 ... NR_VECTORS - 1] = -1
> -};
> +DEFINE_PER_CPU(vector_irq_t, vector_irq);
>  
>  static void __init init_ISA_irqs (void)
>  {
>  	int i;
> +	int cpu;
> +	unsigned int vector;
>  
>  	init_bsp_APIC();
>  	init_8259A(0);
> @@ -150,6 +133,12 @@ static void __init init_ISA_irqs (void)
>  		desc->action = NULL;
>  		desc->depth = 1;
>  
> +		vector = IRQ0_VECTOR + i;
> +		for_each_possible_cpu(cpu) {
> +			BUG_ON(per_cpu(vector_irq, cpu)[vector] != NULL);
> +			per_cpu(vector_irq, cpu)[vector] = desc;
> +		}
> +
>  		/*
>  		 * 16 old-style INTA-cycle interrupts:
>  		 */
> --
> 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 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Yinghai Lu]

On Fri, Sep 19, 2008 at 1:04 PM, Dean Nelson <dcn@sgi.com> wrote:
> Change per_cpu variable vector_irq[] from holding an 'int' irq number to
> holding a 'struct irq_desc' pointer.

why?

Eric wants to change int irq to some struct irq later?

YH

Re: [PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Dean Nelson]

On Fri, Sep 19, 2008 at 01:24:55PM -0700, Mike Travis wrote:
> 
> Note that my recent patch from a couple of weeks ago, stripped almost all of
> the cpumask_t's from the io_apic functions.  Any chance we can apply them
> before we add more?
> 
> I'd be willing to merge in Dean's patch over the top of mine (or vice-versa).

Sorry Mike, I didn't realize you had any patches against io_apic functions.
You might take a look at the second patch in this patchset, I tried to
eliminate some of the cpumask_t's from being defined on the stack and to
pass only pointers to them as args. I didn't do all that could be done,
but just the ones near the main functions I was touching. You may have a
better way of doing it then I did.

Feel free to do whatever with this patchset. (I'm not going to be
reachable until Monday.)

FYI, this patchset was built against linux-next.

Thanks,
Dean

Re: [PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Dean Nelson]

On Fri, Sep 19, 2008 at 01:40:23PM -0700, Yinghai Lu wrote:
> On Fri, Sep 19, 2008 at 1:04 PM, Dean Nelson <dcn@sgi.com> wrote:
> > Change per_cpu variable vector_irq[] from holding an 'int' irq number to
> > holding a 'struct irq_desc' pointer.
> 
> why?
> 
> Eric wants to change int irq to some struct irq later?

Well, it was my take on what Eric was saying in the following:

On Mon, Aug 11, 2008 at 12:39:22PM -0700, Eric W. Biederman wrote:
>
> We create a common factor of assign_irq_vector that looks something like:
> 
> bool __grab_irq_vector(struct irq_desc *desc, unsigned vector, cpumask_t new_domain)
> {
>         /* Must be called with vector lock */
>         struct irq_cfg *cfg;
>         bool grabbed = false;
>         unsigned int old_vector;
>         cpumask_t mask;
>         int cpu;
> 
>         cfg = get_irqp_cfg(irq);
>         old_vector = cfg->vector;
>         cpus_and(mask, new_domain, cpu_online_map);
> 
>         for_each_cpu_mask_nr(cpu, mask) {
>               if (per_cpu(vector_irq, cpu)[vector])
>                       goto out;
>         }
>         /* Available reserve it */
>         for_each_cpu_mask_nr(cpu, mask)
>             per_cpu(vector_irq, cpu)[vector] = desc;

The previous line made me think that was what he wanted....

>         if (cfg->vector) {
>               cfg->move_in_progress;
>                 cfg->old_domain = cfg->domain;
>         }
>         cfg->vector = vector;
>         cfg->domain = mask;
>         grabbed = true;
>         
> out:
>         return grabbed;
> }

... Along with the following paragraph.

> I think vector_irq should return an irq_desc and have an entry for
> all of the static vectors as well (if we are going to do weird
> things with dynamic high priority vector allocation, and dynamic
> detection of which vectors assign_irq_vector may use).

If that's not what Eric meant, then I got it wrong and you can reject
the patchset.

Dean

Re: [PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Ingo Molnar]

* Mike Travis <travis@sgi.com> wrote:

> Hi Ingo,
> 
> Note that my recent patch from a couple of weeks ago, stripped almost 
> all of the cpumask_t's from the io_apic functions.  Any chance we can 
> apply them before we add more?
> 
> I'd be willing to merge in Dean's patch over the top of mine (or 
> vice-versa).

i'm really looking forwards to Rusty's 'get-rid-of-cpumask_t use' 
patchset, which would solve such cpumask_t proliferation issues once and 
for all.

	Ingo

Re: [PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Mike Travis]

Ingo Molnar wrote:
> * Mike Travis <travis@sgi.com> wrote:
> 
>> Hi Ingo,
>>
>> Note that my recent patch from a couple of weeks ago, stripped almost 
>> all of the cpumask_t's from the io_apic functions.  Any chance we can 
>> apply them before we add more?
>>
>> I'd be willing to merge in Dean's patch over the top of mine (or 
>> vice-versa).
> 
> i'm really looking forwards to Rusty's 'get-rid-of-cpumask_t use' 
> patchset, which would solve such cpumask_t proliferation issues once and 
> for all.
> 
> 	Ingo

Hi Ingo,

The net effect of the "getting rid of cpumask_t's" will be the same.  This
patch reduces considerably the amount of stack space being required right
now.  The new cpumask_t approach will take quite a bit of time to get it
"up and running".  (It took me about a day to get init/main.c and kernel/sched.c
to just compile... it now needs the temp cpumasks to actually function.  And
there are a number of areas that will need to be altered before functional
testing can start.)

Thanks,
Mike

Re: [PATCH 1/3] switch vector_irq[] from irq number to irq_desc pointer v2

[Mike Travis]

Mike Travis wrote:
> Ingo Molnar wrote:
>> * Mike Travis <travis@sgi.com> wrote:
>>
>>> Hi Ingo,
>>>
>>> Note that my recent patch from a couple of weeks ago, stripped almost 
>>> all of the cpumask_t's from the io_apic functions.  Any chance we can 
>>> apply them before we add more?
>>>
>>> I'd be willing to merge in Dean's patch over the top of mine (or 
>>> vice-versa).
>> i'm really looking forwards to Rusty's 'get-rid-of-cpumask_t use' 
>> patchset, which would solve such cpumask_t proliferation issues once and 
>> for all.
>>
>> 	Ingo
> 
> Hi Ingo,
> 
> The net effect of the "getting rid of cpumask_t's" will be the same.  This
> patch reduces considerably the amount of stack space being required right
> now.  

Btw, my patch will still be useful even after cpumask's are passed by reference
in that it optimizes the cpumask's that are being used.  For example, in one
function 3 cpumask_t's become 1.  In some others, 2 were reduced to 1.  Since
the new cpumask approach will require temp cpumask's, this cuts down on the
number of temp cpumask's required.

Thanks,
Mike