]> git.proxmox.com Git - qemu.git/blobdiff - xen-all.c
qom: improve documentation of cast functions
[qemu.git] / xen-all.c
index f2cad8227ed6569b8cf412ecce1e6e15c2459e51..539a1549a5cee3b313dbb67dad636735a194e8a5 100644 (file)
--- a/xen-all.c
+++ b/xen-all.c
 
 #include <sys/mman.h>
 
-#include "hw/pci.h"
-#include "hw/pc.h"
-#include "hw/xen_common.h"
-#include "hw/xen_backend.h"
-
-#include "range.h"
-#include "xen-mapcache.h"
+#include "hw/pci/pci.h"
+#include "hw/i386/pc.h"
+#include "hw/xen/xen_common.h"
+#include "hw/xen/xen_backend.h"
+#include "qmp-commands.h"
+
+#include "sysemu/char.h"
+#include "qemu/range.h"
+#include "sysemu/xen-mapcache.h"
 #include "trace.h"
-#include "exec-memory.h"
+#include "exec/address-spaces.h"
 
 #include <xen/hvm/ioreq.h>
 #include <xen/hvm/params.h>
@@ -36,6 +38,7 @@
 
 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
 static MemoryRegion *framebuffer;
+static bool xen_in_migration;
 
 /* Compatibility with older version */
 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
@@ -59,14 +62,17 @@ 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
 
 typedef struct XenPhysmap {
-    target_phys_addr_t start_addr;
+    hwaddr start_addr;
     ram_addr_t size;
     char *name;
-    target_phys_addr_t phys_offset;
+    hwaddr phys_offset;
 
     QLIST_ENTRY(XenPhysmap) list;
 } XenPhysmap;
@@ -77,6 +83,8 @@ typedef struct XenIOState {
     QEMUTimer *buffered_io_timer;
     /* the evtchn port for polling the notification, */
     evtchn_port_t *ioreq_local_port;
+    /* evtchn local port for buffered io */
+    evtchn_port_t bufioreq_local_port;
     /* the evtchn fd for polling */
     XenEvtchn xce_handle;
     /* which vcpu we are serving */
@@ -85,7 +93,7 @@ typedef struct XenIOState {
     struct xs_handle *xenstore;
     MemoryListener memory_listener;
     QLIST_HEAD(, XenPhysmap) physmap;
-    target_phys_addr_t free_phys_offset;
+    hwaddr free_phys_offset;
     const XenPhysmap *log_for_dirtybit;
 
     Notifier exit;
@@ -122,6 +130,11 @@ void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
     }
 }
 
+void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
+{
+    xen_xc_hvm_inject_msi(xen_xc, xen_domid, addr, data);
+}
+
 static void xen_suspend_notifier(Notifier *notifier, void *data)
 {
     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
@@ -190,6 +203,14 @@ void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
     xen_pfn_t *pfn_list;
     int i;
 
+    if (runstate_check(RUN_STATE_INMIGRATE)) {
+        /* RAM already populated in Xen */
+        fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
+                " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
+                __func__, size, ram_addr); 
+        return;
+    }
+
     if (mr == &ram_memory) {
         return;
     }
@@ -211,7 +232,7 @@ void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
 }
 
 static XenPhysmap *get_physmapping(XenIOState *state,
-                                   target_phys_addr_t start_addr, ram_addr_t size)
+                                   hwaddr start_addr, ram_addr_t size)
 {
     XenPhysmap *physmap = NULL;
 
@@ -225,18 +246,34 @@ static XenPhysmap *get_physmapping(XenIOState *state,
     return NULL;
 }
 
+static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
+                                                   ram_addr_t size, void *opaque)
+{
+    hwaddr addr = start_addr & TARGET_PAGE_MASK;
+    XenIOState *xen_io_state = opaque;
+    XenPhysmap *physmap = NULL;
+
+    QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
+        if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
+            return physmap->start_addr;
+        }
+    }
+
+    return start_addr;
+}
+
 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
 static int xen_add_to_physmap(XenIOState *state,
-                              target_phys_addr_t start_addr,
+                              hwaddr start_addr,
                               ram_addr_t size,
                               MemoryRegion *mr,
-                              target_phys_addr_t offset_within_region)
+                              hwaddr offset_within_region)
 {
     unsigned long i = 0;
     int rc = 0;
     XenPhysmap *physmap = NULL;
-    target_phys_addr_t pfn, start_gpfn;
-    target_phys_addr_t phys_offset = memory_region_get_ram_addr(mr);
+    hwaddr pfn, start_gpfn;
+    hwaddr phys_offset = memory_region_get_ram_addr(mr);
     char path[80], value[17];
 
     if (get_physmapping(state, start_addr, size)) {
@@ -256,7 +293,8 @@ static int xen_add_to_physmap(XenIOState *state,
     return -1;
 
 go_physmap:
-    DPRINTF("mapping vram to %llx - %llx\n", start_addr, start_addr + size);
+    DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
+            start_addr, start_addr + size);
 
     pfn = phys_offset >> TARGET_PAGE_BITS;
     start_gpfn = start_addr >> TARGET_PAGE_BITS;
@@ -313,13 +351,13 @@ go_physmap:
 }
 
 static int xen_remove_from_physmap(XenIOState *state,
-                                   target_phys_addr_t start_addr,
+                                   hwaddr start_addr,
                                    ram_addr_t size)
 {
     unsigned long i = 0;
     int rc = 0;
     XenPhysmap *physmap = NULL;
-    target_phys_addr_t phys_offset = 0;
+    hwaddr phys_offset = 0;
 
     physmap = get_physmapping(state, start_addr, size);
     if (physmap == NULL) {
@@ -329,8 +367,8 @@ static int xen_remove_from_physmap(XenIOState *state,
     phys_offset = physmap->phys_offset;
     size = physmap->size;
 
-    DPRINTF("unmapping vram to %llx - %llx, from %llx\n",
-            phys_offset, phys_offset + size, start_addr);
+    DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", from ",
+            "%"HWADDR_PRIx"\n", phys_offset, phys_offset + size, start_addr);
 
     size >>= TARGET_PAGE_BITS;
     start_addr >>= TARGET_PAGE_BITS;
@@ -358,16 +396,16 @@ static int xen_remove_from_physmap(XenIOState *state,
 
 #else
 static int xen_add_to_physmap(XenIOState *state,
-                              target_phys_addr_t start_addr,
+                              hwaddr start_addr,
                               ram_addr_t size,
                               MemoryRegion *mr,
-                              target_phys_addr_t offset_within_region)
+                              hwaddr offset_within_region)
 {
     return -ENOSYS;
 }
 
 static int xen_remove_from_physmap(XenIOState *state,
-                                   target_phys_addr_t start_addr,
+                                   hwaddr start_addr,
                                    ram_addr_t size)
 {
     return -ENOSYS;
@@ -379,7 +417,7 @@ static void xen_set_memory(struct MemoryListener *listener,
                            bool add)
 {
     XenIOState *state = container_of(listener, XenIOState, memory_listener);
-    target_phys_addr_t start_addr = section->offset_within_address_space;
+    hwaddr start_addr = section->offset_within_address_space;
     ram_addr_t size = section->size;
     bool log_dirty = memory_region_is_logging(section->mr);
     hvmmem_type_t mem_type;
@@ -431,10 +469,10 @@ static void xen_region_del(MemoryListener *listener,
 }
 
 static void xen_sync_dirty_bitmap(XenIOState *state,
-                                  target_phys_addr_t start_addr,
+                                  hwaddr start_addr,
                                   ram_addr_t size)
 {
-    target_phys_addr_t npages = size >> TARGET_PAGE_BITS;
+    hwaddr npages = size >> TARGET_PAGE_BITS;
     const int width = sizeof(unsigned long) * 8;
     unsigned long bitmap[(npages + width - 1) / width];
     int rc, i, j;
@@ -458,7 +496,8 @@ static void xen_sync_dirty_bitmap(XenIOState *state,
                                  bitmap);
     if (rc < 0) {
         if (rc != -ENODATA) {
-            fprintf(stderr, "xen: track_dirty_vram failed (0x" TARGET_FMT_plx
+            memory_region_set_dirty(framebuffer, 0, size);
+            DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
                     ", 0x" TARGET_FMT_plx "): %s\n",
                     start_addr, start_addr + size, strerror(-rc));
         }
@@ -505,10 +544,14 @@ static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
 
 static void xen_log_global_start(MemoryListener *listener)
 {
+    if (xen_enabled()) {
+        xen_in_migration = true;
+    }
 }
 
 static void xen_log_global_stop(MemoryListener *listener)
 {
+    xen_in_migration = false;
 }
 
 static MemoryListener xen_memory_listener = {
@@ -519,24 +562,38 @@ static MemoryListener xen_memory_listener = {
     .log_sync = xen_log_sync,
     .log_global_start = xen_log_global_start,
     .log_global_stop = xen_log_global_stop,
+    .priority = 10,
 };
 
+void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
+{
+    if (enable) {
+        memory_global_dirty_log_start();
+    } else {
+        memory_global_dirty_log_stop();
+    }
+}
+
 /* VCPU Operations, MMIO, IO ring ... */
 
 static void xen_reset_vcpu(void *opaque)
 {
-    CPUState *env = opaque;
+    CPUState *cpu = opaque;
 
-    env->halted = 1;
+    cpu->halted = 1;
 }
 
 void xen_vcpu_init(void)
 {
-    CPUState *first_cpu;
+    if (first_cpu != NULL) {
+        CPUState *cpu = ENV_GET_CPU(first_cpu);
 
-    if ((first_cpu = qemu_get_cpu(0))) {
-        qemu_register_reset(xen_reset_vcpu, first_cpu);
-        xen_reset_vcpu(first_cpu);
+        qemu_register_reset(xen_reset_vcpu, cpu);
+        xen_reset_vcpu(cpu);
+    }
+    /* if rtc_clock is left to default (host_clock), disable it */
+    if (rtc_clock == host_clock) {
+        qemu_clock_enable(rtc_clock, false);
     }
 }
 
@@ -569,6 +626,12 @@ static ioreq_t *cpu_get_ioreq(XenIOState *state)
     evtchn_port_t port;
 
     port = xc_evtchn_pending(state->xce_handle);
+    if (port == state->bufioreq_local_port) {
+        qemu_mod_timer(state->buffered_io_timer,
+                BUFFER_IO_MAX_DELAY + qemu_get_clock_ms(rt_clock));
+        return NULL;
+    }
+
     if (port != -1) {
         for (i = 0; i < smp_cpus; i++) {
             if (state->ioreq_local_port[i] == port) {
@@ -621,11 +684,45 @@ static void do_outp(pio_addr_t addr,
     }
 }
 
-static void cpu_ioreq_pio(ioreq_t *req)
+/*
+ * Helper functions which read/write an object from/to physical guest
+ * memory, as part of the implementation of an ioreq.
+ *
+ * Equivalent to
+ *   cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
+ *                          val, req->size, 0/1)
+ * except without the integer overflow problems.
+ */
+static void rw_phys_req_item(hwaddr addr,
+                             ioreq_t *req, uint32_t i, void *val, int rw)
+{
+    /* Do everything unsigned so overflow just results in a truncated result
+     * and accesses to undesired parts of guest memory, which is up
+     * to the guest */
+    hwaddr offset = (hwaddr)req->size * i;
+    if (req->df) {
+        addr -= offset;
+    } else {
+        addr += offset;
+    }
+    cpu_physical_memory_rw(addr, val, req->size, rw);
+}
+
+static inline void read_phys_req_item(hwaddr addr,
+                                      ioreq_t *req, uint32_t i, void *val)
 {
-    int i, sign;
+    rw_phys_req_item(addr, req, i, val, 0);
+}
+static inline void write_phys_req_item(hwaddr addr,
+                                       ioreq_t *req, uint32_t i, void *val)
+{
+    rw_phys_req_item(addr, req, i, val, 1);
+}
 
-    sign = req->df ? -1 : 1;
+
+static void cpu_ioreq_pio(ioreq_t *req)
+{
+    uint32_t i;
 
     if (req->dir == IOREQ_READ) {
         if (!req->data_is_ptr) {
@@ -635,8 +732,7 @@ static void cpu_ioreq_pio(ioreq_t *req)
 
             for (i = 0; i < req->count; i++) {
                 tmp = do_inp(req->addr, req->size);
-                cpu_physical_memory_write(req->data + (sign * i * req->size),
-                        (uint8_t *) &tmp, req->size);
+                write_phys_req_item(req->data, req, i, &tmp);
             }
         }
     } else if (req->dir == IOREQ_WRITE) {
@@ -646,8 +742,7 @@ static void cpu_ioreq_pio(ioreq_t *req)
             for (i = 0; i < req->count; i++) {
                 uint32_t tmp = 0;
 
-                cpu_physical_memory_read(req->data + (sign * i * req->size),
-                        (uint8_t*) &tmp, req->size);
+                read_phys_req_item(req->data, req, i, &tmp);
                 do_outp(req->addr, req->size, tmp);
             }
         }
@@ -656,20 +751,16 @@ static void cpu_ioreq_pio(ioreq_t *req)
 
 static void cpu_ioreq_move(ioreq_t *req)
 {
-    int i, sign;
-
-    sign = req->df ? -1 : 1;
+    uint32_t i;
 
     if (!req->data_is_ptr) {
         if (req->dir == IOREQ_READ) {
             for (i = 0; i < req->count; i++) {
-                cpu_physical_memory_read(req->addr + (sign * i * req->size),
-                        (uint8_t *) &req->data, req->size);
+                read_phys_req_item(req->addr, req, i, &req->data);
             }
         } else if (req->dir == IOREQ_WRITE) {
             for (i = 0; i < req->count; i++) {
-                cpu_physical_memory_write(req->addr + (sign * i * req->size),
-                        (uint8_t *) &req->data, req->size);
+                write_phys_req_item(req->addr, req, i, &req->data);
             }
         }
     } else {
@@ -677,17 +768,13 @@ static void cpu_ioreq_move(ioreq_t *req)
 
         if (req->dir == IOREQ_READ) {
             for (i = 0; i < req->count; i++) {
-                cpu_physical_memory_read(req->addr + (sign * i * req->size),
-                        (uint8_t*) &tmp, req->size);
-                cpu_physical_memory_write(req->data + (sign * i * req->size),
-                        (uint8_t*) &tmp, req->size);
+                read_phys_req_item(req->addr, req, i, &tmp);
+                write_phys_req_item(req->data, req, i, &tmp);
             }
         } else if (req->dir == IOREQ_WRITE) {
             for (i = 0; i < req->count; i++) {
-                cpu_physical_memory_read(req->data + (sign * i * req->size),
-                        (uint8_t*) &tmp, req->size);
-                cpu_physical_memory_write(req->addr + (sign * i * req->size),
-                        (uint8_t*) &tmp, req->size);
+                read_phys_req_item(req->data, req, i, &tmp);
+                write_phys_req_item(req->addr, req, i, &tmp);
             }
         }
     }
@@ -717,16 +804,18 @@ static void handle_ioreq(ioreq_t *req)
     }
 }
 
-static void handle_buffered_iopage(XenIOState *state)
+static int handle_buffered_iopage(XenIOState *state)
 {
     buf_ioreq_t *buf_req = NULL;
     ioreq_t req;
     int qw;
 
     if (!state->buffered_io_page) {
-        return;
+        return 0;
     }
 
+    memset(&req, 0x00, sizeof(req));
+
     while (state->buffered_io_page->read_pointer != state->buffered_io_page->write_pointer) {
         buf_req = &state->buffered_io_page->buf_ioreq[
             state->buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
@@ -751,15 +840,21 @@ static void handle_buffered_iopage(XenIOState *state)
         xen_mb();
         state->buffered_io_page->read_pointer += qw ? 2 : 1;
     }
+
+    return req.count;
 }
 
 static void handle_buffered_io(void *opaque)
 {
     XenIOState *state = opaque;
 
-    handle_buffered_iopage(state);
-    qemu_mod_timer(state->buffered_io_timer,
-                   BUFFER_IO_MAX_DELAY + qemu_get_clock_ms(rt_clock));
+    if (handle_buffered_iopage(state)) {
+        qemu_mod_timer(state->buffered_io_timer,
+                BUFFER_IO_MAX_DELAY + qemu_get_clock_ms(rt_clock));
+    } else {
+        qemu_del_timer(state->buffered_io_timer);
+        xc_evtchn_unmask(state->xce_handle, state->bufioreq_local_port);
+    }
 }
 
 static void cpu_handle_ioreq(void *opaque)
@@ -777,7 +872,7 @@ static void cpu_handle_ioreq(void *opaque)
                     "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
                     req->state, req->data_is_ptr, req->addr,
                     req->data, req->count, req->size);
-            destroy_hvm_domain();
+            destroy_hvm_domain(false);
             return;
         }
 
@@ -791,10 +886,11 @@ static void cpu_handle_ioreq(void *opaque)
          */
         if (runstate_is_running()) {
             if (qemu_shutdown_requested_get()) {
-                destroy_hvm_domain();
+                destroy_hvm_domain(false);
             }
             if (qemu_reset_requested_get()) {
                 qemu_system_reset(VMRESET_REPORT);
+                destroy_hvm_domain(true);
             }
         }
 
@@ -889,7 +985,6 @@ static void xen_main_loop_prepare(XenIOState *state)
 
     state->buffered_io_timer = qemu_new_timer_ms(rt_clock, handle_buffered_io,
                                                  state);
-    qemu_mod_timer(state->buffered_io_timer, qemu_get_clock_ms(rt_clock));
 
     if (evtchn_fd != -1) {
         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
@@ -983,13 +1078,13 @@ static void xen_read_physmap(XenIOState *state)
         QLIST_INSERT_HEAD(&state->physmap, physmap, list);
     }
     free(entries);
-    return;
 }
 
 int xen_hvm_init(void)
 {
     int i, rc;
     unsigned long ioreq_pfn;
+    unsigned long bufioreq_evtchn;
     XenIOState *state;
 
     state = g_malloc0(sizeof (XenIOState));
@@ -1042,15 +1137,29 @@ int xen_hvm_init(void)
         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);
+    if (rc == -1) {
+        fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
+        return -1;
+    }
+    state->bufioreq_local_port = rc;
+
     /* Init RAM management */
-    xen_map_cache_init();
+    xen_map_cache_init(xen_phys_offset_to_gaddr, state);
     xen_ram_init(ram_size);
 
     qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
 
     state->memory_listener = xen_memory_listener;
     QLIST_INIT(&state->physmap);
-    memory_listener_register(&state->memory_listener);
+    memory_listener_register(&state->memory_listener, &address_space_memory);
     state->log_for_dirtybit = NULL;
 
     /* Initialize backend core & drivers */
@@ -1066,7 +1175,7 @@ int xen_hvm_init(void)
     return 0;
 }
 
-void destroy_hvm_domain(void)
+void destroy_hvm_domain(bool reboot)
 {
     XenXC xc_handle;
     int sts;
@@ -1075,12 +1184,15 @@ void destroy_hvm_domain(void)
     if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
         fprintf(stderr, "Cannot acquire xenctrl handle\n");
     } else {
-        sts = xc_domain_shutdown(xc_handle, xen_domid, SHUTDOWN_poweroff);
+        sts = xc_domain_shutdown(xc_handle, xen_domid,
+                                 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
         if (sts != 0) {
-            fprintf(stderr, "? xc_domain_shutdown failed to issue poweroff, "
-                    "sts %d, %s\n", sts, strerror(errno));
+            fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
+                    "sts %d, %s\n", reboot ? "reboot" : "poweroff",
+                    sts, strerror(errno));
         } else {
-            fprintf(stderr, "Issued domain %d poweroff\n", xen_domid);
+            fprintf(stderr, "Issued domain %d %s\n", xen_domid,
+                    reboot ? "reboot" : "poweroff");
         }
         xc_interface_close(xc_handle);
     }
@@ -1090,3 +1202,36 @@ void xen_register_framebuffer(MemoryRegion *mr)
 {
     framebuffer = mr;
 }
+
+void xen_shutdown_fatal_error(const char *fmt, ...)
+{
+    va_list ap;
+
+    va_start(ap, fmt);
+    vfprintf(stderr, fmt, ap);
+    va_end(ap);
+    fprintf(stderr, "Will destroy the domain.\n");
+    /* destroy the domain */
+    qemu_system_shutdown_request();
+}
+
+void xen_modified_memory(ram_addr_t start, ram_addr_t length)
+{
+    if (unlikely(xen_in_migration)) {
+        int rc;
+        ram_addr_t start_pfn, nb_pages;
+
+        if (length == 0) {
+            length = TARGET_PAGE_SIZE;
+        }
+        start_pfn = start >> TARGET_PAGE_BITS;
+        nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
+            - start_pfn;
+        rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
+        if (rc) {
+            fprintf(stderr,
+                    "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
+                    __func__, start, nb_pages, rc, strerror(-rc));
+        }
+    }
+}