i386/xen: handle VCPUOP_register_vcpu_info

Handle the hypercall to set a per vcpu info, and also wire up the default
vcpu_info in the shared_info page for the first 32 vCPUs.

To avoid deadlock within KVM a vCPU thread must set its *own* vcpu_info
rather than it being set from the context in which the hypercall is
invoked.

Add the vcpu_info (and default) GPA to the vmstate_x86_cpu for migration,
and restore it in kvm_arch_put_registers() appropriately.

Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>

target/i386/cpu.h

@@ -1799,6 +1799,8 @@ typedef struct CPUArchState {
 #endif
 #if defined(CONFIG_KVM)
     struct kvm_nested_state *nested_state;
+    uint64_t xen_vcpu_info_gpa;
+    uint64_t xen_vcpu_info_default_gpa;
 #endif
 #if defined(CONFIG_HVF)
     HVFX86LazyFlags hvf_lflags;

target/i386/kvm/kvm.c

@@ -4750,6 +4750,15 @@ int kvm_arch_put_registers(CPUState *cpu, int level)
         kvm_arch_set_tsc_khz(cpu);
     }
 
+#ifdef CONFIG_XEN_EMU
+    if (xen_mode == XEN_EMULATE && level == KVM_PUT_FULL_STATE) {
+        ret = kvm_put_xen_state(cpu);
+        if (ret < 0) {
+            return ret;
+        }
+    }
+#endif
+
     ret = kvm_getput_regs(x86_cpu, 1);
     if (ret < 0) {
         return ret;
@@ -4849,6 +4858,14 @@ int kvm_arch_get_registers(CPUState *cs)
     if (ret < 0) {
         goto out;
     }
+#ifdef CONFIG_XEN_EMU
+    if (xen_mode == XEN_EMULATE) {
+        ret = kvm_get_xen_state(cs);
+        if (ret < 0) {
+            goto out;
+        }
+    }
+#endif
     ret = 0;
  out:
     cpu_sync_bndcs_hflags(&cpu->env);

target/i386/kvm/trace-events

@@ -10,3 +10,4 @@ kvm_x86_update_msi_routes(int num) "Updated %d MSI routes"
 kvm_xen_hypercall(int cpu, uint8_t cpl, uint64_t input, uint64_t a0, uint64_t a1, uint64_t a2, uint64_t ret) "xen_hypercall: cpu %d cpl %d input %" PRIu64 " a0 0x%" PRIx64 " a1 0x%" PRIx64 " a2 0x%" PRIx64" ret 0x%" PRIx64
 kvm_xen_soft_reset(void) ""
 kvm_xen_set_shared_info(uint64_t gfn) "shared info at gfn 0x%" PRIx64
+kvm_xen_set_vcpu_attr(int cpu, int type, uint64_t gpa) "vcpu attr cpu %d type %d gpa 0x%" PRIx64

target/i386/kvm/xen-emu.c

@@ -119,6 +119,8 @@ int kvm_xen_init(KVMState *s, uint32_t hypercall_msr)
 
 int kvm_xen_init_vcpu(CPUState *cs)
 {
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
     int err;
 
     /*
@@ -142,6 +144,9 @@ int kvm_xen_init_vcpu(CPUState *cs)
         }
     }
 
+    env->xen_vcpu_info_gpa = INVALID_GPA;
+    env->xen_vcpu_info_default_gpa = INVALID_GPA;
+
     return 0;
 }
 
@@ -187,10 +192,58 @@ static bool kvm_xen_hcall_xen_version(struct kvm_xen_exit *exit, X86CPU *cpu,
     return true;
 }
 
+static int kvm_xen_set_vcpu_attr(CPUState *cs, uint16_t type, uint64_t gpa)
+{
+    struct kvm_xen_vcpu_attr xhsi;
+
+    xhsi.type = type;
+    xhsi.u.gpa = gpa;
+
+    trace_kvm_xen_set_vcpu_attr(cs->cpu_index, type, gpa);
+
+    return kvm_vcpu_ioctl(cs, KVM_XEN_VCPU_SET_ATTR, &xhsi);
+}
+
+static void do_set_vcpu_info_default_gpa(CPUState *cs, run_on_cpu_data data)
+{
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+
+    env->xen_vcpu_info_default_gpa = data.host_ulong;
+
+    /* Changing the default does nothing if a vcpu_info was explicitly set. */
+    if (env->xen_vcpu_info_gpa == INVALID_GPA) {
+        kvm_xen_set_vcpu_attr(cs, KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO,
+                              env->xen_vcpu_info_default_gpa);
+    }
+}
+
+static void do_set_vcpu_info_gpa(CPUState *cs, run_on_cpu_data data)
+{
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+
+    env->xen_vcpu_info_gpa = data.host_ulong;
+
+    kvm_xen_set_vcpu_attr(cs, KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO,
+                          env->xen_vcpu_info_gpa);
+}
+
+static void do_vcpu_soft_reset(CPUState *cs, run_on_cpu_data data)
+{
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+
+    env->xen_vcpu_info_gpa = INVALID_GPA;
+    env->xen_vcpu_info_default_gpa = INVALID_GPA;
+
+    kvm_xen_set_vcpu_attr(cs, KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO, INVALID_GPA);
+}
+
 static int xen_set_shared_info(uint64_t gfn)
 {
     uint64_t gpa = gfn << TARGET_PAGE_BITS;
-    int err;
+    int i, err;
 
     QEMU_IOTHREAD_LOCK_GUARD();
 
@@ -207,6 +260,15 @@ static int xen_set_shared_info(uint64_t gfn)
 
     trace_kvm_xen_set_shared_info(gfn);
 
+    for (i = 0; i < XEN_LEGACY_MAX_VCPUS; i++) {
+        CPUState *cpu = qemu_get_cpu(i);
+        if (cpu) {
+            async_run_on_cpu(cpu, do_set_vcpu_info_default_gpa,
+                             RUN_ON_CPU_HOST_ULONG(gpa));
+        }
+        gpa += sizeof(vcpu_info_t);
+    }
+
     return err;
 }
 
@@ -364,15 +426,43 @@ static bool kvm_xen_hcall_hvm_op(struct kvm_xen_exit *exit, X86CPU *cpu,
     }
 }
 
+static int vcpuop_register_vcpu_info(CPUState *cs, CPUState *target,
+                                     uint64_t arg)
+{
+    struct vcpu_register_vcpu_info rvi;
+    uint64_t gpa;
+
+    /* No need for 32/64 compat handling */
+    qemu_build_assert(sizeof(rvi) == 16);
+    qemu_build_assert(sizeof(struct vcpu_info) == 64);
+
+    if (!target) {
+        return -ENOENT;
+    }
+
+    if (kvm_copy_from_gva(cs, arg, &rvi, sizeof(rvi))) {
+        return -EFAULT;
+    }
+
+    if (rvi.offset > TARGET_PAGE_SIZE - sizeof(struct vcpu_info)) {
+        return -EINVAL;
+    }
+
+    gpa = ((rvi.mfn << TARGET_PAGE_BITS) + rvi.offset);
+    async_run_on_cpu(target, do_set_vcpu_info_gpa, RUN_ON_CPU_HOST_ULONG(gpa));
+    return 0;
+}
+
 static bool kvm_xen_hcall_vcpu_op(struct kvm_xen_exit *exit, X86CPU *cpu,
                                   int cmd, int vcpu_id, uint64_t arg)
 {
+    CPUState *dest = qemu_get_cpu(vcpu_id);
+    CPUState *cs = CPU(cpu);
     int err;
 
     switch (cmd) {
     case VCPUOP_register_vcpu_info:
-        /* no vcpu info placement for now */
-        err = -ENOSYS;
+        err = vcpuop_register_vcpu_info(cs, dest, arg);
         break;
 
     default:
@@ -385,12 +475,17 @@ static bool kvm_xen_hcall_vcpu_op(struct kvm_xen_exit *exit, X86CPU *cpu,
 
 int kvm_xen_soft_reset(void)
 {
+    CPUState *cpu;
     int err;
 
     assert(qemu_mutex_iothread_locked());
 
     trace_kvm_xen_soft_reset();
 
+    CPU_FOREACH(cpu) {
+        async_run_on_cpu(cpu, do_vcpu_soft_reset, RUN_ON_CPU_NULL);
+    }
+
     err = xen_overlay_map_shinfo_page(INVALID_GFN);
     if (err) {
         return err;
@@ -539,3 +634,55 @@ int kvm_xen_handle_exit(X86CPU *cpu, struct kvm_xen_exit *exit)
                             exit->u.hcall.result);
     return 0;
 }
+
+int kvm_put_xen_state(CPUState *cs)
+{
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+    uint64_t gpa;
+    int ret;
+
+    gpa = env->xen_vcpu_info_gpa;
+    if (gpa == INVALID_GPA) {
+        gpa = env->xen_vcpu_info_default_gpa;
+    }
+
+    if (gpa != INVALID_GPA) {
+        ret = kvm_xen_set_vcpu_attr(cs, KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO, gpa);
+        if (ret < 0) {
+            return ret;
+        }
+    }
+
+    return 0;
+}
+
+int kvm_get_xen_state(CPUState *cs)
+{
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+    uint64_t gpa;
+
+    /*
+     * The kernel does not mark vcpu_info as dirty when it delivers interrupts
+     * to it. It's up to userspace to *assume* that any page shared thus is
+     * always considered dirty. The shared_info page is different since it's
+     * an overlay and migrated separately anyway.
+     */
+    gpa = env->xen_vcpu_info_gpa;
+    if (gpa == INVALID_GPA) {
+        gpa = env->xen_vcpu_info_default_gpa;
+    }
+    if (gpa != INVALID_GPA) {
+        MemoryRegionSection mrs = memory_region_find(get_system_memory(),
+                                                     gpa,
+                                                     sizeof(struct vcpu_info));
+        if (mrs.mr &&
+            !int128_lt(mrs.size, int128_make64(sizeof(struct vcpu_info)))) {
+            memory_region_set_dirty(mrs.mr, mrs.offset_within_region,
+                                    sizeof(struct vcpu_info));
+        }
+    }
+
+    return 0;
+}

target/i386/kvm/xen-emu.h

@@ -26,5 +26,7 @@
 int kvm_xen_init(KVMState *s, uint32_t hypercall_msr);
 int kvm_xen_init_vcpu(CPUState *cs);
 int kvm_xen_handle_exit(X86CPU *cpu, struct kvm_xen_exit *exit);
+int kvm_put_xen_state(CPUState *cs);
+int kvm_get_xen_state(CPUState *cs);
 
 #endif /* QEMU_I386_KVM_XEN_EMU_H */

target/i386/machine.c

@@ -6,8 +6,10 @@
 #include "kvm/hyperv.h"
 #include "hw/i386/x86.h"
 #include "kvm/kvm_i386.h"
+#include "hw/xen/xen.h"
 
 #include "sysemu/kvm.h"
+#include "sysemu/kvm_xen.h"
 #include "sysemu/tcg.h"
 
 #include "qemu/error-report.h"
@@ -1257,6 +1259,22 @@ static const VMStateDescription vmstate_nested_state = {
     }
 };
 
+static bool xen_vcpu_needed(void *opaque)
+{
+    return (xen_mode == XEN_EMULATE);
+}
+
+static const VMStateDescription vmstate_xen_vcpu = {
+    .name = "cpu/xen_vcpu",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = xen_vcpu_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT64(env.xen_vcpu_info_gpa, X86CPU),
+        VMSTATE_UINT64(env.xen_vcpu_info_default_gpa, X86CPU),
+        VMSTATE_END_OF_LIST()
+    }
+};
 #endif
 
 static bool mcg_ext_ctl_needed(void *opaque)
@@ -1716,6 +1734,7 @@ const VMStateDescription vmstate_x86_cpu = {
 #endif
 #ifdef CONFIG_KVM
         &vmstate_nested_state,
+        &vmstate_xen_vcpu,
 #endif
         &vmstate_msr_tsx_ctrl,
         &vmstate_msr_intel_sgx,