• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *
3  *  Copyright (C) 1995  Linus Torvalds
4  *
5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6  */
7 
8 #include <linux/module.h>
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/bootmem.h>
28 #include <linux/memblock.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
33 #include <linux/gfp.h>
34 
35 #include <asm/asm.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/processor.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
40 #include <asm/dma.h>
41 #include <asm/fixmap.h>
42 #include <asm/e820.h>
43 #include <asm/apic.h>
44 #include <asm/bugs.h>
45 #include <asm/tlb.h>
46 #include <asm/tlbflush.h>
47 #include <asm/olpc_ofw.h>
48 #include <asm/pgalloc.h>
49 #include <asm/sections.h>
50 #include <asm/paravirt.h>
51 #include <asm/setup.h>
52 #include <asm/cacheflush.h>
53 #include <asm/page_types.h>
54 #include <asm/init.h>
55 
56 #include "mm_internal.h"
57 
58 unsigned long highstart_pfn, highend_pfn;
59 
60 static noinline int do_test_wp_bit(void);
61 
62 bool __read_mostly __vmalloc_start_set = false;
63 
64 /*
65  * Creates a middle page table and puts a pointer to it in the
66  * given global directory entry. This only returns the gd entry
67  * in non-PAE compilation mode, since the middle layer is folded.
68  */
one_md_table_init(pgd_t * pgd)69 static pmd_t * __init one_md_table_init(pgd_t *pgd)
70 {
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 		pud = pud_offset(pgd, 0);
80 		BUG_ON(pmd_table != pmd_offset(pud, 0));
81 
82 		return pmd_table;
83 	}
84 #endif
85 	pud = pud_offset(pgd, 0);
86 	pmd_table = pmd_offset(pud, 0);
87 
88 	return pmd_table;
89 }
90 
91 /*
92  * Create a page table and place a pointer to it in a middle page
93  * directory entry:
94  */
one_page_table_init(pmd_t * pmd)95 static pte_t * __init one_page_table_init(pmd_t *pmd)
96 {
97 	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
98 		pte_t *page_table = (pte_t *)alloc_low_page();
99 
100 		paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
101 		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
102 		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
103 	}
104 
105 	return pte_offset_kernel(pmd, 0);
106 }
107 
populate_extra_pmd(unsigned long vaddr)108 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
109 {
110 	int pgd_idx = pgd_index(vaddr);
111 	int pmd_idx = pmd_index(vaddr);
112 
113 	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
114 }
115 
populate_extra_pte(unsigned long vaddr)116 pte_t * __init populate_extra_pte(unsigned long vaddr)
117 {
118 	int pte_idx = pte_index(vaddr);
119 	pmd_t *pmd;
120 
121 	pmd = populate_extra_pmd(vaddr);
122 	return one_page_table_init(pmd) + pte_idx;
123 }
124 
125 static unsigned long __init
page_table_range_init_count(unsigned long start,unsigned long end)126 page_table_range_init_count(unsigned long start, unsigned long end)
127 {
128 	unsigned long count = 0;
129 #ifdef CONFIG_HIGHMEM
130 	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
131 	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
132 	int pgd_idx, pmd_idx;
133 	unsigned long vaddr;
134 
135 	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
136 		return 0;
137 
138 	vaddr = start;
139 	pgd_idx = pgd_index(vaddr);
140 	pmd_idx = pmd_index(vaddr);
141 
142 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
143 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
144 							pmd_idx++) {
145 			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
146 			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
147 				count++;
148 			vaddr += PMD_SIZE;
149 		}
150 		pmd_idx = 0;
151 	}
152 #endif
153 	return count;
154 }
155 
page_table_kmap_check(pte_t * pte,pmd_t * pmd,unsigned long vaddr,pte_t * lastpte,void ** adr)156 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
157 					   unsigned long vaddr, pte_t *lastpte,
158 					   void **adr)
159 {
160 #ifdef CONFIG_HIGHMEM
161 	/*
162 	 * Something (early fixmap) may already have put a pte
163 	 * page here, which causes the page table allocation
164 	 * to become nonlinear. Attempt to fix it, and if it
165 	 * is still nonlinear then we have to bug.
166 	 */
167 	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
168 	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
169 
170 	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
171 	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
172 	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
173 		pte_t *newpte;
174 		int i;
175 
176 		BUG_ON(after_bootmem);
177 		newpte = *adr;
178 		for (i = 0; i < PTRS_PER_PTE; i++)
179 			set_pte(newpte + i, pte[i]);
180 		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
181 
182 		paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
183 		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
184 		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
185 		__flush_tlb_all();
186 
187 		paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
188 		pte = newpte;
189 	}
190 	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
191 	       && vaddr > fix_to_virt(FIX_KMAP_END)
192 	       && lastpte && lastpte + PTRS_PER_PTE != pte);
193 #endif
194 	return pte;
195 }
196 
197 /*
198  * This function initializes a certain range of kernel virtual memory
199  * with new bootmem page tables, everywhere page tables are missing in
200  * the given range.
201  *
202  * NOTE: The pagetables are allocated contiguous on the physical space
203  * so we can cache the place of the first one and move around without
204  * checking the pgd every time.
205  */
206 static void __init
page_table_range_init(unsigned long start,unsigned long end,pgd_t * pgd_base)207 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
208 {
209 	int pgd_idx, pmd_idx;
210 	unsigned long vaddr;
211 	pgd_t *pgd;
212 	pmd_t *pmd;
213 	pte_t *pte = NULL;
214 	unsigned long count = page_table_range_init_count(start, end);
215 	void *adr = NULL;
216 
217 	if (count)
218 		adr = alloc_low_pages(count);
219 
220 	vaddr = start;
221 	pgd_idx = pgd_index(vaddr);
222 	pmd_idx = pmd_index(vaddr);
223 	pgd = pgd_base + pgd_idx;
224 
225 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
226 		pmd = one_md_table_init(pgd);
227 		pmd = pmd + pmd_index(vaddr);
228 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
229 							pmd++, pmd_idx++) {
230 			pte = page_table_kmap_check(one_page_table_init(pmd),
231 						    pmd, vaddr, pte, &adr);
232 
233 			vaddr += PMD_SIZE;
234 		}
235 		pmd_idx = 0;
236 	}
237 }
238 
is_kernel_text(unsigned long addr)239 static inline int is_kernel_text(unsigned long addr)
240 {
241 	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
242 		return 1;
243 	return 0;
244 }
245 
246 /*
247  * This maps the physical memory to kernel virtual address space, a total
248  * of max_low_pfn pages, by creating page tables starting from address
249  * PAGE_OFFSET:
250  */
251 unsigned long __init
kernel_physical_mapping_init(unsigned long start,unsigned long end,unsigned long page_size_mask)252 kernel_physical_mapping_init(unsigned long start,
253 			     unsigned long end,
254 			     unsigned long page_size_mask)
255 {
256 	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
257 	unsigned long last_map_addr = end;
258 	unsigned long start_pfn, end_pfn;
259 	pgd_t *pgd_base = swapper_pg_dir;
260 	int pgd_idx, pmd_idx, pte_ofs;
261 	unsigned long pfn;
262 	pgd_t *pgd;
263 	pmd_t *pmd;
264 	pte_t *pte;
265 	unsigned pages_2m, pages_4k;
266 	int mapping_iter;
267 
268 	start_pfn = start >> PAGE_SHIFT;
269 	end_pfn = end >> PAGE_SHIFT;
270 
271 	/*
272 	 * First iteration will setup identity mapping using large/small pages
273 	 * based on use_pse, with other attributes same as set by
274 	 * the early code in head_32.S
275 	 *
276 	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
277 	 * as desired for the kernel identity mapping.
278 	 *
279 	 * This two pass mechanism conforms to the TLB app note which says:
280 	 *
281 	 *     "Software should not write to a paging-structure entry in a way
282 	 *      that would change, for any linear address, both the page size
283 	 *      and either the page frame or attributes."
284 	 */
285 	mapping_iter = 1;
286 
287 	if (!cpu_has_pse)
288 		use_pse = 0;
289 
290 repeat:
291 	pages_2m = pages_4k = 0;
292 	pfn = start_pfn;
293 	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
294 	pgd = pgd_base + pgd_idx;
295 	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
296 		pmd = one_md_table_init(pgd);
297 
298 		if (pfn >= end_pfn)
299 			continue;
300 #ifdef CONFIG_X86_PAE
301 		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
302 		pmd += pmd_idx;
303 #else
304 		pmd_idx = 0;
305 #endif
306 		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
307 		     pmd++, pmd_idx++) {
308 			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
309 
310 			/*
311 			 * Map with big pages if possible, otherwise
312 			 * create normal page tables:
313 			 */
314 			if (use_pse) {
315 				unsigned int addr2;
316 				pgprot_t prot = PAGE_KERNEL_LARGE;
317 				/*
318 				 * first pass will use the same initial
319 				 * identity mapping attribute + _PAGE_PSE.
320 				 */
321 				pgprot_t init_prot =
322 					__pgprot(PTE_IDENT_ATTR |
323 						 _PAGE_PSE);
324 
325 				pfn &= PMD_MASK >> PAGE_SHIFT;
326 				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
327 					PAGE_OFFSET + PAGE_SIZE-1;
328 
329 				if (is_kernel_text(addr) ||
330 				    is_kernel_text(addr2))
331 					prot = PAGE_KERNEL_LARGE_EXEC;
332 
333 				pages_2m++;
334 				if (mapping_iter == 1)
335 					set_pmd(pmd, pfn_pmd(pfn, init_prot));
336 				else
337 					set_pmd(pmd, pfn_pmd(pfn, prot));
338 
339 				pfn += PTRS_PER_PTE;
340 				continue;
341 			}
342 			pte = one_page_table_init(pmd);
343 
344 			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
345 			pte += pte_ofs;
346 			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
347 			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
348 				pgprot_t prot = PAGE_KERNEL;
349 				/*
350 				 * first pass will use the same initial
351 				 * identity mapping attribute.
352 				 */
353 				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
354 
355 				if (is_kernel_text(addr))
356 					prot = PAGE_KERNEL_EXEC;
357 
358 				pages_4k++;
359 				if (mapping_iter == 1) {
360 					set_pte(pte, pfn_pte(pfn, init_prot));
361 					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
362 				} else
363 					set_pte(pte, pfn_pte(pfn, prot));
364 			}
365 		}
366 	}
367 	if (mapping_iter == 1) {
368 		/*
369 		 * update direct mapping page count only in the first
370 		 * iteration.
371 		 */
372 		update_page_count(PG_LEVEL_2M, pages_2m);
373 		update_page_count(PG_LEVEL_4K, pages_4k);
374 
375 		/*
376 		 * local global flush tlb, which will flush the previous
377 		 * mappings present in both small and large page TLB's.
378 		 */
379 		__flush_tlb_all();
380 
381 		/*
382 		 * Second iteration will set the actual desired PTE attributes.
383 		 */
384 		mapping_iter = 2;
385 		goto repeat;
386 	}
387 	return last_map_addr;
388 }
389 
390 pte_t *kmap_pte;
391 pgprot_t kmap_prot;
392 
kmap_get_fixmap_pte(unsigned long vaddr)393 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
394 {
395 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
396 			vaddr), vaddr), vaddr);
397 }
398 
kmap_init(void)399 static void __init kmap_init(void)
400 {
401 	unsigned long kmap_vstart;
402 
403 	/*
404 	 * Cache the first kmap pte:
405 	 */
406 	kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
407 	kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
408 
409 	kmap_prot = PAGE_KERNEL;
410 }
411 
412 #ifdef CONFIG_HIGHMEM
permanent_kmaps_init(pgd_t * pgd_base)413 static void __init permanent_kmaps_init(pgd_t *pgd_base)
414 {
415 	unsigned long vaddr;
416 	pgd_t *pgd;
417 	pud_t *pud;
418 	pmd_t *pmd;
419 	pte_t *pte;
420 
421 	vaddr = PKMAP_BASE;
422 	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
423 
424 	pgd = swapper_pg_dir + pgd_index(vaddr);
425 	pud = pud_offset(pgd, vaddr);
426 	pmd = pmd_offset(pud, vaddr);
427 	pte = pte_offset_kernel(pmd, vaddr);
428 	pkmap_page_table = pte;
429 }
430 
add_highpages_with_active_regions(int nid,unsigned long start_pfn,unsigned long end_pfn)431 void __init add_highpages_with_active_regions(int nid,
432 			 unsigned long start_pfn, unsigned long end_pfn)
433 {
434 	phys_addr_t start, end;
435 	u64 i;
436 
437 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
438 		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
439 					    start_pfn, end_pfn);
440 		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
441 					      start_pfn, end_pfn);
442 		for ( ; pfn < e_pfn; pfn++)
443 			if (pfn_valid(pfn))
444 				free_highmem_page(pfn_to_page(pfn));
445 	}
446 }
447 #else
permanent_kmaps_init(pgd_t * pgd_base)448 static inline void permanent_kmaps_init(pgd_t *pgd_base)
449 {
450 }
451 #endif /* CONFIG_HIGHMEM */
452 
native_pagetable_init(void)453 void __init native_pagetable_init(void)
454 {
455 	unsigned long pfn, va;
456 	pgd_t *pgd, *base = swapper_pg_dir;
457 	pud_t *pud;
458 	pmd_t *pmd;
459 	pte_t *pte;
460 
461 	/*
462 	 * Remove any mappings which extend past the end of physical
463 	 * memory from the boot time page table.
464 	 * In virtual address space, we should have at least two pages
465 	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
466 	 * definition. And max_low_pfn is set to VMALLOC_END physical
467 	 * address. If initial memory mapping is doing right job, we
468 	 * should have pte used near max_low_pfn or one pmd is not present.
469 	 */
470 	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
471 		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
472 		pgd = base + pgd_index(va);
473 		if (!pgd_present(*pgd))
474 			break;
475 
476 		pud = pud_offset(pgd, va);
477 		pmd = pmd_offset(pud, va);
478 		if (!pmd_present(*pmd))
479 			break;
480 
481 		/* should not be large page here */
482 		if (pmd_large(*pmd)) {
483 			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",
484 				pfn, pmd, __pa(pmd));
485 			BUG_ON(1);
486 		}
487 
488 		pte = pte_offset_kernel(pmd, va);
489 		if (!pte_present(*pte))
490 			break;
491 
492 		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
493 				pfn, pmd, __pa(pmd), pte, __pa(pte));
494 		pte_clear(NULL, va, pte);
495 	}
496 	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
497 	paging_init();
498 }
499 
500 /*
501  * Build a proper pagetable for the kernel mappings.  Up until this
502  * point, we've been running on some set of pagetables constructed by
503  * the boot process.
504  *
505  * If we're booting on native hardware, this will be a pagetable
506  * constructed in arch/x86/kernel/head_32.S.  The root of the
507  * pagetable will be swapper_pg_dir.
508  *
509  * If we're booting paravirtualized under a hypervisor, then there are
510  * more options: we may already be running PAE, and the pagetable may
511  * or may not be based in swapper_pg_dir.  In any case,
512  * paravirt_pagetable_init() will set up swapper_pg_dir
513  * appropriately for the rest of the initialization to work.
514  *
515  * In general, pagetable_init() assumes that the pagetable may already
516  * be partially populated, and so it avoids stomping on any existing
517  * mappings.
518  */
early_ioremap_page_table_range_init(void)519 void __init early_ioremap_page_table_range_init(void)
520 {
521 	pgd_t *pgd_base = swapper_pg_dir;
522 	unsigned long vaddr, end;
523 
524 	/*
525 	 * Fixed mappings, only the page table structure has to be
526 	 * created - mappings will be set by set_fixmap():
527 	 */
528 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
529 	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
530 	page_table_range_init(vaddr, end, pgd_base);
531 	early_ioremap_reset();
532 }
533 
pagetable_init(void)534 static void __init pagetable_init(void)
535 {
536 	pgd_t *pgd_base = swapper_pg_dir;
537 
538 	permanent_kmaps_init(pgd_base);
539 }
540 
541 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
542 EXPORT_SYMBOL_GPL(__supported_pte_mask);
543 
544 /* user-defined highmem size */
545 static unsigned int highmem_pages = -1;
546 
547 /*
548  * highmem=size forces highmem to be exactly 'size' bytes.
549  * This works even on boxes that have no highmem otherwise.
550  * This also works to reduce highmem size on bigger boxes.
551  */
parse_highmem(char * arg)552 static int __init parse_highmem(char *arg)
553 {
554 	if (!arg)
555 		return -EINVAL;
556 
557 	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
558 	return 0;
559 }
560 early_param("highmem", parse_highmem);
561 
562 #define MSG_HIGHMEM_TOO_BIG \
563 	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
564 
565 #define MSG_LOWMEM_TOO_SMALL \
566 	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
567 /*
568  * All of RAM fits into lowmem - but if user wants highmem
569  * artificially via the highmem=x boot parameter then create
570  * it:
571  */
lowmem_pfn_init(void)572 static void __init lowmem_pfn_init(void)
573 {
574 	/* max_low_pfn is 0, we already have early_res support */
575 	max_low_pfn = max_pfn;
576 
577 	if (highmem_pages == -1)
578 		highmem_pages = 0;
579 #ifdef CONFIG_HIGHMEM
580 	if (highmem_pages >= max_pfn) {
581 		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
582 			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
583 		highmem_pages = 0;
584 	}
585 	if (highmem_pages) {
586 		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
587 			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
588 				pages_to_mb(highmem_pages));
589 			highmem_pages = 0;
590 		}
591 		max_low_pfn -= highmem_pages;
592 	}
593 #else
594 	if (highmem_pages)
595 		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
596 #endif
597 }
598 
599 #define MSG_HIGHMEM_TOO_SMALL \
600 	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
601 
602 #define MSG_HIGHMEM_TRIMMED \
603 	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
604 /*
605  * We have more RAM than fits into lowmem - we try to put it into
606  * highmem, also taking the highmem=x boot parameter into account:
607  */
highmem_pfn_init(void)608 static void __init highmem_pfn_init(void)
609 {
610 	max_low_pfn = MAXMEM_PFN;
611 
612 	if (highmem_pages == -1)
613 		highmem_pages = max_pfn - MAXMEM_PFN;
614 
615 	if (highmem_pages + MAXMEM_PFN < max_pfn)
616 		max_pfn = MAXMEM_PFN + highmem_pages;
617 
618 	if (highmem_pages + MAXMEM_PFN > max_pfn) {
619 		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
620 			pages_to_mb(max_pfn - MAXMEM_PFN),
621 			pages_to_mb(highmem_pages));
622 		highmem_pages = 0;
623 	}
624 #ifndef CONFIG_HIGHMEM
625 	/* Maximum memory usable is what is directly addressable */
626 	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
627 	if (max_pfn > MAX_NONPAE_PFN)
628 		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
629 	else
630 		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
631 	max_pfn = MAXMEM_PFN;
632 #else /* !CONFIG_HIGHMEM */
633 #ifndef CONFIG_HIGHMEM64G
634 	if (max_pfn > MAX_NONPAE_PFN) {
635 		max_pfn = MAX_NONPAE_PFN;
636 		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
637 	}
638 #endif /* !CONFIG_HIGHMEM64G */
639 #endif /* !CONFIG_HIGHMEM */
640 }
641 
642 /*
643  * Determine low and high memory ranges:
644  */
find_low_pfn_range(void)645 void __init find_low_pfn_range(void)
646 {
647 	/* it could update max_pfn */
648 
649 	if (max_pfn <= MAXMEM_PFN)
650 		lowmem_pfn_init();
651 	else
652 		highmem_pfn_init();
653 }
654 
655 #ifndef CONFIG_NEED_MULTIPLE_NODES
initmem_init(void)656 void __init initmem_init(void)
657 {
658 #ifdef CONFIG_HIGHMEM
659 	highstart_pfn = highend_pfn = max_pfn;
660 	if (max_pfn > max_low_pfn)
661 		highstart_pfn = max_low_pfn;
662 	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
663 		pages_to_mb(highend_pfn - highstart_pfn));
664 	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
665 #else
666 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
667 #endif
668 
669 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
670 	sparse_memory_present_with_active_regions(0);
671 
672 #ifdef CONFIG_FLATMEM
673 	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
674 #endif
675 	__vmalloc_start_set = true;
676 
677 	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
678 			pages_to_mb(max_low_pfn));
679 
680 	setup_bootmem_allocator();
681 }
682 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
683 
setup_bootmem_allocator(void)684 void __init setup_bootmem_allocator(void)
685 {
686 	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
687 		 max_pfn_mapped<<PAGE_SHIFT);
688 	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
689 }
690 
691 /*
692  * paging_init() sets up the page tables - note that the first 8MB are
693  * already mapped by head.S.
694  *
695  * This routines also unmaps the page at virtual kernel address 0, so
696  * that we can trap those pesky NULL-reference errors in the kernel.
697  */
paging_init(void)698 void __init paging_init(void)
699 {
700 	pagetable_init();
701 
702 	__flush_tlb_all();
703 
704 	kmap_init();
705 
706 	/*
707 	 * NOTE: at this point the bootmem allocator is fully available.
708 	 */
709 	olpc_dt_build_devicetree();
710 	sparse_memory_present_with_active_regions(MAX_NUMNODES);
711 	sparse_init();
712 	zone_sizes_init();
713 }
714 
715 /*
716  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
717  * and also on some strange 486's. All 586+'s are OK. This used to involve
718  * black magic jumps to work around some nasty CPU bugs, but fortunately the
719  * switch to using exceptions got rid of all that.
720  */
test_wp_bit(void)721 static void __init test_wp_bit(void)
722 {
723 	printk(KERN_INFO
724   "Checking if this processor honours the WP bit even in supervisor mode...");
725 
726 	/* Any page-aligned address will do, the test is non-destructive */
727 	__set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_KERNEL_RO);
728 	boot_cpu_data.wp_works_ok = do_test_wp_bit();
729 	clear_fixmap(FIX_WP_TEST);
730 
731 	if (!boot_cpu_data.wp_works_ok) {
732 		printk(KERN_CONT "No.\n");
733 		panic("Linux doesn't support CPUs with broken WP.");
734 	} else {
735 		printk(KERN_CONT "Ok.\n");
736 	}
737 }
738 
mem_init(void)739 void __init mem_init(void)
740 {
741 	pci_iommu_alloc();
742 
743 #ifdef CONFIG_FLATMEM
744 	BUG_ON(!mem_map);
745 #endif
746 	/*
747 	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
748 	 * be done before free_all_bootmem(). Memblock use free low memory for
749 	 * temporary data (see find_range_array()) and for this purpose can use
750 	 * pages that was already passed to the buddy allocator, hence marked as
751 	 * not accessible in the page tables when compiled with
752 	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
753 	 * important here.
754 	 */
755 	set_highmem_pages_init();
756 
757 	/* this will put all low memory onto the freelists */
758 	free_all_bootmem();
759 
760 	after_bootmem = 1;
761 
762 	mem_init_print_info(NULL);
763 	printk(KERN_INFO "virtual kernel memory layout:\n"
764 		"    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
765 #ifdef CONFIG_HIGHMEM
766 		"    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
767 #endif
768 		"    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
769 		"    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
770 		"      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
771 		"      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
772 		"      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
773 		FIXADDR_START, FIXADDR_TOP,
774 		(FIXADDR_TOP - FIXADDR_START) >> 10,
775 
776 #ifdef CONFIG_HIGHMEM
777 		PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
778 		(LAST_PKMAP*PAGE_SIZE) >> 10,
779 #endif
780 
781 		VMALLOC_START, VMALLOC_END,
782 		(VMALLOC_END - VMALLOC_START) >> 20,
783 
784 		(unsigned long)__va(0), (unsigned long)high_memory,
785 		((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
786 
787 		(unsigned long)&__init_begin, (unsigned long)&__init_end,
788 		((unsigned long)&__init_end -
789 		 (unsigned long)&__init_begin) >> 10,
790 
791 		(unsigned long)&_etext, (unsigned long)&_edata,
792 		((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
793 
794 		(unsigned long)&_text, (unsigned long)&_etext,
795 		((unsigned long)&_etext - (unsigned long)&_text) >> 10);
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 #ifdef CONFIG_RANDOMIZE_BASE
811 	BUILD_BUG_ON(CONFIG_RANDOMIZE_BASE_MAX_OFFSET > KERNEL_IMAGE_SIZE);
812 #endif
813 
814 #ifdef CONFIG_HIGHMEM
815 	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
816 	BUG_ON(VMALLOC_END				> PKMAP_BASE);
817 #endif
818 	BUG_ON(VMALLOC_START				>= VMALLOC_END);
819 	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
820 
821 	if (boot_cpu_data.wp_works_ok < 0)
822 		test_wp_bit();
823 }
824 
825 #ifdef CONFIG_MEMORY_HOTPLUG
arch_add_memory(int nid,u64 start,u64 size,bool for_device)826 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
827 {
828 	struct pglist_data *pgdata = NODE_DATA(nid);
829 	struct zone *zone = pgdata->node_zones +
830 		zone_for_memory(nid, start, size, ZONE_HIGHMEM, for_device);
831 	unsigned long start_pfn = start >> PAGE_SHIFT;
832 	unsigned long nr_pages = size >> PAGE_SHIFT;
833 
834 	return __add_pages(nid, zone, start_pfn, nr_pages);
835 }
836 
837 #ifdef CONFIG_MEMORY_HOTREMOVE
arch_remove_memory(u64 start,u64 size)838 int arch_remove_memory(u64 start, u64 size)
839 {
840 	unsigned long start_pfn = start >> PAGE_SHIFT;
841 	unsigned long nr_pages = size >> PAGE_SHIFT;
842 	struct zone *zone;
843 
844 	zone = page_zone(pfn_to_page(start_pfn));
845 	return __remove_pages(zone, start_pfn, nr_pages);
846 }
847 #endif
848 #endif
849 
850 /*
851  * This function cannot be __init, since exceptions don't work in that
852  * section.  Put this after the callers, so that it cannot be inlined.
853  */
do_test_wp_bit(void)854 static noinline int do_test_wp_bit(void)
855 {
856 	char tmp_reg;
857 	int flag;
858 
859 	__asm__ __volatile__(
860 		"	movb %0, %1	\n"
861 		"1:	movb %1, %0	\n"
862 		"	xorl %2, %2	\n"
863 		"2:			\n"
864 		_ASM_EXTABLE(1b,2b)
865 		:"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
866 		 "=q" (tmp_reg),
867 		 "=r" (flag)
868 		:"2" (1)
869 		:"memory");
870 
871 	return flag;
872 }
873 
874 const int rodata_test_data = 0xC3;
875 EXPORT_SYMBOL_GPL(rodata_test_data);
876 
877 int kernel_set_to_readonly __read_mostly;
878 
set_kernel_text_rw(void)879 void set_kernel_text_rw(void)
880 {
881 	unsigned long start = PFN_ALIGN(_text);
882 	unsigned long size = PFN_ALIGN(_etext) - start;
883 
884 	if (!kernel_set_to_readonly)
885 		return;
886 
887 	pr_debug("Set kernel text: %lx - %lx for read write\n",
888 		 start, start+size);
889 
890 	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
891 }
892 
set_kernel_text_ro(void)893 void set_kernel_text_ro(void)
894 {
895 	unsigned long start = PFN_ALIGN(_text);
896 	unsigned long size = PFN_ALIGN(_etext) - start;
897 
898 	if (!kernel_set_to_readonly)
899 		return;
900 
901 	pr_debug("Set kernel text: %lx - %lx for read only\n",
902 		 start, start+size);
903 
904 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
905 }
906 
mark_nxdata_nx(void)907 static void mark_nxdata_nx(void)
908 {
909 	/*
910 	 * When this called, init has already been executed and released,
911 	 * so everything past _etext should be NX.
912 	 */
913 	unsigned long start = PFN_ALIGN(_etext);
914 	/*
915 	 * This comes from is_kernel_text upper limit. Also HPAGE where used:
916 	 */
917 	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
918 
919 	if (__supported_pte_mask & _PAGE_NX)
920 		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
921 	set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
922 }
923 
mark_rodata_ro(void)924 void mark_rodata_ro(void)
925 {
926 	unsigned long start = PFN_ALIGN(_text);
927 	unsigned long size = PFN_ALIGN(_etext) - start;
928 
929 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
930 	printk(KERN_INFO "Write protecting the kernel text: %luk\n",
931 		size >> 10);
932 
933 	kernel_set_to_readonly = 1;
934 
935 #ifdef CONFIG_CPA_DEBUG
936 	printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
937 		start, start+size);
938 	set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
939 
940 	printk(KERN_INFO "Testing CPA: write protecting again\n");
941 	set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
942 #endif
943 
944 	start += size;
945 	size = (unsigned long)__end_rodata - start;
946 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
947 	printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
948 		size >> 10);
949 	rodata_test();
950 
951 #ifdef CONFIG_CPA_DEBUG
952 	printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
953 	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
954 
955 	printk(KERN_INFO "Testing CPA: write protecting again\n");
956 	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
957 #endif
958 	mark_nxdata_nx();
959 	if (__supported_pte_mask & _PAGE_NX)
960 		debug_checkwx();
961 }
962