io_uring: ensure finish_wait() is always called in __io_uring_task_cancel()
[linux/fpc-iii.git] / arch / ia64 / kernel / efi.c
blobf932b25fb817a10afda09f962ccf74dd59c1705e
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Extensible Firmware Interface
5 * Based on Extensible Firmware Interface Specification version 0.9
6 * April 30, 1999
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>
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
20 * Implemented EFI runtime services and virtual mode calls. --davidm
22 * Goutham Rao: <goutham.rao@intel.com>
23 * Skip non-WB memory and ignore empty memory ranges.
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>
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>
45 #define EFI_DEBUG 0
47 #define ESI_TABLE_GUID \
48 EFI_GUID(0x43EA58DC, 0xCF28, 0x4b06, 0xB3, \
49 0x91, 0xB7, 0x50, 0x59, 0x34, 0x2B, 0xD4)
51 static unsigned long mps_phys = EFI_INVALID_TABLE_ADDR;
52 static __initdata unsigned long palo_phys;
54 unsigned long __initdata esi_phys = EFI_INVALID_TABLE_ADDR;
55 unsigned long hcdp_phys = EFI_INVALID_TABLE_ADDR;
56 unsigned long sal_systab_phys = EFI_INVALID_TABLE_ADDR;
58 static 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 {},
67 extern efi_status_t efi_call_phys (void *, ...);
69 static efi_runtime_services_t *runtime;
70 static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
72 #define efi_call_virt(f, args...) (*(f))(args)
74 #define STUB_GET_TIME(prefix, adjust_arg) \
75 static efi_status_t \
76 prefix##_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; \
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; \
91 #define STUB_SET_TIME(prefix, adjust_arg) \
92 static efi_status_t \
93 prefix##_set_time (efi_time_t *tm) \
94 { \
95 struct ia64_fpreg fr[6]; \
96 efi_status_t ret; \
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; \
105 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
106 static efi_status_t \
107 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
108 efi_time_t *tm) \
110 struct ia64_fpreg fr[6]; \
111 efi_status_t ret; \
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; \
121 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
122 static efi_status_t \
123 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
125 struct ia64_fpreg fr[6]; \
126 efi_time_t *atm = NULL; \
127 efi_status_t ret; \
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; \
139 #define STUB_GET_VARIABLE(prefix, adjust_arg) \
140 static efi_status_t \
141 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
142 unsigned long *data_size, void *data) \
144 struct ia64_fpreg fr[6]; \
145 u32 *aattr = NULL; \
146 efi_status_t ret; \
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; \
159 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
160 static efi_status_t \
161 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
162 efi_guid_t *vendor) \
164 struct ia64_fpreg fr[6]; \
165 efi_status_t ret; \
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; \
175 #define STUB_SET_VARIABLE(prefix, adjust_arg) \
176 static efi_status_t \
177 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
178 u32 attr, unsigned long data_size, \
179 void *data) \
181 struct ia64_fpreg fr[6]; \
182 efi_status_t ret; \
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; \
193 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
194 static efi_status_t \
195 prefix##_get_next_high_mono_count (u32 *count) \
197 struct ia64_fpreg fr[6]; \
198 efi_status_t ret; \
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; \
208 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \
209 static void \
210 prefix##_reset_system (int reset_type, efi_status_t status, \
211 unsigned long data_size, efi_char16_t *data) \
213 struct ia64_fpreg fr[6]; \
214 efi_char16_t *adata = NULL; \
216 if (data) \
217 adata = adjust_arg(data); \
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); \
227 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
229 STUB_GET_TIME(phys, phys_ptr)
230 STUB_SET_TIME(phys, phys_ptr)
231 STUB_GET_WAKEUP_TIME(phys, phys_ptr)
232 STUB_SET_WAKEUP_TIME(phys, phys_ptr)
233 STUB_GET_VARIABLE(phys, phys_ptr)
234 STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
235 STUB_SET_VARIABLE(phys, phys_ptr)
236 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
237 STUB_RESET_SYSTEM(phys, phys_ptr)
239 #define id(arg) arg
241 STUB_GET_TIME(virt, id)
242 STUB_SET_TIME(virt, id)
243 STUB_GET_WAKEUP_TIME(virt, id)
244 STUB_SET_WAKEUP_TIME(virt, id)
245 STUB_GET_VARIABLE(virt, id)
246 STUB_GET_NEXT_VARIABLE(virt, id)
247 STUB_SET_VARIABLE(virt, id)
248 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
249 STUB_RESET_SYSTEM(virt, id)
251 void
252 efi_gettimeofday (struct timespec64 *ts)
254 efi_time_t tm;
256 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
257 memset(ts, 0, sizeof(*ts));
258 return;
261 ts->tv_sec = mktime64(tm.year, tm.month, tm.day,
262 tm.hour, tm.minute, tm.second);
263 ts->tv_nsec = tm.nanosecond;
266 static int
267 is_memory_available (efi_memory_desc_t *md)
269 if (!(md->attribute & EFI_MEMORY_WB))
270 return 0;
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;
280 return 0;
283 typedef struct kern_memdesc {
284 u64 attribute;
285 u64 start;
286 u64 num_pages;
287 } kern_memdesc_t;
289 static kern_memdesc_t *kern_memmap;
291 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
293 static inline u64
294 kmd_end(kern_memdesc_t *kmd)
296 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
299 static inline u64
300 efi_md_end(efi_memory_desc_t *md)
302 return (md->phys_addr + efi_md_size(md));
305 static inline int
306 efi_wb(efi_memory_desc_t *md)
308 return (md->attribute & EFI_MEMORY_WB);
311 static inline int
312 efi_uc(efi_memory_desc_t *md)
314 return (md->attribute & EFI_MEMORY_UC);
317 static void
318 walk (efi_freemem_callback_t callback, void *arg, u64 attr)
320 kern_memdesc_t *k;
321 u64 start, end, voff;
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;
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.
339 void
340 efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
342 walk(callback, arg, EFI_MEMORY_WB);
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.
349 void
350 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
352 walk(callback, arg, EFI_MEMORY_UC);
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
360 void *
361 efi_get_pal_addr (void)
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;
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;
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;
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;
384 * The only ITLB entry in region 7 that is used is the one
385 * installed by __start(). That entry covers a 64MB range.
387 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
388 vaddr = PAGE_OFFSET + md->phys_addr;
391 * We must check that the PAL mapping won't overlap with the
392 * kernel mapping.
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.
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;
409 if (efi_md_size(md) > IA64_GRANULE_SIZE)
410 panic("Whoa! PAL code size bigger than a granule!");
412 #if EFI_DEBUG
413 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
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);
423 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
424 __func__);
425 return NULL;
429 static u8 __init palo_checksum(u8 *buffer, u32 length)
431 u8 sum = 0;
432 u8 *end = buffer + length;
434 while (buffer < end)
435 sum = (u8) (sum + *(buffer++));
437 return sum;
441 * Parse and handle PALO table which is published at:
442 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
444 static void __init handle_palo(unsigned long phys_addr)
446 struct palo_table *palo = __va(phys_addr);
447 u8 checksum;
449 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
450 printk(KERN_INFO "PALO signature incorrect.\n");
451 return;
454 checksum = palo_checksum((u8 *)palo, palo->length);
455 if (checksum) {
456 printk(KERN_INFO "PALO checksum incorrect.\n");
457 return;
460 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
463 void
464 efi_map_pal_code (void)
466 void *pal_vaddr = efi_get_pal_addr ();
467 u64 psr;
469 if (!pal_vaddr)
470 return;
473 * Cannot write to CRx with PSR.ic=1
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 */
483 void __init
484 efi_init (void)
486 const efi_system_table_t *efi_systab;
487 void *efi_map_start, *efi_map_end;
488 u64 efi_desc_size;
489 char *cp;
491 set_bit(EFI_BOOT, &efi.flags);
492 set_bit(EFI_64BIT, &efi.flags);
495 * It's too early to be able to use the standard kernel command line
496 * support...
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;
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);
519 efi_systab = __va(ia64_boot_param->efi_systab);
522 * Verify the EFI Table
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");
529 efi_systab_report_header(&efi_systab->hdr, efi_systab->fw_vendor);
531 palo_phys = EFI_INVALID_TABLE_ADDR;
533 if (efi_config_parse_tables(__va(efi_systab->tables),
534 efi_systab->nr_tables,
535 arch_tables) != 0)
536 return;
538 if (palo_phys != EFI_INVALID_TABLE_ADDR)
539 handle_palo(palo_phys);
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;
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;
556 #if EFI_DEBUG
557 /* print EFI memory map: */
559 efi_memory_desc_t *md;
560 void *p;
562 for (i = 0, p = efi_map_start; p < efi_map_end;
563 ++i, p += efi_desc_size)
565 const char *unit;
566 unsigned long size;
567 char buf[64];
569 md = p;
570 size = md->num_pages << EFI_PAGE_SHIFT;
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";
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);
593 #endif
595 efi_map_pal_code();
596 efi_enter_virtual_mode();
599 void
600 efi_enter_virtual_mode (void)
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;
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;
611 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
612 md = p;
613 if (md->attribute & EFI_MEMORY_RUNTIME) {
615 * Some descriptors have multiple bits set, so the
616 * order of the tests is relevant.
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
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;
663 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
666 * Now that EFI is in virtual mode, we call the EFI functions more
667 * efficiently:
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;
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.
685 efi_get_iobase (void)
687 void *efi_map_start, *efi_map_end, *p;
688 efi_memory_desc_t *md;
689 u64 efi_desc_size;
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;
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;
702 return 0;
705 static struct kern_memdesc *
706 kern_memory_descriptor (unsigned long phys_addr)
708 struct kern_memdesc *md;
710 for (md = kern_memmap; md->start != ~0UL; md++) {
711 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
712 return md;
714 return NULL;
717 static efi_memory_desc_t *
718 efi_memory_descriptor (unsigned long phys_addr)
720 void *efi_map_start, *efi_map_end, *p;
721 efi_memory_desc_t *md;
722 u64 efi_desc_size;
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;
728 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
729 md = p;
731 if (phys_addr - md->phys_addr < efi_md_size(md))
732 return md;
734 return NULL;
737 static int
738 efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
740 void *efi_map_start, *efi_map_end, *p;
741 efi_memory_desc_t *md;
742 u64 efi_desc_size;
743 unsigned long end;
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;
749 end = phys_addr + size;
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;
756 return 0;
760 efi_mem_type (unsigned long phys_addr)
762 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
764 if (md)
765 return md->type;
766 return -EINVAL;
770 efi_mem_attributes (unsigned long phys_addr)
772 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
774 if (md)
775 return md->attribute;
776 return 0;
778 EXPORT_SYMBOL(efi_mem_attributes);
781 efi_mem_attribute (unsigned long phys_addr, unsigned long size)
783 unsigned long end = phys_addr + size;
784 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
785 u64 attr;
787 if (!md)
788 return 0;
791 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
792 * the kernel that firmware needs this region mapped.
794 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
795 do {
796 unsigned long md_end = efi_md_end(md);
798 if (end <= md_end)
799 return attr;
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 */
809 kern_mem_attribute (unsigned long phys_addr, unsigned long size)
811 unsigned long end = phys_addr + size;
812 struct kern_memdesc *md;
813 u64 attr;
816 * This is a hack for ioremap calls before we set up kern_memmap.
817 * Maybe we should do efi_memmap_init() earlier instead.
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;
826 md = kern_memory_descriptor(phys_addr);
827 if (!md)
828 return 0;
830 attr = md->attribute;
831 do {
832 unsigned long md_end = kmd_end(md);
834 if (end <= md_end)
835 return attr;
837 md = kern_memory_descriptor(md_end);
838 if (!md || md->attribute != attr)
839 return 0;
840 } while (md);
841 return 0; /* never reached */
845 valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
847 u64 attr;
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.
855 attr = kern_mem_attribute(phys_addr, size);
856 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
857 return 1;
858 return 0;
862 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
864 unsigned long phys_addr = pfn << PAGE_SHIFT;
865 u64 attr;
867 attr = efi_mem_attribute(phys_addr, size);
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.
874 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
875 return 1;
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.
883 if (efi_memmap_intersects(phys_addr, size))
884 return 0;
886 return 1;
889 pgprot_t
890 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
891 pgprot_t vma_prot)
893 unsigned long phys_addr = pfn << PAGE_SHIFT;
894 u64 attr;
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.
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);
908 * Some chipsets don't support UC access to memory. If
909 * WB is supported, we prefer that.
911 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
912 return pgprot_cacheable(vma_prot);
914 return pgprot_noncached(vma_prot);
917 int __init
918 efi_uart_console_only(void)
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;
929 /* Convert to UTF-16 */
930 utf16 = name_utf16;
931 s = name;
932 while (*s)
933 *utf16++ = *s++ & 0x7f;
934 *utf16 = 0;
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;
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;
956 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
958 printk(KERN_ERR "Malformed %s value\n", name);
959 return 0;
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
967 struct kern_memdesc *
968 find_memmap_space (void)
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;
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;
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.
986 space_needed = sizeof(kern_memdesc_t) *
987 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
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;
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);
1007 contig_high = GRANULEROUNDDOWN(contig_high);
1009 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1010 continue;
1012 /* Round ends inward to granule boundaries */
1013 as = max(contig_low, md->phys_addr);
1014 ae = min(contig_high, efi_md_end(md));
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;
1022 /* avoid going over mem= command line arg */
1023 if (total_mem + (ae - as) > mem_limit)
1024 ae -= total_mem + (ae - as) - mem_limit;
1026 if (ae <= as)
1027 continue;
1029 if (ae - as > space_needed)
1030 break;
1032 if (p >= efi_map_end)
1033 panic("Can't allocate space for kernel memory descriptors");
1035 return __va(as);
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.
1043 unsigned long
1044 efi_memmap_init(u64 *s, u64 *e)
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;
1054 k = kern_memmap = find_memmap_space();
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;
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++;
1071 continue;
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);
1086 contig_high = GRANULEROUNDDOWN(contig_high);
1088 if (!is_memory_available(md))
1089 continue;
1092 * Round ends inward to granule boundaries
1093 * Give trimmings to uncached allocator
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++;
1112 as = contig_low;
1113 } else
1114 as = md->phys_addr;
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++;
1131 ae = contig_high;
1132 } else
1133 ae = efi_md_end(md);
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;
1141 /* avoid going over mem= command line arg */
1142 if (total_mem + (ae - as) > mem_limit)
1143 ae -= total_mem + (ae - as) - mem_limit;
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;
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++;
1158 k->start = ~0L; /* end-marker */
1160 /* reserve the memory we are using for kern_memmap */
1161 *s = (u64)kern_memmap;
1162 *e = (u64)++k;
1164 return total_mem;
1167 void
1168 efi_initialize_iomem_resources(struct resource *code_resource,
1169 struct resource *data_resource,
1170 struct resource *bss_resource)
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;
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;
1183 res = NULL;
1185 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1186 md = p;
1188 if (md->num_pages == 0) /* should not happen */
1189 continue;
1191 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1192 desc = IORES_DESC_NONE;
1194 switch (md->type) {
1196 case EFI_MEMORY_MAPPED_IO:
1197 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1198 continue;
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;
1214 break;
1216 case EFI_ACPI_MEMORY_NVS:
1217 name = "ACPI Non-volatile Storage";
1218 desc = IORES_DESC_ACPI_NV_STORAGE;
1219 break;
1221 case EFI_UNUSABLE_MEMORY:
1222 name = "reserved";
1223 flags |= IORESOURCE_DISABLED;
1224 break;
1226 case EFI_PERSISTENT_MEMORY:
1227 name = "Persistent Memory";
1228 desc = IORES_DESC_PERSISTENT_MEMORY;
1229 break;
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;
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;
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;
1253 if (insert_resource(&iomem_resource, res) < 0)
1254 kfree(res);
1255 else {
1257 * We don't know which region contains
1258 * kernel data so we try it repeatedly and
1259 * let the resource manager test it.
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
1274 #ifdef CONFIG_KEXEC
1275 /* find a block of memory aligned to 64M exclude reserved regions
1276 rsvd_regions are sorted
1278 unsigned long __init
1279 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
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;
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;
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;
1310 if (end > start + size)
1311 return start;
1314 printk(KERN_WARNING
1315 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1316 return ~0UL;
1318 #endif
1320 #ifdef CONFIG_CRASH_DUMP
1321 /* locate the size find a the descriptor at a certain address */
1322 unsigned long __init
1323 vmcore_find_descriptor_size (unsigned long address)
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;
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;
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;
1343 if (ret == 0)
1344 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1346 return ret;
1348 #endif
1350 char *efi_systab_show_arch(char *str)
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;