8008db2bddb36aaa071065dc273a8d2a65adc34a
[muen/linux.git] / arch / x86 / mm / init_32.c
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/signal.h>
9 #include <linux/sched.h>
10 #include <linux/kernel.h>
11 #include <linux/errno.h>
12 #include <linux/string.h>
13 #include <linux/types.h>
14 #include <linux/ptrace.h>
15 #include <linux/mman.h>
16 #include <linux/mm.h>
17 #include <linux/hugetlb.h>
18 #include <linux/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/pci.h>
24 #include <linux/pfn.h>
25 #include <linux/poison.h>
26 #include <linux/bootmem.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
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 static inline int is_kernel_text(unsigned long addr)
241 {
242         if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
243                 return 1;
244         return 0;
245 }
246
247 /*
248  * This maps the physical memory to kernel virtual address space, a total
249  * of max_low_pfn pages, by creating page tables starting from address
250  * PAGE_OFFSET:
251  */
252 unsigned long __init
253 kernel_physical_mapping_init(unsigned long start,
254                              unsigned long end,
255                              unsigned long page_size_mask)
256 {
257         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
258         unsigned long last_map_addr = end;
259         unsigned long start_pfn, end_pfn;
260         pgd_t *pgd_base = swapper_pg_dir;
261         int pgd_idx, pmd_idx, pte_ofs;
262         unsigned long pfn;
263         pgd_t *pgd;
264         pmd_t *pmd;
265         pte_t *pte;
266         unsigned pages_2m, pages_4k;
267         int mapping_iter;
268
269         start_pfn = start >> PAGE_SHIFT;
270         end_pfn = end >> PAGE_SHIFT;
271
272         /*
273          * First iteration will setup identity mapping using large/small pages
274          * based on use_pse, with other attributes same as set by
275          * the early code in head_32.S
276          *
277          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
278          * as desired for the kernel identity mapping.
279          *
280          * This two pass mechanism conforms to the TLB app note which says:
281          *
282          *     "Software should not write to a paging-structure entry in a way
283          *      that would change, for any linear address, both the page size
284          *      and either the page frame or attributes."
285          */
286         mapping_iter = 1;
287
288         if (!boot_cpu_has(X86_FEATURE_PSE))
289                 use_pse = 0;
290
291 repeat:
292         pages_2m = pages_4k = 0;
293         pfn = start_pfn;
294         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
295         pgd = pgd_base + pgd_idx;
296         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
297                 pmd = one_md_table_init(pgd);
298
299                 if (pfn >= end_pfn)
300                         continue;
301 #ifdef CONFIG_X86_PAE
302                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
303                 pmd += pmd_idx;
304 #else
305                 pmd_idx = 0;
306 #endif
307                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
308                      pmd++, pmd_idx++) {
309                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
310
311                         /*
312                          * Map with big pages if possible, otherwise
313                          * create normal page tables:
314                          */
315                         if (use_pse) {
316                                 unsigned int addr2;
317                                 pgprot_t prot = PAGE_KERNEL_LARGE;
318                                 /*
319                                  * first pass will use the same initial
320                                  * identity mapping attribute + _PAGE_PSE.
321                                  */
322                                 pgprot_t init_prot =
323                                         __pgprot(PTE_IDENT_ATTR |
324                                                  _PAGE_PSE);
325
326                                 pfn &= PMD_MASK >> PAGE_SHIFT;
327                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
328                                         PAGE_OFFSET + PAGE_SIZE-1;
329
330                                 if (is_kernel_text(addr) ||
331                                     is_kernel_text(addr2))
332                                         prot = PAGE_KERNEL_LARGE_EXEC;
333
334                                 pages_2m++;
335                                 if (mapping_iter == 1)
336                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
337                                 else
338                                         set_pmd(pmd, pfn_pmd(pfn, prot));
339
340                                 pfn += PTRS_PER_PTE;
341                                 continue;
342                         }
343                         pte = one_page_table_init(pmd);
344
345                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
346                         pte += pte_ofs;
347                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
348                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
349                                 pgprot_t prot = PAGE_KERNEL;
350                                 /*
351                                  * first pass will use the same initial
352                                  * identity mapping attribute.
353                                  */
354                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
355
356                                 if (is_kernel_text(addr))
357                                         prot = PAGE_KERNEL_EXEC;
358
359                                 pages_4k++;
360                                 if (mapping_iter == 1) {
361                                         set_pte(pte, pfn_pte(pfn, init_prot));
362                                         last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
363                                 } else
364                                         set_pte(pte, pfn_pte(pfn, prot));
365                         }
366                 }
367         }
368         if (mapping_iter == 1) {
369                 /*
370                  * update direct mapping page count only in the first
371                  * iteration.
372                  */
373                 update_page_count(PG_LEVEL_2M, pages_2m);
374                 update_page_count(PG_LEVEL_4K, pages_4k);
375
376                 /*
377                  * local global flush tlb, which will flush the previous
378                  * mappings present in both small and large page TLB's.
379                  */
380                 __flush_tlb_all();
381
382                 /*
383                  * Second iteration will set the actual desired PTE attributes.
384                  */
385                 mapping_iter = 2;
386                 goto repeat;
387         }
388         return last_map_addr;
389 }
390
391 pte_t *kmap_pte;
392
393 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
394 {
395         pgd_t *pgd = pgd_offset_k(vaddr);
396         p4d_t *p4d = p4d_offset(pgd, vaddr);
397         pud_t *pud = pud_offset(p4d, vaddr);
398         pmd_t *pmd = pmd_offset(pud, vaddr);
399         return pte_offset_kernel(pmd, vaddr);
400 }
401
402 static void __init kmap_init(void)
403 {
404         unsigned long kmap_vstart;
405
406         /*
407          * Cache the first kmap pte:
408          */
409         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
410         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
411 }
412
413 #ifdef CONFIG_HIGHMEM
414 static void __init permanent_kmaps_init(pgd_t *pgd_base)
415 {
416         unsigned long vaddr;
417         pgd_t *pgd;
418         p4d_t *p4d;
419         pud_t *pud;
420         pmd_t *pmd;
421         pte_t *pte;
422
423         vaddr = PKMAP_BASE;
424         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
425
426         pgd = swapper_pg_dir + pgd_index(vaddr);
427         p4d = p4d_offset(pgd, vaddr);
428         pud = pud_offset(p4d, vaddr);
429         pmd = pmd_offset(pud, vaddr);
430         pte = pte_offset_kernel(pmd, vaddr);
431         pkmap_page_table = pte;
432 }
433
434 void __init add_highpages_with_active_regions(int nid,
435                          unsigned long start_pfn, unsigned long end_pfn)
436 {
437         phys_addr_t start, end;
438         u64 i;
439
440         for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
441                 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
442                                             start_pfn, end_pfn);
443                 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
444                                               start_pfn, end_pfn);
445                 for ( ; pfn < e_pfn; pfn++)
446                         if (pfn_valid(pfn))
447                                 free_highmem_page(pfn_to_page(pfn));
448         }
449 }
450 #else
451 static inline void permanent_kmaps_init(pgd_t *pgd_base)
452 {
453 }
454 #endif /* CONFIG_HIGHMEM */
455
456 void __init sync_initial_page_table(void)
457 {
458         clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
459                         swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
460                         KERNEL_PGD_PTRS);
461
462         /*
463          * sync back low identity map too.  It is used for example
464          * in the 32-bit EFI stub.
465          */
466         clone_pgd_range(initial_page_table,
467                         swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
468                         min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
469 }
470
471 void __init native_pagetable_init(void)
472 {
473         unsigned long pfn, va;
474         pgd_t *pgd, *base = swapper_pg_dir;
475         p4d_t *p4d;
476         pud_t *pud;
477         pmd_t *pmd;
478         pte_t *pte;
479
480         /*
481          * Remove any mappings which extend past the end of physical
482          * memory from the boot time page table.
483          * In virtual address space, we should have at least two pages
484          * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
485          * definition. And max_low_pfn is set to VMALLOC_END physical
486          * address. If initial memory mapping is doing right job, we
487          * should have pte used near max_low_pfn or one pmd is not present.
488          */
489         for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
490                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
491                 pgd = base + pgd_index(va);
492                 if (!pgd_present(*pgd))
493                         break;
494
495                 p4d = p4d_offset(pgd, va);
496                 pud = pud_offset(p4d, va);
497                 pmd = pmd_offset(pud, va);
498                 if (!pmd_present(*pmd))
499                         break;
500
501                 /* should not be large page here */
502                 if (pmd_large(*pmd)) {
503                         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",
504                                 pfn, pmd, __pa(pmd));
505                         BUG_ON(1);
506                 }
507
508                 pte = pte_offset_kernel(pmd, va);
509                 if (!pte_present(*pte))
510                         break;
511
512                 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
513                                 pfn, pmd, __pa(pmd), pte, __pa(pte));
514                 pte_clear(NULL, va, pte);
515         }
516         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
517         paging_init();
518 }
519
520 /*
521  * Build a proper pagetable for the kernel mappings.  Up until this
522  * point, we've been running on some set of pagetables constructed by
523  * the boot process.
524  *
525  * If we're booting on native hardware, this will be a pagetable
526  * constructed in arch/x86/kernel/head_32.S.  The root of the
527  * pagetable will be swapper_pg_dir.
528  *
529  * If we're booting paravirtualized under a hypervisor, then there are
530  * more options: we may already be running PAE, and the pagetable may
531  * or may not be based in swapper_pg_dir.  In any case,
532  * paravirt_pagetable_init() will set up swapper_pg_dir
533  * appropriately for the rest of the initialization to work.
534  *
535  * In general, pagetable_init() assumes that the pagetable may already
536  * be partially populated, and so it avoids stomping on any existing
537  * mappings.
538  */
539 void __init early_ioremap_page_table_range_init(void)
540 {
541         pgd_t *pgd_base = swapper_pg_dir;
542         unsigned long vaddr, end;
543
544         /*
545          * Fixed mappings, only the page table structure has to be
546          * created - mappings will be set by set_fixmap():
547          */
548         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
549         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
550         page_table_range_init(vaddr, end, pgd_base);
551         early_ioremap_reset();
552 }
553
554 static void __init pagetable_init(void)
555 {
556         pgd_t *pgd_base = swapper_pg_dir;
557
558         permanent_kmaps_init(pgd_base);
559 }
560
561 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
562 EXPORT_SYMBOL_GPL(__supported_pte_mask);
563
564 /* user-defined highmem size */
565 static unsigned int highmem_pages = -1;
566
567 /*
568  * highmem=size forces highmem to be exactly 'size' bytes.
569  * This works even on boxes that have no highmem otherwise.
570  * This also works to reduce highmem size on bigger boxes.
571  */
572 static int __init parse_highmem(char *arg)
573 {
574         if (!arg)
575                 return -EINVAL;
576
577         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
578         return 0;
579 }
580 early_param("highmem", parse_highmem);
581
582 #define MSG_HIGHMEM_TOO_BIG \
583         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
584
585 #define MSG_LOWMEM_TOO_SMALL \
586         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
587 /*
588  * All of RAM fits into lowmem - but if user wants highmem
589  * artificially via the highmem=x boot parameter then create
590  * it:
591  */
592 static void __init lowmem_pfn_init(void)
593 {
594         /* max_low_pfn is 0, we already have early_res support */
595         max_low_pfn = max_pfn;
596
597         if (highmem_pages == -1)
598                 highmem_pages = 0;
599 #ifdef CONFIG_HIGHMEM
600         if (highmem_pages >= max_pfn) {
601                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
602                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
603                 highmem_pages = 0;
604         }
605         if (highmem_pages) {
606                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
607                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
608                                 pages_to_mb(highmem_pages));
609                         highmem_pages = 0;
610                 }
611                 max_low_pfn -= highmem_pages;
612         }
613 #else
614         if (highmem_pages)
615                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
616 #endif
617 }
618
619 #define MSG_HIGHMEM_TOO_SMALL \
620         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
621
622 #define MSG_HIGHMEM_TRIMMED \
623         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
624 /*
625  * We have more RAM than fits into lowmem - we try to put it into
626  * highmem, also taking the highmem=x boot parameter into account:
627  */
628 static void __init highmem_pfn_init(void)
629 {
630         max_low_pfn = MAXMEM_PFN;
631
632         if (highmem_pages == -1)
633                 highmem_pages = max_pfn - MAXMEM_PFN;
634
635         if (highmem_pages + MAXMEM_PFN < max_pfn)
636                 max_pfn = MAXMEM_PFN + highmem_pages;
637
638         if (highmem_pages + MAXMEM_PFN > max_pfn) {
639                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
640                         pages_to_mb(max_pfn - MAXMEM_PFN),
641                         pages_to_mb(highmem_pages));
642                 highmem_pages = 0;
643         }
644 #ifndef CONFIG_HIGHMEM
645         /* Maximum memory usable is what is directly addressable */
646         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
647         if (max_pfn > MAX_NONPAE_PFN)
648                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
649         else
650                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
651         max_pfn = MAXMEM_PFN;
652 #else /* !CONFIG_HIGHMEM */
653 #ifndef CONFIG_HIGHMEM64G
654         if (max_pfn > MAX_NONPAE_PFN) {
655                 max_pfn = MAX_NONPAE_PFN;
656                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
657         }
658 #endif /* !CONFIG_HIGHMEM64G */
659 #endif /* !CONFIG_HIGHMEM */
660 }
661
662 /*
663  * Determine low and high memory ranges:
664  */
665 void __init find_low_pfn_range(void)
666 {
667         /* it could update max_pfn */
668
669         if (max_pfn <= MAXMEM_PFN)
670                 lowmem_pfn_init();
671         else
672                 highmem_pfn_init();
673 }
674
675 #ifndef CONFIG_NEED_MULTIPLE_NODES
676 void __init initmem_init(void)
677 {
678 #ifdef CONFIG_HIGHMEM
679         highstart_pfn = highend_pfn = max_pfn;
680         if (max_pfn > max_low_pfn)
681                 highstart_pfn = max_low_pfn;
682         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
683                 pages_to_mb(highend_pfn - highstart_pfn));
684         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
685 #else
686         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
687 #endif
688
689         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
690         sparse_memory_present_with_active_regions(0);
691
692 #ifdef CONFIG_FLATMEM
693         max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
694 #endif
695         __vmalloc_start_set = true;
696
697         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
698                         pages_to_mb(max_low_pfn));
699
700         setup_bootmem_allocator();
701 }
702 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
703
704 void __init setup_bootmem_allocator(void)
705 {
706         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
707                  max_pfn_mapped<<PAGE_SHIFT);
708         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
709 }
710
711 /*
712  * paging_init() sets up the page tables - note that the first 8MB are
713  * already mapped by head.S.
714  *
715  * This routines also unmaps the page at virtual kernel address 0, so
716  * that we can trap those pesky NULL-reference errors in the kernel.
717  */
718 void __init paging_init(void)
719 {
720         pagetable_init();
721
722         __flush_tlb_all();
723
724         kmap_init();
725
726         /*
727          * NOTE: at this point the bootmem allocator is fully available.
728          */
729         olpc_dt_build_devicetree();
730         sparse_memory_present_with_active_regions(MAX_NUMNODES);
731         sparse_init();
732         zone_sizes_init();
733 }
734
735 /*
736  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
737  * and also on some strange 486's. All 586+'s are OK. This used to involve
738  * black magic jumps to work around some nasty CPU bugs, but fortunately the
739  * switch to using exceptions got rid of all that.
740  */
741 static void __init test_wp_bit(void)
742 {
743         char z = 0;
744
745         printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
746
747         __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
748
749         if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
750                 clear_fixmap(FIX_WP_TEST);
751                 printk(KERN_CONT "Ok.\n");
752                 return;
753         }
754
755         printk(KERN_CONT "No.\n");
756         panic("Linux doesn't support CPUs with broken WP.");
757 }
758
759 void __init mem_init(void)
760 {
761         pci_iommu_alloc();
762
763 #ifdef CONFIG_FLATMEM
764         BUG_ON(!mem_map);
765 #endif
766         /*
767          * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
768          * be done before free_all_bootmem(). Memblock use free low memory for
769          * temporary data (see find_range_array()) and for this purpose can use
770          * pages that was already passed to the buddy allocator, hence marked as
771          * not accessible in the page tables when compiled with
772          * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
773          * important here.
774          */
775         set_highmem_pages_init();
776
777         /* this will put all low memory onto the freelists */
778         free_all_bootmem();
779
780         after_bootmem = 1;
781         x86_init.hyper.init_after_bootmem();
782
783         mem_init_print_info(NULL);
784         printk(KERN_INFO "virtual kernel memory layout:\n"
785                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
786                 "  cpu_entry : 0x%08lx - 0x%08lx   (%4ld kB)\n"
787 #ifdef CONFIG_HIGHMEM
788                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
789 #endif
790                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
791                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
792                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
793                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
794                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
795                 FIXADDR_START, FIXADDR_TOP,
796                 (FIXADDR_TOP - FIXADDR_START) >> 10,
797
798                 CPU_ENTRY_AREA_BASE,
799                 CPU_ENTRY_AREA_BASE + CPU_ENTRY_AREA_MAP_SIZE,
800                 CPU_ENTRY_AREA_MAP_SIZE >> 10,
801
802 #ifdef CONFIG_HIGHMEM
803                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
804                 (LAST_PKMAP*PAGE_SIZE) >> 10,
805 #endif
806
807                 VMALLOC_START, VMALLOC_END,
808                 (VMALLOC_END - VMALLOC_START) >> 20,
809
810                 (unsigned long)__va(0), (unsigned long)high_memory,
811                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
812
813                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
814                 ((unsigned long)&__init_end -
815                  (unsigned long)&__init_begin) >> 10,
816
817                 (unsigned long)&_etext, (unsigned long)&_edata,
818                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
819
820                 (unsigned long)&_text, (unsigned long)&_etext,
821                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
822
823         /*
824          * Check boundaries twice: Some fundamental inconsistencies can
825          * be detected at build time already.
826          */
827 #define __FIXADDR_TOP (-PAGE_SIZE)
828 #ifdef CONFIG_HIGHMEM
829         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
830         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
831 #endif
832 #define high_memory (-128UL << 20)
833         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
834 #undef high_memory
835 #undef __FIXADDR_TOP
836
837 #ifdef CONFIG_HIGHMEM
838         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
839         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
840 #endif
841         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
842         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
843
844         test_wp_bit();
845 }
846
847 #ifdef CONFIG_MEMORY_HOTPLUG
848 int arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
849                 bool want_memblock)
850 {
851         unsigned long start_pfn = start >> PAGE_SHIFT;
852         unsigned long nr_pages = size >> PAGE_SHIFT;
853
854         return __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
855 }
856
857 #ifdef CONFIG_MEMORY_HOTREMOVE
858 int arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
859 {
860         unsigned long start_pfn = start >> PAGE_SHIFT;
861         unsigned long nr_pages = size >> PAGE_SHIFT;
862         struct zone *zone;
863
864         zone = page_zone(pfn_to_page(start_pfn));
865         return __remove_pages(zone, start_pfn, nr_pages, altmap);
866 }
867 #endif
868 #endif
869
870 int kernel_set_to_readonly __read_mostly;
871
872 void set_kernel_text_rw(void)
873 {
874         unsigned long start = PFN_ALIGN(_text);
875         unsigned long size = PFN_ALIGN(_etext) - start;
876
877         if (!kernel_set_to_readonly)
878                 return;
879
880         pr_debug("Set kernel text: %lx - %lx for read write\n",
881                  start, start+size);
882
883         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
884 }
885
886 void set_kernel_text_ro(void)
887 {
888         unsigned long start = PFN_ALIGN(_text);
889         unsigned long size = PFN_ALIGN(_etext) - start;
890
891         if (!kernel_set_to_readonly)
892                 return;
893
894         pr_debug("Set kernel text: %lx - %lx for read only\n",
895                  start, start+size);
896
897         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
898 }
899
900 static void mark_nxdata_nx(void)
901 {
902         /*
903          * When this called, init has already been executed and released,
904          * so everything past _etext should be NX.
905          */
906         unsigned long start = PFN_ALIGN(_etext);
907         /*
908          * This comes from is_kernel_text upper limit. Also HPAGE where used:
909          */
910         unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
911
912         if (__supported_pte_mask & _PAGE_NX)
913                 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
914         set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
915 }
916
917 void mark_rodata_ro(void)
918 {
919         unsigned long start = PFN_ALIGN(_text);
920         unsigned long size = PFN_ALIGN(_etext) - start;
921
922         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
923         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
924                 size >> 10);
925
926         kernel_set_to_readonly = 1;
927
928 #ifdef CONFIG_CPA_DEBUG
929         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
930                 start, start+size);
931         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
932
933         printk(KERN_INFO "Testing CPA: write protecting again\n");
934         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
935 #endif
936
937         start += size;
938         size = (unsigned long)__end_rodata - start;
939         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
940         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
941                 size >> 10);
942
943 #ifdef CONFIG_CPA_DEBUG
944         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
945         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
946
947         printk(KERN_INFO "Testing CPA: write protecting again\n");
948         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
949 #endif
950         mark_nxdata_nx();
951         if (__supported_pte_mask & _PAGE_NX)
952                 debug_checkwx();
953 }