mm: hugetlb: fix hugepage memory leak caused by wrong reserve count
[linux/fpc-iii.git] / arch / ia64 / kernel / efi.c
blobcaae3f4e4341a40d86f4014871c1ebe929a937ae
1 /*
2 * Extensible Firmware Interface
4 * Based on Extensible Firmware Interface Specification version 0.9
5 * April 30, 1999
7 * Copyright (C) 1999 VA Linux Systems
8 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
9 * Copyright (C) 1999-2003 Hewlett-Packard Co.
10 * David Mosberger-Tang <davidm@hpl.hp.com>
11 * Stephane Eranian <eranian@hpl.hp.com>
12 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P.
13 * Bjorn Helgaas <bjorn.helgaas@hp.com>
15 * All EFI Runtime Services are not implemented yet as EFI only
16 * supports physical mode addressing on SoftSDV. This is to be fixed
17 * in a future version. --drummond 1999-07-20
19 * Implemented EFI runtime services and virtual mode calls. --davidm
21 * Goutham Rao: <goutham.rao@intel.com>
22 * Skip non-WB memory and ignore empty memory ranges.
24 #include <linux/module.h>
25 #include <linux/bootmem.h>
26 #include <linux/crash_dump.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/types.h>
30 #include <linux/slab.h>
31 #include <linux/time.h>
32 #include <linux/efi.h>
33 #include <linux/kexec.h>
34 #include <linux/mm.h>
36 #include <asm/io.h>
37 #include <asm/kregs.h>
38 #include <asm/meminit.h>
39 #include <asm/pgtable.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 static __initdata unsigned long palo_phys;
49 static __initdata efi_config_table_type_t arch_tables[] = {
50 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, "PALO", &palo_phys},
51 {NULL_GUID, NULL, 0},
54 extern efi_status_t efi_call_phys (void *, ...);
56 static efi_runtime_services_t *runtime;
57 static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
59 #define efi_call_virt(f, args...) (*(f))(args)
61 #define STUB_GET_TIME(prefix, adjust_arg) \
62 static efi_status_t \
63 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
64 { \
65 struct ia64_fpreg fr[6]; \
66 efi_time_cap_t *atc = NULL; \
67 efi_status_t ret; \
69 if (tc) \
70 atc = adjust_arg(tc); \
71 ia64_save_scratch_fpregs(fr); \
72 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
73 adjust_arg(tm), atc); \
74 ia64_load_scratch_fpregs(fr); \
75 return ret; \
78 #define STUB_SET_TIME(prefix, adjust_arg) \
79 static efi_status_t \
80 prefix##_set_time (efi_time_t *tm) \
81 { \
82 struct ia64_fpreg fr[6]; \
83 efi_status_t ret; \
85 ia64_save_scratch_fpregs(fr); \
86 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
87 adjust_arg(tm)); \
88 ia64_load_scratch_fpregs(fr); \
89 return ret; \
92 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
93 static efi_status_t \
94 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
95 efi_time_t *tm) \
96 { \
97 struct ia64_fpreg fr[6]; \
98 efi_status_t ret; \
100 ia64_save_scratch_fpregs(fr); \
101 ret = efi_call_##prefix( \
102 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
103 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
104 ia64_load_scratch_fpregs(fr); \
105 return ret; \
108 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
109 static efi_status_t \
110 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
112 struct ia64_fpreg fr[6]; \
113 efi_time_t *atm = NULL; \
114 efi_status_t ret; \
116 if (tm) \
117 atm = adjust_arg(tm); \
118 ia64_save_scratch_fpregs(fr); \
119 ret = efi_call_##prefix( \
120 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
121 enabled, atm); \
122 ia64_load_scratch_fpregs(fr); \
123 return ret; \
126 #define STUB_GET_VARIABLE(prefix, adjust_arg) \
127 static efi_status_t \
128 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
129 unsigned long *data_size, void *data) \
131 struct ia64_fpreg fr[6]; \
132 u32 *aattr = NULL; \
133 efi_status_t ret; \
135 if (attr) \
136 aattr = adjust_arg(attr); \
137 ia64_save_scratch_fpregs(fr); \
138 ret = efi_call_##prefix( \
139 (efi_get_variable_t *) __va(runtime->get_variable), \
140 adjust_arg(name), adjust_arg(vendor), aattr, \
141 adjust_arg(data_size), adjust_arg(data)); \
142 ia64_load_scratch_fpregs(fr); \
143 return ret; \
146 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
147 static efi_status_t \
148 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
149 efi_guid_t *vendor) \
151 struct ia64_fpreg fr[6]; \
152 efi_status_t ret; \
154 ia64_save_scratch_fpregs(fr); \
155 ret = efi_call_##prefix( \
156 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
157 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
158 ia64_load_scratch_fpregs(fr); \
159 return ret; \
162 #define STUB_SET_VARIABLE(prefix, adjust_arg) \
163 static efi_status_t \
164 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
165 u32 attr, unsigned long data_size, \
166 void *data) \
168 struct ia64_fpreg fr[6]; \
169 efi_status_t ret; \
171 ia64_save_scratch_fpregs(fr); \
172 ret = efi_call_##prefix( \
173 (efi_set_variable_t *) __va(runtime->set_variable), \
174 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
175 adjust_arg(data)); \
176 ia64_load_scratch_fpregs(fr); \
177 return ret; \
180 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
181 static efi_status_t \
182 prefix##_get_next_high_mono_count (u32 *count) \
184 struct ia64_fpreg fr[6]; \
185 efi_status_t ret; \
187 ia64_save_scratch_fpregs(fr); \
188 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
189 __va(runtime->get_next_high_mono_count), \
190 adjust_arg(count)); \
191 ia64_load_scratch_fpregs(fr); \
192 return ret; \
195 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \
196 static void \
197 prefix##_reset_system (int reset_type, efi_status_t status, \
198 unsigned long data_size, efi_char16_t *data) \
200 struct ia64_fpreg fr[6]; \
201 efi_char16_t *adata = NULL; \
203 if (data) \
204 adata = adjust_arg(data); \
206 ia64_save_scratch_fpregs(fr); \
207 efi_call_##prefix( \
208 (efi_reset_system_t *) __va(runtime->reset_system), \
209 reset_type, status, data_size, adata); \
210 /* should not return, but just in case... */ \
211 ia64_load_scratch_fpregs(fr); \
214 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
216 STUB_GET_TIME(phys, phys_ptr)
217 STUB_SET_TIME(phys, phys_ptr)
218 STUB_GET_WAKEUP_TIME(phys, phys_ptr)
219 STUB_SET_WAKEUP_TIME(phys, phys_ptr)
220 STUB_GET_VARIABLE(phys, phys_ptr)
221 STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
222 STUB_SET_VARIABLE(phys, phys_ptr)
223 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
224 STUB_RESET_SYSTEM(phys, phys_ptr)
226 #define id(arg) arg
228 STUB_GET_TIME(virt, id)
229 STUB_SET_TIME(virt, id)
230 STUB_GET_WAKEUP_TIME(virt, id)
231 STUB_SET_WAKEUP_TIME(virt, id)
232 STUB_GET_VARIABLE(virt, id)
233 STUB_GET_NEXT_VARIABLE(virt, id)
234 STUB_SET_VARIABLE(virt, id)
235 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
236 STUB_RESET_SYSTEM(virt, id)
238 void
239 efi_gettimeofday (struct timespec *ts)
241 efi_time_t tm;
243 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
244 memset(ts, 0, sizeof(*ts));
245 return;
248 ts->tv_sec = mktime(tm.year, tm.month, tm.day,
249 tm.hour, tm.minute, tm.second);
250 ts->tv_nsec = tm.nanosecond;
253 static int
254 is_memory_available (efi_memory_desc_t *md)
256 if (!(md->attribute & EFI_MEMORY_WB))
257 return 0;
259 switch (md->type) {
260 case EFI_LOADER_CODE:
261 case EFI_LOADER_DATA:
262 case EFI_BOOT_SERVICES_CODE:
263 case EFI_BOOT_SERVICES_DATA:
264 case EFI_CONVENTIONAL_MEMORY:
265 return 1;
267 return 0;
270 typedef struct kern_memdesc {
271 u64 attribute;
272 u64 start;
273 u64 num_pages;
274 } kern_memdesc_t;
276 static kern_memdesc_t *kern_memmap;
278 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
280 static inline u64
281 kmd_end(kern_memdesc_t *kmd)
283 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
286 static inline u64
287 efi_md_end(efi_memory_desc_t *md)
289 return (md->phys_addr + efi_md_size(md));
292 static inline int
293 efi_wb(efi_memory_desc_t *md)
295 return (md->attribute & EFI_MEMORY_WB);
298 static inline int
299 efi_uc(efi_memory_desc_t *md)
301 return (md->attribute & EFI_MEMORY_UC);
304 static void
305 walk (efi_freemem_callback_t callback, void *arg, u64 attr)
307 kern_memdesc_t *k;
308 u64 start, end, voff;
310 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
311 for (k = kern_memmap; k->start != ~0UL; k++) {
312 if (k->attribute != attr)
313 continue;
314 start = PAGE_ALIGN(k->start);
315 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
316 if (start < end)
317 if ((*callback)(start + voff, end + voff, arg) < 0)
318 return;
323 * Walk the EFI memory map and call CALLBACK once for each EFI memory
324 * descriptor that has memory that is available for OS use.
326 void
327 efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
329 walk(callback, arg, EFI_MEMORY_WB);
333 * Walk the EFI memory map and call CALLBACK once for each EFI memory
334 * descriptor that has memory that is available for uncached allocator.
336 void
337 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
339 walk(callback, arg, EFI_MEMORY_UC);
343 * Look for the PAL_CODE region reported by EFI and map it using an
344 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
345 * Abstraction Layer chapter 11 in ADAG
347 void *
348 efi_get_pal_addr (void)
350 void *efi_map_start, *efi_map_end, *p;
351 efi_memory_desc_t *md;
352 u64 efi_desc_size;
353 int pal_code_count = 0;
354 u64 vaddr, mask;
356 efi_map_start = __va(ia64_boot_param->efi_memmap);
357 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
358 efi_desc_size = ia64_boot_param->efi_memdesc_size;
360 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
361 md = p;
362 if (md->type != EFI_PAL_CODE)
363 continue;
365 if (++pal_code_count > 1) {
366 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
367 "dropped @ %llx\n", md->phys_addr);
368 continue;
371 * The only ITLB entry in region 7 that is used is the one
372 * installed by __start(). That entry covers a 64MB range.
374 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
375 vaddr = PAGE_OFFSET + md->phys_addr;
378 * We must check that the PAL mapping won't overlap with the
379 * kernel mapping.
381 * PAL code is guaranteed to be aligned on a power of 2 between
382 * 4k and 256KB and that only one ITR is needed to map it. This
383 * implies that the PAL code is always aligned on its size,
384 * i.e., the closest matching page size supported by the TLB.
385 * Therefore PAL code is guaranteed never to cross a 64MB unless
386 * it is bigger than 64MB (very unlikely!). So for now the
387 * following test is enough to determine whether or not we need
388 * a dedicated ITR for the PAL code.
390 if ((vaddr & mask) == (KERNEL_START & mask)) {
391 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
392 __func__);
393 continue;
396 if (efi_md_size(md) > IA64_GRANULE_SIZE)
397 panic("Whoa! PAL code size bigger than a granule!");
399 #if EFI_DEBUG
400 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
402 printk(KERN_INFO "CPU %d: mapping PAL code "
403 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
404 smp_processor_id(), md->phys_addr,
405 md->phys_addr + efi_md_size(md),
406 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
407 #endif
408 return __va(md->phys_addr);
410 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
411 __func__);
412 return NULL;
416 static u8 __init palo_checksum(u8 *buffer, u32 length)
418 u8 sum = 0;
419 u8 *end = buffer + length;
421 while (buffer < end)
422 sum = (u8) (sum + *(buffer++));
424 return sum;
428 * Parse and handle PALO table which is published at:
429 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
431 static void __init handle_palo(unsigned long phys_addr)
433 struct palo_table *palo = __va(phys_addr);
434 u8 checksum;
436 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
437 printk(KERN_INFO "PALO signature incorrect.\n");
438 return;
441 checksum = palo_checksum((u8 *)palo, palo->length);
442 if (checksum) {
443 printk(KERN_INFO "PALO checksum incorrect.\n");
444 return;
447 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
450 void
451 efi_map_pal_code (void)
453 void *pal_vaddr = efi_get_pal_addr ();
454 u64 psr;
456 if (!pal_vaddr)
457 return;
460 * Cannot write to CRx with PSR.ic=1
462 psr = ia64_clear_ic();
463 ia64_itr(0x1, IA64_TR_PALCODE,
464 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
465 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
466 IA64_GRANULE_SHIFT);
467 ia64_set_psr(psr); /* restore psr */
470 void __init
471 efi_init (void)
473 void *efi_map_start, *efi_map_end;
474 efi_char16_t *c16;
475 u64 efi_desc_size;
476 char *cp, vendor[100] = "unknown";
477 int i;
479 set_bit(EFI_BOOT, &efi.flags);
480 set_bit(EFI_64BIT, &efi.flags);
483 * It's too early to be able to use the standard kernel command line
484 * support...
486 for (cp = boot_command_line; *cp; ) {
487 if (memcmp(cp, "mem=", 4) == 0) {
488 mem_limit = memparse(cp + 4, &cp);
489 } else if (memcmp(cp, "max_addr=", 9) == 0) {
490 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
491 } else if (memcmp(cp, "min_addr=", 9) == 0) {
492 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
493 } else {
494 while (*cp != ' ' && *cp)
495 ++cp;
496 while (*cp == ' ')
497 ++cp;
500 if (min_addr != 0UL)
501 printk(KERN_INFO "Ignoring memory below %lluMB\n",
502 min_addr >> 20);
503 if (max_addr != ~0UL)
504 printk(KERN_INFO "Ignoring memory above %lluMB\n",
505 max_addr >> 20);
507 efi.systab = __va(ia64_boot_param->efi_systab);
510 * Verify the EFI Table
512 if (efi.systab == NULL)
513 panic("Whoa! Can't find EFI system table.\n");
514 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
515 panic("Whoa! EFI system table signature incorrect\n");
516 if ((efi.systab->hdr.revision >> 16) == 0)
517 printk(KERN_WARNING "Warning: EFI system table version "
518 "%d.%02d, expected 1.00 or greater\n",
519 efi.systab->hdr.revision >> 16,
520 efi.systab->hdr.revision & 0xffff);
522 /* Show what we know for posterity */
523 c16 = __va(efi.systab->fw_vendor);
524 if (c16) {
525 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
526 vendor[i] = *c16++;
527 vendor[i] = '\0';
530 printk(KERN_INFO "EFI v%u.%.02u by %s:",
531 efi.systab->hdr.revision >> 16,
532 efi.systab->hdr.revision & 0xffff, vendor);
534 set_bit(EFI_SYSTEM_TABLES, &efi.flags);
536 palo_phys = EFI_INVALID_TABLE_ADDR;
538 if (efi_config_init(arch_tables) != 0)
539 return;
541 if (palo_phys != EFI_INVALID_TABLE_ADDR)
542 handle_palo(palo_phys);
544 runtime = __va(efi.systab->runtime);
545 efi.get_time = phys_get_time;
546 efi.set_time = phys_set_time;
547 efi.get_wakeup_time = phys_get_wakeup_time;
548 efi.set_wakeup_time = phys_set_wakeup_time;
549 efi.get_variable = phys_get_variable;
550 efi.get_next_variable = phys_get_next_variable;
551 efi.set_variable = phys_set_variable;
552 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
553 efi.reset_system = phys_reset_system;
555 efi_map_start = __va(ia64_boot_param->efi_memmap);
556 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
557 efi_desc_size = ia64_boot_param->efi_memdesc_size;
559 #if EFI_DEBUG
560 /* print EFI memory map: */
562 efi_memory_desc_t *md;
563 void *p;
565 for (i = 0, p = efi_map_start; p < efi_map_end;
566 ++i, p += efi_desc_size)
568 const char *unit;
569 unsigned long size;
570 char buf[64];
572 md = p;
573 size = md->num_pages << EFI_PAGE_SHIFT;
575 if ((size >> 40) > 0) {
576 size >>= 40;
577 unit = "TB";
578 } else if ((size >> 30) > 0) {
579 size >>= 30;
580 unit = "GB";
581 } else if ((size >> 20) > 0) {
582 size >>= 20;
583 unit = "MB";
584 } else {
585 size >>= 10;
586 unit = "KB";
589 printk("mem%02d: %s "
590 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
591 i, efi_md_typeattr_format(buf, sizeof(buf), md),
592 md->phys_addr,
593 md->phys_addr + efi_md_size(md), size, unit);
596 #endif
598 efi_map_pal_code();
599 efi_enter_virtual_mode();
602 void
603 efi_enter_virtual_mode (void)
605 void *efi_map_start, *efi_map_end, *p;
606 efi_memory_desc_t *md;
607 efi_status_t status;
608 u64 efi_desc_size;
610 efi_map_start = __va(ia64_boot_param->efi_memmap);
611 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
612 efi_desc_size = ia64_boot_param->efi_memdesc_size;
614 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
615 md = p;
616 if (md->attribute & EFI_MEMORY_RUNTIME) {
618 * Some descriptors have multiple bits set, so the
619 * order of the tests is relevant.
621 if (md->attribute & EFI_MEMORY_WB) {
622 md->virt_addr = (u64) __va(md->phys_addr);
623 } else if (md->attribute & EFI_MEMORY_UC) {
624 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
625 } else if (md->attribute & EFI_MEMORY_WC) {
626 #if 0
627 md->virt_addr = ia64_remap(md->phys_addr,
628 (_PAGE_A |
629 _PAGE_P |
630 _PAGE_D |
631 _PAGE_MA_WC |
632 _PAGE_PL_0 |
633 _PAGE_AR_RW));
634 #else
635 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
636 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
637 #endif
638 } else if (md->attribute & EFI_MEMORY_WT) {
639 #if 0
640 md->virt_addr = ia64_remap(md->phys_addr,
641 (_PAGE_A |
642 _PAGE_P |
643 _PAGE_D |
644 _PAGE_MA_WT |
645 _PAGE_PL_0 |
646 _PAGE_AR_RW));
647 #else
648 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
649 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
650 #endif
655 status = efi_call_phys(__va(runtime->set_virtual_address_map),
656 ia64_boot_param->efi_memmap_size,
657 efi_desc_size,
658 ia64_boot_param->efi_memdesc_version,
659 ia64_boot_param->efi_memmap);
660 if (status != EFI_SUCCESS) {
661 printk(KERN_WARNING "warning: unable to switch EFI into "
662 "virtual mode (status=%lu)\n", status);
663 return;
666 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
669 * Now that EFI is in virtual mode, we call the EFI functions more
670 * efficiently:
672 efi.get_time = virt_get_time;
673 efi.set_time = virt_set_time;
674 efi.get_wakeup_time = virt_get_wakeup_time;
675 efi.set_wakeup_time = virt_set_wakeup_time;
676 efi.get_variable = virt_get_variable;
677 efi.get_next_variable = virt_get_next_variable;
678 efi.set_variable = virt_set_variable;
679 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
680 efi.reset_system = virt_reset_system;
684 * Walk the EFI memory map looking for the I/O port range. There can only be
685 * one entry of this type, other I/O port ranges should be described via ACPI.
688 efi_get_iobase (void)
690 void *efi_map_start, *efi_map_end, *p;
691 efi_memory_desc_t *md;
692 u64 efi_desc_size;
694 efi_map_start = __va(ia64_boot_param->efi_memmap);
695 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
696 efi_desc_size = ia64_boot_param->efi_memdesc_size;
698 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
699 md = p;
700 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
701 if (md->attribute & EFI_MEMORY_UC)
702 return md->phys_addr;
705 return 0;
708 static struct kern_memdesc *
709 kern_memory_descriptor (unsigned long phys_addr)
711 struct kern_memdesc *md;
713 for (md = kern_memmap; md->start != ~0UL; md++) {
714 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
715 return md;
717 return NULL;
720 static efi_memory_desc_t *
721 efi_memory_descriptor (unsigned long phys_addr)
723 void *efi_map_start, *efi_map_end, *p;
724 efi_memory_desc_t *md;
725 u64 efi_desc_size;
727 efi_map_start = __va(ia64_boot_param->efi_memmap);
728 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
729 efi_desc_size = ia64_boot_param->efi_memdesc_size;
731 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
732 md = p;
734 if (phys_addr - md->phys_addr < efi_md_size(md))
735 return md;
737 return NULL;
740 static int
741 efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
743 void *efi_map_start, *efi_map_end, *p;
744 efi_memory_desc_t *md;
745 u64 efi_desc_size;
746 unsigned long end;
748 efi_map_start = __va(ia64_boot_param->efi_memmap);
749 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
750 efi_desc_size = ia64_boot_param->efi_memdesc_size;
752 end = phys_addr + size;
754 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
755 md = p;
756 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
757 return 1;
759 return 0;
763 efi_mem_type (unsigned long phys_addr)
765 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
767 if (md)
768 return md->type;
769 return 0;
773 efi_mem_attributes (unsigned long phys_addr)
775 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
777 if (md)
778 return md->attribute;
779 return 0;
781 EXPORT_SYMBOL(efi_mem_attributes);
784 efi_mem_attribute (unsigned long phys_addr, unsigned long size)
786 unsigned long end = phys_addr + size;
787 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
788 u64 attr;
790 if (!md)
791 return 0;
794 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
795 * the kernel that firmware needs this region mapped.
797 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
798 do {
799 unsigned long md_end = efi_md_end(md);
801 if (end <= md_end)
802 return attr;
804 md = efi_memory_descriptor(md_end);
805 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
806 return 0;
807 } while (md);
808 return 0; /* never reached */
812 kern_mem_attribute (unsigned long phys_addr, unsigned long size)
814 unsigned long end = phys_addr + size;
815 struct kern_memdesc *md;
816 u64 attr;
819 * This is a hack for ioremap calls before we set up kern_memmap.
820 * Maybe we should do efi_memmap_init() earlier instead.
822 if (!kern_memmap) {
823 attr = efi_mem_attribute(phys_addr, size);
824 if (attr & EFI_MEMORY_WB)
825 return EFI_MEMORY_WB;
826 return 0;
829 md = kern_memory_descriptor(phys_addr);
830 if (!md)
831 return 0;
833 attr = md->attribute;
834 do {
835 unsigned long md_end = kmd_end(md);
837 if (end <= md_end)
838 return attr;
840 md = kern_memory_descriptor(md_end);
841 if (!md || md->attribute != attr)
842 return 0;
843 } while (md);
844 return 0; /* never reached */
846 EXPORT_SYMBOL(kern_mem_attribute);
849 valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
851 u64 attr;
854 * /dev/mem reads and writes use copy_to_user(), which implicitly
855 * uses a granule-sized kernel identity mapping. It's really
856 * only safe to do this for regions in kern_memmap. For more
857 * details, see Documentation/ia64/aliasing.txt.
859 attr = kern_mem_attribute(phys_addr, size);
860 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
861 return 1;
862 return 0;
866 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
868 unsigned long phys_addr = pfn << PAGE_SHIFT;
869 u64 attr;
871 attr = efi_mem_attribute(phys_addr, size);
874 * /dev/mem mmap uses normal user pages, so we don't need the entire
875 * granule, but the entire region we're mapping must support the same
876 * attribute.
878 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
879 return 1;
882 * Intel firmware doesn't tell us about all the MMIO regions, so
883 * in general we have to allow mmap requests. But if EFI *does*
884 * tell us about anything inside this region, we should deny it.
885 * The user can always map a smaller region to avoid the overlap.
887 if (efi_memmap_intersects(phys_addr, size))
888 return 0;
890 return 1;
893 pgprot_t
894 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
895 pgprot_t vma_prot)
897 unsigned long phys_addr = pfn << PAGE_SHIFT;
898 u64 attr;
901 * For /dev/mem mmap, we use user mappings, but if the region is
902 * in kern_memmap (and hence may be covered by a kernel mapping),
903 * we must use the same attribute as the kernel mapping.
905 attr = kern_mem_attribute(phys_addr, size);
906 if (attr & EFI_MEMORY_WB)
907 return pgprot_cacheable(vma_prot);
908 else if (attr & EFI_MEMORY_UC)
909 return pgprot_noncached(vma_prot);
912 * Some chipsets don't support UC access to memory. If
913 * WB is supported, we prefer that.
915 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
916 return pgprot_cacheable(vma_prot);
918 return pgprot_noncached(vma_prot);
921 int __init
922 efi_uart_console_only(void)
924 efi_status_t status;
925 char *s, name[] = "ConOut";
926 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
927 efi_char16_t *utf16, name_utf16[32];
928 unsigned char data[1024];
929 unsigned long size = sizeof(data);
930 struct efi_generic_dev_path *hdr, *end_addr;
931 int uart = 0;
933 /* Convert to UTF-16 */
934 utf16 = name_utf16;
935 s = name;
936 while (*s)
937 *utf16++ = *s++ & 0x7f;
938 *utf16 = 0;
940 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
941 if (status != EFI_SUCCESS) {
942 printk(KERN_ERR "No EFI %s variable?\n", name);
943 return 0;
946 hdr = (struct efi_generic_dev_path *) data;
947 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
948 while (hdr < end_addr) {
949 if (hdr->type == EFI_DEV_MSG &&
950 hdr->sub_type == EFI_DEV_MSG_UART)
951 uart = 1;
952 else if (hdr->type == EFI_DEV_END_PATH ||
953 hdr->type == EFI_DEV_END_PATH2) {
954 if (!uart)
955 return 0;
956 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
957 return 1;
958 uart = 0;
960 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
962 printk(KERN_ERR "Malformed %s value\n", name);
963 return 0;
967 * Look for the first granule aligned memory descriptor memory
968 * that is big enough to hold EFI memory map. Make sure this
969 * descriptor is atleast granule sized so it does not get trimmed
971 struct kern_memdesc *
972 find_memmap_space (void)
974 u64 contig_low=0, contig_high=0;
975 u64 as = 0, ae;
976 void *efi_map_start, *efi_map_end, *p, *q;
977 efi_memory_desc_t *md, *pmd = NULL, *check_md;
978 u64 space_needed, efi_desc_size;
979 unsigned long total_mem = 0;
981 efi_map_start = __va(ia64_boot_param->efi_memmap);
982 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
983 efi_desc_size = ia64_boot_param->efi_memdesc_size;
986 * Worst case: we need 3 kernel descriptors for each efi descriptor
987 * (if every entry has a WB part in the middle, and UC head and tail),
988 * plus one for the end marker.
990 space_needed = sizeof(kern_memdesc_t) *
991 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
993 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
994 md = p;
995 if (!efi_wb(md)) {
996 continue;
998 if (pmd == NULL || !efi_wb(pmd) ||
999 efi_md_end(pmd) != md->phys_addr) {
1000 contig_low = GRANULEROUNDUP(md->phys_addr);
1001 contig_high = efi_md_end(md);
1002 for (q = p + efi_desc_size; q < efi_map_end;
1003 q += efi_desc_size) {
1004 check_md = q;
1005 if (!efi_wb(check_md))
1006 break;
1007 if (contig_high != check_md->phys_addr)
1008 break;
1009 contig_high = efi_md_end(check_md);
1011 contig_high = GRANULEROUNDDOWN(contig_high);
1013 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1014 continue;
1016 /* Round ends inward to granule boundaries */
1017 as = max(contig_low, md->phys_addr);
1018 ae = min(contig_high, efi_md_end(md));
1020 /* keep within max_addr= and min_addr= command line arg */
1021 as = max(as, min_addr);
1022 ae = min(ae, max_addr);
1023 if (ae <= as)
1024 continue;
1026 /* avoid going over mem= command line arg */
1027 if (total_mem + (ae - as) > mem_limit)
1028 ae -= total_mem + (ae - as) - mem_limit;
1030 if (ae <= as)
1031 continue;
1033 if (ae - as > space_needed)
1034 break;
1036 if (p >= efi_map_end)
1037 panic("Can't allocate space for kernel memory descriptors");
1039 return __va(as);
1043 * Walk the EFI memory map and gather all memory available for kernel
1044 * to use. We can allocate partial granules only if the unavailable
1045 * parts exist, and are WB.
1047 unsigned long
1048 efi_memmap_init(u64 *s, u64 *e)
1050 struct kern_memdesc *k, *prev = NULL;
1051 u64 contig_low=0, contig_high=0;
1052 u64 as, ae, lim;
1053 void *efi_map_start, *efi_map_end, *p, *q;
1054 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1055 u64 efi_desc_size;
1056 unsigned long total_mem = 0;
1058 k = kern_memmap = find_memmap_space();
1060 efi_map_start = __va(ia64_boot_param->efi_memmap);
1061 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1062 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1064 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1065 md = p;
1066 if (!efi_wb(md)) {
1067 if (efi_uc(md) &&
1068 (md->type == EFI_CONVENTIONAL_MEMORY ||
1069 md->type == EFI_BOOT_SERVICES_DATA)) {
1070 k->attribute = EFI_MEMORY_UC;
1071 k->start = md->phys_addr;
1072 k->num_pages = md->num_pages;
1073 k++;
1075 continue;
1077 if (pmd == NULL || !efi_wb(pmd) ||
1078 efi_md_end(pmd) != md->phys_addr) {
1079 contig_low = GRANULEROUNDUP(md->phys_addr);
1080 contig_high = efi_md_end(md);
1081 for (q = p + efi_desc_size; q < efi_map_end;
1082 q += efi_desc_size) {
1083 check_md = q;
1084 if (!efi_wb(check_md))
1085 break;
1086 if (contig_high != check_md->phys_addr)
1087 break;
1088 contig_high = efi_md_end(check_md);
1090 contig_high = GRANULEROUNDDOWN(contig_high);
1092 if (!is_memory_available(md))
1093 continue;
1096 * Round ends inward to granule boundaries
1097 * Give trimmings to uncached allocator
1099 if (md->phys_addr < contig_low) {
1100 lim = min(efi_md_end(md), contig_low);
1101 if (efi_uc(md)) {
1102 if (k > kern_memmap &&
1103 (k-1)->attribute == EFI_MEMORY_UC &&
1104 kmd_end(k-1) == md->phys_addr) {
1105 (k-1)->num_pages +=
1106 (lim - md->phys_addr)
1107 >> EFI_PAGE_SHIFT;
1108 } else {
1109 k->attribute = EFI_MEMORY_UC;
1110 k->start = md->phys_addr;
1111 k->num_pages = (lim - md->phys_addr)
1112 >> EFI_PAGE_SHIFT;
1113 k++;
1116 as = contig_low;
1117 } else
1118 as = md->phys_addr;
1120 if (efi_md_end(md) > contig_high) {
1121 lim = max(md->phys_addr, contig_high);
1122 if (efi_uc(md)) {
1123 if (lim == md->phys_addr && k > kern_memmap &&
1124 (k-1)->attribute == EFI_MEMORY_UC &&
1125 kmd_end(k-1) == md->phys_addr) {
1126 (k-1)->num_pages += md->num_pages;
1127 } else {
1128 k->attribute = EFI_MEMORY_UC;
1129 k->start = lim;
1130 k->num_pages = (efi_md_end(md) - lim)
1131 >> EFI_PAGE_SHIFT;
1132 k++;
1135 ae = contig_high;
1136 } else
1137 ae = efi_md_end(md);
1139 /* keep within max_addr= and min_addr= command line arg */
1140 as = max(as, min_addr);
1141 ae = min(ae, max_addr);
1142 if (ae <= as)
1143 continue;
1145 /* avoid going over mem= command line arg */
1146 if (total_mem + (ae - as) > mem_limit)
1147 ae -= total_mem + (ae - as) - mem_limit;
1149 if (ae <= as)
1150 continue;
1151 if (prev && kmd_end(prev) == md->phys_addr) {
1152 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1153 total_mem += ae - as;
1154 continue;
1156 k->attribute = EFI_MEMORY_WB;
1157 k->start = as;
1158 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1159 total_mem += ae - as;
1160 prev = k++;
1162 k->start = ~0L; /* end-marker */
1164 /* reserve the memory we are using for kern_memmap */
1165 *s = (u64)kern_memmap;
1166 *e = (u64)++k;
1168 return total_mem;
1171 void
1172 efi_initialize_iomem_resources(struct resource *code_resource,
1173 struct resource *data_resource,
1174 struct resource *bss_resource)
1176 struct resource *res;
1177 void *efi_map_start, *efi_map_end, *p;
1178 efi_memory_desc_t *md;
1179 u64 efi_desc_size;
1180 char *name;
1181 unsigned long flags;
1183 efi_map_start = __va(ia64_boot_param->efi_memmap);
1184 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1185 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1187 res = NULL;
1189 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1190 md = p;
1192 if (md->num_pages == 0) /* should not happen */
1193 continue;
1195 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1196 switch (md->type) {
1198 case EFI_MEMORY_MAPPED_IO:
1199 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1200 continue;
1202 case EFI_LOADER_CODE:
1203 case EFI_LOADER_DATA:
1204 case EFI_BOOT_SERVICES_DATA:
1205 case EFI_BOOT_SERVICES_CODE:
1206 case EFI_CONVENTIONAL_MEMORY:
1207 if (md->attribute & EFI_MEMORY_WP) {
1208 name = "System ROM";
1209 flags |= IORESOURCE_READONLY;
1210 } else if (md->attribute == EFI_MEMORY_UC)
1211 name = "Uncached RAM";
1212 else
1213 name = "System RAM";
1214 break;
1216 case EFI_ACPI_MEMORY_NVS:
1217 name = "ACPI Non-volatile Storage";
1218 break;
1220 case EFI_UNUSABLE_MEMORY:
1221 name = "reserved";
1222 flags |= IORESOURCE_DISABLED;
1223 break;
1225 case EFI_PERSISTENT_MEMORY:
1226 name = "Persistent Memory";
1227 break;
1229 case EFI_RESERVED_TYPE:
1230 case EFI_RUNTIME_SERVICES_CODE:
1231 case EFI_RUNTIME_SERVICES_DATA:
1232 case EFI_ACPI_RECLAIM_MEMORY:
1233 default:
1234 name = "reserved";
1235 break;
1238 if ((res = kzalloc(sizeof(struct resource),
1239 GFP_KERNEL)) == NULL) {
1240 printk(KERN_ERR
1241 "failed to allocate resource for iomem\n");
1242 return;
1245 res->name = name;
1246 res->start = md->phys_addr;
1247 res->end = md->phys_addr + efi_md_size(md) - 1;
1248 res->flags = flags;
1250 if (insert_resource(&iomem_resource, res) < 0)
1251 kfree(res);
1252 else {
1254 * We don't know which region contains
1255 * kernel data so we try it repeatedly and
1256 * let the resource manager test it.
1258 insert_resource(res, code_resource);
1259 insert_resource(res, data_resource);
1260 insert_resource(res, bss_resource);
1261 #ifdef CONFIG_KEXEC
1262 insert_resource(res, &efi_memmap_res);
1263 insert_resource(res, &boot_param_res);
1264 if (crashk_res.end > crashk_res.start)
1265 insert_resource(res, &crashk_res);
1266 #endif
1271 #ifdef CONFIG_KEXEC
1272 /* find a block of memory aligned to 64M exclude reserved regions
1273 rsvd_regions are sorted
1275 unsigned long __init
1276 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1278 int i;
1279 u64 start, end;
1280 u64 alignment = 1UL << _PAGE_SIZE_64M;
1281 void *efi_map_start, *efi_map_end, *p;
1282 efi_memory_desc_t *md;
1283 u64 efi_desc_size;
1285 efi_map_start = __va(ia64_boot_param->efi_memmap);
1286 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1287 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1289 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1290 md = p;
1291 if (!efi_wb(md))
1292 continue;
1293 start = ALIGN(md->phys_addr, alignment);
1294 end = efi_md_end(md);
1295 for (i = 0; i < n; i++) {
1296 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1297 if (__pa(r[i].start) > start + size)
1298 return start;
1299 start = ALIGN(__pa(r[i].end), alignment);
1300 if (i < n-1 &&
1301 __pa(r[i+1].start) < start + size)
1302 continue;
1303 else
1304 break;
1307 if (end > start + size)
1308 return start;
1311 printk(KERN_WARNING
1312 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1313 return ~0UL;
1315 #endif
1317 #ifdef CONFIG_CRASH_DUMP
1318 /* locate the size find a the descriptor at a certain address */
1319 unsigned long __init
1320 vmcore_find_descriptor_size (unsigned long address)
1322 void *efi_map_start, *efi_map_end, *p;
1323 efi_memory_desc_t *md;
1324 u64 efi_desc_size;
1325 unsigned long ret = 0;
1327 efi_map_start = __va(ia64_boot_param->efi_memmap);
1328 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1329 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1331 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1332 md = p;
1333 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1334 && md->phys_addr == address) {
1335 ret = efi_md_size(md);
1336 break;
1340 if (ret == 0)
1341 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1343 return ret;
1345 #endif