Date: Mon, 21 Mar 2016 00:34:23 +0000 (UTC) From: Alexander Motin <mav@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-10@freebsd.org Subject: svn commit: r297126 - in stable/10/sys/dev: nvd nvme Message-ID: <201603210034.u2L0YN9s042029@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: mav Date: Mon Mar 21 00:34:22 2016 New Revision: 297126 URL: https://svnweb.freebsd.org/changeset/base/297126 Log: MFC r296617: Revert r292074 (by smh): Limit stripesize reported from nvd(4) to 4K I believe that this patch handled the problem from the wrong side. Instead of making ZFS properly handle large stripe sizes, it made unrelated driver to lie in reported parameters to workaround that. Alternative solution for this problem from ZFS side was committed at r296615. Modified: stable/10/sys/dev/nvd/nvd.c stable/10/sys/dev/nvme/nvme.h stable/10/sys/dev/nvme/nvme_ns.c stable/10/sys/dev/nvme/nvme_sysctl.c Directory Properties: stable/10/ (props changed) Modified: stable/10/sys/dev/nvd/nvd.c ============================================================================== --- stable/10/sys/dev/nvd/nvd.c Mon Mar 21 00:31:06 2016 (r297125) +++ stable/10/sys/dev/nvd/nvd.c Mon Mar 21 00:34:22 2016 (r297126) @@ -311,7 +311,7 @@ nvd_new_disk(struct nvme_namespace *ns, disk->d_delmaxsize = (off_t)nvme_ns_get_size(ns); if (disk->d_delmaxsize > nvd_delete_max) disk->d_delmaxsize = nvd_delete_max; - disk->d_stripesize = nvme_ns_get_optimal_sector_size(ns); + disk->d_stripesize = nvme_ns_get_stripesize(ns); if (TAILQ_EMPTY(&disk_head)) disk->d_unit = 0; Modified: stable/10/sys/dev/nvme/nvme.h ============================================================================== --- stable/10/sys/dev/nvme/nvme.h Mon Mar 21 00:31:06 2016 (r297125) +++ stable/10/sys/dev/nvme/nvme.h Mon Mar 21 00:34:22 2016 (r297126) @@ -870,7 +870,6 @@ const char * nvme_ns_get_serial_number(s const char * nvme_ns_get_model_number(struct nvme_namespace *ns); const struct nvme_namespace_data * nvme_ns_get_data(struct nvme_namespace *ns); -uint32_t nvme_ns_get_optimal_sector_size(struct nvme_namespace *ns); uint32_t nvme_ns_get_stripesize(struct nvme_namespace *ns); int nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp, Modified: stable/10/sys/dev/nvme/nvme_ns.c ============================================================================== --- stable/10/sys/dev/nvme/nvme_ns.c Mon Mar 21 00:31:06 2016 (r297125) +++ stable/10/sys/dev/nvme/nvme_ns.c Mon Mar 21 00:34:22 2016 (r297126) @@ -45,8 +45,6 @@ __FBSDID("$FreeBSD$"); #include "nvme_private.h" -extern int nvme_max_optimal_sectorsize; - static void nvme_bio_child_inbed(struct bio *parent, int bio_error); static void nvme_bio_child_done(void *arg, const struct nvme_completion *cpl); @@ -219,22 +217,6 @@ nvme_ns_get_stripesize(struct nvme_names return (ns->stripesize); } -uint32_t -nvme_ns_get_optimal_sector_size(struct nvme_namespace *ns) -{ - uint32_t stripesize; - - stripesize = nvme_ns_get_stripesize(ns); - - if (stripesize == 0) - return nvme_ns_get_sector_size(ns); - - if (nvme_max_optimal_sectorsize == 0) - return (stripesize); - - return (MIN(stripesize, nvme_max_optimal_sectorsize)); -} - static void nvme_ns_bio_done(void *arg, const struct nvme_completion *status) { Modified: stable/10/sys/dev/nvme/nvme_sysctl.c ============================================================================== --- stable/10/sys/dev/nvme/nvme_sysctl.c Mon Mar 21 00:31:06 2016 (r297125) +++ stable/10/sys/dev/nvme/nvme_sysctl.c Mon Mar 21 00:34:22 2016 (r297126) @@ -33,22 +33,6 @@ __FBSDID("$FreeBSD$"); #include "nvme_private.h" -SYSCTL_NODE(_kern, OID_AUTO, nvme, CTLFLAG_RD, 0, "NVM Express"); -/* - * Intel NVMe controllers have a slow path for I/Os that span a 128KB - * stripe boundary but ZFS limits ashift, which is derived from - * d_stripesize, to 13 (8KB) so we limit the stripesize reported to - * geom(8) to 4KB by default. - * - * This may result in a small number of additional I/Os to require - * splitting in nvme(4), however the NVMe I/O path is very efficient - * so these additional I/Os will cause very minimal (if any) difference - * in performance or CPU utilisation. - */ -int nvme_max_optimal_sectorsize = 1<<12; -SYSCTL_INT(_kern_nvme, OID_AUTO, max_optimal_sectorsize, CTLFLAG_RWTUN, - &nvme_max_optimal_sectorsize, 0, "The maximum optimal sectorsize reported"); - /* * CTLTYPE_S64 and sysctl_handle_64 were added in r217616. Define these * explicitly here for older kernels that don't include the r217616
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201603210034.u2L0YN9s042029>