svn commit: r301924 - stable/10/sys/dev/hyperv/vmbus

[Sepherosa Ziehau]

Author: sephe
Date: Wed Jun 15 09:39:41 2016
New Revision: 301924
URL: https://svnweb.freebsd.org/changeset/base/301924

Log:
  MFC 297219
  
      hyperv/vmbus: use a better retry method in hv_vmbus_post_message()
  
      Most often, hv_vmbus_post_message() doesn't fail.  However, it fails
      intermittently when GPADLs of large shared memory is to be established
      with the host, e.g. on the hn(4) attach path: a GPADL of 15MB sendbuf
      is created, for which lots of messages will be flooded to the host.
      The host side tries to throttle the message rate by returning
      HV_STATUS_INSUFFICIENT_BUFFERS.
  
      Before this commit, we do several retries for failed messages, but the
      delay between each retry is pretty/too low, which will cause sporadic
      message posting failure.  We now use large delay (>=1ms) between each
      retry to fix the message posting failure.
  
      Submitted by:       Dexuan Cui <decui microsoft com>
      Reviewed by:        sephe
      MFC after:  1 week
      Sponsored by:       Microsoft OSTC
      Differential Revision:      https://reviews.freebsd.org/D5715

Modified:
  stable/10/sys/dev/hyperv/vmbus/hv_connection.c
  stable/10/sys/dev/hyperv/vmbus/hv_vmbus_priv.h
Directory Properties:
  stable/10/   (props changed)

Modified: stable/10/sys/dev/hyperv/vmbus/hv_connection.c
==============================================================================
--- stable/10/sys/dev/hyperv/vmbus/hv_connection.c	Wed Jun 15 09:08:14 2016	(r301923)
+++ stable/10/sys/dev/hyperv/vmbus/hv_connection.c	Wed Jun 15 09:39:41 2016	(r301924)
@@ -367,31 +367,35 @@ hv_vmbus_on_events(int cpu)
 /**
  * Send a msg on the vmbus's message connection
  */
-int hv_vmbus_post_message(void *buffer, size_t bufferLen) {
-	int ret = 0;
+int hv_vmbus_post_message(void *buffer, size_t bufferLen)
+{
 	hv_vmbus_connection_id connId;
-	unsigned retries = 0;
-
-	/* NetScaler delays from previous code were consolidated here */
-	static int delayAmount[] = {100, 100, 100, 500, 500, 5000, 5000, 5000};
-
-	/* for(each entry in delayAmount) try to post message,
-	 *  delay a little bit before retrying
+	sbintime_t time = SBT_1MS;
+	int retries;
+	int ret;
+
+	connId.as_uint32_t = 0;
+	connId.u.id = HV_VMBUS_MESSAGE_CONNECTION_ID;
+
+	/*
+	 * We retry to cope with transient failures caused by host side's
+	 * insufficient resources. 20 times should suffice in practice.
 	 */
-	for (retries = 0;
-	    retries < sizeof(delayAmount)/sizeof(delayAmount[0]); retries++) {
-	    connId.as_uint32_t = 0;
-	    connId.u.id = HV_VMBUS_MESSAGE_CONNECTION_ID;
-	    ret = hv_vmbus_post_msg_via_msg_ipc(connId, 1, buffer, bufferLen);
-	    if (ret != HV_STATUS_INSUFFICIENT_BUFFERS)
-		break;
-	    /* TODO: KYS We should use a blocking wait call */
-	    DELAY(delayAmount[retries]);
+	for (retries = 0; retries < 20; retries++) {
+		ret = hv_vmbus_post_msg_via_msg_ipc(connId, 1, buffer,
+						    bufferLen);
+		if (ret == HV_STATUS_SUCCESS)
+			return (0);
+
+		pause_sbt("pstmsg", time, 0, C_HARDCLOCK);
+		if (time < SBT_1S * 2)
+			time *= 2;
 	}
 
-	KASSERT(ret == 0, ("Error VMBUS: Message Post Failed\n"));
+	KASSERT(ret == HV_STATUS_SUCCESS,
+		("Error VMBUS: Message Post Failed, ret=%d\n", ret));
 
-	return (ret);
+	return (EAGAIN);
 }
 
 /**

Modified: stable/10/sys/dev/hyperv/vmbus/hv_vmbus_priv.h
==============================================================================
--- stable/10/sys/dev/hyperv/vmbus/hv_vmbus_priv.h	Wed Jun 15 09:08:14 2016	(r301923)
+++ stable/10/sys/dev/hyperv/vmbus/hv_vmbus_priv.h	Wed Jun 15 09:39:41 2016	(r301924)
@@ -70,6 +70,7 @@ typedef uint16_t hv_vmbus_status;
  *    You did not supply enough message buffers to send a message.
  */
 
+#define HV_STATUS_SUCCESS                ((uint16_t)0)
 #define HV_STATUS_INSUFFICIENT_BUFFERS   ((uint16_t)0x0013)
 
 typedef void (*hv_vmbus_channel_callback)(void *context);