Xen project Mailing List

[Xen-devel] [PATCH v2] x86/apic/x2apic: Share IRQ vector between cluster members only when no cpumask is specified

From: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>

Date: Tue, 8 Aug 2017 11:59:00 -0400

Cc: andrew.cooper3@xxxxxxxxxx, jbeulich@xxxxxxxx

Delivery-date: Tue, 08 Aug 2017 15:56:49 +0000

List-id: Xen developer discussion <xen-devel.lists.xen.org>

We have limited number (slightly under NR_DYNAMIC_VECTORS=192) of IRQ vectors that are available to each processor. Currently, when x2apic cluster mode is used (which is default), each vector is shared among all processors in the cluster. With many IRQs (as is the case on systems with multiple SR-IOV cards) and few clusters (e.g. single socket) there is a good chance that we will run out of vectors. This patch tries to decrease vector sharing between processors by assigning vector to a single processor if the assignment request (via __assign_irq_vector()) comes without explicitly specifying which processors are expected to share the interrupt. This typically happens during boot time (or possibly PCI hotplug) when create_irq(NUMA_NO_NODE) is called. When __assign_irq_vector() is called from set_desc_affinity() which provides sharing mask, vector sharing will continue to be performed, as before. This patch to some extent mirrors Linux commit d872818dbbee ("x86/apic/x2apic: Use multiple cluster members for the irq destination only with the explicit affinity"). Note that this change still does not guarantee that we never run out of vectors. For example, on a single core system we will be effectively back to the single cluster/socket case of original code. Signed-off-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx> --- Changes in v2: * Instead of relying on having mask set to TARGET_CPUS as indication that the caller doesn't care about how vectors are shared allow passing NULL mask to __assign_irq_vector() (and then to vector_allocation_cpumask()). xen/arch/x86/genapic/delivery.c | 6 ++++-- xen/arch/x86/genapic/x2apic.c | 6 +++++- xen/arch/x86/irq.c | 9 +++++---- xen/include/asm-x86/genapic.h | 9 ++++++--- xen/include/asm-x86/mach-generic/mach_apic.h | 3 ++- 5 files changed, 22 insertions(+), 11 deletions(-) diff --git a/xen/arch/x86/genapic/delivery.c b/xen/arch/x86/genapic/delivery.c index ced92a1..3cb65d2 100644 --- a/xen/arch/x86/genapic/delivery.c +++ b/xen/arch/x86/genapic/delivery.c @@ -30,7 +30,8 @@ void __init clustered_apic_check_flat(void) printk("Enabling APIC mode: Flat. Using %d I/O APICs\n", nr_ioapics); } -const cpumask_t *vector_allocation_cpumask_flat(int cpu) +const cpumask_t *vector_allocation_cpumask_flat(int cpu, + const cpumask_t *cpumask) { return &cpu_online_map; } @@ -58,7 +59,8 @@ void __init clustered_apic_check_phys(void) printk("Enabling APIC mode: Phys. Using %d I/O APICs\n", nr_ioapics); } -const cpumask_t *vector_allocation_cpumask_phys(int cpu) +const cpumask_t *vector_allocation_cpumask_phys(int cpu, + const cpumask_t *cpumask) { return cpumask_of(cpu); } diff --git a/xen/arch/x86/genapic/x2apic.c b/xen/arch/x86/genapic/x2apic.c index 5fffb31..b12d529 100644 --- a/xen/arch/x86/genapic/x2apic.c +++ b/xen/arch/x86/genapic/x2apic.c @@ -72,8 +72,12 @@ static void __init clustered_apic_check_x2apic(void) { } -static const cpumask_t *vector_allocation_cpumask_x2apic_cluster(int cpu) +static const cpumask_t *vector_allocation_cpumask_x2apic_cluster(int cpu, + const cpumask_t *cpumask) { + if ( !cpumask ) + return cpumask_of(cpu); + return per_cpu(cluster_cpus, cpu); } diff --git a/xen/arch/x86/irq.c b/xen/arch/x86/irq.c index 57e6c18..a0385a3 100644 --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -450,11 +450,12 @@ static int __assign_irq_vector( static int current_vector = FIRST_DYNAMIC_VECTOR, current_offset = 0; int cpu, err, old_vector; cpumask_t tmp_mask; + const cpumask_t *initial_mask = mask ?: TARGET_CPUS; vmask_t *irq_used_vectors = NULL; old_vector = irq_to_vector(irq); if (old_vector > 0) { - cpumask_and(&tmp_mask, mask, &cpu_online_map); + cpumask_and(&tmp_mask, initial_mask, &cpu_online_map); if (cpumask_intersects(&tmp_mask, desc->arch.cpu_mask)) { desc->arch.vector = old_vector; return 0; @@ -476,7 +477,7 @@ static int __assign_irq_vector( else irq_used_vectors = irq_get_used_vector_mask(irq); - for_each_cpu(cpu, mask) { + for_each_cpu(cpu, initial_mask) { int new_cpu; int vector, offset; @@ -484,7 +485,7 @@ static int __assign_irq_vector( if (!cpu_online(cpu)) continue; - cpumask_and(&tmp_mask, vector_allocation_cpumask(cpu), + cpumask_and(&tmp_mask, vector_allocation_cpumask(cpu, mask), &cpu_online_map); vector = current_vector; @@ -550,7 +551,7 @@ int assign_irq_vector(int irq, const cpumask_t *mask) BUG_ON(irq >= nr_irqs || irq <0); spin_lock_irqsave(&vector_lock, flags); - ret = __assign_irq_vector(irq, desc, mask ?: TARGET_CPUS); + ret = __assign_irq_vector(irq, desc, mask); if (!ret) { ret = desc->arch.vector; cpumask_copy(desc->affinity, desc->arch.cpu_mask); diff --git a/xen/include/asm-x86/genapic.h b/xen/include/asm-x86/genapic.h index 5496ab0..306544d 100644 --- a/xen/include/asm-x86/genapic.h +++ b/xen/include/asm-x86/genapic.h @@ -34,7 +34,8 @@ struct genapic { void (*init_apic_ldr)(void); void (*clustered_apic_check)(void); const cpumask_t *(*target_cpus)(void); - const cpumask_t *(*vector_allocation_cpumask)(int cpu); + const cpumask_t *(*vector_allocation_cpumask)(int cpu, + const cpumask_t *mask); unsigned int (*cpu_mask_to_apicid)(const cpumask_t *cpumask); void (*send_IPI_mask)(const cpumask_t *mask, int vector); void (*send_IPI_self)(uint8_t vector); @@ -58,7 +59,8 @@ void init_apic_ldr_flat(void); void clustered_apic_check_flat(void); unsigned int cpu_mask_to_apicid_flat(const cpumask_t *cpumask); void send_IPI_mask_flat(const cpumask_t *mask, int vector); -const cpumask_t *vector_allocation_cpumask_flat(int cpu); +const cpumask_t *vector_allocation_cpumask_flat(int cpu, + const cpumask_t *mask); #define GENAPIC_FLAT \ .int_delivery_mode = dest_LowestPrio, \ .int_dest_mode = 1 /* logical delivery */, \ @@ -74,7 +76,8 @@ void init_apic_ldr_phys(void); void clustered_apic_check_phys(void); unsigned int cpu_mask_to_apicid_phys(const cpumask_t *cpumask); void send_IPI_mask_phys(const cpumask_t *mask, int vector); -const cpumask_t *vector_allocation_cpumask_phys(int cpu); +const cpumask_t *vector_allocation_cpumask_phys(int cpu, + const cpumask_t *mask); #define GENAPIC_PHYS \ .int_delivery_mode = dest_Fixed, \ .int_dest_mode = 0 /* physical delivery */, \ diff --git a/xen/include/asm-x86/mach-generic/mach_apic.h b/xen/include/asm-x86/mach-generic/mach_apic.h index 03e9e8a..60a32c5 100644 --- a/xen/include/asm-x86/mach-generic/mach_apic.h +++ b/xen/include/asm-x86/mach-generic/mach_apic.h @@ -16,7 +16,8 @@ #define init_apic_ldr (genapic->init_apic_ldr) #define clustered_apic_check (genapic->clustered_apic_check) #define cpu_mask_to_apicid (genapic->cpu_mask_to_apicid) -#define vector_allocation_cpumask(cpu) (genapic->vector_allocation_cpumask(cpu)) +#define vector_allocation_cpumask(cpu, mask) \ + (genapic->vector_allocation_cpumask(cpu, mask)) static inline void enable_apic_mode(void) { -- 1.8.3.1 _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxx https://lists.xen.org/xen-devel

©2013 Xen Project, A Linux Foundation Collaborative Project. All Rights Reserved.
Linux Foundation is a registered trademark of The Linux Foundation.
Xen Project is a trademark of The Linux Foundation.