[Qemu-devel] [PATCH v2 2/2] ivshmem: use irqfd to interrupt among VMs

[Liu Ping Fan <qemulist@gmail.com>]

From: Liu Ping Fan <pingfank@linux.vnet.ibm.com>

Using irqfd, so we can avoid switch between kernel and user when
VMs interrupts each other.

Signed-off-by: Liu Ping Fan <pingfank@linux.vnet.ibm.com>
Signed-off-by: Cam Macdonell <cam@cs.ualberta.ca>
---
 hw/ivshmem.c |   54 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 53 insertions(+), 1 deletions(-)

diff --git a/hw/ivshmem.c b/hw/ivshmem.c
index 7c8630c..b394b07 100644
--- a/hw/ivshmem.c
+++ b/hw/ivshmem.c
@@ -19,6 +19,7 @@
 #include "hw.h"
 #include "pc.h"
 #include "pci.h"
+#include "msi.h"
 #include "msix.h"
 #include "kvm.h"
 #include "migration.h"
@@ -83,6 +84,7 @@ typedef struct IVShmemState {
     uint32_t vectors;
     uint32_t features;
     EventfdEntry *eventfd_table;
+    int *vector_virqs;
 
     Error *migration_blocker;
 
@@ -625,16 +627,62 @@ static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
     return 0;
 }
 
+static int ivshmem_vector_use(PCIDevice *dev, unsigned vector,
+                                     MSIMessage msg)
+{
+    IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
+    int virq;
+    EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
+
+    virq = kvm_irqchip_add_msi_route(kvm_state, msg);
+    if (virq >= 0 && kvm_irqchip_add_irqfd_notifier(kvm_state, n, virq) >= 0) {
+        s->vector_virqs[vector] = virq;
+        qemu_chr_add_handlers(s->eventfd_chr[vector], NULL, NULL, NULL, NULL);
+    } else if (virq >= 0) {
+        kvm_irqchip_release_virq(kvm_state, virq);
+        error_report("ivshmem, can not setup irqfd\n");
+        return -1;
+    } else {
+        error_report("ivshmem, no enough msi route to setup irqfd\n");
+        return -1;
+    }
+
+    return 0;
+}
+
+static void ivshmem_vector_release(PCIDevice *dev, unsigned vector)
+{
+    IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
+    EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
+    int virq = s->vector_virqs[vector];
+
+    if (s->vector_virqs[vector] >= 0) {
+        kvm_irqchip_remove_irqfd_notifier(kvm_state, n, virq);
+        kvm_irqchip_release_virq(kvm_state, virq);
+        s->vector_virqs[vector] = -1;
+    }
+}
+
 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 				 uint32_t val, int len)
 {
+    bool is_enabled, was_enabled = msi_enabled(pci_dev);
+
     pci_default_write_config(pci_dev, address, val, len);
+    is_enabled = msix_enabled(pci_dev);
+    if (!was_enabled && is_enabled) {
+        msix_set_vector_notifiers(pci_dev, ivshmem_vector_use,
+            ivshmem_vector_release);
+    } else if (was_enabled && !is_enabled) {
+        msix_unset_vector_notifiers(pci_dev);
+    }
 }
 
 static int pci_ivshmem_init(PCIDevice *dev)
 {
     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
     uint8_t *pci_conf;
+    int i;
 
     if (s->sizearg == NULL)
         s->ivshmem_size = 4 << 20; /* 4 MB default */
@@ -758,7 +806,10 @@ static int pci_ivshmem_init(PCIDevice *dev)
     }
 
     s->dev.config_write = ivshmem_write_config;
-
+    s->vector_virqs = g_new0(int, s->vectors);
+    for (i = 0; i < s->vectors; i++) {
+        s->vector_virqs[i] = -1;
+    }
     return 0;
 }
 
@@ -770,6 +821,7 @@ static void pci_ivshmem_uninit(PCIDevice *dev)
         migrate_del_blocker(s->migration_blocker);
         error_free(s->migration_blocker);
     }
+    g_free(s->vector_virqs);
 
     memory_region_destroy(&s->ivshmem_mmio);
     memory_region_del_subregion(&s->bar, &s->ivshmem);
-- 
1.7.4.4