2 #include <linux/initrd.h>
3 #include <linux/ioport.h>
4 #include <linux/swap.h>
5 #include <linux/memblock.h>
6 #include <linux/bootmem.h> /* for max_low_pfn */
8 #include <asm/cacheflush.h>
12 #include <asm/page_types.h>
13 #include <asm/sections.h>
14 #include <asm/setup.h>
15 #include <asm/tlbflush.h>
17 #include <asm/proto.h>
18 #include <asm/dma.h> /* for MAX_DMA_PFN */
19 #include <asm/microcode.h>
20 #include <asm/kaslr.h>
23 * We need to define the tracepoints somewhere, and tlb.c
24 * is only compied when SMP=y.
26 #define CREATE_TRACE_POINTS
27 #include <trace/events/tlb.h>
29 #include "mm_internal.h"
32 * Tables translating between page_cache_type_t and pte encoding.
34 * The default values are defined statically as minimal supported mode;
35 * WC and WT fall back to UC-. pat_init() updates these values to support
36 * more cache modes, WC and WT, when it is safe to do so. See pat_init()
37 * for the details. Note, __early_ioremap() used during early boot-time
38 * takes pgprot_t (pte encoding) and does not use these tables.
40 * Index into __cachemode2pte_tbl[] is the cachemode.
42 * Index into __pte2cachemode_tbl[] are the caching attribute bits of the pte
43 * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2.
45 uint16_t __cachemode2pte_tbl
[_PAGE_CACHE_MODE_NUM
] = {
46 [_PAGE_CACHE_MODE_WB
] = 0 | 0 ,
47 [_PAGE_CACHE_MODE_WC
] = 0 | _PAGE_PCD
,
48 [_PAGE_CACHE_MODE_UC_MINUS
] = 0 | _PAGE_PCD
,
49 [_PAGE_CACHE_MODE_UC
] = _PAGE_PWT
| _PAGE_PCD
,
50 [_PAGE_CACHE_MODE_WT
] = 0 | _PAGE_PCD
,
51 [_PAGE_CACHE_MODE_WP
] = 0 | _PAGE_PCD
,
53 EXPORT_SYMBOL(__cachemode2pte_tbl
);
55 uint8_t __pte2cachemode_tbl
[8] = {
56 [__pte2cm_idx( 0 | 0 | 0 )] = _PAGE_CACHE_MODE_WB
,
57 [__pte2cm_idx(_PAGE_PWT
| 0 | 0 )] = _PAGE_CACHE_MODE_UC_MINUS
,
58 [__pte2cm_idx( 0 | _PAGE_PCD
| 0 )] = _PAGE_CACHE_MODE_UC_MINUS
,
59 [__pte2cm_idx(_PAGE_PWT
| _PAGE_PCD
| 0 )] = _PAGE_CACHE_MODE_UC
,
60 [__pte2cm_idx( 0 | 0 | _PAGE_PAT
)] = _PAGE_CACHE_MODE_WB
,
61 [__pte2cm_idx(_PAGE_PWT
| 0 | _PAGE_PAT
)] = _PAGE_CACHE_MODE_UC_MINUS
,
62 [__pte2cm_idx(0 | _PAGE_PCD
| _PAGE_PAT
)] = _PAGE_CACHE_MODE_UC_MINUS
,
63 [__pte2cm_idx(_PAGE_PWT
| _PAGE_PCD
| _PAGE_PAT
)] = _PAGE_CACHE_MODE_UC
,
65 EXPORT_SYMBOL(__pte2cachemode_tbl
);
67 static unsigned long __initdata pgt_buf_start
;
68 static unsigned long __initdata pgt_buf_end
;
69 static unsigned long __initdata pgt_buf_top
;
71 static unsigned long min_pfn_mapped
;
73 static bool __initdata can_use_brk_pgt
= true;
76 * Pages returned are already directly mapped.
78 * Changing that is likely to break Xen, see commit:
80 * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
82 * for detailed information.
84 __ref
void *alloc_low_pages(unsigned int num
)
92 order
= get_order((unsigned long)num
<< PAGE_SHIFT
);
93 return (void *)__get_free_pages(GFP_ATOMIC
| __GFP_NOTRACK
|
97 if ((pgt_buf_end
+ num
) > pgt_buf_top
|| !can_use_brk_pgt
) {
99 if (min_pfn_mapped
>= max_pfn_mapped
)
100 panic("alloc_low_pages: ran out of memory");
101 ret
= memblock_find_in_range(min_pfn_mapped
<< PAGE_SHIFT
,
102 max_pfn_mapped
<< PAGE_SHIFT
,
103 PAGE_SIZE
* num
, PAGE_SIZE
);
105 panic("alloc_low_pages: can not alloc memory");
106 memblock_reserve(ret
, PAGE_SIZE
* num
);
107 pfn
= ret
>> PAGE_SHIFT
;
111 printk(KERN_DEBUG
"BRK [%#010lx, %#010lx] PGTABLE\n",
112 pfn
<< PAGE_SHIFT
, (pgt_buf_end
<< PAGE_SHIFT
) - 1);
115 for (i
= 0; i
< num
; i
++) {
118 adr
= __va((pfn
+ i
) << PAGE_SHIFT
);
122 return __va(pfn
<< PAGE_SHIFT
);
125 /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */
126 #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE)
127 RESERVE_BRK(early_pgt_alloc
, INIT_PGT_BUF_SIZE
);
128 void __init
early_alloc_pgt_buf(void)
130 unsigned long tables
= INIT_PGT_BUF_SIZE
;
133 base
= __pa(extend_brk(tables
, PAGE_SIZE
));
135 pgt_buf_start
= base
>> PAGE_SHIFT
;
136 pgt_buf_end
= pgt_buf_start
;
137 pgt_buf_top
= pgt_buf_start
+ (tables
>> PAGE_SHIFT
);
142 early_param_on_off("gbpages", "nogbpages", direct_gbpages
, CONFIG_X86_DIRECT_GBPAGES
);
147 unsigned page_size_mask
;
150 static int page_size_mask
;
152 static void __init
probe_page_size_mask(void)
154 #if !defined(CONFIG_KMEMCHECK)
156 * For CONFIG_KMEMCHECK or pagealloc debugging, identity mapping will
158 * This will simplify cpa(), which otherwise needs to support splitting
159 * large pages into small in interrupt context, etc.
161 if (boot_cpu_has(X86_FEATURE_PSE
) && !debug_pagealloc_enabled())
162 page_size_mask
|= 1 << PG_LEVEL_2M
;
165 /* Enable PSE if available */
166 if (boot_cpu_has(X86_FEATURE_PSE
))
167 cr4_set_bits_and_update_boot(X86_CR4_PSE
);
169 /* Enable PGE if available */
170 if (boot_cpu_has(X86_FEATURE_PGE
)) {
171 cr4_set_bits_and_update_boot(X86_CR4_PGE
);
172 __supported_pte_mask
|= _PAGE_GLOBAL
;
174 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
176 /* Enable 1 GB linear kernel mappings if available: */
177 if (direct_gbpages
&& boot_cpu_has(X86_FEATURE_GBPAGES
)) {
178 printk(KERN_INFO
"Using GB pages for direct mapping\n");
179 page_size_mask
|= 1 << PG_LEVEL_1G
;
186 #define NR_RANGE_MR 3
187 #else /* CONFIG_X86_64 */
188 #define NR_RANGE_MR 5
191 static int __meminit
save_mr(struct map_range
*mr
, int nr_range
,
192 unsigned long start_pfn
, unsigned long end_pfn
,
193 unsigned long page_size_mask
)
195 if (start_pfn
< end_pfn
) {
196 if (nr_range
>= NR_RANGE_MR
)
197 panic("run out of range for init_memory_mapping\n");
198 mr
[nr_range
].start
= start_pfn
<<PAGE_SHIFT
;
199 mr
[nr_range
].end
= end_pfn
<<PAGE_SHIFT
;
200 mr
[nr_range
].page_size_mask
= page_size_mask
;
208 * adjust the page_size_mask for small range to go with
209 * big page size instead small one if nearby are ram too.
211 static void __init_refok
adjust_range_page_size_mask(struct map_range
*mr
,
216 for (i
= 0; i
< nr_range
; i
++) {
217 if ((page_size_mask
& (1<<PG_LEVEL_2M
)) &&
218 !(mr
[i
].page_size_mask
& (1<<PG_LEVEL_2M
))) {
219 unsigned long start
= round_down(mr
[i
].start
, PMD_SIZE
);
220 unsigned long end
= round_up(mr
[i
].end
, PMD_SIZE
);
223 if ((end
>> PAGE_SHIFT
) > max_low_pfn
)
227 if (memblock_is_region_memory(start
, end
- start
))
228 mr
[i
].page_size_mask
|= 1<<PG_LEVEL_2M
;
230 if ((page_size_mask
& (1<<PG_LEVEL_1G
)) &&
231 !(mr
[i
].page_size_mask
& (1<<PG_LEVEL_1G
))) {
232 unsigned long start
= round_down(mr
[i
].start
, PUD_SIZE
);
233 unsigned long end
= round_up(mr
[i
].end
, PUD_SIZE
);
235 if (memblock_is_region_memory(start
, end
- start
))
236 mr
[i
].page_size_mask
|= 1<<PG_LEVEL_1G
;
241 static const char *page_size_string(struct map_range
*mr
)
243 static const char str_1g
[] = "1G";
244 static const char str_2m
[] = "2M";
245 static const char str_4m
[] = "4M";
246 static const char str_4k
[] = "4k";
248 if (mr
->page_size_mask
& (1<<PG_LEVEL_1G
))
251 * 32-bit without PAE has a 4M large page size.
252 * PG_LEVEL_2M is misnamed, but we can at least
253 * print out the right size in the string.
255 if (IS_ENABLED(CONFIG_X86_32
) &&
256 !IS_ENABLED(CONFIG_X86_PAE
) &&
257 mr
->page_size_mask
& (1<<PG_LEVEL_2M
))
260 if (mr
->page_size_mask
& (1<<PG_LEVEL_2M
))
266 static int __meminit
split_mem_range(struct map_range
*mr
, int nr_range
,
270 unsigned long start_pfn
, end_pfn
, limit_pfn
;
274 limit_pfn
= PFN_DOWN(end
);
276 /* head if not big page alignment ? */
277 pfn
= start_pfn
= PFN_DOWN(start
);
280 * Don't use a large page for the first 2/4MB of memory
281 * because there are often fixed size MTRRs in there
282 * and overlapping MTRRs into large pages can cause
286 end_pfn
= PFN_DOWN(PMD_SIZE
);
288 end_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
289 #else /* CONFIG_X86_64 */
290 end_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
292 if (end_pfn
> limit_pfn
)
294 if (start_pfn
< end_pfn
) {
295 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
299 /* big page (2M) range */
300 start_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
302 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
));
303 #else /* CONFIG_X86_64 */
304 end_pfn
= round_up(pfn
, PFN_DOWN(PUD_SIZE
));
305 if (end_pfn
> round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
)))
306 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
));
309 if (start_pfn
< end_pfn
) {
310 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
311 page_size_mask
& (1<<PG_LEVEL_2M
));
316 /* big page (1G) range */
317 start_pfn
= round_up(pfn
, PFN_DOWN(PUD_SIZE
));
318 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PUD_SIZE
));
319 if (start_pfn
< end_pfn
) {
320 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
322 ((1<<PG_LEVEL_2M
)|(1<<PG_LEVEL_1G
)));
326 /* tail is not big page (1G) alignment */
327 start_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
328 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
));
329 if (start_pfn
< end_pfn
) {
330 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
331 page_size_mask
& (1<<PG_LEVEL_2M
));
336 /* tail is not big page (2M) alignment */
339 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
342 adjust_range_page_size_mask(mr
, nr_range
);
344 /* try to merge same page size and continuous */
345 for (i
= 0; nr_range
> 1 && i
< nr_range
- 1; i
++) {
346 unsigned long old_start
;
347 if (mr
[i
].end
!= mr
[i
+1].start
||
348 mr
[i
].page_size_mask
!= mr
[i
+1].page_size_mask
)
351 old_start
= mr
[i
].start
;
352 memmove(&mr
[i
], &mr
[i
+1],
353 (nr_range
- 1 - i
) * sizeof(struct map_range
));
354 mr
[i
--].start
= old_start
;
358 for (i
= 0; i
< nr_range
; i
++)
359 pr_debug(" [mem %#010lx-%#010lx] page %s\n",
360 mr
[i
].start
, mr
[i
].end
- 1,
361 page_size_string(&mr
[i
]));
366 struct range pfn_mapped
[E820_X_MAX
];
369 static void add_pfn_range_mapped(unsigned long start_pfn
, unsigned long end_pfn
)
371 nr_pfn_mapped
= add_range_with_merge(pfn_mapped
, E820_X_MAX
,
372 nr_pfn_mapped
, start_pfn
, end_pfn
);
373 nr_pfn_mapped
= clean_sort_range(pfn_mapped
, E820_X_MAX
);
375 max_pfn_mapped
= max(max_pfn_mapped
, end_pfn
);
377 if (start_pfn
< (1UL<<(32-PAGE_SHIFT
)))
378 max_low_pfn_mapped
= max(max_low_pfn_mapped
,
379 min(end_pfn
, 1UL<<(32-PAGE_SHIFT
)));
382 bool pfn_range_is_mapped(unsigned long start_pfn
, unsigned long end_pfn
)
386 for (i
= 0; i
< nr_pfn_mapped
; i
++)
387 if ((start_pfn
>= pfn_mapped
[i
].start
) &&
388 (end_pfn
<= pfn_mapped
[i
].end
))
395 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
396 * This runs before bootmem is initialized and gets pages directly from
397 * the physical memory. To access them they are temporarily mapped.
399 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
402 struct map_range mr
[NR_RANGE_MR
];
403 unsigned long ret
= 0;
406 pr_debug("init_memory_mapping: [mem %#010lx-%#010lx]\n",
409 memset(mr
, 0, sizeof(mr
));
410 nr_range
= split_mem_range(mr
, 0, start
, end
);
412 for (i
= 0; i
< nr_range
; i
++)
413 ret
= kernel_physical_mapping_init(mr
[i
].start
, mr
[i
].end
,
414 mr
[i
].page_size_mask
);
416 add_pfn_range_mapped(start
>> PAGE_SHIFT
, ret
>> PAGE_SHIFT
);
418 return ret
>> PAGE_SHIFT
;
422 * We need to iterate through the E820 memory map and create direct mappings
423 * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply
424 * create direct mappings for all pfns from [0 to max_low_pfn) and
425 * [4GB to max_pfn) because of possible memory holes in high addresses
426 * that cannot be marked as UC by fixed/variable range MTRRs.
427 * Depending on the alignment of E820 ranges, this may possibly result
428 * in using smaller size (i.e. 4K instead of 2M or 1G) page tables.
430 * init_mem_mapping() calls init_range_memory_mapping() with big range.
431 * That range would have hole in the middle or ends, and only ram parts
432 * will be mapped in init_range_memory_mapping().
434 static unsigned long __init
init_range_memory_mapping(
435 unsigned long r_start
,
438 unsigned long start_pfn
, end_pfn
;
439 unsigned long mapped_ram_size
= 0;
442 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, NULL
) {
443 u64 start
= clamp_val(PFN_PHYS(start_pfn
), r_start
, r_end
);
444 u64 end
= clamp_val(PFN_PHYS(end_pfn
), r_start
, r_end
);
449 * if it is overlapping with brk pgt, we need to
450 * alloc pgt buf from memblock instead.
452 can_use_brk_pgt
= max(start
, (u64
)pgt_buf_end
<<PAGE_SHIFT
) >=
453 min(end
, (u64
)pgt_buf_top
<<PAGE_SHIFT
);
454 init_memory_mapping(start
, end
);
455 mapped_ram_size
+= end
- start
;
456 can_use_brk_pgt
= true;
459 return mapped_ram_size
;
462 static unsigned long __init
get_new_step_size(unsigned long step_size
)
465 * Initial mapped size is PMD_SIZE (2M).
466 * We can not set step_size to be PUD_SIZE (1G) yet.
467 * In worse case, when we cross the 1G boundary, and
468 * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k)
469 * to map 1G range with PTE. Hence we use one less than the
470 * difference of page table level shifts.
472 * Don't need to worry about overflow in the top-down case, on 32bit,
473 * when step_size is 0, round_down() returns 0 for start, and that
474 * turns it into 0x100000000ULL.
475 * In the bottom-up case, round_up(x, 0) returns 0 though too, which
476 * needs to be taken into consideration by the code below.
478 return step_size
<< (PMD_SHIFT
- PAGE_SHIFT
- 1);
482 * memory_map_top_down - Map [map_start, map_end) top down
483 * @map_start: start address of the target memory range
484 * @map_end: end address of the target memory range
486 * This function will setup direct mapping for memory range
487 * [map_start, map_end) in top-down. That said, the page tables
488 * will be allocated at the end of the memory, and we map the
489 * memory in top-down.
491 static void __init
memory_map_top_down(unsigned long map_start
,
492 unsigned long map_end
)
494 unsigned long real_end
, start
, last_start
;
495 unsigned long step_size
;
497 unsigned long mapped_ram_size
= 0;
499 /* xen has big range in reserved near end of ram, skip it at first.*/
500 addr
= memblock_find_in_range(map_start
, map_end
, PMD_SIZE
, PMD_SIZE
);
501 real_end
= addr
+ PMD_SIZE
;
503 /* step_size need to be small so pgt_buf from BRK could cover it */
504 step_size
= PMD_SIZE
;
505 max_pfn_mapped
= 0; /* will get exact value next */
506 min_pfn_mapped
= real_end
>> PAGE_SHIFT
;
507 last_start
= start
= real_end
;
510 * We start from the top (end of memory) and go to the bottom.
511 * The memblock_find_in_range() gets us a block of RAM from the
512 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
515 while (last_start
> map_start
) {
516 if (last_start
> step_size
) {
517 start
= round_down(last_start
- 1, step_size
);
518 if (start
< map_start
)
522 mapped_ram_size
+= init_range_memory_mapping(start
,
525 min_pfn_mapped
= last_start
>> PAGE_SHIFT
;
526 if (mapped_ram_size
>= step_size
)
527 step_size
= get_new_step_size(step_size
);
530 if (real_end
< map_end
)
531 init_range_memory_mapping(real_end
, map_end
);
535 * memory_map_bottom_up - Map [map_start, map_end) bottom up
536 * @map_start: start address of the target memory range
537 * @map_end: end address of the target memory range
539 * This function will setup direct mapping for memory range
540 * [map_start, map_end) in bottom-up. Since we have limited the
541 * bottom-up allocation above the kernel, the page tables will
542 * be allocated just above the kernel and we map the memory
543 * in [map_start, map_end) in bottom-up.
545 static void __init
memory_map_bottom_up(unsigned long map_start
,
546 unsigned long map_end
)
548 unsigned long next
, start
;
549 unsigned long mapped_ram_size
= 0;
550 /* step_size need to be small so pgt_buf from BRK could cover it */
551 unsigned long step_size
= PMD_SIZE
;
554 min_pfn_mapped
= start
>> PAGE_SHIFT
;
557 * We start from the bottom (@map_start) and go to the top (@map_end).
558 * The memblock_find_in_range() gets us a block of RAM from the
559 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
562 while (start
< map_end
) {
563 if (step_size
&& map_end
- start
> step_size
) {
564 next
= round_up(start
+ 1, step_size
);
571 mapped_ram_size
+= init_range_memory_mapping(start
, next
);
574 if (mapped_ram_size
>= step_size
)
575 step_size
= get_new_step_size(step_size
);
579 void __init
init_mem_mapping(void)
583 probe_page_size_mask();
586 end
= max_pfn
<< PAGE_SHIFT
;
588 end
= max_low_pfn
<< PAGE_SHIFT
;
591 /* the ISA range is always mapped regardless of memory holes */
592 init_memory_mapping(0, ISA_END_ADDRESS
);
594 /* Init the trampoline, possibly with KASLR memory offset */
598 * If the allocation is in bottom-up direction, we setup direct mapping
599 * in bottom-up, otherwise we setup direct mapping in top-down.
601 if (memblock_bottom_up()) {
602 unsigned long kernel_end
= __pa_symbol(_end
);
605 * we need two separate calls here. This is because we want to
606 * allocate page tables above the kernel. So we first map
607 * [kernel_end, end) to make memory above the kernel be mapped
608 * as soon as possible. And then use page tables allocated above
609 * the kernel to map [ISA_END_ADDRESS, kernel_end).
611 memory_map_bottom_up(kernel_end
, end
);
612 memory_map_bottom_up(ISA_END_ADDRESS
, kernel_end
);
614 memory_map_top_down(ISA_END_ADDRESS
, end
);
618 if (max_pfn
> max_low_pfn
) {
619 /* can we preseve max_low_pfn ?*/
620 max_low_pfn
= max_pfn
;
623 early_ioremap_page_table_range_init();
626 load_cr3(swapper_pg_dir
);
629 early_memtest(0, max_pfn_mapped
<< PAGE_SHIFT
);
633 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
634 * is valid. The argument is a physical page number.
637 * On x86, access has to be given to the first megabyte of ram because that area
638 * contains BIOS code and data regions used by X and dosemu and similar apps.
639 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
640 * mmio resources as well as potential bios/acpi data regions.
642 int devmem_is_allowed(unsigned long pagenr
)
646 if (iomem_is_exclusive(pagenr
<< PAGE_SHIFT
))
648 if (!page_is_ram(pagenr
))
653 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
655 unsigned long begin_aligned
, end_aligned
;
657 /* Make sure boundaries are page aligned */
658 begin_aligned
= PAGE_ALIGN(begin
);
659 end_aligned
= end
& PAGE_MASK
;
661 if (WARN_ON(begin_aligned
!= begin
|| end_aligned
!= end
)) {
662 begin
= begin_aligned
;
670 * If debugging page accesses then do not free this memory but
671 * mark them not present - any buggy init-section access will
672 * create a kernel page fault:
674 if (debug_pagealloc_enabled()) {
675 pr_info("debug: unmapping init [mem %#010lx-%#010lx]\n",
677 set_memory_np(begin
, (end
- begin
) >> PAGE_SHIFT
);
680 * We just marked the kernel text read only above, now that
681 * we are going to free part of that, we need to make that
682 * writeable and non-executable first.
684 set_memory_nx(begin
, (end
- begin
) >> PAGE_SHIFT
);
685 set_memory_rw(begin
, (end
- begin
) >> PAGE_SHIFT
);
687 free_reserved_area((void *)begin
, (void *)end
,
688 POISON_FREE_INITMEM
, what
);
692 void free_initmem(void)
694 free_init_pages("unused kernel",
695 (unsigned long)(&__init_begin
),
696 (unsigned long)(&__init_end
));
699 #ifdef CONFIG_BLK_DEV_INITRD
700 void __init
free_initrd_mem(unsigned long start
, unsigned long end
)
703 * Remember, initrd memory may contain microcode or other useful things.
704 * Before we lose initrd mem, we need to find a place to hold them
705 * now that normal virtual memory is enabled.
707 save_microcode_in_initrd();
710 * end could be not aligned, and We can not align that,
711 * decompresser could be confused by aligned initrd_end
712 * We already reserve the end partial page before in
713 * - i386_start_kernel()
714 * - x86_64_start_kernel()
715 * - relocate_initrd()
716 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
718 free_init_pages("initrd", start
, PAGE_ALIGN(end
));
722 void __init
zone_sizes_init(void)
724 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
726 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
728 #ifdef CONFIG_ZONE_DMA
729 max_zone_pfns
[ZONE_DMA
] = min(MAX_DMA_PFN
, max_low_pfn
);
731 #ifdef CONFIG_ZONE_DMA32
732 max_zone_pfns
[ZONE_DMA32
] = min(MAX_DMA32_PFN
, max_low_pfn
);
734 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
735 #ifdef CONFIG_HIGHMEM
736 max_zone_pfns
[ZONE_HIGHMEM
] = max_pfn
;
739 free_area_init_nodes(max_zone_pfns
);
742 DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state
, cpu_tlbstate
) = {
744 .active_mm
= &init_mm
,
747 .cr4
= ~0UL, /* fail hard if we screw up cr4 shadow initialization */
749 EXPORT_SYMBOL_GPL(cpu_tlbstate
);
751 void update_cache_mode_entry(unsigned entry
, enum page_cache_mode cache
)
753 /* entry 0 MUST be WB (hardwired to speed up translations) */
754 BUG_ON(!entry
&& cache
!= _PAGE_CACHE_MODE_WB
);
756 __cachemode2pte_tbl
[cache
] = __cm_idx2pte(entry
);
757 __pte2cachemode_tbl
[entry
] = cache
;