From: Don Dutile <ddutile@redhat.com>
Date: Thu, 11 Dec 2008 13:37:58 -0500
Subject: [xen] pv_hvm: guest hang on FV save/restore
Message-id: 49415E06.2060003@redhat.com
O-Subject: [RHEL5.3 PATCH] Guest hang on FV save/restore
Bugzilla: 475778
RH-Acked-by: Bill Burns <bburns@redhat.com>
RH-Acked-by: Chris Lalancette <clalance@redhat.com>
RH-Acked-by: Rik van Riel <riel@redhat.com>

BZ 475778

While Chris Lalancette was testing his latest patches
for block-attach/detach as well as reboot,
he found that a FV save/restore would hang the restored guest
when the guest was configured w/vcpus=8 (on an 8 cpu system).

The failure is due to a livelock during secondary cpu shutdowns.

Fix is from upstream xen-unstable, cset 18669.

Chris tested an x86_64 kernel-2.6.126 patched FV guest
with the fix, and what once failed right away
worked for over 442 save/restore iterations (with background
kernel make -j 10).

Please review & ack.

Note: This is a regression since a rhel5.2 FV guest on 8vcpu didn't
       hang on restore.

- Don

diff --git a/drivers/xenpv_hvm/platform-pci/machine_reboot.c b/drivers/xenpv_hvm/platform-pci/machine_reboot.c
index 34cb488..0392c61 100644
--- a/drivers/xenpv_hvm/platform-pci/machine_reboot.c
+++ b/drivers/xenpv_hvm/platform-pci/machine_reboot.c
@@ -10,12 +10,6 @@ struct ap_suspend_info {
 	atomic_t nr_spinning;
 };
 
-/*
- * Use a rwlock to protect the hypercall page from being executed in AP context
- * while the BSP is re-initializing it after restore.
- */
-static DEFINE_RWLOCK(suspend_lock);
-
 #ifdef CONFIG_SMP
 /*
  * Spinning prevents, for example, APs touching grant table entries while
@@ -31,12 +25,8 @@ static void ap_suspend(void *_info)
 	atomic_inc(&info->nr_spinning);
 	mb();
 
-	while (info->do_spin) {
+	while (info->do_spin) 
 		cpu_relax();
-		read_lock(&suspend_lock);
-		HYPERVISOR_yield();
-		read_unlock(&suspend_lock);
-	}
 
 	mb();
 	atomic_dec(&info->nr_spinning);
@@ -52,9 +42,7 @@ static int bp_suspend(void)
 	suspend_cancelled = HYPERVISOR_shutdown(SHUTDOWN_suspend);
 
 	if (!suspend_cancelled) {
-		write_lock(&suspend_lock);
 		platform_pci_resume();
-		write_unlock(&suspend_lock);
 		gnttab_resume();
 		irq_resume();
 	}