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>
27 int page_is_ram(unsigned long pagenr
)
29 resource_size_t addr
, end
;
33 * A special case is the first 4Kb of memory;
34 * This is a BIOS owned area, not kernel ram, but generally
35 * not listed as such in the E820 table.
41 * Second special case: Some BIOSen report the PC BIOS
42 * area (640->1Mb) as ram even though it is not.
44 if (pagenr
>= (BIOS_BEGIN
>> PAGE_SHIFT
) &&
45 pagenr
< (BIOS_END
>> PAGE_SHIFT
))
48 for (i
= 0; i
< e820
.nr_map
; i
++) {
52 if (e820
.map
[i
].type
!= E820_RAM
)
54 addr
= (e820
.map
[i
].addr
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
55 end
= (e820
.map
[i
].addr
+ e820
.map
[i
].size
) >> PAGE_SHIFT
;
58 if ((pagenr
>= addr
) && (pagenr
< end
))
65 * Fix up the linear direct mapping of the kernel to avoid cache attribute
68 int ioremap_change_attr(unsigned long vaddr
, unsigned long size
,
69 unsigned long prot_val
)
71 unsigned long nrpages
= size
>> PAGE_SHIFT
;
77 err
= _set_memory_uc(vaddr
, nrpages
);
80 err
= _set_memory_wc(vaddr
, nrpages
);
83 err
= _set_memory_wb(vaddr
, nrpages
);
91 * Remap an arbitrary physical address space into the kernel virtual
92 * address space. Needed when the kernel wants to access high addresses
95 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
96 * have to convert them into an offset in a page-aligned mapping, but the
97 * caller shouldn't need to know that small detail.
99 static void __iomem
*__ioremap_caller(resource_size_t phys_addr
,
100 unsigned long size
, unsigned long prot_val
, void *caller
)
102 unsigned long pfn
, offset
, vaddr
;
103 resource_size_t last_addr
;
104 const resource_size_t unaligned_phys_addr
= phys_addr
;
105 const unsigned long unaligned_size
= size
;
106 struct vm_struct
*area
;
107 unsigned long new_prot_val
;
110 void __iomem
*ret_addr
;
112 /* Don't allow wraparound or zero size */
113 last_addr
= phys_addr
+ size
- 1;
114 if (!size
|| last_addr
< phys_addr
)
117 if (!phys_addr_valid(phys_addr
)) {
118 printk(KERN_WARNING
"ioremap: invalid physical address %llx\n",
119 (unsigned long long)phys_addr
);
125 * Don't remap the low PCI/ISA area, it's always mapped..
127 if (is_ISA_range(phys_addr
, last_addr
))
128 return (__force
void __iomem
*)phys_to_virt(phys_addr
);
131 * Check if the request spans more than any BAR in the iomem resource
134 WARN_ONCE(iomem_map_sanity_check(phys_addr
, size
),
135 KERN_INFO
"Info: mapping multiple BARs. Your kernel is fine.");
138 * Don't allow anybody to remap normal RAM that we're using..
140 for (pfn
= phys_addr
>> PAGE_SHIFT
;
141 (pfn
<< PAGE_SHIFT
) < (last_addr
& PAGE_MASK
);
144 int is_ram
= page_is_ram(pfn
);
146 if (is_ram
&& pfn_valid(pfn
) && !PageReserved(pfn_to_page(pfn
)))
148 WARN_ON_ONCE(is_ram
);
152 * Mappings have to be page-aligned
154 offset
= phys_addr
& ~PAGE_MASK
;
155 phys_addr
&= PAGE_MASK
;
156 size
= PAGE_ALIGN(last_addr
+1) - phys_addr
;
158 retval
= reserve_memtype(phys_addr
, (u64
)phys_addr
+ size
,
159 prot_val
, &new_prot_val
);
161 printk(KERN_ERR
"ioremap reserve_memtype failed %d\n", retval
);
165 if (prot_val
!= new_prot_val
) {
166 if (!is_new_memtype_allowed(phys_addr
, size
,
167 prot_val
, new_prot_val
)) {
169 "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n",
170 (unsigned long long)phys_addr
,
171 (unsigned long long)(phys_addr
+ size
),
172 prot_val
, new_prot_val
);
173 goto err_free_memtype
;
175 prot_val
= new_prot_val
;
181 prot
= PAGE_KERNEL_IO_NOCACHE
;
183 case _PAGE_CACHE_UC_MINUS
:
184 prot
= PAGE_KERNEL_IO_UC_MINUS
;
187 prot
= PAGE_KERNEL_IO_WC
;
190 prot
= PAGE_KERNEL_IO
;
197 area
= get_vm_area_caller(size
, VM_IOREMAP
, caller
);
199 goto err_free_memtype
;
200 area
->phys_addr
= phys_addr
;
201 vaddr
= (unsigned long) area
->addr
;
203 if (kernel_map_sync_memtype(phys_addr
, size
, prot_val
))
206 if (ioremap_page_range(vaddr
, vaddr
+ size
, phys_addr
, prot
))
209 ret_addr
= (void __iomem
*) (vaddr
+ offset
);
210 mmiotrace_ioremap(unaligned_phys_addr
, unaligned_size
, ret_addr
);
216 free_memtype(phys_addr
, phys_addr
+ size
);
221 * ioremap_nocache - map bus memory into CPU space
222 * @offset: bus address of the memory
223 * @size: size of the resource to map
225 * ioremap_nocache performs a platform specific sequence of operations to
226 * make bus memory CPU accessible via the readb/readw/readl/writeb/
227 * writew/writel functions and the other mmio helpers. The returned
228 * address is not guaranteed to be usable directly as a virtual
231 * This version of ioremap ensures that the memory is marked uncachable
232 * on the CPU as well as honouring existing caching rules from things like
233 * the PCI bus. Note that there are other caches and buffers on many
234 * busses. In particular driver authors should read up on PCI writes
236 * It's useful if some control registers are in such an area and
237 * write combining or read caching is not desirable:
239 * Must be freed with iounmap.
241 void __iomem
*ioremap_nocache(resource_size_t phys_addr
, unsigned long size
)
244 * Ideally, this should be:
245 * pat_enabled ? _PAGE_CACHE_UC : _PAGE_CACHE_UC_MINUS;
247 * Till we fix all X drivers to use ioremap_wc(), we will use
250 unsigned long val
= _PAGE_CACHE_UC_MINUS
;
252 return __ioremap_caller(phys_addr
, size
, val
,
253 __builtin_return_address(0));
255 EXPORT_SYMBOL(ioremap_nocache
);
258 * ioremap_wc - map memory into CPU space write combined
259 * @offset: bus address of the memory
260 * @size: size of the resource to map
262 * This version of ioremap ensures that the memory is marked write combining.
263 * Write combining allows faster writes to some hardware devices.
265 * Must be freed with iounmap.
267 void __iomem
*ioremap_wc(resource_size_t phys_addr
, unsigned long size
)
270 return __ioremap_caller(phys_addr
, size
, _PAGE_CACHE_WC
,
271 __builtin_return_address(0));
273 return ioremap_nocache(phys_addr
, size
);
275 EXPORT_SYMBOL(ioremap_wc
);
277 void __iomem
*ioremap_cache(resource_size_t phys_addr
, unsigned long size
)
279 return __ioremap_caller(phys_addr
, size
, _PAGE_CACHE_WB
,
280 __builtin_return_address(0));
282 EXPORT_SYMBOL(ioremap_cache
);
284 static void __iomem
*ioremap_default(resource_size_t phys_addr
,
292 * - WB for WB-able memory and no other conflicting mappings
293 * - UC_MINUS for non-WB-able memory with no other conflicting mappings
294 * - Inherit from confliting mappings otherwise
296 err
= reserve_memtype(phys_addr
, phys_addr
+ size
,
297 _PAGE_CACHE_WB
, &flags
);
301 ret
= __ioremap_caller(phys_addr
, size
, flags
,
302 __builtin_return_address(0));
304 free_memtype(phys_addr
, phys_addr
+ size
);
308 void __iomem
*ioremap_prot(resource_size_t phys_addr
, unsigned long size
,
309 unsigned long prot_val
)
311 return __ioremap_caller(phys_addr
, size
, (prot_val
& _PAGE_CACHE_MASK
),
312 __builtin_return_address(0));
314 EXPORT_SYMBOL(ioremap_prot
);
317 * iounmap - Free a IO remapping
318 * @addr: virtual address from ioremap_*
320 * Caller must ensure there is only one unmapping for the same pointer.
322 void iounmap(volatile void __iomem
*addr
)
324 struct vm_struct
*p
, *o
;
326 if ((void __force
*)addr
<= high_memory
)
330 * __ioremap special-cases the PCI/ISA range by not instantiating a
331 * vm_area and by simply returning an address into the kernel mapping
332 * of ISA space. So handle that here.
334 if ((void __force
*)addr
>= phys_to_virt(ISA_START_ADDRESS
) &&
335 (void __force
*)addr
< phys_to_virt(ISA_END_ADDRESS
))
338 addr
= (volatile void __iomem
*)
339 (PAGE_MASK
& (unsigned long __force
)addr
);
341 mmiotrace_iounmap(addr
);
343 /* Use the vm area unlocked, assuming the caller
344 ensures there isn't another iounmap for the same address
345 in parallel. Reuse of the virtual address is prevented by
346 leaving it in the global lists until we're done with it.
347 cpa takes care of the direct mappings. */
348 read_lock(&vmlist_lock
);
349 for (p
= vmlist
; p
; p
= p
->next
) {
350 if (p
->addr
== (void __force
*)addr
)
353 read_unlock(&vmlist_lock
);
356 printk(KERN_ERR
"iounmap: bad address %p\n", addr
);
361 free_memtype(p
->phys_addr
, p
->phys_addr
+ get_vm_area_size(p
));
363 /* Finally remove it */
364 o
= remove_vm_area((void __force
*)addr
);
365 BUG_ON(p
!= o
|| o
== NULL
);
368 EXPORT_SYMBOL(iounmap
);
371 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
374 void *xlate_dev_mem_ptr(unsigned long phys
)
377 unsigned long start
= phys
& PAGE_MASK
;
379 /* If page is RAM, we can use __va. Otherwise ioremap and unmap. */
380 if (page_is_ram(start
>> PAGE_SHIFT
))
383 addr
= (void __force
*)ioremap_default(start
, PAGE_SIZE
);
385 addr
= (void *)((unsigned long)addr
| (phys
& ~PAGE_MASK
));
390 void unxlate_dev_mem_ptr(unsigned long phys
, void *addr
)
392 if (page_is_ram(phys
>> PAGE_SHIFT
))
395 iounmap((void __iomem
*)((unsigned long)addr
& PAGE_MASK
));
399 static int __initdata early_ioremap_debug
;
401 static int __init
early_ioremap_debug_setup(char *str
)
403 early_ioremap_debug
= 1;
407 early_param("early_ioremap_debug", early_ioremap_debug_setup
);
409 static __initdata
int after_paging_init
;
410 static pte_t bm_pte
[PAGE_SIZE
/sizeof(pte_t
)] __page_aligned_bss
;
412 static inline pmd_t
* __init
early_ioremap_pmd(unsigned long addr
)
414 /* Don't assume we're using swapper_pg_dir at this point */
415 pgd_t
*base
= __va(read_cr3());
416 pgd_t
*pgd
= &base
[pgd_index(addr
)];
417 pud_t
*pud
= pud_offset(pgd
, addr
);
418 pmd_t
*pmd
= pmd_offset(pud
, addr
);
423 static inline pte_t
* __init
early_ioremap_pte(unsigned long addr
)
425 return &bm_pte
[pte_index(addr
)];
428 static unsigned long slot_virt
[FIX_BTMAPS_SLOTS
] __initdata
;
430 void __init
early_ioremap_init(void)
435 if (early_ioremap_debug
)
436 printk(KERN_INFO
"early_ioremap_init()\n");
438 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++)
439 slot_virt
[i
] = __fix_to_virt(FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*i
);
441 pmd
= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN
));
442 memset(bm_pte
, 0, sizeof(bm_pte
));
443 pmd_populate_kernel(&init_mm
, pmd
, bm_pte
);
446 * The boot-ioremap range spans multiple pmds, for which
447 * we are not prepared:
449 if (pmd
!= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END
))) {
451 printk(KERN_WARNING
"pmd %p != %p\n",
452 pmd
, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END
)));
453 printk(KERN_WARNING
"fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
454 fix_to_virt(FIX_BTMAP_BEGIN
));
455 printk(KERN_WARNING
"fix_to_virt(FIX_BTMAP_END): %08lx\n",
456 fix_to_virt(FIX_BTMAP_END
));
458 printk(KERN_WARNING
"FIX_BTMAP_END: %d\n", FIX_BTMAP_END
);
459 printk(KERN_WARNING
"FIX_BTMAP_BEGIN: %d\n",
464 void __init
early_ioremap_reset(void)
466 after_paging_init
= 1;
469 static void __init
__early_set_fixmap(enum fixed_addresses idx
,
470 phys_addr_t phys
, pgprot_t flags
)
472 unsigned long addr
= __fix_to_virt(idx
);
475 if (idx
>= __end_of_fixed_addresses
) {
479 pte
= early_ioremap_pte(addr
);
481 if (pgprot_val(flags
))
482 set_pte(pte
, pfn_pte(phys
>> PAGE_SHIFT
, flags
));
484 pte_clear(&init_mm
, addr
, pte
);
485 __flush_tlb_one(addr
);
488 static inline void __init
early_set_fixmap(enum fixed_addresses idx
,
489 phys_addr_t phys
, pgprot_t prot
)
491 if (after_paging_init
)
492 __set_fixmap(idx
, phys
, prot
);
494 __early_set_fixmap(idx
, phys
, prot
);
497 static inline void __init
early_clear_fixmap(enum fixed_addresses idx
)
499 if (after_paging_init
)
502 __early_set_fixmap(idx
, 0, __pgprot(0));
505 static void __iomem
*prev_map
[FIX_BTMAPS_SLOTS
] __initdata
;
506 static unsigned long prev_size
[FIX_BTMAPS_SLOTS
] __initdata
;
508 static int __init
check_early_ioremap_leak(void)
513 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++)
520 "Debug warning: early ioremap leak of %d areas detected.\n",
523 "please boot with early_ioremap_debug and report the dmesg.\n");
527 late_initcall(check_early_ioremap_leak
);
529 static void __init __iomem
*
530 __early_ioremap(resource_size_t phys_addr
, unsigned long size
, pgprot_t prot
)
532 unsigned long offset
;
533 resource_size_t last_addr
;
534 unsigned int nrpages
;
535 enum fixed_addresses idx0
, idx
;
538 WARN_ON(system_state
!= SYSTEM_BOOTING
);
541 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++) {
549 printk(KERN_INFO
"early_iomap(%08llx, %08lx) not found slot\n",
550 (u64
)phys_addr
, size
);
555 if (early_ioremap_debug
) {
556 printk(KERN_INFO
"early_ioremap(%08llx, %08lx) [%d] => ",
557 (u64
)phys_addr
, size
, slot
);
561 /* Don't allow wraparound or zero size */
562 last_addr
= phys_addr
+ size
- 1;
563 if (!size
|| last_addr
< phys_addr
) {
568 prev_size
[slot
] = size
;
570 * Mappings have to be page-aligned
572 offset
= phys_addr
& ~PAGE_MASK
;
573 phys_addr
&= PAGE_MASK
;
574 size
= PAGE_ALIGN(last_addr
+ 1) - phys_addr
;
577 * Mappings have to fit in the FIX_BTMAP area.
579 nrpages
= size
>> PAGE_SHIFT
;
580 if (nrpages
> NR_FIX_BTMAPS
) {
588 idx0
= FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*slot
;
590 while (nrpages
> 0) {
591 early_set_fixmap(idx
, phys_addr
, prot
);
592 phys_addr
+= PAGE_SIZE
;
596 if (early_ioremap_debug
)
597 printk(KERN_CONT
"%08lx + %08lx\n", offset
, slot_virt
[slot
]);
599 prev_map
[slot
] = (void __iomem
*)(offset
+ slot_virt
[slot
]);
600 return prev_map
[slot
];
603 /* Remap an IO device */
604 void __init __iomem
*
605 early_ioremap(resource_size_t phys_addr
, unsigned long size
)
607 return __early_ioremap(phys_addr
, size
, PAGE_KERNEL_IO
);
611 void __init __iomem
*
612 early_memremap(resource_size_t phys_addr
, unsigned long size
)
614 return __early_ioremap(phys_addr
, size
, PAGE_KERNEL
);
617 void __init
early_iounmap(void __iomem
*addr
, unsigned long size
)
619 unsigned long virt_addr
;
620 unsigned long offset
;
621 unsigned int nrpages
;
622 enum fixed_addresses idx
;
626 for (i
= 0; i
< FIX_BTMAPS_SLOTS
; i
++) {
627 if (prev_map
[i
] == addr
) {
634 printk(KERN_INFO
"early_iounmap(%p, %08lx) not found slot\n",
640 if (prev_size
[slot
] != size
) {
641 printk(KERN_INFO
"early_iounmap(%p, %08lx) [%d] size not consistent %08lx\n",
642 addr
, size
, slot
, prev_size
[slot
]);
647 if (early_ioremap_debug
) {
648 printk(KERN_INFO
"early_iounmap(%p, %08lx) [%d]\n", addr
,
653 virt_addr
= (unsigned long)addr
;
654 if (virt_addr
< fix_to_virt(FIX_BTMAP_BEGIN
)) {
658 offset
= virt_addr
& ~PAGE_MASK
;
659 nrpages
= PAGE_ALIGN(offset
+ size
- 1) >> PAGE_SHIFT
;
661 idx
= FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*slot
;
662 while (nrpages
> 0) {
663 early_clear_fixmap(idx
);
667 prev_map
[slot
] = NULL
;