[SCSI] mpt2sas: Added NUNA IO support in driver which uses multi-reply queue support of the HBA

Support added for controllers capable of multi reply queues.

The following are the modifications to the driver to support NUMA.

1) Create the new structure adapter_reply_queue to contain the reply queue
   info for every msix vector.  This object will contain a
   reply_post_host_index, reply_post_free for each instance, msix_index, among
   other parameters.  We will track all the reply queues on a link list called
   ioc->reply_queue_list. Each reply queue is aligned with each IRQ, and is
   passed to the interrupt via the bus_id parameter.

(2) The driver will figure out the msix_vector_count from the PCIe MSIX
    capabilities register instead of the IOC Facts->MaxMSIxVectors. This is
    because the firmware is not filling in this field until the driver has
    already registered MSIX support.

(3) If the ioc_facts reports that the controller is MSIX compatible in the
    capabilities, then the driver will request for multiple irqs.  This count
    is calculated based on the minimum between the online cpus available and
    the ioc->msix_vector_count.  This count is reported to firmware in the
    ioc_init request.

(4) New routines were added _base_free_irq and _base_request_irq, so
    registering and freeing msix vectors were done thru simple function API.

(5) The new routine _base_assign_reply_queues was added to align the msix
    indexes across cpus. This will initialize the array called
    ioc->cpu_msix_table.  This array is looked up on every MPI request so the
    MSIxIndex is set appropriately.

(6) A new shost sysfs attribute was added to report the reply_queue_count.

(7) User needs to set the affinity cpu mask, so the interrupts occur on the
    same cpu that sent the original request.
Signed-off-by: Nagalakshmi Nandigama <nagalakshmi.nandigama@lsi.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

drivers/scsi/mpt2sas/mpt2sas_base.c

@@ -834,25 +834,31 @@ union reply_descriptor {
 static irqreturn_t
 _base_interrupt(int irq, void *bus_id)
 {
+	struct adapter_reply_queue *reply_q = bus_id;
 	union reply_descriptor rd;
 	u32 completed_cmds;
 	u8 request_desript_type;
 	u16 smid;
 	u8 cb_idx;
 	u32 reply;
-	u8 msix_index;
-	struct MPT2SAS_ADAPTER *ioc = bus_id;
+	u8 msix_index = reply_q->msix_index;
+	struct MPT2SAS_ADAPTER *ioc = reply_q->ioc;
 	Mpi2ReplyDescriptorsUnion_t *rpf;
 	u8 rc;
 
 	if (ioc->mask_interrupts)
 		return IRQ_NONE;
 
-	rpf = &ioc->reply_post_free[ioc->reply_post_host_index];
+	if (!atomic_add_unless(&reply_q->busy, 1, 1))
+		return IRQ_NONE;
+
+	rpf = &reply_q->reply_post_free[reply_q->reply_post_host_index];
 	request_desript_type = rpf->Default.ReplyFlags
 	     & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK;
-	if (request_desript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED)
+	if (request_desript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED) {
+		atomic_dec(&reply_q->busy);
 		return IRQ_NONE;
+	}
 
 	completed_cmds = 0;
 	cb_idx = 0xFF;
@@ -861,9 +867,7 @@ _base_interrupt(int irq, void *bus_id)
 		if (rd.u.low == UINT_MAX || rd.u.high == UINT_MAX)
 			goto out;
 		reply = 0;
-		cb_idx = 0xFF;
 		smid = le16_to_cpu(rpf->Default.DescriptorTypeDependent1);
-		msix_index = rpf->Default.MSIxIndex;
 		if (request_desript_type ==
 		    MPI2_RPY_DESCRIPT_FLAGS_ADDRESS_REPLY) {
 			reply = le32_to_cpu
@@ -907,31 +911,85 @@ _base_interrupt(int irq, void *bus_id)
  next:
 
 		rpf->Words = cpu_to_le64(ULLONG_MAX);
-		ioc->reply_post_host_index = (ioc->reply_post_host_index ==
+		reply_q->reply_post_host_index =
+		    (reply_q->reply_post_host_index ==
 		    (ioc->reply_post_queue_depth - 1)) ? 0 :
-		    ioc->reply_post_host_index + 1;
+		    reply_q->reply_post_host_index + 1;
 		request_desript_type =
-		    ioc->reply_post_free[ioc->reply_post_host_index].Default.
-		    ReplyFlags & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK;
+		    reply_q->reply_post_free[reply_q->reply_post_host_index].
+		    Default.ReplyFlags & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK;
 		completed_cmds++;
 		if (request_desript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED)
 			goto out;
-		if (!ioc->reply_post_host_index)
-			rpf = ioc->reply_post_free;
+		if (!reply_q->reply_post_host_index)
+			rpf = reply_q->reply_post_free;
 		else
 			rpf++;
 	} while (1);
 
  out:
 
-	if (!completed_cmds)
+	if (!completed_cmds) {
+		atomic_dec(&reply_q->busy);
 		return IRQ_NONE;
-
+	}
 	wmb();
-	writel(ioc->reply_post_host_index, &ioc->chip->ReplyPostHostIndex);
+	if (ioc->is_warpdrive) {
+		writel(reply_q->reply_post_host_index,
+		ioc->reply_post_host_index[msix_index]);
+		atomic_dec(&reply_q->busy);
+		return IRQ_HANDLED;
+	}
+	writel(reply_q->reply_post_host_index | (msix_index <<
+	    MPI2_RPHI_MSIX_INDEX_SHIFT), &ioc->chip->ReplyPostHostIndex);
+	atomic_dec(&reply_q->busy);
 	return IRQ_HANDLED;
 }
 
+/**
+ * _base_is_controller_msix_enabled - is controller support muli-reply queues
+ * @ioc: per adapter object
+ *
+ */
+static inline int
+_base_is_controller_msix_enabled(struct MPT2SAS_ADAPTER *ioc)
+{
+	return (ioc->facts.IOCCapabilities &
+	    MPI2_IOCFACTS_CAPABILITY_MSI_X_INDEX) && ioc->msix_enable;
+}
+
+/**
+ * mpt2sas_base_flush_reply_queues - flushing the MSIX reply queues
+ * @ioc: per adapter object
+ * Context: ISR conext
+ *
+ * Called when a Task Management request has completed. We want
+ * to flush the other reply queues so all the outstanding IO has been
+ * completed back to OS before we process the TM completetion.
+ *
+ * Return nothing.
+ */
+void
+mpt2sas_base_flush_reply_queues(struct MPT2SAS_ADAPTER *ioc)
+{
+	struct adapter_reply_queue *reply_q;
+
+	/* If MSIX capability is turned off
+	 * then multi-queues are not enabled
+	 */
+	if (!_base_is_controller_msix_enabled(ioc))
+		return;
+
+	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
+		if (ioc->shost_recovery)
+			return;
+		/* TMs are on msix_index == 0 */
+		if (reply_q->msix_index == 0)
+			continue;
+		_base_interrupt(reply_q->vector, (void *)reply_q);
+	}
+}
+
 /**
  * mpt2sas_base_release_callback_handler - clear interrupt callback handler
  * @cb_idx: callback index
@@ -1082,74 +1140,171 @@ _base_config_dma_addressing(struct MPT2SAS_ADAPTER *ioc, struct pci_dev *pdev)
 }
 
 /**
- * _base_save_msix_table - backup msix vector table
+ * _base_check_enable_msix - checks MSIX capabable.
  * @ioc: per adapter object
  *
- * This address an errata where diag reset clears out the table
+ * Check to see if card is capable of MSIX, and set number
+ * of available msix vectors
  */
-static void
-_base_save_msix_table(struct MPT2SAS_ADAPTER *ioc)
+static int
+_base_check_enable_msix(struct MPT2SAS_ADAPTER *ioc)
 {
-	int i;
+	int base;
+	u16 message_control;
 
-	if (!ioc->msix_enable || ioc->msix_table_backup == NULL)
-		return;
 
-	for (i = 0; i < ioc->msix_vector_count; i++)
-		ioc->msix_table_backup[i] = ioc->msix_table[i];
+	base = pci_find_capability(ioc->pdev, PCI_CAP_ID_MSIX);
+	if (!base) {
+		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "msix not "
+		    "supported\n", ioc->name));
+		return -EINVAL;
+	}
+
+	/* get msix vector count */
+	/* NUMA_IO not supported for older controllers */
+	if (ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2004 ||
+	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2008 ||
+	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2108_1 ||
+	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2108_2 ||
+	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2108_3 ||
+	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2116_1 ||
+	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2116_2)
+		ioc->msix_vector_count = 1;
+	else {
+		pci_read_config_word(ioc->pdev, base + 2, &message_control);
+		ioc->msix_vector_count = (message_control & 0x3FF) + 1;
+	}
+	dinitprintk(ioc, printk(MPT2SAS_INFO_FMT "msix is supported, "
+	    "vector_count(%d)\n", ioc->name, ioc->msix_vector_count));
+
+	return 0;
 }
 
 /**
- * _base_restore_msix_table - this restores the msix vector table
+ * _base_free_irq - free irq
  * @ioc: per adapter object
  *
+ * Freeing respective reply_queue from the list.
  */
 static void
-_base_restore_msix_table(struct MPT2SAS_ADAPTER *ioc)
+_base_free_irq(struct MPT2SAS_ADAPTER *ioc)
 {
-	int i;
+	struct adapter_reply_queue *reply_q, *next;
 
-	if (!ioc->msix_enable || ioc->msix_table_backup == NULL)
+	if (list_empty(&ioc->reply_queue_list))
 		return;
 
-	for (i = 0; i < ioc->msix_vector_count; i++)
-		ioc->msix_table[i] = ioc->msix_table_backup[i];
+	list_for_each_entry_safe(reply_q, next, &ioc->reply_queue_list, list) {
+		list_del(&reply_q->list);
+		synchronize_irq(reply_q->vector);
+		free_irq(reply_q->vector, reply_q);
+		kfree(reply_q);
+	}
 }
 
 /**
- * _base_check_enable_msix - checks MSIX capabable.
+ * _base_request_irq - request irq
  * @ioc: per adapter object
+ * @index: msix index into vector table
+ * @vector: irq vector
  *
- * Check to see if card is capable of MSIX, and set number
- * of available msix vectors
+ * Inserting respective reply_queue into the list.
  */
 static int
-_base_check_enable_msix(struct MPT2SAS_ADAPTER *ioc)
+_base_request_irq(struct MPT2SAS_ADAPTER *ioc, u8 index, u32 vector)
 {
-	int base;
-	u16 message_control;
-	u32 msix_table_offset;
+	struct adapter_reply_queue *reply_q;
+	int r;
 
-	base = pci_find_capability(ioc->pdev, PCI_CAP_ID_MSIX);
-	if (!base) {
-		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "msix not "
-		    "supported\n", ioc->name));
-		return -EINVAL;
+	reply_q =  kzalloc(sizeof(struct adapter_reply_queue), GFP_KERNEL);
+	if (!reply_q) {
+		printk(MPT2SAS_ERR_FMT "unable to allocate memory %d!\n",
+		    ioc->name, (int)sizeof(struct adapter_reply_queue));
+		return -ENOMEM;
+	}
+	reply_q->ioc = ioc;
+	reply_q->msix_index = index;
+	reply_q->vector = vector;
+	atomic_set(&reply_q->busy, 0);
+	if (ioc->msix_enable)
+		snprintf(reply_q->name, MPT_NAME_LENGTH, "%s%d-msix%d",
+		    MPT2SAS_DRIVER_NAME, ioc->id, index);
+	else
+		snprintf(reply_q->name, MPT_NAME_LENGTH, "%s%d",
+		    MPT2SAS_DRIVER_NAME, ioc->id);
+	r = request_irq(vector, _base_interrupt, IRQF_SHARED, reply_q->name,
+	    reply_q);
+	if (r) {
+		printk(MPT2SAS_ERR_FMT "unable to allocate interrupt %d!\n",
+		    reply_q->name, vector);
+		kfree(reply_q);
+		return -EBUSY;
 	}
 
-	/* get msix vector count */
-	pci_read_config_word(ioc->pdev, base + 2, &message_control);
-	ioc->msix_vector_count = (message_control & 0x3FF) + 1;
+	INIT_LIST_HEAD(&reply_q->list);
+	list_add_tail(&reply_q->list, &ioc->reply_queue_list);
+	return 0;
+}
 
-	/* get msix table  */
-	pci_read_config_dword(ioc->pdev, base + 4, &msix_table_offset);
-	msix_table_offset &= 0xFFFFFFF8;
-	ioc->msix_table = (u32 *)((void *)ioc->chip + msix_table_offset);
+/**
+ * _base_assign_reply_queues - assigning msix index for each cpu
+ * @ioc: per adapter object
+ *
+ * The enduser would need to set the affinity via /proc/irq/#/smp_affinity
+ *
+ * It would nice if we could call irq_set_affinity, however it is not
+ * an exported symbol
+ */
+static void
+_base_assign_reply_queues(struct MPT2SAS_ADAPTER *ioc)
+{
+	struct adapter_reply_queue *reply_q;
+	int cpu_id;
+	int cpu_grouping, loop, grouping, grouping_mod;
 
-	dinitprintk(ioc, printk(MPT2SAS_INFO_FMT "msix is supported, "
-	    "vector_count(%d), table_offset(0x%08x), table(%p)\n", ioc->name,
-	    ioc->msix_vector_count, msix_table_offset, ioc->msix_table));
-	return 0;
+	if (!_base_is_controller_msix_enabled(ioc))
+		return;
+
+	memset(ioc->cpu_msix_table, 0, ioc->cpu_msix_table_sz);
+	/* when there are more cpus than available msix vectors,
+	 * then group cpus togeather on same irq
+	 */
+	if (ioc->cpu_count > ioc->msix_vector_count) {
+		grouping = ioc->cpu_count / ioc->msix_vector_count;
+		grouping_mod = ioc->cpu_count % ioc->msix_vector_count;
+		if (grouping < 2 || (grouping == 2 && !grouping_mod))
+			cpu_grouping = 2;
+		else if (grouping < 4 || (grouping == 4 && !grouping_mod))
+			cpu_grouping = 4;
+		else if (grouping < 8 || (grouping == 8 && !grouping_mod))
+			cpu_grouping = 8;
+		else
+			cpu_grouping = 16;
+	} else
+		cpu_grouping = 0;
+
+	loop = 0;
+	reply_q = list_entry(ioc->reply_queue_list.next,
+	     struct adapter_reply_queue, list);
+	for_each_online_cpu(cpu_id) {
+		if (!cpu_grouping) {
+			ioc->cpu_msix_table[cpu_id] = reply_q->msix_index;
+			reply_q = list_entry(reply_q->list.next,
+			    struct adapter_reply_queue, list);
+		} else {
+			if (loop < cpu_grouping) {
+				ioc->cpu_msix_table[cpu_id] =
+					reply_q->msix_index;
+				loop++;
+			} else {
+				reply_q = list_entry(reply_q->list.next,
+				    struct adapter_reply_queue, list);
+				ioc->cpu_msix_table[cpu_id] =
+					reply_q->msix_index;
+				loop = 1;
+			}
+		}
+	}
 }
 
 /**
@@ -1162,8 +1317,6 @@ _base_disable_msix(struct MPT2SAS_ADAPTER *ioc)
 {
 	if (ioc->msix_enable) {
 		pci_disable_msix(ioc->pdev);
-		kfree(ioc->msix_table_backup);
-		ioc->msix_table_backup = NULL;
 		ioc->msix_enable = 0;
 	}
 }
@@ -1176,10 +1329,13 @@ _base_disable_msix(struct MPT2SAS_ADAPTER *ioc)
 static int
 _base_enable_msix(struct MPT2SAS_ADAPTER *ioc)
 {
-	struct msix_entry entries;
+	struct msix_entry *entries, *a;
 	int r;
+	int i;
 	u8 try_msix = 0;
 
+	INIT_LIST_HEAD(&ioc->reply_queue_list);
+
 	if (msix_disable == -1 || msix_disable == 0)
 		try_msix = 1;
 
@@ -1189,51 +1345,48 @@ _base_enable_msix(struct MPT2SAS_ADAPTER *ioc)
 	if (_base_check_enable_msix(ioc) != 0)
 		goto try_ioapic;
 
-	ioc->msix_table_backup = kcalloc(ioc->msix_vector_count,
-	    sizeof(u32), GFP_KERNEL);
-	if (!ioc->msix_table_backup) {
-		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "allocation for "
-		    "msix_table_backup failed!!!\n", ioc->name));
+	ioc->reply_queue_count = min_t(u8, ioc->cpu_count,
+	    ioc->msix_vector_count);
+
+	entries = kcalloc(ioc->reply_queue_count, sizeof(struct msix_entry),
+	    GFP_KERNEL);
+	if (!entries) {
+		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "kcalloc "
+		    "failed @ at %s:%d/%s() !!!\n", ioc->name, __FILE__,
+		    __LINE__, __func__));
 		goto try_ioapic;
 	}
 
-	memset(&entries, 0, sizeof(struct msix_entry));
-	r = pci_enable_msix(ioc->pdev, &entries, 1);
+	for (i = 0, a = entries; i < ioc->reply_queue_count; i++, a++)
+		a->entry = i;
+
+	r = pci_enable_msix(ioc->pdev, entries, ioc->reply_queue_count);
 	if (r) {
 		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "pci_enable_msix "
 		    "failed (r=%d) !!!\n", ioc->name, r));
+		kfree(entries);
 		goto try_ioapic;
 	}
 
-	r = request_irq(entries.vector, _base_interrupt, IRQF_SHARED,
-	    ioc->name, ioc);
-	if (r) {
-		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "unable to allocate "
-		    "interrupt %d !!!\n", ioc->name, entries.vector));
-		pci_disable_msix(ioc->pdev);
-		goto try_ioapic;
+	ioc->msix_enable = 1;
+	for (i = 0, a = entries; i < ioc->reply_queue_count; i++, a++) {
+		r = _base_request_irq(ioc, i, a->vector);
+		if (r) {
+			_base_free_irq(ioc);
+			_base_disable_msix(ioc);
+			kfree(entries);
+			goto try_ioapic;
+		}
 	}
 
-	ioc->pci_irq = entries.vector;
-	ioc->msix_enable = 1;
+	kfree(entries);
 	return 0;
 
 /* failback to io_apic interrupt routing */
  try_ioapic:
 
-	r = request_irq(ioc->pdev->irq, _base_interrupt, IRQF_SHARED,
-	    ioc->name, ioc);
-	if (r) {
-		printk(MPT2SAS_ERR_FMT "unable to allocate interrupt %d!\n",
-		    ioc->name, ioc->pdev->irq);
-		r = -EBUSY;
-		goto out_fail;
-	}
+	r = _base_request_irq(ioc, 0, ioc->pdev->irq);
 
-	ioc->pci_irq = ioc->pdev->irq;
-	return 0;
-
- out_fail:
 	return r;
 }
 
@@ -1252,6 +1405,7 @@ mpt2sas_base_map_resources(struct MPT2SAS_ADAPTER *ioc)
 	int i, r = 0;
 	u64 pio_chip = 0;
 	u64 chip_phys = 0;
+	struct adapter_reply_queue *reply_q;
 
 	dinitprintk(ioc, printk(MPT2SAS_INFO_FMT "%s\n",
 	    ioc->name, __func__));
@@ -1314,9 +1468,11 @@ mpt2sas_base_map_resources(struct MPT2SAS_ADAPTER *ioc)
 	if (r)
 		goto out_fail;
 
-	printk(MPT2SAS_INFO_FMT "%s: IRQ %d\n",
-	    ioc->name,  ((ioc->msix_enable) ? "PCI-MSI-X enabled" :
-	    "IO-APIC enabled"), ioc->pci_irq);
+	list_for_each_entry(reply_q, &ioc->reply_queue_list, list)
+		printk(MPT2SAS_INFO_FMT "%s: IRQ %d\n",
+		    reply_q->name,  ((ioc->msix_enable) ? "PCI-MSI-X enabled" :
+		    "IO-APIC enabled"), reply_q->vector);
+
 	printk(MPT2SAS_INFO_FMT "iomem(0x%016llx), mapped(0x%p), size(%d)\n",
 	    ioc->name, (unsigned long long)chip_phys, ioc->chip, memap_sz);
 	printk(MPT2SAS_INFO_FMT "ioport(0x%016llx), size(%d)\n",
@@ -1331,7 +1487,6 @@ mpt2sas_base_map_resources(struct MPT2SAS_ADAPTER *ioc)
 	if (ioc->chip_phys)
 		iounmap(ioc->chip);
 	ioc->chip_phys = 0;
-	ioc->pci_irq = -1;
 	pci_release_selected_regions(ioc->pdev, ioc->bars);
 	pci_disable_pcie_error_reporting(pdev);
 	pci_disable_device(pdev);
@@ -1578,6 +1733,12 @@ static inline void _base_writeq(__u64 b, volatile void __iomem *addr,
 }
 #endif
 
+static inline u8
+_base_get_msix_index(struct MPT2SAS_ADAPTER *ioc)
+{
+	return ioc->cpu_msix_table[smp_processor_id()];
+}
+
 /**
  * mpt2sas_base_put_smid_scsi_io - send SCSI_IO request to firmware
  * @ioc: per adapter object
@@ -1594,7 +1755,7 @@ mpt2sas_base_put_smid_scsi_io(struct MPT2SAS_ADAPTER *ioc, u16 smid, u16 handle)
 
 
 	descriptor.SCSIIO.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO;
-	descriptor.SCSIIO.MSIxIndex = 0; /* TODO */
+	descriptor.SCSIIO.MSIxIndex =  _base_get_msix_index(ioc);
 	descriptor.SCSIIO.SMID = cpu_to_le16(smid);
 	descriptor.SCSIIO.DevHandle = cpu_to_le16(handle);
 	descriptor.SCSIIO.LMID = 0;
@@ -1618,7 +1779,7 @@ mpt2sas_base_put_smid_hi_priority(struct MPT2SAS_ADAPTER *ioc, u16 smid)
 
 	descriptor.HighPriority.RequestFlags =
 	    MPI2_REQ_DESCRIPT_FLAGS_HIGH_PRIORITY;
-	descriptor.HighPriority.MSIxIndex = 0; /* TODO */
+	descriptor.HighPriority.MSIxIndex =  0;
 	descriptor.HighPriority.SMID = cpu_to_le16(smid);
 	descriptor.HighPriority.LMID = 0;
 	descriptor.HighPriority.Reserved1 = 0;
@@ -1640,7 +1801,7 @@ mpt2sas_base_put_smid_default(struct MPT2SAS_ADAPTER *ioc, u16 smid)
 	u64 *request = (u64 *)&descriptor;
 
 	descriptor.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
-	descriptor.Default.MSIxIndex = 0; /* TODO */
+	descriptor.Default.MSIxIndex =  _base_get_msix_index(ioc);
 	descriptor.Default.SMID = cpu_to_le16(smid);
 	descriptor.Default.LMID = 0;
 	descriptor.Default.DescriptorTypeDependent = 0;
@@ -1665,7 +1826,7 @@ mpt2sas_base_put_smid_target_assist(struct MPT2SAS_ADAPTER *ioc, u16 smid,
 
 	descriptor.SCSITarget.RequestFlags =
 	    MPI2_REQ_DESCRIPT_FLAGS_SCSI_TARGET;
-	descriptor.SCSITarget.MSIxIndex = 0; /* TODO */
+	descriptor.SCSITarget.MSIxIndex =  _base_get_msix_index(ioc);
 	descriptor.SCSITarget.SMID = cpu_to_le16(smid);
 	descriptor.SCSITarget.LMID = 0;
 	descriptor.SCSITarget.IoIndex = cpu_to_le16(io_index);
@@ -2172,7 +2333,7 @@ _base_allocate_memory_pools(struct MPT2SAS_ADAPTER *ioc,  int sleep_flag)
 	u16 max_sge_elements;
 	u16 num_of_reply_frames;
 	u16 chains_needed_per_io;
-	u32 sz, total_sz;
+	u32 sz, total_sz, reply_post_free_sz;
 	u32 retry_sz;
 	u16 max_request_credit;
 	int i;
@@ -2499,7 +2660,12 @@ _base_allocate_memory_pools(struct MPT2SAS_ADAPTER *ioc,  int sleep_flag)
 	total_sz += sz;
 
 	/* reply post queue, 16 byte align */
-	sz = ioc->reply_post_queue_depth * sizeof(Mpi2DefaultReplyDescriptor_t);
+	reply_post_free_sz = ioc->reply_post_queue_depth *
+	    sizeof(Mpi2DefaultReplyDescriptor_t);
+	if (_base_is_controller_msix_enabled(ioc))
+		sz = reply_post_free_sz * ioc->reply_queue_count;
+	else
+		sz = reply_post_free_sz;
 	ioc->reply_post_free_dma_pool = pci_pool_create("reply_post_free pool",
 	    ioc->pdev, sz, 16, 0);
 	if (!ioc->reply_post_free_dma_pool) {
@@ -3187,6 +3353,7 @@ _base_get_ioc_facts(struct MPT2SAS_ADAPTER *ioc, int sleep_flag)
 	facts->MaxChainDepth = mpi_reply.MaxChainDepth;
 	facts->WhoInit = mpi_reply.WhoInit;
 	facts->NumberOfPorts = mpi_reply.NumberOfPorts;
+	facts->MaxMSIxVectors = mpi_reply.MaxMSIxVectors;
 	facts->RequestCredit = le16_to_cpu(mpi_reply.RequestCredit);
 	facts->MaxReplyDescriptorPostQueueDepth =
 	    le16_to_cpu(mpi_reply.MaxReplyDescriptorPostQueueDepth);
@@ -3244,7 +3411,8 @@ _base_send_ioc_init(struct MPT2SAS_ADAPTER *ioc, int sleep_flag)
 	mpi_request.MsgVersion = cpu_to_le16(MPI2_VERSION);
 	mpi_request.HeaderVersion = cpu_to_le16(MPI2_HEADER_VERSION);
 
-
+	if (_base_is_controller_msix_enabled(ioc))
+		mpi_request.HostMSIxVectors = ioc->reply_queue_count;
 	mpi_request.SystemRequestFrameSize = cpu_to_le16(ioc->request_sz/4);
 	mpi_request.ReplyDescriptorPostQueueDepth =
 	    cpu_to_le16(ioc->reply_post_queue_depth);
@@ -3513,9 +3681,6 @@ _base_diag_reset(struct MPT2SAS_ADAPTER *ioc, int sleep_flag)
 	u32 hcb_size;
 
 	printk(MPT2SAS_INFO_FMT "sending diag reset !!\n", ioc->name);
-
-	_base_save_msix_table(ioc);
-
 	drsprintk(ioc, printk(MPT2SAS_INFO_FMT "clear interrupts\n",
 	    ioc->name));
 
@@ -3611,7 +3776,6 @@ _base_diag_reset(struct MPT2SAS_ADAPTER *ioc, int sleep_flag)
 		goto out;
 	}
 
-	_base_restore_msix_table(ioc);
 	printk(MPT2SAS_INFO_FMT "diag reset: SUCCESS\n", ioc->name);
 	return 0;
 
@@ -3692,6 +3856,9 @@ _base_make_ioc_operational(struct MPT2SAS_ADAPTER *ioc, int sleep_flag)
 	u16 smid;
 	struct _tr_list *delayed_tr, *delayed_tr_next;
 	u8 hide_flag;
+	struct adapter_reply_queue *reply_q;
+	long reply_post_free;
+	u32 reply_post_free_sz;
 
 	dinitprintk(ioc, printk(MPT2SAS_INFO_FMT "%s\n", ioc->name,
 	    __func__));
@@ -3757,19 +3924,43 @@ _base_make_ioc_operational(struct MPT2SAS_ADAPTER *ioc, int sleep_flag)
 	    ioc->reply_sz)
 		ioc->reply_free[i] = cpu_to_le32(reply_address);
 
+	/* initialize reply queues */
+	_base_assign_reply_queues(ioc);
+
 	/* initialize Reply Post Free Queue */
-	for (i = 0; i < ioc->reply_post_queue_depth; i++)
-		ioc->reply_post_free[i].Words = cpu_to_le64(ULLONG_MAX);
+	reply_post_free = (long)ioc->reply_post_free;
+	reply_post_free_sz = ioc->reply_post_queue_depth *
+	    sizeof(Mpi2DefaultReplyDescriptor_t);
+	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
+		reply_q->reply_post_host_index = 0;
+		reply_q->reply_post_free = (Mpi2ReplyDescriptorsUnion_t *)
+		    reply_post_free;
+		for (i = 0; i < ioc->reply_post_queue_depth; i++)
+			reply_q->reply_post_free[i].Words =
+							cpu_to_le64(ULLONG_MAX);
+		if (!_base_is_controller_msix_enabled(ioc))
+			goto skip_init_reply_post_free_queue;
+		reply_post_free += reply_post_free_sz;
+	}
+ skip_init_reply_post_free_queue:
 
 	r = _base_send_ioc_init(ioc, sleep_flag);
 	if (r)
 		return r;
 
-	/* initialize the index's */
+	/* initialize reply free host index */
 	ioc->reply_free_host_index = ioc->reply_free_queue_depth - 1;
-	ioc->reply_post_host_index = 0;
 	writel(ioc->reply_free_host_index, &ioc->chip->ReplyFreeHostIndex);
-	writel(0, &ioc->chip->ReplyPostHostIndex);
+
+	/* initialize reply post host index */
+	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
+		writel(reply_q->msix_index << MPI2_RPHI_MSIX_INDEX_SHIFT,
+		    &ioc->chip->ReplyPostHostIndex);
+		if (!_base_is_controller_msix_enabled(ioc))
+			goto skip_init_reply_post_host_index;
+	}
+
+ skip_init_reply_post_host_index:
 
 	_base_unmask_interrupts(ioc);
 	r = _base_event_notification(ioc, sleep_flag);
@@ -3820,14 +4011,10 @@ mpt2sas_base_free_resources(struct MPT2SAS_ADAPTER *ioc)
 	ioc->shost_recovery = 1;
 	_base_make_ioc_ready(ioc, CAN_SLEEP, SOFT_RESET);
 	ioc->shost_recovery = 0;
-	if (ioc->pci_irq) {
-		synchronize_irq(pdev->irq);
-		free_irq(ioc->pci_irq, ioc);
-	}
+	_base_free_irq(ioc);
 	_base_disable_msix(ioc);
 	if (ioc->chip_phys)
 		iounmap(ioc->chip);
-	ioc->pci_irq = -1;
 	ioc->chip_phys = 0;
 	pci_release_selected_regions(ioc->pdev, ioc->bars);
 	pci_disable_pcie_error_reporting(pdev);
@@ -3845,14 +4032,50 @@ int
 mpt2sas_base_attach(struct MPT2SAS_ADAPTER *ioc)
 {
 	int r, i;
+	int cpu_id, last_cpu_id = 0;
 
 	dinitprintk(ioc, printk(MPT2SAS_INFO_FMT "%s\n", ioc->name,
 	    __func__));
 
+	/* setup cpu_msix_table */
+	ioc->cpu_count = num_online_cpus();
+	for_each_online_cpu(cpu_id)
+		last_cpu_id = cpu_id;
+	ioc->cpu_msix_table_sz = last_cpu_id + 1;
+	ioc->cpu_msix_table = kzalloc(ioc->cpu_msix_table_sz, GFP_KERNEL);
+	ioc->reply_queue_count = 1;
+	if (!ioc->cpu_msix_table) {
+		dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "allocation for "
+		    "cpu_msix_table failed!!!\n", ioc->name));
+		r = -ENOMEM;
+		goto out_free_resources;
+	}
+
+	if (ioc->is_warpdrive) {
+		ioc->reply_post_host_index = kcalloc(ioc->cpu_msix_table_sz,
+		    sizeof(resource_size_t *), GFP_KERNEL);
+		if (!ioc->reply_post_host_index) {
+			dfailprintk(ioc, printk(MPT2SAS_INFO_FMT "allocation "
+				"for cpu_msix_table failed!!!\n", ioc->name));
+			r = -ENOMEM;
+			goto out_free_resources;
+		}
+	}
+
 	r = mpt2sas_base_map_resources(ioc);
 	if (r)
 		return r;
 
+	if (ioc->is_warpdrive) {
+		ioc->reply_post_host_index[0] =
+		    (resource_size_t *)&ioc->chip->ReplyPostHostIndex;
+
+		for (i = 1; i < ioc->cpu_msix_table_sz; i++)
+			ioc->reply_post_host_index[i] = (resource_size_t *)
+			((u8 *)&ioc->chip->Doorbell + (0x4000 + ((i - 1)
+			* 4)));
+	}
+
 	pci_set_drvdata(ioc->pdev, ioc->shost);
 	r = _base_get_ioc_facts(ioc, CAN_SLEEP);
 	if (r)
@@ -3973,6 +4196,9 @@ mpt2sas_base_attach(struct MPT2SAS_ADAPTER *ioc)
 	mpt2sas_base_free_resources(ioc);
 	_base_release_memory_pools(ioc);
 	pci_set_drvdata(ioc->pdev, NULL);
+	kfree(ioc->cpu_msix_table);
+	if (ioc->is_warpdrive)
+		kfree(ioc->reply_post_host_index);
 	kfree(ioc->pd_handles);
 	kfree(ioc->tm_cmds.reply);
 	kfree(ioc->transport_cmds.reply);
@@ -4010,6 +4236,9 @@ mpt2sas_base_detach(struct MPT2SAS_ADAPTER *ioc)
 	mpt2sas_base_free_resources(ioc);
 	_base_release_memory_pools(ioc);
 	pci_set_drvdata(ioc->pdev, NULL);
+	kfree(ioc->cpu_msix_table);
+	if (ioc->is_warpdrive)
+		kfree(ioc->reply_post_host_index);
 	kfree(ioc->pd_handles);
 	kfree(ioc->pfacts);
 	kfree(ioc->ctl_cmds.reply);

drivers/scsi/mpt2sas/mpt2sas_base.h

@@ -544,6 +544,28 @@ struct _tr_list {
 
 typedef void (*MPT_ADD_SGE)(void *paddr, u32 flags_length, dma_addr_t dma_addr);
 
+/**
+ * struct adapter_reply_queue - the reply queue struct
+ * @ioc: per adapter object
+ * @msix_index: msix index into vector table
+ * @vector: irq vector
+ * @reply_post_host_index: head index in the pool where FW completes IO
+ * @reply_post_free: reply post base virt address
+ * @name: the name registered to request_irq()
+ * @busy: isr is actively processing replies on another cpu
+ * @list: this list
+*/
+struct adapter_reply_queue {
+	struct MPT2SAS_ADAPTER	*ioc;
+	u8			msix_index;
+	unsigned int		vector;
+	u32			reply_post_host_index;
+	Mpi2ReplyDescriptorsUnion_t *reply_post_free;
+	char			name[MPT_NAME_LENGTH];
+	atomic_t		busy;
+	struct list_head	list;
+};
+
 /* IOC Facts and Port Facts converted from little endian to cpu */
 union mpi2_version_union {
 	MPI2_VERSION_STRUCT		Struct;
@@ -606,7 +628,7 @@ enum mutex_type {
  * @list: ioc_list
  * @shost: shost object
  * @id: unique adapter id
- * @pci_irq: irq number
+ * @cpu_count: number online cpus
  * @name: generic ioc string
  * @tmp_string: tmp string used for logging
  * @pdev: pci pdev object
@@ -636,8 +658,8 @@ enum mutex_type {
  * @wait_for_port_enable_to_complete:
  * @msix_enable: flag indicating msix is enabled
  * @msix_vector_count: number msix vectors
- * @msix_table: virt address to the msix table
- * @msix_table_backup: backup msix table
+ * @cpu_msix_table: table for mapping cpus to msix index
+ * @cpu_msix_table_sz: table size
  * @scsi_io_cb_idx: shost generated commands
  * @tm_cb_idx: task management commands
  * @scsih_cb_idx: scsih internal commands
@@ -728,7 +750,8 @@ enum mutex_type {
  * @reply_post_queue_depth: reply post queue depth
  * @reply_post_free: pool for reply post (64bit descriptor)
  * @reply_post_free_dma:
- * @reply_post_free_dma_pool:
+ * @reply_queue_count: number of reply queue's
+ * @reply_queue_list: link list contaning the reply queue info
  * @reply_post_host_index: head index in the pool where FW completes IO
  * @delayed_tr_list: target reset link list
  * @delayed_tr_volume_list: volume target reset link list
@@ -737,7 +760,7 @@ struct MPT2SAS_ADAPTER {
 	struct list_head list;
 	struct Scsi_Host *shost;
 	u8		id;
-	u32		pci_irq;
+	int		cpu_count;
 	char		name[MPT_NAME_LENGTH];
 	char		tmp_string[MPT_STRING_LENGTH];
 	struct pci_dev	*pdev;
@@ -779,8 +802,9 @@ struct MPT2SAS_ADAPTER {
 
 	u8		msix_enable;
 	u16		msix_vector_count;
-	u32		*msix_table;
-	u32		*msix_table_backup;
+	u8		*cpu_msix_table;
+	resource_size_t	**reply_post_host_index;
+	u16		cpu_msix_table_sz;
 	u32		ioc_reset_count;
 
 	/* internal commands, callback index */
@@ -911,7 +935,8 @@ struct MPT2SAS_ADAPTER {
 	Mpi2ReplyDescriptorsUnion_t *reply_post_free;
 	dma_addr_t	reply_post_free_dma;
 	struct dma_pool *reply_post_free_dma_pool;
-	u32		reply_post_host_index;
+	u8		reply_queue_count;
+	struct list_head reply_queue_list;
 
 	struct list_head delayed_tr_list;
 	struct list_head delayed_tr_volume_list;
@@ -955,6 +980,7 @@ void *mpt2sas_base_get_sense_buffer(struct MPT2SAS_ADAPTER *ioc, u16 smid);
 void mpt2sas_base_build_zero_len_sge(struct MPT2SAS_ADAPTER *ioc, void *paddr);
 __le32 mpt2sas_base_get_sense_buffer_dma(struct MPT2SAS_ADAPTER *ioc,
     u16 smid);
+void mpt2sas_base_flush_reply_queues(struct MPT2SAS_ADAPTER *ioc);
 
 /* hi-priority queue */
 u16 mpt2sas_base_get_smid_hpr(struct MPT2SAS_ADAPTER *ioc, u8 cb_idx);

drivers/scsi/mpt2sas/mpt2sas_ctl.c

@@ -2705,6 +2705,33 @@ _ctl_ioc_reset_count_show(struct device *cdev, struct device_attribute *attr,
 static DEVICE_ATTR(ioc_reset_count, S_IRUGO,
     _ctl_ioc_reset_count_show, NULL);
 
+/**
+ * _ctl_ioc_reply_queue_count_show - number of reply queues
+ * @cdev - pointer to embedded class device
+ * @buf - the buffer returned
+ *
+ * This is number of reply queues
+ *
+ * A sysfs 'read-only' shost attribute.
+ */
+static ssize_t
+_ctl_ioc_reply_queue_count_show(struct device *cdev,
+	 struct device_attribute *attr, char *buf)
+{
+	u8 reply_queue_count;
+	struct Scsi_Host *shost = class_to_shost(cdev);
+	struct MPT2SAS_ADAPTER *ioc = shost_priv(shost);
+
+	if ((ioc->facts.IOCCapabilities &
+	    MPI2_IOCFACTS_CAPABILITY_MSI_X_INDEX) && ioc->msix_enable)
+		reply_queue_count = ioc->reply_queue_count;
+	else
+		reply_queue_count = 1;
+	return snprintf(buf, PAGE_SIZE, "%d\n", reply_queue_count);
+}
+static DEVICE_ATTR(reply_queue_count, S_IRUGO,
+	 _ctl_ioc_reply_queue_count_show, NULL);
+
 struct DIAG_BUFFER_START {
 	__le32 Size;
 	__le32 DiagVersion;
@@ -2915,6 +2942,7 @@ struct device_attribute *mpt2sas_host_attrs[] = {
 	&dev_attr_host_trace_buffer_size,
 	&dev_attr_host_trace_buffer,
 	&dev_attr_host_trace_buffer_enable,
+	&dev_attr_reply_queue_count,
 	NULL,
 };
 

drivers/scsi/mpt2sas/mpt2sas_scsih.c

@@ -2162,6 +2162,7 @@ _scsih_tm_done(struct MPT2SAS_ADAPTER *ioc, u16 smid, u8 msix_index, u32 reply)
 		return 1;
 	if (ioc->tm_cmds.smid != smid)
 		return 1;
+	mpt2sas_base_flush_reply_queues(ioc);
 	ioc->tm_cmds.status |= MPT2_CMD_COMPLETE;
 	mpi_reply =  mpt2sas_base_get_reply_virt_addr(ioc, reply);
 	if (mpi_reply) {