[PATCH] x86_64 irq: Allocate a vector across all cpus for genapic_flat.
The problem we can't take advantage of lowest priority delivery mode if
the vectors are allocated for only one cpu at a time. Nor can we work
around hardware that assumes lowest priority delivery mode is always
used with several cpus.
So this patch introduces the concept of a vector_allocation_domain. A
set of cpus that will receive an irq on the same vector. Currently the
code for implementing this is placed in the genapic structure so we can
vary this depending on how we are using the io_apics.
This allows us to restore the previous behaviour of genapic_flat without
removing the benefits of having separate vector allocation for large
machines.
This should also fix the problem report where a hyperthreaded cpu was
receving the irq on the wrong hyperthread when in logical delivery mode
because the previous behaviour is restored.
This patch properly records our allocation of the first 16 irqs to the
first 16 available vectors on all cpus. This should be fine but it may
run into problems with multiple interrupts at the same interrupt level.
Except for some badly maintained comments in the code and the behaviour
of the interrupt allocator I have no real understanding of that problem.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
diff --git a/arch/x86_64/kernel/genapic_cluster.c b/arch/x86_64/kernel/genapic_cluster.c
index cdb90e6..73d7630 100644
--- a/arch/x86_64/kernel/genapic_cluster.c
+++ b/arch/x86_64/kernel/genapic_cluster.c
@@ -63,6 +63,13 @@
return cpumask_of_cpu(0);
}
+static cpumask_t cluster_vector_allocation_domain(int cpu)
+{
+ cpumask_t domain = CPU_MASK_NONE;
+ cpu_set(cpu, domain);
+ return domain;
+}
+
static void cluster_send_IPI_mask(cpumask_t mask, int vector)
{
send_IPI_mask_sequence(mask, vector);
@@ -119,6 +126,7 @@
.int_delivery_mode = dest_Fixed,
.int_dest_mode = (APIC_DEST_PHYSICAL != 0),
.target_cpus = cluster_target_cpus,
+ .vector_allocation_domain = cluster_vector_allocation_domain,
.apic_id_registered = cluster_apic_id_registered,
.init_apic_ldr = cluster_init_apic_ldr,
.send_IPI_all = cluster_send_IPI_all,
diff --git a/arch/x86_64/kernel/genapic_flat.c b/arch/x86_64/kernel/genapic_flat.c
index 50ad153..0dfc223 100644
--- a/arch/x86_64/kernel/genapic_flat.c
+++ b/arch/x86_64/kernel/genapic_flat.c
@@ -22,6 +22,20 @@
return cpu_online_map;
}
+static cpumask_t flat_vector_allocation_domain(int cpu)
+{
+ /* Careful. Some cpus do not strictly honor the set of cpus
+ * specified in the interrupt destination when using lowest
+ * priority interrupt delivery mode.
+ *
+ * In particular there was a hyperthreading cpu observed to
+ * deliver interrupts to the wrong hyperthread when only one
+ * hyperthread was specified in the interrupt desitination.
+ */
+ cpumask_t domain = { { [0] = APIC_ALL_CPUS, } };
+ return domain;
+}
+
/*
* Set up the logical destination ID.
*
@@ -121,6 +135,7 @@
.int_delivery_mode = dest_LowestPrio,
.int_dest_mode = (APIC_DEST_LOGICAL != 0),
.target_cpus = flat_target_cpus,
+ .vector_allocation_domain = flat_vector_allocation_domain,
.apic_id_registered = flat_apic_id_registered,
.init_apic_ldr = flat_init_apic_ldr,
.send_IPI_all = flat_send_IPI_all,
@@ -141,6 +156,14 @@
return cpumask_of_cpu(0);
}
+static cpumask_t physflat_vector_allocation_domain(int cpu)
+{
+ cpumask_t domain = CPU_MASK_NONE;
+ cpu_set(cpu, domain);
+ return domain;
+}
+
+
static void physflat_send_IPI_mask(cpumask_t cpumask, int vector)
{
send_IPI_mask_sequence(cpumask, vector);
@@ -179,6 +202,7 @@
.int_delivery_mode = dest_Fixed,
.int_dest_mode = (APIC_DEST_PHYSICAL != 0),
.target_cpus = physflat_target_cpus,
+ .vector_allocation_domain = physflat_vector_allocation_domain,
.apic_id_registered = flat_apic_id_registered,
.init_apic_ldr = flat_init_apic_ldr,/*not needed, but shouldn't hurt*/
.send_IPI_all = physflat_send_IPI_all,
diff --git a/arch/x86_64/kernel/io_apic.c b/arch/x86_64/kernel/io_apic.c
index 9c3b9b1..771bcf7 100644
--- a/arch/x86_64/kernel/io_apic.c
+++ b/arch/x86_64/kernel/io_apic.c
@@ -47,7 +47,7 @@
#include <asm/msidef.h>
#include <asm/hypertransport.h>
-static int assign_irq_vector(int irq, cpumask_t mask);
+static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result);
#define __apicdebuginit __init
@@ -174,12 +174,10 @@
cpus_and(mask, tmp, CPU_MASK_ALL);
- vector = assign_irq_vector(irq, mask);
+ vector = assign_irq_vector(irq, mask, &tmp);
if (vector < 0)
return;
- cpus_clear(tmp);
- cpu_set(vector >> 8, tmp);
dest = cpu_mask_to_apicid(tmp);
/*
@@ -188,7 +186,7 @@
dest = SET_APIC_LOGICAL_ID(dest);
spin_lock_irqsave(&ioapic_lock, flags);
- __target_IO_APIC_irq(irq, dest, vector & 0xff);
+ __target_IO_APIC_irq(irq, dest, vector);
set_native_irq_info(irq, mask);
spin_unlock_irqrestore(&ioapic_lock, flags);
}
@@ -563,9 +561,45 @@
}
/* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
-unsigned int irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_EXTERNAL_VECTOR, 0 };
+static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = {
+ [0] = FIRST_EXTERNAL_VECTOR + 0,
+ [1] = FIRST_EXTERNAL_VECTOR + 1,
+ [2] = FIRST_EXTERNAL_VECTOR + 2,
+ [3] = FIRST_EXTERNAL_VECTOR + 3,
+ [4] = FIRST_EXTERNAL_VECTOR + 4,
+ [5] = FIRST_EXTERNAL_VECTOR + 5,
+ [6] = FIRST_EXTERNAL_VECTOR + 6,
+ [7] = FIRST_EXTERNAL_VECTOR + 7,
+ [8] = FIRST_EXTERNAL_VECTOR + 8,
+ [9] = FIRST_EXTERNAL_VECTOR + 9,
+ [10] = FIRST_EXTERNAL_VECTOR + 10,
+ [11] = FIRST_EXTERNAL_VECTOR + 11,
+ [12] = FIRST_EXTERNAL_VECTOR + 12,
+ [13] = FIRST_EXTERNAL_VECTOR + 13,
+ [14] = FIRST_EXTERNAL_VECTOR + 14,
+ [15] = FIRST_EXTERNAL_VECTOR + 15,
+};
-static int __assign_irq_vector(int irq, cpumask_t mask)
+static cpumask_t irq_domain[NR_IRQ_VECTORS] __read_mostly = {
+ [0] = CPU_MASK_ALL,
+ [1] = CPU_MASK_ALL,
+ [2] = CPU_MASK_ALL,
+ [3] = CPU_MASK_ALL,
+ [4] = CPU_MASK_ALL,
+ [5] = CPU_MASK_ALL,
+ [6] = CPU_MASK_ALL,
+ [7] = CPU_MASK_ALL,
+ [8] = CPU_MASK_ALL,
+ [9] = CPU_MASK_ALL,
+ [10] = CPU_MASK_ALL,
+ [11] = CPU_MASK_ALL,
+ [12] = CPU_MASK_ALL,
+ [13] = CPU_MASK_ALL,
+ [14] = CPU_MASK_ALL,
+ [15] = CPU_MASK_ALL,
+};
+
+static int __assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
{
/*
* NOTE! The local APIC isn't very good at handling
@@ -589,14 +623,22 @@
if (irq_vector[irq] > 0)
old_vector = irq_vector[irq];
- if ((old_vector > 0) && cpu_isset(old_vector >> 8, mask)) {
- return old_vector;
+ if (old_vector > 0) {
+ cpus_and(*result, irq_domain[irq], mask);
+ if (!cpus_empty(*result))
+ return old_vector;
}
for_each_cpu_mask(cpu, mask) {
+ cpumask_t domain;
+ int first, new_cpu;
int vector, offset;
- vector = pos[cpu].vector;
- offset = pos[cpu].offset;
+
+ domain = vector_allocation_domain(cpu);
+ first = first_cpu(domain);
+
+ vector = pos[first].vector;
+ offset = pos[first].offset;
next:
vector += 8;
if (vector >= FIRST_SYSTEM_VECTOR) {
@@ -604,35 +646,40 @@
offset = (offset + 1) % 8;
vector = FIRST_DEVICE_VECTOR + offset;
}
- if (unlikely(pos[cpu].vector == vector))
+ if (unlikely(pos[first].vector == vector))
continue;
if (vector == IA32_SYSCALL_VECTOR)
goto next;
- if (per_cpu(vector_irq, cpu)[vector] != -1)
- goto next;
+ for_each_cpu_mask(new_cpu, domain)
+ if (per_cpu(vector_irq, cpu)[vector] != -1)
+ goto next;
/* Found one! */
- pos[cpu].vector = vector;
- pos[cpu].offset = offset;
- if (old_vector >= 0) {
- int old_cpu = old_vector >> 8;
- old_vector &= 0xff;
- per_cpu(vector_irq, old_cpu)[old_vector] = -1;
+ for_each_cpu_mask(new_cpu, domain) {
+ pos[cpu].vector = vector;
+ pos[cpu].offset = offset;
}
- per_cpu(vector_irq, cpu)[vector] = irq;
- vector |= cpu << 8;
+ if (old_vector >= 0) {
+ int old_cpu;
+ for_each_cpu_mask(old_cpu, domain)
+ per_cpu(vector_irq, old_cpu)[old_vector] = -1;
+ }
+ for_each_cpu_mask(new_cpu, domain)
+ per_cpu(vector_irq, new_cpu)[vector] = irq;
irq_vector[irq] = vector;
+ irq_domain[irq] = domain;
+ cpus_and(*result, domain, mask);
return vector;
}
return -ENOSPC;
}
-static int assign_irq_vector(int irq, cpumask_t mask)
+static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
{
int vector;
unsigned long flags;
spin_lock_irqsave(&vector_lock, flags);
- vector = __assign_irq_vector(irq, mask);
+ vector = __assign_irq_vector(irq, mask, result);
spin_unlock_irqrestore(&vector_lock, flags);
return vector;
}
@@ -704,14 +751,12 @@
if (IO_APIC_IRQ(irq)) {
cpumask_t mask;
- vector = assign_irq_vector(irq, TARGET_CPUS);
+ vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
if (vector < 0)
continue;
- cpus_clear(mask);
- cpu_set(vector >> 8, mask);
entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
- entry.vector = vector & 0xff;
+ entry.vector = vector;
ioapic_register_intr(irq, vector, IOAPIC_AUTO);
if (!apic && (irq < 16))
@@ -1430,12 +1475,13 @@
{
int apic1, pin1, apic2, pin2;
int vector;
+ cpumask_t mask;
/*
* get/set the timer IRQ vector:
*/
disable_8259A_irq(0);
- vector = assign_irq_vector(0, TARGET_CPUS);
+ vector = assign_irq_vector(0, TARGET_CPUS, &mask);
/*
* Subtle, code in do_timer_interrupt() expects an AEOI
@@ -1667,6 +1713,7 @@
int new;
int vector = 0;
unsigned long flags;
+ cpumask_t mask;
irq = -ENOSPC;
spin_lock_irqsave(&vector_lock, flags);
@@ -1675,7 +1722,7 @@
continue;
if (irq_vector[new] != 0)
continue;
- vector = __assign_irq_vector(new, TARGET_CPUS);
+ vector = __assign_irq_vector(new, TARGET_CPUS, &mask);
if (likely(vector > 0))
irq = new;
break;
@@ -1707,13 +1754,10 @@
{
int vector;
unsigned dest;
+ cpumask_t tmp;
- vector = assign_irq_vector(irq, TARGET_CPUS);
+ vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
if (vector >= 0) {
- cpumask_t tmp;
-
- cpus_clear(tmp);
- cpu_set(vector >> 8, tmp);
dest = cpu_mask_to_apicid(tmp);
msg->address_hi = MSI_ADDR_BASE_HI;
@@ -1752,12 +1796,10 @@
cpus_and(mask, tmp, CPU_MASK_ALL);
- vector = assign_irq_vector(irq, mask);
+ vector = assign_irq_vector(irq, mask, &tmp);
if (vector < 0)
return;
- cpus_clear(tmp);
- cpu_set(vector >> 8, tmp);
dest = cpu_mask_to_apicid(tmp);
read_msi_msg(irq, &msg);
@@ -1844,12 +1886,10 @@
cpus_and(mask, tmp, CPU_MASK_ALL);
- vector = assign_irq_vector(irq, mask);
+ vector = assign_irq_vector(irq, mask, &tmp);
if (vector < 0)
return;
- cpus_clear(tmp);
- cpu_set(vector >> 8, tmp);
dest = cpu_mask_to_apicid(tmp);
target_ht_irq(irq, dest, vector & 0xff);
@@ -1871,15 +1911,13 @@
int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
{
int vector;
+ cpumask_t tmp;
- vector = assign_irq_vector(irq, TARGET_CPUS);
+ vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
if (vector >= 0) {
u32 low, high;
unsigned dest;
- cpumask_t tmp;
- cpus_clear(tmp);
- cpu_set(vector >> 8, tmp);
dest = cpu_mask_to_apicid(tmp);
high = HT_IRQ_HIGH_DEST_ID(dest);
@@ -1945,13 +1983,10 @@
add_pin_to_irq(irq, ioapic, pin);
- vector = assign_irq_vector(irq, TARGET_CPUS);
+ vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
if (vector < 0)
return vector;
- cpus_clear(mask);
- cpu_set(vector >> 8, mask);
-
/*
* Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
* Note that we mask (disable) IRQs now -- these get enabled when the
diff --git a/include/asm-x86_64/genapic.h b/include/asm-x86_64/genapic.h
index 81e7146..a0e9a4b 100644
--- a/include/asm-x86_64/genapic.h
+++ b/include/asm-x86_64/genapic.h
@@ -18,6 +18,7 @@
u32 int_dest_mode;
int (*apic_id_registered)(void);
cpumask_t (*target_cpus)(void);
+ cpumask_t (*vector_allocation_domain)(int cpu);
void (*init_apic_ldr)(void);
/* ipi */
void (*send_IPI_mask)(cpumask_t mask, int vector);
diff --git a/include/asm-x86_64/mach_apic.h b/include/asm-x86_64/mach_apic.h
index d334224..7b7115a 100644
--- a/include/asm-x86_64/mach_apic.h
+++ b/include/asm-x86_64/mach_apic.h
@@ -17,6 +17,7 @@
#define INT_DELIVERY_MODE (genapic->int_delivery_mode)
#define INT_DEST_MODE (genapic->int_dest_mode)
#define TARGET_CPUS (genapic->target_cpus())
+#define vector_allocation_domain (genapic->vector_allocation_domain)
#define apic_id_registered (genapic->apic_id_registered)
#define init_apic_ldr (genapic->init_apic_ldr)
#define send_IPI_mask (genapic->send_IPI_mask)