From cebdd97736df18bef50417a4157d352308f2380a Mon Sep 17 00:00:00 2001 From: Shan Haitao Date: Sat, 16 Jul 2011 09:12:19 +0100 Subject: [PATCH] Remove MSI IRQ storms prevention logic The reason is: 1. The logic has negative impact on 10G NIC performance (assigned to guest) by lowering the interrupt frequency that Xen can handle. 2. Xen already has IRQ rate limit logic, which can also help to prevent IRQ storms. Signed-off-by: Shan Haitao --- xen/arch/x86/hvm/vlapic.c | 2 - xen/arch/x86/irq.c | 75 ++++-------------------------------- xen/drivers/passthrough/io.c | 60 ----------------------------- xen/include/asm-x86/hvm/io.h | 1 - xen/include/xen/irq.h | 1 - 5 files changed, 8 insertions(+), 131 deletions(-) diff --git a/xen/arch/x86/hvm/vlapic.c b/xen/arch/x86/hvm/vlapic.c index 8b05f1adc3..fabe0e91e8 100644 --- a/xen/arch/x86/hvm/vlapic.c +++ b/xen/arch/x86/hvm/vlapic.c @@ -400,8 +400,6 @@ void vlapic_EOI_set(struct vlapic *vlapic) if ( vlapic_test_and_clear_vector(vector, &vlapic->regs->data[APIC_TMR]) ) vioapic_update_EOI(vlapic_domain(vlapic), vector); - - hvm_dpci_msi_eoi(current->domain, vector); } int vlapic_ipi( diff --git a/xen/arch/x86/irq.c b/xen/arch/x86/irq.c index 98f808450c..c89a3c43c2 100644 --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -790,22 +790,6 @@ static inline void clear_pirq_eoi(struct domain *d, unsigned int irq) clear_bit(irq, d->arch.pv_domain.pirq_eoi_map); } -static void _irq_guest_eoi(struct irq_desc *desc) -{ - irq_guest_action_t *action = (irq_guest_action_t *)desc->action; - unsigned int i, irq = desc - irq_desc; - - if ( !(desc->status & IRQ_GUEST_EOI_PENDING) ) - return; - - for ( i = 0; i < action->nr_guests; ++i ) - clear_pirq_eoi(action->guest[i], - domain_irq_to_pirq(action->guest[i], irq)); - - desc->status &= ~(IRQ_INPROGRESS|IRQ_GUEST_EOI_PENDING); - desc->handler->enable(irq); -} - static void set_eoi_ready(void *data); static void irq_guest_eoi_timer_fn(void *data) @@ -849,9 +833,6 @@ static void irq_guest_eoi_timer_fn(void *data) on_selected_cpus(&cpu_eoi_map, set_eoi_ready, desc, 0); spin_lock_irq(&desc->lock); break; - case ACKTYPE_NONE: - _irq_guest_eoi(desc); - break; } out: @@ -863,7 +844,7 @@ static void __do_IRQ_guest(int irq) struct irq_desc *desc = irq_to_desc(irq); irq_guest_action_t *action = (irq_guest_action_t *)desc->action; struct domain *d; - int i, sp, already_pending = 0; + int i, sp; struct pending_eoi *peoi = this_cpu(pending_eoi); int vector = get_irq_regs()->entry_vector; @@ -897,45 +878,16 @@ static void __do_IRQ_guest(int irq) if ( (action->ack_type != ACKTYPE_NONE) && !test_and_set_bool(pirq->masked) ) action->in_flight++; - if ( hvm_do_IRQ_dpci(d, pirq) ) - { - if ( action->ack_type == ACKTYPE_NONE ) - { - already_pending += !!(desc->status & IRQ_INPROGRESS); - desc->status |= IRQ_INPROGRESS; /* cleared during hvm eoi */ - } - } - else if ( send_guest_pirq(d, pirq) && - (action->ack_type == ACKTYPE_NONE) ) - { - already_pending++; - } + if ( !hvm_do_IRQ_dpci(d, pirq) ) + send_guest_pirq(d, pirq); } - stop_timer(&action->eoi_timer); - - if ( (action->ack_type == ACKTYPE_NONE) && - (already_pending == action->nr_guests) ) + if ( action->ack_type != ACKTYPE_NONE ) { - desc->handler->disable(irq); - desc->status |= IRQ_GUEST_EOI_PENDING; - for ( i = 0; i < already_pending; ++i ) - { - d = action->guest[i]; - set_pirq_eoi(d, domain_irq_to_pirq(d, irq)); - /* - * Could check here whether the guest unmasked the event by now - * (or perhaps just re-issue the send_guest_pirq()), and if it - * can now accept the event, - * - clear all the pirq_eoi bits we already set, - * - re-enable the vector, and - * - skip the timer setup below. - */ - } + stop_timer(&action->eoi_timer); + migrate_timer(&action->eoi_timer, smp_processor_id()); + set_timer(&action->eoi_timer, NOW() + MILLISECS(1)); } - - migrate_timer(&action->eoi_timer, smp_processor_id()); - set_timer(&action->eoi_timer, NOW() + MILLISECS(1)); } /* @@ -1183,13 +1135,6 @@ void desc_guest_eoi(struct irq_desc *desc, struct pirq *pirq) action = (irq_guest_action_t *)desc->action; irq = desc - irq_desc; - if ( action->ack_type == ACKTYPE_NONE ) - { - ASSERT(!pirq->masked); - stop_timer(&action->eoi_timer); - _irq_guest_eoi(desc); - } - if ( unlikely(!test_and_clear_bool(pirq->masked)) || unlikely(--action->in_flight != 0) ) { @@ -1468,10 +1413,6 @@ static irq_guest_action_t *__pirq_guest_unbind( spin_lock_irq(&desc->lock); } break; - case ACKTYPE_NONE: - stop_timer(&action->eoi_timer); - _irq_guest_eoi(desc); - break; } /* @@ -1509,7 +1450,7 @@ static irq_guest_action_t *__pirq_guest_unbind( BUG_ON(!cpus_empty(action->cpu_eoi_map)); desc->action = NULL; - desc->status &= ~(IRQ_GUEST|IRQ_GUEST_EOI_PENDING|IRQ_INPROGRESS); + desc->status &= ~(IRQ_GUEST|IRQ_INPROGRESS); desc->handler->shutdown(irq); /* Caller frees the old guest descriptor block. */ diff --git a/xen/drivers/passthrough/io.c b/xen/drivers/passthrough/io.c index 788c8f6e50..bce86adafc 100644 --- a/xen/drivers/passthrough/io.c +++ b/xen/drivers/passthrough/io.c @@ -421,58 +421,6 @@ int hvm_do_IRQ_dpci(struct domain *d, struct pirq *pirq) } #ifdef SUPPORT_MSI_REMAPPING -/* called with d->event_lock held */ -static void __msi_pirq_eoi(struct hvm_pirq_dpci *pirq_dpci) -{ - irq_desc_t *desc; - - if ( (pirq_dpci->flags & HVM_IRQ_DPCI_MAPPED) && - (pirq_dpci->flags & HVM_IRQ_DPCI_MACH_MSI) ) - { - struct pirq *pirq = dpci_pirq(pirq_dpci); - - BUG_ON(!local_irq_is_enabled()); - desc = pirq_spin_lock_irq_desc(pirq, NULL); - if ( !desc ) - return; - - desc->status &= ~IRQ_INPROGRESS; - desc_guest_eoi(desc, pirq); - } -} - -static int _hvm_dpci_msi_eoi(struct domain *d, - struct hvm_pirq_dpci *pirq_dpci, void *arg) -{ - int vector = (long)arg; - - if ( (pirq_dpci->flags & HVM_IRQ_DPCI_MACH_MSI) && - (pirq_dpci->gmsi.gvec == vector) ) - { - int dest = pirq_dpci->gmsi.gflags & VMSI_DEST_ID_MASK; - int dest_mode = !!(pirq_dpci->gmsi.gflags & VMSI_DM_MASK); - - if ( vlapic_match_dest(vcpu_vlapic(current), NULL, 0, dest, - dest_mode) ) - { - __msi_pirq_eoi(pirq_dpci); - return 1; - } - } - - return 0; -} - -void hvm_dpci_msi_eoi(struct domain *d, int vector) -{ - if ( !iommu_enabled || !d->arch.hvm_domain.irq.dpci ) - return; - - spin_lock(&d->event_lock); - pt_pirq_iterate(d, _hvm_dpci_msi_eoi, (void *)(long)vector); - spin_unlock(&d->event_lock); -} - static int hvm_pci_msi_assert(struct domain *d, struct hvm_pirq_dpci *pirq_dpci) { @@ -510,14 +458,6 @@ static int _hvm_dirq_assist(struct domain *d, struct hvm_pirq_dpci *pirq_dpci, else hvm_pci_intx_assert(d, device, intx); pirq_dpci->pending++; - -#ifdef SUPPORT_MSI_REMAPPING - if ( pirq_dpci->flags & HVM_IRQ_DPCI_TRANSLATE ) - { - /* for translated MSI to INTx interrupt, eoi as early as possible */ - __msi_pirq_eoi(pirq_dpci); - } -#endif } /* diff --git a/xen/include/asm-x86/hvm/io.h b/xen/include/asm-x86/hvm/io.h index ce2bcb373f..419f25f41f 100644 --- a/xen/include/asm-x86/hvm/io.h +++ b/xen/include/asm-x86/hvm/io.h @@ -139,6 +139,5 @@ struct hvm_hw_stdvga { void stdvga_init(struct domain *d); void stdvga_deinit(struct domain *d); -extern void hvm_dpci_msi_eoi(struct domain *d, int vector); #endif /* __ASM_X86_HVM_IO_H__ */ diff --git a/xen/include/xen/irq.h b/xen/include/xen/irq.h index 573ace7e29..32251c5b56 100644 --- a/xen/include/xen/irq.h +++ b/xen/include/xen/irq.h @@ -25,7 +25,6 @@ struct irqaction { #define IRQ_PENDING 4 /* IRQ pending - replay on enable */ #define IRQ_REPLAY 8 /* IRQ has been replayed but not acked yet */ #define IRQ_GUEST 16 /* IRQ is handled by guest OS(es) */ -#define IRQ_GUEST_EOI_PENDING 32 /* IRQ was disabled, pending a guest EOI */ #define IRQ_MOVE_PENDING 64 /* IRQ is migrating to another CPUs */ #define IRQ_PER_CPU 256 /* IRQ is per CPU */ -- 2.30.2