From: AMEET M. PARANJAPE <aparanja@redhat.com>
Date: Tue, 5 May 2009 14:50:25 -0400
Subject: [scsi] ibmvscsi: LPAR hang on a multipath device
Message-id: 20090505184741.27402.8239.sendpatchset@squad5-lp1.lab.bos.redhat.com
O-Subject: [PATCH RHEL5.4 BZ498927] Fix LPAR hang on a multipath device while enabling/disabling ports on the Fibre Channel Switch
Bugzilla: 498927
RH-Acked-by: David Howells <dhowells@redhat.com>

RHBZ#:
======
https://bugzilla.redhat.com/show_bug.cgi?id=498927

Description:
===========
Previously we had one timeout that was used for all types of operations.
This patch adds specific timeout values for different operations (init, login,
adapter info MAD, abort task, and LUN reset).

The extremely short timeout present in the previous shipping code will lead
to data loss and possibly system hangs when the root fs is affected.

RHEL Version Found:
================
RHEL 5.3

kABI Status:
============
No symbols were harmed.

Brew:
=====
Built on all platforms.
http://brewweb.devel.redhat.com/brew/taskinfo?taskID=1785055

Upstream Status:
================
There is no GIT ID for mainline acceptance.  IBM has a larger patch set that
we're preparing that has a host of changes for ibmvscsic and this is one of the
patches on the stack.  We'll be submitting the set to the scsi maintainer for
the 2.6.31 change window.

Therefore, I will provide the GIT ID after it becomes available.

Test Status:
============
1) Setup a power machine with 3 VIOS LPARs:

 * A root VIOS, that serves the root filesystem for the client partitions
 * 2 VIOS that provide 2 paths to a same set of disks located on a SAN array
device

2) Configure multipath on the linux client. You will have now disks located
under /dev/mapper/mpathX, that represent each one of the multipath devices
3) Format the multipath devices with ext3
4) Start running the bonnie filesystem benchmark test on one of the multipath
devices.

The VIOS are connected to the SAN array, so their FC adapters are tied to FC
switch ports. So with that in mind

5) Start turning on/off the ports on the switch, using the commands:

portdisable [port-number]
portenable [port-number]

With the patch applied the environment behaves as expected.
===============================================================
Ameet Paranjape 978-392-3903 ext 23903
IBM on-site partner

Proposed Patch:
===============

diff --git a/drivers/scsi/ibmvscsi/ibmvscsi.c b/drivers/scsi/ibmvscsi/ibmvscsi.c
index a23e4e9..bc7af95 100644
--- a/drivers/scsi/ibmvscsi/ibmvscsi.c
+++ b/drivers/scsi/ibmvscsi/ibmvscsi.c
@@ -84,7 +84,11 @@
  */
 static int max_id = 64;
 static int max_channel = 3;
-static int init_timeout = 5;
+static int init_timeout = 120;
+static int login_timeout = 60;
+static int info_timeout = 30;
+static int abort_timeout = 60;
+static int reset_timeout = 60;
 static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
 static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
 
@@ -877,7 +881,7 @@ static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
 	init_event_struct(evt_struct,
 			  adapter_info_rsp,
 			  VIOSRP_MAD_FORMAT,
-			  init_timeout);
+			  info_timeout);
 	
 	req = &evt_struct->iu.mad.adapter_info;
 	memset(req, 0x00, sizeof(*req));
@@ -896,7 +900,7 @@ static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
 	}
 	
 	spin_lock_irqsave(hostdata->host->host_lock, flags);
-	if (ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2)) {
+	if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2)) {
 		dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
 		dma_unmap_single(hostdata->dev,
 				 addr,
@@ -972,7 +976,7 @@ static int send_srp_login(struct ibmvscsi_host_data *hostdata)
 	init_event_struct(evt_struct,
 			  login_rsp,
 			  VIOSRP_SRP_FORMAT,
-			  init_timeout);
+			  login_timeout);
 
 	login = &evt_struct->iu.srp.login_req;
 	memset(login, 0x00, sizeof(struct srp_login_req));
@@ -987,7 +991,7 @@ static int send_srp_login(struct ibmvscsi_host_data *hostdata)
 	 */
 	atomic_set(&hostdata->request_limit, 0);
 
-	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
+	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
 	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 	dev_info(hostdata->dev, "sent SRP login\n");
 	return rc;
@@ -1028,7 +1032,7 @@ static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
 	 * out the correct tag
 	 */
 	spin_lock_irqsave(hostdata->host->host_lock, flags);
-	wait_switch = jiffies + (init_timeout * HZ);
+	wait_switch = jiffies + (abort_timeout * HZ);
 	do {
 		found_evt = NULL;
 		list_for_each_entry(tmp_evt, &hostdata->sent, list) {
@@ -1053,7 +1057,7 @@ static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
 		init_event_struct(evt,
 				  sync_completion,
 				  VIOSRP_SRP_FORMAT,
-				  init_timeout);
+				  abort_timeout);
 
 		tsk_mgmt = &evt->iu.srp.tsk_mgmt;
 	
@@ -1067,7 +1071,8 @@ static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
 		evt->sync_srp = &srp_rsp;
 
 		init_completion(&evt->comp);
-		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
+		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata,
+		                                 abort_timeout * 2);
 
 		if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
 			break;
@@ -1163,7 +1168,7 @@ static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
 	unsigned long wait_switch = 0;
 
 	spin_lock_irqsave(hostdata->host->host_lock, flags);
-	wait_switch = jiffies + (init_timeout * HZ);
+	wait_switch = jiffies + (reset_timeout * HZ);
 	do {
 		evt = get_event_struct(&hostdata->pool);
 		if (evt == NULL) {
@@ -1176,7 +1181,7 @@ static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
 		init_event_struct(evt,
 				  sync_completion,
 				  VIOSRP_SRP_FORMAT,
-				  init_timeout);
+				  reset_timeout);
 
 		tsk_mgmt = &evt->iu.srp.tsk_mgmt;
 
@@ -1189,7 +1194,8 @@ static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
 		evt->sync_srp = &srp_rsp;
 
 		init_completion(&evt->comp);
-		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, init_timeout * 2);
+		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata,
+		                                 reset_timeout * 2);
 
 		if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
 			break;
@@ -1270,7 +1276,7 @@ static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
 
 	ibmvscsi_reset_host(hostdata);
 
-	for (wait_switch = jiffies + (init_timeout * HZ);
+	for (wait_switch = jiffies + (reset_timeout * HZ);
 	     time_before(jiffies, wait_switch) &&
 		     atomic_read(&hostdata->request_limit) < 2;) {
 
@@ -1421,7 +1427,7 @@ static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
 	init_event_struct(evt_struct,
 			  sync_completion,
 			  VIOSRP_MAD_FORMAT,
-			  init_timeout);
+			  info_timeout);
 
 	host_config = &evt_struct->iu.mad.host_config;
 
@@ -1441,7 +1447,7 @@ static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
 
 	init_completion(&evt_struct->comp);
 	spin_lock_irqsave(hostdata->host->host_lock, flags);
-	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
+	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
 	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
 	if (rc == 0)
 		wait_for_completion(&evt_struct->comp);