[PATCH v3 0/7] Fixup mtip32xx for scheduling

[PATCH v3 0/7] Fixup mtip32xx for scheduling

[Jens Axboe]

Sending out a new version, since it'd be great to get this fixed
up before sending out the 4.12 pull requests.

Changes since v2(a):

- Kill RQF_RESERVED. Export helper to help the driver see if a given
  request is reserved or not, and use this insead of RQF_RESERVED.

-- 
Jens Axboe

[PATCH 1/7] mtip32xx: get rid of 'atomic' argument to mtip_exec_internal_command()

[Jens Axboe]

All callers can safely block. Kill the atomic/block argument, and
remove the argument from all callers.

Reviewed-by: Bart Van Assche <Bart.VanAssche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
---
 drivers/block/mtip32xx/mtip32xx.c | 148 +++++++++++---------------------------
 1 file changed, 43 insertions(+), 105 deletions(-)

diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
index 02804cc79d82..d81d797ee65d 100644
--- a/drivers/block/mtip32xx/mtip32xx.c
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -609,11 +609,6 @@ static void mtip_completion(struct mtip_port *port,
 	complete(waiting);
 }
 
-static void mtip_null_completion(struct mtip_port *port,
-			    int tag, struct mtip_cmd *command, int status)
-{
-}
-
 static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer,
 				dma_addr_t buffer_dma, unsigned int sectors);
 static int mtip_get_smart_attr(struct mtip_port *port, unsigned int id,
@@ -1117,7 +1112,6 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 					dma_addr_t buffer,
 					int buf_len,
 					u32 opts,
-					gfp_t atomic,
 					unsigned long timeout)
 {
 	struct mtip_cmd_sg *command_sg;
@@ -1146,30 +1140,22 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 
 	clear_bit(MTIP_PF_DM_ACTIVE_BIT, &port->flags);
 
-	if (atomic == GFP_KERNEL) {
-		if (fis->command != ATA_CMD_STANDBYNOW1) {
-			/* wait for io to complete if non atomic */
-			if (mtip_quiesce_io(port,
-				MTIP_QUIESCE_IO_TIMEOUT_MS, atomic) < 0) {
-				dev_warn(&dd->pdev->dev,
-					"Failed to quiesce IO\n");
-				mtip_put_int_command(dd, int_cmd);
-				clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
-				wake_up_interruptible(&port->svc_wait);
-				return -EBUSY;
-			}
+	if (fis->command != ATA_CMD_STANDBYNOW1) {
+		/* wait for io to complete if non atomic */
+		if (mtip_quiesce_io(port,
+			MTIP_QUIESCE_IO_TIMEOUT_MS, GFP_KERNEL) < 0) {
+			dev_warn(&dd->pdev->dev, "Failed to quiesce IO\n");
+			mtip_put_int_command(dd, int_cmd);
+			clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
+			wake_up_interruptible(&port->svc_wait);
+			return -EBUSY;
 		}
-
-		/* Set the completion function and data for the command. */
-		int_cmd->comp_data = &wait;
-		int_cmd->comp_func = mtip_completion;
-
-	} else {
-		/* Clear completion - we're going to poll */
-		int_cmd->comp_data = NULL;
-		int_cmd->comp_func = mtip_null_completion;
 	}
 
+	/* Set the completion function and data for the command. */
+	int_cmd->comp_data = &wait;
+	int_cmd->comp_func = mtip_completion;
+
 	/* Copy the command to the command table */
 	memcpy(int_cmd->command, fis, fis_len*4);
 
@@ -1198,81 +1184,41 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 	/* Issue the command to the hardware */
 	mtip_issue_non_ncq_command(port, MTIP_TAG_INTERNAL);
 
-	if (atomic == GFP_KERNEL) {
-		/* Wait for the command to complete or timeout. */
-		if ((rv = wait_for_completion_interruptible_timeout(
-				&wait,
-				msecs_to_jiffies(timeout))) <= 0) {
-
-			if (rv == -ERESTARTSYS) { /* interrupted */
-				dev_err(&dd->pdev->dev,
-					"Internal command [%02X] was interrupted after %u ms\n",
-					fis->command,
-					jiffies_to_msecs(jiffies - start));
-				rv = -EINTR;
-				goto exec_ic_exit;
-			} else if (rv == 0) /* timeout */
-				dev_err(&dd->pdev->dev,
-					"Internal command did not complete [%02X] within timeout of  %lu ms\n",
-					fis->command, timeout);
-			else
-				dev_err(&dd->pdev->dev,
-					"Internal command [%02X] wait returned code [%d] after %lu ms - unhandled\n",
-					fis->command, rv, timeout);
-
-			if (mtip_check_surprise_removal(dd->pdev) ||
-				test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
-						&dd->dd_flag)) {
-				dev_err(&dd->pdev->dev,
-					"Internal command [%02X] wait returned due to SR\n",
-					fis->command);
-				rv = -ENXIO;
-				goto exec_ic_exit;
-			}
-			mtip_device_reset(dd); /* recover from timeout issue */
-			rv = -EAGAIN;
+	/* Wait for the command to complete or timeout. */
+	rv = wait_for_completion_interruptible_timeout(&wait,
+				msecs_to_jiffies(timeout));
+	if (rv <= 0) {
+		if (rv == -ERESTARTSYS) { /* interrupted */
+			dev_err(&dd->pdev->dev,
+				"Internal command [%02X] was interrupted after %u ms\n",
+				fis->command,
+				jiffies_to_msecs(jiffies - start));
+			rv = -EINTR;
 			goto exec_ic_exit;
-		}
-	} else {
-		u32 hba_stat, port_stat;
-
-		/* Spin for <timeout> checking if command still outstanding */
-		timeout = jiffies + msecs_to_jiffies(timeout);
-		while ((readl(port->cmd_issue[MTIP_TAG_INTERNAL])
-				& (1 << MTIP_TAG_INTERNAL))
-				&& time_before(jiffies, timeout)) {
-			if (mtip_check_surprise_removal(dd->pdev)) {
-				rv = -ENXIO;
-				goto exec_ic_exit;
-			}
-			if ((fis->command != ATA_CMD_STANDBYNOW1) &&
-				test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
-						&dd->dd_flag)) {
-				rv = -ENXIO;
-				goto exec_ic_exit;
-			}
-			port_stat = readl(port->mmio + PORT_IRQ_STAT);
-			if (!port_stat)
-				continue;
+		} else if (rv == 0) /* timeout */
+			dev_err(&dd->pdev->dev,
+				"Internal command did not complete [%02X] within timeout of  %lu ms\n",
+				fis->command, timeout);
+		else
+			dev_err(&dd->pdev->dev,
+				"Internal command [%02X] wait returned code [%d] after %lu ms - unhandled\n",
+				fis->command, rv, timeout);
 
-			if (port_stat & PORT_IRQ_ERR) {
-				dev_err(&dd->pdev->dev,
-					"Internal command [%02X] failed\n",
-					fis->command);
-				mtip_device_reset(dd);
-				rv = -EIO;
-				goto exec_ic_exit;
-			} else {
-				writel(port_stat, port->mmio + PORT_IRQ_STAT);
-				hba_stat = readl(dd->mmio + HOST_IRQ_STAT);
-				if (hba_stat)
-					writel(hba_stat,
-						dd->mmio + HOST_IRQ_STAT);
-			}
-			break;
+		if (mtip_check_surprise_removal(dd->pdev) ||
+			test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
+					&dd->dd_flag)) {
+			dev_err(&dd->pdev->dev,
+				"Internal command [%02X] wait returned due to SR\n",
+				fis->command);
+			rv = -ENXIO;
+			goto exec_ic_exit;
 		}
+		mtip_device_reset(dd); /* recover from timeout issue */
+		rv = -EAGAIN;
+		goto exec_ic_exit;
 	}
 
+	rv = 0;
 	if (readl(port->cmd_issue[MTIP_TAG_INTERNAL])
 			& (1 << MTIP_TAG_INTERNAL)) {
 		rv = -ENXIO;
@@ -1391,7 +1337,6 @@ static int mtip_get_identify(struct mtip_port *port, void __user *user_buffer)
 				port->identify_dma,
 				sizeof(u16) * ATA_ID_WORDS,
 				0,
-				GFP_KERNEL,
 				MTIP_INT_CMD_TIMEOUT_MS)
 				< 0) {
 		rv = -1;
@@ -1477,7 +1422,6 @@ static int mtip_standby_immediate(struct mtip_port *port)
 					0,
 					0,
 					0,
-					GFP_ATOMIC,
 					timeout);
 	dbg_printk(MTIP_DRV_NAME "Time taken to complete standby cmd: %d ms\n",
 			jiffies_to_msecs(jiffies - start));
@@ -1523,7 +1467,6 @@ static int mtip_read_log_page(struct mtip_port *port, u8 page, u16 *buffer,
 					buffer_dma,
 					sectors * ATA_SECT_SIZE,
 					0,
-					GFP_ATOMIC,
 					MTIP_INT_CMD_TIMEOUT_MS);
 }
 
@@ -1558,7 +1501,6 @@ static int mtip_get_smart_data(struct mtip_port *port, u8 *buffer,
 					buffer_dma,
 					ATA_SECT_SIZE,
 					0,
-					GFP_ATOMIC,
 					15000);
 }
 
@@ -1686,7 +1628,6 @@ static int mtip_send_trim(struct driver_data *dd, unsigned int lba,
 					dma_addr,
 					ATA_SECT_SIZE,
 					0,
-					GFP_KERNEL,
 					MTIP_TRIM_TIMEOUT_MS) < 0)
 		rv = -EIO;
 
@@ -1850,7 +1791,6 @@ static int exec_drive_task(struct mtip_port *port, u8 *command)
 				 0,
 				 0,
 				 0,
-				 GFP_KERNEL,
 				 to) < 0) {
 		return -1;
 	}
@@ -1946,7 +1886,6 @@ static int exec_drive_command(struct mtip_port *port, u8 *command,
 				 (xfer_sz ? dma_addr : 0),
 				 (xfer_sz ? ATA_SECT_SIZE * xfer_sz : 0),
 				 0,
-				 GFP_KERNEL,
 				 to)
 				 < 0) {
 		rv = -EFAULT;
@@ -2189,7 +2128,6 @@ static int exec_drive_taskfile(struct driver_data *dd,
 				 dma_buffer,
 				 transfer_size,
 				 0,
-				 GFP_KERNEL,
 				 timeout) < 0) {
 		err = -EIO;
 		goto abort;
-- 
2.7.4

[PATCH 2/7] mtip32xx: kill atomic argument to mtip_quiesce_io()

[Jens Axboe]

All callers now pass in GFP_KERNEL, get rid of the argument.

Reviewed-by: Bart Van Assche <Bart.VanAssche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
---
 drivers/block/mtip32xx/mtip32xx.c | 20 +++++---------------
 1 file changed, 5 insertions(+), 15 deletions(-)

diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
index d81d797ee65d..36f3d34f2156 100644
--- a/drivers/block/mtip32xx/mtip32xx.c
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -1035,14 +1035,12 @@ static bool mtip_pause_ncq(struct mtip_port *port,
  *
  * @port    Pointer to port data structure
  * @timeout Max duration to wait (ms)
- * @atomic  gfp_t flag to indicate blockable context or not
  *
  * return value
  *	0	Success
  *	-EBUSY  Commands still active
  */
-static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout,
-								gfp_t atomic)
+static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
 {
 	unsigned long to;
 	unsigned int n;
@@ -1053,18 +1051,12 @@ static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout,
 	to = jiffies + msecs_to_jiffies(timeout);
 	do {
 		if (test_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags) &&
-			test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags) &&
-			atomic == GFP_KERNEL) {
+			test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) {
 			msleep(20);
 			continue; /* svc thd is actively issuing commands */
 		}
 
-		if (atomic == GFP_KERNEL)
-			msleep(100);
-		else {
-			cpu_relax();
-			udelay(100);
-		}
+		msleep(100);
 
 		if (mtip_check_surprise_removal(port->dd->pdev))
 			goto err_fault;
@@ -1142,8 +1134,7 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 
 	if (fis->command != ATA_CMD_STANDBYNOW1) {
 		/* wait for io to complete if non atomic */
-		if (mtip_quiesce_io(port,
-			MTIP_QUIESCE_IO_TIMEOUT_MS, GFP_KERNEL) < 0) {
+		if (mtip_quiesce_io(port, MTIP_QUIESCE_IO_TIMEOUT_MS) < 0) {
 			dev_warn(&dd->pdev->dev, "Failed to quiesce IO\n");
 			mtip_put_int_command(dd, int_cmd);
 			clear_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
@@ -4106,8 +4097,7 @@ static int mtip_block_remove(struct driver_data *dd)
 		 * Explicitly wait here for IOs to quiesce,
 		 * as mtip_standby_drive usually won't wait for IOs.
 		 */
-		if (!mtip_quiesce_io(dd->port, MTIP_QUIESCE_IO_TIMEOUT_MS,
-								GFP_KERNEL))
+		if (!mtip_quiesce_io(dd->port, MTIP_QUIESCE_IO_TIMEOUT_MS))
 			mtip_standby_drive(dd);
 	}
 	else
-- 
2.7.4

[PATCH 3/7] mtip32xx: abstract out "are any commands active" helper

[Jens Axboe]

This is a prep patch for backoff in ->queue_rq() for non-ncq commands.

Reviewed-by: Bart Van Assche <Bart.VanAssche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
---
 drivers/block/mtip32xx/mtip32xx.c | 28 ++++++++++++++++++----------
 1 file changed, 18 insertions(+), 10 deletions(-)

diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
index 36f3d34f2156..aee94f260725 100644
--- a/drivers/block/mtip32xx/mtip32xx.c
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -1030,6 +1030,22 @@ static bool mtip_pause_ncq(struct mtip_port *port,
 	return false;
 }
 
+static bool mtip_commands_active(struct mtip_port *port)
+{
+	unsigned int active;
+	unsigned int n;
+
+	/*
+	 * Ignore s_active bit 0 of array element 0.
+	 * This bit will always be set
+	 */
+	active = readl(port->s_active[0]) & 0xFFFFFFFE;
+	for (n = 1; n < port->dd->slot_groups; n++)
+		active |= readl(port->s_active[n]);
+
+	return active != 0;
+}
+
 /*
  * Wait for port to quiesce
  *
@@ -1043,8 +1059,7 @@ static bool mtip_pause_ncq(struct mtip_port *port,
 static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
 {
 	unsigned long to;
-	unsigned int n;
-	unsigned int active = 1;
+	bool active = true;
 
 	blk_mq_stop_hw_queues(port->dd->queue);
 
@@ -1061,14 +1076,7 @@ static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
 		if (mtip_check_surprise_removal(port->dd->pdev))
 			goto err_fault;
 
-		/*
-		 * Ignore s_active bit 0 of array element 0.
-		 * This bit will always be set
-		 */
-		active = readl(port->s_active[0]) & 0xFFFFFFFE;
-		for (n = 1; n < port->dd->slot_groups; n++)
-			active |= readl(port->s_active[n]);
-
+		active = mtip_commands_active(port);
 		if (!active)
 			break;
 	} while (time_before(jiffies, to));
-- 
2.7.4

[PATCH 4/7] blk-mq: add and export helper to tell if a request is reserved

[Jens Axboe]

We only have an internal helper for checking a tag value. Add
an exported helper that takes a request and hardware queue,
and check against the driver tags.

Signed-off-by: Jens Axboe <axboe@fb.com>
---
 block/blk-mq-tag.c     | 6 ++++++
 include/linux/blk-mq.h | 1 +
 2 files changed, 7 insertions(+)

diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index d0be72ccb091..0a04b5c14076 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -476,3 +476,9 @@ u32 blk_mq_unique_tag(struct request *rq)
 		(rq->tag & BLK_MQ_UNIQUE_TAG_MASK);
 }
 EXPORT_SYMBOL(blk_mq_unique_tag);
+
+bool blk_mq_rq_is_reserved(struct blk_mq_hw_ctx *hctx, struct request *rq)
+{
+	return blk_mq_tag_is_reserved(hctx->tags, rq->tag);
+}
+EXPORT_SYMBOL_GPL(blk_mq_rq_is_reserved);
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 32bd8eb5ba67..13e51520fa27 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -224,6 +224,7 @@ static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
 	return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
 }
 
+bool blk_mq_rq_is_reserved(struct blk_mq_hw_ctx *hctx, struct request *rq);
 
 int blk_mq_request_started(struct request *rq);
 void blk_mq_start_request(struct request *rq);
-- 
2.7.4

[PATCH 5/7] mtip32xx: convert internal command issue to block IO path

[Jens Axboe]

The driver special cases certain things for command issue, depending
on whether it's an internal command or not. Make the internal commands
use the regular infrastructure for issuing IO.

Since this is an 8-group souped up AHCI variant, we have to deal
with NCQ vs non-queueable commands. Do this from the queue_rq
handler, by backing off unless the drive is idle.

Signed-off-by: Jens Axboe <axboe@fb.com>
---
 drivers/block/mtip32xx/mtip32xx.c | 106 +++++++++++++++++++++++++++-----------
 1 file changed, 76 insertions(+), 30 deletions(-)

diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
index aee94f260725..9780c4d3697c 100644
--- a/drivers/block/mtip32xx/mtip32xx.c
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -195,7 +195,7 @@ static struct mtip_cmd *mtip_get_int_command(struct driver_data *dd)
 	if (mtip_check_surprise_removal(dd->pdev))
 		return NULL;
 
-	rq = blk_mq_alloc_request(dd->queue, 0, BLK_MQ_REQ_RESERVED);
+	rq = blk_mq_alloc_request(dd->queue, REQ_OP_DRV_IN, BLK_MQ_REQ_RESERVED);
 	if (IS_ERR(rq))
 		return NULL;
 
@@ -1088,6 +1088,13 @@ static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
 	return -EFAULT;
 }
 
+struct mtip_int_cmd {
+	int fis_len;
+	dma_addr_t buffer;
+	int buf_len;
+	u32 opts;
+};
+
 /*
  * Execute an internal command and wait for the completion.
  *
@@ -1114,10 +1121,16 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 					u32 opts,
 					unsigned long timeout)
 {
-	struct mtip_cmd_sg *command_sg;
 	DECLARE_COMPLETION_ONSTACK(wait);
 	struct mtip_cmd *int_cmd;
 	struct driver_data *dd = port->dd;
+	struct request *rq;
+	struct mtip_int_cmd icmd = {
+		.fis_len = fis_len,
+		.buffer = buffer,
+		.buf_len = buf_len,
+		.opts = opts
+	};
 	int rv = 0;
 	unsigned long start;
 
@@ -1132,6 +1145,8 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 		dbg_printk(MTIP_DRV_NAME "Unable to allocate tag for PIO cmd\n");
 		return -EFAULT;
 	}
+	rq = blk_mq_rq_from_pdu(int_cmd);
+	rq->end_io_data = &icmd;
 
 	set_bit(MTIP_PF_IC_ACTIVE_BIT, &port->flags);
 
@@ -1158,35 +1173,16 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 	/* Copy the command to the command table */
 	memcpy(int_cmd->command, fis, fis_len*4);
 
-	/* Populate the SG list */
-	int_cmd->command_header->opts =
-		 __force_bit2int cpu_to_le32(opts | fis_len);
-	if (buf_len) {
-		command_sg = int_cmd->command + AHCI_CMD_TBL_HDR_SZ;
-
-		command_sg->info =
-			__force_bit2int cpu_to_le32((buf_len-1) & 0x3FFFFF);
-		command_sg->dba	=
-			__force_bit2int cpu_to_le32(buffer & 0xFFFFFFFF);
-		command_sg->dba_upper =
-			__force_bit2int cpu_to_le32((buffer >> 16) >> 16);
-
-		int_cmd->command_header->opts |=
-			__force_bit2int cpu_to_le32((1 << 16));
-	}
-
-	/* Populate the command header */
-	int_cmd->command_header->byte_count = 0;
-
 	start = jiffies;
+	rq->timeout = timeout;
 
-	/* Issue the command to the hardware */
-	mtip_issue_non_ncq_command(port, MTIP_TAG_INTERNAL);
+	/* insert request and run queue */
+	blk_execute_rq_nowait(rq->q, NULL, rq, true, NULL);
+
+	wait_for_completion(&wait);
+	rv = int_cmd->status;
 
-	/* Wait for the command to complete or timeout. */
-	rv = wait_for_completion_interruptible_timeout(&wait,
-				msecs_to_jiffies(timeout));
-	if (rv <= 0) {
+	if (rv < 0) {
 		if (rv == -ERESTARTSYS) { /* interrupted */
 			dev_err(&dd->pdev->dev,
 				"Internal command [%02X] was interrupted after %u ms\n",
@@ -1217,7 +1213,6 @@ static int mtip_exec_internal_command(struct mtip_port *port,
 		goto exec_ic_exit;
 	}
 
-	rv = 0;
 	if (readl(port->cmd_issue[MTIP_TAG_INTERNAL])
 			& (1 << MTIP_TAG_INTERNAL)) {
 		rv = -ENXIO;
@@ -3762,6 +3757,44 @@ static bool mtip_check_unal_depth(struct blk_mq_hw_ctx *hctx,
 	return false;
 }
 
+static int mtip_issue_reserved_cmd(struct blk_mq_hw_ctx *hctx,
+				   struct request *rq)
+{
+	struct driver_data *dd = hctx->queue->queuedata;
+	struct mtip_int_cmd *icmd = rq->end_io_data;
+	struct mtip_cmd *cmd = blk_mq_rq_to_pdu(rq);
+	struct mtip_cmd_sg *command_sg;
+
+	if (mtip_commands_active(dd->port))
+		return BLK_MQ_RQ_QUEUE_BUSY;
+
+	rq->end_io_data = NULL;
+
+	/* Populate the SG list */
+	cmd->command_header->opts =
+		 __force_bit2int cpu_to_le32(icmd->opts | icmd->fis_len);
+	if (icmd->buf_len) {
+		command_sg = cmd->command + AHCI_CMD_TBL_HDR_SZ;
+
+		command_sg->info =
+			__force_bit2int cpu_to_le32((icmd->buf_len-1) & 0x3FFFFF);
+		command_sg->dba	=
+			__force_bit2int cpu_to_le32(icmd->buffer & 0xFFFFFFFF);
+		command_sg->dba_upper =
+			__force_bit2int cpu_to_le32((icmd->buffer >> 16) >> 16);
+
+		cmd->command_header->opts |=
+			__force_bit2int cpu_to_le32((1 << 16));
+	}
+
+	/* Populate the command header */
+	cmd->command_header->byte_count = 0;
+
+	blk_mq_start_request(rq);
+	mtip_issue_non_ncq_command(dd->port, rq->tag);
+	return BLK_MQ_RQ_QUEUE_OK;
+}
+
 static int mtip_queue_rq(struct blk_mq_hw_ctx *hctx,
 			 const struct blk_mq_queue_data *bd)
 {
@@ -3770,6 +3803,12 @@ static int mtip_queue_rq(struct blk_mq_hw_ctx *hctx,
 
 	mtip_init_cmd_header(rq);
 
+	/*
+	 * Reserved tags are internal commands
+	 */
+	if (blk_mq_rq_is_reserved(hctx, rq))
+		return mtip_issue_reserved_cmd(hctx, rq);
+
 	if (unlikely(mtip_check_unal_depth(hctx, rq)))
 		return BLK_MQ_RQ_QUEUE_BUSY;
 
@@ -3825,8 +3864,14 @@ static enum blk_eh_timer_return mtip_cmd_timeout(struct request *req,
 {
 	struct driver_data *dd = req->q->queuedata;
 
-	if (reserved)
+	if (reserved) {
+		struct mtip_cmd *cmd = blk_mq_rq_to_pdu(req);
+
+		cmd->status = -ETIME;
+		if (cmd->comp_func)
+			cmd->comp_func(dd->port, MTIP_TAG_INTERNAL, cmd, -ETIME);
 		goto exit_handler;
+	}
 
 	if (test_bit(req->tag, dd->port->cmds_to_issue))
 		goto exit_handler;
@@ -4063,6 +4108,7 @@ static void mtip_no_dev_cleanup(struct request *rq, void *data, bool reserv)
 	} else if (test_bit(MTIP_PF_IC_ACTIVE_BIT, &dd->port->flags)) {
 
 		cmd = mtip_cmd_from_tag(dd, MTIP_TAG_INTERNAL);
+		cmd->status = -ENODEV;
 		if (cmd->comp_func)
 			cmd->comp_func(dd->port, MTIP_TAG_INTERNAL,
 					cmd, -ENODEV);
-- 
2.7.4

[PATCH 6/7] blk-mq-sched: remove hack that bypasses scheduler for reserved requests

[Jens Axboe]

We have update the troublesome driver (mtip32xx) to deal with this
appropriately. So kill the hack that bypassed scheduler allocation
and insertion for reserved requests.

Signed-off-by: Jens Axboe <axboe@fb.com>
---
 block/blk-mq-sched.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index 8b361e192e8a..e79e9f18d7c2 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -82,11 +82,7 @@ struct request *blk_mq_sched_get_request(struct request_queue *q,
 	if (likely(!data->hctx))
 		data->hctx = blk_mq_map_queue(q, data->ctx->cpu);
 
-	/*
-	 * For a reserved tag, allocate a normal request since we might
-	 * have driver dependencies on the value of the internal tag.
-	 */
-	if (e && !(data->flags & BLK_MQ_REQ_RESERVED)) {
+	if (e) {
 		data->flags |= BLK_MQ_REQ_INTERNAL;
 
 		/*
-- 
2.7.4

[PATCH 7/7] Revert "mtip32xx: pass BLK_MQ_F_NO_SCHED"

[Jens Axboe]

This reverts commit 4981d04dd8f1ab19e2cce008da556d7f099b6e68.

The driver has been converted to using the proper infrastructure
for issuing internal commands. This means it's now safe to use with
the scheduling infrastruture, so we can now revert the change
that turned off scheduling for mtip32xx.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
---
 drivers/block/mtip32xx/mtip32xx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
index 9780c4d3697c..c598c5ac2fd0 100644
--- a/drivers/block/mtip32xx/mtip32xx.c
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -3964,7 +3964,7 @@ static int mtip_block_initialize(struct driver_data *dd)
 	dd->tags.reserved_tags = 1;
 	dd->tags.cmd_size = sizeof(struct mtip_cmd);
 	dd->tags.numa_node = dd->numa_node;
-	dd->tags.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_NO_SCHED;
+	dd->tags.flags = BLK_MQ_F_SHOULD_MERGE;
 	dd->tags.driver_data = dd;
 	dd->tags.timeout = MTIP_NCQ_CMD_TIMEOUT_MS;
 
-- 
2.7.4

Re: [PATCH 4/7] blk-mq: add and export helper to tell if a request is reserved

[Bart Van Assche]

On Fri, 2017-04-28 at 08:55 -0600, Jens Axboe wrote:
> We only have an internal helper for checking a tag value. Add
> an exported helper that takes a request and hardware queue,
> and check against the driver tags.
>=20
> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  block/blk-mq-tag.c     | 6 ++++++
>  include/linux/blk-mq.h | 1 +
>  2 files changed, 7 insertions(+)
>=20
> diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
> index d0be72ccb091..0a04b5c14076 100644
> --- a/block/blk-mq-tag.c
> +++ b/block/blk-mq-tag.c
> @@ -476,3 +476,9 @@ u32 blk_mq_unique_tag(struct request *rq)
>  		(rq->tag & BLK_MQ_UNIQUE_TAG_MASK);
>  }
>  EXPORT_SYMBOL(blk_mq_unique_tag);
> +
> +bool blk_mq_rq_is_reserved(struct blk_mq_hw_ctx *hctx, struct request *r=
q)
> +{
> +	return blk_mq_tag_is_reserved(hctx->tags, rq->tag);
> +}
> +EXPORT_SYMBOL_GPL(blk_mq_rq_is_reserved);
> diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
> index 32bd8eb5ba67..13e51520fa27 100644
> --- a/include/linux/blk-mq.h
> +++ b/include/linux/blk-mq.h
> @@ -224,6 +224,7 @@ static inline u16 blk_mq_unique_tag_to_tag(u32 unique=
_tag)
>  	return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
>  }
> =20
> +bool blk_mq_rq_is_reserved(struct blk_mq_hw_ctx *hctx, struct request *r=
q);
> =20
>  int blk_mq_request_started(struct request *rq);
>  void blk_mq_start_request(struct request *rq);

Hello Jens,

Have you considered to derive the hctx information from rq->mq_ctx instead =
of
passing hctx as an argument (hctx =3D blk_mq_map_queue(rq->q, rq->mq_ctx->c=
pu))?
Anyway, since otherwise this patch looks fine to me:

Reviewed-by: Bart Van Assche <Bart.VanAssche@sandisk.com>=

Re: [PATCH 4/7] blk-mq: add and export helper to tell if a request is reserved

[Ming Lei]

On Fri, Apr 28, 2017 at 08:55:37AM -0600, Jens Axboe wrote:
> We only have an internal helper for checking a tag value. Add
> an exported helper that takes a request and hardware queue,
> and check against the driver tags.
> 
> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  block/blk-mq-tag.c     | 6 ++++++
>  include/linux/blk-mq.h | 1 +
>  2 files changed, 7 insertions(+)
> 
> diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
> index d0be72ccb091..0a04b5c14076 100644
> --- a/block/blk-mq-tag.c
> +++ b/block/blk-mq-tag.c
> @@ -476,3 +476,9 @@ u32 blk_mq_unique_tag(struct request *rq)
>  		(rq->tag & BLK_MQ_UNIQUE_TAG_MASK);
>  }
>  EXPORT_SYMBOL(blk_mq_unique_tag);
> +
> +bool blk_mq_rq_is_reserved(struct blk_mq_hw_ctx *hctx, struct request *rq)
> +{
> +	return blk_mq_tag_is_reserved(hctx->tags, rq->tag);
> +}
> +EXPORT_SYMBOL_GPL(blk_mq_rq_is_reserved);
> diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
> index 32bd8eb5ba67..13e51520fa27 100644
> --- a/include/linux/blk-mq.h
> +++ b/include/linux/blk-mq.h
> @@ -224,6 +224,7 @@ static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
>  	return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
>  }
>  
> +bool blk_mq_rq_is_reserved(struct blk_mq_hw_ctx *hctx, struct request *rq);
>  
>  int blk_mq_request_started(struct request *rq);
>  void blk_mq_start_request(struct request *rq);

Reviewed-by: Ming Lei <ming.lei@redhat.com>

Thanks,
Ming

Re: [PATCH 6/7] blk-mq-sched: remove hack that bypasses scheduler for reserved requests

[Ming Lei]

On Fri, Apr 28, 2017 at 08:55:39AM -0600, Jens Axboe wrote:
> We have update the troublesome driver (mtip32xx) to deal with this
> appropriately. So kill the hack that bypassed scheduler allocation
> and insertion for reserved requests.
> 
> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  block/blk-mq-sched.c | 6 +-----
>  1 file changed, 1 insertion(+), 5 deletions(-)
> 
> diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
> index 8b361e192e8a..e79e9f18d7c2 100644
> --- a/block/blk-mq-sched.c
> +++ b/block/blk-mq-sched.c
> @@ -82,11 +82,7 @@ struct request *blk_mq_sched_get_request(struct request_queue *q,
>  	if (likely(!data->hctx))
>  		data->hctx = blk_mq_map_queue(q, data->ctx->cpu);
>  
> -	/*
> -	 * For a reserved tag, allocate a normal request since we might
> -	 * have driver dependencies on the value of the internal tag.
> -	 */
> -	if (e && !(data->flags & BLK_MQ_REQ_RESERVED)) {
> +	if (e) {
>  		data->flags |= BLK_MQ_REQ_INTERNAL;
>  
>  		/*

Reviewed-by: Ming Lei <ming.lei@redhat.com>

Thanks,
Ming

Re: [PATCH 4/7] blk-mq: add and export helper to tell if a request is reserved

[Hannes Reinecke]

On 04/28/2017 04:55 PM, Jens Axboe wrote:
> We only have an internal helper for checking a tag value. Add
> an exported helper that takes a request and hardware queue,
> and check against the driver tags.
> 
> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  block/blk-mq-tag.c     | 6 ++++++
>  include/linux/blk-mq.h | 1 +
>  2 files changed, 7 insertions(+)
> 
Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N�rnberg
GF: F. Imend�rffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG N�rnberg)

Re: [PATCH 5/7] mtip32xx: convert internal command issue to block IO path

[Hannes Reinecke]

On 04/28/2017 04:55 PM, Jens Axboe wrote:
> The driver special cases certain things for command issue, depending
> on whether it's an internal command or not. Make the internal commands
> use the regular infrastructure for issuing IO.
> 
> Since this is an 8-group souped up AHCI variant, we have to deal
> with NCQ vs non-queueable commands. Do this from the queue_rq
> handler, by backing off unless the drive is idle.
> 
> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  drivers/block/mtip32xx/mtip32xx.c | 106 +++++++++++++++++++++++++++-----------
>  1 file changed, 76 insertions(+), 30 deletions(-)
> 
Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N�rnberg
GF: F. Imend�rffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG N�rnberg)

Re: [PATCH 6/7] blk-mq-sched: remove hack that bypasses scheduler for reserved requests

[Hannes Reinecke]

On 04/28/2017 04:55 PM, Jens Axboe wrote:
> We have update the troublesome driver (mtip32xx) to deal with this
> appropriately. So kill the hack that bypassed scheduler allocation
> and insertion for reserved requests.
> 
updated?

> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  block/blk-mq-sched.c | 6 +-----
>  1 file changed, 1 insertion(+), 5 deletions(-)
> 
> diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
> index 8b361e192e8a..e79e9f18d7c2 100644
> --- a/block/blk-mq-sched.c
> +++ b/block/blk-mq-sched.c
> @@ -82,11 +82,7 @@ struct request *blk_mq_sched_get_request(struct request_queue *q,
>  	if (likely(!data->hctx))
>  		data->hctx = blk_mq_map_queue(q, data->ctx->cpu);
>  
> -	/*
> -	 * For a reserved tag, allocate a normal request since we might
> -	 * have driver dependencies on the value of the internal tag.
> -	 */
> -	if (e && !(data->flags & BLK_MQ_REQ_RESERVED)) {
> +	if (e) {
>  		data->flags |= BLK_MQ_REQ_INTERNAL;
>  
>  		/*
> 
Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N�rnberg
GF: F. Imend�rffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG N�rnberg)

Re: [PATCH 7/7] Revert "mtip32xx: pass BLK_MQ_F_NO_SCHED"

[Hannes Reinecke]

On 04/28/2017 04:55 PM, Jens Axboe wrote:
> This reverts commit 4981d04dd8f1ab19e2cce008da556d7f099b6e68.
> 
> The driver has been converted to using the proper infrastructure
> for issuing internal commands. This means it's now safe to use with
> the scheduling infrastruture, so we can now revert the change
> that turned off scheduling for mtip32xx.
> 
> Reviewed-by: Ming Lei <ming.lei@redhat.com>
> Signed-off-by: Jens Axboe <axboe@fb.com>
> ---
>  drivers/block/mtip32xx/mtip32xx.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
> index 9780c4d3697c..c598c5ac2fd0 100644
> --- a/drivers/block/mtip32xx/mtip32xx.c
> +++ b/drivers/block/mtip32xx/mtip32xx.c
> @@ -3964,7 +3964,7 @@ static int mtip_block_initialize(struct driver_data *dd)
>  	dd->tags.reserved_tags = 1;
>  	dd->tags.cmd_size = sizeof(struct mtip_cmd);
>  	dd->tags.numa_node = dd->numa_node;
> -	dd->tags.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_NO_SCHED;
> +	dd->tags.flags = BLK_MQ_F_SHOULD_MERGE;
>  	dd->tags.driver_data = dd;
>  	dd->tags.timeout = MTIP_NCQ_CMD_TIMEOUT_MS;
>  
> 
Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 N�rnberg
GF: F. Imend�rffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG N�rnberg)