xref: /openbmc/linux/arch/x86/mm/init_32.c (revision bd4af432)
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 			     pgprot_t prot)
262 {
263 	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
264 	unsigned long last_map_addr = end;
265 	unsigned long start_pfn, end_pfn;
266 	pgd_t *pgd_base = swapper_pg_dir;
267 	int pgd_idx, pmd_idx, pte_ofs;
268 	unsigned long pfn;
269 	pgd_t *pgd;
270 	pmd_t *pmd;
271 	pte_t *pte;
272 	unsigned pages_2m, pages_4k;
273 	int mapping_iter;
274 
275 	start_pfn = start >> PAGE_SHIFT;
276 	end_pfn = end >> PAGE_SHIFT;
277 
278 	/*
279 	 * First iteration will setup identity mapping using large/small pages
280 	 * based on use_pse, with other attributes same as set by
281 	 * the early code in head_32.S
282 	 *
283 	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
284 	 * as desired for the kernel identity mapping.
285 	 *
286 	 * This two pass mechanism conforms to the TLB app note which says:
287 	 *
288 	 *     "Software should not write to a paging-structure entry in a way
289 	 *      that would change, for any linear address, both the page size
290 	 *      and either the page frame or attributes."
291 	 */
292 	mapping_iter = 1;
293 
294 	if (!boot_cpu_has(X86_FEATURE_PSE))
295 		use_pse = 0;
296 
297 repeat:
298 	pages_2m = pages_4k = 0;
299 	pfn = start_pfn;
300 	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
301 	pgd = pgd_base + pgd_idx;
302 	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
303 		pmd = one_md_table_init(pgd);
304 
305 		if (pfn >= end_pfn)
306 			continue;
307 #ifdef CONFIG_X86_PAE
308 		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
309 		pmd += pmd_idx;
310 #else
311 		pmd_idx = 0;
312 #endif
313 		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
314 		     pmd++, pmd_idx++) {
315 			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
316 
317 			/*
318 			 * Map with big pages if possible, otherwise
319 			 * create normal page tables:
320 			 */
321 			if (use_pse) {
322 				unsigned int addr2;
323 				pgprot_t prot = PAGE_KERNEL_LARGE;
324 				/*
325 				 * first pass will use the same initial
326 				 * identity mapping attribute + _PAGE_PSE.
327 				 */
328 				pgprot_t init_prot =
329 					__pgprot(PTE_IDENT_ATTR |
330 						 _PAGE_PSE);
331 
332 				pfn &= PMD_MASK >> PAGE_SHIFT;
333 				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
334 					PAGE_OFFSET + PAGE_SIZE-1;
335 
336 				if (__is_kernel_text(addr) ||
337 				    __is_kernel_text(addr2))
338 					prot = PAGE_KERNEL_LARGE_EXEC;
339 
340 				pages_2m++;
341 				if (mapping_iter == 1)
342 					set_pmd(pmd, pfn_pmd(pfn, init_prot));
343 				else
344 					set_pmd(pmd, pfn_pmd(pfn, prot));
345 
346 				pfn += PTRS_PER_PTE;
347 				continue;
348 			}
349 			pte = one_page_table_init(pmd);
350 
351 			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
352 			pte += pte_ofs;
353 			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
354 			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
355 				pgprot_t prot = PAGE_KERNEL;
356 				/*
357 				 * first pass will use the same initial
358 				 * identity mapping attribute.
359 				 */
360 				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
361 
362 				if (__is_kernel_text(addr))
363 					prot = PAGE_KERNEL_EXEC;
364 
365 				pages_4k++;
366 				if (mapping_iter == 1) {
367 					set_pte(pte, pfn_pte(pfn, init_prot));
368 					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
369 				} else
370 					set_pte(pte, pfn_pte(pfn, prot));
371 			}
372 		}
373 	}
374 	if (mapping_iter == 1) {
375 		/*
376 		 * update direct mapping page count only in the first
377 		 * iteration.
378 		 */
379 		update_page_count(PG_LEVEL_2M, pages_2m);
380 		update_page_count(PG_LEVEL_4K, pages_4k);
381 
382 		/*
383 		 * local global flush tlb, which will flush the previous
384 		 * mappings present in both small and large page TLB's.
385 		 */
386 		__flush_tlb_all();
387 
388 		/*
389 		 * Second iteration will set the actual desired PTE attributes.
390 		 */
391 		mapping_iter = 2;
392 		goto repeat;
393 	}
394 	return last_map_addr;
395 }
396 
397 pte_t *kmap_pte;
398 
399 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
400 {
401 	pgd_t *pgd = pgd_offset_k(vaddr);
402 	p4d_t *p4d = p4d_offset(pgd, vaddr);
403 	pud_t *pud = pud_offset(p4d, vaddr);
404 	pmd_t *pmd = pmd_offset(pud, vaddr);
405 	return pte_offset_kernel(pmd, vaddr);
406 }
407 
408 static void __init kmap_init(void)
409 {
410 	unsigned long kmap_vstart;
411 
412 	/*
413 	 * Cache the first kmap pte:
414 	 */
415 	kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
416 	kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
417 }
418 
419 #ifdef CONFIG_HIGHMEM
420 static void __init permanent_kmaps_init(pgd_t *pgd_base)
421 {
422 	unsigned long vaddr;
423 	pgd_t *pgd;
424 	p4d_t *p4d;
425 	pud_t *pud;
426 	pmd_t *pmd;
427 	pte_t *pte;
428 
429 	vaddr = PKMAP_BASE;
430 	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
431 
432 	pgd = swapper_pg_dir + pgd_index(vaddr);
433 	p4d = p4d_offset(pgd, vaddr);
434 	pud = pud_offset(p4d, vaddr);
435 	pmd = pmd_offset(pud, vaddr);
436 	pte = pte_offset_kernel(pmd, vaddr);
437 	pkmap_page_table = pte;
438 }
439 
440 void __init add_highpages_with_active_regions(int nid,
441 			 unsigned long start_pfn, unsigned long end_pfn)
442 {
443 	phys_addr_t start, end;
444 	u64 i;
445 
446 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
447 		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
448 					    start_pfn, end_pfn);
449 		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
450 					      start_pfn, end_pfn);
451 		for ( ; pfn < e_pfn; pfn++)
452 			if (pfn_valid(pfn))
453 				free_highmem_page(pfn_to_page(pfn));
454 	}
455 }
456 #else
457 static inline void permanent_kmaps_init(pgd_t *pgd_base)
458 {
459 }
460 #endif /* CONFIG_HIGHMEM */
461 
462 void __init sync_initial_page_table(void)
463 {
464 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
465 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
466 			KERNEL_PGD_PTRS);
467 
468 	/*
469 	 * sync back low identity map too.  It is used for example
470 	 * in the 32-bit EFI stub.
471 	 */
472 	clone_pgd_range(initial_page_table,
473 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
474 			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
475 }
476 
477 void __init native_pagetable_init(void)
478 {
479 	unsigned long pfn, va;
480 	pgd_t *pgd, *base = swapper_pg_dir;
481 	p4d_t *p4d;
482 	pud_t *pud;
483 	pmd_t *pmd;
484 	pte_t *pte;
485 
486 	/*
487 	 * Remove any mappings which extend past the end of physical
488 	 * memory from the boot time page table.
489 	 * In virtual address space, we should have at least two pages
490 	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
491 	 * definition. And max_low_pfn is set to VMALLOC_END physical
492 	 * address. If initial memory mapping is doing right job, we
493 	 * should have pte used near max_low_pfn or one pmd is not present.
494 	 */
495 	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
496 		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
497 		pgd = base + pgd_index(va);
498 		if (!pgd_present(*pgd))
499 			break;
500 
501 		p4d = p4d_offset(pgd, va);
502 		pud = pud_offset(p4d, va);
503 		pmd = pmd_offset(pud, va);
504 		if (!pmd_present(*pmd))
505 			break;
506 
507 		/* should not be large page here */
508 		if (pmd_large(*pmd)) {
509 			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",
510 				pfn, pmd, __pa(pmd));
511 			BUG_ON(1);
512 		}
513 
514 		pte = pte_offset_kernel(pmd, va);
515 		if (!pte_present(*pte))
516 			break;
517 
518 		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
519 				pfn, pmd, __pa(pmd), pte, __pa(pte));
520 		pte_clear(NULL, va, pte);
521 	}
522 	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
523 	paging_init();
524 }
525 
526 /*
527  * Build a proper pagetable for the kernel mappings.  Up until this
528  * point, we've been running on some set of pagetables constructed by
529  * the boot process.
530  *
531  * If we're booting on native hardware, this will be a pagetable
532  * constructed in arch/x86/kernel/head_32.S.  The root of the
533  * pagetable will be swapper_pg_dir.
534  *
535  * If we're booting paravirtualized under a hypervisor, then there are
536  * more options: we may already be running PAE, and the pagetable may
537  * or may not be based in swapper_pg_dir.  In any case,
538  * paravirt_pagetable_init() will set up swapper_pg_dir
539  * appropriately for the rest of the initialization to work.
540  *
541  * In general, pagetable_init() assumes that the pagetable may already
542  * be partially populated, and so it avoids stomping on any existing
543  * mappings.
544  */
545 void __init early_ioremap_page_table_range_init(void)
546 {
547 	pgd_t *pgd_base = swapper_pg_dir;
548 	unsigned long vaddr, end;
549 
550 	/*
551 	 * Fixed mappings, only the page table structure has to be
552 	 * created - mappings will be set by set_fixmap():
553 	 */
554 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
555 	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
556 	page_table_range_init(vaddr, end, pgd_base);
557 	early_ioremap_reset();
558 }
559 
560 static void __init pagetable_init(void)
561 {
562 	pgd_t *pgd_base = swapper_pg_dir;
563 
564 	permanent_kmaps_init(pgd_base);
565 }
566 
567 #define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
568 /* Bits supported by the hardware: */
569 pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
570 /* Bits allowed in normal kernel mappings: */
571 pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
572 EXPORT_SYMBOL_GPL(__supported_pte_mask);
573 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
574 EXPORT_SYMBOL(__default_kernel_pte_mask);
575 
576 /* user-defined highmem size */
577 static unsigned int highmem_pages = -1;
578 
579 /*
580  * highmem=size forces highmem to be exactly 'size' bytes.
581  * This works even on boxes that have no highmem otherwise.
582  * This also works to reduce highmem size on bigger boxes.
583  */
584 static int __init parse_highmem(char *arg)
585 {
586 	if (!arg)
587 		return -EINVAL;
588 
589 	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
590 	return 0;
591 }
592 early_param("highmem", parse_highmem);
593 
594 #define MSG_HIGHMEM_TOO_BIG \
595 	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
596 
597 #define MSG_LOWMEM_TOO_SMALL \
598 	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
599 /*
600  * All of RAM fits into lowmem - but if user wants highmem
601  * artificially via the highmem=x boot parameter then create
602  * it:
603  */
604 static void __init lowmem_pfn_init(void)
605 {
606 	/* max_low_pfn is 0, we already have early_res support */
607 	max_low_pfn = max_pfn;
608 
609 	if (highmem_pages == -1)
610 		highmem_pages = 0;
611 #ifdef CONFIG_HIGHMEM
612 	if (highmem_pages >= max_pfn) {
613 		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
614 			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
615 		highmem_pages = 0;
616 	}
617 	if (highmem_pages) {
618 		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
619 			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
620 				pages_to_mb(highmem_pages));
621 			highmem_pages = 0;
622 		}
623 		max_low_pfn -= highmem_pages;
624 	}
625 #else
626 	if (highmem_pages)
627 		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
628 #endif
629 }
630 
631 #define MSG_HIGHMEM_TOO_SMALL \
632 	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
633 
634 #define MSG_HIGHMEM_TRIMMED \
635 	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
636 /*
637  * We have more RAM than fits into lowmem - we try to put it into
638  * highmem, also taking the highmem=x boot parameter into account:
639  */
640 static void __init highmem_pfn_init(void)
641 {
642 	max_low_pfn = MAXMEM_PFN;
643 
644 	if (highmem_pages == -1)
645 		highmem_pages = max_pfn - MAXMEM_PFN;
646 
647 	if (highmem_pages + MAXMEM_PFN < max_pfn)
648 		max_pfn = MAXMEM_PFN + highmem_pages;
649 
650 	if (highmem_pages + MAXMEM_PFN > max_pfn) {
651 		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
652 			pages_to_mb(max_pfn - MAXMEM_PFN),
653 			pages_to_mb(highmem_pages));
654 		highmem_pages = 0;
655 	}
656 #ifndef CONFIG_HIGHMEM
657 	/* Maximum memory usable is what is directly addressable */
658 	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
659 	if (max_pfn > MAX_NONPAE_PFN)
660 		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
661 	else
662 		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
663 	max_pfn = MAXMEM_PFN;
664 #else /* !CONFIG_HIGHMEM */
665 #ifndef CONFIG_HIGHMEM64G
666 	if (max_pfn > MAX_NONPAE_PFN) {
667 		max_pfn = MAX_NONPAE_PFN;
668 		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
669 	}
670 #endif /* !CONFIG_HIGHMEM64G */
671 #endif /* !CONFIG_HIGHMEM */
672 }
673 
674 /*
675  * Determine low and high memory ranges:
676  */
677 void __init find_low_pfn_range(void)
678 {
679 	/* it could update max_pfn */
680 
681 	if (max_pfn <= MAXMEM_PFN)
682 		lowmem_pfn_init();
683 	else
684 		highmem_pfn_init();
685 }
686 
687 #ifndef CONFIG_NEED_MULTIPLE_NODES
688 void __init initmem_init(void)
689 {
690 #ifdef CONFIG_HIGHMEM
691 	highstart_pfn = highend_pfn = max_pfn;
692 	if (max_pfn > max_low_pfn)
693 		highstart_pfn = max_low_pfn;
694 	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
695 		pages_to_mb(highend_pfn - highstart_pfn));
696 	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
697 #else
698 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
699 #endif
700 
701 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
702 	sparse_memory_present_with_active_regions(0);
703 
704 #ifdef CONFIG_FLATMEM
705 	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
706 #endif
707 	__vmalloc_start_set = true;
708 
709 	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
710 			pages_to_mb(max_low_pfn));
711 
712 	setup_bootmem_allocator();
713 }
714 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
715 
716 void __init setup_bootmem_allocator(void)
717 {
718 	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
719 		 max_pfn_mapped<<PAGE_SHIFT);
720 	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
721 }
722 
723 /*
724  * paging_init() sets up the page tables - note that the first 8MB are
725  * already mapped by head.S.
726  *
727  * This routines also unmaps the page at virtual kernel address 0, so
728  * that we can trap those pesky NULL-reference errors in the kernel.
729  */
730 void __init paging_init(void)
731 {
732 	pagetable_init();
733 
734 	__flush_tlb_all();
735 
736 	kmap_init();
737 
738 	/*
739 	 * NOTE: at this point the bootmem allocator is fully available.
740 	 */
741 	olpc_dt_build_devicetree();
742 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
743 	sparse_init();
744 	zone_sizes_init();
745 }
746 
747 /*
748  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
749  * and also on some strange 486's. All 586+'s are OK. This used to involve
750  * black magic jumps to work around some nasty CPU bugs, but fortunately the
751  * switch to using exceptions got rid of all that.
752  */
753 static void __init test_wp_bit(void)
754 {
755 	char z = 0;
756 
757 	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
758 
759 	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
760 
761 	if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
762 		clear_fixmap(FIX_WP_TEST);
763 		printk(KERN_CONT "Ok.\n");
764 		return;
765 	}
766 
767 	printk(KERN_CONT "No.\n");
768 	panic("Linux doesn't support CPUs with broken WP.");
769 }
770 
771 void __init mem_init(void)
772 {
773 	pci_iommu_alloc();
774 
775 #ifdef CONFIG_FLATMEM
776 	BUG_ON(!mem_map);
777 #endif
778 	/*
779 	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
780 	 * be done before memblock_free_all(). Memblock use free low memory for
781 	 * temporary data (see find_range_array()) and for this purpose can use
782 	 * pages that was already passed to the buddy allocator, hence marked as
783 	 * not accessible in the page tables when compiled with
784 	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
785 	 * important here.
786 	 */
787 	set_highmem_pages_init();
788 
789 	/* this will put all low memory onto the freelists */
790 	memblock_free_all();
791 
792 	after_bootmem = 1;
793 	x86_init.hyper.init_after_bootmem();
794 
795 	mem_init_print_info(NULL);
796 
797 	/*
798 	 * Check boundaries twice: Some fundamental inconsistencies can
799 	 * be detected at build time already.
800 	 */
801 #define __FIXADDR_TOP (-PAGE_SIZE)
802 #ifdef CONFIG_HIGHMEM
803 	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
804 	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
805 #endif
806 #define high_memory (-128UL << 20)
807 	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
808 #undef high_memory
809 #undef __FIXADDR_TOP
810 
811 #ifdef CONFIG_HIGHMEM
812 	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
813 	BUG_ON(VMALLOC_END				> PKMAP_BASE);
814 #endif
815 	BUG_ON(VMALLOC_START				>= VMALLOC_END);
816 	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
817 
818 	test_wp_bit();
819 }
820 
821 #ifdef CONFIG_MEMORY_HOTPLUG
822 int arch_add_memory(int nid, u64 start, u64 size,
823 		    struct mhp_params *params)
824 {
825 	unsigned long start_pfn = start >> PAGE_SHIFT;
826 	unsigned long nr_pages = size >> PAGE_SHIFT;
827 	int ret;
828 
829 	/*
830 	 * The page tables were already mapped at boot so if the caller
831 	 * requests a different mapping type then we must change all the
832 	 * pages with __set_memory_prot().
833 	 */
834 	if (params->pgprot.pgprot != PAGE_KERNEL.pgprot) {
835 		ret = __set_memory_prot(start, nr_pages, params->pgprot);
836 		if (ret)
837 			return ret;
838 	}
839 
840 	return __add_pages(nid, start_pfn, nr_pages, params);
841 }
842 
843 void arch_remove_memory(int nid, u64 start, u64 size,
844 			struct vmem_altmap *altmap)
845 {
846 	unsigned long start_pfn = start >> PAGE_SHIFT;
847 	unsigned long nr_pages = size >> PAGE_SHIFT;
848 
849 	__remove_pages(start_pfn, nr_pages, altmap);
850 }
851 #endif
852 
853 int kernel_set_to_readonly __read_mostly;
854 
855 static void mark_nxdata_nx(void)
856 {
857 	/*
858 	 * When this called, init has already been executed and released,
859 	 * so everything past _etext should be NX.
860 	 */
861 	unsigned long start = PFN_ALIGN(_etext);
862 	/*
863 	 * This comes from __is_kernel_text upper limit. Also HPAGE where used:
864 	 */
865 	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
866 
867 	if (__supported_pte_mask & _PAGE_NX)
868 		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
869 	set_memory_nx(start, size >> PAGE_SHIFT);
870 }
871 
872 void mark_rodata_ro(void)
873 {
874 	unsigned long start = PFN_ALIGN(_text);
875 	unsigned long size = (unsigned long)__end_rodata - start;
876 
877 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
878 	pr_info("Write protecting kernel text and read-only data: %luk\n",
879 		size >> 10);
880 
881 	kernel_set_to_readonly = 1;
882 
883 #ifdef CONFIG_CPA_DEBUG
884 	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
885 	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
886 
887 	pr_info("Testing CPA: write protecting again\n");
888 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
889 #endif
890 	mark_nxdata_nx();
891 	if (__supported_pte_mask & _PAGE_NX)
892 		debug_checkwx();
893 }
894