svn commit: r252222 - in stable/9: etc/mtree include sbin sbin/nvmecontrol sys/amd64/conf sys/conf sys/dev/nvd sys/dev/nvme sys/i386/conf sys/modules sys/modules/nvme
Jim Harris
jim.harris at gmail.com
Tue Jun 25 23:59:19 UTC 2013
On Tue, Jun 25, 2013 at 4:52 PM, Jim Harris <jimharris at freebsd.org> wrote:
> Author: jimharris
> Date: Tue Jun 25 23:52:39 2013
> New Revision: 252222
> URL: http://svnweb.freebsd.org/changeset/base/252222
>
> Log:
> 240618, 240621, 240633, 240671, 240672, 240697, 240700, 241433,
> 241434, 241657, 241658, 241659, 241660, 241661, 241662, 241663,
> 241664, 241665, 241689, 242420, 243951, 244410, 244411, 244413,
> 244549, 245136, 247963, 248729, 248730, 248731, 248732, 248733,
> 248734, 248735, 248736, 248737, 248738, 248739, 248740, 248741,
> 248746, 248747, 248748, 248749, 248754, 248755, 248756, 248757,
> 248758, 248759, 248760, 248761, 248762, 248763, 248764, 248765,
> 248766, 248767, 248768, 248769, 248770, 248771, 248772, 248773,
> 248780, 248834, 248835, 248913, 248977, 249067, 249416, 249417,
> 249418, 249419, 249420, 249421, 249422, 249432
>
>
Bad commit message obviously. Following up with a forced commit for the
complete commit message.
> Deleted:
> stable/9/sys/dev/nvme/nvme_uio.c
> Modified:
> stable/9/etc/mtree/BSD.include.dist
> stable/9/include/Makefile
> stable/9/sbin/Makefile.amd64
> stable/9/sbin/Makefile.i386
> stable/9/sbin/nvmecontrol/nvmecontrol.8
> stable/9/sbin/nvmecontrol/nvmecontrol.c
> stable/9/sys/amd64/conf/NOTES
> stable/9/sys/conf/files.amd64
> stable/9/sys/conf/files.i386
> stable/9/sys/dev/nvd/nvd.c
> stable/9/sys/dev/nvme/nvme.c
> stable/9/sys/dev/nvme/nvme.h
> stable/9/sys/dev/nvme/nvme_ctrlr.c
> stable/9/sys/dev/nvme/nvme_ctrlr_cmd.c
> stable/9/sys/dev/nvme/nvme_ns.c
> stable/9/sys/dev/nvme/nvme_ns_cmd.c
> stable/9/sys/dev/nvme/nvme_private.h
> stable/9/sys/dev/nvme/nvme_qpair.c
> stable/9/sys/dev/nvme/nvme_sysctl.c
> stable/9/sys/dev/nvme/nvme_test.c
> stable/9/sys/i386/conf/NOTES
> stable/9/sys/modules/Makefile
> stable/9/sys/modules/nvme/Makefile
> Directory Properties:
> stable/9/etc/ (props changed)
> stable/9/etc/mtree/ (props changed)
> stable/9/include/ (props changed)
> stable/9/sbin/ (props changed)
> stable/9/sbin/nvmecontrol/ (props changed)
> stable/9/sys/ (props changed)
> stable/9/sys/conf/ (props changed)
> stable/9/sys/dev/ (props changed)
> stable/9/sys/modules/ (props changed)
>
> Modified: stable/9/etc/mtree/BSD.include.dist
>
> ==============================================================================
> --- stable/9/etc/mtree/BSD.include.dist Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/etc/mtree/BSD.include.dist Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -126,6 +126,8 @@
> mpilib
> ..
> ..
> + nvme
> + ..
> ofw
> ..
> pbio
>
> Modified: stable/9/include/Makefile
>
> ==============================================================================
> --- stable/9/include/Makefile Tue Jun 25 23:30:48 2013 (r252221)
> +++ stable/9/include/Makefile Tue Jun 25 23:52:39 2013 (r252222)
> @@ -44,8 +44,8 @@ LDIRS= bsm cam geom net net80211 netatal
> LSUBDIRS= cam/ata cam/scsi \
> dev/acpica dev/agp dev/an dev/bktr dev/ciss dev/filemon
> dev/firewire \
> dev/hwpmc \
> - dev/ic dev/iicbus ${_dev_ieee488} dev/io dev/lmc dev/mfi dev/ofw \
> - dev/pbio dev/pci ${_dev_powermac_nvram} dev/ppbus dev/smbus \
> + dev/ic dev/iicbus ${_dev_ieee488} dev/io dev/lmc dev/mfi dev/nvme \
> + dev/ofw dev/pbio dev/pci ${_dev_powermac_nvram} dev/ppbus
> dev/smbus \
> dev/speaker dev/usb dev/utopia dev/vkbd dev/wi \
> fs/devfs fs/fdescfs fs/fifofs fs/msdosfs fs/nfs fs/ntfs fs/nullfs \
> ${_fs_nwfs} fs/portalfs fs/procfs fs/smbfs fs/udf fs/unionfs \
>
> Modified: stable/9/sbin/Makefile.amd64
>
> ==============================================================================
> --- stable/9/sbin/Makefile.amd64 Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sbin/Makefile.amd64 Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -2,3 +2,4 @@
>
> SUBDIR += bsdlabel
> SUBDIR += fdisk
> +SUBDIR += nvmecontrol
>
> Modified: stable/9/sbin/Makefile.i386
>
> ==============================================================================
> --- stable/9/sbin/Makefile.i386 Tue Jun 25 23:30:48 2013 (r252221)
> +++ stable/9/sbin/Makefile.i386 Tue Jun 25 23:52:39 2013 (r252222)
> @@ -2,4 +2,5 @@
>
> SUBDIR += bsdlabel
> SUBDIR += fdisk
> +SUBDIR += nvmecontrol
> SUBDIR += sconfig
>
> Modified: stable/9/sbin/nvmecontrol/nvmecontrol.8
>
> ==============================================================================
> --- stable/9/sbin/nvmecontrol/nvmecontrol.8 Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sbin/nvmecontrol/nvmecontrol.8 Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -33,7 +33,7 @@
> .\"
> .\" $FreeBSD$
> .\"
> -.Dd September 17, 2012
> +.Dd March 26, 2013
> .Dt NVMECONTROL 8
> .Os
> .Sh NAME
> @@ -54,7 +54,10 @@
> .Op Fl p
> .Aq Fl s Ar size_in_bytes
> .Aq Fl t Ar time_in_sec
> -.Aq device id
> +.Aq namespace id
> +.Nm
> +.Ic reset
> +.Aq controller id
> .Sh DESCRIPTION
> NVM Express (NVMe) is a storage protocol standard, for SSDs and other
> high-speed storage devices over PCI Express.
> @@ -62,6 +65,7 @@ high-speed storage devices over PCI Expr
> .Dl nvmecontrol devlist
> .Pp
> Display a list of NVMe controllers and namespaces along with their device
> nodes.
> +.Pp
> .Dl nvmecontrol identify nvme0
> .Pp
> Display a human-readable summary of the nvme0 IDENTIFY_CONTROLLER data.
> @@ -77,6 +81,9 @@ Run a performance test on nvme0ns1 using
> thread will issue a single 512 byte read command. Results are printed to
> stdout when 30 seconds expires.
> .Pp
> +.Dl nvmecontrol reset nvme0
> +.Pp
> +Perform a controller-level reset of the nvme0 controller.
> .Sh AUTHORS
> .An -nosplit
> .Nm
>
> Modified: stable/9/sbin/nvmecontrol/nvmecontrol.c
>
> ==============================================================================
> --- stable/9/sbin/nvmecontrol/nvmecontrol.c Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sbin/nvmecontrol/nvmecontrol.c Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -56,6 +56,9 @@ __FBSDID("$FreeBSD$");
> " <-i intr|wait> [-f refthread] [-p]\n"
> \
> " <namespace id>\n"
>
> +#define RESET_USAGE
> \
> +" nvmecontrol reset <controller id>\n"
> +
> static void perftest_usage(void);
>
> static void
> @@ -64,6 +67,7 @@ usage(void)
> fprintf(stderr, "usage:\n");
> fprintf(stderr, DEVLIST_USAGE);
> fprintf(stderr, IDENTIFY_USAGE);
> + fprintf(stderr, RESET_USAGE);
> fprintf(stderr, PERFTEST_USAGE);
> exit(EX_USAGE);
> }
> @@ -206,6 +210,53 @@ ns_get_sector_size(struct nvme_namespace
> return (1 << nsdata->lbaf[0].lbads);
> }
>
> +static void
> +read_controller_data(int fd, struct nvme_controller_data *cdata)
> +{
> + struct nvme_pt_command pt;
> +
> + memset(&pt, 0, sizeof(pt));
> + pt.cmd.opc = NVME_OPC_IDENTIFY;
> + pt.cmd.cdw10 = 1;
> + pt.buf = cdata;
> + pt.len = sizeof(*cdata);
> + pt.is_read = 1;
> +
> + if (ioctl(fd, NVME_PASSTHROUGH_CMD, &pt) < 0) {
> + printf("Identify request failed. errno=%d (%s)\n",
> + errno, strerror(errno));
> + exit(EX_IOERR);
> + }
> +
> + if (nvme_completion_is_error(&pt.cpl)) {
> + printf("Passthrough command returned error.\n");
> + exit(EX_IOERR);
> + }
> +}
> +
> +static void
> +read_namespace_data(int fd, int nsid, struct nvme_namespace_data *nsdata)
> +{
> + struct nvme_pt_command pt;
> +
> + memset(&pt, 0, sizeof(pt));
> + pt.cmd.opc = NVME_OPC_IDENTIFY;
> + pt.cmd.nsid = nsid;
> + pt.buf = nsdata;
> + pt.len = sizeof(*nsdata);
> + pt.is_read = 1;
> +
> + if (ioctl(fd, NVME_PASSTHROUGH_CMD, &pt) < 0) {
> + printf("Identify request failed. errno=%d (%s)\n",
> + errno, strerror(errno));
> + exit(EX_IOERR);
> + }
> +
> + if (nvme_completion_is_error(&pt.cpl)) {
> + printf("Passthrough command returned error.\n");
> + exit(EX_IOERR);
> + }
> +}
>
> static void
> devlist(int argc, char *argv[])
> @@ -241,34 +292,18 @@ devlist(int argc, char *argv[])
>
> fd = open(path, O_RDWR);
> if (fd < 0) {
> - printf("Could not open %s.\n", path);
> + printf("Could not open %s. errno=%d (%s)\n", path,
> + errno, strerror(errno));
> exit_code = EX_NOPERM;
> continue;
> }
>
> - if (ioctl(fd, NVME_IDENTIFY_CONTROLLER, &cdata) == -1) {
> - printf("ioctl to %s failed.\n", path);
> - exit_code = EX_IOERR;
> - continue;
> - }
> -
> + read_controller_data(fd, &cdata);
> printf("%6s: %s\n", name, cdata.mn);
>
> for (i = 0; i < cdata.nn; i++) {
> sprintf(name, "nvme%dns%d", ctrlr, i+1);
> - sprintf(path, "/dev/%s", name);
> -
> - fd = open(path, O_RDWR);
> - if (fd < 0) {
> - printf("Could not open %s.\n", path);
> - exit_code = EX_NOPERM;
> - continue;
> - }
> - if (ioctl(fd, NVME_IDENTIFY_NAMESPACE, &nsdata) ==
> -1) {
> - printf("ioctl to %s failed.\n", path);
> - exit_code = EX_IOERR;
> - continue;
> - }
> + read_namespace_data(fd, i+1, &nsdata);
> printf(" %10s (%lldGB)\n",
> name,
> nsdata.nsze *
> @@ -307,21 +342,20 @@ identify_ctrlr(int argc, char *argv[])
>
> sprintf(path, "/dev/%s", argv[optind]);
>
> - if (stat(path, &devstat) != 0) {
> - printf("Invalid device node '%s'.\n", path);
> + if (stat(path, &devstat) < 0) {
> + printf("Invalid device node %s. errno=%d (%s)\n", path,
> errno,
> + strerror(errno));
> exit(EX_IOERR);
> }
>
> fd = open(path, O_RDWR);
> if (fd < 0) {
> - printf("Could not open %s.\n", path);
> + printf("Could not open %s. errno=%d (%s)\n", path, errno,
> + strerror(errno));
> exit(EX_NOPERM);
> }
>
> - if (ioctl(fd, NVME_IDENTIFY_CONTROLLER, &cdata) == -1) {
> - printf("ioctl to %s failed.\n", path);
> - exit(EX_IOERR);
> - }
> + read_controller_data(fd, &cdata);
>
> if (hexflag == 1) {
> if (verboseflag == 1)
> @@ -348,7 +382,8 @@ identify_ns(int argc, char *argv[])
> struct nvme_namespace_data nsdata;
> struct stat devstat;
> char path[64];
> - int ch, fd, hexflag = 0, hexlength;
> + char *nsloc;
> + int ch, fd, hexflag = 0, hexlength,
> nsid;
> int verboseflag = 0;
>
> while ((ch = getopt(argc, argv, "vx")) != -1) {
> @@ -364,23 +399,55 @@ identify_ns(int argc, char *argv[])
> }
> }
>
> + /*
> + * Check if the specified device node exists before continuing.
> + * This is a cleaner check for cases where the correct controller
> + * is specified, but an invalid namespace on that controller.
> + */
> sprintf(path, "/dev/%s", argv[optind]);
> + if (stat(path, &devstat) < 0) {
> + printf("Invalid device node %s. errno=%d (%s)\n", path,
> errno,
> + strerror(errno));
> + exit(EX_IOERR);
> + }
> +
> + nsloc = strstr(argv[optind], "ns");
> + if (nsloc == NULL) {
> + printf("Invalid namepsace %s.\n", argv[optind]);
> + exit(EX_IOERR);
> + }
> +
> + /*
> + * Pull the namespace id from the string. +2 skips past the "ns"
> part
> + * of the string.
> + */
> + nsid = strtol(nsloc + 2, NULL, 10);
> + if (nsid == 0 && errno != 0) {
> + printf("Invalid namespace ID %s.\n", argv[optind]);
> + exit(EX_IOERR);
> + }
>
> - if (stat(path, &devstat) != 0) {
> - printf("Invalid device node '%s'.\n", path);
> + /*
> + * We send IDENTIFY commands to the controller, not the namespace,
> + * since it is an admin cmd. So the path should only include the
> + * nvmeX part of the nvmeXnsY string.
> + */
> + sprintf(path, "/dev/");
> + strncat(path, argv[optind], nsloc - argv[optind]);
> + if (stat(path, &devstat) < 0) {
> + printf("Invalid device node %s. errno=%d (%s)\n", path,
> errno,
> + strerror(errno));
> exit(EX_IOERR);
> }
>
> fd = open(path, O_RDWR);
> if (fd < 0) {
> - printf("Could not open %s.\n", path);
> + printf("Could not open %s. errno=%d (%s)\n", path, errno,
> + strerror(errno));
> exit(EX_NOPERM);
> }
>
> - if (ioctl(fd, NVME_IDENTIFY_NAMESPACE, &nsdata) == -1) {
> - printf("ioctl to %s failed.\n", path);
> - exit(EX_IOERR);
> - }
> + read_namespace_data(fd, nsid, &nsdata);
>
> if (hexflag == 1) {
> if (verboseflag == 1)
> @@ -423,7 +490,7 @@ identify(int argc, char *argv[])
> optind = 1;
>
> /*
> - * If devicde node contains "ns", we consider it a namespace,
> + * If device node contains "ns", we consider it a namespace,
> * otherwise, consider it a controller.
> */
> if (strstr(target, "ns") == NULL)
> @@ -475,7 +542,7 @@ perftest(int argc, char *argv[])
> char path[64];
> u_long ioctl_cmd = NVME_IO_TEST;
> bool nflag, oflag, sflag, tflag;
> - int err, perthread = 0;
> + int perthread = 0;
>
> nflag = oflag = sflag = tflag = false;
> name = NULL;
> @@ -565,14 +632,14 @@ perftest(int argc, char *argv[])
>
> fd = open(path, O_RDWR);
> if (fd < 0) {
> - fprintf(stderr, "%s not valid device.\n", path);
> + fprintf(stderr, "%s not valid device. errno=%d (%s)\n",
> path,
> + errno, strerror(errno));
> perftest_usage();
> }
>
> - err = ioctl(fd, ioctl_cmd, &io_test);
> -
> - if (err) {
> - fprintf(stderr, "NVME_IO_TEST returned %d\n", errno);
> + if (ioctl(fd, ioctl_cmd, &io_test) < 0) {
> + fprintf(stderr, "NVME_IO_TEST failed. errno=%d (%s)\n",
> errno,
> + strerror(errno));
> exit(EX_IOERR);
> }
>
> @@ -580,6 +647,44 @@ perftest(int argc, char *argv[])
> exit(EX_OK);
> }
>
> +static void
> +reset_ctrlr(int argc, char *argv[])
> +{
> + struct stat devstat;
> + char path[64];
> + int ch, fd;
> +
> + while ((ch = getopt(argc, argv, "")) != -1) {
> + switch ((char)ch) {
> + default:
> + usage();
> + }
> + }
> +
> + sprintf(path, "/dev/%s", argv[optind]);
> +
> + if (stat(path, &devstat) < 0) {
> + printf("Invalid device node %s. errno=%d (%s)\n", path,
> errno,
> + strerror(errno));
> + exit(EX_IOERR);
> + }
> +
> + fd = open(path, O_RDWR);
> + if (fd < 0) {
> + printf("Could not open %s. errno=%d (%s)\n", path, errno,
> + strerror(errno));
> + exit(EX_NOPERM);
> + }
> +
> + if (ioctl(fd, NVME_RESET_CONTROLLER) < 0) {
> + printf("Reset request to %s failed. errno=%d (%s)\n", path,
> + errno, strerror(errno));
> + exit(EX_IOERR);
> + }
> +
> + exit(EX_OK);
> +}
> +
> int
> main(int argc, char *argv[])
> {
> @@ -593,6 +698,8 @@ main(int argc, char *argv[])
> identify(argc-1, &argv[1]);
> else if (strcmp(argv[1], "perftest") == 0)
> perftest(argc-1, &argv[1]);
> + else if (strcmp(argv[1], "reset") == 0)
> + reset_ctrlr(argc-1, &argv[1]);
>
> usage();
>
>
> Modified: stable/9/sys/amd64/conf/NOTES
>
> ==============================================================================
> --- stable/9/sys/amd64/conf/NOTES Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sys/amd64/conf/NOTES Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -433,6 +433,11 @@ device isci
> options ISCI_LOGGING # enable debugging in isci HAL
>
> #
> +# NVM Express (NVMe) support
> +device nvme # base NVMe driver
> +device nvd # expose NVMe namespaces as disks, depends on nvme
> +
> +#
> # SafeNet crypto driver: can be moved to the MI NOTES as soon as
> # it's tested on a big-endian machine
> #
>
> Modified: stable/9/sys/conf/files.amd64
>
> ==============================================================================
> --- stable/9/sys/conf/files.amd64 Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sys/conf/files.amd64 Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -213,7 +213,16 @@ dev/kbd/kbd.c optional
> atkbd | sc | uk
> dev/lindev/full.c optional lindev
> dev/lindev/lindev.c optional lindev
> dev/nfe/if_nfe.c optional nfe pci
> +dev/nvd/nvd.c optional nvd nvme
> dev/nve/if_nve.c optional nve pci
> +dev/nvme/nvme.c optional nvme
> +dev/nvme/nvme_ctrlr.c optional nvme
> +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_sysctl.c optional nvme
> +dev/nvme/nvme_test.c optional nvme
> dev/nvram/nvram.c optional nvram isa
> dev/random/ivy.c optional random rdrand_rng
> dev/random/nehemiah.c optional random padlock_rng
>
> Modified: stable/9/sys/conf/files.i386
>
> ==============================================================================
> --- stable/9/sys/conf/files.i386 Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sys/conf/files.i386 Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -222,7 +222,16 @@ dev/lindev/lindev.c optional lindev
> dev/mse/mse.c optional mse
> dev/mse/mse_isa.c optional mse isa
> dev/nfe/if_nfe.c optional nfe pci
> +dev/nvd/nvd.c optional nvd nvme
> dev/nve/if_nve.c optional nve pci
> +dev/nvme/nvme.c optional nvme
> +dev/nvme/nvme_ctrlr.c optional nvme
> +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_sysctl.c optional nvme
> +dev/nvme/nvme_test.c optional nvme
> dev/nvram/nvram.c optional nvram isa
> dev/pcf/pcf_isa.c optional pcf
> dev/random/ivy.c optional random rdrand_rng
>
> Modified: stable/9/sys/dev/nvd/nvd.c
>
> ==============================================================================
> --- stable/9/sys/dev/nvd/nvd.c Tue Jun 25 23:30:48 2013 (r252221)
> +++ stable/9/sys/dev/nvd/nvd.c Tue Jun 25 23:52:39 2013 (r252222)
> @@ -45,9 +45,12 @@ struct nvd_disk;
> static disk_ioctl_t nvd_ioctl;
> static disk_strategy_t nvd_strategy;
>
> -static void create_geom_disk(void *, struct nvme_namespace *ns);
> +static void *nvd_new_disk(struct nvme_namespace *ns, void *ctrlr);
> static void destroy_geom_disk(struct nvd_disk *ndisk);
>
> +static void *nvd_new_controller(struct nvme_controller *ctrlr);
> +static void nvd_controller_fail(void *ctrlr);
> +
> static int nvd_load(void);
> static void nvd_unload(void);
>
> @@ -67,10 +70,18 @@ struct nvd_disk {
>
> uint32_t cur_depth;
>
> - TAILQ_ENTRY(nvd_disk) tailq;
> + TAILQ_ENTRY(nvd_disk) global_tailq;
> + TAILQ_ENTRY(nvd_disk) ctrlr_tailq;
> +};
> +
> +struct nvd_controller {
> +
> + TAILQ_ENTRY(nvd_controller) tailq;
> + TAILQ_HEAD(, nvd_disk) disk_head;
> };
>
> -TAILQ_HEAD(, nvd_disk) nvd_head;
> +static TAILQ_HEAD(, nvd_controller) ctrlr_head;
> +static TAILQ_HEAD(disk_list, nvd_disk) disk_head;
>
> static int nvd_modevent(module_t mod, int type, void *arg)
> {
> @@ -104,8 +115,11 @@ static int
> nvd_load()
> {
>
> - TAILQ_INIT(&nvd_head);
> - consumer_handle = nvme_register_consumer(create_geom_disk, NULL);
> + TAILQ_INIT(&ctrlr_head);
> + TAILQ_INIT(&disk_head);
> +
> + consumer_handle = nvme_register_consumer(nvd_new_disk,
> + nvd_new_controller, NULL, nvd_controller_fail);
>
> return (consumer_handle != NULL ? 0 : -1);
> }
> @@ -113,13 +127,20 @@ nvd_load()
> static void
> nvd_unload()
> {
> - struct nvd_disk *nvd;
> + struct nvd_controller *ctrlr;
> + struct nvd_disk *disk;
> +
> + while (!TAILQ_EMPTY(&ctrlr_head)) {
> + ctrlr = TAILQ_FIRST(&ctrlr_head);
> + TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
> + free(ctrlr, M_NVD);
> + }
>
> - while (!TAILQ_EMPTY(&nvd_head)) {
> - nvd = TAILQ_FIRST(&nvd_head);
> - TAILQ_REMOVE(&nvd_head, nvd, tailq);
> - destroy_geom_disk(nvd);
> - free(nvd, M_NVD);
> + while (!TAILQ_EMPTY(&disk_head)) {
> + disk = TAILQ_FIRST(&disk_head);
> + TAILQ_REMOVE(&disk_head, disk, global_tailq);
> + destroy_geom_disk(disk);
> + free(disk, M_NVD);
> }
>
> nvme_unregister_consumer(consumer_handle);
> @@ -153,7 +174,7 @@ nvd_ioctl(struct disk *ndisk, u_long cmd
> }
>
> static void
> -nvd_done(void *arg, const struct nvme_completion *status)
> +nvd_done(void *arg, const struct nvme_completion *cpl)
> {
> struct bio *bp;
> struct nvd_disk *ndisk;
> @@ -162,14 +183,13 @@ nvd_done(void *arg, const struct nvme_co
>
> ndisk = bp->bio_disk->d_drv1;
>
> - if (atomic_fetchadd_int(&ndisk->cur_depth, -1) == NVME_QD)
> - taskqueue_enqueue(ndisk->tq, &ndisk->bioqtask);
> + atomic_add_int(&ndisk->cur_depth, -1);
>
> /*
> * TODO: add more extensive translation of NVMe status codes
> * to different bio error codes (i.e. EIO, EINVAL, etc.)
> */
> - if (status->sf_sc || status->sf_sct) {
> + if (nvme_completion_is_error(cpl)) {
> bp->bio_error = EIO;
> bp->bio_flags |= BIO_ERROR;
> bp->bio_resid = bp->bio_bcount;
> @@ -187,9 +207,6 @@ nvd_bioq_process(void *arg, int pending)
> int err;
>
> for (;;) {
> - if (atomic_load_acq_int(&ndisk->cur_depth) >= NVME_QD)
> - break;
> -
> mtx_lock(&ndisk->bioqlock);
> bp = bioq_takefirst(&ndisk->bioq);
> mtx_unlock(&ndisk->bioqlock);
> @@ -210,13 +227,13 @@ nvd_bioq_process(void *arg, int pending)
> #endif
>
> bp->bio_driver1 = NULL;
> - atomic_add_acq_int(&ndisk->cur_depth, 1);
> + atomic_add_int(&ndisk->cur_depth, 1);
>
> err = nvme_ns_bio_process(ndisk->ns, bp, nvd_done);
>
> if (err) {
> - atomic_add_acq_int(&ndisk->cur_depth, -1);
> - bp->bio_error = EIO;
> + atomic_add_int(&ndisk->cur_depth, -1);
> + bp->bio_error = err;
> bp->bio_flags |= BIO_ERROR;
> bp->bio_resid = bp->bio_bcount;
> biodone(bp);
> @@ -237,13 +254,28 @@ nvd_bioq_process(void *arg, int pending)
> }
> }
>
> -static void
> -create_geom_disk(void *arg, struct nvme_namespace *ns)
> +static void *
> +nvd_new_controller(struct nvme_controller *ctrlr)
> {
> - struct nvd_disk *ndisk;
> - struct disk *disk;
> + struct nvd_controller *nvd_ctrlr;
> +
> + nvd_ctrlr = malloc(sizeof(struct nvd_controller), M_NVD,
> + M_ZERO | M_WAITOK);
>
> - ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_NOWAIT);
> + TAILQ_INIT(&nvd_ctrlr->disk_head);
> + TAILQ_INSERT_TAIL(&ctrlr_head, nvd_ctrlr, tailq);
> +
> + return (nvd_ctrlr);
> +}
> +
> +static void *
> +nvd_new_disk(struct nvme_namespace *ns, void *ctrlr_arg)
> +{
> + struct nvd_disk *ndisk;
> + struct disk *disk;
> + struct nvd_controller *ctrlr = ctrlr_arg;
> +
> + ndisk = malloc(sizeof(struct nvd_disk), M_NVD, M_ZERO | M_WAITOK);
>
> disk = disk_alloc();
> disk->d_strategy = nvd_strategy;
> @@ -255,10 +287,11 @@ create_geom_disk(void *arg, struct nvme_
> disk->d_sectorsize = nvme_ns_get_sector_size(ns);
> disk->d_mediasize = (off_t)nvme_ns_get_size(ns);
>
> - if (TAILQ_EMPTY(&nvd_head))
> + if (TAILQ_EMPTY(&disk_head))
> disk->d_unit = 0;
> else
> - disk->d_unit = TAILQ_FIRST(&nvd_head)->disk->d_unit + 1;
> + disk->d_unit =
> + TAILQ_LAST(&disk_head, disk_list)->disk->d_unit + 1;
>
> disk->d_flags = 0;
>
> @@ -268,6 +301,11 @@ create_geom_disk(void *arg, struct nvme_
> if (nvme_ns_get_flags(ns) & NVME_NS_FLUSH_SUPPORTED)
> disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
>
> +/* ifdef used here to ease porting to stable branches at a later point. */
> +#ifdef DISKFLAG_UNMAPPED_BIO
> + disk->d_flags |= DISKFLAG_UNMAPPED_BIO;
> +#endif
> +
> strlcpy(disk->d_ident, nvme_ns_get_serial_number(ns),
> sizeof(disk->d_ident));
>
> @@ -290,7 +328,10 @@ create_geom_disk(void *arg, struct nvme_
> taskqueue_thread_enqueue, &ndisk->tq);
> taskqueue_start_threads(&ndisk->tq, 1, PI_DISK, "nvd taskq");
>
> - TAILQ_INSERT_HEAD(&nvd_head, ndisk, tailq);
> + TAILQ_INSERT_TAIL(&disk_head, ndisk, global_tailq);
> + TAILQ_INSERT_TAIL(&ctrlr->disk_head, ndisk, ctrlr_tailq);
> +
> + return (NULL);
> }
>
> static void
> @@ -316,3 +357,22 @@ destroy_geom_disk(struct nvd_disk *ndisk
>
> mtx_destroy(&ndisk->bioqlock);
> }
> +
> +static void
> +nvd_controller_fail(void *ctrlr_arg)
> +{
> + struct nvd_controller *ctrlr = ctrlr_arg;
> + struct nvd_disk *disk;
> +
> + while (!TAILQ_EMPTY(&ctrlr->disk_head)) {
> + disk = TAILQ_FIRST(&ctrlr->disk_head);
> + TAILQ_REMOVE(&disk_head, disk, global_tailq);
> + TAILQ_REMOVE(&ctrlr->disk_head, disk, ctrlr_tailq);
> + destroy_geom_disk(disk);
> + free(disk, M_NVD);
> + }
> +
> + TAILQ_REMOVE(&ctrlr_head, ctrlr, tailq);
> + free(ctrlr, M_NVD);
> +}
> +
>
> Modified: stable/9/sys/dev/nvme/nvme.c
>
> ==============================================================================
> --- stable/9/sys/dev/nvme/nvme.c Tue Jun 25 23:30:48 2013
> (r252221)
> +++ stable/9/sys/dev/nvme/nvme.c Tue Jun 25 23:52:39 2013
> (r252222)
> @@ -32,22 +32,33 @@ __FBSDID("$FreeBSD$");
> #include <sys/conf.h>
> #include <sys/module.h>
>
> +#include <vm/uma.h>
> +
> +#include <dev/pci/pcireg.h>
> #include <dev/pci/pcivar.h>
>
> #include "nvme_private.h"
>
> struct nvme_consumer {
> - nvme_consumer_cb_fn_t cb_fn;
> - void *cb_arg;
> + uint32_t id;
> + nvme_cons_ns_fn_t ns_fn;
> + nvme_cons_ctrlr_fn_t ctrlr_fn;
> + nvme_cons_async_fn_t async_fn;
> + nvme_cons_fail_fn_t fail_fn;
> };
>
> struct nvme_consumer nvme_consumer[NVME_MAX_CONSUMERS];
> +#define INVALID_CONSUMER_ID 0xFFFF
> +
> +uma_zone_t nvme_request_zone;
> +int32_t nvme_retry_count;
>
> MALLOC_DEFINE(M_NVME, "nvme", "nvme(4) memory allocations");
>
> static int nvme_probe(device_t);
> static int nvme_attach(device_t);
> static int nvme_detach(device_t);
> +static int nvme_modevent(module_t mod, int type, void *arg);
>
> static devclass_t nvme_devclass;
>
> @@ -65,7 +76,7 @@ static driver_t nvme_pci_driver = {
> sizeof(struct nvme_controller),
> };
>
> -DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, 0, 0);
> +DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, nvme_modevent,
> 0);
> MODULE_VERSION(nvme, 1);
>
> static struct _pcsid
> @@ -75,15 +86,19 @@ static struct _pcsid
> } pci_ids[] = {
> { 0x01118086, "NVMe Controller" },
> { CHATHAM_PCI_ID, "Chatham Prototype NVMe Controller" },
> - { IDT_PCI_ID, "IDT NVMe Controller" },
> + { IDT32_PCI_ID, "IDT NVMe Controller (32 channel)" },
> + { IDT8_PCI_ID, "IDT NVMe Controller (8 channel)" },
> { 0x00000000, NULL }
> };
>
> static int
> nvme_probe (device_t device)
> {
> - u_int32_t type = pci_get_devid(device);
> - struct _pcsid *ep = pci_ids;
> + struct _pcsid *ep;
> + u_int32_t type;
> +
> + type = pci_get_devid(device);
> + ep = pci_ids;
>
> while (ep->type && ep->type != type)
> ++ep;
> @@ -91,11 +106,43 @@ nvme_probe (device_t device)
> if (ep->desc) {
> device_set_desc(device, ep->desc);
> return (BUS_PROBE_DEFAULT);
> - } else
> - return (ENXIO);
> + }
> +
> +#if defined(PCIS_STORAGE_NVM)
> + if (pci_get_class(device) == PCIC_STORAGE &&
> + pci_get_subclass(device) == PCIS_STORAGE_NVM &&
> + pci_get_progif(device) ==
> PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) {
> + device_set_desc(device, "Generic NVMe Device");
> + return (BUS_PROBE_GENERIC);
> + }
> +#endif
> +
> + return (ENXIO);
> }
>
> static void
> +nvme_init(void)
> +{
> + uint32_t i;
> +
> + nvme_request_zone = uma_zcreate("nvme_request",
> + sizeof(struct nvme_request), NULL, NULL, NULL, NULL, 0, 0);
> +
> + for (i = 0; i < NVME_MAX_CONSUMERS; i++)
> + nvme_consumer[i].id = INVALID_CONSUMER_ID;
> +}
> +
> +SYSINIT(nvme_register, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_init, NULL);
> +
> +static void
> +nvme_uninit(void)
> +{
> + uma_zdestroy(nvme_request_zone);
> +}
> +
> +SYSUNINIT(nvme_unregister, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_uninit,
> NULL);
> +
> +static void
> nvme_load(void)
> {
> }
> @@ -160,24 +207,14 @@ nvme_modevent(module_t mod, int type, vo
> return (0);
> }
>
> -moduledata_t nvme_mod = {
> - "nvme",
> - (modeventhand_t)nvme_modevent,
> - 0
> -};
> -
> -DECLARE_MODULE(nvme, nvme_mod, SI_SUB_DRIVERS, SI_ORDER_FIRST);
> -
> void
> nvme_dump_command(struct nvme_command *cmd)
> {
> - printf("opc:%x f:%x r1:%x cid:%x nsid:%x r2:%x r3:%x "
> - "mptr:%qx prp1:%qx prp2:%qx cdw:%x %x %x %x %x %x\n",
> + printf(
> +"opc:%x f:%x r1:%x cid:%x nsid:%x r2:%x r3:%x mptr:%jx prp1:%jx prp2:%jx
> cdw:%x %x %x %x %x %x\n",
> cmd->opc, cmd->fuse, cmd->rsvd1, cmd->cid, cmd->nsid,
> cmd->rsvd2, cmd->rsvd3,
> - (long long unsigned int)cmd->mptr,
> - (long long unsigned int)cmd->prp1,
> - (long long unsigned int)cmd->prp2,
> + (uintmax_t)cmd->mptr, (uintmax_t)cmd->prp1,
> (uintmax_t)cmd->prp2,
> cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14,
> cmd->cdw15);
> }
> @@ -188,87 +225,8 @@ nvme_dump_completion(struct nvme_complet
> printf("cdw0:%08x sqhd:%04x sqid:%04x "
> "cid:%04x p:%x sc:%02x sct:%x m:%x dnr:%x\n",
> cpl->cdw0, cpl->sqhd, cpl->sqid,
> - cpl->cid, cpl->p, cpl->sf_sc, cpl->sf_sct, cpl->sf_m,
> - cpl->sf_dnr);
> -}
> -
> -void
> -nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
> -{
> - struct nvme_tracker *tr;
> - struct nvme_qpair *qpair;
> - struct nvme_prp_list *prp_list;
> - uint32_t cur_nseg;
> -
> - KASSERT(error == 0, ("nvme_payload_map error != 0\n"));
> -
> - tr = (struct nvme_tracker *)arg;
> - qpair = tr->qpair;
> -
> - /*
> - * Note that we specified PAGE_SIZE for alignment and max
> - * segment size when creating the bus dma tags. So here
> - * we can safely just transfer each segment to its
> - * associated PRP entry.
> - */
> - tr->cmd.prp1 = seg[0].ds_addr;
> -
> - if (nseg == 2) {
> - tr->cmd.prp2 = seg[1].ds_addr;
> - } else if (nseg > 2) {
> - KASSERT(tr->prp_list,
> - ("prp_list needed but not attached to tracker\n"));
> - cur_nseg = 1;
> - prp_list = tr->prp_list;
> - tr->cmd.prp2 = (uint64_t)prp_list->bus_addr;
> - while (cur_nseg < nseg) {
> - prp_list->prp[cur_nseg-1] =
> - (uint64_t)seg[cur_nseg].ds_addr;
> - cur_nseg++;
> - }
> - }
> -
> - nvme_qpair_submit_cmd(qpair, tr);
> -}
> -
> -struct nvme_tracker *
> -nvme_allocate_tracker(struct nvme_controller *ctrlr, boolean_t is_admin,
> - nvme_cb_fn_t cb_fn, void *cb_arg, uint32_t payload_size, void
> *payload)
> -{
> - struct nvme_tracker *tr;
> - struct nvme_qpair *qpair;
> - uint32_t modulo, offset, num_prps;
> - boolean_t alloc_prp_list = FALSE;
> -
> - if (is_admin) {
> - qpair = &ctrlr->adminq;
> - } else {
> - if (ctrlr->per_cpu_io_queues)
> - qpair = &ctrlr->ioq[curcpu];
> - else
> - qpair = &ctrlr->ioq[0];
> - }
> -
> - num_prps = payload_size / PAGE_SIZE;
> - modulo = payload_size % PAGE_SIZE;
> - offset = (uint32_t)((uintptr_t)payload % PAGE_SIZE);
> -
> - if (modulo || offset)
> - num_prps += 1 + (modulo + offset - 1) / PAGE_SIZE;
> -
> - if (num_prps > 2)
> - alloc_prp_list = TRUE;
> -
> - tr = nvme_qpair_allocate_tracker(qpair, alloc_prp_list);
> -
> - memset(&tr->cmd, 0, sizeof(tr->cmd));
> -
> - tr->qpair = qpair;
> - tr->cb_fn = cb_fn;
> - tr->cb_arg = cb_arg;
> - tr->payload_size = payload_size;
> -
> - return (tr);
> + cpl->cid, cpl->status.p, cpl->status.sc, cpl->status.sct,
> + cpl->status.m, cpl->status.dnr);
> }
>
> static int
> @@ -287,15 +245,17 @@ nvme_attach(device_t dev)
> * to cc.en==0. This is because we don't really know what status
> * the controller was left in when boot handed off to OS.
> */
> - status = nvme_ctrlr_reset(ctrlr);
> + status = nvme_ctrlr_hw_reset(ctrlr);
> if (status != 0)
> return (status);
>
> - status = nvme_ctrlr_reset(ctrlr);
> + status = nvme_ctrlr_hw_reset(ctrlr);
> if (status != 0)
> return (status);
>
> - ctrlr->config_hook.ich_func = nvme_ctrlr_start;
> + nvme_sysctl_initialize_ctrlr(ctrlr);
> +
> + ctrlr->config_hook.ich_func = nvme_ctrlr_start_config_hook;
> ctrlr->config_hook.ich_arg = ctrlr;
>
> config_intrhook_establish(&ctrlr->config_hook);
> @@ -307,77 +267,75 @@ static int
> nvme_detach (device_t dev)
> {
> struct nvme_controller *ctrlr = DEVICE2SOFTC(dev);
> - struct nvme_namespace *ns;
> - int i;
> -
> - if (ctrlr->taskqueue) {
> - taskqueue_drain(ctrlr->taskqueue, &ctrlr->task);
> - taskqueue_free(ctrlr->taskqueue);
> - }
> -
> - for (i = 0; i < NVME_MAX_NAMESPACES; i++) {
> - ns = &ctrlr->ns[i];
> - if (ns->cdev)
> - destroy_dev(ns->cdev);
> - }
> -
> - if (ctrlr->cdev)
> - destroy_dev(ctrlr->cdev);
> -
> - for (i = 0; i < ctrlr->num_io_queues; i++) {
> - nvme_io_qpair_destroy(&ctrlr->ioq[i]);
> - }
> -
> - free(ctrlr->ioq, M_NVME);
> -
> - nvme_admin_qpair_destroy(&ctrlr->adminq);
> -
> - if (ctrlr->resource != NULL) {
> - bus_release_resource(dev, SYS_RES_MEMORY,
> - ctrlr->resource_id, ctrlr->resource);
> - }
> -
> -#ifdef CHATHAM2
> - if (ctrlr->chatham_resource != NULL) {
> - bus_release_resource(dev, SYS_RES_MEMORY,
> - ctrlr->chatham_resource_id, ctrlr->chatham_resource);
> - }
> -#endif
> -
> - if (ctrlr->tag)
> - bus_teardown_intr(ctrlr->dev, ctrlr->res, ctrlr->tag);
> -
> - if (ctrlr->res)
> - bus_release_resource(ctrlr->dev, SYS_RES_IRQ,
> - rman_get_rid(ctrlr->res), ctrlr->res);
> -
> - if (ctrlr->msix_enabled)
> - pci_release_msi(dev);
>
> + nvme_ctrlr_destruct(ctrlr, dev);
> return (0);
> }
>
> static void
> -nvme_notify_consumer(struct nvme_consumer *consumer)
> +nvme_notify_consumer(struct nvme_consumer *cons)
> {
> device_t *devlist;
> struct nvme_controller *ctrlr;
> - int dev, ns, devcount;
> + struct nvme_namespace *ns;
> + void *ctrlr_cookie;
> + int dev_idx, ns_idx, devcount;
>
> if (devclass_get_devices(nvme_devclass, &devlist, &devcount))
> return;
>
> - for (dev = 0; dev < devcount; dev++) {
> - ctrlr = DEVICE2SOFTC(devlist[dev]);
> - for (ns = 0; ns < ctrlr->cdata.nn; ns++)
> - (*consumer->cb_fn)(consumer->cb_arg,
> &ctrlr->ns[ns]);
> + for (dev_idx = 0; dev_idx < devcount; dev_idx++) {
> + ctrlr = DEVICE2SOFTC(devlist[dev_idx]);
> + if (cons->ctrlr_fn != NULL)
> + ctrlr_cookie = (*cons->ctrlr_fn)(ctrlr);
> + else
> + ctrlr_cookie = NULL;
> + ctrlr->cons_cookie[cons->id] = ctrlr_cookie;
> + for (ns_idx = 0; ns_idx < ctrlr->cdata.nn; ns_idx++) {
>
> *** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
>
More information about the svn-src-all
mailing list