1/*
2 * Extensible Firmware Interface
3 *
4 * Based on Extensible Firmware Interface Specification version 0.9
5 * April 30, 1999
6 *
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>
14 *
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
18 *
19 * Implemented EFI runtime services and virtual mode calls.  --davidm
20 *
21 * Goutham Rao: <goutham.rao@intel.com>
22 *	Skip non-WB memory and ignore empty memory ranges.
23 */
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>
35
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>
44
45#define EFI_DEBUG	0
46
47static __initdata unsigned long palo_phys;
48
49static __initdata efi_config_table_type_t arch_tables[] = {
50	{PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, "PALO", &palo_phys},
51	{NULL_GUID, NULL, 0},
52};
53
54extern efi_status_t efi_call_phys (void *, ...);
55
56static efi_runtime_services_t *runtime;
57static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
58
59#define efi_call_virt(f, args...)	(*(f))(args)
60
61#define STUB_GET_TIME(prefix, adjust_arg)				       \
62static efi_status_t							       \
63prefix##_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;						       \
68									       \
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;							       \
76}
77
78#define STUB_SET_TIME(prefix, adjust_arg)				       \
79static efi_status_t							       \
80prefix##_set_time (efi_time_t *tm)					       \
81{									       \
82	struct ia64_fpreg fr[6];					       \
83	efi_status_t ret;						       \
84									       \
85	ia64_save_scratch_fpregs(fr);					       \
86	ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time),    \
87				adjust_arg(tm));			       \
88	ia64_load_scratch_fpregs(fr);					       \
89	return ret;							       \
90}
91
92#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg)			       \
93static efi_status_t							       \
94prefix##_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;						       \
99									       \
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;							       \
106}
107
108#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg)			       \
109static efi_status_t							       \
110prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm)		       \
111{									       \
112	struct ia64_fpreg fr[6];					       \
113	efi_time_t *atm = NULL;						       \
114	efi_status_t ret;						       \
115									       \
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;							       \
124}
125
126#define STUB_GET_VARIABLE(prefix, adjust_arg)				       \
127static efi_status_t							       \
128prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr,      \
129		       unsigned long *data_size, void *data)		       \
130{									       \
131	struct ia64_fpreg fr[6];					       \
132	u32 *aattr = NULL;						       \
133	efi_status_t ret;						       \
134									       \
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;							       \
144}
145
146#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg)			       \
147static efi_status_t							       \
148prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name,      \
149			    efi_guid_t *vendor)				       \
150{									       \
151	struct ia64_fpreg fr[6];					       \
152	efi_status_t ret;						       \
153									       \
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;							       \
160}
161
162#define STUB_SET_VARIABLE(prefix, adjust_arg)				       \
163static efi_status_t							       \
164prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor,		       \
165		       u32 attr, unsigned long data_size,		       \
166		       void *data)					       \
167{									       \
168	struct ia64_fpreg fr[6];					       \
169	efi_status_t ret;						       \
170									       \
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;							       \
178}
179
180#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg)		       \
181static efi_status_t							       \
182prefix##_get_next_high_mono_count (u32 *count)				       \
183{									       \
184	struct ia64_fpreg fr[6];					       \
185	efi_status_t ret;						       \
186									       \
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;							       \
193}
194
195#define STUB_RESET_SYSTEM(prefix, adjust_arg)				       \
196static void								       \
197prefix##_reset_system (int reset_type, efi_status_t status,		       \
198		       unsigned long data_size, efi_char16_t *data)	       \
199{									       \
200	struct ia64_fpreg fr[6];					       \
201	efi_char16_t *adata = NULL;					       \
202									       \
203	if (data)							       \
204		adata = adjust_arg(data);				       \
205									       \
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);					       \
212}
213
214#define phys_ptr(arg)	((__typeof__(arg)) ia64_tpa(arg))
215
216STUB_GET_TIME(phys, phys_ptr)
217STUB_SET_TIME(phys, phys_ptr)
218STUB_GET_WAKEUP_TIME(phys, phys_ptr)
219STUB_SET_WAKEUP_TIME(phys, phys_ptr)
220STUB_GET_VARIABLE(phys, phys_ptr)
221STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
222STUB_SET_VARIABLE(phys, phys_ptr)
223STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
224STUB_RESET_SYSTEM(phys, phys_ptr)
225
226#define id(arg)	arg
227
228STUB_GET_TIME(virt, id)
229STUB_SET_TIME(virt, id)
230STUB_GET_WAKEUP_TIME(virt, id)
231STUB_SET_WAKEUP_TIME(virt, id)
232STUB_GET_VARIABLE(virt, id)
233STUB_GET_NEXT_VARIABLE(virt, id)
234STUB_SET_VARIABLE(virt, id)
235STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
236STUB_RESET_SYSTEM(virt, id)
237
238void
239efi_gettimeofday (struct timespec *ts)
240{
241	efi_time_t tm;
242
243	if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
244		memset(ts, 0, sizeof(*ts));
245		return;
246	}
247
248	ts->tv_sec = mktime(tm.year, tm.month, tm.day,
249			    tm.hour, tm.minute, tm.second);
250	ts->tv_nsec = tm.nanosecond;
251}
252
253static int
254is_memory_available (efi_memory_desc_t *md)
255{
256	if (!(md->attribute & EFI_MEMORY_WB))
257		return 0;
258
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;
266	}
267	return 0;
268}
269
270typedef struct kern_memdesc {
271	u64 attribute;
272	u64 start;
273	u64 num_pages;
274} kern_memdesc_t;
275
276static kern_memdesc_t *kern_memmap;
277
278#define efi_md_size(md)	(md->num_pages << EFI_PAGE_SHIFT)
279
280static inline u64
281kmd_end(kern_memdesc_t *kmd)
282{
283	return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
284}
285
286static inline u64
287efi_md_end(efi_memory_desc_t *md)
288{
289	return (md->phys_addr + efi_md_size(md));
290}
291
292static inline int
293efi_wb(efi_memory_desc_t *md)
294{
295	return (md->attribute & EFI_MEMORY_WB);
296}
297
298static inline int
299efi_uc(efi_memory_desc_t *md)
300{
301	return (md->attribute & EFI_MEMORY_UC);
302}
303
304static void
305walk (efi_freemem_callback_t callback, void *arg, u64 attr)
306{
307	kern_memdesc_t *k;
308	u64 start, end, voff;
309
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;
319	}
320}
321
322/*
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.
325 */
326void
327efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
328{
329	walk(callback, arg, EFI_MEMORY_WB);
330}
331
332/*
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.
335 */
336void
337efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
338{
339	walk(callback, arg, EFI_MEMORY_UC);
340}
341
342/*
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
346 */
347void *
348efi_get_pal_addr (void)
349{
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;
355
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;
359
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;
364
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;
369		}
370		/*
371		 * The only ITLB entry in region 7 that is used is the one
372		 * installed by __start().  That entry covers a 64MB range.
373		 */
374		mask  = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
375		vaddr = PAGE_OFFSET + md->phys_addr;
376
377		/*
378		 * We must check that the PAL mapping won't overlap with the
379		 * kernel mapping.
380		 *
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.
389		 */
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;
394		}
395
396		if (efi_md_size(md) > IA64_GRANULE_SIZE)
397			panic("Whoa!  PAL code size bigger than a granule!");
398
399#if EFI_DEBUG
400		mask  = ~((1 << IA64_GRANULE_SHIFT) - 1);
401
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);
409	}
410	printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
411	       __func__);
412	return NULL;
413}
414
415
416static u8 __init palo_checksum(u8 *buffer, u32 length)
417{
418	u8 sum = 0;
419	u8 *end = buffer + length;
420
421	while (buffer < end)
422		sum = (u8) (sum + *(buffer++));
423
424	return sum;
425}
426
427/*
428 * Parse and handle PALO table which is published at:
429 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
430 */
431static void __init handle_palo(unsigned long phys_addr)
432{
433	struct palo_table *palo = __va(phys_addr);
434	u8  checksum;
435
436	if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
437		printk(KERN_INFO "PALO signature incorrect.\n");
438		return;
439	}
440
441	checksum = palo_checksum((u8 *)palo, palo->length);
442	if (checksum) {
443		printk(KERN_INFO "PALO checksum incorrect.\n");
444		return;
445	}
446
447	setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
448}
449
450void
451efi_map_pal_code (void)
452{
453	void *pal_vaddr = efi_get_pal_addr ();
454	u64 psr;
455
456	if (!pal_vaddr)
457		return;
458
459	/*
460	 * Cannot write to CRx with PSR.ic=1
461	 */
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	paravirt_dv_serialize_data();
468	ia64_set_psr(psr);		/* restore psr */
469}
470
471void __init
472efi_init (void)
473{
474	void *efi_map_start, *efi_map_end;
475	efi_char16_t *c16;
476	u64 efi_desc_size;
477	char *cp, vendor[100] = "unknown";
478	int i;
479
480	set_bit(EFI_BOOT, &efi.flags);
481	set_bit(EFI_64BIT, &efi.flags);
482
483	/*
484	 * It's too early to be able to use the standard kernel command line
485	 * support...
486	 */
487	for (cp = boot_command_line; *cp; ) {
488		if (memcmp(cp, "mem=", 4) == 0) {
489			mem_limit = memparse(cp + 4, &cp);
490		} else if (memcmp(cp, "max_addr=", 9) == 0) {
491			max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
492		} else if (memcmp(cp, "min_addr=", 9) == 0) {
493			min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
494		} else {
495			while (*cp != ' ' && *cp)
496				++cp;
497			while (*cp == ' ')
498				++cp;
499		}
500	}
501	if (min_addr != 0UL)
502		printk(KERN_INFO "Ignoring memory below %lluMB\n",
503		       min_addr >> 20);
504	if (max_addr != ~0UL)
505		printk(KERN_INFO "Ignoring memory above %lluMB\n",
506		       max_addr >> 20);
507
508	efi.systab = __va(ia64_boot_param->efi_systab);
509
510	/*
511	 * Verify the EFI Table
512	 */
513	if (efi.systab == NULL)
514		panic("Whoa! Can't find EFI system table.\n");
515	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
516		panic("Whoa! EFI system table signature incorrect\n");
517	if ((efi.systab->hdr.revision >> 16) == 0)
518		printk(KERN_WARNING "Warning: EFI system table version "
519		       "%d.%02d, expected 1.00 or greater\n",
520		       efi.systab->hdr.revision >> 16,
521		       efi.systab->hdr.revision & 0xffff);
522
523	/* Show what we know for posterity */
524	c16 = __va(efi.systab->fw_vendor);
525	if (c16) {
526		for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
527			vendor[i] = *c16++;
528		vendor[i] = '\0';
529	}
530
531	printk(KERN_INFO "EFI v%u.%.02u by %s:",
532	       efi.systab->hdr.revision >> 16,
533	       efi.systab->hdr.revision & 0xffff, vendor);
534
535	set_bit(EFI_SYSTEM_TABLES, &efi.flags);
536
537	palo_phys      = EFI_INVALID_TABLE_ADDR;
538
539	if (efi_config_init(arch_tables) != 0)
540		return;
541
542	if (palo_phys != EFI_INVALID_TABLE_ADDR)
543		handle_palo(palo_phys);
544
545	runtime = __va(efi.systab->runtime);
546	efi.get_time = phys_get_time;
547	efi.set_time = phys_set_time;
548	efi.get_wakeup_time = phys_get_wakeup_time;
549	efi.set_wakeup_time = phys_set_wakeup_time;
550	efi.get_variable = phys_get_variable;
551	efi.get_next_variable = phys_get_next_variable;
552	efi.set_variable = phys_set_variable;
553	efi.get_next_high_mono_count = phys_get_next_high_mono_count;
554	efi.reset_system = phys_reset_system;
555
556	efi_map_start = __va(ia64_boot_param->efi_memmap);
557	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
558	efi_desc_size = ia64_boot_param->efi_memdesc_size;
559
560#if EFI_DEBUG
561	/* print EFI memory map: */
562	{
563		efi_memory_desc_t *md;
564		void *p;
565
566		for (i = 0, p = efi_map_start; p < efi_map_end;
567		     ++i, p += efi_desc_size)
568		{
569			const char *unit;
570			unsigned long size;
571			char buf[64];
572
573			md = p;
574			size = md->num_pages << EFI_PAGE_SHIFT;
575
576			if ((size >> 40) > 0) {
577				size >>= 40;
578				unit = "TB";
579			} else if ((size >> 30) > 0) {
580				size >>= 30;
581				unit = "GB";
582			} else if ((size >> 20) > 0) {
583				size >>= 20;
584				unit = "MB";
585			} else {
586				size >>= 10;
587				unit = "KB";
588			}
589
590			printk("mem%02d: %s "
591			       "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
592			       i, efi_md_typeattr_format(buf, sizeof(buf), md),
593			       md->phys_addr,
594			       md->phys_addr + efi_md_size(md), size, unit);
595		}
596	}
597#endif
598
599	efi_map_pal_code();
600	efi_enter_virtual_mode();
601}
602
603void
604efi_enter_virtual_mode (void)
605{
606	void *efi_map_start, *efi_map_end, *p;
607	efi_memory_desc_t *md;
608	efi_status_t status;
609	u64 efi_desc_size;
610
611	efi_map_start = __va(ia64_boot_param->efi_memmap);
612	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
613	efi_desc_size = ia64_boot_param->efi_memdesc_size;
614
615	for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
616		md = p;
617		if (md->attribute & EFI_MEMORY_RUNTIME) {
618			/*
619			 * Some descriptors have multiple bits set, so the
620			 * order of the tests is relevant.
621			 */
622			if (md->attribute & EFI_MEMORY_WB) {
623				md->virt_addr = (u64) __va(md->phys_addr);
624			} else if (md->attribute & EFI_MEMORY_UC) {
625				md->virt_addr = (u64) ioremap(md->phys_addr, 0);
626			} else if (md->attribute & EFI_MEMORY_WC) {
627#if 0
628				md->virt_addr = ia64_remap(md->phys_addr,
629							   (_PAGE_A |
630							    _PAGE_P |
631							    _PAGE_D |
632							    _PAGE_MA_WC |
633							    _PAGE_PL_0 |
634							    _PAGE_AR_RW));
635#else
636				printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
637				md->virt_addr = (u64) ioremap(md->phys_addr, 0);
638#endif
639			} else if (md->attribute & EFI_MEMORY_WT) {
640#if 0
641				md->virt_addr = ia64_remap(md->phys_addr,
642							   (_PAGE_A |
643							    _PAGE_P |
644							    _PAGE_D |
645							    _PAGE_MA_WT |
646							    _PAGE_PL_0 |
647							    _PAGE_AR_RW));
648#else
649				printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
650				md->virt_addr = (u64) ioremap(md->phys_addr, 0);
651#endif
652			}
653		}
654	}
655
656	status = efi_call_phys(__va(runtime->set_virtual_address_map),
657			       ia64_boot_param->efi_memmap_size,
658			       efi_desc_size,
659			       ia64_boot_param->efi_memdesc_version,
660			       ia64_boot_param->efi_memmap);
661	if (status != EFI_SUCCESS) {
662		printk(KERN_WARNING "warning: unable to switch EFI into "
663		       "virtual mode (status=%lu)\n", status);
664		return;
665	}
666
667	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
668
669	/*
670	 * Now that EFI is in virtual mode, we call the EFI functions more
671	 * efficiently:
672	 */
673	efi.get_time = virt_get_time;
674	efi.set_time = virt_set_time;
675	efi.get_wakeup_time = virt_get_wakeup_time;
676	efi.set_wakeup_time = virt_set_wakeup_time;
677	efi.get_variable = virt_get_variable;
678	efi.get_next_variable = virt_get_next_variable;
679	efi.set_variable = virt_set_variable;
680	efi.get_next_high_mono_count = virt_get_next_high_mono_count;
681	efi.reset_system = virt_reset_system;
682}
683
684/*
685 * Walk the EFI memory map looking for the I/O port range.  There can only be
686 * one entry of this type, other I/O port ranges should be described via ACPI.
687 */
688u64
689efi_get_iobase (void)
690{
691	void *efi_map_start, *efi_map_end, *p;
692	efi_memory_desc_t *md;
693	u64 efi_desc_size;
694
695	efi_map_start = __va(ia64_boot_param->efi_memmap);
696	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
697	efi_desc_size = ia64_boot_param->efi_memdesc_size;
698
699	for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
700		md = p;
701		if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
702			if (md->attribute & EFI_MEMORY_UC)
703				return md->phys_addr;
704		}
705	}
706	return 0;
707}
708
709static struct kern_memdesc *
710kern_memory_descriptor (unsigned long phys_addr)
711{
712	struct kern_memdesc *md;
713
714	for (md = kern_memmap; md->start != ~0UL; md++) {
715		if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
716			 return md;
717	}
718	return NULL;
719}
720
721static efi_memory_desc_t *
722efi_memory_descriptor (unsigned long phys_addr)
723{
724	void *efi_map_start, *efi_map_end, *p;
725	efi_memory_desc_t *md;
726	u64 efi_desc_size;
727
728	efi_map_start = __va(ia64_boot_param->efi_memmap);
729	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
730	efi_desc_size = ia64_boot_param->efi_memdesc_size;
731
732	for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
733		md = p;
734
735		if (phys_addr - md->phys_addr < efi_md_size(md))
736			 return md;
737	}
738	return NULL;
739}
740
741static int
742efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
743{
744	void *efi_map_start, *efi_map_end, *p;
745	efi_memory_desc_t *md;
746	u64 efi_desc_size;
747	unsigned long end;
748
749	efi_map_start = __va(ia64_boot_param->efi_memmap);
750	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
751	efi_desc_size = ia64_boot_param->efi_memdesc_size;
752
753	end = phys_addr + size;
754
755	for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
756		md = p;
757		if (md->phys_addr < end && efi_md_end(md) > phys_addr)
758			return 1;
759	}
760	return 0;
761}
762
763u32
764efi_mem_type (unsigned long phys_addr)
765{
766	efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
767
768	if (md)
769		return md->type;
770	return 0;
771}
772
773u64
774efi_mem_attributes (unsigned long phys_addr)
775{
776	efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
777
778	if (md)
779		return md->attribute;
780	return 0;
781}
782EXPORT_SYMBOL(efi_mem_attributes);
783
784u64
785efi_mem_attribute (unsigned long phys_addr, unsigned long size)
786{
787	unsigned long end = phys_addr + size;
788	efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
789	u64 attr;
790
791	if (!md)
792		return 0;
793
794	/*
795	 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
796	 * the kernel that firmware needs this region mapped.
797	 */
798	attr = md->attribute & ~EFI_MEMORY_RUNTIME;
799	do {
800		unsigned long md_end = efi_md_end(md);
801
802		if (end <= md_end)
803			return attr;
804
805		md = efi_memory_descriptor(md_end);
806		if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
807			return 0;
808	} while (md);
809	return 0;	/* never reached */
810}
811
812u64
813kern_mem_attribute (unsigned long phys_addr, unsigned long size)
814{
815	unsigned long end = phys_addr + size;
816	struct kern_memdesc *md;
817	u64 attr;
818
819	/*
820	 * This is a hack for ioremap calls before we set up kern_memmap.
821	 * Maybe we should do efi_memmap_init() earlier instead.
822	 */
823	if (!kern_memmap) {
824		attr = efi_mem_attribute(phys_addr, size);
825		if (attr & EFI_MEMORY_WB)
826			return EFI_MEMORY_WB;
827		return 0;
828	}
829
830	md = kern_memory_descriptor(phys_addr);
831	if (!md)
832		return 0;
833
834	attr = md->attribute;
835	do {
836		unsigned long md_end = kmd_end(md);
837
838		if (end <= md_end)
839			return attr;
840
841		md = kern_memory_descriptor(md_end);
842		if (!md || md->attribute != attr)
843			return 0;
844	} while (md);
845	return 0;	/* never reached */
846}
847EXPORT_SYMBOL(kern_mem_attribute);
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.txt.
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 atleast 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;
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		switch (md->type) {
1198
1199			case EFI_MEMORY_MAPPED_IO:
1200			case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1201				continue;
1202
1203			case EFI_LOADER_CODE:
1204			case EFI_LOADER_DATA:
1205			case EFI_BOOT_SERVICES_DATA:
1206			case EFI_BOOT_SERVICES_CODE:
1207			case EFI_CONVENTIONAL_MEMORY:
1208				if (md->attribute & EFI_MEMORY_WP) {
1209					name = "System ROM";
1210					flags |= IORESOURCE_READONLY;
1211				} else if (md->attribute == EFI_MEMORY_UC)
1212					name = "Uncached RAM";
1213				else
1214					name = "System RAM";
1215				break;
1216
1217			case EFI_ACPI_MEMORY_NVS:
1218				name = "ACPI Non-volatile Storage";
1219				break;
1220
1221			case EFI_UNUSABLE_MEMORY:
1222				name = "reserved";
1223				flags |= IORESOURCE_DISABLED;
1224				break;
1225
1226			case EFI_RESERVED_TYPE:
1227			case EFI_RUNTIME_SERVICES_CODE:
1228			case EFI_RUNTIME_SERVICES_DATA:
1229			case EFI_ACPI_RECLAIM_MEMORY:
1230			default:
1231				name = "reserved";
1232				break;
1233		}
1234
1235		if ((res = kzalloc(sizeof(struct resource),
1236				   GFP_KERNEL)) == NULL) {
1237			printk(KERN_ERR
1238			       "failed to allocate resource for iomem\n");
1239			return;
1240		}
1241
1242		res->name = name;
1243		res->start = md->phys_addr;
1244		res->end = md->phys_addr + efi_md_size(md) - 1;
1245		res->flags = flags;
1246
1247		if (insert_resource(&iomem_resource, res) < 0)
1248			kfree(res);
1249		else {
1250			/*
1251			 * We don't know which region contains
1252			 * kernel data so we try it repeatedly and
1253			 * let the resource manager test it.
1254			 */
1255			insert_resource(res, code_resource);
1256			insert_resource(res, data_resource);
1257			insert_resource(res, bss_resource);
1258#ifdef CONFIG_KEXEC
1259                        insert_resource(res, &efi_memmap_res);
1260                        insert_resource(res, &boot_param_res);
1261			if (crashk_res.end > crashk_res.start)
1262				insert_resource(res, &crashk_res);
1263#endif
1264		}
1265	}
1266}
1267
1268#ifdef CONFIG_KEXEC
1269/* find a block of memory aligned to 64M exclude reserved regions
1270   rsvd_regions are sorted
1271 */
1272unsigned long __init
1273kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1274{
1275	int i;
1276	u64 start, end;
1277	u64 alignment = 1UL << _PAGE_SIZE_64M;
1278	void *efi_map_start, *efi_map_end, *p;
1279	efi_memory_desc_t *md;
1280	u64 efi_desc_size;
1281
1282	efi_map_start = __va(ia64_boot_param->efi_memmap);
1283	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
1284	efi_desc_size = ia64_boot_param->efi_memdesc_size;
1285
1286	for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1287		md = p;
1288		if (!efi_wb(md))
1289			continue;
1290		start = ALIGN(md->phys_addr, alignment);
1291		end = efi_md_end(md);
1292		for (i = 0; i < n; i++) {
1293			if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1294				if (__pa(r[i].start) > start + size)
1295					return start;
1296				start = ALIGN(__pa(r[i].end), alignment);
1297				if (i < n-1 &&
1298				    __pa(r[i+1].start) < start + size)
1299					continue;
1300				else
1301					break;
1302			}
1303		}
1304		if (end > start + size)
1305			return start;
1306	}
1307
1308	printk(KERN_WARNING
1309	       "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1310	return ~0UL;
1311}
1312#endif
1313
1314#ifdef CONFIG_CRASH_DUMP
1315/* locate the size find a the descriptor at a certain address */
1316unsigned long __init
1317vmcore_find_descriptor_size (unsigned long address)
1318{
1319	void *efi_map_start, *efi_map_end, *p;
1320	efi_memory_desc_t *md;
1321	u64 efi_desc_size;
1322	unsigned long ret = 0;
1323
1324	efi_map_start = __va(ia64_boot_param->efi_memmap);
1325	efi_map_end   = efi_map_start + ia64_boot_param->efi_memmap_size;
1326	efi_desc_size = ia64_boot_param->efi_memdesc_size;
1327
1328	for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1329		md = p;
1330		if (efi_wb(md) && md->type == EFI_LOADER_DATA
1331		    && md->phys_addr == address) {
1332			ret = efi_md_size(md);
1333			break;
1334		}
1335	}
1336
1337	if (ret == 0)
1338		printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1339
1340	return ret;
1341}
1342#endif
1343