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>
22 * We need to define the tracepoints somewhere, and tlb.c
23 * is only compied when SMP=y.
25 #define CREATE_TRACE_POINTS
26 #include <trace/events/tlb.h>
28 #include "mm_internal.h"
30 static unsigned long __initdata pgt_buf_start
;
31 static unsigned long __initdata pgt_buf_end
;
32 static unsigned long __initdata pgt_buf_top
;
34 static unsigned long min_pfn_mapped
;
36 static bool __initdata can_use_brk_pgt
= true;
39 * Pages returned are already directly mapped.
41 * Changing that is likely to break Xen, see commit:
43 * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
45 * for detailed information.
47 __ref
void *alloc_low_pages(unsigned int num
)
55 order
= get_order((unsigned long)num
<< PAGE_SHIFT
);
56 return (void *)__get_free_pages(GFP_ATOMIC
| __GFP_NOTRACK
|
60 if ((pgt_buf_end
+ num
) > pgt_buf_top
|| !can_use_brk_pgt
) {
62 if (min_pfn_mapped
>= max_pfn_mapped
)
63 panic("alloc_low_pages: ran out of memory");
64 ret
= memblock_find_in_range(min_pfn_mapped
<< PAGE_SHIFT
,
65 max_pfn_mapped
<< PAGE_SHIFT
,
66 PAGE_SIZE
* num
, PAGE_SIZE
);
68 panic("alloc_low_pages: can not alloc memory");
69 memblock_reserve(ret
, PAGE_SIZE
* num
);
70 pfn
= ret
>> PAGE_SHIFT
;
74 printk(KERN_DEBUG
"BRK [%#010lx, %#010lx] PGTABLE\n",
75 pfn
<< PAGE_SHIFT
, (pgt_buf_end
<< PAGE_SHIFT
) - 1);
78 for (i
= 0; i
< num
; i
++) {
81 adr
= __va((pfn
+ i
) << PAGE_SHIFT
);
85 return __va(pfn
<< PAGE_SHIFT
);
88 /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */
89 #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE)
90 RESERVE_BRK(early_pgt_alloc
, INIT_PGT_BUF_SIZE
);
91 void __init
early_alloc_pgt_buf(void)
93 unsigned long tables
= INIT_PGT_BUF_SIZE
;
96 base
= __pa(extend_brk(tables
, PAGE_SIZE
));
98 pgt_buf_start
= base
>> PAGE_SHIFT
;
99 pgt_buf_end
= pgt_buf_start
;
100 pgt_buf_top
= pgt_buf_start
+ (tables
>> PAGE_SHIFT
);
106 #ifdef CONFIG_DIRECT_GBPAGES
111 static void __init
init_gbpages(void)
114 if (direct_gbpages
&& cpu_has_gbpages
)
115 printk(KERN_INFO
"Using GB pages for direct mapping\n");
124 unsigned page_size_mask
;
127 static int page_size_mask
;
129 static void __init
probe_page_size_mask(void)
133 #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK)
135 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
136 * This will simplify cpa(), which otherwise needs to support splitting
137 * large pages into small in interrupt context, etc.
140 page_size_mask
|= 1 << PG_LEVEL_1G
;
142 page_size_mask
|= 1 << PG_LEVEL_2M
;
145 /* Enable PSE if available */
147 set_in_cr4(X86_CR4_PSE
);
149 /* Enable PGE if available */
151 set_in_cr4(X86_CR4_PGE
);
152 __supported_pte_mask
|= _PAGE_GLOBAL
;
157 #define NR_RANGE_MR 3
158 #else /* CONFIG_X86_64 */
159 #define NR_RANGE_MR 5
162 static int __meminit
save_mr(struct map_range
*mr
, int nr_range
,
163 unsigned long start_pfn
, unsigned long end_pfn
,
164 unsigned long page_size_mask
)
166 if (start_pfn
< end_pfn
) {
167 if (nr_range
>= NR_RANGE_MR
)
168 panic("run out of range for init_memory_mapping\n");
169 mr
[nr_range
].start
= start_pfn
<<PAGE_SHIFT
;
170 mr
[nr_range
].end
= end_pfn
<<PAGE_SHIFT
;
171 mr
[nr_range
].page_size_mask
= page_size_mask
;
179 * adjust the page_size_mask for small range to go with
180 * big page size instead small one if nearby are ram too.
182 static void __init_refok
adjust_range_page_size_mask(struct map_range
*mr
,
187 for (i
= 0; i
< nr_range
; i
++) {
188 if ((page_size_mask
& (1<<PG_LEVEL_2M
)) &&
189 !(mr
[i
].page_size_mask
& (1<<PG_LEVEL_2M
))) {
190 unsigned long start
= round_down(mr
[i
].start
, PMD_SIZE
);
191 unsigned long end
= round_up(mr
[i
].end
, PMD_SIZE
);
194 if ((end
>> PAGE_SHIFT
) > max_low_pfn
)
198 if (memblock_is_region_memory(start
, end
- start
))
199 mr
[i
].page_size_mask
|= 1<<PG_LEVEL_2M
;
201 if ((page_size_mask
& (1<<PG_LEVEL_1G
)) &&
202 !(mr
[i
].page_size_mask
& (1<<PG_LEVEL_1G
))) {
203 unsigned long start
= round_down(mr
[i
].start
, PUD_SIZE
);
204 unsigned long end
= round_up(mr
[i
].end
, PUD_SIZE
);
206 if (memblock_is_region_memory(start
, end
- start
))
207 mr
[i
].page_size_mask
|= 1<<PG_LEVEL_1G
;
212 static int __meminit
split_mem_range(struct map_range
*mr
, int nr_range
,
216 unsigned long start_pfn
, end_pfn
, limit_pfn
;
220 limit_pfn
= PFN_DOWN(end
);
222 /* head if not big page alignment ? */
223 pfn
= start_pfn
= PFN_DOWN(start
);
226 * Don't use a large page for the first 2/4MB of memory
227 * because there are often fixed size MTRRs in there
228 * and overlapping MTRRs into large pages can cause
232 end_pfn
= PFN_DOWN(PMD_SIZE
);
234 end_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
235 #else /* CONFIG_X86_64 */
236 end_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
238 if (end_pfn
> limit_pfn
)
240 if (start_pfn
< end_pfn
) {
241 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
245 /* big page (2M) range */
246 start_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
248 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
));
249 #else /* CONFIG_X86_64 */
250 end_pfn
= round_up(pfn
, PFN_DOWN(PUD_SIZE
));
251 if (end_pfn
> round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
)))
252 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
));
255 if (start_pfn
< end_pfn
) {
256 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
257 page_size_mask
& (1<<PG_LEVEL_2M
));
262 /* big page (1G) range */
263 start_pfn
= round_up(pfn
, PFN_DOWN(PUD_SIZE
));
264 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PUD_SIZE
));
265 if (start_pfn
< end_pfn
) {
266 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
268 ((1<<PG_LEVEL_2M
)|(1<<PG_LEVEL_1G
)));
272 /* tail is not big page (1G) alignment */
273 start_pfn
= round_up(pfn
, PFN_DOWN(PMD_SIZE
));
274 end_pfn
= round_down(limit_pfn
, PFN_DOWN(PMD_SIZE
));
275 if (start_pfn
< end_pfn
) {
276 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
,
277 page_size_mask
& (1<<PG_LEVEL_2M
));
282 /* tail is not big page (2M) alignment */
285 nr_range
= save_mr(mr
, nr_range
, start_pfn
, end_pfn
, 0);
288 adjust_range_page_size_mask(mr
, nr_range
);
290 /* try to merge same page size and continuous */
291 for (i
= 0; nr_range
> 1 && i
< nr_range
- 1; i
++) {
292 unsigned long old_start
;
293 if (mr
[i
].end
!= mr
[i
+1].start
||
294 mr
[i
].page_size_mask
!= mr
[i
+1].page_size_mask
)
297 old_start
= mr
[i
].start
;
298 memmove(&mr
[i
], &mr
[i
+1],
299 (nr_range
- 1 - i
) * sizeof(struct map_range
));
300 mr
[i
--].start
= old_start
;
304 for (i
= 0; i
< nr_range
; i
++)
305 printk(KERN_DEBUG
" [mem %#010lx-%#010lx] page %s\n",
306 mr
[i
].start
, mr
[i
].end
- 1,
307 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_1G
))?"1G":(
308 (mr
[i
].page_size_mask
& (1<<PG_LEVEL_2M
))?"2M":"4k"));
313 struct range pfn_mapped
[E820_X_MAX
];
316 static void add_pfn_range_mapped(unsigned long start_pfn
, unsigned long end_pfn
)
318 nr_pfn_mapped
= add_range_with_merge(pfn_mapped
, E820_X_MAX
,
319 nr_pfn_mapped
, start_pfn
, end_pfn
);
320 nr_pfn_mapped
= clean_sort_range(pfn_mapped
, E820_X_MAX
);
322 max_pfn_mapped
= max(max_pfn_mapped
, end_pfn
);
324 if (start_pfn
< (1UL<<(32-PAGE_SHIFT
)))
325 max_low_pfn_mapped
= max(max_low_pfn_mapped
,
326 min(end_pfn
, 1UL<<(32-PAGE_SHIFT
)));
329 bool pfn_range_is_mapped(unsigned long start_pfn
, unsigned long end_pfn
)
333 for (i
= 0; i
< nr_pfn_mapped
; i
++)
334 if ((start_pfn
>= pfn_mapped
[i
].start
) &&
335 (end_pfn
<= pfn_mapped
[i
].end
))
342 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
343 * This runs before bootmem is initialized and gets pages directly from
344 * the physical memory. To access them they are temporarily mapped.
346 unsigned long __init_refok
init_memory_mapping(unsigned long start
,
349 struct map_range mr
[NR_RANGE_MR
];
350 unsigned long ret
= 0;
353 pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n",
356 memset(mr
, 0, sizeof(mr
));
357 nr_range
= split_mem_range(mr
, 0, start
, end
);
359 for (i
= 0; i
< nr_range
; i
++)
360 ret
= kernel_physical_mapping_init(mr
[i
].start
, mr
[i
].end
,
361 mr
[i
].page_size_mask
);
363 add_pfn_range_mapped(start
>> PAGE_SHIFT
, ret
>> PAGE_SHIFT
);
365 return ret
>> PAGE_SHIFT
;
369 * We need to iterate through the E820 memory map and create direct mappings
370 * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply
371 * create direct mappings for all pfns from [0 to max_low_pfn) and
372 * [4GB to max_pfn) because of possible memory holes in high addresses
373 * that cannot be marked as UC by fixed/variable range MTRRs.
374 * Depending on the alignment of E820 ranges, this may possibly result
375 * in using smaller size (i.e. 4K instead of 2M or 1G) page tables.
377 * init_mem_mapping() calls init_range_memory_mapping() with big range.
378 * That range would have hole in the middle or ends, and only ram parts
379 * will be mapped in init_range_memory_mapping().
381 static unsigned long __init
init_range_memory_mapping(
382 unsigned long r_start
,
385 unsigned long start_pfn
, end_pfn
;
386 unsigned long mapped_ram_size
= 0;
389 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, NULL
) {
390 u64 start
= clamp_val(PFN_PHYS(start_pfn
), r_start
, r_end
);
391 u64 end
= clamp_val(PFN_PHYS(end_pfn
), r_start
, r_end
);
396 * if it is overlapping with brk pgt, we need to
397 * alloc pgt buf from memblock instead.
399 can_use_brk_pgt
= max(start
, (u64
)pgt_buf_end
<<PAGE_SHIFT
) >=
400 min(end
, (u64
)pgt_buf_top
<<PAGE_SHIFT
);
401 init_memory_mapping(start
, end
);
402 mapped_ram_size
+= end
- start
;
403 can_use_brk_pgt
= true;
406 return mapped_ram_size
;
409 static unsigned long __init
get_new_step_size(unsigned long step_size
)
412 * Explain why we shift by 5 and why we don't have to worry about
413 * 'step_size << 5' overflowing:
415 * initial mapped size is PMD_SIZE (2M).
416 * We can not set step_size to be PUD_SIZE (1G) yet.
417 * In worse case, when we cross the 1G boundary, and
418 * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k)
419 * to map 1G range with PTE. Use 5 as shift for now.
421 * Don't need to worry about overflow, on 32bit, when step_size
422 * is 0, round_down() returns 0 for start, and that turns it
423 * into 0x100000000ULL.
425 return step_size
<< 5;
429 * memory_map_top_down - Map [map_start, map_end) top down
430 * @map_start: start address of the target memory range
431 * @map_end: end address of the target memory range
433 * This function will setup direct mapping for memory range
434 * [map_start, map_end) in top-down. That said, the page tables
435 * will be allocated at the end of the memory, and we map the
436 * memory in top-down.
438 static void __init
memory_map_top_down(unsigned long map_start
,
439 unsigned long map_end
)
441 unsigned long real_end
, start
, last_start
;
442 unsigned long step_size
;
444 unsigned long mapped_ram_size
= 0;
445 unsigned long new_mapped_ram_size
;
447 /* xen has big range in reserved near end of ram, skip it at first.*/
448 addr
= memblock_find_in_range(map_start
, map_end
, PMD_SIZE
, PMD_SIZE
);
449 real_end
= addr
+ PMD_SIZE
;
451 /* step_size need to be small so pgt_buf from BRK could cover it */
452 step_size
= PMD_SIZE
;
453 max_pfn_mapped
= 0; /* will get exact value next */
454 min_pfn_mapped
= real_end
>> PAGE_SHIFT
;
455 last_start
= start
= real_end
;
458 * We start from the top (end of memory) and go to the bottom.
459 * The memblock_find_in_range() gets us a block of RAM from the
460 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
463 while (last_start
> map_start
) {
464 if (last_start
> step_size
) {
465 start
= round_down(last_start
- 1, step_size
);
466 if (start
< map_start
)
470 new_mapped_ram_size
= init_range_memory_mapping(start
,
473 min_pfn_mapped
= last_start
>> PAGE_SHIFT
;
474 /* only increase step_size after big range get mapped */
475 if (new_mapped_ram_size
> mapped_ram_size
)
476 step_size
= get_new_step_size(step_size
);
477 mapped_ram_size
+= new_mapped_ram_size
;
480 if (real_end
< map_end
)
481 init_range_memory_mapping(real_end
, map_end
);
485 * memory_map_bottom_up - Map [map_start, map_end) bottom up
486 * @map_start: start address of the target memory range
487 * @map_end: end address of the target memory range
489 * This function will setup direct mapping for memory range
490 * [map_start, map_end) in bottom-up. Since we have limited the
491 * bottom-up allocation above the kernel, the page tables will
492 * be allocated just above the kernel and we map the memory
493 * in [map_start, map_end) in bottom-up.
495 static void __init
memory_map_bottom_up(unsigned long map_start
,
496 unsigned long map_end
)
498 unsigned long next
, new_mapped_ram_size
, start
;
499 unsigned long mapped_ram_size
= 0;
500 /* step_size need to be small so pgt_buf from BRK could cover it */
501 unsigned long step_size
= PMD_SIZE
;
504 min_pfn_mapped
= start
>> PAGE_SHIFT
;
507 * We start from the bottom (@map_start) and go to the top (@map_end).
508 * The memblock_find_in_range() gets us a block of RAM from the
509 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages
512 while (start
< map_end
) {
513 if (map_end
- start
> step_size
) {
514 next
= round_up(start
+ 1, step_size
);
520 new_mapped_ram_size
= init_range_memory_mapping(start
, next
);
523 if (new_mapped_ram_size
> mapped_ram_size
)
524 step_size
= get_new_step_size(step_size
);
525 mapped_ram_size
+= new_mapped_ram_size
;
529 void __init
init_mem_mapping(void)
533 probe_page_size_mask();
536 end
= max_pfn
<< PAGE_SHIFT
;
538 end
= max_low_pfn
<< PAGE_SHIFT
;
541 /* the ISA range is always mapped regardless of memory holes */
542 init_memory_mapping(0, ISA_END_ADDRESS
);
545 * If the allocation is in bottom-up direction, we setup direct mapping
546 * in bottom-up, otherwise we setup direct mapping in top-down.
548 if (memblock_bottom_up()) {
549 unsigned long kernel_end
= __pa_symbol(_end
);
552 * we need two separate calls here. This is because we want to
553 * allocate page tables above the kernel. So we first map
554 * [kernel_end, end) to make memory above the kernel be mapped
555 * as soon as possible. And then use page tables allocated above
556 * the kernel to map [ISA_END_ADDRESS, kernel_end).
558 memory_map_bottom_up(kernel_end
, end
);
559 memory_map_bottom_up(ISA_END_ADDRESS
, kernel_end
);
561 memory_map_top_down(ISA_END_ADDRESS
, end
);
565 if (max_pfn
> max_low_pfn
) {
566 /* can we preseve max_low_pfn ?*/
567 max_low_pfn
= max_pfn
;
570 early_ioremap_page_table_range_init();
573 load_cr3(swapper_pg_dir
);
576 early_memtest(0, max_pfn_mapped
<< PAGE_SHIFT
);
580 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
581 * is valid. The argument is a physical page number.
584 * On x86, access has to be given to the first megabyte of ram because that area
585 * contains bios code and data regions used by X and dosemu and similar apps.
586 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
587 * mmio resources as well as potential bios/acpi data regions.
589 int devmem_is_allowed(unsigned long pagenr
)
593 if (iomem_is_exclusive(pagenr
<< PAGE_SHIFT
))
595 if (!page_is_ram(pagenr
))
600 void free_init_pages(char *what
, unsigned long begin
, unsigned long end
)
602 unsigned long begin_aligned
, end_aligned
;
604 /* Make sure boundaries are page aligned */
605 begin_aligned
= PAGE_ALIGN(begin
);
606 end_aligned
= end
& PAGE_MASK
;
608 if (WARN_ON(begin_aligned
!= begin
|| end_aligned
!= end
)) {
609 begin
= begin_aligned
;
617 * If debugging page accesses then do not free this memory but
618 * mark them not present - any buggy init-section access will
619 * create a kernel page fault:
621 #ifdef CONFIG_DEBUG_PAGEALLOC
622 printk(KERN_INFO
"debug: unmapping init [mem %#010lx-%#010lx]\n",
624 set_memory_np(begin
, (end
- begin
) >> PAGE_SHIFT
);
627 * We just marked the kernel text read only above, now that
628 * we are going to free part of that, we need to make that
629 * writeable and non-executable first.
631 set_memory_nx(begin
, (end
- begin
) >> PAGE_SHIFT
);
632 set_memory_rw(begin
, (end
- begin
) >> PAGE_SHIFT
);
634 free_reserved_area((void *)begin
, (void *)end
, POISON_FREE_INITMEM
, what
);
638 void free_initmem(void)
640 free_init_pages("unused kernel",
641 (unsigned long)(&__init_begin
),
642 (unsigned long)(&__init_end
));
645 #ifdef CONFIG_BLK_DEV_INITRD
646 void __init
free_initrd_mem(unsigned long start
, unsigned long end
)
648 #ifdef CONFIG_MICROCODE_EARLY
650 * Remember, initrd memory may contain microcode or other useful things.
651 * Before we lose initrd mem, we need to find a place to hold them
652 * now that normal virtual memory is enabled.
654 save_microcode_in_initrd();
658 * end could be not aligned, and We can not align that,
659 * decompresser could be confused by aligned initrd_end
660 * We already reserve the end partial page before in
661 * - i386_start_kernel()
662 * - x86_64_start_kernel()
663 * - relocate_initrd()
664 * So here We can do PAGE_ALIGN() safely to get partial page to be freed
666 free_init_pages("initrd", start
, PAGE_ALIGN(end
));
670 void __init
zone_sizes_init(void)
672 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
674 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
676 #ifdef CONFIG_ZONE_DMA
677 max_zone_pfns
[ZONE_DMA
] = MAX_DMA_PFN
;
679 #ifdef CONFIG_ZONE_DMA32
680 max_zone_pfns
[ZONE_DMA32
] = MAX_DMA32_PFN
;
682 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
683 #ifdef CONFIG_HIGHMEM
684 max_zone_pfns
[ZONE_HIGHMEM
] = max_pfn
;
687 free_area_init_nodes(max_zone_pfns
);