hv: change clockevents unbind tactics

To get prepared to CPU offlining support we need co change the way how we
unbind clockevent devices. As one CPU may go online/offline multiple times
we need to bind it in hv_synic_init() and unbind it in hv_synic_cleanup().
There is an additional corner case: when we unload the module completely we
need to switch to some other clockevent mechanism before stopping VMBus or
we will hang. We can't call hv_synic_cleanup() before unloading VMBus as
we won't be able to send UNLOAD request and get a response so
hv_synic_clockevents_cleanup() has to live. Luckily, we can always call
clockevents_unbind_device(), even if it wasn't bound before and there is
no issue if we call it twice.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 4 insertions, 2 deletions

diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
index 60dbd6c..446802a 100644
--- a/drivers/hv/hv.c
+++ b/drivers/hv/hv.c
@@ -575,7 +575,7 @@ void hv_synic_clockevents_cleanup(void)
 	if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
 		return;
 
-	for_each_online_cpu(cpu)
+	for_each_present_cpu(cpu)
 		clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
 }
 
@@ -594,8 +594,10 @@ void hv_synic_cleanup(void *arg)
 		return;
 
 	/* Turn off clockevent device */
-	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
+	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) {
+		clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
 		hv_ce_shutdown(hv_context.clk_evt[cpu]);
+	}
 
 	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);