Merge tag 'hwmon-for-v6.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux.git] / mm / sparse-vmemmap.c
blobcec67c5f37d884a248032de8878de3ba5f3925a3
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Virtual Memory Map support
5 * (C) 2007 sgi. Christoph Lameter.
7 * Virtual memory maps allow VM primitives pfn_to_page, page_to_pfn,
8 * virt_to_page, page_address() to be implemented as a base offset
9 * calculation without memory access.
11 * However, virtual mappings need a page table and TLBs. Many Linux
12 * architectures already map their physical space using 1-1 mappings
13 * via TLBs. For those arches the virtual memory map is essentially
14 * for free if we use the same page size as the 1-1 mappings. In that
15 * case the overhead consists of a few additional pages that are
16 * allocated to create a view of memory for vmemmap.
18 * The architecture is expected to provide a vmemmap_populate() function
19 * to instantiate the mapping.
21 #include <linux/mm.h>
22 #include <linux/mmzone.h>
23 #include <linux/memblock.h>
24 #include <linux/memremap.h>
25 #include <linux/highmem.h>
26 #include <linux/slab.h>
27 #include <linux/spinlock.h>
28 #include <linux/vmalloc.h>
29 #include <linux/sched.h>
31 #include <asm/dma.h>
32 #include <asm/pgalloc.h>
35 * Allocate a block of memory to be used to back the virtual memory map
36 * or to back the page tables that are used to create the mapping.
37 * Uses the main allocators if they are available, else bootmem.
40 static void * __ref __earlyonly_bootmem_alloc(int node,
41 unsigned long size,
42 unsigned long align,
43 unsigned long goal)
45 return memblock_alloc_try_nid_raw(size, align, goal,
46 MEMBLOCK_ALLOC_ACCESSIBLE, node);
49 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
51 /* If the main allocator is up use that, fallback to bootmem. */
52 if (slab_is_available()) {
53 gfp_t gfp_mask = GFP_KERNEL|__GFP_RETRY_MAYFAIL|__GFP_NOWARN;
54 int order = get_order(size);
55 static bool warned;
56 struct page *page;
58 page = alloc_pages_node(node, gfp_mask, order);
59 if (page)
60 return page_address(page);
62 if (!warned) {
63 warn_alloc(gfp_mask & ~__GFP_NOWARN, NULL,
64 "vmemmap alloc failure: order:%u", order);
65 warned = true;
67 return NULL;
68 } else
69 return __earlyonly_bootmem_alloc(node, size, size,
70 __pa(MAX_DMA_ADDRESS));
73 static void * __meminit altmap_alloc_block_buf(unsigned long size,
74 struct vmem_altmap *altmap);
76 /* need to make sure size is all the same during early stage */
77 void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node,
78 struct vmem_altmap *altmap)
80 void *ptr;
82 if (altmap)
83 return altmap_alloc_block_buf(size, altmap);
85 ptr = sparse_buffer_alloc(size);
86 if (!ptr)
87 ptr = vmemmap_alloc_block(size, node);
88 return ptr;
91 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
93 return altmap->base_pfn + altmap->reserve + altmap->alloc
94 + altmap->align;
97 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
99 unsigned long allocated = altmap->alloc + altmap->align;
101 if (altmap->free > allocated)
102 return altmap->free - allocated;
103 return 0;
106 static void * __meminit altmap_alloc_block_buf(unsigned long size,
107 struct vmem_altmap *altmap)
109 unsigned long pfn, nr_pfns, nr_align;
111 if (size & ~PAGE_MASK) {
112 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
113 __func__, size);
114 return NULL;
117 pfn = vmem_altmap_next_pfn(altmap);
118 nr_pfns = size >> PAGE_SHIFT;
119 nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
120 nr_align = ALIGN(pfn, nr_align) - pfn;
121 if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
122 return NULL;
124 altmap->alloc += nr_pfns;
125 altmap->align += nr_align;
126 pfn += nr_align;
128 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
129 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
130 return __va(__pfn_to_phys(pfn));
133 void __meminit vmemmap_verify(pte_t *pte, int node,
134 unsigned long start, unsigned long end)
136 unsigned long pfn = pte_pfn(ptep_get(pte));
137 int actual_node = early_pfn_to_nid(pfn);
139 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
140 pr_warn_once("[%lx-%lx] potential offnode page_structs\n",
141 start, end - 1);
144 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
145 struct vmem_altmap *altmap,
146 struct page *reuse)
148 pte_t *pte = pte_offset_kernel(pmd, addr);
149 if (pte_none(ptep_get(pte))) {
150 pte_t entry;
151 void *p;
153 if (!reuse) {
154 p = vmemmap_alloc_block_buf(PAGE_SIZE, node, altmap);
155 if (!p)
156 return NULL;
157 } else {
159 * When a PTE/PMD entry is freed from the init_mm
160 * there's a free_pages() call to this page allocated
161 * above. Thus this get_page() is paired with the
162 * put_page_testzero() on the freeing path.
163 * This can only called by certain ZONE_DEVICE path,
164 * and through vmemmap_populate_compound_pages() when
165 * slab is available.
167 get_page(reuse);
168 p = page_to_virt(reuse);
170 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
171 set_pte_at(&init_mm, addr, pte, entry);
173 return pte;
176 static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
178 void *p = vmemmap_alloc_block(size, node);
180 if (!p)
181 return NULL;
182 memset(p, 0, size);
184 return p;
187 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
189 pmd_t *pmd = pmd_offset(pud, addr);
190 if (pmd_none(*pmd)) {
191 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
192 if (!p)
193 return NULL;
194 kernel_pte_init(p);
195 pmd_populate_kernel(&init_mm, pmd, p);
197 return pmd;
200 pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
202 pud_t *pud = pud_offset(p4d, addr);
203 if (pud_none(*pud)) {
204 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
205 if (!p)
206 return NULL;
207 pmd_init(p);
208 pud_populate(&init_mm, pud, p);
210 return pud;
213 p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
215 p4d_t *p4d = p4d_offset(pgd, addr);
216 if (p4d_none(*p4d)) {
217 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
218 if (!p)
219 return NULL;
220 pud_init(p);
221 p4d_populate(&init_mm, p4d, p);
223 return p4d;
226 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
228 pgd_t *pgd = pgd_offset_k(addr);
229 if (pgd_none(*pgd)) {
230 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
231 if (!p)
232 return NULL;
233 pgd_populate(&init_mm, pgd, p);
235 return pgd;
238 static pte_t * __meminit vmemmap_populate_address(unsigned long addr, int node,
239 struct vmem_altmap *altmap,
240 struct page *reuse)
242 pgd_t *pgd;
243 p4d_t *p4d;
244 pud_t *pud;
245 pmd_t *pmd;
246 pte_t *pte;
248 pgd = vmemmap_pgd_populate(addr, node);
249 if (!pgd)
250 return NULL;
251 p4d = vmemmap_p4d_populate(pgd, addr, node);
252 if (!p4d)
253 return NULL;
254 pud = vmemmap_pud_populate(p4d, addr, node);
255 if (!pud)
256 return NULL;
257 pmd = vmemmap_pmd_populate(pud, addr, node);
258 if (!pmd)
259 return NULL;
260 pte = vmemmap_pte_populate(pmd, addr, node, altmap, reuse);
261 if (!pte)
262 return NULL;
263 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
265 return pte;
268 static int __meminit vmemmap_populate_range(unsigned long start,
269 unsigned long end, int node,
270 struct vmem_altmap *altmap,
271 struct page *reuse)
273 unsigned long addr = start;
274 pte_t *pte;
276 for (; addr < end; addr += PAGE_SIZE) {
277 pte = vmemmap_populate_address(addr, node, altmap, reuse);
278 if (!pte)
279 return -ENOMEM;
282 return 0;
285 int __meminit vmemmap_populate_basepages(unsigned long start, unsigned long end,
286 int node, struct vmem_altmap *altmap)
288 return vmemmap_populate_range(start, end, node, altmap, NULL);
291 void __weak __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
292 unsigned long addr, unsigned long next)
296 int __weak __meminit vmemmap_check_pmd(pmd_t *pmd, int node,
297 unsigned long addr, unsigned long next)
299 return 0;
302 int __meminit vmemmap_populate_hugepages(unsigned long start, unsigned long end,
303 int node, struct vmem_altmap *altmap)
305 unsigned long addr;
306 unsigned long next;
307 pgd_t *pgd;
308 p4d_t *p4d;
309 pud_t *pud;
310 pmd_t *pmd;
312 for (addr = start; addr < end; addr = next) {
313 next = pmd_addr_end(addr, end);
315 pgd = vmemmap_pgd_populate(addr, node);
316 if (!pgd)
317 return -ENOMEM;
319 p4d = vmemmap_p4d_populate(pgd, addr, node);
320 if (!p4d)
321 return -ENOMEM;
323 pud = vmemmap_pud_populate(p4d, addr, node);
324 if (!pud)
325 return -ENOMEM;
327 pmd = pmd_offset(pud, addr);
328 if (pmd_none(READ_ONCE(*pmd))) {
329 void *p;
331 p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
332 if (p) {
333 vmemmap_set_pmd(pmd, p, node, addr, next);
334 continue;
335 } else if (altmap) {
337 * No fallback: In any case we care about, the
338 * altmap should be reasonably sized and aligned
339 * such that vmemmap_alloc_block_buf() will always
340 * succeed. For consistency with the PTE case,
341 * return an error here as failure could indicate
342 * a configuration issue with the size of the altmap.
344 return -ENOMEM;
346 } else if (vmemmap_check_pmd(pmd, node, addr, next))
347 continue;
348 if (vmemmap_populate_basepages(addr, next, node, altmap))
349 return -ENOMEM;
351 return 0;
354 #ifndef vmemmap_populate_compound_pages
356 * For compound pages bigger than section size (e.g. x86 1G compound
357 * pages with 2M subsection size) fill the rest of sections as tail
358 * pages.
360 * Note that memremap_pages() resets @nr_range value and will increment
361 * it after each range successful onlining. Thus the value or @nr_range
362 * at section memmap populate corresponds to the in-progress range
363 * being onlined here.
365 static bool __meminit reuse_compound_section(unsigned long start_pfn,
366 struct dev_pagemap *pgmap)
368 unsigned long nr_pages = pgmap_vmemmap_nr(pgmap);
369 unsigned long offset = start_pfn -
370 PHYS_PFN(pgmap->ranges[pgmap->nr_range].start);
372 return !IS_ALIGNED(offset, nr_pages) && nr_pages > PAGES_PER_SUBSECTION;
375 static pte_t * __meminit compound_section_tail_page(unsigned long addr)
377 pte_t *pte;
379 addr -= PAGE_SIZE;
382 * Assuming sections are populated sequentially, the previous section's
383 * page data can be reused.
385 pte = pte_offset_kernel(pmd_off_k(addr), addr);
386 if (!pte)
387 return NULL;
389 return pte;
392 static int __meminit vmemmap_populate_compound_pages(unsigned long start_pfn,
393 unsigned long start,
394 unsigned long end, int node,
395 struct dev_pagemap *pgmap)
397 unsigned long size, addr;
398 pte_t *pte;
399 int rc;
401 if (reuse_compound_section(start_pfn, pgmap)) {
402 pte = compound_section_tail_page(start);
403 if (!pte)
404 return -ENOMEM;
407 * Reuse the page that was populated in the prior iteration
408 * with just tail struct pages.
410 return vmemmap_populate_range(start, end, node, NULL,
411 pte_page(ptep_get(pte)));
414 size = min(end - start, pgmap_vmemmap_nr(pgmap) * sizeof(struct page));
415 for (addr = start; addr < end; addr += size) {
416 unsigned long next, last = addr + size;
418 /* Populate the head page vmemmap page */
419 pte = vmemmap_populate_address(addr, node, NULL, NULL);
420 if (!pte)
421 return -ENOMEM;
423 /* Populate the tail pages vmemmap page */
424 next = addr + PAGE_SIZE;
425 pte = vmemmap_populate_address(next, node, NULL, NULL);
426 if (!pte)
427 return -ENOMEM;
430 * Reuse the previous page for the rest of tail pages
431 * See layout diagram in Documentation/mm/vmemmap_dedup.rst
433 next += PAGE_SIZE;
434 rc = vmemmap_populate_range(next, last, node, NULL,
435 pte_page(ptep_get(pte)));
436 if (rc)
437 return -ENOMEM;
440 return 0;
443 #endif
445 struct page * __meminit __populate_section_memmap(unsigned long pfn,
446 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
447 struct dev_pagemap *pgmap)
449 unsigned long start = (unsigned long) pfn_to_page(pfn);
450 unsigned long end = start + nr_pages * sizeof(struct page);
451 int r;
453 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, PAGES_PER_SUBSECTION) ||
454 !IS_ALIGNED(nr_pages, PAGES_PER_SUBSECTION)))
455 return NULL;
457 if (vmemmap_can_optimize(altmap, pgmap))
458 r = vmemmap_populate_compound_pages(pfn, start, end, nid, pgmap);
459 else
460 r = vmemmap_populate(start, end, nid, altmap);
462 if (r < 0)
463 return NULL;
465 if (system_state == SYSTEM_BOOTING)
466 memmap_boot_pages_add(DIV_ROUND_UP(end - start, PAGE_SIZE));
467 else
468 memmap_pages_add(DIV_ROUND_UP(end - start, PAGE_SIZE));
470 return pfn_to_page(pfn);