mm/hugetlb: do pre-HVO for bootmem allocated pages

For large systems, the overhead of vmemmap pages for hugetlb is
substantial.  It's about 1.5% of memory, which is about 45G for a 3T
system.  If you want to configure most of that system for hugetlb (e.g.
to use as backing memory for VMs), there is a chance of running out of
memory on boot, even though you know that the 45G will become available
later.

To avoid this scenario, and since it's a waste to first allocate and then
free that 45G during boot, do pre-HVO for hugetlb bootmem allocated pages
('gigantic' pages).

pre-HVO is done by adding functions that are called from
sparse_init_nid_early and sparse_init_nid_late.  The first is called
before memmap allocation, so it takes care of allocating memmap HVO-style.
The second verifies that all bootmem pages look good, specifically it
checks that they do not intersect with multiple zones.  This can only be
done from sparse_init_nid_late path, when zones have been initialized.

The hugetlb page size must be aligned to the section size, and aligned to
the size of memory described by the number of page structures contained in
one PMD (since pre-HVO is not prepared to split PMDs).  This should be
true for most 'gigantic' pages, it is for 1G pages on x86, where both of
these alignment requirements are 128M.

This will only have an effect if hugetlb_bootmem_alloc was called early in
boot.  If not, it won't do anything, and HVO for bootmem hugetlb pages
works as before.

Link: https://lkml.kernel.org/r/20250228182928.2645936-20-fvdl@google.com
Signed-off-by: Frank van der Linden <fvdl@google.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Joao Martins <joao.m.martins@oracle.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roman Gushchin (Cruise) <roman.gushchin@linux.dev>
Cc: Usama Arif <usamaarif642@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 3353d7775c975ecb16b78e7787a91a0b46858200..eb768f498dfd834d68e42d1e8d80465b208d7cf0 100644 (file)
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -687,6 +687,8 @@ struct huge_bootmem_page {
 #define HUGE_BOOTMEM_HVO               0x0001
 #define HUGE_BOOTMEM_ZONES_VALID       0x0002
 
+bool hugetlb_bootmem_page_zones_valid(int nid, struct huge_bootmem_page *m);
+
 int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
 int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
 void wait_for_freed_hugetlb_folios(void);

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 4ed9c7bcea42debaff53b4d5c1764a704d307dc2..c934e6cf80c2ecf568ab531d8d76355c0f64c50b 100644 (file)
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3219,7 +3219,18 @@ found:
         */
        memblock_reserved_mark_noinit(virt_to_phys((void *)m + PAGE_SIZE),
                huge_page_size(h) - PAGE_SIZE);
-       /* Put them into a private list first because mem_map is not up yet */
+
+       /*
+        * Put them into a private list first because mem_map is not up yet.
+        *
+        * For pre-HVO to work correctly, pages need to be on the list for
+        * the node they were actually allocated from. That node may be
+        * different in the case of fallback by memblock_alloc_try_nid_raw.
+        * So, extract the actual node first.
+        */
+       if (nid == NUMA_NO_NODE)
+               node = early_pfn_to_nid(PHYS_PFN(virt_to_phys(m)));
+
        INIT_LIST_HEAD(&m->list);
        list_add(&m->list, &huge_boot_pages[node]);
        m->hstate = h;
@@ -3314,8 +3325,8 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
        }
 }
 
-static bool __init hugetlb_bootmem_page_zones_valid(int nid,
-                                                   struct huge_bootmem_page *m)
+bool __init hugetlb_bootmem_page_zones_valid(int nid,
+                                            struct huge_bootmem_page *m)
 {
        unsigned long start_pfn;
        bool valid;

diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index be6b33ecbc8ece6eff5a87c837b5a89b57b5fa64..9a99dfa3c495831f88d5e2c2281bfadb08e00102 100644 (file)
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -743,6 +743,149 @@ void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h, struct list_head
        __hugetlb_vmemmap_optimize_folios(h, folio_list, true);
 }
 
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+
+/* Return true of a bootmem allocated HugeTLB page should be pre-HVO-ed */
+static bool vmemmap_should_optimize_bootmem_page(struct huge_bootmem_page *m)
+{
+       unsigned long section_size, psize, pmd_vmemmap_size;
+       phys_addr_t paddr;
+
+       if (!READ_ONCE(vmemmap_optimize_enabled))
+               return false;
+
+       if (!hugetlb_vmemmap_optimizable(m->hstate))
+               return false;
+
+       psize = huge_page_size(m->hstate);
+       paddr = virt_to_phys(m);
+
+       /*
+        * Pre-HVO only works if the bootmem huge page
+        * is aligned to the section size.
+        */
+       section_size = (1UL << PA_SECTION_SHIFT);
+       if (!IS_ALIGNED(paddr, section_size) ||
+           !IS_ALIGNED(psize, section_size))
+               return false;
+
+       /*
+        * The pre-HVO code does not deal with splitting PMDS,
+        * so the bootmem page must be aligned to the number
+        * of base pages that can be mapped with one vmemmap PMD.
+        */
+       pmd_vmemmap_size = (PMD_SIZE / (sizeof(struct page))) << PAGE_SHIFT;
+       if (!IS_ALIGNED(paddr, pmd_vmemmap_size) ||
+           !IS_ALIGNED(psize, pmd_vmemmap_size))
+               return false;
+
+       return true;
+}
+
+/*
+ * Initialize memmap section for a gigantic page, HVO-style.
+ */
+void __init hugetlb_vmemmap_init_early(int nid)
+{
+       unsigned long psize, paddr, section_size;
+       unsigned long ns, i, pnum, pfn, nr_pages;
+       unsigned long start, end;
+       struct huge_bootmem_page *m = NULL;
+       void *map;
+
+       /*
+        * Noting to do if bootmem pages were not allocated
+        * early in boot, or if HVO wasn't enabled in the
+        * first place.
+        */
+       if (!hugetlb_bootmem_allocated())
+               return;
+
+       if (!READ_ONCE(vmemmap_optimize_enabled))
+               return;
+
+       section_size = (1UL << PA_SECTION_SHIFT);
+
+       list_for_each_entry(m, &huge_boot_pages[nid], list) {
+               if (!vmemmap_should_optimize_bootmem_page(m))
+                       continue;
+
+               nr_pages = pages_per_huge_page(m->hstate);
+               psize = nr_pages << PAGE_SHIFT;
+               paddr = virt_to_phys(m);
+               pfn = PHYS_PFN(paddr);
+               map = pfn_to_page(pfn);
+               start = (unsigned long)map;
+               end = start + nr_pages * sizeof(struct page);
+
+               if (vmemmap_populate_hvo(start, end, nid,
+                                       HUGETLB_VMEMMAP_RESERVE_SIZE) < 0)
+                       continue;
+
+               memmap_boot_pages_add(HUGETLB_VMEMMAP_RESERVE_SIZE / PAGE_SIZE);
+
+               pnum = pfn_to_section_nr(pfn);
+               ns = psize / section_size;
+
+               for (i = 0; i < ns; i++) {
+                       sparse_init_early_section(nid, map, pnum,
+                                       SECTION_IS_VMEMMAP_PREINIT);
+                       map += section_map_size();
+                       pnum++;
+               }
+
+               m->flags |= HUGE_BOOTMEM_HVO;
+       }
+}
+
+void __init hugetlb_vmemmap_init_late(int nid)
+{
+       struct huge_bootmem_page *m, *tm;
+       unsigned long phys, nr_pages, start, end;
+       unsigned long pfn, nr_mmap;
+       struct hstate *h;
+       void *map;
+
+       if (!hugetlb_bootmem_allocated())
+               return;
+
+       if (!READ_ONCE(vmemmap_optimize_enabled))
+               return;
+
+       list_for_each_entry_safe(m, tm, &huge_boot_pages[nid], list) {
+               if (!(m->flags & HUGE_BOOTMEM_HVO))
+                       continue;
+
+               phys = virt_to_phys(m);
+               h = m->hstate;
+               pfn = PHYS_PFN(phys);
+               nr_pages = pages_per_huge_page(h);
+
+               if (!hugetlb_bootmem_page_zones_valid(nid, m)) {
+                       /*
+                        * Oops, the hugetlb page spans multiple zones.
+                        * Remove it from the list, and undo HVO.
+                        */
+                       list_del(&m->list);
+
+                       map = pfn_to_page(pfn);
+
+                       start = (unsigned long)map;
+                       end = start + nr_pages * sizeof(struct page);
+
+                       vmemmap_undo_hvo(start, end, nid,
+                                        HUGETLB_VMEMMAP_RESERVE_SIZE);
+                       nr_mmap = end - start - HUGETLB_VMEMMAP_RESERVE_SIZE;
+                       memmap_boot_pages_add(DIV_ROUND_UP(nr_mmap, PAGE_SIZE));
+
+                       memblock_phys_free(phys, huge_page_size(h));
+                       continue;
+               } else
+                       m->flags |= HUGE_BOOTMEM_ZONES_VALID;
+       }
+}
+#endif
+
 static const struct ctl_table hugetlb_vmemmap_sysctls[] = {
        {
                .procname       = "hugetlb_optimize_vmemmap",

diff --git a/mm/hugetlb_vmemmap.h b/mm/hugetlb_vmemmap.h
index 62d3d645a793c73ac1b0180168f674f21ce7932f..18b4908252153d5312e0c8c5ad7400d754e1e3cb 100644 (file)
--- a/mm/hugetlb_vmemmap.h
+++ b/mm/hugetlb_vmemmap.h
@@ -9,6 +9,8 @@
 #ifndef _LINUX_HUGETLB_VMEMMAP_H
 #define _LINUX_HUGETLB_VMEMMAP_H
 #include <linux/hugetlb.h>
+#include <linux/io.h>
+#include <linux/memblock.h>
 
 /*
  * Reserve one vmemmap page, all vmemmap addresses are mapped to it. See
@@ -25,6 +27,10 @@ long hugetlb_vmemmap_restore_folios(const struct hstate *h,
 void hugetlb_vmemmap_optimize_folio(const struct hstate *h, struct folio *folio);
 void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_list);
 void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h, struct list_head *folio_list);
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+void hugetlb_vmemmap_init_early(int nid);
+void hugetlb_vmemmap_init_late(int nid);
+#endif
 
 
 static inline unsigned int hugetlb_vmemmap_size(const struct hstate *h)
@@ -71,6 +77,14 @@ static inline void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h,
 {
 }
 
+static inline void hugetlb_vmemmap_init_early(int nid)
+{
+}
+
+static inline void hugetlb_vmemmap_init_late(int nid)
+{
+}
+
 static inline unsigned int hugetlb_vmemmap_optimizable_size(const struct hstate *h)
 {
        return 0;

diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index 8cc848c4b17c20c6b3120fcf89d090aae3e5f670..fd2ab5118e13dfc71ef39abf63cbc040fe597c76 100644 (file)
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -32,6 +32,8 @@
 #include <asm/pgalloc.h>
 #include <asm/tlbflush.h>
 
+#include "hugetlb_vmemmap.h"
+
 /*
  * Flags for vmemmap_populate_range and friends.
  */
@@ -594,6 +596,7 @@ struct page * __meminit __populate_section_memmap(unsigned long pfn,
  */
 void __init sparse_vmemmap_init_nid_early(int nid)
 {
+       hugetlb_vmemmap_init_early(nid);
 }
 
 /*
@@ -604,5 +607,6 @@ void __init sparse_vmemmap_init_nid_early(int nid)
  */
 void __init sparse_vmemmap_init_nid_late(int nid)
 {
+       hugetlb_vmemmap_init_late(nid);
 }
 #endif