[SCSI] aacraid: Series 7 Async. (performance) mode support

- Series 7 Async. (performance) mode support added
- New scatter/gather list format for Series 7
- Driver converts s/g list to a firmware suitable list for best performance on
  Series 7, this can be disabled with driver parameter "aac_convert_sgl" for
  testing purposes
- New container read/write command structure for Series 7
- Fast response support for the SCSI pass-through path added
- Async. status response buffer changes

Signed-off-by: Mahesh Rajashekhara <Mahesh_Rajashekhara@pmc-sierra.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

8 files changed, 357 insertions, 125 deletions

diff --git a/drivers/scsi/aacraid/aachba.c b/drivers/scsi/aacraid/aachba.c
index 52551662d107..d79457ac8bef 100644
--- a/drivers/scsi/aacraid/aachba.c
+++ b/drivers/scsi/aacraid/aachba.c
@@ -135,6 +135,8 @@ struct inquiry_data {
 static unsigned long aac_build_sg(struct scsi_cmnd* scsicmd, struct sgmap* sgmap);
 static unsigned long aac_build_sg64(struct scsi_cmnd* scsicmd, struct sgmap64* psg);
 static unsigned long aac_build_sgraw(struct scsi_cmnd* scsicmd, struct sgmapraw* psg);
+static unsigned long aac_build_sgraw2(struct scsi_cmnd *scsicmd, struct aac_raw_io2 *rio2, int sg_max);
+static int aac_convert_sgraw2(struct aac_raw_io2 *rio2, int pages, int nseg, int nseg_new);
 static int aac_send_srb_fib(struct scsi_cmnd* scsicmd);
 #ifdef AAC_DETAILED_STATUS_INFO
 static char *aac_get_status_string(u32 status);
@@ -152,10 +154,14 @@ int aac_commit = -1;
 int startup_timeout = 180;
 int aif_timeout = 120;
 int aac_sync_mode;  /* Only Sync. transfer - disabled */
+int aac_convert_sgl = 1;	/* convert non-conformable s/g list - enabled */
 
 module_param(aac_sync_mode, int, S_IRUGO|S_IWUSR);
 MODULE_PARM_DESC(aac_sync_mode, "Force sync. transfer mode"
 	" 0=off, 1=on");
+module_param(aac_convert_sgl, int, S_IRUGO|S_IWUSR);
+MODULE_PARM_DESC(aac_convert_sgl, "Convert non-conformable s/g list"
+	" 0=off, 1=on");
 module_param(nondasd, int, S_IRUGO|S_IWUSR);
 MODULE_PARM_DESC(nondasd, "Control scanning of hba for nondasd devices."
 	" 0=off, 1=on");
@@ -963,25 +969,44 @@ static void io_callback(void *context, struct fib * fibptr);
 
 static int aac_read_raw_io(struct fib * fib, struct scsi_cmnd * cmd, u64 lba, u32 count)
 {
-	u16 fibsize;
-	struct aac_raw_io *readcmd;
+	struct aac_dev *dev = fib->dev;
+	u16 fibsize, command;
+
 	aac_fib_init(fib);
-	readcmd = (struct aac_raw_io *) fib_data(fib);
-	readcmd->block[0] = cpu_to_le32((u32)(lba&0xffffffff));
-	readcmd->block[1] = cpu_to_le32((u32)((lba&0xffffffff00000000LL)>>32));
-	readcmd->count = cpu_to_le32(count<<9);
-	readcmd->cid = cpu_to_le16(scmd_id(cmd));
-	readcmd->flags = cpu_to_le16(IO_TYPE_READ);
-	readcmd->bpTotal = 0;
-	readcmd->bpComplete = 0;
+	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE2 && !dev->sync_mode) {
+		struct aac_raw_io2 *readcmd2;
+		readcmd2 = (struct aac_raw_io2 *) fib_data(fib);
+		memset(readcmd2, 0, sizeof(struct aac_raw_io2));
+		readcmd2->blockLow = cpu_to_le32((u32)(lba&0xffffffff));
+		readcmd2->blockHigh = cpu_to_le32((u32)((lba&0xffffffff00000000LL)>>32));
+		readcmd2->byteCount = cpu_to_le32(count<<9);
+		readcmd2->cid = cpu_to_le16(scmd_id(cmd));
+		readcmd2->flags = cpu_to_le16(RIO2_IO_TYPE_READ);
+		aac_build_sgraw2(cmd, readcmd2, dev->scsi_host_ptr->sg_tablesize);
+		command = ContainerRawIo2;
+		fibsize = sizeof(struct aac_raw_io2) +
+			((le32_to_cpu(readcmd2->sgeCnt)-1) * sizeof(struct sge_ieee1212));
+	} else {
+		struct aac_raw_io *readcmd;
+		readcmd = (struct aac_raw_io *) fib_data(fib);
+		readcmd->block[0] = cpu_to_le32((u32)(lba&0xffffffff));
+		readcmd->block[1] = cpu_to_le32((u32)((lba&0xffffffff00000000LL)>>32));
+		readcmd->count = cpu_to_le32(count<<9);
+		readcmd->cid = cpu_to_le16(scmd_id(cmd));
+		readcmd->flags = cpu_to_le16(RIO_TYPE_READ);
+		readcmd->bpTotal = 0;
+		readcmd->bpComplete = 0;
+		aac_build_sgraw(cmd, &readcmd->sg);
+		command = ContainerRawIo;
+		fibsize = sizeof(struct aac_raw_io) +
+			((le32_to_cpu(readcmd->sg.count)-1) * sizeof(struct sgentryraw));
+	}
 
-	aac_build_sgraw(cmd, &readcmd->sg);
-	fibsize = sizeof(struct aac_raw_io) + ((le32_to_cpu(readcmd->sg.count) - 1) * sizeof (struct sgentryraw));
 	BUG_ON(fibsize > (fib->dev->max_fib_size - sizeof(struct aac_fibhdr)));
 	/*
 	 *	Now send the Fib to the adapter
 	 */
-	return aac_fib_send(ContainerRawIo,
+	return aac_fib_send(command,
 			  fib,
 			  fibsize,
 			  FsaNormal,
@@ -1052,28 +1077,50 @@ static int aac_read_block(struct fib * fib, struct scsi_cmnd * cmd, u64 lba, u32
 
 static int aac_write_raw_io(struct fib * fib, struct scsi_cmnd * cmd, u64 lba, u32 count, int fua)
 {
-	u16 fibsize;
-	struct aac_raw_io *writecmd;
+	struct aac_dev *dev = fib->dev;
+	u16 fibsize, command;
+
 	aac_fib_init(fib);
-	writecmd = (struct aac_raw_io *) fib_data(fib);
-	writecmd->block[0] = cpu_to_le32((u32)(lba&0xffffffff));
-	writecmd->block[1] = cpu_to_le32((u32)((lba&0xffffffff00000000LL)>>32));
-	writecmd->count = cpu_to_le32(count<<9);
-	writecmd->cid = cpu_to_le16(scmd_id(cmd));
-	writecmd->flags = (fua && ((aac_cache & 5) != 1) &&
-	  (((aac_cache & 5) != 5) || !fib->dev->cache_protected)) ?
-		cpu_to_le16(IO_TYPE_WRITE|IO_SUREWRITE) :
-		cpu_to_le16(IO_TYPE_WRITE);
-	writecmd->bpTotal = 0;
-	writecmd->bpComplete = 0;
-
-	aac_build_sgraw(cmd, &writecmd->sg);
-	fibsize = sizeof(struct aac_raw_io) + ((le32_to_cpu(writecmd->sg.count) - 1) * sizeof (struct sgentryraw));
+	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE2 && !dev->sync_mode) {
+		struct aac_raw_io2 *writecmd2;
+		writecmd2 = (struct aac_raw_io2 *) fib_data(fib);
+		memset(writecmd2, 0, sizeof(struct aac_raw_io2));
+		writecmd2->blockLow = cpu_to_le32((u32)(lba&0xffffffff));
+		writecmd2->blockHigh = cpu_to_le32((u32)((lba&0xffffffff00000000LL)>>32));
+		writecmd2->byteCount = cpu_to_le32(count<<9);
+		writecmd2->cid = cpu_to_le16(scmd_id(cmd));
+		writecmd2->flags = (fua && ((aac_cache & 5) != 1) &&
+						   (((aac_cache & 5) != 5) || !fib->dev->cache_protected)) ?
+			cpu_to_le16(RIO2_IO_TYPE_WRITE|RIO2_IO_SUREWRITE) :
+			cpu_to_le16(RIO2_IO_TYPE_WRITE);
+		aac_build_sgraw2(cmd, writecmd2, dev->scsi_host_ptr->sg_tablesize);
+		command = ContainerRawIo2;
+		fibsize = sizeof(struct aac_raw_io2) +
+			((le32_to_cpu(writecmd2->sgeCnt)-1) * sizeof(struct sge_ieee1212));
+	} else {
+		struct aac_raw_io *writecmd;
+		writecmd = (struct aac_raw_io *) fib_data(fib);
+		writecmd->block[0] = cpu_to_le32((u32)(lba&0xffffffff));
+		writecmd->block[1] = cpu_to_le32((u32)((lba&0xffffffff00000000LL)>>32));
+		writecmd->count = cpu_to_le32(count<<9);
+		writecmd->cid = cpu_to_le16(scmd_id(cmd));
+		writecmd->flags = (fua && ((aac_cache & 5) != 1) &&
+						   (((aac_cache & 5) != 5) || !fib->dev->cache_protected)) ?
+			cpu_to_le16(RIO_TYPE_WRITE|RIO_SUREWRITE) :
+			cpu_to_le16(RIO_TYPE_WRITE);
+		writecmd->bpTotal = 0;
+		writecmd->bpComplete = 0;
+		aac_build_sgraw(cmd, &writecmd->sg);
+		command = ContainerRawIo;
+		fibsize = sizeof(struct aac_raw_io) +
+			((le32_to_cpu(writecmd->sg.count)-1) * sizeof (struct sgentryraw));
+	}
+
 	BUG_ON(fibsize > (fib->dev->max_fib_size - sizeof(struct aac_fibhdr)));
 	/*
 	 *	Now send the Fib to the adapter
 	 */
-	return aac_fib_send(ContainerRawIo,
+	return aac_fib_send(command,
 			  fib,
 			  fibsize,
 			  FsaNormal,
@@ -1492,8 +1539,6 @@ int aac_get_adapter_info(struct aac_dev* dev)
 			dev->a_ops.adapter_write = aac_write_block;
 		}
 		dev->scsi_host_ptr->max_sectors = AAC_MAX_32BIT_SGBCOUNT;
-		if (dev->adapter_info.options & AAC_OPT_NEW_COMM_TYPE1)
-			dev->adapter_info.options |= AAC_OPT_NEW_COMM;
 		if (!(dev->adapter_info.options & AAC_OPT_NEW_COMM)) {
 			/*
 			 * Worst case size that could cause sg overflow when
@@ -2616,12 +2661,18 @@ static void aac_srb_callback(void *context, struct fib * fibptr)
 	srbreply = (struct aac_srb_reply *) fib_data(fibptr);
 
 	scsicmd->sense_buffer[0] = '\0';  /* Initialize sense valid flag to false */
-	/*
-	 *	Calculate resid for sg
-	 */
 
-	scsi_set_resid(scsicmd, scsi_bufflen(scsicmd)
-		       - le32_to_cpu(srbreply->data_xfer_length));
+	if (fibptr->flags & FIB_CONTEXT_FLAG_FASTRESP) {
+		/* fast response */
+		srbreply->srb_status = cpu_to_le32(SRB_STATUS_SUCCESS);
+		srbreply->scsi_status = cpu_to_le32(SAM_STAT_GOOD);
+	} else {
+		/*
+		 *	Calculate resid for sg
+		 */
+		scsi_set_resid(scsicmd, scsi_bufflen(scsicmd)
+				   - le32_to_cpu(srbreply->data_xfer_length));
+	}
 
 	scsi_dma_unmap(scsicmd);
 
@@ -2954,6 +3005,118 @@ static unsigned long aac_build_sgraw(struct scsi_cmnd* scsicmd, struct sgmapraw*
 	return byte_count;
 }
 
+static unsigned long aac_build_sgraw2(struct scsi_cmnd *scsicmd, struct aac_raw_io2 *rio2, int sg_max)
+{
+	unsigned long byte_count = 0;
+	int nseg;
+
+	nseg = scsi_dma_map(scsicmd);
+	BUG_ON(nseg < 0);
+	if (nseg) {
+		struct scatterlist *sg;
+		int i, conformable = 0;
+		u32 min_size = PAGE_SIZE, cur_size;
+
+		scsi_for_each_sg(scsicmd, sg, nseg, i) {
+			int count = sg_dma_len(sg);
+			u64 addr = sg_dma_address(sg);
+
+			BUG_ON(i >= sg_max);
+			rio2->sge[i].addrHigh = cpu_to_le32((u32)(addr>>32));
+			rio2->sge[i].addrLow = cpu_to_le32((u32)(addr & 0xffffffff));
+			cur_size = cpu_to_le32(count);
+			rio2->sge[i].length = cur_size;
+			rio2->sge[i].flags = 0;
+			if (i == 0) {
+				conformable = 1;
+				rio2->sgeFirstSize = cur_size;
+			} else if (i == 1) {
+				rio2->sgeNominalSize = cur_size;
+				min_size = cur_size;
+			} else if ((i+1) < nseg && cur_size != rio2->sgeNominalSize) {
+				conformable = 0;
+				if (cur_size < min_size)
+					min_size = cur_size;
+			}
+			byte_count += count;
+		}
+
+		/* hba wants the size to be exact */
+		if (byte_count > scsi_bufflen(scsicmd)) {
+			u32 temp = le32_to_cpu(rio2->sge[i-1].length) -
+				(byte_count - scsi_bufflen(scsicmd));
+			rio2->sge[i-1].length = cpu_to_le32(temp);
+			byte_count = scsi_bufflen(scsicmd);
+		}
+
+		rio2->sgeCnt = cpu_to_le32(nseg);
+		rio2->flags |= cpu_to_le16(RIO2_SG_FORMAT_IEEE1212);
+		/* not conformable: evaluate required sg elements */
+		if (!conformable) {
+			int j, nseg_new = nseg, err_found;
+			for (i = min_size / PAGE_SIZE; i >= 1; --i) {
+				err_found = 0;
+				nseg_new = 2;
+				for (j = 1; j < nseg - 1; ++j) {
+					if (rio2->sge[j].length % (i*PAGE_SIZE)) {
+						err_found = 1;
+						break;
+					}
+					nseg_new += (rio2->sge[j].length / (i*PAGE_SIZE));
+				}
+				if (!err_found)
+					break;
+			}
+			if (i > 0 && nseg_new <= sg_max)
+				aac_convert_sgraw2(rio2, i, nseg, nseg_new);
+		} else
+			rio2->flags |= cpu_to_le16(RIO2_SGL_CONFORMANT);
+
+		/* Check for command underflow */
+		if (scsicmd->underflow && (byte_count < scsicmd->underflow)) {
+			printk(KERN_WARNING"aacraid: cmd len %08lX cmd underflow %08X\n",
+					byte_count, scsicmd->underflow);
+		}
+	}
+
+	return byte_count;
+}
+
+static int aac_convert_sgraw2(struct aac_raw_io2 *rio2, int pages, int nseg, int nseg_new)
+{
+	struct sge_ieee1212 *sge;
+	int i, j, pos;
+	u32 addr_low;
+
+	if (aac_convert_sgl == 0)
+		return 0;
+
+	sge = kmalloc(nseg_new * sizeof(struct sge_ieee1212), GFP_ATOMIC);
+	if (sge == NULL)
+		return -1;
+
+	for (i = 1, pos = 1; i < nseg-1; ++i) {
+		for (j = 0; j < rio2->sge[i].length / (pages * PAGE_SIZE); ++j) {
+			addr_low = rio2->sge[i].addrLow + j * pages * PAGE_SIZE;
+			sge[pos].addrLow = addr_low;
+			sge[pos].addrHigh = rio2->sge[i].addrHigh;
+			if (addr_low < rio2->sge[i].addrLow)
+				sge[pos].addrHigh++;
+			sge[pos].length = pages * PAGE_SIZE;
+			sge[pos].flags = 0;
+			pos++;
+		}
+	}
+	sge[pos] = rio2->sge[nseg-1];
+	memcpy(&rio2->sge[1], &sge[1], (nseg_new-1)*sizeof(struct sge_ieee1212));
+
+	kfree(sge);
+	rio2->sgeCnt = cpu_to_le32(nseg_new);
+	rio2->flags |= cpu_to_le16(RIO2_SGL_CONFORMANT);
+	rio2->sgeNominalSize = pages * PAGE_SIZE;
+	return 0;
+}
+
 #ifdef AAC_DETAILED_STATUS_INFO
 
 struct aac_srb_status_info {
diff --git a/drivers/scsi/aacraid/aacraid.h b/drivers/scsi/aacraid/aacraid.h
index 6ab32db73bb6..9e933a88a8bc 100644
--- a/drivers/scsi/aacraid/aacraid.h
+++ b/drivers/scsi/aacraid/aacraid.h
@@ -12,7 +12,7 @@
  *----------------------------------------------------------------------------*/
 
 #ifndef AAC_DRIVER_BUILD
-# define AAC_DRIVER_BUILD 28900
+# define AAC_DRIVER_BUILD 29800
 # define AAC_DRIVER_BRANCH "-ms"
 #endif
 #define MAXIMUM_NUM_CONTAINERS	32
@@ -100,6 +100,13 @@ struct user_sgentryraw {
 	u32		flags;	/* reserved for F/W use */
 };
 
+struct sge_ieee1212 {
+	u32	addrLow;
+	u32	addrHigh;
+	u32	length;
+	u32	flags;
+};
+
 /*
  *	SGMAP
  *
@@ -270,6 +277,8 @@ enum aac_queue_types {
  */
 
 #define		FIB_MAGIC	0x0001
+#define		FIB_MAGIC2	0x0004
+#define		FIB_MAGIC2_64	0x0005
 
 /*
  *	Define the priority levels the FSA communication routines support.
@@ -296,22 +305,20 @@ struct aac_fibhdr {
 	__le32 XferState;	/* Current transfer state for this CCB */
 	__le16 Command;		/* Routing information for the destination */
 	u8 StructType;		/* Type FIB */
-	u8 Flags;		/* Flags for FIB */
+	u8 Unused;		/* Unused */
 	__le16 Size;		/* Size of this FIB in bytes */
 	__le16 SenderSize;	/* Size of the FIB in the sender
 				   (for response sizing) */
 	__le32 SenderFibAddress;  /* Host defined data in the FIB */
-	__le32 ReceiverFibAddress;/* Logical address of this FIB for
-				     the adapter */
-	u32 SenderData;		/* Place holder for the sender to store data */
 	union {
-		struct {
-		    __le32 _ReceiverTimeStart;	/* Timestamp for
-						   receipt of fib */
-		    __le32 _ReceiverTimeDone;	/* Timestamp for
-						   completion of fib */
-		} _s;
-	} _u;
+		__le32 ReceiverFibAddress;/* Logical address of this FIB for
+				     the adapter (old) */
+		__le32 SenderFibAddressHigh;/* upper 32bit of phys. FIB address */
+		__le32 TimeStamp;	/* otherwise timestamp for FW internal use */
+	} u;
+	u32 Handle;		/* FIB handle used for MSGU commnunication */
+	u32 Previous;		/* FW internal use */
+	u32 Next;		/* FW internal use */
 };
 
 struct hw_fib {
@@ -361,6 +368,7 @@ struct hw_fib {
 #define		ContainerCommand		500
 #define		ContainerCommand64		501
 #define		ContainerRawIo			502
+#define		ContainerRawIo2			503
 /*
  *	Scsi Port commands (scsi passthrough)
  */
@@ -417,6 +425,7 @@ enum fib_xfer_state {
 #define ADAPTER_INIT_STRUCT_REVISION		3
 #define ADAPTER_INIT_STRUCT_REVISION_4		4 // rocket science
 #define ADAPTER_INIT_STRUCT_REVISION_6		6 /* PMC src */
+#define ADAPTER_INIT_STRUCT_REVISION_7		7 /* Denali */
 
 struct aac_init
 {
@@ -441,7 +450,9 @@ struct aac_init
 #define INITFLAGS_NEW_COMM_SUPPORTED	0x00000001
 #define INITFLAGS_DRIVER_USES_UTC_TIME	0x00000010
 #define INITFLAGS_DRIVER_SUPPORTS_PM	0x00000020
-#define INITFLAGS_NEW_COMM_TYPE1_SUPPORTED	0x00000041
+#define INITFLAGS_NEW_COMM_TYPE1_SUPPORTED	0x00000040
+#define INITFLAGS_FAST_JBOD_SUPPORTED	0x00000080
+#define INITFLAGS_NEW_COMM_TYPE2_SUPPORTED	0x00000100
 	__le32	MaxIoCommands;	/* max outstanding commands */
 	__le32	MaxIoSize;	/* largest I/O command */
 	__le32	MaxFibSize;	/* largest FIB to adapter */
@@ -1124,6 +1135,7 @@ struct aac_dev
 #	define AAC_COMM_PRODUCER 0
 #	define AAC_COMM_MESSAGE  1
 #	define AAC_COMM_MESSAGE_TYPE1	3
+#	define AAC_COMM_MESSAGE_TYPE2	4
 	u8			raw_io_interface;
 	u8			raw_io_64;
 	u8			printf_enabled;
@@ -1182,6 +1194,7 @@ struct aac_dev
 #define FIB_CONTEXT_FLAG_TIMED_OUT		(0x00000001)
 #define FIB_CONTEXT_FLAG			(0x00000002)
 #define FIB_CONTEXT_FLAG_WAIT			(0x00000004)
+#define FIB_CONTEXT_FLAG_FASTRESP		(0x00000008)
 
 /*
  *	Define the command values
@@ -1288,6 +1301,22 @@ struct aac_dev
 #define CMDATA_SYNCH		4
 #define CMUNSTABLE		5
 
+#define	RIO_TYPE_WRITE 			0x0000
+#define	RIO_TYPE_READ			0x0001
+#define	RIO_SUREWRITE			0x0008
+
+#define RIO2_IO_TYPE			0x0003
+#define RIO2_IO_TYPE_WRITE		0x0000
+#define RIO2_IO_TYPE_READ		0x0001
+#define RIO2_IO_TYPE_VERIFY		0x0002
+#define RIO2_IO_ERROR			0x0004
+#define RIO2_IO_SUREWRITE		0x0008
+#define RIO2_SGL_CONFORMANT		0x0010
+#define RIO2_SG_FORMAT			0xF000
+#define RIO2_SG_FORMAT_ARC		0x0000
+#define RIO2_SG_FORMAT_SRL		0x1000
+#define RIO2_SG_FORMAT_IEEE1212		0x2000
+
 struct aac_read
 {
 	__le32		command;
@@ -1332,9 +1361,6 @@ struct aac_write64
 	__le32		block;
 	__le16		pad;
 	__le16		flags;
-#define	IO_TYPE_WRITE 0x00000000
-#define	IO_TYPE_READ  0x00000001
-#define	IO_SUREWRITE  0x00000008
 	struct sgmap64	sg;	// Must be last in struct because it is variable
 };
 struct aac_write_reply
@@ -1355,6 +1381,22 @@ struct aac_raw_io
 	struct sgmapraw	sg;
 };
 
+struct aac_raw_io2 {
+	__le32		blockLow;
+	__le32		blockHigh;
+	__le32		byteCount;
+	__le16		cid;
+	__le16		flags;		/* RIO2 flags */
+	__le32		sgeFirstSize;	/* size of first sge el. */
+	__le32		sgeNominalSize;	/* size of 2nd sge el. (if conformant) */
+	u8		sgeCnt;		/* only 8 bits required */
+	u8		bpTotal;	/* reserved for F/W use */
+	u8		bpComplete;	/* reserved for F/W use */
+	u8		sgeFirstIndex;	/* reserved for F/W use */
+	u8		unused[4];
+	struct sge_ieee1212	sge[1];
+};
+
 #define CT_FLUSH_CACHE 129
 struct aac_synchronize {
 	__le32		command;	/* VM_ContainerConfig */
diff --git a/drivers/scsi/aacraid/commctrl.c b/drivers/scsi/aacraid/commctrl.c
index 0bd38da4ada0..1ef041bc60c8 100644
--- a/drivers/scsi/aacraid/commctrl.c
+++ b/drivers/scsi/aacraid/commctrl.c
@@ -498,6 +498,8 @@ static int aac_send_raw_srb(struct aac_dev* dev, void __user * arg)
 		return -ENOMEM;
 	}
 	aac_fib_init(srbfib);
+	/* raw_srb FIB is not FastResponseCapable */
+	srbfib->hw_fib_va->header.XferState &= ~cpu_to_le32(FastResponseCapable);
 
 	srbcmd = (struct aac_srb*) fib_data(srbfib);
 
diff --git a/drivers/scsi/aacraid/comminit.c b/drivers/scsi/aacraid/comminit.c
index 8e4b525b1b79..8e5d3be16127 100644
--- a/drivers/scsi/aacraid/comminit.c
+++ b/drivers/scsi/aacraid/comminit.c
@@ -58,7 +58,8 @@ static int aac_alloc_comm(struct aac_dev *dev, void **commaddr, unsigned long co
 	dma_addr_t phys;
 	unsigned long aac_max_hostphysmempages;
 
-	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1)
+	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1 ||
+	    dev->comm_interface == AAC_COMM_MESSAGE_TYPE2)
 		host_rrq_size = (dev->scsi_host_ptr->can_queue
 			+ AAC_NUM_MGT_FIB) * sizeof(u32);
 	size = fibsize + sizeof(struct aac_init) + commsize +
@@ -75,7 +76,8 @@ static int aac_alloc_comm(struct aac_dev *dev, void **commaddr, unsigned long co
 	dev->comm_phys = phys;
 	dev->comm_size = size;
 	
-	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1) {
+	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1 ||
+	    dev->comm_interface == AAC_COMM_MESSAGE_TYPE2) {
 		dev->host_rrq = (u32 *)(base + fibsize);
 		dev->host_rrq_pa = phys + fibsize;
 		memset(dev->host_rrq, 0, host_rrq_size);
@@ -115,26 +117,32 @@ static int aac_alloc_comm(struct aac_dev *dev, void **commaddr, unsigned long co
 	else
 		init->HostPhysMemPages = cpu_to_le32(AAC_MAX_HOSTPHYSMEMPAGES);
 
-	init->InitFlags = 0;
+	init->InitFlags = cpu_to_le32(INITFLAGS_DRIVER_USES_UTC_TIME |
+		INITFLAGS_DRIVER_SUPPORTS_PM);
+	init->MaxIoCommands = cpu_to_le32(dev->scsi_host_ptr->can_queue + AAC_NUM_MGT_FIB);
+	init->MaxIoSize = cpu_to_le32(dev->scsi_host_ptr->max_sectors << 9);
+	init->MaxFibSize = cpu_to_le32(dev->max_fib_size);
+	init->MaxNumAif = cpu_to_le32(dev->max_num_aif);
+
 	if (dev->comm_interface == AAC_COMM_MESSAGE) {
 		init->InitFlags |= cpu_to_le32(INITFLAGS_NEW_COMM_SUPPORTED);
 		dprintk((KERN_WARNING"aacraid: New Comm Interface enabled\n"));
 	} else if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1) {
 		init->InitStructRevision = cpu_to_le32(ADAPTER_INIT_STRUCT_REVISION_6);
-		init->InitFlags |= cpu_to_le32(INITFLAGS_NEW_COMM_TYPE1_SUPPORTED);
-		dprintk((KERN_WARNING
-			"aacraid: New Comm Interface type1 enabled\n"));
+		init->InitFlags |= cpu_to_le32(INITFLAGS_NEW_COMM_SUPPORTED |
+			INITFLAGS_NEW_COMM_TYPE1_SUPPORTED | INITFLAGS_FAST_JBOD_SUPPORTED);
+		init->HostRRQ_AddrHigh = cpu_to_le32((u32)((u64)dev->host_rrq_pa >> 32));
+		init->HostRRQ_AddrLow = cpu_to_le32((u32)(dev->host_rrq_pa & 0xffffffff));
+		dprintk((KERN_WARNING"aacraid: New Comm Interface type1 enabled\n"));
+	} else if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE2) {
+		init->InitStructRevision = cpu_to_le32(ADAPTER_INIT_STRUCT_REVISION_7);
+		init->InitFlags |= cpu_to_le32(INITFLAGS_NEW_COMM_SUPPORTED |
+			INITFLAGS_NEW_COMM_TYPE2_SUPPORTED | INITFLAGS_FAST_JBOD_SUPPORTED);
+		init->HostRRQ_AddrHigh = cpu_to_le32((u32)((u64)dev->host_rrq_pa >> 32));
+		init->HostRRQ_AddrLow = cpu_to_le32((u32)(dev->host_rrq_pa & 0xffffffff));
+		init->MiniPortRevision = cpu_to_le32(0L);		/* number of MSI-X */
+		dprintk((KERN_WARNING"aacraid: New Comm Interface type2 enabled\n"));
 	}
-	init->InitFlags |= cpu_to_le32(INITFLAGS_DRIVER_USES_UTC_TIME |
-				       INITFLAGS_DRIVER_SUPPORTS_PM);
-	init->MaxIoCommands = cpu_to_le32(dev->scsi_host_ptr->can_queue + AAC_NUM_MGT_FIB);
-	init->MaxIoSize = cpu_to_le32(dev->scsi_host_ptr->max_sectors << 9);
-	init->MaxFibSize = cpu_to_le32(dev->max_fib_size);
-
-	init->MaxNumAif = cpu_to_le32(dev->max_num_aif);
-	init->HostRRQ_AddrHigh = cpu_to_le32((u32)((u64)dev->host_rrq_pa >> 32));
-	init->HostRRQ_AddrLow = cpu_to_le32((u32)(dev->host_rrq_pa & 0xffffffff));
-
 
 	/*
 	 * Increment the base address by the amount already used
@@ -354,13 +362,15 @@ struct aac_dev *aac_init_adapter(struct aac_dev *dev)
 			if ((status[1] & le32_to_cpu(AAC_OPT_NEW_COMM_TYPE1))) {
 				/* driver supports TYPE1 (Tupelo) */
 				dev->comm_interface = AAC_COMM_MESSAGE_TYPE1;
+			} else if ((status[1] & le32_to_cpu(AAC_OPT_NEW_COMM_TYPE2))) {
+				/* driver supports TYPE2 (Denali) */
+				dev->comm_interface = AAC_COMM_MESSAGE_TYPE2;
 			} else if ((status[1] & le32_to_cpu(AAC_OPT_NEW_COMM_TYPE4)) ||
-				  (status[1] & le32_to_cpu(AAC_OPT_NEW_COMM_TYPE3)) ||
-				  (status[1] & le32_to_cpu(AAC_OPT_NEW_COMM_TYPE2))) {
-					/* driver doesn't support TYPE2 (Series7), TYPE3 and TYPE4 */
-					/* switch to sync. mode */
-					dev->comm_interface = AAC_COMM_MESSAGE_TYPE1;
-					dev->sync_mode = 1;
+				  (status[1] & le32_to_cpu(AAC_OPT_NEW_COMM_TYPE3))) {
+				/* driver doesn't TYPE3 and TYPE4 */
+				/* switch to sync. mode */
+				dev->comm_interface = AAC_COMM_MESSAGE_TYPE2;
+				dev->sync_mode = 1;
 			}
 		}
 		if ((dev->comm_interface == AAC_COMM_MESSAGE) &&
diff --git a/drivers/scsi/aacraid/commsup.c b/drivers/scsi/aacraid/commsup.c
index 906a5013edae..1be0776a80c4 100644
--- a/drivers/scsi/aacraid/commsup.c
+++ b/drivers/scsi/aacraid/commsup.c
@@ -136,6 +136,7 @@ int aac_fib_setup(struct aac_dev * dev)
 		i < (dev->scsi_host_ptr->can_queue + AAC_NUM_MGT_FIB);
 		i++, fibptr++)
 	{
+		fibptr->flags = 0;
 		fibptr->dev = dev;
 		fibptr->hw_fib_va = hw_fib;
 		fibptr->data = (void *) fibptr->hw_fib_va->data;
@@ -240,11 +241,11 @@ void aac_fib_init(struct fib *fibptr)
 {
 	struct hw_fib *hw_fib = fibptr->hw_fib_va;
 
+	memset(&hw_fib->header, 0, sizeof(struct aac_fibhdr));
 	hw_fib->header.StructType = FIB_MAGIC;
 	hw_fib->header.Size = cpu_to_le16(fibptr->dev->max_fib_size);
 	hw_fib->header.XferState = cpu_to_le32(HostOwned | FibInitialized | FibEmpty | FastResponseCapable);
-	hw_fib->header.SenderFibAddress = 0; /* Filled in later if needed */
-	hw_fib->header.ReceiverFibAddress = cpu_to_le32(fibptr->hw_fib_pa);
+	hw_fib->header.u.ReceiverFibAddress = cpu_to_le32(fibptr->hw_fib_pa);
 	hw_fib->header.SenderSize = cpu_to_le16(fibptr->dev->max_fib_size);
 }
 
@@ -259,7 +260,6 @@ void aac_fib_init(struct fib *fibptr)
 static void fib_dealloc(struct fib * fibptr)
 {
 	struct hw_fib *hw_fib = fibptr->hw_fib_va;
-	BUG_ON(hw_fib->header.StructType != FIB_MAGIC);
 	hw_fib->header.XferState = 0;
 }
 
@@ -370,7 +370,7 @@ int aac_queue_get(struct aac_dev * dev, u32 * index, u32 qid, struct hw_fib * hw
 		entry->size = cpu_to_le32(le16_to_cpu(hw_fib->header.Size));
 		entry->addr = hw_fib->header.SenderFibAddress;
 			/* Restore adapters pointer to the FIB */
-		hw_fib->header.ReceiverFibAddress = hw_fib->header.SenderFibAddress;	/* Let the adapter now where to find its data */
+		hw_fib->header.u.ReceiverFibAddress = hw_fib->header.SenderFibAddress;  /* Let the adapter now where to find its data */
 		map = 0;
 	}
 	/*
@@ -450,7 +450,7 @@ int aac_fib_send(u16 command, struct fib *fibptr, unsigned long size,
 	 */
 
 	hw_fib->header.SenderFibAddress = cpu_to_le32(((u32)(fibptr - dev->fibs)) << 2);
-	hw_fib->header.SenderData = (u32)(fibptr - dev->fibs);
+	hw_fib->header.Handle = (u32)(fibptr - dev->fibs) + 1;
 	/*
 	 *	Set FIB state to indicate where it came from and if we want a
 	 *	response from the adapter. Also load the command from the
@@ -460,7 +460,6 @@ int aac_fib_send(u16 command, struct fib *fibptr, unsigned long size,
 	 */
 	hw_fib->header.Command = cpu_to_le16(command);
 	hw_fib->header.XferState |= cpu_to_le32(SentFromHost);
-	fibptr->hw_fib_va->header.Flags = 0;	/* 0 the flags field - internal only*/
 	/*
 	 *	Set the size of the Fib we want to send to the adapter
 	 */
@@ -711,7 +710,8 @@ int aac_fib_adapter_complete(struct fib *fibptr, unsigned short size)
 	unsigned long nointr = 0;
 	unsigned long qflags;
 
-	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1) {
+	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE1 ||
+	    dev->comm_interface == AAC_COMM_MESSAGE_TYPE2) {
 		kfree(hw_fib);
 		return 0;
 	}
@@ -724,7 +724,9 @@ int aac_fib_adapter_complete(struct fib *fibptr, unsigned short size)
 	/*
 	 *	If we plan to do anything check the structure type first.
 	 */
-	if (hw_fib->header.StructType != FIB_MAGIC) {
+	if (hw_fib->header.StructType != FIB_MAGIC &&
+	    hw_fib->header.StructType != FIB_MAGIC2 &&
+	    hw_fib->header.StructType != FIB_MAGIC2_64) {
 		if (dev->comm_interface == AAC_COMM_MESSAGE)
 			kfree(hw_fib);
 		return -EINVAL;
@@ -786,7 +788,9 @@ int aac_fib_complete(struct fib *fibptr)
 	 *	If we plan to do anything check the structure type first.
 	 */
 
-	if (hw_fib->header.StructType != FIB_MAGIC)
+	if (hw_fib->header.StructType != FIB_MAGIC &&
+	    hw_fib->header.StructType != FIB_MAGIC2 &&
+	    hw_fib->header.StructType != FIB_MAGIC2_64)
 		return -EINVAL;
 	/*
 	 *	This block completes a cdb which orginated on the host and we
diff --git a/drivers/scsi/aacraid/dpcsup.c b/drivers/scsi/aacraid/dpcsup.c
index f0c66a80ad13..d81b2810f0f7 100644
--- a/drivers/scsi/aacraid/dpcsup.c
+++ b/drivers/scsi/aacraid/dpcsup.c
@@ -101,6 +101,7 @@ unsigned int aac_response_normal(struct aac_queue * q)
 			 */
 			*(__le32 *)hwfib->data = cpu_to_le32(ST_OK);
 			hwfib->header.XferState |= cpu_to_le32(AdapterProcessed);
+			fib->flags |= FIB_CONTEXT_FLAG_FASTRESP;
 		}
 
 		FIB_COUNTER_INCREMENT(aac_config.FibRecved);
@@ -121,7 +122,7 @@ unsigned int aac_response_normal(struct aac_queue * q)
 			 *	NOTE:  we cannot touch the fib after this
 			 *	    call, because it may have been deallocated.
 			 */
-			fib->flags = 0;
+			fib->flags &= FIB_CONTEXT_FLAG_FASTRESP;
 			fib->callback(fib->callback_data, fib);
 		} else {
 			unsigned long flagv;
@@ -367,6 +368,7 @@ unsigned int aac_intr_normal(struct aac_dev *dev, u32 index,
 			 */
 			*(__le32 *)hwfib->data = cpu_to_le32(ST_OK);
 			hwfib->header.XferState |= cpu_to_le32(AdapterProcessed);
+			fib->flags |= FIB_CONTEXT_FLAG_FASTRESP;
 		}
 
 		FIB_COUNTER_INCREMENT(aac_config.FibRecved);
@@ -387,7 +389,7 @@ unsigned int aac_intr_normal(struct aac_dev *dev, u32 index,
 			 *	NOTE:  we cannot touch the fib after this
 			 *	    call, because it may have been deallocated.
 			 */
-			fib->flags = 0;
+			fib->flags &= FIB_CONTEXT_FLAG_FASTRESP;
 			fib->callback(fib->callback_data, fib);
 		} else {
 			unsigned long flagv;
diff --git a/drivers/scsi/aacraid/linit.c b/drivers/scsi/aacraid/linit.c
index 2be612b0caa5..7199534cd07d 100644
--- a/drivers/scsi/aacraid/linit.c
+++ b/drivers/scsi/aacraid/linit.c
@@ -1166,7 +1166,7 @@ static int __devinit aac_probe_one(struct pci_dev *pdev,
 	aac->cardtype = index;
 	INIT_LIST_HEAD(&aac->entry);
 
-	aac->fibs = kmalloc(sizeof(struct fib) * (shost->can_queue + AAC_NUM_MGT_FIB), GFP_KERNEL);
+	aac->fibs = kzalloc(sizeof(struct fib) * (shost->can_queue + AAC_NUM_MGT_FIB), GFP_KERNEL);
 	if (!aac->fibs)
 		goto out_free_host;
 	spin_lock_init(&aac->fib_lock);
diff --git a/drivers/scsi/aacraid/src.c b/drivers/scsi/aacraid/src.c
index 0fb1f5507cd8..3b021ec63255 100644
--- a/drivers/scsi/aacraid/src.c
+++ b/drivers/scsi/aacraid/src.c
@@ -56,21 +56,10 @@ static irqreturn_t aac_src_intr_message(int irq, void *dev_id)
 	if (bellbits & PmDoorBellResponseSent) {
 		bellbits = PmDoorBellResponseSent;
 		/* handle async. status */
+		src_writel(dev, MUnit.ODR_C, bellbits);
+		src_readl(dev, MUnit.ODR_C);
 		our_interrupt = 1;
 		index = dev->host_rrq_idx;
-		if (dev->host_rrq[index] == 0) {
-			u32 old_index = index;
-			/* adjust index */
-			do {
-				index++;
-				if (index == dev->scsi_host_ptr->can_queue +
-							AAC_NUM_MGT_FIB)
-					index = 0;
-				if (dev->host_rrq[index] != 0)
-					break;
-			} while (index != old_index);
-			dev->host_rrq_idx = index;
-		}
 		for (;;) {
 			isFastResponse = 0;
 			/* remove toggle bit (31) */
@@ -93,6 +82,8 @@ static irqreturn_t aac_src_intr_message(int irq, void *dev_id)
 	} else {
 		bellbits_shifted = (bellbits >> SRC_ODR_SHIFT);
 		if (bellbits_shifted & DoorBellAifPending) {
+			src_writel(dev, MUnit.ODR_C, bellbits);
+			src_readl(dev, MUnit.ODR_C);
 			our_interrupt = 1;
 			/* handle AIF */
 			aac_intr_normal(dev, 0, 2, 0, NULL);
@@ -100,6 +91,13 @@ static irqreturn_t aac_src_intr_message(int irq, void *dev_id)
 			unsigned long sflags;
 			struct list_head *entry;
 			int send_it = 0;
+			extern int aac_sync_mode;
+
+			if (!aac_sync_mode) {
+				src_writel(dev, MUnit.ODR_C, bellbits);
+				src_readl(dev, MUnit.ODR_C);
+				our_interrupt = 1;
+			}
 
 			if (dev->sync_fib) {
 				our_interrupt = 1;
@@ -132,7 +130,6 @@ static irqreturn_t aac_src_intr_message(int irq, void *dev_id)
 	}
 
 	if (our_interrupt) {
-		src_writel(dev, MUnit.ODR_C, bellbits);
 		return IRQ_HANDLED;
 	}
 	return IRQ_NONE;
@@ -336,6 +333,9 @@ static void aac_src_start_adapter(struct aac_dev *dev)
 {
 	struct aac_init *init;
 
+	 /* reset host_rrq_idx first */
+	dev->host_rrq_idx = 0;
+
 	init = dev->init;
 	init->HostElapsedSeconds = cpu_to_le32(get_seconds());
 
@@ -397,31 +397,40 @@ static int aac_src_deliver_message(struct fib *fib)
 	q->numpending++;
 	spin_unlock_irqrestore(q->lock, qflags);
 
-	/* Calculate the amount to the fibsize bits */
-	fibsize = (sizeof(struct aac_fib_xporthdr) + hdr_size + 127) / 128 - 1;
-	if (fibsize > (ALIGN32 - 1))
-		return -EMSGSIZE;
+	if (dev->comm_interface == AAC_COMM_MESSAGE_TYPE2) {
+		/* Calculate the amount to the fibsize bits */
+		fibsize = (hdr_size + 127) / 128 - 1;
+		if (fibsize > (ALIGN32 - 1))
+			return -EMSGSIZE;
+		/* New FIB header, 32-bit */
+		address = fib->hw_fib_pa;
+		fib->hw_fib_va->header.StructType = FIB_MAGIC2;
+		fib->hw_fib_va->header.SenderFibAddress = (u32)address;
+		fib->hw_fib_va->header.u.TimeStamp = 0;
+		BUG_ON((u32)(address >> 32) != 0L);
+		address |= fibsize;
+	} else {
+		/* Calculate the amount to the fibsize bits */
+		fibsize = (sizeof(struct aac_fib_xporthdr) + hdr_size + 127) / 128 - 1;
+		if (fibsize > (ALIGN32 - 1))
+			return -EMSGSIZE;
+
+		/* Fill XPORT header */
+		pFibX = (void *)fib->hw_fib_va - sizeof(struct aac_fib_xporthdr);
+		pFibX->Handle = cpu_to_le32(fib->hw_fib_va->header.Handle);
+		pFibX->HostAddress = cpu_to_le64(fib->hw_fib_pa);
+		pFibX->Size = cpu_to_le32(hdr_size);
 
-	/* Fill XPORT header */
-	pFibX = (void *)fib->hw_fib_va - sizeof(struct aac_fib_xporthdr);
-	/*
-	 * This was stored by aac_fib_send() and it is the index into
-	 * dev->fibs. Not sure why we add 1 to it, but I suspect that it's
-	 * because it can't be zero when we pass it to the hardware. Note that
-	 * it was stored in native endian, hence the lack of swapping. -- BenC
-	 */
-	pFibX->Handle = cpu_to_le32(fib->hw_fib_va->header.SenderData + 1);
-	pFibX->HostAddress = cpu_to_le64(fib->hw_fib_pa);
-	pFibX->Size = cpu_to_le32(hdr_size);
+		/*
+		 * The xport header has been 32-byte aligned for us so that fibsize
+		 * can be masked out of this address by hardware. -- BenC
+		 */
+		address = fib->hw_fib_pa - sizeof(struct aac_fib_xporthdr);
+		if (address & (ALIGN32 - 1))
+			return -EINVAL;
+		address |= fibsize;
+	}
 
-	/*
-	 * The xport header has been 32-byte aligned for us so that fibsize
-	 * can be masked out of this address by hardware. -- BenC
-	 */
-	address = fib->hw_fib_pa - sizeof(struct aac_fib_xporthdr);
-	if (address & (ALIGN32 - 1))
-		return -EINVAL;
-	address |= fibsize;
 	src_writel(dev, MUnit.IQ_H, (address >> 32) & 0xffffffff);
 	src_writel(dev, MUnit.IQ_L, address & 0xffffffff);
 
@@ -764,7 +773,7 @@ int aac_srcv_init(struct aac_dev *dev)
 
 	if (aac_init_adapter(dev) == NULL)
 		goto error_iounmap;
-	if (dev->comm_interface != AAC_COMM_MESSAGE_TYPE1)
+	if (dev->comm_interface != AAC_COMM_MESSAGE_TYPE2)
 		goto error_iounmap;
 	dev->msi = aac_msi && !pci_enable_msi(dev->pdev);
 	if (request_irq(dev->pdev->irq, dev->a_ops.adapter_intr,