xref: /linux/arch/x86/mm/init_32.c (revision 9d106c6dd81bb26ad7fc3ee89cb1d62557c8e2c9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  *  Copyright (C) 1995  Linus Torvalds
5  *
6  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
7  */
8 
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/mm.h>
18 #include <linux/hugetlb.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/memblock.h>
28 #include <linux/proc_fs.h>
29 #include <linux/memory_hotplug.h>
30 #include <linux/initrd.h>
31 #include <linux/cpumask.h>
32 #include <linux/gfp.h>
33 
34 #include <asm/asm.h>
35 #include <asm/bios_ebda.h>
36 #include <asm/processor.h>
37 #include <linux/uaccess.h>
38 #include <asm/pgtable.h>
39 #include <asm/dma.h>
40 #include <asm/fixmap.h>
41 #include <asm/e820/api.h>
42 #include <asm/apic.h>
43 #include <asm/bugs.h>
44 #include <asm/tlb.h>
45 #include <asm/tlbflush.h>
46 #include <asm/olpc_ofw.h>
47 #include <asm/pgalloc.h>
48 #include <asm/sections.h>
49 #include <asm/paravirt.h>
50 #include <asm/setup.h>
51 #include <asm/set_memory.h>
52 #include <asm/page_types.h>
53 #include <asm/cpu_entry_area.h>
54 #include <asm/init.h>
55 #include <asm/pgtable_areas.h>
56 
57 #include "mm_internal.h"
58 
59 unsigned long highstart_pfn, highend_pfn;
60 
61 bool __read_mostly __vmalloc_start_set = false;
62 
63 /*
64  * Creates a middle page table and puts a pointer to it in the
65  * given global directory entry. This only returns the gd entry
66  * in non-PAE compilation mode, since the middle layer is folded.
67  */
68 static pmd_t * __init one_md_table_init(pgd_t *pgd)
69 {
70 	p4d_t *p4d;
71 	pud_t *pud;
72 	pmd_t *pmd_table;
73 
74 #ifdef CONFIG_X86_PAE
75 	if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
76 		pmd_table = (pmd_t *)alloc_low_page();
77 		paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
78 		set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
79 		p4d = p4d_offset(pgd, 0);
80 		pud = pud_offset(p4d, 0);
81 		BUG_ON(pmd_table != pmd_offset(pud, 0));
82 
83 		return pmd_table;
84 	}
85 #endif
86 	p4d = p4d_offset(pgd, 0);
87 	pud = pud_offset(p4d, 0);
88 	pmd_table = pmd_offset(pud, 0);
89 
90 	return pmd_table;
91 }
92 
93 /*
94  * Create a page table and place a pointer to it in a middle page
95  * directory entry:
96  */
97 static pte_t * __init one_page_table_init(pmd_t *pmd)
98 {
99 	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
100 		pte_t *page_table = (pte_t *)alloc_low_page();
101 
102 		paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
103 		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
104 		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
105 	}
106 
107 	return pte_offset_kernel(pmd, 0);
108 }
109 
110 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
111 {
112 	int pgd_idx = pgd_index(vaddr);
113 	int pmd_idx = pmd_index(vaddr);
114 
115 	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
116 }
117 
118 pte_t * __init populate_extra_pte(unsigned long vaddr)
119 {
120 	int pte_idx = pte_index(vaddr);
121 	pmd_t *pmd;
122 
123 	pmd = populate_extra_pmd(vaddr);
124 	return one_page_table_init(pmd) + pte_idx;
125 }
126 
127 static unsigned long __init
128 page_table_range_init_count(unsigned long start, unsigned long end)
129 {
130 	unsigned long count = 0;
131 #ifdef CONFIG_HIGHMEM
132 	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
133 	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
134 	int pgd_idx, pmd_idx;
135 	unsigned long vaddr;
136 
137 	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
138 		return 0;
139 
140 	vaddr = start;
141 	pgd_idx = pgd_index(vaddr);
142 	pmd_idx = pmd_index(vaddr);
143 
144 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
145 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
146 							pmd_idx++) {
147 			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
148 			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
149 				count++;
150 			vaddr += PMD_SIZE;
151 		}
152 		pmd_idx = 0;
153 	}
154 #endif
155 	return count;
156 }
157 
158 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
159 					   unsigned long vaddr, pte_t *lastpte,
160 					   void **adr)
161 {
162 #ifdef CONFIG_HIGHMEM
163 	/*
164 	 * Something (early fixmap) may already have put a pte
165 	 * page here, which causes the page table allocation
166 	 * to become nonlinear. Attempt to fix it, and if it
167 	 * is still nonlinear then we have to bug.
168 	 */
169 	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
170 	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
171 
172 	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
173 	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
174 	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
175 		pte_t *newpte;
176 		int i;
177 
178 		BUG_ON(after_bootmem);
179 		newpte = *adr;
180 		for (i = 0; i < PTRS_PER_PTE; i++)
181 			set_pte(newpte + i, pte[i]);
182 		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
183 
184 		paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
185 		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
186 		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
187 		__flush_tlb_all();
188 
189 		paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
190 		pte = newpte;
191 	}
192 	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
193 	       && vaddr > fix_to_virt(FIX_KMAP_END)
194 	       && lastpte && lastpte + PTRS_PER_PTE != pte);
195 #endif
196 	return pte;
197 }
198 
199 /*
200  * This function initializes a certain range of kernel virtual memory
201  * with new bootmem page tables, everywhere page tables are missing in
202  * the given range.
203  *
204  * NOTE: The pagetables are allocated contiguous on the physical space
205  * so we can cache the place of the first one and move around without
206  * checking the pgd every time.
207  */
208 static void __init
209 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
210 {
211 	int pgd_idx, pmd_idx;
212 	unsigned long vaddr;
213 	pgd_t *pgd;
214 	pmd_t *pmd;
215 	pte_t *pte = NULL;
216 	unsigned long count = page_table_range_init_count(start, end);
217 	void *adr = NULL;
218 
219 	if (count)
220 		adr = alloc_low_pages(count);
221 
222 	vaddr = start;
223 	pgd_idx = pgd_index(vaddr);
224 	pmd_idx = pmd_index(vaddr);
225 	pgd = pgd_base + pgd_idx;
226 
227 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
228 		pmd = one_md_table_init(pgd);
229 		pmd = pmd + pmd_index(vaddr);
230 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
231 							pmd++, pmd_idx++) {
232 			pte = page_table_kmap_check(one_page_table_init(pmd),
233 						    pmd, vaddr, pte, &adr);
234 
235 			vaddr += PMD_SIZE;
236 		}
237 		pmd_idx = 0;
238 	}
239 }
240 
241 /*
242  * The <linux/kallsyms.h> already defines is_kernel_text,
243  * using '__' prefix not to get in conflict.
244  */
245 static inline int __is_kernel_text(unsigned long addr)
246 {
247 	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
248 		return 1;
249 	return 0;
250 }
251 
252 /*
253  * This maps the physical memory to kernel virtual address space, a total
254  * of max_low_pfn pages, by creating page tables starting from address
255  * PAGE_OFFSET:
256  */
257 unsigned long __init
258 kernel_physical_mapping_init(unsigned long start,
259 			     unsigned long end,
260 			     unsigned long page_size_mask)
261 {
262 	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
263 	unsigned long last_map_addr = end;
264 	unsigned long start_pfn, end_pfn;
265 	pgd_t *pgd_base = swapper_pg_dir;
266 	int pgd_idx, pmd_idx, pte_ofs;
267 	unsigned long pfn;
268 	pgd_t *pgd;
269 	pmd_t *pmd;
270 	pte_t *pte;
271 	unsigned pages_2m, pages_4k;
272 	int mapping_iter;
273 
274 	start_pfn = start >> PAGE_SHIFT;
275 	end_pfn = end >> PAGE_SHIFT;
276 
277 	/*
278 	 * First iteration will setup identity mapping using large/small pages
279 	 * based on use_pse, with other attributes same as set by
280 	 * the early code in head_32.S
281 	 *
282 	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
283 	 * as desired for the kernel identity mapping.
284 	 *
285 	 * This two pass mechanism conforms to the TLB app note which says:
286 	 *
287 	 *     "Software should not write to a paging-structure entry in a way
288 	 *      that would change, for any linear address, both the page size
289 	 *      and either the page frame or attributes."
290 	 */
291 	mapping_iter = 1;
292 
293 	if (!boot_cpu_has(X86_FEATURE_PSE))
294 		use_pse = 0;
295 
296 repeat:
297 	pages_2m = pages_4k = 0;
298 	pfn = start_pfn;
299 	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
300 	pgd = pgd_base + pgd_idx;
301 	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
302 		pmd = one_md_table_init(pgd);
303 
304 		if (pfn >= end_pfn)
305 			continue;
306 #ifdef CONFIG_X86_PAE
307 		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
308 		pmd += pmd_idx;
309 #else
310 		pmd_idx = 0;
311 #endif
312 		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
313 		     pmd++, pmd_idx++) {
314 			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
315 
316 			/*
317 			 * Map with big pages if possible, otherwise
318 			 * create normal page tables:
319 			 */
320 			if (use_pse) {
321 				unsigned int addr2;
322 				pgprot_t prot = PAGE_KERNEL_LARGE;
323 				/*
324 				 * first pass will use the same initial
325 				 * identity mapping attribute + _PAGE_PSE.
326 				 */
327 				pgprot_t init_prot =
328 					__pgprot(PTE_IDENT_ATTR |
329 						 _PAGE_PSE);
330 
331 				pfn &= PMD_MASK >> PAGE_SHIFT;
332 				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
333 					PAGE_OFFSET + PAGE_SIZE-1;
334 
335 				if (__is_kernel_text(addr) ||
336 				    __is_kernel_text(addr2))
337 					prot = PAGE_KERNEL_LARGE_EXEC;
338 
339 				pages_2m++;
340 				if (mapping_iter == 1)
341 					set_pmd(pmd, pfn_pmd(pfn, init_prot));
342 				else
343 					set_pmd(pmd, pfn_pmd(pfn, prot));
344 
345 				pfn += PTRS_PER_PTE;
346 				continue;
347 			}
348 			pte = one_page_table_init(pmd);
349 
350 			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
351 			pte += pte_ofs;
352 			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
353 			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
354 				pgprot_t prot = PAGE_KERNEL;
355 				/*
356 				 * first pass will use the same initial
357 				 * identity mapping attribute.
358 				 */
359 				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
360 
361 				if (__is_kernel_text(addr))
362 					prot = PAGE_KERNEL_EXEC;
363 
364 				pages_4k++;
365 				if (mapping_iter == 1) {
366 					set_pte(pte, pfn_pte(pfn, init_prot));
367 					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
368 				} else
369 					set_pte(pte, pfn_pte(pfn, prot));
370 			}
371 		}
372 	}
373 	if (mapping_iter == 1) {
374 		/*
375 		 * update direct mapping page count only in the first
376 		 * iteration.
377 		 */
378 		update_page_count(PG_LEVEL_2M, pages_2m);
379 		update_page_count(PG_LEVEL_4K, pages_4k);
380 
381 		/*
382 		 * local global flush tlb, which will flush the previous
383 		 * mappings present in both small and large page TLB's.
384 		 */
385 		__flush_tlb_all();
386 
387 		/*
388 		 * Second iteration will set the actual desired PTE attributes.
389 		 */
390 		mapping_iter = 2;
391 		goto repeat;
392 	}
393 	return last_map_addr;
394 }
395 
396 pte_t *kmap_pte;
397 
398 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
399 {
400 	pgd_t *pgd = pgd_offset_k(vaddr);
401 	p4d_t *p4d = p4d_offset(pgd, vaddr);
402 	pud_t *pud = pud_offset(p4d, vaddr);
403 	pmd_t *pmd = pmd_offset(pud, vaddr);
404 	return pte_offset_kernel(pmd, vaddr);
405 }
406 
407 static void __init kmap_init(void)
408 {
409 	unsigned long kmap_vstart;
410 
411 	/*
412 	 * Cache the first kmap pte:
413 	 */
414 	kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
415 	kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
416 }
417 
418 #ifdef CONFIG_HIGHMEM
419 static void __init permanent_kmaps_init(pgd_t *pgd_base)
420 {
421 	unsigned long vaddr;
422 	pgd_t *pgd;
423 	p4d_t *p4d;
424 	pud_t *pud;
425 	pmd_t *pmd;
426 	pte_t *pte;
427 
428 	vaddr = PKMAP_BASE;
429 	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
430 
431 	pgd = swapper_pg_dir + pgd_index(vaddr);
432 	p4d = p4d_offset(pgd, vaddr);
433 	pud = pud_offset(p4d, vaddr);
434 	pmd = pmd_offset(pud, vaddr);
435 	pte = pte_offset_kernel(pmd, vaddr);
436 	pkmap_page_table = pte;
437 }
438 
439 void __init add_highpages_with_active_regions(int nid,
440 			 unsigned long start_pfn, unsigned long end_pfn)
441 {
442 	phys_addr_t start, end;
443 	u64 i;
444 
445 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
446 		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
447 					    start_pfn, end_pfn);
448 		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
449 					      start_pfn, end_pfn);
450 		for ( ; pfn < e_pfn; pfn++)
451 			if (pfn_valid(pfn))
452 				free_highmem_page(pfn_to_page(pfn));
453 	}
454 }
455 #else
456 static inline void permanent_kmaps_init(pgd_t *pgd_base)
457 {
458 }
459 #endif /* CONFIG_HIGHMEM */
460 
461 void __init sync_initial_page_table(void)
462 {
463 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
464 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
465 			KERNEL_PGD_PTRS);
466 
467 	/*
468 	 * sync back low identity map too.  It is used for example
469 	 * in the 32-bit EFI stub.
470 	 */
471 	clone_pgd_range(initial_page_table,
472 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
473 			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
474 }
475 
476 void __init native_pagetable_init(void)
477 {
478 	unsigned long pfn, va;
479 	pgd_t *pgd, *base = swapper_pg_dir;
480 	p4d_t *p4d;
481 	pud_t *pud;
482 	pmd_t *pmd;
483 	pte_t *pte;
484 
485 	/*
486 	 * Remove any mappings which extend past the end of physical
487 	 * memory from the boot time page table.
488 	 * In virtual address space, we should have at least two pages
489 	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
490 	 * definition. And max_low_pfn is set to VMALLOC_END physical
491 	 * address. If initial memory mapping is doing right job, we
492 	 * should have pte used near max_low_pfn or one pmd is not present.
493 	 */
494 	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
495 		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
496 		pgd = base + pgd_index(va);
497 		if (!pgd_present(*pgd))
498 			break;
499 
500 		p4d = p4d_offset(pgd, va);
501 		pud = pud_offset(p4d, va);
502 		pmd = pmd_offset(pud, va);
503 		if (!pmd_present(*pmd))
504 			break;
505 
506 		/* should not be large page here */
507 		if (pmd_large(*pmd)) {
508 			pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
509 				pfn, pmd, __pa(pmd));
510 			BUG_ON(1);
511 		}
512 
513 		pte = pte_offset_kernel(pmd, va);
514 		if (!pte_present(*pte))
515 			break;
516 
517 		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
518 				pfn, pmd, __pa(pmd), pte, __pa(pte));
519 		pte_clear(NULL, va, pte);
520 	}
521 	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
522 	paging_init();
523 }
524 
525 /*
526  * Build a proper pagetable for the kernel mappings.  Up until this
527  * point, we've been running on some set of pagetables constructed by
528  * the boot process.
529  *
530  * If we're booting on native hardware, this will be a pagetable
531  * constructed in arch/x86/kernel/head_32.S.  The root of the
532  * pagetable will be swapper_pg_dir.
533  *
534  * If we're booting paravirtualized under a hypervisor, then there are
535  * more options: we may already be running PAE, and the pagetable may
536  * or may not be based in swapper_pg_dir.  In any case,
537  * paravirt_pagetable_init() will set up swapper_pg_dir
538  * appropriately for the rest of the initialization to work.
539  *
540  * In general, pagetable_init() assumes that the pagetable may already
541  * be partially populated, and so it avoids stomping on any existing
542  * mappings.
543  */
544 void __init early_ioremap_page_table_range_init(void)
545 {
546 	pgd_t *pgd_base = swapper_pg_dir;
547 	unsigned long vaddr, end;
548 
549 	/*
550 	 * Fixed mappings, only the page table structure has to be
551 	 * created - mappings will be set by set_fixmap():
552 	 */
553 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
554 	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
555 	page_table_range_init(vaddr, end, pgd_base);
556 	early_ioremap_reset();
557 }
558 
559 static void __init pagetable_init(void)
560 {
561 	pgd_t *pgd_base = swapper_pg_dir;
562 
563 	permanent_kmaps_init(pgd_base);
564 }
565 
566 #define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
567 /* Bits supported by the hardware: */
568 pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
569 /* Bits allowed in normal kernel mappings: */
570 pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
571 EXPORT_SYMBOL_GPL(__supported_pte_mask);
572 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
573 EXPORT_SYMBOL(__default_kernel_pte_mask);
574 
575 /* user-defined highmem size */
576 static unsigned int highmem_pages = -1;
577 
578 /*
579  * highmem=size forces highmem to be exactly 'size' bytes.
580  * This works even on boxes that have no highmem otherwise.
581  * This also works to reduce highmem size on bigger boxes.
582  */
583 static int __init parse_highmem(char *arg)
584 {
585 	if (!arg)
586 		return -EINVAL;
587 
588 	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
589 	return 0;
590 }
591 early_param("highmem", parse_highmem);
592 
593 #define MSG_HIGHMEM_TOO_BIG \
594 	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
595 
596 #define MSG_LOWMEM_TOO_SMALL \
597 	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
598 /*
599  * All of RAM fits into lowmem - but if user wants highmem
600  * artificially via the highmem=x boot parameter then create
601  * it:
602  */
603 static void __init lowmem_pfn_init(void)
604 {
605 	/* max_low_pfn is 0, we already have early_res support */
606 	max_low_pfn = max_pfn;
607 
608 	if (highmem_pages == -1)
609 		highmem_pages = 0;
610 #ifdef CONFIG_HIGHMEM
611 	if (highmem_pages >= max_pfn) {
612 		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
613 			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
614 		highmem_pages = 0;
615 	}
616 	if (highmem_pages) {
617 		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
618 			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
619 				pages_to_mb(highmem_pages));
620 			highmem_pages = 0;
621 		}
622 		max_low_pfn -= highmem_pages;
623 	}
624 #else
625 	if (highmem_pages)
626 		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
627 #endif
628 }
629 
630 #define MSG_HIGHMEM_TOO_SMALL \
631 	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
632 
633 #define MSG_HIGHMEM_TRIMMED \
634 	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
635 /*
636  * We have more RAM than fits into lowmem - we try to put it into
637  * highmem, also taking the highmem=x boot parameter into account:
638  */
639 static void __init highmem_pfn_init(void)
640 {
641 	max_low_pfn = MAXMEM_PFN;
642 
643 	if (highmem_pages == -1)
644 		highmem_pages = max_pfn - MAXMEM_PFN;
645 
646 	if (highmem_pages + MAXMEM_PFN < max_pfn)
647 		max_pfn = MAXMEM_PFN + highmem_pages;
648 
649 	if (highmem_pages + MAXMEM_PFN > max_pfn) {
650 		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
651 			pages_to_mb(max_pfn - MAXMEM_PFN),
652 			pages_to_mb(highmem_pages));
653 		highmem_pages = 0;
654 	}
655 #ifndef CONFIG_HIGHMEM
656 	/* Maximum memory usable is what is directly addressable */
657 	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
658 	if (max_pfn > MAX_NONPAE_PFN)
659 		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
660 	else
661 		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
662 	max_pfn = MAXMEM_PFN;
663 #else /* !CONFIG_HIGHMEM */
664 #ifndef CONFIG_HIGHMEM64G
665 	if (max_pfn > MAX_NONPAE_PFN) {
666 		max_pfn = MAX_NONPAE_PFN;
667 		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
668 	}
669 #endif /* !CONFIG_HIGHMEM64G */
670 #endif /* !CONFIG_HIGHMEM */
671 }
672 
673 /*
674  * Determine low and high memory ranges:
675  */
676 void __init find_low_pfn_range(void)
677 {
678 	/* it could update max_pfn */
679 
680 	if (max_pfn <= MAXMEM_PFN)
681 		lowmem_pfn_init();
682 	else
683 		highmem_pfn_init();
684 }
685 
686 #ifndef CONFIG_NEED_MULTIPLE_NODES
687 void __init initmem_init(void)
688 {
689 #ifdef CONFIG_HIGHMEM
690 	highstart_pfn = highend_pfn = max_pfn;
691 	if (max_pfn > max_low_pfn)
692 		highstart_pfn = max_low_pfn;
693 	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
694 		pages_to_mb(highend_pfn - highstart_pfn));
695 	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
696 #else
697 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
698 #endif
699 
700 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
701 	sparse_memory_present_with_active_regions(0);
702 
703 #ifdef CONFIG_FLATMEM
704 	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
705 #endif
706 	__vmalloc_start_set = true;
707 
708 	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
709 			pages_to_mb(max_low_pfn));
710 
711 	setup_bootmem_allocator();
712 }
713 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
714 
715 void __init setup_bootmem_allocator(void)
716 {
717 	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
718 		 max_pfn_mapped<<PAGE_SHIFT);
719 	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
720 }
721 
722 /*
723  * paging_init() sets up the page tables - note that the first 8MB are
724  * already mapped by head.S.
725  *
726  * This routines also unmaps the page at virtual kernel address 0, so
727  * that we can trap those pesky NULL-reference errors in the kernel.
728  */
729 void __init paging_init(void)
730 {
731 	pagetable_init();
732 
733 	__flush_tlb_all();
734 
735 	kmap_init();
736 
737 	/*
738 	 * NOTE: at this point the bootmem allocator is fully available.
739 	 */
740 	olpc_dt_build_devicetree();
741 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
742 	sparse_init();
743 	zone_sizes_init();
744 }
745 
746 /*
747  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
748  * and also on some strange 486's. All 586+'s are OK. This used to involve
749  * black magic jumps to work around some nasty CPU bugs, but fortunately the
750  * switch to using exceptions got rid of all that.
751  */
752 static void __init test_wp_bit(void)
753 {
754 	char z = 0;
755 
756 	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
757 
758 	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
759 
760 	if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
761 		clear_fixmap(FIX_WP_TEST);
762 		printk(KERN_CONT "Ok.\n");
763 		return;
764 	}
765 
766 	printk(KERN_CONT "No.\n");
767 	panic("Linux doesn't support CPUs with broken WP.");
768 }
769 
770 void __init mem_init(void)
771 {
772 	pci_iommu_alloc();
773 
774 #ifdef CONFIG_FLATMEM
775 	BUG_ON(!mem_map);
776 #endif
777 	/*
778 	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
779 	 * be done before memblock_free_all(). Memblock use free low memory for
780 	 * temporary data (see find_range_array()) and for this purpose can use
781 	 * pages that was already passed to the buddy allocator, hence marked as
782 	 * not accessible in the page tables when compiled with
783 	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
784 	 * important here.
785 	 */
786 	set_highmem_pages_init();
787 
788 	/* this will put all low memory onto the freelists */
789 	memblock_free_all();
790 
791 	after_bootmem = 1;
792 	x86_init.hyper.init_after_bootmem();
793 
794 	mem_init_print_info(NULL);
795 	printk(KERN_INFO "virtual kernel memory layout:\n"
796 		"    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
797 		"  cpu_entry : 0x%08lx - 0x%08lx   (%4ld kB)\n"
798 #ifdef CONFIG_HIGHMEM
799 		"    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
800 #endif
801 		"    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
802 		"    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
803 		"      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
804 		"      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
805 		"      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
806 		FIXADDR_START, FIXADDR_TOP,
807 		(FIXADDR_TOP - FIXADDR_START) >> 10,
808 
809 		CPU_ENTRY_AREA_BASE,
810 		CPU_ENTRY_AREA_BASE + CPU_ENTRY_AREA_MAP_SIZE,
811 		CPU_ENTRY_AREA_MAP_SIZE >> 10,
812 
813 #ifdef CONFIG_HIGHMEM
814 		PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
815 		(LAST_PKMAP*PAGE_SIZE) >> 10,
816 #endif
817 
818 		VMALLOC_START, VMALLOC_END,
819 		(VMALLOC_END - VMALLOC_START) >> 20,
820 
821 		(unsigned long)__va(0), (unsigned long)high_memory,
822 		((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
823 
824 		(unsigned long)&__init_begin, (unsigned long)&__init_end,
825 		((unsigned long)&__init_end -
826 		 (unsigned long)&__init_begin) >> 10,
827 
828 		(unsigned long)&_etext, (unsigned long)&_edata,
829 		((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
830 
831 		(unsigned long)&_text, (unsigned long)&_etext,
832 		((unsigned long)&_etext - (unsigned long)&_text) >> 10);
833 
834 	/*
835 	 * Check boundaries twice: Some fundamental inconsistencies can
836 	 * be detected at build time already.
837 	 */
838 #define __FIXADDR_TOP (-PAGE_SIZE)
839 #ifdef CONFIG_HIGHMEM
840 	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
841 	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
842 #endif
843 #define high_memory (-128UL << 20)
844 	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
845 #undef high_memory
846 #undef __FIXADDR_TOP
847 
848 #ifdef CONFIG_HIGHMEM
849 	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
850 	BUG_ON(VMALLOC_END				> PKMAP_BASE);
851 #endif
852 	BUG_ON(VMALLOC_START				>= VMALLOC_END);
853 	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
854 
855 	test_wp_bit();
856 }
857 
858 #ifdef CONFIG_MEMORY_HOTPLUG
859 int arch_add_memory(int nid, u64 start, u64 size,
860 			struct mhp_restrictions *restrictions)
861 {
862 	unsigned long start_pfn = start >> PAGE_SHIFT;
863 	unsigned long nr_pages = size >> PAGE_SHIFT;
864 
865 	return __add_pages(nid, start_pfn, nr_pages, restrictions);
866 }
867 
868 void arch_remove_memory(int nid, u64 start, u64 size,
869 			struct vmem_altmap *altmap)
870 {
871 	unsigned long start_pfn = start >> PAGE_SHIFT;
872 	unsigned long nr_pages = size >> PAGE_SHIFT;
873 
874 	__remove_pages(start_pfn, nr_pages, altmap);
875 }
876 #endif
877 
878 int kernel_set_to_readonly __read_mostly;
879 
880 static void mark_nxdata_nx(void)
881 {
882 	/*
883 	 * When this called, init has already been executed and released,
884 	 * so everything past _etext should be NX.
885 	 */
886 	unsigned long start = PFN_ALIGN(_etext);
887 	/*
888 	 * This comes from __is_kernel_text upper limit. Also HPAGE where used:
889 	 */
890 	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
891 
892 	if (__supported_pte_mask & _PAGE_NX)
893 		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
894 	set_memory_nx(start, size >> PAGE_SHIFT);
895 }
896 
897 void mark_rodata_ro(void)
898 {
899 	unsigned long start = PFN_ALIGN(_text);
900 	unsigned long size = (unsigned long)__end_rodata - start;
901 
902 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
903 	pr_info("Write protecting kernel text and read-only data: %luk\n",
904 		size >> 10);
905 
906 	kernel_set_to_readonly = 1;
907 
908 #ifdef CONFIG_CPA_DEBUG
909 	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
910 	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
911 
912 	pr_info("Testing CPA: write protecting again\n");
913 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
914 #endif
915 	mark_nxdata_nx();
916 	if (__supported_pte_mask & _PAGE_NX)
917 		debug_checkwx();
918 }
919