Merge patch series "Add poll support for hisi_sas v3 hw"

chenxiang <chenxiang66@hisilicon.com> says:

To support IO_URING IOPOLL support for hisi_sas, we need to:

 - Add and fill mq_poll interface to poll queue;

 - Ensure internal I/Os (including internal abort I/Os) are delivered and
   completed through non-iopoll queue (queue 0);

Sending internal abort commands to non-poll queue actually requires to
sending the abort command to every queue. This carries a a risk. Make iopoll
support module parameter "experimental".

I have tested performance on v3 hw with different modes as follows.  4K
READs and 4K WRITEs both see an improvement when enabling poll mode:

			4K READ	    4K RANDREAD	    4K WRITE	4K RANDWRITE
interrupt + libaio	1770k	    1316k	    1197k	831k
interrupt + io_uring	1848k	    1390k	    1238k	857k
iopoll + io_uring	2117k	    1364k	    1874k	849k

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

3 files changed, 167 insertions, 36 deletions

diff --git a/drivers/scsi/hisi_sas/hisi_sas.h b/drivers/scsi/hisi_sas/hisi_sas.h
index 6f8a52a1b808..3a5fc3658d3a 100644
--- a/drivers/scsi/hisi_sas/hisi_sas.h
+++ b/drivers/scsi/hisi_sas/hisi_sas.h
@@ -207,6 +207,7 @@ struct hisi_sas_cq {
 	int	rd_point;
 	int	id;
 	int	irq_no;
+	spinlock_t poll_lock;
 };
 
 struct hisi_sas_dq {
@@ -484,6 +485,8 @@ struct hisi_hba {
 	struct dentry *debugfs_dump_dentry;
 	struct dentry *debugfs_bist_dentry;
 	struct dentry *debugfs_fifo_dentry;
+
+	int iopoll_q_cnt;
 };
 
 /* Generic HW DMA host memory structures */
@@ -657,12 +660,13 @@ extern void hisi_sas_slot_task_free(struct hisi_hba *hisi_hba,
 extern void hisi_sas_init_mem(struct hisi_hba *hisi_hba);
 extern void hisi_sas_rst_work_handler(struct work_struct *work);
 extern void hisi_sas_sync_rst_work_handler(struct work_struct *work);
-extern void hisi_sas_sync_irqs(struct hisi_hba *hisi_hba);
 extern void hisi_sas_phy_oob_ready(struct hisi_hba *hisi_hba, int phy_no);
 extern bool hisi_sas_notify_phy_event(struct hisi_sas_phy *phy,
 				enum hisi_sas_phy_event event);
 extern void hisi_sas_release_tasks(struct hisi_hba *hisi_hba);
 extern u8 hisi_sas_get_prog_phy_linkrate_mask(enum sas_linkrate max);
+extern void hisi_sas_sync_cqs(struct hisi_hba *hisi_hba);
+extern void hisi_sas_sync_poll_cqs(struct hisi_hba *hisi_hba);
 extern void hisi_sas_controller_reset_prepare(struct hisi_hba *hisi_hba);
 extern void hisi_sas_controller_reset_done(struct hisi_hba *hisi_hba);
 #endif
diff --git a/drivers/scsi/hisi_sas/hisi_sas_main.c b/drivers/scsi/hisi_sas/hisi_sas_main.c
index 8c038ccf1c09..325d6d6a21c3 100644
--- a/drivers/scsi/hisi_sas/hisi_sas_main.c
+++ b/drivers/scsi/hisi_sas/hisi_sas_main.c
@@ -529,10 +529,21 @@ static int hisi_sas_queue_command(struct sas_task *task, gfp_t gfp_flags)
 			dq_index = blk_mq_unique_tag_to_hwq(blk_tag);
 			dq = &hisi_hba->dq[dq_index];
 		} else {
-			struct Scsi_Host *shost = hisi_hba->shost;
-			struct blk_mq_queue_map *qmap = &shost->tag_set.map[HCTX_TYPE_DEFAULT];
-			int queue = qmap->mq_map[raw_smp_processor_id()];
+			int queue;
+
+			if (hisi_hba->iopoll_q_cnt) {
+				/*
+				 * Use interrupt queue (queue 0) to deliver and complete
+				 * internal IOs of libsas or libata when there is at least
+				 * one iopoll queue
+				 */
+				queue = 0;
+			} else {
+				struct Scsi_Host *shost = hisi_hba->shost;
+				struct blk_mq_queue_map *qmap = &shost->tag_set.map[HCTX_TYPE_DEFAULT];
 
+				queue = qmap->mq_map[raw_smp_processor_id()];
+			}
 			dq = &hisi_hba->dq[queue];
 		}
 		break;
@@ -672,6 +683,55 @@ static struct hisi_sas_device *hisi_sas_alloc_dev(struct domain_device *device)
 	return sas_dev;
 }
 
+static void hisi_sas_sync_poll_cq(struct hisi_sas_cq *cq)
+{
+	/* make sure CQ entries being processed are processed to completion */
+	spin_lock(&cq->poll_lock);
+	spin_unlock(&cq->poll_lock);
+}
+
+static bool hisi_sas_queue_is_poll(struct hisi_sas_cq *cq)
+{
+	struct hisi_hba *hisi_hba = cq->hisi_hba;
+
+	if (cq->id < hisi_hba->queue_count - hisi_hba->iopoll_q_cnt)
+		return false;
+	return true;
+}
+
+static void hisi_sas_sync_cq(struct hisi_sas_cq *cq)
+{
+	if (hisi_sas_queue_is_poll(cq))
+		hisi_sas_sync_poll_cq(cq);
+	else
+		synchronize_irq(cq->irq_no);
+}
+
+void hisi_sas_sync_poll_cqs(struct hisi_hba *hisi_hba)
+{
+	int i;
+
+	for (i = 0; i < hisi_hba->queue_count; i++) {
+		struct hisi_sas_cq *cq = &hisi_hba->cq[i];
+
+		if (hisi_sas_queue_is_poll(cq))
+			hisi_sas_sync_poll_cq(cq);
+	}
+}
+EXPORT_SYMBOL_GPL(hisi_sas_sync_poll_cqs);
+
+void hisi_sas_sync_cqs(struct hisi_hba *hisi_hba)
+{
+	int i;
+
+	for (i = 0; i < hisi_hba->queue_count; i++) {
+		struct hisi_sas_cq *cq = &hisi_hba->cq[i];
+
+		hisi_sas_sync_cq(cq);
+	}
+}
+EXPORT_SYMBOL_GPL(hisi_sas_sync_cqs);
+
 static void hisi_sas_tmf_aborted(struct sas_task *task)
 {
 	struct hisi_sas_slot *slot = task->lldd_task;
@@ -683,10 +743,10 @@ static void hisi_sas_tmf_aborted(struct sas_task *task)
 		struct hisi_sas_cq *cq =
 			   &hisi_hba->cq[slot->dlvry_queue];
 		/*
-		 * sync irq to avoid free'ing task
+		 * sync irq or poll queue to avoid free'ing task
 		 * before using task in IO completion
 		 */
-		synchronize_irq(cq->irq_no);
+		hisi_sas_sync_cq(cq);
 		slot->task = NULL;
 	}
 }
@@ -1540,11 +1600,11 @@ static int hisi_sas_abort_task(struct sas_task *task)
 
 		if (slot) {
 			/*
-			 * sync irq to avoid free'ing task
+			 * sync irq or poll queue to avoid free'ing task
 			 * before using task in IO completion
 			 */
 			cq = &hisi_hba->cq[slot->dlvry_queue];
-			synchronize_irq(cq->irq_no);
+			hisi_sas_sync_cq(cq);
 		}
 		spin_unlock_irqrestore(&task->task_state_lock, flags);
 		rc = TMF_RESP_FUNC_COMPLETE;
@@ -1611,10 +1671,10 @@ static int hisi_sas_abort_task(struct sas_task *task)
 		if (((rc < 0) || (rc == TMF_RESP_FUNC_FAILED)) &&
 					task->lldd_task) {
 			/*
-			 * sync irq to avoid free'ing task
+			 * sync irq or poll queue to avoid free'ing task
 			 * before using task in IO completion
 			 */
-			synchronize_irq(cq->irq_no);
+			hisi_sas_sync_cq(cq);
 			slot->task = NULL;
 		}
 	}
@@ -1885,10 +1945,10 @@ static bool hisi_sas_internal_abort_timeout(struct sas_task *task,
 			struct hisi_sas_cq *cq =
 				&hisi_hba->cq[slot->dlvry_queue];
 			/*
-			 * sync irq to avoid free'ing task
+			 * sync irq or poll queue to avoid free'ing task
 			 * before using task in IO completion
 			 */
-			synchronize_irq(cq->irq_no);
+			hisi_sas_sync_cq(cq);
 			slot->task = NULL;
 		}
 
@@ -1992,18 +2052,6 @@ void hisi_sas_phy_bcast(struct hisi_sas_phy *phy)
 }
 EXPORT_SYMBOL_GPL(hisi_sas_phy_bcast);
 
-void hisi_sas_sync_irqs(struct hisi_hba *hisi_hba)
-{
-	int i;
-
-	for (i = 0; i < hisi_hba->cq_nvecs; i++) {
-		struct hisi_sas_cq *cq = &hisi_hba->cq[i];
-
-		synchronize_irq(cq->irq_no);
-	}
-}
-EXPORT_SYMBOL_GPL(hisi_sas_sync_irqs);
-
 int hisi_sas_host_reset(struct Scsi_Host *shost, int reset_type)
 {
 	struct hisi_hba *hisi_hba = shost_priv(shost);
@@ -2101,6 +2149,7 @@ int hisi_sas_alloc(struct hisi_hba *hisi_hba)
 		/* Completion queue structure */
 		cq->id = i;
 		cq->hisi_hba = hisi_hba;
+		spin_lock_init(&cq->poll_lock);
 
 		/* Delivery queue structure */
 		spin_lock_init(&dq->lock);
diff --git a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
index 0c3fcb807806..8d0e2dd6207a 100644
--- a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
+++ b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
@@ -552,6 +552,11 @@ static int prot_mask;
 module_param(prot_mask, int, 0444);
 MODULE_PARM_DESC(prot_mask, " host protection capabilities mask, def=0x0 ");
 
+/* the index of iopoll queues are bigger than interrupt queues' */
+static int experimental_iopoll_q_cnt;
+module_param(experimental_iopoll_q_cnt, int, 0444);
+MODULE_PARM_DESC(experimental_iopoll_q_cnt, "number of queues to be used as poll mode, def=0");
+
 static void debugfs_work_handler_v3_hw(struct work_struct *work);
 static void debugfs_snapshot_regs_v3_hw(struct hisi_hba *hisi_hba);
 
@@ -2391,23 +2396,25 @@ out:
 		task->task_done(task);
 }
 
-static irqreturn_t  cq_thread_v3_hw(int irq_no, void *p)
+static int complete_v3_hw(struct hisi_sas_cq *cq)
 {
-	struct hisi_sas_cq *cq = p;
-	struct hisi_hba *hisi_hba = cq->hisi_hba;
-	struct hisi_sas_slot *slot;
 	struct hisi_sas_complete_v3_hdr *complete_queue;
-	u32 rd_point = cq->rd_point, wr_point;
+	struct hisi_hba *hisi_hba = cq->hisi_hba;
+	u32 rd_point, wr_point;
 	int queue = cq->id;
+	int completed;
 
+	rd_point = cq->rd_point;
 	complete_queue = hisi_hba->complete_hdr[queue];
 
 	wr_point = hisi_sas_read32(hisi_hba, COMPL_Q_0_WR_PTR +
 				   (0x14 * queue));
+	completed = (wr_point + HISI_SAS_QUEUE_SLOTS - rd_point) % HISI_SAS_QUEUE_SLOTS;
 
 	while (rd_point != wr_point) {
 		struct hisi_sas_complete_v3_hdr *complete_hdr;
 		struct device *dev = hisi_hba->dev;
+		struct hisi_sas_slot *slot;
 		u32 dw0, dw1, dw3;
 		int iptt;
 
@@ -2451,6 +2458,28 @@ static irqreturn_t  cq_thread_v3_hw(int irq_no, void *p)
 	cq->rd_point = rd_point;
 	hisi_sas_write32(hisi_hba, COMPL_Q_0_RD_PTR + (0x14 * queue), rd_point);
 
+	return completed;
+}
+
+static int queue_complete_v3_hw(struct Scsi_Host *shost, unsigned int queue)
+{
+	struct hisi_hba *hisi_hba = shost_priv(shost);
+	struct hisi_sas_cq *cq = &hisi_hba->cq[queue];
+	int completed;
+
+	spin_lock(&cq->poll_lock);
+	completed = complete_v3_hw(cq);
+	spin_unlock(&cq->poll_lock);
+
+	return completed;
+}
+
+static irqreturn_t cq_thread_v3_hw(int irq_no, void *p)
+{
+	struct hisi_sas_cq *cq = p;
+
+	complete_v3_hw(cq);
+
 	return IRQ_HANDLED;
 }
 
@@ -2474,8 +2503,9 @@ static void hisi_sas_v3_free_vectors(void *data)
 
 static int interrupt_preinit_v3_hw(struct hisi_hba *hisi_hba)
 {
-	int vectors;
-	int max_msi = HISI_SAS_MSI_COUNT_V3_HW, min_msi;
+	/* Allocate all MSI vectors to avoid re-insertion issue */
+	int max_msi = HISI_SAS_MSI_COUNT_V3_HW;
+	int vectors, min_msi;
 	struct Scsi_Host *shost = hisi_hba->shost;
 	struct pci_dev *pdev = hisi_hba->pci_dev;
 	struct irq_affinity desc = {
@@ -2492,8 +2522,8 @@ static int interrupt_preinit_v3_hw(struct hisi_hba *hisi_hba)
 		return -ENOENT;
 
 
-	hisi_hba->cq_nvecs = vectors - BASE_VECTORS_V3_HW;
-	shost->nr_hw_queues = hisi_hba->cq_nvecs;
+	hisi_hba->cq_nvecs = vectors - BASE_VECTORS_V3_HW - hisi_hba->iopoll_q_cnt;
+	shost->nr_hw_queues = hisi_hba->cq_nvecs + hisi_hba->iopoll_q_cnt;
 
 	devm_add_action(&pdev->dev, hisi_sas_v3_free_vectors, pdev);
 	return 0;
@@ -2627,6 +2657,7 @@ static int disable_host_v3_hw(struct hisi_hba *hisi_hba)
 	int rc;
 
 	interrupt_disable_v3_hw(hisi_hba);
+	hisi_sas_sync_poll_cqs(hisi_hba);
 	hisi_sas_write32(hisi_hba, DLVRY_QUEUE_ENABLE, 0x0);
 
 	hisi_sas_stop_phys(hisi_hba);
@@ -2824,6 +2855,18 @@ static ssize_t intr_coal_count_v3_hw_store(struct device *dev,
 }
 static DEVICE_ATTR_RW(intr_coal_count_v3_hw);
 
+static ssize_t iopoll_q_cnt_v3_hw_show(struct device *dev,
+					  struct device_attribute
+					  *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(dev);
+	struct hisi_hba *hisi_hba = shost_priv(shost);
+
+	return scnprintf(buf, PAGE_SIZE, "%u\n",
+			 hisi_hba->iopoll_q_cnt);
+}
+static DEVICE_ATTR_RO(iopoll_q_cnt_v3_hw);
+
 static int slave_configure_v3_hw(struct scsi_device *sdev)
 {
 	struct Scsi_Host *shost = dev_to_shost(&sdev->sdev_gendev);
@@ -2853,6 +2896,7 @@ static struct attribute *host_v3_hw_attrs[] = {
 	&dev_attr_intr_conv_v3_hw.attr,
 	&dev_attr_intr_coal_ticks_v3_hw.attr,
 	&dev_attr_intr_coal_count_v3_hw.attr,
+	&dev_attr_iopoll_q_cnt_v3_hw.attr,
 	NULL
 };
 
@@ -3039,7 +3083,7 @@ static void debugfs_snapshot_prepare_v3_hw(struct hisi_hba *hisi_hba)
 
 	wait_cmds_complete_timeout_v3_hw(hisi_hba, 100, 5000);
 
-	hisi_sas_sync_irqs(hisi_hba);
+	hisi_sas_sync_cqs(hisi_hba);
 }
 
 static void debugfs_snapshot_restore_v3_hw(struct hisi_hba *hisi_hba)
@@ -3211,9 +3255,31 @@ static int debugfs_set_bist_v3_hw(struct hisi_hba *hisi_hba, bool enable)
 static void hisi_sas_map_queues(struct Scsi_Host *shost)
 {
 	struct hisi_hba *hisi_hba = shost_priv(shost);
-	struct blk_mq_queue_map *qmap = &shost->tag_set.map[HCTX_TYPE_DEFAULT];
+	struct blk_mq_queue_map *qmap;
+	int i, qoff;
+
+	for (i = 0, qoff = 0; i < shost->nr_maps; i++) {
+		qmap = &shost->tag_set.map[i];
+		if (i == HCTX_TYPE_DEFAULT) {
+			qmap->nr_queues = hisi_hba->cq_nvecs;
+		} else if (i == HCTX_TYPE_POLL) {
+			qmap->nr_queues = hisi_hba->iopoll_q_cnt;
+		} else {
+			qmap->nr_queues = 0;
+			continue;
+		}
 
-	blk_mq_pci_map_queues(qmap, hisi_hba->pci_dev, BASE_VECTORS_V3_HW);
+		/* At least one interrupt hardware queue */
+		if (!qmap->nr_queues)
+			WARN_ON(i == HCTX_TYPE_DEFAULT);
+		qmap->queue_offset = qoff;
+		if (i == HCTX_TYPE_POLL)
+			blk_mq_map_queues(qmap);
+		else
+			blk_mq_pci_map_queues(qmap, hisi_hba->pci_dev,
+					      BASE_VECTORS_V3_HW);
+		qoff += qmap->nr_queues;
+	}
 }
 
 static struct scsi_host_template sht_v3_hw = {
@@ -3245,6 +3311,7 @@ static struct scsi_host_template sht_v3_hw = {
 	.tag_alloc_policy	= BLK_TAG_ALLOC_RR,
 	.host_reset             = hisi_sas_host_reset,
 	.host_tagset		= 1,
+	.mq_poll		= queue_complete_v3_hw,
 };
 
 static const struct hisi_sas_hw hisi_sas_v3_hw = {
@@ -3304,6 +3371,13 @@ hisi_sas_shost_alloc_pci(struct pci_dev *pdev)
 	if (hisi_sas_get_fw_info(hisi_hba) < 0)
 		goto err_out;
 
+	if (experimental_iopoll_q_cnt < 0 ||
+		experimental_iopoll_q_cnt >= hisi_hba->queue_count)
+		dev_err(dev, "iopoll queue count %d cannot exceed or equal 16, using default 0\n",
+			experimental_iopoll_q_cnt);
+	else
+		hisi_hba->iopoll_q_cnt = experimental_iopoll_q_cnt;
+
 	if (hisi_sas_alloc(hisi_hba)) {
 		hisi_sas_free(hisi_hba);
 		goto err_out;
@@ -4859,6 +4933,10 @@ hisi_sas_v3_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	shost->max_cmd_len = 16;
 	shost->can_queue = HISI_SAS_UNRESERVED_IPTT;
 	shost->cmd_per_lun = HISI_SAS_UNRESERVED_IPTT;
+	if (hisi_hba->iopoll_q_cnt)
+		shost->nr_maps = 3;
+	else
+		shost->nr_maps = 1;
 
 	sha->sas_ha_name = DRV_NAME;
 	sha->dev = dev;