]> git.proxmox.com Git - mirror_qemu.git/blobdiff - xen-hvm.c
iotests, parallels: test for newly created parallels image via qemu-img
[mirror_qemu.git] / xen-hvm.c
index 05e522cd73f31b06356a5eb5de27468ffb76df61..315864ca705f3bac0f6ff856dbae400c5145c217 100644 (file)
--- a/xen-hvm.c
+++ b/xen-hvm.c
@@ -41,6 +41,29 @@ static MemoryRegion *framebuffer;
 static bool xen_in_migration;
 
 /* Compatibility with older version */
+
+/* This allows QEMU to build on a system that has Xen 4.5 or earlier
+ * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
+ * needs to be included before this block and hw/xen/xen_common.h needs to
+ * be included before xen/hvm/ioreq.h
+ */
+#ifndef IOREQ_TYPE_VMWARE_PORT
+#define IOREQ_TYPE_VMWARE_PORT  3
+struct vmware_regs {
+    uint32_t esi;
+    uint32_t edi;
+    uint32_t ebx;
+    uint32_t ecx;
+    uint32_t edx;
+};
+typedef struct vmware_regs vmware_regs_t;
+
+struct shared_vmport_iopage {
+    struct vmware_regs vcpu_vmport_regs[1];
+};
+typedef struct shared_vmport_iopage shared_vmport_iopage_t;
+#endif
+
 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
 {
@@ -62,11 +85,14 @@ static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
 }
 #  define FMT_ioreq_size "u"
 #endif
-#ifndef HVM_PARAM_BUFIOREQ_EVTCHN
-#define HVM_PARAM_BUFIOREQ_EVTCHN 26
-#endif
 
 #define BUFFER_IO_MAX_DELAY  100
+/* Leave some slack so that hvmloader does not complain about lack of
+ * memory at boot time ("Could not allocate order=0 extent").
+ * Once hvmloader is modified to cope with that situation without
+ * printing warning messages, QEMU_SPARE_PAGES can be removed.
+ */
+#define QEMU_SPARE_PAGES 16
 
 typedef struct XenPhysmap {
     hwaddr start_addr;
@@ -78,9 +104,12 @@ typedef struct XenPhysmap {
 } XenPhysmap;
 
 typedef struct XenIOState {
+    ioservid_t ioservid;
     shared_iopage_t *shared_page;
+    shared_vmport_iopage_t *shared_vmport_page;
     buffered_iopage_t *buffered_io_page;
     QEMUTimer *buffered_io_timer;
+    CPUState **cpu_by_vcpu_id;
     /* the evtchn port for polling the notification, */
     evtchn_port_t *ioreq_local_port;
     /* evtchn local port for buffered io */
@@ -92,6 +121,8 @@ typedef struct XenIOState {
 
     struct xs_handle *xenstore;
     MemoryListener memory_listener;
+    MemoryListener io_listener;
+    DeviceListener device_listener;
     QLIST_HEAD(, XenPhysmap) physmap;
     hwaddr free_phys_offset;
     const XenPhysmap *log_for_dirtybit;
@@ -219,6 +250,8 @@ void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
     unsigned long nr_pfn;
     xen_pfn_t *pfn_list;
     int i;
+    xc_domaininfo_t info;
+    unsigned long free_pages;
 
     if (runstate_check(RUN_STATE_INMIGRATE)) {
         /* RAM already populated in Xen */
@@ -241,6 +274,22 @@ void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
         pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
     }
 
+    if ((xc_domain_getinfolist(xen_xc, xen_domid, 1, &info) != 1) ||
+        (info.domain != xen_domid)) {
+        hw_error("xc_domain_getinfolist failed");
+    }
+    free_pages = info.max_pages - info.tot_pages;
+    if (free_pages > QEMU_SPARE_PAGES) {
+        free_pages -= QEMU_SPARE_PAGES;
+    } else {
+        free_pages = 0;
+    }
+    if ((free_pages < nr_pfn) &&
+        (xc_domain_setmaxmem(xen_xc, xen_domid,
+                             ((info.max_pages + nr_pfn - free_pages)
+                              << (XC_PAGE_SHIFT - 10))) < 0)) {
+        hw_error("xc_domain_setmaxmem failed");
+    }
     if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
         hw_error("xen: failed to populate ram at " RAM_ADDR_FMT, ram_addr);
     }
@@ -442,12 +491,23 @@ static void xen_set_memory(struct MemoryListener *listener,
     bool log_dirty = memory_region_is_logging(section->mr);
     hvmmem_type_t mem_type;
 
+    if (section->mr == &ram_memory) {
+        return;
+    } else {
+        if (add) {
+            xen_map_memory_section(xen_xc, xen_domid, state->ioservid,
+                                   section);
+        } else {
+            xen_unmap_memory_section(xen_xc, xen_domid, state->ioservid,
+                                     section);
+        }
+    }
+
     if (!memory_region_is_ram(section->mr)) {
         return;
     }
 
-    if (!(section->mr != &ram_memory
-          && ( (log_dirty && add) || (!log_dirty && !add)))) {
+    if (log_dirty != add) {
         return;
     }
 
@@ -490,6 +550,50 @@ static void xen_region_del(MemoryListener *listener,
     memory_region_unref(section->mr);
 }
 
+static void xen_io_add(MemoryListener *listener,
+                       MemoryRegionSection *section)
+{
+    XenIOState *state = container_of(listener, XenIOState, io_listener);
+
+    memory_region_ref(section->mr);
+
+    xen_map_io_section(xen_xc, xen_domid, state->ioservid, section);
+}
+
+static void xen_io_del(MemoryListener *listener,
+                       MemoryRegionSection *section)
+{
+    XenIOState *state = container_of(listener, XenIOState, io_listener);
+
+    xen_unmap_io_section(xen_xc, xen_domid, state->ioservid, section);
+
+    memory_region_unref(section->mr);
+}
+
+static void xen_device_realize(DeviceListener *listener,
+                              DeviceState *dev)
+{
+    XenIOState *state = container_of(listener, XenIOState, device_listener);
+
+    if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
+        PCIDevice *pci_dev = PCI_DEVICE(dev);
+
+        xen_map_pcidev(xen_xc, xen_domid, state->ioservid, pci_dev);
+    }
+}
+
+static void xen_device_unrealize(DeviceListener *listener,
+                                DeviceState *dev)
+{
+    XenIOState *state = container_of(listener, XenIOState, device_listener);
+
+    if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
+        PCIDevice *pci_dev = PCI_DEVICE(dev);
+
+        xen_unmap_pcidev(xen_xc, xen_domid, state->ioservid, pci_dev);
+    }
+}
+
 static void xen_sync_dirty_bitmap(XenIOState *state,
                                   hwaddr start_addr,
                                   ram_addr_t size)
@@ -590,6 +694,17 @@ static MemoryListener xen_memory_listener = {
     .priority = 10,
 };
 
+static MemoryListener xen_io_listener = {
+    .region_add = xen_io_add,
+    .region_del = xen_io_del,
+    .priority = 10,
+};
+
+static DeviceListener xen_device_listener = {
+    .realize = xen_device_realize,
+    .unrealize = xen_device_unrealize,
+};
+
 /* get the ioreq packets from share mem */
 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
 {
@@ -773,7 +888,50 @@ static void cpu_ioreq_move(ioreq_t *req)
     }
 }
 
-static void handle_ioreq(ioreq_t *req)
+static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
+{
+    X86CPU *cpu;
+    CPUX86State *env;
+
+    cpu = X86_CPU(current_cpu);
+    env = &cpu->env;
+    env->regs[R_EAX] = req->data;
+    env->regs[R_EBX] = vmport_regs->ebx;
+    env->regs[R_ECX] = vmport_regs->ecx;
+    env->regs[R_EDX] = vmport_regs->edx;
+    env->regs[R_ESI] = vmport_regs->esi;
+    env->regs[R_EDI] = vmport_regs->edi;
+}
+
+static void regs_from_cpu(vmware_regs_t *vmport_regs)
+{
+    X86CPU *cpu = X86_CPU(current_cpu);
+    CPUX86State *env = &cpu->env;
+
+    vmport_regs->ebx = env->regs[R_EBX];
+    vmport_regs->ecx = env->regs[R_ECX];
+    vmport_regs->edx = env->regs[R_EDX];
+    vmport_regs->esi = env->regs[R_ESI];
+    vmport_regs->edi = env->regs[R_EDI];
+}
+
+static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
+{
+    vmware_regs_t *vmport_regs;
+
+    assert(state->shared_vmport_page);
+    vmport_regs =
+        &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
+    QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
+
+    current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
+    regs_to_cpu(vmport_regs, req);
+    cpu_ioreq_pio(req);
+    regs_from_cpu(vmport_regs);
+    current_cpu = NULL;
+}
+
+static void handle_ioreq(XenIOState *state, ioreq_t *req)
 {
     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
             (req->size < sizeof (target_ulong))) {
@@ -787,11 +945,35 @@ static void handle_ioreq(ioreq_t *req)
         case IOREQ_TYPE_COPY:
             cpu_ioreq_move(req);
             break;
+        case IOREQ_TYPE_VMWARE_PORT:
+            handle_vmport_ioreq(state, req);
+            break;
         case IOREQ_TYPE_TIMEOFFSET:
             break;
         case IOREQ_TYPE_INVALIDATE:
             xen_invalidate_map_cache();
             break;
+        case IOREQ_TYPE_PCI_CONFIG: {
+            uint32_t sbdf = req->addr >> 32;
+            uint32_t val;
+
+            /* Fake a write to port 0xCF8 so that
+             * the config space access will target the
+             * correct device model.
+             */
+            val = (1u << 31) |
+                  ((req->addr & 0x0f00) << 16) |
+                  ((sbdf & 0xffff) << 8) |
+                  (req->addr & 0xfc);
+            do_outp(0xcf8, 4, val);
+
+            /* Now issue the config space access via
+             * port 0xCFC
+             */
+            req->addr = 0xcfc | (req->addr & 0x03);
+            cpu_ioreq_pio(req);
+            break;
+        }
         default:
             hw_error("Invalid ioreq type 0x%x\n", req->type);
     }
@@ -828,7 +1010,7 @@ static int handle_buffered_iopage(XenIOState *state)
             req.data |= ((uint64_t)buf_req->data) << 32;
         }
 
-        handle_ioreq(&req);
+        handle_ioreq(state, &req);
 
         xen_mb();
         state->buffered_io_page->read_pointer += qw ? 2 : 1;
@@ -857,14 +1039,16 @@ static void cpu_handle_ioreq(void *opaque)
 
     handle_buffered_iopage(state);
     if (req) {
-        handle_ioreq(req);
+        handle_ioreq(state, req);
 
         if (req->state != STATE_IOREQ_INPROCESS) {
             fprintf(stderr, "Badness in I/O request ... not in service?!: "
                     "%x, ptr: %x, port: %"PRIx64", "
-                    "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
+                    "data: %"PRIx64", count: %" FMT_ioreq_size
+                    ", size: %" FMT_ioreq_size
+                    ", type: %"FMT_ioreq_size"\n",
                     req->state, req->data_is_ptr, req->addr,
-                    req->data, req->count, req->size);
+                    req->data, req->count, req->size, req->type);
             destroy_hvm_domain(false);
             return;
         }
@@ -904,6 +1088,14 @@ static void xen_main_loop_prepare(XenIOState *state)
                                                  state);
 
     if (evtchn_fd != -1) {
+        CPUState *cpu_state;
+
+        DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
+        CPU_FOREACH(cpu_state) {
+            DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
+                    __func__, cpu_state->cpu_index, cpu_state);
+            state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
+        }
         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
     }
 }
@@ -912,10 +1104,15 @@ static void xen_main_loop_prepare(XenIOState *state)
 static void xen_hvm_change_state_handler(void *opaque, int running,
                                          RunState rstate)
 {
-    XenIOState *xstate = opaque;
+    XenIOState *state = opaque;
+
     if (running) {
-        xen_main_loop_prepare(xstate);
+        xen_main_loop_prepare(state);
     }
+
+    xen_set_ioreq_server_state(xen_xc, xen_domid,
+                               state->ioservid,
+                               (rstate == RUN_STATE_RUNNING));
 }
 
 static void xen_exit_notifier(Notifier *n, void *data)
@@ -984,8 +1181,9 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
                  MemoryRegion **ram_memory)
 {
     int i, rc;
-    unsigned long ioreq_pfn;
-    unsigned long bufioreq_evtchn;
+    xen_pfn_t ioreq_pfn;
+    xen_pfn_t bufioreq_pfn;
+    evtchn_port_t bufioreq_evtchn;
     XenIOState *state;
 
     state = g_malloc0(sizeof (XenIOState));
@@ -1002,6 +1200,12 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
         return -1;
     }
 
+    rc = xen_create_ioreq_server(xen_xc, xen_domid, &state->ioservid);
+    if (rc < 0) {
+        perror("xen: ioreq server create");
+        return -1;
+    }
+
     state->exit.notify = xen_exit_notifier;
     qemu_add_exit_notifier(&state->exit);
 
@@ -1011,8 +1215,18 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
     state->wakeup.notify = xen_wakeup_notifier;
     qemu_register_wakeup_notifier(&state->wakeup);
 
-    xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_IOREQ_PFN, &ioreq_pfn);
+    rc = xen_get_ioreq_server_info(xen_xc, xen_domid, state->ioservid,
+                                   &ioreq_pfn, &bufioreq_pfn,
+                                   &bufioreq_evtchn);
+    if (rc < 0) {
+        hw_error("failed to get ioreq server info: error %d handle=" XC_INTERFACE_FMT,
+                 errno, xen_xc);
+    }
+
     DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
+    DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
+    DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
+
     state->shared_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
                                               PROT_READ|PROT_WRITE, ioreq_pfn);
     if (state->shared_page == NULL) {
@@ -1020,14 +1234,37 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
                  errno, xen_xc);
     }
 
-    xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
-    DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
-    state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
-                                                   PROT_READ|PROT_WRITE, ioreq_pfn);
+    rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
+    if (!rc) {
+        DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
+        state->shared_vmport_page =
+            xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
+                                 PROT_READ|PROT_WRITE, ioreq_pfn);
+        if (state->shared_vmport_page == NULL) {
+            hw_error("map shared vmport IO page returned error %d handle="
+                     XC_INTERFACE_FMT, errno, xen_xc);
+        }
+    } else if (rc != -ENOSYS) {
+        hw_error("get vmport regs pfn returned error %d, rc=%d", errno, rc);
+    }
+
+    state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid,
+                                                   XC_PAGE_SIZE,
+                                                   PROT_READ|PROT_WRITE,
+                                                   bufioreq_pfn);
     if (state->buffered_io_page == NULL) {
         hw_error("map buffered IO page returned error %d", errno);
     }
 
+    /* Note: cpus is empty at this point in init */
+    state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
+
+    rc = xen_set_ioreq_server_state(xen_xc, xen_domid, state->ioservid, true);
+    if (rc < 0) {
+        hw_error("failed to enable ioreq server info: error %d handle=" XC_INTERFACE_FMT,
+                 errno, xen_xc);
+    }
+
     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
 
     /* FIXME: how about if we overflow the page here? */
@@ -1035,22 +1272,16 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
         rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
                                         xen_vcpu_eport(state->shared_page, i));
         if (rc == -1) {
-            fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
+            fprintf(stderr, "shared evtchn %d bind error %d\n", i, errno);
             return -1;
         }
         state->ioreq_local_port[i] = rc;
     }
 
-    rc = xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_EVTCHN,
-            &bufioreq_evtchn);
-    if (rc < 0) {
-        fprintf(stderr, "failed to get HVM_PARAM_BUFIOREQ_EVTCHN\n");
-        return -1;
-    }
     rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
-            (uint32_t)bufioreq_evtchn);
+                                    bufioreq_evtchn);
     if (rc == -1) {
-        fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
+        fprintf(stderr, "buffered evtchn bind error %d\n", errno);
         return -1;
     }
     state->bufioreq_local_port = rc;
@@ -1066,6 +1297,12 @@ int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
     memory_listener_register(&state->memory_listener, &address_space_memory);
     state->log_for_dirtybit = NULL;
 
+    state->io_listener = xen_io_listener;
+    memory_listener_register(&state->io_listener, &address_space_io);
+
+    state->device_listener = xen_device_listener;
+    device_listener_register(&state->device_listener);
+
     /* Initialize backend core & drivers */
     if (xen_be_init() != 0) {
         fprintf(stderr, "%s: xen backend core setup failed\n", __FUNCTION__);