From: Chip Coldwell <coldwell@redhat.com>
Subject: Re: [RHEL5.1 PATCH] bz245184 megaraid_sas: intercept cmd timeout 	and throttle io
Date: Tue, 4 Sep 2007 14:15:31 -0400 (Eastern Daylight Time)
Bugzilla: 245184 247581
Message-Id: <Pine.WNT.4.64.0709041412280.604@chiptop>
Changelog: [scsi] megaraid_sas: intercept cmd timeout and throttle io

This patch resolves an issue originally raised by 225221.  It was
omitted during the omnibus megaraid_sas update because it required a
KABI-breaking change to the scsi_host struct.  I have papered over
this in the usual way (move the new struct element to the end and
surround it with a preprocessor #ifndef __GENKSYMS___).
Compile-tested by me

functionality testing by LSI.  I've also rolled in the fix for
bz247581, which just bumps the version number in the driver.

On Mon, 20 Aug 2007, Pete Zaitcev wrote:
> 
> BTW, if the bifield trick falls through, you can play further tricks.
> For example, put 0x1dead1af into a function pointer (like proc_info),
> and have the core detect that. If found, it means "extended" host
> template. It actually may be safer than a bitfield, I think.

I have implemented this recommendation, and tested that the megaraid_sas 
driver can be loaded on an unpatched kernel (2.6.18-36.el5) and the driver 
functions as expected.

The kernel has been build for all architectures, with no build issues:

Please review and ACK/NACK as apropriate.

 drivers/scsi/megaraid/megaraid_sas.c |   68 ++++++++++++++++++++++++++++++---
 drivers/scsi/megaraid/megaraid_sas.h |   14 ++++--
 drivers/scsi/scsi_error.c            |   12 +++++-
 include/scsi/scsi_host.h             |   19 +++++++++
 4 files changed, 100 insertions(+), 13 deletions(-)

diff --git a/drivers/scsi/megaraid/megaraid_sas.c b/drivers/scsi/megaraid/megaraid_sas.c
index 00598b0..ec799fe 100644
--- a/drivers/scsi/megaraid/megaraid_sas.c
+++ b/drivers/scsi/megaraid/megaraid_sas.c
@@ -10,7 +10,7 @@
  *	   2 of the License, or (at your option) any later version.
  *
  * FILE		: megaraid_sas.c
- * Version	: v00.00.03.05
+ * Version	: v00.00.03.10
  *
  * Authors:
  * 	Sreenivas Bagalkote	<Sreenivas.Bagalkote@lsi.com>
@@ -884,6 +884,7 @@ megasas_queue_command(struct scsi_cmnd *scmd, void (*done) (struct scsi_cmnd *))
 		goto out_return_cmd;
 
 	cmd->scmd = scmd;
+	scmd->SCp.ptr = (char *)cmd;
 
 	/*
 	 * Issue the command to the FW
@@ -917,7 +918,7 @@ static int megasas_slave_configure(struct scsi_device *sdev)
 	 * The RAID firmware may require extended timeouts.
 	 */
 	if (sdev->channel >= MEGASAS_MAX_PD_CHANNELS)
-		sdev->timeout = 90 * HZ;
+		sdev->timeout = MEGASAS_DEFAULT_CMD_TIMEOUT * HZ;
 	return 0;
 }
 
@@ -979,8 +980,8 @@ static int megasas_generic_reset(struct scsi_cmnd *scmd)
 
 	instance = (struct megasas_instance *)scmd->device->host->hostdata;
 
-	scmd_printk(KERN_NOTICE, scmd, "megasas: RESET -%ld cmd=%x\n",
-	       scmd->serial_number, scmd->cmnd[0]);
+	scmd_printk(KERN_NOTICE, scmd, "megasas: RESET -%ld cmd=%x retries=%x\n",
+		 scmd->serial_number, scmd->cmnd[0], scmd->retries);
 
 	if (instance->hw_crit_error) {
 		printk(KERN_ERR "megasas: cannot recover from previous reset "
@@ -998,6 +999,39 @@ static int megasas_generic_reset(struct scsi_cmnd *scmd)
 }
 
 /**
+ * megasas_reset_timer - quiesce the adapter if required
+ * @scmd:		scsi cmnd
+ *
+ * Sets the FW busy flag and reduces the host->can_queue if the
+ * cmd has not been completed within the timeout period.
+ */
+static enum
+scsi_eh_timer_return megasas_reset_timer(struct scsi_cmnd *scmd)
+{
+	struct megasas_cmd *cmd = (struct megasas_cmd *)scmd->SCp.ptr;
+	struct megasas_instance *instance;
+	unsigned long flags;
+
+	if (time_after(jiffies, scmd->jiffies_at_alloc +
+				(MEGASAS_DEFAULT_CMD_TIMEOUT * 2) * HZ)) {
+		return EH_NOT_HANDLED;
+	}
+
+	instance = cmd->instance;
+	if (!(instance->flag & MEGASAS_FW_BUSY)) {
+		/* FW is busy, throttle IO */
+		spin_lock_irqsave(instance->host->host_lock, flags);
+
+		instance->host->can_queue = 16;
+		instance->last_time = jiffies;
+		instance->flag |= MEGASAS_FW_BUSY;
+
+		spin_unlock_irqrestore(instance->host->host_lock, flags);
+	}
+	return EH_RESET_TIMER;
+}
+
+/**
  * megasas_reset_device -	Device reset handler entry point
  */
 static int megasas_reset_device(struct scsi_cmnd *scmd)
@@ -1110,6 +1144,8 @@ static struct scsi_host_template megasas_template = {
 	.eh_device_reset_handler = megasas_reset_device,
 	.eh_bus_reset_handler = megasas_reset_bus_host,
 	.eh_host_reset_handler = megasas_reset_bus_host,
+	.proc_info = (void *) RH_EXTENDED_MAGIC,
+	.eh_timed_out = megasas_reset_timer,
 	.bios_param = megasas_bios_param,
 	.use_clustering = ENABLE_CLUSTERING,
 };
@@ -1213,9 +1249,8 @@ megasas_complete_cmd(struct megasas_instance *instance, struct megasas_cmd *cmd,
 	int exception = 0;
 	struct megasas_header *hdr = &cmd->frame->hdr;
 
-	if (cmd->scmd) {
-		cmd->scmd->SCp.ptr = (char *)0;
-	}
+	if (cmd->scmd)
+		cmd->scmd->SCp.ptr = NULL;
 
 	switch (hdr->cmd) {
 
@@ -1804,6 +1839,7 @@ static void megasas_complete_cmd_dpc(unsigned long instance_addr)
 	u32 context;
 	struct megasas_cmd *cmd;
 	struct megasas_instance *instance = (struct megasas_instance *)instance_addr;
+	unsigned long flags;
 
 	/* If we have already declared adapter dead, donot complete cmds */
 	if (instance->hw_crit_error)
@@ -1826,6 +1862,22 @@ static void megasas_complete_cmd_dpc(unsigned long instance_addr)
 	}
 
 	*instance->consumer = producer;
+
+	/*
+	 * Check if we can restore can_queue
+	 */
+	if (instance->flag & MEGASAS_FW_BUSY
+		&& time_after(jiffies, instance->last_time + 5 * HZ)
+		&& atomic_read(&instance->fw_outstanding) < 17) {
+
+		spin_lock_irqsave(instance->host->host_lock, flags);
+		instance->flag &= ~MEGASAS_FW_BUSY;
+		instance->host->can_queue =
+				instance->max_fw_cmds - MEGASAS_INT_CMDS;
+
+		spin_unlock_irqrestore(instance->host->host_lock, flags);
+	}
+
 }
 
 /**
@@ -2396,6 +2448,8 @@ megasas_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	instance->init_id = MEGASAS_DEFAULT_INIT_ID;
 
 	megasas_dbg_lvl = 0;
+	instance->flag = 0;
+	instance->last_time = 0;
 
 	/*
 	 * Initialize MFI Firmware
diff --git a/drivers/scsi/megaraid/megaraid_sas.h b/drivers/scsi/megaraid/megaraid_sas.h
index 55eddcf..86604bf 100644
--- a/drivers/scsi/megaraid/megaraid_sas.h
+++ b/drivers/scsi/megaraid/megaraid_sas.h
@@ -18,9 +18,9 @@
 /**
  * MegaRAID SAS Driver meta data
  */
-#define MEGASAS_VERSION				"00.00.03.05"
-#define MEGASAS_RELDATE				"Oct 02, 2006"
-#define MEGASAS_EXT_VERSION			"Mon Oct 02 11:21:32 PDT 2006"
+#define MEGASAS_VERSION				"00.00.03.10"
+#define MEGASAS_RELDATE				"Mar 28, 2007"
+#define MEGASAS_EXT_VERSION			"Wed Mar 28 10:25:52 PST 2007"
 
 /*
  * Device IDs
@@ -539,6 +539,8 @@ struct megasas_ctrl_info {
 
 #define MEGASAS_DBG_LVL				1
 
+#define MEGASAS_FW_BUSY				1
+
 /*
  * When SCSI mid-layer calls driver's reset routine, driver waits for
  * MEGASAS_RESET_WAIT_TIME seconds for all outstanding IO to complete. Note
@@ -549,8 +551,8 @@ struct megasas_ctrl_info {
 #define MEGASAS_RESET_WAIT_TIME			180
 #define MEGASAS_INTERNAL_CMD_WAIT_TIME		180
 #define	MEGASAS_RESET_NOTICE_INTERVAL		5
-
 #define MEGASAS_IOCTL_CMD			0
+#define MEGASAS_DEFAULT_CMD_TIMEOUT		90
 
 /*
  * FW reports the maximum of number of commands that it can accept (maximum
@@ -1073,7 +1075,6 @@ struct megasas_instance {
 	struct megasas_register_set __iomem *reg_set;
 
 	s8 init_id;
-	u8 reserved[3];
 
 	u16 max_num_sge;
 	u16 max_fw_cmds;
@@ -1104,6 +1105,9 @@ struct megasas_instance {
 
 	struct megasas_instance_template *instancet;
 	struct tasklet_struct isr_tasklet;
+
+	u8 flag;
+	unsigned long last_time;
 };
 
 #define MEGASAS_IS_LOGICAL(scp)						\
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 5c07a3b..b775066 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -184,10 +184,20 @@ int scsi_delete_timer(struct scsi_cmnd *scmd)
  **/
 void scsi_times_out(struct scsi_cmnd *scmd)
 {
+	enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
+
 	scsi_log_completion(scmd, TIMEOUT_ERROR);
 
 	if (scmd->device->host->transportt->eh_timed_out)
-		switch (scmd->device->host->transportt->eh_timed_out(scmd)) {
+		eh_timed_out = scmd->device->host->transportt->eh_timed_out;
+	else if ((scmd->device->host->hostt->proc_info == (void *) RH_EXTENDED_MAGIC) &&
+		 scmd->device->host->hostt->eh_timed_out)
+		eh_timed_out = scmd->device->host->hostt->eh_timed_out;
+	else
+		eh_timed_out = NULL;
+
+	if (eh_timed_out)
+		switch (eh_timed_out(scmd)) {
 		case EH_HANDLED:
 			__scsi_done(scmd);
 			return;
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index e05a37e..ab322bc 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -285,6 +285,10 @@ struct scsi_host_template {
 	 *
 	 * Status: OBSOLETE
 	 */
+	/* Red Hat KABI: set proc_info to RH_EXTENDED_MAGIC to enable
+	   Red Hat extensions of the scsi_host_template structure, namely
+	   the eh_timed_out function pointer. */
+#define RH_EXTENDED_MAGIC 0x1dead1af
 	int (*proc_info)(struct Scsi_Host *, char *, char **, off_t, int, int);
 
 	/*
@@ -429,6 +433,21 @@ struct scsi_host_template {
 	 * module_init/module_exit.
 	 */
 	struct list_head legacy_hosts;
+
+#ifndef __GENKSYMS__
+	/*
+	 * This is an optional routine that allows the transport to become
+	 * involved when a scsi io timer fires. The return value tells the
+	 * timer routine how to finish the io timeout handling:
+	 * EH_HANDLED:		I fixed the error, please complete the command
+	 * EH_RESET_TIMER:	I need more time, reset the timer and
+	 *			begin counting again
+	 * EH_NOT_HANDLED	Begin normal error recovery
+	 *
+	 * Status: OPTIONAL
+	 */
+	enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
+#endif
 };
 
 /*

-- 
Charles M. "Chip" Coldwell
Senior Software Engineer
Red Hat, Inc
978-392-2426