projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[IA64] IA64 Kexec/kdump
[linux-block.git] / arch / ia64 / kernel / efi.c
CommitLineData
1da177e4
LT		 1/*
2 * Extensible Firmware Interface
3 *
4 * Based on Extensible Firmware Interface Specification version 0.9 April 30, 1999
5 *
6 * Copyright (C) 1999 VA Linux Systems
7 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
8 * Copyright (C) 1999-2003 Hewlett-Packard Co.
9 * David Mosberger-Tang <davidm@hpl.hp.com>
10 * Stephane Eranian <eranian@hpl.hp.com>
32e62c63
BH		 11 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P.
12 * Bjorn Helgaas <bjorn.helgaas@hp.com>
1da177e4
LT		 13 *
14 * All EFI Runtime Services are not implemented yet as EFI only
15 * supports physical mode addressing on SoftSDV. This is to be fixed
16 * in a future version. --drummond 1999-07-20
17 *
18 * Implemented EFI runtime services and virtual mode calls. --davidm
19 *
20 * Goutham Rao: <goutham.rao@intel.com>
21 * Skip non-WB memory and ignore empty memory ranges.
22 */
1da177e4
LT		 23#include <linux/module.h>
24#include <linux/kernel.h>
25#include <linux/init.h>
26#include <linux/types.h>
27#include <linux/time.h>
28#include <linux/efi.h>
a7956113 29#include <linux/kexec.h>
1da177e4
LT		 30
31#include <asm/io.h>
32#include <asm/kregs.h>
33#include <asm/meminit.h>
34#include <asm/pgtable.h>
35#include <asm/processor.h>
36#include <asm/mca.h>
37
38#define EFI_DEBUG 0
39
40extern efi_status_t efi_call_phys (void *, ...);
41
42struct efi efi;
43EXPORT_SYMBOL(efi);
44static efi_runtime_services_t *runtime;
a7956113 45static unsigned long mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
1da177e4
LT		 46
47#define efi_call_virt(f, args...) (*(f))(args)
48
49#define STUB_GET_TIME(prefix, adjust_arg) \
50static efi_status_t \
51prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
52{ \
53 struct ia64_fpreg fr[6]; \
54 efi_time_cap_t *atc = NULL; \
55 efi_status_t ret; \
56 \
57 if (tc) \
58 atc = adjust_arg(tc); \
59 ia64_save_scratch_fpregs(fr); \
60 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), adjust_arg(tm), atc); \
61 ia64_load_scratch_fpregs(fr); \
62 return ret; \
63}
64
65#define STUB_SET_TIME(prefix, adjust_arg) \
66static efi_status_t \
67prefix##_set_time (efi_time_t *tm) \
68{ \
69 struct ia64_fpreg fr[6]; \
70 efi_status_t ret; \
71 \
72 ia64_save_scratch_fpregs(fr); \
73 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), adjust_arg(tm)); \
74 ia64_load_scratch_fpregs(fr); \
75 return ret; \
76}
77
78#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
79static efi_status_t \
80prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, efi_time_t *tm) \
81{ \
82 struct ia64_fpreg fr[6]; \
83 efi_status_t ret; \
84 \
85 ia64_save_scratch_fpregs(fr); \
86 ret = efi_call_##prefix((efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
87 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
88 ia64_load_scratch_fpregs(fr); \
89 return ret; \
90}
91
92#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
93static efi_status_t \
94prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
95{ \
96 struct ia64_fpreg fr[6]; \
97 efi_time_t *atm = NULL; \
98 efi_status_t ret; \
99 \
100 if (tm) \
101 atm = adjust_arg(tm); \
102 ia64_save_scratch_fpregs(fr); \
103 ret = efi_call_##prefix((efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
104 enabled, atm); \
105 ia64_load_scratch_fpregs(fr); \
106 return ret; \
107}
108
109#define STUB_GET_VARIABLE(prefix, adjust_arg) \
110static efi_status_t \
111prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
112 unsigned long *data_size, void *data) \
113{ \
114 struct ia64_fpreg fr[6]; \
115 u32 *aattr = NULL; \
116 efi_status_t ret; \
117 \
118 if (attr) \
119 aattr = adjust_arg(attr); \
120 ia64_save_scratch_fpregs(fr); \
121 ret = efi_call_##prefix((efi_get_variable_t *) __va(runtime->get_variable), \
122 adjust_arg(name), adjust_arg(vendor), aattr, \
123 adjust_arg(data_size), adjust_arg(data)); \
124 ia64_load_scratch_fpregs(fr); \
125 return ret; \
126}
127
128#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
129static efi_status_t \
130prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, efi_guid_t *vendor) \
131{ \
132 struct ia64_fpreg fr[6]; \
133 efi_status_t ret; \
134 \
135 ia64_save_scratch_fpregs(fr); \
136 ret = efi_call_##prefix((efi_get_next_variable_t *) __va(runtime->get_next_variable), \
137 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
138 ia64_load_scratch_fpregs(fr); \
139 return ret; \
140}
141
142#define STUB_SET_VARIABLE(prefix, adjust_arg) \
143static efi_status_t \
144prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, unsigned long attr, \
145 unsigned long data_size, void *data) \
146{ \
147 struct ia64_fpreg fr[6]; \
148 efi_status_t ret; \
149 \
150 ia64_save_scratch_fpregs(fr); \
151 ret = efi_call_##prefix((efi_set_variable_t *) __va(runtime->set_variable), \
152 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
153 adjust_arg(data)); \
154 ia64_load_scratch_fpregs(fr); \
155 return ret; \
156}
157
158#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
159static efi_status_t \
160prefix##_get_next_high_mono_count (u32 *count) \
161{ \
162 struct ia64_fpreg fr[6]; \
163 efi_status_t ret; \
164 \
165 ia64_save_scratch_fpregs(fr); \
166 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
167 __va(runtime->get_next_high_mono_count), adjust_arg(count)); \
168 ia64_load_scratch_fpregs(fr); \
169 return ret; \
170}
171
172#define STUB_RESET_SYSTEM(prefix, adjust_arg) \
173static void \
174prefix##_reset_system (int reset_type, efi_status_t status, \
175 unsigned long data_size, efi_char16_t *data) \
176{ \
177 struct ia64_fpreg fr[6]; \
178 efi_char16_t *adata = NULL; \
179 \
180 if (data) \
181 adata = adjust_arg(data); \
182 \
183 ia64_save_scratch_fpregs(fr); \
184 efi_call_##prefix((efi_reset_system_t *) __va(runtime->reset_system), \
185 reset_type, status, data_size, adata); \
186 /* should not return, but just in case... */ \
187 ia64_load_scratch_fpregs(fr); \
188}
189
190#define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
191
192STUB_GET_TIME(phys, phys_ptr)
193STUB_SET_TIME(phys, phys_ptr)
194STUB_GET_WAKEUP_TIME(phys, phys_ptr)
195STUB_SET_WAKEUP_TIME(phys, phys_ptr)
196STUB_GET_VARIABLE(phys, phys_ptr)
197STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
198STUB_SET_VARIABLE(phys, phys_ptr)
199STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
200STUB_RESET_SYSTEM(phys, phys_ptr)
201
202#define id(arg) arg
203
204STUB_GET_TIME(virt, id)
205STUB_SET_TIME(virt, id)
206STUB_GET_WAKEUP_TIME(virt, id)
207STUB_SET_WAKEUP_TIME(virt, id)
208STUB_GET_VARIABLE(virt, id)
209STUB_GET_NEXT_VARIABLE(virt, id)
210STUB_SET_VARIABLE(virt, id)
211STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
212STUB_RESET_SYSTEM(virt, id)
213
214void
215efi_gettimeofday (struct timespec *ts)
216{
217 efi_time_t tm;
218
219 memset(ts, 0, sizeof(ts));
220 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS)
221 return;
222
223 ts->tv_sec = mktime(tm.year, tm.month, tm.day, tm.hour, tm.minute, tm.second);
224 ts->tv_nsec = tm.nanosecond;
225}
226
227static int
228is_available_memory (efi_memory_desc_t *md)
229{
230 if (!(md->attribute & EFI_MEMORY_WB))
231 return 0;
232
233 switch (md->type) {
234 case EFI_LOADER_CODE:
235 case EFI_LOADER_DATA:
236 case EFI_BOOT_SERVICES_CODE:
237 case EFI_BOOT_SERVICES_DATA:
238 case EFI_CONVENTIONAL_MEMORY:
239 return 1;
240 }
241 return 0;
242}
243
d8c97d5f
TL		 244typedef struct kern_memdesc {
245 u64 attribute;
246 u64 start;
247 u64 num_pages;
248} kern_memdesc_t;
1da177e4 249
d8c97d5f 250static kern_memdesc_t *kern_memmap;
1da177e4 251
80851ef2
BH		 252#define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
253
254static inline u64
255kmd_end(kern_memdesc_t *kmd)
256{
257 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
258}
259
260static inline u64
261efi_md_end(efi_memory_desc_t *md)
262{
263 return (md->phys_addr + efi_md_size(md));
264}
265
266static inline int
267efi_wb(efi_memory_desc_t *md)
268{
269 return (md->attribute & EFI_MEMORY_WB);
270}
271
272static inline int
273efi_uc(efi_memory_desc_t *md)
274{
275 return (md->attribute & EFI_MEMORY_UC);
276}
277
1da177e4 278static void
d8c97d5f 279walk (efi_freemem_callback_t callback, void *arg, u64 attr)
1da177e4 280{
d8c97d5f
TL		 281 kern_memdesc_t *k;
282 u64 start, end, voff;
1da177e4 283
d8c97d5f
TL		 284 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
285 for (k = kern_memmap; k->start != ~0UL; k++) {
286 if (k->attribute != attr)
287 continue;
288 start = PAGE_ALIGN(k->start);
289 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
290 if (start < end)
291 if ((*callback)(start + voff, end + voff, arg) < 0)
292 return;
293 }
1da177e4
LT		 294}
295
296/*
297 * Walks the EFI memory map and calls CALLBACK once for each EFI memory descriptor that
298 * has memory that is available for OS use.
299 */
300void
301efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
302{
d8c97d5f 303 walk(callback, arg, EFI_MEMORY_WB);
1da177e4
LT		 304}
305
f14f75b8 306/*
d8c97d5f
TL		 307 * Walks the EFI memory map and calls CALLBACK once for each EFI memory descriptor that
308 * has memory that is available for uncached allocator.
f14f75b8 309 */
d8c97d5f
TL		 310void
311efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
f14f75b8 312{
d8c97d5f 313 walk(callback, arg, EFI_MEMORY_UC);
f14f75b8
JS		 314}
315
1da177e4
LT		 316/*
317 * Look for the PAL_CODE region reported by EFI and maps it using an
318 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
319 * Abstraction Layer chapter 11 in ADAG
320 */
321
322void *
323efi_get_pal_addr (void)
324{
325 void *efi_map_start, *efi_map_end, *p;
326 efi_memory_desc_t *md;
327 u64 efi_desc_size;
328 int pal_code_count = 0;
329 u64 vaddr, mask;
330
331 efi_map_start = __va(ia64_boot_param->efi_memmap);
332 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
333 efi_desc_size = ia64_boot_param->efi_memdesc_size;
334
335 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
336 md = p;
337 if (md->type != EFI_PAL_CODE)
338 continue;
339
340 if (++pal_code_count > 1) {
341 printk(KERN_ERR "Too many EFI Pal Code memory ranges, dropped @ %lx\n",
342 md->phys_addr);
343 continue;
344 }
345 /*
346 * The only ITLB entry in region 7 that is used is the one installed by
347 * __start(). That entry covers a 64MB range.
348 */
349 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
350 vaddr = PAGE_OFFSET + md->phys_addr;
351
352 /*
353 * We must check that the PAL mapping won't overlap with the kernel
354 * mapping.
355 *
356 * PAL code is guaranteed to be aligned on a power of 2 between 4k and
357 * 256KB and that only one ITR is needed to map it. This implies that the
358 * PAL code is always aligned on its size, i.e., the closest matching page
359 * size supported by the TLB. Therefore PAL code is guaranteed never to
360 * cross a 64MB unless it is bigger than 64MB (very unlikely!). So for
361 * now the following test is enough to determine whether or not we need a
362 * dedicated ITR for the PAL code.
363 */
364 if ((vaddr & mask) == (KERNEL_START & mask)) {
365 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
366 __FUNCTION__);
367 continue;
368 }
369
370 if (md->num_pages << EFI_PAGE_SHIFT > IA64_GRANULE_SIZE)
371 panic("Woah! PAL code size bigger than a granule!");
372
373#if EFI_DEBUG
374 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
375
376 printk(KERN_INFO "CPU %d: mapping PAL code [0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
377 smp_processor_id(), md->phys_addr,
378 md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
379 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
380#endif
381 return __va(md->phys_addr);
382 }
383 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found",
384 __FUNCTION__);
385 return NULL;
386}
387
388void
389efi_map_pal_code (void)
390{
391 void *pal_vaddr = efi_get_pal_addr ();
392 u64 psr;
393
394 if (!pal_vaddr)
395 return;
396
397 /*
398 * Cannot write to CRx with PSR.ic=1
399 */
400 psr = ia64_clear_ic();
401 ia64_itr(0x1, IA64_TR_PALCODE, GRANULEROUNDDOWN((unsigned long) pal_vaddr),
402 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
403 IA64_GRANULE_SHIFT);
404 ia64_set_psr(psr); /* restore psr */
405 ia64_srlz_i();
406}
407
408void __init
409efi_init (void)
410{
411 void *efi_map_start, *efi_map_end;
412 efi_config_table_t *config_tables;
413 efi_char16_t *c16;
414 u64 efi_desc_size;
9d78f43d 415 char *cp, vendor[100] = "unknown";
1da177e4
LT		 416 extern char saved_command_line[];
417 int i;
418
419 /* it's too early to be able to use the standard kernel command line support... */
420 for (cp = saved_command_line; *cp; ) {
421 if (memcmp(cp, "mem=", 4) == 0) {
9d78f43d 422 mem_limit = memparse(cp + 4, &cp);
1da177e4 423 } else if (memcmp(cp, "max_addr=", 9) == 0) {
9d78f43d 424 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
a7956113
ZN		 425 } else if (memcmp(cp, "min_addr=", 9) == 0) {
426 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
1da177e4
LT		 427 } else {
428 while (*cp != ' ' && *cp)
429 ++cp;
430 while (*cp == ' ')
431 ++cp;
432 }
433 }
a7956113
ZN		 434 if (min_addr != 0UL)
435 printk(KERN_INFO "Ignoring memory below %luMB\n", min_addr >> 20);
1da177e4
LT		 436 if (max_addr != ~0UL)
437 printk(KERN_INFO "Ignoring memory above %luMB\n", max_addr >> 20);
438
439 efi.systab = __va(ia64_boot_param->efi_systab);
440
441 /*
442 * Verify the EFI Table
443 */
444 if (efi.systab == NULL)
445 panic("Woah! Can't find EFI system table.\n");
446 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
447 panic("Woah! EFI system table signature incorrect\n");
448 if ((efi.systab->hdr.revision ^ EFI_SYSTEM_TABLE_REVISION) >> 16 != 0)
449 printk(KERN_WARNING "Warning: EFI system table major version mismatch: "
450 "got %d.%02d, expected %d.%02d\n",
451 efi.systab->hdr.revision >> 16, efi.systab->hdr.revision & 0xffff,
452 EFI_SYSTEM_TABLE_REVISION >> 16, EFI_SYSTEM_TABLE_REVISION & 0xffff);
453
454 config_tables = __va(efi.systab->tables);
455
456 /* Show what we know for posterity */
457 c16 = __va(efi.systab->fw_vendor);
458 if (c16) {
ecdd5dab 459 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
1da177e4
LT		 460 vendor[i] = *c16++;
461 vendor[i] = '\0';
462 }
463
464 printk(KERN_INFO "EFI v%u.%.02u by %s:",
465 efi.systab->hdr.revision >> 16, efi.systab->hdr.revision & 0xffff, vendor);
466
b2c99e3c
BH		 467 efi.mps = EFI_INVALID_TABLE_ADDR;
468 efi.acpi = EFI_INVALID_TABLE_ADDR;
469 efi.acpi20 = EFI_INVALID_TABLE_ADDR;
470 efi.smbios = EFI_INVALID_TABLE_ADDR;
471 efi.sal_systab = EFI_INVALID_TABLE_ADDR;
472 efi.boot_info = EFI_INVALID_TABLE_ADDR;
473 efi.hcdp = EFI_INVALID_TABLE_ADDR;
474 efi.uga = EFI_INVALID_TABLE_ADDR;
475
1da177e4
LT		 476 for (i = 0; i < (int) efi.systab->nr_tables; i++) {
477 if (efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID) == 0) {
b2c99e3c 478 efi.mps = config_tables[i].table;
1da177e4
LT		 479 printk(" MPS=0x%lx", config_tables[i].table);
480 } else if (efi_guidcmp(config_tables[i].guid, ACPI_20_TABLE_GUID) == 0) {
b2c99e3c 481 efi.acpi20 = config_tables[i].table;
1da177e4
LT		 482 printk(" ACPI 2.0=0x%lx", config_tables[i].table);
483 } else if (efi_guidcmp(config_tables[i].guid, ACPI_TABLE_GUID) == 0) {
b2c99e3c 484 efi.acpi = config_tables[i].table;
1da177e4
LT		 485 printk(" ACPI=0x%lx", config_tables[i].table);
486 } else if (efi_guidcmp(config_tables[i].guid, SMBIOS_TABLE_GUID) == 0) {
b2c99e3c 487 efi.smbios = config_tables[i].table;
1da177e4
LT		 488 printk(" SMBIOS=0x%lx", config_tables[i].table);
489 } else if (efi_guidcmp(config_tables[i].guid, SAL_SYSTEM_TABLE_GUID) == 0) {
b2c99e3c 490 efi.sal_systab = config_tables[i].table;
1da177e4
LT		 491 printk(" SALsystab=0x%lx", config_tables[i].table);
492 } else if (efi_guidcmp(config_tables[i].guid, HCDP_TABLE_GUID) == 0) {
b2c99e3c 493 efi.hcdp = config_tables[i].table;
1da177e4
LT		 494 printk(" HCDP=0x%lx", config_tables[i].table);
495 }
496 }
497 printk("\n");
498
499 runtime = __va(efi.systab->runtime);
500 efi.get_time = phys_get_time;
501 efi.set_time = phys_set_time;
502 efi.get_wakeup_time = phys_get_wakeup_time;
503 efi.set_wakeup_time = phys_set_wakeup_time;
504 efi.get_variable = phys_get_variable;
505 efi.get_next_variable = phys_get_next_variable;
506 efi.set_variable = phys_set_variable;
507 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
508 efi.reset_system = phys_reset_system;
509
510 efi_map_start = __va(ia64_boot_param->efi_memmap);
511 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
512 efi_desc_size = ia64_boot_param->efi_memdesc_size;
513
514#if EFI_DEBUG
515 /* print EFI memory map: */
516 {
517 efi_memory_desc_t *md;
518 void *p;
519
520 for (i = 0, p = efi_map_start; p < efi_map_end; ++i, p += efi_desc_size) {
521 md = p;
522 printk("mem%02u: type=%u, attr=0x%lx, range=[0x%016lx-0x%016lx) (%luMB)\n",
523 i, md->type, md->attribute, md->phys_addr,
524 md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
525 md->num_pages >> (20 - EFI_PAGE_SHIFT));
526 }
527 }
528#endif
529
530 efi_map_pal_code();
531 efi_enter_virtual_mode();
532}
533
534void
535efi_enter_virtual_mode (void)
536{
537 void *efi_map_start, *efi_map_end, *p;
538 efi_memory_desc_t *md;
539 efi_status_t status;
540 u64 efi_desc_size;
541
542 efi_map_start = __va(ia64_boot_param->efi_memmap);
543 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
544 efi_desc_size = ia64_boot_param->efi_memdesc_size;
545
546 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
547 md = p;
548 if (md->attribute & EFI_MEMORY_RUNTIME) {
549 /*
550 * Some descriptors have multiple bits set, so the order of
551 * the tests is relevant.
552 */
553 if (md->attribute & EFI_MEMORY_WB) {
554 md->virt_addr = (u64) __va(md->phys_addr);
555 } else if (md->attribute & EFI_MEMORY_UC) {
556 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
557 } else if (md->attribute & EFI_MEMORY_WC) {
558#if 0
559 md->virt_addr = ia64_remap(md->phys_addr, (_PAGE_A | _PAGE_P
560 | _PAGE_D
561 | _PAGE_MA_WC
562 | _PAGE_PL_0
563 | _PAGE_AR_RW));
564#else
565 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
566 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
567#endif
568 } else if (md->attribute & EFI_MEMORY_WT) {
569#if 0
570 md->virt_addr = ia64_remap(md->phys_addr, (_PAGE_A | _PAGE_P
571 | _PAGE_D | _PAGE_MA_WT
572 | _PAGE_PL_0
573 | _PAGE_AR_RW));
574#else
575 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
576 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
577#endif
578 }
579 }
580 }
581
582 status = efi_call_phys(__va(runtime->set_virtual_address_map),
583 ia64_boot_param->efi_memmap_size,
584 efi_desc_size, ia64_boot_param->efi_memdesc_version,
585 ia64_boot_param->efi_memmap);
586 if (status != EFI_SUCCESS) {
587 printk(KERN_WARNING "warning: unable to switch EFI into virtual mode "
588 "(status=%lu)\n", status);
589 return;
590 }
591
592 /*
593 * Now that EFI is in virtual mode, we call the EFI functions more efficiently:
594 */
595 efi.get_time = virt_get_time;
596 efi.set_time = virt_set_time;
597 efi.get_wakeup_time = virt_get_wakeup_time;
598 efi.set_wakeup_time = virt_set_wakeup_time;
599 efi.get_variable = virt_get_variable;
600 efi.get_next_variable = virt_get_next_variable;
601 efi.set_variable = virt_set_variable;
602 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
603 efi.reset_system = virt_reset_system;
604}
605
606/*
607 * Walk the EFI memory map looking for the I/O port range. There can only be one entry of
608 * this type, other I/O port ranges should be described via ACPI.
609 */
610u64
611efi_get_iobase (void)
612{
613 void *efi_map_start, *efi_map_end, *p;
614 efi_memory_desc_t *md;
615 u64 efi_desc_size;
616
617 efi_map_start = __va(ia64_boot_param->efi_memmap);
618 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
619 efi_desc_size = ia64_boot_param->efi_memdesc_size;
620
621 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
622 md = p;
623 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
624 if (md->attribute & EFI_MEMORY_UC)
625 return md->phys_addr;
626 }
627 }
628 return 0;
629}
630
32e62c63
BH		 631static struct kern_memdesc *
632kern_memory_descriptor (unsigned long phys_addr)
1da177e4 633{
32e62c63 634 struct kern_memdesc *md;
1da177e4 635
32e62c63
BH		 636 for (md = kern_memmap; md->start != ~0UL; md++) {
637 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
80851ef2 638 return md;
1da177e4 639 }
e037cda5 640 return NULL;
1da177e4
LT		 641}
642
32e62c63
BH		 643static efi_memory_desc_t *
644efi_memory_descriptor (unsigned long phys_addr)
1da177e4
LT		 645{
646 void *efi_map_start, *efi_map_end, *p;
647 efi_memory_desc_t *md;
648 u64 efi_desc_size;
649
650 efi_map_start = __va(ia64_boot_param->efi_memmap);
651 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
652 efi_desc_size = ia64_boot_param->efi_memdesc_size;
653
654 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
655 md = p;
656
32e62c63
BH		 657 if (phys_addr - md->phys_addr < (md->num_pages << EFI_PAGE_SHIFT))
658 return md;
1da177e4 659 }
e037cda5 660 return NULL;
1da177e4 661}
80851ef2
BH		 662
663u32
664efi_mem_type (unsigned long phys_addr)
665{
666 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
667
668 if (md)
669 return md->type;
670 return 0;
671}
672
673u64
674efi_mem_attributes (unsigned long phys_addr)
675{
676 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
677
678 if (md)
679 return md->attribute;
680 return 0;
681}
1da177e4
LT		 682EXPORT_SYMBOL(efi_mem_attributes);
683
32e62c63
BH		 684u64
685efi_mem_attribute (unsigned long phys_addr, unsigned long size)
80851ef2 686{
136939a2 687 unsigned long end = phys_addr + size;
80851ef2 688 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
32e62c63
BH		 689 u64 attr;
690
691 if (!md)
692 return 0;
693
694 /*
695 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
696 * the kernel that firmware needs this region mapped.
697 */
698 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
699 do {
700 unsigned long md_end = efi_md_end(md);
701
702 if (end <= md_end)
703 return attr;
704
705 md = efi_memory_descriptor(md_end);
706 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
707 return 0;
708 } while (md);
709 return 0;
710}
711
712u64
713kern_mem_attribute (unsigned long phys_addr, unsigned long size)
714{
715 unsigned long end = phys_addr + size;
716 struct kern_memdesc *md;
717 u64 attr;
80851ef2 718
136939a2 719 /*
32e62c63
BH		 720 * This is a hack for ioremap calls before we set up kern_memmap.
721 * Maybe we should do efi_memmap_init() earlier instead.
136939a2 722 */
32e62c63
BH		 723 if (!kern_memmap) {
724 attr = efi_mem_attribute(phys_addr, size);
725 if (attr & EFI_MEMORY_WB)
726 return EFI_MEMORY_WB;
80851ef2 727 return 0;
136939a2 728 }
80851ef2 729
32e62c63
BH		 730 md = kern_memory_descriptor(phys_addr);
731 if (!md)
732 return 0;
733
734 attr = md->attribute;
80851ef2 735 do {
32e62c63 736 unsigned long md_end = kmd_end(md);
136939a2
BH		 737
738 if (end <= md_end)
32e62c63 739 return attr;
80851ef2 740
32e62c63
BH		 741 md = kern_memory_descriptor(md_end);
742 if (!md || md->attribute != attr)
136939a2 743 return 0;
80851ef2
BH		 744 } while (md);
745 return 0;
746}
32e62c63 747EXPORT_SYMBOL(kern_mem_attribute);
80851ef2 748
1da177e4 749int
136939a2 750valid_phys_addr_range (unsigned long phys_addr, unsigned long size)
1da177e4 751{
32e62c63
BH		 752 u64 attr;
753
754 /*
755 * /dev/mem reads and writes use copy_to_user(), which implicitly
756 * uses a granule-sized kernel identity mapping. It's really
757 * only safe to do this for regions in kern_memmap. For more
758 * details, see Documentation/ia64/aliasing.txt.
759 */
760 attr = kern_mem_attribute(phys_addr, size);
761 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
762 return 1;
763 return 0;
80851ef2 764}
1da177e4 765
80851ef2 766int
06c67bef 767valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
80851ef2 768{
32e62c63
BH		 769 /*
770 * MMIO regions are often missing from the EFI memory map.
771 * We must allow mmap of them for programs like X, so we
772 * currently can't do any useful validation.
773 */
774 return 1;
775}
1da177e4 776
32e62c63
BH		 777pgprot_t
778phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
779 pgprot_t vma_prot)
780{
781 unsigned long phys_addr = pfn << PAGE_SHIFT;
782 u64 attr;
1da177e4 783
32e62c63
BH		 784 /*
785 * For /dev/mem mmap, we use user mappings, but if the region is
786 * in kern_memmap (and hence may be covered by a kernel mapping),
787 * we must use the same attribute as the kernel mapping.
788 */
789 attr = kern_mem_attribute(phys_addr, size);
790 if (attr & EFI_MEMORY_WB)
791 return pgprot_cacheable(vma_prot);
792 else if (attr & EFI_MEMORY_UC)
793 return pgprot_noncached(vma_prot);
794
795 /*
796 * Some chipsets don't support UC access to memory. If
797 * WB is supported, we prefer that.
798 */
799 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
800 return pgprot_cacheable(vma_prot);
801
802 return pgprot_noncached(vma_prot);
1da177e4
LT		 803}
804
805int __init
806efi_uart_console_only(void)
807{
808 efi_status_t status;
809 char *s, name[] = "ConOut";
810 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
811 efi_char16_t *utf16, name_utf16[32];
812 unsigned char data[1024];
813 unsigned long size = sizeof(data);
814 struct efi_generic_dev_path *hdr, *end_addr;
815 int uart = 0;
816
817 /* Convert to UTF-16 */
818 utf16 = name_utf16;
819 s = name;
820 while (*s)
821 *utf16++ = *s++ & 0x7f;
822 *utf16 = 0;
823
824 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
825 if (status != EFI_SUCCESS) {
826 printk(KERN_ERR "No EFI %s variable?\n", name);
827 return 0;
828 }
829
830 hdr = (struct efi_generic_dev_path *) data;
831 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
832 while (hdr < end_addr) {
833 if (hdr->type == EFI_DEV_MSG &&
834 hdr->sub_type == EFI_DEV_MSG_UART)
835 uart = 1;
836 else if (hdr->type == EFI_DEV_END_PATH ||
837 hdr->type == EFI_DEV_END_PATH2) {
838 if (!uart)
839 return 0;
840 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
841 return 1;
842 uart = 0;
843 }
844 hdr = (struct efi_generic_dev_path *) ((u8 *) hdr + hdr->length);
845 }
846 printk(KERN_ERR "Malformed %s value\n", name);
847 return 0;
848}
d8c97d5f 849
d8c97d5f
TL		 850/*
851 * Look for the first granule aligned memory descriptor memory
852 * that is big enough to hold EFI memory map. Make sure this
853 * descriptor is atleast granule sized so it does not get trimmed
854 */
855struct kern_memdesc *
856find_memmap_space (void)
857{
858 u64 contig_low=0, contig_high=0;
859 u64 as = 0, ae;
860 void *efi_map_start, *efi_map_end, *p, *q;
861 efi_memory_desc_t *md, *pmd = NULL, *check_md;
862 u64 space_needed, efi_desc_size;
863 unsigned long total_mem = 0;
864
865 efi_map_start = __va(ia64_boot_param->efi_memmap);
866 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
867 efi_desc_size = ia64_boot_param->efi_memdesc_size;
868
869 /*
870 * Worst case: we need 3 kernel descriptors for each efi descriptor
871 * (if every entry has a WB part in the middle, and UC head and tail),
872 * plus one for the end marker.
873 */
874 space_needed = sizeof(kern_memdesc_t) *
875 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
876
877 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
878 md = p;
879 if (!efi_wb(md)) {
880 continue;
881 }
882 if (pmd == NULL || !efi_wb(pmd) || efi_md_end(pmd) != md->phys_addr) {
883 contig_low = GRANULEROUNDUP(md->phys_addr);
884 contig_high = efi_md_end(md);
885 for (q = p + efi_desc_size; q < efi_map_end; q += efi_desc_size) {
886 check_md = q;
887 if (!efi_wb(check_md))
888 break;
889 if (contig_high != check_md->phys_addr)
890 break;
891 contig_high = efi_md_end(check_md);
892 }
893 contig_high = GRANULEROUNDDOWN(contig_high);
894 }
895 if (!is_available_memory(md) || md->type == EFI_LOADER_DATA)
896 continue;
897
898 /* Round ends inward to granule boundaries */
899 as = max(contig_low, md->phys_addr);
900 ae = min(contig_high, efi_md_end(md));
901
a7956113
ZN		 902 /* keep within max_addr= and min_addr= command line arg */
903 as = max(as, min_addr);
d8c97d5f
TL		 904 ae = min(ae, max_addr);
905 if (ae <= as)
906 continue;
907
908 /* avoid going over mem= command line arg */
909 if (total_mem + (ae - as) > mem_limit)
910 ae -= total_mem + (ae - as) - mem_limit;
911
912 if (ae <= as)
913 continue;
914
915 if (ae - as > space_needed)
916 break;
917 }
918 if (p >= efi_map_end)
919 panic("Can't allocate space for kernel memory descriptors");
920
921 return __va(as);
922}
923
924/*
925 * Walk the EFI memory map and gather all memory available for kernel
926 * to use. We can allocate partial granules only if the unavailable
927 * parts exist, and are WB.
928 */
929void
930efi_memmap_init(unsigned long *s, unsigned long *e)
931{
e037cda5 932 struct kern_memdesc *k, *prev = NULL;
d8c97d5f
TL		 933 u64 contig_low=0, contig_high=0;
934 u64 as, ae, lim;
935 void *efi_map_start, *efi_map_end, *p, *q;
936 efi_memory_desc_t *md, *pmd = NULL, *check_md;
937 u64 efi_desc_size;
938 unsigned long total_mem = 0;
939
940 k = kern_memmap = find_memmap_space();
941
942 efi_map_start = __va(ia64_boot_param->efi_memmap);
943 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
944 efi_desc_size = ia64_boot_param->efi_memdesc_size;
945
946 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
947 md = p;
948 if (!efi_wb(md)) {
949 if (efi_uc(md) && (md->type == EFI_CONVENTIONAL_MEMORY ||
950 md->type == EFI_BOOT_SERVICES_DATA)) {
951 k->attribute = EFI_MEMORY_UC;
952 k->start = md->phys_addr;
953 k->num_pages = md->num_pages;
954 k++;
955 }
956 continue;
957 }
958 if (pmd == NULL || !efi_wb(pmd) || efi_md_end(pmd) != md->phys_addr) {
959 contig_low = GRANULEROUNDUP(md->phys_addr);
960 contig_high = efi_md_end(md);
961 for (q = p + efi_desc_size; q < efi_map_end; q += efi_desc_size) {
962 check_md = q;
963 if (!efi_wb(check_md))
964 break;
965 if (contig_high != check_md->phys_addr)
966 break;
967 contig_high = efi_md_end(check_md);
968 }
969 contig_high = GRANULEROUNDDOWN(contig_high);
970 }
971 if (!is_available_memory(md))
972 continue;
973
974 /*
975 * Round ends inward to granule boundaries
976 * Give trimmings to uncached allocator
977 */
978 if (md->phys_addr < contig_low) {
979 lim = min(efi_md_end(md), contig_low);
980 if (efi_uc(md)) {
981 if (k > kern_memmap && (k-1)->attribute == EFI_MEMORY_UC &&
982 kmd_end(k-1) == md->phys_addr) {
983 (k-1)->num_pages += (lim - md->phys_addr) >> EFI_PAGE_SHIFT;
984 } else {
985 k->attribute = EFI_MEMORY_UC;
986 k->start = md->phys_addr;
987 k->num_pages = (lim - md->phys_addr) >> EFI_PAGE_SHIFT;
988 k++;
989 }
990 }
991 as = contig_low;
992 } else
993 as = md->phys_addr;
994
995 if (efi_md_end(md) > contig_high) {
996 lim = max(md->phys_addr, contig_high);
997 if (efi_uc(md)) {
998 if (lim == md->phys_addr && k > kern_memmap &&
999 (k-1)->attribute == EFI_MEMORY_UC &&
1000 kmd_end(k-1) == md->phys_addr) {
1001 (k-1)->num_pages += md->num_pages;
1002 } else {
1003 k->attribute = EFI_MEMORY_UC;
1004 k->start = lim;
1005 k->num_pages = (efi_md_end(md) - lim) >> EFI_PAGE_SHIFT;
1006 k++;
1007 }
1008 }
1009 ae = contig_high;
1010 } else
1011 ae = efi_md_end(md);
1012
a7956113
ZN		 1013 /* keep within max_addr= and min_addr= command line arg */
1014 as = max(as, min_addr);
d8c97d5f
TL		 1015 ae = min(ae, max_addr);
1016 if (ae <= as)
1017 continue;
1018
1019 /* avoid going over mem= command line arg */
1020 if (total_mem + (ae - as) > mem_limit)
1021 ae -= total_mem + (ae - as) - mem_limit;
1022
1023 if (ae <= as)
1024 continue;
1025 if (prev && kmd_end(prev) == md->phys_addr) {
1026 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1027 total_mem += ae - as;
1028 continue;
1029 }
1030 k->attribute = EFI_MEMORY_WB;
1031 k->start = as;
1032 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1033 total_mem += ae - as;
1034 prev = k++;
1035 }
1036 k->start = ~0L; /* end-marker */
1037
1038 /* reserve the memory we are using for kern_memmap */
1039 *s = (u64)kern_memmap;
1040 *e = (u64)++k;
1041}
be379124
KA		 1042
1043void
1044efi_initialize_iomem_resources(struct resource *code_resource,
1045 struct resource *data_resource)
1046{
1047 struct resource *res;
1048 void *efi_map_start, *efi_map_end, *p;
1049 efi_memory_desc_t *md;
1050 u64 efi_desc_size;
1051 char *name;
1052 unsigned long flags;
1053
1054 efi_map_start = __va(ia64_boot_param->efi_memmap);
1055 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1056 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1057
1058 res = NULL;
1059
1060 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1061 md = p;
1062
1063 if (md->num_pages == 0) /* should not happen */
1064 continue;
1065
1066 flags = IORESOURCE_MEM;
1067 switch (md->type) {
1068
1069 case EFI_MEMORY_MAPPED_IO:
1070 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1071 continue;
1072
1073 case EFI_LOADER_CODE:
1074 case EFI_LOADER_DATA:
1075 case EFI_BOOT_SERVICES_DATA:
1076 case EFI_BOOT_SERVICES_CODE:
1077 case EFI_CONVENTIONAL_MEMORY:
1078 if (md->attribute & EFI_MEMORY_WP) {
1079 name = "System ROM";
1080 flags |= IORESOURCE_READONLY;
1081 } else {
1082 name = "System RAM";
1083 }
1084 break;
1085
1086 case EFI_ACPI_MEMORY_NVS:
1087 name = "ACPI Non-volatile Storage";
1088 flags |= IORESOURCE_BUSY;
1089 break;
1090
1091 case EFI_UNUSABLE_MEMORY:
1092 name = "reserved";
1093 flags |= IORESOURCE_BUSY | IORESOURCE_DISABLED;
1094 break;
1095
1096 case EFI_RESERVED_TYPE:
1097 case EFI_RUNTIME_SERVICES_CODE:
1098 case EFI_RUNTIME_SERVICES_DATA:
1099 case EFI_ACPI_RECLAIM_MEMORY:
1100 default:
1101 name = "reserved";
1102 flags |= IORESOURCE_BUSY;
1103 break;
1104 }
1105
baf47fb6 1106 if ((res = kzalloc(sizeof(struct resource), GFP_KERNEL)) == NULL) {
be379124
KA		 1107 printk(KERN_ERR "failed to alocate resource for iomem\n");
1108 return;
1109 }
1110
1111 res->name = name;
1112 res->start = md->phys_addr;
1113 res->end = md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1;
1114 res->flags = flags;
1115
1116 if (insert_resource(&iomem_resource, res) < 0)
1117 kfree(res);
1118 else {
1119 /*
1120 * We don't know which region contains
1121 * kernel data so we try it repeatedly and
1122 * let the resource manager test it.
1123 */
1124 insert_resource(res, code_resource);
1125 insert_resource(res, data_resource);
a7956113
ZN		 1126#ifdef CONFIG_KEXEC
1127 insert_resource(res, &efi_memmap_res);
1128 insert_resource(res, &boot_param_res);
1129 if (crashk_res.end > crashk_res.start)
1130 insert_resource(res, &crashk_res);
1131#endif
be379124
KA		 1132 }
1133 }
1134}
a7956113
ZN		 1135
1136#ifdef CONFIG_KEXEC
1137/* find a block of memory aligned to 64M exclude reserved regions
1138 rsvd_regions are sorted
1139 */
1140unsigned long
1141kdump_find_rsvd_region (unsigned long size,
1142 struct rsvd_region *r, int n)
1143{
1144 int i;
1145 u64 start, end;
1146 u64 alignment = 1UL << _PAGE_SIZE_64M;
1147 void *efi_map_start, *efi_map_end, *p;
1148 efi_memory_desc_t *md;
1149 u64 efi_desc_size;
1150
1151 efi_map_start = __va(ia64_boot_param->efi_memmap);
1152 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1153 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1154
1155 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1156 md = p;
1157 if (!efi_wb(md))
1158 continue;
1159 start = ALIGN(md->phys_addr, alignment);
1160 end = efi_md_end(md);
1161 for (i = 0; i < n; i++) {
1162 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1163 if (__pa(r[i].start) > start + size)
1164 return start;
1165 start = ALIGN(__pa(r[i].end), alignment);
1166 if (i < n-1 && __pa(r[i+1].start) < start + size)
1167 continue;
1168 else
1169 break;
1170 }
1171 }
1172 if (end > start + size)
1173 return start;
1174 }
1175
1176 printk(KERN_WARNING "Cannot reserve 0x%lx byte of memory for crashdump\n",
1177 size);
1178 return ~0UL;
1179}
1180#endif
Linux 6.x block layer and io_uring tree(s)