Merge branch 'stable/for-linus-3.3' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Tue, 10 Jan 2012 18:09:59 +0000 (10:09 -0800)]
* 'stable/for-linus-3.3' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen: (37 commits)
  xen/pciback: Expand the warning message to include domain id.
  xen/pciback: Fix "device has been assigned to X domain!" warning
  xen/pciback: Move the PCI_DEV_FLAGS_ASSIGNED ops to the "[un|]bind"
  xen/xenbus: don't reimplement kvasprintf via a fixed size buffer
  xenbus: maximum buffer size is XENSTORE_PAYLOAD_MAX
  xen/xenbus: Reject replies with payload > XENSTORE_PAYLOAD_MAX.
  Xen: consolidate and simplify struct xenbus_driver instantiation
  xen-gntalloc: introduce missing kfree
  xen/xenbus: Fix compile error - missing header for xen_initial_domain()
  xen/netback: Enable netback on HVM guests
  xen/grant-table: Support mappings required by blkback
  xenbus: Use grant-table wrapper functions
  xenbus: Support HVM backends
  xen/xenbus-frontend: Fix compile error with randconfig
  xen/xenbus-frontend: Make error message more clear
  xen/privcmd: Remove unused support for arch specific privcmp mmap
  xen: Add xenbus_backend device
  xen: Add xenbus device driver
  xen: Add privcmd device driver
  xen/gntalloc: fix reference counts on multi-page mappings
  ...

44 files changed:
Documentation/ABI/stable/sysfs-bus-xen-backend [new file with mode: 0644]
Documentation/ABI/stable/sysfs-devices-system-xen_memory [new file with mode: 0644]
arch/ia64/include/asm/xen/interface.h
arch/x86/xen/Kconfig
arch/x86/xen/grant-table.c
drivers/block/xen-blkback/xenbus.c
drivers/block/xen-blkfront.c
drivers/input/misc/xen-kbdfront.c
drivers/net/xen-netback/netback.c
drivers/net/xen-netback/xenbus.c
drivers/net/xen-netfront.c
drivers/pci/xen-pcifront.c
drivers/video/xen-fbfront.c
drivers/xen/Kconfig
drivers/xen/Makefile
drivers/xen/events.c
drivers/xen/evtchn.c
drivers/xen/gntalloc.c
drivers/xen/gntdev.c
drivers/xen/grant-table.c
drivers/xen/privcmd.c [moved from drivers/xen/xenfs/privcmd.c with 92% similarity]
drivers/xen/privcmd.h [new file with mode: 0644]
drivers/xen/xen-pciback/pci_stub.c
drivers/xen/xen-pciback/xenbus.c
drivers/xen/xenbus/Makefile
drivers/xen/xenbus/xenbus_client.c
drivers/xen/xenbus/xenbus_comms.h
drivers/xen/xenbus/xenbus_dev_backend.c [new file with mode: 0644]
drivers/xen/xenbus/xenbus_dev_frontend.c [moved from drivers/xen/xenfs/xenbus.c with 95% similarity]
drivers/xen/xenbus/xenbus_probe.c
drivers/xen/xenbus/xenbus_probe.h
drivers/xen/xenbus/xenbus_probe_backend.c
drivers/xen/xenbus/xenbus_probe_frontend.c
drivers/xen/xenbus/xenbus_xs.c
drivers/xen/xenfs/Makefile
drivers/xen/xenfs/super.c
drivers/xen/xenfs/xenfs.h
include/xen/events.h
include/xen/grant_table.h
include/xen/interface/grant_table.h
include/xen/interface/io/xs_wire.h
include/xen/interface/xen.h
include/xen/xenbus.h
include/xen/xenbus_dev.h [new file with mode: 0644]

diff --git a/Documentation/ABI/stable/sysfs-bus-xen-backend b/Documentation/ABI/stable/sysfs-bus-xen-backend
new file mode 100644 (file)
index 0000000..3d5951c
--- /dev/null
@@ -0,0 +1,75 @@
+What:          /sys/bus/xen-backend/devices/*/devtype
+Date:          Feb 2009
+KernelVersion: 2.6.38
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                The type of the device.  e.g., one of: 'vbd' (block),
+                'vif' (network), or 'vfb' (framebuffer).
+
+What:          /sys/bus/xen-backend/devices/*/nodename
+Date:          Feb 2009
+KernelVersion: 2.6.38
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                XenStore node (under /local/domain/NNN/) for this
+                backend device.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/physical_device
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                The major:minor number (in hexidecimal) of the
+                physical device providing the storage for this backend
+                block device.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/mode
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Whether the block device is read-only ('r') or
+                read-write ('w').
+
+What:          /sys/bus/xen-backend/devices/vbd-*/statistics/f_req
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Number of flush requests from the frontend.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/statistics/oo_req
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Number of requests delayed because the backend was too
+                busy processing previous requests.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/statistics/rd_req
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Number of read requests from the frontend.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/statistics/rd_sect
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Number of sectors read by the frontend.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/statistics/wr_req
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Number of write requests from the frontend.
+
+What:          /sys/bus/xen-backend/devices/vbd-*/statistics/wr_sect
+Date:          April 2011
+KernelVersion: 3.0
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+                Number of sectors written by the frontend.
diff --git a/Documentation/ABI/stable/sysfs-devices-system-xen_memory b/Documentation/ABI/stable/sysfs-devices-system-xen_memory
new file mode 100644 (file)
index 0000000..caa311d
--- /dev/null
@@ -0,0 +1,77 @@
+What:          /sys/devices/system/xen_memory/xen_memory0/max_retry_count
+Date:          May 2011
+KernelVersion: 2.6.39
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               The maximum number of times the balloon driver will
+               attempt to increase the balloon before giving up.  See
+               also 'retry_count' below.
+               A value of zero means retry forever and is the default one.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/max_schedule_delay
+Date:          May 2011
+KernelVersion: 2.6.39
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               The limit that 'schedule_delay' (see below) will be
+               increased to. The default value is 32 seconds.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/retry_count
+Date:          May 2011
+KernelVersion: 2.6.39
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               The current number of times that the balloon driver
+               has attempted to increase the size of the balloon.
+               The default value is one. With max_retry_count being
+               zero (unlimited), this means that the driver will attempt
+               to retry with a 'schedule_delay' delay.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/schedule_delay
+Date:          May 2011
+KernelVersion: 2.6.39
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               The time (in seconds) to wait between attempts to
+               increase the balloon.  Each time the balloon cannot be
+               increased, 'schedule_delay' is increased (until
+               'max_schedule_delay' is reached at which point it
+               will use the max value).
+
+What:          /sys/devices/system/xen_memory/xen_memory0/target
+Date:          April 2008
+KernelVersion: 2.6.26
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               The target number of pages to adjust this domain's
+               memory reservation to.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/target_kb
+Date:          April 2008
+KernelVersion: 2.6.26
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               As target above, except the value is in KiB.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/info/current_kb
+Date:          April 2008
+KernelVersion: 2.6.26
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               Current size (in KiB) of this domain's memory
+               reservation.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/info/high_kb
+Date:          April 2008
+KernelVersion: 2.6.26
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               Amount (in KiB) of high memory in the balloon.
+
+What:          /sys/devices/system/xen_memory/xen_memory0/info/low_kb
+Date:          April 2008
+KernelVersion: 2.6.26
+Contact:       Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Description:
+               Amount (in KiB) of low (or normal) memory in the
+               balloon.
index 1d2427d..fbb5198 100644 (file)
@@ -71,7 +71,7 @@
 __DEFINE_GUEST_HANDLE(uchar, unsigned char);
 __DEFINE_GUEST_HANDLE(uint, unsigned int);
 __DEFINE_GUEST_HANDLE(ulong, unsigned long);
-__DEFINE_GUEST_HANDLE(u64, unsigned long);
+
 DEFINE_GUEST_HANDLE(char);
 DEFINE_GUEST_HANDLE(int);
 DEFINE_GUEST_HANDLE(long);
index 26c731a..fdce49c 100644 (file)
@@ -29,7 +29,8 @@ config XEN_PVHVM
 
 config XEN_MAX_DOMAIN_MEMORY
        int
-       default 128
+       default 500 if X86_64
+       default 64 if X86_32
        depends on XEN
        help
          This only affects the sizing of some bss arrays, the unused
@@ -48,3 +49,4 @@ config XEN_DEBUG_FS
        help
          Enable statistics output and various tuning options in debugfs.
          Enabling this option may incur a significant performance overhead.
+
index 5a40d24..3a5f55d 100644 (file)
@@ -54,6 +54,20 @@ static int map_pte_fn(pte_t *pte, struct page *pmd_page,
        return 0;
 }
 
+/*
+ * This function is used to map shared frames to store grant status. It is
+ * different from map_pte_fn above, the frames type here is uint64_t.
+ */
+static int map_pte_fn_status(pte_t *pte, struct page *pmd_page,
+                            unsigned long addr, void *data)
+{
+       uint64_t **frames = (uint64_t **)data;
+
+       set_pte_at(&init_mm, addr, pte, mfn_pte((*frames)[0], PAGE_KERNEL));
+       (*frames)++;
+       return 0;
+}
+
 static int unmap_pte_fn(pte_t *pte, struct page *pmd_page,
                        unsigned long addr, void *data)
 {
@@ -64,10 +78,10 @@ static int unmap_pte_fn(pte_t *pte, struct page *pmd_page,
 
 int arch_gnttab_map_shared(unsigned long *frames, unsigned long nr_gframes,
                           unsigned long max_nr_gframes,
-                          struct grant_entry **__shared)
+                          void **__shared)
 {
        int rc;
-       struct grant_entry *shared = *__shared;
+       void *shared = *__shared;
 
        if (shared == NULL) {
                struct vm_struct *area =
@@ -83,8 +97,30 @@ int arch_gnttab_map_shared(unsigned long *frames, unsigned long nr_gframes,
        return rc;
 }
 
-void arch_gnttab_unmap_shared(struct grant_entry *shared,
-                             unsigned long nr_gframes)
+int arch_gnttab_map_status(uint64_t *frames, unsigned long nr_gframes,
+                          unsigned long max_nr_gframes,
+                          grant_status_t **__shared)
+{
+       int rc;
+       grant_status_t *shared = *__shared;
+
+       if (shared == NULL) {
+               /* No need to pass in PTE as we are going to do it
+                * in apply_to_page_range anyhow. */
+               struct vm_struct *area =
+                       alloc_vm_area(PAGE_SIZE * max_nr_gframes, NULL);
+               BUG_ON(area == NULL);
+               shared = area->addr;
+               *__shared = shared;
+       }
+
+       rc = apply_to_page_range(&init_mm, (unsigned long)shared,
+                                PAGE_SIZE * nr_gframes,
+                                map_pte_fn_status, &frames);
+       return rc;
+}
+
+void arch_gnttab_unmap(void *shared, unsigned long nr_gframes)
 {
        apply_to_page_range(&init_mm, (unsigned long)shared,
                            PAGE_SIZE * nr_gframes, unmap_pte_fn, NULL);
index 8069322..37c794d 100644 (file)
@@ -787,17 +787,14 @@ static const struct xenbus_device_id xen_blkbk_ids[] = {
 };
 
 
-static struct xenbus_driver xen_blkbk = {
-       .name = "vbd",
-       .owner = THIS_MODULE,
-       .ids = xen_blkbk_ids,
+static DEFINE_XENBUS_DRIVER(xen_blkbk, ,
        .probe = xen_blkbk_probe,
        .remove = xen_blkbk_remove,
        .otherend_changed = frontend_changed
-};
+);
 
 
 int xen_blkif_xenbus_init(void)
 {
-       return xenbus_register_backend(&xen_blkbk);
+       return xenbus_register_backend(&xen_blkbk_driver);
 }
index 7b2ec59..9fd3ee2 100644 (file)
@@ -1437,16 +1437,13 @@ static const struct xenbus_device_id blkfront_ids[] = {
        { "" }
 };
 
-static struct xenbus_driver blkfront = {
-       .name = "vbd",
-       .owner = THIS_MODULE,
-       .ids = blkfront_ids,
+static DEFINE_XENBUS_DRIVER(blkfront, ,
        .probe = blkfront_probe,
        .remove = blkfront_remove,
        .resume = blkfront_resume,
        .otherend_changed = blkback_changed,
        .is_ready = blkfront_is_ready,
-};
+);
 
 static int __init xlblk_init(void)
 {
@@ -1461,7 +1458,7 @@ static int __init xlblk_init(void)
                return -ENODEV;
        }
 
-       ret = xenbus_register_frontend(&blkfront);
+       ret = xenbus_register_frontend(&blkfront_driver);
        if (ret) {
                unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
                return ret;
@@ -1474,7 +1471,7 @@ module_init(xlblk_init);
 
 static void __exit xlblk_exit(void)
 {
-       return xenbus_unregister_driver(&blkfront);
+       return xenbus_unregister_driver(&blkfront_driver);
 }
 module_exit(xlblk_exit);
 
index ad2e51c..02ca868 100644 (file)
@@ -361,15 +361,12 @@ static const struct xenbus_device_id xenkbd_ids[] = {
        { "" }
 };
 
-static struct xenbus_driver xenkbd_driver = {
-       .name = "vkbd",
-       .owner = THIS_MODULE,
-       .ids = xenkbd_ids,
+static DEFINE_XENBUS_DRIVER(xenkbd, ,
        .probe = xenkbd_probe,
        .remove = xenkbd_remove,
        .resume = xenkbd_resume,
        .otherend_changed = xenkbd_backend_changed,
-};
+);
 
 static int __init xenkbd_init(void)
 {
index 639cf8a..59effac 100644 (file)
@@ -1634,7 +1634,7 @@ static int __init netback_init(void)
        int rc = 0;
        int group;
 
-       if (!xen_pv_domain())
+       if (!xen_domain())
                return -ENODEV;
 
        xen_netbk_group_nr = num_online_cpus();
index 1ce729d..410018c 100644 (file)
@@ -474,17 +474,14 @@ static const struct xenbus_device_id netback_ids[] = {
 };
 
 
-static struct xenbus_driver netback = {
-       .name = "vif",
-       .owner = THIS_MODULE,
-       .ids = netback_ids,
+static DEFINE_XENBUS_DRIVER(netback, ,
        .probe = netback_probe,
        .remove = netback_remove,
        .uevent = netback_uevent,
        .otherend_changed = frontend_changed,
-};
+);
 
 int xenvif_xenbus_init(void)
 {
-       return xenbus_register_backend(&netback);
+       return xenbus_register_backend(&netback_driver);
 }
index 0a59c57..fa67905 100644 (file)
@@ -1914,7 +1914,7 @@ static void xennet_sysfs_delif(struct net_device *netdev)
 
 #endif /* CONFIG_SYSFS */
 
-static struct xenbus_device_id netfront_ids[] = {
+static const struct xenbus_device_id netfront_ids[] = {
        { "vif" },
        { "" }
 };
@@ -1941,15 +1941,12 @@ static int __devexit xennet_remove(struct xenbus_device *dev)
        return 0;
 }
 
-static struct xenbus_driver netfront_driver = {
-       .name = "vif",
-       .owner = THIS_MODULE,
-       .ids = netfront_ids,
+static DEFINE_XENBUS_DRIVER(netfront, ,
        .probe = netfront_probe,
        .remove = __devexit_p(xennet_remove),
        .resume = netfront_resume,
        .otherend_changed = netback_changed,
-};
+);
 
 static int __init netif_init(void)
 {
index 90832a9..7cf3d2f 100644 (file)
@@ -1126,14 +1126,11 @@ static const struct xenbus_device_id xenpci_ids[] = {
        {""},
 };
 
-static struct xenbus_driver xenbus_pcifront_driver = {
-       .name                   = "pcifront",
-       .owner                  = THIS_MODULE,
-       .ids                    = xenpci_ids,
+static DEFINE_XENBUS_DRIVER(xenpci, "pcifront",
        .probe                  = pcifront_xenbus_probe,
        .remove                 = pcifront_xenbus_remove,
        .otherend_changed       = pcifront_backend_changed,
-};
+);
 
 static int __init pcifront_init(void)
 {
@@ -1142,12 +1139,12 @@ static int __init pcifront_init(void)
 
        pci_frontend_registrar(1 /* enable */);
 
-       return xenbus_register_frontend(&xenbus_pcifront_driver);
+       return xenbus_register_frontend(&xenpci_driver);
 }
 
 static void __exit pcifront_cleanup(void)
 {
-       xenbus_unregister_driver(&xenbus_pcifront_driver);
+       xenbus_unregister_driver(&xenpci_driver);
        pci_frontend_registrar(0 /* disable */);
 }
 module_init(pcifront_init);
index beac52f..cb4529c 100644 (file)
@@ -671,20 +671,17 @@ InitWait:
        }
 }
 
-static struct xenbus_device_id xenfb_ids[] = {
+static const struct xenbus_device_id xenfb_ids[] = {
        { "vfb" },
        { "" }
 };
 
-static struct xenbus_driver xenfb_driver = {
-       .name = "vfb",
-       .owner = THIS_MODULE,
-       .ids = xenfb_ids,
+static DEFINE_XENBUS_DRIVER(xenfb, ,
        .probe = xenfb_probe,
        .remove = xenfb_remove,
        .resume = xenfb_resume,
        .otherend_changed = xenfb_backend_changed,
-};
+);
 
 static int __init xenfb_init(void)
 {
index 8795480..a1ced52 100644 (file)
@@ -86,6 +86,7 @@ config XEN_BACKEND
 
 config XENFS
        tristate "Xen filesystem"
+       select XEN_PRIVCMD
        default y
        help
          The xen filesystem provides a way for domains to share
@@ -171,4 +172,10 @@ config XEN_PCIDEV_BACKEND
          xen-pciback.hide=(03:00.0)(04:00.0)
 
          If in doubt, say m.
+
+config XEN_PRIVCMD
+       tristate
+       depends on XEN
+       default m
+
 endmenu
index 974fffd..aa31337 100644 (file)
@@ -19,7 +19,9 @@ obj-$(CONFIG_XEN_TMEM)                        += tmem.o
 obj-$(CONFIG_SWIOTLB_XEN)              += swiotlb-xen.o
 obj-$(CONFIG_XEN_DOM0)                 += pci.o
 obj-$(CONFIG_XEN_PCIDEV_BACKEND)       += xen-pciback/
+obj-$(CONFIG_XEN_PRIVCMD)              += xen-privcmd.o
 
 xen-evtchn-y                           := evtchn.o
 xen-gntdev-y                           := gntdev.o
 xen-gntalloc-y                         := gntalloc.o
+xen-privcmd-y                          := privcmd.o
index 6e075cd..e5e5812 100644 (file)
@@ -87,6 +87,7 @@ enum xen_irq_type {
  */
 struct irq_info {
        struct list_head list;
+       int refcnt;
        enum xen_irq_type type; /* type */
        unsigned irq;
        unsigned short evtchn;  /* event channel */
@@ -406,6 +407,7 @@ static void xen_irq_init(unsigned irq)
                panic("Unable to allocate metadata for IRQ%d\n", irq);
 
        info->type = IRQT_UNBOUND;
+       info->refcnt = -1;
 
        irq_set_handler_data(irq, info);
 
@@ -469,6 +471,8 @@ static void xen_free_irq(unsigned irq)
 
        irq_set_handler_data(irq, NULL);
 
+       WARN_ON(info->refcnt > 0);
+
        kfree(info);
 
        /* Legacy IRQ descriptors are managed by the arch. */
@@ -637,7 +641,7 @@ int xen_bind_pirq_gsi_to_irq(unsigned gsi,
        if (irq != -1) {
                printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
                       irq, gsi);
-               goto out;       /* XXX need refcount? */
+               goto out;
        }
 
        irq = xen_allocate_irq_gsi(gsi);
@@ -939,9 +943,16 @@ static void unbind_from_irq(unsigned int irq)
 {
        struct evtchn_close close;
        int evtchn = evtchn_from_irq(irq);
+       struct irq_info *info = irq_get_handler_data(irq);
 
        mutex_lock(&irq_mapping_update_lock);
 
+       if (info->refcnt > 0) {
+               info->refcnt--;
+               if (info->refcnt != 0)
+                       goto done;
+       }
+
        if (VALID_EVTCHN(evtchn)) {
                close.port = evtchn;
                if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
@@ -970,6 +981,7 @@ static void unbind_from_irq(unsigned int irq)
 
        xen_free_irq(irq);
 
+ done:
        mutex_unlock(&irq_mapping_update_lock);
 }
 
@@ -1065,6 +1077,69 @@ void unbind_from_irqhandler(unsigned int irq, void *dev_id)
 }
 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
 
+int evtchn_make_refcounted(unsigned int evtchn)
+{
+       int irq = evtchn_to_irq[evtchn];
+       struct irq_info *info;
+
+       if (irq == -1)
+               return -ENOENT;
+
+       info = irq_get_handler_data(irq);
+
+       if (!info)
+               return -ENOENT;
+
+       WARN_ON(info->refcnt != -1);
+
+       info->refcnt = 1;
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
+
+int evtchn_get(unsigned int evtchn)
+{
+       int irq;
+       struct irq_info *info;
+       int err = -ENOENT;
+
+       if (evtchn >= NR_EVENT_CHANNELS)
+               return -EINVAL;
+
+       mutex_lock(&irq_mapping_update_lock);
+
+       irq = evtchn_to_irq[evtchn];
+       if (irq == -1)
+               goto done;
+
+       info = irq_get_handler_data(irq);
+
+       if (!info)
+               goto done;
+
+       err = -EINVAL;
+       if (info->refcnt <= 0)
+               goto done;
+
+       info->refcnt++;
+       err = 0;
+ done:
+       mutex_unlock(&irq_mapping_update_lock);
+
+       return err;
+}
+EXPORT_SYMBOL_GPL(evtchn_get);
+
+void evtchn_put(unsigned int evtchn)
+{
+       int irq = evtchn_to_irq[evtchn];
+       if (WARN_ON(irq == -1))
+               return;
+       unbind_from_irq(irq);
+}
+EXPORT_SYMBOL_GPL(evtchn_put);
+
 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
 {
        int irq = per_cpu(ipi_to_irq, cpu)[vector];
index dbc13e9..b1f60a0 100644 (file)
@@ -268,7 +268,7 @@ static int evtchn_bind_to_user(struct per_user_data *u, int port)
        rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, IRQF_DISABLED,
                                       u->name, (void *)(unsigned long)port);
        if (rc >= 0)
-               rc = 0;
+               rc = evtchn_make_refcounted(port);
 
        return rc;
 }
index e1c4c6e..934985d 100644 (file)
@@ -74,7 +74,7 @@ MODULE_PARM_DESC(limit, "Maximum number of grants that may be allocated by "
                "the gntalloc device");
 
 static LIST_HEAD(gref_list);
-static DEFINE_SPINLOCK(gref_lock);
+static DEFINE_MUTEX(gref_mutex);
 static int gref_size;
 
 struct notify_info {
@@ -99,6 +99,12 @@ struct gntalloc_file_private_data {
        uint64_t index;
 };
 
+struct gntalloc_vma_private_data {
+       struct gntalloc_gref *gref;
+       int users;
+       int count;
+};
+
 static void __del_gref(struct gntalloc_gref *gref);
 
 static void do_cleanup(void)
@@ -143,15 +149,15 @@ static int add_grefs(struct ioctl_gntalloc_alloc_gref *op,
        }
 
        /* Add to gref lists. */
-       spin_lock(&gref_lock);
+       mutex_lock(&gref_mutex);
        list_splice_tail(&queue_gref, &gref_list);
        list_splice_tail(&queue_file, &priv->list);
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
 
        return 0;
 
 undo:
-       spin_lock(&gref_lock);
+       mutex_lock(&gref_mutex);
        gref_size -= (op->count - i);
 
        list_for_each_entry(gref, &queue_file, next_file) {
@@ -167,7 +173,7 @@ undo:
         */
        if (unlikely(!list_empty(&queue_gref)))
                list_splice_tail(&queue_gref, &gref_list);
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
        return rc;
 }
 
@@ -178,8 +184,10 @@ static void __del_gref(struct gntalloc_gref *gref)
                tmp[gref->notify.pgoff] = 0;
                kunmap(gref->page);
        }
-       if (gref->notify.flags & UNMAP_NOTIFY_SEND_EVENT)
+       if (gref->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
                notify_remote_via_evtchn(gref->notify.event);
+               evtchn_put(gref->notify.event);
+       }
 
        gref->notify.flags = 0;
 
@@ -189,6 +197,8 @@ static void __del_gref(struct gntalloc_gref *gref)
 
                if (!gnttab_end_foreign_access_ref(gref->gref_id, 0))
                        return;
+
+               gnttab_free_grant_reference(gref->gref_id);
        }
 
        gref_size--;
@@ -251,7 +261,7 @@ static int gntalloc_release(struct inode *inode, struct file *filp)
 
        pr_debug("%s: priv %p\n", __func__, priv);
 
-       spin_lock(&gref_lock);
+       mutex_lock(&gref_mutex);
        while (!list_empty(&priv->list)) {
                gref = list_entry(priv->list.next,
                        struct gntalloc_gref, next_file);
@@ -261,7 +271,7 @@ static int gntalloc_release(struct inode *inode, struct file *filp)
                        __del_gref(gref);
        }
        kfree(priv);
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
 
        return 0;
 }
@@ -286,21 +296,21 @@ static long gntalloc_ioctl_alloc(struct gntalloc_file_private_data *priv,
                goto out;
        }
 
-       spin_lock(&gref_lock);
+       mutex_lock(&gref_mutex);
        /* Clean up pages that were at zero (local) users but were still mapped
         * by remote domains. Since those pages count towards the limit that we
         * are about to enforce, removing them here is a good idea.
         */
        do_cleanup();
        if (gref_size + op.count > limit) {
-               spin_unlock(&gref_lock);
+               mutex_unlock(&gref_mutex);
                rc = -ENOSPC;
                goto out_free;
        }
        gref_size += op.count;
        op.index = priv->index;
        priv->index += op.count * PAGE_SIZE;
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
 
        rc = add_grefs(&op, gref_ids, priv);
        if (rc < 0)
@@ -343,7 +353,7 @@ static long gntalloc_ioctl_dealloc(struct gntalloc_file_private_data *priv,
                goto dealloc_grant_out;
        }
 
-       spin_lock(&gref_lock);
+       mutex_lock(&gref_mutex);
        gref = find_grefs(priv, op.index, op.count);
        if (gref) {
                /* Remove from the file list only, and decrease reference count.
@@ -363,7 +373,7 @@ static long gntalloc_ioctl_dealloc(struct gntalloc_file_private_data *priv,
 
        do_cleanup();
 
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
 dealloc_grant_out:
        return rc;
 }
@@ -383,7 +393,7 @@ static long gntalloc_ioctl_unmap_notify(struct gntalloc_file_private_data *priv,
        index = op.index & ~(PAGE_SIZE - 1);
        pgoff = op.index & (PAGE_SIZE - 1);
 
-       spin_lock(&gref_lock);
+       mutex_lock(&gref_mutex);
 
        gref = find_grefs(priv, index, 1);
        if (!gref) {
@@ -396,12 +406,30 @@ static long gntalloc_ioctl_unmap_notify(struct gntalloc_file_private_data *priv,
                goto unlock_out;
        }
 
+       /* We need to grab a reference to the event channel we are going to use
+        * to send the notify before releasing the reference we may already have
+        * (if someone has called this ioctl twice). This is required so that
+        * it is possible to change the clear_byte part of the notification
+        * without disturbing the event channel part, which may now be the last
+        * reference to that event channel.
+        */
+       if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
+               if (evtchn_get(op.event_channel_port)) {
+                       rc = -EINVAL;
+                       goto unlock_out;
+               }
+       }
+
+       if (gref->notify.flags & UNMAP_NOTIFY_SEND_EVENT)
+               evtchn_put(gref->notify.event);
+
        gref->notify.flags = op.action;
        gref->notify.pgoff = pgoff;
        gref->notify.event = op.event_channel_port;
        rc = 0;
+
  unlock_out:
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
        return rc;
 }
 
@@ -429,26 +457,40 @@ static long gntalloc_ioctl(struct file *filp, unsigned int cmd,
 
 static void gntalloc_vma_open(struct vm_area_struct *vma)
 {
-       struct gntalloc_gref *gref = vma->vm_private_data;
-       if (!gref)
+       struct gntalloc_vma_private_data *priv = vma->vm_private_data;
+
+       if (!priv)
                return;
 
-       spin_lock(&gref_lock);
-       gref->users++;
-       spin_unlock(&gref_lock);
+       mutex_lock(&gref_mutex);
+       priv->users++;
+       mutex_unlock(&gref_mutex);
 }
 
 static void gntalloc_vma_close(struct vm_area_struct *vma)
 {
-       struct gntalloc_gref *gref = vma->vm_private_data;
-       if (!gref)
+       struct gntalloc_vma_private_data *priv = vma->vm_private_data;
+       struct gntalloc_gref *gref, *next;
+       int i;
+
+       if (!priv)
                return;
 
-       spin_lock(&gref_lock);
-       gref->users--;
-       if (gref->users == 0)
-               __del_gref(gref);
-       spin_unlock(&gref_lock);
+       mutex_lock(&gref_mutex);
+       priv->users--;
+       if (priv->users == 0) {
+               gref = priv->gref;
+               for (i = 0; i < priv->count; i++) {
+                       gref->users--;
+                       next = list_entry(gref->next_gref.next,
+                                         struct gntalloc_gref, next_gref);
+                       if (gref->users == 0)
+                               __del_gref(gref);
+                       gref = next;
+               }
+               kfree(priv);
+       }
+       mutex_unlock(&gref_mutex);
 }
 
 static struct vm_operations_struct gntalloc_vmops = {
@@ -459,30 +501,41 @@ static struct vm_operations_struct gntalloc_vmops = {
 static int gntalloc_mmap(struct file *filp, struct vm_area_struct *vma)
 {
        struct gntalloc_file_private_data *priv = filp->private_data;
+       struct gntalloc_vma_private_data *vm_priv;
        struct gntalloc_gref *gref;
        int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
        int rv, i;
 
-       pr_debug("%s: priv %p, page %lu+%d\n", __func__,
-                      priv, vma->vm_pgoff, count);
-
        if (!(vma->vm_flags & VM_SHARED)) {
                printk(KERN_ERR "%s: Mapping must be shared.\n", __func__);
                return -EINVAL;
        }
 
-       spin_lock(&gref_lock);
+       vm_priv = kmalloc(sizeof(*vm_priv), GFP_KERNEL);
+       if (!vm_priv)
+               return -ENOMEM;
+
+       mutex_lock(&gref_mutex);
+
+       pr_debug("%s: priv %p,%p, page %lu+%d\n", __func__,
+                      priv, vm_priv, vma->vm_pgoff, count);
+
        gref = find_grefs(priv, vma->vm_pgoff << PAGE_SHIFT, count);
        if (gref == NULL) {
                rv = -ENOENT;
                pr_debug("%s: Could not find grant reference",
                                __func__);
+               kfree(vm_priv);
                goto out_unlock;
        }
 
-       vma->vm_private_data = gref;
+       vm_priv->gref = gref;
+       vm_priv->users = 1;
+       vm_priv->count = count;
+
+       vma->vm_private_data = vm_priv;
 
-       vma->vm_flags |= VM_RESERVED;
+       vma->vm_flags |= VM_RESERVED | VM_DONTEXPAND;
 
        vma->vm_ops = &gntalloc_vmops;
 
@@ -499,7 +552,7 @@ static int gntalloc_mmap(struct file *filp, struct vm_area_struct *vma)
        rv = 0;
 
 out_unlock:
-       spin_unlock(&gref_lock);
+       mutex_unlock(&gref_mutex);
        return rv;
 }
 
index afca14d..99d8151 100644 (file)
@@ -193,8 +193,10 @@ static void gntdev_put_map(struct grant_map *map)
 
        atomic_sub(map->count, &pages_mapped);
 
-       if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT)
+       if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
                notify_remote_via_evtchn(map->notify.event);
+               evtchn_put(map->notify.event);
+       }
 
        if (map->pages) {
                if (!use_ptemod)
@@ -312,7 +314,8 @@ static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
                }
        }
 
-       err = gnttab_unmap_refs(map->unmap_ops + offset, map->pages + offset, pages);
+       err = gnttab_unmap_refs(map->unmap_ops + offset, map->pages + offset,
+                               pages, true);
        if (err)
                return err;
 
@@ -599,6 +602,8 @@ static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
        struct ioctl_gntdev_unmap_notify op;
        struct grant_map *map;
        int rc;
+       int out_flags;
+       unsigned int out_event;
 
        if (copy_from_user(&op, u, sizeof(op)))
                return -EFAULT;
@@ -606,6 +611,21 @@ static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
        if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
                return -EINVAL;
 
+       /* We need to grab a reference to the event channel we are going to use
+        * to send the notify before releasing the reference we may already have
+        * (if someone has called this ioctl twice). This is required so that
+        * it is possible to change the clear_byte part of the notification
+        * without disturbing the event channel part, which may now be the last
+        * reference to that event channel.
+        */
+       if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
+               if (evtchn_get(op.event_channel_port))
+                       return -EINVAL;
+       }
+
+       out_flags = op.action;
+       out_event = op.event_channel_port;
+
        spin_lock(&priv->lock);
 
        list_for_each_entry(map, &priv->maps, next) {
@@ -624,12 +644,22 @@ static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
                goto unlock_out;
        }
 
+       out_flags = map->notify.flags;
+       out_event = map->notify.event;
+
        map->notify.flags = op.action;
        map->notify.addr = op.index - (map->index << PAGE_SHIFT);
        map->notify.event = op.event_channel_port;
+
        rc = 0;
+
  unlock_out:
        spin_unlock(&priv->lock);
+
+       /* Drop the reference to the event channel we did not save in the map */
+       if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
+               evtchn_put(out_event);
+
        return rc;
 }
 
index bf1c094..1cd94da 100644 (file)
 #include <xen/page.h>
 #include <xen/grant_table.h>
 #include <xen/interface/memory.h>
+#include <xen/hvc-console.h>
 #include <asm/xen/hypercall.h>
 
 #include <asm/pgtable.h>
 #include <asm/sync_bitops.h>
 
-
 /* External tools reserve first few grant table entries. */
 #define NR_RESERVED_ENTRIES 8
 #define GNTTAB_LIST_END 0xffffffff
-#define GREFS_PER_GRANT_FRAME (PAGE_SIZE / sizeof(struct grant_entry))
+#define GREFS_PER_GRANT_FRAME \
+(grant_table_version == 1 ?                      \
+(PAGE_SIZE / sizeof(struct grant_entry_v1)) :   \
+(PAGE_SIZE / sizeof(union grant_entry_v2)))
 
 static grant_ref_t **gnttab_list;
 static unsigned int nr_grant_frames;
@@ -64,13 +67,97 @@ static DEFINE_SPINLOCK(gnttab_list_lock);
 unsigned long xen_hvm_resume_frames;
 EXPORT_SYMBOL_GPL(xen_hvm_resume_frames);
 
-static struct grant_entry *shared;
+static union {
+       struct grant_entry_v1 *v1;
+       union grant_entry_v2 *v2;
+       void *addr;
+} gnttab_shared;
+
+/*This is a structure of function pointers for grant table*/
+struct gnttab_ops {
+       /*
+        * Mapping a list of frames for storing grant entries. Frames parameter
+        * is used to store grant table address when grant table being setup,
+        * nr_gframes is the number of frames to map grant table. Returning
+        * GNTST_okay means success and negative value means failure.
+        */
+       int (*map_frames)(unsigned long *frames, unsigned int nr_gframes);
+       /*
+        * Release a list of frames which are mapped in map_frames for grant
+        * entry status.
+        */
+       void (*unmap_frames)(void);
+       /*
+        * Introducing a valid entry into the grant table, granting the frame of
+        * this grant entry to domain for accessing or transfering. Ref
+        * parameter is reference of this introduced grant entry, domid is id of
+        * granted domain, frame is the page frame to be granted, and flags is
+        * status of the grant entry to be updated.
+        */
+       void (*update_entry)(grant_ref_t ref, domid_t domid,
+                            unsigned long frame, unsigned flags);
+       /*
+        * Stop granting a grant entry to domain for accessing. Ref parameter is
+        * reference of a grant entry whose grant access will be stopped,
+        * readonly is not in use in this function. If the grant entry is
+        * currently mapped for reading or writing, just return failure(==0)
+        * directly and don't tear down the grant access. Otherwise, stop grant
+        * access for this entry and return success(==1).
+        */
+       int (*end_foreign_access_ref)(grant_ref_t ref, int readonly);
+       /*
+        * Stop granting a grant entry to domain for transfer. Ref parameter is
+        * reference of a grant entry whose grant transfer will be stopped. If
+        * tranfer has not started, just reclaim the grant entry and return
+        * failure(==0). Otherwise, wait for the transfer to complete and then
+        * return the frame.
+        */
+       unsigned long (*end_foreign_transfer_ref)(grant_ref_t ref);
+       /*
+        * Query the status of a grant entry. Ref parameter is reference of
+        * queried grant entry, return value is the status of queried entry.
+        * Detailed status(writing/reading) can be gotten from the return value
+        * by bit operations.
+        */
+       int (*query_foreign_access)(grant_ref_t ref);
+       /*
+        * Grant a domain to access a range of bytes within the page referred by
+        * an available grant entry. Ref parameter is reference of a grant entry
+        * which will be sub-page accessed, domid is id of grantee domain, frame
+        * is frame address of subpage grant, flags is grant type and flag
+        * information, page_off is offset of the range of bytes, and length is
+        * length of bytes to be accessed.
+        */
+       void (*update_subpage_entry)(grant_ref_t ref, domid_t domid,
+                                    unsigned long frame, int flags,
+                                    unsigned page_off, unsigned length);
+       /*
+        * Redirect an available grant entry on domain A to another grant
+        * reference of domain B, then allow domain C to use grant reference
+        * of domain B transitively. Ref parameter is an available grant entry
+        * reference on domain A, domid is id of domain C which accesses grant
+        * entry transitively, flags is grant type and flag information,
+        * trans_domid is id of domain B whose grant entry is finally accessed
+        * transitively, trans_gref is grant entry transitive reference of
+        * domain B.
+        */
+       void (*update_trans_entry)(grant_ref_t ref, domid_t domid, int flags,
+                                  domid_t trans_domid, grant_ref_t trans_gref);
+};
+
+static struct gnttab_ops *gnttab_interface;
+
+/*This reflects status of grant entries, so act as a global value*/
+static grant_status_t *grstatus;
+
+static int grant_table_version;
 
 static struct gnttab_free_callback *gnttab_free_callback_list;
 
 static int gnttab_expand(unsigned int req_entries);
 
 #define RPP (PAGE_SIZE / sizeof(grant_ref_t))
+#define SPP (PAGE_SIZE / sizeof(grant_status_t))
 
 static inline grant_ref_t *__gnttab_entry(grant_ref_t entry)
 {
@@ -142,23 +229,33 @@ static void put_free_entry(grant_ref_t ref)
        spin_unlock_irqrestore(&gnttab_list_lock, flags);
 }
 
-static void update_grant_entry(grant_ref_t ref, domid_t domid,
-                              unsigned long frame, unsigned flags)
+/*
+ * Following applies to gnttab_update_entry_v1 and gnttab_update_entry_v2.
+ * Introducing a valid entry into the grant table:
+ *  1. Write ent->domid.
+ *  2. Write ent->frame:
+ *      GTF_permit_access:   Frame to which access is permitted.
+ *      GTF_accept_transfer: Pseudo-phys frame slot being filled by new
+ *                           frame, or zero if none.
+ *  3. Write memory barrier (WMB).
+ *  4. Write ent->flags, inc. valid type.
+ */
+static void gnttab_update_entry_v1(grant_ref_t ref, domid_t domid,
+                                  unsigned long frame, unsigned flags)
+{
+       gnttab_shared.v1[ref].domid = domid;
+       gnttab_shared.v1[ref].frame = frame;
+       wmb();
+       gnttab_shared.v1[ref].flags = flags;
+}
+
+static void gnttab_update_entry_v2(grant_ref_t ref, domid_t domid,
+                                  unsigned long frame, unsigned flags)
 {
-       /*
-        * Introducing a valid entry into the grant table:
-        *  1. Write ent->domid.
-        *  2. Write ent->frame:
-        *      GTF_permit_access:   Frame to which access is permitted.
-        *      GTF_accept_transfer: Pseudo-phys frame slot being filled by new
-        *                           frame, or zero if none.
-        *  3. Write memory barrier (WMB).
-        *  4. Write ent->flags, inc. valid type.
-        */
-       shared[ref].frame = frame;
-       shared[ref].domid = domid;
+       gnttab_shared.v2[ref].hdr.domid = domid;
+       gnttab_shared.v2[ref].full_page.frame = frame;
        wmb();
-       shared[ref].flags = flags;
+       gnttab_shared.v2[ref].hdr.flags = GTF_permit_access | flags;
 }
 
 /*
@@ -167,7 +264,7 @@ static void update_grant_entry(grant_ref_t ref, domid_t domid,
 void gnttab_grant_foreign_access_ref(grant_ref_t ref, domid_t domid,
                                     unsigned long frame, int readonly)
 {
-       update_grant_entry(ref, domid, frame,
+       gnttab_interface->update_entry(ref, domid, frame,
                           GTF_permit_access | (readonly ? GTF_readonly : 0));
 }
 EXPORT_SYMBOL_GPL(gnttab_grant_foreign_access_ref);
@@ -187,31 +284,184 @@ int gnttab_grant_foreign_access(domid_t domid, unsigned long frame,
 }
 EXPORT_SYMBOL_GPL(gnttab_grant_foreign_access);
 
-int gnttab_query_foreign_access(grant_ref_t ref)
+void gnttab_update_subpage_entry_v2(grant_ref_t ref, domid_t domid,
+                                   unsigned long frame, int flags,
+                                   unsigned page_off,
+                                   unsigned length)
+{
+       gnttab_shared.v2[ref].sub_page.frame = frame;
+       gnttab_shared.v2[ref].sub_page.page_off = page_off;
+       gnttab_shared.v2[ref].sub_page.length = length;
+       gnttab_shared.v2[ref].hdr.domid = domid;
+       wmb();
+       gnttab_shared.v2[ref].hdr.flags =
+                               GTF_permit_access | GTF_sub_page | flags;
+}
+
+int gnttab_grant_foreign_access_subpage_ref(grant_ref_t ref, domid_t domid,
+                                           unsigned long frame, int flags,
+                                           unsigned page_off,
+                                           unsigned length)
 {
-       u16 nflags;
+       if (flags & (GTF_accept_transfer | GTF_reading |
+                    GTF_writing | GTF_transitive))
+               return -EPERM;
 
-       nflags = shared[ref].flags;
+       if (gnttab_interface->update_subpage_entry == NULL)
+               return -ENOSYS;
 
-       return nflags & (GTF_reading|GTF_writing);
+       gnttab_interface->update_subpage_entry(ref, domid, frame, flags,
+                                              page_off, length);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(gnttab_grant_foreign_access_subpage_ref);
+
+int gnttab_grant_foreign_access_subpage(domid_t domid, unsigned long frame,
+                                       int flags, unsigned page_off,
+                                       unsigned length)
+{
+       int ref, rc;
+
+       ref = get_free_entries(1);
+       if (unlikely(ref < 0))
+               return -ENOSPC;
+
+       rc = gnttab_grant_foreign_access_subpage_ref(ref, domid, frame, flags,
+                                                    page_off, length);
+       if (rc < 0) {
+               put_free_entry(ref);
+               return rc;
+       }
+
+       return ref;
+}
+EXPORT_SYMBOL_GPL(gnttab_grant_foreign_access_subpage);
+
+bool gnttab_subpage_grants_available(void)
+{
+       return gnttab_interface->update_subpage_entry != NULL;
+}
+EXPORT_SYMBOL_GPL(gnttab_subpage_grants_available);
+
+void gnttab_update_trans_entry_v2(grant_ref_t ref, domid_t domid,
+                                 int flags, domid_t trans_domid,
+                                 grant_ref_t trans_gref)
+{
+       gnttab_shared.v2[ref].transitive.trans_domid = trans_domid;
+       gnttab_shared.v2[ref].transitive.gref = trans_gref;
+       gnttab_shared.v2[ref].hdr.domid = domid;
+       wmb();
+       gnttab_shared.v2[ref].hdr.flags =
+                               GTF_permit_access | GTF_transitive | flags;
+}
+
+int gnttab_grant_foreign_access_trans_ref(grant_ref_t ref, domid_t domid,
+                                         int flags, domid_t trans_domid,
+                                         grant_ref_t trans_gref)
+{
+       if (flags & (GTF_accept_transfer | GTF_reading |
+                    GTF_writing | GTF_sub_page))
+               return -EPERM;
+
+       if (gnttab_interface->update_trans_entry == NULL)
+               return -ENOSYS;
+
+       gnttab_interface->update_trans_entry(ref, domid, flags, trans_domid,
+                                            trans_gref);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(gnttab_grant_foreign_access_trans_ref);
+
+int gnttab_grant_foreign_access_trans(domid_t domid, int flags,
+                                     domid_t trans_domid,
+                                     grant_ref_t trans_gref)
+{
+       int ref, rc;
+
+       ref = get_free_entries(1);
+       if (unlikely(ref < 0))
+               return -ENOSPC;
+
+       rc = gnttab_grant_foreign_access_trans_ref(ref, domid, flags,
+                                                  trans_domid, trans_gref);
+       if (rc < 0) {
+               put_free_entry(ref);
+               return rc;
+       }
+
+       return ref;
+}
+EXPORT_SYMBOL_GPL(gnttab_grant_foreign_access_trans);
+
+bool gnttab_trans_grants_available(void)
+{
+       return gnttab_interface->update_trans_entry != NULL;
+}
+EXPORT_SYMBOL_GPL(gnttab_trans_grants_available);
+
+static int gnttab_query_foreign_access_v1(grant_ref_t ref)
+{
+       return gnttab_shared.v1[ref].flags & (GTF_reading|GTF_writing);
+}
+
+static int gnttab_query_foreign_access_v2(grant_ref_t ref)
+{
+       return grstatus[ref] & (GTF_reading|GTF_writing);
+}
+
+int gnttab_query_foreign_access(grant_ref_t ref)
+{
+       return gnttab_interface->query_foreign_access(ref);
 }
 EXPORT_SYMBOL_GPL(gnttab_query_foreign_access);
 
-int gnttab_end_foreign_access_ref(grant_ref_t ref, int readonly)
+static int gnttab_end_foreign_access_ref_v1(grant_ref_t ref, int readonly)
 {
        u16 flags, nflags;
+       u16 *pflags;
 
-       nflags = shared[ref].flags;
+       pflags = &gnttab_shared.v1[ref].flags;
+       nflags = *pflags;
        do {
                flags = nflags;
                if (flags & (GTF_reading|GTF_writing)) {
                        printk(KERN_ALERT "WARNING: g.e. still in use!\n");
                        return 0;
                }
-       } while ((nflags = sync_cmpxchg(&shared[ref].flags, flags, 0)) != flags);
+       } while ((nflags = sync_cmpxchg(pflags, flags, 0)) != flags);
+
+       return 1;
+}
+
+static int gnttab_end_foreign_access_ref_v2(grant_ref_t ref, int readonly)
+{
+       gnttab_shared.v2[ref].hdr.flags = 0;
+       mb();
+       if (grstatus[ref] & (GTF_reading|GTF_writing)) {
+               return 0;
+       } else {
+               /* The read of grstatus needs to have acquire
+               semantics.  On x86, reads already have
+               that, and we just need to protect against
+               compiler reorderings.  On other
+               architectures we may need a full
+               barrier. */
+#ifdef CONFIG_X86
+               barrier();
+#else
+               mb();
+#endif
+       }
 
        return 1;
 }
+
+int gnttab_end_foreign_access_ref(grant_ref_t ref, int readonly)
+{
+       return gnttab_interface->end_foreign_access_ref(ref, readonly);
+}
 EXPORT_SYMBOL_GPL(gnttab_end_foreign_access_ref);
 
 void gnttab_end_foreign_access(grant_ref_t ref, int readonly,
@@ -246,37 +496,76 @@ EXPORT_SYMBOL_GPL(gnttab_grant_foreign_transfer);
 void gnttab_grant_foreign_transfer_ref(grant_ref_t ref, domid_t domid,
                                       unsigned long pfn)
 {
-       update_grant_entry(ref, domid, pfn, GTF_accept_transfer);
+       gnttab_interface->update_entry(ref, domid, pfn, GTF_accept_transfer);
 }
 EXPORT_SYMBOL_GPL(gnttab_grant_foreign_transfer_ref);
 
-unsigned long gnttab_end_foreign_transfer_ref(grant_ref_t ref)
+static unsigned long gnttab_end_foreign_transfer_ref_v1(grant_ref_t ref)
 {
        unsigned long frame;
        u16           flags;
+       u16          *pflags;
+
+       pflags = &gnttab_shared.v1[ref].flags;
 
        /*
         * If a transfer is not even yet started, try to reclaim the grant
         * reference and return failure (== 0).
         */
-       while (!((flags = shared[ref].flags) & GTF_transfer_committed)) {
-               if (sync_cmpxchg(&shared[ref].flags, flags, 0) == flags)
+       while (!((flags = *pflags) & GTF_transfer_committed)) {
+               if (sync_cmpxchg(pflags, flags, 0) == flags)
                        return 0;
                cpu_relax();
        }
 
        /* If a transfer is in progress then wait until it is completed. */
        while (!(flags & GTF_transfer_completed)) {
-               flags = shared[ref].flags;
+               flags = *pflags;
                cpu_relax();
        }
 
        rmb();  /* Read the frame number /after/ reading completion status. */
-       frame = shared[ref].frame;
+       frame = gnttab_shared.v1[ref].frame;
+       BUG_ON(frame == 0);
+
+       return frame;
+}
+
+static unsigned long gnttab_end_foreign_transfer_ref_v2(grant_ref_t ref)
+{
+       unsigned long frame;
+       u16           flags;
+       u16          *pflags;
+
+       pflags = &gnttab_shared.v2[ref].hdr.flags;
+
+       /*
+        * If a transfer is not even yet started, try to reclaim the grant
+        * reference and return failure (== 0).
+        */
+       while (!((flags = *pflags) & GTF_transfer_committed)) {
+               if (sync_cmpxchg(pflags, flags, 0) == flags)
+                       return 0;
+               cpu_relax();
+       }
+
+       /* If a transfer is in progress then wait until it is completed. */
+       while (!(flags & GTF_transfer_completed)) {
+               flags = *pflags;
+               cpu_relax();
+       }
+
+       rmb();  /* Read the frame number /after/ reading completion status. */
+       frame = gnttab_shared.v2[ref].full_page.frame;
        BUG_ON(frame == 0);
 
        return frame;
 }
+
+unsigned long gnttab_end_foreign_transfer_ref(grant_ref_t ref)
+{
+       return gnttab_interface->end_foreign_transfer_ref(ref);
+}
 EXPORT_SYMBOL_GPL(gnttab_end_foreign_transfer_ref);
 
 unsigned long gnttab_end_foreign_transfer(grant_ref_t ref)
@@ -448,8 +737,8 @@ unsigned int gnttab_max_grant_frames(void)
 EXPORT_SYMBOL_GPL(gnttab_max_grant_frames);
 
 int gnttab_map_refs(struct gnttab_map_grant_ref *map_ops,
-                       struct gnttab_map_grant_ref *kmap_ops,
-                       struct page **pages, unsigned int count)
+                   struct gnttab_map_grant_ref *kmap_ops,
+                   struct page **pages, unsigned int count)
 {
        int i, ret;
        pte_t *pte;
@@ -472,24 +761,10 @@ int gnttab_map_refs(struct gnttab_map_grant_ref *map_ops,
                                (map_ops[i].host_addr & ~PAGE_MASK));
                        mfn = pte_mfn(*pte);
                } else {
-                       /* If you really wanted to do this:
-                        * mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
-                        *
-                        * The reason we do not implement it is b/c on the
-                        * unmap path (gnttab_unmap_refs) we have no means of
-                        * checking whether the page is !GNTMAP_contains_pte.
-                        *
-                        * That is without some extra data-structure to carry
-                        * the struct page, bool clear_pte, and list_head next
-                        * tuples and deal with allocation/delallocation, etc.
-                        *
-                        * The users of this API set the GNTMAP_contains_pte
-                        * flag so lets just return not supported until it
-                        * becomes neccessary to implement.
-                        */
-                       return -EOPNOTSUPP;
+                       mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
                }
-               ret = m2p_add_override(mfn, pages[i], &kmap_ops[i]);
+               ret = m2p_add_override(mfn, pages[i], kmap_ops ?
+                                      &kmap_ops[i] : NULL);
                if (ret)
                        return ret;
        }
@@ -499,7 +774,7 @@ int gnttab_map_refs(struct gnttab_map_grant_ref *map_ops,
 EXPORT_SYMBOL_GPL(gnttab_map_refs);
 
 int gnttab_unmap_refs(struct gnttab_unmap_grant_ref *unmap_ops,
-               struct page **pages, unsigned int count)
+                     struct page **pages, unsigned int count, bool clear_pte)
 {
        int i, ret;
 
@@ -511,7 +786,7 @@ int gnttab_unmap_refs(struct gnttab_unmap_grant_ref *unmap_ops,
                return ret;
 
        for (i = 0; i < count; i++) {
-               ret = m2p_remove_override(pages[i], true /* clear the PTE */);
+               ret = m2p_remove_override(pages[i], clear_pte);
                if (ret)
                        return ret;
        }
@@ -520,6 +795,77 @@ int gnttab_unmap_refs(struct gnttab_unmap_grant_ref *unmap_ops,
 }
 EXPORT_SYMBOL_GPL(gnttab_unmap_refs);
 
+static unsigned nr_status_frames(unsigned nr_grant_frames)
+{
+       return (nr_grant_frames * GREFS_PER_GRANT_FRAME + SPP - 1) / SPP;
+}
+
+static int gnttab_map_frames_v1(unsigned long *frames, unsigned int nr_gframes)
+{
+       int rc;
+
+       rc = arch_gnttab_map_shared(frames, nr_gframes,
+                                   gnttab_max_grant_frames(),
+                                   &gnttab_shared.addr);
+       BUG_ON(rc);
+
+       return 0;
+}
+
+static void gnttab_unmap_frames_v1(void)
+{
+       arch_gnttab_unmap(gnttab_shared.addr, nr_grant_frames);
+}
+
+static int gnttab_map_frames_v2(unsigned long *frames, unsigned int nr_gframes)
+{
+       uint64_t *sframes;
+       unsigned int nr_sframes;
+       struct gnttab_get_status_frames getframes;
+       int rc;
+
+       nr_sframes = nr_status_frames(nr_gframes);
+
+       /* No need for kzalloc as it is initialized in following hypercall
+        * GNTTABOP_get_status_frames.
+        */
+       sframes = kmalloc(nr_sframes  * sizeof(uint64_t), GFP_ATOMIC);
+       if (!sframes)
+               return -ENOMEM;
+
+       getframes.dom        = DOMID_SELF;
+       getframes.nr_frames  = nr_sframes;
+       set_xen_guest_handle(getframes.frame_list, sframes);
+
+       rc = HYPERVISOR_grant_table_op(GNTTABOP_get_status_frames,
+                                      &getframes, 1);
+       if (rc == -ENOSYS) {
+               kfree(sframes);
+               return -ENOSYS;
+       }
+
+       BUG_ON(rc || getframes.status);
+
+       rc = arch_gnttab_map_status(sframes, nr_sframes,
+                                   nr_status_frames(gnttab_max_grant_frames()),
+                                   &grstatus);
+       BUG_ON(rc);
+       kfree(sframes);
+
+       rc = arch_gnttab_map_shared(frames, nr_gframes,
+                                   gnttab_max_grant_frames(),
+                                   &gnttab_shared.addr);
+       BUG_ON(rc);
+
+       return 0;
+}
+
+static void gnttab_unmap_frames_v2(void)
+{
+       arch_gnttab_unmap(gnttab_shared.addr, nr_grant_frames);
+       arch_gnttab_unmap(grstatus, nr_status_frames(nr_grant_frames));
+}
+
 static int gnttab_map(unsigned int start_idx, unsigned int end_idx)
 {
        struct gnttab_setup_table setup;
@@ -551,6 +897,9 @@ static int gnttab_map(unsigned int start_idx, unsigned int end_idx)
                return rc;
        }
 
+       /* No need for kzalloc as it is initialized in following hypercall
+        * GNTTABOP_setup_table.
+        */
        frames = kmalloc(nr_gframes * sizeof(unsigned long), GFP_ATOMIC);
        if (!frames)
                return -ENOMEM;
@@ -567,19 +916,65 @@ static int gnttab_map(unsigned int start_idx, unsigned int end_idx)
 
        BUG_ON(rc || setup.status);
 
-       rc = arch_gnttab_map_shared(frames, nr_gframes, gnttab_max_grant_frames(),
-                                   &shared);
-       BUG_ON(rc);
+       rc = gnttab_interface->map_frames(frames, nr_gframes);
 
        kfree(frames);
 
-       return 0;
+       return rc;
+}
+
+static struct gnttab_ops gnttab_v1_ops = {
+       .map_frames                     = gnttab_map_frames_v1,
+       .unmap_frames                   = gnttab_unmap_frames_v1,
+       .update_entry                   = gnttab_update_entry_v1,
+       .end_foreign_access_ref         = gnttab_end_foreign_access_ref_v1,
+       .end_foreign_transfer_ref       = gnttab_end_foreign_transfer_ref_v1,
+       .query_foreign_access           = gnttab_query_foreign_access_v1,
+};
+
+static struct gnttab_ops gnttab_v2_ops = {
+       .map_frames                     = gnttab_map_frames_v2,
+       .unmap_frames                   = gnttab_unmap_frames_v2,
+       .update_entry                   = gnttab_update_entry_v2,
+       .end_foreign_access_ref         = gnttab_end_foreign_access_ref_v2,
+       .end_foreign_transfer_ref       = gnttab_end_foreign_transfer_ref_v2,
+       .query_foreign_access           = gnttab_query_foreign_access_v2,
+       .update_subpage_entry           = gnttab_update_subpage_entry_v2,
+       .update_trans_entry             = gnttab_update_trans_entry_v2,
+};
+
+static void gnttab_request_version(void)
+{
+       int rc;
+       struct gnttab_set_version gsv;
+
+       gsv.version = 2;
+       rc = HYPERVISOR_grant_table_op(GNTTABOP_set_version, &gsv, 1);
+       if (rc == 0) {
+               grant_table_version = 2;
+               gnttab_interface = &gnttab_v2_ops;
+       } else if (grant_table_version == 2) {
+               /*
+                * If we've already used version 2 features,
+                * but then suddenly discover that they're not
+                * available (e.g. migrating to an older
+                * version of Xen), almost unbounded badness
+                * can happen.
+                */
+               panic("we need grant tables version 2, but only version 1 is available");
+       } else {
+               grant_table_version = 1;
+               gnttab_interface = &gnttab_v1_ops;
+       }
+       printk(KERN_INFO "Grant tables using version %d layout.\n",
+               grant_table_version);
 }
 
 int gnttab_resume(void)
 {
        unsigned int max_nr_gframes;
 
+       gnttab_request_version();
        max_nr_gframes = gnttab_max_grant_frames();
        if (max_nr_gframes < nr_grant_frames)
                return -ENOSYS;
@@ -587,9 +982,10 @@ int gnttab_resume(void)
        if (xen_pv_domain())
                return gnttab_map(0, nr_grant_frames - 1);
 
-       if (!shared) {
-               shared = ioremap(xen_hvm_resume_frames, PAGE_SIZE * max_nr_gframes);
-               if (shared == NULL) {
+       if (gnttab_shared.addr == NULL) {
+               gnttab_shared.addr = ioremap(xen_hvm_resume_frames,
+                                               PAGE_SIZE * max_nr_gframes);
+               if (gnttab_shared.addr == NULL) {
                        printk(KERN_WARNING
                                        "Failed to ioremap gnttab share frames!");
                        return -ENOMEM;
@@ -603,7 +999,7 @@ int gnttab_resume(void)
 
 int gnttab_suspend(void)
 {
-       arch_gnttab_unmap_shared(shared, nr_grant_frames);
+       gnttab_interface->unmap_frames();
        return 0;
 }
 
similarity index 92%
rename from drivers/xen/xenfs/privcmd.c
rename to drivers/xen/privcmd.c
index dbd3b16..ccee0f1 100644 (file)
@@ -7,6 +7,7 @@
  */
 
 #include <linux/kernel.h>
+#include <linux/module.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/string.h>
@@ -18,6 +19,7 @@
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
 #include <linux/seq_file.h>
+#include <linux/miscdevice.h>
 
 #include <asm/pgalloc.h>
 #include <asm/pgtable.h>
 #include <xen/page.h>
 #include <xen/xen-ops.h>
 
+#include "privcmd.h"
+
+MODULE_LICENSE("GPL");
+
 #ifndef HAVE_ARCH_PRIVCMD_MMAP
 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma);
 #endif
@@ -359,7 +365,6 @@ static long privcmd_ioctl(struct file *file,
        return ret;
 }
 
-#ifndef HAVE_ARCH_PRIVCMD_MMAP
 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
        printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
@@ -392,9 +397,39 @@ static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
 {
        return (xchg(&vma->vm_private_data, (void *)1) == NULL);
 }
-#endif
 
-const struct file_operations privcmd_file_ops = {
+const struct file_operations xen_privcmd_fops = {
+       .owner = THIS_MODULE,
        .unlocked_ioctl = privcmd_ioctl,
        .mmap = privcmd_mmap,
 };
+EXPORT_SYMBOL_GPL(xen_privcmd_fops);
+
+static struct miscdevice privcmd_dev = {
+       .minor = MISC_DYNAMIC_MINOR,
+       .name = "xen/privcmd",
+       .fops = &xen_privcmd_fops,
+};
+
+static int __init privcmd_init(void)
+{
+       int err;
+
+       if (!xen_domain())
+               return -ENODEV;
+
+       err = misc_register(&privcmd_dev);
+       if (err != 0) {
+               printk(KERN_ERR "Could not register Xen privcmd device\n");
+               return err;
+       }
+       return 0;
+}
+
+static void __exit privcmd_exit(void)
+{
+       misc_deregister(&privcmd_dev);
+}
+
+module_init(privcmd_init);
+module_exit(privcmd_exit);
diff --git a/drivers/xen/privcmd.h b/drivers/xen/privcmd.h
new file mode 100644 (file)
index 0000000..14facae
--- /dev/null
@@ -0,0 +1,3 @@
+#include <linux/fs.h>
+
+extern const struct file_operations xen_privcmd_fops;
index 8f06e1e..7944a17 100644 (file)
@@ -99,6 +99,7 @@ static void pcistub_device_release(struct kref *kref)
        kfree(pci_get_drvdata(psdev->dev));
        pci_set_drvdata(psdev->dev, NULL);
 
+       psdev->dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
        pci_dev_put(psdev->dev);
 
        kfree(psdev);
@@ -234,6 +235,8 @@ void pcistub_put_pci_dev(struct pci_dev *dev)
        xen_pcibk_config_free_dyn_fields(found_psdev->dev);
        xen_pcibk_config_reset_dev(found_psdev->dev);
 
+       xen_unregister_device_domain_owner(found_psdev->dev);
+
        spin_lock_irqsave(&found_psdev->lock, flags);
        found_psdev->pdev = NULL;
        spin_unlock_irqrestore(&found_psdev->lock, flags);
@@ -331,6 +334,7 @@ static int __devinit pcistub_init_device(struct pci_dev *dev)
        dev_dbg(&dev->dev, "reset device\n");
        xen_pcibk_reset_device(dev);
 
+       dev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
        return 0;
 
 config_release:
index 0755259..8e1c44d 100644 (file)
@@ -241,11 +241,10 @@ static int xen_pcibk_export_device(struct xen_pcibk_device *pdev,
                goto out;
 
        dev_dbg(&dev->dev, "registering for %d\n", pdev->xdev->otherend_id);
-       dev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
        if (xen_register_device_domain_owner(dev,
                                             pdev->xdev->otherend_id) != 0) {
-               dev_err(&dev->dev, "device has been assigned to another " \
-                       "domain! Over-writting the ownership, but beware.\n");
+               dev_err(&dev->dev, "Stealing ownership from dom%d.\n",
+                       xen_find_device_domain_owner(dev));
                xen_unregister_device_domain_owner(dev);
                xen_register_device_domain_owner(dev, pdev->xdev->otherend_id);
        }
@@ -281,7 +280,6 @@ static int xen_pcibk_remove_device(struct xen_pcibk_device *pdev,
        }
 
        dev_dbg(&dev->dev, "unregistering for %d\n", pdev->xdev->otherend_id);
-       dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
        xen_unregister_device_domain_owner(dev);
 
        xen_pcibk_release_pci_dev(pdev, dev);
@@ -707,19 +705,16 @@ static int xen_pcibk_xenbus_remove(struct xenbus_device *dev)
        return 0;
 }
 
-static const struct xenbus_device_id xenpci_ids[] = {
+static const struct xenbus_device_id xen_pcibk_ids[] = {
        {"pci"},
        {""},
 };
 
-static struct xenbus_driver xenbus_xen_pcibk_driver = {
-       .name                   = DRV_NAME,
-       .owner                  = THIS_MODULE,
-       .ids                    = xenpci_ids,
+static DEFINE_XENBUS_DRIVER(xen_pcibk, DRV_NAME,
        .probe                  = xen_pcibk_xenbus_probe,
        .remove                 = xen_pcibk_xenbus_remove,
        .otherend_changed       = xen_pcibk_frontend_changed,
-};
+);
 
 const struct xen_pcibk_backend *__read_mostly xen_pcibk_backend;
 
@@ -735,11 +730,11 @@ int __init xen_pcibk_xenbus_register(void)
        if (passthrough)
                xen_pcibk_backend = &xen_pcibk_passthrough_backend;
        pr_info(DRV_NAME ": backend is %s\n", xen_pcibk_backend->name);
-       return xenbus_register_backend(&xenbus_xen_pcibk_driver);
+       return xenbus_register_backend(&xen_pcibk_driver);
 }
 
 void __exit xen_pcibk_xenbus_unregister(void)
 {
        destroy_workqueue(xen_pcibk_wq);
-       xenbus_unregister_driver(&xenbus_xen_pcibk_driver);
+       xenbus_unregister_driver(&xen_pcibk_driver);
 }
index 8dca685..31e2e90 100644 (file)
@@ -1,4 +1,5 @@
 obj-y  += xenbus.o
+obj-y  += xenbus_dev_frontend.o
 
 xenbus-objs =
 xenbus-objs += xenbus_client.o
@@ -9,4 +10,5 @@ xenbus-objs += xenbus_probe.o
 xenbus-be-objs-$(CONFIG_XEN_BACKEND) += xenbus_probe_backend.o
 xenbus-objs += $(xenbus-be-objs-y)
 
+obj-$(CONFIG_XEN_BACKEND) += xenbus_dev_backend.o
 obj-$(CONFIG_XEN_XENBUS_FRONTEND) += xenbus_probe_frontend.o
index 1906125..566d2ad 100644 (file)
 
 #include <linux/slab.h>
 #include <linux/types.h>
+#include <linux/spinlock.h>
 #include <linux/vmalloc.h>
 #include <linux/export.h>
 #include <asm/xen/hypervisor.h>
 #include <asm/xen/page.h>
 #include <xen/interface/xen.h>
 #include <xen/interface/event_channel.h>
+#include <xen/balloon.h>
 #include <xen/events.h>
 #include <xen/grant_table.h>
 #include <xen/xenbus.h>
+#include <xen/xen.h>
+
+#include "xenbus_probe.h"
+
+struct xenbus_map_node {
+       struct list_head next;
+       union {
+               struct vm_struct *area; /* PV */
+               struct page *page;     /* HVM */
+       };
+       grant_handle_t handle;
+};
+
+static DEFINE_SPINLOCK(xenbus_valloc_lock);
+static LIST_HEAD(xenbus_valloc_pages);
+
+struct xenbus_ring_ops {
+       int (*map)(struct xenbus_device *dev, int gnt, void **vaddr);
+       int (*unmap)(struct xenbus_device *dev, void *vaddr);
+};
+
+static const struct xenbus_ring_ops *ring_ops __read_mostly;
 
 const char *xenbus_strstate(enum xenbus_state state)
 {
@@ -436,19 +460,33 @@ EXPORT_SYMBOL_GPL(xenbus_free_evtchn);
  */
 int xenbus_map_ring_valloc(struct xenbus_device *dev, int gnt_ref, void **vaddr)
 {
+       return ring_ops->map(dev, gnt_ref, vaddr);
+}
+EXPORT_SYMBOL_GPL(xenbus_map_ring_valloc);
+
+static int xenbus_map_ring_valloc_pv(struct xenbus_device *dev,
+                                    int gnt_ref, void **vaddr)
+{
        struct gnttab_map_grant_ref op = {
                .flags = GNTMAP_host_map | GNTMAP_contains_pte,
                .ref   = gnt_ref,
                .dom   = dev->otherend_id,
        };
+       struct xenbus_map_node *node;
        struct vm_struct *area;
        pte_t *pte;
 
        *vaddr = NULL;
 
+       node = kzalloc(sizeof(*node), GFP_KERNEL);
+       if (!node)
+               return -ENOMEM;
+
        area = alloc_vm_area(PAGE_SIZE, &pte);
-       if (!area)
+       if (!area) {
+               kfree(node);
                return -ENOMEM;
+       }
 
        op.host_addr = arbitrary_virt_to_machine(pte).maddr;
 
@@ -457,19 +495,59 @@ int xenbus_map_ring_valloc(struct xenbus_device *dev, int gnt_ref, void **vaddr)
 
        if (op.status != GNTST_okay) {
                free_vm_area(area);
+               kfree(node);
                xenbus_dev_fatal(dev, op.status,
                                 "mapping in shared page %d from domain %d",
                                 gnt_ref, dev->otherend_id);
                return op.status;
        }
 
-       /* Stuff the handle in an unused field */
-       area->phys_addr = (unsigned long)op.handle;
+       node->handle = op.handle;
+       node->area = area;
+
+       spin_lock(&xenbus_valloc_lock);
+       list_add(&node->next, &xenbus_valloc_pages);
+       spin_unlock(&xenbus_valloc_lock);
 
        *vaddr = area->addr;
        return 0;
 }
-EXPORT_SYMBOL_GPL(xenbus_map_ring_valloc);
+
+static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev,
+                                     int gnt_ref, void **vaddr)
+{
+       struct xenbus_map_node *node;
+       int err;
+       void *addr;
+
+       *vaddr = NULL;
+
+       node = kzalloc(sizeof(*node), GFP_KERNEL);
+       if (!node)
+               return -ENOMEM;
+
+       err = alloc_xenballooned_pages(1, &node->page, false /* lowmem */);
+       if (err)
+               goto out_err;
+
+       addr = pfn_to_kaddr(page_to_pfn(node->page));
+
+       err = xenbus_map_ring(dev, gnt_ref, &node->handle, addr);
+       if (err)
+               goto out_err;
+
+       spin_lock(&xenbus_valloc_lock);
+       list_add(&node->next, &xenbus_valloc_pages);
+       spin_unlock(&xenbus_valloc_lock);
+
+       *vaddr = addr;
+       return 0;
+
+ out_err:
+       free_xenballooned_pages(1, &node->page);
+       kfree(node);
+       return err;
+}
 
 
 /**
@@ -489,12 +567,10 @@ EXPORT_SYMBOL_GPL(xenbus_map_ring_valloc);
 int xenbus_map_ring(struct xenbus_device *dev, int gnt_ref,
                    grant_handle_t *handle, void *vaddr)
 {
-       struct gnttab_map_grant_ref op = {
-               .host_addr = (unsigned long)vaddr,
-               .flags     = GNTMAP_host_map,
-               .ref       = gnt_ref,
-               .dom       = dev->otherend_id,
-       };
+       struct gnttab_map_grant_ref op;
+
+       gnttab_set_map_op(&op, (phys_addr_t)vaddr, GNTMAP_host_map, gnt_ref,
+                         dev->otherend_id);
 
        if (HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, &op, 1))
                BUG();
@@ -525,32 +601,36 @@ EXPORT_SYMBOL_GPL(xenbus_map_ring);
  */
 int xenbus_unmap_ring_vfree(struct xenbus_device *dev, void *vaddr)
 {
-       struct vm_struct *area;
+       return ring_ops->unmap(dev, vaddr);
+}
+EXPORT_SYMBOL_GPL(xenbus_unmap_ring_vfree);
+
+static int xenbus_unmap_ring_vfree_pv(struct xenbus_device *dev, void *vaddr)
+{
+       struct xenbus_map_node *node;
        struct gnttab_unmap_grant_ref op = {
                .host_addr = (unsigned long)vaddr,
        };
        unsigned int level;
 
-       /* It'd be nice if linux/vmalloc.h provided a find_vm_area(void *addr)
-        * method so that we don't have to muck with vmalloc internals here.
-        * We could force the user to hang on to their struct vm_struct from
-        * xenbus_map_ring_valloc, but these 6 lines considerably simplify
-        * this API.
-        */
-       read_lock(&vmlist_lock);
-       for (area = vmlist; area != NULL; area = area->next) {
-               if (area->addr == vaddr)
-                       break;
+       spin_lock(&xenbus_valloc_lock);
+       list_for_each_entry(node, &xenbus_valloc_pages, next) {
+               if (node->area->addr == vaddr) {
+                       list_del(&node->next);
+                       goto found;
+               }
        }
-       read_unlock(&vmlist_lock);
+       node = NULL;
+ found:
+       spin_unlock(&xenbus_valloc_lock);
 
-       if (!area) {
+       if (!node) {
                xenbus_dev_error(dev, -ENOENT,
                                 "can't find mapped virtual address %p", vaddr);
                return GNTST_bad_virt_addr;
        }
 
-       op.handle = (grant_handle_t)area->phys_addr;
+       op.handle = node->handle;
        op.host_addr = arbitrary_virt_to_machine(
                lookup_address((unsigned long)vaddr, &level)).maddr;
 
@@ -558,16 +638,50 @@ int xenbus_unmap_ring_vfree(struct xenbus_device *dev, void *vaddr)
                BUG();
 
        if (op.status == GNTST_okay)
-               free_vm_area(area);
+               free_vm_area(node->area);
        else
                xenbus_dev_error(dev, op.status,
                                 "unmapping page at handle %d error %d",
-                                (int16_t)area->phys_addr, op.status);
+                                node->handle, op.status);
 
+       kfree(node);
        return op.status;
 }
-EXPORT_SYMBOL_GPL(xenbus_unmap_ring_vfree);
 
+static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr)
+{
+       int rv;
+       struct xenbus_map_node *node;
+       void *addr;
+
+       spin_lock(&xenbus_valloc_lock);
+       list_for_each_entry(node, &xenbus_valloc_pages, next) {
+               addr = pfn_to_kaddr(page_to_pfn(node->page));
+               if (addr == vaddr) {
+                       list_del(&node->next);
+                       goto found;
+               }
+       }
+       node = NULL;
+ found:
+       spin_unlock(&xenbus_valloc_lock);
+
+       if (!node) {
+               xenbus_dev_error(dev, -ENOENT,
+                                "can't find mapped virtual address %p", vaddr);
+               return GNTST_bad_virt_addr;
+       }
+
+       rv = xenbus_unmap_ring(dev, node->handle, addr);
+
+       if (!rv)
+               free_xenballooned_pages(1, &node->page);
+       else
+               WARN(1, "Leaking %p\n", vaddr);
+
+       kfree(node);
+       return rv;
+}
 
 /**
  * xenbus_unmap_ring
@@ -582,10 +696,9 @@ EXPORT_SYMBOL_GPL(xenbus_unmap_ring_vfree);
 int xenbus_unmap_ring(struct xenbus_device *dev,
                      grant_handle_t handle, void *vaddr)
 {
-       struct gnttab_unmap_grant_ref op = {
-               .host_addr = (unsigned long)vaddr,
-               .handle    = handle,
-       };
+       struct gnttab_unmap_grant_ref op;
+
+       gnttab_set_unmap_op(&op, (phys_addr_t)vaddr, GNTMAP_host_map, handle);
 
        if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
                BUG();
@@ -617,3 +730,21 @@ enum xenbus_state xenbus_read_driver_state(const char *path)
        return result;
 }
 EXPORT_SYMBOL_GPL(xenbus_read_driver_state);
+
+static const struct xenbus_ring_ops ring_ops_pv = {
+       .map = xenbus_map_ring_valloc_pv,
+       .unmap = xenbus_unmap_ring_vfree_pv,
+};
+
+static const struct xenbus_ring_ops ring_ops_hvm = {
+       .map = xenbus_map_ring_valloc_hvm,
+       .unmap = xenbus_unmap_ring_vfree_hvm,
+};
+
+void __init xenbus_ring_ops_init(void)
+{
+       if (xen_pv_domain())
+               ring_ops = &ring_ops_pv;
+       else
+               ring_ops = &ring_ops_hvm;
+}
index c21db75..6e42800 100644 (file)
@@ -31,6 +31,8 @@
 #ifndef _XENBUS_COMMS_H
 #define _XENBUS_COMMS_H
 
+#include <linux/fs.h>
+
 int xs_init(void);
 int xb_init_comms(void);
 
@@ -43,4 +45,6 @@ int xs_input_avail(void);
 extern struct xenstore_domain_interface *xen_store_interface;
 extern int xen_store_evtchn;
 
+extern const struct file_operations xen_xenbus_fops;
+
 #endif /* _XENBUS_COMMS_H */
diff --git a/drivers/xen/xenbus/xenbus_dev_backend.c b/drivers/xen/xenbus/xenbus_dev_backend.c
new file mode 100644 (file)
index 0000000..3d3be78
--- /dev/null
@@ -0,0 +1,90 @@
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/fs.h>
+#include <linux/miscdevice.h>
+#include <linux/module.h>
+#include <linux/capability.h>
+
+#include <xen/xen.h>
+#include <xen/page.h>
+#include <xen/xenbus_dev.h>
+
+#include "xenbus_comms.h"
+
+MODULE_LICENSE("GPL");
+
+static int xenbus_backend_open(struct inode *inode, struct file *filp)
+{
+       if (!capable(CAP_SYS_ADMIN))
+               return -EPERM;
+
+       return nonseekable_open(inode, filp);
+}
+
+static long xenbus_backend_ioctl(struct file *file, unsigned int cmd, unsigned long data)
+{
+       if (!capable(CAP_SYS_ADMIN))
+               return -EPERM;
+
+       switch (cmd) {
+               case IOCTL_XENBUS_BACKEND_EVTCHN:
+                       if (xen_store_evtchn > 0)
+                               return xen_store_evtchn;
+                       return -ENODEV;
+
+               default:
+                       return -ENOTTY;
+       }
+}
+
+static int xenbus_backend_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       size_t size = vma->vm_end - vma->vm_start;
+
+       if (!capable(CAP_SYS_ADMIN))
+               return -EPERM;
+
+       if ((size > PAGE_SIZE) || (vma->vm_pgoff != 0))
+               return -EINVAL;
+
+       if (remap_pfn_range(vma, vma->vm_start,
+                           virt_to_pfn(xen_store_interface),
+                           size, vma->vm_page_prot))
+               return -EAGAIN;
+
+       return 0;
+}
+
+const struct file_operations xenbus_backend_fops = {
+       .open = xenbus_backend_open,
+       .mmap = xenbus_backend_mmap,
+       .unlocked_ioctl = xenbus_backend_ioctl,
+};
+
+static struct miscdevice xenbus_backend_dev = {
+       .minor = MISC_DYNAMIC_MINOR,
+       .name = "xen/xenbus_backend",
+       .fops = &xenbus_backend_fops,
+};
+
+static int __init xenbus_backend_init(void)
+{
+       int err;
+
+       if (!xen_initial_domain())
+               return -ENODEV;
+
+       err = misc_register(&xenbus_backend_dev);
+       if (err)
+               printk(KERN_ERR "Could not register xenbus backend device\n");
+       return err;
+}
+
+static void __exit xenbus_backend_exit(void)
+{
+       misc_deregister(&xenbus_backend_dev);
+}
+
+module_init(xenbus_backend_init);
+module_exit(xenbus_backend_exit);
similarity index 95%
rename from drivers/xen/xenfs/xenbus.c
rename to drivers/xen/xenbus/xenbus_dev_frontend.c
index bbd000f..527dc2a 100644 (file)
 #include <linux/namei.h>
 #include <linux/string.h>
 #include <linux/slab.h>
+#include <linux/miscdevice.h>
+#include <linux/module.h>
 
-#include "xenfs.h"
-#include "../xenbus/xenbus_comms.h"
+#include "xenbus_comms.h"
 
 #include <xen/xenbus.h>
+#include <xen/xen.h>
 #include <asm/xen/hypervisor.h>
 
+MODULE_LICENSE("GPL");
+
 /*
  * An element of a list of outstanding transactions, for which we're
  * still waiting a reply.
@@ -101,7 +105,7 @@ struct xenbus_file_priv {
        unsigned int len;
        union {
                struct xsd_sockmsg msg;
-               char buffer[PAGE_SIZE];
+               char buffer[XENSTORE_PAYLOAD_MAX];
        } u;
 
        /* Response queue. */
@@ -583,7 +587,7 @@ static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
        return 0;
 }
 
-const struct file_operations xenbus_file_ops = {
+const struct file_operations xen_xenbus_fops = {
        .read = xenbus_file_read,
        .write = xenbus_file_write,
        .open = xenbus_file_open,
@@ -591,3 +595,31 @@ const struct file_operations xenbus_file_ops = {
        .poll = xenbus_file_poll,
        .llseek = no_llseek,
 };
+EXPORT_SYMBOL_GPL(xen_xenbus_fops);
+
+static struct miscdevice xenbus_dev = {
+       .minor = MISC_DYNAMIC_MINOR,
+       .name = "xen/xenbus",
+       .fops = &xen_xenbus_fops,
+};
+
+static int __init xenbus_init(void)
+{
+       int err;
+
+       if (!xen_domain())
+               return -ENODEV;
+
+       err = misc_register(&xenbus_dev);
+       if (err)
+               printk(KERN_ERR "Could not register xenbus frontend device\n");
+       return err;
+}
+
+static void __exit xenbus_exit(void)
+{
+       misc_deregister(&xenbus_dev);
+}
+
+module_init(xenbus_init);
+module_exit(xenbus_exit);
index 1b178c6..3864967 100644 (file)
@@ -291,14 +291,9 @@ void xenbus_dev_shutdown(struct device *_dev)
 EXPORT_SYMBOL_GPL(xenbus_dev_shutdown);
 
 int xenbus_register_driver_common(struct xenbus_driver *drv,
-                                 struct xen_bus_type *bus,
-                                 struct module *owner,
-                                 const char *mod_name)
+                                 struct xen_bus_type *bus)
 {
-       drv->driver.name = drv->name;
        drv->driver.bus = &bus->bus;
-       drv->driver.owner = owner;
-       drv->driver.mod_name = mod_name;
 
        return driver_register(&drv->driver);
 }
@@ -730,6 +725,8 @@ static int __init xenbus_init(void)
        if (!xen_domain())
                return -ENODEV;
 
+       xenbus_ring_ops_init();
+
        if (xen_hvm_domain()) {
                uint64_t v = 0;
                err = hvm_get_parameter(HVM_PARAM_STORE_EVTCHN, &v);
index 9b1de4e..bb4f92e 100644 (file)
@@ -53,9 +53,7 @@ extern int xenbus_match(struct device *_dev, struct device_driver *_drv);
 extern int xenbus_dev_probe(struct device *_dev);
 extern int xenbus_dev_remove(struct device *_dev);
 extern int xenbus_register_driver_common(struct xenbus_driver *drv,
-                                        struct xen_bus_type *bus,
-                                        struct module *owner,
-                                        const char *mod_name);
+                                        struct xen_bus_type *bus);
 extern int xenbus_probe_node(struct xen_bus_type *bus,
                             const char *type,
                             const char *nodename);
@@ -76,4 +74,6 @@ extern void xenbus_otherend_changed(struct xenbus_watch *watch,
 extern int xenbus_read_otherend_details(struct xenbus_device *xendev,
                                        char *id_node, char *path_node);
 
+void xenbus_ring_ops_init(void);
+
 #endif
index c3c7cd1..257be37 100644 (file)
@@ -232,15 +232,13 @@ int xenbus_dev_is_online(struct xenbus_device *dev)
 }
 EXPORT_SYMBOL_GPL(xenbus_dev_is_online);
 
-int __xenbus_register_backend(struct xenbus_driver *drv,
-                             struct module *owner, const char *mod_name)
+int xenbus_register_backend(struct xenbus_driver *drv)
 {
        drv->read_otherend_details = read_frontend_details;
 
-       return xenbus_register_driver_common(drv, &xenbus_backend,
-                                            owner, mod_name);
+       return xenbus_register_driver_common(drv, &xenbus_backend);
 }
-EXPORT_SYMBOL_GPL(__xenbus_register_backend);
+EXPORT_SYMBOL_GPL(xenbus_register_backend);
 
 static int backend_probe_and_watch(struct notifier_block *notifier,
                                   unsigned long event,
index 2f73195..9c57819 100644 (file)
@@ -230,15 +230,13 @@ static void wait_for_devices(struct xenbus_driver *xendrv)
                         print_device_status);
 }
 
-int __xenbus_register_frontend(struct xenbus_driver *drv,
-                              struct module *owner, const char *mod_name)
+int xenbus_register_frontend(struct xenbus_driver *drv)
 {
        int ret;
 
        drv->read_otherend_details = read_backend_details;
 
-       ret = xenbus_register_driver_common(drv, &xenbus_frontend,
-                                           owner, mod_name);
+       ret = xenbus_register_driver_common(drv, &xenbus_frontend);
        if (ret)
                return ret;
 
@@ -247,7 +245,7 @@ int __xenbus_register_frontend(struct xenbus_driver *drv,
 
        return 0;
 }
-EXPORT_SYMBOL_GPL(__xenbus_register_frontend);
+EXPORT_SYMBOL_GPL(xenbus_register_frontend);
 
 static DECLARE_WAIT_QUEUE_HEAD(backend_state_wq);
 static int backend_state;
index ede860f..d1c217b 100644 (file)
@@ -532,21 +532,18 @@ int xenbus_printf(struct xenbus_transaction t,
 {
        va_list ap;
        int ret;
-#define PRINTF_BUFFER_SIZE 4096
-       char *printf_buffer;
-
-       printf_buffer = kmalloc(PRINTF_BUFFER_SIZE, GFP_NOIO | __GFP_HIGH);
-       if (printf_buffer == NULL)
-               return -ENOMEM;
+       char *buf;
 
        va_start(ap, fmt);
-       ret = vsnprintf(printf_buffer, PRINTF_BUFFER_SIZE, fmt, ap);
+       buf = kvasprintf(GFP_NOIO | __GFP_HIGH, fmt, ap);
        va_end(ap);
 
-       BUG_ON(ret > PRINTF_BUFFER_SIZE-1);
-       ret = xenbus_write(t, dir, node, printf_buffer);
+       if (!buf)
+               return -ENOMEM;
+
+       ret = xenbus_write(t, dir, node, buf);
 
-       kfree(printf_buffer);
+       kfree(buf);
 
        return ret;
 }
@@ -801,6 +798,12 @@ static int process_msg(void)
                goto out;
        }
 
+       if (msg->hdr.len > XENSTORE_PAYLOAD_MAX) {
+               kfree(msg);
+               err = -EINVAL;
+               goto out;
+       }
+
        body = kmalloc(msg->hdr.len + 1, GFP_NOIO | __GFP_HIGH);
        if (body == NULL) {
                kfree(msg);
index 4fde944..b019865 100644 (file)
@@ -1,4 +1,4 @@
 obj-$(CONFIG_XENFS) += xenfs.o
 
-xenfs-y                          = super.o xenbus.o privcmd.o
+xenfs-y                          = super.o
 xenfs-$(CONFIG_XEN_DOM0) += xenstored.o
index 1aa3897..a84b53c 100644 (file)
@@ -16,6 +16,8 @@
 #include <xen/xen.h>
 
 #include "xenfs.h"
+#include "../privcmd.h"
+#include "../xenbus/xenbus_comms.h"
 
 #include <asm/xen/hypervisor.h>
 
@@ -82,9 +84,9 @@ static int xenfs_fill_super(struct super_block *sb, void *data, int silent)
 {
        static struct tree_descr xenfs_files[] = {
                [1] = {},
-               { "xenbus", &xenbus_file_ops, S_IRUSR|S_IWUSR },
+               { "xenbus", &xen_xenbus_fops, S_IRUSR|S_IWUSR },
                { "capabilities", &capabilities_file_ops, S_IRUGO },
-               { "privcmd", &privcmd_file_ops, S_IRUSR|S_IWUSR },
+               { "privcmd", &xen_privcmd_fops, S_IRUSR|S_IWUSR },
                {""},
        };
        int rc;
index b68aa62..6b80c77 100644 (file)
@@ -1,8 +1,6 @@
 #ifndef _XENFS_XENBUS_H
 #define _XENFS_XENBUS_H
 
-extern const struct file_operations xenbus_file_ops;
-extern const struct file_operations privcmd_file_ops;
 extern const struct file_operations xsd_kva_file_ops;
 extern const struct file_operations xsd_port_file_ops;
 
index d287997..0f77370 100644 (file)
@@ -37,6 +37,13 @@ int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
  */
 void unbind_from_irqhandler(unsigned int irq, void *dev_id);
 
+/*
+ * Allow extra references to event channels exposed to userspace by evtchn
+ */
+int evtchn_make_refcounted(unsigned int evtchn);
+int evtchn_get(unsigned int evtchn);
+void evtchn_put(unsigned int evtchn);
+
 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector);
 int resend_irq_on_evtchn(unsigned int irq);
 void rebind_evtchn_irq(int evtchn, int irq);
index 11e2dfc..15f8a00 100644 (file)
@@ -62,6 +62,24 @@ int gnttab_resume(void);
 
 int gnttab_grant_foreign_access(domid_t domid, unsigned long frame,
                                int readonly);
+int gnttab_grant_foreign_access_subpage(domid_t domid, unsigned long frame,
+                                       int flags, unsigned page_off,
+                                       unsigned length);
+int gnttab_grant_foreign_access_trans(domid_t domid, int flags,
+                                     domid_t trans_domid,
+                                     grant_ref_t trans_gref);
+
+/*
+ * Are sub-page grants available on this version of Xen?  Returns true if they
+ * are, and false if they're not.
+ */
+bool gnttab_subpage_grants_available(void);
+
+/*
+ * Are transitive grants available on this version of Xen?  Returns true if they
+ * are, and false if they're not.
+ */
+bool gnttab_trans_grants_available(void);
 
 /*
  * End access through the given grant reference, iff the grant entry is no
@@ -108,6 +126,13 @@ void gnttab_cancel_free_callback(struct gnttab_free_callback *callback);
 
 void gnttab_grant_foreign_access_ref(grant_ref_t ref, domid_t domid,
                                     unsigned long frame, int readonly);
+int gnttab_grant_foreign_access_subpage_ref(grant_ref_t ref, domid_t domid,
+                                           unsigned long frame, int flags,
+                                           unsigned page_off,
+                                           unsigned length);
+int gnttab_grant_foreign_access_trans_ref(grant_ref_t ref, domid_t domid,
+                                         int flags, domid_t trans_domid,
+                                         grant_ref_t trans_gref);
 
 void gnttab_grant_foreign_transfer_ref(grant_ref_t, domid_t domid,
                                       unsigned long pfn);
@@ -145,9 +170,11 @@ gnttab_set_unmap_op(struct gnttab_unmap_grant_ref *unmap, phys_addr_t addr,
 
 int arch_gnttab_map_shared(unsigned long *frames, unsigned long nr_gframes,
                           unsigned long max_nr_gframes,
-                          struct grant_entry **__shared);
-void arch_gnttab_unmap_shared(struct grant_entry *shared,
-                             unsigned long nr_gframes);
+                          void **__shared);
+int arch_gnttab_map_status(uint64_t *frames, unsigned long nr_gframes,
+                          unsigned long max_nr_gframes,
+                          grant_status_t **__shared);
+void arch_gnttab_unmap(void *shared, unsigned long nr_gframes);
 
 extern unsigned long xen_hvm_resume_frames;
 unsigned int gnttab_max_grant_frames(void);
@@ -155,9 +182,9 @@ unsigned int gnttab_max_grant_frames(void);
 #define gnttab_map_vaddr(map) ((void *)(map.host_virt_addr))
 
 int gnttab_map_refs(struct gnttab_map_grant_ref *map_ops,
-                       struct gnttab_map_grant_ref *kmap_ops,
+                   struct gnttab_map_grant_ref *kmap_ops,
                    struct page **pages, unsigned int count);
 int gnttab_unmap_refs(struct gnttab_unmap_grant_ref *unmap_ops,
-                     struct page **pages, unsigned int count);
+                     struct page **pages, unsigned int count, bool clear_pte);
 
 #endif /* __ASM_GNTTAB_H__ */
index 39e5717..a17d844 100644 (file)
  */
 
 /*
+ * Reference to a grant entry in a specified domain's grant table.
+ */
+typedef uint32_t grant_ref_t;
+
+/*
  * A grant table comprises a packed array of grant entries in one or more
  * page frames shared between Xen and a guest.
  * [XEN]: This field is written by Xen and read by the sharing guest.
  * [GST]: This field is written by the guest and read by Xen.
  */
-struct grant_entry {
+
+/*
+ * Version 1 of the grant table entry structure is maintained purely
+ * for backwards compatibility.  New guests should use version 2.
+ */
+struct grant_entry_v1 {
     /* GTF_xxx: various type and flag information.  [XEN,GST] */
     uint16_t flags;
     /* The domain being granted foreign privileges. [GST] */
@@ -108,10 +118,13 @@ struct grant_entry {
  *  GTF_permit_access: Allow @domid to map/access @frame.
  *  GTF_accept_transfer: Allow @domid to transfer ownership of one page frame
  *                       to this guest. Xen writes the page number to @frame.
+ *  GTF_transitive: Allow @domid to transitively access a subrange of
+ *                  @trans_grant in @trans_domid.  No mappings are allowed.
  */
 #define GTF_invalid         (0U<<0)
 #define GTF_permit_access   (1U<<0)
 #define GTF_accept_transfer (2U<<0)
+#define GTF_transitive      (3U<<0)
 #define GTF_type_mask       (3U<<0)
 
 /*
@@ -119,6 +132,9 @@ struct grant_entry {
  *  GTF_readonly: Restrict @domid to read-only mappings and accesses. [GST]
  *  GTF_reading: Grant entry is currently mapped for reading by @domid. [XEN]
  *  GTF_writing: Grant entry is currently mapped for writing by @domid. [XEN]
+ *  GTF_sub_page: Grant access to only a subrange of the page.  @domid
+ *                will only be allowed to copy from the grant, and not
+ *                map it. [GST]
  */
 #define _GTF_readonly       (2)
 #define GTF_readonly        (1U<<_GTF_readonly)
@@ -126,6 +142,8 @@ struct grant_entry {
 #define GTF_reading         (1U<<_GTF_reading)
 #define _GTF_writing        (4)
 #define GTF_writing         (1U<<_GTF_writing)
+#define _GTF_sub_page       (8)
+#define GTF_sub_page        (1U<<_GTF_sub_page)
 
 /*
  * Subflags for GTF_accept_transfer:
@@ -142,15 +160,81 @@ struct grant_entry {
 #define _GTF_transfer_completed (3)
 #define GTF_transfer_completed  (1U<<_GTF_transfer_completed)
 
+/*
+ * Version 2 grant table entries.  These fulfil the same role as
+ * version 1 entries, but can represent more complicated operations.
+ * Any given domain will have either a version 1 or a version 2 table,
+ * and every entry in the table will be the same version.
+ *
+ * The interface by which domains use grant references does not depend
+ * on the grant table version in use by the other domain.
+ */
 
-/***********************************
- * GRANT TABLE QUERIES AND USES
+/*
+ * Version 1 and version 2 grant entries share a common prefix.  The
+ * fields of the prefix are documented as part of struct
+ * grant_entry_v1.
  */
+struct grant_entry_header {
+    uint16_t flags;
+    domid_t  domid;
+};
 
 /*
- * Reference to a grant entry in a specified domain's grant table.
+ * Version 2 of the grant entry structure, here is an union because three
+ * different types are suppotted: full_page, sub_page and transitive.
+ */
+union grant_entry_v2 {
+    struct grant_entry_header hdr;
+
+    /*
+     * This member is used for V1-style full page grants, where either:
+     *
+     * -- hdr.type is GTF_accept_transfer, or
+     * -- hdr.type is GTF_permit_access and GTF_sub_page is not set.
+     *
+     * In that case, the frame field has the same semantics as the
+     * field of the same name in the V1 entry structure.
+     */
+    struct {
+       struct grant_entry_header hdr;
+       uint32_t pad0;
+       uint64_t frame;
+    } full_page;
+
+    /*
+     * If the grant type is GTF_grant_access and GTF_sub_page is set,
+     * @domid is allowed to access bytes [@page_off,@page_off+@length)
+     * in frame @frame.
+     */
+    struct {
+       struct grant_entry_header hdr;
+       uint16_t page_off;
+       uint16_t length;
+       uint64_t frame;
+    } sub_page;
+
+    /*
+     * If the grant is GTF_transitive, @domid is allowed to use the
+     * grant @gref in domain @trans_domid, as if it was the local
+     * domain.  Obviously, the transitive access must be compatible
+     * with the original grant.
+     */
+    struct {
+       struct grant_entry_header hdr;
+       domid_t trans_domid;
+       uint16_t pad0;
+       grant_ref_t gref;
+    } transitive;
+
+    uint32_t __spacer[4]; /* Pad to a power of two */
+};
+
+typedef uint16_t grant_status_t;
+
+/***********************************
+ * GRANT TABLE QUERIES AND USES
  */
-typedef uint32_t grant_ref_t;
 
 /*
  * Handle to track a mapping created via a grant reference.
@@ -322,6 +406,79 @@ struct gnttab_query_size {
 DEFINE_GUEST_HANDLE_STRUCT(gnttab_query_size);
 
 /*
+ * GNTTABOP_unmap_and_replace: Destroy one or more grant-reference mappings
+ * tracked by <handle> but atomically replace the page table entry with one
+ * pointing to the machine address under <new_addr>.  <new_addr> will be
+ * redirected to the null entry.
+ * NOTES:
+ *  1. The call may fail in an undefined manner if either mapping is not
+ *     tracked by <handle>.
+ *  2. After executing a batch of unmaps, it is guaranteed that no stale
+ *     mappings will remain in the device or host TLBs.
+ */
+#define GNTTABOP_unmap_and_replace    7
+struct gnttab_unmap_and_replace {
+    /* IN parameters. */
+    uint64_t host_addr;
+    uint64_t new_addr;
+    grant_handle_t handle;
+    /* OUT parameters. */
+    int16_t  status;              /* GNTST_* */
+};
+DEFINE_GUEST_HANDLE_STRUCT(gnttab_unmap_and_replace);
+
+/*
+ * GNTTABOP_set_version: Request a particular version of the grant
+ * table shared table structure.  This operation can only be performed
+ * once in any given domain.  It must be performed before any grants
+ * are activated; otherwise, the domain will be stuck with version 1.
+ * The only defined versions are 1 and 2.
+ */
+#define GNTTABOP_set_version          8
+struct gnttab_set_version {
+    /* IN parameters */
+    uint32_t version;
+};
+DEFINE_GUEST_HANDLE_STRUCT(gnttab_set_version);
+
+/*
+ * GNTTABOP_get_status_frames: Get the list of frames used to store grant
+ * status for <dom>. In grant format version 2, the status is separated
+ * from the other shared grant fields to allow more efficient synchronization
+ * using barriers instead of atomic cmpexch operations.
+ * <nr_frames> specify the size of vector <frame_list>.
+ * The frame addresses are returned in the <frame_list>.
+ * Only <nr_frames> addresses are returned, even if the table is larger.
+ * NOTES:
+ *  1. <dom> may be specified as DOMID_SELF.
+ *  2. Only a sufficiently-privileged domain may specify <dom> != DOMID_SELF.
+ */
+#define GNTTABOP_get_status_frames     9
+struct gnttab_get_status_frames {
+    /* IN parameters. */
+    uint32_t nr_frames;
+    domid_t  dom;
+    /* OUT parameters. */
+    int16_t  status;              /* GNTST_* */
+    GUEST_HANDLE(uint64_t) frame_list;
+};
+DEFINE_GUEST_HANDLE_STRUCT(gnttab_get_status_frames);
+
+/*
+ * GNTTABOP_get_version: Get the grant table version which is in
+ * effect for domain <dom>.
+ */
+#define GNTTABOP_get_version          10
+struct gnttab_get_version {
+    /* IN parameters */
+    domid_t dom;
+    uint16_t pad;
+    /* OUT parameters */
+    uint32_t version;
+};
+DEFINE_GUEST_HANDLE_STRUCT(gnttab_get_version);
+
+/*
  * Bitfield values for update_pin_status.flags.
  */
  /* Map the grant entry for access by I/O devices. */
index f6f07aa..7cdfca2 100644 (file)
@@ -87,4 +87,7 @@ struct xenstore_domain_interface {
     XENSTORE_RING_IDX rsp_cons, rsp_prod;
 };
 
+/* Violating this is very bad.  See docs/misc/xenstore.txt. */
+#define XENSTORE_PAYLOAD_MAX 4096
+
 #endif /* _XS_WIRE_H */
index 6a6e914..a890804 100644 (file)
@@ -523,6 +523,8 @@ struct tmem_op {
        } u;
 };
 
+DEFINE_GUEST_HANDLE(u64);
+
 #else /* __ASSEMBLY__ */
 
 /* In assembly code we cannot use C numeric constant suffixes. */
index b1b6676..e8c599b 100644 (file)
@@ -85,8 +85,6 @@ struct xenbus_device_id
 
 /* A xenbus driver. */
 struct xenbus_driver {
-       char *name;
-       struct module *owner;
        const struct xenbus_device_id *ids;
        int (*probe)(struct xenbus_device *dev,
                     const struct xenbus_device_id *id);
@@ -101,31 +99,20 @@ struct xenbus_driver {
        int (*is_ready)(struct xenbus_device *dev);
 };
 
-static inline struct xenbus_driver *to_xenbus_driver(struct device_driver *drv)
-{
-       return container_of(drv, struct xenbus_driver, driver);
+#define DEFINE_XENBUS_DRIVER(var, drvname, methods...)         \
+struct xenbus_driver var ## _driver = {                                \
+       .driver.name = drvname + 0 ?: var ## _ids->devicetype,  \
+       .driver.owner = THIS_MODULE,                            \
+       .ids = var ## _ids, ## methods                          \
 }
 
-int __must_check __xenbus_register_frontend(struct xenbus_driver *drv,
-                                           struct module *owner,
-                                           const char *mod_name);
-
-static inline int __must_check
-xenbus_register_frontend(struct xenbus_driver *drv)
+static inline struct xenbus_driver *to_xenbus_driver(struct device_driver *drv)
 {
-       WARN_ON(drv->owner != THIS_MODULE);
-       return __xenbus_register_frontend(drv, THIS_MODULE, KBUILD_MODNAME);
+       return container_of(drv, struct xenbus_driver, driver);
 }
 
-int __must_check __xenbus_register_backend(struct xenbus_driver *drv,
-                                          struct module *owner,
-                                          const char *mod_name);
-static inline int __must_check
-xenbus_register_backend(struct xenbus_driver *drv)
-{
-       WARN_ON(drv->owner != THIS_MODULE);
-       return __xenbus_register_backend(drv, THIS_MODULE, KBUILD_MODNAME);
-}
+int __must_check xenbus_register_frontend(struct xenbus_driver *);
+int __must_check xenbus_register_backend(struct xenbus_driver *);
 
 void xenbus_unregister_driver(struct xenbus_driver *drv);
 
diff --git a/include/xen/xenbus_dev.h b/include/xen/xenbus_dev.h
new file mode 100644 (file)
index 0000000..ac5f0fe
--- /dev/null
@@ -0,0 +1,41 @@
+/******************************************************************************
+ * evtchn.h
+ *
+ * Interface to /dev/xen/xenbus_backend.
+ *
+ * Copyright (c) 2011 Bastian Blank <waldi@debian.org>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version 2
+ * as published by the Free Software Foundation; or, when distributed
+ * separately from the Linux kernel or incorporated into other
+ * software packages, subject to the following license:
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this source file (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use, copy, modify,
+ * merge, publish, distribute, sublicense, and/or sell copies of the Software,
+ * and to permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef __LINUX_XEN_XENBUS_DEV_H__
+#define __LINUX_XEN_XENBUS_DEV_H__
+
+#include <linux/ioctl.h>
+
+#define IOCTL_XENBUS_BACKEND_EVTCHN                    \
+       _IOC(_IOC_NONE, 'B', 0, 0)
+
+#endif /* __LINUX_XEN_XENBUS_DEV_H__ */