cxlflash: Base support for IBM CXL Flash Adapter

SCSI device driver to support filesystem access on the IBM CXL Flash adapter.

Supported-by: Stephen Bates <stephen.bates@pmcs.com>
Reviewed-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <JBottomley@Odin.com>

6 files changed, 3056 insertions, 0 deletions

diff --git a/drivers/scsi/cxlflash/Kconfig b/drivers/scsi/cxlflash/Kconfig
new file mode 100644
index 000000000000..c7075084cfdb
--- /dev/null
+++ b/drivers/scsi/cxlflash/Kconfig
@@ -0,0 +1,11 @@
+#
+# IBM CXL-attached Flash Accelerator SCSI Driver
+#
+
+config CXLFLASH
+	tristate "Support for IBM CAPI Flash"
+	depends on PCI && SCSI && CXL
+	default m
+	help
+	  Allows CAPI Accelerated IO to Flash
+	  If unsure, say N.
diff --git a/drivers/scsi/cxlflash/Makefile b/drivers/scsi/cxlflash/Makefile
new file mode 100644
index 000000000000..dc95e203e3af
--- /dev/null
+++ b/drivers/scsi/cxlflash/Makefile
@@ -0,0 +1,2 @@
+obj-$(CONFIG_CXLFLASH) += cxlflash.o
+cxlflash-y += main.o
diff --git a/drivers/scsi/cxlflash/common.h b/drivers/scsi/cxlflash/common.h
new file mode 100644
index 000000000000..5f43608dc8a1
--- /dev/null
+++ b/drivers/scsi/cxlflash/common.h
@@ -0,0 +1,180 @@
+/*
+ * CXL Flash Device Driver
+ *
+ * Written by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>, IBM Corporation
+ *             Matthew R. Ochs <mrochs@linux.vnet.ibm.com>, IBM Corporation
+ *
+ * Copyright (C) 2015 IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#ifndef _CXLFLASH_COMMON_H
+#define _CXLFLASH_COMMON_H
+
+#include <linux/list.h>
+#include <linux/types.h>
+#include <scsi/scsi.h>
+#include <scsi/scsi_device.h>
+
+
+#define MAX_CONTEXT  CXLFLASH_MAX_CONTEXT       /* num contexts per afu */
+
+#define CXLFLASH_BLOCK_SIZE	4096	/* 4K blocks */
+#define CXLFLASH_MAX_XFER_SIZE	16777216	/* 16MB transfer */
+#define CXLFLASH_MAX_SECTORS	(CXLFLASH_MAX_XFER_SIZE/512)	/* SCSI wants
+								   max_sectors
+								   in units of
+								   512 byte
+								   sectors
+								*/
+
+#define NUM_RRQ_ENTRY    16     /* for master issued cmds */
+#define MAX_RHT_PER_CONTEXT (PAGE_SIZE / sizeof(struct sisl_rht_entry))
+
+/* AFU command retry limit */
+#define MC_RETRY_CNT         5	/* sufficient for SCSI check and
+				   certain AFU errors */
+
+/* Command management definitions */
+#define CXLFLASH_NUM_CMDS	(2 * CXLFLASH_MAX_CMDS)	/* Must be a pow2 for
+							   alignment and more
+							   efficient array
+							   index derivation
+							 */
+
+#define CXLFLASH_MAX_CMDS               16
+#define CXLFLASH_MAX_CMDS_PER_LUN       CXLFLASH_MAX_CMDS
+
+
+static inline void check_sizes(void)
+{
+	BUILD_BUG_ON_NOT_POWER_OF_2(CXLFLASH_NUM_CMDS);
+}
+
+/* AFU defines a fixed size of 4K for command buffers (borrow 4K page define) */
+#define CMD_BUFSIZE     SIZE_4K
+
+/* flags in IOA status area for host use */
+#define B_DONE       0x01
+#define B_ERROR      0x02	/* set with B_DONE */
+#define B_TIMEOUT    0x04	/* set with B_DONE & B_ERROR */
+
+enum cxlflash_lr_state {
+	LINK_RESET_INVALID,
+	LINK_RESET_REQUIRED,
+	LINK_RESET_COMPLETE
+};
+
+enum cxlflash_init_state {
+	INIT_STATE_NONE,
+	INIT_STATE_PCI,
+	INIT_STATE_AFU,
+	INIT_STATE_SCSI
+};
+
+/*
+ * Each context has its own set of resource handles that is visible
+ * only from that context.
+ */
+
+struct cxlflash_cfg {
+	struct afu *afu;
+	struct cxl_context *mcctx;
+
+	struct pci_dev *dev;
+	struct pci_device_id *dev_id;
+	struct Scsi_Host *host;
+
+	ulong cxlflash_regs_pci;
+
+	wait_queue_head_t eeh_waitq;
+
+	struct work_struct work_q;
+	enum cxlflash_init_state init_state;
+	enum cxlflash_lr_state lr_state;
+	int lr_port;
+
+	struct cxl_afu *cxl_afu;
+
+	struct pci_pool *cxlflash_cmd_pool;
+	struct pci_dev *parent_dev;
+
+	wait_queue_head_t tmf_waitq;
+	bool tmf_active;
+	u8 err_recovery_active:1;
+};
+
+struct afu_cmd {
+	struct sisl_ioarcb rcb;	/* IOARCB (cache line aligned) */
+	struct sisl_ioasa sa;	/* IOASA must follow IOARCB */
+	spinlock_t slock;
+	struct completion cevent;
+	char *buf;		/* per command buffer */
+	struct afu *parent;
+	int slot;
+	atomic_t free;
+
+	u8 cmd_tmf:1;
+
+	/* As per the SISLITE spec the IOARCB EA has to be 16-byte aligned.
+	 * However for performance reasons the IOARCB/IOASA should be
+	 * cache line aligned.
+	 */
+} __aligned(cache_line_size());
+
+struct afu {
+	/* Stuff requiring alignment go first. */
+
+	u64 rrq_entry[NUM_RRQ_ENTRY];	/* 128B RRQ */
+	/*
+	 * Command & data for AFU commands.
+	 */
+	struct afu_cmd cmd[CXLFLASH_NUM_CMDS];
+
+	/* Beware of alignment till here. Preferably introduce new
+	 * fields after this point
+	 */
+
+	/* AFU HW */
+	struct cxl_ioctl_start_work work;
+	struct cxlflash_afu_map *afu_map;	/* entire MMIO map */
+	struct sisl_host_map *host_map;		/* MC host map */
+	struct sisl_ctrl_map *ctrl_map;		/* MC control map */
+
+	ctx_hndl_t ctx_hndl;	/* master's context handle */
+	u64 *hrrq_start;
+	u64 *hrrq_end;
+	u64 *hrrq_curr;
+	bool toggle;
+	bool read_room;
+	atomic64_t room;
+	u64 hb;
+	u32 cmd_couts;		/* Number of command checkouts */
+	u32 internal_lun;	/* User-desired LUN mode for this AFU */
+
+	char version[8];
+	u64 interface_version;
+
+	struct cxlflash_cfg *parent; /* Pointer back to parent cxlflash_cfg */
+
+};
+
+static inline u64 lun_to_lunid(u64 lun)
+{
+	u64 lun_id;
+
+	int_to_scsilun(lun, (struct scsi_lun *)&lun_id);
+	return swab64(lun_id);
+}
+
+int cxlflash_send_cmd(struct afu *, struct afu_cmd *);
+void cxlflash_wait_resp(struct afu *, struct afu_cmd *);
+int cxlflash_afu_reset(struct cxlflash_cfg *);
+struct afu_cmd *cxlflash_cmd_checkout(struct afu *);
+void cxlflash_cmd_checkin(struct afu_cmd *);
+int cxlflash_afu_sync(struct afu *, ctx_hndl_t, res_hndl_t, u8);
+#endif /* ifndef _CXLFLASH_COMMON_H */
diff --git a/drivers/scsi/cxlflash/main.c b/drivers/scsi/cxlflash/main.c
new file mode 100644
index 000000000000..0720d2f13c2a
--- /dev/null
+++ b/drivers/scsi/cxlflash/main.c
@@ -0,0 +1,2294 @@
+/*
+ * CXL Flash Device Driver
+ *
+ * Written by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>, IBM Corporation
+ *             Matthew R. Ochs <mrochs@linux.vnet.ibm.com>, IBM Corporation
+ *
+ * Copyright (C) 2015 IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/delay.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+
+#include <asm/unaligned.h>
+
+#include <misc/cxl.h>
+
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_host.h>
+
+#include "main.h"
+#include "sislite.h"
+#include "common.h"
+
+MODULE_DESCRIPTION(CXLFLASH_ADAPTER_NAME);
+MODULE_AUTHOR("Manoj N. Kumar <manoj@linux.vnet.ibm.com>");
+MODULE_AUTHOR("Matthew R. Ochs <mrochs@linux.vnet.ibm.com>");
+MODULE_LICENSE("GPL");
+
+
+/**
+ * cxlflash_cmd_checkout() - checks out an AFU command
+ * @afu:	AFU to checkout from.
+ *
+ * Commands are checked out in a round-robin fashion. Note that since
+ * the command pool is larger than the hardware queue, the majority of
+ * times we will only loop once or twice before getting a command. The
+ * buffer and CDB within the command are initialized (zeroed) prior to
+ * returning.
+ *
+ * Return: The checked out command or NULL when command pool is empty.
+ */
+struct afu_cmd *cxlflash_cmd_checkout(struct afu *afu)
+{
+	int k, dec = CXLFLASH_NUM_CMDS;
+	struct afu_cmd *cmd;
+
+	while (dec--) {
+		k = (afu->cmd_couts++ & (CXLFLASH_NUM_CMDS - 1));
+
+		cmd = &afu->cmd[k];
+
+		if (!atomic_dec_if_positive(&cmd->free)) {
+			pr_debug("%s: returning found index=%d\n",
+				 __func__, cmd->slot);
+			memset(cmd->buf, 0, CMD_BUFSIZE);
+			memset(cmd->rcb.cdb, 0, sizeof(cmd->rcb.cdb));
+			return cmd;
+		}
+	}
+
+	return NULL;
+}
+
+/**
+ * cxlflash_cmd_checkin() - checks in an AFU command
+ * @cmd:	AFU command to checkin.
+ *
+ * Safe to pass commands that have already been checked in. Several
+ * internal tracking fields are reset as part of the checkin. Note
+ * that these are intentionally reset prior to toggling the free bit
+ * to avoid clobbering values in the event that the command is checked
+ * out right away.
+ */
+void cxlflash_cmd_checkin(struct afu_cmd *cmd)
+{
+	cmd->rcb.scp = NULL;
+	cmd->rcb.timeout = 0;
+	cmd->sa.ioasc = 0;
+	cmd->cmd_tmf = false;
+	cmd->sa.host_use[0] = 0; /* clears both completion and retry bytes */
+
+	if (unlikely(atomic_inc_return(&cmd->free) != 1)) {
+		pr_err("%s: Freeing cmd (%d) that is not in use!\n",
+		       __func__, cmd->slot);
+		return;
+	}
+
+	pr_debug("%s: released cmd %p index=%d\n", __func__, cmd, cmd->slot);
+}
+
+/**
+ * process_cmd_err() - command error handler
+ * @cmd:	AFU command that experienced the error.
+ * @scp:	SCSI command associated with the AFU command in error.
+ *
+ * Translates error bits from AFU command to SCSI command results.
+ */
+static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
+{
+	struct sisl_ioarcb *ioarcb;
+	struct sisl_ioasa *ioasa;
+
+	if (unlikely(!cmd))
+		return;
+
+	ioarcb = &(cmd->rcb);
+	ioasa = &(cmd->sa);
+
+	if (ioasa->rc.flags & SISL_RC_FLAGS_UNDERRUN) {
+		pr_debug("%s: cmd underrun cmd = %p scp = %p\n",
+			 __func__, cmd, scp);
+		scp->result = (DID_ERROR << 16);
+	}
+
+	if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
+		pr_debug("%s: cmd underrun cmd = %p scp = %p\n",
+			 __func__, cmd, scp);
+		scp->result = (DID_ERROR << 16);
+	}
+
+	pr_debug("%s: cmd failed afu_rc=%d scsi_rc=%d fc_rc=%d "
+		 "afu_extra=0x%X, scsi_entra=0x%X, fc_extra=0x%X\n",
+		 __func__, ioasa->rc.afu_rc, ioasa->rc.scsi_rc,
+		 ioasa->rc.fc_rc, ioasa->afu_extra, ioasa->scsi_extra,
+		 ioasa->fc_extra);
+
+	if (ioasa->rc.scsi_rc) {
+		/* We have a SCSI status */
+		if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
+			memcpy(scp->sense_buffer, ioasa->sense_data,
+			       SISL_SENSE_DATA_LEN);
+			scp->result = ioasa->rc.scsi_rc;
+		} else
+			scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
+	}
+
+	/*
+	 * We encountered an error. Set scp->result based on nature
+	 * of error.
+	 */
+	if (ioasa->rc.fc_rc) {
+		/* We have an FC status */
+		switch (ioasa->rc.fc_rc) {
+		case SISL_FC_RC_LINKDOWN:
+			scp->result = (DID_REQUEUE << 16);
+			break;
+		case SISL_FC_RC_RESID:
+			/* This indicates an FCP resid underrun */
+			if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
+				/* If the SISL_RC_FLAGS_OVERRUN flag was set,
+				 * then we will handle this error else where.
+				 * If not then we must handle it here.
+				 * This is probably an AFU bug. We will
+				 * attempt a retry to see if that resolves it.
+				 */
+				scp->result = (DID_ERROR << 16);
+			}
+			break;
+		case SISL_FC_RC_RESIDERR:
+			/* Resid mismatch between adapter and device */
+		case SISL_FC_RC_TGTABORT:
+		case SISL_FC_RC_ABORTOK:
+		case SISL_FC_RC_ABORTFAIL:
+		case SISL_FC_RC_NOLOGI:
+		case SISL_FC_RC_ABORTPEND:
+		case SISL_FC_RC_WRABORTPEND:
+		case SISL_FC_RC_NOEXP:
+		case SISL_FC_RC_INUSE:
+			scp->result = (DID_ERROR << 16);
+			break;
+		}
+	}
+
+	if (ioasa->rc.afu_rc) {
+		/* We have an AFU error */
+		switch (ioasa->rc.afu_rc) {
+		case SISL_AFU_RC_NO_CHANNELS:
+			scp->result = (DID_MEDIUM_ERROR << 16);
+			break;
+		case SISL_AFU_RC_DATA_DMA_ERR:
+			switch (ioasa->afu_extra) {
+			case SISL_AFU_DMA_ERR_PAGE_IN:
+				/* Retry */
+				scp->result = (DID_IMM_RETRY << 16);
+				break;
+			case SISL_AFU_DMA_ERR_INVALID_EA:
+			default:
+				scp->result = (DID_ERROR << 16);
+			}
+			break;
+		case SISL_AFU_RC_OUT_OF_DATA_BUFS:
+			/* Retry */
+			scp->result = (DID_ALLOC_FAILURE << 16);
+			break;
+		default:
+			scp->result = (DID_ERROR << 16);
+		}
+	}
+}
+
+/**
+ * cmd_complete() - command completion handler
+ * @cmd:	AFU command that has completed.
+ *
+ * Prepares and submits command that has either completed or timed out to
+ * the SCSI stack. Checks AFU command back into command pool for non-internal
+ * (rcb.scp populated) commands.
+ */
+static void cmd_complete(struct afu_cmd *cmd)
+{
+	struct scsi_cmnd *scp;
+	u32 resid;
+	ulong lock_flags;
+	struct afu *afu = cmd->parent;
+	struct cxlflash_cfg *cfg = afu->parent;
+	bool cmd_is_tmf;
+
+	spin_lock_irqsave(&cmd->slock, lock_flags);
+	cmd->sa.host_use_b[0] |= B_DONE;
+	spin_unlock_irqrestore(&cmd->slock, lock_flags);
+
+	if (cmd->rcb.scp) {
+		scp = cmd->rcb.scp;
+		if (unlikely(cmd->sa.rc.afu_rc ||
+			     cmd->sa.rc.scsi_rc ||
+			     cmd->sa.rc.fc_rc))
+			process_cmd_err(cmd, scp);
+		else
+			scp->result = (DID_OK << 16);
+
+		resid = cmd->sa.resid;
+		cmd_is_tmf = cmd->cmd_tmf;
+		cxlflash_cmd_checkin(cmd); /* Don't use cmd after here */
+
+		pr_debug("%s: calling scsi_set_resid, scp=%p "
+			 "result=%X resid=%d\n", __func__,
+			 scp, scp->result, resid);
+
+		scsi_set_resid(scp, resid);
+		scsi_dma_unmap(scp);
+		scp->scsi_done(scp);
+
+		if (cmd_is_tmf) {
+			spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
+			cfg->tmf_active = false;
+			wake_up_all_locked(&cfg->tmf_waitq);
+			spin_unlock_irqrestore(&cfg->tmf_waitq.lock,
+					       lock_flags);
+		}
+	} else
+		complete(&cmd->cevent);
+}
+
+/**
+ * send_tmf() - sends a Task Management Function (TMF)
+ * @afu:	AFU to checkout from.
+ * @scp:	SCSI command from stack.
+ * @tmfcmd:	TMF command to send.
+ *
+ * Return:
+ *	0 on success
+ *	SCSI_MLQUEUE_HOST_BUSY when host is busy
+ */
+static int send_tmf(struct afu *afu, struct scsi_cmnd *scp, u64 tmfcmd)
+{
+	struct afu_cmd *cmd;
+
+	u32 port_sel = scp->device->channel + 1;
+	short lflag = 0;
+	struct Scsi_Host *host = scp->device->host;
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
+	ulong lock_flags;
+	int rc = 0;
+
+	cmd = cxlflash_cmd_checkout(afu);
+	if (unlikely(!cmd)) {
+		pr_err("%s: could not get a free command\n", __func__);
+		rc = SCSI_MLQUEUE_HOST_BUSY;
+		goto out;
+	}
+
+	/* If a Task Management Function is active, do not send one more.
+	 */
+	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
+	if (cfg->tmf_active)
+		wait_event_interruptible_locked_irq(cfg->tmf_waitq,
+						    !cfg->tmf_active);
+	cfg->tmf_active = true;
+	cmd->cmd_tmf = true;
+	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
+
+	cmd->rcb.ctx_id = afu->ctx_hndl;
+	cmd->rcb.port_sel = port_sel;
+	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
+
+	lflag = SISL_REQ_FLAGS_TMF_CMD;
+
+	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
+			      SISL_REQ_FLAGS_SUP_UNDERRUN | lflag);
+
+	/* Stash the scp in the reserved field, for reuse during interrupt */
+	cmd->rcb.scp = scp;
+
+	/* Copy the CDB from the cmd passed in */
+	memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
+
+	/* Send the command */
+	rc = cxlflash_send_cmd(afu, cmd);
+	if (unlikely(rc)) {
+		cxlflash_cmd_checkin(cmd);
+		spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
+		cfg->tmf_active = false;
+		spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
+		goto out;
+	}
+
+	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
+	wait_event_interruptible_locked_irq(cfg->tmf_waitq, !cfg->tmf_active);
+	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
+out:
+	return rc;
+}
+
+/**
+ * cxlflash_driver_info() - information handler for this host driver
+ * @host:	SCSI host associated with device.
+ *
+ * Return: A string describing the device.
+ */
+static const char *cxlflash_driver_info(struct Scsi_Host *host)
+{
+	return CXLFLASH_ADAPTER_NAME;
+}
+
+/**
+ * cxlflash_queuecommand() - sends a mid-layer request
+ * @host:	SCSI host associated with device.
+ * @scp:	SCSI command to send.
+ *
+ * Return:
+ *	0 on success
+ *	SCSI_MLQUEUE_HOST_BUSY when host is busy
+ */
+static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
+{
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
+	struct afu *afu = cfg->afu;
+	struct pci_dev *pdev = cfg->dev;
+	struct afu_cmd *cmd;
+	u32 port_sel = scp->device->channel + 1;
+	int nseg, i, ncount;
+	struct scatterlist *sg;
+	ulong lock_flags;
+	short lflag = 0;
+	int rc = 0;
+
+	pr_debug("%s: (scp=%p) %d/%d/%d/%llu cdb=(%08X-%08X-%08X-%08X)\n",
+		 __func__, scp, host->host_no, scp->device->channel,
+		 scp->device->id, scp->device->lun,
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
+
+	/* If a Task Management Function is active, wait for it to complete
+	 * before continuing with regular commands.
+	 */
+	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
+	if (cfg->tmf_active) {
+		spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
+		rc = SCSI_MLQUEUE_HOST_BUSY;
+		goto out;
+	}
+	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
+
+	cmd = cxlflash_cmd_checkout(afu);
+	if (unlikely(!cmd)) {
+		pr_err("%s: could not get a free command\n", __func__);
+		rc = SCSI_MLQUEUE_HOST_BUSY;
+		goto out;
+	}
+
+	cmd->rcb.ctx_id = afu->ctx_hndl;
+	cmd->rcb.port_sel = port_sel;
+	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
+
+	if (scp->sc_data_direction == DMA_TO_DEVICE)
+		lflag = SISL_REQ_FLAGS_HOST_WRITE;
+	else
+		lflag = SISL_REQ_FLAGS_HOST_READ;
+
+	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
+			      SISL_REQ_FLAGS_SUP_UNDERRUN | lflag);
+
+	/* Stash the scp in the reserved field, for reuse during interrupt */
+	cmd->rcb.scp = scp;
+
+	nseg = scsi_dma_map(scp);
+	if (unlikely(nseg < 0)) {
+		dev_err(&pdev->dev, "%s: Fail DMA map! nseg=%d\n",
+			__func__, nseg);
+		rc = SCSI_MLQUEUE_HOST_BUSY;
+		goto out;
+	}
+
+	ncount = scsi_sg_count(scp);
+	scsi_for_each_sg(scp, sg, ncount, i) {
+		cmd->rcb.data_len = sg_dma_len(sg);
+		cmd->rcb.data_ea = sg_dma_address(sg);
+	}
+
+	/* Copy the CDB from the scsi_cmnd passed in */
+	memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
+
+	/* Send the command */
+	rc = cxlflash_send_cmd(afu, cmd);
+	if (unlikely(rc)) {
+		cxlflash_cmd_checkin(cmd);
+		scsi_dma_unmap(scp);
+	}
+
+out:
+	return rc;
+}
+
+/**
+ * cxlflash_eh_device_reset_handler() - reset a single LUN
+ * @scp:	SCSI command to send.
+ *
+ * Return:
+ *	SUCCESS as defined in scsi/scsi.h
+ *	FAILED as defined in scsi/scsi.h
+ */
+static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
+{
+	int rc = SUCCESS;
+	struct Scsi_Host *host = scp->device->host;
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
+	struct afu *afu = cfg->afu;
+	int rcr = 0;
+
+	pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
+		 "cdb=(%08X-%08X-%08X-%08X)\n", __func__, scp,
+		 host->host_no, scp->device->channel,
+		 scp->device->id, scp->device->lun,
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
+
+	rcr = send_tmf(afu, scp, TMF_LUN_RESET);
+	if (unlikely(rcr))
+		rc = FAILED;
+
+	pr_debug("%s: returning rc=%d\n", __func__, rc);
+	return rc;
+}
+
+/**
+ * cxlflash_eh_host_reset_handler() - reset the host adapter
+ * @scp:	SCSI command from stack identifying host.
+ *
+ * Return:
+ *	SUCCESS as defined in scsi/scsi.h
+ *	FAILED as defined in scsi/scsi.h
+ */
+static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
+{
+	int rc = SUCCESS;
+	int rcr = 0;
+	struct Scsi_Host *host = scp->device->host;
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
+
+	pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
+		 "cdb=(%08X-%08X-%08X-%08X)\n", __func__, scp,
+		 host->host_no, scp->device->channel,
+		 scp->device->id, scp->device->lun,
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
+		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
+
+	rcr = cxlflash_afu_reset(cfg);
+	if (rcr == 0)
+		rc = SUCCESS;
+	else
+		rc = FAILED;
+
+	pr_debug("%s: returning rc=%d\n", __func__, rc);
+	return rc;
+}
+
+/**
+ * cxlflash_change_queue_depth() - change the queue depth for the device
+ * @sdev:	SCSI device destined for queue depth change.
+ * @qdepth:	Requested queue depth value to set.
+ *
+ * The requested queue depth is capped to the maximum supported value.
+ *
+ * Return: The actual queue depth set.
+ */
+static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
+{
+
+	if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
+		qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
+
+	scsi_change_queue_depth(sdev, qdepth);
+	return sdev->queue_depth;
+}
+
+/**
+ * cxlflash_show_port_status() - queries and presents the current port status
+ * @dev:	Generic device associated with the host owning the port.
+ * @attr:	Device attribute representing the port.
+ * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
+ *
+ * Return: The size of the ASCII string returned in @buf.
+ */
+static ssize_t cxlflash_show_port_status(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(dev);
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
+	struct afu *afu = cfg->afu;
+
+	char *disp_status;
+	int rc;
+	u32 port;
+	u64 status;
+	u64 *fc_regs;
+
+	rc = kstrtouint((attr->attr.name + 4), 10, &port);
+	if (rc || (port > NUM_FC_PORTS))
+		return 0;
+
+	fc_regs = &afu->afu_map->global.fc_regs[port][0];
+	status =
+	    (readq_be(&fc_regs[FC_MTIP_STATUS / 8]) & FC_MTIP_STATUS_MASK);
+
+	if (status == FC_MTIP_STATUS_ONLINE)
+		disp_status = "online";
+	else if (status == FC_MTIP_STATUS_OFFLINE)
+		disp_status = "offline";
+	else
+		disp_status = "unknown";
+
+	return snprintf(buf, PAGE_SIZE, "%s\n", disp_status);
+}
+
+/**
+ * cxlflash_show_lun_mode() - presents the current LUN mode of the host
+ * @dev:	Generic device associated with the host.
+ * @attr:	Device attribute representing the lun mode.
+ * @buf:	Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
+ *
+ * Return: The size of the ASCII string returned in @buf.
+ */
+static ssize_t cxlflash_show_lun_mode(struct device *dev,
+				      struct device_attribute *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(dev);
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
+	struct afu *afu = cfg->afu;
+
+	return snprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
+}
+
+/**
+ * cxlflash_store_lun_mode() - sets the LUN mode of the host
+ * @dev:	Generic device associated with the host.
+ * @attr:	Device attribute representing the lun mode.
+ * @buf:	Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
+ * @count:	Length of data resizing in @buf.
+ *
+ * The CXL Flash AFU supports a dummy LUN mode where the external
+ * links and storage are not required. Space on the FPGA is used
+ * to create 1 or 2 small LUNs which are presented to the system
+ * as if they were a normal storage device. This feature is useful
+ * during development and also provides manufacturing with a way
+ * to test the AFU without an actual device.
+ *
+ * 0 = external LUN[s] (default)
+ * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
+ * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
+ * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
+ * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
+ *
+ * Return: The size of the ASCII string returned in @buf.
+ */
+static ssize_t cxlflash_store_lun_mode(struct device *dev,
+				       struct device_attribute *attr,
+				       const char *buf, size_t count)
+{
+	struct Scsi_Host *shost = class_to_shost(dev);
+	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
+	struct afu *afu = cfg->afu;
+	int rc;
+	u32 lun_mode;
+
+	rc = kstrtouint(buf, 10, &lun_mode);
+	if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
+		afu->internal_lun = lun_mode;
+		cxlflash_afu_reset(cfg);
+		scsi_scan_host(cfg->host);
+	}
+
+	return count;
+}
+
+/**
+ * cxlflash_show_dev_mode() - presents the current mode of the device
+ * @dev:	Generic device associated with the device.
+ * @attr:	Device attribute representing the device mode.
+ * @buf:	Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
+ *
+ * Return: The size of the ASCII string returned in @buf.
+ */
+static ssize_t cxlflash_show_dev_mode(struct device *dev,
+				      struct device_attribute *attr, char *buf)
+{
+	struct scsi_device *sdev = to_scsi_device(dev);
+
+	return snprintf(buf, PAGE_SIZE, "%s\n",
+			sdev->hostdata ? "superpipe" : "legacy");
+}
+
+/**
+ * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
+ * @cxlflash:	Internal structure associated with the host.
+ */
+static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
+{
+	struct pci_dev *pdev = cfg->dev;
+
+	if (pci_channel_offline(pdev))
+		wait_event_timeout(cfg->eeh_waitq,
+				   !pci_channel_offline(pdev),
+				   CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
+}
+
+/*
+ * Host attributes
+ */
+static DEVICE_ATTR(port0, S_IRUGO, cxlflash_show_port_status, NULL);
+static DEVICE_ATTR(port1, S_IRUGO, cxlflash_show_port_status, NULL);
+static DEVICE_ATTR(lun_mode, S_IRUGO | S_IWUSR, cxlflash_show_lun_mode,
+		   cxlflash_store_lun_mode);
+
+static struct device_attribute *cxlflash_host_attrs[] = {
+	&dev_attr_port0,
+	&dev_attr_port1,
+	&dev_attr_lun_mode,
+	NULL
+};
+
+/*
+ * Device attributes
+ */
+static DEVICE_ATTR(mode, S_IRUGO, cxlflash_show_dev_mode, NULL);
+
+static struct device_attribute *cxlflash_dev_attrs[] = {
+	&dev_attr_mode,
+	NULL
+};
+
+/*
+ * Host template
+ */
+static struct scsi_host_template driver_template = {
+	.module = THIS_MODULE,
+	.name = CXLFLASH_ADAPTER_NAME,
+	.info = cxlflash_driver_info,
+	.proc_name = CXLFLASH_NAME,
+	.queuecommand = cxlflash_queuecommand,
+	.eh_device_reset_handler = cxlflash_eh_device_reset_handler,
+	.eh_host_reset_handler = cxlflash_eh_host_reset_handler,
+	.change_queue_depth = cxlflash_change_queue_depth,
+	.cmd_per_lun = 16,
+	.can_queue = CXLFLASH_MAX_CMDS,
+	.this_id = -1,
+	.sg_tablesize = SG_NONE,	/* No scatter gather support. */
+	.max_sectors = CXLFLASH_MAX_SECTORS,
+	.use_clustering = ENABLE_CLUSTERING,
+	.shost_attrs = cxlflash_host_attrs,
+	.sdev_attrs = cxlflash_dev_attrs,
+};
+
+/*
+ * Device dependent values
+ */
+static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS };
+
+/*
+ * PCI device binding table
+ */
+static struct pci_device_id cxlflash_pci_table[] = {
+	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
+	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
+	{}
+};
+
+MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
+
+/**
+ * free_mem() - free memory associated with the AFU
+ * @cxlflash:	Internal structure associated with the host.
+ */
+static void free_mem(struct cxlflash_cfg *cfg)
+{
+	int i;
+	char *buf = NULL;
+	struct afu *afu = cfg->afu;
+
+	if (cfg->afu) {
+		for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
+			buf = afu->cmd[i].buf;
+			if (!((u64)buf & (PAGE_SIZE - 1)))
+				free_page((ulong)buf);
+		}
+
+		free_pages((ulong)afu, get_order(sizeof(struct afu)));
+		cfg->afu = NULL;
+	}
+}
+
+/**
+ * stop_afu() - stops the AFU command timers and unmaps the MMIO space
+ * @cxlflash:	Internal structure associated with the host.
+ *
+ * Safe to call with AFU in a partially allocated/initialized state.
+ */
+static void stop_afu(struct cxlflash_cfg *cfg)
+{
+	int i;
+	struct afu *afu = cfg->afu;
+
+	if (likely(afu)) {
+		for (i = 0; i < CXLFLASH_NUM_CMDS; i++)
+			complete(&afu->cmd[i].cevent);
+
+		if (likely(afu->afu_map)) {
+			cxl_psa_unmap((void *)afu->afu_map);
+			afu->afu_map = NULL;
+		}
+	}
+}
+
+/**
+ * term_mc() - terminates the master context
+ * @cxlflash:	Internal structure associated with the host.
+ * @level:	Depth of allocation, where to begin waterfall tear down.
+ *
+ * Safe to call with AFU/MC in partially allocated/initialized state.
+ */
+static void term_mc(struct cxlflash_cfg *cfg, enum undo_level level)
+{
+	int rc = 0;
+	struct afu *afu = cfg->afu;
+
+	if (!afu || !cfg->mcctx) {
+		pr_err("%s: returning from term_mc with NULL afu or MC\n",
+		       __func__);
+		return;
+	}
+
+	switch (level) {
+	case UNDO_START:
+		rc = cxl_stop_context(cfg->mcctx);
+		BUG_ON(rc);
+	case UNMAP_THREE:
+		cxl_unmap_afu_irq(cfg->mcctx, 3, afu);
+	case UNMAP_TWO:
+		cxl_unmap_afu_irq(cfg->mcctx, 2, afu);
+	case UNMAP_ONE:
+		cxl_unmap_afu_irq(cfg->mcctx, 1, afu);
+	case FREE_IRQ:
+		cxl_free_afu_irqs(cfg->mcctx);
+	case RELEASE_CONTEXT:
+		cfg->mcctx = NULL;
+	}
+}
+
+/**
+ * term_afu() - terminates the AFU
+ * @cxlflash:	Internal structure associated with the host.
+ *
+ * Safe to call with AFU/MC in partially allocated/initialized state.
+ */
+static void term_afu(struct cxlflash_cfg *cfg)
+{
+	term_mc(cfg, UNDO_START);
+
+	if (cfg->afu)
+		stop_afu(cfg);
+
+	pr_debug("%s: returning\n", __func__);
+}
+
+/**
+ * cxlflash_remove() - PCI entry point to tear down host
+ * @pdev:	PCI device associated with the host.
+ *
+ * Safe to use as a cleanup in partially allocated/initialized state.
+ */
+static void cxlflash_remove(struct pci_dev *pdev)
+{
+	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
+	ulong lock_flags;
+
+	/* If a Task Management Function is active, wait for it to complete
+	 * before continuing with remove.
+	 */
+	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
+	if (cfg->tmf_active)
+		wait_event_interruptible_locked_irq(cfg->tmf_waitq,
+						    !cfg->tmf_active);
+	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
+
+	switch (cfg->init_state) {
+	case INIT_STATE_SCSI:
+		scsi_remove_host(cfg->host);
+		scsi_host_put(cfg->host);
+		/* Fall through */
+	case INIT_STATE_AFU:
+		term_afu(cfg);
+	case INIT_STATE_PCI:
+		pci_release_regions(cfg->dev);
+		pci_disable_device(pdev);
+	case INIT_STATE_NONE:
+		flush_work(&cfg->work_q);
+		free_mem(cfg);
+		break;
+	}
+
+	pr_debug("%s: returning\n", __func__);
+}
+
+/**
+ * alloc_mem() - allocates the AFU and its command pool
+ * @cxlflash:	Internal structure associated with the host.
+ *
+ * A partially allocated state remains on failure.
+ *
+ * Return:
+ *	0 on success
+ *	-ENOMEM on failure to allocate memory
+ */
+static int alloc_mem(struct cxlflash_cfg *cfg)
+{
+	int rc = 0;
+	int i;
+	char *buf = NULL;
+
+	/* This allocation is about 12K, i.e. only 1 64k page
+	 * and upto 4 4k pages
+	 */
+	cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
+					    get_order(sizeof(struct afu)));
+	if (unlikely(!cfg->afu)) {
+		pr_err("%s: cannot get %d free pages\n",
+		       __func__, get_order(sizeof(struct afu)));
+		rc = -ENOMEM;
+		goto out;
+	}
+	cfg->afu->parent = cfg;
+	cfg->afu->afu_map = NULL;
+
+	for (i = 0; i < CXLFLASH_NUM_CMDS; buf += CMD_BUFSIZE, i++) {
+		if (!((u64)buf & (PAGE_SIZE - 1))) {
+			buf = (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
+			if (unlikely(!buf)) {
+				pr_err("%s: Allocate command buffers fail!\n",
+				       __func__);
+				rc = -ENOMEM;
+				free_mem(cfg);
+				goto out;
+			}
+		}
+
+		cfg->afu->cm