xref: /openbmc/linux/arch/x86/mm/init_32.c (revision f20c7d91)
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/dma.h>
39 #include <asm/fixmap.h>
40 #include <asm/e820/api.h>
41 #include <asm/apic.h>
42 #include <asm/bugs.h>
43 #include <asm/tlb.h>
44 #include <asm/tlbflush.h>
45 #include <asm/olpc_ofw.h>
46 #include <asm/pgalloc.h>
47 #include <asm/sections.h>
48 #include <asm/paravirt.h>
49 #include <asm/setup.h>
50 #include <asm/set_memory.h>
51 #include <asm/page_types.h>
52 #include <asm/cpu_entry_area.h>
53 #include <asm/init.h>
54 #include <asm/pgtable_areas.h>
55 
56 #include "mm_internal.h"
57 
58 unsigned long highstart_pfn, highend_pfn;
59 
60 bool __read_mostly __vmalloc_start_set = false;
61 
62 /*
63  * Creates a middle page table and puts a pointer to it in the
64  * given global directory entry. This only returns the gd entry
65  * in non-PAE compilation mode, since the middle layer is folded.
66  */
67 static pmd_t * __init one_md_table_init(pgd_t *pgd)
68 {
69 	p4d_t *p4d;
70 	pud_t *pud;
71 	pmd_t *pmd_table;
72 
73 #ifdef CONFIG_X86_PAE
74 	if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
75 		pmd_table = (pmd_t *)alloc_low_page();
76 		paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
77 		set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
78 		p4d = p4d_offset(pgd, 0);
79 		pud = pud_offset(p4d, 0);
80 		BUG_ON(pmd_table != pmd_offset(pud, 0));
81 
82 		return pmd_table;
83 	}
84 #endif
85 	p4d = p4d_offset(pgd, 0);
86 	pud = pud_offset(p4d, 0);
87 	pmd_table = pmd_offset(pud, 0);
88 
89 	return pmd_table;
90 }
91 
92 /*
93  * Create a page table and place a pointer to it in a middle page
94  * directory entry:
95  */
96 static pte_t * __init one_page_table_init(pmd_t *pmd)
97 {
98 	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
99 		pte_t *page_table = (pte_t *)alloc_low_page();
100 
101 		paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
102 		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
103 		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
104 	}
105 
106 	return pte_offset_kernel(pmd, 0);
107 }
108 
109 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
110 {
111 	int pgd_idx = pgd_index(vaddr);
112 	int pmd_idx = pmd_index(vaddr);
113 
114 	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
115 }
116 
117 pte_t * __init populate_extra_pte(unsigned long vaddr)
118 {
119 	int pte_idx = pte_index(vaddr);
120 	pmd_t *pmd;
121 
122 	pmd = populate_extra_pmd(vaddr);
123 	return one_page_table_init(pmd) + pte_idx;
124 }
125 
126 static unsigned long __init
127 page_table_range_init_count(unsigned long start, unsigned long end)
128 {
129 	unsigned long count = 0;
130 #ifdef CONFIG_HIGHMEM
131 	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
132 	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
133 	int pgd_idx, pmd_idx;
134 	unsigned long vaddr;
135 
136 	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
137 		return 0;
138 
139 	vaddr = start;
140 	pgd_idx = pgd_index(vaddr);
141 	pmd_idx = pmd_index(vaddr);
142 
143 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
144 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
145 							pmd_idx++) {
146 			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
147 			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
148 				count++;
149 			vaddr += PMD_SIZE;
150 		}
151 		pmd_idx = 0;
152 	}
153 #endif
154 	return count;
155 }
156 
157 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
158 					   unsigned long vaddr, pte_t *lastpte,
159 					   void **adr)
160 {
161 #ifdef CONFIG_HIGHMEM
162 	/*
163 	 * Something (early fixmap) may already have put a pte
164 	 * page here, which causes the page table allocation
165 	 * to become nonlinear. Attempt to fix it, and if it
166 	 * is still nonlinear then we have to bug.
167 	 */
168 	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
169 	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
170 
171 	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
172 	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
173 	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
174 		pte_t *newpte;
175 		int i;
176 
177 		BUG_ON(after_bootmem);
178 		newpte = *adr;
179 		for (i = 0; i < PTRS_PER_PTE; i++)
180 			set_pte(newpte + i, pte[i]);
181 		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
182 
183 		paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
184 		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
185 		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
186 		__flush_tlb_all();
187 
188 		paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
189 		pte = newpte;
190 	}
191 	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
192 	       && vaddr > fix_to_virt(FIX_KMAP_END)
193 	       && lastpte && lastpte + PTRS_PER_PTE != pte);
194 #endif
195 	return pte;
196 }
197 
198 /*
199  * This function initializes a certain range of kernel virtual memory
200  * with new bootmem page tables, everywhere page tables are missing in
201  * the given range.
202  *
203  * NOTE: The pagetables are allocated contiguous on the physical space
204  * so we can cache the place of the first one and move around without
205  * checking the pgd every time.
206  */
207 static void __init
208 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
209 {
210 	int pgd_idx, pmd_idx;
211 	unsigned long vaddr;
212 	pgd_t *pgd;
213 	pmd_t *pmd;
214 	pte_t *pte = NULL;
215 	unsigned long count = page_table_range_init_count(start, end);
216 	void *adr = NULL;
217 
218 	if (count)
219 		adr = alloc_low_pages(count);
220 
221 	vaddr = start;
222 	pgd_idx = pgd_index(vaddr);
223 	pmd_idx = pmd_index(vaddr);
224 	pgd = pgd_base + pgd_idx;
225 
226 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
227 		pmd = one_md_table_init(pgd);
228 		pmd = pmd + pmd_index(vaddr);
229 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
230 							pmd++, pmd_idx++) {
231 			pte = page_table_kmap_check(one_page_table_init(pmd),
232 						    pmd, vaddr, pte, &adr);
233 
234 			vaddr += PMD_SIZE;
235 		}
236 		pmd_idx = 0;
237 	}
238 }
239 
240 /*
241  * The <linux/kallsyms.h> already defines is_kernel_text,
242  * using '__' prefix not to get in conflict.
243  */
244 static inline int __is_kernel_text(unsigned long addr)
245 {
246 	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
247 		return 1;
248 	return 0;
249 }
250 
251 /*
252  * This maps the physical memory to kernel virtual address space, a total
253  * of max_low_pfn pages, by creating page tables starting from address
254  * PAGE_OFFSET:
255  */
256 unsigned long __init
257 kernel_physical_mapping_init(unsigned long start,
258 			     unsigned long end,
259 			     unsigned long page_size_mask,
260 			     pgprot_t prot)
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 void __init kmap_init(void)
399 {
400 	unsigned long kmap_vstart;
401 
402 	/*
403 	 * Cache the first kmap pte:
404 	 */
405 	kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
406 	kmap_pte = virt_to_kpte(kmap_vstart);
407 }
408 
409 #ifdef CONFIG_HIGHMEM
410 static void __init permanent_kmaps_init(pgd_t *pgd_base)
411 {
412 	unsigned long vaddr = PKMAP_BASE;
413 
414 	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
415 
416 	pkmap_page_table = virt_to_kpte(vaddr);
417 }
418 
419 void __init add_highpages_with_active_regions(int nid,
420 			 unsigned long start_pfn, unsigned long end_pfn)
421 {
422 	phys_addr_t start, end;
423 	u64 i;
424 
425 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
426 		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
427 					    start_pfn, end_pfn);
428 		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
429 					      start_pfn, end_pfn);
430 		for ( ; pfn < e_pfn; pfn++)
431 			if (pfn_valid(pfn))
432 				free_highmem_page(pfn_to_page(pfn));
433 	}
434 }
435 #else
436 static inline void permanent_kmaps_init(pgd_t *pgd_base)
437 {
438 }
439 #endif /* CONFIG_HIGHMEM */
440 
441 void __init sync_initial_page_table(void)
442 {
443 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
444 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
445 			KERNEL_PGD_PTRS);
446 
447 	/*
448 	 * sync back low identity map too.  It is used for example
449 	 * in the 32-bit EFI stub.
450 	 */
451 	clone_pgd_range(initial_page_table,
452 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
453 			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
454 }
455 
456 void __init native_pagetable_init(void)
457 {
458 	unsigned long pfn, va;
459 	pgd_t *pgd, *base = swapper_pg_dir;
460 	p4d_t *p4d;
461 	pud_t *pud;
462 	pmd_t *pmd;
463 	pte_t *pte;
464 
465 	/*
466 	 * Remove any mappings which extend past the end of physical
467 	 * memory from the boot time page table.
468 	 * In virtual address space, we should have at least two pages
469 	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
470 	 * definition. And max_low_pfn is set to VMALLOC_END physical
471 	 * address. If initial memory mapping is doing right job, we
472 	 * should have pte used near max_low_pfn or one pmd is not present.
473 	 */
474 	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
475 		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
476 		pgd = base + pgd_index(va);
477 		if (!pgd_present(*pgd))
478 			break;
479 
480 		p4d = p4d_offset(pgd, va);
481 		pud = pud_offset(p4d, va);
482 		pmd = pmd_offset(pud, va);
483 		if (!pmd_present(*pmd))
484 			break;
485 
486 		/* should not be large page here */
487 		if (pmd_large(*pmd)) {
488 			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",
489 				pfn, pmd, __pa(pmd));
490 			BUG_ON(1);
491 		}
492 
493 		pte = pte_offset_kernel(pmd, va);
494 		if (!pte_present(*pte))
495 			break;
496 
497 		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
498 				pfn, pmd, __pa(pmd), pte, __pa(pte));
499 		pte_clear(NULL, va, pte);
500 	}
501 	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
502 	paging_init();
503 }
504 
505 /*
506  * Build a proper pagetable for the kernel mappings.  Up until this
507  * point, we've been running on some set of pagetables constructed by
508  * the boot process.
509  *
510  * If we're booting on native hardware, this will be a pagetable
511  * constructed in arch/x86/kernel/head_32.S.  The root of the
512  * pagetable will be swapper_pg_dir.
513  *
514  * If we're booting paravirtualized under a hypervisor, then there are
515  * more options: we may already be running PAE, and the pagetable may
516  * or may not be based in swapper_pg_dir.  In any case,
517  * paravirt_pagetable_init() will set up swapper_pg_dir
518  * appropriately for the rest of the initialization to work.
519  *
520  * In general, pagetable_init() assumes that the pagetable may already
521  * be partially populated, and so it avoids stomping on any existing
522  * mappings.
523  */
524 void __init early_ioremap_page_table_range_init(void)
525 {
526 	pgd_t *pgd_base = swapper_pg_dir;
527 	unsigned long vaddr, end;
528 
529 	/*
530 	 * Fixed mappings, only the page table structure has to be
531 	 * created - mappings will be set by set_fixmap():
532 	 */
533 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
534 	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
535 	page_table_range_init(vaddr, end, pgd_base);
536 	early_ioremap_reset();
537 }
538 
539 static void __init pagetable_init(void)
540 {
541 	pgd_t *pgd_base = swapper_pg_dir;
542 
543 	permanent_kmaps_init(pgd_base);
544 }
545 
546 #define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
547 /* Bits supported by the hardware: */
548 pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
549 /* Bits allowed in normal kernel mappings: */
550 pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
551 EXPORT_SYMBOL_GPL(__supported_pte_mask);
552 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
553 EXPORT_SYMBOL(__default_kernel_pte_mask);
554 
555 /* user-defined highmem size */
556 static unsigned int highmem_pages = -1;
557 
558 /*
559  * highmem=size forces highmem to be exactly 'size' bytes.
560  * This works even on boxes that have no highmem otherwise.
561  * This also works to reduce highmem size on bigger boxes.
562  */
563 static int __init parse_highmem(char *arg)
564 {
565 	if (!arg)
566 		return -EINVAL;
567 
568 	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
569 	return 0;
570 }
571 early_param("highmem", parse_highmem);
572 
573 #define MSG_HIGHMEM_TOO_BIG \
574 	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
575 
576 #define MSG_LOWMEM_TOO_SMALL \
577 	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
578 /*
579  * All of RAM fits into lowmem - but if user wants highmem
580  * artificially via the highmem=x boot parameter then create
581  * it:
582  */
583 static void __init lowmem_pfn_init(void)
584 {
585 	/* max_low_pfn is 0, we already have early_res support */
586 	max_low_pfn = max_pfn;
587 
588 	if (highmem_pages == -1)
589 		highmem_pages = 0;
590 #ifdef CONFIG_HIGHMEM
591 	if (highmem_pages >= max_pfn) {
592 		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
593 			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
594 		highmem_pages = 0;
595 	}
596 	if (highmem_pages) {
597 		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
598 			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
599 				pages_to_mb(highmem_pages));
600 			highmem_pages = 0;
601 		}
602 		max_low_pfn -= highmem_pages;
603 	}
604 #else
605 	if (highmem_pages)
606 		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
607 #endif
608 }
609 
610 #define MSG_HIGHMEM_TOO_SMALL \
611 	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
612 
613 #define MSG_HIGHMEM_TRIMMED \
614 	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
615 /*
616  * We have more RAM than fits into lowmem - we try to put it into
617  * highmem, also taking the highmem=x boot parameter into account:
618  */
619 static void __init highmem_pfn_init(void)
620 {
621 	max_low_pfn = MAXMEM_PFN;
622 
623 	if (highmem_pages == -1)
624 		highmem_pages = max_pfn - MAXMEM_PFN;
625 
626 	if (highmem_pages + MAXMEM_PFN < max_pfn)
627 		max_pfn = MAXMEM_PFN + highmem_pages;
628 
629 	if (highmem_pages + MAXMEM_PFN > max_pfn) {
630 		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
631 			pages_to_mb(max_pfn - MAXMEM_PFN),
632 			pages_to_mb(highmem_pages));
633 		highmem_pages = 0;
634 	}
635 #ifndef CONFIG_HIGHMEM
636 	/* Maximum memory usable is what is directly addressable */
637 	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
638 	if (max_pfn > MAX_NONPAE_PFN)
639 		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
640 	else
641 		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
642 	max_pfn = MAXMEM_PFN;
643 #else /* !CONFIG_HIGHMEM */
644 #ifndef CONFIG_HIGHMEM64G
645 	if (max_pfn > MAX_NONPAE_PFN) {
646 		max_pfn = MAX_NONPAE_PFN;
647 		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
648 	}
649 #endif /* !CONFIG_HIGHMEM64G */
650 #endif /* !CONFIG_HIGHMEM */
651 }
652 
653 /*
654  * Determine low and high memory ranges:
655  */
656 void __init find_low_pfn_range(void)
657 {
658 	/* it could update max_pfn */
659 
660 	if (max_pfn <= MAXMEM_PFN)
661 		lowmem_pfn_init();
662 	else
663 		highmem_pfn_init();
664 }
665 
666 #ifndef CONFIG_NEED_MULTIPLE_NODES
667 void __init initmem_init(void)
668 {
669 #ifdef CONFIG_HIGHMEM
670 	highstart_pfn = highend_pfn = max_pfn;
671 	if (max_pfn > max_low_pfn)
672 		highstart_pfn = max_low_pfn;
673 	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
674 		pages_to_mb(highend_pfn - highstart_pfn));
675 	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
676 #else
677 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
678 #endif
679 
680 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
681 	sparse_memory_present_with_active_regions(0);
682 
683 #ifdef CONFIG_FLATMEM
684 	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
685 #endif
686 	__vmalloc_start_set = true;
687 
688 	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
689 			pages_to_mb(max_low_pfn));
690 
691 	setup_bootmem_allocator();
692 }
693 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
694 
695 void __init setup_bootmem_allocator(void)
696 {
697 	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
698 		 max_pfn_mapped<<PAGE_SHIFT);
699 	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
700 }
701 
702 /*
703  * paging_init() sets up the page tables - note that the first 8MB are
704  * already mapped by head.S.
705  *
706  * This routines also unmaps the page at virtual kernel address 0, so
707  * that we can trap those pesky NULL-reference errors in the kernel.
708  */
709 void __init paging_init(void)
710 {
711 	pagetable_init();
712 
713 	__flush_tlb_all();
714 
715 	kmap_init();
716 
717 	/*
718 	 * NOTE: at this point the bootmem allocator is fully available.
719 	 */
720 	olpc_dt_build_devicetree();
721 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
722 	sparse_init();
723 	zone_sizes_init();
724 }
725 
726 /*
727  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
728  * and also on some strange 486's. All 586+'s are OK. This used to involve
729  * black magic jumps to work around some nasty CPU bugs, but fortunately the
730  * switch to using exceptions got rid of all that.
731  */
732 static void __init test_wp_bit(void)
733 {
734 	char z = 0;
735 
736 	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
737 
738 	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
739 
740 	if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
741 		clear_fixmap(FIX_WP_TEST);
742 		printk(KERN_CONT "Ok.\n");
743 		return;
744 	}
745 
746 	printk(KERN_CONT "No.\n");
747 	panic("Linux doesn't support CPUs with broken WP.");
748 }
749 
750 void __init mem_init(void)
751 {
752 	pci_iommu_alloc();
753 
754 #ifdef CONFIG_FLATMEM
755 	BUG_ON(!mem_map);
756 #endif
757 	/*
758 	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
759 	 * be done before memblock_free_all(). Memblock use free low memory for
760 	 * temporary data (see find_range_array()) and for this purpose can use
761 	 * pages that was already passed to the buddy allocator, hence marked as
762 	 * not accessible in the page tables when compiled with
763 	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
764 	 * important here.
765 	 */
766 	set_highmem_pages_init();
767 
768 	/* this will put all low memory onto the freelists */
769 	memblock_free_all();
770 
771 	after_bootmem = 1;
772 	x86_init.hyper.init_after_bootmem();
773 
774 	mem_init_print_info(NULL);
775 
776 	/*
777 	 * Check boundaries twice: Some fundamental inconsistencies can
778 	 * be detected at build time already.
779 	 */
780 #define __FIXADDR_TOP (-PAGE_SIZE)
781 #ifdef CONFIG_HIGHMEM
782 	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
783 	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
784 #endif
785 #define high_memory (-128UL << 20)
786 	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
787 #undef high_memory
788 #undef __FIXADDR_TOP
789 
790 #ifdef CONFIG_HIGHMEM
791 	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
792 	BUG_ON(VMALLOC_END				> PKMAP_BASE);
793 #endif
794 	BUG_ON(VMALLOC_START				>= VMALLOC_END);
795 	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
796 
797 	test_wp_bit();
798 }
799 
800 #ifdef CONFIG_MEMORY_HOTPLUG
801 int arch_add_memory(int nid, u64 start, u64 size,
802 		    struct mhp_params *params)
803 {
804 	unsigned long start_pfn = start >> PAGE_SHIFT;
805 	unsigned long nr_pages = size >> PAGE_SHIFT;
806 	int ret;
807 
808 	/*
809 	 * The page tables were already mapped at boot so if the caller
810 	 * requests a different mapping type then we must change all the
811 	 * pages with __set_memory_prot().
812 	 */
813 	if (params->pgprot.pgprot != PAGE_KERNEL.pgprot) {
814 		ret = __set_memory_prot(start, nr_pages, params->pgprot);
815 		if (ret)
816 			return ret;
817 	}
818 
819 	return __add_pages(nid, start_pfn, nr_pages, params);
820 }
821 
822 void arch_remove_memory(int nid, u64 start, u64 size,
823 			struct vmem_altmap *altmap)
824 {
825 	unsigned long start_pfn = start >> PAGE_SHIFT;
826 	unsigned long nr_pages = size >> PAGE_SHIFT;
827 
828 	__remove_pages(start_pfn, nr_pages, altmap);
829 }
830 #endif
831 
832 int kernel_set_to_readonly __read_mostly;
833 
834 static void mark_nxdata_nx(void)
835 {
836 	/*
837 	 * When this called, init has already been executed and released,
838 	 * so everything past _etext should be NX.
839 	 */
840 	unsigned long start = PFN_ALIGN(_etext);
841 	/*
842 	 * This comes from __is_kernel_text upper limit. Also HPAGE where used:
843 	 */
844 	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
845 
846 	if (__supported_pte_mask & _PAGE_NX)
847 		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
848 	set_memory_nx(start, size >> PAGE_SHIFT);
849 }
850 
851 void mark_rodata_ro(void)
852 {
853 	unsigned long start = PFN_ALIGN(_text);
854 	unsigned long size = (unsigned long)__end_rodata - start;
855 
856 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
857 	pr_info("Write protecting kernel text and read-only data: %luk\n",
858 		size >> 10);
859 
860 	kernel_set_to_readonly = 1;
861 
862 #ifdef CONFIG_CPA_DEBUG
863 	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
864 	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
865 
866 	pr_info("Testing CPA: write protecting again\n");
867 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
868 #endif
869 	mark_nxdata_nx();
870 	if (__supported_pte_mask & _PAGE_NX)
871 		debug_checkwx();
872 }
873