[PATCH 2/2] Limit retry in ControllerGetResponse

["Tu Dinh" <ngoc-tu.dinh@xxxxxxxxxx>]

During unplug, NDIS may issue a hash reconfiguration. However, the
backend may disappear before the request has had a chance to complete.
This is observable by e.g. adding a delay at the beginning of
VifReceiverSetHashAlgorithm and unplugging. If this happens, Xenvif will
get stuck waiting forever in ControllerGetResponse, at DISPATCH_LEVEL to
boot.

Limit the EvtchnWait retry to 10 attempts (10ms each to keep the
previous 100ms limit) and return an error code to the caller if the wait
has failed.

Signed-off-by: Tu Dinh <ngoc-tu.dinh@xxxxxxxxxx>
---
 src/xenvif/controller.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/src/xenvif/controller.c b/src/xenvif/controller.c
index e032972..109c030 100644
--- a/src/xenvif/controller.c
+++ b/src/xenvif/controller.c
@@ -248,7 +248,7 @@ fail1:
 #define TIME_MS(_ms)        (TIME_US((_ms) * 1000))
 #define TIME_RELATIVE(_t)   (-(_t))
 
-#define XENVIF_CONTROLLER_POLL_PERIOD 100 // ms
+#define XENVIF_CONTROLLER_POLL_PERIOD 10 // ms
 
 _IRQL_requires_(DISPATCH_LEVEL)
 static NTSTATUS
@@ -258,11 +258,12 @@ ControllerGetResponse(
     )
 {
     LARGE_INTEGER                   Timeout;
+    ULONG                           Attempt;
     NTSTATUS                        status;
 
     Timeout.QuadPart = TIME_RELATIVE(TIME_MS(XENVIF_CONTROLLER_POLL_PERIOD));
 
-    for (;;) {
+    for (Attempt = 0; Attempt < 10; Attempt++) {
         ULONG   Count;
 
         Count = XENBUS_EVTCHN(GetCount,
@@ -284,6 +285,13 @@ ControllerGetResponse(
             __ControllerSend(Controller);
     }
 
+    // Use STATUS_TRANSACTION_TIMED_OUT as an error code since STATUS_TIMEOUT 
is
+    // a success code.
+    if (Controller->Response.id != Controller->Request.id) {
+        status = STATUS_TRANSACTION_TIMED_OUT;
+        goto done;
+    }
+
     ASSERT3U(Controller->Response.type, ==, Controller->Request.type);
 
     switch (Controller->Response.status) {
@@ -311,6 +319,7 @@ ControllerGetResponse(
     if (NT_SUCCESS(status) && Data != NULL)
         *Data = Controller->Response.data;
 
+done:
     RtlZeroMemory(&Controller->Request,
                   sizeof (struct xen_netif_ctrl_request));
     RtlZeroMemory(&Controller->Response,
-- 
2.51.0.windows.2



--
Ngoc Tu Dinh | Vates XCP-ng Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech