Loading...
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/io.h>
38#include <asm/kregs.h>
39#include <asm/meminit.h>
40#include <asm/processor.h>
41#include <asm/mca.h>
42#include <asm/setup.h>
43#include <asm/tlbflush.h>
44
45#define EFI_DEBUG 0
46
47#define ESI_TABLE_GUID \
48 EFI_GUID(0x43EA58DC, 0xCF28, 0x4b06, 0xB3, \
49 0x91, 0xB7, 0x50, 0x59, 0x34, 0x2B, 0xD4)
50
51static unsigned long mps_phys = EFI_INVALID_TABLE_ADDR;
52static __initdata unsigned long palo_phys;
53
54unsigned long __initdata esi_phys = EFI_INVALID_TABLE_ADDR;
55unsigned long hcdp_phys = EFI_INVALID_TABLE_ADDR;
56unsigned long sal_systab_phys = EFI_INVALID_TABLE_ADDR;
57
58static const efi_config_table_type_t arch_tables[] __initconst = {
59 {ESI_TABLE_GUID, &esi_phys, "ESI" },
60 {HCDP_TABLE_GUID, &hcdp_phys, "HCDP" },
61 {MPS_TABLE_GUID, &mps_phys, "MPS" },
62 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, &palo_phys, "PALO" },
63 {SAL_SYSTEM_TABLE_GUID, &sal_systab_phys, "SALsystab" },
64 {},
65};
66
67extern efi_status_t efi_call_phys (void *, ...);
68
69static efi_runtime_services_t *runtime;
70static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
71
72#define efi_call_virt(f, args...) (*(f))(args)
73
74#define STUB_GET_TIME(prefix, adjust_arg) \
75static efi_status_t \
76prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
77{ \
78 struct ia64_fpreg fr[6]; \
79 efi_time_cap_t *atc = NULL; \
80 efi_status_t ret; \
81 \
82 if (tc) \
83 atc = adjust_arg(tc); \
84 ia64_save_scratch_fpregs(fr); \
85 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
86 adjust_arg(tm), atc); \
87 ia64_load_scratch_fpregs(fr); \
88 return ret; \
89}
90
91#define STUB_SET_TIME(prefix, adjust_arg) \
92static efi_status_t \
93prefix##_set_time (efi_time_t *tm) \
94{ \
95 struct ia64_fpreg fr[6]; \
96 efi_status_t ret; \
97 \
98 ia64_save_scratch_fpregs(fr); \
99 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
100 adjust_arg(tm)); \
101 ia64_load_scratch_fpregs(fr); \
102 return ret; \
103}
104
105#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
106static efi_status_t \
107prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
108 efi_time_t *tm) \
109{ \
110 struct ia64_fpreg fr[6]; \
111 efi_status_t ret; \
112 \
113 ia64_save_scratch_fpregs(fr); \
114 ret = efi_call_##prefix( \
115 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
116 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
117 ia64_load_scratch_fpregs(fr); \
118 return ret; \
119}
120
121#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
122static efi_status_t \
123prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
124{ \
125 struct ia64_fpreg fr[6]; \
126 efi_time_t *atm = NULL; \
127 efi_status_t ret; \
128 \
129 if (tm) \
130 atm = adjust_arg(tm); \
131 ia64_save_scratch_fpregs(fr); \
132 ret = efi_call_##prefix( \
133 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
134 enabled, atm); \
135 ia64_load_scratch_fpregs(fr); \
136 return ret; \
137}
138
139#define STUB_GET_VARIABLE(prefix, adjust_arg) \
140static efi_status_t \
141prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
142 unsigned long *data_size, void *data) \
143{ \
144 struct ia64_fpreg fr[6]; \
145 u32 *aattr = NULL; \
146 efi_status_t ret; \
147 \
148 if (attr) \
149 aattr = adjust_arg(attr); \
150 ia64_save_scratch_fpregs(fr); \
151 ret = efi_call_##prefix( \
152 (efi_get_variable_t *) __va(runtime->get_variable), \
153 adjust_arg(name), adjust_arg(vendor), aattr, \
154 adjust_arg(data_size), adjust_arg(data)); \
155 ia64_load_scratch_fpregs(fr); \
156 return ret; \
157}
158
159#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
160static efi_status_t \
161prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
162 efi_guid_t *vendor) \
163{ \
164 struct ia64_fpreg fr[6]; \
165 efi_status_t ret; \
166 \
167 ia64_save_scratch_fpregs(fr); \
168 ret = efi_call_##prefix( \
169 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
170 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
171 ia64_load_scratch_fpregs(fr); \
172 return ret; \
173}
174
175#define STUB_SET_VARIABLE(prefix, adjust_arg) \
176static efi_status_t \
177prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
178 u32 attr, unsigned long data_size, \
179 void *data) \
180{ \
181 struct ia64_fpreg fr[6]; \
182 efi_status_t ret; \
183 \
184 ia64_save_scratch_fpregs(fr); \
185 ret = efi_call_##prefix( \
186 (efi_set_variable_t *) __va(runtime->set_variable), \
187 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
188 adjust_arg(data)); \
189 ia64_load_scratch_fpregs(fr); \
190 return ret; \
191}
192
193#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
194static efi_status_t \
195prefix##_get_next_high_mono_count (u32 *count) \
196{ \
197 struct ia64_fpreg fr[6]; \
198 efi_status_t ret; \
199 \
200 ia64_save_scratch_fpregs(fr); \
201 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
202 __va(runtime->get_next_high_mono_count), \
203 adjust_arg(count)); \
204 ia64_load_scratch_fpregs(fr); \
205 return ret; \
206}
207
208#define STUB_RESET_SYSTEM(prefix, adjust_arg) \
209static void \
210prefix##_reset_system (int reset_type, efi_status_t status, \
211 unsigned long data_size, efi_char16_t *data) \
212{ \
213 struct ia64_fpreg fr[6]; \
214 efi_char16_t *adata = NULL; \
215 \
216 if (data) \
217 adata = adjust_arg(data); \
218 \
219 ia64_save_scratch_fpregs(fr); \
220 efi_call_##prefix( \
221 (efi_reset_system_t *) __va(runtime->reset_system), \
222 reset_type, status, data_size, adata); \
223 /* should not return, but just in case... */ \
224 ia64_load_scratch_fpregs(fr); \
225}
226
227#define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
228
229STUB_GET_TIME(phys, phys_ptr)
230STUB_SET_TIME(phys, phys_ptr)
231STUB_GET_WAKEUP_TIME(phys, phys_ptr)
232STUB_SET_WAKEUP_TIME(phys, phys_ptr)
233STUB_GET_VARIABLE(phys, phys_ptr)
234STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
235STUB_SET_VARIABLE(phys, phys_ptr)
236STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
237STUB_RESET_SYSTEM(phys, phys_ptr)
238
239#define id(arg) arg
240
241STUB_GET_TIME(virt, id)
242STUB_SET_TIME(virt, id)
243STUB_GET_WAKEUP_TIME(virt, id)
244STUB_SET_WAKEUP_TIME(virt, id)
245STUB_GET_VARIABLE(virt, id)
246STUB_GET_NEXT_VARIABLE(virt, id)
247STUB_SET_VARIABLE(virt, id)
248STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
249STUB_RESET_SYSTEM(virt, id)
250
251void
252efi_gettimeofday (struct timespec64 *ts)
253{
254 efi_time_t tm;
255
256 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
257 memset(ts, 0, sizeof(*ts));
258 return;
259 }
260
261 ts->tv_sec = mktime64(tm.year, tm.month, tm.day,
262 tm.hour, tm.minute, tm.second);
263 ts->tv_nsec = tm.nanosecond;
264}
265
266static int
267is_memory_available (efi_memory_desc_t *md)
268{
269 if (!(md->attribute & EFI_MEMORY_WB))
270 return 0;
271
272 switch (md->type) {
273 case EFI_LOADER_CODE:
274 case EFI_LOADER_DATA:
275 case EFI_BOOT_SERVICES_CODE:
276 case EFI_BOOT_SERVICES_DATA:
277 case EFI_CONVENTIONAL_MEMORY:
278 return 1;
279 }
280 return 0;
281}
282
283typedef struct kern_memdesc {
284 u64 attribute;
285 u64 start;
286 u64 num_pages;
287} kern_memdesc_t;
288
289static kern_memdesc_t *kern_memmap;
290
291#define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
292
293static inline u64
294kmd_end(kern_memdesc_t *kmd)
295{
296 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
297}
298
299static inline u64
300efi_md_end(efi_memory_desc_t *md)
301{
302 return (md->phys_addr + efi_md_size(md));
303}
304
305static inline int
306efi_wb(efi_memory_desc_t *md)
307{
308 return (md->attribute & EFI_MEMORY_WB);
309}
310
311static inline int
312efi_uc(efi_memory_desc_t *md)
313{
314 return (md->attribute & EFI_MEMORY_UC);
315}
316
317static void
318walk (efi_freemem_callback_t callback, void *arg, u64 attr)
319{
320 kern_memdesc_t *k;
321 u64 start, end, voff;
322
323 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
324 for (k = kern_memmap; k->start != ~0UL; k++) {
325 if (k->attribute != attr)
326 continue;
327 start = PAGE_ALIGN(k->start);
328 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
329 if (start < end)
330 if ((*callback)(start + voff, end + voff, arg) < 0)
331 return;
332 }
333}
334
335/*
336 * Walk the EFI memory map and call CALLBACK once for each EFI memory
337 * descriptor that has memory that is available for OS use.
338 */
339void
340efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
341{
342 walk(callback, arg, EFI_MEMORY_WB);
343}
344
345/*
346 * Walk the EFI memory map and call CALLBACK once for each EFI memory
347 * descriptor that has memory that is available for uncached allocator.
348 */
349void
350efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
351{
352 walk(callback, arg, EFI_MEMORY_UC);
353}
354
355/*
356 * Look for the PAL_CODE region reported by EFI and map it using an
357 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
358 * Abstraction Layer chapter 11 in ADAG
359 */
360void *
361efi_get_pal_addr (void)
362{
363 void *efi_map_start, *efi_map_end, *p;
364 efi_memory_desc_t *md;
365 u64 efi_desc_size;
366 int pal_code_count = 0;
367 u64 vaddr, mask;
368
369 efi_map_start = __va(ia64_boot_param->efi_memmap);
370 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
371 efi_desc_size = ia64_boot_param->efi_memdesc_size;
372
373 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
374 md = p;
375 if (md->type != EFI_PAL_CODE)
376 continue;
377
378 if (++pal_code_count > 1) {
379 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
380 "dropped @ %llx\n", md->phys_addr);
381 continue;
382 }
383 /*
384 * The only ITLB entry in region 7 that is used is the one
385 * installed by __start(). That entry covers a 64MB range.
386 */
387 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
388 vaddr = PAGE_OFFSET + md->phys_addr;
389
390 /*
391 * We must check that the PAL mapping won't overlap with the
392 * kernel mapping.
393 *
394 * PAL code is guaranteed to be aligned on a power of 2 between
395 * 4k and 256KB and that only one ITR is needed to map it. This
396 * implies that the PAL code is always aligned on its size,
397 * i.e., the closest matching page size supported by the TLB.
398 * Therefore PAL code is guaranteed never to cross a 64MB unless
399 * it is bigger than 64MB (very unlikely!). So for now the
400 * following test is enough to determine whether or not we need
401 * a dedicated ITR for the PAL code.
402 */
403 if ((vaddr & mask) == (KERNEL_START & mask)) {
404 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
405 __func__);
406 continue;
407 }
408
409 if (efi_md_size(md) > IA64_GRANULE_SIZE)
410 panic("Whoa! PAL code size bigger than a granule!");
411
412#if EFI_DEBUG
413 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
414
415 printk(KERN_INFO "CPU %d: mapping PAL code "
416 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
417 smp_processor_id(), md->phys_addr,
418 md->phys_addr + efi_md_size(md),
419 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
420#endif
421 return __va(md->phys_addr);
422 }
423 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
424 __func__);
425 return NULL;
426}
427
428
429static u8 __init palo_checksum(u8 *buffer, u32 length)
430{
431 u8 sum = 0;
432 u8 *end = buffer + length;
433
434 while (buffer < end)
435 sum = (u8) (sum + *(buffer++));
436
437 return sum;
438}
439
440/*
441 * Parse and handle PALO table which is published at:
442 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
443 */
444static void __init handle_palo(unsigned long phys_addr)
445{
446 struct palo_table *palo = __va(phys_addr);
447 u8 checksum;
448
449 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
450 printk(KERN_INFO "PALO signature incorrect.\n");
451 return;
452 }
453
454 checksum = palo_checksum((u8 *)palo, palo->length);
455 if (checksum) {
456 printk(KERN_INFO "PALO checksum incorrect.\n");
457 return;
458 }
459
460 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
461}
462
463void
464efi_map_pal_code (void)
465{
466 void *pal_vaddr = efi_get_pal_addr ();
467 u64 psr;
468
469 if (!pal_vaddr)
470 return;
471
472 /*
473 * Cannot write to CRx with PSR.ic=1
474 */
475 psr = ia64_clear_ic();
476 ia64_itr(0x1, IA64_TR_PALCODE,
477 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
478 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
479 IA64_GRANULE_SHIFT);
480 ia64_set_psr(psr); /* restore psr */
481}
482
483void __init
484efi_init (void)
485{
486 const efi_system_table_t *efi_systab;
487 void *efi_map_start, *efi_map_end;
488 u64 efi_desc_size;
489 char *cp;
490
491 set_bit(EFI_BOOT, &efi.flags);
492 set_bit(EFI_64BIT, &efi.flags);
493
494 /*
495 * It's too early to be able to use the standard kernel command line
496 * support...
497 */
498 for (cp = boot_command_line; *cp; ) {
499 if (memcmp(cp, "mem=", 4) == 0) {
500 mem_limit = memparse(cp + 4, &cp);
501 } else if (memcmp(cp, "max_addr=", 9) == 0) {
502 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
503 } else if (memcmp(cp, "min_addr=", 9) == 0) {
504 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
505 } else {
506 while (*cp != ' ' && *cp)
507 ++cp;
508 while (*cp == ' ')
509 ++cp;
510 }
511 }
512 if (min_addr != 0UL)
513 printk(KERN_INFO "Ignoring memory below %lluMB\n",
514 min_addr >> 20);
515 if (max_addr != ~0UL)
516 printk(KERN_INFO "Ignoring memory above %lluMB\n",
517 max_addr >> 20);
518
519 efi_systab = __va(ia64_boot_param->efi_systab);
520
521 /*
522 * Verify the EFI Table
523 */
524 if (efi_systab == NULL)
525 panic("Whoa! Can't find EFI system table.\n");
526 if (efi_systab_check_header(&efi_systab->hdr, 1))
527 panic("Whoa! EFI system table signature incorrect\n");
528
529 efi_systab_report_header(&efi_systab->hdr, efi_systab->fw_vendor);
530
531 palo_phys = EFI_INVALID_TABLE_ADDR;
532
533 if (efi_config_parse_tables(__va(efi_systab->tables),
534 efi_systab->nr_tables,
535 arch_tables) != 0)
536 return;
537
538 if (palo_phys != EFI_INVALID_TABLE_ADDR)
539 handle_palo(palo_phys);
540
541 runtime = __va(efi_systab->runtime);
542 efi.get_time = phys_get_time;
543 efi.set_time = phys_set_time;
544 efi.get_wakeup_time = phys_get_wakeup_time;
545 efi.set_wakeup_time = phys_set_wakeup_time;
546 efi.get_variable = phys_get_variable;
547 efi.get_next_variable = phys_get_next_variable;
548 efi.set_variable = phys_set_variable;
549 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
550 efi.reset_system = phys_reset_system;
551
552 efi_map_start = __va(ia64_boot_param->efi_memmap);
553 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
554 efi_desc_size = ia64_boot_param->efi_memdesc_size;
555
556#if EFI_DEBUG
557 /* print EFI memory map: */
558 {
559 efi_memory_desc_t *md;
560 void *p;
561
562 for (i = 0, p = efi_map_start; p < efi_map_end;
563 ++i, p += efi_desc_size)
564 {
565 const char *unit;
566 unsigned long size;
567 char buf[64];
568
569 md = p;
570 size = md->num_pages << EFI_PAGE_SHIFT;
571
572 if ((size >> 40) > 0) {
573 size >>= 40;
574 unit = "TB";
575 } else if ((size >> 30) > 0) {
576 size >>= 30;
577 unit = "GB";
578 } else if ((size >> 20) > 0) {
579 size >>= 20;
580 unit = "MB";
581 } else {
582 size >>= 10;
583 unit = "KB";
584 }
585
586 printk("mem%02d: %s "
587 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
588 i, efi_md_typeattr_format(buf, sizeof(buf), md),
589 md->phys_addr,
590 md->phys_addr + efi_md_size(md), size, unit);
591 }
592 }
593#endif
594
595 efi_map_pal_code();
596 efi_enter_virtual_mode();
597}
598
599void
600efi_enter_virtual_mode (void)
601{
602 void *efi_map_start, *efi_map_end, *p;
603 efi_memory_desc_t *md;
604 efi_status_t status;
605 u64 efi_desc_size;
606
607 efi_map_start = __va(ia64_boot_param->efi_memmap);
608 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
609 efi_desc_size = ia64_boot_param->efi_memdesc_size;
610
611 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
612 md = p;
613 if (md->attribute & EFI_MEMORY_RUNTIME) {
614 /*
615 * Some descriptors have multiple bits set, so the
616 * order of the tests is relevant.
617 */
618 if (md->attribute & EFI_MEMORY_WB) {
619 md->virt_addr = (u64) __va(md->phys_addr);
620 } else if (md->attribute & EFI_MEMORY_UC) {
621 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
622 } else if (md->attribute & EFI_MEMORY_WC) {
623#if 0
624 md->virt_addr = ia64_remap(md->phys_addr,
625 (_PAGE_A |
626 _PAGE_P |
627 _PAGE_D |
628 _PAGE_MA_WC |
629 _PAGE_PL_0 |
630 _PAGE_AR_RW));
631#else
632 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
633 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
634#endif
635 } else if (md->attribute & EFI_MEMORY_WT) {
636#if 0
637 md->virt_addr = ia64_remap(md->phys_addr,
638 (_PAGE_A |
639 _PAGE_P |
640 _PAGE_D |
641 _PAGE_MA_WT |
642 _PAGE_PL_0 |
643 _PAGE_AR_RW));
644#else
645 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
646 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
647#endif
648 }
649 }
650 }
651
652 status = efi_call_phys(__va(runtime->set_virtual_address_map),
653 ia64_boot_param->efi_memmap_size,
654 efi_desc_size,
655 ia64_boot_param->efi_memdesc_version,
656 ia64_boot_param->efi_memmap);
657 if (status != EFI_SUCCESS) {
658 printk(KERN_WARNING "warning: unable to switch EFI into "
659 "virtual mode (status=%lu)\n", status);
660 return;
661 }
662
663 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
664
665 /*
666 * Now that EFI is in virtual mode, we call the EFI functions more
667 * efficiently:
668 */
669 efi.get_time = virt_get_time;
670 efi.set_time = virt_set_time;
671 efi.get_wakeup_time = virt_get_wakeup_time;
672 efi.set_wakeup_time = virt_set_wakeup_time;
673 efi.get_variable = virt_get_variable;
674 efi.get_next_variable = virt_get_next_variable;
675 efi.set_variable = virt_set_variable;
676 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
677 efi.reset_system = virt_reset_system;
678}
679
680/*
681 * Walk the EFI memory map looking for the I/O port range. There can only be
682 * one entry of this type, other I/O port ranges should be described via ACPI.
683 */
684u64
685efi_get_iobase (void)
686{
687 void *efi_map_start, *efi_map_end, *p;
688 efi_memory_desc_t *md;
689 u64 efi_desc_size;
690
691 efi_map_start = __va(ia64_boot_param->efi_memmap);
692 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
693 efi_desc_size = ia64_boot_param->efi_memdesc_size;
694
695 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
696 md = p;
697 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
698 if (md->attribute & EFI_MEMORY_UC)
699 return md->phys_addr;
700 }
701 }
702 return 0;
703}
704
705static struct kern_memdesc *
706kern_memory_descriptor (unsigned long phys_addr)
707{
708 struct kern_memdesc *md;
709
710 for (md = kern_memmap; md->start != ~0UL; md++) {
711 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
712 return md;
713 }
714 return NULL;
715}
716
717static efi_memory_desc_t *
718efi_memory_descriptor (unsigned long phys_addr)
719{
720 void *efi_map_start, *efi_map_end, *p;
721 efi_memory_desc_t *md;
722 u64 efi_desc_size;
723
724 efi_map_start = __va(ia64_boot_param->efi_memmap);
725 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
726 efi_desc_size = ia64_boot_param->efi_memdesc_size;
727
728 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
729 md = p;
730
731 if (phys_addr - md->phys_addr < efi_md_size(md))
732 return md;
733 }
734 return NULL;
735}
736
737static int
738efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
739{
740 void *efi_map_start, *efi_map_end, *p;
741 efi_memory_desc_t *md;
742 u64 efi_desc_size;
743 unsigned long end;
744
745 efi_map_start = __va(ia64_boot_param->efi_memmap);
746 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
747 efi_desc_size = ia64_boot_param->efi_memdesc_size;
748
749 end = phys_addr + size;
750
751 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
752 md = p;
753 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
754 return 1;
755 }
756 return 0;
757}
758
759int
760efi_mem_type (unsigned long phys_addr)
761{
762 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
763
764 if (md)
765 return md->type;
766 return -EINVAL;
767}
768
769u64
770efi_mem_attributes (unsigned long phys_addr)
771{
772 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
773
774 if (md)
775 return md->attribute;
776 return 0;
777}
778EXPORT_SYMBOL(efi_mem_attributes);
779
780u64
781efi_mem_attribute (unsigned long phys_addr, unsigned long size)
782{
783 unsigned long end = phys_addr + size;
784 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
785 u64 attr;
786
787 if (!md)
788 return 0;
789
790 /*
791 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
792 * the kernel that firmware needs this region mapped.
793 */
794 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
795 do {
796 unsigned long md_end = efi_md_end(md);
797
798 if (end <= md_end)
799 return attr;
800
801 md = efi_memory_descriptor(md_end);
802 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
803 return 0;
804 } while (md);
805 return 0; /* never reached */
806}
807
808u64
809kern_mem_attribute (unsigned long phys_addr, unsigned long size)
810{
811 unsigned long end = phys_addr + size;
812 struct kern_memdesc *md;
813 u64 attr;
814
815 /*
816 * This is a hack for ioremap calls before we set up kern_memmap.
817 * Maybe we should do efi_memmap_init() earlier instead.
818 */
819 if (!kern_memmap) {
820 attr = efi_mem_attribute(phys_addr, size);
821 if (attr & EFI_MEMORY_WB)
822 return EFI_MEMORY_WB;
823 return 0;
824 }
825
826 md = kern_memory_descriptor(phys_addr);
827 if (!md)
828 return 0;
829
830 attr = md->attribute;
831 do {
832 unsigned long md_end = kmd_end(md);
833
834 if (end <= md_end)
835 return attr;
836
837 md = kern_memory_descriptor(md_end);
838 if (!md || md->attribute != attr)
839 return 0;
840 } while (md);
841 return 0; /* never reached */
842}
843
844int
845valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
846{
847 u64 attr;
848
849 /*
850 * /dev/mem reads and writes use copy_to_user(), which implicitly
851 * uses a granule-sized kernel identity mapping. It's really
852 * only safe to do this for regions in kern_memmap. For more
853 * details, see Documentation/ia64/aliasing.rst.
854 */
855 attr = kern_mem_attribute(phys_addr, size);
856 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
857 return 1;
858 return 0;
859}
860
861int
862valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
863{
864 unsigned long phys_addr = pfn << PAGE_SHIFT;
865 u64 attr;
866
867 attr = efi_mem_attribute(phys_addr, size);
868
869 /*
870 * /dev/mem mmap uses normal user pages, so we don't need the entire
871 * granule, but the entire region we're mapping must support the same
872 * attribute.
873 */
874 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
875 return 1;
876
877 /*
878 * Intel firmware doesn't tell us about all the MMIO regions, so
879 * in general we have to allow mmap requests. But if EFI *does*
880 * tell us about anything inside this region, we should deny it.
881 * The user can always map a smaller region to avoid the overlap.
882 */
883 if (efi_memmap_intersects(phys_addr, size))
884 return 0;
885
886 return 1;
887}
888
889pgprot_t
890phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
891 pgprot_t vma_prot)
892{
893 unsigned long phys_addr = pfn << PAGE_SHIFT;
894 u64 attr;
895
896 /*
897 * For /dev/mem mmap, we use user mappings, but if the region is
898 * in kern_memmap (and hence may be covered by a kernel mapping),
899 * we must use the same attribute as the kernel mapping.
900 */
901 attr = kern_mem_attribute(phys_addr, size);
902 if (attr & EFI_MEMORY_WB)
903 return pgprot_cacheable(vma_prot);
904 else if (attr & EFI_MEMORY_UC)
905 return pgprot_noncached(vma_prot);
906
907 /*
908 * Some chipsets don't support UC access to memory. If
909 * WB is supported, we prefer that.
910 */
911 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
912 return pgprot_cacheable(vma_prot);
913
914 return pgprot_noncached(vma_prot);
915}
916
917int __init
918efi_uart_console_only(void)
919{
920 efi_status_t status;
921 char *s, name[] = "ConOut";
922 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
923 efi_char16_t *utf16, name_utf16[32];
924 unsigned char data[1024];
925 unsigned long size = sizeof(data);
926 struct efi_generic_dev_path *hdr, *end_addr;
927 int uart = 0;
928
929 /* Convert to UTF-16 */
930 utf16 = name_utf16;
931 s = name;
932 while (*s)
933 *utf16++ = *s++ & 0x7f;
934 *utf16 = 0;
935
936 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
937 if (status != EFI_SUCCESS) {
938 printk(KERN_ERR "No EFI %s variable?\n", name);
939 return 0;
940 }
941
942 hdr = (struct efi_generic_dev_path *) data;
943 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
944 while (hdr < end_addr) {
945 if (hdr->type == EFI_DEV_MSG &&
946 hdr->sub_type == EFI_DEV_MSG_UART)
947 uart = 1;
948 else if (hdr->type == EFI_DEV_END_PATH ||
949 hdr->type == EFI_DEV_END_PATH2) {
950 if (!uart)
951 return 0;
952 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
953 return 1;
954 uart = 0;
955 }
956 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
957 }
958 printk(KERN_ERR "Malformed %s value\n", name);
959 return 0;
960}
961
962/*
963 * Look for the first granule aligned memory descriptor memory
964 * that is big enough to hold EFI memory map. Make sure this
965 * descriptor is at least granule sized so it does not get trimmed
966 */
967struct kern_memdesc *
968find_memmap_space (void)
969{
970 u64 contig_low=0, contig_high=0;
971 u64 as = 0, ae;
972 void *efi_map_start, *efi_map_end, *p, *q;
973 efi_memory_desc_t *md, *pmd = NULL, *check_md;
974 u64 space_needed, efi_desc_size;
975 unsigned long total_mem = 0;
976
977 efi_map_start = __va(ia64_boot_param->efi_memmap);
978 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
979 efi_desc_size = ia64_boot_param->efi_memdesc_size;
980
981 /*
982 * Worst case: we need 3 kernel descriptors for each efi descriptor
983 * (if every entry has a WB part in the middle, and UC head and tail),
984 * plus one for the end marker.
985 */
986 space_needed = sizeof(kern_memdesc_t) *
987 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
988
989 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
990 md = p;
991 if (!efi_wb(md)) {
992 continue;
993 }
994 if (pmd == NULL || !efi_wb(pmd) ||
995 efi_md_end(pmd) != md->phys_addr) {
996 contig_low = GRANULEROUNDUP(md->phys_addr);
997 contig_high = efi_md_end(md);
998 for (q = p + efi_desc_size; q < efi_map_end;
999 q += efi_desc_size) {
1000 check_md = q;
1001 if (!efi_wb(check_md))
1002 break;
1003 if (contig_high != check_md->phys_addr)
1004 break;
1005 contig_high = efi_md_end(check_md);
1006 }
1007 contig_high = GRANULEROUNDDOWN(contig_high);
1008 }
1009 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1010 continue;
1011
1012 /* Round ends inward to granule boundaries */
1013 as = max(contig_low, md->phys_addr);
1014 ae = min(contig_high, efi_md_end(md));
1015
1016 /* keep within max_addr= and min_addr= command line arg */
1017 as = max(as, min_addr);
1018 ae = min(ae, max_addr);
1019 if (ae <= as)
1020 continue;
1021
1022 /* avoid going over mem= command line arg */
1023 if (total_mem + (ae - as) > mem_limit)
1024 ae -= total_mem + (ae - as) - mem_limit;
1025
1026 if (ae <= as)
1027 continue;
1028
1029 if (ae - as > space_needed)
1030 break;
1031 }
1032 if (p >= efi_map_end)
1033 panic("Can't allocate space for kernel memory descriptors");
1034
1035 return __va(as);
1036}
1037
1038/*
1039 * Walk the EFI memory map and gather all memory available for kernel
1040 * to use. We can allocate partial granules only if the unavailable
1041 * parts exist, and are WB.
1042 */
1043unsigned long
1044efi_memmap_init(u64 *s, u64 *e)
1045{
1046 struct kern_memdesc *k, *prev = NULL;
1047 u64 contig_low=0, contig_high=0;
1048 u64 as, ae, lim;
1049 void *efi_map_start, *efi_map_end, *p, *q;
1050 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1051 u64 efi_desc_size;
1052 unsigned long total_mem = 0;
1053
1054 k = kern_memmap = find_memmap_space();
1055
1056 efi_map_start = __va(ia64_boot_param->efi_memmap);
1057 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1058 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1059
1060 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1061 md = p;
1062 if (!efi_wb(md)) {
1063 if (efi_uc(md) &&
1064 (md->type == EFI_CONVENTIONAL_MEMORY ||
1065 md->type == EFI_BOOT_SERVICES_DATA)) {
1066 k->attribute = EFI_MEMORY_UC;
1067 k->start = md->phys_addr;
1068 k->num_pages = md->num_pages;
1069 k++;
1070 }
1071 continue;
1072 }
1073 if (pmd == NULL || !efi_wb(pmd) ||
1074 efi_md_end(pmd) != md->phys_addr) {
1075 contig_low = GRANULEROUNDUP(md->phys_addr);
1076 contig_high = efi_md_end(md);
1077 for (q = p + efi_desc_size; q < efi_map_end;
1078 q += efi_desc_size) {
1079 check_md = q;
1080 if (!efi_wb(check_md))
1081 break;
1082 if (contig_high != check_md->phys_addr)
1083 break;
1084 contig_high = efi_md_end(check_md);
1085 }
1086 contig_high = GRANULEROUNDDOWN(contig_high);
1087 }
1088 if (!is_memory_available(md))
1089 continue;
1090
1091 /*
1092 * Round ends inward to granule boundaries
1093 * Give trimmings to uncached allocator
1094 */
1095 if (md->phys_addr < contig_low) {
1096 lim = min(efi_md_end(md), contig_low);
1097 if (efi_uc(md)) {
1098 if (k > kern_memmap &&
1099 (k-1)->attribute == EFI_MEMORY_UC &&
1100 kmd_end(k-1) == md->phys_addr) {
1101 (k-1)->num_pages +=
1102 (lim - md->phys_addr)
1103 >> EFI_PAGE_SHIFT;
1104 } else {
1105 k->attribute = EFI_MEMORY_UC;
1106 k->start = md->phys_addr;
1107 k->num_pages = (lim - md->phys_addr)
1108 >> EFI_PAGE_SHIFT;
1109 k++;
1110 }
1111 }
1112 as = contig_low;
1113 } else
1114 as = md->phys_addr;
1115
1116 if (efi_md_end(md) > contig_high) {
1117 lim = max(md->phys_addr, contig_high);
1118 if (efi_uc(md)) {
1119 if (lim == md->phys_addr && k > kern_memmap &&
1120 (k-1)->attribute == EFI_MEMORY_UC &&
1121 kmd_end(k-1) == md->phys_addr) {
1122 (k-1)->num_pages += md->num_pages;
1123 } else {
1124 k->attribute = EFI_MEMORY_UC;
1125 k->start = lim;
1126 k->num_pages = (efi_md_end(md) - lim)
1127 >> EFI_PAGE_SHIFT;
1128 k++;
1129 }
1130 }
1131 ae = contig_high;
1132 } else
1133 ae = efi_md_end(md);
1134
1135 /* keep within max_addr= and min_addr= command line arg */
1136 as = max(as, min_addr);
1137 ae = min(ae, max_addr);
1138 if (ae <= as)
1139 continue;
1140
1141 /* avoid going over mem= command line arg */
1142 if (total_mem + (ae - as) > mem_limit)
1143 ae -= total_mem + (ae - as) - mem_limit;
1144
1145 if (ae <= as)
1146 continue;
1147 if (prev && kmd_end(prev) == md->phys_addr) {
1148 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1149 total_mem += ae - as;
1150 continue;
1151 }
1152 k->attribute = EFI_MEMORY_WB;
1153 k->start = as;
1154 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1155 total_mem += ae - as;
1156 prev = k++;
1157 }
1158 k->start = ~0L; /* end-marker */
1159
1160 /* reserve the memory we are using for kern_memmap */
1161 *s = (u64)kern_memmap;
1162 *e = (u64)++k;
1163
1164 return total_mem;
1165}
1166
1167void
1168efi_initialize_iomem_resources(struct resource *code_resource,
1169 struct resource *data_resource,
1170 struct resource *bss_resource)
1171{
1172 struct resource *res;
1173 void *efi_map_start, *efi_map_end, *p;
1174 efi_memory_desc_t *md;
1175 u64 efi_desc_size;
1176 char *name;
1177 unsigned long flags, desc;
1178
1179 efi_map_start = __va(ia64_boot_param->efi_memmap);
1180 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1181 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1182
1183 res = NULL;
1184
1185 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1186 md = p;
1187
1188 if (md->num_pages == 0) /* should not happen */
1189 continue;
1190
1191 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1192 desc = IORES_DESC_NONE;
1193
1194 switch (md->type) {
1195
1196 case EFI_MEMORY_MAPPED_IO:
1197 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1198 continue;
1199
1200 case EFI_LOADER_CODE:
1201 case EFI_LOADER_DATA:
1202 case EFI_BOOT_SERVICES_DATA:
1203 case EFI_BOOT_SERVICES_CODE:
1204 case EFI_CONVENTIONAL_MEMORY:
1205 if (md->attribute & EFI_MEMORY_WP) {
1206 name = "System ROM";
1207 flags |= IORESOURCE_READONLY;
1208 } else if (md->attribute == EFI_MEMORY_UC) {
1209 name = "Uncached RAM";
1210 } else {
1211 name = "System RAM";
1212 flags |= IORESOURCE_SYSRAM;
1213 }
1214 break;
1215
1216 case EFI_ACPI_MEMORY_NVS:
1217 name = "ACPI Non-volatile Storage";
1218 desc = IORES_DESC_ACPI_NV_STORAGE;
1219 break;
1220
1221 case EFI_UNUSABLE_MEMORY:
1222 name = "reserved";
1223 flags |= IORESOURCE_DISABLED;
1224 break;
1225
1226 case EFI_PERSISTENT_MEMORY:
1227 name = "Persistent Memory";
1228 desc = IORES_DESC_PERSISTENT_MEMORY;
1229 break;
1230
1231 case EFI_RESERVED_TYPE:
1232 case EFI_RUNTIME_SERVICES_CODE:
1233 case EFI_RUNTIME_SERVICES_DATA:
1234 case EFI_ACPI_RECLAIM_MEMORY:
1235 default:
1236 name = "reserved";
1237 break;
1238 }
1239
1240 if ((res = kzalloc(sizeof(struct resource),
1241 GFP_KERNEL)) == NULL) {
1242 printk(KERN_ERR
1243 "failed to allocate resource for iomem\n");
1244 return;
1245 }
1246
1247 res->name = name;
1248 res->start = md->phys_addr;
1249 res->end = md->phys_addr + efi_md_size(md) - 1;
1250 res->flags = flags;
1251 res->desc = desc;
1252
1253 if (insert_resource(&iomem_resource, res) < 0)
1254 kfree(res);
1255 else {
1256 /*
1257 * We don't know which region contains
1258 * kernel data so we try it repeatedly and
1259 * let the resource manager test it.
1260 */
1261 insert_resource(res, code_resource);
1262 insert_resource(res, data_resource);
1263 insert_resource(res, bss_resource);
1264#ifdef CONFIG_KEXEC
1265 insert_resource(res, &efi_memmap_res);
1266 insert_resource(res, &boot_param_res);
1267 if (crashk_res.end > crashk_res.start)
1268 insert_resource(res, &crashk_res);
1269#endif
1270 }
1271 }
1272}
1273
1274#ifdef CONFIG_KEXEC
1275/* find a block of memory aligned to 64M exclude reserved regions
1276 rsvd_regions are sorted
1277 */
1278unsigned long __init
1279kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1280{
1281 int i;
1282 u64 start, end;
1283 u64 alignment = 1UL << _PAGE_SIZE_64M;
1284 void *efi_map_start, *efi_map_end, *p;
1285 efi_memory_desc_t *md;
1286 u64 efi_desc_size;
1287
1288 efi_map_start = __va(ia64_boot_param->efi_memmap);
1289 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1290 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1291
1292 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1293 md = p;
1294 if (!efi_wb(md))
1295 continue;
1296 start = ALIGN(md->phys_addr, alignment);
1297 end = efi_md_end(md);
1298 for (i = 0; i < n; i++) {
1299 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1300 if (__pa(r[i].start) > start + size)
1301 return start;
1302 start = ALIGN(__pa(r[i].end), alignment);
1303 if (i < n-1 &&
1304 __pa(r[i+1].start) < start + size)
1305 continue;
1306 else
1307 break;
1308 }
1309 }
1310 if (end > start + size)
1311 return start;
1312 }
1313
1314 printk(KERN_WARNING
1315 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1316 return ~0UL;
1317}
1318#endif
1319
1320#ifdef CONFIG_CRASH_DUMP
1321/* locate the size find a the descriptor at a certain address */
1322unsigned long __init
1323vmcore_find_descriptor_size (unsigned long address)
1324{
1325 void *efi_map_start, *efi_map_end, *p;
1326 efi_memory_desc_t *md;
1327 u64 efi_desc_size;
1328 unsigned long ret = 0;
1329
1330 efi_map_start = __va(ia64_boot_param->efi_memmap);
1331 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1332 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1333
1334 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1335 md = p;
1336 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1337 && md->phys_addr == address) {
1338 ret = efi_md_size(md);
1339 break;
1340 }
1341 }
1342
1343 if (ret == 0)
1344 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1345
1346 return ret;
1347}
1348#endif
1349
1350char *efi_systab_show_arch(char *str)
1351{
1352 if (mps_phys != EFI_INVALID_TABLE_ADDR)
1353 str += sprintf(str, "MPS=0x%lx\n", mps_phys);
1354 if (hcdp_phys != EFI_INVALID_TABLE_ADDR)
1355 str += sprintf(str, "HCDP=0x%lx\n", hcdp_phys);
1356 return str;
1357}
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/io.h>
38#include <asm/kregs.h>
39#include <asm/meminit.h>
40#include <asm/pgtable.h>
41#include <asm/processor.h>
42#include <asm/mca.h>
43#include <asm/setup.h>
44#include <asm/tlbflush.h>
45
46#define EFI_DEBUG 0
47
48static __initdata unsigned long palo_phys;
49
50unsigned long sal_systab_phys = EFI_INVALID_TABLE_ADDR;
51
52static __initdata efi_config_table_type_t arch_tables[] = {
53 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, "PALO", &palo_phys},
54 {SAL_SYSTEM_TABLE_GUID, "SALsystab", &sal_systab_phys},
55 {NULL_GUID, NULL, 0},
56};
57
58extern efi_status_t efi_call_phys (void *, ...);
59
60static efi_runtime_services_t *runtime;
61static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
62
63#define efi_call_virt(f, args...) (*(f))(args)
64
65#define STUB_GET_TIME(prefix, adjust_arg) \
66static efi_status_t \
67prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
68{ \
69 struct ia64_fpreg fr[6]; \
70 efi_time_cap_t *atc = NULL; \
71 efi_status_t ret; \
72 \
73 if (tc) \
74 atc = adjust_arg(tc); \
75 ia64_save_scratch_fpregs(fr); \
76 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
77 adjust_arg(tm), atc); \
78 ia64_load_scratch_fpregs(fr); \
79 return ret; \
80}
81
82#define STUB_SET_TIME(prefix, adjust_arg) \
83static efi_status_t \
84prefix##_set_time (efi_time_t *tm) \
85{ \
86 struct ia64_fpreg fr[6]; \
87 efi_status_t ret; \
88 \
89 ia64_save_scratch_fpregs(fr); \
90 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
91 adjust_arg(tm)); \
92 ia64_load_scratch_fpregs(fr); \
93 return ret; \
94}
95
96#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
97static efi_status_t \
98prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
99 efi_time_t *tm) \
100{ \
101 struct ia64_fpreg fr[6]; \
102 efi_status_t ret; \
103 \
104 ia64_save_scratch_fpregs(fr); \
105 ret = efi_call_##prefix( \
106 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
107 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
108 ia64_load_scratch_fpregs(fr); \
109 return ret; \
110}
111
112#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
113static efi_status_t \
114prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
115{ \
116 struct ia64_fpreg fr[6]; \
117 efi_time_t *atm = NULL; \
118 efi_status_t ret; \
119 \
120 if (tm) \
121 atm = adjust_arg(tm); \
122 ia64_save_scratch_fpregs(fr); \
123 ret = efi_call_##prefix( \
124 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
125 enabled, atm); \
126 ia64_load_scratch_fpregs(fr); \
127 return ret; \
128}
129
130#define STUB_GET_VARIABLE(prefix, adjust_arg) \
131static efi_status_t \
132prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
133 unsigned long *data_size, void *data) \
134{ \
135 struct ia64_fpreg fr[6]; \
136 u32 *aattr = NULL; \
137 efi_status_t ret; \
138 \
139 if (attr) \
140 aattr = adjust_arg(attr); \
141 ia64_save_scratch_fpregs(fr); \
142 ret = efi_call_##prefix( \
143 (efi_get_variable_t *) __va(runtime->get_variable), \
144 adjust_arg(name), adjust_arg(vendor), aattr, \
145 adjust_arg(data_size), adjust_arg(data)); \
146 ia64_load_scratch_fpregs(fr); \
147 return ret; \
148}
149
150#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
151static efi_status_t \
152prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
153 efi_guid_t *vendor) \
154{ \
155 struct ia64_fpreg fr[6]; \
156 efi_status_t ret; \
157 \
158 ia64_save_scratch_fpregs(fr); \
159 ret = efi_call_##prefix( \
160 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
161 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
162 ia64_load_scratch_fpregs(fr); \
163 return ret; \
164}
165
166#define STUB_SET_VARIABLE(prefix, adjust_arg) \
167static efi_status_t \
168prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
169 u32 attr, unsigned long data_size, \
170 void *data) \
171{ \
172 struct ia64_fpreg fr[6]; \
173 efi_status_t ret; \
174 \
175 ia64_save_scratch_fpregs(fr); \
176 ret = efi_call_##prefix( \
177 (efi_set_variable_t *) __va(runtime->set_variable), \
178 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
179 adjust_arg(data)); \
180 ia64_load_scratch_fpregs(fr); \
181 return ret; \
182}
183
184#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
185static efi_status_t \
186prefix##_get_next_high_mono_count (u32 *count) \
187{ \
188 struct ia64_fpreg fr[6]; \
189 efi_status_t ret; \
190 \
191 ia64_save_scratch_fpregs(fr); \
192 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
193 __va(runtime->get_next_high_mono_count), \
194 adjust_arg(count)); \
195 ia64_load_scratch_fpregs(fr); \
196 return ret; \
197}
198
199#define STUB_RESET_SYSTEM(prefix, adjust_arg) \
200static void \
201prefix##_reset_system (int reset_type, efi_status_t status, \
202 unsigned long data_size, efi_char16_t *data) \
203{ \
204 struct ia64_fpreg fr[6]; \
205 efi_char16_t *adata = NULL; \
206 \
207 if (data) \
208 adata = adjust_arg(data); \
209 \
210 ia64_save_scratch_fpregs(fr); \
211 efi_call_##prefix( \
212 (efi_reset_system_t *) __va(runtime->reset_system), \
213 reset_type, status, data_size, adata); \
214 /* should not return, but just in case... */ \
215 ia64_load_scratch_fpregs(fr); \
216}
217
218#define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
219
220STUB_GET_TIME(phys, phys_ptr)
221STUB_SET_TIME(phys, phys_ptr)
222STUB_GET_WAKEUP_TIME(phys, phys_ptr)
223STUB_SET_WAKEUP_TIME(phys, phys_ptr)
224STUB_GET_VARIABLE(phys, phys_ptr)
225STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
226STUB_SET_VARIABLE(phys, phys_ptr)
227STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
228STUB_RESET_SYSTEM(phys, phys_ptr)
229
230#define id(arg) arg
231
232STUB_GET_TIME(virt, id)
233STUB_SET_TIME(virt, id)
234STUB_GET_WAKEUP_TIME(virt, id)
235STUB_SET_WAKEUP_TIME(virt, id)
236STUB_GET_VARIABLE(virt, id)
237STUB_GET_NEXT_VARIABLE(virt, id)
238STUB_SET_VARIABLE(virt, id)
239STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
240STUB_RESET_SYSTEM(virt, id)
241
242void
243efi_gettimeofday (struct timespec64 *ts)
244{
245 efi_time_t tm;
246
247 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
248 memset(ts, 0, sizeof(*ts));
249 return;
250 }
251
252 ts->tv_sec = mktime64(tm.year, tm.month, tm.day,
253 tm.hour, tm.minute, tm.second);
254 ts->tv_nsec = tm.nanosecond;
255}
256
257static int
258is_memory_available (efi_memory_desc_t *md)
259{
260 if (!(md->attribute & EFI_MEMORY_WB))
261 return 0;
262
263 switch (md->type) {
264 case EFI_LOADER_CODE:
265 case EFI_LOADER_DATA:
266 case EFI_BOOT_SERVICES_CODE:
267 case EFI_BOOT_SERVICES_DATA:
268 case EFI_CONVENTIONAL_MEMORY:
269 return 1;
270 }
271 return 0;
272}
273
274typedef struct kern_memdesc {
275 u64 attribute;
276 u64 start;
277 u64 num_pages;
278} kern_memdesc_t;
279
280static kern_memdesc_t *kern_memmap;
281
282#define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
283
284static inline u64
285kmd_end(kern_memdesc_t *kmd)
286{
287 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
288}
289
290static inline u64
291efi_md_end(efi_memory_desc_t *md)
292{
293 return (md->phys_addr + efi_md_size(md));
294}
295
296static inline int
297efi_wb(efi_memory_desc_t *md)
298{
299 return (md->attribute & EFI_MEMORY_WB);
300}
301
302static inline int
303efi_uc(efi_memory_desc_t *md)
304{
305 return (md->attribute & EFI_MEMORY_UC);
306}
307
308static void
309walk (efi_freemem_callback_t callback, void *arg, u64 attr)
310{
311 kern_memdesc_t *k;
312 u64 start, end, voff;
313
314 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
315 for (k = kern_memmap; k->start != ~0UL; k++) {
316 if (k->attribute != attr)
317 continue;
318 start = PAGE_ALIGN(k->start);
319 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
320 if (start < end)
321 if ((*callback)(start + voff, end + voff, arg) < 0)
322 return;
323 }
324}
325
326/*
327 * Walk the EFI memory map and call CALLBACK once for each EFI memory
328 * descriptor that has memory that is available for OS use.
329 */
330void
331efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
332{
333 walk(callback, arg, EFI_MEMORY_WB);
334}
335
336/*
337 * Walk the EFI memory map and call CALLBACK once for each EFI memory
338 * descriptor that has memory that is available for uncached allocator.
339 */
340void
341efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
342{
343 walk(callback, arg, EFI_MEMORY_UC);
344}
345
346/*
347 * Look for the PAL_CODE region reported by EFI and map it using an
348 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
349 * Abstraction Layer chapter 11 in ADAG
350 */
351void *
352efi_get_pal_addr (void)
353{
354 void *efi_map_start, *efi_map_end, *p;
355 efi_memory_desc_t *md;
356 u64 efi_desc_size;
357 int pal_code_count = 0;
358 u64 vaddr, mask;
359
360 efi_map_start = __va(ia64_boot_param->efi_memmap);
361 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
362 efi_desc_size = ia64_boot_param->efi_memdesc_size;
363
364 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
365 md = p;
366 if (md->type != EFI_PAL_CODE)
367 continue;
368
369 if (++pal_code_count > 1) {
370 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
371 "dropped @ %llx\n", md->phys_addr);
372 continue;
373 }
374 /*
375 * The only ITLB entry in region 7 that is used is the one
376 * installed by __start(). That entry covers a 64MB range.
377 */
378 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
379 vaddr = PAGE_OFFSET + md->phys_addr;
380
381 /*
382 * We must check that the PAL mapping won't overlap with the
383 * kernel mapping.
384 *
385 * PAL code is guaranteed to be aligned on a power of 2 between
386 * 4k and 256KB and that only one ITR is needed to map it. This
387 * implies that the PAL code is always aligned on its size,
388 * i.e., the closest matching page size supported by the TLB.
389 * Therefore PAL code is guaranteed never to cross a 64MB unless
390 * it is bigger than 64MB (very unlikely!). So for now the
391 * following test is enough to determine whether or not we need
392 * a dedicated ITR for the PAL code.
393 */
394 if ((vaddr & mask) == (KERNEL_START & mask)) {
395 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
396 __func__);
397 continue;
398 }
399
400 if (efi_md_size(md) > IA64_GRANULE_SIZE)
401 panic("Whoa! PAL code size bigger than a granule!");
402
403#if EFI_DEBUG
404 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
405
406 printk(KERN_INFO "CPU %d: mapping PAL code "
407 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
408 smp_processor_id(), md->phys_addr,
409 md->phys_addr + efi_md_size(md),
410 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
411#endif
412 return __va(md->phys_addr);
413 }
414 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
415 __func__);
416 return NULL;
417}
418
419
420static u8 __init palo_checksum(u8 *buffer, u32 length)
421{
422 u8 sum = 0;
423 u8 *end = buffer + length;
424
425 while (buffer < end)
426 sum = (u8) (sum + *(buffer++));
427
428 return sum;
429}
430
431/*
432 * Parse and handle PALO table which is published at:
433 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
434 */
435static void __init handle_palo(unsigned long phys_addr)
436{
437 struct palo_table *palo = __va(phys_addr);
438 u8 checksum;
439
440 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
441 printk(KERN_INFO "PALO signature incorrect.\n");
442 return;
443 }
444
445 checksum = palo_checksum((u8 *)palo, palo->length);
446 if (checksum) {
447 printk(KERN_INFO "PALO checksum incorrect.\n");
448 return;
449 }
450
451 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
452}
453
454void
455efi_map_pal_code (void)
456{
457 void *pal_vaddr = efi_get_pal_addr ();
458 u64 psr;
459
460 if (!pal_vaddr)
461 return;
462
463 /*
464 * Cannot write to CRx with PSR.ic=1
465 */
466 psr = ia64_clear_ic();
467 ia64_itr(0x1, IA64_TR_PALCODE,
468 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
469 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
470 IA64_GRANULE_SHIFT);
471 ia64_set_psr(psr); /* restore psr */
472}
473
474void __init
475efi_init (void)
476{
477 void *efi_map_start, *efi_map_end;
478 efi_char16_t *c16;
479 u64 efi_desc_size;
480 char *cp, vendor[100] = "unknown";
481 int i;
482
483 set_bit(EFI_BOOT, &efi.flags);
484 set_bit(EFI_64BIT, &efi.flags);
485
486 /*
487 * It's too early to be able to use the standard kernel command line
488 * support...
489 */
490 for (cp = boot_command_line; *cp; ) {
491 if (memcmp(cp, "mem=", 4) == 0) {
492 mem_limit = memparse(cp + 4, &cp);
493 } else if (memcmp(cp, "max_addr=", 9) == 0) {
494 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
495 } else if (memcmp(cp, "min_addr=", 9) == 0) {
496 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
497 } else {
498 while (*cp != ' ' && *cp)
499 ++cp;
500 while (*cp == ' ')
501 ++cp;
502 }
503 }
504 if (min_addr != 0UL)
505 printk(KERN_INFO "Ignoring memory below %lluMB\n",
506 min_addr >> 20);
507 if (max_addr != ~0UL)
508 printk(KERN_INFO "Ignoring memory above %lluMB\n",
509 max_addr >> 20);
510
511 efi.systab = __va(ia64_boot_param->efi_systab);
512
513 /*
514 * Verify the EFI Table
515 */
516 if (efi.systab == NULL)
517 panic("Whoa! Can't find EFI system table.\n");
518 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
519 panic("Whoa! EFI system table signature incorrect\n");
520 if ((efi.systab->hdr.revision >> 16) == 0)
521 printk(KERN_WARNING "Warning: EFI system table version "
522 "%d.%02d, expected 1.00 or greater\n",
523 efi.systab->hdr.revision >> 16,
524 efi.systab->hdr.revision & 0xffff);
525
526 /* Show what we know for posterity */
527 c16 = __va(efi.systab->fw_vendor);
528 if (c16) {
529 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
530 vendor[i] = *c16++;
531 vendor[i] = '\0';
532 }
533
534 printk(KERN_INFO "EFI v%u.%.02u by %s:",
535 efi.systab->hdr.revision >> 16,
536 efi.systab->hdr.revision & 0xffff, vendor);
537
538 palo_phys = EFI_INVALID_TABLE_ADDR;
539
540 if (efi_config_init(arch_tables) != 0)
541 return;
542
543 if (palo_phys != EFI_INVALID_TABLE_ADDR)
544 handle_palo(palo_phys);
545
546 runtime = __va(efi.systab->runtime);
547 efi.get_time = phys_get_time;
548 efi.set_time = phys_set_time;
549 efi.get_wakeup_time = phys_get_wakeup_time;
550 efi.set_wakeup_time = phys_set_wakeup_time;
551 efi.get_variable = phys_get_variable;
552 efi.get_next_variable = phys_get_next_variable;
553 efi.set_variable = phys_set_variable;
554 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
555 efi.reset_system = phys_reset_system;
556
557 efi_map_start = __va(ia64_boot_param->efi_memmap);
558 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
559 efi_desc_size = ia64_boot_param->efi_memdesc_size;
560
561#if EFI_DEBUG
562 /* print EFI memory map: */
563 {
564 efi_memory_desc_t *md;
565 void *p;
566
567 for (i = 0, p = efi_map_start; p < efi_map_end;
568 ++i, p += efi_desc_size)
569 {
570 const char *unit;
571 unsigned long size;
572 char buf[64];
573
574 md = p;
575 size = md->num_pages << EFI_PAGE_SHIFT;
576
577 if ((size >> 40) > 0) {
578 size >>= 40;
579 unit = "TB";
580 } else if ((size >> 30) > 0) {
581 size >>= 30;
582 unit = "GB";
583 } else if ((size >> 20) > 0) {
584 size >>= 20;
585 unit = "MB";
586 } else {
587 size >>= 10;
588 unit = "KB";
589 }
590
591 printk("mem%02d: %s "
592 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
593 i, efi_md_typeattr_format(buf, sizeof(buf), md),
594 md->phys_addr,
595 md->phys_addr + efi_md_size(md), size, unit);
596 }
597 }
598#endif
599
600 efi_map_pal_code();
601 efi_enter_virtual_mode();
602}
603
604void
605efi_enter_virtual_mode (void)
606{
607 void *efi_map_start, *efi_map_end, *p;
608 efi_memory_desc_t *md;
609 efi_status_t status;
610 u64 efi_desc_size;
611
612 efi_map_start = __va(ia64_boot_param->efi_memmap);
613 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
614 efi_desc_size = ia64_boot_param->efi_memdesc_size;
615
616 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
617 md = p;
618 if (md->attribute & EFI_MEMORY_RUNTIME) {
619 /*
620 * Some descriptors have multiple bits set, so the
621 * order of the tests is relevant.
622 */
623 if (md->attribute & EFI_MEMORY_WB) {
624 md->virt_addr = (u64) __va(md->phys_addr);
625 } else if (md->attribute & EFI_MEMORY_UC) {
626 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
627 } else if (md->attribute & EFI_MEMORY_WC) {
628#if 0
629 md->virt_addr = ia64_remap(md->phys_addr,
630 (_PAGE_A |
631 _PAGE_P |
632 _PAGE_D |
633 _PAGE_MA_WC |
634 _PAGE_PL_0 |
635 _PAGE_AR_RW));
636#else
637 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
638 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
639#endif
640 } else if (md->attribute & EFI_MEMORY_WT) {
641#if 0
642 md->virt_addr = ia64_remap(md->phys_addr,
643 (_PAGE_A |
644 _PAGE_P |
645 _PAGE_D |
646 _PAGE_MA_WT |
647 _PAGE_PL_0 |
648 _PAGE_AR_RW));
649#else
650 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
651 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
652#endif
653 }
654 }
655 }
656
657 status = efi_call_phys(__va(runtime->set_virtual_address_map),
658 ia64_boot_param->efi_memmap_size,
659 efi_desc_size,
660 ia64_boot_param->efi_memdesc_version,
661 ia64_boot_param->efi_memmap);
662 if (status != EFI_SUCCESS) {
663 printk(KERN_WARNING "warning: unable to switch EFI into "
664 "virtual mode (status=%lu)\n", status);
665 return;
666 }
667
668 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
669
670 /*
671 * Now that EFI is in virtual mode, we call the EFI functions more
672 * efficiently:
673 */
674 efi.get_time = virt_get_time;
675 efi.set_time = virt_set_time;
676 efi.get_wakeup_time = virt_get_wakeup_time;
677 efi.set_wakeup_time = virt_set_wakeup_time;
678 efi.get_variable = virt_get_variable;
679 efi.get_next_variable = virt_get_next_variable;
680 efi.set_variable = virt_set_variable;
681 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
682 efi.reset_system = virt_reset_system;
683}
684
685/*
686 * Walk the EFI memory map looking for the I/O port range. There can only be
687 * one entry of this type, other I/O port ranges should be described via ACPI.
688 */
689u64
690efi_get_iobase (void)
691{
692 void *efi_map_start, *efi_map_end, *p;
693 efi_memory_desc_t *md;
694 u64 efi_desc_size;
695
696 efi_map_start = __va(ia64_boot_param->efi_memmap);
697 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
698 efi_desc_size = ia64_boot_param->efi_memdesc_size;
699
700 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
701 md = p;
702 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
703 if (md->attribute & EFI_MEMORY_UC)
704 return md->phys_addr;
705 }
706 }
707 return 0;
708}
709
710static struct kern_memdesc *
711kern_memory_descriptor (unsigned long phys_addr)
712{
713 struct kern_memdesc *md;
714
715 for (md = kern_memmap; md->start != ~0UL; md++) {
716 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
717 return md;
718 }
719 return NULL;
720}
721
722static efi_memory_desc_t *
723efi_memory_descriptor (unsigned long phys_addr)
724{
725 void *efi_map_start, *efi_map_end, *p;
726 efi_memory_desc_t *md;
727 u64 efi_desc_size;
728
729 efi_map_start = __va(ia64_boot_param->efi_memmap);
730 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
731 efi_desc_size = ia64_boot_param->efi_memdesc_size;
732
733 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
734 md = p;
735
736 if (phys_addr - md->phys_addr < efi_md_size(md))
737 return md;
738 }
739 return NULL;
740}
741
742static int
743efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
744{
745 void *efi_map_start, *efi_map_end, *p;
746 efi_memory_desc_t *md;
747 u64 efi_desc_size;
748 unsigned long end;
749
750 efi_map_start = __va(ia64_boot_param->efi_memmap);
751 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
752 efi_desc_size = ia64_boot_param->efi_memdesc_size;
753
754 end = phys_addr + size;
755
756 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
757 md = p;
758 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
759 return 1;
760 }
761 return 0;
762}
763
764int
765efi_mem_type (unsigned long phys_addr)
766{
767 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
768
769 if (md)
770 return md->type;
771 return -EINVAL;
772}
773
774u64
775efi_mem_attributes (unsigned long phys_addr)
776{
777 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
778
779 if (md)
780 return md->attribute;
781 return 0;
782}
783EXPORT_SYMBOL(efi_mem_attributes);
784
785u64
786efi_mem_attribute (unsigned long phys_addr, unsigned long size)
787{
788 unsigned long end = phys_addr + size;
789 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
790 u64 attr;
791
792 if (!md)
793 return 0;
794
795 /*
796 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
797 * the kernel that firmware needs this region mapped.
798 */
799 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
800 do {
801 unsigned long md_end = efi_md_end(md);
802
803 if (end <= md_end)
804 return attr;
805
806 md = efi_memory_descriptor(md_end);
807 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
808 return 0;
809 } while (md);
810 return 0; /* never reached */
811}
812
813u64
814kern_mem_attribute (unsigned long phys_addr, unsigned long size)
815{
816 unsigned long end = phys_addr + size;
817 struct kern_memdesc *md;
818 u64 attr;
819
820 /*
821 * This is a hack for ioremap calls before we set up kern_memmap.
822 * Maybe we should do efi_memmap_init() earlier instead.
823 */
824 if (!kern_memmap) {
825 attr = efi_mem_attribute(phys_addr, size);
826 if (attr & EFI_MEMORY_WB)
827 return EFI_MEMORY_WB;
828 return 0;
829 }
830
831 md = kern_memory_descriptor(phys_addr);
832 if (!md)
833 return 0;
834
835 attr = md->attribute;
836 do {
837 unsigned long md_end = kmd_end(md);
838
839 if (end <= md_end)
840 return attr;
841
842 md = kern_memory_descriptor(md_end);
843 if (!md || md->attribute != attr)
844 return 0;
845 } while (md);
846 return 0; /* never reached */
847}
848
849int
850valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
851{
852 u64 attr;
853
854 /*
855 * /dev/mem reads and writes use copy_to_user(), which implicitly
856 * uses a granule-sized kernel identity mapping. It's really
857 * only safe to do this for regions in kern_memmap. For more
858 * details, see Documentation/ia64/aliasing.rst.
859 */
860 attr = kern_mem_attribute(phys_addr, size);
861 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
862 return 1;
863 return 0;
864}
865
866int
867valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
868{
869 unsigned long phys_addr = pfn << PAGE_SHIFT;
870 u64 attr;
871
872 attr = efi_mem_attribute(phys_addr, size);
873
874 /*
875 * /dev/mem mmap uses normal user pages, so we don't need the entire
876 * granule, but the entire region we're mapping must support the same
877 * attribute.
878 */
879 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
880 return 1;
881
882 /*
883 * Intel firmware doesn't tell us about all the MMIO regions, so
884 * in general we have to allow mmap requests. But if EFI *does*
885 * tell us about anything inside this region, we should deny it.
886 * The user can always map a smaller region to avoid the overlap.
887 */
888 if (efi_memmap_intersects(phys_addr, size))
889 return 0;
890
891 return 1;
892}
893
894pgprot_t
895phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
896 pgprot_t vma_prot)
897{
898 unsigned long phys_addr = pfn << PAGE_SHIFT;
899 u64 attr;
900
901 /*
902 * For /dev/mem mmap, we use user mappings, but if the region is
903 * in kern_memmap (and hence may be covered by a kernel mapping),
904 * we must use the same attribute as the kernel mapping.
905 */
906 attr = kern_mem_attribute(phys_addr, size);
907 if (attr & EFI_MEMORY_WB)
908 return pgprot_cacheable(vma_prot);
909 else if (attr & EFI_MEMORY_UC)
910 return pgprot_noncached(vma_prot);
911
912 /*
913 * Some chipsets don't support UC access to memory. If
914 * WB is supported, we prefer that.
915 */
916 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
917 return pgprot_cacheable(vma_prot);
918
919 return pgprot_noncached(vma_prot);
920}
921
922int __init
923efi_uart_console_only(void)
924{
925 efi_status_t status;
926 char *s, name[] = "ConOut";
927 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
928 efi_char16_t *utf16, name_utf16[32];
929 unsigned char data[1024];
930 unsigned long size = sizeof(data);
931 struct efi_generic_dev_path *hdr, *end_addr;
932 int uart = 0;
933
934 /* Convert to UTF-16 */
935 utf16 = name_utf16;
936 s = name;
937 while (*s)
938 *utf16++ = *s++ & 0x7f;
939 *utf16 = 0;
940
941 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
942 if (status != EFI_SUCCESS) {
943 printk(KERN_ERR "No EFI %s variable?\n", name);
944 return 0;
945 }
946
947 hdr = (struct efi_generic_dev_path *) data;
948 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
949 while (hdr < end_addr) {
950 if (hdr->type == EFI_DEV_MSG &&
951 hdr->sub_type == EFI_DEV_MSG_UART)
952 uart = 1;
953 else if (hdr->type == EFI_DEV_END_PATH ||
954 hdr->type == EFI_DEV_END_PATH2) {
955 if (!uart)
956 return 0;
957 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
958 return 1;
959 uart = 0;
960 }
961 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
962 }
963 printk(KERN_ERR "Malformed %s value\n", name);
964 return 0;
965}
966
967/*
968 * Look for the first granule aligned memory descriptor memory
969 * that is big enough to hold EFI memory map. Make sure this
970 * descriptor is at least granule sized so it does not get trimmed
971 */
972struct kern_memdesc *
973find_memmap_space (void)
974{
975 u64 contig_low=0, contig_high=0;
976 u64 as = 0, ae;
977 void *efi_map_start, *efi_map_end, *p, *q;
978 efi_memory_desc_t *md, *pmd = NULL, *check_md;
979 u64 space_needed, efi_desc_size;
980 unsigned long total_mem = 0;
981
982 efi_map_start = __va(ia64_boot_param->efi_memmap);
983 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
984 efi_desc_size = ia64_boot_param->efi_memdesc_size;
985
986 /*
987 * Worst case: we need 3 kernel descriptors for each efi descriptor
988 * (if every entry has a WB part in the middle, and UC head and tail),
989 * plus one for the end marker.
990 */
991 space_needed = sizeof(kern_memdesc_t) *
992 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
993
994 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
995 md = p;
996 if (!efi_wb(md)) {
997 continue;
998 }
999 if (pmd == NULL || !efi_wb(pmd) ||
1000 efi_md_end(pmd) != md->phys_addr) {
1001 contig_low = GRANULEROUNDUP(md->phys_addr);
1002 contig_high = efi_md_end(md);
1003 for (q = p + efi_desc_size; q < efi_map_end;
1004 q += efi_desc_size) {
1005 check_md = q;
1006 if (!efi_wb(check_md))
1007 break;
1008 if (contig_high != check_md->phys_addr)
1009 break;
1010 contig_high = efi_md_end(check_md);
1011 }
1012 contig_high = GRANULEROUNDDOWN(contig_high);
1013 }
1014 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1015 continue;
1016
1017 /* Round ends inward to granule boundaries */
1018 as = max(contig_low, md->phys_addr);
1019 ae = min(contig_high, efi_md_end(md));
1020
1021 /* keep within max_addr= and min_addr= command line arg */
1022 as = max(as, min_addr);
1023 ae = min(ae, max_addr);
1024 if (ae <= as)
1025 continue;
1026
1027 /* avoid going over mem= command line arg */
1028 if (total_mem + (ae - as) > mem_limit)
1029 ae -= total_mem + (ae - as) - mem_limit;
1030
1031 if (ae <= as)
1032 continue;
1033
1034 if (ae - as > space_needed)
1035 break;
1036 }
1037 if (p >= efi_map_end)
1038 panic("Can't allocate space for kernel memory descriptors");
1039
1040 return __va(as);
1041}
1042
1043/*
1044 * Walk the EFI memory map and gather all memory available for kernel
1045 * to use. We can allocate partial granules only if the unavailable
1046 * parts exist, and are WB.
1047 */
1048unsigned long
1049efi_memmap_init(u64 *s, u64 *e)
1050{
1051 struct kern_memdesc *k, *prev = NULL;
1052 u64 contig_low=0, contig_high=0;
1053 u64 as, ae, lim;
1054 void *efi_map_start, *efi_map_end, *p, *q;
1055 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1056 u64 efi_desc_size;
1057 unsigned long total_mem = 0;
1058
1059 k = kern_memmap = find_memmap_space();
1060
1061 efi_map_start = __va(ia64_boot_param->efi_memmap);
1062 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1063 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1064
1065 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1066 md = p;
1067 if (!efi_wb(md)) {
1068 if (efi_uc(md) &&
1069 (md->type == EFI_CONVENTIONAL_MEMORY ||
1070 md->type == EFI_BOOT_SERVICES_DATA)) {
1071 k->attribute = EFI_MEMORY_UC;
1072 k->start = md->phys_addr;
1073 k->num_pages = md->num_pages;
1074 k++;
1075 }
1076 continue;
1077 }
1078 if (pmd == NULL || !efi_wb(pmd) ||
1079 efi_md_end(pmd) != md->phys_addr) {
1080 contig_low = GRANULEROUNDUP(md->phys_addr);
1081 contig_high = efi_md_end(md);
1082 for (q = p + efi_desc_size; q < efi_map_end;
1083 q += efi_desc_size) {
1084 check_md = q;
1085 if (!efi_wb(check_md))
1086 break;
1087 if (contig_high != check_md->phys_addr)
1088 break;
1089 contig_high = efi_md_end(check_md);
1090 }
1091 contig_high = GRANULEROUNDDOWN(contig_high);
1092 }
1093 if (!is_memory_available(md))
1094 continue;
1095
1096 /*
1097 * Round ends inward to granule boundaries
1098 * Give trimmings to uncached allocator
1099 */
1100 if (md->phys_addr < contig_low) {
1101 lim = min(efi_md_end(md), contig_low);
1102 if (efi_uc(md)) {
1103 if (k > kern_memmap &&
1104 (k-1)->attribute == EFI_MEMORY_UC &&
1105 kmd_end(k-1) == md->phys_addr) {
1106 (k-1)->num_pages +=
1107 (lim - md->phys_addr)
1108 >> EFI_PAGE_SHIFT;
1109 } else {
1110 k->attribute = EFI_MEMORY_UC;
1111 k->start = md->phys_addr;
1112 k->num_pages = (lim - md->phys_addr)
1113 >> EFI_PAGE_SHIFT;
1114 k++;
1115 }
1116 }
1117 as = contig_low;
1118 } else
1119 as = md->phys_addr;
1120
1121 if (efi_md_end(md) > contig_high) {
1122 lim = max(md->phys_addr, contig_high);
1123 if (efi_uc(md)) {
1124 if (lim == md->phys_addr && k > kern_memmap &&
1125 (k-1)->attribute == EFI_MEMORY_UC &&
1126 kmd_end(k-1) == md->phys_addr) {
1127 (k-1)->num_pages += md->num_pages;
1128 } else {
1129 k->attribute = EFI_MEMORY_UC;
1130 k->start = lim;
1131 k->num_pages = (efi_md_end(md) - lim)
1132 >> EFI_PAGE_SHIFT;
1133 k++;
1134 }
1135 }
1136 ae = contig_high;
1137 } else
1138 ae = efi_md_end(md);
1139
1140 /* keep within max_addr= and min_addr= command line arg */
1141 as = max(as, min_addr);
1142 ae = min(ae, max_addr);
1143 if (ae <= as)
1144 continue;
1145
1146 /* avoid going over mem= command line arg */
1147 if (total_mem + (ae - as) > mem_limit)
1148 ae -= total_mem + (ae - as) - mem_limit;
1149
1150 if (ae <= as)
1151 continue;
1152 if (prev && kmd_end(prev) == md->phys_addr) {
1153 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1154 total_mem += ae - as;
1155 continue;
1156 }
1157 k->attribute = EFI_MEMORY_WB;
1158 k->start = as;
1159 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1160 total_mem += ae - as;
1161 prev = k++;
1162 }
1163 k->start = ~0L; /* end-marker */
1164
1165 /* reserve the memory we are using for kern_memmap */
1166 *s = (u64)kern_memmap;
1167 *e = (u64)++k;
1168
1169 return total_mem;
1170}
1171
1172void
1173efi_initialize_iomem_resources(struct resource *code_resource,
1174 struct resource *data_resource,
1175 struct resource *bss_resource)
1176{
1177 struct resource *res;
1178 void *efi_map_start, *efi_map_end, *p;
1179 efi_memory_desc_t *md;
1180 u64 efi_desc_size;
1181 char *name;
1182 unsigned long flags, desc;
1183
1184 efi_map_start = __va(ia64_boot_param->efi_memmap);
1185 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1186 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1187
1188 res = NULL;
1189
1190 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1191 md = p;
1192
1193 if (md->num_pages == 0) /* should not happen */
1194 continue;
1195
1196 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1197 desc = IORES_DESC_NONE;
1198
1199 switch (md->type) {
1200
1201 case EFI_MEMORY_MAPPED_IO:
1202 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1203 continue;
1204
1205 case EFI_LOADER_CODE:
1206 case EFI_LOADER_DATA:
1207 case EFI_BOOT_SERVICES_DATA:
1208 case EFI_BOOT_SERVICES_CODE:
1209 case EFI_CONVENTIONAL_MEMORY:
1210 if (md->attribute & EFI_MEMORY_WP) {
1211 name = "System ROM";
1212 flags |= IORESOURCE_READONLY;
1213 } else if (md->attribute == EFI_MEMORY_UC) {
1214 name = "Uncached RAM";
1215 } else {
1216 name = "System RAM";
1217 flags |= IORESOURCE_SYSRAM;
1218 }
1219 break;
1220
1221 case EFI_ACPI_MEMORY_NVS:
1222 name = "ACPI Non-volatile Storage";
1223 desc = IORES_DESC_ACPI_NV_STORAGE;
1224 break;
1225
1226 case EFI_UNUSABLE_MEMORY:
1227 name = "reserved";
1228 flags |= IORESOURCE_DISABLED;
1229 break;
1230
1231 case EFI_PERSISTENT_MEMORY:
1232 name = "Persistent Memory";
1233 desc = IORES_DESC_PERSISTENT_MEMORY;
1234 break;
1235
1236 case EFI_RESERVED_TYPE:
1237 case EFI_RUNTIME_SERVICES_CODE:
1238 case EFI_RUNTIME_SERVICES_DATA:
1239 case EFI_ACPI_RECLAIM_MEMORY:
1240 default:
1241 name = "reserved";
1242 break;
1243 }
1244
1245 if ((res = kzalloc(sizeof(struct resource),
1246 GFP_KERNEL)) == NULL) {
1247 printk(KERN_ERR
1248 "failed to allocate resource for iomem\n");
1249 return;
1250 }
1251
1252 res->name = name;
1253 res->start = md->phys_addr;
1254 res->end = md->phys_addr + efi_md_size(md) - 1;
1255 res->flags = flags;
1256 res->desc = desc;
1257
1258 if (insert_resource(&iomem_resource, res) < 0)
1259 kfree(res);
1260 else {
1261 /*
1262 * We don't know which region contains
1263 * kernel data so we try it repeatedly and
1264 * let the resource manager test it.
1265 */
1266 insert_resource(res, code_resource);
1267 insert_resource(res, data_resource);
1268 insert_resource(res, bss_resource);
1269#ifdef CONFIG_KEXEC
1270 insert_resource(res, &efi_memmap_res);
1271 insert_resource(res, &boot_param_res);
1272 if (crashk_res.end > crashk_res.start)
1273 insert_resource(res, &crashk_res);
1274#endif
1275 }
1276 }
1277}
1278
1279#ifdef CONFIG_KEXEC
1280/* find a block of memory aligned to 64M exclude reserved regions
1281 rsvd_regions are sorted
1282 */
1283unsigned long __init
1284kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1285{
1286 int i;
1287 u64 start, end;
1288 u64 alignment = 1UL << _PAGE_SIZE_64M;
1289 void *efi_map_start, *efi_map_end, *p;
1290 efi_memory_desc_t *md;
1291 u64 efi_desc_size;
1292
1293 efi_map_start = __va(ia64_boot_param->efi_memmap);
1294 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1295 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1296
1297 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1298 md = p;
1299 if (!efi_wb(md))
1300 continue;
1301 start = ALIGN(md->phys_addr, alignment);
1302 end = efi_md_end(md);
1303 for (i = 0; i < n; i++) {
1304 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1305 if (__pa(r[i].start) > start + size)
1306 return start;
1307 start = ALIGN(__pa(r[i].end), alignment);
1308 if (i < n-1 &&
1309 __pa(r[i+1].start) < start + size)
1310 continue;
1311 else
1312 break;
1313 }
1314 }
1315 if (end > start + size)
1316 return start;
1317 }
1318
1319 printk(KERN_WARNING
1320 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1321 return ~0UL;
1322}
1323#endif
1324
1325#ifdef CONFIG_CRASH_DUMP
1326/* locate the size find a the descriptor at a certain address */
1327unsigned long __init
1328vmcore_find_descriptor_size (unsigned long address)
1329{
1330 void *efi_map_start, *efi_map_end, *p;
1331 efi_memory_desc_t *md;
1332 u64 efi_desc_size;
1333 unsigned long ret = 0;
1334
1335 efi_map_start = __va(ia64_boot_param->efi_memmap);
1336 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1337 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1338
1339 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1340 md = p;
1341 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1342 && md->phys_addr == address) {
1343 ret = efi_md_size(md);
1344 break;
1345 }
1346 }
1347
1348 if (ret == 0)
1349 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1350
1351 return ret;
1352}
1353#endif