2 files changed, 133 insertions, 40 deletions

diff --git a/drivers/xen/xenbus/xenbus_client.c b/drivers/xen/xenbus/xenbus_client.c
index df6890681231..d6fdd2d209d3 100644
--- a/drivers/xen/xenbus/xenbus_client.c
+++ b/drivers/xen/xenbus/xenbus_client.c
@@ -363,50 +363,92 @@ static void xenbus_switch_fatal(struct xenbus_device *dev, int depth, int err,
 		__xenbus_switch_state(dev, XenbusStateClosing, 1);
 }
 
-/**
- * xenbus_grant_ring
+/*
+ * xenbus_setup_ring
  * @dev: xenbus device
- * @vaddr: starting virtual address of the ring
+ * @vaddr: pointer to starting virtual address of the ring
  * @nr_pages: number of pages to be granted
  * @grefs: grant reference array to be filled in
  *
- * Grant access to the given @vaddr to the peer of the given device.
- * Then fill in @grefs with grant references.  Return 0 on success, or
- * -errno on error.  On error, the device will switch to
- * XenbusStateClosing, and the error will be saved in the store.
+ * Allocate physically contiguous pages for a shared ring buffer and grant it
+ * to the peer of the given device. The ring buffer is initially filled with
+ * zeroes. The virtual address of the ring is stored at @vaddr and the
+ * grant references are stored in the @grefs array. In case of error @vaddr
+ * will be set to NULL and @grefs will be filled with INVALID_GRANT_REF.
  */
-int xenbus_grant_ring(struct xenbus_device *dev, void *vaddr,
+int xenbus_setup_ring(struct xenbus_device *dev, gfp_t gfp, void **vaddr,
 		      unsigned int nr_pages, grant_ref_t *grefs)
 {
-	int err;
-	unsigned int i;
+	unsigned long ring_size = nr_pages * XEN_PAGE_SIZE;
 	grant_ref_t gref_head;
+	unsigned int i;
+	int ret;
 
-	err = gnttab_alloc_grant_references(nr_pages, &gref_head);
-	if (err) {
-		xenbus_dev_fatal(dev, err, "granting access to ring page");
-		return err;
+	*vaddr = alloc_pages_exact(ring_size, gfp | __GFP_ZERO);
+	if (!*vaddr) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	ret = gnttab_alloc_grant_references(nr_pages, &gref_head);
+	if (ret) {
+		xenbus_dev_fatal(dev, ret, "granting access to %u ring pages",
+				 nr_pages);
+		goto err;
 	}
 
 	for (i = 0; i < nr_pages; i++) {
 		unsigned long gfn;
 
-		if (is_vmalloc_addr(vaddr))
-			gfn = pfn_to_gfn(vmalloc_to_pfn(vaddr));
+		if (is_vmalloc_addr(*vaddr))
+			gfn = pfn_to_gfn(vmalloc_to_pfn(vaddr[i]));
 		else
-			gfn = virt_to_gfn(vaddr);
+			gfn = virt_to_gfn(vaddr[i]);
 
 		grefs[i] = gnttab_claim_grant_reference(&gref_head);
 		gnttab_grant_foreign_access_ref(grefs[i], dev->otherend_id,
 						gfn, 0);
-
-		vaddr = vaddr + XEN_PAGE_SIZE;
 	}
 
 	return 0;
+
+ err:
+	if (*vaddr)
+		free_pages_exact(*vaddr, ring_size);
+	for (i = 0; i < nr_pages; i++)
+		grefs[i] = INVALID_GRANT_REF;
+	*vaddr = NULL;
+
+	return ret;
 }
-EXPORT_SYMBOL_GPL(xenbus_grant_ring);
+EXPORT_SYMBOL_GPL(xenbus_setup_ring);
 
+/*
+ * xenbus_teardown_ring
+ * @vaddr: starting virtual address of the ring
+ * @nr_pages: number of pages
+ * @grefs: grant reference array
+ *
+ * Remove grants for the shared ring buffer and free the associated memory.
+ * On return the grant reference array is filled with INVALID_GRANT_REF.
+ */
+void xenbus_teardown_ring(void **vaddr, unsigned int nr_pages,
+			  grant_ref_t *grefs)
+{
+	unsigned int i;
+
+	for (i = 0; i < nr_pages; i++) {
+		if (grefs[i] != INVALID_GRANT_REF) {
+			gnttab_end_foreign_access(grefs[i], 0);
+			grefs[i] = INVALID_GRANT_REF;
+		}
+	}
+
+	if (*vaddr)
+		free_pages_exact(*vaddr, nr_pages * XEN_PAGE_SIZE);
+	*vaddr = NULL;
+}
+EXPORT_SYMBOL_GPL(xenbus_teardown_ring);
 
 /**
  * Allocate an event channel for the given xenbus_device, assigning the newly
diff --git a/drivers/xen/xenbus/xenbus_probe.c b/drivers/xen/xenbus/xenbus_probe.c
index fe360c33ce71..d367f2bd2b93 100644
--- a/drivers/xen/xenbus/xenbus_probe.c
+++ b/drivers/xen/xenbus/xenbus_probe.c
@@ -65,6 +65,7 @@
 #include "xenbus.h"
 
 
+static int xs_init_irq;
 int xen_store_evtchn;
 EXPORT_SYMBOL_GPL(xen_store_evtchn);
 
@@ -750,6 +751,20 @@ static void xenbus_probe(void)
 {
 	xenstored_ready = 1;
 
+	if (!xen_store_interface) {
+		xen_store_interface = xen_remap(xen_store_gfn << XEN_PAGE_SHIFT,
+						XEN_PAGE_SIZE);
+		/*
+		 * Now it is safe to free the IRQ used for xenstore late
+		 * initialization. No need to unbind: it is about to be
+		 * bound again from xb_init_comms. Note that calling
+		 * unbind_from_irqhandler now would result in xen_evtchn_close()
+		 * being called and the event channel not being enabled again
+		 * afterwards, resulting in missed event notifications.
+		 */
+		free_irq(xs_init_irq, &xb_waitq);
+	}
+
 	/*
 	 * In the HVM case, xenbus_init() deferred its call to
 	 * xs_init() in case callbacks were not operational yet.
@@ -798,20 +813,22 @@ static int __init xenbus_probe_initcall(void)
 {
 	/*
 	 * Probe XenBus here in the XS_PV case, and also XS_HVM unless we
-	 * need to wait for the platform PCI device to come up.
+	 * need to wait for the platform PCI device to come up or
+	 * xen_store_interface is not ready.
 	 */
 	if (xen_store_domain_type == XS_PV ||
 	    (xen_store_domain_type == XS_HVM &&
-	     !xs_hvm_defer_init_for_callback()))
+	     !xs_hvm_defer_init_for_callback() &&
+	     xen_store_interface != NULL))
 		xenbus_probe();
 
 	/*
-	 * For XS_LOCAL, spawn a thread which will wait for xenstored
-	 * or a xenstore-stubdom to be started, then probe. It will be
-	 * triggered when communication starts happening, by waiting
-	 * on xb_waitq.
+	 * For XS_LOCAL or when xen_store_interface is not ready, spawn a
+	 * thread which will wait for xenstored or a xenstore-stubdom to be
+	 * started, then probe.  It will be triggered when communication
+	 * starts happening, by waiting on xb_waitq.
 	 */
-	if (xen_store_domain_type == XS_LOCAL) {
+	if (xen_store_domain_type == XS_LOCAL || xen_store_interface == NULL) {
 		struct task_struct *probe_task;
 
 		probe_task = kthread_run(xenbus_probe_thread, NULL,
@@ -907,10 +924,25 @@ static struct notifier_block xenbus_resume_nb = {
 	.notifier_call = xenbus_resume_cb,
 };
 
+static irqreturn_t xenbus_late_init(int irq, void *unused)
+{
+	int err;
+	uint64_t v = 0;
+
+	err = hvm_get_parameter(HVM_PARAM_STORE_PFN, &v);
+	if (err || !v || !~v)
+		return IRQ_HANDLED;
+	xen_store_gfn = (unsigned long)v;
+
+	wake_up(&xb_waitq);
+	return IRQ_HANDLED;
+}
+
 static int __init xenbus_init(void)
 {
 	int err;
 	uint64_t v = 0;
+	bool wait = false;
 	xen_store_domain_type = XS_UNKNOWN;
 
 	if (!xen_domain())
@@ -957,25 +989,44 @@ static int __init xenbus_init(void)
 		 * been properly initialized. Instead of attempting to map a
 		 * wrong guest physical address return error.
 		 *
-		 * Also recognize all bits set as an invalid value.
+		 * Also recognize all bits set as an invalid/uninitialized value.
 		 */
-		if (!v || !~v) {
+		if (!v) {
 			err = -ENOENT;
 			goto out_error;
 		}
-		/* Avoid truncation on 32-bit. */
+		if (v == ~0ULL) {
+			wait = true;
+		} else {
+			/* Avoid truncation on 32-bit. */
 #if BITS_PER_LONG == 32
-		if (v > ULONG_MAX) {
-			pr_err("%s: cannot handle HVM_PARAM_STORE_PFN=%llx > ULONG_MAX\n",
-			       __func__, v);
-			err = -EINVAL;
-			goto out_error;
-		}
+			if (v > ULONG_MAX) {
+				pr_err("%s: cannot handle HVM_PARAM_STORE_PFN=%llx > ULONG_MAX\n",
+				       __func__, v);
+				err = -EINVAL;
+				goto out_error;
+			}
 #endif
-		xen_store_gfn = (unsigned long)v;
-		xen_store_interface =
-			xen_remap(xen_store_gfn << XEN_PAGE_SHIFT,
-				  XEN_PAGE_SIZE);
+			xen_store_gfn = (unsigned long)v;
+			xen_store_interface =
+				xen_remap(xen_store_gfn << XEN_PAGE_SHIFT,
+					  XEN_PAGE_SIZE);
+			if (xen_store_interface->connection != XENSTORE_CONNECTED)
+				wait = true;
+		}
+		if (wait) {
+			err = bind_evtchn_to_irqhandler(xen_store_evtchn,
+							xenbus_late_init,
+							0, "xenstore_late_init",
+							&xb_waitq);
+			if (err < 0) {
+				pr_err("xenstore_late_init couldn't bind irq err=%d\n",
+				       err);
+				return err;
+			}
+
+			xs_init_irq = err;
+		}
 		break;
 	default:
 		pr_warn("Xenstore state unknown\n");