Skip to contentSkip to conversation

scsi: enabled reserved commands for LLDDs

PATCH Superseded Hannes Reinecke 4 participants 55 messages May 3, 2021 – Jun 24, 2021

Subsystems (5)

v8 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v8 cover scsi: enabled reserved commands for LLDDs

Hi all,

quite some drivers use internal commands for various purposes, most commonly sending TMFs or querying the HBA status.
While these commands use the same submission mechanism than normal I/O commands, they will not be counted as outstanding commands, requiring those drivers to implement their own mechanism to figure out outstanding commands.
The block layer already has the concept of 'reserved' tags for precisely this purpose, namely non-I/O tags which live off a separate tag pool. That guarantees that these commands can always be sent, and won't be influenced by tag starvation from the I/O tag pool. This patchset enables the use of reserved tags for the SCSI midlayer by allocating a virtual LUN for the HBA itself which just serves as a resource to allocate valid tags from.
This removes quite some hacks which were required for some drivers (eg. fnic or snic), and allows the use of tagset iterators within the drivers.

Command allocation currently ignores the hardware queues, as none of the modified drivers is mq-capable.

The entire patchset can be found at

git://git.kernel.org/pub/scm/linux/kernel/git/hare/scsi-devel.git
reserved-tags.v8

This patchset also includes the busy_iter patches for fnic, which were also sent as a separate patchset. So if they are applied separately they can be dropped from this patchset.

As usual, comments and reviews are welcome.

Changes to v7:
- Drop changes to hisi_sas, pm8001, and mv_sas
- Drop patch to introduce REQ_INTERNAL flag
- Include reviews from John Garry

Changes to v6:
- Remove patch to drop gdth
- Rework libsas to use a tag per slow task
- Update hisi_sas, pm8001, and mv_sas

Changes to v5:
- Remove patch for csiostor
- Warn on normal commands in scsi_put_reserved_cmd()
- Fixup aacraid to not only scsi_put_internal_cmd() for reserved commands
- Add 'nr_reserved_cmds' field to host template
- Reshuffle patches

Changes to v4:
- Fixup kbuild warning
- Include reviews from Bart

Changes to v3:
- Kill gdth
- Only convert fnic, snic, hpsa, and aacraid
- Drop command emulation for pseudo host device
- make 'can_queue' exclude the number or reserved tags
- Drop persistent commands proposal
- Sanitize host device handling

Changes to v2:
- Update patches from John Garry
- Use virtual LUN as suggested by Christoph
- Improve SCSI Host device to present a real SCSI device
- Implement 'persistent' commands for AENs
- Convert Megaraid SAS

Changes to v1:
- Make scsi_{get, put}_reserved_cmd() for Scsi host
- Previously we separate scsi_{get, put}_reserved_cmd() for sdev and scsi_host_get_reserved_cmd() for the host
- Fix how Scsi_Host.can_queue is set in the virtio-scsi change
- Drop Scsi_Host.use_reserved_cmd_q
- Drop scsi_is_reserved_cmd()
- Add support in libsas and associated HBA drivers
- Allocate reserved command in slow task
- Switch hisi_sas to use reserved Scsi command
- Reorder the series a little
- Some tidying

Hannes Reinecke (18):
fnic: kill 'exclude_id' argument to fnic_cleanup_io()
fnic: use scsi_host_busy_iter() to traverse commands
scsi: add scsi_{get,put}_internal_cmd() helper
fnic: use internal commands
scsi: use real inquiry data when initialising devices
scsi: Use dummy inquiry data for the host device
scsi: revamp host device handling
snic: use reserved commands
snic: use tagset iter for traversing commands
scsi: implement reserved command handling
hpsa: move hpsa_hba_inquiry after scsi_add_host()
hpsa: use reserved commands
hpsa: use scsi_host_busy_iter() to traverse outstanding commands
hpsa: drop refcount field from CommandList
aacraid: move scsi_add_host()
aacraid: store target id in host_scribble
aacraid: use scsi_get_internal_cmd()
aacraid: use scsi_host_busy_iter() to traverse outstanding commands

 drivers/scsi/aacraid/aachba.c   | 137 ++---
 drivers/scsi/aacraid/aacraid.h  |  10 +-
 drivers/scsi/aacraid/commctrl.c |  25 +-
 drivers/scsi/aacraid/comminit.c |   2 +-
 drivers/scsi/aacraid/commsup.c  | 118 ++--
 drivers/scsi/aacraid/dpcsup.c   |   2 +-
 drivers/scsi/aacraid/linit.c    | 175 +++---
 drivers/scsi/fnic/fnic_scsi.c   | 927 +++++++++++++++-----------------
 drivers/scsi/hosts.c            |   3 +
 drivers/scsi/hpsa.c             | 365 ++++++-------
 drivers/scsi/hpsa.h             |   3 +-
 drivers/scsi/hpsa_cmd.h         |  10 -
 drivers/scsi/scsi_devinfo.c     |   1 +
 drivers/scsi/scsi_lib.c         |  48 +-
 drivers/scsi/scsi_scan.c        |  96 ++--
 drivers/scsi/scsi_sysfs.c       |   5 +-
 drivers/scsi/snic/snic.h        |   4 +-
 drivers/scsi/snic/snic_main.c   |   7 +
 drivers/scsi/snic/snic_scsi.c   | 524 +++++++++---------
 include/scsi/scsi_device.h      |   3 +
 include/scsi/scsi_host.h        |  36 +-
 21 files changed, 1234 insertions(+), 1267 deletions(-)
Signature · 1 line
2.29.2
#1 lore raw

PATCH v8 1/18 fnic: kill 'exclude_id' argument to fnic_cleanup_io()

'exclude_id' is always SCSI_NO_TAG, which will never be reached when traversing the list of tags.

---
 drivers/scsi/fnic/fnic_scsi.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
drivers/scsi/fnic/fnic_scsi.c +3 -6
index e619a82f921b..f41d1b1c2e39 100644
@@ -102,7 +102,7 @@ static const char *fnic_fcpio_status_to_str(unsigned int status)
 	return fcpio_status_str[status];
 }
 
-static void fnic_cleanup_io(struct fnic *fnic, int exclude_id);
+static void fnic_cleanup_io(struct fnic *fnic);
 
 static inline spinlock_t *fnic_io_lock_hash(struct fnic *fnic,
 					    struct scsi_cmnd *sc)
@@ -638,7 +638,7 @@ static int fnic_fcpio_fw_reset_cmpl_handler(struct fnic *fnic,
 	atomic64_inc(&reset_stats->fw_reset_completions);
 
 	/* Clean up all outstanding io requests */
-	fnic_cleanup_io(fnic, SCSI_NO_TAG);
+	fnic_cleanup_io(fnic);
 
 	atomic64_set(&fnic->fnic_stats.fw_stats.active_fw_reqs, 0);
 	atomic64_set(&fnic->fnic_stats.io_stats.active_ios, 0);
@@ -1361,7 +1361,7 @@ int fnic_wq_copy_cmpl_handler(struct fnic *fnic, int copy_work_to_do)
 	return wq_work_done;
 }
 
-static void fnic_cleanup_io(struct fnic *fnic, int exclude_id)
+static void fnic_cleanup_io(struct fnic *fnic)
 {
 	int i;
 	struct fnic_io_req *io_req;
@@ -1372,9 +1372,6 @@ static void fnic_cleanup_io(struct fnic *fnic, int exclude_id)
 	struct fnic_stats *fnic_stats = &fnic->fnic_stats;
 
 	for (i = 0; i < fnic->fnic_max_tag_id; i++) {
-		if (i == exclude_id)
-			continue;
-
 		io_lock = fnic_io_lock_tag(fnic, i);
 		spin_lock_irqsave(io_lock, flags);
 		sc = scsi_host_find_tag(fnic->lport->host, i);
Signature · 1 line
2.29.2
Reviewed-by on v8 1/18
#2 lore raw

Re: [PATCH 01/18] fnic: kill 'exclude_id' argument to fnic_cleanup_io()

Quoted 2 lines · Hannes Reinecke

'exclude_id' is always SCSI_NO_TAG, which will never be reached when traversing the list of tags.

#4 lore raw

PATCH v8 2/18 fnic: use scsi_host_busy_iter() to traverse commands

In reply to Hannes Reinecke

Use scsi_host_busy_iter() to traverse commands instead of hand-crafted routines walking the command list.

---
 drivers/scsi/fnic/fnic_scsi.c | 821 ++++++++++++++++------------------
 1 file changed, 375 insertions(+), 446 deletions(-)
drivers/scsi/fnic/fnic_scsi.c +375 -446
index f41d1b1c2e39..762cc8bd2653 100644
@@ -1361,90 +1361,90 @@ int fnic_wq_copy_cmpl_handler(struct fnic *fnic, int copy_work_to_do)
 	return wq_work_done;
 }
 
-static void fnic_cleanup_io(struct fnic *fnic)
+static bool fnic_cleanup_io_iter(struct scsi_cmnd *sc, void *data,
+				 bool reserved)
 {
-	int i;
+	struct fnic *fnic = data;
 	struct fnic_io_req *io_req;
 	unsigned long flags = 0;
-	struct scsi_cmnd *sc;
 	spinlock_t *io_lock;
 	unsigned long start_time = 0;
 	struct fnic_stats *fnic_stats = &fnic->fnic_stats;
 
-	for (i = 0; i < fnic->fnic_max_tag_id; i++) {
-		io_lock = fnic_io_lock_tag(fnic, i);
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(fnic->lport->host, i);
-		if (!sc) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
-		if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
-			!(CMD_FLAGS(sc) & FNIC_DEV_RST_DONE)) {
-			/*
-			 * We will be here only when FW completes reset
-			 * without sending completions for outstanding ios.
-			 */
-			CMD_FLAGS(sc) |= FNIC_DEV_RST_DONE;
-			if (io_req && io_req->dr_done)
-				complete(io_req->dr_done);
-			else if (io_req && io_req->abts_done)
-				complete(io_req->abts_done);
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		} else if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-		if (!io_req) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-
-		CMD_SP(sc) = NULL;
-
-		spin_unlock_irqrestore(io_lock, flags);
+	io_lock = fnic_io_lock_tag(fnic, sc->request->tag);
+	spin_lock_irqsave(io_lock, flags);
 
+	io_req = (struct fnic_io_req *)CMD_SP(sc);
+	if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
+	    !(CMD_FLAGS(sc) & FNIC_DEV_RST_DONE)) {
 		/*
-		 * If there is a scsi_cmnd associated with this io_req, then
-		 * free the corresponding state
+		 * We will be here only when FW completes reset
+		 * without sending completions for outstanding ios.
 		 */
-		start_time = io_req->start_time;
-		fnic_release_ioreq_buf(fnic, io_req, sc);
-		mempool_free(io_req, fnic->io_req_pool);
+		CMD_FLAGS(sc) |= FNIC_DEV_RST_DONE;
+		if (io_req && io_req->dr_done)
+			complete(io_req->dr_done);
+		else if (io_req && io_req->abts_done)
+			complete(io_req->abts_done);
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	} else if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
+	if (!io_req) {
+		spin_unlock_irqrestore(io_lock, flags);
+		goto cleanup_scsi_cmd;
+	}
 
-		sc->result = DID_TRANSPORT_DISRUPTED << 16;
-		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-			      "%s: tag:0x%x : sc:0x%p duration = %lu DID_TRANSPORT_DISRUPTED\n",
-			      __func__, sc->request->tag, sc,
-			      (jiffies - start_time));
+	CMD_SP(sc) = NULL;
 
-		if (atomic64_read(&fnic->io_cmpl_skip))
-			atomic64_dec(&fnic->io_cmpl_skip);
-		else
-			atomic64_inc(&fnic_stats->io_stats.io_completions);
+	spin_unlock_irqrestore(io_lock, flags);
 
-		/* Complete the command to SCSI */
-		if (sc->scsi_done) {
-			if (!(CMD_FLAGS(sc) & FNIC_IO_ISSUED))
-				shost_printk(KERN_ERR, fnic->lport->host,
-				"Calling done for IO not issued to fw: tag:0x%x sc:0x%p\n",
-				 sc->request->tag, sc);
+	/*
+	 * If there is a scsi_cmnd associated with this io_req, then
+	 * free the corresponding state
+	 */
+	start_time = io_req->start_time;
+	fnic_release_ioreq_buf(fnic, io_req, sc);
+	mempool_free(io_req, fnic->io_req_pool);
 
-			FNIC_TRACE(fnic_cleanup_io,
-				  sc->device->host->host_no, i, sc,
-				  jiffies_to_msecs(jiffies - start_time),
-				  0, ((u64)sc->cmnd[0] << 32 |
-				  (u64)sc->cmnd[2] << 24 |
-				  (u64)sc->cmnd[3] << 16 |
-				  (u64)sc->cmnd[4] << 8 | sc->cmnd[5]),
-				  (((u64)CMD_FLAGS(sc) << 32) | CMD_STATE(sc)));
+cleanup_scsi_cmd:
+	sc->result = DID_TRANSPORT_DISRUPTED << 16;
+	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
+		      "fnic_cleanup_io: tag:0x%x : sc:0x%p duration = %lu DID_TRANSPORT_DISRUPTED\n",
+		      sc->request->tag, sc, (jiffies - start_time));
 
-			sc->scsi_done(sc);
-		}
+	if (atomic64_read(&fnic->io_cmpl_skip))
+		atomic64_dec(&fnic->io_cmpl_skip);
+	else
+		atomic64_inc(&fnic_stats->io_stats.io_completions);
+
+	/* Complete the command to SCSI */
+	if (sc->scsi_done) {
+		if (!(CMD_FLAGS(sc) & FNIC_IO_ISSUED))
+			shost_printk(KERN_ERR, fnic->lport->host,
+				     "Calling done for IO not issued to fw: tag:0x%x sc:0x%p\n",
+				     sc->request->tag, sc);
+
+		FNIC_TRACE(fnic_cleanup_io,
+			   sc->device->host->host_no, sc->request->tag, sc,
+			   jiffies_to_msecs(jiffies - start_time),
+			   0, ((u64)sc->cmnd[0] << 32 |
+			       (u64)sc->cmnd[2] << 24 |
+			       (u64)sc->cmnd[3] << 16 |
+			       (u64)sc->cmnd[4] << 8 | sc->cmnd[5]),
+			   (((u64)CMD_FLAGS(sc) << 32) | CMD_STATE(sc)));
+
+		sc->scsi_done(sc);
 	}
+	return true;
+}
+
+static void fnic_cleanup_io(struct fnic *fnic)
+{
+	scsi_host_busy_iter(fnic->lport->host,
+			    fnic_cleanup_io_iter, fnic);
 }
 
 void fnic_wq_copy_cleanup_handler(struct vnic_wq_copy *wq,
@@ -1555,143 +1555,141 @@ static inline int fnic_queue_abort_io_req(struct fnic *fnic, int tag,
 	return 0;
 }
 
-static void fnic_rport_exch_reset(struct fnic *fnic, u32 port_id)
+struct fnic_rport_abort_io_iter_data {
+	struct fnic *fnic;
+	u32 port_id;
+	int term_cnt;
+};
+
+static bool fnic_rport_abort_io_iter(struct scsi_cmnd *sc, void *data,
+				     bool reserved)
 {
-	int tag;
-	int abt_tag;
-	int term_cnt = 0;
+	struct fnic_rport_abort_io_iter_data *iter_data = data;
+	struct fnic *fnic = iter_data->fnic;
+	int abt_tag = sc->request->tag;
 	struct fnic_io_req *io_req;
 	spinlock_t *io_lock;
 	unsigned long flags;
-	struct scsi_cmnd *sc;
 	struct reset_stats *reset_stats = &fnic->fnic_stats.reset_stats;
 	struct terminate_stats *term_stats = &fnic->fnic_stats.term_stats;
 	struct scsi_lun fc_lun;
 	enum fnic_ioreq_state old_ioreq_state;
 
-	FNIC_SCSI_DBG(KERN_DEBUG,
-		      fnic->lport->host,
-		      "fnic_rport_exch_reset called portid 0x%06x\n",
-		      port_id);
-
-	if (fnic->in_remove)
-		return;
-
-	for (tag = 0; tag < fnic->fnic_max_tag_id; tag++) {
-		abt_tag = tag;
-		io_lock = fnic_io_lock_tag(fnic, tag);
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(fnic->lport->host, tag);
-		if (!sc) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
+	io_lock = fnic_io_lock_tag(fnic, abt_tag);
+	spin_lock_irqsave(io_lock, flags);
 
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
+	io_req = (struct fnic_io_req *)CMD_SP(sc);
 
-		if (!io_req || io_req->port_id != port_id) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
+	if (!io_req || io_req->port_id != iter_data->port_id) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
-		if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
-			(!(CMD_FLAGS(sc) & FNIC_DEV_RST_ISSUED))) {
-			FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
+	if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
+	    (!(CMD_FLAGS(sc) & FNIC_DEV_RST_ISSUED))) {
+		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
 			"fnic_rport_exch_reset dev rst not pending sc 0x%p\n",
 			sc);
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
-		/*
-		 * Found IO that is still pending with firmware and
-		 * belongs to rport that went away
-		 */
-		if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-		if (io_req->abts_done) {
-			shost_printk(KERN_ERR, fnic->lport->host,
+	/*
+	 * Found IO that is still pending with firmware and
+	 * belongs to rport that went away
+	 */
+	if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
+	if (io_req->abts_done) {
+		shost_printk(KERN_ERR, fnic->lport->host,
 			"fnic_rport_exch_reset: io_req->abts_done is set "
 			"state is %s\n",
 			fnic_ioreq_state_to_str(CMD_STATE(sc)));
-		}
+	}
 
-		if (!(CMD_FLAGS(sc) & FNIC_IO_ISSUED)) {
-			shost_printk(KERN_ERR, fnic->lport->host,
-				  "rport_exch_reset "
-				  "IO not yet issued %p tag 0x%x flags "
-				  "%x state %d\n",
-				  sc, tag, CMD_FLAGS(sc), CMD_STATE(sc));
-		}
-		old_ioreq_state = CMD_STATE(sc);
-		CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
-		CMD_ABTS_STATUS(sc) = FCPIO_INVALID_CODE;
-		if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
-			atomic64_inc(&reset_stats->device_reset_terminates);
-			abt_tag = (tag | FNIC_TAG_DEV_RST);
-			FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-			"fnic_rport_exch_reset dev rst sc 0x%p\n",
-			sc);
-		}
+	if (!(CMD_FLAGS(sc) & FNIC_IO_ISSUED)) {
+		shost_printk(KERN_ERR, fnic->lport->host,
+			     "rport_exch_reset "
+			     "IO not yet issued %p tag 0x%x flags "
+			     "%x state %d\n",
+			     sc, abt_tag, CMD_FLAGS(sc), CMD_STATE(sc));
+	}
+	old_ioreq_state = CMD_STATE(sc);
+	CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
+	CMD_ABTS_STATUS(sc) = FCPIO_INVALID_CODE;
+	if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
+		atomic64_inc(&reset_stats->device_reset_terminates);
+		abt_tag |= FNIC_TAG_DEV_RST;
+	}
+	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
+		      "fnic_rport_exch_reset dev rst sc 0x%p\n", sc);
+	BUG_ON(io_req->abts_done);
 
-		BUG_ON(io_req->abts_done);
+	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
+		      "fnic_rport_reset_exch: Issuing abts\n");
 
-		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-			      "fnic_rport_reset_exch: Issuing abts\n");
+	spin_unlock_irqrestore(io_lock, flags);
 
+	/* Now queue the abort command to firmware */
+	int_to_scsilun(sc->device->lun, &fc_lun);
+
+	if (fnic_queue_abort_io_req(fnic, abt_tag,
+				    FCPIO_ITMF_ABT_TASK_TERM,
+				    fc_lun.scsi_lun, io_req)) {
+		/*
+		 * Revert the cmd state back to old state, if
+		 * it hasn't changed in between. This cmd will get
+		 * aborted later by scsi_eh, or cleaned up during
+		 * lun reset
+		 */
+		spin_lock_irqsave(io_lock, flags);
+		if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING)
+			CMD_STATE(sc) = old_ioreq_state;
+		spin_unlock_irqrestore(io_lock, flags);
+	} else {
+		spin_lock_irqsave(io_lock, flags);
+		if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET)
+			CMD_FLAGS(sc) |= FNIC_DEV_RST_TERM_ISSUED;
+		else
+			CMD_FLAGS(sc) |= FNIC_IO_INTERNAL_TERM_ISSUED;
 		spin_unlock_irqrestore(io_lock, flags);
+		atomic64_inc(&term_stats->terminates);
+		iter_data->term_cnt++;
+	}
+	return true;
+}
 
-		/* Now queue the abort command to firmware */
-		int_to_scsilun(sc->device->lun, &fc_lun);
+static void fnic_rport_exch_reset(struct fnic *fnic, u32 port_id)
+{
+	struct terminate_stats *term_stats = &fnic->fnic_stats.term_stats;
+	struct fnic_rport_abort_io_iter_data iter_data = {
+		.fnic = fnic,
+		.port_id = port_id,
+		.term_cnt = 0,
+	};
 
-		if (fnic_queue_abort_io_req(fnic, abt_tag,
-					    FCPIO_ITMF_ABT_TASK_TERM,
-					    fc_lun.scsi_lun, io_req)) {
-			/*
-			 * Revert the cmd state back to old state, if
-			 * it hasn't changed in between. This cmd will get
-			 * aborted later by scsi_eh, or cleaned up during
-			 * lun reset
-			 */
-			spin_lock_irqsave(io_lock, flags);
-			if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING)
-				CMD_STATE(sc) = old_ioreq_state;
-			spin_unlock_irqrestore(io_lock, flags);
-		} else {
-			spin_lock_irqsave(io_lock, flags);
-			if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET)
-				CMD_FLAGS(sc) |= FNIC_DEV_RST_TERM_ISSUED;
-			else
-				CMD_FLAGS(sc) |= FNIC_IO_INTERNAL_TERM_ISSUED;
-			spin_unlock_irqrestore(io_lock, flags);
-			atomic64_inc(&term_stats->terminates);
-			term_cnt++;
-		}
-	}
-	if (term_cnt > atomic64_read(&term_stats->max_terminates))
-		atomic64_set(&term_stats->max_terminates, term_cnt);
+	FNIC_SCSI_DBG(KERN_DEBUG,
+		      fnic->lport->host,
+		      "fnic_rport_exch_reset called portid 0x%06x\n",
+		      port_id);
+
+	if (fnic->in_remove)
+		return;
+
+	scsi_host_busy_iter(fnic->lport->host, fnic_rport_abort_io_iter,
+			    &iter_data);
+	if (iter_data.term_cnt > atomic64_read(&term_stats->max_terminates))
+		atomic64_set(&term_stats->max_terminates, iter_data.term_cnt);
 
 }
 
 void fnic_terminate_rport_io(struct fc_rport *rport)
 {
-	int tag;
-	int abt_tag;
-	int term_cnt = 0;
-	struct fnic_io_req *io_req;
-	spinlock_t *io_lock;
-	unsigned long flags;
-	struct scsi_cmnd *sc;
-	struct scsi_lun fc_lun;
 	struct fc_rport_libfc_priv *rdata;
 	struct fc_lport *lport;
 	struct fnic *fnic;
-	struct fc_rport *cmd_rport;
-	struct reset_stats *reset_stats;
-	struct terminate_stats *term_stats;
-	enum fnic_ioreq_state old_ioreq_state;
 
 	if (!rport) {
 		printk(KERN_ERR "fnic_terminate_rport_io: rport is NULL\n");
@@ -1719,108 +1717,7 @@ void fnic_terminate_rport_io(struct fc_rport *rport)
 	if (fnic->in_remove)
 		return;
 
-	reset_stats = &fnic->fnic_stats.reset_stats;
-	term_stats = &fnic->fnic_stats.term_stats;
-
-	for (tag = 0; tag < fnic->fnic_max_tag_id; tag++) {
-		abt_tag = tag;
-		io_lock = fnic_io_lock_tag(fnic, tag);
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(fnic->lport->host, tag);
-		if (!sc) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
-		if (!io_req) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-
-		cmd_rport = starget_to_rport(scsi_target(sc->device));
-		if (rport != cmd_rport) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-
-		if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
-			(!(CMD_FLAGS(sc) & FNIC_DEV_RST_ISSUED))) {
-			FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-			"fnic_terminate_rport_io dev rst not pending sc 0x%p\n",
-			sc);
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-		/*
-		 * Found IO that is still pending with firmware and
-		 * belongs to rport that went away
-		 */
-		if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-		if (io_req->abts_done) {
-			shost_printk(KERN_ERR, fnic->lport->host,
-			"fnic_terminate_rport_io: io_req->abts_done is set "
-			"state is %s\n",
-			fnic_ioreq_state_to_str(CMD_STATE(sc)));
-		}
-		if (!(CMD_FLAGS(sc) & FNIC_IO_ISSUED)) {
-			FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
-				  "fnic_terminate_rport_io "
-				  "IO not yet issued %p tag 0x%x flags "
-				  "%x state %d\n",
-				  sc, tag, CMD_FLAGS(sc), CMD_STATE(sc));
-		}
-		old_ioreq_state = CMD_STATE(sc);
-		CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
-		CMD_ABTS_STATUS(sc) = FCPIO_INVALID_CODE;
-		if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
-			atomic64_inc(&reset_stats->device_reset_terminates);
-			abt_tag = (tag | FNIC_TAG_DEV_RST);
-			FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-			"fnic_terminate_rport_io dev rst sc 0x%p\n", sc);
-		}
-
-		BUG_ON(io_req->abts_done);
-
-		FNIC_SCSI_DBG(KERN_DEBUG,
-			      fnic->lport->host,
-			      "fnic_terminate_rport_io: Issuing abts\n");
-
-		spin_unlock_irqrestore(io_lock, flags);
-
-		/* Now queue the abort command to firmware */
-		int_to_scsilun(sc->device->lun, &fc_lun);
-
-		if (fnic_queue_abort_io_req(fnic, abt_tag,
-					    FCPIO_ITMF_ABT_TASK_TERM,
-					    fc_lun.scsi_lun, io_req)) {
-			/*
-			 * Revert the cmd state back to old state, if
-			 * it hasn't changed in between. This cmd will get
-			 * aborted later by scsi_eh, or cleaned up during
-			 * lun reset
-			 */
-			spin_lock_irqsave(io_lock, flags);
-			if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING)
-				CMD_STATE(sc) = old_ioreq_state;
-			spin_unlock_irqrestore(io_lock, flags);
-		} else {
-			spin_lock_irqsave(io_lock, flags);
-			if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET)
-				CMD_FLAGS(sc) |= FNIC_DEV_RST_TERM_ISSUED;
-			else
-				CMD_FLAGS(sc) |= FNIC_IO_INTERNAL_TERM_ISSUED;
-			spin_unlock_irqrestore(io_lock, flags);
-			atomic64_inc(&term_stats->terminates);
-			term_cnt++;
-		}
-	}
-	if (term_cnt > atomic64_read(&term_stats->max_terminates))
-		atomic64_set(&term_stats->max_terminates, term_cnt);
-
+	fnic_rport_exch_reset(fnic, rport->port_id);
 }
 
 /*
@@ -2115,165 +2012,183 @@ static inline int fnic_queue_dr_io_req(struct fnic *fnic,
 	return ret;
 }
 
-/*
- * Clean up any pending aborts on the lun
- * For each outstanding IO on this lun, whose abort is not completed by fw,
- * issue a local abort. Wait for abort to complete. Return 0 if all commands
- * successfully aborted, 1 otherwise
- */
-static int fnic_clean_pending_aborts(struct fnic *fnic,
-				     struct scsi_cmnd *lr_sc,
-					 bool new_sc)
+struct fnic_pending_aborts_iter_data {
+	struct fnic *fnic;
+	struct scsi_cmnd *lr_sc;
+	struct scsi_device *lun_dev;
+	int ret;
+};
 
+static bool fnic_pending_aborts_iter(struct scsi_cmnd *sc,
+				     void *data, bool reserved)
 {
-	int tag, abt_tag;
+	struct fnic_pending_aborts_iter_data *iter_data = data;
+	struct fnic *fnic = iter_data->fnic;
+	struct scsi_device *lun_dev = iter_data->lun_dev;
+	int abt_tag = sc->request->tag;
 	struct fnic_io_req *io_req;
 	spinlock_t *io_lock;
 	unsigned long flags;
-	int ret = 0;
-	struct scsi_cmnd *sc;
 	struct scsi_lun fc_lun;
-	struct scsi_device *lun_dev = lr_sc->device;
 	DECLARE_COMPLETION_ONSTACK(tm_done);
 	enum fnic_ioreq_state old_ioreq_state;
 
-	for (tag = 0; tag < fnic->fnic_max_tag_id; tag++) {
-		io_lock = fnic_io_lock_tag(fnic, tag);
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(fnic->lport->host, tag);
-		/*
-		 * ignore this lun reset cmd if issued using new SC
-		 * or cmds that do not belong to this lun
-		 */
-		if (!sc || ((sc == lr_sc) && new_sc) || sc->device != lun_dev) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
+	if (sc == iter_data->lr_sc || sc->device != lun_dev)
+		return true;
+	if (reserved)
+		return true;
 
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
-
-		if (!io_req || sc->device != lun_dev) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-
-		/*
-		 * Found IO that is still pending with firmware and
-		 * belongs to the LUN that we are resetting
-		 */
-		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-			      "Found IO in %s on lun\n",
-			      fnic_ioreq_state_to_str(CMD_STATE(sc)));
-
-		if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
-		if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
-			(!(CMD_FLAGS(sc) & FNIC_DEV_RST_ISSUED))) {
-			FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
-				"%s dev rst not pending sc 0x%p\n", __func__,
-				sc);
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
+	io_lock = fnic_io_lock_tag(fnic, abt_tag);
+	spin_lock_irqsave(io_lock, flags);
+	io_req = (struct fnic_io_req *)CMD_SP(sc);
+	if (!io_req) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
-		if (io_req->abts_done)
-			shost_printk(KERN_ERR, fnic->lport->host,
-			  "%s: io_req->abts_done is set state is %s\n",
-			  __func__, fnic_ioreq_state_to_str(CMD_STATE(sc)));
-		old_ioreq_state = CMD_STATE(sc);
-		/*
-		 * Any pending IO issued prior to reset is expected to be
-		 * in abts pending state, if not we need to set
-		 * FNIC_IOREQ_ABTS_PENDING to indicate the IO is abort pending.
-		 * When IO is completed, the IO will be handed over and
-		 * handled in this function.
-		 */
-		CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
+	/*
+	 * Found IO that is still pending with firmware and
+	 * belongs to the LUN that we are resetting
+	 */
+	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
+		      "Found IO in %s on lun\n",
+		      fnic_ioreq_state_to_str(CMD_STATE(sc)));
 
-		BUG_ON(io_req->abts_done);
+	if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
+	if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
+	    (!(CMD_FLAGS(sc) & FNIC_DEV_RST_ISSUED))) {
+		FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
+			      "%s dev rst not pending sc 0x%p\n", __func__,
+			      sc);
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
-		abt_tag = tag;
-		if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
-			abt_tag |= FNIC_TAG_DEV_RST;
-			FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
-				  "%s: dev rst sc 0x%p\n", __func__, sc);
-		}
+	if (io_req->abts_done)
+		shost_printk(KERN_ERR, fnic->lport->host,
+			     "%s: io_req->abts_done is set state is %s\n",
+			     __func__, fnic_ioreq_state_to_str(CMD_STATE(sc)));
+	old_ioreq_state = CMD_STATE(sc);
+	/*
+	 * Any pending IO issued prior to reset is expected to be
+	 * in abts pending state, if not we need to set
+	 * FNIC_IOREQ_ABTS_PENDING to indicate the IO is abort pending.
+	 * When IO is completed, the IO will be handed over and
+	 * handled in this function.
+	 */
+	CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
 
-		CMD_ABTS_STATUS(sc) = FCPIO_INVALID_CODE;
-		io_req->abts_done = &tm_done;
-		spin_unlock_irqrestore(io_lock, flags);
+	BUG_ON(io_req->abts_done);
 
-		/* Now queue the abort command to firmware */
-		int_to_scsilun(sc->device->lun, &fc_lun);
+	if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET) {
+		abt_tag |= FNIC_TAG_DEV_RST;
+		FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
+			      "%s: dev rst sc 0x%p\n", __func__, sc);
+	}
 
-		if (fnic_queue_abort_io_req(fnic, abt_tag,
-					    FCPIO_ITMF_ABT_TASK_TERM,
-					    fc_lun.scsi_lun, io_req)) {
-			spin_lock_irqsave(io_lock, flags);
-			io_req = (struct fnic_io_req *)CMD_SP(sc);
-			if (io_req)
-				io_req->abts_done = NULL;
-			if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING)
-				CMD_STATE(sc) = old_ioreq_state;
-			spin_unlock_irqrestore(io_lock, flags);
-			ret = 1;
-			goto clean_pending_aborts_end;
-		} else {
-			spin_lock_irqsave(io_lock, flags);
-			if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET)
-				CMD_FLAGS(sc) |= FNIC_DEV_RST_TERM_ISSUED;
-			spin_unlock_irqrestore(io_lock, flags);
-		}
-		CMD_FLAGS(sc) |= FNIC_IO_INTERNAL_TERM_ISSUED;
+	CMD_ABTS_STATUS(sc) = FCPIO_INVALID_CODE;
+	io_req->abts_done = &tm_done;
+	spin_unlock_irqrestore(io_lock, flags);
 
-		wait_for_completion_timeout(&tm_done,
-					    msecs_to_jiffies
-					    (fnic->config.ed_tov));
+	/* Now queue the abort command to firmware */
+	int_to_scsilun(sc->device->lun, &fc_lun);
 
-		/* Recheck cmd state to check if it is now aborted */
+	if (fnic_queue_abort_io_req(fnic, abt_tag,
+				    FCPIO_ITMF_ABT_TASK_TERM,
+				    fc_lun.scsi_lun, io_req)) {
 		spin_lock_irqsave(io_lock, flags);
 		io_req = (struct fnic_io_req *)CMD_SP(sc);
-		if (!io_req) {
-			spin_unlock_irqrestore(io_lock, flags);
-			CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_REQ_NULL;
-			continue;
-		}
+		if (io_req)
+			io_req->abts_done = NULL;
+		if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING)
+			CMD_STATE(sc) = old_ioreq_state;
+		spin_unlock_irqrestore(io_lock, flags);
+		iter_data->ret = FAILED;
+		return false;
+	} else {
+		spin_lock_irqsave(io_lock, flags);
+		if (CMD_FLAGS(sc) & FNIC_DEVICE_RESET)
+			CMD_FLAGS(sc) |= FNIC_DEV_RST_TERM_ISSUED;
+		spin_unlock_irqrestore(io_lock, flags);
+	}
+	CMD_FLAGS(sc) |= FNIC_IO_INTERNAL_TERM_ISSUED;
 
-		io_req->abts_done = NULL;
+	wait_for_completion_timeout(&tm_done, msecs_to_jiffies
+				    (fnic->config.ed_tov));
 
-		/* if abort is still pending with fw, fail */
-		if (CMD_ABTS_STATUS(sc) == FCPIO_INVALID_CODE) {
-			spin_unlock_irqrestore(io_lock, flags);
-			CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_DONE;
-			ret = 1;
-			goto clean_pending_aborts_end;
-		}
-		CMD_STATE(sc) = FNIC_IOREQ_ABTS_COMPLETE;
+	/* Recheck cmd state to check if it is now aborted */
+	spin_lock_irqsave(io_lock, flags);
+	io_req = (struct fnic_io_req *)CMD_SP(sc);
+	if (!io_req) {
+		spin_unlock_irqrestore(io_lock, flags);
+		CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_REQ_NULL;
+		return true;
+	}
 
-		/* original sc used for lr is handled by dev reset code */
-		if (sc != lr_sc)
-			CMD_SP(sc) = NULL;
+	io_req->abts_done = NULL;
+
+	/* if abort is still pending with fw, fail */
+	if (CMD_ABTS_STATUS(sc) == FCPIO_INVALID_CODE) {
 		spin_unlock_irqrestore(io_lock, flags);
+		CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_DONE;
+		iter_data->ret = FAILED;
+		return false;
+	}
+	CMD_STATE(sc) = FNIC_IOREQ_ABTS_COMPLETE;
 
-		/* original sc used for lr is handled by dev reset code */
-		if (sc != lr_sc) {
-			fnic_release_ioreq_buf(fnic, io_req, sc);
-			mempool_free(io_req, fnic->io_req_pool);
-		}
+	/* original sc used for lr is handled by dev reset code */
+	if (sc != iter_data->lr_sc)
+		CMD_SP(sc) = NULL;
+	spin_unlock_irqrestore(io_lock, flags);
 
-		/*
-		 * Any IO is returned during reset, it needs to call scsi_done
-		 * to return the scsi_cmnd to upper layer.
-		 */
-		if (sc->scsi_done) {
-			/* Set result to let upper SCSI layer retry */
-			sc->result = DID_RESET << 16;
-			sc->scsi_done(sc);
-		}
+	/* original sc used for lr is handled by dev reset code */
+	if (sc != iter_data->lr_sc) {
+		fnic_release_ioreq_buf(fnic, io_req, sc);
+		mempool_free(io_req, fnic->io_req_pool);
 	}
 
+	/*
+	 * Any IO is returned during reset, it needs to call scsi_done
+	 * to return the scsi_cmnd to upper layer.
+	 */
+	if (sc->scsi_done) {
+		/* Set result to let upper SCSI layer retry */
+		sc->result = DID_RESET << 16;
+		sc->scsi_done(sc);
+	}
+	return true;
+}
+
+/*
+ * Clean up any pending aborts on the lun
+ * For each outstanding IO on this lun, whose abort is not completed by fw,
+ * issue a local abort. Wait for abort to complete. Return 0 if all commands
+ * successfully aborted, 1 otherwise
+ */
+static int fnic_clean_pending_aborts(struct fnic *fnic,
+				     struct scsi_cmnd *lr_sc,
+				     bool new_sc)
+
+{
+	int ret = SUCCESS;
+	struct fnic_pending_aborts_iter_data iter_data = {
+		.fnic = fnic,
+		.lun_dev = lr_sc->device,
+		.ret = SUCCESS,
+	};
+
+	if (new_sc)
+		iter_data.lr_sc = lr_sc;
+
+	scsi_host_busy_iter(fnic->lport->host,
+			    fnic_pending_aborts_iter, &iter_data);
+	if (iter_data.ret == FAILED) {
+		ret = iter_data.ret;
+		goto clean_pending_aborts_end;
+	}
 	schedule_timeout(msecs_to_jiffies(2 * fnic->config.ed_tov));
 
 	/* walk again to check, if IOs are still pending in fw */
@@ -2772,58 +2687,72 @@ void fnic_exch_mgr_reset(struct fc_lport *lp, u32 sid, u32 did)
 
 }
 
-/*
- * fnic_is_abts_pending() is a helper function that
- * walks through tag map to check if there is any IOs pending,if there is one,
- * then it returns 1 (true), otherwise 0 (false)
- * if @lr_sc is non NULL, then it checks IOs specific to particular LUN,
- * otherwise, it checks for all IOs.
- */
-int fnic_is_abts_pending(struct fnic *fnic, struct scsi_cmnd *lr_sc)
+static bool fnic_abts_pending_iter(struct scsi_cmnd *sc, void *data,
+				   bool reserved)
 {
-	int tag;
+	struct fnic_pending_aborts_iter_data *iter_data = data;
+	struct fnic *fnic = iter_data->fnic;
+	int cmd_state;
 	struct fnic_io_req *io_req;
 	spinlock_t *io_lock;
 	unsigned long flags;
-	int ret = 0;
-	struct scsi_cmnd *sc;
-	struct scsi_device *lun_dev = NULL;
 
-	if (lr_sc)
-		lun_dev = lr_sc->device;
+	/*
+	 * ignore this lun reset cmd or cmds that do not belong to
+	 * this lun
+	 */
+	if (iter_data->lr_sc && sc == iter_data->lr_sc)
+		return true;
+	if (iter_data->lun_dev && sc->device != iter_data->lun_dev)
+		return true;
 
-	/* walk again to check, if IOs are still pending in fw */
-	for (tag = 0; tag < fnic->fnic_max_tag_id; tag++) {
-		sc = scsi_host_find_tag(fnic->lport->host, tag);
-		/*
-		 * ignore this lun reset cmd or cmds that do not belong to
-		 * this lun
-		 */
-		if (!sc || (lr_sc && (sc->device != lun_dev || sc == lr_sc)))
-			continue;
+	io_lock = fnic_io_lock_hash(fnic, sc);
+	spin_lock_irqsave(io_lock, flags);
 
-		io_lock = fnic_io_lock_hash(fnic, sc);
-		spin_lock_irqsave(io_lock, flags);
+	io_req = (struct fnic_io_req *)CMD_SP(sc);
+	if (!io_req) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
+	/*
+	 * Found IO that is still pending with firmware and
+	 * belongs to the LUN that we are resetting
+	 */
+	FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
+		      "Found IO in %s on lun\n",
+		      fnic_ioreq_state_to_str(CMD_STATE(sc)));
+	cmd_state = CMD_STATE(sc);
+	spin_unlock_irqrestore(io_lock, flags);
+	if (cmd_state == FNIC_IOREQ_ABTS_PENDING)
+		iter_data->ret = 1;
 
-		if (!io_req || sc->device != lun_dev) {
-			spin_unlock_irqrestore(io_lock, flags);
-			continue;
-		}
+	return iter_data->ret ? false : true;
+}
 
-		/*
-		 * Found IO that is still pending with firmware and
-		 * belongs to the LUN that we are resetting
-		 */
-		FNIC_SCSI_DBG(KERN_INFO, fnic->lport->host,
-			      "Found IO in %s on lun\n",
-			      fnic_ioreq_state_to_str(CMD_STATE(sc)));
+/*
+ * fnic_is_abts_pending() is a helper function that
+ * walks through tag map to check if there is any IOs pending,if there is one,
+ * then it returns 1 (true), otherwise 0 (false)
+ * if @lr_sc is non NULL, then it checks IOs specific to particular LUN,
+ * otherwise, it checks for all IOs.
+ */
+int fnic_is_abts_pending(struct fnic *fnic, struct scsi_cmnd *lr_sc)
+{
+	struct fnic_pending_aborts_iter_data iter_data = {
+		.fnic = fnic,
+		.lun_dev = NULL,
+		.ret = 0,
+	};
 
-		if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING)
-			ret = 1;
-		spin_unlock_irqrestore(io_lock, flags);
+	if (lr_sc) {
+		iter_data.lun_dev = lr_sc->device;
+		iter_data.lr_sc = lr_sc;
 	}
 
-	return ret;
+	/* walk again to check, if IOs are still pending in fw */
+	scsi_host_busy_iter(fnic->lport->host,
+			    fnic_abts_pending_iter, &iter_data);
+
+	return iter_data.ret;
 }
Signature · 1 line
2.29.2
#5 lore raw

Re: [PATCH 02/18] fnic: use scsi_host_busy_iter() to traverse commands

Quoted 2 lines · Hannes Reinecke

Use scsi_host_busy_iter() to traverse commands instead of hand-crafted routines walking the command list.

While the replacement looks like the right thing to do at the micro level, can we take one step back?

Shouldn't completing commands be left entirely to the SCSI EH code instead of messing with it in the driver?

#6 lore raw

Re: [PATCH 02/18] fnic: use scsi_host_busy_iter() to traverse commands

Quoted 10 lines · Christoph Hellwig
Quoted 2 lines · Hannes Reinecke

Use scsi_host_busy_iter() to traverse commands instead of hand-crafted routines walking the command list.

While the replacement looks like the right thing to do at the micro level, can we take one step back?

Shouldn't completing commands be left entirely to the SCSI EH code instead of messing with it in the driver?

Not sure if that'd be easily possible.

Most drivers operate under the assumption that upon return from
eh_device_reset_handler() all commands for that device (except for the reset command itself and the command inducing the reset) will have to be aborted or completed.
As the call is synchronous, the LLDDs have to _wait_ for completion to happen (which, incidentally, is the cause for most of the issues we're having with SCSI EH...), and only then can they return from the callback.

So I can't see how we could delegate this to SCSI EH, as this would mean a break with the current behaviour.

Mind you, I would _love_ to be able to do an asynchronous device reset, seeing that we had _really_ good experience with the asynchronous command abort, but that will be an incompatible change.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#7 lore raw

PATCH v8 3/18 scsi: add scsi_{get,put}_internal_cmd() helper

In reply to Hannes Reinecke

Add helper functions to allow LLDDs to allocate and free internal commands.

---
 drivers/scsi/scsi_lib.c    | 38 ++++++++++++++++++++++++++++++++++++++
 include/scsi/scsi_device.h |  3 +++
 2 files changed, 41 insertions(+)
drivers/scsi/scsi_lib.c +38 -0
index d7c0d5a5f263..f83b04e49bae 100644
@@ -1989,6 +1989,44 @@ void scsi_mq_destroy_tags(struct Scsi_Host *shost)
 	blk_mq_free_tag_set(&shost->tag_set);
 }
 
+/**
+ * scsi_get_internal_cmd - allocate an internal SCSI command
+ * @sdev: SCSI device from which to allocate the command
+ * @op: operation (REQ_OP_SCSI_IN or REQ_OP_SCSI_OUT)
+ * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
+ *
+ * Allocates a SCSI command for internal LLDD use.
+ */
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags)
+{
+	struct request *rq;
+	struct scsi_cmnd *scmd;
+
+	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
+		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+	if (IS_ERR(rq))
+		return NULL;
+	scmd = blk_mq_rq_to_pdu(rq);
+	scmd->request = rq;
+	scmd->device = sdev;
+	return scmd;
+}
+EXPORT_SYMBOL_GPL(scsi_get_internal_cmd);
+
+/**
+ * scsi_put_internal_cmd - free an internal SCSI command
+ * @scmd: SCSI command to be freed
+ */
+void scsi_put_internal_cmd(struct scsi_cmnd *scmd)
+{
+	struct request *rq = blk_mq_rq_from_pdu(scmd);
+
+	blk_mq_free_request(rq);
+}
+EXPORT_SYMBOL_GPL(scsi_put_internal_cmd);
+
 /**
  * scsi_device_from_queue - return sdev associated with a request_queue
  * @q: The request queue to return the sdev from
include/scsi/scsi_device.h +3 -0
index ac6ab16abee7..eaa5414ff220 100644
@@ -462,6 +462,9 @@ static inline int scsi_execute_req(struct scsi_device *sdev,
 	return scsi_execute(sdev, cmd, data_direction, buffer,
 		bufflen, NULL, sshdr, timeout, retries,  0, 0, resid);
 }
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags);
+void scsi_put_internal_cmd(struct scsi_cmnd *scmd);
 extern void sdev_disable_disk_events(struct scsi_device *sdev);
 extern void sdev_enable_disk_events(struct scsi_device *sdev);
 extern int scsi_vpd_lun_id(struct scsi_device *, char *, size_t);
Signature · 1 line
2.29.2
#8 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

Quoted 15 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +15 -0
+/**
+ * scsi_get_internal_cmd - allocate an internal SCSI command
+ * @sdev: SCSI device from which to allocate the command
+ * @op: operation (REQ_OP_SCSI_IN or REQ_OP_SCSI_OUT)
+ * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
+ *
+ * Allocates a SCSI command for internal LLDD use.
+ */
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags)
+{
+	struct request *rq;
+	struct scsi_cmnd *scmd;
+
+	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2016

Quoted 10 lines
drivers/scsi/scsi_lib.c +10 -0
+		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+	if (IS_ERR(rq))
+		return NULL;
+	scmd = blk_mq_rq_to_pdu(rq);
+	scmd->request = rq;
+	scmd->device = sdev;
+	return scmd;
+}
+EXPORT_SYMBOL_GPL(scsi_get_internal_cmd);

Multiple fields that are initialized by the regular command submission path are not initialized by the above function, e.g. scmd->tag. Has it been considered to call scsi_init_command() instead of adding yet another code path that initializes scmd->request?

Thanks,

Bart.

#9 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

Quoted 23 lines · Bart Van Assche
Quoted 22 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +22 -0
+/**
+ * scsi_get_internal_cmd - allocate an internal SCSI command
+ * @sdev: SCSI device from which to allocate the command
+ * @op: operation (REQ_OP_SCSI_IN or REQ_OP_SCSI_OUT)
+ * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
+ *
+ * Allocates a SCSI command for internal LLDD use.
+ */
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags)
+{
+	struct request *rq;
+	struct scsi_cmnd *scmd;
+
+	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
+		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+	if (IS_ERR(rq))
+		return NULL;
+	scmd = blk_mq_rq_to_pdu(rq);
+	scmd->request = rq;
+	scmd->device = sdev;
Quoted 9 lines

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2016

Quoted 3 lines
drivers/scsi/scsi_lib.c +3 -0
+	return scmd;
+}
+EXPORT_SYMBOL_GPL(scsi_get_internal_cmd);

Multiple fields that are initialized by the regular command submission path are not initialized by the above function, e.g. scmd->tag. Has it been considered to call scsi_init_command() instead of adding yet another code path that initializes scmd->request?

Hmm. No, I don't think it's a good idea.
Basic idea is that the SCSI request serves as a container for (non-SCSI) management commands, _and_ that they are submitted directly from within the driver, ie never ever enter ->queue_rq(). As such we don't need an initialisation vie scsi_init_request(), and it would actually be counter-productive as we would be setting up fields which we'll never need anyway, and might need to tear down afterwards.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#10 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

Quoted 33 lines · Hannes Reinecke
Quoted 23 lines · Bart Van Assche
Quoted 22 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +22 -0
+/**
+ * scsi_get_internal_cmd - allocate an internal SCSI command
+ * @sdev: SCSI device from which to allocate the command
+ * @op: operation (REQ_OP_SCSI_IN or REQ_OP_SCSI_OUT)
+ * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
+ *
+ * Allocates a SCSI command for internal LLDD use.
+ */
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+    unsigned int op, blk_mq_req_flags_t flags)
+{
+    struct request *rq;
+    struct scsi_cmnd *scmd;
+
+    WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
+             ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+    rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+    if (IS_ERR(rq))
+        return NULL;
+    scmd = blk_mq_rq_to_pdu(rq);
+    scmd->request = rq;
+    scmd->device = sdev;
Quoted 9 lines

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2016

Quoted 3 lines

+    return scmd;

drivers/scsi/scsi_lib.c +2 -0
+}
+EXPORT_SYMBOL_GPL(scsi_get_internal_cmd);

Multiple fields that are initialized by the regular command submission path are not initialized by the above function, e.g. scmd->tag. Has it been considered to call scsi_init_command() instead of adding yet another code path that initializes scmd->request?

Quoted 3 lines

Hmm. No, I don't think it's a good idea.
Basic idea is that the SCSI request serves as a container for (non-SCSI) management commands, _and_ that they are submitted directly from within the driver, ie never ever enter ->queue_rq(). As such we don't need an initialisation vie scsi_init_request(), and it would actually be counter-productive as we would be setting up fields which we'll never need anyway, and might need to tear down afterwards.

I will note that we also bypass the queue budgeting in blk_mq_ops.{get,put}_budget. I figure that is not an issue...

BTW, any chance of a new version?

Thanks,
John

#11 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

Quoted 34 lines · John Garry
Quoted 33 lines · Hannes Reinecke
Quoted 32 lines · Bart Van Assche
Quoted 15 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +15 -0
+/**
+ * scsi_get_internal_cmd - allocate an internal SCSI command
+ * @sdev: SCSI device from which to allocate the command
+ * @op: operation (REQ_OP_SCSI_IN or REQ_OP_SCSI_OUT)
+ * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
+ *
+ * Allocates a SCSI command for internal LLDD use.
+ */
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+    unsigned int op, blk_mq_req_flags_t flags)
+{
+    struct request *rq;
+    struct scsi_cmnd *scmd;
+
+    WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2016

Quoted 10 lines
drivers/scsi/scsi_lib.c +10 -0
+             ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+    rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+    if (IS_ERR(rq))
+        return NULL;
+    scmd = blk_mq_rq_to_pdu(rq);
+    scmd->request = rq;
+    scmd->device = sdev;
+    return scmd;
+}
+EXPORT_SYMBOL_GPL(scsi_get_internal_cmd);

Multiple fields that are initialized by the regular command submission path are not initialized by the above function, e.g. scmd->tag. Has it been considered to call scsi_init_command() instead of adding yet another code path that initializes scmd->request?

Quoted 8 lines
Quoted 3 lines

Hmm. No, I don't think it's a good idea.
Basic idea is that the SCSI request serves as a container for (non-SCSI) management commands, _and_ that they are submitted directly from within the driver, ie never ever enter ->queue_rq(). As such we don't need an initialisation vie scsi_init_request(), and it would actually be counter-productive as we would be setting up fields which we'll never need anyway, and might need to tear down afterwards.

I will note that we also bypass the queue budgeting in blk_mq_ops.{get,put}_budget. I figure that is not an issue...

BTW, any chance of a new version?

I have _so_ no idea.

The review from Christoph to patch 07/18 he (apparently) changed his mind for the current implementation of using scsi_get_host_dev(), citing an approach I had been implemented for v2 (and which got changed due to his reviews for v2). So no I'm not sure if he retracted on his earlier review, or if he just had forgotten about it.
And before I get clarification from him I can't really move forward, as both reviews contradict each other.

Christoph?

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#12 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

Quoted 5 lines · Hannes Reinecke
Quoted 5 lines

I will note that we also bypass the queue budgeting in blk_mq_ops.{get,put}_budget. I figure that is not an issue...

BTW, any chance of a new version?

Quoted 7 lines

I have _so_ no idea.

The review from Christoph to patch 07/18 he (apparently) changed his mind for the current implementation of using scsi_get_host_dev(), citing an approach I had been implemented for v2 (and which got changed due to his reviews for v2). So no I'm not sure if he retracted on his earlier review, or if he just had forgotten about it.
And before I get clarification from him I can't really move forward, as both reviews contradict each other.

Christoph?

I see.

FWIW, not using scsi_get_host_dev() means that internal command scsi_cmnd.device cannot be set. But then many fields are not set there.

I suppose that if we delete scsi_get_host_dev() now [as it has no user], then that will narrow the choice...

BTW, I was just checking the v7, and we have no queue sysfs folder for that host scsi_device. Regardless of how the queue is created - v2 vs v7 method - I think that we should have that somehow.

Thanks,
John

#13 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

In reply to Hannes Reinecke

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2001

Quoted 2 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +2 -0
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags)

Weird indentation - prototype continuations either use two tabs or are aligned after the opening brace (I generally prefer the former).

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2007

Quoted 6 lines
drivers/scsi/scsi_lib.c +6 -0
+{
+	struct request *rq;
+	struct scsi_cmnd *scmd;
+
+	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
+		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));

Woudn't a simple bool write command make more sense than passing the actual op here?

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2013

Quoted 6 lines
drivers/scsi/scsi_lib.c +6 -0
+	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+	if (IS_ERR(rq))
+		return NULL;
+	scmd = blk_mq_rq_to_pdu(rq);
+	scmd->request = rq;
+	scmd->device = sdev;

Maybe a comment that explains what part of the scmd are initialized and which not would be useful.

#14 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

Quoted 7 lines · Christoph Hellwig

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2001

Quoted 2 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +2 -0
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags)

Weird indentation - prototype continuations either use two tabs or are aligned after the opening brace (I generally prefer the former).

Yeah, I'm never sure how one should be indenting the second line for a function declaration. But I'll fix it.

Quoted 10 lines

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2007

Quoted 6 lines
drivers/scsi/scsi_lib.c +6 -0
+{
+	struct request *rq;
+	struct scsi_cmnd *scmd;
+
+	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
+		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));

Woudn't a simple bool write command make more sense than passing the actual op here?

Yep, can do.

Quoted 10 lines

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2013

Quoted 6 lines
drivers/scsi/scsi_lib.c +6 -0
+	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
+	if (IS_ERR(rq))
+		return NULL;
+	scmd = blk_mq_rq_to_pdu(rq);
+	scmd->request = rq;
+	scmd->device = sdev;

Maybe a comment that explains what part of the scmd are initialized and which not would be useful.

Okay.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#15 lore raw

Re: [PATCH 03/18] scsi: add scsi_{get,put}_internal_cmd() helper

In reply to Hannes Reinecke

drivers/scsi/scsi_lib.c void scsi_mq_destroy_tags(struct Scsi_Host *shost) · line 2007

Quoted 8 lines · Hannes Reinecke
drivers/scsi/scsi_lib.c +8 -0
+struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
+	unsigned int op, blk_mq_req_flags_t flags)
+{
+	struct request *rq;
+	struct scsi_cmnd *scmd;
+
+	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
+		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));

Consider using blk_op_is_scsi() instead of open-coding it.

Thanks,

Bart.

#16 lore raw

PATCH v8 4/18 fnic: use internal commands

In reply to Hannes Reinecke

Remove hack to get tag for the reset command by using internal commands.

---
 drivers/scsi/fnic/fnic_scsi.c | 111 +++++++++++++++-------------------
 1 file changed, 48 insertions(+), 63 deletions(-)
drivers/scsi/fnic/fnic_scsi.c +48 -63
index 762cc8bd2653..77a85fff8fa1 100644
@@ -2169,20 +2169,16 @@ static bool fnic_pending_aborts_iter(struct scsi_cmnd *sc,
  * successfully aborted, 1 otherwise
  */
 static int fnic_clean_pending_aborts(struct fnic *fnic,
-				     struct scsi_cmnd *lr_sc,
-				     bool new_sc)
-
+				     struct scsi_cmnd *lr_sc)
 {
 	int ret = SUCCESS;
 	struct fnic_pending_aborts_iter_data iter_data = {
 		.fnic = fnic,
 		.lun_dev = lr_sc->device,
+		.lr_sc = lr_sc,
 		.ret = SUCCESS,
 	};
 
-	if (new_sc)
-		iter_data.lr_sc = lr_sc;
-
 	scsi_host_busy_iter(fnic->lport->host,
 			    fnic_pending_aborts_iter, &iter_data);
 	if (iter_data.ret == FAILED) {
@@ -2247,19 +2243,19 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 	spinlock_t *io_lock;
 	unsigned long flags;
 	unsigned long start_time = 0;
+	struct scsi_device *sdev = sc->device;
 	struct scsi_lun fc_lun;
 	struct fnic_stats *fnic_stats;
 	struct reset_stats *reset_stats;
 	int tag = 0;
 	DECLARE_COMPLETION_ONSTACK(tm_done);
-	int tag_gen_flag = 0;   /*to track tags allocated by fnic driver*/
-	bool new_sc = 0;
+	struct scsi_cmnd *reset_sc = NULL;
 
 	/* Wait for rport to unblock */
 	fc_block_scsi_eh(sc);
 
 	/* Get local-port, check ready and link up */
-	lp = shost_priv(sc->device->host);
+	lp = shost_priv(sdev->host);
 
 	fnic = lport_priv(lp);
 	fnic_stats = &fnic->fnic_stats;
@@ -2267,10 +2263,10 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 
 	atomic64_inc(&reset_stats->device_resets);
 
-	rport = starget_to_rport(scsi_target(sc->device));
+	rport = starget_to_rport(scsi_target(sdev));
 	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
 		      "Device reset called FCID 0x%x, LUN 0x%llx sc 0x%p\n",
-		      rport->port_id, sc->device->lun, sc);
+		      rport->port_id, sdev->lun, sc);
 
 	if (lp->state != LPORT_ST_READY || !(lp->link_up))
 		goto fnic_device_reset_end;
@@ -2281,42 +2277,31 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 		goto fnic_device_reset_end;
 	}
 
-	CMD_FLAGS(sc) = FNIC_DEVICE_RESET;
-	/* Allocate tag if not present */
+	reset_sc = scsi_get_internal_cmd(sdev, REQ_OP_SCSI_IN,
+					 BLK_MQ_REQ_NOWAIT);
+	if (unlikely(!reset_sc))
+		goto fnic_device_reset_end;
 
-	tag = sc->request->tag;
-	if (unlikely(tag < 0)) {
-		/*
-		 * Really should fix the midlayer to pass in a proper
-		 * request for ioctls...
-		 */
-		tag = fnic_scsi_host_start_tag(fnic, sc);
-		if (unlikely(tag == SCSI_NO_TAG))
-			goto fnic_device_reset_end;
-		tag_gen_flag = 1;
-		new_sc = 1;
-	}
-	io_lock = fnic_io_lock_hash(fnic, sc);
+	CMD_FLAGS(reset_sc) = FNIC_DEVICE_RESET;
+	tag = reset_sc->request->tag;
+	io_lock = fnic_io_lock_hash(fnic, reset_sc);
 	spin_lock_irqsave(io_lock, flags);
-	io_req = (struct fnic_io_req *)CMD_SP(sc);
 
 	/*
-	 * If there is a io_req attached to this command, then use it,
-	 * else allocate a new one.
+	 * Allocate a new io_req.
 	 */
+	io_req = mempool_alloc(fnic->io_req_pool, GFP_ATOMIC);
 	if (!io_req) {
-		io_req = mempool_alloc(fnic->io_req_pool, GFP_ATOMIC);
-		if (!io_req) {
-			spin_unlock_irqrestore(io_lock, flags);
-			goto fnic_device_reset_end;
-		}
-		memset(io_req, 0, sizeof(*io_req));
-		io_req->port_id = rport->port_id;
-		CMD_SP(sc) = (char *)io_req;
+		spin_unlock_irqrestore(io_lock, flags);
+		goto fnic_device_reset_end;
 	}
+	memset(io_req, 0, sizeof(*io_req));
+	io_req->port_id = rport->port_id;
+	CMD_SP(reset_sc) = (char *)io_req;
+
 	io_req->dr_done = &tm_done;
-	CMD_STATE(sc) = FNIC_IOREQ_CMD_PENDING;
-	CMD_LR_STATUS(sc) = FCPIO_INVALID_CODE;
+	CMD_STATE(reset_sc) = FNIC_IOREQ_CMD_PENDING;
+	CMD_LR_STATUS(reset_sc) = FCPIO_INVALID_CODE;
 	spin_unlock_irqrestore(io_lock, flags);
 
 	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host, "TAG %x\n", tag);
@@ -2325,15 +2310,15 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 	 * issue the device reset, if enqueue failed, clean up the ioreq
 	 * and break assoc with scsi cmd
 	 */
-	if (fnic_queue_dr_io_req(fnic, sc, io_req)) {
+	if (fnic_queue_dr_io_req(fnic, reset_sc, io_req)) {
 		spin_lock_irqsave(io_lock, flags);
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
+		io_req = (struct fnic_io_req *)CMD_SP(reset_sc);
 		if (io_req)
 			io_req->dr_done = NULL;
 		goto fnic_device_reset_clean;
 	}
 	spin_lock_irqsave(io_lock, flags);
-	CMD_FLAGS(sc) |= FNIC_DEV_RST_ISSUED;
+	CMD_FLAGS(reset_sc) |= FNIC_DEV_RST_ISSUED;
 	spin_unlock_irqrestore(io_lock, flags);
 
 	/*
@@ -2344,16 +2329,16 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 				    msecs_to_jiffies(FNIC_LUN_RESET_TIMEOUT));
 
 	spin_lock_irqsave(io_lock, flags);
-	io_req = (struct fnic_io_req *)CMD_SP(sc);
+	io_req = (struct fnic_io_req *)CMD_SP(reset_sc);
 	if (!io_req) {
 		spin_unlock_irqrestore(io_lock, flags);
 		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
-				"io_req is null tag 0x%x sc 0x%p\n", tag, sc);
+				"io_req is null tag 0x%x sc 0x%p\n", tag, reset_sc);
 		goto fnic_device_reset_end;
 	}
 	io_req->dr_done = NULL;
 
-	status = CMD_LR_STATUS(sc);
+	status = CMD_LR_STATUS(reset_sc);
 
 	/*
 	 * If lun reset not completed, bail out with failed. io_req
@@ -2363,16 +2348,16 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 		atomic64_inc(&reset_stats->device_reset_timeouts);
 		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
 			      "Device reset timed out\n");
-		CMD_FLAGS(sc) |= FNIC_DEV_RST_TIMED_OUT;
+		CMD_FLAGS(reset_sc) |= FNIC_DEV_RST_TIMED_OUT;
 		spin_unlock_irqrestore(io_lock, flags);
-		int_to_scsilun(sc->device->lun, &fc_lun);
+		int_to_scsilun(sdev->lun, &fc_lun);
 		/*
 		 * Issue abort and terminate on device reset request.
 		 * If q'ing of terminate fails, retry it after a delay.
 		 */
 		while (1) {
 			spin_lock_irqsave(io_lock, flags);
-			if (CMD_FLAGS(sc) & FNIC_DEV_RST_TERM_ISSUED) {
+			if (CMD_FLAGS(reset_sc) & FNIC_DEV_RST_TERM_ISSUED) {
 				spin_unlock_irqrestore(io_lock, flags);
 				break;
 			}
@@ -2385,13 +2370,13 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 				msecs_to_jiffies(FNIC_ABT_TERM_DELAY_TIMEOUT));
 			} else {
 				spin_lock_irqsave(io_lock, flags);
-				CMD_FLAGS(sc) |= FNIC_DEV_RST_TERM_ISSUED;
-				CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
+				CMD_FLAGS(reset_sc) |= FNIC_DEV_RST_TERM_ISSUED;
+				CMD_STATE(reset_sc) = FNIC_IOREQ_ABTS_PENDING;
 				io_req->abts_done = &tm_done;
 				spin_unlock_irqrestore(io_lock, flags);
 				FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
 				"Abort and terminate issued on Device reset "
-				"tag 0x%x sc 0x%p\n", tag, sc);
+				"tag 0x%x sc 0x%p\n", tag, reset_sc);
 				break;
 			}
 		}
@@ -2403,7 +2388,7 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 				msecs_to_jiffies(FNIC_LUN_RESET_TIMEOUT));
 				break;
 			} else {
-				io_req = (struct fnic_io_req *)CMD_SP(sc);
+				io_req = (struct fnic_io_req *)CMD_SP(reset_sc);
 				io_req->abts_done = NULL;
 				goto fnic_device_reset_clean;
 			}
@@ -2418,7 +2403,7 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 		FNIC_SCSI_DBG(KERN_DEBUG,
 			      fnic->lport->host,
 			      "Device reset completed - failed\n");
-		io_req = (struct fnic_io_req *)CMD_SP(sc);
+		io_req = (struct fnic_io_req *)CMD_SP(reset_sc);
 		goto fnic_device_reset_clean;
 	}
 
@@ -2429,7 +2414,7 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 	 * the lun reset cmd. If all cmds get cleaned, the lun reset
 	 * succeeds
 	 */
-	if (fnic_clean_pending_aborts(fnic, sc, new_sc)) {
+	if (fnic_clean_pending_aborts(fnic, reset_sc)) {
 		spin_lock_irqsave(io_lock, flags);
 		io_req = (struct fnic_io_req *)CMD_SP(sc);
 		FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
@@ -2440,35 +2425,35 @@ int fnic_device_reset(struct scsi_cmnd *sc)
 
 	/* Clean lun reset command */
 	spin_lock_irqsave(io_lock, flags);
-	io_req = (struct fnic_io_req *)CMD_SP(sc);
+	io_req = (struct fnic_io_req *)CMD_SP(reset_sc);
 	if (io_req)
 		/* Completed, and successful */
 		ret = SUCCESS;
 
 fnic_device_reset_clean:
 	if (io_req)
-		CMD_SP(sc) = NULL;
+		CMD_SP(reset_sc) = NULL;
 
 	spin_unlock_irqrestore(io_lock, flags);
 
 	if (io_req) {
 		start_time = io_req->start_time;
-		fnic_release_ioreq_buf(fnic, io_req, sc);
+		fnic_release_ioreq_buf(fnic, io_req, reset_sc);
 		mempool_free(io_req, fnic->io_req_pool);
 	}
 
 fnic_device_reset_end:
-	FNIC_TRACE(fnic_device_reset, sc->device->host->host_no,
-		  sc->request->tag, sc,
+	FNIC_TRACE(fnic_device_reset, sdev->host->host_no,
+		   reset_sc->request->tag, reset_sc,
 		  jiffies_to_msecs(jiffies - start_time),
-		  0, ((u64)sc->cmnd[0] << 32 |
+		  0, ((u64)reset_sc->cmnd[0] << 32 |
 		  (u64)sc->cmnd[2] << 24 | (u64)sc->cmnd[3] << 16 |
 		  (u64)sc->cmnd[4] << 8 | sc->cmnd[5]),
 		  (((u64)CMD_FLAGS(sc) << 32) | CMD_STATE(sc)));
 
-	/* free tag if it is allocated */
-	if (unlikely(tag_gen_flag))
-		fnic_scsi_host_end_tag(fnic, sc);
+	/* free internal command if it is allocated */
+	if (reset_sc)
+		scsi_put_internal_cmd(reset_sc);
 
 	FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
 		      "Returning from device reset %s\n",
Signature · 1 line
2.29.2
#17 lore raw

Re: [PATCH 04/18] fnic: use internal commands

drivers/scsi/fnic/fnic_scsi.c int fnic_device_reset(struct scsi_cmnd *sc) · line 2298

Quoted 12 lines · Hannes Reinecke
drivers/scsi/fnic/fnic_scsi.c +0 -12
-	tag = sc->request->tag;
-	if (unlikely(tag < 0)) {
-		/*
-		 * Really should fix the midlayer to pass in a proper
-		 * request for ioctls...
-		 */
-		tag = fnic_scsi_host_start_tag(fnic, sc);
-		if (unlikely(tag == SCSI_NO_TAG))
-			goto fnic_device_reset_end;
-		tag_gen_flag = 1;
-		new_sc = 1;
-	}

Since this patch removes the only callers of fnic_scsi_host_start_tag() and fnic_scsi_host_end_tag(), please modify this patch such that it also removes these functions.

Thanks,

Bart.

#18 lore raw

Re: [PATCH 04/18] fnic: use internal commands

Quoted 18 lines · Bart Van Assche

drivers/scsi/fnic/fnic_scsi.c int fnic_device_reset(struct scsi_cmnd *sc) · line 2298

Quoted 12 lines · Hannes Reinecke
drivers/scsi/fnic/fnic_scsi.c +0 -12
-	tag = sc->request->tag;
-	if (unlikely(tag < 0)) {
-		/*
-		 * Really should fix the midlayer to pass in a proper
-		 * request for ioctls...
-		 */
-		tag = fnic_scsi_host_start_tag(fnic, sc);
-		if (unlikely(tag == SCSI_NO_TAG))
-			goto fnic_device_reset_end;
-		tag_gen_flag = 1;
-		new_sc = 1;
-	}

Since this patch removes the only callers of fnic_scsi_host_start_tag() and fnic_scsi_host_end_tag(), please modify this patch such that it also removes these functions.

Of course.

Will do in the next round.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#19 lore raw

PATCH v8 5/18 scsi: use real inquiry data when initialising devices

In reply to Hannes Reinecke

Use dummy inquiry data when initialising devices and not just some 'nullnullnull' string.

---
 drivers/scsi/scsi_scan.c | 34 +++++++++++++++++++++++++---------
 1 file changed, 25 insertions(+), 9 deletions(-)
drivers/scsi/scsi_scan.c +25 -9
index 9f1b7f3c650a..a50abba41ac1 100644
@@ -81,7 +81,27 @@
 #define SCSI_SCAN_TARGET_PRESENT	1
 #define SCSI_SCAN_LUN_PRESENT		2
 
-static const char *scsi_null_device_strs = "nullnullnullnull";
+/*
+ * Dummy inquiry for virtual LUNs:
+ *
+ * standard INQUIRY: [qualifier indicates no connected LU]
+ *  PQual=1  Device_type=31  RMB=0  LU_CONG=0  version=0x05  [SPC-3]
+ *  [AERC=0]  [TrmTsk=0]  NormACA=0  HiSUP=0  Resp_data_format=2
+ *  SCCS=0  ACC=0  TPGS=0  3PC=0  Protect=0  [BQue=0]
+ *  EncServ=0  MultiP=0  [MChngr=0]  [ACKREQQ=0]  Addr16=0
+ *  [RelAdr=0]  WBus16=0  Sync=0  [Linked=0]  [TranDis=0]  CmdQue=0
+ *    length=36 (0x24)   Peripheral device type: no physical device on this lu
+ * Vendor identification: LINUX
+ * Product identification: VIRTUALLUN
+ * Product revision level: 1.0
+ */
+static const unsigned char scsi_null_inquiry[36] = {
+	0x3f, 0x00, 0x05, 0x02, 0x1f, 0x00, 0x00, 0x00,
+	0x4c, 0x49, 0x4e, 0x55, 0x58, 0x20, 0x20, 0x20,
+	0x56, 0x49, 0x52, 0x54, 0x55, 0x41, 0x4c, 0x4c,
+	0x55, 0x4e, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20,
+	0x31, 0x2e, 0x30, 0x20
+};
 
 #define MAX_SCSI_LUNS	512
 
@@ -225,9 +245,10 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
 	if (!sdev)
 		goto out;
 
-	sdev->vendor = scsi_null_device_strs;
-	sdev->model = scsi_null_device_strs;
-	sdev->rev = scsi_null_device_strs;
+	sdev->type = scsi_null_inquiry[0] & 0x1f;
+	sdev->vendor = scsi_null_inquiry + 8;
+	sdev->model = scsi_null_inquiry + 16;
+	sdev->rev = scsi_null_inquiry + 32;
 	sdev->host = shost;
 	sdev->queue_ramp_up_period = SCSI_DEFAULT_RAMP_UP_PERIOD;
 	sdev->id = starget->id;
@@ -254,11 +275,6 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
 	 * slave_configure function */
 	sdev->max_device_blocked = SCSI_DEFAULT_DEVICE_BLOCKED;
 
-	/*
-	 * Some low level driver could use device->type
-	 */
-	sdev->type = -1;
-
 	/*
 	 * Assume that the device will have handshaking problems,
 	 * and then fix this field later if it turns out it
Signature · 1 line
2.29.2
Reviewed-by on v8 5/18
#20 lore raw

Re: [PATCH 05/18] scsi: use real inquiry data when initialising devices

Quoted 2 lines · Hannes Reinecke

Use dummy inquiry data when initialising devices and not just some 'nullnullnull' string.

#21 lore raw

Re: [PATCH 05/18] scsi: use real inquiry data when initialising devices

In reply to Hannes Reinecke
Quoted 2 lines · Hannes Reinecke

Use dummy inquiry data when initialising devices and not just some 'nullnullnull' string.

Why?

Quoted 9 lines

+/*
+ * Dummy inquiry for virtual LUNs:
+ *
+ * standard INQUIRY: [qualifier indicates no connected LU]

+ *  PQual=1  Device_type=31  RMB=0  LU_CONG=0  version=0x05  [SPC-3]
+ *  [AERC=0]  [TrmTsk=0]  NormACA=0  HiSUP=0  Resp_data_format=2
+ *  SCCS=0  ACC=0  TPGS=0  3PC=0  Protect=0  [BQue=0]
+ *  EncServ=0  MultiP=0  [MChngr=0]  [ACKREQQ=0]  Addr16=0
+ *  [RelAdr=0]  WBus16=0  Sync=0  [Linked=0]  [TranDis=0]  CmdQue=0
Quoted 5 lines

+ * length=36 (0x24) Peripheral device type: no physical device on this lu
+ * Vendor identification: LINUX
+ * Product identification: VIRTUALLUN
+ * Product revision level: 1.0
+ */

You don't juse set this up for virtual Luns, but as a default for all scsi_devices before calling inquirty. I'd much helper with a helper to fill out fake inquiry data rather than having seemingly valid data for all devices before inquirty is called or if it fails.

#22 lore raw

Re: [PATCH 05/18] scsi: use real inquiry data when initialising devices

Quoted 6 lines · Christoph Hellwig
Quoted 2 lines · Hannes Reinecke

Use dummy inquiry data when initialising devices and not just some 'nullnullnull' string.

Why?

Because it's really weird if you start up scsi_debug with thousands of devices and then call 'lsscsi' repeatedly. That will print out several devices with 'nullnullnull', only to be replaced with the 'real' inquiry data during device discovery.
I'd rather have a valid inquiry right from the start.

Quoted 15 lines
Quoted 9 lines

+/*
+ * Dummy inquiry for virtual LUNs:
+ *
+ * standard INQUIRY: [qualifier indicates no connected LU]

+ *  PQual=1  Device_type=31  RMB=0  LU_CONG=0  version=0x05  [SPC-3]
+ *  [AERC=0]  [TrmTsk=0]  NormACA=0  HiSUP=0  Resp_data_format=2
+ *  SCCS=0  ACC=0  TPGS=0  3PC=0  Protect=0  [BQue=0]
+ *  EncServ=0  MultiP=0  [MChngr=0]  [ACKREQQ=0]  Addr16=0
+ *  [RelAdr=0]  WBus16=0  Sync=0  [Linked=0]  [TranDis=0]  CmdQue=0
Quoted 5 lines

+ * length=36 (0x24) Peripheral device type: no physical device on this lu
+ * Vendor identification: LINUX
+ * Product identification: VIRTUALLUN
+ * Product revision level: 1.0
+ */

Quoted 5 lines

You don't juse set this up for virtual Luns, but as a default for all scsi_devices before calling inquirty. I'd much helper with a helper to fill out fake inquiry data rather than having seemingly valid data for all devices before inquirty is called or if it fails.

Right. Will be doing so.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#23 lore raw

PATCH v8 6/18 scsi: Use dummy inquiry data for the host device

In reply to Hannes Reinecke

Attach the dummy inquiry data to the scsi host device and use it to check if any given device is a scsi host device.

---
 drivers/scsi/scsi_scan.c  | 17 +++++++++++++++--
 drivers/scsi/scsi_sysfs.c |  3 ++-
 include/scsi/scsi_host.h  |  7 ++++---
 3 files changed, 21 insertions(+), 6 deletions(-)
drivers/scsi/scsi_scan.c +15 -2
index a50abba41ac1..55fa7f7317a2 100644
@@ -1940,9 +1940,11 @@ struct scsi_device *scsi_get_host_dev(struct Scsi_Host *shost)
 		goto out;
 
 	sdev = scsi_alloc_sdev(starget, 0, NULL);
-	if (sdev)
+	if (sdev) {
 		sdev->borken = 0;
-	else
+		sdev->inquiry = (unsigned char *)scsi_null_inquiry;
+		sdev->inquiry_len = sizeof(scsi_null_inquiry);
+	} else
 		scsi_target_reap(starget);
 	put_device(&starget->dev);
  out:
@@ -1967,3 +1969,14 @@ void scsi_free_host_dev(struct scsi_device *sdev)
 }
 EXPORT_SYMBOL(scsi_free_host_dev);
 
+/**
+ * scsi_device_is_host_dev - Check if a scsi device is a host device
+ * @sdev: SCSI device to test
+ *
+ * Returns: true if @sdev is a host device, false otherwise
+ */
+bool scsi_device_is_host_dev(struct scsi_device *sdev)
+{
+	return ((const unsigned char *)sdev->inquiry == scsi_null_inquiry);
+}
+EXPORT_SYMBOL_GPL(scsi_device_is_host_dev);
drivers/scsi/scsi_sysfs.c +2 -1
index 6d2a28294092..d5260d1b7b38 100644
@@ -498,7 +498,8 @@ static void scsi_device_dev_release_usercontext(struct work_struct *work)
 		kfree_rcu(vpd_pg80, rcu);
 	if (vpd_pg89)
 		kfree_rcu(vpd_pg89, rcu);
-	kfree(sdev->inquiry);
+	if (!scsi_device_is_host_dev(sdev))
+		kfree(sdev->inquiry);
 	kfree(sdev);
 
 	if (parent)
include/scsi/scsi_host.h +4 -3
index 3f3ebfdedeb2..f492e9034a62 100644
@@ -804,14 +804,15 @@ void scsi_host_busy_iter(struct Scsi_Host *,
 struct class_container;
 
 /*
- * These two functions are used to allocate and free a pseudo device
- * which will connect to the host adapter itself rather than any
- * physical device.  You must deallocate when you are done with the
+ * These three functions are used to allocate, free, and test for
+ * a pseudo device which will connect to the host adapter itself rather
+ * than any physical device.  You must deallocate when you are done with the
  * thing.  This physical pseudo-device isn't real and won't be available
  * from any high-level drivers.
  */
 extern void scsi_free_host_dev(struct scsi_device *);
 extern struct scsi_device *scsi_get_host_dev(struct Scsi_Host *);
+bool scsi_device_is_host_dev(struct scsi_device *);
 
 /*
  * DIF defines the exchange of protection information between
Signature · 1 line
2.29.2
#24 lore raw

Re: [PATCH 06/18] scsi: Use dummy inquiry data for the host device

drivers/scsi/scsi_scan.c void scsi_free_host_dev(struct scsi_device *sdev) · line 1981

Quoted 10 lines · Hannes Reinecke
drivers/scsi/scsi_scan.c +10 -0
+/**
+ * scsi_device_is_host_dev - Check if a scsi device is a host device
+ * @sdev: SCSI device to test
+ *
+ * Returns: true if @sdev is a host device, false otherwise
+ */
+bool scsi_device_is_host_dev(struct scsi_device *sdev)
+{
+	return ((const unsigned char *)sdev->inquiry == scsi_null_inquiry);
+}

No parentheses around the expression in a return statement please.

Quoted 1 line

+EXPORT_SYMBOL_GPL(scsi_device_is_host_dev);

Does this mean that this function will be used outside the SCSI core?

Thanks,

Bart.

#25 lore raw

Re: [PATCH 06/18] scsi: Use dummy inquiry data for the host device

Quoted 14 lines · Bart Van Assche

drivers/scsi/scsi_scan.c void scsi_free_host_dev(struct scsi_device *sdev) · line 1981

Quoted 10 lines · Hannes Reinecke
drivers/scsi/scsi_scan.c +10 -0
+/**
+ * scsi_device_is_host_dev - Check if a scsi device is a host device
+ * @sdev: SCSI device to test
+ *
+ * Returns: true if @sdev is a host device, false otherwise
+ */
+bool scsi_device_is_host_dev(struct scsi_device *sdev)
+{
+	return ((const unsigned char *)sdev->inquiry == scsi_null_inquiry);
+}

No parentheses around the expression in a return statement please.

Okay.

Quoted 4 lines
Quoted 1 line

+EXPORT_SYMBOL_GPL(scsi_device_is_host_dev);

Does this mean that this function will be used outside the SCSI core?

Will have to check; might be a left-over from previous iterations.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#26 lore raw

PATCH v8 7/18 scsi: revamp host device handling

In reply to Hannes Reinecke

Ensure that the host device is excluded from scanning by setting the BLIST_NOLUN flag, and avoid it being presented in sysfs. Also move the device id from using the ->this_id value as target id (which is a bit odd as it's typically is set to -1 anyway) to using ->max_channel + 1 as the channel number and '0' as the target id.
With that the host device is now handled like any other scsi device, which means we can drop the scsi_put_host_dev() function and let scsi_forget_host() etc handle the deallocation.
We only need to ensure that the host device is deallocated last is the driver might need it to send commands during teardown.

---
 drivers/scsi/scsi_devinfo.c |  1 +
 drivers/scsi/scsi_scan.c    | 55 +++++++++++++++++++------------------
 include/scsi/scsi_host.h    | 13 +++++----
 3 files changed, 37 insertions(+), 32 deletions(-)
drivers/scsi/scsi_devinfo.c +1 -0
index d92cec12454c..e70d87c5342b 100644
@@ -195,6 +195,7 @@ static struct {
 	{"Intel", "Multi-Flex", NULL, BLIST_NO_RSOC},
 	{"iRiver", "iFP Mass Driver", NULL, BLIST_NOT_LOCKABLE | BLIST_INQUIRY_36},
 	{"LASOUND", "CDX7405", "3.10", BLIST_MAX5LUN | BLIST_SINGLELUN},
+	{"LINUX", "VIRTUALLUN", NULL, BLIST_NOLUN},
 	{"Marvell", "Console", NULL, BLIST_SKIP_VPD_PAGES},
 	{"Marvell", "91xx Config", "1.01", BLIST_SKIP_VPD_PAGES},
 	{"MATSHITA", "PD-1", NULL, BLIST_FORCELUN | BLIST_SINGLELUN},
drivers/scsi/scsi_scan.c +29 -26

105 lines not on this page · raw message

include/scsi/scsi_host.h +7 -6
index f492e9034a62..f115150559ca 100644
@@ -804,14 +804,15 @@ void scsi_host_busy_iter(struct Scsi_Host *,
 struct class_container;
 
 /*
- * These three functions are used to allocate, free, and test for
- * a pseudo device which will connect to the host adapter itself rather
- * than any physical device.  You must deallocate when you are done with the
- * thing.  This physical pseudo-device isn't real and won't be available
+ * These functions are used to allocate and test a pseudo device
+ * which will refer to the host adapter itself rather than any
+ * physical device.  The device will be deallocated together with
+ * all other scsi devices, so there is no need to have a separate
+ * function to free it.
+ * This device will not show up in sysfs and won't be available
  * from any high-level drivers.
  */
-extern void scsi_free_host_dev(struct scsi_device *);
-extern struct scsi_device *scsi_get_host_dev(struct Scsi_Host *);
+struct scsi_device *scsi_get_host_dev(struct Scsi_Host *);
 bool scsi_device_is_host_dev(struct scsi_device *);
 
 /*
Signature · 1 line
2.29.2
#27 lore raw

Re: [PATCH 07/18] scsi: revamp host device handling

Quoted 4 lines · Hannes Reinecke

+ * Attach a single scsi_device to the Scsi_Host. The primary aim
+ * for this device is to serve as a container from which valid
+ * scsi commands can be allocated from. Each scsi command will carry
+ * an unused/free command tag,

Shouldn't introducing this comment be deferred to the patch that actually reserves a tag for internal commands (this does not mean that I'm convinced that this is the right approach)?

Quoted 3 lines

+ * which then can be used by the LLDD to
+ * send internal or passthrough commands without having to find a
+ * valid command tag internally.

Please change "valid SCSI commands" into "internal SCSI commands".

Thanks,

Bart.

#28 lore raw

Re: [PATCH 07/18] scsi: revamp host device handling

Quoted 10 lines · Bart Van Assche
Quoted 4 lines · Hannes Reinecke

+ * Attach a single scsi_device to the Scsi_Host. The primary aim
+ * for this device is to serve as a container from which valid
+ * scsi commands can be allocated from. Each scsi command will carry
+ * an unused/free command tag,

Shouldn't introducing this comment be deferred to the patch that actually reserves a tag for internal commands (this does not mean that I'm convinced that this is the right approach)?

Will do.

Quoted 6 lines
Quoted 3 lines

+ * which then can be used by the LLDD to
+ * send internal or passthrough commands without having to find a
+ * valid command tag internally.

Please change "valid SCSI commands" into "internal SCSI commands".

Okay.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#29 lore raw

Re: [PATCH 07/18] scsi: revamp host device handling

In reply to Hannes Reinecke

So right now scsi_get_host_dev/scsi_free_host_dev is entirely unused.

І'd rather just kill them off rather than giving them a new life and hacking all over the core code for them.

What do you need the scsi_device for instead of just having a request_queue for comands to the controller? If we need a scsi_device can we somehow make sure it doesn't hit the scanning and sysfs code from a much higher level?

#30 lore raw

Re: [PATCH 07/18] scsi: revamp host device handling

Quoted 10 lines · Christoph Hellwig

So right now scsi_get_host_dev/scsi_free_host_dev is entirely unused.

І'd rather just kill them off rather than giving them a new life and hacking all over the core code for them.

What do you need the scsi_device for instead of just having a request_queue for comands to the controller? If we need a scsi_device can we somehow make sure it doesn't hit the scanning and sysfs code from a much higher level?

That is what I did in v2, and got the response:

Quoted 7 lines

That was just a question on why virtio uses the per-device tags, which didn't look like it made any sense. What I'm worried about here is mixing up the concept of reserved tags in the tagset, and queues to use them. Note that we already have the scsi_get_host_dev to allocate a scsi_device and thus a request_queue for the host itself. That seems like the better interface to use a tag for a host wide command vs introducing a parallel path.

(To be found at https://lore.kernel.org/linux-scsi/[email protected]/ if you are curious.)

So do you retract on that statement?
Or did I misinterpret something there?

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#31 lore raw

PATCH v8 8/18 snic: use reserved commands

In reply to Hannes Reinecke

Allocate a host device and use internal commands for host and device reset.
This allows us to remove the special handling of host and bus reset commands.

---
 drivers/scsi/snic/snic.h      |   4 +-
 drivers/scsi/snic/snic_main.c |   7 ++
 drivers/scsi/snic/snic_scsi.c | 160 ++++++++++++++--------------------
 3 files changed, 74 insertions(+), 97 deletions(-)
drivers/scsi/snic/snic.h +2 -2
index f4c666285bba..5a2428033190 100644
@@ -59,7 +59,6 @@
  */
 #define SNIC_TAG_ABORT		BIT(30)		/* Tag indicating abort */
 #define SNIC_TAG_DEV_RST	BIT(29)		/* Tag for device reset */
-#define SNIC_TAG_IOCTL_DEV_RST	BIT(28)		/* Tag for User Device Reset */
 #define SNIC_TAG_MASK		(BIT(24) - 1)	/* Mask for lookup */
 #define SNIC_NO_TAG		-1
 
@@ -278,6 +277,7 @@ struct snic {
 
 	/* Scsi Host info */
 	struct Scsi_Host *shost;
+	struct scsi_device *shost_dev;
 
 	/* vnic related structures */
 	struct vnic_dev_bar bar0;
@@ -380,7 +380,7 @@ int snic_queuecommand(struct Scsi_Host *, struct scsi_cmnd *);
 int snic_abort_cmd(struct scsi_cmnd *);
 int snic_device_reset(struct scsi_cmnd *);
 int snic_host_reset(struct scsi_cmnd *);
-int snic_reset(struct Scsi_Host *, struct scsi_cmnd *);
+int snic_reset(struct Scsi_Host *);
 void snic_shutdown_scsi_cleanup(struct snic *);
 
 
drivers/scsi/snic/snic_main.c +7 -0
index 14f4ce665e58..ce0e7ab4ef1d 100644
@@ -303,6 +303,7 @@ static int
 snic_add_host(struct Scsi_Host *shost, struct pci_dev *pdev)
 {
 	int ret = 0;
+	struct snic *snic = shost_priv(shost);
 
 	ret = scsi_add_host(shost, &pdev->dev);
 	if (ret) {
@@ -313,6 +314,12 @@ snic_add_host(struct Scsi_Host *shost, struct pci_dev *pdev)
 		return ret;
 	}
 
+	snic->shost_dev = scsi_get_host_dev(shost);
+	if (!snic->shost_dev) {
+		SNIC_HOST_ERR(shost,
+			      "snic: scsi_get_virtual_dev failed\n");
+		return -ENOMEM;
+	}
 	SNIC_BUG_ON(shost->work_q != NULL);
 	snprintf(shost->work_q_name, sizeof(shost->work_q_name), "scsi_wq_%d",
 		 shost->host_no);
drivers/scsi/snic/snic_scsi.c +65 -95

361 lines not on this page · raw message

Signature · 1 line
2.29.2
#32 lore raw

Re: [PATCH 08/18] snic: use reserved commands

drivers/scsi/snic/snic_scsi.c snic_device_reset(struct scsi_cmnd *sc) · line 2150

Quoted 12 lines · Hannes Reinecke
drivers/scsi/snic/snic_scsi.c +4 -4
@@ -2170,42 +2144,42 @@ snic_device_reset(struct scsi_cmnd *sc)
 		goto dev_rst_end;
 	}
 
-	/* There is no tag when lun reset is issue through ioctl. */
-	if (unlikely(tag <= SNIC_NO_TAG)) {
-		SNIC_HOST_INFO(snic->shost,
-			       "Devrst: LUN Reset Recvd thru IOCTL.\n");
+	reset_sc = scsi_get_internal_cmd(sc->device, REQ_OP_SCSI_IN,
+					 BLK_MQ_REQ_NOWAIT);
+	if (!reset_sc)
+		goto dev_rst_end;

The SCSI error handler may call .eh_device_reset_handler and other error handling callbacks if no tags are available. If no tags are available, scsi_get_internal_cmd() will fail. If scsi_get_internal_cmd() fails, snic_device_reset() will fail. Does that count as a regression?

Thanks,

Bart.

#33 lore raw

Re: [PATCH 08/18] snic: use reserved commands

Quoted 5 lines · Bart Van Assche
Quoted 4 lines · Hannes Reinecke
drivers/scsi/snic/snic_scsi.c +0 -0
@@ -2170,42 +2144,42 @@ snic_device_reset(struct scsi_cmnd *sc)
  		goto dev_rst_end;
  	}
 
Quoted 14 lines

drivers/scsi/snic/snic_scsi.c snic_device_reset(struct scsi_cmnd *sc) · line 2150

Quoted 8 lines
drivers/scsi/snic/snic_scsi.c +4 -4
-	/* There is no tag when lun reset is issue through ioctl. */
-	if (unlikely(tag <= SNIC_NO_TAG)) {
-		SNIC_HOST_INFO(snic->shost,
-			       "Devrst: LUN Reset Recvd thru IOCTL.\n");
+	reset_sc = scsi_get_internal_cmd(sc->device, REQ_OP_SCSI_IN,
+					 BLK_MQ_REQ_NOWAIT);
+	if (!reset_sc)
+		goto dev_rst_end;

The SCSI error handler may call .eh_device_reset_handler and other error handling callbacks if no tags are available. If no tags are available, scsi_get_internal_cmd() will fail. If scsi_get_internal_cmd() fails, snic_device_reset() will fail. Does that count as a regression?

The snic driver _requires_ a free tag to send a reset.
If no free tags are available the driver cannot send a reset, even in the old code. So no, this is not a regression.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#34 lore raw

PATCH v8 9/18 snic: use tagset iter for traversing commands

In reply to Hannes Reinecke

Use the tagset iter to traverse active commands during device and hba reset.

---
 drivers/scsi/snic/snic_scsi.c | 366 +++++++++++++++++-----------------
 1 file changed, 187 insertions(+), 179 deletions(-)
drivers/scsi/snic/snic_scsi.c +187 -179
index 63cb9f81f82b..7bffad6626f8 100644
@@ -1641,90 +1641,97 @@ snic_abort_cmd(struct scsi_cmnd *sc)
 	return ret;
 }
 
+struct snic_cmd_pending_iter_data {
+	struct snic *snic;
+	struct scsi_device *sdev;
+	int ret;
+};
 
-
-static int
-snic_is_abts_pending(struct snic *snic, struct scsi_cmnd *lr_sc)
+static bool
+snic_is_abts_pending_iter(struct scsi_cmnd *sc, void *data, bool reserved)
 {
+	struct snic_cmd_pending_iter_data *iter_data = data;
 	struct snic_req_info *rqi = NULL;
-	struct scsi_cmnd *sc = NULL;
-	struct scsi_device *lr_sdev = NULL;
 	spinlock_t *io_lock = NULL;
-	u32 tag;
 	unsigned long flags;
 
-	if (lr_sc)
-		lr_sdev = lr_sc->device;
+	if (reserved)
+		return true;
 
-	/* walk through the tag map, an dcheck if IOs are still pending in fw*/
-	for (tag = 0; tag < snic->max_tag_id; tag++) {
-		io_lock = snic_io_lock_tag(snic, tag);
-
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(snic->shost, tag);
-
-		if (!sc || (lr_sc && (sc->device != lr_sdev || sc == lr_sc))) {
-			spin_unlock_irqrestore(io_lock, flags);
+	if (iter_data->sdev && iter_data->sdev != sc->device)
+		return true;
 
-			continue;
-		}
+	io_lock = snic_io_lock_hash(iter_data->snic, sc);
+	spin_lock_irqsave(io_lock, flags);
 
-		rqi = (struct snic_req_info *) CMD_SP(sc);
-		if (!rqi) {
-			spin_unlock_irqrestore(io_lock, flags);
+	rqi = (struct snic_req_info *) CMD_SP(sc);
+	if (!rqi) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
-			continue;
-		}
+	/*
+	 * Found IO that is still pending w/ firmware and belongs to
+	 * the LUN that is under reset, if lr_sc != NULL
+	 */
+	SNIC_SCSI_DBG(iter_data->snic->shost, "Found IO in %s on LUN\n",
+		      snic_ioreq_state_to_str(CMD_STATE(sc)));
 
-		/*
-		 * Found IO that is still pending w/ firmware and belongs to
-		 * the LUN that is under reset, if lr_sc != NULL
-		 */
-		SNIC_SCSI_DBG(snic->shost, "Found IO in %s on LUN\n",
-			      snic_ioreq_state_to_str(CMD_STATE(sc)));
+	if (CMD_STATE(sc) == SNIC_IOREQ_ABTS_PENDING)
+		iter_data->ret = 1;
+	spin_unlock_irqrestore(io_lock, flags);
 
-		if (CMD_STATE(sc) == SNIC_IOREQ_ABTS_PENDING) {
-			spin_unlock_irqrestore(io_lock, flags);
+	return true;
+}
 
-			return 1;
-		}
+static int
+snic_is_abts_pending(struct snic *snic, struct scsi_device *lr_sdev)
+{
+	struct snic_cmd_pending_iter_data iter_data = {
+		.snic = snic,
+		.sdev = lr_sdev,
+		.ret = 0,
+	};
 
-		spin_unlock_irqrestore(io_lock, flags);
-	}
+	/* walk through the tag map, an dcheck if IOs are still pending in fw*/
+	scsi_host_busy_iter(snic->shost,
+			    snic_is_abts_pending_iter, &iter_data);
 
-	return 0;
+	return iter_data.ret;
 } /* end of snic_is_abts_pending */
 
-static int
-snic_dr_clean_single_req(struct snic *snic,
-			 u32 tag,
-			 struct scsi_device *lr_sdev)
+static bool
+snic_dr_clean_single_req(struct scsi_cmnd *sc, void *data, bool reserved)
 {
 	struct snic_req_info *rqi = NULL;
 	struct snic_tgt *tgt = NULL;
-	struct scsi_cmnd *sc = NULL;
 	spinlock_t *io_lock = NULL;
 	u32 sv_state = 0, tmf = 0;
 	DECLARE_COMPLETION_ONSTACK(tm_done);
 	unsigned long flags;
 	int ret = 0;
+	struct snic_cmd_pending_iter_data *iter_data = data;
+	struct snic *snic = iter_data->snic;
 
-	io_lock = snic_io_lock_tag(snic, tag);
-	spin_lock_irqsave(io_lock, flags);
-	sc = scsi_host_find_tag(snic->shost, tag);
+	if (reserved)
+		return true;
 
-	/* Ignore Cmd that don't belong to Lun Reset device */
-	if (!sc || sc->device != lr_sdev)
-		goto skip_clean;
+	if (sc->device != iter_data->sdev)
+		return true;
 
+	io_lock = snic_io_lock_hash(snic, sc);
+	spin_lock_irqsave(io_lock, flags);
 	rqi = (struct snic_req_info *) CMD_SP(sc);
-
-	if (!rqi)
-		goto skip_clean;
+	if (!rqi) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
 
-	if (CMD_STATE(sc) == SNIC_IOREQ_ABTS_PENDING)
-		goto skip_clean;
+	if (CMD_STATE(sc) == SNIC_IOREQ_ABTS_PENDING) {
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
+	}
 
 
 	if ((CMD_FLAGS(sc) & SNIC_DEVICE_RESET) &&
@@ -1733,8 +1740,8 @@ snic_dr_clean_single_req(struct snic *snic,
 		SNIC_SCSI_DBG(snic->shost,
 			      "clean_single_req: devrst is not pending sc 0x%p\n",
 			      sc);
-
-		goto skip_clean;
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;
 	}
 
 	SNIC_SCSI_DBG(snic->shost,
@@ -1777,7 +1784,7 @@ snic_dr_clean_single_req(struct snic *snic,
 	if (ret) {
 		SNIC_HOST_ERR(snic->shost,
 			      "clean_single_req_err:sc %p, tag %d abt failed. tm_tag %d flags 0x%llx\n",
-			      sc, tag, rqi->tm_tag, CMD_FLAGS(sc));
+			      sc, snic_cmd_tag(sc), rqi->tm_tag, CMD_FLAGS(sc));
 
 		spin_lock_irqsave(io_lock, flags);
 		rqi = (struct snic_req_info *) CMD_SP(sc);
@@ -1788,8 +1795,9 @@ snic_dr_clean_single_req(struct snic *snic,
 		if (CMD_STATE(sc) == SNIC_IOREQ_ABTS_PENDING)
 			CMD_STATE(sc) = sv_state;
 
-		ret = 1;
-		goto skip_clean;
+		iter_data->ret = 1;
+		spin_unlock_irqrestore(io_lock, flags);
+		return false;
 	}
 
 	spin_lock_irqsave(io_lock, flags);
@@ -1806,7 +1814,8 @@ snic_dr_clean_single_req(struct snic *snic,
 	rqi = (struct snic_req_info *) CMD_SP(sc);
 	if (!rqi) {
 		CMD_FLAGS(sc) |= SNIC_IO_ABTS_TERM_REQ_NULL;
-		goto skip_clean;
+		spin_unlock_irqrestore(io_lock, flags);
+		return true;;
 	}
 	rqi->abts_done = NULL;
 
@@ -1814,14 +1823,13 @@ snic_dr_clean_single_req(struct snic *snic,
 	if (CMD_ABTS_STATUS(sc) == SNIC_INVALID_CODE) {
 		SNIC_HOST_ERR(snic->shost,
 			      "clean_single_req_err:sc %p tag %d abt still pending w/ fw, tm_tag %d flags 0x%llx\n",
-			      sc, tag, rqi->tm_tag, CMD_FLAGS(sc));
+			      sc, snic_cmd_tag(sc), rqi->tm_tag, CMD_FLAGS(sc));
 
 		CMD_FLAGS(sc) |= SNIC_IO_ABTS_TERM_DONE;
-		ret = 1;
-
-		goto skip_clean;
+		iter_data->ret = 1;
+		spin_unlock_irqrestore(io_lock, flags);
+		return false;
 	}
-
 	CMD_STATE(sc) = SNIC_IOREQ_ABTS_COMPLETE;
 	CMD_SP(sc) = NULL;
 	spin_unlock_irqrestore(io_lock, flags);
@@ -1831,39 +1839,31 @@ snic_dr_clean_single_req(struct snic *snic,
 	sc->result = (DID_ERROR << 16);
 	sc->scsi_done(sc);
 
-	ret = 0;
-
-	return ret;
-
-skip_clean:
-	spin_unlock_irqrestore(io_lock, flags);
-
-	return ret;
+	return true;
 } /* end of snic_dr_clean_single_req */
 
 static int
-snic_dr_clean_pending_req(struct snic *snic, struct scsi_cmnd *lr_sc)
+snic_dr_clean_pending_req(struct snic *snic, struct scsi_device *lr_sdev)
 {
-	struct scsi_device *lr_sdev = lr_sc->device;
-	u32 tag = 0;
 	int ret = FAILED;
+	struct snic_cmd_pending_iter_data iter_data = {
+		.snic = snic,
+		.sdev = lr_sdev,
+		.ret = 0,
+	};
 
-	for (tag = 0; tag < snic->max_tag_id; tag++) {
-		if (tag == snic_cmd_tag(lr_sc))
-			continue;
+	scsi_host_busy_iter(snic->shost,
+			    snic_dr_clean_single_req, &iter_data);
+	if (iter_data.ret) {
+		SNIC_HOST_ERR(snic->shost, "clean_err = %d\n", iter_data.ret);
 
-		ret = snic_dr_clean_single_req(snic, tag, lr_sdev);
-		if (ret) {
-			SNIC_HOST_ERR(snic->shost, "clean_err:tag = %d\n", tag);
-
-			goto clean_err;
-		}
+		goto clean_err;
 	}
 
 	schedule_timeout(msecs_to_jiffies(100));
 
 	/* Walk through all the cmds and check abts status. */
-	if (snic_is_abts_pending(snic, lr_sc)) {
+	if (snic_is_abts_pending(snic, lr_sdev)) {
 		ret = FAILED;
 
 		goto clean_err;
@@ -1950,7 +1950,7 @@ snic_dr_finish(struct snic *snic, struct scsi_cmnd *sc)
 	 * succeeds.
 	 */
 
-	ret = snic_dr_clean_pending_req(snic, sc);
+	ret = snic_dr_clean_pending_req(snic, sc->device);
 	if (ret) {
 		spin_lock_irqsave(io_lock, flags);
 		SNIC_SCSI_DBG(snic->shost,
@@ -2408,77 +2408,79 @@ snic_cmpl_pending_tmreq(struct snic *snic, struct scsi_cmnd *sc)
 		complete(rqi->abts_done);
 }
 
-/*
- * snic_scsi_cleanup: Walks through tag map and releases the reqs
- */
-static void
-snic_scsi_cleanup(struct snic *snic)
+static bool
+snic_scsi_cleanup_iter(struct scsi_cmnd *sc, void *data, bool rsvd)
 {
+	struct snic *snic = data;
 	struct snic_req_info *rqi = NULL;
-	struct scsi_cmnd *sc = NULL;
 	spinlock_t *io_lock = NULL;
 	unsigned long flags;
-	int tag;
+	int tag = snic_cmd_tag(sc);
 	u64 st_time = 0;
 
-	SNIC_SCSI_DBG(snic->shost, "sc_clean: scsi cleanup.\n");
-
-	for (tag = 0; tag < snic->max_tag_id; tag++) {
-		io_lock = snic_io_lock_tag(snic, tag);
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(snic->shost, tag);
-		if (unlikely(snic_tmreq_pending(sc))) {
-			/*
-			 * When FW Completes reset w/o sending completions
-			 * for outstanding ios.
-			 */
-			snic_cmpl_pending_tmreq(snic, sc);
-			spin_unlock_irqrestore(io_lock, flags);
-
-			continue;
-		}
-
-		rqi = (struct snic_req_info *) CMD_SP(sc);
-		if (!rqi) {
-			spin_unlock_irqrestore(io_lock, flags);
+	io_lock = snic_io_lock_hash(snic, sc);
+	spin_lock_irqsave(io_lock, flags);
+	if (unlikely(snic_tmreq_pending(sc))) {
+		/*
+		 * When FW Completes reset w/o sending completions
+		 * for outstanding ios.
+		 */
+		snic_cmpl_pending_tmreq(snic, sc);
+		spin_unlock_irqrestore(io_lock, flags);
 
-			goto cleanup;
-		}
+		return true;
+	}
+	rqi = (struct snic_req_info *) CMD_SP(sc);
+	if (!rqi) {
+		spin_unlock_irqrestore(io_lock, flags);
+		goto cleanup;
+	}
 
-		SNIC_SCSI_DBG(snic->shost,
-			      "sc_clean: sc %p, rqi %p, tag %d flags 0x%llx\n",
-			      sc, rqi, tag, CMD_FLAGS(sc));
+	SNIC_SCSI_DBG(snic->shost,
+		      "sc_clean: sc %p, rqi %p, tag %d flags 0x%llx\n",
+		      sc, rqi, tag, CMD_FLAGS(sc));
 
-		CMD_SP(sc) = NULL;
-		CMD_FLAGS(sc) |= SNIC_SCSI_CLEANUP;
-		spin_unlock_irqrestore(io_lock, flags);
-		st_time = rqi->start_time;
+	CMD_SP(sc) = NULL;
+	CMD_FLAGS(sc) |= SNIC_SCSI_CLEANUP;
+	spin_unlock_irqrestore(io_lock, flags);
+	st_time = rqi->start_time;
 
-		SNIC_HOST_INFO(snic->shost,
-			       "sc_clean: Releasing rqi %p : flags 0x%llx\n",
-			       rqi, CMD_FLAGS(sc));
+	SNIC_HOST_INFO(snic->shost,
+		       "sc_clean: Releasing rqi %p : flags 0x%llx\n",
+		       rqi, CMD_FLAGS(sc));
 
-		snic_release_req_buf(snic, rqi, sc);
+	snic_release_req_buf(snic, rqi, sc);
 
 cleanup:
-		sc->result = DID_TRANSPORT_DISRUPTED << 16;
-		SNIC_HOST_INFO(snic->shost,
-			       "sc_clean: DID_TRANSPORT_DISRUPTED for sc %p, Tag %d flags 0x%llx rqi %p duration %u msecs\n",
-			       sc, sc->request->tag, CMD_FLAGS(sc), rqi,
-			       jiffies_to_msecs(jiffies - st_time));
+	sc->result = DID_TRANSPORT_DISRUPTED << 16;
+	SNIC_HOST_INFO(snic->shost,
+		       "sc_clean: DID_TRANSPORT_DISRUPTED for sc %p, Tag %d flags 0x%llx rqi %p duration %u msecs\n",
+		       sc, tag, CMD_FLAGS(sc), rqi,
+		       jiffies_to_msecs(jiffies - st_time));
 
-		/* Update IO stats */
-		snic_stats_update_io_cmpl(&snic->s_stats);
+	/* Update IO stats */
+	snic_stats_update_io_cmpl(&snic->s_stats);
 
-		if (sc->scsi_done) {
-			SNIC_TRC(snic->shost->host_no, tag, (ulong) sc,
-				 jiffies_to_msecs(jiffies - st_time), 0,
-				 SNIC_TRC_CMD(sc),
-				 SNIC_TRC_CMD_STATE_FLAGS(sc));
+	if (sc->scsi_done) {
+		SNIC_TRC(snic->shost->host_no, tag, (ulong) sc,
+			 jiffies_to_msecs(jiffies - st_time), 0,
+			 SNIC_TRC_CMD(sc),
+			 SNIC_TRC_CMD_STATE_FLAGS(sc));
 
-			sc->scsi_done(sc);
-		}
+		sc->scsi_done(sc);
 	}
+	return true;
+}
+
+/*
+ * snic_scsi_cleanup: Walks through tag map and releases the reqs
+ */
+static void
+snic_scsi_cleanup(struct snic *snic)
+{
+	SNIC_SCSI_DBG(snic->shost, "sc_clean: scsi cleanup.\n");
+
+	scsi_host_busy_iter(snic->shost, snic_scsi_cleanup_iter, snic);
 } /* end of snic_scsi_cleanup */
 
 void
@@ -2575,6 +2577,40 @@ snic_internal_abort_io(struct snic *snic, struct scsi_cmnd *sc, int tmf)
 	return ret;
 } /* end of snic_internal_abort_io */
 
+struct snic_tgt_scsi_abort_io_iter_data {
+	struct snic *snic;
+	struct snic_tgt *tgt;
+	int tmf;
+	int abt_cnt;
+};
+
+static bool
+snic_tgt_scsi_abort_io_iter(struct scsi_cmnd *sc, void *data, bool reserved)
+{
+	struct snic_tgt_scsi_abort_io_iter_data *iter_data = data;
+	struct snic_tgt *sc_tgt = NULL;
+	int ret;
+
+	if (reserved)
+		return true;
+
+	sc_tgt = starget_to_tgt(scsi_target(sc->device));
+	if (sc_tgt != iter_data->tgt)
+		return true;
+
+	ret = snic_internal_abort_io(iter_data->snic, sc, iter_data->tmf);
+	if (ret < 0) {
+		SNIC_HOST_ERR(iter_data->snic->shost,
+			      "tgt_abt_io: Tag %x, Failed w err = %d\n",
+			      snic_cmd_tag(sc), ret);
+		return true;
+	}
+
+	if (ret == SUCCESS)
+		iter_data->abt_cnt++;
+	return true;
+}
+
 /*
  * snic_tgt_scsi_abort_io : called by snic_tgt_del
  */
@@ -2582,11 +2618,10 @@ int
 snic_tgt_scsi_abort_io(struct snic_tgt *tgt)
 {
 	struct snic *snic = NULL;
-	struct scsi_cmnd *sc = NULL;
-	struct snic_tgt *sc_tgt = NULL;
-	spinlock_t *io_lock = NULL;
-	unsigned long flags;
-	int ret = 0, tag, abt_cnt = 0, tmf = 0;
+	struct snic_tgt_scsi_abort_io_iter_data iter_data = {
+		.tgt = tgt,
+		.abt_cnt = 0,
+	};
 
 	if (!tgt)
 		return -1;
@@ -2595,43 +2630,16 @@ snic_tgt_scsi_abort_io(struct snic_tgt *tgt)
 	SNIC_SCSI_DBG(snic->shost, "tgt_abt_io: Cleaning Pending IOs.\n");
 
 	if (tgt->tdata.typ == SNIC_TGT_DAS)
-		tmf = SNIC_ITMF_ABTS_TASK;
+		iter_data.tmf = SNIC_ITMF_ABTS_TASK;
 	else
-		tmf = SNIC_ITMF_ABTS_TASK_TERM;
-
-	for (tag = 0; tag < snic->max_tag_id; tag++) {
-		io_lock = snic_io_lock_tag(snic, tag);
-
-		spin_lock_irqsave(io_lock, flags);
-		sc = scsi_host_find_tag(snic->shost, tag);
-		if (!sc) {
-			spin_unlock_irqrestore(io_lock, flags);
-
-			continue;
-		}
+		iter_data.tmf = SNIC_ITMF_ABTS_TASK_TERM;
+	iter_data.snic = snic;
 
-		sc_tgt = starget_to_tgt(scsi_target(sc->device));
-		if (sc_tgt != tgt) {
-			spin_unlock_irqrestore(io_lock, flags);
-
-			continue;
-		}
-		spin_unlock_irqrestore(io_lock, flags);
-
-		ret = snic_internal_abort_io(snic, sc, tmf);
-		if (ret < 0) {
-			SNIC_HOST_ERR(snic->shost,
-				      "tgt_abt_io: Tag %x, Failed w err = %d\n",
-				      tag, ret);
-
-			continue;
-		}
-
-		if (ret == SUCCESS)
-			abt_cnt++;
-	}
+	scsi_host_busy_iter(snic->shost,
+			    snic_tgt_scsi_abort_io_iter, &iter_data);
 
-	SNIC_SCSI_DBG(snic->shost, "tgt_abt_io: abt_cnt = %d\n", abt_cnt);
+	SNIC_SCSI_DBG(snic->shost, "tgt_abt_io: abt_cnt = %d\n",
+		      iter_data.abt_cnt);
 
 	return 0;
 } /* end of snic_tgt_scsi_abort_io */
Signature · 1 line
2.29.2
#35 lore raw

Re: [PATCH 09/18] snic: use tagset iter for traversing commands

Same questions as for fnic: why does the driver implements its own command completion in the EH path?

#36 lore raw

PATCH v8 10/18 scsi: implement reserved command handling

In reply to Hannes Reinecke

Quite some drivers are using management commands internally, which typically use the same hardware tag pool (ie they are being allocated from the same hardware resources) as the 'normal' I/O commands. These commands are set aside before allocating the block-mq tag bitmap, so they'll never show up as busy in the tag map.
The block-layer, OTOH, already has 'reserved_tags' to handle precisely this situation.
So this patch adds a new field 'nr_reserved_cmds' to the SCSI host template to instruct the block layer to set aside a tag space for these management commands by using reserved tags.

---
 drivers/scsi/hosts.c      |  3 +++
 drivers/scsi/scsi_lib.c   | 10 +++++++++-
 drivers/scsi/scsi_sysfs.c |  2 ++
 include/scsi/scsi_host.h  | 22 +++++++++++++++++++++-
 4 files changed, 35 insertions(+), 2 deletions(-)
drivers/scsi/hosts.c +3 -0
index 2f162603876f..661ed7696562 100644
@@ -469,6 +469,9 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
 	if (sht->virt_boundary_mask)
 		shost->virt_boundary_mask = sht->virt_boundary_mask;
 
+	if (sht->nr_reserved_cmds)
+		shost->nr_reserved_cmds = sht->nr_reserved_cmds;
+
 	device_initialize(&shost->shost_gendev);
 	dev_set_name(&shost->shost_gendev, "host%d", shost->host_no);
 	shost->shost_gendev.bus = &scsi_bus_type;
drivers/scsi/scsi_lib.c +9 -1
index f83b04e49bae..3c83b0fabefb 100644
@@ -1971,7 +1971,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost)
 		tag_set->ops = &scsi_mq_ops_no_commit;
 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
 	tag_set->nr_maps = shost->nr_maps ? : 1;
-	tag_set->queue_depth = shost->can_queue;
+	tag_set->queue_depth = shost->can_queue + shost->nr_reserved_cmds;
+	tag_set->reserved_tags = shost->nr_reserved_cmds;
 	tag_set->cmd_size = cmd_size;
 	tag_set->numa_node = NUMA_NO_NODE;
 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
@@ -1996,6 +1997,9 @@ void scsi_mq_destroy_tags(struct Scsi_Host *shost)
  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
  *
  * Allocates a SCSI command for internal LLDD use.
+ * If 'nr_reserved_commands' is specified by the host the
+ * command will be allocated from the reserved tag pool;
+ * otherwise the normal tag pool will be used.
  */
 struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
 	unsigned int op, blk_mq_req_flags_t flags)
@@ -2005,6 +2009,10 @@ struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
 
 	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
 		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+
+	if (sdev->host->nr_reserved_cmds)
+		flags |= BLK_MQ_REQ_RESERVED;
+
 	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
 	if (IS_ERR(rq))
 		return NULL;
drivers/scsi/scsi_sysfs.c +2 -0
index d5260d1b7b38..f4119999a402 100644
@@ -371,6 +371,7 @@ static DEVICE_ATTR(eh_deadline, S_IRUGO | S_IWUSR, show_shost_eh_deadline, store
 shost_rd_attr(unique_id, "%u\n");
 shost_rd_attr(cmd_per_lun, "%hd\n");
 shost_rd_attr(can_queue, "%d\n");
+shost_rd_attr(nr_reserved_cmds, "%d\n");
 shost_rd_attr(sg_tablesize, "%hu\n");
 shost_rd_attr(sg_prot_tablesize, "%hu\n");
 shost_rd_attr(unchecked_isa_dma, "%d\n");
@@ -422,6 +423,7 @@ static struct attribute *scsi_sysfs_shost_attrs[] = {
 	&dev_attr_host_reset.attr,
 	&dev_attr_eh_deadline.attr,
 	&dev_attr_nr_hw_queues.attr,
+	&dev_attr_nr_reserved_cmds.attr,
 	NULL
 };
 
include/scsi/scsi_host.h +21 -1
index f115150559ca..0831b33ee186 100644
@@ -367,10 +367,19 @@ struct scsi_host_template {
 	/*
 	 * This determines if we will use a non-interrupt driven
 	 * or an interrupt driven scheme.  It is set to the maximum number
-	 * of simultaneous commands a single hw queue in HBA will accept.
+	 * of simultaneous commands a single hw queue in HBA will accept
+	 * excluding internal commands.
 	 */
 	int can_queue;
 
+	/*
+	 * This determines how many commands the HBA will set aside
+	 * for internal commands. This number will be added to
+	 * @can_queue to calculate the maximum number of simultaneous
+	 * commands sent to the host.
+	 */
+	int nr_reserved_cmds;
+
 	/*
 	 * In many instances, especially where disconnect / reconnect are
 	 * supported, our host also has an ID on the SCSI bus.  If this is
@@ -614,6 +623,11 @@ struct Scsi_Host {
 	unsigned short max_cmd_len;
 
 	int this_id;
+
+	/*
+	 * Number of commands this host can handle at the same time.
+	 * This excludes reserved commands as specified by nr_reserved_cmds.
+	 */
 	int can_queue;
 	short cmd_per_lun;
 	short unsigned int sg_tablesize;
@@ -632,6 +646,12 @@ struct Scsi_Host {
 	 */
 	unsigned nr_hw_queues;
 	unsigned nr_maps;
+
+	/*
+	 * Number of reserved commands to allocate, if any.
+	 */
+	unsigned nr_reserved_cmds;
+
 	unsigned active_mode:2;
 	unsigned unchecked_isa_dma:1;
 
Signature · 1 line
2.29.2
#37 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 2 lines · Hannes Reinecke

These commands are set aside before allocating the block-mq tag bitmap, so they'll never show up as busy in the tag map.

That doesn't sound correct to me. Should the above perhaps be changed into "blk_mq_start_request() is never called for internal commands so they'll never show up as busy in the tag map"?

Thanks,

Bart.

#38 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 8 lines · Bart Van Assche
Quoted 2 lines · Hannes Reinecke

These commands are set aside before allocating the block-mq tag bitmap, so they'll never show up as busy in the tag map.

That doesn't sound correct to me. Should the above perhaps be changed into "blk_mq_start_request() is never called for internal commands so they'll never show up as busy in the tag map"?

Yes, will do.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#39 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 10 lines · Hannes Reinecke
Quoted 8 lines · Bart Van Assche
Quoted 2 lines · Hannes Reinecke

These commands are set aside before allocating the block-mq tag bitmap, so they'll never show up as busy in the tag map.

That doesn't sound correct to me. Should the above perhaps be changed into "blk_mq_start_request() is never called for internal commands so they'll never show up as busy in the tag map"?

Yes, will do.

So why don't these - or shouldn't these - turn up in the busy tag map?

One of the motivations to use these block requests for internal commands is that we can take advantage of the block layer handling for CPU hotplug for MQ hosts, i.e. if blk-mq can't see these are inflight, then they would be missed in blk_mq_hctx_notify_offline() -> blk_mq_hctx_has_requests(), right? And who knows what else...

Thanks,
John

#40 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 10 lines · John Garry
Quoted 9 lines · Hannes Reinecke
Quoted 8 lines · Bart Van Assche
Quoted 2 lines · Hannes Reinecke

These commands are set aside before allocating the block-mq tag bitmap, so they'll never show up as busy in the tag map.

That doesn't sound correct to me. Should the above perhaps be changed into "blk_mq_start_request() is never called for internal commands so they'll never show up as busy in the tag map"?

Quoted 6 lines
Quoted 1 line

Yes, will do.

So why don't these - or shouldn't these - turn up in the busy tag map?

One of the motivations to use these block requests for internal commands is that we can take advantage of the block layer handling for CPU hotplug for MQ hosts, i.e. if blk-mq can't see these are inflight, then they would be missed in blk_mq_hctx_notify_offline() -> blk_mq_hctx_has_requests(), right? And who knows what else...

Oh, but of course it's possible to call 'start' on these requests to have them counted in the busy map. I just didn't see the need for it until now, that's all.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#41 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 22 lines · Hannes Reinecke
Quoted 13 lines · John Garry
Quoted 11 lines · Hannes Reinecke
Quoted 9 lines · Bart Van Assche
Quoted 3 lines · Hannes Reinecke

These commands are set aside before allocating the block-mq tag bitmap,
so they'll never show up as busy in the tag map.

That doesn't sound correct to me. Should the above perhaps be changed into "blk_mq_start_request() is never called for internal commands so they'll never show up as busy in the tag map"?

Yes, will do.

Quoted 8 lines

So why don't these - or shouldn't these - turn up in the busy tag map?

One of the motivations to use these block requests for internal commands is that we can take advantage of the block layer handling for CPU hotplug for MQ hosts, i.e. if blk-mq can't see these are inflight, then they would be missed in blk_mq_hctx_notify_offline() -> blk_mq_hctx_has_requests(), right? And who knows what else...

Quoted 3 lines

Oh, but of course it's possible to call 'start' on these requests to have them counted in the busy map.
I just didn't see the need for it until now, that's all.

This is possible but this will require careful review of at least the following code paths such that nothing unexpected happens for internal commands:
* The SCSI timeout code.
* All blk_mq_tagset_busy_iter() and scsi_host_busy_iter() callers. As an example, scsi_host_busy() must not include LLD-internal commands.

Thanks,

Bart.

#42 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 27 lines · Bart Van Assche
Quoted 22 lines · Hannes Reinecke
Quoted 13 lines · John Garry
Quoted 11 lines · Hannes Reinecke
Quoted 9 lines · Bart Van Assche
Quoted 3 lines · Hannes Reinecke

These commands are set aside before allocating the block-mq tag bitmap,
so they'll never show up as busy in the tag map.

That doesn't sound correct to me. Should the above perhaps be changed into "blk_mq_start_request() is never called for internal commands so they'll never show up as busy in the tag map"?

Yes, will do.

Quoted 8 lines

So why don't these - or shouldn't these - turn up in the busy tag map?

One of the motivations to use these block requests for internal commands is that we can take advantage of the block layer handling for CPU hotplug for MQ hosts, i.e. if blk-mq can't see these are inflight, then they would be missed in blk_mq_hctx_notify_offline() -> blk_mq_hctx_has_requests(), right? And who knows what else...

Quoted 3 lines

Oh, but of course it's possible to call 'start' on these requests to have them counted in the busy map.
I just didn't see the need for it until now, that's all.

Quoted 7 lines

This is possible but this will require careful review of at least the following code paths such that nothing unexpected happens for internal commands:
* The SCSI timeout code.
* All blk_mq_tagset_busy_iter() and scsi_host_busy_iter() callers. As an example, scsi_host_busy() must not include LLD-internal commands.

Oh, _that_ is easy. These are reserved commands, which will have the last bool argument to the iter functions set to 'true'.

bool (*fn)(struct scsi_cmnd *, void *, bool)

So we just need to return from the iter if the last argument is true.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		        Kernel Storage Architect
[email protected]			               +49 911 74053 688
SUSE Software Solutions Germany GmbH, 90409 Nürnberg
GF: F. Imendörffer, HRB 36809 (AG Nürnberg)
#43 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

In reply to Hannes Reinecke

drivers/scsi/scsi_lib.c struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev, · line 2018

Quoted 13 lines · Hannes Reinecke

struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,

drivers/scsi/scsi_lib.c +4 -0
 	unsigned int op, blk_mq_req_flags_t flags)
@@ -2005,6 +2009,10 @@ struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
 
 	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
 		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+
+	if (sdev->host->nr_reserved_cmds)
+		flags |= BLK_MQ_REQ_RESERVED;
+
 	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
 	if (IS_ERR(rq))
 		return NULL;

Can the if-statement be removed such that scsi_get_internal_cmd() fails if sdev->host->nr_reserved_cmds == 0? I'm concerned that otherwise it will be very hard to determine which requests are internal and which ones not from inside a blk_mq_tagset_busy_iter() callback.

Thanks,

Bart.

#44 lore raw

Re: [PATCH 10/18] scsi: implement reserved command handling

Quoted 5 lines · Bart Van Assche
Quoted 4 lines · Hannes Reinecke

struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,

drivers/scsi/scsi_lib.c +0 -0
 	unsigned int op, blk_mq_req_flags_t flags)
@@ -2005,6 +2009,10 @@ struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
 
Quoted 15 lines

drivers/scsi/scsi_lib.c struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev, · line 2018

Quoted 9 lines
drivers/scsi/scsi_lib.c +4 -0
 	WARN_ON_ONCE(((op & REQ_OP_MASK) != REQ_OP_SCSI_IN) &&
 		     ((op & REQ_OP_MASK) != REQ_OP_SCSI_OUT));
+
+	if (sdev->host->nr_reserved_cmds)
+		flags |= BLK_MQ_REQ_RESERVED;
+
 	rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
 	if (IS_ERR(rq))
 		return NULL;

Can the if-statement be removed such that scsi_get_internal_cmd() fails if sdev->host->nr_reserved_cmds == 0? I'm concerned that otherwise it will be very hard to determine which requests are internal and which ones not from inside a blk_mq_tagset_busy_iter() callback.

Original idea was that one could use scsi_get_internal_cmd() even with nr_reserved_cmds == 0, but you are right that this will probably just lead to confusion.

Will be modifying it for the next round.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		        Kernel Storage Architect
[email protected]			               +49 911 74053 688
SUSE Software Solutions Germany GmbH, 90409 Nürnberg
GF: F. Imendörffer, HRB 36809 (AG Nürnberg)
#45 lore raw

PATCH v8 11/18 hpsa: move hpsa_hba_inquiry after scsi_add_host()

In reply to Hannes Reinecke

Move hpsa_hba_inquiry to after scsi_add_host() so that the host is fully initialized.

---
 drivers/scsi/hpsa.c | 37 +++++++++++++++++++------------------
 1 file changed, 19 insertions(+), 18 deletions(-)
drivers/scsi/hpsa.c +19 -18
index 38369766511c..c82f218cd1f6 100644
@@ -5874,6 +5874,22 @@ static int hpsa_scsi_host_alloc(struct ctlr_info *h)
 	return 0;
 }
 
+static void hpsa_hba_inquiry(struct ctlr_info *h)
+{
+	int rc;
+
+#define HBA_INQUIRY_BYTE_COUNT 64
+	h->hba_inquiry_data = kmalloc(HBA_INQUIRY_BYTE_COUNT, GFP_KERNEL);
+	if (!h->hba_inquiry_data)
+		return;
+	rc = hpsa_scsi_do_inquiry(h, RAID_CTLR_LUNID, 0,
+		h->hba_inquiry_data, HBA_INQUIRY_BYTE_COUNT);
+	if (rc != 0) {
+		kfree(h->hba_inquiry_data);
+		h->hba_inquiry_data = NULL;
+	}
+}
+
 static int hpsa_scsi_add_host(struct ctlr_info *h)
 {
 	int rv;
@@ -5883,6 +5899,9 @@ static int hpsa_scsi_add_host(struct ctlr_info *h)
 		dev_err(&h->pdev->dev, "scsi_add_host failed\n");
 		return rv;
 	}
+
+	hpsa_hba_inquiry(h);
+
 	scsi_scan_host(h->scsi_host);
 	return 0;
 }
@@ -7949,22 +7968,6 @@ static int hpsa_pci_init(struct ctlr_info *h)
 	return err;
 }
 
-static void hpsa_hba_inquiry(struct ctlr_info *h)
-{
-	int rc;
-
-#define HBA_INQUIRY_BYTE_COUNT 64
-	h->hba_inquiry_data = kmalloc(HBA_INQUIRY_BYTE_COUNT, GFP_KERNEL);
-	if (!h->hba_inquiry_data)
-		return;
-	rc = hpsa_scsi_do_inquiry(h, RAID_CTLR_LUNID, 0,
-		h->hba_inquiry_data, HBA_INQUIRY_BYTE_COUNT);
-	if (rc != 0) {
-		kfree(h->hba_inquiry_data);
-		h->hba_inquiry_data = NULL;
-	}
-}
-
 static int hpsa_init_reset_devices(struct pci_dev *pdev, u32 board_id)
 {
 	int rc, i;
@@ -8869,8 +8872,6 @@ static int hpsa_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	/* Turn the interrupts on so we can service requests */
 	h->access.set_intr_mask(h, HPSA_INTR_ON);
 
-	hpsa_hba_inquiry(h);
-
 	h->lastlogicals = kzalloc(sizeof(*(h->lastlogicals)), GFP_KERNEL);
 	if (!h->lastlogicals)
 		dev_info(&h->pdev->dev,
Signature · 1 line
2.29.2
#46 lore raw

PATCH v8 12/18 hpsa: use reserved commands

In reply to Hannes Reinecke

Enable the use of reserved commands, and drop the hand-crafted command allocation.

---
 drivers/scsi/hpsa.c | 202 ++++++++++++++++----------------------------
 drivers/scsi/hpsa.h |   3 +-
 2 files changed, 76 insertions(+), 129 deletions(-)
drivers/scsi/hpsa.c +74 -128
index c82f218cd1f6..7a4fcd40364b 100644
@@ -245,10 +245,6 @@ static struct hpsa_scsi_dev_t
 	*hpsa_find_device_by_sas_rphy(struct ctlr_info *h,
 		struct sas_rphy *rphy);
 
-#define SCSI_CMD_BUSY ((struct scsi_cmnd *)&hpsa_cmd_busy)
-static const struct scsi_cmnd hpsa_cmd_busy;
-#define SCSI_CMD_IDLE ((struct scsi_cmnd *)&hpsa_cmd_idle)
-static const struct scsi_cmnd hpsa_cmd_idle;
 static int number_of_controllers;
 
 static irqreturn_t do_hpsa_intr_intx(int irq, void *dev_id);
@@ -266,7 +262,7 @@ static int hpsa_compat_ioctl(struct scsi_device *dev, unsigned int cmd,
 #endif
 
 static void cmd_free(struct ctlr_info *h, struct CommandList *c);
-static struct CommandList *cmd_alloc(struct ctlr_info *h);
+static struct CommandList *cmd_alloc(struct ctlr_info *h, u8 direction);
 static void cmd_tagged_free(struct ctlr_info *h, struct CommandList *c);
 static struct CommandList *cmd_tagged_alloc(struct ctlr_info *h,
 					    struct scsi_cmnd *scmd);
@@ -347,7 +343,7 @@ static inline struct ctlr_info *shost_to_hba(struct Scsi_Host *sh)
 
 static inline bool hpsa_is_cmd_idle(struct CommandList *c)
 {
-	return c->scsi_cmd == SCSI_CMD_IDLE;
+	return c->scsi_cmd == NULL;
 }
 
 /* extract sense key, asc, and ascq from sense data.  -1 means invalid. */
@@ -2462,7 +2458,12 @@ static void hpsa_cmd_resolve_events(struct ctlr_info *h,
 	 * this command has completed.  Then, check to see if the handler is
 	 * waiting for this command, and, if so, wake it.
 	 */
-	c->scsi_cmd = SCSI_CMD_IDLE;
+	if (c->scsi_cmd && c->cmd_type == CMD_IOCTL_PEND) {
+		struct scsi_cmnd *scmd = c->scsi_cmd;
+
+		scsi_put_internal_cmd(scmd);
+	}
+	c->scsi_cmd = NULL;
 	mb();	/* Declare command idle before checking for pending events. */
 	if (dev) {
 		atomic_dec(&dev->commands_outstanding);
@@ -3004,7 +3005,7 @@ static int hpsa_do_receive_diagnostic(struct ctlr_info *h, u8 *scsi3addr,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 	if (fill_cmd(c, RECEIVE_DIAGNOSTIC, h, buf, bufsize,
 			page, scsi3addr, TYPE_CMD)) {
 		rc = -1;
@@ -3056,7 +3057,7 @@ static int hpsa_scsi_do_inquiry(struct ctlr_info *h, unsigned char *scsi3addr,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	if (fill_cmd(c, HPSA_INQUIRY, h, buf, bufsize,
 			page, scsi3addr, TYPE_CMD)) {
@@ -3084,7 +3085,7 @@ static int hpsa_send_reset(struct ctlr_info *h, struct hpsa_scsi_dev_t *dev,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_NONE);
 	c->device = dev;
 
 	/* fill_cmd can't fail here, no data buffer to map. */
@@ -3310,7 +3311,7 @@ static int hpsa_get_raid_map(struct ctlr_info *h,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	if (fill_cmd(c, HPSA_GET_RAID_MAP, h, &this_device->raid_map,
 			sizeof(this_device->raid_map), 0,
@@ -3352,7 +3353,7 @@ static int hpsa_bmic_sense_subsystem_information(struct ctlr_info *h,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	rc = fill_cmd(c, BMIC_SENSE_SUBSYSTEM_INFORMATION, h, buf, bufsize,
 		0, RAID_CTLR_LUNID, TYPE_CMD);
@@ -3383,7 +3384,7 @@ static int hpsa_bmic_id_controller(struct ctlr_info *h,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	rc = fill_cmd(c, BMIC_IDENTIFY_CONTROLLER, h, buf, bufsize,
 		0, RAID_CTLR_LUNID, TYPE_CMD);
@@ -3412,7 +3413,7 @@ static int hpsa_bmic_id_physical_device(struct ctlr_info *h,
 	struct CommandList *c;
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 	rc = fill_cmd(c, BMIC_IDENTIFY_PHYSICAL_DEVICE, h, buf, bufsize,
 		0, RAID_CTLR_LUNID, TYPE_CMD);
 	if (rc)
@@ -3489,7 +3490,7 @@ static void hpsa_get_enclosure_info(struct ctlr_info *h,
 		goto out;
 	}
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	rc = fill_cmd(c, BMIC_SENSE_STORAGE_BOX_PARAMS, h, bssbp,
 			sizeof(*bssbp), 0, RAID_CTLR_LUNID, TYPE_CMD);
@@ -3745,7 +3746,7 @@ static int hpsa_scsi_do_report_luns(struct ctlr_info *h, int logical,
 	unsigned char scsi3addr[8];
 	struct ErrorInfo *ei;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	/* address the controller */
 	memset(scsi3addr, 0, sizeof(scsi3addr));
@@ -3886,7 +3887,7 @@ static unsigned char hpsa_volume_offline(struct ctlr_info *h,
 #define ASCQ_LUN_NOT_READY_FORMAT_IN_PROGRESS 0x04
 #define ASCQ_LUN_NOT_READY_INITIALIZING_CMD_REQ 0x02
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_NONE);
 
 	(void) fill_cmd(c, TEST_UNIT_READY, h, NULL, 0, 0, scsi3addr, TYPE_CMD);
 	rc = hpsa_scsi_do_simple_cmd(h, c, DEFAULT_REPLY_QUEUE,
@@ -5542,7 +5543,6 @@ static void hpsa_cmd_init(struct ctlr_info *h, int index,
 	c->ErrDesc.Addr = cpu_to_le64((u64) err_dma_handle);
 	c->ErrDesc.Len = cpu_to_le32((u32) sizeof(*c->err_info));
 	c->h = h;
-	c->scsi_cmd = SCSI_CMD_IDLE;
 }
 
 static void hpsa_preinitialize_commands(struct ctlr_info *h)
@@ -5857,12 +5857,12 @@ static int hpsa_scsi_host_alloc(struct ctlr_info *h)
 
 	sh->io_port = 0;
 	sh->n_io_port = 0;
-	sh->this_id = -1;
 	sh->max_channel = 3;
 	sh->max_cmd_len = MAX_COMMAND_SIZE;
 	sh->max_lun = HPSA_MAX_LUN;
 	sh->max_id = HPSA_MAX_LUN;
 	sh->can_queue = h->nr_cmds - HPSA_NRESERVED_CMDS;
+	sh->nr_reserved_cmds = HPSA_NRESERVED_CMDS;
 	sh->cmd_per_lun = sh->can_queue;
 	sh->sg_tablesize = h->maxsgentries;
 	sh->transportt = hpsa_sas_transport_template;
@@ -5899,30 +5899,18 @@ static int hpsa_scsi_add_host(struct ctlr_info *h)
 		dev_err(&h->pdev->dev, "scsi_add_host failed\n");
 		return rv;
 	}
-
+	h->raid_ctrl_sdev = scsi_get_host_dev(h->scsi_host);
+	if (!h->raid_ctrl_sdev) {
+		dev_err(&h->pdev->dev,
+			"allocate raid controller device failed\n");
+		return -ENOMEM;
+	}
 	hpsa_hba_inquiry(h);
 
 	scsi_scan_host(h->scsi_host);
 	return 0;
 }
 
-/*
- * The block layer has already gone to the trouble of picking out a unique,
- * small-integer tag for this request.  We use an offset from that value as
- * an index to select our command block.  (The offset allows us to reserve the
- * low-numbered entries for our own uses.)
- */
-static int hpsa_get_cmd_index(struct scsi_cmnd *scmd)
-{
-	int idx = scmd->request->tag;
-
-	if (idx < 0)
-		return idx;
-
-	/* Offset to leave space for internal cmds. */
-	return idx += HPSA_NRESERVED_CMDS;
-}
-
 /*
  * Send a TEST_UNIT_READY command to the specified LUN using the specified
  * reply queue; returns zero if the unit is ready, and non-zero otherwise.
@@ -6006,7 +5994,7 @@ static int wait_for_device_to_become_ready(struct ctlr_info *h,
 	int rc = 0;
 	struct CommandList *c;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_NONE);
 
 	/*
 	 * If no specific reply queue was requested, then send the TUR
@@ -6079,7 +6067,7 @@ static int hpsa_eh_device_reset_handler(struct scsi_cmnd *scsicmd)
 	if (lockup_detected(h)) {
 		snprintf(msg, sizeof(msg),
 			 "cmd %d RESET FAILED, lockup detected",
-			 hpsa_get_cmd_index(scsicmd));
+			 scsicmd->request->tag);
 		hpsa_show_dev_msg(KERN_WARNING, h, dev, msg);
 		rc = FAILED;
 		goto return_reset_status;
@@ -6089,7 +6077,7 @@ static int hpsa_eh_device_reset_handler(struct scsi_cmnd *scsicmd)
 	if (detect_controller_lockup(h)) {
 		snprintf(msg, sizeof(msg),
 			 "cmd %d RESET FAILED, new lockup detected",
-			 hpsa_get_cmd_index(scsicmd));
+			 scsicmd->request->tag);
 		hpsa_show_dev_msg(KERN_WARNING, h, dev, msg);
 		rc = FAILED;
 		goto return_reset_status;
@@ -6152,12 +6140,12 @@ static int hpsa_eh_device_reset_handler(struct scsi_cmnd *scsicmd)
 static struct CommandList *cmd_tagged_alloc(struct ctlr_info *h,
 					    struct scsi_cmnd *scmd)
 {
-	int idx = hpsa_get_cmd_index(scmd);
+	int idx = scmd->request->tag;
 	struct CommandList *c = h->cmd_pool + idx;
 
-	if (idx < HPSA_NRESERVED_CMDS || idx >= h->nr_cmds) {
+	if (idx < 0 || idx >= h->nr_cmds) {
 		dev_err(&h->pdev->dev, "Bad block tag: %d not in [%d..%d]\n",
-			idx, HPSA_NRESERVED_CMDS, h->nr_cmds - 1);
+			idx, 0, h->nr_cmds - 1);
 		/* The index value comes from the block layer, so if it's out of
 		 * bounds, it's probably not our bug.
 		 */
@@ -6200,62 +6188,32 @@ static void cmd_tagged_free(struct ctlr_info *h, struct CommandList *c)
 	 * else to free it, because it is accessed by index.
 	 */
 	(void)atomic_dec(&c->refcount);
+	c->scsi_cmd = NULL;
 }
 
-/*
- * For operations that cannot sleep, a command block is allocated at init,
- * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
- * which ones are free or in use.  Lock must be held when calling this.
- * cmd_free() is the complement.
- * This function never gives up and returns NULL.  If it hangs,
- * another thread must call cmd_free() to free some tags.
- */
-
-static struct CommandList *cmd_alloc(struct ctlr_info *h)
+static struct CommandList *cmd_alloc(struct ctlr_info *h, u8 direction)
 {
+	struct scsi_cmnd *scmd;
 	struct CommandList *c;
-	int refcount, i;
-	int offset = 0;
+	int idx;
 
-	/*
-	 * There is some *extremely* small but non-zero chance that that
-	 * multiple threads could get in here, and one thread could
-	 * be scanning through the list of bits looking for a free
-	 * one, but the free ones are always behind him, and other
-	 * threads sneak in behind him and eat them before he can
-	 * get to them, so that while there is always a free one, a
-	 * very unlucky thread might be starved anyway, never able to
-	 * beat the other threads.  In reality, this happens so
-	 * infrequently as to be indistinguishable from never.
-	 *
-	 * Note that we start allocating commands before the SCSI host structure
-	 * is initialized.  Since the search starts at bit zero, this
-	 * all works, since we have at least one command structure available;
-	 * however, it means that the structures with the low indexes have to be
-	 * reserved for driver-initiated requests, while requests from the block
-	 * layer will use the higher indexes.
-	 */
-
-	for (;;) {
-		i = find_next_zero_bit(h->cmd_pool_bits,
-					HPSA_NRESERVED_CMDS,
-					offset);
-		if (unlikely(i >= HPSA_NRESERVED_CMDS)) {
-			offset = 0;
-			continue;
-		}
-		c = h->cmd_pool + i;
-		refcount = atomic_inc_return(&c->refcount);
-		if (unlikely(refcount > 1)) {
-			cmd_free(h, c); /* already in use */
-			offset = (i + 1) % HPSA_NRESERVED_CMDS;
-			continue;
-		}
-		set_bit(i & (BITS_PER_LONG - 1),
-			h->cmd_pool_bits + (i / BITS_PER_LONG));
-		break; /* it's ours now. */
+	scmd = scsi_get_internal_cmd(h->raid_ctrl_sdev,
+				     (direction & XFER_WRITE) ?
+				     REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
+	if (!scmd) {
+		dev_warn(&h->pdev->dev, "failed to allocate reserved cmd\n");
+		return NULL;
 	}
-	hpsa_cmd_partial_init(h, i, c);
+	idx = scmd->request->tag;
+	c = cmd_tagged_alloc(h, scmd);
+	if (!c) {
+		dev_warn(&h->pdev->dev, "failed to allocate reserved cmd %u\n",
+			 idx);
+		scsi_put_internal_cmd(scmd);
+		return NULL;
+	}
+	hpsa_cmd_partial_init(h, idx, c);
+	c->scsi_cmd = scmd;
 	c->device = NULL;
 
 	/*
@@ -6263,24 +6221,24 @@ static struct CommandList *cmd_alloc(struct ctlr_info *h)
 	 * retried.
 	 */
 	c->retry_pending = false;
-
+	c->cmd_type = CMD_IOCTL_PEND;
+	dev_dbg(&h->pdev->dev, "using reserved cmd %u\n", idx);
 	return c;
 }
 
-/*
- * This is the complementary operation to cmd_alloc().  Note, however, in some
- * corner cases it may also be used to free blocks allocated by
- * cmd_tagged_alloc() in which case the ref-count decrement does the trick and
- * the clear-bit is harmless.
- */
 static void cmd_free(struct ctlr_info *h, struct CommandList *c)
 {
-	if (atomic_dec_and_test(&c->refcount)) {
-		int i;
+	struct scsi_cmnd *scmd = c->scsi_cmd;
 
-		i = c - h->cmd_pool;
-		clear_bit(i & (BITS_PER_LONG - 1),
-			  h->cmd_pool_bits + (i / BITS_PER_LONG));
+	if (!scmd) {
+		dev_warn(&h->pdev->dev, "freeing idle cmd\n");
+		return;
+	}
+	cmd_tagged_free(h, c);
+	if (c->cmd_type == CMD_IOCTL_PEND) {
+		dev_dbg(&h->pdev->dev, "returning reserved cmd %u\n",
+			scmd->request->tag);
+		scsi_put_internal_cmd(scmd);
 	}
 }
 
@@ -6447,11 +6405,8 @@ static int hpsa_passthru_ioctl(struct ctlr_info *h,
 			memset(buff, 0, iocommand->buf_size);
 		}
 	}
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, iocommand->Request.Type.Direction);
 
-	/* Fill in the command type */
-	c->cmd_type = CMD_IOCTL_PEND;
-	c->scsi_cmd = SCSI_CMD_BUSY;
 	/* Fill in Command Header */
 	c->Header.ReplyQueue = 0; /* unused in simple mode */
 	if (iocommand->buf_size > 0) {	/* buffer to fill */
@@ -6564,10 +6519,8 @@ static int hpsa_big_passthru_ioctl(struct ctlr_info *h,
 		data_ptr += sz;
 		sg_used++;
 	}
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, ioc->Request.Type.Direction);
 
-	c->cmd_type = CMD_IOCTL_PEND;
-	c->scsi_cmd = SCSI_CMD_BUSY;
 	c->Header.ReplyQueue = 0;
 	c->Header.SGList = (u8) sg_used;
 	c->Header.SGTotal = cpu_to_le16(sg_used);
@@ -6698,7 +6651,7 @@ static void hpsa_send_host_reset(struct ctlr_info *h, u8 reset_type)
 {
 	struct CommandList *c;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_NONE);
 
 	/* fill_cmd can't fail here, no data buffer to map */
 	(void) fill_cmd(c, HPSA_DEVICE_RESET_MSG, h, NULL, 0, 0,
@@ -6719,8 +6672,6 @@ static int fill_cmd(struct CommandList *c, u8 cmd, struct ctlr_info *h,
 {
 	enum dma_data_direction dir = DMA_NONE;
 
-	c->cmd_type = CMD_IOCTL_PEND;
-	c->scsi_cmd = SCSI_CMD_BUSY;
 	c->Header.ReplyQueue = 0;
 	if (buff != NULL && size > 0) {
 		c->Header.SGList = 1;
@@ -8032,8 +7983,6 @@ static int hpsa_init_reset_devices(struct pci_dev *pdev, u32 board_id)
 
 static void hpsa_free_cmd_pool(struct ctlr_info *h)
 {
-	kfree(h->cmd_pool_bits);
-	h->cmd_pool_bits = NULL;
 	if (h->cmd_pool) {
 		dma_free_coherent(&h->pdev->dev,
 				h->nr_cmds * sizeof(struct CommandList),
@@ -8054,17 +8003,13 @@ static void hpsa_free_cmd_pool(struct ctlr_info *h)
 
 static int hpsa_alloc_cmd_pool(struct ctlr_info *h)
 {
-	h->cmd_pool_bits = kcalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG),
-				   sizeof(unsigned long),
-				   GFP_KERNEL);
 	h->cmd_pool = dma_alloc_coherent(&h->pdev->dev,
 		    h->nr_cmds * sizeof(*h->cmd_pool),
 		    &h->cmd_pool_dhandle, GFP_KERNEL);
 	h->errinfo_pool = dma_alloc_coherent(&h->pdev->dev,
 		    h->nr_cmds * sizeof(*h->errinfo_pool),
 		    &h->errinfo_pool_dhandle, GFP_KERNEL);
-	if ((h->cmd_pool_bits == NULL)
-	    || (h->cmd_pool == NULL)
+	if ((h->cmd_pool == NULL)
 	    || (h->errinfo_pool == NULL)) {
 		dev_err(&h->pdev->dev, "out of memory in %s", __func__);
 		goto clean_up;
@@ -8945,7 +8890,7 @@ static void hpsa_flush_cache(struct ctlr_info *h)
 	if (!flush_buf)
 		return;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_NONE);
 
 	if (fill_cmd(c, HPSA_CACHE_FLUSH, h, flush_buf, 4, 0,
 		RAID_CTLR_LUNID, TYPE_CMD)) {
@@ -8980,7 +8925,7 @@ static void hpsa_disable_rld_caching(struct ctlr_info *h)
 	if (!options)
 		return;
 
-	c = cmd_alloc(h);
+	c = cmd_alloc(h, XFER_READ);
 
 	/* first, get the current diag options settings */
 	if (fill_cmd(c, BMIC_SENSE_DIAG_OPTIONS, h, options, 4, 0,
@@ -9030,11 +8975,10 @@ static void __hpsa_shutdown(struct pci_dev *pdev)
 	struct ctlr_info *h;
 
 	h = pci_get_drvdata(pdev);
-	/* Turn board interrupts off  and send the flush cache command
-	 * sendcmd will turn off interrupt, and send the flush...
-	 * To write all data in the battery backed cache to disks
+	/*
+	 * Turn board interrupts off;
+	 * flush cache command has already been sent.
 	 */
-	hpsa_flush_cache(h);
 	h->access.set_intr_mask(h, HPSA_INTR_OFF);
 	hpsa_free_irqs(h);			/* init_one 4 */
 	hpsa_disable_interrupt_mode(h);		/* pci_init 2 */
@@ -9042,6 +8986,7 @@ static void __hpsa_shutdown(struct pci_dev *pdev)
 
 static void hpsa_shutdown(struct pci_dev *pdev)
 {
+	hpsa_flush_cache(pci_get_drvdata(pdev));
 	__hpsa_shutdown(pdev);
 	pci_disable_device(pdev);
 }
@@ -9086,6 +9031,7 @@ static void hpsa_remove_one(struct pci_dev *pdev)
 	 * when multipath is enabled. There can be SYNCHRONIZE CACHE
 	 * operations which cannot complete and will hang the system.
 	 */
+	hpsa_flush_cache(h);
 	if (h->scsi_host)
 		scsi_remove_host(h->scsi_host);		/* init_one 8 */
 	/* includes hpsa_free_irqs - init_one 4 */
drivers/scsi/hpsa.h +2 -1
index 99b0750850b2..93e9fd9b931c 100644
@@ -206,7 +206,6 @@ struct ctlr_info {
 	dma_addr_t		ioaccel2_cmd_pool_dhandle;
 	struct ErrorInfo 	*errinfo_pool;
 	dma_addr_t		errinfo_pool_dhandle;
-	unsigned long  		*cmd_pool_bits;
 	int			scan_finished;
 	u8			scan_waiting : 1;
 	spinlock_t		scan_lock;
@@ -216,6 +215,8 @@ struct ctlr_info {
 	spinlock_t devlock; /* to protect hba[ctlr]->dev[];  */
 	int ndevices; /* number of used elements in .dev[] array. */
 	struct hpsa_scsi_dev_t *dev[HPSA_MAX_DEVICES];
+	struct scsi_device *raid_ctrl_sdev;
+
 	/*
 	 * Performant mode tables.
 	 */
Signature · 1 line
2.29.2
#47 lore raw

PATCH v8 13/18 hpsa: use scsi_host_busy_iter() to traverse outstanding commands

In reply to Hannes Reinecke

Replace all hand-crafted command iterations with
scsi_host_busy_iter() calls.

---
 drivers/scsi/hpsa.c | 117 +++++++++++++++++++++++++-------------------
 1 file changed, 67 insertions(+), 50 deletions(-)
drivers/scsi/hpsa.c +67 -50
index 7a4fcd40364b..282338944a85 100644
@@ -1820,30 +1820,26 @@ static int hpsa_add_device(struct ctlr_info *h, struct hpsa_scsi_dev_t *device)
 	return rc;
 }
 
-static int hpsa_find_outstanding_commands_for_dev(struct ctlr_info *h,
-						struct hpsa_scsi_dev_t *dev)
-{
-	int i;
-	int count = 0;
-
-	for (i = 0; i < h->nr_cmds; i++) {
-		struct CommandList *c = h->cmd_pool + i;
-		int refcount = atomic_inc_return(&c->refcount);
-
-		if (refcount > 1 && hpsa_cmd_dev_match(h, c, dev,
-				dev->scsi3addr)) {
-			unsigned long flags;
+struct hpsa_command_iter_data {
+	struct ctlr_info *h;
+	struct hpsa_scsi_dev_t *dev;
+	unsigned char *scsi3addr;
+	int count;
+};
 
-			spin_lock_irqsave(&h->lock, flags);	/* Implied MB */
-			if (!hpsa_is_cmd_idle(c))
-				++count;
-			spin_unlock_irqrestore(&h->lock, flags);
-		}
+static bool hpsa_find_outstanding_commands_iter(struct scsi_cmnd *sc,
+						void *data, bool reserved)
+{
+	struct hpsa_command_iter_data *iter_data = data;
+	struct ctlr_info *h = iter_data->h;
+	struct hpsa_scsi_dev_t *dev = iter_data->dev;
+	struct CommandList *c = h->cmd_pool + sc->request->tag;
 
-		cmd_free(h, c);
+	if (hpsa_cmd_dev_match(h, c, dev, dev->scsi3addr)) {
+		iter_data->count++;
+		return false;
 	}
-
-	return count;
+	return true;
 }
 
 #define NUM_WAIT 20
@@ -1853,13 +1849,20 @@ static void hpsa_wait_for_outstanding_commands_for_dev(struct ctlr_info *h,
 	int cmds = 0;
 	int waits = 0;
 	int num_wait = NUM_WAIT;
+	struct hpsa_command_iter_data iter_data = {
+		.h = h,
+		.dev = device,
+	};
 
 	if (device->external)
 		num_wait = HPSA_EH_PTRAID_TIMEOUT;
 
 	while (1) {
-		cmds = hpsa_find_outstanding_commands_for_dev(h, device);
-		if (cmds == 0)
+		iter_data.count = 0;
+		scsi_host_busy_iter(h->scsi_host,
+				    hpsa_find_outstanding_commands_iter,
+				    &iter_data);
+		if (iter_data.count == 0)
 			break;
 		if (++waits > num_wait)
 			break;
@@ -8180,27 +8183,34 @@ static void hpsa_undo_allocations_after_kdump_soft_reset(struct ctlr_info *h)
 	kfree(h);				/* init_one 1 */
 }
 
+static bool fail_all_outstanding_cmds_iter(struct scsi_cmnd *sc, void *data,
+					   bool reserved)
+{
+	struct hpsa_command_iter_data *iter_data = data;
+	struct ctlr_info *h = iter_data->h;
+	struct CommandList *c = h->cmd_pool + sc->request->tag;
+
+	c->err_info->CommandStatus = CMD_CTLR_LOCKUP;
+	finish_cmd(c);
+	atomic_dec(&h->commands_outstanding);
+	iter_data->count++;
+
+	return true;
+}
+
 /* Called when controller lockup detected. */
 static void fail_all_outstanding_cmds(struct ctlr_info *h)
 {
-	int i, refcount;
-	struct CommandList *c;
-	int failcount = 0;
+	struct hpsa_command_iter_data iter_data = {
+		.h = h,
+		.count = 0,
+	};
 
 	flush_workqueue(h->resubmit_wq); /* ensure all cmds are fully built */
-	for (i = 0; i < h->nr_cmds; i++) {
-		c = h->cmd_pool + i;
-		refcount = atomic_inc_return(&c->refcount);
-		if (refcount > 1) {
-			c->err_info->CommandStatus = CMD_CTLR_LOCKUP;
-			finish_cmd(c);
-			atomic_dec(&h->commands_outstanding);
-			failcount++;
-		}
-		cmd_free(h, c);
-	}
+	scsi_host_busy_iter(h->scsi_host,
+			    fail_all_outstanding_cmds_iter, &iter_data);
 	dev_warn(&h->pdev->dev,
-		"failed %d commands in fail_all\n", failcount);
+		"failed %d commands in fail_all\n", iter_data.count);
 }
 
 static void set_lockup_detected_for_all_cpus(struct ctlr_info *h, u32 value)
@@ -9499,22 +9509,29 @@ static int is_accelerated_cmd(struct CommandList *c)
 	return c->cmd_type == CMD_IOACCEL1 || c->cmd_type == CMD_IOACCEL2;
 }
 
+static bool hpsa_drain_accel_commands_iter(struct scsi_cmnd *sc, void *data,
+					   bool reserved)
+{
+	struct hpsa_command_iter_data *iter_data = data;
+	struct ctlr_info *h = iter_data->h;
+	struct CommandList *c = h->cmd_pool + sc->request->tag;
+
+	iter_data->count += is_accelerated_cmd(c);
+	return true;
+}
+
 static void hpsa_drain_accel_commands(struct ctlr_info *h)
 {
-	struct CommandList *c = NULL;
-	int i, accel_cmds_out;
-	int refcount;
+	struct hpsa_command_iter_data iter_data = {
+		.h = h,
+	};
 
 	do { /* wait for all outstanding ioaccel commands to drain out */
-		accel_cmds_out = 0;
-		for (i = 0; i < h->nr_cmds; i++) {
-			c = h->cmd_pool + i;
-			refcount = atomic_inc_return(&c->refcount);
-			if (refcount > 1) /* Command is allocated */
-				accel_cmds_out += is_accelerated_cmd(c);
-			cmd_free(h, c);
-		}
-		if (accel_cmds_out <= 0)
+		iter_data.count = 0;
+		scsi_host_busy_iter(h->scsi_host,
+				    hpsa_drain_accel_commands_iter,
+				    &iter_data);
+		if (iter_data.count <= 0)
 			break;
 		msleep(100);
 	} while (1);
Signature · 1 line
2.29.2
#48 lore raw

PATCH v8 14/18 hpsa: drop refcount field from CommandList

In reply to Hannes Reinecke

Field is now unused, so drop it.

---
 drivers/scsi/hpsa.c     | 11 ++---------
 drivers/scsi/hpsa_cmd.h | 10 ----------
 2 files changed, 2 insertions(+), 19 deletions(-)
drivers/scsi/hpsa.c +2 -9
index 282338944a85..61f993704e23 100644
@@ -5533,8 +5533,8 @@ static void hpsa_cmd_init(struct ctlr_info *h, int index,
 {
 	dma_addr_t cmd_dma_handle, err_dma_handle;
 
-	/* Zero out all of commandlist except the last field, refcount */
-	memset(c, 0, offsetof(struct CommandList, refcount));
+	/* Zero out all of commandlist */
+	memset(c, 0, sizeof(struct CommandList));
 	c->Header.tag = cpu_to_le64((u64) (index << DIRECT_LOOKUP_SHIFT));
 	cmd_dma_handle = h->cmd_pool_dhandle + index * sizeof(*c);
 	c->err_info = h->errinfo_pool + index;
@@ -5556,7 +5556,6 @@ static void hpsa_preinitialize_commands(struct ctlr_info *h)
 		struct CommandList *c = h->cmd_pool + i;
 
 		hpsa_cmd_init(h, i, c);
-		atomic_set(&c->refcount, 0);
 	}
 }
 
@@ -6172,7 +6171,6 @@ static struct CommandList *cmd_tagged_alloc(struct ctlr_info *h,
 		return NULL;
 	}
 
-	atomic_inc(&c->refcount);
 	hpsa_cmd_partial_init(h, idx, c);
 
 	/*
@@ -6186,11 +6184,6 @@ static struct CommandList *cmd_tagged_alloc(struct ctlr_info *h,
 
 static void cmd_tagged_free(struct ctlr_info *h, struct CommandList *c)
 {
-	/*
-	 * Release our reference to the block.  We don't need to do anything
-	 * else to free it, because it is accessed by index.
-	 */
-	(void)atomic_dec(&c->refcount);
 	c->scsi_cmd = NULL;
 }
 
drivers/scsi/hpsa_cmd.h +0 -10
index ba6a3aa8d954..04c92c94cc6c 100644
@@ -454,18 +454,8 @@ struct CommandList {
 
 	bool retry_pending;
 	struct hpsa_scsi_dev_t *device;
-	atomic_t refcount; /* Must be last to avoid memset in hpsa_cmd_init() */
 } __aligned(COMMANDLIST_ALIGNMENT);
 
-/*
- * Make sure our embedded atomic variable is aligned. Otherwise we break atomic
- * operations on architectures that don't support unaligned atomics like IA64.
- *
- * The assert guards against reintroductin against unwanted __packed to
- * the struct CommandList.
- */
-static_assert(offsetof(struct CommandList, refcount) % __alignof__(atomic_t) == 0);
-
 /* Max S/G elements in I/O accelerator command */
 #define IOACCEL1_MAXSGENTRIES           24
 #define IOACCEL2_MAXSGENTRIES		28
Signature · 1 line
2.29.2
#49 lore raw

PATCH v8 15/18 aacraid: move scsi_add_host()

In reply to Hannes Reinecke

Move the call to scsi_add_host() so that the Scsi_Host structure is initialized before any I/O is sent.

---
 drivers/scsi/aacraid/linit.c | 31 +++++++++++++++----------------
 1 file changed, 15 insertions(+), 16 deletions(-)
drivers/scsi/aacraid/linit.c +15 -16
index 3168915adaa7..e5d89b309c3a 100644
@@ -1639,6 +1639,9 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	shost->irq = pdev->irq;
 	shost->unique_id = unique_id;
 	shost->max_cmd_len = 16;
+	shost->max_id = MAXIMUM_NUM_CONTAINERS;
+	shost->max_lun = AAC_MAX_LUN;
+	shost->sg_tablesize = HBA_MAX_SG_SEPARATE;
 
 	if (aac_cfg_major == AAC_CHARDEV_NEEDS_REINIT)
 		aac_init_char();
@@ -1677,7 +1680,7 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	aac->base_size = AAC_MIN_FOOTPRINT_SIZE;
 	if ((*aac_drivers[index].init)(aac)) {
 		error = -ENODEV;
-		goto out_unmap;
+		goto out_free_fibs;
 	}
 
 	if (aac->sync_mode) {
@@ -1703,9 +1706,15 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 		printk(KERN_ERR "aacraid: Unable to create command thread.\n");
 		error = PTR_ERR(aac->thread);
 		aac->thread = NULL;
-		goto out_deinit;
+		goto out_unmap;
 	}
 
+	pci_set_drvdata(pdev, shost);
+
+	error = scsi_add_host(shost, &pdev->dev);
+	if (error)
+		goto out_deinit;
+
 	aac->maximum_num_channels = aac_drivers[index].channels;
 	error = aac_get_adapter_info(aac);
 	if (error < 0)
@@ -1764,18 +1773,6 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (!aac->sa_firmware && aac_drivers[index].quirks & AAC_QUIRK_SRC)
 		aac_intr_normal(aac, 0, 2, 0, NULL);
 
-	/*
-	 * dmb - we may need to move the setting of these parms somewhere else once
-	 * we get a fib that can report the actual numbers
-	 */
-	shost->max_lun = AAC_MAX_LUN;
-
-	pci_set_drvdata(pdev, shost);
-
-	error = scsi_add_host(shost, &pdev->dev);
-	if (error)
-		goto out_deinit;
-
 	aac_scan_host(aac);
 
 	pci_enable_pcie_error_reporting(pdev);
@@ -1792,10 +1789,12 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 				  aac->comm_addr, aac->comm_phys);
 	kfree(aac->queues);
 	aac_adapter_ioremap(aac, 0);
-	kfree(aac->fibs);
 	kfree(aac->fsa_dev);
+ out_free_fibs:
+	kfree(aac->fibs);
  out_free_host:
 	scsi_host_put(shost);
+	pci_set_drvdata(pdev, NULL);
  out_disable_pdev:
 	pci_disable_device(pdev);
  out:
@@ -1903,9 +1902,9 @@ static void aac_remove_one(struct pci_dev *pdev)
 	struct aac_dev *aac = (struct aac_dev *)shost->hostdata;
 
 	aac_cancel_rescan_worker(aac);
-	scsi_remove_host(shost);
 
 	__aac_shutdown(aac);
+	scsi_remove_host(shost);
 	aac_fib_map_free(aac);
 	dma_free_coherent(&aac->pdev->dev, aac->comm_size, aac->comm_addr,
 			  aac->comm_phys);
Signature · 1 line
2.29.2
#50 lore raw

PATCH v8 16/18 aacraid: store target id in host_scribble

In reply to Hannes Reinecke

The probe_container mechanism requires a target id to be present, even if the device itself isn't present (yet).
As we're now allocating internal commands the target id of the device is immutable, so store the requested target id in the host_scribble field.

---
 drivers/scsi/aacraid/aachba.c | 53 +++++++++++++++++++++++++----------
 1 file changed, 38 insertions(+), 15 deletions(-)
drivers/scsi/aacraid/aachba.c +38 -15
index f1f62b5da8b7..ef59303db9b9 100644
@@ -609,9 +609,11 @@ static int aac_get_container_name(struct scsi_cmnd * scsicmd)
 
 static int aac_probe_container_callback2(struct scsi_cmnd * scsicmd)
 {
-	struct fsa_dev_info *fsa_dev_ptr = ((struct aac_dev *)(scsicmd->device->host->hostdata))->fsa_dev;
+	struct aac_dev *dev = (struct aac_dev *)(scsicmd->device->host->hostdata);
+	struct fsa_dev_info *fsa_dev_ptr = dev->fsa_dev;
 
-	if ((fsa_dev_ptr[scmd_id(scsicmd)].valid & 1))
+	if (scmd_id(scsicmd) < dev->maximum_num_containers &&
+	    (fsa_dev_ptr[scmd_id(scsicmd)].valid & 1))
 		return aac_scsi_cmd(scsicmd);
 
 	scsicmd->result = DID_NO_CONNECT << 16;
@@ -624,6 +626,7 @@ static void _aac_probe_container2(void * context, struct fib * fibptr)
 	struct fsa_dev_info *fsa_dev_ptr;
 	int (*callback)(struct scsi_cmnd *);
 	struct scsi_cmnd * scsicmd = (struct scsi_cmnd *)context;
+	int cid = scmd_id(scsicmd);
 	int i;
 
 
@@ -631,12 +634,15 @@ static void _aac_probe_container2(void * context, struct fib * fibptr)
 		return;
 
 	scsicmd->SCp.Status = 0;
+	if (scsicmd->host_scribble)
+		cid = *(int *)scsicmd->host_scribble;
+
 	fsa_dev_ptr = fibptr->dev->fsa_dev;
-	if (fsa_dev_ptr) {
+	if (fsa_dev_ptr && cid < fibptr->dev->maximum_num_containers) {
 		struct aac_mount * dresp = (struct aac_mount *) fib_data(fibptr);
 		__le32 sup_options2;
 
-		fsa_dev_ptr += scmd_id(scsicmd);
+		fsa_dev_ptr += cid;
 		sup_options2 =
 			fibptr->dev->supplement_adapter_info.supported_options2;
 
@@ -671,7 +677,6 @@ static void _aac_probe_container2(void * context, struct fib * fibptr)
 		scsicmd->SCp.Status = le32_to_cpu(dresp->count);
 	}
 	aac_fib_complete(fibptr);
-	aac_fib_free(fibptr);
 	callback = (int (*)(struct scsi_cmnd *))(scsicmd->SCp.ptr);
 	scsicmd->SCp.ptr = NULL;
 	(*callback)(scsicmd);
@@ -683,6 +688,7 @@ static void _aac_probe_container1(void * context, struct fib * fibptr)
 	struct scsi_cmnd * scsicmd;
 	struct aac_mount * dresp;
 	struct aac_query_mount *dinfo;
+	int cid;
 	int status;
 
 	dresp = (struct aac_mount *) fib_data(fibptr);
@@ -695,10 +701,15 @@ static void _aac_probe_container1(void * context, struct fib * fibptr)
 		}
 	}
 	scsicmd = (struct scsi_cmnd *) context;
-
 	if (!aac_valid_context(scsicmd, fibptr))
 		return;
-
+	cid = scmd_id(scsicmd);
+	if (scsicmd->host_scribble)
+		cid = *(int *)scsicmd->host_scribble;
+	if (cid >= fibptr->dev->maximum_num_containers) {
+		_aac_probe_container2(context, fibptr);
+		return;
+	}
 	aac_fib_init(fibptr);
 
 	dinfo = (struct aac_query_mount *)fib_data(fibptr);
@@ -709,7 +720,7 @@ static void _aac_probe_container1(void * context, struct fib * fibptr)
 	else
 		dinfo->command = cpu_to_le32(VM_NameServe64);
 
-	dinfo->count = cpu_to_le32(scmd_id(scsicmd));
+	dinfo->count = cpu_to_le32(cid);
 	dinfo->type = cpu_to_le32(FT_FILESYS);
 	scsicmd->SCp.phase = AAC_OWNER_FIRMWARE;
 
@@ -732,10 +743,20 @@ static void _aac_probe_container1(void * context, struct fib * fibptr)
 
 static int _aac_probe_container(struct scsi_cmnd * scsicmd, int (*callback)(struct scsi_cmnd *))
 {
+	struct aac_dev * dev;
 	struct fib * fibptr;
 	int status = -ENOMEM;
+	int cid = scmd_id(scsicmd);
 
-	if ((fibptr = aac_fib_alloc((struct aac_dev *)scsicmd->device->host->hostdata))) {
+	dev = (struct aac_dev *)scsicmd->device->host->hostdata;
+	if (scsicmd->host_scribble) {
+		cid = *(int *)scsicmd->host_scribble;
+		if (cid > dev->maximum_num_containers) {
+			status = -ENODEV;
+			goto out_status;
+		}
+	}
+	if ((fibptr = aac_fib_alloc(dev))) {
 		struct aac_query_mount *dinfo;
 
 		aac_fib_init(fibptr);
@@ -748,7 +769,7 @@ static int _aac_probe_container(struct scsi_cmnd * scsicmd, int (*callback)(stru
 		else
 			dinfo->command = cpu_to_le32(VM_NameServe);
 
-		dinfo->count = cpu_to_le32(scmd_id(scsicmd));
+		dinfo->count = cpu_to_le32(cid);
 		dinfo->type = cpu_to_le32(FT_FILESYS);
 		scsicmd->SCp.ptr = (char *)callback;
 		scsicmd->SCp.phase = AAC_OWNER_FIRMWARE;
@@ -772,10 +793,11 @@ static int _aac_probe_container(struct scsi_cmnd * scsicmd, int (*callback)(stru
 			aac_fib_free(fibptr);
 		}
 	}
+out_status:
 	if (status < 0) {
-		struct fsa_dev_info *fsa_dev_ptr = ((struct aac_dev *)(scsicmd->device->host->hostdata))->fsa_dev;
-		if (fsa_dev_ptr) {
-			fsa_dev_ptr += scmd_id(scsicmd);
+		struct fsa_dev_info *fsa_dev_ptr = dev->fsa_dev;
+		if (fsa_dev_ptr && cid < dev->maximum_num_containers) {
+			fsa_dev_ptr += cid;
 			if ((fsa_dev_ptr->valid & 1) == 0) {
 				fsa_dev_ptr->valid = 0;
 				return (*callback)(scsicmd);
@@ -794,7 +816,7 @@ static int _aac_probe_container(struct scsi_cmnd * scsicmd, int (*callback)(stru
  */
 static int aac_probe_container_callback1(struct scsi_cmnd * scsicmd)
 {
-	scsicmd->device = NULL;
+	scsicmd->host_scribble = NULL;
 	return 0;
 }
 
@@ -815,6 +837,7 @@ int aac_probe_container(struct aac_dev *dev, int cid)
 		return -ENOMEM;
 	}
 	scsicmd->scsi_done = aac_probe_container_scsi_done;
+	scsicmd->host_scribble = (unsigned char *)&cid;
 
 	scsicmd->device = scsidev;
 	scsidev->sdev_state = 0;
@@ -822,7 +845,7 @@ int aac_probe_container(struct aac_dev *dev, int cid)
 	scsidev->host = dev->scsi_host_ptr;
 
 	if (_aac_probe_container(scsicmd, aac_probe_container_callback1) == 0)
-		while (scsicmd->device == scsidev)
+		while (scsicmd->host_scribble == (unsigned char *)&cid)
 			schedule();
 	kfree(scsidev);
 	status = scsicmd->SCp.Status;
Signature · 1 line
2.29.2
#51 lore raw

Re: [PATCH 16/18] aacraid: store target id in host_scribble

Quoted 5 lines · Hannes Reinecke

The probe_container mechanism requires a target id to be present, even if the device itself isn't present (yet).
As we're now allocating internal commands the target id of the device is immutable, so store the requested target id in the host_scribble field.

A more elegant solution is probably to introduce private data per SCSI command and to set the .cmd_size member in the SCSI host template. I'd like to get rid of the host_scribble field because it makes the SCSI command data structure larger than necessary for SCSI LLDs that don't use 'host_scribble'.

Thanks,

Bart.

#52 lore raw

Re: [PATCH 16/18] aacraid: store target id in host_scribble

Quoted 7 lines · Bart Van Assche
Quoted 5 lines · Hannes Reinecke

The probe_container mechanism requires a target id to be present, even if the device itself isn't present (yet).
As we're now allocating internal commands the target id of the device is immutable, so store the requested target id in the host_scribble field.

Quoted 6 lines

A more elegant solution is probably to introduce private data per SCSI command and to set the .cmd_size member in the SCSI host template. I'd like to get rid of the host_scribble field because it makes the SCSI command data structure larger than necessary for SCSI LLDs that don't use 'host_scribble'.

Ah. Good idea, both with using the .cmd_size and removing the host_scribble field.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
#53 lore raw

PATCH v8 17/18 aacraid: use scsi_get_internal_cmd()

In reply to Hannes Reinecke

From: Hannes Reinecke <[email protected]>

Use scsi_get_internal_cmd() to allocate internal commands. Internal commands are marked with a new FIB_CONTEXT_FLAG_INTERNAL_CMD, so we need to ensure that the 'flags' field is not cleared completely in aac_fib_send() to avoid command allocation failures.

---
 drivers/scsi/aacraid/aachba.c   | 92 +++++++++++++++------------------
 drivers/scsi/aacraid/aacraid.h  | 10 ++--
 drivers/scsi/aacraid/commctrl.c | 25 +++++----
 drivers/scsi/aacraid/comminit.c |  2 +-
 drivers/scsi/aacraid/commsup.c  | 69 ++++++++++++-------------
 drivers/scsi/aacraid/dpcsup.c   |  2 +-
 drivers/scsi/aacraid/linit.c    | 13 +++--
 7 files changed, 108 insertions(+), 105 deletions(-)
drivers/scsi/aacraid/aachba.c +42 -50
index ef59303db9b9..ce60b7344be9 100644
@@ -360,7 +360,7 @@ int aac_get_config_status(struct aac_dev *dev, int commit_flag)
 	int status = 0;
 	struct fib * fibptr;
 
-	if (!(fibptr = aac_fib_alloc(dev)))
+	if (!(fibptr = aac_fib_alloc(dev, DMA_FROM_DEVICE)))
 		return -ENOMEM;
 
 	aac_fib_init(fibptr);
@@ -457,7 +457,7 @@ int aac_get_containers(struct aac_dev *dev)
 	struct aac_get_container_count_resp *dresp;
 	int maximum_num_containers = MAXIMUM_NUM_CONTAINERS;
 
-	if (!(fibptr = aac_fib_alloc(dev)))
+	if (!(fibptr = aac_fib_alloc(dev, DMA_FROM_DEVICE)))
 		return -ENOMEM;
 
 	aac_fib_init(fibptr);
@@ -741,61 +741,56 @@ static void _aac_probe_container1(void * context, struct fib * fibptr)
 	}
 }
 
-static int _aac_probe_container(struct scsi_cmnd * scsicmd, int (*callback)(struct scsi_cmnd *))
+static int _aac_probe_container(struct fib * fibptr, int (*callback)(struct scsi_cmnd *))
 {
+	struct scsi_cmnd *scsicmd = fibptr->scmd;
 	struct aac_dev * dev;
-	struct fib * fibptr;
 	int status = -ENOMEM;
+	struct aac_query_mount *dinfo;
 	int cid = scmd_id(scsicmd);
 
 	dev = (struct aac_dev *)scsicmd->device->host->hostdata;
 	if (scsicmd->host_scribble) {
-		cid = *(int *)scsicmd->host_scribble;
-		if (cid > dev->maximum_num_containers) {
+		cid = *(unsigned int *)scsicmd->host_scribble;
+		if (cid >= dev->maximum_num_containers) {
 			status = -ENODEV;
 			goto out_status;
 		}
 	}
-	if ((fibptr = aac_fib_alloc(dev))) {
-		struct aac_query_mount *dinfo;
 
-		aac_fib_init(fibptr);
+	aac_fib_init(fibptr);
 
-		dinfo = (struct aac_query_mount *)fib_data(fibptr);
+	dinfo = (struct aac_query_mount *)fib_data(fibptr);
 
-		if (fibptr->dev->supplement_adapter_info.supported_options2 &
-		    AAC_OPTION_VARIABLE_BLOCK_SIZE)
-			dinfo->command = cpu_to_le32(VM_NameServeAllBlk);
-		else
-			dinfo->command = cpu_to_le32(VM_NameServe);
+	if (fibptr->dev->supplement_adapter_info.supported_options2 &
+	    AAC_OPTION_VARIABLE_BLOCK_SIZE)
+		dinfo->command = cpu_to_le32(VM_NameServeAllBlk);
+	else
+		dinfo->command = cpu_to_le32(VM_NameServe);
 
-		dinfo->count = cpu_to_le32(cid);
-		dinfo->type = cpu_to_le32(FT_FILESYS);
-		scsicmd->SCp.ptr = (char *)callback;
-		scsicmd->SCp.phase = AAC_OWNER_FIRMWARE;
+	dinfo->count = cpu_to_le32(cid);
+	dinfo->type = cpu_to_le32(FT_FILESYS);
+	scsicmd->SCp.ptr = (char *)callback;
+	scsicmd->SCp.phase = AAC_OWNER_FIRMWARE;
 
-		status = aac_fib_send(ContainerCommand,
+	status = aac_fib_send(ContainerCommand,
 			  fibptr,
 			  sizeof(struct aac_query_mount),
 			  FsaNormal,
 			  0, 1,
 			  _aac_probe_container1,
 			  (void *) scsicmd);
-		/*
-		 *	Check that the command queued to the controller
-		 */
-		if (status == -EINPROGRESS)
-			return 0;
-
-		if (status < 0) {
-			scsicmd->SCp.ptr = NULL;
-			aac_fib_complete(fibptr);
-			aac_fib_free(fibptr);
-		}
-	}
+	/*
+	 *	Check that the command queued to the controller
+	 */
+	if (status == -EINPROGRESS)
+		return 0;
 out_status:
 	if (status < 0) {
 		struct fsa_dev_info *fsa_dev_ptr = dev->fsa_dev;
+
+		scsicmd->SCp.ptr = NULL;
+		aac_fib_complete(fibptr);
 		if (fsa_dev_ptr && cid < dev->maximum_num_containers) {
 			fsa_dev_ptr += cid;
 			if ((fsa_dev_ptr->valid & 1) == 0) {
@@ -827,29 +822,23 @@ static void aac_probe_container_scsi_done(struct scsi_cmnd *scsi_cmnd)
 
 int aac_probe_container(struct aac_dev *dev, int cid)
 {
-	struct scsi_cmnd *scsicmd = kmalloc(sizeof(*scsicmd), GFP_KERNEL);
-	struct scsi_device *scsidev = kmalloc(sizeof(*scsidev), GFP_KERNEL);
+	struct fib *fibptr;
+	struct scsi_cmnd *scsicmd;
 	int status;
 
-	if (!scsicmd || !scsidev) {
-		kfree(scsicmd);
-		kfree(scsidev);
+	fibptr = aac_fib_alloc(dev, DMA_FROM_DEVICE);
+	if (!fibptr)
 		return -ENOMEM;
-	}
+
+	scsicmd = fibptr->scmd;
 	scsicmd->scsi_done = aac_probe_container_scsi_done;
 	scsicmd->host_scribble = (unsigned char *)&cid;
 
-	scsicmd->device = scsidev;
-	scsidev->sdev_state = 0;
-	scsidev->id = cid;
-	scsidev->host = dev->scsi_host_ptr;
-
-	if (_aac_probe_container(scsicmd, aac_probe_container_callback1) == 0)
+	if (_aac_probe_container(fibptr, aac_probe_container_callback1) == 0)
 		while (scsicmd->host_scribble == (unsigned char *)&cid)
 			schedule();
-	kfree(scsidev);
 	status = scsicmd->SCp.Status;
-	kfree(scsicmd);
+	aac_fib_free(fibptr);
 	return status;
 }
 
@@ -1694,7 +1683,7 @@ static int aac_send_safw_bmic_cmd(struct aac_dev *dev,
 		return 0;
 
 	/* allocate FIB */
-	fibptr = aac_fib_alloc(dev);
+	fibptr = aac_fib_alloc(dev, DMA_BIDIRECTIONAL);
 	if (!fibptr)
 		return -ENOMEM;
 
@@ -2054,7 +2043,7 @@ int aac_get_adapter_info(struct aac_dev* dev)
 	struct aac_bus_info *command;
 	struct aac_bus_info_response *bus_info;
 
-	if (!(fibptr = aac_fib_alloc(dev)))
+	if (!(fibptr = aac_fib_alloc(dev, DMA_FROM_DEVICE)))
 		return -ENOMEM;
 
 	aac_fib_init(fibptr);
@@ -2101,7 +2090,7 @@ int aac_get_adapter_info(struct aac_dev* dev)
 		if (rcode >= 0)
 			memcpy(&dev->supplement_adapter_info, sinfo, sizeof(*sinfo));
 		if (rcode == -ERESTARTSYS) {
-			fibptr = aac_fib_alloc(dev);
+			fibptr = aac_fib_alloc(dev, DMA_FROM_DEVICE);
 			if (!fibptr)
 				return -ENOMEM;
 		}
@@ -2811,6 +2800,8 @@ int aac_scsi_cmd(struct scsi_cmnd * scsicmd)
 			if (((fsa_dev_ptr[cid].valid & 1) == 0) ||
 			  (fsa_dev_ptr[cid].sense_data.sense_key ==
 			   NOT_READY)) {
+				struct fib * fibptr;
+
 				switch (scsicmd->cmnd[0]) {
 				case SERVICE_ACTION_IN_16:
 					if (!(dev->raw_io_interface) ||
@@ -2823,7 +2814,8 @@ int aac_scsi_cmd(struct scsi_cmnd * scsicmd)
 				case TEST_UNIT_READY:
 					if (dev->in_reset)
 						return -1;
-					return _aac_probe_container(scsicmd,
+					fibptr = aac_fib_alloc_tag(dev, scsicmd);
+					return _aac_probe_container(fibptr,
 							aac_probe_container_callback2);
 				default:
 					break;
drivers/scsi/aacraid/aacraid.h +7 -3
index e3e4ecbea726..3cd7f33497d8 100644
@@ -1291,13 +1291,16 @@ struct fsa_dev_info {
 };
 
 struct fib {
-	void			*next;	/* this is used by the allocator */
 	s16			type;
 	s16			size;
 	/*
 	 *	The Adapter that this I/O is destined for.
 	 */
 	struct aac_dev		*dev;
+	/*
+	 * The associated scsi command
+	 */
+	struct scsi_cmnd 	*scmd;
 	/*
 	 *	This is the event the sendfib routine will wait on if the
 	 *	caller did not pass one and this is synch io.
@@ -1552,7 +1555,6 @@ struct aac_dev
 	 */
 	struct fib              *fibs;
 
-	struct fib		*free_fib;
 	spinlock_t		fib_lock;
 
 	struct mutex		ioctl_mutex;
@@ -1597,6 +1599,7 @@ struct aac_dev
 	size_t			comm_size;
 
 	struct Scsi_Host	*scsi_host_ptr;
+	struct scsi_device	*scsi_host_dev;
 	int			maximum_num_containers;
 	int			maximum_num_physicals;
 	int			maximum_num_channels;
@@ -1729,6 +1732,7 @@ struct aac_dev
 #define FIB_CONTEXT_FLAG_NATIVE_HBA_TMF	(0x00000020)
 #define FIB_CONTEXT_FLAG_SCSI_CMD	(0x00000040)
 #define FIB_CONTEXT_FLAG_EH_RESET	(0x00000080)
+#define FIB_CONTEXT_FLAG_INTERNAL_CMD	(0x00000100)
 
 /*
  *	Define the command values
@@ -2686,7 +2690,7 @@ void aac_free_irq(struct aac_dev *dev);
 int aac_setup_safw_adapter(struct aac_dev *dev);
 const char *aac_driverinfo(struct Scsi_Host *);
 void aac_fib_vector_assign(struct aac_dev *dev);
-struct fib *aac_fib_alloc(struct aac_dev *dev);
+struct fib *aac_fib_alloc(struct aac_dev *dev, int direction);
 struct fib *aac_fib_alloc_tag(struct aac_dev *dev, struct scsi_cmnd *scmd);
 int aac_fib_setup(struct aac_dev *dev);
 void aac_fib_map_free(struct aac_dev *dev);
drivers/scsi/aacraid/commctrl.c +15 -10
index e7cc927ed952..bc5b81696476 100644
@@ -55,7 +55,7 @@ static int ioctl_send_fib(struct aac_dev * dev, void __user *arg)
 	if (dev->in_reset) {
 		return -EBUSY;
 	}
-	fibptr = aac_fib_alloc(dev);
+	fibptr = aac_fib_alloc(dev, DMA_BIDIRECTIONAL);
 	if(fibptr == NULL) {
 		return -ENOMEM;
 	}
@@ -478,7 +478,7 @@ static int check_revision(struct aac_dev *dev, void __user *arg)
  */
 static int aac_send_raw_srb(struct aac_dev* dev, void __user * arg)
 {
-	struct fib* srbfib;
+	struct fib* srbfib = NULL;
 	int status;
 	struct aac_srb *srbcmd = NULL;
 	struct aac_hba_cmd_req *hbacmd = NULL;
@@ -509,12 +509,6 @@ static int aac_send_raw_srb(struct aac_dev* dev, void __user * arg)
 		dprintk((KERN_DEBUG"aacraid: No permission to send raw srb\n"));
 		return -EPERM;
 	}
-	/*
-	 *	Allocate and initialize a Fib then setup a SRB command
-	 */
-	if (!(srbfib = aac_fib_alloc(dev))) {
-		return -ENOMEM;
-	}
 
 	memset(sg_list, 0, sizeof(sg_list)); /* cleanup may take issue */
 	if(copy_from_user(&fibsize, &user_srb->count,sizeof(u32))){
@@ -561,6 +555,15 @@ static int aac_send_raw_srb(struct aac_dev* dev, void __user * arg)
 		rcode = -EINVAL;
 		goto cleanup;
 	}
+
+	/*
+	 *	Allocate and initialize a Fib
+	 */
+	if (!(srbfib = aac_fib_alloc(dev, data_dir))) {
+		rcode = -ENOMEM;
+		goto cleanup;
+	}
+
 	actual_fibsize = sizeof(struct aac_srb) - sizeof(struct sgentry) +
 		((user_srbcmd->sg.count & 0xff) * sizeof(struct sgentry));
 	actual_fibsize64 = actual_fibsize + (user_srbcmd->sg.count & 0xff) *
@@ -988,8 +991,10 @@ static int aac_send_raw_srb(struct aac_dev* dev, void __user * arg)
 	if (rcode != -ERESTARTSYS) {
 		for (i = 0; i <= sg_indx; i++)
 			kfree(sg_list[i]);
-		aac_fib_complete(srbfib);
-		aac_fib_free(srbfib);
+		if (srbfib) {
+			aac_fib_complete(srbfib);
+			aac_fib_free(srbfib);
+		}
 	}
 
 	return rcode;
drivers/scsi/aacraid/comminit.c +1 -1
index 355b16f0b145..0d5be76891b0 100644
@@ -327,7 +327,7 @@ int aac_send_shutdown(struct aac_dev * dev)
 
 	aac_wait_for_io_completion(dev);
 
-	fibctx = aac_fib_alloc(dev);
+	fibctx = aac_fib_alloc(dev, DMA_NONE);
 	if (!fibctx)
 		return -ENOMEM;
 	aac_fib_init(fibctx);
drivers/scsi/aacraid/commsup.c +32 -37

146 lines not on this page · raw message

drivers/scsi/aacraid/dpcsup.c +1 -1
index fbe334c59f37..9424792432ec 100644
@@ -315,7 +315,7 @@ unsigned int aac_intr_normal(struct aac_dev *dev, u32 index, int isAif,
 		struct fib *fibctx;
 		struct aac_aifcmd *cmd;
 
-		fibctx = aac_fib_alloc(dev);
+		fibctx = aac_fib_alloc(dev, DMA_FROM_DEVICE);
 		if (!fibctx)
 			return 1;
 		aac_fib_init(fibctx);
drivers/scsi/aacraid/linit.c +10 -3
index e5d89b309c3a..ab3762c01738 100644
@@ -714,7 +714,7 @@ static int aac_eh_abort(struct scsi_cmnd* cmd)
 			return ret;
 
 		/* start a HBA_TMF_ABORT_TASK TMF request */
-		fib = aac_fib_alloc(aac);
+		fib = aac_fib_alloc(aac, DMA_NONE);
 		if (!fib)
 			return ret;
 
@@ -925,7 +925,7 @@ static int aac_eh_dev_reset(struct scsi_cmnd *cmd)
 	pr_err("%s: Host device reset request. SCSI hang ?\n",
 	       AAC_DRIVERNAME);
 
-	fib = aac_fib_alloc(aac);
+	fib = aac_fib_alloc(aac, DMA_NONE);
 	if (!fib)
 		return ret;
 
@@ -988,7 +988,7 @@ static int aac_eh_target_reset(struct scsi_cmnd *cmd)
 	pr_err("%s: Host target reset request. SCSI hang ?\n",
 	       AAC_DRIVERNAME);
 
-	fib = aac_fib_alloc(aac);
+	fib = aac_fib_alloc(aac, DMA_NONE);
 	if (!fib)
 		return ret;
 
@@ -1641,6 +1641,7 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	shost->max_cmd_len = 16;
 	shost->max_id = MAXIMUM_NUM_CONTAINERS;
 	shost->max_lun = AAC_MAX_LUN;
+	shost->nr_reserved_cmds = AAC_NUM_MGT_FIB;
 	shost->sg_tablesize = HBA_MAX_SG_SEPARATE;
 
 	if (aac_cfg_major == AAC_CHARDEV_NEEDS_REINIT)
@@ -1715,6 +1716,10 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (error)
 		goto out_deinit;
 
+	aac->scsi_host_dev = scsi_get_host_dev(shost);
+	if (!aac->scsi_host_dev)
+		goto out_remove_host;
+
 	aac->maximum_num_channels = aac_drivers[index].channels;
 	error = aac_get_adapter_info(aac);
 	if (error < 0)
@@ -1790,6 +1795,8 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	kfree(aac->queues);
 	aac_adapter_ioremap(aac, 0);
 	kfree(aac->fsa_dev);
+ out_remove_host:
+	scsi_remove_host(shost);
  out_free_fibs:
 	kfree(aac->fibs);
  out_free_host:
Signature · 1 line
2.29.2
#54 lore raw

PATCH v8 18/18 aacraid: use scsi_host_busy_iter() to traverse outstanding commands

In reply to Hannes Reinecke

Instead of walking the array of potential commands and trying to figure out which one might be pending the driver should be using scsi_host_busy_iter() to traverse all outstanding commands.
And for command abort we can now lookup the fibs directly as we now have a 1:1 mapping between request tags and fibs.

---
 drivers/scsi/aacraid/commsup.c |  49 ++++++------
 drivers/scsi/aacraid/linit.c   | 131 ++++++++++++++-------------------
 2 files changed, 84 insertions(+), 96 deletions(-)
drivers/scsi/aacraid/commsup.c +27 -22
index 6e2e381b12d4..73a3a90d8114 100644
@@ -1468,6 +1468,32 @@ static void aac_schedule_bus_scan(struct aac_dev *aac)
 		aac_schedule_src_reinit_aif_worker(aac);
 }
 
+static bool aac_close_sync_fib_iter(struct scsi_cmnd *command, void *data,
+				    bool reserved)
+{
+	struct Scsi_Host *host = command->device->host;
+	struct aac_dev *aac = (struct aac_dev *)host->hostdata;
+	struct fib *fib = &aac->fibs[command->request->tag];
+	int *retval = data;
+	__le32 XferState = fib->hw_fib_va->header.XferState;
+	bool is_response_expected = false;
+
+	if (!(XferState & cpu_to_le32(NoResponseExpected | Async)) &&
+	    (XferState & cpu_to_le32(ResponseExpected)))
+		is_response_expected = true;
+
+	if (is_response_expected
+	    || fib->flags & FIB_CONTEXT_FLAG_WAIT) {
+		unsigned long flagv;
+		spin_lock_irqsave(&fib->event_lock, flagv);
+		complete(&fib->event_wait);
+		spin_unlock_irqrestore(&fib->event_lock, flagv);
+		schedule();
+		*retval = 0;
+	}
+	return true;
+}
+
 static int _aac_reset_adapter(struct aac_dev *aac, int forced, u8 reset_type)
 {
 	int index, quirks;
@@ -1476,7 +1502,6 @@ static int _aac_reset_adapter(struct aac_dev *aac, int forced, u8 reset_type)
 	int jafo = 0;
 	int bled;
 	u64 dmamask;
-	int num_of_fibs = 0;
 
 	/*
 	 * Assumptions:
@@ -1510,27 +1535,7 @@ static int _aac_reset_adapter(struct aac_dev *aac, int forced, u8 reset_type)
 	 *	Loop through the fibs, close the synchronous FIBS
 	 */
 	retval = 1;
-	num_of_fibs = aac->scsi_host_ptr->can_queue + AAC_NUM_MGT_FIB;
-	for (index = 0; index <  num_of_fibs; index++) {
-
-		struct fib *fib = &aac->fibs[index];
-		__le32 XferState = fib->hw_fib_va->header.XferState;
-		bool is_response_expected = false;
-
-		if (!(XferState & cpu_to_le32(NoResponseExpected | Async)) &&
-		   (XferState & cpu_to_le32(ResponseExpected)))
-			is_response_expected = true;
-
-		if (is_response_expected
-		  || fib->flags & FIB_CONTEXT_FLAG_WAIT) {
-			unsigned long flagv;
-			spin_lock_irqsave(&fib->event_lock, flagv);
-			complete(&fib->event_wait);
-			spin_unlock_irqrestore(&fib->event_lock, flagv);
-			schedule();
-			retval = 0;
-		}
-	}
+	scsi_host_busy_iter(host, aac_close_sync_fib_iter, &retval);
 	/* Give some extra time for ioctls to complete. */
 	if (retval == 0)
 		ssleep(2);
drivers/scsi/aacraid/linit.c +57 -74

172 lines not on this page · raw message

Signature · 1 line
2.29.2