History

mm/sparsemem: support sub-section hotplug

The libnvdimm sub-system has suffered a series of hacks and broken workarounds for the memory-hotplug implementation's awkward section-aligned (128MB) granularity. For example the following backtrace is emitted when attempting arch_add_memory() with physical address ranges that intersect 'System RAM' (RAM) with 'Persistent Memory' (PMEM) within a given section: # cat /proc/iomem | grep -A1 -B1 Persistent\ Memory 100000000-1ffffffff : System RAM 200000000-303ffffff : Persistent Memory (legacy) 304000000-43fffffff : System RAM 440000000-23ffffffff : Persistent Memory 2400000000-43bfffffff : Persistent Memory 2400000000-43bfffffff : namespace2.0 WARNING: CPU: 38 PID: 928 at arch/x86/mm/init_64.c:850 add_pages+0x5c/0x60 [..] RIP: 0010:add_pages+0x5c/0x60 [..] Call Trace: devm_memremap_pages+0x460/0x6e0 pmem_attach_disk+0x29e/0x680 [nd_pmem] ? nd_dax_probe+0xfc/0x120 [libnvdimm] nvdimm_bus_probe+0x66/0x160 [libnvdimm] It was discovered that the problem goes beyond RAM vs PMEM collisions as some platform produce PMEM vs PMEM collisions within a given section. The libnvdimm workaround for that case revealed that the libnvdimm section-alignment-padding implementation has been broken for a long while. A fix for that long-standing breakage introduces as many problems as it solves as it would require a backward-incompatible change to the namespace metadata interpretation. Instead of that dubious route [1], address the root problem in the memory-hotplug implementation. Note that EEXIST is no longer treated as success as that is how sparse_add_section() reports subsection collisions, it was also obviated by recent changes to perform the request_region() for 'System RAM' before arch_add_memory() in the add_memory() sequence. [1] https://lore.kernel.org/r/155000671719.348031.2347363160141119237.stgit@dwillia2-desk3.amr.corp.intel.com [osalvador@suse.de: fix deactivate_section for early sections] Link: http://lkml.kernel.org/r/20190715081549.32577-2-osalvador@suse.de Link: http://lkml.kernel.org/r/156092354368.979959.6232443923440952359.stgit@dwillia2-desk3.amr.corp.intel.com Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Oscar Salvador <osalvador@suse.de> Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64] Reviewed-by: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@suse.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Logan Gunthorpe <logang@deltatee.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: David Hildenbrand <david@redhat.com> Cc: Jane Chu <jane.chu@oracle.com> Cc: Jeff Moyer <jmoyer@redhat.com> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Toshi Kani <toshi.kani@hpe.com> Cc: Wei Yang <richardw.yang@linux.intel.com> Cc: Jason Gunthorpe <jgg@mellanox.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Dan Williams committed 7 years ago

Commit ba72b4c1 parent 7ea62164 files · +141 −96Browse files

Unified · Split

4 files
  1. include/linux/memory_hotplug.h +1 −1
  2. mm/memory_hotplug.c +4 −23
  3. mm/page_alloc.c +1 −1
  4. mm/sparse.c +135 −71
include/linux/memory_hotplug.h+1−1
@@ -348,7 +348,7 @@ extern void move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, · 347 unmodified lines
348348 extern bool is_memblock_offlined(struct memory_block *mem);
349349 extern int sparse_add_section(int nid, unsigned long pfn,
350350 unsigned long nr_pages, struct vmem_altmap *altmap);
351−extern void sparse_remove_one_section(struct mem_section *ms,
351+extern void sparse_remove_section(struct mem_section *ms,
352352 unsigned long pfn, unsigned long nr_pages,
353353 unsigned long map_offset, struct vmem_altmap *altmap);
354354 extern struct page *sparse_decode_mem_map(unsigned long coded_mem_map,
mm/memory_hotplug.c+4−23
@@ -252,18 +252,6 @@ void __init register_page_bootmem_info_node(struct pglist_data *pgdat) · 251 unmodified lines
252252 }
253253 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
254254
255−static int __meminit __add_section(int nid, unsigned long pfn,
256− unsigned long nr_pages, struct vmem_altmap *altmap)
257−{
258− int ret;
259−
260− if (pfn_valid(pfn))
261− return -EEXIST;
262−
263− ret = sparse_add_section(nid, pfn, nr_pages, altmap);
264− return ret < 0 ? ret : 0;
265−}
266−
267255 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
268256 const char *reason)
269257 {
@@ -327,18 +315,11 @@ int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, · 57 unmodified lines
327315
328316 pfns = min(nr_pages, PAGES_PER_SECTION
329317 - (pfn & ~PAGE_SECTION_MASK));
330− err = __add_section(nid, pfn, pfns, altmap);
318+ err = sparse_add_section(nid, pfn, pfns, altmap);
319+ if (err)
320+ break;
331321 pfn += pfns;
332322 nr_pages -= pfns;
333−
334− /*
335− * EEXIST is finally dealt with by ioresource collision
336− * check. see add_memory() => register_memory_resource()
337− * Warning will be printed if there is collision.
338− */
339− if (err && (err != -EEXIST))
340− break;
341− err = 0;
342323 cond_resched();
343324 }
344325 vmemmap_populate_print_last();
@@ -541,7 +522,7 @@ static void __remove_section(struct zone *zone, unsigned long pfn, · 196 unmodified lines
541522 return;
542523
543524 __remove_zone(zone, pfn, nr_pages);
544− sparse_remove_one_section(ms, pfn, nr_pages, map_offset, altmap);
525+ sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
545526 }
546527
547528 /**
mm/page_alloc.c+1−1
@@ -5974,7 +5974,7 @@ void __ref memmap_init_zone_device(struct zone *zone, · 5973 unmodified lines
59745974 * pfn out of zone.
59755975 *
59765976 * Please note that MEMMAP_HOTPLUG path doesn't clear memmap
5977− * because this is done early in sparse_add_one_section
5977+ * because this is done early in section_activate()
59785978 */
59795979 if (!(pfn & (pageblock_nr_pages - 1))) {
59805980 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
mm/sparse.c+135−71
@@ -83,8 +83,15 @@ static int __meminit sparse_index_init(unsigned long section_nr, int nid) · 82 unmodified lines
8383 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
8484 struct mem_section *section;
8585
86+ /*
87+ * An existing section is possible in the sub-section hotplug
88+ * case. First hot-add instantiates, follow-on hot-add reuses
89+ * the existing section.
90+ *
91+ * The mem_hotplug_lock resolves the apparent race below.
92+ */
8693 if (mem_section[root])
87− return -EEXIST;
94+ return 0;
8895
8996 section = sparse_index_alloc(nid);
9097 if (!section)
@@ -715,10 +722,120 @@ static void free_map_bootmem(struct page *memmap) · 624 unmodified lines
715722 }
716723 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
717724
725+static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
726+ struct vmem_altmap *altmap)
727+{
728+ DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
729+ DECLARE_BITMAP(tmp, SUBSECTIONS_PER_SECTION) = { 0 };
730+ struct mem_section *ms = __pfn_to_section(pfn);
731+ bool section_is_early = early_section(ms);
732+ struct page *memmap = NULL;
733+ unsigned long *subsection_map = ms->usage
734+ ? &ms->usage->subsection_map[0] : NULL;
735+
736+ subsection_mask_set(map, pfn, nr_pages);
737+ if (subsection_map)
738+ bitmap_and(tmp, map, subsection_map, SUBSECTIONS_PER_SECTION);
739+
740+ if (WARN(!subsection_map || !bitmap_equal(tmp, map, SUBSECTIONS_PER_SECTION),
741+ "section already deactivated (%#lx + %ld)\n",
742+ pfn, nr_pages))
743+ return;
744+
745+ /*
746+ * There are 3 cases to handle across two configurations
747+ * (SPARSEMEM_VMEMMAP={y,n}):
748+ *
749+ * 1/ deactivation of a partial hot-added section (only possible
750+ * in the SPARSEMEM_VMEMMAP=y case).
751+ * a/ section was present at memory init
752+ * b/ section was hot-added post memory init
753+ * 2/ deactivation of a complete hot-added section
754+ * 3/ deactivation of a complete section from memory init
755+ *
756+ * For 1/, when subsection_map does not empty we will not be
757+ * freeing the usage map, but still need to free the vmemmap
758+ * range.
759+ *
760+ * For 2/ and 3/ the SPARSEMEM_VMEMMAP={y,n} cases are unified
761+ */
762+ bitmap_xor(subsection_map, map, subsection_map, SUBSECTIONS_PER_SECTION);
763+ if (bitmap_empty(subsection_map, SUBSECTIONS_PER_SECTION)) {
764+ unsigned long section_nr = pfn_to_section_nr(pfn);
765+
766+ if (!section_is_early) {
767+ kfree(ms->usage);
768+ ms->usage = NULL;
769+ }
770+ memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
771+ ms->section_mem_map = sparse_encode_mem_map(NULL, section_nr);
772+ }
773+
774+ if (section_is_early && memmap)
775+ free_map_bootmem(memmap);
776+ else
777+ depopulate_section_memmap(pfn, nr_pages, altmap);
778+}
779+
780+static struct page * __meminit section_activate(int nid, unsigned long pfn,
781+ unsigned long nr_pages, struct vmem_altmap *altmap)
782+{
783+ DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
784+ struct mem_section *ms = __pfn_to_section(pfn);
785+ struct mem_section_usage *usage = NULL;
786+ unsigned long *subsection_map;
787+ struct page *memmap;
788+ int rc = 0;
789+
790+ subsection_mask_set(map, pfn, nr_pages);
791+
792+ if (!ms->usage) {
793+ usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
794+ if (!usage)
795+ return ERR_PTR(-ENOMEM);
796+ ms->usage = usage;
797+ }
798+ subsection_map = &ms->usage->subsection_map[0];
799+
800+ if (bitmap_empty(map, SUBSECTIONS_PER_SECTION))
801+ rc = -EINVAL;
802+ else if (bitmap_intersects(map, subsection_map, SUBSECTIONS_PER_SECTION))
803+ rc = -EEXIST;
804+ else
805+ bitmap_or(subsection_map, map, subsection_map,
806+ SUBSECTIONS_PER_SECTION);
807+
808+ if (rc) {
809+ if (usage)
810+ ms->usage = NULL;
811+ kfree(usage);
812+ return ERR_PTR(rc);
813+ }
814+
815+ /*
816+ * The early init code does not consider partially populated
817+ * initial sections, it simply assumes that memory will never be
818+ * referenced. If we hot-add memory into such a section then we
819+ * do not need to populate the memmap and can simply reuse what
820+ * is already there.
821+ */
822+ if (nr_pages < PAGES_PER_SECTION && early_section(ms))
823+ return pfn_to_page(pfn);
824+
825+ memmap = populate_section_memmap(pfn, nr_pages, nid, altmap);
826+ if (!memmap) {
827+ section_deactivate(pfn, nr_pages, altmap);
828+ return ERR_PTR(-ENOMEM);
829+ }
830+
831+ return memmap;
832+}
833+
718834 /**
719− * sparse_add_one_section - add a memory section
835+ * sparse_add_section - add a memory section, or populate an existing one
720836 * @nid: The node to add section on
721837 * @start_pfn: start pfn of the memory range
838+ * @nr_pages: number of pfns to add in the section
722839 * @altmap: device page map
723840 *
724841 * This is only intended for hotplug.
@@ -732,51 +849,34 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn, · 7 unmodified lines
732849 unsigned long nr_pages, struct vmem_altmap *altmap)
733850 {
734851 unsigned long section_nr = pfn_to_section_nr(start_pfn);
735− struct mem_section_usage *usage;
736852 struct mem_section *ms;
737853 struct page *memmap;
738854 int ret;
739855
740− /*
741− * no locking for this, because it does its own
742− * plus, it does a kmalloc
743− */
744856 ret = sparse_index_init(section_nr, nid);
745− if (ret < 0 && ret != -EEXIST)
857+ if (ret < 0)
746858 return ret;
747− ret = 0;
748− memmap = populate_section_memmap(start_pfn, PAGES_PER_SECTION, nid,
749− altmap);
750− if (!memmap)
751− return -ENOMEM;
752− usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
753− if (!usage) {
754− depopulate_section_memmap(start_pfn, PAGES_PER_SECTION, altmap);
755− return -ENOMEM;
756− }
757859
758− ms = __pfn_to_section(start_pfn);
759− if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
760− ret = -EEXIST;
761− goto out;
762− }
860+ memmap = section_activate(nid, start_pfn, nr_pages, altmap);
861+ if (IS_ERR(memmap))
862+ return PTR_ERR(memmap);
763863
764864 /*
765865 * Poison uninitialized struct pages in order to catch invalid flags
766866 * combinations.
767867 */
768− page_init_poison(memmap, sizeof(struct page) * PAGES_PER_SECTION);
868+ page_init_poison(pfn_to_page(start_pfn), sizeof(struct page) * nr_pages);
769869
870+ ms = __pfn_to_section(start_pfn);
770871 set_section_nid(section_nr, nid);
771872 section_mark_present(ms);
772− sparse_init_one_section(ms, section_nr, memmap, usage, 0);
773873
774−out:
775− if (ret < 0) {
776− kfree(usage);
777− depopulate_section_memmap(start_pfn, PAGES_PER_SECTION, altmap);
778− }
779− return ret;
874+ /* Align memmap to section boundary in the subsection case */
875+ if (section_nr_to_pfn(section_nr) != start_pfn)
876+ memmap = pfn_to_kaddr(section_nr_to_pfn(section_nr));
877+ sparse_init_one_section(ms, section_nr, memmap, ms->usage, 0);
878+
879+ return 0;
780880 }
781881
782882 #ifdef CONFIG_MEMORY_FAILURE
@@ -809,48 +909,12 @@ static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) · 26 unmodified lines
809909 }
810910 #endif
811911
812−static void free_section_usage(struct mem_section *ms, struct page *memmap,
813− struct mem_section_usage *usage, unsigned long pfn,
814− unsigned long nr_pages, struct vmem_altmap *altmap)
815−{
816− if (!usage)
817− return;
818−
819− /*
820− * Check to see if allocation came from hot-plug-add
821− */
822− if (!early_section(ms)) {
823− kfree(usage);
824− if (memmap)
825− depopulate_section_memmap(pfn, nr_pages, altmap);
826− return;
827− }
828−
829− /*
830− * The usemap came from bootmem. This is packed with other usemaps
831− * on the section which has pgdat at boot time. Just keep it as is now.
832− */
833−
834− if (memmap)
835− free_map_bootmem(memmap);
836−}
837−
838−void sparse_remove_one_section(struct mem_section *ms, unsigned long pfn,
912+void sparse_remove_section(struct mem_section *ms, unsigned long pfn,
839913 unsigned long nr_pages, unsigned long map_offset,
840914 struct vmem_altmap *altmap)
841915 {
842− struct page *memmap = NULL;
843− struct mem_section_usage *usage = NULL;
844−
845− if (ms->section_mem_map) {
846− usage = ms->usage;
847− memmap = sparse_decode_mem_map(ms->section_mem_map,
848− __section_nr(ms));
849− ms->section_mem_map = 0;
850− ms->usage = NULL;
851− }
852−
853− clear_hwpoisoned_pages(memmap + map_offset, nr_pages - map_offset);
854− free_section_usage(ms, memmap, usage, pfn, nr_pages, altmap);
916+ clear_hwpoisoned_pages(pfn_to_page(pfn) + map_offset,
917+ nr_pages - map_offset);
918+ section_deactivate(pfn, nr_pages, altmap);
855919 }
856920 #endif /* CONFIG_MEMORY_HOTPLUG */