xen: sched: optimize exclusive pinning case (Credit1 & 2)

Exclusive pinning of vCPUs is used, sometimes, for
achieving the highest level of determinism, and the
least possible overhead, for the vCPUs in question.

Although static 1:1 pinning is not recommended, for
general use cases, optimizing the tickling code (of
Credit1 and Credit2) is easy and cheap enough, so go
for it.

Signed-off-by: Dario Faggioli <dfaggioli@suse.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>

diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
index a618d303b4b2160b5eec497ebe4f4ab885d5fd8f..fb7638f8099f8acae2d2700d0052559118ff1135 100644 (file)
--- a/xen/common/sched_credit.c
+++ b/xen/common/sched_credit.c
@@ -73,6 +73,7 @@
 #define CSCHED_FLAG_VCPU_PARKED    0x0  /* VCPU over capped credits */
 #define CSCHED_FLAG_VCPU_YIELD     0x1  /* VCPU yielding */
 #define CSCHED_FLAG_VCPU_MIGRATING 0x2  /* VCPU may have moved to a new pcpu */
+#define CSCHED_FLAG_VCPU_PINNED    0x4  /* VCPU can run only on 1 pcpu */
 
 
 /*
@@ -361,6 +362,25 @@ static inline void __runq_tickle(struct csched_vcpu *new)
     cpumask_and(&idle_mask, prv->idlers, online);
     idlers_empty = cpumask_empty(&idle_mask);
 
+    /*
+     * Exclusive pinning is when a vcpu has hard-affinity with only one
+     * cpu, and there is no other vcpu that has hard-affinity with that
+     * same cpu. This is infrequent, but if it happens, is for achieving
+     * the most possible determinism, and least possible overhead for
+     * the vcpus in question.
+     *
+     * Try to identify the vast majority of these situations, and deal
+     * with them quickly.
+     */
+    if ( unlikely(test_bit(CSCHED_FLAG_VCPU_PINNED, &new->flags) &&
+                  cpumask_test_cpu(cpu, &idle_mask)) )
+    {
+        ASSERT(cpumask_cycle(cpu, new->vcpu->cpu_hard_affinity) == cpu);
+        SCHED_STAT_CRANK(tickled_idle_cpu_excl);
+        __cpumask_set_cpu(cpu, &mask);
+        goto tickle;
+    }
+
     /*
      * If the pcpu is idle, or there are no idlers and the new
      * vcpu is a higher priority than the old vcpu, run it here.
@@ -457,6 +477,7 @@ static inline void __runq_tickle(struct csched_vcpu *new)
         }
     }
 
+ tickle:
     if ( !cpumask_empty(&mask) )
     {
         if ( unlikely(tb_init_done) )
@@ -1213,6 +1234,22 @@ csched_dom_cntl(
     return rc;
 }
 
+static void
+csched_aff_cntl(const struct scheduler *ops, struct vcpu *v,
+                const cpumask_t *hard, const cpumask_t *soft)
+{
+    struct csched_vcpu *svc = CSCHED_VCPU(v);
+
+    if ( !hard )
+        return;
+
+    /* Are we becoming exclusively pinned? */
+    if ( cpumask_weight(hard) == 1 )
+        set_bit(CSCHED_FLAG_VCPU_PINNED, &svc->flags);
+    else
+        clear_bit(CSCHED_FLAG_VCPU_PINNED, &svc->flags);
+}
+
 static inline void
 __csched_set_tslice(struct csched_private *prv, unsigned int timeslice_ms)
 {
@@ -2247,6 +2284,7 @@ static const struct scheduler sched_credit_def = {
     .yield          = csched_vcpu_yield,
 
     .adjust         = csched_dom_cntl,
+    .adjust_affinity= csched_aff_cntl,
     .adjust_global  = csched_sys_cntl,
 
     .pick_cpu       = csched_cpu_pick,

diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
index 5a635e8c4ced3839156ecc813f7139b2cac793f0..0e5a8fc54140948984f089acdbe35c67aa3bc1b2 100644 (file)
--- a/xen/common/sched_credit2.c
+++ b/xen/common/sched_credit2.c
@@ -299,6 +299,12 @@
  */
 #define __CSFLAG_vcpu_yield 4
 #define CSFLAG_vcpu_yield (1U<<__CSFLAG_vcpu_yield)
+/*
+ * CSFLAGS_pinned: this vcpu is currently 'pinned', i.e., has its hard
+ * affinity set to one and only 1 cpu (and, hence, can only run there).
+ */
+#define __CSFLAG_pinned 5
+#define CSFLAG_pinned (1U<<__CSFLAG_pinned)
 
 static unsigned int __read_mostly opt_migrate_resist = 500;
 integer_param("sched_credit2_migrate_resist", opt_migrate_resist);
@@ -1455,6 +1461,26 @@ runq_tickle(const struct scheduler *ops, struct csched2_vcpu *new, s_time_t now)
                     (unsigned char *)&d);
     }
 
+    /*
+     * Exclusive pinning is when a vcpu has hard-affinity with only one
+     * cpu, and there is no other vcpu that has hard-affinity with that
+     * same cpu. This is infrequent, but if it happens, is for achieving
+     * the most possible determinism, and least possible overhead for
+     * the vcpus in question.
+     *
+     * Try to identify the vast majority of these situations, and deal
+     * with them quickly.
+     */
+    if ( unlikely((new->flags & CSFLAG_pinned) &&
+                  cpumask_test_cpu(cpu, &rqd->idle) &&
+                  !cpumask_test_cpu(cpu, &rqd->tickled)) )
+    {
+        ASSERT(cpumask_cycle(cpu, new->vcpu->cpu_hard_affinity) == cpu);
+        SCHED_STAT_CRANK(tickled_idle_cpu_excl);
+        ipid = cpu;
+        goto tickle;
+    }
+
     for_each_affinity_balance_step( bs )
     {
         /* Just skip first step, if we don't have a soft affinity */
@@ -2971,6 +2997,22 @@ csched2_dom_cntl(
     return rc;
 }
 
+static void
+csched2_aff_cntl(const struct scheduler *ops, struct vcpu *v,
+                 const cpumask_t *hard, const cpumask_t *soft)
+{
+    struct csched2_vcpu *svc = csched2_vcpu(v);
+
+    if ( !hard )
+        return;
+
+    /* Are we becoming exclusively pinned? */
+    if ( cpumask_weight(hard) == 1 )
+        __set_bit(__CSFLAG_pinned, &svc->flags);
+    else
+        __clear_bit(__CSFLAG_pinned, &svc->flags);
+}
+
 static int csched2_sys_cntl(const struct scheduler *ops,
                             struct xen_sysctl_scheduler_op *sc)
 {
@@ -3999,6 +4041,7 @@ static const struct scheduler sched_credit2_def = {
     .yield          = csched2_vcpu_yield,
 
     .adjust         = csched2_dom_cntl,
+    .adjust_affinity= csched2_aff_cntl,
     .adjust_global  = csched2_sys_cntl,
 
     .pick_cpu       = csched2_cpu_pick,

diff --git a/xen/include/xen/perfc_defn.h b/xen/include/xen/perfc_defn.h
index e8ff5653219c8f896cd8f93fc685515c865c1865..ef6f86b91e83550c139a8432c156b78790bae4c0 100644 (file)
--- a/xen/include/xen/perfc_defn.h
+++ b/xen/include/xen/perfc_defn.h
@@ -32,6 +32,7 @@ PERFCOUNTER(vcpu_wake_runnable,     "sched: vcpu_wake_runnable")
 PERFCOUNTER(vcpu_wake_not_runnable, "sched: vcpu_wake_not_runnable")
 PERFCOUNTER(tickled_no_cpu,         "sched: tickled_no_cpu")
 PERFCOUNTER(tickled_idle_cpu,       "sched: tickled_idle_cpu")
+PERFCOUNTER(tickled_idle_cpu_excl,  "sched: tickled_idle_cpu_exclusive")
 PERFCOUNTER(tickled_busy_cpu,       "sched: tickled_busy_cpu")
 PERFCOUNTER(vcpu_check,             "sched: vcpu_check")