2 * linux/arch/arm/mm/mmu.c
4 * Copyright (C) 1995-2005 Russell King
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
10 #include <linux/module.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/mman.h>
15 #include <linux/nodemask.h>
16 #include <linux/memblock.h>
18 #include <linux/vmalloc.h>
19 #include <linux/sizes.h>
22 #include <asm/cputype.h>
23 #include <asm/sections.h>
24 #include <asm/cachetype.h>
25 #include <asm/setup.h>
26 #include <asm/smp_plat.h>
28 #include <asm/highmem.h>
29 #include <asm/system_info.h>
30 #include <asm/traps.h>
32 #include <asm/mach/arch.h>
33 #include <asm/mach/map.h>
34 #include <asm/mach/pci.h>
40 * empty_zero_page is a special page that is used for
41 * zero-initialized data and COW.
43 struct page
*empty_zero_page
;
44 EXPORT_SYMBOL(empty_zero_page
);
47 * The pmd table for the upper-most set of pages.
51 #define CPOLICY_UNCACHED 0
52 #define CPOLICY_BUFFERED 1
53 #define CPOLICY_WRITETHROUGH 2
54 #define CPOLICY_WRITEBACK 3
55 #define CPOLICY_WRITEALLOC 4
57 static unsigned int cachepolicy __initdata
= CPOLICY_WRITEBACK
;
58 static unsigned int ecc_mask __initdata
= 0;
60 pgprot_t pgprot_kernel
;
61 pgprot_t pgprot_hyp_device
;
63 pgprot_t pgprot_s2_device
;
65 EXPORT_SYMBOL(pgprot_user
);
66 EXPORT_SYMBOL(pgprot_kernel
);
69 const char policy
[16];
76 #ifdef CONFIG_ARM_LPAE
77 #define s2_policy(policy) policy
79 #define s2_policy(policy) 0
82 static struct cachepolicy cache_policies
[] __initdata
= {
86 .pmd
= PMD_SECT_UNCACHED
,
87 .pte
= L_PTE_MT_UNCACHED
,
88 .pte_s2
= s2_policy(L_PTE_S2_MT_UNCACHED
),
92 .pmd
= PMD_SECT_BUFFERED
,
93 .pte
= L_PTE_MT_BUFFERABLE
,
94 .pte_s2
= s2_policy(L_PTE_S2_MT_UNCACHED
),
96 .policy
= "writethrough",
99 .pte
= L_PTE_MT_WRITETHROUGH
,
100 .pte_s2
= s2_policy(L_PTE_S2_MT_WRITETHROUGH
),
102 .policy
= "writeback",
105 .pte
= L_PTE_MT_WRITEBACK
,
106 .pte_s2
= s2_policy(L_PTE_S2_MT_WRITEBACK
),
108 .policy
= "writealloc",
110 .pmd
= PMD_SECT_WBWA
,
111 .pte
= L_PTE_MT_WRITEALLOC
,
112 .pte_s2
= s2_policy(L_PTE_S2_MT_WRITEBACK
),
116 #ifdef CONFIG_CPU_CP15
118 * These are useful for identifying cache coherency
119 * problems by allowing the cache or the cache and
120 * writebuffer to be turned off. (Note: the write
121 * buffer should not be on and the cache off).
123 static int __init
early_cachepolicy(char *p
)
127 for (i
= 0; i
< ARRAY_SIZE(cache_policies
); i
++) {
128 int len
= strlen(cache_policies
[i
].policy
);
130 if (memcmp(p
, cache_policies
[i
].policy
, len
) == 0) {
132 cr_alignment
&= ~cache_policies
[i
].cr_mask
;
133 cr_no_alignment
&= ~cache_policies
[i
].cr_mask
;
137 if (i
== ARRAY_SIZE(cache_policies
))
138 printk(KERN_ERR
"ERROR: unknown or unsupported cache policy\n");
140 * This restriction is partly to do with the way we boot; it is
141 * unpredictable to have memory mapped using two different sets of
142 * memory attributes (shared, type, and cache attribs). We can not
143 * change these attributes once the initial assembly has setup the
146 if (cpu_architecture() >= CPU_ARCH_ARMv6
) {
147 printk(KERN_WARNING
"Only cachepolicy=writeback supported on ARMv6 and later\n");
148 cachepolicy
= CPOLICY_WRITEBACK
;
151 set_cr(cr_alignment
);
154 early_param("cachepolicy", early_cachepolicy
);
156 static int __init
early_nocache(char *__unused
)
158 char *p
= "buffered";
159 printk(KERN_WARNING
"nocache is deprecated; use cachepolicy=%s\n", p
);
160 early_cachepolicy(p
);
163 early_param("nocache", early_nocache
);
165 static int __init
early_nowrite(char *__unused
)
167 char *p
= "uncached";
168 printk(KERN_WARNING
"nowb is deprecated; use cachepolicy=%s\n", p
);
169 early_cachepolicy(p
);
172 early_param("nowb", early_nowrite
);
174 #ifndef CONFIG_ARM_LPAE
175 static int __init
early_ecc(char *p
)
177 if (memcmp(p
, "on", 2) == 0)
178 ecc_mask
= PMD_PROTECTION
;
179 else if (memcmp(p
, "off", 3) == 0)
183 early_param("ecc", early_ecc
);
186 static int __init
noalign_setup(char *__unused
)
188 cr_alignment
&= ~CR_A
;
189 cr_no_alignment
&= ~CR_A
;
190 set_cr(cr_alignment
);
193 __setup("noalign", noalign_setup
);
196 void adjust_cr(unsigned long mask
, unsigned long set
)
204 local_irq_save(flags
);
206 cr_no_alignment
= (cr_no_alignment
& ~mask
) | set
;
207 cr_alignment
= (cr_alignment
& ~mask
) | set
;
209 set_cr((get_cr() & ~mask
) | set
);
211 local_irq_restore(flags
);
215 #else /* ifdef CONFIG_CPU_CP15 */
217 static int __init
early_cachepolicy(char *p
)
219 pr_warning("cachepolicy kernel parameter not supported without cp15\n");
221 early_param("cachepolicy", early_cachepolicy
);
223 static int __init
noalign_setup(char *__unused
)
225 pr_warning("noalign kernel parameter not supported without cp15\n");
227 __setup("noalign", noalign_setup
);
229 #endif /* ifdef CONFIG_CPU_CP15 / else */
231 #define PROT_PTE_DEVICE L_PTE_PRESENT|L_PTE_YOUNG|L_PTE_DIRTY|L_PTE_XN
232 #define PROT_SECT_DEVICE PMD_TYPE_SECT|PMD_SECT_AP_WRITE
234 static struct mem_type mem_types
[] = {
235 [MT_DEVICE
] = { /* Strongly ordered / ARMv6 shared device */
236 .prot_pte
= PROT_PTE_DEVICE
| L_PTE_MT_DEV_SHARED
|
238 .prot_l1
= PMD_TYPE_TABLE
,
239 .prot_sect
= PROT_SECT_DEVICE
| PMD_SECT_S
,
242 [MT_DEVICE_NONSHARED
] = { /* ARMv6 non-shared device */
243 .prot_pte
= PROT_PTE_DEVICE
| L_PTE_MT_DEV_NONSHARED
,
244 .prot_l1
= PMD_TYPE_TABLE
,
245 .prot_sect
= PROT_SECT_DEVICE
,
248 [MT_DEVICE_CACHED
] = { /* ioremap_cached */
249 .prot_pte
= PROT_PTE_DEVICE
| L_PTE_MT_DEV_CACHED
,
250 .prot_l1
= PMD_TYPE_TABLE
,
251 .prot_sect
= PROT_SECT_DEVICE
| PMD_SECT_WB
,
254 [MT_DEVICE_WC
] = { /* ioremap_wc */
255 .prot_pte
= PROT_PTE_DEVICE
| L_PTE_MT_DEV_WC
,
256 .prot_l1
= PMD_TYPE_TABLE
,
257 .prot_sect
= PROT_SECT_DEVICE
,
261 .prot_pte
= PROT_PTE_DEVICE
,
262 .prot_l1
= PMD_TYPE_TABLE
,
263 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_XN
,
267 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_XN
,
268 .domain
= DOMAIN_KERNEL
,
270 #ifndef CONFIG_ARM_LPAE
272 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_XN
| PMD_SECT_MINICACHE
,
273 .domain
= DOMAIN_KERNEL
,
277 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
|
279 .prot_l1
= PMD_TYPE_TABLE
,
280 .domain
= DOMAIN_USER
,
282 [MT_HIGH_VECTORS
] = {
283 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
|
284 L_PTE_USER
| L_PTE_RDONLY
,
285 .prot_l1
= PMD_TYPE_TABLE
,
286 .domain
= DOMAIN_USER
,
289 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
,
290 .prot_l1
= PMD_TYPE_TABLE
,
291 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_AP_WRITE
,
292 .domain
= DOMAIN_KERNEL
,
295 .prot_sect
= PMD_TYPE_SECT
,
296 .domain
= DOMAIN_KERNEL
,
298 [MT_MEMORY_NONCACHED
] = {
299 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
|
301 .prot_l1
= PMD_TYPE_TABLE
,
302 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_AP_WRITE
,
303 .domain
= DOMAIN_KERNEL
,
306 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
|
308 .prot_l1
= PMD_TYPE_TABLE
,
309 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_XN
,
310 .domain
= DOMAIN_KERNEL
,
313 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
,
314 .prot_l1
= PMD_TYPE_TABLE
,
315 .domain
= DOMAIN_KERNEL
,
318 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
|
319 L_PTE_MT_UNCACHED
| L_PTE_XN
,
320 .prot_l1
= PMD_TYPE_TABLE
,
321 .prot_sect
= PMD_TYPE_SECT
| PMD_SECT_AP_WRITE
| PMD_SECT_S
|
322 PMD_SECT_UNCACHED
| PMD_SECT_XN
,
323 .domain
= DOMAIN_KERNEL
,
325 [MT_MEMORY_DMA_READY
] = {
326 .prot_pte
= L_PTE_PRESENT
| L_PTE_YOUNG
| L_PTE_DIRTY
,
327 .prot_l1
= PMD_TYPE_TABLE
,
328 .domain
= DOMAIN_KERNEL
,
332 const struct mem_type
*get_mem_type(unsigned int type
)
334 return type
< ARRAY_SIZE(mem_types
) ? &mem_types
[type
] : NULL
;
336 EXPORT_SYMBOL(get_mem_type
);
339 * Adjust the PMD section entries according to the CPU in use.
341 static void __init
build_mem_type_table(void)
343 struct cachepolicy
*cp
;
344 unsigned int cr
= get_cr();
345 pteval_t user_pgprot
, kern_pgprot
, vecs_pgprot
;
346 pteval_t hyp_device_pgprot
, s2_pgprot
, s2_device_pgprot
;
347 int cpu_arch
= cpu_architecture();
350 if (cpu_arch
< CPU_ARCH_ARMv6
) {
351 #if defined(CONFIG_CPU_DCACHE_DISABLE)
352 if (cachepolicy
> CPOLICY_BUFFERED
)
353 cachepolicy
= CPOLICY_BUFFERED
;
354 #elif defined(CONFIG_CPU_DCACHE_WRITETHROUGH)
355 if (cachepolicy
> CPOLICY_WRITETHROUGH
)
356 cachepolicy
= CPOLICY_WRITETHROUGH
;
359 if (cpu_arch
< CPU_ARCH_ARMv5
) {
360 if (cachepolicy
>= CPOLICY_WRITEALLOC
)
361 cachepolicy
= CPOLICY_WRITEBACK
;
365 cachepolicy
= CPOLICY_WRITEALLOC
;
368 * Strip out features not present on earlier architectures.
369 * Pre-ARMv5 CPUs don't have TEX bits. Pre-ARMv6 CPUs or those
370 * without extended page tables don't have the 'Shared' bit.
372 if (cpu_arch
< CPU_ARCH_ARMv5
)
373 for (i
= 0; i
< ARRAY_SIZE(mem_types
); i
++)
374 mem_types
[i
].prot_sect
&= ~PMD_SECT_TEX(7);
375 if ((cpu_arch
< CPU_ARCH_ARMv6
|| !(cr
& CR_XP
)) && !cpu_is_xsc3())
376 for (i
= 0; i
< ARRAY_SIZE(mem_types
); i
++)
377 mem_types
[i
].prot_sect
&= ~PMD_SECT_S
;
380 * ARMv5 and lower, bit 4 must be set for page tables (was: cache
381 * "update-able on write" bit on ARM610). However, Xscale and
382 * Xscale3 require this bit to be cleared.
384 if (cpu_is_xscale() || cpu_is_xsc3()) {
385 for (i
= 0; i
< ARRAY_SIZE(mem_types
); i
++) {
386 mem_types
[i
].prot_sect
&= ~PMD_BIT4
;
387 mem_types
[i
].prot_l1
&= ~PMD_BIT4
;
389 } else if (cpu_arch
< CPU_ARCH_ARMv6
) {
390 for (i
= 0; i
< ARRAY_SIZE(mem_types
); i
++) {
391 if (mem_types
[i
].prot_l1
)
392 mem_types
[i
].prot_l1
|= PMD_BIT4
;
393 if (mem_types
[i
].prot_sect
)
394 mem_types
[i
].prot_sect
|= PMD_BIT4
;
399 * Mark the device areas according to the CPU/architecture.
401 if (cpu_is_xsc3() || (cpu_arch
>= CPU_ARCH_ARMv6
&& (cr
& CR_XP
))) {
402 if (!cpu_is_xsc3()) {
404 * Mark device regions on ARMv6+ as execute-never
405 * to prevent speculative instruction fetches.
407 mem_types
[MT_DEVICE
].prot_sect
|= PMD_SECT_XN
;
408 mem_types
[MT_DEVICE_NONSHARED
].prot_sect
|= PMD_SECT_XN
;
409 mem_types
[MT_DEVICE_CACHED
].prot_sect
|= PMD_SECT_XN
;
410 mem_types
[MT_DEVICE_WC
].prot_sect
|= PMD_SECT_XN
;
412 if (cpu_arch
>= CPU_ARCH_ARMv7
&& (cr
& CR_TRE
)) {
414 * For ARMv7 with TEX remapping,
415 * - shared device is SXCB=1100
416 * - nonshared device is SXCB=0100
417 * - write combine device mem is SXCB=0001
418 * (Uncached Normal memory)
420 mem_types
[MT_DEVICE
].prot_sect
|= PMD_SECT_TEX(1);
421 mem_types
[MT_DEVICE_NONSHARED
].prot_sect
|= PMD_SECT_TEX(1);
422 mem_types
[MT_DEVICE_WC
].prot_sect
|= PMD_SECT_BUFFERABLE
;
423 } else if (cpu_is_xsc3()) {
426 * - shared device is TEXCB=00101
427 * - nonshared device is TEXCB=01000
428 * - write combine device mem is TEXCB=00100
429 * (Inner/Outer Uncacheable in xsc3 parlance)
431 mem_types
[MT_DEVICE
].prot_sect
|= PMD_SECT_TEX(1) | PMD_SECT_BUFFERED
;
432 mem_types
[MT_DEVICE_NONSHARED
].prot_sect
|= PMD_SECT_TEX(2);
433 mem_types
[MT_DEVICE_WC
].prot_sect
|= PMD_SECT_TEX(1);
436 * For ARMv6 and ARMv7 without TEX remapping,
437 * - shared device is TEXCB=00001
438 * - nonshared device is TEXCB=01000
439 * - write combine device mem is TEXCB=00100
440 * (Uncached Normal in ARMv6 parlance).
442 mem_types
[MT_DEVICE
].prot_sect
|= PMD_SECT_BUFFERED
;
443 mem_types
[MT_DEVICE_NONSHARED
].prot_sect
|= PMD_SECT_TEX(2);
444 mem_types
[MT_DEVICE_WC
].prot_sect
|= PMD_SECT_TEX(1);
448 * On others, write combining is "Uncached/Buffered"
450 mem_types
[MT_DEVICE_WC
].prot_sect
|= PMD_SECT_BUFFERABLE
;
454 * Now deal with the memory-type mappings
456 cp
= &cache_policies
[cachepolicy
];
457 vecs_pgprot
= kern_pgprot
= user_pgprot
= cp
->pte
;
458 s2_pgprot
= cp
->pte_s2
;
459 hyp_device_pgprot
= s2_device_pgprot
= mem_types
[MT_DEVICE
].prot_pte
;
462 * ARMv6 and above have extended page tables.
464 if (cpu_arch
>= CPU_ARCH_ARMv6
&& (cr
& CR_XP
)) {
465 #ifndef CONFIG_ARM_LPAE
467 * Mark cache clean areas and XIP ROM read only
468 * from SVC mode and no access from userspace.
470 mem_types
[MT_ROM
].prot_sect
|= PMD_SECT_APX
|PMD_SECT_AP_WRITE
;
471 mem_types
[MT_MINICLEAN
].prot_sect
|= PMD_SECT_APX
|PMD_SECT_AP_WRITE
;
472 mem_types
[MT_CACHECLEAN
].prot_sect
|= PMD_SECT_APX
|PMD_SECT_AP_WRITE
;
477 * Mark memory with the "shared" attribute
480 user_pgprot
|= L_PTE_SHARED
;
481 kern_pgprot
|= L_PTE_SHARED
;
482 vecs_pgprot
|= L_PTE_SHARED
;
483 s2_pgprot
|= L_PTE_SHARED
;
484 mem_types
[MT_DEVICE_WC
].prot_sect
|= PMD_SECT_S
;
485 mem_types
[MT_DEVICE_WC
].prot_pte
|= L_PTE_SHARED
;
486 mem_types
[MT_DEVICE_CACHED
].prot_sect
|= PMD_SECT_S
;
487 mem_types
[MT_DEVICE_CACHED
].prot_pte
|= L_PTE_SHARED
;
488 mem_types
[MT_MEMORY
].prot_sect
|= PMD_SECT_S
;
489 mem_types
[MT_MEMORY
].prot_pte
|= L_PTE_SHARED
;
490 mem_types
[MT_MEMORY_DMA_READY
].prot_pte
|= L_PTE_SHARED
;
491 mem_types
[MT_MEMORY_NONCACHED
].prot_sect
|= PMD_SECT_S
;
492 mem_types
[MT_MEMORY_NONCACHED
].prot_pte
|= L_PTE_SHARED
;
497 * Non-cacheable Normal - intended for memory areas that must
498 * not cause dirty cache line writebacks when used
500 if (cpu_arch
>= CPU_ARCH_ARMv6
) {
501 if (cpu_arch
>= CPU_ARCH_ARMv7
&& (cr
& CR_TRE
)) {
502 /* Non-cacheable Normal is XCB = 001 */
503 mem_types
[MT_MEMORY_NONCACHED
].prot_sect
|=
506 /* For both ARMv6 and non-TEX-remapping ARMv7 */
507 mem_types
[MT_MEMORY_NONCACHED
].prot_sect
|=
511 mem_types
[MT_MEMORY_NONCACHED
].prot_sect
|= PMD_SECT_BUFFERABLE
;
514 #ifdef CONFIG_ARM_LPAE
516 * Do not generate access flag faults for the kernel mappings.
518 for (i
= 0; i
< ARRAY_SIZE(mem_types
); i
++) {
519 mem_types
[i
].prot_pte
|= PTE_EXT_AF
;
520 if (mem_types
[i
].prot_sect
)
521 mem_types
[i
].prot_sect
|= PMD_SECT_AF
;
523 kern_pgprot
|= PTE_EXT_AF
;
524 vecs_pgprot
|= PTE_EXT_AF
;
527 for (i
= 0; i
< 16; i
++) {
528 pteval_t v
= pgprot_val(protection_map
[i
]);
529 protection_map
[i
] = __pgprot(v
| user_pgprot
);
532 mem_types
[MT_LOW_VECTORS
].prot_pte
|= vecs_pgprot
;
533 mem_types
[MT_HIGH_VECTORS
].prot_pte
|= vecs_pgprot
;
535 pgprot_user
= __pgprot(L_PTE_PRESENT
| L_PTE_YOUNG
| user_pgprot
);
536 pgprot_kernel
= __pgprot(L_PTE_PRESENT
| L_PTE_YOUNG
|
537 L_PTE_DIRTY
| kern_pgprot
);
538 pgprot_s2
= __pgprot(L_PTE_PRESENT
| L_PTE_YOUNG
| s2_pgprot
);
539 pgprot_s2_device
= __pgprot(s2_device_pgprot
);
540 pgprot_hyp_device
= __pgprot(hyp_device_pgprot
);
542 mem_types
[MT_LOW_VECTORS
].prot_l1
|= ecc_mask
;
543 mem_types
[MT_HIGH_VECTORS
].prot_l1
|= ecc_mask
;
544 mem_types
[MT_MEMORY
].prot_sect
|= ecc_mask
| cp
->pmd
;
545 mem_types
[MT_MEMORY
].prot_pte
|= kern_pgprot
;
546 mem_types
[MT_MEMORY_DMA_READY
].prot_pte
|= kern_pgprot
;
547 mem_types
[MT_MEMORY_NONCACHED
].prot_sect
|= ecc_mask
;
548 mem_types
[MT_ROM
].prot_sect
|= cp
->pmd
;
552 mem_types
[MT_CACHECLEAN
].prot_sect
|= PMD_SECT_WT
;
556 mem_types
[MT_CACHECLEAN
].prot_sect
|= PMD_SECT_WB
;
559 printk("Memory policy: ECC %sabled, Data cache %s\n",
560 ecc_mask
? "en" : "dis", cp
->policy
);
562 for (i
= 0; i
< ARRAY_SIZE(mem_types
); i
++) {
563 struct mem_type
*t
= &mem_types
[i
];
565 t
->prot_l1
|= PMD_DOMAIN(t
->domain
);
567 t
->prot_sect
|= PMD_DOMAIN(t
->domain
);
571 #ifdef CONFIG_ARM_DMA_MEM_BUFFERABLE
572 pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
573 unsigned long size
, pgprot_t vma_prot
)
576 return pgprot_noncached(vma_prot
);
577 else if (file
->f_flags
& O_SYNC
)
578 return pgprot_writecombine(vma_prot
);
581 EXPORT_SYMBOL(phys_mem_access_prot
);
584 #define vectors_base() (vectors_high() ? 0xffff0000 : 0)
586 static void __init
*early_alloc_aligned(unsigned long sz
, unsigned long align
)
588 void *ptr
= __va(memblock_alloc(sz
, align
));
593 static void __init
*early_alloc(unsigned long sz
)
595 return early_alloc_aligned(sz
, sz
);
598 static pte_t
* __init
early_pte_alloc(pmd_t
*pmd
, unsigned long addr
, unsigned long prot
)
600 if (pmd_none(*pmd
)) {
601 pte_t
*pte
= early_alloc(PTE_HWTABLE_OFF
+ PTE_HWTABLE_SIZE
);
602 __pmd_populate(pmd
, __pa(pte
), prot
);
604 BUG_ON(pmd_bad(*pmd
));
605 return pte_offset_kernel(pmd
, addr
);
608 static void __init
alloc_init_pte(pmd_t
*pmd
, unsigned long addr
,
609 unsigned long end
, unsigned long pfn
,
610 const struct mem_type
*type
)
612 pte_t
*pte
= early_pte_alloc(pmd
, addr
, type
->prot_l1
);
614 set_pte_ext(pte
, pfn_pte(pfn
, __pgprot(type
->prot_pte
)), 0);
616 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
619 static void __init
map_init_section(pmd_t
*pmd
, unsigned long addr
,
620 unsigned long end
, phys_addr_t phys
,
621 const struct mem_type
*type
)
623 #ifndef CONFIG_ARM_LPAE
625 * In classic MMU format, puds and pmds are folded in to
626 * the pgds. pmd_offset gives the PGD entry. PGDs refer to a
627 * group of L1 entries making up one logical pointer to
628 * an L2 table (2MB), where as PMDs refer to the individual
629 * L1 entries (1MB). Hence increment to get the correct
630 * offset for odd 1MB sections.
631 * (See arch/arm/include/asm/pgtable-2level.h)
633 if (addr
& SECTION_SIZE
)
637 *pmd
= __pmd(phys
| type
->prot_sect
);
638 phys
+= SECTION_SIZE
;
639 } while (pmd
++, addr
+= SECTION_SIZE
, addr
!= end
);
641 flush_pmd_entry(pmd
);
644 static void __init
alloc_init_pmd(pud_t
*pud
, unsigned long addr
,
645 unsigned long end
, phys_addr_t phys
,
646 const struct mem_type
*type
)
648 pmd_t
*pmd
= pmd_offset(pud
, addr
);
653 * With LPAE, we must loop over to map
654 * all the pmds for the given range.
656 next
= pmd_addr_end(addr
, end
);
659 * Try a section mapping - addr, next and phys must all be
660 * aligned to a section boundary.
662 if (type
->prot_sect
&&
663 ((addr
| next
| phys
) & ~SECTION_MASK
) == 0) {
664 map_init_section(pmd
, addr
, next
, phys
, type
);
666 alloc_init_pte(pmd
, addr
, next
,
667 __phys_to_pfn(phys
), type
);
672 } while (pmd
++, addr
= next
, addr
!= end
);
675 static void __init
alloc_init_pud(pgd_t
*pgd
, unsigned long addr
,
676 unsigned long end
, unsigned long phys
, const struct mem_type
*type
)
678 pud_t
*pud
= pud_offset(pgd
, addr
);
682 next
= pud_addr_end(addr
, end
);
683 alloc_init_pmd(pud
, addr
, next
, phys
, type
);
685 } while (pud
++, addr
= next
, addr
!= end
);
688 #ifndef CONFIG_ARM_LPAE
689 static void __init
create_36bit_mapping(struct map_desc
*md
,
690 const struct mem_type
*type
)
692 unsigned long addr
, length
, end
;
697 phys
= __pfn_to_phys(md
->pfn
);
698 length
= PAGE_ALIGN(md
->length
);
700 if (!(cpu_architecture() >= CPU_ARCH_ARMv6
|| cpu_is_xsc3())) {
701 printk(KERN_ERR
"MM: CPU does not support supersection "
702 "mapping for 0x%08llx at 0x%08lx\n",
703 (long long)__pfn_to_phys((u64
)md
->pfn
), addr
);
707 /* N.B. ARMv6 supersections are only defined to work with domain 0.
708 * Since domain assignments can in fact be arbitrary, the
709 * 'domain == 0' check below is required to insure that ARMv6
710 * supersections are only allocated for domain 0 regardless
711 * of the actual domain assignments in use.
714 printk(KERN_ERR
"MM: invalid domain in supersection "
715 "mapping for 0x%08llx at 0x%08lx\n",
716 (long long)__pfn_to_phys((u64
)md
->pfn
), addr
);
720 if ((addr
| length
| __pfn_to_phys(md
->pfn
)) & ~SUPERSECTION_MASK
) {
721 printk(KERN_ERR
"MM: cannot create mapping for 0x%08llx"
722 " at 0x%08lx invalid alignment\n",
723 (long long)__pfn_to_phys((u64
)md
->pfn
), addr
);
728 * Shift bits [35:32] of address into bits [23:20] of PMD
731 phys
|= (((md
->pfn
>> (32 - PAGE_SHIFT
)) & 0xF) << 20);
733 pgd
= pgd_offset_k(addr
);
736 pud_t
*pud
= pud_offset(pgd
, addr
);
737 pmd_t
*pmd
= pmd_offset(pud
, addr
);
740 for (i
= 0; i
< 16; i
++)
741 *pmd
++ = __pmd(phys
| type
->prot_sect
| PMD_SECT_SUPER
);
743 addr
+= SUPERSECTION_SIZE
;
744 phys
+= SUPERSECTION_SIZE
;
745 pgd
+= SUPERSECTION_SIZE
>> PGDIR_SHIFT
;
746 } while (addr
!= end
);
748 #endif /* !CONFIG_ARM_LPAE */
751 * Create the page directory entries and any necessary
752 * page tables for the mapping specified by `md'. We
753 * are able to cope here with varying sizes and address
754 * offsets, and we take full advantage of sections and
757 static void __init
create_mapping(struct map_desc
*md
)
759 unsigned long addr
, length
, end
;
761 const struct mem_type
*type
;
764 if (md
->virtual != vectors_base() && md
->virtual < TASK_SIZE
) {
765 printk(KERN_WARNING
"BUG: not creating mapping for 0x%08llx"
766 " at 0x%08lx in user region\n",
767 (long long)__pfn_to_phys((u64
)md
->pfn
), md
->virtual);
771 if ((md
->type
== MT_DEVICE
|| md
->type
== MT_ROM
) &&
772 md
->virtual >= PAGE_OFFSET
&&
773 (md
->virtual < VMALLOC_START
|| md
->virtual >= VMALLOC_END
)) {
774 printk(KERN_WARNING
"BUG: mapping for 0x%08llx"
775 " at 0x%08lx out of vmalloc space\n",
776 (long long)__pfn_to_phys((u64
)md
->pfn
), md
->virtual);
779 type
= &mem_types
[md
->type
];
781 #ifndef CONFIG_ARM_LPAE
783 * Catch 36-bit addresses
785 if (md
->pfn
>= 0x100000) {
786 create_36bit_mapping(md
, type
);
791 addr
= md
->virtual & PAGE_MASK
;
792 phys
= __pfn_to_phys(md
->pfn
);
793 length
= PAGE_ALIGN(md
->length
+ (md
->virtual & ~PAGE_MASK
));
795 if (type
->prot_l1
== 0 && ((addr
| phys
| length
) & ~SECTION_MASK
)) {
796 printk(KERN_WARNING
"BUG: map for 0x%08llx at 0x%08lx can not "
797 "be mapped using pages, ignoring.\n",
798 (long long)__pfn_to_phys(md
->pfn
), addr
);
802 pgd
= pgd_offset_k(addr
);
805 unsigned long next
= pgd_addr_end(addr
, end
);
807 alloc_init_pud(pgd
, addr
, next
, phys
, type
);
811 } while (pgd
++, addr
!= end
);
815 * Create the architecture specific mappings
817 void __init
iotable_init(struct map_desc
*io_desc
, int nr
)
820 struct vm_struct
*vm
;
821 struct static_vm
*svm
;
826 svm
= early_alloc_aligned(sizeof(*svm
) * nr
, __alignof__(*svm
));
828 for (md
= io_desc
; nr
; md
++, nr
--) {
832 vm
->addr
= (void *)(md
->virtual & PAGE_MASK
);
833 vm
->size
= PAGE_ALIGN(md
->length
+ (md
->virtual & ~PAGE_MASK
));
834 vm
->phys_addr
= __pfn_to_phys(md
->pfn
);
835 vm
->flags
= VM_IOREMAP
| VM_ARM_STATIC_MAPPING
;
836 vm
->flags
|= VM_ARM_MTYPE(md
->type
);
837 vm
->caller
= iotable_init
;
838 add_static_vm_early(svm
++);
842 void __init
vm_reserve_area_early(unsigned long addr
, unsigned long size
,
845 struct vm_struct
*vm
;
846 struct static_vm
*svm
;
848 svm
= early_alloc_aligned(sizeof(*svm
), __alignof__(*svm
));
851 vm
->addr
= (void *)addr
;
853 vm
->flags
= VM_IOREMAP
| VM_ARM_EMPTY_MAPPING
;
855 add_static_vm_early(svm
);
858 #ifndef CONFIG_ARM_LPAE
861 * The Linux PMD is made of two consecutive section entries covering 2MB
862 * (see definition in include/asm/pgtable-2level.h). However a call to
863 * create_mapping() may optimize static mappings by using individual
864 * 1MB section mappings. This leaves the actual PMD potentially half
865 * initialized if the top or bottom section entry isn't used, leaving it
866 * open to problems if a subsequent ioremap() or vmalloc() tries to use
867 * the virtual space left free by that unused section entry.
869 * Let's avoid the issue by inserting dummy vm entries covering the unused
870 * PMD halves once the static mappings are in place.
873 static void __init
pmd_empty_section_gap(unsigned long addr
)
875 vm_reserve_area_early(addr
, SECTION_SIZE
, pmd_empty_section_gap
);
878 static void __init
fill_pmd_gaps(void)
880 struct static_vm
*svm
;
881 struct vm_struct
*vm
;
882 unsigned long addr
, next
= 0;
885 list_for_each_entry(svm
, &static_vmlist
, list
) {
887 addr
= (unsigned long)vm
->addr
;
892 * Check if this vm starts on an odd section boundary.
893 * If so and the first section entry for this PMD is free
894 * then we block the corresponding virtual address.
896 if ((addr
& ~PMD_MASK
) == SECTION_SIZE
) {
897 pmd
= pmd_off_k(addr
);
899 pmd_empty_section_gap(addr
& PMD_MASK
);
903 * Then check if this vm ends on an odd section boundary.
904 * If so and the second section entry for this PMD is empty
905 * then we block the corresponding virtual address.
908 if ((addr
& ~PMD_MASK
) == SECTION_SIZE
) {
909 pmd
= pmd_off_k(addr
) + 1;
911 pmd_empty_section_gap(addr
);
914 /* no need to look at any vm entry until we hit the next PMD */
915 next
= (addr
+ PMD_SIZE
- 1) & PMD_MASK
;
920 #define fill_pmd_gaps() do { } while (0)
923 #if defined(CONFIG_PCI) && !defined(CONFIG_NEED_MACH_IO_H)
924 static void __init
pci_reserve_io(void)
926 struct static_vm
*svm
;
928 svm
= find_static_vm_vaddr((void *)PCI_IO_VIRT_BASE
);
932 vm_reserve_area_early(PCI_IO_VIRT_BASE
, SZ_2M
, pci_reserve_io
);
935 #define pci_reserve_io() do { } while (0)
938 #ifdef CONFIG_DEBUG_LL
939 void __init
debug_ll_io_init(void)
943 debug_ll_addr(&map
.pfn
, &map
.virtual);
944 if (!map
.pfn
|| !map
.virtual)
946 map
.pfn
= __phys_to_pfn(map
.pfn
);
947 map
.virtual &= PAGE_MASK
;
948 map
.length
= PAGE_SIZE
;
949 map
.type
= MT_DEVICE
;
950 create_mapping(&map
);
954 static void * __initdata vmalloc_min
=
955 (void *)(VMALLOC_END
- (240 << 20) - VMALLOC_OFFSET
);
958 * vmalloc=size forces the vmalloc area to be exactly 'size'
959 * bytes. This can be used to increase (or decrease) the vmalloc
960 * area - the default is 240m.
962 static int __init
early_vmalloc(char *arg
)
964 unsigned long vmalloc_reserve
= memparse(arg
, NULL
);
966 if (vmalloc_reserve
< SZ_16M
) {
967 vmalloc_reserve
= SZ_16M
;
969 "vmalloc area too small, limiting to %luMB\n",
970 vmalloc_reserve
>> 20);
973 if (vmalloc_reserve
> VMALLOC_END
- (PAGE_OFFSET
+ SZ_32M
)) {
974 vmalloc_reserve
= VMALLOC_END
- (PAGE_OFFSET
+ SZ_32M
);
976 "vmalloc area is too big, limiting to %luMB\n",
977 vmalloc_reserve
>> 20);
980 vmalloc_min
= (void *)(VMALLOC_END
- vmalloc_reserve
);
983 early_param("vmalloc", early_vmalloc
);
985 phys_addr_t arm_lowmem_limit __initdata
= 0;
987 void __init
sanity_check_meminfo(void)
989 int i
, j
, highmem
= 0;
991 for (i
= 0, j
= 0; i
< meminfo
.nr_banks
; i
++) {
992 struct membank
*bank
= &meminfo
.bank
[j
];
993 *bank
= meminfo
.bank
[i
];
995 if (bank
->start
> ULONG_MAX
)
998 #ifdef CONFIG_HIGHMEM
999 if (__va(bank
->start
) >= vmalloc_min
||
1000 __va(bank
->start
) < (void *)PAGE_OFFSET
)
1003 bank
->highmem
= highmem
;
1006 * Split those memory banks which are partially overlapping
1007 * the vmalloc area greatly simplifying things later.
1009 if (!highmem
&& __va(bank
->start
) < vmalloc_min
&&
1010 bank
->size
> vmalloc_min
- __va(bank
->start
)) {
1011 if (meminfo
.nr_banks
>= NR_BANKS
) {
1012 printk(KERN_CRIT
"NR_BANKS too low, "
1013 "ignoring high memory\n");
1015 memmove(bank
+ 1, bank
,
1016 (meminfo
.nr_banks
- i
) * sizeof(*bank
));
1019 bank
[1].size
-= vmalloc_min
- __va(bank
->start
);
1020 bank
[1].start
= __pa(vmalloc_min
- 1) + 1;
1021 bank
[1].highmem
= highmem
= 1;
1024 bank
->size
= vmalloc_min
- __va(bank
->start
);
1027 bank
->highmem
= highmem
;
1030 * Highmem banks not allowed with !CONFIG_HIGHMEM.
1033 printk(KERN_NOTICE
"Ignoring RAM at %.8llx-%.8llx "
1034 "(!CONFIG_HIGHMEM).\n",
1035 (unsigned long long)bank
->start
,
1036 (unsigned long long)bank
->start
+ bank
->size
- 1);
1041 * Check whether this memory bank would entirely overlap
1044 if (__va(bank
->start
) >= vmalloc_min
||
1045 __va(bank
->start
) < (void *)PAGE_OFFSET
) {
1046 printk(KERN_NOTICE
"Ignoring RAM at %.8llx-%.8llx "
1047 "(vmalloc region overlap).\n",
1048 (unsigned long long)bank
->start
,
1049 (unsigned long long)bank
->start
+ bank
->size
- 1);
1054 * Check whether this memory bank would partially overlap
1057 if (__va(bank
->start
+ bank
->size
- 1) >= vmalloc_min
||
1058 __va(bank
->start
+ bank
->size
- 1) <= __va(bank
->start
)) {
1059 unsigned long newsize
= vmalloc_min
- __va(bank
->start
);
1060 printk(KERN_NOTICE
"Truncating RAM at %.8llx-%.8llx "
1061 "to -%.8llx (vmalloc region overlap).\n",
1062 (unsigned long long)bank
->start
,
1063 (unsigned long long)bank
->start
+ bank
->size
- 1,
1064 (unsigned long long)bank
->start
+ newsize
- 1);
1065 bank
->size
= newsize
;
1068 if (!bank
->highmem
&& bank
->start
+ bank
->size
> arm_lowmem_limit
)
1069 arm_lowmem_limit
= bank
->start
+ bank
->size
;
1073 #ifdef CONFIG_HIGHMEM
1075 const char *reason
= NULL
;
1077 if (cache_is_vipt_aliasing()) {
1079 * Interactions between kmap and other mappings
1080 * make highmem support with aliasing VIPT caches
1083 reason
= "with VIPT aliasing cache";
1086 printk(KERN_CRIT
"HIGHMEM is not supported %s, ignoring high memory\n",
1088 while (j
> 0 && meminfo
.bank
[j
- 1].highmem
)
1093 meminfo
.nr_banks
= j
;
1094 high_memory
= __va(arm_lowmem_limit
- 1) + 1;
1095 memblock_set_current_limit(arm_lowmem_limit
);
1098 static inline void prepare_page_table(void)
1104 * Clear out all the mappings below the kernel image.
1106 for (addr
= 0; addr
< MODULES_VADDR
; addr
+= PMD_SIZE
)
1107 pmd_clear(pmd_off_k(addr
));
1109 #ifdef CONFIG_XIP_KERNEL
1110 /* The XIP kernel is mapped in the module area -- skip over it */
1111 addr
= ((unsigned long)_etext
+ PMD_SIZE
- 1) & PMD_MASK
;
1113 for ( ; addr
< PAGE_OFFSET
; addr
+= PMD_SIZE
)
1114 pmd_clear(pmd_off_k(addr
));
1117 * Find the end of the first block of lowmem.
1119 end
= memblock
.memory
.regions
[0].base
+ memblock
.memory
.regions
[0].size
;
1120 if (end
>= arm_lowmem_limit
)
1121 end
= arm_lowmem_limit
;
1124 * Clear out all the kernel space mappings, except for the first
1125 * memory bank, up to the vmalloc region.
1127 for (addr
= __phys_to_virt(end
);
1128 addr
< VMALLOC_START
; addr
+= PMD_SIZE
)
1129 pmd_clear(pmd_off_k(addr
));
1132 #ifdef CONFIG_ARM_LPAE
1133 /* the first page is reserved for pgd */
1134 #define SWAPPER_PG_DIR_SIZE (PAGE_SIZE + \
1135 PTRS_PER_PGD * PTRS_PER_PMD * sizeof(pmd_t))
1137 #define SWAPPER_PG_DIR_SIZE (PTRS_PER_PGD * sizeof(pgd_t))
1141 * Reserve the special regions of memory
1143 void __init
arm_mm_memblock_reserve(void)
1146 * Reserve the page tables. These are already in use,
1147 * and can only be in node 0.
1149 memblock_reserve(__pa(swapper_pg_dir
), SWAPPER_PG_DIR_SIZE
);
1151 #ifdef CONFIG_SA1111
1153 * Because of the SA1111 DMA bug, we want to preserve our
1154 * precious DMA-able memory...
1156 memblock_reserve(PHYS_OFFSET
, __pa(swapper_pg_dir
) - PHYS_OFFSET
);
1161 * Set up the device mappings. Since we clear out the page tables for all
1162 * mappings above VMALLOC_START, we will remove any debug device mappings.
1163 * This means you have to be careful how you debug this function, or any
1164 * called function. This means you can't use any function or debugging
1165 * method which may touch any device, otherwise the kernel _will_ crash.
1167 static void __init
devicemaps_init(struct machine_desc
*mdesc
)
1169 struct map_desc map
;
1174 * Allocate the vector page early.
1176 vectors
= early_alloc(PAGE_SIZE
);
1178 early_trap_init(vectors
);
1180 for (addr
= VMALLOC_START
; addr
; addr
+= PMD_SIZE
)
1181 pmd_clear(pmd_off_k(addr
));
1184 * Map the kernel if it is XIP.
1185 * It is always first in the modulearea.
1187 #ifdef CONFIG_XIP_KERNEL
1188 map
.pfn
= __phys_to_pfn(CONFIG_XIP_PHYS_ADDR
& SECTION_MASK
);
1189 map
.virtual = MODULES_VADDR
;
1190 map
.length
= ((unsigned long)_etext
- map
.virtual + ~SECTION_MASK
) & SECTION_MASK
;
1192 create_mapping(&map
);
1196 * Map the cache flushing regions.
1199 map
.pfn
= __phys_to_pfn(FLUSH_BASE_PHYS
);
1200 map
.virtual = FLUSH_BASE
;
1202 map
.type
= MT_CACHECLEAN
;
1203 create_mapping(&map
);
1205 #ifdef FLUSH_BASE_MINICACHE
1206 map
.pfn
= __phys_to_pfn(FLUSH_BASE_PHYS
+ SZ_1M
);
1207 map
.virtual = FLUSH_BASE_MINICACHE
;
1209 map
.type
= MT_MINICLEAN
;
1210 create_mapping(&map
);
1214 * Create a mapping for the machine vectors at the high-vectors
1215 * location (0xffff0000). If we aren't using high-vectors, also
1216 * create a mapping at the low-vectors virtual address.
1218 map
.pfn
= __phys_to_pfn(virt_to_phys(vectors
));
1219 map
.virtual = 0xffff0000;
1220 map
.length
= PAGE_SIZE
;
1221 map
.type
= MT_HIGH_VECTORS
;
1222 create_mapping(&map
);
1224 if (!vectors_high()) {
1226 map
.type
= MT_LOW_VECTORS
;
1227 create_mapping(&map
);
1231 * Ask the machine support to map in the statically mapped devices.
1237 /* Reserve fixed i/o space in VMALLOC region */
1241 * Finally flush the caches and tlb to ensure that we're in a
1242 * consistent state wrt the writebuffer. This also ensures that
1243 * any write-allocated cache lines in the vector page are written
1244 * back. After this point, we can start to touch devices again.
1246 local_flush_tlb_all();
1250 static void __init
kmap_init(void)
1252 #ifdef CONFIG_HIGHMEM
1253 pkmap_page_table
= early_pte_alloc(pmd_off_k(PKMAP_BASE
),
1254 PKMAP_BASE
, _PAGE_KERNEL_TABLE
);
1258 static void __init
map_lowmem(void)
1260 struct memblock_region
*reg
;
1262 /* Map all the lowmem memory banks. */
1263 for_each_memblock(memory
, reg
) {
1264 phys_addr_t start
= reg
->base
;
1265 phys_addr_t end
= start
+ reg
->size
;
1266 struct map_desc map
;
1268 if (end
> arm_lowmem_limit
)
1269 end
= arm_lowmem_limit
;
1273 map
.pfn
= __phys_to_pfn(start
);
1274 map
.virtual = __phys_to_virt(start
);
1275 map
.length
= end
- start
;
1276 map
.type
= MT_MEMORY
;
1278 create_mapping(&map
);
1283 * paging_init() sets up the page tables, initialises the zone memory
1284 * maps, and sets up the zero page, bad page and bad page tables.
1286 void __init
paging_init(struct machine_desc
*mdesc
)
1290 memblock_set_current_limit(arm_lowmem_limit
);
1292 build_mem_type_table();
1293 prepare_page_table();
1295 dma_contiguous_remap();
1296 devicemaps_init(mdesc
);
1300 top_pmd
= pmd_off_k(0xffff0000);
1302 /* allocate the zero page. */
1303 zero_page
= early_alloc(PAGE_SIZE
);
1307 empty_zero_page
= virt_to_page(zero_page
);
1308 __flush_dcache_page(NULL
, empty_zero_page
);