[Qemu-devel] [PATCH 13/13] xen: pv domain builder.

[Gerd Hoffmann <kraxel@redhat.com>]

This adds domain building support for paravirtual domains to qemu.
This allows booting xen guests directly with qemu, without Xend
and the management stack.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
---
 Makefile.target      |    2 +-
 configure            |    2 +-
 hw/xen_backend.h     |    3 +
 hw/xen_devconfig.c   |   29 +++++
 hw/xen_domainbuild.c |  281 ++++++++++++++++++++++++++++++++++++++++++++++++++
 hw/xen_domainbuild.h |    9 ++
 hw/xen_machine_pv.c  |   34 ++++---
 7 files changed, 344 insertions(+), 16 deletions(-)
 create mode 100644 hw/xen_domainbuild.c
 create mode 100644 hw/xen_domainbuild.h

diff --git a/Makefile.target b/Makefile.target
index 7e507e4..47bfc77 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -516,7 +516,7 @@ LIBS += $(CONFIG_VNC_TLS_LIBS)
 endif
 
 # xen backend driver support
-XEN_OBJS := xen_machine_pv.o xen_backend.o xen_devconfig.o
+XEN_OBJS := xen_machine_pv.o xen_backend.o xen_devconfig.o xen_domainbuild.o
 XEN_OBJS += xen_console.o xen_framebuffer.o xen_disk.o xen_nic.o
 ifeq ($(CONFIG_XEN), yes)
   OBJS += $(XEN_OBJS)
diff --git a/configure b/configure
index 14689d2..cec0b48 100755
--- a/configure
+++ b/configure
@@ -1228,7 +1228,7 @@ fi
 if test "$xen" = "yes" ; then
   echo "CONFIG_XEN=yes" >> $config_mak
   echo "#define CONFIG_XEN 1" >> $config_h
-  echo "XEN_LIBS=-lxenstore -lxenctrl" >> $config_mak
+  echo "XEN_LIBS=-lxenstore -lxenctrl -lxenguest" >> $config_mak
 fi
 if test "$aio" = "yes" ; then
   echo "#define CONFIG_AIO 1" >> $config_h
diff --git a/hw/xen_backend.h b/hw/xen_backend.h
index f591743..70ad016 100644
--- a/hw/xen_backend.h
+++ b/hw/xen_backend.h
@@ -99,5 +99,8 @@ void xen_set_display(int domid, DisplayState *ds);
 void xen_config_cleanup(void);
 int xen_config_dev_blk(DriveInfo *disk);
 int xen_config_dev_nic(NICInfo *nic);
+int xen_config_dev_vfb(int vdev, char *type);
+int xen_config_dev_vkbd(int vdev);
+int xen_config_dev_console(int vdev);
 
 #endif /* QEMU_XEN_BACKEND_H */
diff --git a/hw/xen_devconfig.c b/hw/xen_devconfig.c
index 25d28b4..0d506ca 100644
--- a/hw/xen_devconfig.c
+++ b/hw/xen_devconfig.c
@@ -142,3 +142,32 @@ int xen_config_dev_nic(NICInfo *nic)
     /* common stuff */
     return xen_config_dev_all(fe, be);
 }
+
+int xen_config_dev_vfb(int vdev, char *type)
+{
+    char fe[256], be[256];
+
+    xen_config_dev_dirs("vfb", "vfb", vdev, fe, be, sizeof(fe));
+
+    /* backend */
+    xenstore_write_str(be, "type",  type);
+
+    /* common stuff */
+    return xen_config_dev_all(fe, be);
+}
+
+int xen_config_dev_vkbd(int vdev)
+{
+    char fe[256], be[256];
+
+    xen_config_dev_dirs("vkbd", "vkbd", vdev, fe, be, sizeof(fe));
+    return xen_config_dev_all(fe, be);
+}
+
+int xen_config_dev_console(int vdev)
+{
+    char fe[256], be[256];
+
+    xen_config_dev_dirs("console", "console", vdev, fe, be, sizeof(fe));
+    return xen_config_dev_all(fe, be);
+}
diff --git a/hw/xen_domainbuild.c b/hw/xen_domainbuild.c
new file mode 100644
index 0000000..deba98d
--- /dev/null
+++ b/hw/xen_domainbuild.c
@@ -0,0 +1,281 @@
+#include "xen_backend.h"
+#include "xen_domainbuild.h"
+#include "sysemu.h"
+#include "qemu-timer.h"
+
+#include <xenguest.h>
+
+/* temporary */
+static const char qemu_uuid[] = "f0c9d298-3dd9-4c20-8c90-0205ac932241";
+
+static int xenstore_domain_mkdir(char *path)
+{
+    struct xs_permissions perms = {
+	.id    = xen_domid,
+	.perms = XS_PERM_WRITE,
+    };
+
+    if (!xs_mkdir(xenstore, 0, path)) {
+        fprintf(stderr, "%s: xs_mkdir %s: failed\n", __FUNCTION__, path);
+	return -1;
+    }
+    if (!xs_set_permissions(xenstore, 0, path, &perms, 1)) {
+        fprintf(stderr, "%s: xs_set_permissions failed\n", __FUNCTION__);
+	return -1;
+    }
+    return 0;
+}
+
+static int xenstore_domain_init1(const char *kernel, const char *ramdisk,
+                                 const char *cmdline)
+{
+    char *dom, vm[256], path[256];
+    int i;
+
+    dom = xs_get_domain_path(xenstore, xen_domid);
+    snprintf(vm,  sizeof(vm),  "/vm/%s", qemu_uuid);
+
+    xenstore_domain_mkdir(dom);
+
+    xenstore_write_str(vm, "image/ostype",  "linux");
+    if (kernel)
+        xenstore_write_str(vm, "image/kernel",  kernel);
+    if (ramdisk)
+        xenstore_write_str(vm, "image/ramdisk", ramdisk);
+    if (cmdline)
+        xenstore_write_str(vm, "image/cmdline", cmdline);
+
+    /* name + id */
+    xenstore_write_str(vm,  "name",   qemu_name ? qemu_name : "no-name");
+    xenstore_write_str(vm,  "uuid",   qemu_uuid);
+    xenstore_write_str(dom, "name",   qemu_name ? qemu_name : "no-name");
+    xenstore_write_int(dom, "domid",  xen_domid);
+    xenstore_write_str(dom, "vm",     vm);
+
+    /* memory */
+    xenstore_write_int(dom, "memory/target", ram_size >> 10);  // kB
+    xenstore_write_int(vm, "memory",         ram_size >> 20);  // MB
+    xenstore_write_int(vm, "maxmem",         ram_size >> 20);  // MB
+
+    /* cpus */
+    for (i = 0; i < smp_cpus; i++) {
+	snprintf(path, sizeof(path), "cpu/%d/availability",i);
+	xenstore_write_str(dom, path, "online");
+    }
+    xenstore_write_int(vm, "vcpu_avail",  smp_cpus);
+    xenstore_write_int(vm, "vcpus",       smp_cpus);
+
+    /* vnc password */
+    xenstore_write_str(vm, "vncpassword", "" /* FIXME */);
+
+    free(dom);
+    return 0;
+}
+
+static int xenstore_domain_init2(int xenstore_port, int xenstore_mfn,
+                                 int console_port, int console_mfn)
+{
+    char *dom;
+
+    dom = xs_get_domain_path(xenstore, xen_domid);
+
+    /* signal new domain */
+    xs_introduce_domain(xenstore,
+                        xen_domid,
+                        xenstore_mfn,
+                        xenstore_port);
+
+    /* xenstore */
+    xenstore_write_int(dom, "store/ring-ref",   xenstore_mfn);
+    xenstore_write_int(dom, "store/port",       xenstore_port);
+
+    /* console */
+    xenstore_write_str(dom, "console/type",     "ioemu");
+    xenstore_write_int(dom, "console/limit",    128 * 1024);
+    xenstore_write_int(dom, "console/ring-ref", console_mfn);
+    xenstore_write_int(dom, "console/port",     console_port);
+    xen_config_dev_console(0);
+
+    /* devices */
+    if (1 /* FIXME */) {
+	xen_config_dev_vfb(0, "vnc");
+	xen_config_dev_vkbd(0);
+    }
+
+    free(dom);
+    return 0;
+}
+
+/* ------------------------------------------------------------- */
+
+static QEMUTimer *xen_poll;
+
+/* check domain state once per second */
+static void xen_domain_poll(void *opaque)
+{
+    struct xc_dominfo info;
+    int rc;
+
+    rc = xc_domain_getinfo(xen_xc, xen_domid, 1, &info);
+    if ((1 != rc) || (info.domid != xen_domid)) {
+        fprintf(stderr, "xen: domain %d is gone\n", xen_domid);
+        goto quit;
+    }
+    if (info.dying) {
+        fprintf(stderr, "xen: domain %d is dying (%s%s)\n", xen_domid,
+                info.crashed  ? "crashed"  : "",
+                info.shutdown ? "shutdown" : "");
+        goto quit;
+    }
+
+    qemu_mod_timer(xen_poll, qemu_get_clock(rt_clock) + 1000);
+    return;
+
+quit:
+    qemu_system_shutdown_request();
+    return;
+}
+
+static void xen_domain_watcher(void)
+{
+    int qemu_running = 1;
+    int fd[2], i, rc;
+    char byte;
+
+    pipe(fd);
+    if (0 != fork())
+        return; /* not child */
+
+    /* close all file handles, except stdio/out/err,
+     * our watch pipe and the xen interface handle */
+    for (i = 3; i < 256; i++) {
+        if (i == fd[0])
+            continue;
+        if (i == xen_xc)
+            continue;
+        close(i);
+    }
+
+    /* wait for qemu exiting */
+    while (qemu_running) {
+        rc = read(fd[0], &byte, 1);
+        switch (rc) {
+        case -1:
+            if (EINTR == errno)
+                continue;
+            fprintf(stderr, "%s: Huh? read error: %s\n", __FUNCTION__, strerror(errno));
+            qemu_running = 0;
+            break;
+        case 0:
+            /* EOF -> qemu exited */
+            qemu_running = 0;
+            break;
+        default:
+            fprintf(stderr, "%s: Huh? data on the watch pipe?\n", __FUNCTION__);
+            break;
+        }
+    }
+
+    /* cleanup */
+    fprintf(stderr, "%s: destroy domain %d\n", __FUNCTION__, xen_domid);
+    xc_domain_destroy(xen_xc, xen_domid);
+    _exit(0);
+}
+
+/* normal cleanup */
+static void xen_domain_cleanup(void)
+{
+    char *dom;
+
+    dom = xs_get_domain_path(xenstore, xen_domid);
+    if (dom) {
+        xs_rm(xenstore, 0, dom);
+        free(dom);
+    }
+    xs_release_domain(xenstore, xen_domid);
+}
+
+int xen_domain_build_pv(const char *kernel, const char *ramdisk,
+                        const char *cmdline)
+{
+    uint32_t ssidref = 0;
+    uint32_t flags = 0;
+    xen_domain_handle_t uuid;
+    unsigned int xenstore_port = 0, console_port = 0;
+    unsigned long xenstore_mfn = 0, console_mfn = 0;
+    int rc, i, pos, val;
+
+    for (i = 0, pos = 0; i < 16; i++, pos += 2) {
+        if (i == 4 || i == 6 || i == 8 || i == 10) {
+            /* skip dashes */
+            if ('-' != qemu_uuid[pos]) {
+                fprintf(stderr, "xen: uuid parse error");
+                goto err;
+            }
+            pos++;
+        }
+        if (1 != sscanf(qemu_uuid+pos, "%02x", &val)) {
+            fprintf(stderr, "xen: uuid parse error");
+            goto err;
+        }
+        uuid[i] = val;
+    }
+
+    rc = xc_domain_create(xen_xc, ssidref, uuid, flags, &xen_domid);
+    if (rc < 0) {
+        fprintf(stderr, "xen: xc_domain_create() failed\n");
+        goto err;
+    }
+    fprintf(stderr, "xen: created domain %d\n", xen_domid);
+    atexit(xen_domain_cleanup);
+    xen_domain_watcher();
+
+    xenstore_domain_init1(kernel, ramdisk, cmdline);
+
+    rc = xc_domain_max_vcpus(xen_xc, xen_domid, smp_cpus);
+    if (rc < 0) {
+        fprintf(stderr, "xen: xc_domain_max_vcpus() failed\n");
+        goto err;
+    }
+
+    rc = xc_domain_setcpuweight(xen_xc, xen_domid, 256);
+    if (rc < 0) {
+        fprintf(stderr, "xen: xc_domain_setcpuweight() failed\n");
+        goto err;
+    }
+
+    rc = xc_domain_setmaxmem(xen_xc, xen_domid, ram_size >> 10);
+    if (rc < 0) {
+        fprintf(stderr, "xen: xc_domain_setmaxmem() failed\n");
+        goto err;
+    }
+
+    xenstore_port = xc_evtchn_alloc_unbound(xen_xc, xen_domid, 0);
+    console_port = xc_evtchn_alloc_unbound(xen_xc, xen_domid, 0);
+
+    rc = xc_linux_build(xen_xc, xen_domid, ram_size >> 20,
+                        kernel, ramdisk, cmdline,
+                        0, flags,
+                        xenstore_port, &xenstore_mfn,
+                        console_port, &console_mfn);
+    if (rc < 0) {
+        fprintf(stderr, "xen: xc_linux_build() failed\n");
+        goto err;
+    }
+
+    xenstore_domain_init2(xenstore_port, xenstore_mfn,
+                          console_port, console_mfn);
+
+    rc = xc_domain_unpause(xen_xc, xen_domid);
+    if (rc < 0) {
+        fprintf(stderr, "xen: xc_domain_unpause() failed\n");
+        goto err;
+    }
+
+    xen_poll = qemu_new_timer(rt_clock, xen_domain_poll, NULL);
+    qemu_mod_timer(xen_poll, qemu_get_clock(rt_clock) + 1000);
+    return 0;
+
+err:
+    return -1;
+}
diff --git a/hw/xen_domainbuild.h b/hw/xen_domainbuild.h
new file mode 100644
index 0000000..21ee180
--- /dev/null
+++ b/hw/xen_domainbuild.h
@@ -0,0 +1,9 @@
+#ifndef QEMU_XEN_DOMAINBUILD_H
+#define QEMU_XEN_DOMAINBUILD_H 1
+
+#include "xen_common.h"
+
+int xen_domain_build_pv(const char *kernel, const char *ramdisk,
+                        const char *cmdline);
+
+#endif /* QEMU_XEN_DOMAINBUILD_H */
diff --git a/hw/xen_machine_pv.c b/hw/xen_machine_pv.c
index 9f18742..8c9f3c1 100644
--- a/hw/xen_machine_pv.c
+++ b/hw/xen_machine_pv.c
@@ -27,6 +27,7 @@
 #include "console.h"
 
 #include "xen_backend.h"
+#include "xen_domainbuild.h"
 
 /* -------------------------------------------------------------------- */
 /* variables                                                            */
@@ -90,7 +91,9 @@ static int xen_init(void)
     return 0;
 }
 
-static int xen_init_pv(DisplayState *ds)
+static int xen_init_pv(const char *kernel_filename,
+		       const char *kernel_cmdline,
+		       const char *initrd_filename)
 {
     int rc;
 
@@ -98,6 +101,19 @@ static int xen_init_pv(DisplayState *ds)
     if (rc < 0)
         return rc;
 
+    if (xen_emulate) {
+        fprintf(stderr, "xen pv emulation not implemented yet\n");
+        return -1;
+    }
+
+    if (xen_domainbuild) {
+        if (xen_domain_build_pv(kernel_filename, initrd_filename,
+                                kernel_cmdline) < 0) {
+            fprintf(stderr, "xen pv domain creation failed\n");
+            return -1;
+        }
+    }
+
     /* xenbus backend drivers */
     xen_be_register("console", &xen_console_ops);
     xen_be_register("vkbd", &xen_kbdmouse_ops);
@@ -105,9 +121,6 @@ static int xen_init_pv(DisplayState *ds)
     xen_be_register("qdisk", &xen_blkdev_ops);
     xen_be_register("qnic", &xen_netdev_ops);
 
-    /* setup framebuffer */
-    xen_set_display(xen_domid, ds);
-
     return 0;
 }
 
@@ -147,19 +160,12 @@ static void xenpv_init(ram_addr_t ram_size, int vga_ram_size,
     CPUState *env;
     int index,i,rc;
 
-    rc = xen_init_pv(ds);
+    rc = xen_init_pv(kernel_filename, kernel_cmdline, initrd_filename);
     if (-1 == rc)
         goto err;
 
-    if (xen_emulate) {
-        fprintf(stderr, "xen pv emulation not implemented yet\n");
-        goto err;
-    }
-    if (xen_domainbuild) {
-        fprintf(stderr, "xen pv domain creation not implemented yet\n");
-        goto err;
-    }
-
+    /* setup framebuffer */
+    xen_set_display(xen_domid, ds);
     xen_init_vnc();
 
     /* create dummy cpu, halted */
-- 
1.5.5.1