]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Re-map IO memory to kernel address space so that we can access it. |
4 | * This is needed for high PCI addresses that aren't mapped in the | |
5 | * 640k-1MB IO memory area on PC's | |
6 | * | |
7 | * (C) Copyright 1995 1996 Linus Torvalds | |
8 | */ | |
9 | ||
57c8a661 | 10 | #include <linux/memblock.h> |
1da177e4 | 11 | #include <linux/init.h> |
a148ecfd | 12 | #include <linux/io.h> |
9de94dbb | 13 | #include <linux/ioport.h> |
3cbd09e4 TG |
14 | #include <linux/slab.h> |
15 | #include <linux/vmalloc.h> | |
d61fc448 | 16 | #include <linux/mmiotrace.h> |
8f716c9b TL |
17 | #include <linux/mem_encrypt.h> |
18 | #include <linux/efi.h> | |
3cbd09e4 | 19 | |
d1163651 | 20 | #include <asm/set_memory.h> |
66441bd3 | 21 | #include <asm/e820/api.h> |
e55f31a5 | 22 | #include <asm/efi.h> |
3cbd09e4 | 23 | #include <asm/fixmap.h> |
ca5999fd | 24 | #include <linux/pgtable.h> |
3cbd09e4 | 25 | #include <asm/tlbflush.h> |
f6df72e7 | 26 | #include <asm/pgalloc.h> |
eb243d1d | 27 | #include <asm/memtype.h> |
8f716c9b | 28 | #include <asm/setup.h> |
1da177e4 | 29 | |
78c86e5e | 30 | #include "physaddr.h" |
240d3a7c | 31 | |
5da04cc8 LJ |
32 | /* |
33 | * Descriptor controlling ioremap() behavior. | |
34 | */ | |
35 | struct ioremap_desc { | |
36 | unsigned int flags; | |
0e4c12b4 TL |
37 | }; |
38 | ||
e9332cac TG |
39 | /* |
40 | * Fix up the linear direct mapping of the kernel to avoid cache attribute | |
41 | * conflicts. | |
42 | */ | |
3a96ce8c | 43 | int ioremap_change_attr(unsigned long vaddr, unsigned long size, |
b14097bd | 44 | enum page_cache_mode pcm) |
e9332cac | 45 | { |
d806e5ee | 46 | unsigned long nrpages = size >> PAGE_SHIFT; |
93809be8 | 47 | int err; |
e9332cac | 48 | |
b14097bd JG |
49 | switch (pcm) { |
50 | case _PAGE_CACHE_MODE_UC: | |
d806e5ee | 51 | default: |
1219333d | 52 | err = _set_memory_uc(vaddr, nrpages); |
d806e5ee | 53 | break; |
b14097bd | 54 | case _PAGE_CACHE_MODE_WC: |
b310f381 | 55 | err = _set_memory_wc(vaddr, nrpages); |
56 | break; | |
623dffb2 TK |
57 | case _PAGE_CACHE_MODE_WT: |
58 | err = _set_memory_wt(vaddr, nrpages); | |
59 | break; | |
b14097bd | 60 | case _PAGE_CACHE_MODE_WB: |
1219333d | 61 | err = _set_memory_wb(vaddr, nrpages); |
d806e5ee TG |
62 | break; |
63 | } | |
e9332cac TG |
64 | |
65 | return err; | |
66 | } | |
67 | ||
5da04cc8 LJ |
68 | /* Does the range (or a subset of) contain normal RAM? */ |
69 | static unsigned int __ioremap_check_ram(struct resource *res) | |
c81c8a1e | 70 | { |
0e4c12b4 | 71 | unsigned long start_pfn, stop_pfn; |
c81c8a1e RD |
72 | unsigned long i; |
73 | ||
0e4c12b4 | 74 | if ((res->flags & IORESOURCE_SYSTEM_RAM) != IORESOURCE_SYSTEM_RAM) |
5da04cc8 | 75 | return 0; |
0e4c12b4 TL |
76 | |
77 | start_pfn = (res->start + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
78 | stop_pfn = (res->end + 1) >> PAGE_SHIFT; | |
79 | if (stop_pfn > start_pfn) { | |
80 | for (i = 0; i < (stop_pfn - start_pfn); ++i) | |
81 | if (pfn_valid(start_pfn + i) && | |
82 | !PageReserved(pfn_to_page(start_pfn + i))) | |
5da04cc8 | 83 | return IORES_MAP_SYSTEM_RAM; |
0e4c12b4 TL |
84 | } |
85 | ||
5da04cc8 | 86 | return 0; |
0e4c12b4 TL |
87 | } |
88 | ||
5da04cc8 LJ |
89 | /* |
90 | * In a SEV guest, NONE and RESERVED should not be mapped encrypted because | |
91 | * there the whole memory is already encrypted. | |
92 | */ | |
93 | static unsigned int __ioremap_check_encrypted(struct resource *res) | |
0e4c12b4 | 94 | { |
5da04cc8 LJ |
95 | if (!sev_active()) |
96 | return 0; | |
97 | ||
98 | switch (res->desc) { | |
99 | case IORES_DESC_NONE: | |
100 | case IORES_DESC_RESERVED: | |
101 | break; | |
102 | default: | |
103 | return IORES_MAP_ENCRYPTED; | |
104 | } | |
105 | ||
106 | return 0; | |
0e4c12b4 TL |
107 | } |
108 | ||
985e537a TL |
109 | /* |
110 | * The EFI runtime services data area is not covered by walk_mem_res(), but must | |
111 | * be mapped encrypted when SEV is active. | |
112 | */ | |
113 | static void __ioremap_check_other(resource_size_t addr, struct ioremap_desc *desc) | |
114 | { | |
115 | if (!sev_active()) | |
116 | return; | |
117 | ||
870b4333 BP |
118 | if (!IS_ENABLED(CONFIG_EFI)) |
119 | return; | |
120 | ||
985e537a TL |
121 | if (efi_mem_type(addr) == EFI_RUNTIME_SERVICES_DATA) |
122 | desc->flags |= IORES_MAP_ENCRYPTED; | |
123 | } | |
124 | ||
5da04cc8 | 125 | static int __ioremap_collect_map_flags(struct resource *res, void *arg) |
0e4c12b4 | 126 | { |
5da04cc8 | 127 | struct ioremap_desc *desc = arg; |
0e4c12b4 | 128 | |
5da04cc8 LJ |
129 | if (!(desc->flags & IORES_MAP_SYSTEM_RAM)) |
130 | desc->flags |= __ioremap_check_ram(res); | |
0e4c12b4 | 131 | |
5da04cc8 LJ |
132 | if (!(desc->flags & IORES_MAP_ENCRYPTED)) |
133 | desc->flags |= __ioremap_check_encrypted(res); | |
c81c8a1e | 134 | |
5da04cc8 LJ |
135 | return ((desc->flags & (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED)) == |
136 | (IORES_MAP_SYSTEM_RAM | IORES_MAP_ENCRYPTED)); | |
0e4c12b4 TL |
137 | } |
138 | ||
139 | /* | |
140 | * To avoid multiple resource walks, this function walks resources marked as | |
141 | * IORESOURCE_MEM and IORESOURCE_BUSY and looking for system RAM and/or a | |
142 | * resource described not as IORES_DESC_NONE (e.g. IORES_DESC_ACPI_TABLES). | |
985e537a TL |
143 | * |
144 | * After that, deal with misc other ranges in __ioremap_check_other() which do | |
145 | * not fall into the above category. | |
0e4c12b4 TL |
146 | */ |
147 | static void __ioremap_check_mem(resource_size_t addr, unsigned long size, | |
5da04cc8 | 148 | struct ioremap_desc *desc) |
0e4c12b4 TL |
149 | { |
150 | u64 start, end; | |
151 | ||
152 | start = (u64)addr; | |
153 | end = start + size - 1; | |
5da04cc8 | 154 | memset(desc, 0, sizeof(struct ioremap_desc)); |
0e4c12b4 | 155 | |
5da04cc8 | 156 | walk_mem_res(start, end, desc, __ioremap_collect_map_flags); |
985e537a TL |
157 | |
158 | __ioremap_check_other(addr, desc); | |
c81c8a1e RD |
159 | } |
160 | ||
1da177e4 LT |
161 | /* |
162 | * Remap an arbitrary physical address space into the kernel virtual | |
5d72b4fb TK |
163 | * address space. It transparently creates kernel huge I/O mapping when |
164 | * the physical address is aligned by a huge page size (1GB or 2MB) and | |
165 | * the requested size is at least the huge page size. | |
166 | * | |
167 | * NOTE: MTRRs can override PAT memory types with a 4KB granularity. | |
168 | * Therefore, the mapping code falls back to use a smaller page toward 4KB | |
169 | * when a mapping range is covered by non-WB type of MTRRs. | |
1da177e4 LT |
170 | * |
171 | * NOTE! We need to allow non-page-aligned mappings too: we will obviously | |
172 | * have to convert them into an offset in a page-aligned mapping, but the | |
173 | * caller shouldn't need to know that small detail. | |
174 | */ | |
5da04cc8 LJ |
175 | static void __iomem * |
176 | __ioremap_caller(resource_size_t phys_addr, unsigned long size, | |
177 | enum page_cache_mode pcm, void *caller, bool encrypted) | |
1da177e4 | 178 | { |
ffa71f33 | 179 | unsigned long offset, vaddr; |
0e4c12b4 | 180 | resource_size_t last_addr; |
87e547fe PP |
181 | const resource_size_t unaligned_phys_addr = phys_addr; |
182 | const unsigned long unaligned_size = size; | |
5da04cc8 | 183 | struct ioremap_desc io_desc; |
91eebf40 | 184 | struct vm_struct *area; |
b14097bd | 185 | enum page_cache_mode new_pcm; |
d806e5ee | 186 | pgprot_t prot; |
dee7cbb2 | 187 | int retval; |
d61fc448 | 188 | void __iomem *ret_addr; |
1da177e4 LT |
189 | |
190 | /* Don't allow wraparound or zero size */ | |
191 | last_addr = phys_addr + size - 1; | |
192 | if (!size || last_addr < phys_addr) | |
193 | return NULL; | |
194 | ||
e3100c82 | 195 | if (!phys_addr_valid(phys_addr)) { |
6997ab49 | 196 | printk(KERN_WARNING "ioremap: invalid physical address %llx\n", |
4c8337ac | 197 | (unsigned long long)phys_addr); |
e3100c82 TG |
198 | WARN_ON_ONCE(1); |
199 | return NULL; | |
200 | } | |
201 | ||
5da04cc8 | 202 | __ioremap_check_mem(phys_addr, size, &io_desc); |
0e4c12b4 | 203 | |
1da177e4 LT |
204 | /* |
205 | * Don't allow anybody to remap normal RAM that we're using.. | |
206 | */ | |
5da04cc8 | 207 | if (io_desc.flags & IORES_MAP_SYSTEM_RAM) { |
8a0a5da6 TG |
208 | WARN_ONCE(1, "ioremap on RAM at %pa - %pa\n", |
209 | &phys_addr, &last_addr); | |
9a58eebe | 210 | return NULL; |
906e36c5 | 211 | } |
9a58eebe | 212 | |
d7677d40 | 213 | /* |
214 | * Mappings have to be page-aligned | |
215 | */ | |
216 | offset = phys_addr & ~PAGE_MASK; | |
ffa71f33 | 217 | phys_addr &= PHYSICAL_PAGE_MASK; |
d7677d40 | 218 | size = PAGE_ALIGN(last_addr+1) - phys_addr; |
219 | ||
ecdd6ee7 | 220 | retval = memtype_reserve(phys_addr, (u64)phys_addr + size, |
e00c8cc9 | 221 | pcm, &new_pcm); |
dee7cbb2 | 222 | if (retval) { |
ecdd6ee7 | 223 | printk(KERN_ERR "ioremap memtype_reserve failed %d\n", retval); |
dee7cbb2 VP |
224 | return NULL; |
225 | } | |
226 | ||
b14097bd JG |
227 | if (pcm != new_pcm) { |
228 | if (!is_new_memtype_allowed(phys_addr, size, pcm, new_pcm)) { | |
279e669b | 229 | printk(KERN_ERR |
b14097bd | 230 | "ioremap error for 0x%llx-0x%llx, requested 0x%x, got 0x%x\n", |
4c8337ac RD |
231 | (unsigned long long)phys_addr, |
232 | (unsigned long long)(phys_addr + size), | |
b14097bd | 233 | pcm, new_pcm); |
de2a47cf | 234 | goto err_free_memtype; |
d7677d40 | 235 | } |
b14097bd | 236 | pcm = new_pcm; |
d7677d40 | 237 | } |
238 | ||
0e4c12b4 TL |
239 | /* |
240 | * If the page being mapped is in memory and SEV is active then | |
241 | * make sure the memory encryption attribute is enabled in the | |
242 | * resulting mapping. | |
243 | */ | |
b14097bd | 244 | prot = PAGE_KERNEL_IO; |
5da04cc8 | 245 | if ((io_desc.flags & IORES_MAP_ENCRYPTED) || encrypted) |
0e4c12b4 TL |
246 | prot = pgprot_encrypted(prot); |
247 | ||
b14097bd JG |
248 | switch (pcm) { |
249 | case _PAGE_CACHE_MODE_UC: | |
d806e5ee | 250 | default: |
b14097bd JG |
251 | prot = __pgprot(pgprot_val(prot) | |
252 | cachemode2protval(_PAGE_CACHE_MODE_UC)); | |
d806e5ee | 253 | break; |
b14097bd JG |
254 | case _PAGE_CACHE_MODE_UC_MINUS: |
255 | prot = __pgprot(pgprot_val(prot) | | |
256 | cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS)); | |
de33c442 | 257 | break; |
b14097bd JG |
258 | case _PAGE_CACHE_MODE_WC: |
259 | prot = __pgprot(pgprot_val(prot) | | |
260 | cachemode2protval(_PAGE_CACHE_MODE_WC)); | |
b310f381 | 261 | break; |
d838270e TK |
262 | case _PAGE_CACHE_MODE_WT: |
263 | prot = __pgprot(pgprot_val(prot) | | |
264 | cachemode2protval(_PAGE_CACHE_MODE_WT)); | |
265 | break; | |
b14097bd | 266 | case _PAGE_CACHE_MODE_WB: |
d806e5ee TG |
267 | break; |
268 | } | |
a148ecfd | 269 | |
1da177e4 LT |
270 | /* |
271 | * Ok, go for it.. | |
272 | */ | |
23016969 | 273 | area = get_vm_area_caller(size, VM_IOREMAP, caller); |
1da177e4 | 274 | if (!area) |
de2a47cf | 275 | goto err_free_memtype; |
1da177e4 | 276 | area->phys_addr = phys_addr; |
e66aadbe | 277 | vaddr = (unsigned long) area->addr; |
43a432b1 | 278 | |
ecdd6ee7 | 279 | if (memtype_kernel_map_sync(phys_addr, size, pcm)) |
de2a47cf | 280 | goto err_free_area; |
e9332cac | 281 | |
de2a47cf XF |
282 | if (ioremap_page_range(vaddr, vaddr + size, phys_addr, prot)) |
283 | goto err_free_area; | |
e9332cac | 284 | |
d61fc448 | 285 | ret_addr = (void __iomem *) (vaddr + offset); |
87e547fe | 286 | mmiotrace_ioremap(unaligned_phys_addr, unaligned_size, ret_addr); |
d61fc448 | 287 | |
c7a7b814 TG |
288 | /* |
289 | * Check if the request spans more than any BAR in the iomem resource | |
290 | * tree. | |
291 | */ | |
9abb0ecd LA |
292 | if (iomem_map_sanity_check(unaligned_phys_addr, unaligned_size)) |
293 | pr_warn("caller %pS mapping multiple BARs\n", caller); | |
c7a7b814 | 294 | |
d61fc448 | 295 | return ret_addr; |
de2a47cf XF |
296 | err_free_area: |
297 | free_vm_area(area); | |
298 | err_free_memtype: | |
ecdd6ee7 | 299 | memtype_free(phys_addr, phys_addr + size); |
de2a47cf | 300 | return NULL; |
1da177e4 | 301 | } |
1da177e4 LT |
302 | |
303 | /** | |
c0d94aa5 | 304 | * ioremap - map bus memory into CPU space |
9efc31b8 | 305 | * @phys_addr: bus address of the memory |
1da177e4 LT |
306 | * @size: size of the resource to map |
307 | * | |
c0d94aa5 | 308 | * ioremap performs a platform specific sequence of operations to |
1da177e4 LT |
309 | * make bus memory CPU accessible via the readb/readw/readl/writeb/ |
310 | * writew/writel functions and the other mmio helpers. The returned | |
311 | * address is not guaranteed to be usable directly as a virtual | |
91eebf40 | 312 | * address. |
1da177e4 LT |
313 | * |
314 | * This version of ioremap ensures that the memory is marked uncachable | |
315 | * on the CPU as well as honouring existing caching rules from things like | |
91eebf40 | 316 | * the PCI bus. Note that there are other caches and buffers on many |
1da177e4 LT |
317 | * busses. In particular driver authors should read up on PCI writes |
318 | * | |
319 | * It's useful if some control registers are in such an area and | |
320 | * write combining or read caching is not desirable: | |
91eebf40 | 321 | * |
1da177e4 LT |
322 | * Must be freed with iounmap. |
323 | */ | |
c0d94aa5 | 324 | void __iomem *ioremap(resource_size_t phys_addr, unsigned long size) |
1da177e4 | 325 | { |
de33c442 SS |
326 | /* |
327 | * Ideally, this should be: | |
cb32edf6 | 328 | * pat_enabled() ? _PAGE_CACHE_MODE_UC : _PAGE_CACHE_MODE_UC_MINUS; |
de33c442 SS |
329 | * |
330 | * Till we fix all X drivers to use ioremap_wc(), we will use | |
e4b6be33 LR |
331 | * UC MINUS. Drivers that are certain they need or can already |
332 | * be converted over to strong UC can use ioremap_uc(). | |
de33c442 | 333 | */ |
b14097bd | 334 | enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC_MINUS; |
de33c442 | 335 | |
b14097bd | 336 | return __ioremap_caller(phys_addr, size, pcm, |
c3a7a61c | 337 | __builtin_return_address(0), false); |
1da177e4 | 338 | } |
c0d94aa5 | 339 | EXPORT_SYMBOL(ioremap); |
1da177e4 | 340 | |
e4b6be33 LR |
341 | /** |
342 | * ioremap_uc - map bus memory into CPU space as strongly uncachable | |
343 | * @phys_addr: bus address of the memory | |
344 | * @size: size of the resource to map | |
345 | * | |
346 | * ioremap_uc performs a platform specific sequence of operations to | |
347 | * make bus memory CPU accessible via the readb/readw/readl/writeb/ | |
348 | * writew/writel functions and the other mmio helpers. The returned | |
349 | * address is not guaranteed to be usable directly as a virtual | |
350 | * address. | |
351 | * | |
352 | * This version of ioremap ensures that the memory is marked with a strong | |
353 | * preference as completely uncachable on the CPU when possible. For non-PAT | |
354 | * systems this ends up setting page-attribute flags PCD=1, PWT=1. For PAT | |
355 | * systems this will set the PAT entry for the pages as strong UC. This call | |
356 | * will honor existing caching rules from things like the PCI bus. Note that | |
357 | * there are other caches and buffers on many busses. In particular driver | |
358 | * authors should read up on PCI writes. | |
359 | * | |
360 | * It's useful if some control registers are in such an area and | |
361 | * write combining or read caching is not desirable: | |
362 | * | |
363 | * Must be freed with iounmap. | |
364 | */ | |
365 | void __iomem *ioremap_uc(resource_size_t phys_addr, unsigned long size) | |
366 | { | |
367 | enum page_cache_mode pcm = _PAGE_CACHE_MODE_UC; | |
368 | ||
369 | return __ioremap_caller(phys_addr, size, pcm, | |
c3a7a61c | 370 | __builtin_return_address(0), false); |
e4b6be33 LR |
371 | } |
372 | EXPORT_SYMBOL_GPL(ioremap_uc); | |
373 | ||
b310f381 | 374 | /** |
375 | * ioremap_wc - map memory into CPU space write combined | |
9efc31b8 | 376 | * @phys_addr: bus address of the memory |
b310f381 | 377 | * @size: size of the resource to map |
378 | * | |
379 | * This version of ioremap ensures that the memory is marked write combining. | |
380 | * Write combining allows faster writes to some hardware devices. | |
381 | * | |
382 | * Must be freed with iounmap. | |
383 | */ | |
d639bab8 | 384 | void __iomem *ioremap_wc(resource_size_t phys_addr, unsigned long size) |
b310f381 | 385 | { |
7202fdb1 | 386 | return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WC, |
c3a7a61c | 387 | __builtin_return_address(0), false); |
b310f381 | 388 | } |
389 | EXPORT_SYMBOL(ioremap_wc); | |
390 | ||
d838270e TK |
391 | /** |
392 | * ioremap_wt - map memory into CPU space write through | |
393 | * @phys_addr: bus address of the memory | |
394 | * @size: size of the resource to map | |
395 | * | |
396 | * This version of ioremap ensures that the memory is marked write through. | |
397 | * Write through stores data into memory while keeping the cache up-to-date. | |
398 | * | |
399 | * Must be freed with iounmap. | |
400 | */ | |
401 | void __iomem *ioremap_wt(resource_size_t phys_addr, unsigned long size) | |
402 | { | |
403 | return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WT, | |
c3a7a61c | 404 | __builtin_return_address(0), false); |
d838270e TK |
405 | } |
406 | EXPORT_SYMBOL(ioremap_wt); | |
407 | ||
c3a7a61c LJ |
408 | void __iomem *ioremap_encrypted(resource_size_t phys_addr, unsigned long size) |
409 | { | |
410 | return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB, | |
411 | __builtin_return_address(0), true); | |
412 | } | |
413 | EXPORT_SYMBOL(ioremap_encrypted); | |
414 | ||
b9e76a00 | 415 | void __iomem *ioremap_cache(resource_size_t phys_addr, unsigned long size) |
5f868152 | 416 | { |
b14097bd | 417 | return __ioremap_caller(phys_addr, size, _PAGE_CACHE_MODE_WB, |
c3a7a61c | 418 | __builtin_return_address(0), false); |
5f868152 TG |
419 | } |
420 | EXPORT_SYMBOL(ioremap_cache); | |
421 | ||
28b2ee20 RR |
422 | void __iomem *ioremap_prot(resource_size_t phys_addr, unsigned long size, |
423 | unsigned long prot_val) | |
424 | { | |
b14097bd JG |
425 | return __ioremap_caller(phys_addr, size, |
426 | pgprot2cachemode(__pgprot(prot_val)), | |
c3a7a61c | 427 | __builtin_return_address(0), false); |
28b2ee20 RR |
428 | } |
429 | EXPORT_SYMBOL(ioremap_prot); | |
430 | ||
bf5421c3 AK |
431 | /** |
432 | * iounmap - Free a IO remapping | |
433 | * @addr: virtual address from ioremap_* | |
434 | * | |
435 | * Caller must ensure there is only one unmapping for the same pointer. | |
436 | */ | |
1da177e4 LT |
437 | void iounmap(volatile void __iomem *addr) |
438 | { | |
bf5421c3 | 439 | struct vm_struct *p, *o; |
c23a4e96 AM |
440 | |
441 | if ((void __force *)addr <= high_memory) | |
1da177e4 LT |
442 | return; |
443 | ||
444 | /* | |
33c2b803 TL |
445 | * The PCI/ISA range special-casing was removed from __ioremap() |
446 | * so this check, in theory, can be removed. However, there are | |
447 | * cases where iounmap() is called for addresses not obtained via | |
448 | * ioremap() (vga16fb for example). Add a warning so that these | |
449 | * cases can be caught and fixed. | |
1da177e4 | 450 | */ |
6e92a5a6 | 451 | if ((void __force *)addr >= phys_to_virt(ISA_START_ADDRESS) && |
33c2b803 TL |
452 | (void __force *)addr < phys_to_virt(ISA_END_ADDRESS)) { |
453 | WARN(1, "iounmap() called for ISA range not obtained using ioremap()\n"); | |
1da177e4 | 454 | return; |
33c2b803 | 455 | } |
1da177e4 | 456 | |
6d60ce38 KH |
457 | mmiotrace_iounmap(addr); |
458 | ||
91eebf40 TG |
459 | addr = (volatile void __iomem *) |
460 | (PAGE_MASK & (unsigned long __force)addr); | |
bf5421c3 AK |
461 | |
462 | /* Use the vm area unlocked, assuming the caller | |
463 | ensures there isn't another iounmap for the same address | |
464 | in parallel. Reuse of the virtual address is prevented by | |
465 | leaving it in the global lists until we're done with it. | |
466 | cpa takes care of the direct mappings. */ | |
ef932473 | 467 | p = find_vm_area((void __force *)addr); |
bf5421c3 AK |
468 | |
469 | if (!p) { | |
91eebf40 | 470 | printk(KERN_ERR "iounmap: bad address %p\n", addr); |
c23a4e96 | 471 | dump_stack(); |
bf5421c3 | 472 | return; |
1da177e4 LT |
473 | } |
474 | ||
ecdd6ee7 | 475 | memtype_free(p->phys_addr, p->phys_addr + get_vm_area_size(p)); |
d7677d40 | 476 | |
bf5421c3 | 477 | /* Finally remove it */ |
6e92a5a6 | 478 | o = remove_vm_area((void __force *)addr); |
bf5421c3 | 479 | BUG_ON(p != o || o == NULL); |
91eebf40 | 480 | kfree(p); |
1da177e4 | 481 | } |
129f6946 | 482 | EXPORT_SYMBOL(iounmap); |
1da177e4 | 483 | |
0f472d04 AK |
484 | int __init arch_ioremap_p4d_supported(void) |
485 | { | |
486 | return 0; | |
487 | } | |
488 | ||
1e6277de | 489 | int __init arch_ioremap_pud_supported(void) |
5d72b4fb TK |
490 | { |
491 | #ifdef CONFIG_X86_64 | |
b8291adc | 492 | return boot_cpu_has(X86_FEATURE_GBPAGES); |
5d72b4fb TK |
493 | #else |
494 | return 0; | |
495 | #endif | |
496 | } | |
497 | ||
1e6277de | 498 | int __init arch_ioremap_pmd_supported(void) |
5d72b4fb | 499 | { |
16bf9226 | 500 | return boot_cpu_has(X86_FEATURE_PSE); |
5d72b4fb TK |
501 | } |
502 | ||
e045fb2a | 503 | /* |
504 | * Convert a physical pointer to a virtual kernel pointer for /dev/mem | |
505 | * access | |
506 | */ | |
4707a341 | 507 | void *xlate_dev_mem_ptr(phys_addr_t phys) |
e045fb2a | 508 | { |
94d4b476 IM |
509 | unsigned long start = phys & PAGE_MASK; |
510 | unsigned long offset = phys & ~PAGE_MASK; | |
562bfca4 | 511 | void *vaddr; |
e045fb2a | 512 | |
8458bf94 TL |
513 | /* memremap() maps if RAM, otherwise falls back to ioremap() */ |
514 | vaddr = memremap(start, PAGE_SIZE, MEMREMAP_WB); | |
e045fb2a | 515 | |
8458bf94 | 516 | /* Only add the offset on success and return NULL if memremap() failed */ |
94d4b476 IM |
517 | if (vaddr) |
518 | vaddr += offset; | |
e045fb2a | 519 | |
562bfca4 | 520 | return vaddr; |
e045fb2a | 521 | } |
522 | ||
4707a341 | 523 | void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr) |
e045fb2a | 524 | { |
8458bf94 | 525 | memunmap((void *)((unsigned long)addr & PAGE_MASK)); |
e045fb2a | 526 | } |
527 | ||
8f716c9b TL |
528 | /* |
529 | * Examine the physical address to determine if it is an area of memory | |
530 | * that should be mapped decrypted. If the memory is not part of the | |
531 | * kernel usable area it was accessed and created decrypted, so these | |
1de32862 TL |
532 | * areas should be mapped decrypted. And since the encryption key can |
533 | * change across reboots, persistent memory should also be mapped | |
534 | * decrypted. | |
072f58c6 TL |
535 | * |
536 | * If SEV is active, that implies that BIOS/UEFI also ran encrypted so | |
537 | * only persistent memory should be mapped decrypted. | |
8f716c9b TL |
538 | */ |
539 | static bool memremap_should_map_decrypted(resource_size_t phys_addr, | |
540 | unsigned long size) | |
541 | { | |
1de32862 TL |
542 | int is_pmem; |
543 | ||
544 | /* | |
545 | * Check if the address is part of a persistent memory region. | |
546 | * This check covers areas added by E820, EFI and ACPI. | |
547 | */ | |
548 | is_pmem = region_intersects(phys_addr, size, IORESOURCE_MEM, | |
549 | IORES_DESC_PERSISTENT_MEMORY); | |
550 | if (is_pmem != REGION_DISJOINT) | |
551 | return true; | |
552 | ||
553 | /* | |
554 | * Check if the non-volatile attribute is set for an EFI | |
555 | * reserved area. | |
556 | */ | |
557 | if (efi_enabled(EFI_BOOT)) { | |
558 | switch (efi_mem_type(phys_addr)) { | |
559 | case EFI_RESERVED_TYPE: | |
560 | if (efi_mem_attributes(phys_addr) & EFI_MEMORY_NV) | |
561 | return true; | |
562 | break; | |
563 | default: | |
564 | break; | |
565 | } | |
566 | } | |
567 | ||
8f716c9b TL |
568 | /* Check if the address is outside kernel usable area */ |
569 | switch (e820__get_entry_type(phys_addr, phys_addr + size - 1)) { | |
570 | case E820_TYPE_RESERVED: | |
571 | case E820_TYPE_ACPI: | |
572 | case E820_TYPE_NVS: | |
573 | case E820_TYPE_UNUSABLE: | |
072f58c6 TL |
574 | /* For SEV, these areas are encrypted */ |
575 | if (sev_active()) | |
576 | break; | |
577 | /* Fallthrough */ | |
578 | ||
1de32862 | 579 | case E820_TYPE_PRAM: |
8f716c9b TL |
580 | return true; |
581 | default: | |
582 | break; | |
583 | } | |
584 | ||
585 | return false; | |
586 | } | |
587 | ||
588 | /* | |
589 | * Examine the physical address to determine if it is EFI data. Check | |
590 | * it against the boot params structure and EFI tables and memory types. | |
591 | */ | |
592 | static bool memremap_is_efi_data(resource_size_t phys_addr, | |
593 | unsigned long size) | |
594 | { | |
595 | u64 paddr; | |
596 | ||
597 | /* Check if the address is part of EFI boot/runtime data */ | |
598 | if (!efi_enabled(EFI_BOOT)) | |
599 | return false; | |
600 | ||
601 | paddr = boot_params.efi_info.efi_memmap_hi; | |
602 | paddr <<= 32; | |
603 | paddr |= boot_params.efi_info.efi_memmap; | |
604 | if (phys_addr == paddr) | |
605 | return true; | |
606 | ||
607 | paddr = boot_params.efi_info.efi_systab_hi; | |
608 | paddr <<= 32; | |
609 | paddr |= boot_params.efi_info.efi_systab; | |
610 | if (phys_addr == paddr) | |
611 | return true; | |
612 | ||
613 | if (efi_is_table_address(phys_addr)) | |
614 | return true; | |
615 | ||
616 | switch (efi_mem_type(phys_addr)) { | |
617 | case EFI_BOOT_SERVICES_DATA: | |
618 | case EFI_RUNTIME_SERVICES_DATA: | |
619 | return true; | |
620 | default: | |
621 | break; | |
622 | } | |
623 | ||
624 | return false; | |
625 | } | |
626 | ||
627 | /* | |
628 | * Examine the physical address to determine if it is boot data by checking | |
629 | * it against the boot params setup_data chain. | |
630 | */ | |
631 | static bool memremap_is_setup_data(resource_size_t phys_addr, | |
632 | unsigned long size) | |
633 | { | |
634 | struct setup_data *data; | |
635 | u64 paddr, paddr_next; | |
636 | ||
637 | paddr = boot_params.hdr.setup_data; | |
638 | while (paddr) { | |
639 | unsigned int len; | |
640 | ||
641 | if (phys_addr == paddr) | |
642 | return true; | |
643 | ||
644 | data = memremap(paddr, sizeof(*data), | |
645 | MEMREMAP_WB | MEMREMAP_DEC); | |
646 | ||
647 | paddr_next = data->next; | |
648 | len = data->len; | |
649 | ||
b3c72fc9 DK |
650 | if ((phys_addr > paddr) && (phys_addr < (paddr + len))) { |
651 | memunmap(data); | |
652 | return true; | |
653 | } | |
654 | ||
655 | if (data->type == SETUP_INDIRECT && | |
656 | ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) { | |
657 | paddr = ((struct setup_indirect *)data->data)->addr; | |
658 | len = ((struct setup_indirect *)data->data)->len; | |
659 | } | |
660 | ||
8f716c9b TL |
661 | memunmap(data); |
662 | ||
663 | if ((phys_addr > paddr) && (phys_addr < (paddr + len))) | |
664 | return true; | |
665 | ||
666 | paddr = paddr_next; | |
667 | } | |
668 | ||
669 | return false; | |
670 | } | |
671 | ||
672 | /* | |
673 | * Examine the physical address to determine if it is boot data by checking | |
674 | * it against the boot params setup_data chain (early boot version). | |
675 | */ | |
676 | static bool __init early_memremap_is_setup_data(resource_size_t phys_addr, | |
677 | unsigned long size) | |
678 | { | |
679 | struct setup_data *data; | |
680 | u64 paddr, paddr_next; | |
681 | ||
682 | paddr = boot_params.hdr.setup_data; | |
683 | while (paddr) { | |
684 | unsigned int len; | |
685 | ||
686 | if (phys_addr == paddr) | |
687 | return true; | |
688 | ||
689 | data = early_memremap_decrypted(paddr, sizeof(*data)); | |
690 | ||
691 | paddr_next = data->next; | |
692 | len = data->len; | |
693 | ||
694 | early_memunmap(data, sizeof(*data)); | |
695 | ||
696 | if ((phys_addr > paddr) && (phys_addr < (paddr + len))) | |
697 | return true; | |
698 | ||
699 | paddr = paddr_next; | |
700 | } | |
701 | ||
702 | return false; | |
703 | } | |
704 | ||
705 | /* | |
706 | * Architecture function to determine if RAM remap is allowed. By default, a | |
707 | * RAM remap will map the data as encrypted. Determine if a RAM remap should | |
708 | * not be done so that the data will be mapped decrypted. | |
709 | */ | |
710 | bool arch_memremap_can_ram_remap(resource_size_t phys_addr, unsigned long size, | |
711 | unsigned long flags) | |
712 | { | |
072f58c6 | 713 | if (!mem_encrypt_active()) |
8f716c9b TL |
714 | return true; |
715 | ||
716 | if (flags & MEMREMAP_ENC) | |
717 | return true; | |
718 | ||
719 | if (flags & MEMREMAP_DEC) | |
720 | return false; | |
721 | ||
072f58c6 TL |
722 | if (sme_active()) { |
723 | if (memremap_is_setup_data(phys_addr, size) || | |
724 | memremap_is_efi_data(phys_addr, size)) | |
725 | return false; | |
726 | } | |
8f716c9b | 727 | |
072f58c6 | 728 | return !memremap_should_map_decrypted(phys_addr, size); |
8f716c9b TL |
729 | } |
730 | ||
731 | /* | |
732 | * Architecture override of __weak function to adjust the protection attributes | |
733 | * used when remapping memory. By default, early_memremap() will map the data | |
734 | * as encrypted. Determine if an encrypted mapping should not be done and set | |
735 | * the appropriate protection attributes. | |
736 | */ | |
737 | pgprot_t __init early_memremap_pgprot_adjust(resource_size_t phys_addr, | |
738 | unsigned long size, | |
739 | pgprot_t prot) | |
740 | { | |
072f58c6 TL |
741 | bool encrypted_prot; |
742 | ||
743 | if (!mem_encrypt_active()) | |
8f716c9b TL |
744 | return prot; |
745 | ||
072f58c6 TL |
746 | encrypted_prot = true; |
747 | ||
748 | if (sme_active()) { | |
749 | if (early_memremap_is_setup_data(phys_addr, size) || | |
750 | memremap_is_efi_data(phys_addr, size)) | |
751 | encrypted_prot = false; | |
752 | } | |
753 | ||
754 | if (encrypted_prot && memremap_should_map_decrypted(phys_addr, size)) | |
755 | encrypted_prot = false; | |
8f716c9b | 756 | |
072f58c6 TL |
757 | return encrypted_prot ? pgprot_encrypted(prot) |
758 | : pgprot_decrypted(prot); | |
8f716c9b TL |
759 | } |
760 | ||
8458bf94 TL |
761 | bool phys_mem_access_encrypted(unsigned long phys_addr, unsigned long size) |
762 | { | |
763 | return arch_memremap_can_ram_remap(phys_addr, size, 0); | |
764 | } | |
765 | ||
ce9084ba | 766 | #ifdef CONFIG_AMD_MEM_ENCRYPT |
f88a68fa TL |
767 | /* Remap memory with encryption */ |
768 | void __init *early_memremap_encrypted(resource_size_t phys_addr, | |
769 | unsigned long size) | |
770 | { | |
771 | return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC); | |
772 | } | |
773 | ||
774 | /* | |
775 | * Remap memory with encryption and write-protected - cannot be called | |
776 | * before pat_init() is called | |
777 | */ | |
778 | void __init *early_memremap_encrypted_wp(resource_size_t phys_addr, | |
779 | unsigned long size) | |
780 | { | |
1f6f655e | 781 | if (!x86_has_pat_wp()) |
f88a68fa | 782 | return NULL; |
f88a68fa TL |
783 | return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_ENC_WP); |
784 | } | |
785 | ||
786 | /* Remap memory without encryption */ | |
787 | void __init *early_memremap_decrypted(resource_size_t phys_addr, | |
788 | unsigned long size) | |
789 | { | |
790 | return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC); | |
791 | } | |
792 | ||
793 | /* | |
794 | * Remap memory without encryption and write-protected - cannot be called | |
795 | * before pat_init() is called | |
796 | */ | |
797 | void __init *early_memremap_decrypted_wp(resource_size_t phys_addr, | |
798 | unsigned long size) | |
799 | { | |
1f6f655e | 800 | if (!x86_has_pat_wp()) |
f88a68fa | 801 | return NULL; |
f88a68fa TL |
802 | return early_memremap_prot(phys_addr, size, __PAGE_KERNEL_NOENC_WP); |
803 | } | |
ce9084ba | 804 | #endif /* CONFIG_AMD_MEM_ENCRYPT */ |
f88a68fa | 805 | |
45c7b28f | 806 | static pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)] __page_aligned_bss; |
0947b2f3 | 807 | |
551889a6 | 808 | static inline pmd_t * __init early_ioremap_pmd(unsigned long addr) |
0947b2f3 | 809 | { |
37cc8d7f | 810 | /* Don't assume we're using swapper_pg_dir at this point */ |
6c690ee1 | 811 | pgd_t *base = __va(read_cr3_pa()); |
37cc8d7f | 812 | pgd_t *pgd = &base[pgd_index(addr)]; |
e0c4f675 KS |
813 | p4d_t *p4d = p4d_offset(pgd, addr); |
814 | pud_t *pud = pud_offset(p4d, addr); | |
551889a6 IC |
815 | pmd_t *pmd = pmd_offset(pud, addr); |
816 | ||
817 | return pmd; | |
0947b2f3 HY |
818 | } |
819 | ||
551889a6 | 820 | static inline pte_t * __init early_ioremap_pte(unsigned long addr) |
0947b2f3 | 821 | { |
551889a6 | 822 | return &bm_pte[pte_index(addr)]; |
0947b2f3 HY |
823 | } |
824 | ||
fef5ba79 JF |
825 | bool __init is_early_ioremap_ptep(pte_t *ptep) |
826 | { | |
827 | return ptep >= &bm_pte[0] && ptep < &bm_pte[PAGE_SIZE/sizeof(pte_t)]; | |
828 | } | |
829 | ||
beacfaac | 830 | void __init early_ioremap_init(void) |
0947b2f3 | 831 | { |
551889a6 | 832 | pmd_t *pmd; |
0947b2f3 | 833 | |
73159fdc AL |
834 | #ifdef CONFIG_X86_64 |
835 | BUILD_BUG_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1)); | |
836 | #else | |
837 | WARN_ON((fix_to_virt(0) + PAGE_SIZE) & ((1 << PMD_SHIFT) - 1)); | |
838 | #endif | |
839 | ||
5b7c73e0 | 840 | early_ioremap_setup(); |
8827247f | 841 | |
551889a6 | 842 | pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)); |
45c7b28f JF |
843 | memset(bm_pte, 0, sizeof(bm_pte)); |
844 | pmd_populate_kernel(&init_mm, pmd, bm_pte); | |
551889a6 | 845 | |
0e3a9549 | 846 | /* |
551889a6 | 847 | * The boot-ioremap range spans multiple pmds, for which |
0e3a9549 IM |
848 | * we are not prepared: |
849 | */ | |
499a5f1e JB |
850 | #define __FIXADDR_TOP (-PAGE_SIZE) |
851 | BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) | |
852 | != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); | |
853 | #undef __FIXADDR_TOP | |
551889a6 | 854 | if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) { |
0e3a9549 | 855 | WARN_ON(1); |
551889a6 IC |
856 | printk(KERN_WARNING "pmd %p != %p\n", |
857 | pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))); | |
91eebf40 | 858 | printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", |
551889a6 | 859 | fix_to_virt(FIX_BTMAP_BEGIN)); |
91eebf40 | 860 | printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END): %08lx\n", |
551889a6 | 861 | fix_to_virt(FIX_BTMAP_END)); |
91eebf40 TG |
862 | |
863 | printk(KERN_WARNING "FIX_BTMAP_END: %d\n", FIX_BTMAP_END); | |
864 | printk(KERN_WARNING "FIX_BTMAP_BEGIN: %d\n", | |
865 | FIX_BTMAP_BEGIN); | |
0e3a9549 | 866 | } |
0947b2f3 HY |
867 | } |
868 | ||
5b7c73e0 MS |
869 | void __init __early_set_fixmap(enum fixed_addresses idx, |
870 | phys_addr_t phys, pgprot_t flags) | |
0947b2f3 | 871 | { |
551889a6 IC |
872 | unsigned long addr = __fix_to_virt(idx); |
873 | pte_t *pte; | |
0947b2f3 HY |
874 | |
875 | if (idx >= __end_of_fixed_addresses) { | |
876 | BUG(); | |
877 | return; | |
878 | } | |
beacfaac | 879 | pte = early_ioremap_pte(addr); |
4583ed51 | 880 | |
fb43d6cb | 881 | /* Sanitize 'prot' against any unsupported bits: */ |
510bb96f | 882 | pgprot_val(flags) &= __supported_pte_mask; |
fb43d6cb | 883 | |
0947b2f3 | 884 | if (pgprot_val(flags)) |
551889a6 | 885 | set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags)); |
0947b2f3 | 886 | else |
4f9c11dd | 887 | pte_clear(&init_mm, addr, pte); |
58430c5d | 888 | flush_tlb_one_kernel(addr); |
0947b2f3 | 889 | } |