New NVMe front end (nda).

8 files changed, 1973 insertions, 0 deletions

diff --git a/sys/cam/cam_ccb.h b/sys/cam/cam_ccb.h
index 4b9b8e6..6286c43 100644
--- a/sys/cam/cam_ccb.h
+++ b/sys/cam/cam_ccb.h
@@ -41,6 +41,7 @@
 #include <cam/cam_debug.h>
 #include <cam/scsi/scsi_all.h>
 #include <cam/ata/ata_all.h>
+#include <cam/nvme/nvme_all.h>
 
 /* General allocation length definitions for CCB structures */
 #define	IOCDBLEN	CAM_MAX_CDBLEN	/* Space for CDB bytes/pointer */
@@ -265,6 +266,7 @@ typedef enum {
 	PROTO_ATAPI,	/* AT Attachment Packetized Interface */
 	PROTO_SATAPM,	/* SATA Port Multiplier */
 	PROTO_SEMB,	/* SATA Enclosure Management Bridge */
+	PROTO_NVME,	/* NVME */
 } cam_proto;
 
 typedef enum {
@@ -280,6 +282,7 @@ typedef enum {
 	XPORT_SATA,	/* Serial AT Attachment */
 	XPORT_ISCSI,	/* iSCSI */
 	XPORT_SRP,	/* SCSI RDMA Protocol */
+	XPORT_NVME,	/* NVMe over PCIe */
 } cam_xport;
 
 #define XPORT_IS_ATA(t)		((t) == XPORT_ATA || (t) == XPORT_SATA)
@@ -783,6 +786,19 @@ struct ccb_relsim {
 };
 
 /*
+ * NVMe I/O Request CCB used for the XPT_NVME_IO function code.
+ */
+struct ccb_nvmeio {
+	struct	   ccb_hdr ccb_h;
+	union	   ccb *next_ccb;	/* Ptr for next CCB for action */
+	struct nvme_command cmd;	/* NVME command, per NVME standard */
+	struct nvme_completion cpl;	/* NVME completion, per NVME standard */
+	uint8_t   *data_ptr;		/* Ptr to the data buf/SG list */
+	uint32_t  dxfer_len;		/* Data transfer length */
+	uint32_t  resid;		/* Transfer residual length: 2's comp unused ?*/
+};
+
+/*
  * Definitions for the asynchronous callback CCB fields.
  */
 typedef enum {
@@ -1234,6 +1250,7 @@ union ccb {
 	struct	ccb_ataio		ataio;
 	struct	ccb_dev_advinfo		cdai;
 	struct	ccb_async		casync;
+	struct	ccb_nvmeio		nvmeio;
 };
 
 #define CCB_CLEAR_ALL_EXCEPT_HDR(ccbp)			\
@@ -1250,6 +1267,12 @@ cam_fill_csio(struct ccb_scsiio *csio, u_int32_t retries,
 	      u_int32_t timeout);
 
 static __inline void
+cam_fill_nvmeio(struct ccb_nvmeio *nvmeio, u_int32_t retries,
+	      void (*cbfcnp)(struct cam_periph *, union ccb *),
+	      u_int32_t flags, u_int8_t *data_ptr, u_int32_t dxfer_len,
+	      u_int32_t timeout);
+
+static __inline void
 cam_fill_ctio(struct ccb_scsiio *csio, u_int32_t retries,
 	      void (*cbfcnp)(struct cam_periph *, union ccb *),
 	      u_int32_t flags, u_int tag_action, u_int tag_id,
@@ -1370,6 +1393,20 @@ cam_ccb_status(union ccb *ccb)
 
 void cam_calc_geometry(struct ccb_calc_geometry *ccg, int extended);
 
+static __inline void
+cam_fill_nvmeio(struct ccb_nvmeio *nvmeio, u_int32_t retries,
+	      void (*cbfcnp)(struct cam_periph *, union ccb *),
+	      u_int32_t flags, u_int8_t *data_ptr, u_int32_t dxfer_len,
+	      u_int32_t timeout)
+{
+	nvmeio->ccb_h.func_code = XPT_NVME_IO;
+	nvmeio->ccb_h.flags = flags;
+	nvmeio->ccb_h.retry_count = retries;
+	nvmeio->ccb_h.cbfcnp = cbfcnp;
+	nvmeio->ccb_h.timeout = timeout;
+	nvmeio->data_ptr = data_ptr;
+	nvmeio->dxfer_len = dxfer_len;
+}
 __END_DECLS
 
 #endif /* _CAM_CAM_CCB_H */
diff --git a/sys/cam/cam_xpt_internal.h b/sys/cam/cam_xpt_internal.h
index 23d6d34..b0624af 100644
--- a/sys/cam/cam_xpt_internal.h
+++ b/sys/cam/cam_xpt_internal.h
@@ -117,6 +117,8 @@ struct cam_ed {
 	STAILQ_ENTRY(cam_ed) highpowerq_entry;
 	struct mtx	 device_mtx;
 	struct task	 device_destroy_task;
+	const struct	 nvme_controller_data *nvme_cdata;
+	const struct	 nvme_namespace_data *nvme_data;
 };
 
 /*
@@ -167,6 +169,7 @@ struct cam_path {
 
 struct xpt_xport *	scsi_get_xport(void);
 struct xpt_xport *	ata_get_xport(void);
+struct xpt_xport *	nvme_get_xport(void);
 
 struct cam_ed *		xpt_alloc_device(struct cam_eb *bus,
 					 struct cam_et *target,
diff --git a/sys/cam/nvme/nvme_all.c b/sys/cam/nvme/nvme_all.c
new file mode 100644
index 0000000..3904891
--- /dev/null
+++ b/sys/cam/nvme/nvme_all.c
@@ -0,0 +1,124 @@
+/*-
+ * Copyright (c) 2015 Netflix, Inc
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer,
+ *    without modification, immediately at the beginning of the file.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/param.h>
+
+#ifdef _KERNEL
+#include "opt_scsi.h"
+
+#include <sys/systm.h>
+#include <sys/libkern.h>
+#include <sys/kernel.h>
+#include <sys/sysctl.h>
+#else
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#ifndef min
+#define min(a,b) (((a)<(b))?(a):(b))
+#endif
+#endif
+
+#include <cam/cam.h>
+#include <cam/cam_ccb.h>
+#include <cam/cam_queue.h>
+#include <cam/cam_xpt.h>
+#include <cam/nvme/nvme_all.h>
+#include <sys/sbuf.h>
+#include <sys/endian.h>
+
+void
+nvme_ns_cmd(struct ccb_nvmeio *nvmeio, uint8_t cmd, uint32_t nsid,
+    uint32_t cdw10, uint32_t cdw11, uint32_t cdw12, uint32_t cdw13,
+    uint32_t cdw14, uint32_t cdw15)
+{
+	bzero(&nvmeio->cmd, sizeof(struct nvme_command));
+	nvmeio->cmd.opc = cmd;
+	nvmeio->cmd.nsid = nsid;
+	nvmeio->cmd.cdw10 = cdw10;
+	nvmeio->cmd.cdw11 = cdw11;
+	nvmeio->cmd.cdw12 = cdw12;
+	nvmeio->cmd.cdw13 = cdw13;
+	nvmeio->cmd.cdw14 = cdw14;
+	nvmeio->cmd.cdw15 = cdw15;
+}
+
+int
+nvme_identify_match(caddr_t identbuffer, caddr_t table_entry)
+{
+	return 0;
+}
+
+
+void
+nvme_print_ident(const struct nvme_controller_data *cdata,
+    const struct nvme_namespace_data *data)
+{
+	printf("I'm a pretty NVME drive\n");
+}
+
+/* XXX need to do nvme admin opcodes too, but those aren't used yet by nda */
+static const char *
+nvme_opc2str[] = {
+	"FLUSH",
+	"WRITE",
+	"READ",
+	"RSVD-3",
+	"WRITE_UNCORRECTABLE",
+	"COMPARE",
+	"RSVD-6",
+	"RSVD-7",
+	"DATASET_MANAGEMENT"
+};
+
+const char *
+nvme_op_string(const struct nvme_command *cmd)
+{
+	if (cmd->opc > nitems(nvme_opc2str))
+		return "UNKNOWN";
+
+	return nvme_opc2str[cmd->opc];
+}
+
+const char *
+nvme_cmd_string(const struct nvme_command *cmd, char *cmd_string, size_t len)
+{
+	/*
+	 * cid, rsvd areas and mptr not printed, since they are used
+	 * only internally by the SIM.
+	 */
+	snprintf(cmd_string, len,
+	    "opc=%x fuse=%x nsid=%x prp1=%llx prp2=%llx cdw=%x %x %x %x %x %x",
+	    cmd->opc, cmd->fuse, cmd->nsid,
+	    (unsigned long long)cmd->prp1, (unsigned long long)cmd->prp2,
+	    cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14, cmd->cdw15);
+
+	return cmd_string;
+}
diff --git a/sys/cam/nvme/nvme_all.h b/sys/cam/nvme/nvme_all.h
new file mode 100644
index 0000000..3cff74d
--- /dev/null
+++ b/sys/cam/nvme/nvme_all.h
@@ -0,0 +1,48 @@
+/*-
+ * Copyright (c) 2015 Netflix, Inc
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer,
+ *    without modification, immediately at the beginning of the file.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+
+#ifndef CAM_NVME_NVME_ALL_H
+#define CAM_NVME_NVME_ALL_H 1
+
+#include <dev/nvme/nvme.h>
+
+struct ccb_nvmeio;
+
+#define NVME_REV_1	1	/* Supports NVMe 1.2 or earlier */
+
+void	nvme_ns_cmd(struct ccb_nvmeio *nvmeio, uint8_t cmd, uint32_t nsid,
+    uint32_t cdw10, uint32_t cdw11, uint32_t cdw12, uint32_t cdw13,
+    uint32_t cdw14, uint32_t cdw15);
+
+int	nvme_identify_match(caddr_t identbuffer, caddr_t table_entry);
+
+void	nvme_print_ident(const struct nvme_controller_data *, const struct nvme_namespace_data *);
+const char *nvme_op_string(const struct nvme_command *);
+const char *nvme_cmd_string(const struct nvme_command *, char *, size_t);
+
+#endif /* CAM_NVME_NVME_ALL_H */
diff --git a/sys/cam/nvme/nvme_da.c b/sys/cam/nvme/nvme_da.c
new file mode 100644
index 0000000..9628530
--- /dev/null
+++ b/sys/cam/nvme/nvme_da.c
@@ -0,0 +1,1152 @@
+/*-
+ * Copyright (c) 2015 Netflix, Inc
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer,
+ *    without modification, immediately at the beginning of the file.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Derived from ata_da.c:
+ * Copyright (c) 2009 Alexander Motin <mav@FreeBSD.org>
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/param.h>
+
+#ifdef _KERNEL
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/bio.h>
+#include <sys/sysctl.h>
+#include <sys/taskqueue.h>
+#include <sys/lock.h>
+#include <sys/mutex.h>
+#include <sys/conf.h>
+#include <sys/devicestat.h>
+#include <sys/eventhandler.h>
+#include <sys/malloc.h>
+#include <sys/cons.h>
+#include <sys/proc.h>
+#include <sys/reboot.h>
+#include <geom/geom_disk.h>
+#endif /* _KERNEL */
+
+#ifndef _KERNEL
+#include <stdio.h>
+#include <string.h>
+#endif /* _KERNEL */
+
+#include <cam/cam.h>
+#include <cam/cam_ccb.h>
+#include <cam/cam_periph.h>
+#include <cam/cam_xpt_periph.h>
+#include <cam/cam_sim.h>
+#include <cam/cam_iosched.h>
+
+#include <cam/nvme/nvme_all.h>
+
+typedef enum {
+	NDA_STATE_NORMAL
+} nda_state;
+
+typedef enum {
+	NDA_FLAG_OPEN		= 0x0001,
+	NDA_FLAG_DIRTY		= 0x0002,
+	NDA_FLAG_SCTX_INIT	= 0x0004,
+} nda_flags;
+
+typedef enum {
+	NDA_Q_4K   = 0x01,
+	NDA_Q_NONE = 0x00,
+} nda_quirks;
+	
+#define NDA_Q_BIT_STRING	\
+	"\020"			\
+	"\001Bit 0"
+
+typedef enum {
+	NDA_CCB_BUFFER_IO	= 0x01,
+	NDA_CCB_DUMP            = 0x02,
+	NDA_CCB_TRIM            = 0x03,
+	NDA_CCB_TYPE_MASK	= 0x0F,
+} nda_ccb_state;
+
+/* Offsets into our private area for storing information */
+#define ccb_state	ppriv_field0
+#define ccb_bp		ppriv_ptr1
+
+struct trim_request {
+	TAILQ_HEAD(, bio) bps;
+};
+struct nda_softc {
+	struct   cam_iosched_softc *cam_iosched;
+	int	 outstanding_cmds;	/* Number of active commands */
+	int	 refcount;		/* Active xpt_action() calls */
+	nda_state state;
+	nda_flags flags;
+	nda_quirks quirks;
+	int	 unmappedio;
+	uint32_t  nsid;			/* Namespace ID for this nda device */
+	struct disk *disk;
+	struct task		sysctl_task;
+	struct sysctl_ctx_list	sysctl_ctx;
+	struct sysctl_oid	*sysctl_tree;
+	struct trim_request	trim_req;
+#ifdef CAM_IO_STATS
+	struct sysctl_ctx_list	sysctl_stats_ctx;
+	struct sysctl_oid	*sysctl_stats_tree;
+	u_int	timeouts;
+	u_int	errors;
+	u_int	invalidations;
+#endif
+};
+
+/* Need quirk table */
+
+static	disk_strategy_t	ndastrategy;
+static	dumper_t	ndadump;
+static	periph_init_t	ndainit;
+static	void		ndaasync(void *callback_arg, u_int32_t code,
+				struct cam_path *path, void *arg);
+static	void		ndasysctlinit(void *context, int pending);
+static	periph_ctor_t	ndaregister;
+static	periph_dtor_t	ndacleanup;
+static	periph_start_t	ndastart;
+static	periph_oninv_t	ndaoninvalidate;
+static	void		ndadone(struct cam_periph *periph,
+			       union ccb *done_ccb);
+static  int		ndaerror(union ccb *ccb, u_int32_t cam_flags,
+				u_int32_t sense_flags);
+static void		ndashutdown(void *arg, int howto);
+static void		ndasuspend(void *arg);
+
+#ifndef	NDA_DEFAULT_SEND_ORDERED
+#define	NDA_DEFAULT_SEND_ORDERED	1
+#endif
+#ifndef NDA_DEFAULT_TIMEOUT
+#define NDA_DEFAULT_TIMEOUT 30	/* Timeout in seconds */
+#endif
+#ifndef	NDA_DEFAULT_RETRY
+#define	NDA_DEFAULT_RETRY	4
+#endif
+
+
+//static int nda_retry_count = NDA_DEFAULT_RETRY;
+static int nda_send_ordered = NDA_DEFAULT_SEND_ORDERED;
+static int nda_default_timeout = NDA_DEFAULT_TIMEOUT;
+
+/*
+ * All NVMe media is non-rotational, so all nvme device instances
+ * share this to implement the sysctl.
+ */
+static int nda_rotating_media = 0;
+
+static SYSCTL_NODE(_kern_cam, OID_AUTO, nda, CTLFLAG_RD, 0,
+            "CAM Direct Access Disk driver");
+
+static struct periph_driver ndadriver =
+{
+	ndainit, "nda",
+	TAILQ_HEAD_INITIALIZER(ndadriver.units), /* generation */ 0
+};
+
+PERIPHDRIVER_DECLARE(nda, ndadriver);
+
+static MALLOC_DEFINE(M_NVMEDA, "nvme_da", "nvme_da buffers");
+
+/*
+ * nice wrappers. Maybe these belong in nvme_all.c instead of
+ * here, but this is the only place that uses these. Should
+ * we ever grow another NVME periph, we should move them
+ * all there wholesale.
+ */
+
+static void
+nda_nvme_flush(struct nda_softc *softc, struct ccb_nvmeio *nvmeio)
+{
+	cam_fill_nvmeio(nvmeio,
+	    0,			/* retries */
+	    ndadone,		/* cbfcnp */
+	    CAM_DIR_NONE,	/* flags */
+	    NULL,		/* data_ptr */
+	    0,			/* dxfer_len */
+	    nda_default_timeout * 1000); /* timeout 5s */
+	nvme_ns_flush_cmd(&nvmeio->cmd, softc->nsid);
+}
+
+static void
+nda_nvme_trim(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
+    void *payload, uint32_t num_ranges)
+{
+	cam_fill_nvmeio(nvmeio,
+	    0,			/* retries */
+	    ndadone,		/* cbfcnp */
+	    CAM_DIR_OUT,	/* flags */
+	    payload,		/* data_ptr */
+	    num_ranges * sizeof(struct nvme_dsm_range), /* dxfer_len */
+	    nda_default_timeout * 1000); /* timeout 5s */
+	nvme_ns_trim_cmd(&nvmeio->cmd, softc->nsid, num_ranges);
+}
+
+static void
+nda_nvme_write(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
+    void *payload, uint64_t lba, uint32_t len, uint32_t count)
+{
+	cam_fill_nvmeio(nvmeio,
+	    0,			/* retries */
+	    ndadone,		/* cbfcnp */
+	    CAM_DIR_OUT,	/* flags */
+	    payload,		/* data_ptr */
+	    len,		/* dxfer_len */
+	    nda_default_timeout * 1000); /* timeout 5s */
+	nvme_ns_write_cmd(&nvmeio->cmd, softc->nsid, lba, count);
+}
+
+static void
+nda_nvme_rw_bio(struct nda_softc *softc, struct ccb_nvmeio *nvmeio,
+    struct bio *bp, uint32_t rwcmd)
+{
+	int flags = rwcmd == NVME_OPC_READ ? CAM_DIR_IN : CAM_DIR_OUT;
+	void *payload;
+	uint64_t lba;
+	uint32_t count;
+
+	if (bp->bio_flags & BIO_UNMAPPED) {
+		flags |= CAM_DATA_BIO;
+		payload = bp;
+	} else {
+		payload = bp->bio_data;
+	}
+
+	lba = bp->bio_pblkno;
+	count = bp->bio_bcount / softc->disk->d_sectorsize;
+
+	cam_fill_nvmeio(nvmeio,
+	    0,			/* retries */
+	    ndadone,		/* cbfcnp */
+	    flags,		/* flags */
+	    payload,		/* data_ptr */
+	    bp->bio_bcount,	/* dxfer_len */
+	    nda_default_timeout * 1000);		/* timeout 5s */
+	nvme_ns_rw_cmd(&nvmeio->cmd, rwcmd, softc->nsid, lba, count);
+}
+
+static int
+ndaopen(struct disk *dp)
+{
+	struct cam_periph *periph;
+	struct nda_softc *softc;
+	int error;
+
+	periph = (struct cam_periph *)dp->d_drv1;
+	if (cam_periph_acquire(periph) != CAM_REQ_CMP) {
+		return(ENXIO);
+	}
+
+	cam_periph_lock(periph);
+	if ((error = cam_periph_hold(periph, PRIBIO|PCATCH)) != 0) {
+		cam_periph_unlock(periph);
+		cam_periph_release(periph);
+		return (error);
+	}
+
+	CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
+	    ("ndaopen\n"));
+
+	softc = (struct nda_softc *)periph->softc;
+	softc->flags |= NDA_FLAG_OPEN;
+
+	cam_periph_unhold(periph);
+	cam_periph_unlock(periph);
+	return (0);
+}
+
+static int
+ndaclose(struct disk *dp)
+{
+	struct	cam_periph *periph;
+	struct	nda_softc *softc;
+	union ccb *ccb;
+	int error;
+
+	periph = (struct cam_periph *)dp->d_drv1;
+	softc = (struct nda_softc *)periph->softc;
+	cam_periph_lock(periph);
+
+	CAM_DEBUG(periph->path, CAM_DEBUG_TRACE | CAM_DEBUG_PERIPH,
+	    ("ndaclose\n"));
+
+	if ((softc->flags & NDA_FLAG_DIRTY) != 0 &&
+	    (periph->flags & CAM_PERIPH_INVALID) == 0 &&
+	    cam_periph_hold(periph, PRIBIO) == 0) {
+
+		ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
+		nda_nvme_flush(softc, &ccb->nvmeio);
+		error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
+		    /*sense_flags*/0, softc->disk->d_devstat);
+
+		if (error != 0)
+			xpt_print(periph->path, "Synchronize cache failed\n");
+		else
+			softc->flags &= ~NDA_FLAG_DIRTY;
+		xpt_release_ccb(ccb);
+		cam_periph_unhold(periph);
+	}
+
+	softc->flags &= ~NDA_FLAG_OPEN;
+
+	while (softc->refcount != 0)
+		cam_periph_sleep(periph, &softc->refcount, PRIBIO, "ndaclose", 1);
+	cam_periph_unlock(periph);
+	cam_periph_release(periph);
+	return (0);	
+}
+
+static void
+ndaschedule(struct cam_periph *periph)
+{
+	struct nda_softc *softc = (struct nda_softc *)periph->softc;
+
+	if (softc->state != NDA_STATE_NORMAL)
+		return;
+
+	cam_iosched_schedule(softc->cam_iosched, periph);
+}
+
+/*
+ * Actually translate the requested transfer into one the physical driver
+ * can understand.  The transfer is described by a buf and will include
+ * only one physical transfer.
+ */
+static void
+ndastrategy(struct bio *bp)
+{
+	struct cam_periph *periph;
+	struct nda_softc *softc;
+	
+	periph = (struct cam_periph *)bp->bio_disk->d_drv1;
+	softc = (struct nda_softc *)periph->softc;
+
+	cam_periph_lock(periph);
+
+	CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastrategy(%p)\n", bp));
+
+	/*
+	 * If the device has been made invalid, error out
+	 */
+	if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
+		cam_periph_unlock(periph);
+		biofinish(bp, NULL, ENXIO);
+		return;
+	}
+	
+	/*
+	 * Place it in the queue of disk activities for this disk
+	 */
+	cam_iosched_queue_work(softc->cam_iosched, bp);
+
+	/*
+	 * Schedule ourselves for performing the work.
+	 */
+	ndaschedule(periph);
+	cam_periph_unlock(periph);
+
+	return;
+}
+
+static int
+ndadump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length)
+{
+	struct	    cam_periph *periph;
+	struct	    nda_softc *softc;
+	u_int	    secsize;
+	union	    ccb ccb;
+	struct	    disk *dp;
+	uint64_t    lba;
+	uint32_t    count;
+	int	    error = 0;
+
+	dp = arg;
+	periph = dp->d_drv1;
+	softc = (struct nda_softc *)periph->softc;
+	cam_periph_lock(periph);
+	secsize = softc->disk->d_sectorsize;
+	lba = offset / secsize;
+	count = length / secsize;
+	
+	if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
+		cam_periph_unlock(periph);
+		return (ENXIO);
+	}
+
+	if (length > 0) {
+		xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
+		ccb.ccb_h.ccb_state = NDA_CCB_DUMP;
+		nda_nvme_write(softc, &ccb.nvmeio, virtual, lba, length, count);
+		xpt_polled_action(&ccb);
+
+		error = cam_periph_error(&ccb,
+		    0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
+		if ((ccb.ccb_h.status & CAM_DEV_QFRZN) != 0)
+			cam_release_devq(ccb.ccb_h.path, /*relsim_flags*/0,
+			    /*reduction*/0, /*timeout*/0, /*getcount_only*/0);
+		if (error != 0)
+			printf("Aborting dump due to I/O error.\n");
+
+		cam_periph_unlock(periph);
+		return (error);
+	}
+	
+	/* Flush */
+	xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL);
+
+	ccb.ccb_h.ccb_state = NDA_CCB_DUMP;
+	nda_nvme_flush(softc, &ccb.nvmeio);
+	xpt_polled_action(&ccb);
+
+	error = cam_periph_error(&ccb,
+	    0, SF_NO_RECOVERY | SF_NO_RETRY, NULL);
+	if ((ccb.ccb_h.status & CAM_DEV_QFRZN) != 0)
+		cam_release_devq(ccb.ccb_h.path, /*relsim_flags*/0,
+		    /*reduction*/0, /*timeout*/0, /*getcount_only*/0);
+	if (error != 0)
+		xpt_print(periph->path, "flush cmd failed\n");
+	cam_periph_unlock(periph);
+	return (error);
+}
+
+static void
+ndainit(void)
+{
+	cam_status status;
+
+	/*
+	 * Install a global async callback.  This callback will
+	 * receive async callbacks like "new device found".
+	 */
+	status = xpt_register_async(AC_FOUND_DEVICE, ndaasync, NULL, NULL);
+
+	if (status != CAM_REQ_CMP) {
+		printf("nda: Failed to attach master async callback "
+		       "due to status 0x%x!\n", status);
+	} else if (nda_send_ordered) {
+
+		/* Register our event handlers */
+		if ((EVENTHANDLER_REGISTER(power_suspend, ndasuspend,
+					   NULL, EVENTHANDLER_PRI_LAST)) == NULL)
+		    printf("ndainit: power event registration failed!\n");
+		if ((EVENTHANDLER_REGISTER(shutdown_post_sync, ndashutdown,
+					   NULL, SHUTDOWN_PRI_DEFAULT)) == NULL)
+		    printf("ndainit: shutdown event registration failed!\n");
+	}
+}
+
+/*
+ * Callback from GEOM, called when it has finished cleaning up its
+ * resources.
+ */
+static void
+ndadiskgonecb(struct disk *dp)
+{
+	struct cam_periph *periph;
+
+	periph = (struct cam_periph *)dp->d_drv1;
+
+	cam_periph_release(periph);
+}
+
+static void
+ndaoninvalidate(struct cam_periph *periph)
+{
+	struct nda_softc *softc;
+
+	softc = (struct nda_softc *)periph->softc;
+
+	/*
+	 * De-register any async callbacks.
+	 */
+	xpt_register_async(0, ndaasync, periph, periph->path);
+#ifdef CAM_IO_STATS
+	softc->invalidations++;
+#endif
+
+	/*
+	 * Return all queued I/O with ENXIO.
+	 * XXX Handle any transactions queued to the card
+	 *     with XPT_ABORT_CCB.
+	 */
+	cam_iosched_flush(softc->cam_iosched, NULL, ENXIO);
+
+	disk_gone(softc->disk);
+}
+
+static void
+ndacleanup(struct cam_periph *periph)
+{
+	struct nda_softc *softc;
+
+	softc = (struct nda_softc *)periph->softc;
+
+	cam_periph_unlock(periph);
+
+	cam_iosched_fini(softc->cam_iosched);
+
+	/*
+	 * If we can't free the sysctl tree, oh well...
+	 */
+	if ((softc->flags & NDA_FLAG_SCTX_INIT) != 0) {
+#ifdef CAM_IO_STATS
+		if (sysctl_ctx_free(&softc->sysctl_stats_ctx) != 0)
+			xpt_print(periph->path,
+			    "can't remove sysctl stats context\n");
+#endif
+		if (sysctl_ctx_free(&softc->sysctl_ctx) != 0)
+			xpt_print(periph->path,
+			    "can't remove sysctl context\n");
+	}
+
+	disk_destroy(softc->disk);
+	free(softc, M_DEVBUF);
+	cam_periph_lock(periph);
+}
+
+static void
+ndaasync(void *callback_arg, u_int32_t code,
+	struct cam_path *path, void *arg)
+{
+	struct cam_periph *periph;
+
+	periph = (struct cam_periph *)callback_arg;
+	switch (code) {
+	case AC_FOUND_DEVICE:
+	{
+		struct ccb_getdev *cgd;
+		cam_status status;
+ 
+		cgd = (struct ccb_getdev *)arg;
+		if (cgd == NULL)
+			break;
+
+		if (cgd->protocol != PROTO_NVME)
+			break;
+
+		/*
+		 * Allocate a peripheral instance for
+		 * this device and start the probe
+		 * process.
+		 */
+		status = cam_periph_alloc(ndaregister, ndaoninvalidate,
+					  ndacleanup, ndastart,
+					  "nda", CAM_PERIPH_BIO,
+					  path, ndaasync,
+					  AC_FOUND_DEVICE, cgd);
+
+		if (status != CAM_REQ_CMP
+		 && status != CAM_REQ_INPROG)
+			printf("ndaasync: Unable to attach to new device "
+				"due to status 0x%x\n", status);
+		break;
+	}
+	case AC_ADVINFO_CHANGED:
+	{
+		uintptr_t buftype;
+
+		buftype = (uintptr_t)arg;
+		if (buftype == CDAI_TYPE_PHYS_PATH) {
+			struct nda_softc *softc;
+
+			softc = periph->softc;
+			disk_attr_changed(softc->disk, "GEOM::physpath",
+					  M_NOWAIT);
+		}
+		break;
+	}
+	case AC_LOST_DEVICE:
+	default:
+		cam_periph_async(periph, code, path, arg);
+		break;
+	}
+}
+
+static void
+ndasysctlinit(void *context, int pending)
+{
+	struct cam_periph *periph;
+	struct nda_softc *softc;
+	char tmpstr[80], tmpstr2[80];
+
+	periph = (struct cam_periph *)context;
+
+	/* periph was held for us when this task was enqueued */
+	if ((periph->flags & CAM_PERIPH_INVALID) != 0) {
+		cam_periph_release(periph);
+		return;
+	}
+
+	softc = (struct nda_softc *)periph->softc;
+	snprintf(tmpstr, sizeof(tmpstr), "CAM NDA unit %d", periph->unit_number);
+	snprintf(tmpstr2, sizeof(tmpstr2), "%d", periph->unit_number);
+
+	sysctl_ctx_init(&softc->sysctl_ctx);
+	softc->flags |= NDA_FLAG_SCTX_INIT;
+	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
+		SYSCTL_STATIC_CHILDREN(_kern_cam_nda), OID_AUTO, tmpstr2,
+		CTLFLAG_RD, 0, tmpstr);
+	if (softc->sysctl_tree == NULL) {
+		printf("ndasysctlinit: unable to allocate sysctl tree\n");
+		cam_periph_release(periph);
+		return;
+	}
+
+	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
+		OID_AUTO, "unmapped_io", CTLFLAG_RD | CTLFLAG_MPSAFE,
+		&softc->unmappedio, 0, "Unmapped I/O leaf");
+
+	SYSCTL_ADD_INT(&softc->sysctl_ctx,
+		       SYSCTL_CHILDREN(softc->sysctl_tree),
+		       OID_AUTO,
+		       "rotating",
+		       CTLFLAG_RD | CTLFLAG_MPSAFE, 
+		       &nda_rotating_media,
+		       0,
+		       "Rotating media");
+
+#ifdef CAM_IO_STATS
+	softc->sysctl_stats_tree = SYSCTL_ADD_NODE(&softc->sysctl_stats_ctx,
+		SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "stats",
+		CTLFLAG_RD, 0, "Statistics");
+	if (softc->sysctl_stats_tree == NULL) {
+		printf("ndasysctlinit: unable to allocate sysctl tree for stats\n");
+		cam_periph_release(periph);
+		return;
+	}
+	SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
+		SYSCTL_CHILDREN(softc->sysctl_stats_tree),
+		OID_AUTO, "timeouts", CTLFLAG_RD | CTLFLAG_MPSAFE,
+		&softc->timeouts, 0,
+		"Device timeouts reported by the SIM");
+	SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
+		SYSCTL_CHILDREN(softc->sysctl_stats_tree),
+		OID_AUTO, "errors", CTLFLAG_RD | CTLFLAG_MPSAFE,
+		&softc->errors, 0,
+		"Transport errors reported by the SIM.");
+	SYSCTL_ADD_INT(&softc->sysctl_stats_ctx,
+		SYSCTL_CHILDREN(softc->sysctl_stats_tree),
+		OID_AUTO, "pack_invalidations", CTLFLAG_RD | CTLFLAG_MPSAFE,
+		&softc->invalidations, 0,
+		"Device pack invalidations.");
+#endif
+
+	cam_iosched_sysctl_init(softc->cam_iosched, &softc->sysctl_ctx,
+	    softc->sysctl_tree);
+
+	cam_periph_release(periph);
+}
+
+static int
+ndagetattr(struct bio *bp)
+{
+	int ret;
+	struct cam_periph *periph;
+
+	periph = (struct cam_periph *)bp->bio_disk->d_drv1;
+	cam_periph_lock(periph);
+	ret = xpt_getattr(bp->bio_data, bp->bio_length, bp->bio_attribute,
+	    periph->path);
+	cam_periph_unlock(periph);
+	if (ret == 0)
+		bp->bio_completed = bp->bio_length;
+	return ret;
+}
+
+static cam_status
+ndaregister(struct cam_periph *periph, void *arg)
+{
+	struct nda_softc *softc;
+	struct disk *disk;
+	struct ccb_pathinq cpi;
+	struct ccb_getdev *cgd;
+	const struct nvme_namespace_data *nsd;
+	const struct nvme_controller_data *cd;
+	char   announce_buf[80];
+//	caddr_t match;
+	u_int maxio;
+	int quirks;
+
+	cgd = (struct ccb_getdev *)arg;
+	if (cgd == NULL) {
+		printf("ndaregister: no getdev CCB, can't register device\n");
+		return(CAM_REQ_CMP_ERR);
+	}
+	nsd = cgd->nvme_data;
+	cd = cgd->nvme_cdata;
+
+	softc = (struct nda_softc *)malloc(sizeof(*softc), M_DEVBUF,
+	    M_NOWAIT | M_ZERO);
+
+	if (softc == NULL) {
+		printf("ndaregister: Unable to probe new device. "
+		    "Unable to allocate softc\n");
+		return(CAM_REQ_CMP_ERR);
+	}
+
+	if (cam_iosched_init(&softc->cam_iosched, periph) != 0) {
+		printf("ndaregister: Unable to probe new device. "
+		       "Unable to allocate iosched memory\n");
+		return(CAM_REQ_CMP_ERR);
+	}
+
+	/* ident_data parsing */
+
+	periph->softc = softc;
+
+#if 0
+	/*
+	 * See if this device has any quirks.
+	 */
+	match = cam_quirkmatch((caddr_t)&cgd->ident_data,
+			       (caddr_t)nda_quirk_table,
+			       sizeof(nda_quirk_table)/sizeof(*nda_quirk_table),
+			       sizeof(*nda_quirk_table), ata_identify_match);
+	if (match != NULL)
+		softc->quirks = ((struct nda_quirk_entry *)match)->quirks;
+	else
+#endif
+		softc->quirks = NDA_Q_NONE;
+
+	bzero(&cpi, sizeof(cpi));
+	xpt_setup_ccb(&cpi.ccb_h, periph->path, CAM_PRIORITY_NONE);
+	cpi.ccb_h.func_code = XPT_PATH_INQ;
+	xpt_action((union ccb *)&cpi);
+
+	TASK_INIT(&softc->sysctl_task, 0, ndasysctlinit, periph);
+
+	/*
+	 * The name space ID is the lun, save it for later I/O
+	 */
+	softc->nsid = (uint16_t)xpt_path_lun_id(periph->path);
+
+	/*
+	 * Register this media as a disk
+	 */
+	(void)cam_periph_hold(periph, PRIBIO);
+	cam_periph_unlock(periph);
+	snprintf(announce_buf, sizeof(announce_buf),
+	    "kern.cam.nda.%d.quirks", periph->unit_number);
+	quirks = softc->quirks;
+	TUNABLE_INT_FETCH(announce_buf, &quirks);
+	softc->quirks = quirks;
+	cam_iosched_set_sort_queue(softc->cam_iosched, 0);
+	softc->disk = disk = disk_alloc();
+	strlcpy(softc->disk->d_descr, cd->mn,
+	    MIN(sizeof(softc->disk->d_descr), sizeof(cd->mn)));
+	strlcpy(softc->disk->d_ident, cd->sn,
+	    MIN(sizeof(softc->disk->d_ident), sizeof(cd->sn)));
+	disk->d_rotation_rate = 0;	/* Spinning rust need not apply */
+	disk->d_open = ndaopen;
+	disk->d_close = ndaclose;
+	disk->d_strategy = ndastrategy;
+	disk->d_getattr = ndagetattr;
+	disk->d_dump = ndadump;
+	disk->d_gone = ndadiskgonecb;
+	disk->d_name = "nda";
+	disk->d_drv1 = periph;
+	disk->d_unit = periph->unit_number;
+	maxio = cpi.maxio;		/* Honor max I/O size of SIM */
+	if (maxio == 0)
+		maxio = DFLTPHYS;	/* traditional default */
+	else if (maxio > MAXPHYS)
+		maxio = MAXPHYS;	/* for safety */
+	disk->d_maxsize = maxio;
+	disk->d_sectorsize = 1 << nsd->lbaf[nsd->flbas.format].lbads;
+	disk->d_mediasize = (off_t)(disk->d_sectorsize * nsd->nsze);
+	disk->d_delmaxsize = disk->d_mediasize;
+	disk->d_flags = DISKFLAG_DIRECT_COMPLETION;
+//	if (cd->oncs.dsm) // XXX broken?
+		disk->d_flags |= DISKFLAG_CANDELETE;
+	if (cd->vwc.present)
+		disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
+	if ((cpi.hba_misc & PIM_UNMAPPED) != 0) {
+		disk->d_flags |= DISKFLAG_UNMAPPED_BIO;
+		softc->unmappedio = 1;
+	}
+	/*
+	 * d_ident and d_descr are both far bigger than the length of either
+	 *  the serial or model number strings.
+	 */
+	nvme_strvis(disk->d_descr, cd->mn,
+	    sizeof(disk->d_descr), NVME_MODEL_NUMBER_LENGTH);
+	nvme_strvis(disk->d_ident, cd->sn,
+	    sizeof(disk->d_ident), NVME_SERIAL_NUMBER_LENGTH);
+	disk->d_hba_vendor = cpi.hba_vendor;
+	disk->d_hba_device = cpi.hba_device;
+	disk->d_hba_subvendor = cpi.hba_subvendor;
+	disk->d_hba_subdevice = cpi.hba_subdevice;
+	disk->d_stripesize = disk->d_sectorsize;
+	disk->d_stripeoffset = 0;
+	disk->d_devstat = devstat_new_entry(periph->periph_name,
+	    periph->unit_number, disk->d_sectorsize,
+	    DEVSTAT_ALL_SUPPORTED,
+	    DEVSTAT_TYPE_DIRECT | XPORT_DEVSTAT_TYPE(cpi.transport),
+	    DEVSTAT_PRIORITY_DISK);
+
+	/*
+	 * Acquire a reference to the periph before we register with GEOM.
+	 * We'll release this reference once GEOM calls us back (via
+	 * ndadiskgonecb()) telling us that our provider has been freed.
+	 */
+	if (cam_periph_acquire(periph) != CAM_REQ_CMP) {
+		xpt_print(periph->path, "%s: lost periph during "
+			  "registration!\n", __func__);
+		cam_periph_lock(periph);
+		return (CAM_REQ_CMP_ERR);
+	}
+	disk_create(softc->disk, DISK_VERSION);
+	cam_periph_lock(periph);
+	cam_periph_unhold(periph);
+
+	snprintf(announce_buf, sizeof(announce_buf),
+		"%juMB (%ju %u byte sectors)",
+	    (uintmax_t)((uintmax_t)disk->d_mediasize / (1024*1024)),
+		(uintmax_t)disk->d_mediasize / disk->d_sectorsize,
+		disk->d_sectorsize);
+	xpt_announce_periph(periph, announce_buf);
+	xpt_announce_quirks(periph, softc->quirks, NDA_Q_BIT_STRING);
+
+	/*
+	 * Create our sysctl variables, now that we know
+	 * we have successfully attached.
+	 */
+	if (cam_periph_acquire(periph) == CAM_REQ_CMP)
+		taskqueue_enqueue(taskqueue_thread, &softc->sysctl_task);
+
+	/*
+	 * Register for device going away and info about the drive
+	 * changing (though with NVMe, it can't)
+	 */
+	xpt_register_async(AC_LOST_DEVICE | AC_ADVINFO_CHANGED,
+	    ndaasync, periph, periph->path);
+
+	softc->state = NDA_STATE_NORMAL;
+	return(CAM_REQ_CMP);
+}
+
+static void
+ndastart(struct cam_periph *periph, union ccb *start_ccb)
+{
+	struct nda_softc *softc = (struct nda_softc *)periph->softc;
+	struct ccb_nvmeio *nvmeio = &start_ccb->nvmeio;
+
+	CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart\n"));
+
+	switch (softc->state) {
+	case NDA_STATE_NORMAL:
+	{
+		struct bio *bp;
+
+		bp = cam_iosched_next_bio(softc->cam_iosched);
+		CAM_DEBUG(periph->path, CAM_DEBUG_TRACE, ("ndastart: bio %p\n", bp));
+		if (bp == NULL) {
+			xpt_release_ccb(start_ccb);
+			break;
+		}
+
+		switch (bp->bio_cmd) {
+		case BIO_WRITE:
+			softc->flags |= NDA_FLAG_DIRTY;
+			/* FALLTHROUGH */
+		case BIO_READ:
+		{
+#ifdef NDA_TEST_FAILURE
+			int fail = 0;
+
+			/*
+			 * Support the failure ioctls.  If the command is a
+			 * read, and there are pending forced read errors, or
+			 * if a write and pending write errors, then fail this
+			 * operation with EIO.  This is useful for testing
+			 * purposes.  Also, support having every Nth read fail.
+			 *
+			 * This is a rather blunt tool.
+			 */
+			if (bp->bio_cmd == BIO_READ) {
+				if (softc->force_read_error) {
+					softc->force_read_error--;
+					fail = 1;
+				}
+				if (softc->periodic_read_error > 0) {
+					if (++softc->periodic_read_count >=
+					    softc->periodic_read_error) {
+						softc->periodic_read_count = 0;
+						fail = 1;
+					}
+				}
+			} else {
+				if (softc->force_write_error) {
+					softc->force_write_error--;
+					fail = 1;
+				}
+			}
+			if (fail) {
+				biofinish(bp, NULL, EIO);
+				xpt_release_ccb(start_ccb);
+				ndaschedule(periph);
+				return;
+			}
+#endif
+			KASSERT((bp->bio_flags & BIO_UNMAPPED) == 0 ||
+			    round_page(bp->bio_bcount + bp->bio_ma_offset) /
+			    PAGE_SIZE == bp->bio_ma_n,
+			    ("Short bio %p", bp));
+			nda_nvme_rw_bio(softc, &start_ccb->nvmeio, bp, bp->bio_cmd == BIO_READ ?
+			    NVME_OPC_READ : NVME_OPC_WRITE);
+			break;
+		}
+		case BIO_DELETE:
+		{
+			struct nvme_dsm_range *dsm_range;
+
+			dsm_range =
+			    malloc(sizeof(*dsm_range), M_NVMEDA, M_ZERO | M_WAITOK);
+			dsm_range->length =
+			    bp->bio_bcount / softc->disk->d_sectorsize;
+			dsm_range->starting_lba =
+			    bp->bio_offset / softc->disk->d_sectorsize;
+			bp->bio_driver2 = dsm_range;
+			nda_nvme_trim(softc, &start_ccb->nvmeio, dsm_range, 1);
+			start_ccb->ccb_h.ccb_state = NDA_CCB_TRIM;
+			start_ccb->ccb_h.flags |= CAM_UNLOCKED;
+			cam_iosched_submit_trim(softc->cam_iosched);	/* XXX */
+			goto out;
+		}
+		case BIO_FLUSH:
+			nda_nvme_flush(softc, nvmeio);
+			break;
+		}
+		start_ccb->ccb_h.ccb_state = NDA_CCB_BUFFER_IO;
+		start_ccb->ccb_h.flags |= CAM_UNLOCKED;
+out:
+		start_ccb->ccb_h.ccb_bp = bp;
+		softc->outstanding_cmds++;
+		softc->refcount++;
+		cam_periph_unlock(periph);
+		xpt_action(start_ccb);
+		cam_periph_lock(periph);
+		softc->refcount--;
+
+		/* May have more work to do, so ensure we stay scheduled */
+		ndaschedule(periph);
+		break;
+		}
+	}
+}
+
+static void
+ndadone(struct cam_periph *periph, union ccb *done_ccb)
+{
+	struct nda_softc *softc;
+	struct ccb_nvmeio *nvmeio = &done_ccb->nvmeio;
+	struct cam_path *path;
+	int state;
+
+	softc = (struct nda_softc *)periph->softc;
+	path = done_ccb->ccb_h.path;
+
+	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("ndadone\n"));
+
+	state = nvmeio->ccb_h.ccb_state & NDA_CCB_TYPE_MASK;
+	switch (state) {
+	case NDA_CCB_BUFFER_IO:
+	case NDA_CCB_TRIM:
+	{
+		struct bio *bp;
+		int error;
+
+		cam_periph_lock(periph);
+		bp = (struct bio *)done_ccb->ccb_h.ccb_bp;
+		if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
+			error = ndaerror(done_ccb, 0, 0);
+			if (error == ERESTART) {
+				/* A retry was scheduled, so just return. */
+				cam_periph_unlock(periph);
+				return;
+			}
+			if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
+				cam_release_devq(path,
+						 /*relsim_flags*/0,
+						 /*reduction*/0,
+						 /*timeout*/0,
+						 /*getcount_only*/0);
+		} else {
+			if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0)
+				panic("REQ_CMP with QFRZN");
+			error = 0;
+		}
+		bp->bio_error = error;
+		if (error != 0) {
+			bp->bio_resid = bp->bio_bcount;
+			bp->bio_flags |= BIO_ERROR;
+		} else {
+			if (state == NDA_CCB_TRIM)
+				bp->bio_resid = 0;
+			else
+				bp->bio_resid = nvmeio->resid;
+			if (bp->bio_resid > 0)
+				bp->bio_flags |= BIO_ERROR;
+		}
+		if (state == NDA_CCB_TRIM)
+			free(bp->bio_driver2, M_NVMEDA);
+		softc->outstanding_cmds--;
+
+		cam_iosched_bio_complete(softc->cam_iosched, bp, done_ccb);
+		xpt_release_ccb(done_ccb);
+		if (state == NDA_CCB_TRIM) {
+#ifdef notyet
+			TAILQ_HEAD(, bio) queue;
+			struct bio *bp1;
+
+			TAILQ_INIT(&queue);
+			TAILQ_CONCAT(&queue, &softc->trim_req.bps, bio_queue);
+#endif
+			cam_iosched_trim_done(softc->cam_iosched);
+			ndaschedule(periph);
+			cam_periph_unlock(periph);
+#ifdef notyet
+/* Not yet collapsing several BIO_DELETE requests into one TRIM */
+			while ((bp1 = TAILQ_FIRST(&queue)) != NULL) {
+				TAILQ_REMOVE(&queue, bp1, bio_queue);
+				bp1->bio_error = error;
+				if (error != 0) {
+					bp1->bio_flags |= BIO_ERROR;
+					bp1->bio_resid = bp1->bio_bcount;
+				} else
+					bp1->bio_resid = 0;
+				biodone(bp1);
+			}
+#else
+			biodone(bp);
+#endif
+		} else {
+			ndaschedule(periph);
+			cam_periph_unlock(periph);
+			biodone(bp);
+		}
+		return;
+	}
+	case NDA_CCB_DUMP:
+		/* No-op.  We're polling */
+		return;
+	default:
+		break;
+	}
+	xpt_release_ccb(done_ccb);
+}
+
+static int
+ndaerror(union ccb *ccb, u_int32_t cam_flags, u_int32_t sense_flags)
+{
+	struct nda_softc *softc;
+	struct cam_periph *periph;
+
+	periph = xpt_path_periph(ccb->ccb_h.path);
+	softc = (struct nda_softc *)periph->softc;
+
+	switch (ccb->ccb_h.status & CAM_STATUS_MASK) {
+	case CAM_CMD_TIMEOUT:
+#ifdef CAM_IO_STATS
+		softc->timeouts++;
+#endif
+		break;
+	case CAM_REQ_ABORTED:
+	case CAM_REQ_CMP_ERR:
+	case CAM_REQ_TERMIO:
+	case CAM_UNREC_HBA_ERROR:
+	case CAM_DATA_RUN_ERR:
+	case CAM_ATA_STATUS_ERROR:
+#ifdef CAM_IO_STATS
+		softc->errors++;
+#endif
+		break;
+	default:
+		break;
+	}
+
+	return(cam_periph_error(ccb, cam_flags, sense_flags, NULL));
+}
+
+/*
+ * Step through all NDA peripheral drivers, and if the device is still open,
+ * sync the disk cache to physical media.
+ */
+static void
+ndaflush(void)
+{
+	struct cam_periph *periph;
+	struct nda_softc *softc;
+	union ccb *ccb;
+	int error;
+
+	CAM_PERIPH_FOREACH(periph, &ndadriver) {
+		softc = (struct nda_softc *)periph->softc;
+		if (SCHEDULER_STOPPED()) {
+			/* If we paniced with the lock held, do not recurse. */
+			if (!cam_periph_owned(periph) &&
+			    (softc->flags & NDA_FLAG_OPEN)) {
+				ndadump(softc->disk, NULL, 0, 0, 0);
+			}
+			continue;
+		}
+		cam_periph_lock(periph);
+		/*
+		 * We only sync the cache if the drive is still open, and
+		 * if the drive is capable of it..
+		 */
+		if ((softc->flags & NDA_FLAG_OPEN) == 0) {
+			cam_periph_unlock(periph);
+			continue;
+		}
+
+		ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL);
+		nda_nvme_flush(softc, &ccb->nvmeio);
+		error = cam_periph_runccb(ccb, ndaerror, /*cam_flags*/0,
+		    /*sense_flags*/ SF_NO_RECOVERY | SF_NO_RETRY,
+		    softc->disk->d_devstat);
+		if (error != 0)
+			xpt_print(periph->path, "Synchronize cache failed\n");
+		xpt_release_ccb(ccb);
+		cam_periph_unlock(periph);
+	}
+}
+
+static void
+ndashutdown(void *arg, int howto)
+{
+
+	ndaflush();
+}
+
+static void
+ndasuspend(void *arg)
+{
+
+	ndaflush();
+}
diff --git a/sys/cam/nvme/nvme_xpt.c b/sys/cam/nvme/nvme_xpt.c
new file mode 100644
index 0000000..9139e29
--- /dev/null
+++ b/sys/cam/nvme/nvme_xpt.c
@@ -0,0 +1,605 @@
+/*-
+ * Copyright (c) 2015 Netflix, Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer,
+ *    without modification, immediately at the beginning of the file.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * derived from ata_xpt.c: Copyright (c) 2009 Alexander Motin <mav@FreeBSD.org>
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/param.h>
+#include <sys/bus.h>
+#include <sys/endian.h>
+#include <sys/systm.h>
+#include <sys/types.h>
+#include <sys/malloc.h>
+#include <sys/kernel.h>
+#include <sys/time.h>
+#include <sys/conf.h>
+#include <sys/fcntl.h>
+#include <sys/interrupt.h>
+#include <sys/sbuf.h>
+
+#include <sys/lock.h>
+#include <sys/mutex.h>
+#include <sys/sysctl.h>
+
+#include <cam/cam.h>
+#include <cam/cam_ccb.h>
+#include <cam/cam_queue.h>
+#include <cam/cam_periph.h>
+#include <cam/cam_sim.h>
+#include <cam/cam_xpt.h>
+#include <cam/cam_xpt_sim.h>
+#include <cam/cam_xpt_periph.h>
+#include <cam/cam_xpt_internal.h>
+#include <cam/cam_debug.h>
+
+#include <cam/scsi/scsi_all.h>
+#include <cam/scsi/scsi_message.h>
+#include <cam/nvme/nvme_all.h>
+#include <machine/stdarg.h>	/* for xpt_print below */
+#include "opt_cam.h"
+
+struct nvme_quirk_entry {
+	u_int quirks;
+#define CAM_QUIRK_MAXTAGS 1
+	u_int mintags;
+	u_int maxtags;
+};
+
+/* Not even sure why we need this */
+static periph_init_t nvme_probe_periph_init;
+
+static struct periph_driver nvme_probe_driver =
+{
+	nvme_probe_periph_init, "nvme_probe",
+	TAILQ_HEAD_INITIALIZER(nvme_probe_driver.units), /* generation */ 0,
+	CAM_PERIPH_DRV_EARLY
+};
+
+PERIPHDRIVER_DECLARE(nvme_probe, nvme_probe_driver);
+
+typedef enum {
+	NVME_PROBE_IDENTIFY,
+	NVME_PROBE_DONE,
+	NVME_PROBE_INVALID,
+	NVME_PROBE_RESET
+} nvme_probe_action;
+
+static char *nvme_probe_action_text[] = {
+	"NVME_PROBE_IDENTIFY",
+	"NVME_PROBE_DONE",
+	"NVME_PROBE_INVALID",
+	"NVME_PROBE_RESET",
+};
+
+#define NVME_PROBE_SET_ACTION(softc, newaction)	\
+do {									\
+	char **text;							\
+	text = nvme_probe_action_text;					\
+	CAM_DEBUG((softc)->periph->path, CAM_DEBUG_PROBE,		\
+	    ("Probe %s to %s\n", text[(softc)->action],			\
+	    text[(newaction)]));					\
+	(softc)->action = (newaction);					\
+} while(0)
+
+typedef enum {
+	NVME_PROBE_NO_ANNOUNCE	= 0x04
+} nvme_probe_flags;
+
+typedef struct {
+	TAILQ_HEAD(, ccb_hdr) request_ccbs;
+	nvme_probe_action	action;
+	nvme_probe_flags	flags;
+	int		restart;
+	struct cam_periph *periph;
+} nvme_probe_softc;
+
+static struct nvme_quirk_entry nvme_quirk_table[] =
+{
+	{
+//		{
+//		  T_ANY, SIP_MEDIA_REMOVABLE|SIP_MEDIA_FIXED,
+//		  /*vendor*/"*", /*product*/"*", /*revision*/"*"
+//		},
+		.quirks = 0, .mintags = 0, .maxtags = 0
+	},
+};
+
+static const int nvme_quirk_table_size =
+	sizeof(nvme_quirk_table) / sizeof(*nvme_quirk_table);
+
+static cam_status	nvme_probe_register(struct cam_periph *periph,
+				      void *arg);
+static void	 nvme_probe_schedule(struct cam_periph *nvme_probe_periph);
+static void	 nvme_probe_start(struct cam_periph *periph, union ccb *start_ccb);
+static void	 nvme_probe_cleanup(struct cam_periph *periph);
+static void	 nvme_find_quirk(struct cam_ed *device);
+static void	 nvme_scan_lun(struct cam_periph *periph,
+			       struct cam_path *path, cam_flags flags,
+			       union ccb *ccb);
+static struct cam_ed *
+		 nvme_alloc_device(struct cam_eb *bus, struct cam_et *target,
+				   lun_id_t lun_id);
+static void	 nvme_device_transport(struct cam_path *path);
+static void	 nvme_dev_async(u_int32_t async_code,
+				struct cam_eb *bus,
+				struct cam_et *target,
+				struct cam_ed *device,
+				void *async_arg);
+static void	 nvme_action(union ccb *start_ccb);
+static void	 nvme_announce_periph(struct cam_periph *periph);
+
+static struct xpt_xport nvme_xport = {
+	.alloc_device = nvme_alloc_device,
+	.action = nvme_action,
+	.async = nvme_dev_async,
+	.announce = nvme_announce_periph,
+};
+
+struct xpt_xport *
+nvme_get_xport(void)
+{
+	return (&nvme_xport);
+}
+
+static void
+nvme_probe_periph_init()
+{
+	printf("nvme cam probe device init\n");
+}
+
+static cam_status
+nvme_probe_register(struct cam_periph *periph, void *arg)
+{
+	union ccb *request_ccb;	/* CCB representing the probe request */
+	cam_status status;
+	nvme_probe_softc *softc;
+
+	request_ccb = (union ccb *)arg;
+	if (request_ccb == NULL) {
+		printf("nvme_probe_register: no probe CCB, "
+		       "can't register device\n");
+		return(CAM_REQ_CMP_ERR);
+	}
+
+	softc = (nvme_probe_softc *)malloc(sizeof(*softc), M_CAMXPT, M_ZERO | M_NOWAIT);
+
+	if (softc == NULL) {
+		printf("nvme_probe_register: Unable to probe new device. "
+		       "Unable to allocate softc\n");
+		return(CAM_REQ_CMP_ERR);
+	}
+	TAILQ_INIT(&softc->request_ccbs);
+	TAILQ_INSERT_TAIL(&softc->request_ccbs, &request_ccb->ccb_h,
+			  periph_links.tqe);
+	softc->flags = 0;
+	periph->softc = softc;
+	softc->periph = periph;
+	softc->action = NVME_PROBE_INVALID;
+	status = cam_periph_acquire(periph);
+	if (status != CAM_REQ_CMP) {
+		return (status);
+	}
+	CAM_DEBUG(periph->path, CAM_DEBUG_PROBE, ("Probe started\n"));
+
+//	nvme_device_transport(periph->path);
+	nvme_probe_schedule(periph);
+
+	return(CAM_REQ_CMP);
+}
+
+static void
+nvme_probe_schedule(struct cam_periph *periph)
+{
+	union ccb *ccb;
+	nvme_probe_softc *softc;
+
+	softc = (nvme_probe_softc *)periph->softc;
+	ccb = (union ccb *)TAILQ_FIRST(&softc->request_ccbs);
+
+	NVME_PROBE_SET_ACTION(softc, NVME_PROBE_IDENTIFY);
+
+	if (ccb->crcn.flags & CAM_EXPECT_INQ_CHANGE)
+		softc->flags |= NVME_PROBE_NO_ANNOUNCE;
+	else
+		softc->flags &= ~NVME_PROBE_NO_ANNOUNCE;
+
+	xpt_schedule(periph, CAM_PRIORITY_XPT);
+}
+
+static void
+nvme_probe_start(struct cam_periph *periph, union ccb *start_ccb)
+{
+	struct ccb_nvmeio *nvmeio;
+	struct ccb_scsiio *csio;
+	nvme_probe_softc *softc;
+	struct cam_path *path;
+	const struct nvme_namespace_data *nvme_data;
+	lun_id_t lun;
+
+	CAM_DEBUG(start_ccb->ccb_h.path, CAM_DEBUG_TRACE, ("nvme_probe_start\n"));
+
+	softc = (nvme_probe_softc *)periph->softc;
+	path = start_ccb->ccb_h.path;
+	nvmeio = &start_ccb->nvmeio;
+	csio = &start_ccb->csio;
+	nvme_data = periph->path->device->nvme_data;
+
+	if (softc->restart) {
+		softc->restart = 0;
+		if (periph->path->device->flags & CAM_DEV_UNCONFIGURED)
+			NVME_PROBE_SET_ACTION(softc, NVME_PROBE_RESET);
+		else
+			NVME_PROBE_SET_ACTION(softc, NVME_PROBE_IDENTIFY);
+	}
+
+	/*
+	 * Other transports have to ask their SIM to do a lot of action.
+	 * NVMe doesn't, so don't do the dance. Just do things
+	 * directly.
+	 */
+	switch (softc->action) {
+	case NVME_PROBE_RESET:
+		/* FALLTHROUGH */
+	case NVME_PROBE_IDENTIFY:
+		nvme_device_transport(path);
+		/*
+		 * Test for lun == CAM_LUN_WILDCARD is lame, but
+		 * appears to be necessary here. XXX
+		 */
+		lun = xpt_path_lun_id(periph->path);
+		if (lun == CAM_LUN_WILDCARD ||
+		    periph->path->device->flags & CAM_DEV_UNCONFIGURED) {
+			path->device->flags &= ~CAM_DEV_UNCONFIGURED;
+			xpt_acquire_device(path->device);
+			start_ccb->ccb_h.func_code = XPT_GDEV_TYPE;
+			xpt_action(start_ccb);
+			xpt_async(AC_FOUND_DEVICE, path, start_ccb);
+		}
+		NVME_PROBE_SET_ACTION(softc, NVME_PROBE_DONE);
+		break;
+	default:
+		panic("nvme_probe_start: invalid action state 0x%x\n", softc->action);
+	}
+	/*
+	 * Probing is now done. We need to complete any lingering items
+	 * in the queue, though there shouldn't be any.
+	 */
+	xpt_release_ccb(start_ccb);
+	CAM_DEBUG(periph->path, CAM_DEBUG_PROBE, ("Probe completed\n"));
+	while ((start_ccb = (union ccb *)TAILQ_FIRST(&softc->request_ccbs))) {
+		TAILQ_REMOVE(&softc->request_ccbs,
+		    &start_ccb->ccb_h, periph_links.tqe);
+		start_ccb->ccb_h.status = CAM_REQ_CMP;
+		xpt_done(start_ccb);
+	}
+// XXX not sure I need this
+// XXX unlike other XPTs, we never freeze the queue since we have a super-simple
+// XXX state machine
+	/* Drop freeze taken due to CAM_DEV_QFREEZE flag set. -- did we really do this? */
+//	cam_release_devq(path, 0, 0, 0, FALSE);
+	cam_periph_invalidate(periph);
+	cam_periph_release_locked(periph);
+}
+
+static void
+nvme_probe_cleanup(struct cam_periph *periph)
+{
+	free(periph->softc, M_CAMXPT);
+}
+
+/* XXX should be used, don't delete */
+static void
+nvme_find_quirk(struct cam_ed *device)
+{
+	struct nvme_quirk_entry *quirk;
+	caddr_t	match;
+
+	match = cam_quirkmatch((caddr_t)&device->nvme_data,
+			       (caddr_t)nvme_quirk_table,
+			       nvme_quirk_table_size,
+			       sizeof(*nvme_quirk_table), nvme_identify_match);
+
+	if (match == NULL)
+		panic("xpt_find_quirk: device didn't match wildcard entry!!");
+
+	quirk = (struct nvme_quirk_entry *)match;
+	device->quirk = quirk;
+	if (quirk->quirks & CAM_QUIRK_MAXTAGS) {
+		device->mintags = quirk->mintags;
+		device->maxtags = quirk->maxtags;
+	}
+}
+
+static void
+nvme_scan_lun(struct cam_periph *periph, struct cam_path *path,
+	     cam_flags flags, union ccb *request_ccb)
+{
+	struct ccb_pathinq cpi;
+	cam_status status;
+	struct cam_periph *old_periph;
+	int lock;
+
+	CAM_DEBUG(path, CAM_DEBUG_TRACE, ("nvme_scan_lun\n"));
+
+	xpt_setup_ccb(&cpi.ccb_h, path, CAM_PRIORITY_NONE);
+	cpi.ccb_h.func_code = XPT_PATH_INQ;
+	xpt_action((union ccb *)&cpi);
+
+	if (cpi.ccb_h.status != CAM_REQ_CMP) {
+		if (request_ccb != NULL) {
+			request_ccb->ccb_h.status = cpi.ccb_h.status;
+			xpt_done(request_ccb);
+		}
+		return;
+	}
+
+	if (xpt_path_lun_id(path) == CAM_LUN_WILDCARD) {
+		CAM_DEBUG(path, CAM_DEBUG_TRACE, ("nvme_scan_lun ignoring bus\n"));
+		request_ccb->ccb_h.status = CAM_REQ_CMP;	/* XXX signal error ? */
+		xpt_done(request_ccb);
+		return;
+	}
+
+	lock = (xpt_path_owned(path) == 0);
+	if (lock)
+		xpt_path_lock(path);
+	if ((old_periph = cam_periph_find(path, "nvme_probe")) != NULL) {
+		if ((old_periph->flags & CAM_PERIPH_INVALID) == 0) {
+			nvme_probe_softc *softc;
+
+			softc = (nvme_probe_softc *)old_periph->softc;
+			TAILQ_INSERT_TAIL(&softc->request_ccbs,
+				&request_ccb->ccb_h, periph_links.tqe);
+			softc->restart = 1;
+			CAM_DEBUG(path, CAM_DEBUG_TRACE,
+			    ("restarting nvme_probe device\n"));
+		} else {
+			request_ccb->ccb_h.status = CAM_REQ_CMP_ERR;
+			CAM_DEBUG(path, CAM_DEBUG_TRACE,
+			    ("Failing to restart nvme_probe device\n"));
+			xpt_done(request_ccb);
+		}
+	} else {
+		CAM_DEBUG(path, CAM_DEBUG_TRACE,
+		    ("Adding nvme_probe device\n"));
+		status = cam_periph_alloc(nvme_probe_register, NULL, nvme_probe_cleanup,
+					  nvme_probe_start, "nvme_probe",
+					  CAM_PERIPH_BIO,
+					  request_ccb->ccb_h.path, NULL, 0,
+					  request_ccb);
+
+		if (status != CAM_REQ_CMP) {
+			xpt_print(path, "xpt_scan_lun: cam_alloc_periph "
+			    "returned an error, can't continue probe\n");
+			request_ccb->ccb_h.status = status;
+			xpt_done(request_ccb);
+		}
+	}
+	if (lock)
+		xpt_path_unlock(path);
+}
+
+static struct cam_ed *
+nvme_alloc_device(struct cam_eb *bus, struct cam_et *target, lun_id_t lun_id)
+{
+	struct nvme_quirk_entry *quirk;
+	struct cam_ed *device;
+
+	device = xpt_alloc_device(bus, target, lun_id);
+	if (device == NULL)
+		return (NULL);
+
+	/*
+	 * Take the default quirk entry until we have inquiry
+	 * data from nvme and can determine a better quirk to use.
+	 */
+	quirk = &nvme_quirk_table[nvme_quirk_table_size - 1];
+	device->quirk = (void *)quirk;
+	device->mintags = 0;
+	device->maxtags = 0;
+	device->inq_flags = 0;
+	device->queue_flags = 0;
+	device->device_id = NULL;	/* XXX Need to set this somewhere */
+	device->device_id_len = 0;
+	device->serial_num = NULL;	/* XXX Need to set this somewhere */
+	device->serial_num_len = 0;
+	return (device);
+}
+
+static void
+nvme_device_transport(struct cam_path *path)
+{
+	struct ccb_pathinq cpi;
+	struct ccb_trans_settings cts;
+	/* XXX get data from nvme namespace and other info ??? */
+
+	/* Get transport information from the SIM */
+	xpt_setup_ccb(&cpi.ccb_h, path, CAM_PRIORITY_NONE);
+	cpi.ccb_h.func_code = XPT_PATH_INQ;
+	xpt_action((union ccb *)&cpi);
+
+	path->device->transport = cpi.transport;
+	path->device->transport_version = cpi.transport_version;
+
+	path->device->protocol = cpi.protocol;
+	path->device->protocol_version = cpi.protocol_version;
+
+	/* Tell the controller what we think */
+	xpt_setup_ccb(&cts.ccb_h, path, CAM_PRIORITY_NONE);
+	cts.ccb_h.func_code = XPT_SET_TRAN_SETTINGS;
+	cts.type = CTS_TYPE_CURRENT_SETTINGS;
+	cts.transport = path->device->transport;
+	cts.transport_version = path->device->transport_version;
+	cts.protocol = path->device->protocol;
+	cts.protocol_version = path->device->protocol_version;
+	cts.proto_specific.valid = 0;
+	cts.xport_specific.valid = 0;
+	xpt_action((union ccb *)&cts);
+}
+
+static void
+nvme_dev_advinfo(union ccb *start_ccb)
+{
+	struct cam_ed *device;
+	struct ccb_dev_advinfo *cdai;
+	off_t amt; 
+
+	start_ccb->ccb_h.status = CAM_REQ_INVALID;
+	device = start_ccb->ccb_h.path->device;
+	cdai = &start_ccb->cdai;
+	switch(cdai->buftype) {
+	case CDAI_TYPE_SCSI_DEVID:
+		if (cdai->flags & CDAI_FLAG_STORE)
+			return;
+		cdai->provsiz = device->device_id_len;
+		if (device->device_id_len == 0)
+			break;
+		amt = device->device_id_len;
+		if (cdai->provsiz > cdai->bufsiz)
+			amt = cdai->bufsiz;
+		memcpy(cdai->buf, device->device_id, amt);
+		break;
+	case CDAI_TYPE_SERIAL_NUM:
+		if (cdai->flags & CDAI_FLAG_STORE)
+			return;
+		cdai->provsiz = device->serial_num_len;
+		if (device->serial_num_len == 0)
+			break;
+		amt = device->serial_num_len;
+		if (cdai->provsiz > cdai->bufsiz)
+			amt = cdai->bufsiz;
+		memcpy(cdai->buf, device->serial_num, amt);
+		break;
+	case CDAI_TYPE_PHYS_PATH:
+		if (cdai->flags & CDAI_FLAG_STORE) {
+			if (device->physpath != NULL)
+				free(device->physpath, M_CAMXPT);
+			device->physpath_len = cdai->bufsiz;
+			/* Clear existing buffer if zero length */
+			if (cdai->bufsiz == 0)
+				break;
+			device->physpath = malloc(cdai->bufsiz, M_CAMXPT, M_NOWAIT);
+			if (device->physpath == NULL) {
+				start_ccb->ccb_h.status = CAM_REQ_ABORTED;
+				return;
+			}
+			memcpy(device->physpath, cdai->buf, cdai->bufsiz);
+		} else {
+			cdai->provsiz = device->physpath_len;
+			if (device->physpath_len == 0)
+				break;
+			amt = device->physpath_len;
+			if (cdai->provsiz > cdai->bufsiz)
+				amt = cdai->bufsiz;
+			memcpy(cdai->buf, device->physpath, amt);
+		}
+		break;
+	default:
+		return;
+	}
+	start_ccb->ccb_h.status = CAM_REQ_CMP;
+
+	if (cdai->flags & CDAI_FLAG_STORE) {
+		xpt_async(AC_ADVINFO_CHANGED, start_ccb->ccb_h.path,
+			  (void *)(uintptr_t)cdai->buftype);
+	}
+}
+
+static void
+nvme_action(union ccb *start_ccb)
+{
+	CAM_DEBUG(start_ccb->ccb_h.path, CAM_DEBUG_TRACE,
+	    ("nvme_action: func= %#x\n", start_ccb->ccb_h.func_code));
+
+	switch (start_ccb->ccb_h.func_code) {
+	case XPT_SCAN_BUS:
+		printf("NVME scan BUS started -- ignored\n");
+//		break;
+	case XPT_SCAN_TGT:
+		printf("NVME scan TGT started -- ignored\n");
+//		break;
+	case XPT_SCAN_LUN:
+		printf("NVME scan started\n");
+		nvme_scan_lun(start_ccb->ccb_h.path->periph,
+			      start_ccb->ccb_h.path, start_ccb->crcn.flags,
+			      start_ccb);
+		break;
+	case XPT_DEV_ADVINFO:
+		nvme_dev_advinfo(start_ccb);
+		break;
+
+	default:
+		xpt_action_default(start_ccb);
+		break;
+	}
+}
+
+/*
+ * Handle any per-device event notifications that require action by the XPT.
+ */
+static void
+nvme_dev_async(u_int32_t async_code, struct cam_eb *bus, struct cam_et *target,
+	      struct cam_ed *device, void *async_arg)
+{
+
+	/*
+	 * We only need to handle events for real devices.
+	 */
+	if (target->target_id == CAM_TARGET_WILDCARD
+	 || device->lun_id == CAM_LUN_WILDCARD)
+		return;
+
+	if (async_code == AC_LOST_DEVICE &&
+	    (device->flags & CAM_DEV_UNCONFIGURED) == 0) {
+		device->flags |= CAM_DEV_UNCONFIGURED;
+		xpt_release_device(device);
+	}
+}
+
+static void
+nvme_announce_periph(struct cam_periph *periph)
+{
+	struct	ccb_pathinq cpi;
+	struct	ccb_trans_settings cts;
+	struct	cam_path *path = periph->path;
+
+	cam_periph_assert(periph, MA_OWNED);
+
+	xpt_setup_ccb(&cts.ccb_h, path, CAM_PRIORITY_NORMAL);
+	cts.ccb_h.func_code = XPT_GET_TRAN_SETTINGS;
+	cts.type = CTS_TYPE_CURRENT_SETTINGS;
+	xpt_action((union ccb*)&cts);
+	if ((cts.ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP)
+		return;
+	/* Ask the SIM for its base transfer speed */
+	xpt_setup_ccb(&cpi.ccb_h, path, CAM_PRIORITY_NORMAL);
+	cpi.ccb_h.func_code = XPT_PATH_INQ;
+	xpt_action((union ccb *)&cpi);
+	/* XXX NVME STUFF HERE */
+	printf("\n");
+}
diff --git a/sys/conf/files b/sys/conf/files
index a364a62..49dca44 100644
--- a/sys/conf/files
+++ b/sys/conf/files
@@ -86,6 +86,9 @@ cam/cam_xpt.c			optional scbus
 cam/ata/ata_all.c		optional scbus
 cam/ata/ata_xpt.c		optional scbus
 cam/ata/ata_pmp.c		optional scbus
+cam/nvme/nvme_all.c		optional scbus nvme 
+cam/nvme/nvme_da.c		optional scbus nvme da
+cam/nvme/nvme_xpt.c		optional scbus nvme
 cam/scsi/scsi_xpt.c		optional scbus
 cam/scsi/scsi_all.c		optional scbus
 cam/scsi/scsi_cd.c		optional cd
diff --git a/sys/conf/files.amd64 b/sys/conf/files.amd64
index 53c299f..4949d71 100644
--- a/sys/conf/files.amd64
+++ b/sys/conf/files.amd64
@@ -290,6 +290,7 @@ dev/nvme/nvme_ctrlr_cmd.c	optional	nvme
 dev/nvme/nvme_ns.c		optional	nvme
 dev/nvme/nvme_ns_cmd.c		optional	nvme
 dev/nvme/nvme_qpair.c		optional	nvme
+dev/nvme/nvme_sim.c		optional	nvme scbus !nvd
 dev/nvme/nvme_sysctl.c		optional	nvme
 dev/nvme/nvme_test.c		optional	nvme
 dev/nvme/nvme_util.c		optional	nvme