2 * Re-map IO memory to kernel address space so that we can access it.
3 * This is needed for high PCI addresses that aren't mapped in the
4 * 640k-1MB IO memory area on PC's
6 * (C) Copyright 1995 1996 Linus Torvalds
9 #include <linux/bootmem.h>
10 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/vmalloc.h>
15 #include <linux/mmiotrace.h>
17 #include <asm/cacheflush.h>
19 #include <asm/fixmap.h>
20 #include <asm/pgtable.h>
21 #include <asm/tlbflush.h>
22 #include <asm/pgalloc.h>
25 static inline int phys_addr_valid(resource_size_t addr
)
27 #ifdef CONFIG_PHYS_ADDR_T_64BIT
28 return !(addr
>> boot_cpu_data
.x86_phys_bits
);
36 unsigned long __phys_addr(unsigned long x
)
38 if (x
>= __START_KERNEL_map
) {
39 x
-= __START_KERNEL_map
;
40 VIRTUAL_BUG_ON(x
>= KERNEL_IMAGE_SIZE
);
43 VIRTUAL_BUG_ON(x
< PAGE_OFFSET
);
45 VIRTUAL_BUG_ON(!phys_addr_valid(x
));
49 EXPORT_SYMBOL(__phys_addr
);
51 bool __virt_addr_valid(unsigned long x
)
53 if (x
>= __START_KERNEL_map
) {
54 x
-= __START_KERNEL_map
;
55 if (x
>= KERNEL_IMAGE_SIZE
)
62 if (!phys_addr_valid(x
))
66 return pfn_valid(x
>> PAGE_SHIFT
);
68 EXPORT_SYMBOL(__virt_addr_valid
);
72 #ifdef CONFIG_DEBUG_VIRTUAL
73 unsigned long __phys_addr(unsigned long x
)
75 /* VMALLOC_* aren't constants */
76 VIRTUAL_BUG_ON(x
< PAGE_OFFSET
);
77 VIRTUAL_BUG_ON(__vmalloc_start_set
&& is_vmalloc_addr((void *) x
));
78 return x
- PAGE_OFFSET
;
80 EXPORT_SYMBOL(__phys_addr
);
83 bool __virt_addr_valid(unsigned long x
)
87 if (__vmalloc_start_set
&& is_vmalloc_addr((void *) x
))
89 if (x
>= FIXADDR_START
)
91 return pfn_valid((x
- PAGE_OFFSET
) >> PAGE_SHIFT
);
93 EXPORT_SYMBOL(__virt_addr_valid
);
97 int page_is_ram(unsigned long pagenr
)
99 resource_size_t addr
, end
;
103 * A special case is the first 4Kb of memory;
104 * This is a BIOS owned area, not kernel ram, but generally
105 * not listed as such in the E820 table.
111 * Second special case: Some BIOSen report the PC BIOS
112 * area (640->1Mb) as ram even though it is not.
114 if (pagenr
>= (BIOS_BEGIN
>> PAGE_SHIFT
) &&
115 pagenr
< (BIOS_END
>> PAGE_SHIFT
))
118 for (i
= 0; i
< e820
.nr_map
; i
++) {
122 if (e820
.map
[i
].type
!= E820_RAM
)
124 addr
= (e820
.map
[i
].addr
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
125 end
= (e820
.map
[i
].addr
+ e820
.map
[i
].size
) >> PAGE_SHIFT
;
128 if ((pagenr
>= addr
) && (pagenr
< end
))
135 * Fix up the linear direct mapping of the kernel to avoid cache attribute
138 int ioremap_change_attr(unsigned long vaddr
, unsigned long size
,
139 unsigned long prot_val
)
141 unsigned long nrpages
= size
>> PAGE_SHIFT
;
147 err
= _set_memory_uc(vaddr
, nrpages
);
150 err
= _set_memory_wc(vaddr
, nrpages
);
153 err
= _set_memory_wb(vaddr
, nrpages
);
161 * Remap an arbitrary physical address space into the kernel virtual
162 * address space. Needed when the kernel wants to access high addresses
165 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
166 * have to convert them into an offset in a page-aligned mapping, but the
167 * caller shouldn't need to know that small detail.
169 static void __iomem
*__ioremap_caller(resource_size_t phys_addr
,
170 unsigned long size
, unsigned long prot_val
, void *caller
)
172 unsigned long pfn
, offset
, vaddr
;
173 resource_size_t last_addr
;
174 const resource_size_t unaligned_phys_addr
= phys_addr
;
175 const unsigned long unaligned_size
= size
;
176 struct vm_struct
*area
;
177 unsigned long new_prot_val
;
180 void __iomem
*ret_addr
;
182 /* Don't allow wraparound or zero size */
183 last_addr
= phys_addr
+ size
- 1;
184 if (!size
|| last_addr
< phys_addr
)
187 if (!phys_addr_valid(phys_addr
)) {
188 printk(KERN_WARNING
"ioremap: invalid physical address %llx\n",
189 (unsigned long long)phys_addr
);
195 * Don't remap the low PCI/ISA area, it's always mapped..
197 if (is_ISA_range(phys_addr
, last_addr
))
198 return (__force
void __iomem
*)phys_to_virt(phys_addr
);
201 * Check if the request spans more than any BAR in the iomem resource
204 WARN_ONCE(iomem_map_sanity_check(phys_addr
, size
),
205 KERN_INFO
"Info: mapping multiple BARs. Your kernel is fine.");
208 * Don't allow anybody to remap normal RAM that we're using..
210 for (pfn
= phys_addr
>> PAGE_SHIFT
;
211 (pfn
<< PAGE_SHIFT
) < (last_addr
& PAGE_MASK
);
214 int is_ram
= page_is_ram(pfn
);
216 if (is_ram
&& pfn_valid(pfn
) && !PageReserved(pfn_to_page(pfn
)))
218 WARN_ON_ONCE(is_ram
);
222 * Mappings have to be page-aligned
224 offset
= phys_addr
& ~PAGE_MASK
;
225 phys_addr
&= PAGE_MASK
;
226 size
= PAGE_ALIGN(last_addr
+1) - phys_addr
;
228 retval
= reserve_memtype(phys_addr
, (u64
)phys_addr
+ size
,
229 prot_val
, &new_prot_val
);
231 printk(KERN_ERR
"ioremap reserve_memtype failed %d\n", retval
);
235 if (prot_val
!= new_prot_val
) {
236 if (!is_new_memtype_allowed(phys_addr
, size
,
237 prot_val
, new_prot_val
)) {
239 "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n",
240 (unsigned long long)phys_addr
,
241 (unsigned long long)(phys_addr
+ size
),
242 prot_val
, new_prot_val
);
243 free_memtype(phys_addr
, phys_addr
+ size
);
246 prot_val
= new_prot_val
;
252 prot
= PAGE_KERNEL_IO_NOCACHE
;
254 case _PAGE_CACHE_UC_MINUS
:
255 prot
= PAGE_KERNEL_IO_UC_MINUS
;
258 prot
= PAGE_KERNEL_IO_WC
;
261 prot
= PAGE_KERNEL_IO
;
268 area
= get_vm_area_caller(size
, VM_IOREMAP
, caller
);
271 area
->phys_addr
= phys_addr
;
272 vaddr
= (unsigned long) area
->addr
;
274 if (kernel_map_sync_memtype(phys_addr
, size
, prot_val
)) {
275 free_memtype(phys_addr
, phys_addr
+ size
);
280 if (ioremap_page_range(vaddr
, vaddr
+ size
, phys_addr
, prot
)) {
281 free_memtype(phys_addr
, phys_addr
+ size
);
286 ret_addr
= (void __iomem
*) (vaddr
+ offset
);
287 mmiotrace_ioremap(unaligned_phys_addr
, unaligned_size
, ret_addr
);
293 * ioremap_nocache - map bus memory into CPU space
294 * @offset: bus address of the memory
295 * @size: size of the resource to map
297 * ioremap_nocache performs a platform specific sequence of operations to
298 * make bus memory CPU accessible via the readb/readw/readl/writeb/
299 * writew/writel functions and the other mmio helpers. The returned
300 * address is not guaranteed to be usable directly as a virtual
303 * This version of ioremap ensures that the memory is marked uncachable
304 * on the CPU as well as honouring existing caching rules from things like
305 * the PCI bus. Note that there are other caches and buffers on many
306 * busses. In particular driver authors should read up on PCI writes
308 * It's useful if some control registers are in such an area and
309 * write combining or read caching is not desirable:
311 * Must be freed with iounmap.
313 void __iomem
*ioremap_nocache(resource_size_t phys_addr
, unsigned long size
)
316 * Ideally, this should be:
317 * pat_enabled ? _PAGE_CACHE_UC : _PAGE_CACHE_UC_MINUS;
319 * Till we fix all X drivers to use ioremap_wc(), we will use
322 unsigned long val
= _PAGE_CACHE_UC_MINUS
;
324 return __ioremap_caller(phys_addr
, size
, val
,
325 __builtin_return_address(0));
327 EXPORT_SYMBOL(ioremap_nocache
);
330 * ioremap_wc - map memory into CPU space write combined
331 * @offset: bus address of the memory
332 * @size: size of the resource to map
334 * This version of ioremap ensures that the memory is marked write combining.
335 * Write combining allows faster writes to some hardware devices.
337 * Must be freed with iounmap.
339 void __iomem
*ioremap_wc(resource_size_t phys_addr
, unsigned long size
)
342 return __ioremap_caller(phys_addr
, size
, _PAGE_CACHE_WC
,
343 __builtin_return_address(0));
345 return ioremap_nocache(phys_addr
, size
);
347 EXPORT_SYMBOL(ioremap_wc
);
349 void __iomem
*ioremap_cache(resource_size_t phys_addr
, unsigned long size
)
351 return __ioremap_caller(phys_addr
, size
, _PAGE_CACHE_WB
,
352 __builtin_return_address(0));
354 EXPORT_SYMBOL(ioremap_cache
);
356 static void __iomem
*ioremap_default(resource_size_t phys_addr
,
364 * - WB for WB-able memory and no other conflicting mappings
365 * - UC_MINUS for non-WB-able memory with no other conflicting mappings
366 * - Inherit from confliting mappings otherwise
368 err
= reserve_memtype(phys_addr
, phys_addr
+ size
,
369 _PAGE_CACHE_WB
, &flags
);
373 ret
= __ioremap_caller(phys_addr
, size
, flags
,
374 __builtin_return_address(0));
376 free_memtype(phys_addr
, phys_addr
+ size
);
380 void __iomem
*ioremap_prot(resource_size_t phys_addr
, unsigned long size
,
381 unsigned long prot_val
)
383 return __ioremap_caller(phys_addr
, size
, (prot_val
& _PAGE_CACHE_MASK
),
384 __builtin_return_address(0));
386 EXPORT_SYMBOL(ioremap_prot
);
389 * iounmap - Free a IO remapping
390 * @addr: virtual address from ioremap_*
392 * Caller must ensure there is only one unmapping for the same pointer.
394 void iounmap(volatile void __iomem
*addr
)
396 struct vm_struct
*p
, *o
;
398 if ((void __force
*)addr
<= high_memory
)
402 * __ioremap special-cases the PCI/ISA range by not instantiating a
403 * vm_area and by simply returning an address into the kernel mapping
404 * of ISA space. So handle that here.
406 if ((void __force
*)addr
>= phys_to_virt(ISA_START_ADDRESS
) &&
407 (void __force
*)addr
< phys_to_virt(ISA_END_ADDRESS
))
410 addr
= (volatile void __iomem
*)
411 (PAGE_MASK
& (unsigned long __force
)addr
);
413 mmiotrace_iounmap(addr
);
415 /* Use the vm area unlocked, assuming the caller
416 ensures there isn't another iounmap for the same address
417 in parallel. Reuse of the virtual address is prevented by
418 leaving it in the global lists until we're done with it.
419 cpa takes care of the direct mappings. */
420 read_lock(&vmlist_lock
);
421 for (p
= vmlist
; p
; p
= p
->next
) {
422 if (p
->addr
== (void __force
*)addr
)
425 read_unlock(&vmlist_lock
);
428 printk(KERN_ERR
"iounmap: bad address %p\n", addr
);
433 free_memtype(p
->phys_addr
, p
->phys_addr
+ get_vm_area_size(p
));
435 /* Finally remove it */
436 o
= remove_vm_area((void __force
*)addr
);
437 BUG_ON(p
!= o
|| o
== NULL
);
440 EXPORT_SYMBOL(iounmap
);
443 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
446 void *xlate_dev_mem_ptr(unsigned long phys
)
449 unsigned long start
= phys
& PAGE_MASK
;
451 /* If page is RAM, we can use __va. Otherwise ioremap and unmap. */
452 if (page_is_ram(start
>> PAGE_SHIFT
))
455 addr
= (void __force
*)ioremap_default(start
, PAGE_SIZE
);
457 addr
= (void *)((unsigned long)addr
| (phys
& ~PAGE_MASK
));
462 void unxlate_dev_mem_ptr(unsigned long phys
, void *addr
)
464 if (page_is_ram(phys
>> PAGE_SHIFT
))
467 iounmap((void __iomem
*)((unsigned long)addr
& PAGE_MASK
));
471 static int __initdata early_ioremap_debug
;
473 static int __init
early_ioremap_debug_setup(char *str
)
475 early_ioremap_debug
= 1;
479 early_param("early_ioremap_debug", early_ioremap_debug_setup
);
481 static __initdata
int after_paging_init
;
482 static pte_t bm_pte
[PAGE_SIZE
/sizeof(pte_t
)] __page_aligned_bss
;
484 static inline pmd_t
* __init
early_ioremap_pmd(unsigned long addr
)
486 /* Don't assume we're using swapper_pg_dir at this point */
487 pgd_t
*base
= __va(read_cr3());
488 pgd_t
*pgd
= &base
[pgd_index(addr
)];
489 pud_t
*pud
= pud_offset(pgd
, addr
);
490 pmd_t
*pmd
= pmd_offset(pud
, addr
);
495 static inline pte_t
* __init
early_ioremap_pte(unsigned long addr
)
497 return &bm_pte
[pte_index(addr
)];
500 static unsigned long slot_virt
[FIX_BTMAPS_SLOTS
] __initdata
;
502 void __init
early_ioremap_init(void)
507 if (early_ioremap_debug
)
508 printk(KERN_INFO
"early_ioremap_init()\n");
510 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++)
511 slot_virt
[i
] = __fix_to_virt(FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*i
);
513 pmd
= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN
));
514 memset(bm_pte
, 0, sizeof(bm_pte
));
515 pmd_populate_kernel(&init_mm
, pmd
, bm_pte
);
518 * The boot-ioremap range spans multiple pmds, for which
519 * we are not prepared:
521 if (pmd
!= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END
))) {
523 printk(KERN_WARNING
"pmd %p != %p\n",
524 pmd
, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END
)));
525 printk(KERN_WARNING
"fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
526 fix_to_virt(FIX_BTMAP_BEGIN
));
527 printk(KERN_WARNING
"fix_to_virt(FIX_BTMAP_END): %08lx\n",
528 fix_to_virt(FIX_BTMAP_END
));
530 printk(KERN_WARNING
"FIX_BTMAP_END: %d\n", FIX_BTMAP_END
);
531 printk(KERN_WARNING
"FIX_BTMAP_BEGIN: %d\n",
536 void __init
early_ioremap_reset(void)
538 after_paging_init
= 1;
541 static void __init
__early_set_fixmap(enum fixed_addresses idx
,
542 phys_addr_t phys
, pgprot_t flags
)
544 unsigned long addr
= __fix_to_virt(idx
);
547 if (idx
>= __end_of_fixed_addresses
) {
551 pte
= early_ioremap_pte(addr
);
553 if (pgprot_val(flags
))
554 set_pte(pte
, pfn_pte(phys
>> PAGE_SHIFT
, flags
));
556 pte_clear(&init_mm
, addr
, pte
);
557 __flush_tlb_one(addr
);
560 static inline void __init
early_set_fixmap(enum fixed_addresses idx
,
561 phys_addr_t phys
, pgprot_t prot
)
563 if (after_paging_init
)
564 __set_fixmap(idx
, phys
, prot
);
566 __early_set_fixmap(idx
, phys
, prot
);
569 static inline void __init
early_clear_fixmap(enum fixed_addresses idx
)
571 if (after_paging_init
)
574 __early_set_fixmap(idx
, 0, __pgprot(0));
577 static void __iomem
*prev_map
[FIX_BTMAPS_SLOTS
] __initdata
;
578 static unsigned long prev_size
[FIX_BTMAPS_SLOTS
] __initdata
;
580 static int __init
check_early_ioremap_leak(void)
585 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++)
592 "Debug warning: early ioremap leak of %d areas detected.\n",
595 "please boot with early_ioremap_debug and report the dmesg.\n");
599 late_initcall(check_early_ioremap_leak
);
601 static void __init __iomem
*
602 __early_ioremap(resource_size_t phys_addr
, unsigned long size
, pgprot_t prot
)
604 unsigned long offset
;
605 resource_size_t last_addr
;
606 unsigned int nrpages
;
607 enum fixed_addresses idx0
, idx
;
610 WARN_ON(system_state
!= SYSTEM_BOOTING
);
613 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++) {
621 printk(KERN_INFO
"early_iomap(%08llx, %08lx) not found slot\n",
622 (u64
)phys_addr
, size
);
627 if (early_ioremap_debug
) {
628 printk(KERN_INFO
"early_ioremap(%08llx, %08lx) [%d] => ",
629 (u64
)phys_addr
, size
, slot
);
633 /* Don't allow wraparound or zero size */
634 last_addr
= phys_addr
+ size
- 1;
635 if (!size
|| last_addr
< phys_addr
) {
640 prev_size
[slot
] = size
;
642 * Mappings have to be page-aligned
644 offset
= phys_addr
& ~PAGE_MASK
;
645 phys_addr
&= PAGE_MASK
;
646 size
= PAGE_ALIGN(last_addr
+ 1) - phys_addr
;
649 * Mappings have to fit in the FIX_BTMAP area.
651 nrpages
= size
>> PAGE_SHIFT
;
652 if (nrpages
> NR_FIX_BTMAPS
) {
660 idx0
= FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*slot
;
662 while (nrpages
> 0) {
663 early_set_fixmap(idx
, phys_addr
, prot
);
664 phys_addr
+= PAGE_SIZE
;
668 if (early_ioremap_debug
)
669 printk(KERN_CONT
"%08lx + %08lx\n", offset
, slot_virt
[slot
]);
671 prev_map
[slot
] = (void __iomem
*)(offset
+ slot_virt
[slot
]);
672 return prev_map
[slot
];
675 /* Remap an IO device */
676 void __init __iomem
*
677 early_ioremap(resource_size_t phys_addr
, unsigned long size
)
679 return __early_ioremap(phys_addr
, size
, PAGE_KERNEL_IO
);
683 void __init __iomem
*
684 early_memremap(resource_size_t phys_addr
, unsigned long size
)
686 return __early_ioremap(phys_addr
, size
, PAGE_KERNEL
);
689 void __init
early_iounmap(void __iomem
*addr
, unsigned long size
)
691 unsigned long virt_addr
;
692 unsigned long offset
;
693 unsigned int nrpages
;
694 enum fixed_addresses idx
;
698 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++) {
699 if (prev_map
[i
] == addr
) {
706 printk(KERN_INFO
"early_iounmap(%p, %08lx) not found slot\n",
712 if (prev_size
[slot
] != size
) {
713 printk(KERN_INFO
"early_iounmap(%p, %08lx) [%d] size not consistent %08lx\n",
714 addr
, size
, slot
, prev_size
[slot
]);
719 if (early_ioremap_debug
) {
720 printk(KERN_INFO
"early_iounmap(%p, %08lx) [%d]\n", addr
,
725 virt_addr
= (unsigned long)addr
;
726 if (virt_addr
< fix_to_virt(FIX_BTMAP_BEGIN
)) {
730 offset
= virt_addr
& ~PAGE_MASK
;
731 nrpages
= PAGE_ALIGN(offset
+ size
- 1) >> PAGE_SHIFT
;
733 idx
= FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*slot
;
734 while (nrpages
> 0) {
735 early_clear_fixmap(idx
);
739 prev_map
[slot
] = NULL
;