xen/events: move 2-level specific code into its own file
authorDavid Vrabel <david.vrabel@citrix.com>
Wed, 13 Mar 2013 15:29:25 +0000 (15:29 +0000)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Mon, 6 Jan 2014 15:07:41 +0000 (10:07 -0500)
In preparation for alternative event channel ABIs, move all the
functions accessing the shared data structures into their own file.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
drivers/xen/events/Makefile
drivers/xen/events/events_2l.c [new file with mode: 0644]
drivers/xen/events/events_base.c
drivers/xen/events/events_internal.h [new file with mode: 0644]
include/xen/events.h

index f0bc6071fd84b2d226c02c55d6abd0181dc592f5..08179fe04612b90f2b8a1d613bc4f73e0ac86e04 100644 (file)
@@ -1,3 +1,4 @@
 obj-y += events.o
 
 events-y += events_base.o
+events-y += events_2l.o
diff --git a/drivers/xen/events/events_2l.c b/drivers/xen/events/events_2l.c
new file mode 100644 (file)
index 0000000..a77e98d
--- /dev/null
@@ -0,0 +1,348 @@
+/*
+ * Xen event channels (2-level ABI)
+ *
+ * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
+ */
+
+#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
+
+#include <linux/linkage.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/module.h>
+
+#include <asm/sync_bitops.h>
+#include <asm/xen/hypercall.h>
+#include <asm/xen/hypervisor.h>
+
+#include <xen/xen.h>
+#include <xen/xen-ops.h>
+#include <xen/events.h>
+#include <xen/interface/xen.h>
+#include <xen/interface/event_channel.h>
+
+#include "events_internal.h"
+
+/*
+ * Note sizeof(xen_ulong_t) can be more than sizeof(unsigned long). Be
+ * careful to only use bitops which allow for this (e.g
+ * test_bit/find_first_bit and friends but not __ffs) and to pass
+ * BITS_PER_EVTCHN_WORD as the bitmask length.
+ */
+#define BITS_PER_EVTCHN_WORD (sizeof(xen_ulong_t)*8)
+/*
+ * Make a bitmask (i.e. unsigned long *) of a xen_ulong_t
+ * array. Primarily to avoid long lines (hence the terse name).
+ */
+#define BM(x) (unsigned long *)(x)
+/* Find the first set bit in a evtchn mask */
+#define EVTCHN_FIRST_BIT(w) find_first_bit(BM(&(w)), BITS_PER_EVTCHN_WORD)
+
+static DEFINE_PER_CPU(xen_ulong_t [NR_EVENT_CHANNELS/BITS_PER_EVTCHN_WORD],
+                     cpu_evtchn_mask);
+
+void xen_evtchn_port_bind_to_cpu(struct irq_info *info, int cpu)
+{
+       clear_bit(info->evtchn, BM(per_cpu(cpu_evtchn_mask, info->cpu)));
+       set_bit(info->evtchn, BM(per_cpu(cpu_evtchn_mask, cpu)));
+}
+
+void clear_evtchn(int port)
+{
+       struct shared_info *s = HYPERVISOR_shared_info;
+       sync_clear_bit(port, BM(&s->evtchn_pending[0]));
+}
+
+void set_evtchn(int port)
+{
+       struct shared_info *s = HYPERVISOR_shared_info;
+       sync_set_bit(port, BM(&s->evtchn_pending[0]));
+}
+
+int test_evtchn(int port)
+{
+       struct shared_info *s = HYPERVISOR_shared_info;
+       return sync_test_bit(port, BM(&s->evtchn_pending[0]));
+}
+
+int test_and_set_mask(int port)
+{
+       struct shared_info *s = HYPERVISOR_shared_info;
+       return sync_test_and_set_bit(port, BM(&s->evtchn_mask[0]));
+}
+
+void mask_evtchn(int port)
+{
+       struct shared_info *s = HYPERVISOR_shared_info;
+       sync_set_bit(port, BM(&s->evtchn_mask[0]));
+}
+
+void unmask_evtchn(int port)
+{
+       struct shared_info *s = HYPERVISOR_shared_info;
+       unsigned int cpu = get_cpu();
+       int do_hypercall = 0, evtchn_pending = 0;
+
+       BUG_ON(!irqs_disabled());
+
+       if (unlikely((cpu != cpu_from_evtchn(port))))
+               do_hypercall = 1;
+       else {
+               /*
+                * Need to clear the mask before checking pending to
+                * avoid a race with an event becoming pending.
+                *
+                * EVTCHNOP_unmask will only trigger an upcall if the
+                * mask bit was set, so if a hypercall is needed
+                * remask the event.
+                */
+               sync_clear_bit(port, BM(&s->evtchn_mask[0]));
+               evtchn_pending = sync_test_bit(port, BM(&s->evtchn_pending[0]));
+
+               if (unlikely(evtchn_pending && xen_hvm_domain())) {
+                       sync_set_bit(port, BM(&s->evtchn_mask[0]));
+                       do_hypercall = 1;
+               }
+       }
+
+       /* Slow path (hypercall) if this is a non-local port or if this is
+        * an hvm domain and an event is pending (hvm domains don't have
+        * their own implementation of irq_enable). */
+       if (do_hypercall) {
+               struct evtchn_unmask unmask = { .port = port };
+               (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
+       } else {
+               struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
+
+               /*
+                * The following is basically the equivalent of
+                * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
+                * the interrupt edge' if the channel is masked.
+                */
+               if (evtchn_pending &&
+                   !sync_test_and_set_bit(port / BITS_PER_EVTCHN_WORD,
+                                          BM(&vcpu_info->evtchn_pending_sel)))
+                       vcpu_info->evtchn_upcall_pending = 1;
+       }
+
+       put_cpu();
+}
+
+static DEFINE_PER_CPU(unsigned int, current_word_idx);
+static DEFINE_PER_CPU(unsigned int, current_bit_idx);
+
+/*
+ * Mask out the i least significant bits of w
+ */
+#define MASK_LSBS(w, i) (w & ((~((xen_ulong_t)0UL)) << i))
+
+static inline xen_ulong_t active_evtchns(unsigned int cpu,
+                                        struct shared_info *sh,
+                                        unsigned int idx)
+{
+       return sh->evtchn_pending[idx] &
+               per_cpu(cpu_evtchn_mask, cpu)[idx] &
+               ~sh->evtchn_mask[idx];
+}
+
+/*
+ * Search the CPU's pending events bitmasks.  For each one found, map
+ * the event number to an irq, and feed it into do_IRQ() for handling.
+ *
+ * Xen uses a two-level bitmap to speed searching.  The first level is
+ * a bitset of words which contain pending event bits.  The second
+ * level is a bitset of pending events themselves.
+ */
+void xen_evtchn_handle_events(int cpu)
+{
+       int irq;
+       xen_ulong_t pending_words;
+       xen_ulong_t pending_bits;
+       int start_word_idx, start_bit_idx;
+       int word_idx, bit_idx;
+       int i;
+       struct irq_desc *desc;
+       struct shared_info *s = HYPERVISOR_shared_info;
+       struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
+
+       /* Timer interrupt has highest priority. */
+       irq = irq_from_virq(cpu, VIRQ_TIMER);
+       if (irq != -1) {
+               unsigned int evtchn = evtchn_from_irq(irq);
+               word_idx = evtchn / BITS_PER_LONG;
+               bit_idx = evtchn % BITS_PER_LONG;
+               if (active_evtchns(cpu, s, word_idx) & (1ULL << bit_idx)) {
+                       desc = irq_to_desc(irq);
+                       if (desc)
+                               generic_handle_irq_desc(irq, desc);
+               }
+       }
+
+       /*
+        * Master flag must be cleared /before/ clearing
+        * selector flag. xchg_xen_ulong must contain an
+        * appropriate barrier.
+        */
+       pending_words = xchg_xen_ulong(&vcpu_info->evtchn_pending_sel, 0);
+
+       start_word_idx = __this_cpu_read(current_word_idx);
+       start_bit_idx = __this_cpu_read(current_bit_idx);
+
+       word_idx = start_word_idx;
+
+       for (i = 0; pending_words != 0; i++) {
+               xen_ulong_t words;
+
+               words = MASK_LSBS(pending_words, word_idx);
+
+               /*
+                * If we masked out all events, wrap to beginning.
+                */
+               if (words == 0) {
+                       word_idx = 0;
+                       bit_idx = 0;
+                       continue;
+               }
+               word_idx = EVTCHN_FIRST_BIT(words);
+
+               pending_bits = active_evtchns(cpu, s, word_idx);
+               bit_idx = 0; /* usually scan entire word from start */
+               /*
+                * We scan the starting word in two parts.
+                *
+                * 1st time: start in the middle, scanning the
+                * upper bits.
+                *
+                * 2nd time: scan the whole word (not just the
+                * parts skipped in the first pass) -- if an
+                * event in the previously scanned bits is
+                * pending again it would just be scanned on
+                * the next loop anyway.
+                */
+               if (word_idx == start_word_idx) {
+                       if (i == 0)
+                               bit_idx = start_bit_idx;
+               }
+
+               do {
+                       xen_ulong_t bits;
+                       int port;
+
+                       bits = MASK_LSBS(pending_bits, bit_idx);
+
+                       /* If we masked out all events, move on. */
+                       if (bits == 0)
+                               break;
+
+                       bit_idx = EVTCHN_FIRST_BIT(bits);
+
+                       /* Process port. */
+                       port = (word_idx * BITS_PER_EVTCHN_WORD) + bit_idx;
+                       irq = evtchn_to_irq[port];
+
+                       if (irq != -1) {
+                               desc = irq_to_desc(irq);
+                               if (desc)
+                                       generic_handle_irq_desc(irq, desc);
+                       }
+
+                       bit_idx = (bit_idx + 1) % BITS_PER_EVTCHN_WORD;
+
+                       /* Next caller starts at last processed + 1 */
+                       __this_cpu_write(current_word_idx,
+                                        bit_idx ? word_idx :
+                                        (word_idx+1) % BITS_PER_EVTCHN_WORD);
+                       __this_cpu_write(current_bit_idx, bit_idx);
+               } while (bit_idx != 0);
+
+               /* Scan start_l1i twice; all others once. */
+               if ((word_idx != start_word_idx) || (i != 0))
+                       pending_words &= ~(1UL << word_idx);
+
+               word_idx = (word_idx + 1) % BITS_PER_EVTCHN_WORD;
+       }
+}
+
+irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
+{
+       struct shared_info *sh = HYPERVISOR_shared_info;
+       int cpu = smp_processor_id();
+       xen_ulong_t *cpu_evtchn = per_cpu(cpu_evtchn_mask, cpu);
+       int i;
+       unsigned long flags;
+       static DEFINE_SPINLOCK(debug_lock);
+       struct vcpu_info *v;
+
+       spin_lock_irqsave(&debug_lock, flags);
+
+       printk("\nvcpu %d\n  ", cpu);
+
+       for_each_online_cpu(i) {
+               int pending;
+               v = per_cpu(xen_vcpu, i);
+               pending = (get_irq_regs() && i == cpu)
+                       ? xen_irqs_disabled(get_irq_regs())
+                       : v->evtchn_upcall_mask;
+               printk("%d: masked=%d pending=%d event_sel %0*"PRI_xen_ulong"\n  ", i,
+                      pending, v->evtchn_upcall_pending,
+                      (int)(sizeof(v->evtchn_pending_sel)*2),
+                      v->evtchn_pending_sel);
+       }
+       v = per_cpu(xen_vcpu, cpu);
+
+       printk("\npending:\n   ");
+       for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
+               printk("%0*"PRI_xen_ulong"%s",
+                      (int)sizeof(sh->evtchn_pending[0])*2,
+                      sh->evtchn_pending[i],
+                      i % 8 == 0 ? "\n   " : " ");
+       printk("\nglobal mask:\n   ");
+       for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
+               printk("%0*"PRI_xen_ulong"%s",
+                      (int)(sizeof(sh->evtchn_mask[0])*2),
+                      sh->evtchn_mask[i],
+                      i % 8 == 0 ? "\n   " : " ");
+
+       printk("\nglobally unmasked:\n   ");
+       for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
+               printk("%0*"PRI_xen_ulong"%s",
+                      (int)(sizeof(sh->evtchn_mask[0])*2),
+                      sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
+                      i % 8 == 0 ? "\n   " : " ");
+
+       printk("\nlocal cpu%d mask:\n   ", cpu);
+       for (i = (NR_EVENT_CHANNELS/BITS_PER_EVTCHN_WORD)-1; i >= 0; i--)
+               printk("%0*"PRI_xen_ulong"%s", (int)(sizeof(cpu_evtchn[0])*2),
+                      cpu_evtchn[i],
+                      i % 8 == 0 ? "\n   " : " ");
+
+       printk("\nlocally unmasked:\n   ");
+       for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
+               xen_ulong_t pending = sh->evtchn_pending[i]
+                       & ~sh->evtchn_mask[i]
+                       & cpu_evtchn[i];
+               printk("%0*"PRI_xen_ulong"%s",
+                      (int)(sizeof(sh->evtchn_mask[0])*2),
+                      pending, i % 8 == 0 ? "\n   " : " ");
+       }
+
+       printk("\npending list:\n");
+       for (i = 0; i < NR_EVENT_CHANNELS; i++) {
+               if (sync_test_bit(i, BM(sh->evtchn_pending))) {
+                       int word_idx = i / BITS_PER_EVTCHN_WORD;
+                       printk("  %d: event %d -> irq %d%s%s%s\n",
+                              cpu_from_evtchn(i), i,
+                              evtchn_to_irq[i],
+                              sync_test_bit(word_idx, BM(&v->evtchn_pending_sel))
+                              ? "" : " l2-clear",
+                              !sync_test_bit(i, BM(sh->evtchn_mask))
+                              ? "" : " globally-masked",
+                              sync_test_bit(i, BM(cpu_evtchn))
+                              ? "" : " locally-masked");
+               }
+       }
+
+       spin_unlock_irqrestore(&debug_lock, flags);
+
+       return IRQ_HANDLED;
+}
index fec5da4ff3a0fc2f87e87f59ae616c36851dcf2a..8771b740e30fce82add0f8fbf9a85cbb874eed01 100644 (file)
@@ -59,6 +59,8 @@
 #include <xen/interface/vcpu.h>
 #include <asm/hw_irq.h>
 
+#include "events_internal.h"
+
 /*
  * This lock protects updates to the following mapping and reference-count
  * arrays. The lock does not need to be acquired to read the mapping tables.
@@ -73,72 +75,12 @@ static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
 /* IRQ <-> IPI mapping */
 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
 
-/* Interrupt types. */
-enum xen_irq_type {
-       IRQT_UNBOUND = 0,
-       IRQT_PIRQ,
-       IRQT_VIRQ,
-       IRQT_IPI,
-       IRQT_EVTCHN
-};
-
-/*
- * Packed IRQ information:
- * type - enum xen_irq_type
- * event channel - irq->event channel mapping
- * cpu - cpu this event channel is bound to
- * index - type-specific information:
- *    PIRQ - physical IRQ, GSI, flags, and owner domain
- *    VIRQ - virq number
- *    IPI - IPI vector
- *    EVTCHN -
- */
-struct irq_info {
-       struct list_head list;
-       int refcnt;
-       enum xen_irq_type type; /* type */
-       unsigned irq;
-       unsigned short evtchn;  /* event channel */
-       unsigned short cpu;     /* cpu bound */
-
-       union {
-               unsigned short virq;
-               enum ipi_vector ipi;
-               struct {
-                       unsigned short pirq;
-                       unsigned short gsi;
-                       unsigned char flags;
-                       uint16_t domid;
-               } pirq;
-       } u;
-};
-#define PIRQ_NEEDS_EOI (1 << 0)
-#define PIRQ_SHAREABLE (1 << 1)
-
-static int *evtchn_to_irq;
+int *evtchn_to_irq;
 #ifdef CONFIG_X86
 static unsigned long *pirq_eoi_map;
 #endif
 static bool (*pirq_needs_eoi)(unsigned irq);
 
-/*
- * Note sizeof(xen_ulong_t) can be more than sizeof(unsigned long). Be
- * careful to only use bitops which allow for this (e.g
- * test_bit/find_first_bit and friends but not __ffs) and to pass
- * BITS_PER_EVTCHN_WORD as the bitmask length.
- */
-#define BITS_PER_EVTCHN_WORD (sizeof(xen_ulong_t)*8)
-/*
- * Make a bitmask (i.e. unsigned long *) of a xen_ulong_t
- * array. Primarily to avoid long lines (hence the terse name).
- */
-#define BM(x) (unsigned long *)(x)
-/* Find the first set bit in a evtchn mask */
-#define EVTCHN_FIRST_BIT(w) find_first_bit(BM(&(w)), BITS_PER_EVTCHN_WORD)
-
-static DEFINE_PER_CPU(xen_ulong_t [NR_EVENT_CHANNELS/BITS_PER_EVTCHN_WORD],
-                     cpu_evtchn_mask);
-
 /* Xen will never allocate port zero for any purpose. */
 #define VALID_EVTCHN(chn)      ((chn) != 0)
 
@@ -149,7 +91,7 @@ static void enable_dynirq(struct irq_data *data);
 static void disable_dynirq(struct irq_data *data);
 
 /* Get info for IRQ */
-static struct irq_info *info_for_irq(unsigned irq)
+struct irq_info *info_for_irq(unsigned irq)
 {
        return irq_get_handler_data(irq);
 }
@@ -230,7 +172,7 @@ static void xen_irq_info_pirq_init(unsigned irq,
 /*
  * Accessors for packed IRQ information.
  */
-static unsigned int evtchn_from_irq(unsigned irq)
+unsigned int evtchn_from_irq(unsigned irq)
 {
        if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
                return 0;
@@ -244,6 +186,11 @@ unsigned irq_from_evtchn(unsigned int evtchn)
 }
 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 
+int irq_from_virq(unsigned int cpu, unsigned int virq)
+{
+       return per_cpu(virq_to_irq, cpu)[virq];
+}
+
 static enum ipi_vector ipi_from_irq(unsigned irq)
 {
        struct irq_info *info = info_for_irq(irq);
@@ -279,12 +226,12 @@ static enum xen_irq_type type_from_irq(unsigned irq)
        return info_for_irq(irq)->type;
 }
 
-static unsigned cpu_from_irq(unsigned irq)
+unsigned cpu_from_irq(unsigned irq)
 {
        return info_for_irq(irq)->cpu;
 }
 
-static unsigned int cpu_from_evtchn(unsigned int evtchn)
+unsigned int cpu_from_evtchn(unsigned int evtchn)
 {
        int irq = evtchn_to_irq[evtchn];
        unsigned ret = 0;
@@ -310,55 +257,21 @@ static bool pirq_needs_eoi_flag(unsigned irq)
        return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 }
 
-static inline xen_ulong_t active_evtchns(unsigned int cpu,
-                                        struct shared_info *sh,
-                                        unsigned int idx)
-{
-       return sh->evtchn_pending[idx] &
-               per_cpu(cpu_evtchn_mask, cpu)[idx] &
-               ~sh->evtchn_mask[idx];
-}
-
 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 {
        int irq = evtchn_to_irq[chn];
+       struct irq_info *info = info_for_irq(irq);
 
        BUG_ON(irq == -1);
 #ifdef CONFIG_SMP
        cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
 #endif
 
-       clear_bit(chn, BM(per_cpu(cpu_evtchn_mask, cpu_from_irq(irq))));
-       set_bit(chn, BM(per_cpu(cpu_evtchn_mask, cpu)));
+       xen_evtchn_port_bind_to_cpu(info, cpu);
 
-       info_for_irq(irq)->cpu = cpu;
-}
-
-static inline void clear_evtchn(int port)
-{
-       struct shared_info *s = HYPERVISOR_shared_info;
-       sync_clear_bit(port, BM(&s->evtchn_pending[0]));
-}
-
-static inline void set_evtchn(int port)
-{
-       struct shared_info *s = HYPERVISOR_shared_info;
-       sync_set_bit(port, BM(&s->evtchn_pending[0]));
-}
-
-static inline int test_evtchn(int port)
-{
-       struct shared_info *s = HYPERVISOR_shared_info;
-       return sync_test_bit(port, BM(&s->evtchn_pending[0]));
-}
-
-static inline int test_and_set_mask(int port)
-{
-       struct shared_info *s = HYPERVISOR_shared_info;
-       return sync_test_and_set_bit(port, BM(&s->evtchn_mask[0]));
+       info->cpu = cpu;
 }
 
-
 /**
  * notify_remote_via_irq - send event to remote end of event channel via irq
  * @irq: irq of event channel to send event to
@@ -376,63 +289,6 @@ void notify_remote_via_irq(int irq)
 }
 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 
-static void mask_evtchn(int port)
-{
-       struct shared_info *s = HYPERVISOR_shared_info;
-       sync_set_bit(port, BM(&s->evtchn_mask[0]));
-}
-
-static void unmask_evtchn(int port)
-{
-       struct shared_info *s = HYPERVISOR_shared_info;
-       unsigned int cpu = get_cpu();
-       int do_hypercall = 0, evtchn_pending = 0;
-
-       BUG_ON(!irqs_disabled());
-
-       if (unlikely((cpu != cpu_from_evtchn(port))))
-               do_hypercall = 1;
-       else {
-               /*
-                * Need to clear the mask before checking pending to
-                * avoid a race with an event becoming pending.
-                *
-                * EVTCHNOP_unmask will only trigger an upcall if the
-                * mask bit was set, so if a hypercall is needed
-                * remask the event.
-                */
-               sync_clear_bit(port, BM(&s->evtchn_mask[0]));
-               evtchn_pending = sync_test_bit(port, BM(&s->evtchn_pending[0]));
-
-               if (unlikely(evtchn_pending && xen_hvm_domain())) {
-                       sync_set_bit(port, BM(&s->evtchn_mask[0]));
-                       do_hypercall = 1;
-               }
-       }
-
-       /* Slow path (hypercall) if this is a non-local port or if this is
-        * an hvm domain and an event is pending (hvm domains don't have
-        * their own implementation of irq_enable). */
-       if (do_hypercall) {
-               struct evtchn_unmask unmask = { .port = port };
-               (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
-       } else {
-               struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
-
-               /*
-                * The following is basically the equivalent of
-                * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
-                * the interrupt edge' if the channel is masked.
-                */
-               if (evtchn_pending &&
-                   !sync_test_and_set_bit(port / BITS_PER_EVTCHN_WORD,
-                                          BM(&vcpu_info->evtchn_pending_sel)))
-                       vcpu_info->evtchn_upcall_pending = 1;
-       }
-
-       put_cpu();
-}
-
 static void xen_irq_init(unsigned irq)
 {
        struct irq_info *info;
@@ -1216,222 +1072,21 @@ void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
        notify_remote_via_irq(irq);
 }
 
-irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
-{
-       struct shared_info *sh = HYPERVISOR_shared_info;
-       int cpu = smp_processor_id();
-       xen_ulong_t *cpu_evtchn = per_cpu(cpu_evtchn_mask, cpu);
-       int i;
-       unsigned long flags;
-       static DEFINE_SPINLOCK(debug_lock);
-       struct vcpu_info *v;
-
-       spin_lock_irqsave(&debug_lock, flags);
-
-       printk("\nvcpu %d\n  ", cpu);
-
-       for_each_online_cpu(i) {
-               int pending;
-               v = per_cpu(xen_vcpu, i);
-               pending = (get_irq_regs() && i == cpu)
-                       ? xen_irqs_disabled(get_irq_regs())
-                       : v->evtchn_upcall_mask;
-               printk("%d: masked=%d pending=%d event_sel %0*"PRI_xen_ulong"\n  ", i,
-                      pending, v->evtchn_upcall_pending,
-                      (int)(sizeof(v->evtchn_pending_sel)*2),
-                      v->evtchn_pending_sel);
-       }
-       v = per_cpu(xen_vcpu, cpu);
-
-       printk("\npending:\n   ");
-       for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
-               printk("%0*"PRI_xen_ulong"%s",
-                      (int)sizeof(sh->evtchn_pending[0])*2,
-                      sh->evtchn_pending[i],
-                      i % 8 == 0 ? "\n   " : " ");
-       printk("\nglobal mask:\n   ");
-       for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
-               printk("%0*"PRI_xen_ulong"%s",
-                      (int)(sizeof(sh->evtchn_mask[0])*2),
-                      sh->evtchn_mask[i],
-                      i % 8 == 0 ? "\n   " : " ");
-
-       printk("\nglobally unmasked:\n   ");
-       for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
-               printk("%0*"PRI_xen_ulong"%s",
-                      (int)(sizeof(sh->evtchn_mask[0])*2),
-                      sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
-                      i % 8 == 0 ? "\n   " : " ");
-
-       printk("\nlocal cpu%d mask:\n   ", cpu);
-       for (i = (NR_EVENT_CHANNELS/BITS_PER_EVTCHN_WORD)-1; i >= 0; i--)
-               printk("%0*"PRI_xen_ulong"%s", (int)(sizeof(cpu_evtchn[0])*2),
-                      cpu_evtchn[i],
-                      i % 8 == 0 ? "\n   " : " ");
-
-       printk("\nlocally unmasked:\n   ");
-       for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
-               xen_ulong_t pending = sh->evtchn_pending[i]
-                       & ~sh->evtchn_mask[i]
-                       & cpu_evtchn[i];
-               printk("%0*"PRI_xen_ulong"%s",
-                      (int)(sizeof(sh->evtchn_mask[0])*2),
-                      pending, i % 8 == 0 ? "\n   " : " ");
-       }
-
-       printk("\npending list:\n");
-       for (i = 0; i < NR_EVENT_CHANNELS; i++) {
-               if (sync_test_bit(i, BM(sh->evtchn_pending))) {
-                       int word_idx = i / BITS_PER_EVTCHN_WORD;
-                       printk("  %d: event %d -> irq %d%s%s%s\n",
-                              cpu_from_evtchn(i), i,
-                              evtchn_to_irq[i],
-                              sync_test_bit(word_idx, BM(&v->evtchn_pending_sel))
-                                            ? "" : " l2-clear",
-                              !sync_test_bit(i, BM(sh->evtchn_mask))
-                                            ? "" : " globally-masked",
-                              sync_test_bit(i, BM(cpu_evtchn))
-                                            ? "" : " locally-masked");
-               }
-       }
-
-       spin_unlock_irqrestore(&debug_lock, flags);
-
-       return IRQ_HANDLED;
-}
-
 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
-static DEFINE_PER_CPU(unsigned int, current_word_idx);
-static DEFINE_PER_CPU(unsigned int, current_bit_idx);
 
-/*
- * Mask out the i least significant bits of w
- */
-#define MASK_LSBS(w, i) (w & ((~((xen_ulong_t)0UL)) << i))
-
-/*
- * Search the CPUs pending events bitmasks.  For each one found, map
- * the event number to an irq, and feed it into do_IRQ() for
- * handling.
- *
- * Xen uses a two-level bitmap to speed searching.  The first level is
- * a bitset of words which contain pending event bits.  The second
- * level is a bitset of pending events themselves.
- */
 static void __xen_evtchn_do_upcall(void)
 {
-       int start_word_idx, start_bit_idx;
-       int word_idx, bit_idx;
-       int i, irq;
-       int cpu = get_cpu();
-       struct shared_info *s = HYPERVISOR_shared_info;
        struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
+       int cpu = get_cpu();
        unsigned count;
 
        do {
-               xen_ulong_t pending_words;
-               xen_ulong_t pending_bits;
-               struct irq_desc *desc;
-
                vcpu_info->evtchn_upcall_pending = 0;
 
                if (__this_cpu_inc_return(xed_nesting_count) - 1)
                        goto out;
 
-               /*
-                * Master flag must be cleared /before/ clearing
-                * selector flag. xchg_xen_ulong must contain an
-                * appropriate barrier.
-                */
-               if ((irq = per_cpu(virq_to_irq, cpu)[VIRQ_TIMER]) != -1) {
-                       int evtchn = evtchn_from_irq(irq);
-                       word_idx = evtchn / BITS_PER_LONG;
-                       pending_bits = evtchn % BITS_PER_LONG;
-                       if (active_evtchns(cpu, s, word_idx) & (1ULL << pending_bits)) {
-                               desc = irq_to_desc(irq);
-                               if (desc)
-                                       generic_handle_irq_desc(irq, desc);
-                       }
-               }
-
-               pending_words = xchg_xen_ulong(&vcpu_info->evtchn_pending_sel, 0);
-
-               start_word_idx = __this_cpu_read(current_word_idx);
-               start_bit_idx = __this_cpu_read(current_bit_idx);
-
-               word_idx = start_word_idx;
-
-               for (i = 0; pending_words != 0; i++) {
-                       xen_ulong_t words;
-
-                       words = MASK_LSBS(pending_words, word_idx);
-
-                       /*
-                        * If we masked out all events, wrap to beginning.
-                        */
-                       if (words == 0) {
-                               word_idx = 0;
-                               bit_idx = 0;
-                               continue;
-                       }
-                       word_idx = EVTCHN_FIRST_BIT(words);
-
-                       pending_bits = active_evtchns(cpu, s, word_idx);
-                       bit_idx = 0; /* usually scan entire word from start */
-                       /*
-                        * We scan the starting word in two parts.
-                        *
-                        * 1st time: start in the middle, scanning the
-                        * upper bits.
-                        *
-                        * 2nd time: scan the whole word (not just the
-                        * parts skipped in the first pass) -- if an
-                        * event in the previously scanned bits is
-                        * pending again it would just be scanned on
-                        * the next loop anyway.
-                        */
-                       if (word_idx == start_word_idx) {
-                               if (i == 0)
-                                       bit_idx = start_bit_idx;
-                       }
-
-                       do {
-                               xen_ulong_t bits;
-                               int port;
-
-                               bits = MASK_LSBS(pending_bits, bit_idx);
-
-                               /* If we masked out all events, move on. */
-                               if (bits == 0)
-                                       break;
-
-                               bit_idx = EVTCHN_FIRST_BIT(bits);
-
-                               /* Process port. */
-                               port = (word_idx * BITS_PER_EVTCHN_WORD) + bit_idx;
-                               irq = evtchn_to_irq[port];
-
-                               if (irq != -1) {
-                                       desc = irq_to_desc(irq);
-                                       if (desc)
-                                               generic_handle_irq_desc(irq, desc);
-                               }
-
-                               bit_idx = (bit_idx + 1) % BITS_PER_EVTCHN_WORD;
-
-                               /* Next caller starts at last processed + 1 */
-                               __this_cpu_write(current_word_idx,
-                                                bit_idx ? word_idx :
-                                                (word_idx+1) % BITS_PER_EVTCHN_WORD);
-                               __this_cpu_write(current_bit_idx, bit_idx);
-                       } while (bit_idx != 0);
-
-                       /* Scan start_l1i twice; all others once. */
-                       if ((word_idx != start_word_idx) || (i != 0))
-                               pending_words &= ~(1UL << word_idx);
-
-                       word_idx = (word_idx + 1) % BITS_PER_EVTCHN_WORD;
-               }
+               xen_evtchn_handle_events(cpu);
 
                BUG_ON(!irqs_disabled());
 
diff --git a/drivers/xen/events/events_internal.h b/drivers/xen/events/events_internal.h
new file mode 100644 (file)
index 0000000..79ac70b
--- /dev/null
@@ -0,0 +1,74 @@
+/*
+ * Xen Event Channels (internal header)
+ *
+ * Copyright (C) 2013 Citrix Systems R&D Ltd.
+ *
+ * This source code is licensed under the GNU General Public License,
+ * Version 2 or later.  See the file COPYING for more details.
+ */
+#ifndef __EVENTS_INTERNAL_H__
+#define __EVENTS_INTERNAL_H__
+
+/* Interrupt types. */
+enum xen_irq_type {
+       IRQT_UNBOUND = 0,
+       IRQT_PIRQ,
+       IRQT_VIRQ,
+       IRQT_IPI,
+       IRQT_EVTCHN
+};
+
+/*
+ * Packed IRQ information:
+ * type - enum xen_irq_type
+ * event channel - irq->event channel mapping
+ * cpu - cpu this event channel is bound to
+ * index - type-specific information:
+ *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
+ *           guest, or GSI (real passthrough IRQ) of the device.
+ *    VIRQ - virq number
+ *    IPI - IPI vector
+ *    EVTCHN -
+ */
+struct irq_info {
+       struct list_head list;
+       int refcnt;
+       enum xen_irq_type type; /* type */
+       unsigned irq;
+       unsigned short evtchn;  /* event channel */
+       unsigned short cpu;     /* cpu bound */
+
+       union {
+               unsigned short virq;
+               enum ipi_vector ipi;
+               struct {
+                       unsigned short pirq;
+                       unsigned short gsi;
+                       unsigned char vector;
+                       unsigned char flags;
+                       uint16_t domid;
+               } pirq;
+       } u;
+};
+
+#define PIRQ_NEEDS_EOI (1 << 0)
+#define PIRQ_SHAREABLE (1 << 1)
+
+extern int *evtchn_to_irq;
+
+struct irq_info *info_for_irq(unsigned irq);
+unsigned cpu_from_irq(unsigned irq);
+unsigned cpu_from_evtchn(unsigned int evtchn);
+
+void xen_evtchn_port_bind_to_cpu(struct irq_info *info, int cpu);
+
+void clear_evtchn(int port);
+void set_evtchn(int port);
+int test_evtchn(int port);
+int test_and_set_mask(int port);
+void mask_evtchn(int port);
+void unmask_evtchn(int port);
+
+void xen_evtchn_handle_events(int cpu);
+
+#endif /* #ifndef __EVENTS_INTERNAL_H__ */
index c9ea10ee227358b5e4deadc5b21c9612a9d15214..32ae0f2637491ebe709ba24635a1c51982b773db 100644 (file)
@@ -73,6 +73,8 @@ void xen_poll_irq_timeout(int irq, u64 timeout);
 
 /* Determine the IRQ which is bound to an event channel */
 unsigned irq_from_evtchn(unsigned int evtchn);
+int irq_from_virq(unsigned int cpu, unsigned int virq);
+unsigned int evtchn_from_irq(unsigned irq);
 
 /* Xen HVM evtchn vector callback */
 void xen_hvm_callback_vector(void);