From: Tom Coughlan <coughlan@redhat.com>
Subject: [RHEL 5.0 PATCH] cciss bugfixes
Date: Thu, 14 Dec 2006 18:34:20 -0500
Bugzilla: 185021
Message-Id: <1166139261.6591.37.camel@bianchi.boston.redhat.com>
Changelog: cciss bugfixes


This is a set of patches that HP considers critical for cciss. This
represents some cherry picking from their latest upstream submissions.
All are bug fixes.  All are in Linus' tree. 

- Remove the NR_CMDS define and replace it with a per controller value.
Most Smart Array controllers can support up to 1024 commands but the
E200 family can only support 128. To prevent annoying "fifo full"
messages we define nr_cmds on a per controller basis by adding it the
product table.

- Change the SSID on the E500, as a workaround for a firmware bug.

- Disable DMA prefetch on the P600, another bug workaround.

- Add a check to make sure we don't try to start a queue on a disk which
is configuring. Otherwise, there was a small window when the interrupt
handler could fire and we would try to process the command, causing a
panic.

- Cleanup for cciss_interrupt_mode.

- Remove certain calls to pci_disable_device. These were causing insmod
after rmmod to fail. Important for ease of debugging.  

- Version number change.

HP has tested this on RHEL 5. I did some regression testing here with
some of the older CCISS hardware. 

Bug 185021.

Tom

Common subdirectories: linux-2.6.18.noarch/drivers/block-stock/aoe and linux-2.6.18.noarch/drivers/block/aoe
diff -up linux-2.6.18.noarch/drivers/block-stock/cciss.c linux-2.6.18.noarch/drivers/block/cciss.c
--- linux-2.6.18.noarch/drivers/block-stock/cciss.c
+++ linux-2.6.18.noarch/drivers/block/cciss.c
@@ -47,16 +47,16 @@
 #include <linux/completion.h>
 
 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
-#define DRIVER_NAME "HP CISS Driver (v 3.6.10)"
-#define DRIVER_VERSION CCISS_DRIVER_VERSION(3,6,10)
+#define DRIVER_NAME "HP CISS Driver (v 3.6.14-RH1)"
+#define DRIVER_VERSION CCISS_DRIVER_VERSION(3,6,14)
 
 /* Embedded module documentation macros - see modules.h */
 MODULE_AUTHOR("Hewlett-Packard Company");
-MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 3.6.10");
+MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 3.6.14-RH1");
 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
 			" SA6i P600 P800 P400 P400i E200 E200i E500");
+MODULE_VERSION("3.6.14-RH1");
 MODULE_LICENSE("GPL");
-MODULE_VERSION("2.6.8");
 
 #include "cciss_cmd.h"
 #include "cciss.h"
@@ -82,7 +82,7 @@ static const struct pci_device_id cciss_
 	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
 	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
-	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3233},
+	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 	{0,}
 };
 
@@ -91,27 +91,28 @@ MODULE_DEVICE_TABLE(pci, cciss_pci_devic
 /*  board_id = Subsystem Device ID & Vendor ID
  *  product = Marketing Name for the board
  *  access = Address of the struct of function pointers
+ *  nr_cmds = Number of commands supported by controller
  */
 static struct board_type products[] = {
-	{0x40700E11, "Smart Array 5300", &SA5_access},
-	{0x40800E11, "Smart Array 5i", &SA5B_access},
-	{0x40820E11, "Smart Array 532", &SA5B_access},
-	{0x40830E11, "Smart Array 5312", &SA5B_access},
-	{0x409A0E11, "Smart Array 641", &SA5_access},
-	{0x409B0E11, "Smart Array 642", &SA5_access},
-	{0x409C0E11, "Smart Array 6400", &SA5_access},
-	{0x409D0E11, "Smart Array 6400 EM", &SA5_access},
-	{0x40910E11, "Smart Array 6i", &SA5_access},
-	{0x3225103C, "Smart Array P600", &SA5_access},
-	{0x3223103C, "Smart Array P800", &SA5_access},
-	{0x3234103C, "Smart Array P400", &SA5_access},
-	{0x3235103C, "Smart Array P400i", &SA5_access},
-	{0x3211103C, "Smart Array E200i", &SA5_access},
-	{0x3212103C, "Smart Array E200", &SA5_access},
-	{0x3213103C, "Smart Array E200i", &SA5_access},
-	{0x3214103C, "Smart Array E200i", &SA5_access},
-	{0x3215103C, "Smart Array E200i", &SA5_access},
-	{0x3233103C, "Smart Array E500", &SA5_access},
+	{0x40700E11, "Smart Array 5300", &SA5_access, 512},
+	{0x40800E11, "Smart Array 5i", &SA5B_access, 512},
+	{0x40820E11, "Smart Array 532", &SA5B_access, 512},
+	{0x40830E11, "Smart Array 5312", &SA5B_access, 512},
+	{0x409A0E11, "Smart Array 641", &SA5_access, 512},
+	{0x409B0E11, "Smart Array 642", &SA5_access, 512},
+	{0x409C0E11, "Smart Array 6400", &SA5_access, 512},
+	{0x409D0E11, "Smart Array 6400 EM", &SA5_access, 512},
+	{0x40910E11, "Smart Array 6i", &SA5_access, 512},
+	{0x3225103C, "Smart Array P600", &SA5_access, 512},
+	{0x3223103C, "Smart Array P800", &SA5_access, 512},
+	{0x3234103C, "Smart Array P400", &SA5_access, 512},
+	{0x3235103C, "Smart Array P400i", &SA5_access, 512},
+	{0x3211103C, "Smart Array E200i", &SA5_access, 120},
+	{0x3212103C, "Smart Array E200", &SA5_access, 120},
+	{0x3213103C, "Smart Array E200i", &SA5_access, 120},
+	{0x3214103C, "Smart Array E200i", &SA5_access, 120},
+	{0x3215103C, "Smart Array E200i", &SA5_access, 120},
+	{0x3237103C, "Smart Array E500", &SA5_access, 512},
 };
 
 /* How long to wait (in milliseconds) for board to go into simple mode */
@@ -122,7 +123,6 @@ static struct board_type products[] = {
 #define MAX_CMD_RETRIES 3
 
 #define READ_AHEAD 	 1024
-#define NR_CMDS		 384	/* #commands that can be outstanding */
 #define MAX_CTLR	32
 
 /* Originally cciss driver only supports 8 major numbers */
@@ -401,8 +401,8 @@ static CommandList_struct *cmd_alloc(ctl
 	} else {		/* get it out of the controllers pool */
 
 		do {
-			i = find_first_zero_bit(h->cmd_pool_bits, NR_CMDS);
-			if (i == NR_CMDS)
+			i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
+			if (i == h->nr_cmds)
 				return NULL;
 		} while (test_and_set_bit
 			 (i & (BITS_PER_LONG - 1),
@@ -1245,7 +1245,7 @@ static void cciss_check_queues(ctlr_info
 	 * in case the interrupt we serviced was from an ioctl and did not
 	 * free any new commands.
 	 */
-	if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
+	if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
 		return;
 
 	/* We have room on the queue for more commands.  Now we need to queue
@@ -1257,14 +1257,17 @@ static void cciss_check_queues(ctlr_info
 		/* make sure the disk has been added and the drive is real
 		 * because this can be called from the middle of init_one.
 		 */
-		if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
+		if (!(h->drv[curr_queue].queue) ||
+		    !(h->drv[curr_queue].heads) ||
+		    h->drv[curr_queue].busy_configuring)
 			continue;
+
 		blk_start_queue(h->gendisk[curr_queue]->queue);
 
 		/* check to see if we have maxed out the number of commands
 		 * that can be placed on the queue.
 		 */
-		if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS) {
+		if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
 			if (curr_queue == start_queue) {
 				h->next_to_run =
 				    (start_queue + 1) % (h->highest_lun + 1);
@@ -2076,7 +2079,7 @@ static int add_sendcmd_reject(__u8 cmd, 
 
 	/* We've sent down an abort or reset, but something else
 	   has completed */
-	if (srl->ncompletions >= (NR_CMDS + 2)) {
+	if (srl->ncompletions >= (hba[ctlr]->nr_cmds + 2)) {
 		/* Uh oh.  No room to save it for later... */
 		printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
 		       "reject list overflow, command lost!\n", ctlr);
@@ -2594,7 +2597,7 @@ static irqreturn_t do_cciss_intr(int irq
 			a1 = a;
 			if ((a & 0x04)) {
 				a2 = (a >> 3);
-				if (a2 >= NR_CMDS) {
+				if (a2 >= h->nr_cmds) {
 					printk(KERN_WARNING
 					       "cciss: controller cciss%d failed, stopping.\n",
 					       h->ctlr);
@@ -2748,23 +2751,21 @@ static void __devinit cciss_interrupt_mo
 		if (err > 0) {
 			printk(KERN_WARNING "cciss: only %d MSI-X vectors "
 			       "available\n", err);
+			goto default_int_mode;
 		} else {
 			printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
 			       err);
+			goto default_int_mode;
 		}
 	}
 	if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
 		if (!pci_enable_msi(pdev)) {
-			c->intr[SIMPLE_MODE_INT] = pdev->irq;
 			c->msi_vector = 1;
-			return;
 		} else {
 			printk(KERN_WARNING "cciss: MSI init failed\n");
-			c->intr[SIMPLE_MODE_INT] = pdev->irq;
-			return;
 		}
 	}
-      default_int_mode:
+default_int_mode:
 #endif				/* CONFIG_PCI_MSI */
 	/* if we get here we're going to use the default interrupt mode */
 	c->intr[SIMPLE_MODE_INT] = pdev->irq;
@@ -2799,7 +2800,7 @@ static int cciss_pci_init(ctlr_info_t *c
 	if (err) {
 		printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
 		       "aborting\n");
-		goto err_out_disable_pdev;
+		return err;
 	}
 
 	subsystem_vendor_id = pdev->subsystem_vendor;
@@ -2877,6 +2878,7 @@ static int cciss_pci_init(ctlr_info_t *c
 		if (board_id == products[i].board_id) {
 			c->product_name = products[i].product_name;
 			c->access = *(products[i].access);
+			c->nr_cmds = products[i].nr_cmds;
 			break;
 		}
 	}
@@ -2905,6 +2907,17 @@ static int cciss_pci_init(ctlr_info_t *c
 	}
 #endif
 
+	/* Disabling DMA prefetch for the P600
+	 * An ASIC bug may result in a prefetch beyond
+	 * physical memory.
+	 */
+	if(board_id == 0x3225103C) {
+		__u32 dma_prefetch;
+		dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
+		dma_prefetch |= 0x8000;
+		writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
+	}
+
 #ifdef CCISS_DEBUG
 	printk("Trying to put board into Simple mode\n");
 #endif				/* CCISS_DEBUG */
@@ -2941,10 +2954,11 @@ static int cciss_pci_init(ctlr_info_t *c
 	return 0;
 
       err_out_free_res:
+	/*
+	 * Deliberately omit pci_disable_device(): it does something nasty to
+	 * Smart Array controllers that pci_enable_device does not undo
+	 */
 	pci_release_regions(pdev);
-
-      err_out_disable_pdev:
-	pci_disable_device(pdev);
 	return err;
 }
 
@@ -3182,15 +3196,15 @@ static int __devinit cciss_init_one(stru
 	       hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
 
 	hba[i]->cmd_pool_bits =
-	    kmalloc(((NR_CMDS + BITS_PER_LONG -
+	    kmalloc(((hba[i]->nr_cmds + BITS_PER_LONG -
 		      1) / BITS_PER_LONG) * sizeof(unsigned long), GFP_KERNEL);
 	hba[i]->cmd_pool = (CommandList_struct *)
 	    pci_alloc_consistent(hba[i]->pdev,
-		    NR_CMDS * sizeof(CommandList_struct),
+		    hba[i]->nr_cmds * sizeof(CommandList_struct),
 		    &(hba[i]->cmd_pool_dhandle));
 	hba[i]->errinfo_pool = (ErrorInfo_struct *)
 	    pci_alloc_consistent(hba[i]->pdev,
-		    NR_CMDS * sizeof(ErrorInfo_struct),
+		    hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
 		    &(hba[i]->errinfo_pool_dhandle));
 	if ((hba[i]->cmd_pool_bits == NULL)
 	    || (hba[i]->cmd_pool == NULL)
@@ -3201,7 +3215,7 @@ static int __devinit cciss_init_one(stru
 #ifdef CONFIG_CISS_SCSI_TAPE
 	hba[i]->scsi_rejects.complete =
 	    kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
-		    (NR_CMDS + 5), GFP_KERNEL);
+		    (hba[i]->nr_cmds + 5), GFP_KERNEL);
 	if (hba[i]->scsi_rejects.complete == NULL) {
 		printk(KERN_ERR "cciss: out of memory");
 		goto clean4;
@@ -3215,7 +3229,7 @@ static int __devinit cciss_init_one(stru
 	/* command and error info recs zeroed out before
 	   they are used */
 	memset(hba[i]->cmd_pool_bits, 0,
-	       ((NR_CMDS + BITS_PER_LONG -
+	       ((hba[i]->nr_cmds + BITS_PER_LONG -
 		 1) / BITS_PER_LONG) * sizeof(unsigned long));
 
 #ifdef CCISS_DEBUG
@@ -3284,11 +3298,11 @@ static int __devinit cciss_init_one(stru
 	kfree(hba[i]->cmd_pool_bits);
 	if (hba[i]->cmd_pool)
 		pci_free_consistent(hba[i]->pdev,
-				    NR_CMDS * sizeof(CommandList_struct),
+				    hba[i]->nr_cmds * sizeof(CommandList_struct),
 				    hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
 	if (hba[i]->errinfo_pool)
 		pci_free_consistent(hba[i]->pdev,
-				    NR_CMDS * sizeof(ErrorInfo_struct),
+				    hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
 				    hba[i]->errinfo_pool,
 				    hba[i]->errinfo_pool_dhandle);
 	free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
@@ -3355,16 +3369,19 @@ static void __devexit cciss_remove_one(s
 		}
 	}
 
-	pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
+	pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
 			    hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
-	pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(ErrorInfo_struct),
+	pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
 			    hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
 	kfree(hba[i]->cmd_pool_bits);
 #ifdef CONFIG_CISS_SCSI_TAPE
 	kfree(hba[i]->scsi_rejects.complete);
 #endif
+	/*
+	 * Deliberately omit pci_disable_device(): it does something nasty to
+	 * Smart Array controllers that pci_enable_device does not undo
+	 */
 	pci_release_regions(pdev);
-	pci_disable_device(pdev);
 	pci_set_drvdata(pdev, NULL);
 	free_hba(i);
 }
diff -up linux-2.6.18.noarch/drivers/block-stock/cciss_cmd.h linux-2.6.18.noarch/drivers/block/cciss_cmd.h
--- linux-2.6.18.noarch/drivers/block-stock/cciss_cmd.h
+++ linux-2.6.18.noarch/drivers/block/cciss_cmd.h
@@ -55,6 +55,7 @@
 #define I2O_INT_MASK            0x34
 #define I2O_IBPOST_Q            0x40
 #define I2O_OBPOST_Q            0x44
+#define I2O_DMA1_CFG		0x214
 
 //Configuration Table
 #define CFGTBL_ChangeReq        0x00000001l
diff -up linux-2.6.18.noarch/drivers/block-stock/cciss.h linux-2.6.18.noarch/drivers/block/cciss.h
--- linux-2.6.18.noarch/drivers/block-stock/cciss.h
+++ linux-2.6.18.noarch/drivers/block/cciss.h
@@ -60,6 +60,7 @@ struct ctlr_info 
 	__u32	board_id;
 	void __iomem *vaddr;
 	unsigned long paddr;
+	int 	nr_cmds; /* Number of commands allowed on this controller */
 	CfgTable_struct __iomem *cfgtable;
 	int	interrupts_enabled;
 	int	major;
@@ -279,6 +280,7 @@ struct board_type {
 	__u32	board_id;
 	char	*product_name;
 	struct access_method *access;
+	int nr_cmds; /* Max cmds this kind of ctlr can handle. */
 };
 
 #define CCISS_LOCK(i)	(&hba[i]->lock)
Common subdirectories: linux-2.6.18.noarch/drivers/block-stock/paride and linux-2.6.18.noarch/drivers/block/paride

Date: Mon, 18 Dec 2006 19:14:35 -0500
From: Tom Coughlan <coughlan@redhat.com>
X-Mailer: Evolution 2.8.0 (2.8.0-28.el5) 
To: rhkernel-list@redhat.com
Reply-To: rhkernel-list@redhat.com
Subject: Re: [RHEL 5.0 PATCH] cciss bugfixes
Message-Id: <1166487275.1111.11.camel@bianchi.boston.redhat.com>

On Thu, 2006-12-14 at 18:34 -0500, Tom Coughlan wrote:

One of the changes in this patch:

> - Disable DMA prefetch on the P600, another bug workaround.

was incomplete. That fix has:

+       if(board_id == 0x3225103C) {
+               __u32 dma_prefetch;
+               dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
+               dma_prefetch |= 0x8000;
+               writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);

and:

+#define I2O_DMA1_CFG           0x214

In order to do this, it is necessary to map out more of the config
table, to get to this bit. The attached patch fixes it (incremental to
the original patch).

This patch is posted upstream:

http://marc.theaimsgroup.com/?l=linux-scsi&m=116595333400191&w=2

I built and ran with it. I do not know if we have one of these boards. 

Tom


--- linux-2.6.18.noarch/drivers/block/cciss.c.old
+++ linux-2.6.18.noarch/drivers/block/cciss.c
@@ -2828,7 +2828,7 @@ static int cciss_pci_init(ctlr_info_t *c
 #ifdef CCISS_DEBUG
 	printk("address 0 = %x\n", c->paddr);
 #endif				/* CCISS_DEBUG */
-	c->vaddr = remap_pci_mem(c->paddr, 200);
+	c->vaddr = remap_pci_mem(c->paddr, 0x250);
 
 	/* Wait for the board to become ready.  (PCI hotplug needs this.)
 	 * We poll for up to 120 secs, once per 100ms. */