kvm/irqchip: Speed up KVM_SET_GSI_ROUTING

When starting lots of dataplane devices the bootup takes very long on
Christian's s390 with irqfd patches. With larger setups he is even
able to trigger some timeouts in some components.  Turns out that the
KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec)
when having multiple CPUs.  This is caused by the  synchronize_rcu and
the HZ=100 of s390.  By changing the code to use a private srcu we can
speed things up.  This patch reduces the boot time till mounting root
from 8 to 2 seconds on my s390 guest with 100 disks.

Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu
are fine because they do not have lockdep checks (hlist_for_each_entry_rcu
uses rcu_dereference_raw rather than rcu_dereference, and write-sides
do not do rcu lockdep at all).

Note that we're hardly relying on the "sleepable" part of srcu.  We just
want SRCU's faster detection of grace periods.

Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS
and RR.  The difference between results "before" and "after" the patch
has mean -0.2% and standard deviation 0.6%.  Using a paired t-test on the
data points says that there is a 2.5% probability that the patch is the
cause of the performance difference (rather than a random fluctuation).

(Restricting the t-test to RR, which is the most likely to be affected,
changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8%
probability that the numbers actually say something about the patch.
The probability increases mostly because there are fewer data points).

Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

4 files changed, 50 insertions, 39 deletions

diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
index 912ec5a95e2c..20c3af7692c5 100644
--- a/virt/kvm/eventfd.c
+++ b/virt/kvm/eventfd.c
@@ -31,6 +31,7 @@
 #include <linux/list.h>
 #include <linux/eventfd.h>
 #include <linux/kernel.h>
+#include <linux/srcu.h>
 #include <linux/slab.h>
 
 #include "iodev.h"
@@ -118,19 +119,22 @@ static void
 irqfd_resampler_ack(struct kvm_irq_ack_notifier *kian)
 {
         struct _irqfd_resampler *resampler;
+        struct kvm *kvm;
         struct _irqfd *irqfd;
+        int idx;
 
         resampler = container_of(kian, struct _irqfd_resampler, notifier);
+        kvm = resampler->kvm;
 
-        kvm_set_irq(resampler->kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
+        kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
                     resampler->notifier.gsi, 0, false);
 
-        rcu_read_lock();
+        idx = srcu_read_lock(&kvm->irq_srcu);
 
         list_for_each_entry_rcu(irqfd, &resampler->list, resampler_link)
                 eventfd_signal(irqfd->resamplefd, 1);
 
-        rcu_read_unlock();
+        srcu_read_unlock(&kvm->irq_srcu, idx);
 }
 
 static void
@@ -142,7 +146,7 @@ irqfd_resampler_shutdown(struct _irqfd *irqfd)
         mutex_lock(&kvm->irqfds.resampler_lock);
 
         list_del_rcu(&irqfd->resampler_link);
-        synchronize_rcu();
+        synchronize_srcu(&kvm->irq_srcu);
 
         if (list_empty(&resampler->list)) {
                 list_del(&resampler->link);
@@ -221,17 +225,18 @@ irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key)
         unsigned long flags = (unsigned long)key;
         struct kvm_kernel_irq_routing_entry *irq;
         struct kvm *kvm = irqfd->kvm;
+        int idx;
 
         if (flags & POLLIN) {
-                rcu_read_lock();
-                irq = rcu_dereference(irqfd->irq_entry);
+                idx = srcu_read_lock(&kvm->irq_srcu);
+                irq = srcu_dereference(irqfd->irq_entry, &kvm->irq_srcu);
                 /* An event has been signaled, inject an interrupt */
                 if (irq)
                         kvm_set_msi(irq, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1,
                                         false);
                 else
                         schedule_work(&irqfd->inject);
-                rcu_read_unlock();
+                srcu_read_unlock(&kvm->irq_srcu, idx);
         }
 
         if (flags & POLLHUP) {
@@ -363,7 +368,7 @@ kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args)
                 }
 
                 list_add_rcu(&irqfd->resampler_link, &irqfd->resampler->list);
-                synchronize_rcu();
+                synchronize_srcu(&kvm->irq_srcu);
 
                 mutex_unlock(&kvm->irqfds.resampler_lock);
         }
@@ -465,7 +470,7 @@ kvm_irqfd_deassign(struct kvm *kvm, struct kvm_irqfd *args)
                          * another thread calls kvm_irq_routing_update before
                          * we flush workqueue below (we synchronize with
                          * kvm_irq_routing_update using irqfds.lock).
-                         * It is paired with synchronize_rcu done by caller
+                         * It is paired with synchronize_srcu done by caller
                          * of that function.
                          */
                         rcu_assign_pointer(irqfd->irq_entry, NULL);
@@ -524,7 +529,7 @@ kvm_irqfd_release(struct kvm *kvm)
 
 /*
  * Change irq_routing and irqfd.
- * Caller must invoke synchronize_rcu afterwards.
+ * Caller must invoke synchronize_srcu(&kvm->irq_srcu) afterwards.
  */
 void kvm_irq_routing_update(struct kvm *kvm,
                             struct kvm_irq_routing_table *irq_rt)
diff --git a/virt/kvm/irq_comm.c b/virt/kvm/irq_comm.c
index e2e6b4473a96..ced4a542a031 100644
--- a/virt/kvm/irq_comm.c
+++ b/virt/kvm/irq_comm.c
@@ -163,6 +163,7 @@ int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level)
         struct kvm_kernel_irq_routing_entry *e;
         int ret = -EINVAL;
         struct kvm_irq_routing_table *irq_rt;
+        int idx;
 
         trace_kvm_set_irq(irq, level, irq_source_id);
 
@@ -174,8 +175,8 @@ int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level)
          * Since there's no easy way to do this, we only support injecting MSI
          * which is limited to 1:1 GSI mapping.
          */
-        rcu_read_lock();
-        irq_rt = rcu_dereference(kvm->irq_routing);
+        idx = srcu_read_lock(&kvm->irq_srcu);
+        irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
         if (irq < irq_rt->nr_rt_entries)
                 hlist_for_each_entry(e, &irq_rt->map[irq], link) {
                         if (likely(e->type == KVM_IRQ_ROUTING_MSI))
@@ -184,7 +185,7 @@ int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level)
                                 ret = -EWOULDBLOCK;
                         break;
                 }
-        rcu_read_unlock();
+        srcu_read_unlock(&kvm->irq_srcu, idx);
         return ret;
 }
 
@@ -253,22 +254,22 @@ void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
         mutex_lock(&kvm->irq_lock);
         hlist_del_rcu(&kimn->link);
         mutex_unlock(&kvm->irq_lock);
-        synchronize_rcu();
+        synchronize_srcu(&kvm->irq_srcu);
 }
 
 void kvm_fire_mask_notifiers(struct kvm *kvm, unsigned irqchip, unsigned pin,
                              bool mask)
 {
         struct kvm_irq_mask_notifier *kimn;
-        int gsi;
+        int idx, gsi;
 
-        rcu_read_lock();
-        gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
+        idx = srcu_read_lock(&kvm->irq_srcu);
+        gsi = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu)->chip[irqchip][pin];
         if (gsi != -1)
                 hlist_for_each_entry_rcu(kimn, &kvm->mask_notifier_list, link)
                         if (kimn->irq == gsi)
                                 kimn->func(kimn, mask);