Signed-off-by: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>
---
arch/ia64/kernel/irq_ia64.c | 363 ++++++++++++++++++++++++++++++++++++++++++-
arch/ia64/xen/hypercall.S | 10 ++
include/asm-ia64/hw_irq.h | 10 ++
include/asm-ia64/irq.h | 33 ++++
4 files changed, 414 insertions(+), 2 deletions(-)
diff --git a/arch/ia64/kernel/irq_ia64.c b/arch/ia64/kernel/irq_ia64.c
index 0b52f19..b5dcb49 100644
--- a/arch/ia64/kernel/irq_ia64.c
+++ b/arch/ia64/kernel/irq_ia64.c
@@ -30,6 +30,9 @@
#include <linux/threads.h>
#include <linux/bitops.h>
#include <linux/irq.h>
+#ifdef CONFIG_XEN
+#include <linux/cpu.h>
+#endif
#include <asm/delay.h>
#include <asm/intrinsics.h>
@@ -204,6 +207,13 @@ assign_irq_vector (int irq)
vector = -ENOSPC;
+#ifdef CONFIG_XEN
+ if (is_running_on_xen()) {
+ extern int xen_assign_irq_vector(int);
+ return xen_assign_irq_vector(irq);
+ }
+#endif
+
spin_lock_irqsave(&vector_lock, flags);
for_each_online_cpu(cpu) {
domain = vector_allocation_domain(cpu);
@@ -227,6 +237,14 @@ free_irq_vector (int vector)
if (vector < IA64_FIRST_DEVICE_VECTOR ||
vector > IA64_LAST_DEVICE_VECTOR)
return;
+
+#ifdef CONFIG_XEN
+ if (is_running_on_xen()) {
+ extern void xen_free_irq_vector(int);
+ xen_free_irq_vector(vector);
+ return;
+ }
+#endif
clear_irq_vector(vector);
}
@@ -555,12 +573,343 @@ static struct irqaction tlb_irqaction = {
#endif
+#ifdef CONFIG_XEN
+#include <xen/events.h>
+#include <xen/interface/callback.h>
+
+static DEFINE_PER_CPU(int, timer_irq) = -1;
+static DEFINE_PER_CPU(int, ipi_irq) = -1;
+static DEFINE_PER_CPU(int, resched_irq) = -1;
+static DEFINE_PER_CPU(int, cmc_irq) = -1;
+static DEFINE_PER_CPU(int, cmcp_irq) = -1;
+static DEFINE_PER_CPU(int, cpep_irq) = -1;
+static char timer_name[NR_CPUS][15];
+static char ipi_name[NR_CPUS][15];
+static char resched_name[NR_CPUS][15];
+static char cmc_name[NR_CPUS][15];
+static char cmcp_name[NR_CPUS][15];
+static char cpep_name[NR_CPUS][15];
+
+struct saved_irq {
+ unsigned int irq;
+ struct irqaction *action;
+};
+/* 16 should be far optimistic value, since only several percpu irqs
+ * are registered early.
+ */
+#define MAX_LATE_IRQ 16
+static struct saved_irq saved_percpu_irqs[MAX_LATE_IRQ];
+static unsigned short late_irq_cnt = 0;
+static unsigned short saved_irq_cnt = 0;
+static int xen_slab_ready = 0;
+
+#ifdef CONFIG_SMP
+/* Dummy stub. Though we may check RESCHEDULE_VECTOR before __do_IRQ,
+ * it ends up to issue several memory accesses upon percpu data and
+ * thus adds unnecessary traffic to other paths.
+ */
+static irqreturn_t
+xen_dummy_handler(int irq, void *dev_id)
+{
+
+ return IRQ_HANDLED;
+}
+
+static struct irqaction xen_resched_irqaction = {
+ .handler = xen_dummy_handler,
+ .flags = IRQF_DISABLED,
+ .name = "resched"
+};
+
+static struct irqaction xen_tlb_irqaction = {
+ .handler = xen_dummy_handler,
+ .flags = IRQF_DISABLED,
+ .name = "tlb_flush"
+};
+#endif
+
+/*
+ * This is xen version percpu irq registration, which needs bind
+ * to xen specific evtchn sub-system. One trick here is that xen
+ * evtchn binding interface depends on kmalloc because related
+ * port needs to be freed at device/cpu down. So we cache the
+ * registration on BSP before slab is ready and then deal them
+ * at later point. For rest instances happening after slab ready,
+ * we hook them to xen evtchn immediately.
+ *
+ * FIXME: MCA is not supported by far, and thus "nomca" boot param is
+ * required.
+ */
+static void
+xen_register_percpu_irq(unsigned int cpu, unsigned int vec,
+ struct irqaction *action, int save)
+{
+ irq_desc_t *desc;
+ int irq = 0;
+
+ if (xen_slab_ready) {
+ switch (vec) {
+ case IA64_TIMER_VECTOR:
+ sprintf(timer_name[cpu], "%s%d", action->name, cpu);
+ irq = bind_virq_to_irqhandler(VIRQ_ITC, cpu,
+ action->handler, action->flags,
+ timer_name[cpu], action->dev_id);
+ per_cpu(timer_irq, cpu) = irq;
+ break;
+ case IA64_IPI_RESCHEDULE:
+ sprintf(resched_name[cpu], "%s%d", action->name, cpu);
+ irq = bind_ipi_to_irqhandler(RESCHEDULE_VECTOR, cpu,
+ action->handler, action->flags,
+ resched_name[cpu], action->dev_id);
+ per_cpu(resched_irq, cpu) = irq;
+ break;
+ case IA64_IPI_VECTOR:
+ sprintf(ipi_name[cpu], "%s%d", action->name, cpu);
+ irq = bind_ipi_to_irqhandler(IPI_VECTOR, cpu,
+ action->handler, action->flags,
+ ipi_name[cpu], action->dev_id);
+ per_cpu(ipi_irq, cpu) = irq;
+ break;
+ case IA64_CMC_VECTOR:
+ sprintf(cmc_name[cpu], "%s%d", action->name, cpu);
+ irq = bind_virq_to_irqhandler(VIRQ_MCA_CMC, cpu,
+ action->handler,
+ action->flags,
+ cmc_name[cpu],
+ action->dev_id);
+ per_cpu(cmc_irq, cpu) = irq;
+ break;
+ case IA64_CMCP_VECTOR:
+ sprintf(cmcp_name[cpu], "%s%d", action->name, cpu);
+ irq = bind_ipi_to_irqhandler(CMCP_VECTOR, cpu,
+ action->handler,
+ action->flags,
+ cmcp_name[cpu],
+ action->dev_id);
+ per_cpu(cmcp_irq, cpu) = irq;
+ break;
+ case IA64_CPEP_VECTOR:
+ sprintf(cpep_name[cpu], "%s%d", action->name, cpu);
+ irq = bind_ipi_to_irqhandler(CPEP_VECTOR, cpu,
+ action->handler,
+ action->flags,
+ cpep_name[cpu],
+ action->dev_id);
+ per_cpu(cpep_irq, cpu) = irq;
+ break;
+ case IA64_CPE_VECTOR:
+ case IA64_MCA_RENDEZ_VECTOR:
+ case IA64_PERFMON_VECTOR:
+ case IA64_MCA_WAKEUP_VECTOR:
+ case IA64_SPURIOUS_INT_VECTOR:
+ /* No need to complain, these aren't supported. */
+ break;
+ default:
+ printk(KERN_WARNING "Percpu irq %d is unsupported "
+ "by xen!\n", vec);
+ break;
+ }
+ BUG_ON(irq < 0);
+
+ if (irq > 0) {
+ /*
+ * Mark percpu. Without this, migrate_irqs() will
+ * mark the interrupt for migrations and trigger it
+ * on cpu hotplug.
+ */
+ desc = irq_desc + irq;
+ desc->status |= IRQ_PER_CPU;
+ }
+ }
+
+ /* For BSP, we cache registered percpu irqs, and then re-walk
+ * them when initializing APs
+ */
+ if (!cpu && save) {
+ BUG_ON(saved_irq_cnt == MAX_LATE_IRQ);
+ saved_percpu_irqs[saved_irq_cnt].irq = vec;
+ saved_percpu_irqs[saved_irq_cnt].action = action;
+ saved_irq_cnt++;
+ if (!xen_slab_ready)
+ late_irq_cnt++;
+ }
+}
+
+static void
+xen_bind_early_percpu_irq(void)
+{
+ int i;
+
+ xen_slab_ready = 1;
+ /* There's no race when accessing this cached array, since only
+ * BSP will face with such step shortly
+ */
+ for (i = 0; i < late_irq_cnt; i++)
+ xen_register_percpu_irq(smp_processor_id(),
+ saved_percpu_irqs[i].irq,
+ saved_percpu_irqs[i].action, 0);
+}
+
+/* FIXME: There's no obvious point to check whether slab is ready. So
+ * a hack is used here by utilizing a late time hook.
+ */
+extern void (*late_time_init)(void);
+extern char xen_event_callback;
+extern void xen_init_IRQ(void);
+
+#ifdef CONFIG_HOTPLUG_CPU
+static int __devinit
+unbind_evtchn_callback(struct notifier_block *nfb,
+ unsigned long action, void *hcpu)
+{
+ unsigned int cpu = (unsigned long)hcpu;
+
+ if (action == CPU_DEAD) {
+ /* Unregister evtchn. */
+ if (per_cpu(cpep_irq, cpu) >= 0) {
+ unbind_from_irqhandler(per_cpu(cpep_irq, cpu), NULL);
+ per_cpu(cpep_irq, cpu) = -1;
+ }
+ if (per_cpu(cmcp_irq, cpu) >= 0) {
+ unbind_from_irqhandler(per_cpu(cmcp_irq, cpu), NULL);
+ per_cpu(cmcp_irq, cpu) = -1;
+ }
+ if (per_cpu(cmc_irq, cpu) >= 0) {
+ unbind_from_irqhandler(per_cpu(cmc_irq, cpu), NULL);
+ per_cpu(cmc_irq, cpu) = -1;
+ }
+ if (per_cpu(ipi_irq, cpu) >= 0) {
+ unbind_from_irqhandler(per_cpu(ipi_irq, cpu), NULL);
+ per_cpu(ipi_irq, cpu) = -1;
+ }
+ if (per_cpu(resched_irq, cpu) >= 0) {
+ unbind_from_irqhandler(per_cpu(resched_irq, cpu),
+ NULL);
+ per_cpu(resched_irq, cpu) = -1;
+ }
+ if (per_cpu(timer_irq, cpu) >= 0) {
+ unbind_from_irqhandler(per_cpu(timer_irq, cpu), NULL);
+ per_cpu(timer_irq, cpu) = -1;
+ }
+ }
+ return NOTIFY_OK;
+}
+
+static struct notifier_block unbind_evtchn_notifier = {
+ .notifier_call = unbind_evtchn_callback,
+ .priority = 0
+};
+#endif
+
+DECLARE_PER_CPU(int, ipi_to_irq[NR_IPIS]);
+void xen_smp_intr_init_early(unsigned int cpu)
+{
+#ifdef CONFIG_SMP
+ unsigned int i;
+
+ for (i = 0; i < saved_irq_cnt; i++)
+ xen_register_percpu_irq(cpu, saved_percpu_irqs[i].irq,
+ saved_percpu_irqs[i].action, 0);
+#endif
+}
+
+void xen_smp_intr_init(void)
+{
+#ifdef CONFIG_SMP
+ unsigned int cpu = smp_processor_id();
+ struct callback_register event = {
+ .type = CALLBACKTYPE_event,
+ .address = (unsigned long)&xen_event_callback,
+ };
+
+ if (cpu == 0) {
+ /* Initialization was already done for boot cpu. */
+#ifdef CONFIG_HOTPLUG_CPU
+ /* Register the notifier only once. */
+ register_cpu_notifier(&unbind_evtchn_notifier);
+#endif
+ return;
+ }
+
+ /* This should be piggyback when setup vcpu guest context */
+ BUG_ON(HYPERVISOR_callback_op(CALLBACKOP_register, &event));
+#endif /* CONFIG_SMP */
+}
+
+void __init
+xen_irq_init(void)
+{
+ struct callback_register event = {
+ .type = CALLBACKTYPE_event,
+ .address = (unsigned long)&xen_event_callback,
+ };
+
+ xen_init_IRQ();
+ BUG_ON(HYPERVISOR_callback_op(CALLBACKOP_register, &event));
+ late_time_init = xen_bind_early_percpu_irq;
+}
+
+void
+xen_platform_send_ipi(int cpu, int vector, int delivery_mode, int redirect)
+{
+ int irq = -1;
+
+#ifdef CONFIG_SMP
+ /* TODO: we need to call vcpu_up here */
+ if (unlikely(vector == ap_wakeup_vector)) {
+ extern void xen_send_ipi(int cpu, int vec);
+
+ /* XXX
+ * This should be in __cpu_up(cpu) in ia64 smpboot.c
+ * like x86. But don't want to modify it,
+ * keep it untouched.
+ */
+ xen_smp_intr_init_early(cpu);
+
+ xen_send_ipi(cpu, vector);
+ /* vcpu_prepare_and_up(cpu); */
+ return;
+ }
+#endif
+
+ switch (vector) {
+ case IA64_IPI_VECTOR:
+ irq = per_cpu(ipi_to_irq, cpu)[IPI_VECTOR];
+ break;
+ case IA64_IPI_RESCHEDULE:
+ irq = per_cpu(ipi_to_irq, cpu)[RESCHEDULE_VECTOR];
+ break;
+ case IA64_CMCP_VECTOR:
+ irq = per_cpu(ipi_to_irq, cpu)[CMCP_VECTOR];
+ break;
+ case IA64_CPEP_VECTOR:
+ irq = per_cpu(ipi_to_irq, cpu)[CPEP_VECTOR];
+ break;
+ default:
+ printk(KERN_WARNING "Unsupported IPI type 0x%x\n",
+ vector);
+ irq = 0;
+ break;
+ }
+
+ BUG_ON(irq < 0);
+ notify_remote_via_irq(irq);
+ return;
+}
+#endif /* CONFIG_XEN */
+
void
register_percpu_irq (ia64_vector vec, struct irqaction *action)
{
irq_desc_t *desc;
unsigned int irq;
+#ifdef CONFIG_XEN
+ if (is_running_on_xen())
+ return xen_register_percpu_irq(smp_processor_id(),
+ vec, action, 1);
+#endif
+
irq = vec;
BUG_ON(bind_irq_vector(irq, vec, CPU_MASK_ALL));
desc = irq_desc + irq;
@@ -576,8 +925,18 @@ init_IRQ (void)
register_percpu_irq(IA64_SPURIOUS_INT_VECTOR, NULL);
#ifdef CONFIG_SMP
register_percpu_irq(IA64_IPI_VECTOR, &ipi_irqaction);
- register_percpu_irq(IA64_IPI_RESCHEDULE, &resched_irqaction);
- register_percpu_irq(IA64_IPI_LOCAL_TLB_FLUSH, &tlb_irqaction);
+#ifdef CONFIG_XEN
+ if (is_running_on_xen()) {
+ register_percpu_irq(IA64_IPI_RESCHEDULE,
+ &xen_resched_irqaction);
+ register_percpu_irq(IA64_IPI_LOCAL_TLB_FLUSH,
+ &xen_tlb_irqaction);
+ } else
+#endif
+ {
+ register_percpu_irq(IA64_IPI_RESCHEDULE, &resched_irqaction);
+ register_percpu_irq(IA64_IPI_LOCAL_TLB_FLUSH, &tlb_irqaction);
+ }
#endif
#ifdef CONFIG_PERFMON
pfm_init_percpu();
diff --git a/arch/ia64/xen/hypercall.S b/arch/ia64/xen/hypercall.S
index 7c5242b..3fad2fe 100644
--- a/arch/ia64/xen/hypercall.S
+++ b/arch/ia64/xen/hypercall.S
@@ -123,6 +123,16 @@ END(xen_set_eflag)
#endif /* CONFIG_IA32_SUPPORT */
#endif /* ASM_SUPPORTED */
+GLOBAL_ENTRY(xen_send_ipi)
+ mov r14=r32
+ mov r15=r33
+ mov r2=0x400
+ break 0x1000
+ ;;
+ br.ret.sptk.many rp
+ ;;
+END(xen_send_ipi)
+
GLOBAL_ENTRY(__hypercall)
mov r2=r37
break 0x1000
diff --git a/include/asm-ia64/hw_irq.h b/include/asm-ia64/hw_irq.h
index 7e6e377..357f5a8 100644
--- a/include/asm-ia64/hw_irq.h
+++ b/include/asm-ia64/hw_irq.h
@@ -15,7 +15,11 @@
#include <asm/ptrace.h>
#include <asm/smp.h>
+#ifndef CONFIG_XEN
typedef u8 ia64_vector;
+#else
+typedef u16 ia64_vector;
+#endif
/*
* 0 special
@@ -114,6 +118,12 @@ extern void destroy_and_reserve_irq (unsigned int irq);
static inline void ia64_resend_irq(unsigned int vector)
{
+#ifdef CONFIG_XEN
+ extern int resend_irq_on_evtchn(unsigned int i);
+ if (is_running_on_xen())
+ resend_irq_on_evtchn(vector);
+ else
+#endif /* CONFIG_XEN */
platform_send_ipi(smp_processor_id(), vector, IA64_IPI_DM_INT, 0);
}
diff --git a/include/asm-ia64/irq.h b/include/asm-ia64/irq.h
index a66d268..aead249 100644
--- a/include/asm-ia64/irq.h
+++ b/include/asm-ia64/irq.h
@@ -14,6 +14,7 @@
#include <linux/types.h>
#include <linux/cpumask.h>
+#ifndef CONFIG_XEN
#define NR_VECTORS 256
#if (NR_VECTORS + 32 * NR_CPUS) < 1024
@@ -21,6 +22,38 @@
#else
#define NR_IRQS 1024
#endif
+#else
+/*
+ * The flat IRQ space is divided into two regions:
+ * 1. A one-to-one mapping of real physical IRQs. This space is only used
+ * if we have physical device-access privilege. This region is at the
+ * start of the IRQ space so that existing device drivers do not need
+ * to be modified to translate physical IRQ numbers into our IRQ space.
+ * 3. A dynamic mapping of inter-domain and Xen-sourced virtual IRQs. These
+ * are bound using the provided bind/unbind functions.
+ */
+
+#define PIRQ_BASE 0
+#define NR_PIRQS 256
+
+#define DYNIRQ_BASE (PIRQ_BASE + NR_PIRQS)
+#define NR_DYNIRQS (CONFIG_NR_CPUS * 8)
+
+#define NR_IRQS (NR_PIRQS + NR_DYNIRQS)
+#define NR_IRQ_VECTORS NR_IRQS
+
+#define pirq_to_irq(_x) ((_x) + PIRQ_BASE)
+#define irq_to_pirq(_x) ((_x) - PIRQ_BASE)
+
+#define dynirq_to_irq(_x) ((_x) + DYNIRQ_BASE)
+#define irq_to_dynirq(_x) ((_x) - DYNIRQ_BASE)
+
+#define RESCHEDULE_VECTOR 0
+#define IPI_VECTOR 1
+#define CMCP_VECTOR 2
+#define CPEP_VECTOR 3
+#define NR_IPIS 4
+#endif /* CONFIG_XEN */
static __inline__ int
irq_canonicalize (int irq)
--
1.5.3
--
yamahata
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/virtualization
