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>
16 #include <asm/cacheflush.h>
18 #include <asm/fixmap.h>
19 #include <asm/pgtable.h>
20 #include <asm/tlbflush.h>
21 #include <asm/pgalloc.h>
26 unsigned long __phys_addr(unsigned long x
)
28 if (x
>= __START_KERNEL_map
)
29 return x
- __START_KERNEL_map
+ phys_base
;
30 return x
- PAGE_OFFSET
;
32 EXPORT_SYMBOL(__phys_addr
);
34 static inline int phys_addr_valid(unsigned long addr
)
36 return addr
< (1UL << boot_cpu_data
.x86_phys_bits
);
41 static inline int phys_addr_valid(unsigned long addr
)
48 int page_is_ram(unsigned long pagenr
)
50 resource_size_t addr
, end
;
54 * A special case is the first 4Kb of memory;
55 * This is a BIOS owned area, not kernel ram, but generally
56 * not listed as such in the E820 table.
62 * Second special case: Some BIOSen report the PC BIOS
63 * area (640->1Mb) as ram even though it is not.
65 if (pagenr
>= (BIOS_BEGIN
>> PAGE_SHIFT
) &&
66 pagenr
< (BIOS_END
>> PAGE_SHIFT
))
69 for (i
= 0; i
< e820
.nr_map
; i
++) {
73 if (e820
.map
[i
].type
!= E820_RAM
)
75 addr
= (e820
.map
[i
].addr
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
76 end
= (e820
.map
[i
].addr
+ e820
.map
[i
].size
) >> PAGE_SHIFT
;
79 if ((pagenr
>= addr
) && (pagenr
< end
))
86 * Fix up the linear direct mapping of the kernel to avoid cache attribute
89 int ioremap_change_attr(unsigned long vaddr
, unsigned long size
,
90 unsigned long prot_val
)
92 unsigned long nrpages
= size
>> PAGE_SHIFT
;
98 err
= _set_memory_uc(vaddr
, nrpages
);
101 err
= _set_memory_wc(vaddr
, nrpages
);
104 err
= _set_memory_wb(vaddr
, nrpages
);
112 * Remap an arbitrary physical address space into the kernel virtual
113 * address space. Needed when the kernel wants to access high addresses
116 * NOTE! We need to allow non-page-aligned mappings too: we will obviously
117 * have to convert them into an offset in a page-aligned mapping, but the
118 * caller shouldn't need to know that small detail.
120 static void __iomem
*__ioremap_caller(resource_size_t phys_addr
,
121 unsigned long size
, unsigned long prot_val
, void *caller
)
123 unsigned long pfn
, offset
, vaddr
;
124 resource_size_t last_addr
;
125 struct vm_struct
*area
;
126 unsigned long new_prot_val
;
130 /* Don't allow wraparound or zero size */
131 last_addr
= phys_addr
+ size
- 1;
132 if (!size
|| last_addr
< phys_addr
)
135 if (!phys_addr_valid(phys_addr
)) {
136 printk(KERN_WARNING
"ioremap: invalid physical address %llx\n",
137 (unsigned long long)phys_addr
);
143 * Don't remap the low PCI/ISA area, it's always mapped..
145 if (is_ISA_range(phys_addr
, last_addr
))
146 return (__force
void __iomem
*)phys_to_virt(phys_addr
);
149 * Don't allow anybody to remap normal RAM that we're using..
151 for (pfn
= phys_addr
>> PAGE_SHIFT
;
152 (pfn
<< PAGE_SHIFT
) < (last_addr
& PAGE_MASK
);
155 int is_ram
= page_is_ram(pfn
);
157 if (is_ram
&& pfn_valid(pfn
) && !PageReserved(pfn_to_page(pfn
)))
159 WARN_ON_ONCE(is_ram
);
163 * Mappings have to be page-aligned
165 offset
= phys_addr
& ~PAGE_MASK
;
166 phys_addr
&= PAGE_MASK
;
167 size
= PAGE_ALIGN(last_addr
+1) - phys_addr
;
169 retval
= reserve_memtype(phys_addr
, phys_addr
+ size
,
170 prot_val
, &new_prot_val
);
172 pr_debug("Warning: reserve_memtype returned %d\n", retval
);
176 if (prot_val
!= new_prot_val
) {
178 * Do not fallback to certain memory types with certain
180 * - request is uc-, return cannot be write-back
181 * - request is uc-, return cannot be write-combine
182 * - request is write-combine, return cannot be write-back
184 if ((prot_val
== _PAGE_CACHE_UC_MINUS
&&
185 (new_prot_val
== _PAGE_CACHE_WB
||
186 new_prot_val
== _PAGE_CACHE_WC
)) ||
187 (prot_val
== _PAGE_CACHE_WC
&&
188 new_prot_val
== _PAGE_CACHE_WB
)) {
190 "ioremap error for 0x%llx-0x%llx, requested 0x%lx, got 0x%lx\n",
191 (unsigned long long)phys_addr
,
192 (unsigned long long)(phys_addr
+ size
),
193 prot_val
, new_prot_val
);
194 free_memtype(phys_addr
, phys_addr
+ size
);
197 prot_val
= new_prot_val
;
203 prot
= PAGE_KERNEL_NOCACHE
;
205 case _PAGE_CACHE_UC_MINUS
:
206 prot
= PAGE_KERNEL_UC_MINUS
;
209 prot
= PAGE_KERNEL_WC
;
219 area
= get_vm_area_caller(size
, VM_IOREMAP
, caller
);
222 area
->phys_addr
= phys_addr
;
223 vaddr
= (unsigned long) area
->addr
;
224 if (ioremap_page_range(vaddr
, vaddr
+ size
, phys_addr
, prot
)) {
225 free_memtype(phys_addr
, phys_addr
+ size
);
230 if (ioremap_change_attr(vaddr
, size
, prot_val
) < 0) {
231 free_memtype(phys_addr
, phys_addr
+ size
);
236 return (void __iomem
*) (vaddr
+ offset
);
240 * ioremap_nocache - map bus memory into CPU space
241 * @offset: bus address of the memory
242 * @size: size of the resource to map
244 * ioremap_nocache performs a platform specific sequence of operations to
245 * make bus memory CPU accessible via the readb/readw/readl/writeb/
246 * writew/writel functions and the other mmio helpers. The returned
247 * address is not guaranteed to be usable directly as a virtual
250 * This version of ioremap ensures that the memory is marked uncachable
251 * on the CPU as well as honouring existing caching rules from things like
252 * the PCI bus. Note that there are other caches and buffers on many
253 * busses. In particular driver authors should read up on PCI writes
255 * It's useful if some control registers are in such an area and
256 * write combining or read caching is not desirable:
258 * Must be freed with iounmap.
260 void __iomem
*ioremap_nocache(resource_size_t phys_addr
, unsigned long size
)
263 * Ideally, this should be:
264 * pat_enabled ? _PAGE_CACHE_UC : _PAGE_CACHE_UC_MINUS;
266 * Till we fix all X drivers to use ioremap_wc(), we will use
269 unsigned long val
= _PAGE_CACHE_UC_MINUS
;
271 return __ioremap_caller(phys_addr
, size
, val
,
272 __builtin_return_address(0));
274 EXPORT_SYMBOL(ioremap_nocache
);
277 * ioremap_wc - map memory into CPU space write combined
278 * @offset: bus address of the memory
279 * @size: size of the resource to map
281 * This version of ioremap ensures that the memory is marked write combining.
282 * Write combining allows faster writes to some hardware devices.
284 * Must be freed with iounmap.
286 void __iomem
*ioremap_wc(unsigned long phys_addr
, unsigned long size
)
289 return __ioremap_caller(phys_addr
, size
, _PAGE_CACHE_WC
,
290 __builtin_return_address(0));
292 return ioremap_nocache(phys_addr
, size
);
294 EXPORT_SYMBOL(ioremap_wc
);
296 void __iomem
*ioremap_cache(resource_size_t phys_addr
, unsigned long size
)
298 return __ioremap_caller(phys_addr
, size
, _PAGE_CACHE_WB
,
299 __builtin_return_address(0));
301 EXPORT_SYMBOL(ioremap_cache
);
304 * iounmap - Free a IO remapping
305 * @addr: virtual address from ioremap_*
307 * Caller must ensure there is only one unmapping for the same pointer.
309 void iounmap(volatile void __iomem
*addr
)
311 struct vm_struct
*p
, *o
;
313 if ((void __force
*)addr
<= high_memory
)
317 * __ioremap special-cases the PCI/ISA range by not instantiating a
318 * vm_area and by simply returning an address into the kernel mapping
319 * of ISA space. So handle that here.
321 if ((void __force
*)addr
>= phys_to_virt(ISA_START_ADDRESS
) &&
322 (void __force
*)addr
< phys_to_virt(ISA_END_ADDRESS
))
325 addr
= (volatile void __iomem
*)
326 (PAGE_MASK
& (unsigned long __force
)addr
);
328 /* Use the vm area unlocked, assuming the caller
329 ensures there isn't another iounmap for the same address
330 in parallel. Reuse of the virtual address is prevented by
331 leaving it in the global lists until we're done with it.
332 cpa takes care of the direct mappings. */
333 read_lock(&vmlist_lock
);
334 for (p
= vmlist
; p
; p
= p
->next
) {
335 if (p
->addr
== (void __force
*)addr
)
338 read_unlock(&vmlist_lock
);
341 printk(KERN_ERR
"iounmap: bad address %p\n", addr
);
346 free_memtype(p
->phys_addr
, p
->phys_addr
+ get_vm_area_size(p
));
348 /* Finally remove it */
349 o
= remove_vm_area((void __force
*)addr
);
350 BUG_ON(p
!= o
|| o
== NULL
);
353 EXPORT_SYMBOL(iounmap
);
356 * Convert a physical pointer to a virtual kernel pointer for /dev/mem
359 void *xlate_dev_mem_ptr(unsigned long phys
)
362 unsigned long start
= phys
& PAGE_MASK
;
364 /* If page is RAM, we can use __va. Otherwise ioremap and unmap. */
365 if (page_is_ram(start
>> PAGE_SHIFT
))
368 addr
= (void __force
*)ioremap(start
, PAGE_SIZE
);
370 addr
= (void *)((unsigned long)addr
| (phys
& ~PAGE_MASK
));
375 void unxlate_dev_mem_ptr(unsigned long phys
, void *addr
)
377 if (page_is_ram(phys
>> PAGE_SHIFT
))
380 iounmap((void __iomem
*)((unsigned long)addr
& PAGE_MASK
));
384 int __initdata early_ioremap_debug
;
386 static int __init
early_ioremap_debug_setup(char *str
)
388 early_ioremap_debug
= 1;
392 early_param("early_ioremap_debug", early_ioremap_debug_setup
);
394 static __initdata
int after_paging_init
;
395 static pte_t bm_pte
[PAGE_SIZE
/sizeof(pte_t
)] __page_aligned_bss
;
397 static inline pmd_t
* __init
early_ioremap_pmd(unsigned long addr
)
399 /* Don't assume we're using swapper_pg_dir at this point */
400 pgd_t
*base
= __va(read_cr3());
401 pgd_t
*pgd
= &base
[pgd_index(addr
)];
402 pud_t
*pud
= pud_offset(pgd
, addr
);
403 pmd_t
*pmd
= pmd_offset(pud
, addr
);
408 static inline pte_t
* __init
early_ioremap_pte(unsigned long addr
)
410 return &bm_pte
[pte_index(addr
)];
413 void __init
early_ioremap_init(void)
417 if (early_ioremap_debug
)
418 printk(KERN_INFO
"early_ioremap_init()\n");
420 pmd
= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN
));
421 memset(bm_pte
, 0, sizeof(bm_pte
));
422 pmd_populate_kernel(&init_mm
, pmd
, bm_pte
);
425 * The boot-ioremap range spans multiple pmds, for which
426 * we are not prepared:
428 if (pmd
!= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END
))) {
430 printk(KERN_WARNING
"pmd %p != %p\n",
431 pmd
, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END
)));
432 printk(KERN_WARNING
"fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
433 fix_to_virt(FIX_BTMAP_BEGIN
));
434 printk(KERN_WARNING
"fix_to_virt(FIX_BTMAP_END): %08lx\n",
435 fix_to_virt(FIX_BTMAP_END
));
437 printk(KERN_WARNING
"FIX_BTMAP_END: %d\n", FIX_BTMAP_END
);
438 printk(KERN_WARNING
"FIX_BTMAP_BEGIN: %d\n",
443 void __init
early_ioremap_clear(void)
447 if (early_ioremap_debug
)
448 printk(KERN_INFO
"early_ioremap_clear()\n");
450 pmd
= early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN
));
452 paravirt_release_pte(__pa(bm_pte
) >> PAGE_SHIFT
);
456 void __init
early_ioremap_reset(void)
458 enum fixed_addresses idx
;
459 unsigned long addr
, phys
;
462 after_paging_init
= 1;
463 for (idx
= FIX_BTMAP_BEGIN
; idx
>= FIX_BTMAP_END
; idx
--) {
464 addr
= fix_to_virt(idx
);
465 pte
= early_ioremap_pte(addr
);
466 if (pte_present(*pte
)) {
467 phys
= pte_val(*pte
) & PAGE_MASK
;
468 set_fixmap(idx
, phys
);
473 static void __init
__early_set_fixmap(enum fixed_addresses idx
,
474 unsigned long phys
, pgprot_t flags
)
476 unsigned long addr
= __fix_to_virt(idx
);
479 if (idx
>= __end_of_fixed_addresses
) {
483 pte
= early_ioremap_pte(addr
);
485 if (pgprot_val(flags
))
486 set_pte(pte
, pfn_pte(phys
>> PAGE_SHIFT
, flags
));
488 pte_clear(&init_mm
, addr
, pte
);
489 __flush_tlb_one(addr
);
492 static inline void __init
early_set_fixmap(enum fixed_addresses idx
,
495 if (after_paging_init
)
496 set_fixmap(idx
, phys
);
498 __early_set_fixmap(idx
, phys
, PAGE_KERNEL
);
501 static inline void __init
early_clear_fixmap(enum fixed_addresses idx
)
503 if (after_paging_init
)
506 __early_set_fixmap(idx
, 0, __pgprot(0));
510 int __initdata early_ioremap_nested
;
512 static int __init
check_early_ioremap_leak(void)
514 if (!early_ioremap_nested
)
518 "Debug warning: early ioremap leak of %d areas detected.\n",
519 early_ioremap_nested
);
521 "please boot with early_ioremap_debug and report the dmesg.\n");
526 late_initcall(check_early_ioremap_leak
);
528 void __init
*early_ioremap(unsigned long phys_addr
, unsigned long size
)
530 unsigned long offset
, last_addr
;
531 unsigned int nrpages
, nesting
;
532 enum fixed_addresses idx0
, idx
;
534 WARN_ON(system_state
!= SYSTEM_BOOTING
);
536 nesting
= early_ioremap_nested
;
537 if (early_ioremap_debug
) {
538 printk(KERN_INFO
"early_ioremap(%08lx, %08lx) [%d] => ",
539 phys_addr
, size
, nesting
);
543 /* Don't allow wraparound or zero size */
544 last_addr
= phys_addr
+ size
- 1;
545 if (!size
|| last_addr
< phys_addr
) {
550 if (nesting
>= FIX_BTMAPS_NESTING
) {
554 early_ioremap_nested
++;
556 * Mappings have to be page-aligned
558 offset
= phys_addr
& ~PAGE_MASK
;
559 phys_addr
&= PAGE_MASK
;
560 size
= PAGE_ALIGN(last_addr
) - phys_addr
;
563 * Mappings have to fit in the FIX_BTMAP area.
565 nrpages
= size
>> PAGE_SHIFT
;
566 if (nrpages
> NR_FIX_BTMAPS
) {
574 idx0
= FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*nesting
;
576 while (nrpages
> 0) {
577 early_set_fixmap(idx
, phys_addr
);
578 phys_addr
+= PAGE_SIZE
;
582 if (early_ioremap_debug
)
583 printk(KERN_CONT
"%08lx + %08lx\n", offset
, fix_to_virt(idx0
));
585 return (void *) (offset
+ fix_to_virt(idx0
));
588 void __init
early_iounmap(void *addr
, unsigned long size
)
590 unsigned long virt_addr
;
591 unsigned long offset
;
592 unsigned int nrpages
;
593 enum fixed_addresses idx
;
596 nesting
= --early_ioremap_nested
;
597 if (WARN_ON(nesting
< 0))
600 if (early_ioremap_debug
) {
601 printk(KERN_INFO
"early_iounmap(%p, %08lx) [%d]\n", addr
,
606 virt_addr
= (unsigned long)addr
;
607 if (virt_addr
< fix_to_virt(FIX_BTMAP_BEGIN
)) {
611 offset
= virt_addr
& ~PAGE_MASK
;
612 nrpages
= PAGE_ALIGN(offset
+ size
- 1) >> PAGE_SHIFT
;
614 idx
= FIX_BTMAP_BEGIN
- NR_FIX_BTMAPS
*nesting
;
615 while (nrpages
> 0) {
616 early_clear_fixmap(idx
);
622 void __this_fixmap_does_not_exist(void)