Loading...
Note: File does not exist in v6.8.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Extensible Firmware Interface
4 *
5 * Based on Extensible Firmware Interface Specification version 0.9
6 * April 30, 1999
7 *
8 * Copyright (C) 1999 VA Linux Systems
9 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
10 * Copyright (C) 1999-2003 Hewlett-Packard Co.
11 * David Mosberger-Tang <davidm@hpl.hp.com>
12 * Stephane Eranian <eranian@hpl.hp.com>
13 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P.
14 * Bjorn Helgaas <bjorn.helgaas@hp.com>
15 *
16 * All EFI Runtime Services are not implemented yet as EFI only
17 * supports physical mode addressing on SoftSDV. This is to be fixed
18 * in a future version. --drummond 1999-07-20
19 *
20 * Implemented EFI runtime services and virtual mode calls. --davidm
21 *
22 * Goutham Rao: <goutham.rao@intel.com>
23 * Skip non-WB memory and ignore empty memory ranges.
24 */
25#include <linux/module.h>
26#include <linux/memblock.h>
27#include <linux/crash_dump.h>
28#include <linux/kernel.h>
29#include <linux/init.h>
30#include <linux/types.h>
31#include <linux/slab.h>
32#include <linux/time.h>
33#include <linux/efi.h>
34#include <linux/kexec.h>
35#include <linux/mm.h>
36
37#include <asm/efi.h>
38#include <asm/io.h>
39#include <asm/kregs.h>
40#include <asm/meminit.h>
41#include <asm/processor.h>
42#include <asm/mca.h>
43#include <asm/sal.h>
44#include <asm/setup.h>
45#include <asm/tlbflush.h>
46
47#define EFI_DEBUG 0
48
49#define ESI_TABLE_GUID \
50 EFI_GUID(0x43EA58DC, 0xCF28, 0x4b06, 0xB3, \
51 0x91, 0xB7, 0x50, 0x59, 0x34, 0x2B, 0xD4)
52
53static unsigned long mps_phys = EFI_INVALID_TABLE_ADDR;
54static __initdata unsigned long palo_phys;
55
56unsigned long __initdata esi_phys = EFI_INVALID_TABLE_ADDR;
57unsigned long hcdp_phys = EFI_INVALID_TABLE_ADDR;
58unsigned long sal_systab_phys = EFI_INVALID_TABLE_ADDR;
59
60static const efi_config_table_type_t arch_tables[] __initconst = {
61 {ESI_TABLE_GUID, &esi_phys, "ESI" },
62 {HCDP_TABLE_GUID, &hcdp_phys, "HCDP" },
63 {MPS_TABLE_GUID, &mps_phys, "MPS" },
64 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, &palo_phys, "PALO" },
65 {SAL_SYSTEM_TABLE_GUID, &sal_systab_phys, "SALsystab" },
66 {},
67};
68
69extern efi_status_t efi_call_phys (void *, ...);
70
71static efi_runtime_services_t *runtime;
72static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
73
74#define efi_call_virt(f, args...) (*(f))(args)
75
76#define STUB_GET_TIME(prefix, adjust_arg) \
77static efi_status_t \
78prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
79{ \
80 struct ia64_fpreg fr[6]; \
81 efi_time_cap_t *atc = NULL; \
82 efi_status_t ret; \
83 \
84 if (tc) \
85 atc = adjust_arg(tc); \
86 ia64_save_scratch_fpregs(fr); \
87 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
88 adjust_arg(tm), atc); \
89 ia64_load_scratch_fpregs(fr); \
90 return ret; \
91}
92
93#define STUB_SET_TIME(prefix, adjust_arg) \
94static efi_status_t \
95prefix##_set_time (efi_time_t *tm) \
96{ \
97 struct ia64_fpreg fr[6]; \
98 efi_status_t ret; \
99 \
100 ia64_save_scratch_fpregs(fr); \
101 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
102 adjust_arg(tm)); \
103 ia64_load_scratch_fpregs(fr); \
104 return ret; \
105}
106
107#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
108static efi_status_t \
109prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
110 efi_time_t *tm) \
111{ \
112 struct ia64_fpreg fr[6]; \
113 efi_status_t ret; \
114 \
115 ia64_save_scratch_fpregs(fr); \
116 ret = efi_call_##prefix( \
117 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
118 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
119 ia64_load_scratch_fpregs(fr); \
120 return ret; \
121}
122
123#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
124static efi_status_t \
125prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
126{ \
127 struct ia64_fpreg fr[6]; \
128 efi_time_t *atm = NULL; \
129 efi_status_t ret; \
130 \
131 if (tm) \
132 atm = adjust_arg(tm); \
133 ia64_save_scratch_fpregs(fr); \
134 ret = efi_call_##prefix( \
135 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
136 enabled, atm); \
137 ia64_load_scratch_fpregs(fr); \
138 return ret; \
139}
140
141#define STUB_GET_VARIABLE(prefix, adjust_arg) \
142static efi_status_t \
143prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
144 unsigned long *data_size, void *data) \
145{ \
146 struct ia64_fpreg fr[6]; \
147 u32 *aattr = NULL; \
148 efi_status_t ret; \
149 \
150 if (attr) \
151 aattr = adjust_arg(attr); \
152 ia64_save_scratch_fpregs(fr); \
153 ret = efi_call_##prefix( \
154 (efi_get_variable_t *) __va(runtime->get_variable), \
155 adjust_arg(name), adjust_arg(vendor), aattr, \
156 adjust_arg(data_size), adjust_arg(data)); \
157 ia64_load_scratch_fpregs(fr); \
158 return ret; \
159}
160
161#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
162static efi_status_t \
163prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
164 efi_guid_t *vendor) \
165{ \
166 struct ia64_fpreg fr[6]; \
167 efi_status_t ret; \
168 \
169 ia64_save_scratch_fpregs(fr); \
170 ret = efi_call_##prefix( \
171 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
172 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
173 ia64_load_scratch_fpregs(fr); \
174 return ret; \
175}
176
177#define STUB_SET_VARIABLE(prefix, adjust_arg) \
178static efi_status_t \
179prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
180 u32 attr, unsigned long data_size, \
181 void *data) \
182{ \
183 struct ia64_fpreg fr[6]; \
184 efi_status_t ret; \
185 \
186 ia64_save_scratch_fpregs(fr); \
187 ret = efi_call_##prefix( \
188 (efi_set_variable_t *) __va(runtime->set_variable), \
189 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
190 adjust_arg(data)); \
191 ia64_load_scratch_fpregs(fr); \
192 return ret; \
193}
194
195#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
196static efi_status_t \
197prefix##_get_next_high_mono_count (u32 *count) \
198{ \
199 struct ia64_fpreg fr[6]; \
200 efi_status_t ret; \
201 \
202 ia64_save_scratch_fpregs(fr); \
203 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
204 __va(runtime->get_next_high_mono_count), \
205 adjust_arg(count)); \
206 ia64_load_scratch_fpregs(fr); \
207 return ret; \
208}
209
210#define STUB_RESET_SYSTEM(prefix, adjust_arg) \
211static void \
212prefix##_reset_system (int reset_type, efi_status_t status, \
213 unsigned long data_size, efi_char16_t *data) \
214{ \
215 struct ia64_fpreg fr[6]; \
216 efi_char16_t *adata = NULL; \
217 \
218 if (data) \
219 adata = adjust_arg(data); \
220 \
221 ia64_save_scratch_fpregs(fr); \
222 efi_call_##prefix( \
223 (efi_reset_system_t *) __va(runtime->reset_system), \
224 reset_type, status, data_size, adata); \
225 /* should not return, but just in case... */ \
226 ia64_load_scratch_fpregs(fr); \
227}
228
229#define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
230
231STUB_GET_TIME(phys, phys_ptr)
232STUB_SET_TIME(phys, phys_ptr)
233STUB_GET_WAKEUP_TIME(phys, phys_ptr)
234STUB_SET_WAKEUP_TIME(phys, phys_ptr)
235STUB_GET_VARIABLE(phys, phys_ptr)
236STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
237STUB_SET_VARIABLE(phys, phys_ptr)
238STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
239STUB_RESET_SYSTEM(phys, phys_ptr)
240
241#define id(arg) arg
242
243STUB_GET_TIME(virt, id)
244STUB_SET_TIME(virt, id)
245STUB_GET_WAKEUP_TIME(virt, id)
246STUB_SET_WAKEUP_TIME(virt, id)
247STUB_GET_VARIABLE(virt, id)
248STUB_GET_NEXT_VARIABLE(virt, id)
249STUB_SET_VARIABLE(virt, id)
250STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
251STUB_RESET_SYSTEM(virt, id)
252
253void
254efi_gettimeofday (struct timespec64 *ts)
255{
256 efi_time_t tm;
257
258 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
259 memset(ts, 0, sizeof(*ts));
260 return;
261 }
262
263 ts->tv_sec = mktime64(tm.year, tm.month, tm.day,
264 tm.hour, tm.minute, tm.second);
265 ts->tv_nsec = tm.nanosecond;
266}
267
268static int
269is_memory_available (efi_memory_desc_t *md)
270{
271 if (!(md->attribute & EFI_MEMORY_WB))
272 return 0;
273
274 switch (md->type) {
275 case EFI_LOADER_CODE:
276 case EFI_LOADER_DATA:
277 case EFI_BOOT_SERVICES_CODE:
278 case EFI_BOOT_SERVICES_DATA:
279 case EFI_CONVENTIONAL_MEMORY:
280 return 1;
281 }
282 return 0;
283}
284
285typedef struct kern_memdesc {
286 u64 attribute;
287 u64 start;
288 u64 num_pages;
289} kern_memdesc_t;
290
291static kern_memdesc_t *kern_memmap;
292
293#define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
294
295static inline u64
296kmd_end(kern_memdesc_t *kmd)
297{
298 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
299}
300
301static inline u64
302efi_md_end(efi_memory_desc_t *md)
303{
304 return (md->phys_addr + efi_md_size(md));
305}
306
307static inline int
308efi_wb(efi_memory_desc_t *md)
309{
310 return (md->attribute & EFI_MEMORY_WB);
311}
312
313static inline int
314efi_uc(efi_memory_desc_t *md)
315{
316 return (md->attribute & EFI_MEMORY_UC);
317}
318
319static void
320walk (efi_freemem_callback_t callback, void *arg, u64 attr)
321{
322 kern_memdesc_t *k;
323 u64 start, end, voff;
324
325 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
326 for (k = kern_memmap; k->start != ~0UL; k++) {
327 if (k->attribute != attr)
328 continue;
329 start = PAGE_ALIGN(k->start);
330 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
331 if (start < end)
332 if ((*callback)(start + voff, end + voff, arg) < 0)
333 return;
334 }
335}
336
337/*
338 * Walk the EFI memory map and call CALLBACK once for each EFI memory
339 * descriptor that has memory that is available for OS use.
340 */
341void
342efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
343{
344 walk(callback, arg, EFI_MEMORY_WB);
345}
346
347/*
348 * Walk the EFI memory map and call CALLBACK once for each EFI memory
349 * descriptor that has memory that is available for uncached allocator.
350 */
351void
352efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
353{
354 walk(callback, arg, EFI_MEMORY_UC);
355}
356
357/*
358 * Look for the PAL_CODE region reported by EFI and map it using an
359 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
360 * Abstraction Layer chapter 11 in ADAG
361 */
362void *
363efi_get_pal_addr (void)
364{
365 void *efi_map_start, *efi_map_end, *p;
366 efi_memory_desc_t *md;
367 u64 efi_desc_size;
368 int pal_code_count = 0;
369 u64 vaddr, mask;
370
371 efi_map_start = __va(ia64_boot_param->efi_memmap);
372 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
373 efi_desc_size = ia64_boot_param->efi_memdesc_size;
374
375 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
376 md = p;
377 if (md->type != EFI_PAL_CODE)
378 continue;
379
380 if (++pal_code_count > 1) {
381 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
382 "dropped @ %llx\n", md->phys_addr);
383 continue;
384 }
385 /*
386 * The only ITLB entry in region 7 that is used is the one
387 * installed by __start(). That entry covers a 64MB range.
388 */
389 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
390 vaddr = PAGE_OFFSET + md->phys_addr;
391
392 /*
393 * We must check that the PAL mapping won't overlap with the
394 * kernel mapping.
395 *
396 * PAL code is guaranteed to be aligned on a power of 2 between
397 * 4k and 256KB and that only one ITR is needed to map it. This
398 * implies that the PAL code is always aligned on its size,
399 * i.e., the closest matching page size supported by the TLB.
400 * Therefore PAL code is guaranteed never to cross a 64MB unless
401 * it is bigger than 64MB (very unlikely!). So for now the
402 * following test is enough to determine whether or not we need
403 * a dedicated ITR for the PAL code.
404 */
405 if ((vaddr & mask) == (KERNEL_START & mask)) {
406 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
407 __func__);
408 continue;
409 }
410
411 if (efi_md_size(md) > IA64_GRANULE_SIZE)
412 panic("Whoa! PAL code size bigger than a granule!");
413
414#if EFI_DEBUG
415 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
416
417 printk(KERN_INFO "CPU %d: mapping PAL code "
418 "[0x%llx-0x%llx) into [0x%llx-0x%llx)\n",
419 smp_processor_id(), md->phys_addr,
420 md->phys_addr + efi_md_size(md),
421 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
422#endif
423 return __va(md->phys_addr);
424 }
425 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
426 __func__);
427 return NULL;
428}
429
430
431static u8 __init palo_checksum(u8 *buffer, u32 length)
432{
433 u8 sum = 0;
434 u8 *end = buffer + length;
435
436 while (buffer < end)
437 sum = (u8) (sum + *(buffer++));
438
439 return sum;
440}
441
442/*
443 * Parse and handle PALO table which is published at:
444 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
445 */
446static void __init handle_palo(unsigned long phys_addr)
447{
448 struct palo_table *palo = __va(phys_addr);
449 u8 checksum;
450
451 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
452 printk(KERN_INFO "PALO signature incorrect.\n");
453 return;
454 }
455
456 checksum = palo_checksum((u8 *)palo, palo->length);
457 if (checksum) {
458 printk(KERN_INFO "PALO checksum incorrect.\n");
459 return;
460 }
461
462 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
463}
464
465void
466efi_map_pal_code (void)
467{
468 void *pal_vaddr = efi_get_pal_addr ();
469 u64 psr;
470
471 if (!pal_vaddr)
472 return;
473
474 /*
475 * Cannot write to CRx with PSR.ic=1
476 */
477 psr = ia64_clear_ic();
478 ia64_itr(0x1, IA64_TR_PALCODE,
479 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
480 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
481 IA64_GRANULE_SHIFT);
482 ia64_set_psr(psr); /* restore psr */
483}
484
485void __init
486efi_init (void)
487{
488 const efi_system_table_t *efi_systab;
489 void *efi_map_start, *efi_map_end;
490 u64 efi_desc_size;
491 char *cp;
492
493 set_bit(EFI_BOOT, &efi.flags);
494 set_bit(EFI_64BIT, &efi.flags);
495
496 /*
497 * It's too early to be able to use the standard kernel command line
498 * support...
499 */
500 for (cp = boot_command_line; *cp; ) {
501 if (memcmp(cp, "mem=", 4) == 0) {
502 mem_limit = memparse(cp + 4, &cp);
503 } else if (memcmp(cp, "max_addr=", 9) == 0) {
504 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
505 } else if (memcmp(cp, "min_addr=", 9) == 0) {
506 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
507 } else {
508 while (*cp != ' ' && *cp)
509 ++cp;
510 while (*cp == ' ')
511 ++cp;
512 }
513 }
514 if (min_addr != 0UL)
515 printk(KERN_INFO "Ignoring memory below %lluMB\n",
516 min_addr >> 20);
517 if (max_addr != ~0UL)
518 printk(KERN_INFO "Ignoring memory above %lluMB\n",
519 max_addr >> 20);
520
521 efi_systab = __va(ia64_boot_param->efi_systab);
522
523 /*
524 * Verify the EFI Table
525 */
526 if (efi_systab == NULL)
527 panic("Whoa! Can't find EFI system table.\n");
528 if (efi_systab_check_header(&efi_systab->hdr, 1))
529 panic("Whoa! EFI system table signature incorrect\n");
530
531 efi_systab_report_header(&efi_systab->hdr, efi_systab->fw_vendor);
532
533 palo_phys = EFI_INVALID_TABLE_ADDR;
534
535 if (efi_config_parse_tables(__va(efi_systab->tables),
536 efi_systab->nr_tables,
537 arch_tables) != 0)
538 return;
539
540 if (palo_phys != EFI_INVALID_TABLE_ADDR)
541 handle_palo(palo_phys);
542
543 runtime = __va(efi_systab->runtime);
544 efi.get_time = phys_get_time;
545 efi.set_time = phys_set_time;
546 efi.get_wakeup_time = phys_get_wakeup_time;
547 efi.set_wakeup_time = phys_set_wakeup_time;
548 efi.get_variable = phys_get_variable;
549 efi.get_next_variable = phys_get_next_variable;
550 efi.set_variable = phys_set_variable;
551 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
552 efi.reset_system = phys_reset_system;
553
554 efi_map_start = __va(ia64_boot_param->efi_memmap);
555 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
556 efi_desc_size = ia64_boot_param->efi_memdesc_size;
557
558#if EFI_DEBUG
559 /* print EFI memory map: */
560 {
561 efi_memory_desc_t *md;
562 void *p;
563 unsigned int i;
564
565 for (i = 0, p = efi_map_start; p < efi_map_end;
566 ++i, p += efi_desc_size)
567 {
568 const char *unit;
569 unsigned long size;
570 char buf[64];
571
572 md = p;
573 size = md->num_pages << EFI_PAGE_SHIFT;
574
575 if ((size >> 40) > 0) {
576 size >>= 40;
577 unit = "TB";
578 } else if ((size >> 30) > 0) {
579 size >>= 30;
580 unit = "GB";
581 } else if ((size >> 20) > 0) {
582 size >>= 20;
583 unit = "MB";
584 } else {
585 size >>= 10;
586 unit = "KB";
587 }
588
589 printk("mem%02d: %s "
590 "range=[0x%016llx-0x%016llx) (%4lu%s)\n",
591 i, efi_md_typeattr_format(buf, sizeof(buf), md),
592 md->phys_addr,
593 md->phys_addr + efi_md_size(md), size, unit);
594 }
595 }
596#endif
597
598 efi_map_pal_code();
599 efi_enter_virtual_mode();
600}
601
602void
603efi_enter_virtual_mode (void)
604{
605 void *efi_map_start, *efi_map_end, *p;
606 efi_memory_desc_t *md;
607 efi_status_t status;
608 u64 efi_desc_size;
609
610 efi_map_start = __va(ia64_boot_param->efi_memmap);
611 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
612 efi_desc_size = ia64_boot_param->efi_memdesc_size;
613
614 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
615 md = p;
616 if (md->attribute & EFI_MEMORY_RUNTIME) {
617 /*
618 * Some descriptors have multiple bits set, so the
619 * order of the tests is relevant.
620 */
621 if (md->attribute & EFI_MEMORY_WB) {
622 md->virt_addr = (u64) __va(md->phys_addr);
623 } else if (md->attribute & EFI_MEMORY_UC) {
624 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
625 } else if (md->attribute & EFI_MEMORY_WC) {
626#if 0
627 md->virt_addr = ia64_remap(md->phys_addr,
628 (_PAGE_A |
629 _PAGE_P |
630 _PAGE_D |
631 _PAGE_MA_WC |
632 _PAGE_PL_0 |
633 _PAGE_AR_RW));
634#else
635 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
636 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
637#endif
638 } else if (md->attribute & EFI_MEMORY_WT) {
639#if 0
640 md->virt_addr = ia64_remap(md->phys_addr,
641 (_PAGE_A |
642 _PAGE_P |
643 _PAGE_D |
644 _PAGE_MA_WT |
645 _PAGE_PL_0 |
646 _PAGE_AR_RW));
647#else
648 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
649 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
650#endif
651 }
652 }
653 }
654
655 status = efi_call_phys(__va(runtime->set_virtual_address_map),
656 ia64_boot_param->efi_memmap_size,
657 efi_desc_size,
658 ia64_boot_param->efi_memdesc_version,
659 ia64_boot_param->efi_memmap);
660 if (status != EFI_SUCCESS) {
661 printk(KERN_WARNING "warning: unable to switch EFI into "
662 "virtual mode (status=%lu)\n", status);
663 return;
664 }
665
666 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
667
668 /*
669 * Now that EFI is in virtual mode, we call the EFI functions more
670 * efficiently:
671 */
672 efi.get_time = virt_get_time;
673 efi.set_time = virt_set_time;
674 efi.get_wakeup_time = virt_get_wakeup_time;
675 efi.set_wakeup_time = virt_set_wakeup_time;
676 efi.get_variable = virt_get_variable;
677 efi.get_next_variable = virt_get_next_variable;
678 efi.set_variable = virt_set_variable;
679 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
680 efi.reset_system = virt_reset_system;
681}
682
683/*
684 * Walk the EFI memory map looking for the I/O port range. There can only be
685 * one entry of this type, other I/O port ranges should be described via ACPI.
686 */
687u64
688efi_get_iobase (void)
689{
690 void *efi_map_start, *efi_map_end, *p;
691 efi_memory_desc_t *md;
692 u64 efi_desc_size;
693
694 efi_map_start = __va(ia64_boot_param->efi_memmap);
695 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
696 efi_desc_size = ia64_boot_param->efi_memdesc_size;
697
698 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
699 md = p;
700 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
701 if (md->attribute & EFI_MEMORY_UC)
702 return md->phys_addr;
703 }
704 }
705 return 0;
706}
707
708static struct kern_memdesc *
709kern_memory_descriptor (unsigned long phys_addr)
710{
711 struct kern_memdesc *md;
712
713 for (md = kern_memmap; md->start != ~0UL; md++) {
714 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
715 return md;
716 }
717 return NULL;
718}
719
720static efi_memory_desc_t *
721efi_memory_descriptor (unsigned long phys_addr)
722{
723 void *efi_map_start, *efi_map_end, *p;
724 efi_memory_desc_t *md;
725 u64 efi_desc_size;
726
727 efi_map_start = __va(ia64_boot_param->efi_memmap);
728 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
729 efi_desc_size = ia64_boot_param->efi_memdesc_size;
730
731 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
732 md = p;
733
734 if (phys_addr - md->phys_addr < efi_md_size(md))
735 return md;
736 }
737 return NULL;
738}
739
740static int
741efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
742{
743 void *efi_map_start, *efi_map_end, *p;
744 efi_memory_desc_t *md;
745 u64 efi_desc_size;
746 unsigned long end;
747
748 efi_map_start = __va(ia64_boot_param->efi_memmap);
749 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
750 efi_desc_size = ia64_boot_param->efi_memdesc_size;
751
752 end = phys_addr + size;
753
754 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
755 md = p;
756 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
757 return 1;
758 }
759 return 0;
760}
761
762int
763efi_mem_type (unsigned long phys_addr)
764{
765 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
766
767 if (md)
768 return md->type;
769 return -EINVAL;
770}
771
772u64
773efi_mem_attributes (unsigned long phys_addr)
774{
775 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
776
777 if (md)
778 return md->attribute;
779 return 0;
780}
781EXPORT_SYMBOL(efi_mem_attributes);
782
783u64
784efi_mem_attribute (unsigned long phys_addr, unsigned long size)
785{
786 unsigned long end = phys_addr + size;
787 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
788 u64 attr;
789
790 if (!md)
791 return 0;
792
793 /*
794 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
795 * the kernel that firmware needs this region mapped.
796 */
797 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
798 do {
799 unsigned long md_end = efi_md_end(md);
800
801 if (end <= md_end)
802 return attr;
803
804 md = efi_memory_descriptor(md_end);
805 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
806 return 0;
807 } while (md);
808 return 0; /* never reached */
809}
810
811u64
812kern_mem_attribute (unsigned long phys_addr, unsigned long size)
813{
814 unsigned long end = phys_addr + size;
815 struct kern_memdesc *md;
816 u64 attr;
817
818 /*
819 * This is a hack for ioremap calls before we set up kern_memmap.
820 * Maybe we should do efi_memmap_init() earlier instead.
821 */
822 if (!kern_memmap) {
823 attr = efi_mem_attribute(phys_addr, size);
824 if (attr & EFI_MEMORY_WB)
825 return EFI_MEMORY_WB;
826 return 0;
827 }
828
829 md = kern_memory_descriptor(phys_addr);
830 if (!md)
831 return 0;
832
833 attr = md->attribute;
834 do {
835 unsigned long md_end = kmd_end(md);
836
837 if (end <= md_end)
838 return attr;
839
840 md = kern_memory_descriptor(md_end);
841 if (!md || md->attribute != attr)
842 return 0;
843 } while (md);
844 return 0; /* never reached */
845}
846
847int
848valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
849{
850 u64 attr;
851
852 /*
853 * /dev/mem reads and writes use copy_to_user(), which implicitly
854 * uses a granule-sized kernel identity mapping. It's really
855 * only safe to do this for regions in kern_memmap. For more
856 * details, see Documentation/ia64/aliasing.rst.
857 */
858 attr = kern_mem_attribute(phys_addr, size);
859 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
860 return 1;
861 return 0;
862}
863
864int
865valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
866{
867 unsigned long phys_addr = pfn << PAGE_SHIFT;
868 u64 attr;
869
870 attr = efi_mem_attribute(phys_addr, size);
871
872 /*
873 * /dev/mem mmap uses normal user pages, so we don't need the entire
874 * granule, but the entire region we're mapping must support the same
875 * attribute.
876 */
877 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
878 return 1;
879
880 /*
881 * Intel firmware doesn't tell us about all the MMIO regions, so
882 * in general we have to allow mmap requests. But if EFI *does*
883 * tell us about anything inside this region, we should deny it.
884 * The user can always map a smaller region to avoid the overlap.
885 */
886 if (efi_memmap_intersects(phys_addr, size))
887 return 0;
888
889 return 1;
890}
891
892pgprot_t
893phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
894 pgprot_t vma_prot)
895{
896 unsigned long phys_addr = pfn << PAGE_SHIFT;
897 u64 attr;
898
899 /*
900 * For /dev/mem mmap, we use user mappings, but if the region is
901 * in kern_memmap (and hence may be covered by a kernel mapping),
902 * we must use the same attribute as the kernel mapping.
903 */
904 attr = kern_mem_attribute(phys_addr, size);
905 if (attr & EFI_MEMORY_WB)
906 return pgprot_cacheable(vma_prot);
907 else if (attr & EFI_MEMORY_UC)
908 return pgprot_noncached(vma_prot);
909
910 /*
911 * Some chipsets don't support UC access to memory. If
912 * WB is supported, we prefer that.
913 */
914 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
915 return pgprot_cacheable(vma_prot);
916
917 return pgprot_noncached(vma_prot);
918}
919
920int __init
921efi_uart_console_only(void)
922{
923 efi_status_t status;
924 char *s, name[] = "ConOut";
925 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
926 efi_char16_t *utf16, name_utf16[32];
927 unsigned char data[1024];
928 unsigned long size = sizeof(data);
929 struct efi_generic_dev_path *hdr, *end_addr;
930 int uart = 0;
931
932 /* Convert to UTF-16 */
933 utf16 = name_utf16;
934 s = name;
935 while (*s)
936 *utf16++ = *s++ & 0x7f;
937 *utf16 = 0;
938
939 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
940 if (status != EFI_SUCCESS) {
941 printk(KERN_ERR "No EFI %s variable?\n", name);
942 return 0;
943 }
944
945 hdr = (struct efi_generic_dev_path *) data;
946 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
947 while (hdr < end_addr) {
948 if (hdr->type == EFI_DEV_MSG &&
949 hdr->sub_type == EFI_DEV_MSG_UART)
950 uart = 1;
951 else if (hdr->type == EFI_DEV_END_PATH ||
952 hdr->type == EFI_DEV_END_PATH2) {
953 if (!uart)
954 return 0;
955 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
956 return 1;
957 uart = 0;
958 }
959 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
960 }
961 printk(KERN_ERR "Malformed %s value\n", name);
962 return 0;
963}
964
965/*
966 * Look for the first granule aligned memory descriptor memory
967 * that is big enough to hold EFI memory map. Make sure this
968 * descriptor is at least granule sized so it does not get trimmed
969 */
970struct kern_memdesc *
971find_memmap_space (void)
972{
973 u64 contig_low=0, contig_high=0;
974 u64 as = 0, ae;
975 void *efi_map_start, *efi_map_end, *p, *q;
976 efi_memory_desc_t *md, *pmd = NULL, *check_md;
977 u64 space_needed, efi_desc_size;
978 unsigned long total_mem = 0;
979
980 efi_map_start = __va(ia64_boot_param->efi_memmap);
981 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
982 efi_desc_size = ia64_boot_param->efi_memdesc_size;
983
984 /*
985 * Worst case: we need 3 kernel descriptors for each efi descriptor
986 * (if every entry has a WB part in the middle, and UC head and tail),
987 * plus one for the end marker.
988 */
989 space_needed = sizeof(kern_memdesc_t) *
990 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
991
992 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
993 md = p;
994 if (!efi_wb(md)) {
995 continue;
996 }
997 if (pmd == NULL || !efi_wb(pmd) ||
998 efi_md_end(pmd) != md->phys_addr) {
999 contig_low = GRANULEROUNDUP(md->phys_addr);
1000 contig_high = efi_md_end(md);
1001 for (q = p + efi_desc_size; q < efi_map_end;
1002 q += efi_desc_size) {
1003 check_md = q;
1004 if (!efi_wb(check_md))
1005 break;
1006 if (contig_high != check_md->phys_addr)
1007 break;
1008 contig_high = efi_md_end(check_md);
1009 }
1010 contig_high = GRANULEROUNDDOWN(contig_high);
1011 }
1012 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1013 continue;
1014
1015 /* Round ends inward to granule boundaries */
1016 as = max(contig_low, md->phys_addr);
1017 ae = min(contig_high, efi_md_end(md));
1018
1019 /* keep within max_addr= and min_addr= command line arg */
1020 as = max(as, min_addr);
1021 ae = min(ae, max_addr);
1022 if (ae <= as)
1023 continue;
1024
1025 /* avoid going over mem= command line arg */
1026 if (total_mem + (ae - as) > mem_limit)
1027 ae -= total_mem + (ae - as) - mem_limit;
1028
1029 if (ae <= as)
1030 continue;
1031
1032 if (ae - as > space_needed)
1033 break;
1034 }
1035 if (p >= efi_map_end)
1036 panic("Can't allocate space for kernel memory descriptors");
1037
1038 return __va(as);
1039}
1040
1041/*
1042 * Walk the EFI memory map and gather all memory available for kernel
1043 * to use. We can allocate partial granules only if the unavailable
1044 * parts exist, and are WB.
1045 */
1046unsigned long
1047efi_memmap_init(u64 *s, u64 *e)
1048{
1049 struct kern_memdesc *k, *prev = NULL;
1050 u64 contig_low=0, contig_high=0;
1051 u64 as, ae, lim;
1052 void *efi_map_start, *efi_map_end, *p, *q;
1053 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1054 u64 efi_desc_size;
1055 unsigned long total_mem = 0;
1056
1057 k = kern_memmap = find_memmap_space();
1058
1059 efi_map_start = __va(ia64_boot_param->efi_memmap);
1060 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1061 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1062
1063 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1064 md = p;
1065 if (!efi_wb(md)) {
1066 if (efi_uc(md) &&
1067 (md->type == EFI_CONVENTIONAL_MEMORY ||
1068 md->type == EFI_BOOT_SERVICES_DATA)) {
1069 k->attribute = EFI_MEMORY_UC;
1070 k->start = md->phys_addr;
1071 k->num_pages = md->num_pages;
1072 k++;
1073 }
1074 continue;
1075 }
1076 if (pmd == NULL || !efi_wb(pmd) ||
1077 efi_md_end(pmd) != md->phys_addr) {
1078 contig_low = GRANULEROUNDUP(md->phys_addr);
1079 contig_high = efi_md_end(md);
1080 for (q = p + efi_desc_size; q < efi_map_end;
1081 q += efi_desc_size) {
1082 check_md = q;
1083 if (!efi_wb(check_md))
1084 break;
1085 if (contig_high != check_md->phys_addr)
1086 break;
1087 contig_high = efi_md_end(check_md);
1088 }
1089 contig_high = GRANULEROUNDDOWN(contig_high);
1090 }
1091 if (!is_memory_available(md))
1092 continue;
1093
1094 /*
1095 * Round ends inward to granule boundaries
1096 * Give trimmings to uncached allocator
1097 */
1098 if (md->phys_addr < contig_low) {
1099 lim = min(efi_md_end(md), contig_low);
1100 if (efi_uc(md)) {
1101 if (k > kern_memmap &&
1102 (k-1)->attribute == EFI_MEMORY_UC &&
1103 kmd_end(k-1) == md->phys_addr) {
1104 (k-1)->num_pages +=
1105 (lim - md->phys_addr)
1106 >> EFI_PAGE_SHIFT;
1107 } else {
1108 k->attribute = EFI_MEMORY_UC;
1109 k->start = md->phys_addr;
1110 k->num_pages = (lim - md->phys_addr)
1111 >> EFI_PAGE_SHIFT;
1112 k++;
1113 }
1114 }
1115 as = contig_low;
1116 } else
1117 as = md->phys_addr;
1118
1119 if (efi_md_end(md) > contig_high) {
1120 lim = max(md->phys_addr, contig_high);
1121 if (efi_uc(md)) {
1122 if (lim == md->phys_addr && k > kern_memmap &&
1123 (k-1)->attribute == EFI_MEMORY_UC &&
1124 kmd_end(k-1) == md->phys_addr) {
1125 (k-1)->num_pages += md->num_pages;
1126 } else {
1127 k->attribute = EFI_MEMORY_UC;
1128 k->start = lim;
1129 k->num_pages = (efi_md_end(md) - lim)
1130 >> EFI_PAGE_SHIFT;
1131 k++;
1132 }
1133 }
1134 ae = contig_high;
1135 } else
1136 ae = efi_md_end(md);
1137
1138 /* keep within max_addr= and min_addr= command line arg */
1139 as = max(as, min_addr);
1140 ae = min(ae, max_addr);
1141 if (ae <= as)
1142 continue;
1143
1144 /* avoid going over mem= command line arg */
1145 if (total_mem + (ae - as) > mem_limit)
1146 ae -= total_mem + (ae - as) - mem_limit;
1147
1148 if (ae <= as)
1149 continue;
1150 if (prev && kmd_end(prev) == md->phys_addr) {
1151 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1152 total_mem += ae - as;
1153 continue;
1154 }
1155 k->attribute = EFI_MEMORY_WB;
1156 k->start = as;
1157 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1158 total_mem += ae - as;
1159 prev = k++;
1160 }
1161 k->start = ~0L; /* end-marker */
1162
1163 /* reserve the memory we are using for kern_memmap */
1164 *s = (u64)kern_memmap;
1165 *e = (u64)++k;
1166
1167 return total_mem;
1168}
1169
1170void
1171efi_initialize_iomem_resources(struct resource *code_resource,
1172 struct resource *data_resource,
1173 struct resource *bss_resource)
1174{
1175 struct resource *res;
1176 void *efi_map_start, *efi_map_end, *p;
1177 efi_memory_desc_t *md;
1178 u64 efi_desc_size;
1179 char *name;
1180 unsigned long flags, desc;
1181
1182 efi_map_start = __va(ia64_boot_param->efi_memmap);
1183 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1184 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1185
1186 res = NULL;
1187
1188 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1189 md = p;
1190
1191 if (md->num_pages == 0) /* should not happen */
1192 continue;
1193
1194 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1195 desc = IORES_DESC_NONE;
1196
1197 switch (md->type) {
1198
1199 case EFI_MEMORY_MAPPED_IO:
1200 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1201 continue;
1202
1203 case EFI_LOADER_CODE:
1204 case EFI_LOADER_DATA:
1205 case EFI_BOOT_SERVICES_DATA:
1206 case EFI_BOOT_SERVICES_CODE:
1207 case EFI_CONVENTIONAL_MEMORY:
1208 if (md->attribute & EFI_MEMORY_WP) {
1209 name = "System ROM";
1210 flags |= IORESOURCE_READONLY;
1211 } else if (md->attribute == EFI_MEMORY_UC) {
1212 name = "Uncached RAM";
1213 } else {
1214 name = "System RAM";
1215 flags |= IORESOURCE_SYSRAM;
1216 }
1217 break;
1218
1219 case EFI_ACPI_MEMORY_NVS:
1220 name = "ACPI Non-volatile Storage";
1221 desc = IORES_DESC_ACPI_NV_STORAGE;
1222 break;
1223
1224 case EFI_UNUSABLE_MEMORY:
1225 name = "reserved";
1226 flags |= IORESOURCE_DISABLED;
1227 break;
1228
1229 case EFI_PERSISTENT_MEMORY:
1230 name = "Persistent Memory";
1231 desc = IORES_DESC_PERSISTENT_MEMORY;
1232 break;
1233
1234 case EFI_RESERVED_TYPE:
1235 case EFI_RUNTIME_SERVICES_CODE:
1236 case EFI_RUNTIME_SERVICES_DATA:
1237 case EFI_ACPI_RECLAIM_MEMORY:
1238 default:
1239 name = "reserved";
1240 break;
1241 }
1242
1243 if ((res = kzalloc(sizeof(struct resource),
1244 GFP_KERNEL)) == NULL) {
1245 printk(KERN_ERR
1246 "failed to allocate resource for iomem\n");
1247 return;
1248 }
1249
1250 res->name = name;
1251 res->start = md->phys_addr;
1252 res->end = md->phys_addr + efi_md_size(md) - 1;
1253 res->flags = flags;
1254 res->desc = desc;
1255
1256 if (insert_resource(&iomem_resource, res) < 0)
1257 kfree(res);
1258 else {
1259 /*
1260 * We don't know which region contains
1261 * kernel data so we try it repeatedly and
1262 * let the resource manager test it.
1263 */
1264 insert_resource(res, code_resource);
1265 insert_resource(res, data_resource);
1266 insert_resource(res, bss_resource);
1267#ifdef CONFIG_KEXEC
1268 insert_resource(res, &efi_memmap_res);
1269 insert_resource(res, &boot_param_res);
1270 if (crashk_res.end > crashk_res.start)
1271 insert_resource(res, &crashk_res);
1272#endif
1273 }
1274 }
1275}
1276
1277#ifdef CONFIG_KEXEC
1278/* find a block of memory aligned to 64M exclude reserved regions
1279 rsvd_regions are sorted
1280 */
1281unsigned long __init
1282kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1283{
1284 int i;
1285 u64 start, end;
1286 u64 alignment = 1UL << _PAGE_SIZE_64M;
1287 void *efi_map_start, *efi_map_end, *p;
1288 efi_memory_desc_t *md;
1289 u64 efi_desc_size;
1290
1291 efi_map_start = __va(ia64_boot_param->efi_memmap);
1292 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1293 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1294
1295 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1296 md = p;
1297 if (!efi_wb(md))
1298 continue;
1299 start = ALIGN(md->phys_addr, alignment);
1300 end = efi_md_end(md);
1301 for (i = 0; i < n; i++) {
1302 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1303 if (__pa(r[i].start) > start + size)
1304 return start;
1305 start = ALIGN(__pa(r[i].end), alignment);
1306 if (i < n-1 &&
1307 __pa(r[i+1].start) < start + size)
1308 continue;
1309 else
1310 break;
1311 }
1312 }
1313 if (end > start + size)
1314 return start;
1315 }
1316
1317 printk(KERN_WARNING
1318 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1319 return ~0UL;
1320}
1321#endif
1322
1323#ifdef CONFIG_CRASH_DUMP
1324/* locate the size find a the descriptor at a certain address */
1325unsigned long __init
1326vmcore_find_descriptor_size (unsigned long address)
1327{
1328 void *efi_map_start, *efi_map_end, *p;
1329 efi_memory_desc_t *md;
1330 u64 efi_desc_size;
1331 unsigned long ret = 0;
1332
1333 efi_map_start = __va(ia64_boot_param->efi_memmap);
1334 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1335 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1336
1337 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1338 md = p;
1339 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1340 && md->phys_addr == address) {
1341 ret = efi_md_size(md);
1342 break;
1343 }
1344 }
1345
1346 if (ret == 0)
1347 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1348
1349 return ret;
1350}
1351#endif
1352
1353char *efi_systab_show_arch(char *str)
1354{
1355 if (mps_phys != EFI_INVALID_TABLE_ADDR)
1356 str += sprintf(str, "MPS=0x%lx\n", mps_phys);
1357 if (hcdp_phys != EFI_INVALID_TABLE_ADDR)
1358 str += sprintf(str, "HCDP=0x%lx\n", hcdp_phys);
1359 return str;
1360}