[SCSI] megaraid_sas: Add multiple MSI-X vector/multiple reply queue support

Signed-off-by: Adam Radford <aradford@gmail.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

drivers/scsi/megaraid/megaraid_sas.h

@@ -757,6 +757,7 @@ struct megasas_ctrl_info {
 #define MEGASAS_INT_CMDS			32
 #define MEGASAS_SKINNY_INT_CMDS			5
 
+#define MEGASAS_MAX_MSIX_QUEUES			16
 /*
  * FW can accept both 32 and 64 bit SGLs. We want to allocate 32/64 bit
  * SGLs based on the size of dma_addr_t
@@ -1278,6 +1279,11 @@ struct megasas_aen_event {
 	struct megasas_instance *instance;
 };
 
+struct megasas_irq_context {
+	struct megasas_instance *instance;
+	u32 MSIxIndex;
+};
+
 struct megasas_instance {
 
 	u32 *producer;
@@ -1351,8 +1357,9 @@ struct megasas_instance {
 
 	/* Ptr to hba specific information */
 	void *ctrl_context;
-	u8	msi_flag;
-	struct msix_entry msixentry;
+	unsigned int msix_vectors;
+	struct msix_entry msixentry[MEGASAS_MAX_MSIX_QUEUES];
+	struct megasas_irq_context irq_context[MEGASAS_MAX_MSIX_QUEUES];
 	u64 map_id;
 	struct megasas_cmd *map_update_cmd;
 	unsigned long bar;

drivers/scsi/megaraid/megaraid_sas_base.c

@@ -2536,7 +2536,7 @@ megasas_deplete_reply_queue(struct megasas_instance *instance,
 						instance->reg_set)
 						) == 0) {
 		/* Hardware may not set outbound_intr_status in MSI-X mode */
-		if (!instance->msi_flag)
+		if (!instance->msix_vectors)
 			return IRQ_NONE;
 	}
 
@@ -2594,16 +2594,14 @@ megasas_deplete_reply_queue(struct megasas_instance *instance,
  */
 static irqreturn_t megasas_isr(int irq, void *devp)
 {
-	struct megasas_instance *instance;
+	struct megasas_irq_context *irq_context = devp;
+	struct megasas_instance *instance = irq_context->instance;
 	unsigned long flags;
 	irqreturn_t	rc;
 
-	if (atomic_read(
-		&(((struct megasas_instance *)devp)->fw_reset_no_pci_access)))
+	if (atomic_read(&instance->fw_reset_no_pci_access))
 		return IRQ_HANDLED;
 
-	instance = (struct megasas_instance *)devp;
-
 	spin_lock_irqsave(&instance->hba_lock, flags);
 	rc =  megasas_deplete_reply_queue(instance, DID_OK);
 	spin_unlock_irqrestore(&instance->hba_lock, flags);
@@ -3488,6 +3486,7 @@ static int megasas_init_fw(struct megasas_instance *instance)
 	struct megasas_register_set __iomem *reg_set;
 	struct megasas_ctrl_info *ctrl_info;
 	unsigned long bar_list;
+	int i;
 
 	/* Find first memory bar */
 	bar_list = pci_select_bars(instance->pdev, IORESOURCE_MEM);
@@ -3541,9 +3540,33 @@ static int megasas_init_fw(struct megasas_instance *instance)
 	/* Check if MSI-X is supported while in ready state */
 	msix_enable = (instance->instancet->read_fw_status_reg(reg_set) &
 		       0x4000000) >> 0x1a;
-	if (msix_enable && !msix_disable &&
-	    !pci_enable_msix(instance->pdev, &instance->msixentry, 1))
-		instance->msi_flag = 1;
+	if (msix_enable && !msix_disable) {
+		/* Check max MSI-X vectors */
+		if ((instance->pdev->device == PCI_DEVICE_ID_LSI_FUSION) ||
+		    (instance->pdev->device == PCI_DEVICE_ID_LSI_INVADER)) {
+			instance->msix_vectors = (readl(&instance->reg_set->
+							outbound_scratch_pad_2
+							  ) & 0x1F) + 1;
+		} else
+			instance->msix_vectors = 1;
+		/* Don't bother allocating more MSI-X vectors than cpus */
+		instance->msix_vectors = min(instance->msix_vectors,
+					     (unsigned int)num_online_cpus());
+		for (i = 0; i < instance->msix_vectors; i++)
+			instance->msixentry[i].entry = i;
+		i = pci_enable_msix(instance->pdev, instance->msixentry,
+				    instance->msix_vectors);
+		if (i >= 0) {
+			if (i) {
+				if (!pci_enable_msix(instance->pdev,
+						     instance->msixentry, i))
+					instance->msix_vectors = i;
+				else
+					instance->msix_vectors = 0;
+			}
+		} else
+			instance->msix_vectors = 0;
+	}
 
 	/* Get operational params, sge flags, send init cmd to controller */
 	if (instance->instancet->init_adapter(instance))
@@ -3958,7 +3981,7 @@ megasas_set_dma_mask(struct pci_dev *pdev)
 static int __devinit
 megasas_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 {
-	int rval, pos;
+	int rval, pos, i, j;
 	struct Scsi_Host *host;
 	struct megasas_instance *instance;
 	u16 control = 0;
@@ -4126,11 +4149,32 @@ megasas_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	/*
 	 * Register IRQ
 	 */
-	if (request_irq(instance->msi_flag ? instance->msixentry.vector :
-			pdev->irq, instance->instancet->service_isr,
-			IRQF_SHARED, "megasas", instance)) {
-		printk(KERN_DEBUG "megasas: Failed to register IRQ\n");
-		goto fail_irq;
+	if (instance->msix_vectors) {
+		for (i = 0 ; i < instance->msix_vectors; i++) {
+			instance->irq_context[i].instance = instance;
+			instance->irq_context[i].MSIxIndex = i;
+			if (request_irq(instance->msixentry[i].vector,
+					instance->instancet->service_isr, 0,
+					"megasas",
+					&instance->irq_context[i])) {
+				printk(KERN_DEBUG "megasas: Failed to "
+				       "register IRQ for vector %d.\n", i);
+				for (j = 0 ; j < i ; j++)
+					free_irq(
+						instance->msixentry[j].vector,
+						&instance->irq_context[j]);
+				goto fail_irq;
+			}
+		}
+	} else {
+		instance->irq_context[0].instance = instance;
+		instance->irq_context[0].MSIxIndex = 0;
+		if (request_irq(pdev->irq, instance->instancet->service_isr,
+				IRQF_SHARED, "megasas",
+				&instance->irq_context[0])) {
+			printk(KERN_DEBUG "megasas: Failed to register IRQ\n");
+			goto fail_irq;
+		}
 	}
 
 	instance->instancet->enable_intr(instance->reg_set);
@@ -4174,8 +4218,12 @@ megasas_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 
 	pci_set_drvdata(pdev, NULL);
 	instance->instancet->disable_intr(instance->reg_set);
-	free_irq(instance->msi_flag ? instance->msixentry.vector :
-		 instance->pdev->irq, instance);
+	if (instance->msix_vectors)
+		for (i = 0 ; i < instance->msix_vectors; i++)
+			free_irq(instance->msixentry[i].vector,
+				 &instance->irq_context[i]);
+	else
+		free_irq(instance->pdev->irq, &instance->irq_context[0]);
 fail_irq:
 	if ((instance->pdev->device == PCI_DEVICE_ID_LSI_FUSION) ||
 	    (instance->pdev->device == PCI_DEVICE_ID_LSI_INVADER))
@@ -4183,7 +4231,7 @@ megasas_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	else
 		megasas_release_mfi(instance);
       fail_init_mfi:
-	if (instance->msi_flag)
+	if (instance->msix_vectors)
 		pci_disable_msix(instance->pdev);
       fail_alloc_dma_buf:
 	if (instance->evt_detail)
@@ -4299,6 +4347,7 @@ megasas_suspend(struct pci_dev *pdev, pm_message_t state)
 {
 	struct Scsi_Host *host;
 	struct megasas_instance *instance;
+	int i;
 
 	instance = pci_get_drvdata(pdev);
 	host = instance->host;
@@ -4322,9 +4371,14 @@ megasas_suspend(struct pci_dev *pdev, pm_message_t state)
 
 	pci_set_drvdata(instance->pdev, instance);
 	instance->instancet->disable_intr(instance->reg_set);
-	free_irq(instance->msi_flag ? instance->msixentry.vector :
-		 instance->pdev->irq, instance);
-	if (instance->msi_flag)
+
+	if (instance->msix_vectors)
+		for (i = 0 ; i < instance->msix_vectors; i++)
+			free_irq(instance->msixentry[i].vector,
+				 &instance->irq_context[i]);
+	else
+		free_irq(instance->pdev->irq, &instance->irq_context[0]);
+	if (instance->msix_vectors)
 		pci_disable_msix(instance->pdev);
 
 	pci_save_state(pdev);
@@ -4342,7 +4396,7 @@ megasas_suspend(struct pci_dev *pdev, pm_message_t state)
 static int
 megasas_resume(struct pci_dev *pdev)
 {
-	int rval;
+	int rval, i, j;
 	struct Scsi_Host *host;
 	struct megasas_instance *instance;
 
@@ -4380,8 +4434,9 @@ megasas_resume(struct pci_dev *pdev)
 		goto fail_ready_state;
 
 	/* Now re-enable MSI-X */
-	if (instance->msi_flag)
-		pci_enable_msix(instance->pdev, &instance->msixentry, 1);
+	if (instance->msix_vectors)
+		pci_enable_msix(instance->pdev, instance->msixentry,
+				instance->msix_vectors);
 
 	switch (instance->pdev->device) {
 	case PCI_DEVICE_ID_LSI_FUSION:
@@ -4411,11 +4466,32 @@ megasas_resume(struct pci_dev *pdev)
 	/*
 	 * Register IRQ
 	 */
-	if (request_irq(instance->msi_flag ? instance->msixentry.vector :
-			pdev->irq, instance->instancet->service_isr,
-			IRQF_SHARED, "megasas", instance)) {
-		printk(KERN_ERR "megasas: Failed to register IRQ\n");
-		goto fail_irq;
+	if (instance->msix_vectors) {
+		for (i = 0 ; i < instance->msix_vectors; i++) {
+			instance->irq_context[i].instance = instance;
+			instance->irq_context[i].MSIxIndex = i;
+			if (request_irq(instance->msixentry[i].vector,
+					instance->instancet->service_isr, 0,
+					"megasas",
+					&instance->irq_context[i])) {
+				printk(KERN_DEBUG "megasas: Failed to "
+				       "register IRQ for vector %d.\n", i);
+				for (j = 0 ; j < i ; j++)
+					free_irq(
+						instance->msixentry[j].vector,
+						&instance->irq_context[j]);
+				goto fail_irq;
+			}
+		}
+	} else {
+		instance->irq_context[0].instance = instance;
+		instance->irq_context[0].MSIxIndex = 0;
+		if (request_irq(pdev->irq, instance->instancet->service_isr,
+				IRQF_SHARED, "megasas",
+				&instance->irq_context[0])) {
+			printk(KERN_DEBUG "megasas: Failed to register IRQ\n");
+			goto fail_irq;
+		}
 	}
 
 	instance->instancet->enable_intr(instance->reg_set);
@@ -4512,9 +4588,13 @@ static void __devexit megasas_detach_one(struct pci_dev *pdev)
 
 	instance->instancet->disable_intr(instance->reg_set);
 
-	free_irq(instance->msi_flag ? instance->msixentry.vector :
-		 instance->pdev->irq, instance);
-	if (instance->msi_flag)
+	if (instance->msix_vectors)
+		for (i = 0 ; i < instance->msix_vectors; i++)
+			free_irq(instance->msixentry[i].vector,
+				 &instance->irq_context[i]);
+	else
+		free_irq(instance->pdev->irq, &instance->irq_context[0]);
+	if (instance->msix_vectors)
 		pci_disable_msix(instance->pdev);
 
 	switch (instance->pdev->device) {
@@ -4560,14 +4640,20 @@ static void __devexit megasas_detach_one(struct pci_dev *pdev)
  */
 static void megasas_shutdown(struct pci_dev *pdev)
 {
+	int i;
 	struct megasas_instance *instance = pci_get_drvdata(pdev);
+
 	instance->unload = 1;
 	megasas_flush_cache(instance);
 	megasas_shutdown_controller(instance, MR_DCMD_CTRL_SHUTDOWN);
 	instance->instancet->disable_intr(instance->reg_set);
-	free_irq(instance->msi_flag ? instance->msixentry.vector :
-		 instance->pdev->irq, instance);
-	if (instance->msi_flag)
+	if (instance->msix_vectors)
+		for (i = 0 ; i < instance->msix_vectors; i++)
+			free_irq(instance->msixentry[i].vector,
+				 &instance->irq_context[i]);
+	else
+		free_irq(instance->pdev->irq, &instance->irq_context[0]);
+	if (instance->msix_vectors)
 		pci_disable_msix(instance->pdev);
 }
 

drivers/scsi/megaraid/megaraid_sas_fusion.c

@@ -385,7 +385,7 @@ static int megasas_create_frame_pool_fusion(struct megasas_instance *instance)
 int
 megasas_alloc_cmds_fusion(struct megasas_instance *instance)
 {
-	int i, j;
+	int i, j, count;
 	u32 max_cmd, io_frames_sz;
 	struct fusion_context *fusion;
 	struct megasas_cmd_fusion *cmd;
@@ -409,9 +409,10 @@ megasas_alloc_cmds_fusion(struct megasas_instance *instance)
 		goto fail_req_desc;
 	}
 
+	count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
 	fusion->reply_frames_desc_pool =
 		pci_pool_create("reply_frames pool", instance->pdev,
-				fusion->reply_alloc_sz, 16, 0);
+				fusion->reply_alloc_sz * count, 16, 0);
 
 	if (!fusion->reply_frames_desc_pool) {
 		printk(KERN_ERR "megasas; Could not allocate memory for "
@@ -430,7 +431,7 @@ megasas_alloc_cmds_fusion(struct megasas_instance *instance)
 	}
 
 	reply_desc = fusion->reply_frames_desc;
-	for (i = 0; i < fusion->reply_q_depth; i++, reply_desc++)
+	for (i = 0; i < fusion->reply_q_depth * count; i++, reply_desc++)
 		reply_desc->Words = ULLONG_MAX;
 
 	io_frames_sz = fusion->io_frames_alloc_sz;
@@ -633,7 +634,9 @@ megasas_ioc_init_fusion(struct megasas_instance *instance)
 		fusion->reply_frames_desc_phys;
 	IOCInitMessage->SystemRequestFrameBaseAddress =
 		fusion->io_request_frames_phys;
-
+	/* Set to 0 for none or 1 MSI-X vectors */
+	IOCInitMessage->HostMSIxVectors = (instance->msix_vectors > 0 ?
+					   instance->msix_vectors : 0);
 	init_frame = (struct megasas_init_frame *)cmd->frame;
 	memset(init_frame, 0, MEGAMFI_FRAME_SIZE);
 
@@ -877,7 +880,7 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
 	struct megasas_register_set __iomem *reg_set;
 	struct fusion_context *fusion;
 	u32 max_cmd;
-	int i = 0;
+	int i = 0, count;
 
 	fusion = instance->ctrl_context;
 
@@ -929,7 +932,9 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
 		(MEGA_MPI2_RAID_DEFAULT_IO_FRAME_SIZE -
 		 sizeof(union MPI2_SGE_IO_UNION))/16;
 
-	fusion->last_reply_idx = 0;
+	count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
+	for (i = 0 ; i < count; i++)
+		fusion->last_reply_idx[i] = 0;
 
 	/*
 	 * Allocate memory for descriptors
@@ -1421,6 +1426,12 @@ megasas_build_ldio_fusion(struct megasas_instance *instance,
 			fp_possible = io_info.fpOkForIo;
 	}
 
+	/* Use smp_processor_id() for now until cmd->request->cpu is CPU
+	   id by default, not CPU group id, otherwise all MSI-X queues won't
+	   be utilized */
+	cmd->request_desc->SCSIIO.MSIxIndex = instance->msix_vectors ?
+		smp_processor_id() % instance->msix_vectors : 0;
+
 	if (fp_possible) {
 		megasas_set_pd_lba(io_request, scp->cmd_len, &io_info, scp,
 				   local_map_ptr, start_lba_lo);
@@ -1691,7 +1702,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,
  * Completes all commands that is in reply descriptor queue
  */
 int
-complete_cmd_fusion(struct megasas_instance *instance)
+complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex)
 {
 	union MPI2_REPLY_DESCRIPTORS_UNION *desc;
 	struct MPI2_SCSI_IO_SUCCESS_REPLY_DESCRIPTOR *reply_desc;
@@ -1711,7 +1722,9 @@ complete_cmd_fusion(struct megasas_instance *instance)
 		return IRQ_HANDLED;
 
 	desc = fusion->reply_frames_desc;
-	desc += fusion->last_reply_idx;
+	desc += ((MSIxIndex * fusion->reply_alloc_sz)/
+		 sizeof(union MPI2_REPLY_DESCRIPTORS_UNION)) +
+		fusion->last_reply_idx[MSIxIndex];
 
 	reply_desc = (struct MPI2_SCSI_IO_SUCCESS_REPLY_DESCRIPTOR *)desc;
 
@@ -1784,16 +1797,19 @@ complete_cmd_fusion(struct megasas_instance *instance)
 			break;
 		}
 
-		fusion->last_reply_idx++;
-		if (fusion->last_reply_idx >= fusion->reply_q_depth)
-			fusion->last_reply_idx = 0;
+		fusion->last_reply_idx[MSIxIndex]++;
+		if (fusion->last_reply_idx[MSIxIndex] >=
+		    fusion->reply_q_depth)
+			fusion->last_reply_idx[MSIxIndex] = 0;
 
 		desc->Words = ULLONG_MAX;
 		num_completed++;
 
 		/* Get the next reply descriptor */
-		if (!fusion->last_reply_idx)
-			desc = fusion->reply_frames_desc;
+		if (!fusion->last_reply_idx[MSIxIndex])
+			desc = fusion->reply_frames_desc +
+				((MSIxIndex * fusion->reply_alloc_sz)/
+				 sizeof(union MPI2_REPLY_DESCRIPTORS_UNION));
 		else
 			desc++;
 
@@ -1813,7 +1829,7 @@ complete_cmd_fusion(struct megasas_instance *instance)
 		return IRQ_NONE;
 
 	wmb();
-	writel(fusion->last_reply_idx,
+	writel((MSIxIndex << 24) | fusion->last_reply_idx[MSIxIndex],
 	       &instance->reg_set->reply_post_host_index);
 	megasas_check_and_restore_queue_depth(instance);
 	return IRQ_HANDLED;
@@ -1831,6 +1847,9 @@ megasas_complete_cmd_dpc_fusion(unsigned long instance_addr)
 	struct megasas_instance *instance =
 		(struct megasas_instance *)instance_addr;
 	unsigned long flags;
+	u32 count, MSIxIndex;
+
+	count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
 
 	/* If we have already declared adapter dead, donot complete cmds */
 	spin_lock_irqsave(&instance->hba_lock, flags);
@@ -1841,7 +1860,8 @@ megasas_complete_cmd_dpc_fusion(unsigned long instance_addr)
 	spin_unlock_irqrestore(&instance->hba_lock, flags);
 
 	spin_lock_irqsave(&instance->completion_lock, flags);
-	complete_cmd_fusion(instance);
+	for (MSIxIndex = 0 ; MSIxIndex < count; MSIxIndex++)
+		complete_cmd_fusion(instance, MSIxIndex);
 	spin_unlock_irqrestore(&instance->completion_lock, flags);
 }
 
@@ -1850,10 +1870,11 @@ megasas_complete_cmd_dpc_fusion(unsigned long instance_addr)
  */
 irqreturn_t megasas_isr_fusion(int irq, void *devp)
 {
-	struct megasas_instance *instance = (struct megasas_instance *)devp;
+	struct megasas_irq_context *irq_context = devp;
+	struct megasas_instance *instance = irq_context->instance;
 	u32 mfiStatus, fw_state;
 
-	if (!instance->msi_flag) {
+	if (!instance->msix_vectors) {
 		mfiStatus = instance->instancet->clear_intr(instance->reg_set);
 		if (!mfiStatus)
 			return IRQ_NONE;
@@ -1865,7 +1886,7 @@ irqreturn_t megasas_isr_fusion(int irq, void *devp)
 		return IRQ_HANDLED;
 	}
 
-	if (!complete_cmd_fusion(instance)) {
+	if (!complete_cmd_fusion(instance, irq_context->MSIxIndex)) {
 		instance->instancet->clear_intr(instance->reg_set);
 		/* If we didn't complete any commands, check for FW fault */
 		fw_state = instance->instancet->read_fw_status_reg(
@@ -2081,14 +2102,16 @@ int megasas_wait_for_outstanding_fusion(struct megasas_instance *instance)
 
 void  megasas_reset_reply_desc(struct megasas_instance *instance)
 {
-	int i;
+	int i, count;
 	struct fusion_context *fusion;
 	union MPI2_REPLY_DESCRIPTORS_UNION *reply_desc;
 
 	fusion = instance->ctrl_context;
-	fusion->last_reply_idx = 0;
+	count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
+	for (i = 0 ; i < count ; i++)
+		fusion->last_reply_idx[i] = 0;
 	reply_desc = fusion->reply_frames_desc;
-	for (i = 0 ; i < fusion->reply_q_depth; i++, reply_desc++)
+	for (i = 0 ; i < fusion->reply_q_depth * count; i++, reply_desc++)
 		reply_desc->Words = ULLONG_MAX;
 }
 

drivers/scsi/megaraid/megaraid_sas_fusion.h

@@ -43,6 +43,7 @@
 #define HOST_DIAG_WRITE_ENABLE			    0x80
 #define HOST_DIAG_RESET_ADAPTER			    0x4
 #define MEGASAS_FUSION_MAX_RESET_TRIES		    3
+#define MAX_MSIX_QUEUES_FUSION			    16
 
 /* Invader defines */
 #define MPI2_TYPE_CUDA				    0x2
@@ -673,7 +674,7 @@ struct fusion_context {
 	union MPI2_REPLY_DESCRIPTORS_UNION *reply_frames_desc;
 	struct dma_pool *reply_frames_desc_pool;
 
-	u16 last_reply_idx;
+	u16 last_reply_idx[MAX_MSIX_QUEUES_FUSION];
 
 	u32 reply_q_depth;
 	u32 request_alloc_sz;