From 701189e31140a7c82ec02a7f4ca632cfd6a8559d Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:20:39 +0300
Subject: hyperv: factor out arch-independent API into hw/hyperv

A significant part of hyperv.c is not actually tied to x86, and can
be moved to hw/.

This will allow to maintain most of Hyper-V and VMBus
target-independent, and to avoid conflicts with inclusion of
arch-specific headers down the road in VMBus implementation.

Also this stuff can now be opt-out with CONFIG_HYPERV.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082041.29380-4-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c | 138 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 138 insertions(+)
 create mode 100644 hw/hyperv/hyperv.c

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
new file mode 100644
index 0000000000..97db87561e
--- /dev/null
+++ b/hw/hyperv/hyperv.c
@@ -0,0 +1,138 @@
+/*
+ * Hyper-V guest/hypervisor interaction
+ *
+ * Copyright (c) 2015-2018 Virtuozzo International GmbH.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/main-loop.h"
+#include "sysemu/kvm.h"
+#include "hw/hyperv/hyperv.h"
+
+struct HvSintRoute {
+    uint32_t sint;
+    CPUState *cs;
+    int gsi;
+    EventNotifier sint_set_notifier;
+    EventNotifier sint_ack_notifier;
+    HvSintAckClb sint_ack_clb;
+    void *sint_ack_clb_data;
+    unsigned refcount;
+};
+
+static CPUState *hyperv_find_vcpu(uint32_t vp_index)
+{
+    CPUState *cs = qemu_get_cpu(vp_index);
+    assert(hyperv_vp_index(cs) == vp_index);
+    return cs;
+}
+
+static void kvm_hv_sint_ack_handler(EventNotifier *notifier)
+{
+    HvSintRoute *sint_route = container_of(notifier, HvSintRoute,
+                                           sint_ack_notifier);
+    event_notifier_test_and_clear(notifier);
+    sint_route->sint_ack_clb(sint_route->sint_ack_clb_data);
+}
+
+HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
+                                   HvSintAckClb sint_ack_clb,
+                                   void *sint_ack_clb_data)
+{
+    HvSintRoute *sint_route;
+    EventNotifier *ack_notifier;
+    int r, gsi;
+    CPUState *cs;
+
+    cs = hyperv_find_vcpu(vp_index);
+    if (!cs) {
+        return NULL;
+    }
+
+    sint_route = g_new0(HvSintRoute, 1);
+    r = event_notifier_init(&sint_route->sint_set_notifier, false);
+    if (r) {
+        goto err;
+    }
+
+    ack_notifier = sint_ack_clb ? &sint_route->sint_ack_notifier : NULL;
+    if (ack_notifier) {
+        r = event_notifier_init(ack_notifier, false);
+        if (r) {
+            goto err_sint_set_notifier;
+        }
+
+        event_notifier_set_handler(ack_notifier, kvm_hv_sint_ack_handler);
+    }
+
+    gsi = kvm_irqchip_add_hv_sint_route(kvm_state, vp_index, sint);
+    if (gsi < 0) {
+        goto err_gsi;
+    }
+
+    r = kvm_irqchip_add_irqfd_notifier_gsi(kvm_state,
+                                           &sint_route->sint_set_notifier,
+                                           ack_notifier, gsi);
+    if (r) {
+        goto err_irqfd;
+    }
+    sint_route->gsi = gsi;
+    sint_route->sint_ack_clb = sint_ack_clb;
+    sint_route->sint_ack_clb_data = sint_ack_clb_data;
+    sint_route->cs = cs;
+    sint_route->sint = sint;
+    sint_route->refcount = 1;
+
+    return sint_route;
+
+err_irqfd:
+    kvm_irqchip_release_virq(kvm_state, gsi);
+err_gsi:
+    if (ack_notifier) {
+        event_notifier_set_handler(ack_notifier, NULL);
+        event_notifier_cleanup(ack_notifier);
+    }
+err_sint_set_notifier:
+    event_notifier_cleanup(&sint_route->sint_set_notifier);
+err:
+    g_free(sint_route);
+
+    return NULL;
+}
+
+void hyperv_sint_route_ref(HvSintRoute *sint_route)
+{
+    sint_route->refcount++;
+}
+
+void hyperv_sint_route_unref(HvSintRoute *sint_route)
+{
+    if (!sint_route) {
+        return;
+    }
+
+    assert(sint_route->refcount > 0);
+
+    if (--sint_route->refcount) {
+        return;
+    }
+
+    kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state,
+                                          &sint_route->sint_set_notifier,
+                                          sint_route->gsi);
+    kvm_irqchip_release_virq(kvm_state, sint_route->gsi);
+    if (sint_route->sint_ack_clb) {
+        event_notifier_set_handler(&sint_route->sint_ack_notifier, NULL);
+        event_notifier_cleanup(&sint_route->sint_ack_notifier);
+    }
+    event_notifier_cleanup(&sint_route->sint_set_notifier);
+    g_free(sint_route);
+}
+
+int hyperv_sint_route_set_sint(HvSintRoute *sint_route)
+{
+    return event_notifier_set(&sint_route->sint_set_notifier);
+}
-- 
cgit v1.2.3-55-g7522


From 606c34bfd57a0ecda67b395bea022bb307a5384e Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:09 +0300
Subject: hyperv: qom-ify SynIC

Make Hyper-V SynIC a device which is attached as a child to a CPU.  For
now it only makes SynIC visibile in the qom hierarchy, and maintains its
internal fields in sync with the respecitve msrs of the parent cpu (the
fields will be used in followup patches).

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-3-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c         | 101 ++++++++++++++++++++++++++++++++++++++++++++-
 include/hw/hyperv/hyperv.h |   5 +++
 target/i386/hyperv-stub.c  |  13 ++++++
 target/i386/hyperv.c       |  25 +++++++++++
 target/i386/hyperv.h       |   4 ++
 target/i386/kvm.c          |   9 ++++
 target/i386/machine.c      |   9 ++++
 7 files changed, 164 insertions(+), 2 deletions(-)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index 97db87561e..3d6f044282 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -9,12 +9,103 @@
 
 #include "qemu/osdep.h"
 #include "qemu/main-loop.h"
+#include "qapi/error.h"
 #include "sysemu/kvm.h"
 #include "hw/hyperv/hyperv.h"
 
+typedef struct SynICState {
+    DeviceState parent_obj;
+
+    CPUState *cs;
+
+    bool enabled;
+    hwaddr msg_page_addr;
+    hwaddr event_page_addr;
+} SynICState;
+
+#define TYPE_SYNIC "hyperv-synic"
+#define SYNIC(obj) OBJECT_CHECK(SynICState, (obj), TYPE_SYNIC)
+
+static SynICState *get_synic(CPUState *cs)
+{
+    return SYNIC(object_resolve_path_component(OBJECT(cs), "synic"));
+}
+
+static void synic_update(SynICState *synic, bool enable,
+                         hwaddr msg_page_addr, hwaddr event_page_addr)
+{
+
+    synic->enabled = enable;
+    synic->msg_page_addr = msg_page_addr;
+    synic->event_page_addr = event_page_addr;
+}
+
+void hyperv_synic_update(CPUState *cs, bool enable,
+                         hwaddr msg_page_addr, hwaddr event_page_addr)
+{
+    SynICState *synic = get_synic(cs);
+
+    if (!synic) {
+        return;
+    }
+
+    synic_update(synic, enable, msg_page_addr, event_page_addr);
+}
+
+static void synic_realize(DeviceState *dev, Error **errp)
+{
+}
+
+static void synic_reset(DeviceState *dev)
+{
+    SynICState *synic = SYNIC(dev);
+    synic_update(synic, false, 0, 0);
+}
+
+static void synic_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+
+    dc->realize = synic_realize;
+    dc->reset = synic_reset;
+    dc->user_creatable = false;
+}
+
+void hyperv_synic_add(CPUState *cs)
+{
+    Object *obj;
+    SynICState *synic;
+
+    obj = object_new(TYPE_SYNIC);
+    synic = SYNIC(obj);
+    synic->cs = cs;
+    object_property_add_child(OBJECT(cs), "synic", obj, &error_abort);
+    object_unref(obj);
+    object_property_set_bool(obj, true, "realized", &error_abort);
+}
+
+void hyperv_synic_reset(CPUState *cs)
+{
+    device_reset(DEVICE(get_synic(cs)));
+}
+
+static const TypeInfo synic_type_info = {
+    .name = TYPE_SYNIC,
+    .parent = TYPE_DEVICE,
+    .instance_size = sizeof(SynICState),
+    .class_init = synic_class_init,
+};
+
+static void synic_register_types(void)
+{
+    type_register_static(&synic_type_info);
+}
+
+type_init(synic_register_types)
+
 struct HvSintRoute {
     uint32_t sint;
-    CPUState *cs;
+    SynICState *synic;
     int gsi;
     EventNotifier sint_set_notifier;
     EventNotifier sint_ack_notifier;
@@ -46,12 +137,18 @@ HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
     EventNotifier *ack_notifier;
     int r, gsi;
     CPUState *cs;
+    SynICState *synic;
 
     cs = hyperv_find_vcpu(vp_index);
     if (!cs) {
         return NULL;
     }
 
+    synic = get_synic(cs);
+    if (!synic) {
+        return NULL;
+    }
+
     sint_route = g_new0(HvSintRoute, 1);
     r = event_notifier_init(&sint_route->sint_set_notifier, false);
     if (r) {
@@ -82,7 +179,7 @@ HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
     sint_route->gsi = gsi;
     sint_route->sint_ack_clb = sint_ack_clb;
     sint_route->sint_ack_clb_data = sint_ack_clb_data;
-    sint_route->cs = cs;
+    sint_route->synic = synic;
     sint_route->sint = sint;
     sint_route->refcount = 1;
 
diff --git a/include/hw/hyperv/hyperv.h b/include/hw/hyperv/hyperv.h
index d6c8d78353..6fba4762c8 100644
--- a/include/hw/hyperv/hyperv.h
+++ b/include/hw/hyperv/hyperv.h
@@ -28,4 +28,9 @@ static inline uint32_t hyperv_vp_index(CPUState *cs)
     return cs->cpu_index;
 }
 
+void hyperv_synic_add(CPUState *cs);
+void hyperv_synic_reset(CPUState *cs);
+void hyperv_synic_update(CPUState *cs, bool enable,
+                         hwaddr msg_page_addr, hwaddr event_page_addr);
+
 #endif
diff --git a/target/i386/hyperv-stub.c b/target/i386/hyperv-stub.c
index 5919ba851c..fe548cbae2 100644
--- a/target/i386/hyperv-stub.c
+++ b/target/i386/hyperv-stub.c
@@ -33,3 +33,16 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
     }
 }
 #endif
+
+int hyperv_x86_synic_add(X86CPU *cpu)
+{
+    return -ENOSYS;
+}
+
+void hyperv_x86_synic_reset(X86CPU *cpu)
+{
+}
+
+void hyperv_x86_synic_update(X86CPU *cpu)
+{
+}
diff --git a/target/i386/hyperv.c b/target/i386/hyperv.c
index 1eac727774..0216735d67 100644
--- a/target/i386/hyperv.c
+++ b/target/i386/hyperv.c
@@ -16,6 +16,28 @@
 #include "hw/hyperv/hyperv.h"
 #include "hyperv-proto.h"
 
+int hyperv_x86_synic_add(X86CPU *cpu)
+{
+    hyperv_synic_add(CPU(cpu));
+    return 0;
+}
+
+void hyperv_x86_synic_reset(X86CPU *cpu)
+{
+    hyperv_synic_reset(CPU(cpu));
+}
+
+void hyperv_x86_synic_update(X86CPU *cpu)
+{
+    CPUX86State *env = &cpu->env;
+    bool enable = env->msr_hv_synic_control & HV_SYNIC_ENABLE;
+    hwaddr msg_page_addr = (env->msr_hv_synic_msg_page & HV_SIMP_ENABLE) ?
+        (env->msr_hv_synic_msg_page & TARGET_PAGE_MASK) : 0;
+    hwaddr event_page_addr = (env->msr_hv_synic_evt_page & HV_SIEFP_ENABLE) ?
+        (env->msr_hv_synic_evt_page & TARGET_PAGE_MASK) : 0;
+    hyperv_synic_update(CPU(cpu), enable, msg_page_addr, event_page_addr);
+}
+
 int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
 {
     CPUX86State *env = &cpu->env;
@@ -44,6 +66,9 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
         default:
             return -1;
         }
+
+        hyperv_x86_synic_update(cpu);
+
         return 0;
     case KVM_EXIT_HYPERV_HCALL: {
         uint16_t code;
diff --git a/target/i386/hyperv.h b/target/i386/hyperv.h
index f0a27c3d73..67543296c3 100644
--- a/target/i386/hyperv.h
+++ b/target/i386/hyperv.h
@@ -22,4 +22,8 @@
 int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit);
 #endif
 
+int hyperv_x86_synic_add(X86CPU *cpu);
+void hyperv_x86_synic_reset(X86CPU *cpu);
+void hyperv_x86_synic_update(X86CPU *cpu);
+
 #endif
diff --git a/target/i386/kvm.c b/target/i386/kvm.c
index 2e5b9f63eb..cf6270ae39 100644
--- a/target/i386/kvm.c
+++ b/target/i386/kvm.c
@@ -790,6 +790,13 @@ static int hyperv_init_vcpu(X86CPU *cpu)
                          strerror(-ret));
             return ret;
         }
+
+        ret = hyperv_x86_synic_add(cpu);
+        if (ret < 0) {
+            error_report("failed to create HyperV SynIC: %s",
+                         strerror(-ret));
+            return ret;
+        }
     }
 
     return 0;
@@ -1250,6 +1257,8 @@ void kvm_arch_reset_vcpu(X86CPU *cpu)
         for (i = 0; i < ARRAY_SIZE(env->msr_hv_synic_sint); i++) {
             env->msr_hv_synic_sint[i] = HV_SINT_MASKED;
         }
+
+        hyperv_x86_synic_reset(cpu);
     }
 }
 
diff --git a/target/i386/machine.c b/target/i386/machine.c
index 084c2c73a8..225b5d433b 100644
--- a/target/i386/machine.c
+++ b/target/i386/machine.c
@@ -7,6 +7,7 @@
 #include "hw/i386/pc.h"
 #include "hw/isa/isa.h"
 #include "migration/cpu.h"
+#include "hyperv.h"
 
 #include "sysemu/kvm.h"
 
@@ -672,11 +673,19 @@ static bool hyperv_synic_enable_needed(void *opaque)
     return false;
 }
 
+static int hyperv_synic_post_load(void *opaque, int version_id)
+{
+    X86CPU *cpu = opaque;
+    hyperv_x86_synic_update(cpu);
+    return 0;
+}
+
 static const VMStateDescription vmstate_msr_hyperv_synic = {
     .name = "cpu/msr_hyperv_synic",
     .version_id = 1,
     .minimum_version_id = 1,
     .needed = hyperv_synic_enable_needed,
+    .post_load = hyperv_synic_post_load,
     .fields = (VMStateField[]) {
         VMSTATE_UINT64(env.msr_hv_synic_control, X86CPU),
         VMSTATE_UINT64(env.msr_hv_synic_evt_page, X86CPU),
-- 
cgit v1.2.3-55-g7522


From 267e071bd6d675c15e7ffbf8aaf44d488ebd5c83 Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:11 +0300
Subject: hyperv: make overlay pages for SynIC

Per Hyper-V spec, SynIC message and event flag pages are to be
implemented as so called overlay pages.  That is, they are owned by the
hypervisor and, when mapped into the guest physical address space,
overlay the guest physical pages such that

1) the overlaid guest page becomes invisible to the guest CPUs until the
   overlay page is turned off
2) the contents of the overlay page is preserved when it's turned off
   and back on, even at a different address; it's only zeroed at vcpu
   reset

This particular nature of SynIC message and event flag pages is ignored
in the current code, and guest physical pages are used directly instead.
This happens to (mostly) work because the actual guests seem not to
depend on the features listed above.

This patch implements those pages as the spec mandates.

Since the extra RAM regions, which introduce migration incompatibility,
are only added at SynIC object creation which only happens when
hyperv_synic_kvm_only == false, no extra compat logic is necessary.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-5-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c   | 51 ++++++++++++++++++++++++++++++++++++++++++++++++---
 target/i386/hyperv.c | 20 ++++++++++++++------
 2 files changed, 62 insertions(+), 9 deletions(-)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index 3d6f044282..70cf129d04 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -10,6 +10,7 @@
 #include "qemu/osdep.h"
 #include "qemu/main-loop.h"
 #include "qapi/error.h"
+#include "exec/address-spaces.h"
 #include "sysemu/kvm.h"
 #include "hw/hyperv/hyperv.h"
 
@@ -21,6 +22,10 @@ typedef struct SynICState {
     bool enabled;
     hwaddr msg_page_addr;
     hwaddr event_page_addr;
+    MemoryRegion msg_page_mr;
+    MemoryRegion event_page_mr;
+    struct hyperv_message_page *msg_page;
+    struct hyperv_event_flags_page *event_page;
 } SynICState;
 
 #define TYPE_SYNIC "hyperv-synic"
@@ -36,8 +41,28 @@ static void synic_update(SynICState *synic, bool enable,
 {
 
     synic->enabled = enable;
-    synic->msg_page_addr = msg_page_addr;
-    synic->event_page_addr = event_page_addr;
+    if (synic->msg_page_addr != msg_page_addr) {
+        if (synic->msg_page_addr) {
+            memory_region_del_subregion(get_system_memory(),
+                                        &synic->msg_page_mr);
+        }
+        if (msg_page_addr) {
+            memory_region_add_subregion(get_system_memory(), msg_page_addr,
+                                        &synic->msg_page_mr);
+        }
+        synic->msg_page_addr = msg_page_addr;
+    }
+    if (synic->event_page_addr != event_page_addr) {
+        if (synic->event_page_addr) {
+            memory_region_del_subregion(get_system_memory(),
+                                        &synic->event_page_mr);
+        }
+        if (event_page_addr) {
+            memory_region_add_subregion(get_system_memory(), event_page_addr,
+                                        &synic->event_page_mr);
+        }
+        synic->event_page_addr = event_page_addr;
+    }
 }
 
 void hyperv_synic_update(CPUState *cs, bool enable,
@@ -54,11 +79,31 @@ void hyperv_synic_update(CPUState *cs, bool enable,
 
 static void synic_realize(DeviceState *dev, Error **errp)
 {
+    Object *obj = OBJECT(dev);
+    SynICState *synic = SYNIC(dev);
+    char *msgp_name, *eventp_name;
+    uint32_t vp_index;
+
+    /* memory region names have to be globally unique */
+    vp_index = hyperv_vp_index(synic->cs);
+    msgp_name = g_strdup_printf("synic-%u-msg-page", vp_index);
+    eventp_name = g_strdup_printf("synic-%u-event-page", vp_index);
+
+    memory_region_init_ram(&synic->msg_page_mr, obj, msgp_name,
+                           sizeof(*synic->msg_page), &error_abort);
+    memory_region_init_ram(&synic->event_page_mr, obj, eventp_name,
+                           sizeof(*synic->event_page), &error_abort);
+    synic->msg_page = memory_region_get_ram_ptr(&synic->msg_page_mr);
+    synic->event_page = memory_region_get_ram_ptr(&synic->event_page_mr);
+
+    g_free(msgp_name);
+    g_free(eventp_name);
 }
-
 static void synic_reset(DeviceState *dev)
 {
     SynICState *synic = SYNIC(dev);
+    memset(synic->msg_page, 0, sizeof(*synic->msg_page));
+    memset(synic->event_page, 0, sizeof(*synic->event_page));
     synic_update(synic, false, 0, 0);
 }
 
diff --git a/target/i386/hyperv.c b/target/i386/hyperv.c
index 0216735d67..3f76c3e266 100644
--- a/target/i386/hyperv.c
+++ b/target/i386/hyperv.c
@@ -12,6 +12,7 @@
  */
 
 #include "qemu/osdep.h"
+#include "qemu/main-loop.h"
 #include "hyperv.h"
 #include "hw/hyperv/hyperv.h"
 #include "hyperv-proto.h"
@@ -38,6 +39,13 @@ void hyperv_x86_synic_update(X86CPU *cpu)
     hyperv_synic_update(CPU(cpu), enable, msg_page_addr, event_page_addr);
 }
 
+static void async_synic_update(CPUState *cs, run_on_cpu_data data)
+{
+    qemu_mutex_lock_iothread();
+    hyperv_x86_synic_update(X86_CPU(cs));
+    qemu_mutex_unlock_iothread();
+}
+
 int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
 {
     CPUX86State *env = &cpu->env;
@@ -48,11 +56,6 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
             return -1;
         }
 
-        /*
-         * For now just track changes in SynIC control and msg/evt pages msr's.
-         * When SynIC messaging/events processing will be added in future
-         * here we will do messages queues flushing and pages remapping.
-         */
         switch (exit->u.synic.msr) {
         case HV_X64_MSR_SCONTROL:
             env->msr_hv_synic_control = exit->u.synic.control;
@@ -67,7 +70,12 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
             return -1;
         }
 
-        hyperv_x86_synic_update(cpu);
+        /*
+         * this will run in this cpu thread before it returns to KVM, but in a
+         * safe environment (i.e. when all cpus are quiescent) -- this is
+         * necessary because memory hierarchy is being changed
+         */
+        async_safe_run_on_cpu(CPU(cpu), async_synic_update, RUN_ON_CPU_NULL);
 
         return 0;
     case KVM_EXIT_HYPERV_HCALL: {
-- 
cgit v1.2.3-55-g7522


From 4cbaf3c13300b79d0386b567630f8e9c91ac5099 Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:12 +0300
Subject: hyperv: add synic message delivery

Add infrastructure to deliver SynIC messages to the SynIC message page.

Note that KVM may also want to deliver (SynIC timer) messages to the
same message slot.

The problem is that the access to a SynIC message slot is controlled by
the value of its .msg_type field which indicates if the slot is being
owned by the hypervisor (zero) or by the guest (non-zero).

This leaves no room for synchronizing multiple concurrent producers.

The simplest way to deal with this for both KVM and QEMU is to only
deliver messages in the vcpu thread.  KVM already does this; this patch
makes it for QEMU, too.

Specifically,

 - add a function for posting messages, which only copies the message
   into the staging buffer if its free, and schedules a work on the
   corresponding vcpu to actually deliver it to the guest slot;

 - instead of a sint ack callback, set up the sint route with a message
   status callback.  This function is called in a bh whenever there are
   updates to the message slot status: either the vcpu made definitive
   progress delivering the message from the staging buffer (succeeded or
   failed) or the guest issued EOM; the status is passed as an argument
   to the callback.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-6-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c         | 162 ++++++++++++++++++++++++++++++++++++++++++---
 include/hw/hyperv/hyperv.h |  18 ++++-
 2 files changed, 166 insertions(+), 14 deletions(-)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index 70cf129d04..654ca4ffc5 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -148,14 +148,51 @@ static void synic_register_types(void)
 
 type_init(synic_register_types)
 
+/*
+ * KVM has its own message producers (SynIC timers).  To guarantee
+ * serialization with both KVM vcpu and the guest cpu, the messages are first
+ * staged in an intermediate area and then posted to the SynIC message page in
+ * the vcpu thread.
+ */
+typedef struct HvSintStagedMessage {
+    /* message content staged by hyperv_post_msg */
+    struct hyperv_message msg;
+    /* callback + data (r/o) to complete the processing in a BH */
+    HvSintMsgCb cb;
+    void *cb_data;
+    /* message posting status filled by cpu_post_msg */
+    int status;
+    /* passing the buck: */
+    enum {
+        /* initial state */
+        HV_STAGED_MSG_FREE,
+        /*
+         * hyperv_post_msg (e.g. in main loop) grabs the staged area (FREE ->
+         * BUSY), copies msg, and schedules cpu_post_msg on the assigned cpu
+         */
+        HV_STAGED_MSG_BUSY,
+        /*
+         * cpu_post_msg (vcpu thread) tries to copy staged msg to msg slot,
+         * notify the guest, records the status, marks the posting done (BUSY
+         * -> POSTED), and schedules sint_msg_bh BH
+         */
+        HV_STAGED_MSG_POSTED,
+        /*
+         * sint_msg_bh (BH) verifies that the posting is done, runs the
+         * callback, and starts over (POSTED -> FREE)
+         */
+    } state;
+} HvSintStagedMessage;
+
 struct HvSintRoute {
     uint32_t sint;
     SynICState *synic;
     int gsi;
     EventNotifier sint_set_notifier;
     EventNotifier sint_ack_notifier;
-    HvSintAckClb sint_ack_clb;
-    void *sint_ack_clb_data;
+
+    HvSintStagedMessage *staged_msg;
+
     unsigned refcount;
 };
 
@@ -166,17 +203,115 @@ static CPUState *hyperv_find_vcpu(uint32_t vp_index)
     return cs;
 }
 
-static void kvm_hv_sint_ack_handler(EventNotifier *notifier)
+/*
+ * BH to complete the processing of a staged message.
+ */
+static void sint_msg_bh(void *opaque)
+{
+    HvSintRoute *sint_route = opaque;
+    HvSintStagedMessage *staged_msg = sint_route->staged_msg;
+
+    if (atomic_read(&staged_msg->state) != HV_STAGED_MSG_POSTED) {
+        /* status nor ready yet (spurious ack from guest?), ignore */
+        return;
+    }
+
+    staged_msg->cb(staged_msg->cb_data, staged_msg->status);
+    staged_msg->status = 0;
+
+    /* staged message processing finished, ready to start over */
+    atomic_set(&staged_msg->state, HV_STAGED_MSG_FREE);
+    /* drop the reference taken in hyperv_post_msg */
+    hyperv_sint_route_unref(sint_route);
+}
+
+/*
+ * Worker to transfer the message from the staging area into the SynIC message
+ * page in vcpu context.
+ */
+static void cpu_post_msg(CPUState *cs, run_on_cpu_data data)
+{
+    HvSintRoute *sint_route = data.host_ptr;
+    HvSintStagedMessage *staged_msg = sint_route->staged_msg;
+    SynICState *synic = sint_route->synic;
+    struct hyperv_message *dst_msg;
+    bool wait_for_sint_ack = false;
+
+    assert(staged_msg->state == HV_STAGED_MSG_BUSY);
+
+    if (!synic->enabled || !synic->msg_page_addr) {
+        staged_msg->status = -ENXIO;
+        goto posted;
+    }
+
+    dst_msg = &synic->msg_page->slot[sint_route->sint];
+
+    if (dst_msg->header.message_type != HV_MESSAGE_NONE) {
+        dst_msg->header.message_flags |= HV_MESSAGE_FLAG_PENDING;
+        staged_msg->status = -EAGAIN;
+        wait_for_sint_ack = true;
+    } else {
+        memcpy(dst_msg, &staged_msg->msg, sizeof(*dst_msg));
+        staged_msg->status = hyperv_sint_route_set_sint(sint_route);
+    }
+
+    memory_region_set_dirty(&synic->msg_page_mr, 0, sizeof(*synic->msg_page));
+
+posted:
+    atomic_set(&staged_msg->state, HV_STAGED_MSG_POSTED);
+    /*
+     * Notify the msg originator of the progress made; if the slot was busy we
+     * set msg_pending flag in it so it will be the guest who will do EOM and
+     * trigger the notification from KVM via sint_ack_notifier
+     */
+    if (!wait_for_sint_ack) {
+        aio_bh_schedule_oneshot(qemu_get_aio_context(), sint_msg_bh,
+                                sint_route);
+    }
+}
+
+/*
+ * Post a Hyper-V message to the staging area, for delivery to guest in the
+ * vcpu thread.
+ */
+int hyperv_post_msg(HvSintRoute *sint_route, struct hyperv_message *src_msg)
+{
+    HvSintStagedMessage *staged_msg = sint_route->staged_msg;
+
+    assert(staged_msg);
+
+    /* grab the staging area */
+    if (atomic_cmpxchg(&staged_msg->state, HV_STAGED_MSG_FREE,
+                       HV_STAGED_MSG_BUSY) != HV_STAGED_MSG_FREE) {
+        return -EAGAIN;
+    }
+
+    memcpy(&staged_msg->msg, src_msg, sizeof(*src_msg));
+
+    /* hold a reference on sint_route until the callback is finished */
+    hyperv_sint_route_ref(sint_route);
+
+    /* schedule message posting attempt in vcpu thread */
+    async_run_on_cpu(sint_route->synic->cs, cpu_post_msg,
+                     RUN_ON_CPU_HOST_PTR(sint_route));
+    return 0;
+}
+
+static void sint_ack_handler(EventNotifier *notifier)
 {
     HvSintRoute *sint_route = container_of(notifier, HvSintRoute,
                                            sint_ack_notifier);
     event_notifier_test_and_clear(notifier);
-    sint_route->sint_ack_clb(sint_route->sint_ack_clb_data);
+
+    /*
+     * the guest consumed the previous message so complete the current one with
+     * -EAGAIN and let the msg originator retry
+     */
+    aio_bh_schedule_oneshot(qemu_get_aio_context(), sint_msg_bh, sint_route);
 }
 
 HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
-                                   HvSintAckClb sint_ack_clb,
-                                   void *sint_ack_clb_data)
+                                   HvSintMsgCb cb, void *cb_data)
 {
     HvSintRoute *sint_route;
     EventNotifier *ack_notifier;
@@ -200,14 +335,19 @@ HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
         goto err;
     }
 
-    ack_notifier = sint_ack_clb ? &sint_route->sint_ack_notifier : NULL;
+
+    ack_notifier = cb ? &sint_route->sint_ack_notifier : NULL;
     if (ack_notifier) {
+        sint_route->staged_msg = g_new0(HvSintStagedMessage, 1);
+        sint_route->staged_msg->cb = cb;
+        sint_route->staged_msg->cb_data = cb_data;
+
         r = event_notifier_init(ack_notifier, false);
         if (r) {
             goto err_sint_set_notifier;
         }
 
-        event_notifier_set_handler(ack_notifier, kvm_hv_sint_ack_handler);
+        event_notifier_set_handler(ack_notifier, sint_ack_handler);
     }
 
     gsi = kvm_irqchip_add_hv_sint_route(kvm_state, vp_index, sint);
@@ -222,8 +362,6 @@ HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
         goto err_irqfd;
     }
     sint_route->gsi = gsi;
-    sint_route->sint_ack_clb = sint_ack_clb;
-    sint_route->sint_ack_clb_data = sint_ack_clb_data;
     sint_route->synic = synic;
     sint_route->sint = sint;
     sint_route->refcount = 1;
@@ -236,6 +374,7 @@ err_gsi:
     if (ack_notifier) {
         event_notifier_set_handler(ack_notifier, NULL);
         event_notifier_cleanup(ack_notifier);
+        g_free(sint_route->staged_msg);
     }
 err_sint_set_notifier:
     event_notifier_cleanup(&sint_route->sint_set_notifier);
@@ -266,9 +405,10 @@ void hyperv_sint_route_unref(HvSintRoute *sint_route)
                                           &sint_route->sint_set_notifier,
                                           sint_route->gsi);
     kvm_irqchip_release_virq(kvm_state, sint_route->gsi);
-    if (sint_route->sint_ack_clb) {
+    if (sint_route->staged_msg) {
         event_notifier_set_handler(&sint_route->sint_ack_notifier, NULL);
         event_notifier_cleanup(&sint_route->sint_ack_notifier);
+        g_free(sint_route->staged_msg);
     }
     event_notifier_cleanup(&sint_route->sint_set_notifier);
     g_free(sint_route);
diff --git a/include/hw/hyperv/hyperv.h b/include/hw/hyperv/hyperv.h
index 6fba4762c8..82d561fc88 100644
--- a/include/hw/hyperv/hyperv.h
+++ b/include/hw/hyperv/hyperv.h
@@ -11,18 +11,30 @@
 #define HW_HYPERV_HYPERV_H
 
 #include "cpu-qom.h"
+#include "hw/hyperv/hyperv-proto.h"
 
 typedef struct HvSintRoute HvSintRoute;
-typedef void (*HvSintAckClb)(void *data);
+
+/*
+ * Callback executed in a bottom-half when the status of posting the message
+ * becomes known, before unblocking the connection for further messages
+ */
+typedef void (*HvSintMsgCb)(void *data, int status);
 
 HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
-                                   HvSintAckClb sint_ack_clb,
-                                   void *sint_ack_clb_data);
+                                   HvSintMsgCb cb, void *cb_data);
 void hyperv_sint_route_ref(HvSintRoute *sint_route);
 void hyperv_sint_route_unref(HvSintRoute *sint_route);
 
 int hyperv_sint_route_set_sint(HvSintRoute *sint_route);
 
+/*
+ * Submit a message to be posted in vcpu context.  If the submission succeeds,
+ * the status of posting the message is reported via the callback associated
+ * with the @sint_route; until then no more messages are accepted.
+ */
+int hyperv_post_msg(HvSintRoute *sint_route, struct hyperv_message *msg);
+
 static inline uint32_t hyperv_vp_index(CPUState *cs)
 {
     return cs->cpu_index;
-- 
cgit v1.2.3-55-g7522


From f5642f8b458ba578c1ea94b9ad773e1e5c6cb615 Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:13 +0300
Subject: hyperv: add synic event flag signaling

Add infrastructure to signal SynIC event flags by atomically setting the
corresponding bit in the event flags page and firing a SINT if
necessary.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-7-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c         | 32 ++++++++++++++++++++++++++++++++
 include/hw/hyperv/hyperv.h |  4 ++++
 2 files changed, 36 insertions(+)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index 654ca4ffc5..2b0e593bf9 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -12,6 +12,7 @@
 #include "qapi/error.h"
 #include "exec/address-spaces.h"
 #include "sysemu/kvm.h"
+#include "qemu/bitops.h"
 #include "hw/hyperv/hyperv.h"
 
 typedef struct SynICState {
@@ -310,6 +311,37 @@ static void sint_ack_handler(EventNotifier *notifier)
     aio_bh_schedule_oneshot(qemu_get_aio_context(), sint_msg_bh, sint_route);
 }
 
+/*
+ * Set given event flag for a given sint on a given vcpu, and signal the sint.
+ */
+int hyperv_set_event_flag(HvSintRoute *sint_route, unsigned eventno)
+{
+    int ret;
+    SynICState *synic = sint_route->synic;
+    unsigned long *flags, set_mask;
+    unsigned set_idx;
+
+    if (eventno > HV_EVENT_FLAGS_COUNT) {
+        return -EINVAL;
+    }
+    if (!synic->enabled || !synic->event_page_addr) {
+        return -ENXIO;
+    }
+
+    set_idx = BIT_WORD(eventno);
+    set_mask = BIT_MASK(eventno);
+    flags = synic->event_page->slot[sint_route->sint].flags;
+
+    if ((atomic_fetch_or(&flags[set_idx], set_mask) & set_mask) != set_mask) {
+        memory_region_set_dirty(&synic->event_page_mr, 0,
+                                sizeof(*synic->event_page));
+        ret = hyperv_sint_route_set_sint(sint_route);
+    } else {
+        ret = 0;
+    }
+    return ret;
+}
+
 HvSintRoute *hyperv_sint_route_new(uint32_t vp_index, uint32_t sint,
                                    HvSintMsgCb cb, void *cb_data)
 {
diff --git a/include/hw/hyperv/hyperv.h b/include/hw/hyperv/hyperv.h
index 82d561fc88..757c85eb8f 100644
--- a/include/hw/hyperv/hyperv.h
+++ b/include/hw/hyperv/hyperv.h
@@ -34,6 +34,10 @@ int hyperv_sint_route_set_sint(HvSintRoute *sint_route);
  * with the @sint_route; until then no more messages are accepted.
  */
 int hyperv_post_msg(HvSintRoute *sint_route, struct hyperv_message *msg);
+/*
+ * Set event flag @eventno, and signal the SINT if the flag has changed.
+ */
+int hyperv_set_event_flag(HvSintRoute *sint_route, unsigned eventno);
 
 static inline uint32_t hyperv_vp_index(CPUState *cs)
 {
-- 
cgit v1.2.3-55-g7522


From e6ea9f45b72fe83d49adda948ff397dafc00c68f Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:14 +0300
Subject: hyperv: process SIGNAL_EVENT hypercall

Add handling of SIGNAL_EVENT hypercall.  For that, provide an interface
to associate an EventNotifier with an event connection number, so that
it's signaled when the SIGNAL_EVENT hypercall with the matching
connection ID is called by the guest.

Support for using KVM functionality for this will be added in a followup
patch.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-8-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c               | 93 ++++++++++++++++++++++++++++++++++++++++
 include/hw/hyperv/hyperv-proto.h |  1 +
 include/hw/hyperv/hyperv.h       | 13 ++++++
 target/i386/hyperv.c             | 10 +++--
 4 files changed, 113 insertions(+), 4 deletions(-)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index 2b0e593bf9..d745016c17 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -13,6 +13,9 @@
 #include "exec/address-spaces.h"
 #include "sysemu/kvm.h"
 #include "qemu/bitops.h"
+#include "qemu/queue.h"
+#include "qemu/rcu.h"
+#include "qemu/rcu_queue.h"
 #include "hw/hyperv/hyperv.h"
 
 typedef struct SynICState {
@@ -450,3 +453,93 @@ int hyperv_sint_route_set_sint(HvSintRoute *sint_route)
 {
     return event_notifier_set(&sint_route->sint_set_notifier);
 }
+
+typedef struct EventFlagHandler {
+    struct rcu_head rcu;
+    QLIST_ENTRY(EventFlagHandler) link;
+    uint32_t conn_id;
+    EventNotifier *notifier;
+} EventFlagHandler;
+
+static QLIST_HEAD(, EventFlagHandler) event_flag_handlers;
+static QemuMutex handlers_mutex;
+
+static void __attribute__((constructor)) hv_init(void)
+{
+    QLIST_INIT(&event_flag_handlers);
+    qemu_mutex_init(&handlers_mutex);
+}
+
+int hyperv_set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier)
+{
+    int ret;
+    EventFlagHandler *handler;
+
+    qemu_mutex_lock(&handlers_mutex);
+    QLIST_FOREACH(handler, &event_flag_handlers, link) {
+        if (handler->conn_id == conn_id) {
+            if (notifier) {
+                ret = -EEXIST;
+            } else {
+                QLIST_REMOVE_RCU(handler, link);
+                g_free_rcu(handler, rcu);
+                ret = 0;
+            }
+            goto unlock;
+        }
+    }
+
+    if (notifier) {
+        handler = g_new(EventFlagHandler, 1);
+        handler->conn_id = conn_id;
+        handler->notifier = notifier;
+        QLIST_INSERT_HEAD_RCU(&event_flag_handlers, handler, link);
+        ret = 0;
+    } else {
+        ret = -ENOENT;
+    }
+unlock:
+    qemu_mutex_unlock(&handlers_mutex);
+    return ret;
+}
+
+uint16_t hyperv_hcall_signal_event(uint64_t param, bool fast)
+{
+    uint16_t ret;
+    EventFlagHandler *handler;
+
+    if (unlikely(!fast)) {
+        hwaddr addr = param;
+
+        if (addr & (__alignof__(addr) - 1)) {
+            return HV_STATUS_INVALID_ALIGNMENT;
+        }
+
+        param = ldq_phys(&address_space_memory, addr);
+    }
+
+    /*
+     * Per spec, bits 32-47 contain the extra "flag number".  However, we
+     * have no use for it, and in all known usecases it is zero, so just
+     * report lookup failure if it isn't.
+     */
+    if (param & 0xffff00000000ULL) {
+        return HV_STATUS_INVALID_PORT_ID;
+    }
+    /* remaining bits are reserved-zero */
+    if (param & ~HV_CONNECTION_ID_MASK) {
+        return HV_STATUS_INVALID_HYPERCALL_INPUT;
+    }
+
+    ret = HV_STATUS_INVALID_CONNECTION_ID;
+    rcu_read_lock();
+    QLIST_FOREACH_RCU(handler, &event_flag_handlers, link) {
+        if (handler->conn_id == param) {
+            event_notifier_set(handler->notifier);
+            ret = 0;
+            break;
+        }
+    }
+    rcu_read_unlock();
+    return ret;
+}
diff --git a/include/hw/hyperv/hyperv-proto.h b/include/hw/hyperv/hyperv-proto.h
index 2dc78eeafb..21dc28aee9 100644
--- a/include/hw/hyperv/hyperv-proto.h
+++ b/include/hw/hyperv/hyperv-proto.h
@@ -21,6 +21,7 @@
 #define HV_STATUS_INVALID_ALIGNMENT           4
 #define HV_STATUS_INVALID_PARAMETER           5
 #define HV_STATUS_INSUFFICIENT_MEMORY         11
+#define HV_STATUS_INVALID_PORT_ID             17
 #define HV_STATUS_INVALID_CONNECTION_ID       18
 #define HV_STATUS_INSUFFICIENT_BUFFERS        19
 
diff --git a/include/hw/hyperv/hyperv.h b/include/hw/hyperv/hyperv.h
index 757c85eb8f..df92ed7e66 100644
--- a/include/hw/hyperv/hyperv.h
+++ b/include/hw/hyperv/hyperv.h
@@ -39,6 +39,19 @@ int hyperv_post_msg(HvSintRoute *sint_route, struct hyperv_message *msg);
  */
 int hyperv_set_event_flag(HvSintRoute *sint_route, unsigned eventno);
 
+/*
+ * Associate @notifier with the event connection @conn_id, such that @notifier
+ * is signaled when the guest executes HV_SIGNAL_EVENT hypercall on @conn_id.
+ * If @notifier is NULL clear the association.
+ */
+int hyperv_set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier);
+
+/*
+ * Process HV_SIGNAL_EVENT hypercall: signal the EventNotifier associated with
+ * the connection as specified in @param.
+ */
+uint16_t hyperv_hcall_signal_event(uint64_t param, bool fast);
+
 static inline uint32_t hyperv_vp_index(CPUState *cs)
 {
     return cs->cpu_index;
diff --git a/target/i386/hyperv.c b/target/i386/hyperv.c
index 3f76c3e266..96b3b5ad7f 100644
--- a/target/i386/hyperv.c
+++ b/target/i386/hyperv.c
@@ -79,16 +79,18 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
 
         return 0;
     case KVM_EXIT_HYPERV_HCALL: {
-        uint16_t code;
+        uint16_t code = exit->u.hcall.input & 0xffff;
+        bool fast = exit->u.hcall.input & HV_HYPERCALL_FAST;
+        uint64_t param = exit->u.hcall.params[0];
 
-        code  = exit->u.hcall.input & 0xffff;
         switch (code) {
-        case HV_POST_MESSAGE:
         case HV_SIGNAL_EVENT:
+            exit->u.hcall.result = hyperv_hcall_signal_event(param, fast);
+            break;
         default:
             exit->u.hcall.result = HV_STATUS_INVALID_HYPERCALL_CODE;
-            return 0;
         }
+        return 0;
     }
     default:
         return -1;
-- 
cgit v1.2.3-55-g7522


From 8d3bc0b75318dce70928b8bde1a49e632adf5137 Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:15 +0300
Subject: hyperv: add support for KVM_HYPERV_EVENTFD

When setting up a notifier for Hyper-V event connection, try to use the
KVM-assisted one first, and fall back to userspace handling of the
hypercall if the kernel doesn't provide the requested feature.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-9-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c | 27 ++++++++++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index d745016c17..7a1c1bbee4 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -13,6 +13,7 @@
 #include "exec/address-spaces.h"
 #include "sysemu/kvm.h"
 #include "qemu/bitops.h"
+#include "qemu/error-report.h"
 #include "qemu/queue.h"
 #include "qemu/rcu.h"
 #include "qemu/rcu_queue.h"
@@ -470,7 +471,7 @@ static void __attribute__((constructor)) hv_init(void)
     qemu_mutex_init(&handlers_mutex);
 }
 
-int hyperv_set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier)
+static int set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier)
 {
     int ret;
     EventFlagHandler *handler;
@@ -503,6 +504,30 @@ unlock:
     return ret;
 }
 
+static bool process_event_flags_userspace;
+
+int hyperv_set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier)
+{
+    if (!process_event_flags_userspace &&
+        !kvm_check_extension(kvm_state, KVM_CAP_HYPERV_EVENTFD)) {
+        process_event_flags_userspace = true;
+
+        warn_report("Hyper-V event signaling is not supported by this kernel; "
+                    "using slower userspace hypercall processing");
+    }
+
+    if (!process_event_flags_userspace) {
+        struct kvm_hyperv_eventfd hvevfd = {
+            .conn_id = conn_id,
+            .fd = notifier ? event_notifier_get_fd(notifier) : -1,
+            .flags = notifier ? 0 : KVM_HYPERV_EVENTFD_DEASSIGN,
+        };
+
+        return kvm_vm_ioctl(kvm_state, KVM_HYPERV_EVENTFD, &hvevfd);
+    }
+    return set_event_flag_handler(conn_id, notifier);
+}
+
 uint16_t hyperv_hcall_signal_event(uint64_t param, bool fast)
 {
     uint16_t ret;
-- 
cgit v1.2.3-55-g7522


From 76036a5fc7ca632f805748aeef416355b1d212a3 Mon Sep 17 00:00:00 2001
From: Roman Kagan
Date: Fri, 21 Sep 2018 11:22:16 +0300
Subject: hyperv: process POST_MESSAGE hypercall

Add handling of POST_MESSAGE hypercall.  For that, add an interface to
regsiter a handler for the messages arrived from the guest on a
particular connection id (IOW set up a message connection in Hyper-V
speak).

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Message-Id: <20180921082217.29481-10-rkagan@virtuozzo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/hyperv/hyperv.c         | 84 ++++++++++++++++++++++++++++++++++++++++++++++
 include/hw/hyperv/hyperv.h | 18 ++++++++++
 target/i386/hyperv.c       |  3 ++
 3 files changed, 105 insertions(+)

(limited to 'hw/hyperv/hyperv.c')

diff --git a/hw/hyperv/hyperv.c b/hw/hyperv/hyperv.c
index 7a1c1bbee4..a28e7249d8 100644
--- a/hw/hyperv/hyperv.c
+++ b/hw/hyperv/hyperv.c
@@ -455,6 +455,14 @@ int hyperv_sint_route_set_sint(HvSintRoute *sint_route)
     return event_notifier_set(&sint_route->sint_set_notifier);
 }
 
+typedef struct MsgHandler {
+    struct rcu_head rcu;
+    QLIST_ENTRY(MsgHandler) link;
+    uint32_t conn_id;
+    HvMsgHandler handler;
+    void *data;
+} MsgHandler;
+
 typedef struct EventFlagHandler {
     struct rcu_head rcu;
     QLIST_ENTRY(EventFlagHandler) link;
@@ -462,15 +470,91 @@ typedef struct EventFlagHandler {
     EventNotifier *notifier;
 } EventFlagHandler;
 
+static QLIST_HEAD(, MsgHandler) msg_handlers;
 static QLIST_HEAD(, EventFlagHandler) event_flag_handlers;
 static QemuMutex handlers_mutex;
 
 static void __attribute__((constructor)) hv_init(void)
 {
+    QLIST_INIT(&msg_handlers);
     QLIST_INIT(&event_flag_handlers);
     qemu_mutex_init(&handlers_mutex);
 }
 
+int hyperv_set_msg_handler(uint32_t conn_id, HvMsgHandler handler, void *data)
+{
+    int ret;
+    MsgHandler *mh;
+
+    qemu_mutex_lock(&handlers_mutex);
+    QLIST_FOREACH(mh, &msg_handlers, link) {
+        if (mh->conn_id == conn_id) {
+            if (handler) {
+                ret = -EEXIST;
+            } else {
+                QLIST_REMOVE_RCU(mh, link);
+                g_free_rcu(mh, rcu);
+                ret = 0;
+            }
+            goto unlock;
+        }
+    }
+
+    if (handler) {
+        mh = g_new(MsgHandler, 1);
+        mh->conn_id = conn_id;
+        mh->handler = handler;
+        mh->data = data;
+        QLIST_INSERT_HEAD_RCU(&msg_handlers, mh, link);
+        ret = 0;
+    } else {
+        ret = -ENOENT;
+    }
+unlock:
+    qemu_mutex_unlock(&handlers_mutex);
+    return ret;
+}
+
+uint16_t hyperv_hcall_post_message(uint64_t param, bool fast)
+{
+    uint16_t ret;
+    hwaddr len;
+    struct hyperv_post_message_input *msg;
+    MsgHandler *mh;
+
+    if (fast) {
+        return HV_STATUS_INVALID_HYPERCALL_CODE;
+    }
+    if (param & (__alignof__(*msg) - 1)) {
+        return HV_STATUS_INVALID_ALIGNMENT;
+    }
+
+    len = sizeof(*msg);
+    msg = cpu_physical_memory_map(param, &len, 0);
+    if (len < sizeof(*msg)) {
+        ret = HV_STATUS_INSUFFICIENT_MEMORY;
+        goto unmap;
+    }
+    if (msg->payload_size > sizeof(msg->payload)) {
+        ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
+        goto unmap;
+    }
+
+    ret = HV_STATUS_INVALID_CONNECTION_ID;
+    rcu_read_lock();
+    QLIST_FOREACH_RCU(mh, &msg_handlers, link) {
+        if (mh->conn_id == (msg->connection_id & HV_CONNECTION_ID_MASK)) {
+            ret = mh->handler(msg, mh->data);
+            break;
+        }
+    }
+    rcu_read_unlock();
+
+unmap:
+    cpu_physical_memory_unmap(msg, len, 0, 0);
+    return ret;
+}
+
 static int set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier)
 {
     int ret;
diff --git a/include/hw/hyperv/hyperv.h b/include/hw/hyperv/hyperv.h
index df92ed7e66..597381cb01 100644
--- a/include/hw/hyperv/hyperv.h
+++ b/include/hw/hyperv/hyperv.h
@@ -39,6 +39,18 @@ int hyperv_post_msg(HvSintRoute *sint_route, struct hyperv_message *msg);
  */
 int hyperv_set_event_flag(HvSintRoute *sint_route, unsigned eventno);
 
+/*
+ * Handler for messages arriving from the guest via HV_POST_MESSAGE hypercall.
+ * Executed in vcpu context.
+ */
+typedef uint16_t (*HvMsgHandler)(const struct hyperv_post_message_input *msg,
+                                 void *data);
+/*
+ * Associate @handler with the message connection @conn_id, such that @handler
+ * is called with @data when the guest executes HV_POST_MESSAGE hypercall on
+ * @conn_id.  If @handler is NULL clear the association.
+ */
+int hyperv_set_msg_handler(uint32_t conn_id, HvMsgHandler handler, void *data);
 /*
  * Associate @notifier with the event connection @conn_id, such that @notifier
  * is signaled when the guest executes HV_SIGNAL_EVENT hypercall on @conn_id.
@@ -46,6 +58,12 @@ int hyperv_set_event_flag(HvSintRoute *sint_route, unsigned eventno);
  */
 int hyperv_set_event_flag_handler(uint32_t conn_id, EventNotifier *notifier);
 
+/*
+ * Process HV_POST_MESSAGE hypercall: parse the data in the guest memory as
+ * specified in @param, and call the HvMsgHandler associated with the
+ * connection on the message contained therein.
+ */
+uint16_t hyperv_hcall_post_message(uint64_t param, bool fast);
 /*
  * Process HV_SIGNAL_EVENT hypercall: signal the EventNotifier associated with
  * the connection as specified in @param.
diff --git a/target/i386/hyperv.c b/target/i386/hyperv.c
index 96b3b5ad7f..b264a28620 100644
--- a/target/i386/hyperv.c
+++ b/target/i386/hyperv.c
@@ -84,6 +84,9 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
         uint64_t param = exit->u.hcall.params[0];
 
         switch (code) {
+        case HV_POST_MESSAGE:
+            exit->u.hcall.result = hyperv_hcall_post_message(param, fast);
+            break;
         case HV_SIGNAL_EVENT:
             exit->u.hcall.result = hyperv_hcall_signal_event(param, fast);
             break;
-- 
cgit v1.2.3-55-g7522