aboutsummaryrefslogtreecommitdiffstats
path: root/sys/cam/nvme/nvme_da.c
diff options
context:
space:
mode:
authorWarner Losh <imp@FreeBSD.org>2016-06-09 22:39:02 +0000
committerWarner Losh <imp@FreeBSD.org>2016-06-09 22:39:02 +0000
commitbaabaca31be9676473d74dea3fbc90b32f0be35d (patch)
tree588e4979faf0585ef8448c0fdb718ede05c837f6 /sys/cam/nvme/nvme_da.c
parentc7310cbc6a1a1d47ea575e5f45eeb3efe31178d2 (diff)
downloadsrc-baabaca31be9676473d74dea3fbc90b32f0be35d.tar.gz
src-baabaca31be9676473d74dea3fbc90b32f0be35d.zip
New NVMe front end (nda).
Notes
Notes: svn path=/head/; revision=301771
Diffstat (limited to 'sys/cam/nvme/nvme_da.c')
-rw-r--r--sys/cam/nvme/nvme_da.c1152
1 files changed, 1152 insertions, 0 deletions
diff --git a/sys/cam/nvme/nvme_da.c b/sys/cam/nvme/nvme_da.c
new file mode 100644
index 000000000000..9628530edb20
--- /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();
+}