[RFC 2/4] introduce dynamically allocated system vectors

From: Dean Nelson
Date: Thu Sep 11 2008 - 11:27:44 EST

Introduce the dynamic allocation and deallocation of system vectors which
are mapped to irq numbers allowing the use of request_irq()/free_irq().

Signed-off-by: Dean Nelson <dcn@xxxxxxx>

---

arch/x86/kernel/apic.c | 3
arch/x86/kernel/io_apic.c | 264 +++++++++++++++++++++++++++++++++-----
arch/x86/kernel/irqinit_64.c | 4
include/asm-x86/desc.h | 13 +
include/asm-x86/irq_vectors.h | 1
include/linux/irq.h | 13 +
6 files changed, 258 insertions(+), 40 deletions(-)

Index: linux/arch/x86/kernel/io_apic.c
===================================================================
--- linux.orig/arch/x86/kernel/io_apic.c 2008-09-10 12:08:46.000000000 -0500
+++ linux/arch/x86/kernel/io_apic.c 2008-09-11 07:17:33.000000000 -0500
@@ -1205,7 +1205,34 @@ void unlock_vector_lock(void)
spin_unlock(&vector_lock);
}

-static int __assign_irq_vector(int irq, cpumask_t mask)
+bool __grab_irq_vector(struct irq_desc *desc, unsigned int vector,
+ cpumask_t *new_domain_mask)
+{
+ /* Must be called with vector lock */
+ struct irq_cfg *cfg;
+ int cpu;
+
+ for_each_cpu_mask_nr(cpu, *new_domain_mask) {
+ if (per_cpu(vector_irq, cpu)[vector] != NULL)
+ return false;
+ }
+
+ /* Available reserve it */
+ for_each_cpu_mask_nr(cpu, *new_domain_mask)
+ per_cpu(vector_irq, cpu)[vector] = desc;
+
+ cfg = irq_cfg(desc->irq);
+ if (cfg->vector) {
+ cfg->move_in_progress = 1;
+ cfg->old_domain = cfg->domain;
+ }
+ cfg->vector = vector;
+ cfg->domain = *new_domain_mask;
+
+ return true;
+}
+
+static int __assign_irq_vector(int irq, cpumask_t *mask)
{
/*
* NOTE! The local APIC isn't very good at handling
@@ -1219,42 +1246,40 @@ 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;
+ cpumask_t target_cpus_mask;
int cpu;
struct irq_cfg *cfg;
struct irq_desc *desc;

cfg = irq_cfg(irq);

- /* Only try and allocate irqs on cpus that are present */
- cpus_and(mask, mask, cpu_online_map);
-
if ((cfg->move_in_progress) || cfg->move_cleanup_count)
return -EBUSY;

- old_vector = cfg->vector;
- if (old_vector) {
+ /* Only try and allocate irqs on cpus that are present */
+ cpus_and(target_cpus_mask, *mask, cpu_online_map);
+
+ if (cfg->vector) {
cpumask_t tmp;
- cpus_and(tmp, cfg->domain, mask);
+ cpus_and(tmp, cfg->domain, target_cpus_mask);
if (!cpus_empty(tmp))
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, target_cpus_mask) {
+ cpumask_t domain, new_domain_mask;
int vector, offset;

domain = vector_allocation_domain(cpu);
- cpus_and(new_mask, domain, cpu_online_map);
+ cpus_and(new_domain_mask, domain, cpu_online_map);

vector = current_vector;
offset = current_offset;
next:
vector += 8;
- if (vector >= first_system_vector) {
+ if (vector > last_device_vector) {
/* If we run out of vectors on large boxen, must share them. */
offset = (offset + 1) % 8;
vector = FIRST_DEVICE_VECTOR + offset;
@@ -1268,20 +1293,12 @@ 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;
+ if (!__grab_irq_vector(desc, vector, &new_domain_mask))
+ 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;
@@ -1293,11 +1310,51 @@ static int assign_irq_vector(int irq, cp
unsigned long flags;

spin_lock_irqsave(&vector_lock, flags);
- err = __assign_irq_vector(irq, mask);
+ err = __assign_irq_vector(irq, &mask);
spin_unlock_irqrestore(&vector_lock, flags);
return err;
}

+static int __assign_irq_system_vector(int irq, cpumask_t *mask, int priority)
+{
+ int vector;
+ cpumask_t target_cpus_mask;
+ int cpu;
+ cpumask_t domain;
+ cpumask_t new_domain_mask = CPU_MASK_NONE;
+ struct irq_desc *desc;
+
+ if (priority == IRQ_PRIORITY_HIGH)
+ vector = first_static_system_vector;
+ else if (priority == IRQ_PRIORITY_LOW)
+ vector = FIRST_DEVICE_VECTOR - 1;
+ else
+ BUG();
+
+ cpus_and(target_cpus_mask, *mask, cpu_possible_map);
+ for_each_cpu_mask_nr(cpu, target_cpus_mask) {
+ domain = vector_allocation_domain(cpu);
+ cpus_and(domain, domain, cpu_possible_map);
+ cpus_or(new_domain_mask, new_domain_mask, domain);
+ }
+
+ desc = irq_to_desc_alloc(irq);
+
+ do {
+ if (priority == IRQ_PRIORITY_HIGH) {
+ if (--vector < FIRST_DEVICE_VECTOR)
+ return -ENOSPC;
+ } else { /* IRQ_PRIORITY_LOW */
+ if (++vector == first_static_system_vector)
+ return -ENOSPC;
+ }
+
+ } while (!__grab_irq_vector(desc, vector, &new_domain_mask));
+
+ /* found one */
+ return 0;
+}
+
static void __clear_irq_vector(int irq)
{
struct irq_cfg *cfg;
@@ -3045,21 +3102,22 @@ static int __init ioapic_init_sysfs(void

device_initcall(ioapic_init_sysfs);

-/*
- * Dynamic irq allocate and deallocation
- */
-unsigned int create_irq_nr(unsigned int irq_want)
+#define DEVICE_VECTOR 1
+#define SYSTEM_VECTOR 2
+
+static unsigned int __create_irq_nr(int vector_type, unsigned int irq_want,
+ cpumask_t *mask, int priority)
{
/* Allocate an unused irq */
unsigned int irq;
unsigned int new;
unsigned long flags;
struct irq_cfg *cfg_new;
+ int ret;

#ifndef CONFIG_HAVE_SPARSE_IRQ
irq_want = nr_irqs - 1;
#endif
-
irq = 0;
spin_lock_irqsave(&vector_lock, flags);
for (new = irq_want; new > 0; new--) {
@@ -3071,18 +3129,34 @@ 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 (vector_type == DEVICE_VECTOR)
+ ret = __assign_irq_vector(new, mask);
+ else
+ ret = __assign_irq_system_vector(new, mask, priority);
+
+ if (ret == 0)
irq = new;
break;
}
spin_unlock_irqrestore(&vector_lock, flags);

- if (irq > 0) {
+ if (irq > 0)
dynamic_irq_init(irq);
- }
+
return irq;
}

+unsigned int create_irq_nr(unsigned int irq_want)
+{
+ cpumask_t mask = TARGET_CPUS;
+
+ return __create_irq_nr(DEVICE_VECTOR, irq_want, &mask,
+ IRQ_PRIORITY_LOW);
+}
+
+/*
+ * Dynamic irq device vector allocation.
+ */
int create_irq(void)
{
int irq;
@@ -3095,6 +3169,9 @@ int create_irq(void)
return irq;
}

+/*
+ * Dynamic irq device vector deallocation.
+ */
void destroy_irq(unsigned int irq)
{
unsigned long flags;
@@ -3109,6 +3186,127 @@ void destroy_irq(unsigned int irq)
spin_unlock_irqrestore(&vector_lock, flags);
}

+static void noop(unsigned int irq)
+{
+}
+
+static unsigned int noop_ret(unsigned int irq)
+{
+ return 0;
+}
+
+static void ack_apic(unsigned int irq)
+{
+ ack_APIC_irq();
+}
+
+static struct irq_chip ack_apic_chip = {
+ .name = "ack_apic",
+ .startup = noop_ret,
+ .shutdown = noop,
+ .enable = noop,
+ .disable = noop,
+ .ack = noop,
+ .mask = noop,
+ .unmask = noop,
+ .eoi = ack_apic,
+ .end = noop,
+};
+
+unsigned int create_irq_system_vector_nr(unsigned int irq_want, cpumask_t *mask,
+ int priority)
+{
+ return __create_irq_nr(SYSTEM_VECTOR, irq_want, mask, priority);
+}
+
+/*
+ * Dynamic irq system vector allocation.
+ */
+unsigned int create_irq_system_vector(cpumask_t *mask, int priority,
+ char *irq_name, int *assigned_vector)
+{
+ unsigned long flags;
+ struct irq_cfg *cfg;
+ int irq;
+
+ /* allocate an available irq and vector mapping */
+ irq = create_irq_system_vector_nr(nr_irqs - 1, mask, priority);
+ if (irq == 0)
+ return -1;
+
+ spin_lock_irqsave(&vector_lock, flags);
+ set_irq_chip_and_handler_name(irq, &ack_apic_chip, handle_percpu_irq,
+ irq_name);
+ spin_unlock_irqrestore(&vector_lock, flags);
+
+ cfg = irq_cfg(irq);
+ *assigned_vector = cfg->vector;
+ return irq;
+}
+EXPORT_SYMBOL(create_irq_system_vector);
+
+/*
+ * Dynamic irq system vector deallocation.
+ */
+void destroy_irq_system_vector(unsigned int irq)
+{
+ unsigned long flags;
+ struct irq_cfg *cfg;
+ int cpu;
+
+ if (irq >= nr_irqs)
+ return;
+ cfg = irq_cfg(irq);
+ if (cfg->vector == 0)
+ return;
+
+#ifdef CONFIG_SMP
+ synchronize_irq(irq);
+#endif
+ dynamic_irq_cleanup(irq);
+ disable_irq(irq);
+
+ spin_lock_irqsave(&vector_lock, flags);
+
+ for_each_cpu_mask_nr(cpu, cfg->domain)
+ per_cpu(vector_irq, cpu)[cfg->vector] = NULL;
+
+ cfg->vector = 0;
+ cpus_clear(cfg->domain);
+
+ spin_unlock_irqrestore(&vector_lock, flags);
+}
+EXPORT_SYMBOL(destroy_irq_system_vector);
+
+int reserve_system_vectors(int number)
+{
+ unsigned long flags;
+ int new_last_device_vector;
+ int vector;
+ int cpu;
+ int ret = -EBUSY;
+
+ spin_lock_irqsave(&vector_lock, flags);
+
+ new_last_device_vector = last_device_vector - number;
+ if (new_last_device_vector < MIN_LAST_DEVICE_VECTOR)
+ goto out;
+
+ for (vector = last_device_vector; vector > new_last_device_vector;
+ vector--) {
+ for_each_cpu_mask_nr(cpu, cpu_possible_map) {
+ if (per_cpu(vector_irq, cpu)[vector] != NULL)
+ goto out;
+ }
+ }
+
+ last_device_vector = new_last_device_vector;
+ ret = 0;
+out:
+ spin_unlock_irqrestore(&vector_lock, flags);
+ return ret;
+}
+
/*
* MSI message composition
*/
Index: linux/include/linux/irq.h
===================================================================
--- linux.orig/include/linux/irq.h 2008-09-10 12:08:46.000000000 -0500
+++ linux/include/linux/irq.h 2008-09-11 06:53:16.000000000 -0500
@@ -390,11 +390,22 @@ set_irq_chained_handler(unsigned int irq
extern void set_irq_noprobe(unsigned int irq);
extern void set_irq_probe(unsigned int irq);

-/* Handle dynamic irq creation and destruction */
+/* Handle dynamic irq device vector allocation and deallocation */
extern unsigned int create_irq_nr(unsigned int irq_want);
extern int create_irq(void);
extern void destroy_irq(unsigned int irq);

+/* Handle dynamic irq system vector allocation and deallocation */
+extern unsigned int create_irq_system_vector(cpumask_t *mask, int priority,
+ char *irq_name,
+ int *assigned_vector);
+#define IRQ_PRIORITY_LOW 1
+#define IRQ_PRIORITY_HIGH 2
+
+extern void destroy_irq_system_vector(unsigned int irq);
+
+extern int reserve_system_vectors(int number);
+
/* Test to see if a driver has successfully requested an irq */
static inline int irq_has_action(unsigned int irq)
{
Index: linux/arch/x86/kernel/apic.c
===================================================================
--- linux.orig/arch/x86/kernel/apic.c 2008-09-10 12:08:46.000000000 -0500
+++ linux/arch/x86/kernel/apic.c 2008-09-11 06:42:34.000000000 -0500
@@ -116,7 +116,8 @@ static int disable_apic_timer __cpuinitd
int local_apic_timer_c2_ok;
EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);

-int first_system_vector = 0xfe;
+int first_static_system_vector = 0xfe;
+int last_device_vector = 0xfd;

char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE};

Index: linux/include/asm-x86/desc.h
===================================================================
--- linux.orig/include/asm-x86/desc.h 2008-09-10 12:08:46.000000000 -0500
+++ linux/include/asm-x86/desc.h 2008-09-11 06:42:34.000000000 -0500
@@ -323,22 +323,25 @@ static inline void set_intr_gate(unsigne
#define SYS_VECTOR_FREE 0
#define SYS_VECTOR_ALLOCED 1

-extern int first_system_vector;
+extern int first_static_system_vector;
+extern int last_device_vector;
extern char system_vectors[];

-static inline void alloc_system_vector(int vector)
+static inline void alloc_static_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;
+ if (first_static_system_vector > vector)
+ first_static_system_vector = vector;
+ if (last_device_vector > vector - 1)
+ last_device_vector = vector - 1;
} else
BUG();
}

static inline void alloc_intr_gate(unsigned int n, void *addr)
{
- alloc_system_vector(n);
+ alloc_static_system_vector(n);
set_intr_gate(n, addr);
}

Index: linux/include/asm-x86/irq_vectors.h
===================================================================
--- linux.orig/include/asm-x86/irq_vectors.h 2008-09-05 08:38:48.000000000 -0500
+++ linux/include/asm-x86/irq_vectors.h 2008-09-11 07:14:54.000000000 -0500
@@ -92,6 +92,7 @@
* levels. (0x80 is the syscall vector)
*/
#define FIRST_DEVICE_VECTOR (IRQ15_VECTOR + 2)
+#define MIN_LAST_DEVICE_VECTOR (LOCAL_TIMER_VECTOR - 16)

#define NR_VECTORS 256

Index: linux/arch/x86/kernel/irqinit_64.c
===================================================================
--- linux.orig/arch/x86/kernel/irqinit_64.c 2008-09-09 12:57:13.000000000 -0500
+++ linux/arch/x86/kernel/irqinit_64.c 2008-09-11 07:21:41.000000000 -0500
@@ -22,6 +22,7 @@
#include <asm/desc.h>
#include <asm/apic.h>
#include <asm/i8259.h>
+#include <asm/genapic.h>

/*
* Common place to define all x86 IRQ vectors
@@ -202,6 +203,9 @@ void __init native_init_IRQ(void)
alloc_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt);

+ if (is_uv_system())
+ reserve_system_vectors(8);
+
if (!acpi_ioapic)
setup_irq(2, &irq2);
}
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/