1/*
2 * x86_64 specific EFI support functions
3 * Based on Extensible Firmware Interface Specification version 1.0
4 *
5 * Copyright (C) 2005-2008 Intel Co.
6 *	Fenghua Yu <fenghua.yu@intel.com>
7 *	Bibo Mao <bibo.mao@intel.com>
8 *	Chandramouli Narayanan <mouli@linux.intel.com>
9 *	Huang Ying <ying.huang@intel.com>
10 *
11 * Code to convert EFI to E820 map has been implemented in elilo bootloader
12 * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table
13 * is setup appropriately for EFI runtime code.
14 * - mouli 06/14/2007.
15 *
16 */
17
18#include <linux/kernel.h>
19#include <linux/init.h>
20#include <linux/mm.h>
21#include <linux/types.h>
22#include <linux/spinlock.h>
23#include <linux/bootmem.h>
24#include <linux/ioport.h>
25#include <linux/module.h>
26#include <linux/efi.h>
27#include <linux/uaccess.h>
28#include <linux/io.h>
29#include <linux/reboot.h>
30#include <linux/slab.h>
31
32#include <asm/setup.h>
33#include <asm/page.h>
34#include <asm/e820.h>
35#include <asm/pgtable.h>
36#include <asm/tlbflush.h>
37#include <asm/proto.h>
38#include <asm/efi.h>
39#include <asm/cacheflush.h>
40#include <asm/fixmap.h>
41#include <asm/realmode.h>
42#include <asm/time.h>
43
44/*
45 * We allocate runtime services regions bottom-up, starting from -4G, i.e.
46 * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
47 */
48static u64 efi_va = EFI_VA_START;
49
50/*
51 * Scratch space used for switching the pagetable in the EFI stub
52 */
53struct efi_scratch {
54	u64 r15;
55	u64 prev_cr3;
56	pgd_t *efi_pgt;
57	bool use_pgd;
58	u64 phys_stack;
59} __packed;
60
61static void __init early_code_mapping_set_exec(int executable)
62{
63	efi_memory_desc_t *md;
64	void *p;
65
66	if (!(__supported_pte_mask & _PAGE_NX))
67		return;
68
69	/* Make EFI service code area executable */
70	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
71		md = p;
72		if (md->type == EFI_RUNTIME_SERVICES_CODE ||
73		    md->type == EFI_BOOT_SERVICES_CODE)
74			efi_set_executable(md, executable);
75	}
76}
77
78pgd_t * __init efi_call_phys_prolog(void)
79{
80	unsigned long vaddress;
81	pgd_t *save_pgd;
82
83	int pgd;
84	int n_pgds;
85
86	if (!efi_enabled(EFI_OLD_MEMMAP))
87		return NULL;
88
89	early_code_mapping_set_exec(1);
90
91	n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
92	save_pgd = kmalloc(n_pgds * sizeof(pgd_t), GFP_KERNEL);
93
94	for (pgd = 0; pgd < n_pgds; pgd++) {
95		save_pgd[pgd] = *pgd_offset_k(pgd * PGDIR_SIZE);
96		vaddress = (unsigned long)__va(pgd * PGDIR_SIZE);
97		set_pgd(pgd_offset_k(pgd * PGDIR_SIZE), *pgd_offset_k(vaddress));
98	}
99	__flush_tlb_all();
100
101	return save_pgd;
102}
103
104void __init efi_call_phys_epilog(pgd_t *save_pgd)
105{
106	/*
107	 * After the lock is released, the original page table is restored.
108	 */
109	int pgd_idx;
110	int nr_pgds;
111
112	if (!save_pgd)
113		return;
114
115	nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
116
117	for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++)
118		set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
119
120	kfree(save_pgd);
121
122	__flush_tlb_all();
123	early_code_mapping_set_exec(0);
124}
125
126/*
127 * Add low kernel mappings for passing arguments to EFI functions.
128 */
129void efi_sync_low_kernel_mappings(void)
130{
131	unsigned num_pgds;
132	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
133
134	if (efi_enabled(EFI_OLD_MEMMAP))
135		return;
136
137	num_pgds = pgd_index(MODULES_END - 1) - pgd_index(PAGE_OFFSET);
138
139	memcpy(pgd + pgd_index(PAGE_OFFSET),
140		init_mm.pgd + pgd_index(PAGE_OFFSET),
141		sizeof(pgd_t) * num_pgds);
142}
143
144int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
145{
146	unsigned long text;
147	struct page *page;
148	unsigned npages;
149	pgd_t *pgd;
150
151	if (efi_enabled(EFI_OLD_MEMMAP))
152		return 0;
153
154	efi_scratch.efi_pgt = (pgd_t *)(unsigned long)real_mode_header->trampoline_pgd;
155	pgd = __va(efi_scratch.efi_pgt);
156
157	/*
158	 * It can happen that the physical address of new_memmap lands in memory
159	 * which is not mapped in the EFI page table. Therefore we need to go
160	 * and ident-map those pages containing the map before calling
161	 * phys_efi_set_virtual_address_map().
162	 */
163	if (kernel_map_pages_in_pgd(pgd, pa_memmap, pa_memmap, num_pages, _PAGE_NX)) {
164		pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
165		return 1;
166	}
167
168	efi_scratch.use_pgd = true;
169
170	/*
171	 * When making calls to the firmware everything needs to be 1:1
172	 * mapped and addressable with 32-bit pointers. Map the kernel
173	 * text and allocate a new stack because we can't rely on the
174	 * stack pointer being < 4GB.
175	 */
176	if (!IS_ENABLED(CONFIG_EFI_MIXED))
177		return 0;
178
179	page = alloc_page(GFP_KERNEL|__GFP_DMA32);
180	if (!page)
181		panic("Unable to allocate EFI runtime stack < 4GB\n");
182
183	efi_scratch.phys_stack = virt_to_phys(page_address(page));
184	efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
185
186	npages = (_end - _text) >> PAGE_SHIFT;
187	text = __pa(_text);
188
189	if (kernel_map_pages_in_pgd(pgd, text >> PAGE_SHIFT, text, npages, 0)) {
190		pr_err("Failed to map kernel text 1:1\n");
191		return 1;
192	}
193
194	return 0;
195}
196
197void __init efi_cleanup_page_tables(unsigned long pa_memmap, unsigned num_pages)
198{
199	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
200
201	kernel_unmap_pages_in_pgd(pgd, pa_memmap, num_pages);
202}
203
204static void __init __map_region(efi_memory_desc_t *md, u64 va)
205{
206	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
207	unsigned long pf = 0;
208
209	if (!(md->attribute & EFI_MEMORY_WB))
210		pf |= _PAGE_PCD;
211
212	if (kernel_map_pages_in_pgd(pgd, md->phys_addr, va, md->num_pages, pf))
213		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
214			   md->phys_addr, va);
215}
216
217void __init efi_map_region(efi_memory_desc_t *md)
218{
219	unsigned long size = md->num_pages << PAGE_SHIFT;
220	u64 pa = md->phys_addr;
221
222	if (efi_enabled(EFI_OLD_MEMMAP))
223		return old_map_region(md);
224
225	/*
226	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
227	 * firmware which doesn't update all internal pointers after switching
228	 * to virtual mode and would otherwise crap on us.
229	 */
230	__map_region(md, md->phys_addr);
231
232	/*
233	 * Enforce the 1:1 mapping as the default virtual address when
234	 * booting in EFI mixed mode, because even though we may be
235	 * running a 64-bit kernel, the firmware may only be 32-bit.
236	 */
237	if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
238		md->virt_addr = md->phys_addr;
239		return;
240	}
241
242	efi_va -= size;
243
244	/* Is PA 2M-aligned? */
245	if (!(pa & (PMD_SIZE - 1))) {
246		efi_va &= PMD_MASK;
247	} else {
248		u64 pa_offset = pa & (PMD_SIZE - 1);
249		u64 prev_va = efi_va;
250
251		/* get us the same offset within this 2M page */
252		efi_va = (efi_va & PMD_MASK) + pa_offset;
253
254		if (efi_va > prev_va)
255			efi_va -= PMD_SIZE;
256	}
257
258	if (efi_va < EFI_VA_END) {
259		pr_warn(FW_WARN "VA address range overflow!\n");
260		return;
261	}
262
263	/* Do the VA map */
264	__map_region(md, efi_va);
265	md->virt_addr = efi_va;
266}
267
268/*
269 * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
270 * md->virt_addr is the original virtual address which had been mapped in kexec
271 * 1st kernel.
272 */
273void __init efi_map_region_fixed(efi_memory_desc_t *md)
274{
275	__map_region(md, md->virt_addr);
276}
277
278void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
279				 u32 type, u64 attribute)
280{
281	unsigned long last_map_pfn;
282
283	if (type == EFI_MEMORY_MAPPED_IO)
284		return ioremap(phys_addr, size);
285
286	last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
287	if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
288		unsigned long top = last_map_pfn << PAGE_SHIFT;
289		efi_ioremap(top, size - (top - phys_addr), type, attribute);
290	}
291
292	if (!(attribute & EFI_MEMORY_WB))
293		efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
294
295	return (void __iomem *)__va(phys_addr);
296}
297
298void __init parse_efi_setup(u64 phys_addr, u32 data_len)
299{
300	efi_setup = phys_addr + sizeof(struct setup_data);
301}
302
303void __init efi_runtime_mkexec(void)
304{
305	if (!efi_enabled(EFI_OLD_MEMMAP))
306		return;
307
308	if (__supported_pte_mask & _PAGE_NX)
309		runtime_code_page_mkexec();
310}
311
312void __init efi_dump_pagetable(void)
313{
314#ifdef CONFIG_EFI_PGT_DUMP
315	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
316
317	ptdump_walk_pgd_level(NULL, pgd);
318#endif
319}
320
321#ifdef CONFIG_EFI_MIXED
322extern efi_status_t efi64_thunk(u32, ...);
323
324#define runtime_service32(func)						 \
325({									 \
326	u32 table = (u32)(unsigned long)efi.systab;			 \
327	u32 *rt, *___f;							 \
328									 \
329	rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));	 \
330	___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
331	*___f;								 \
332})
333
334/*
335 * Switch to the EFI page tables early so that we can access the 1:1
336 * runtime services mappings which are not mapped in any other page
337 * tables. This function must be called before runtime_service32().
338 *
339 * Also, disable interrupts because the IDT points to 64-bit handlers,
340 * which aren't going to function correctly when we switch to 32-bit.
341 */
342#define efi_thunk(f, ...)						\
343({									\
344	efi_status_t __s;						\
345	unsigned long flags;						\
346	u32 func;							\
347									\
348	efi_sync_low_kernel_mappings();					\
349	local_irq_save(flags);						\
350									\
351	efi_scratch.prev_cr3 = read_cr3();				\
352	write_cr3((unsigned long)efi_scratch.efi_pgt);			\
353	__flush_tlb_all();						\
354									\
355	func = runtime_service32(f);					\
356	__s = efi64_thunk(func, __VA_ARGS__);			\
357									\
358	write_cr3(efi_scratch.prev_cr3);				\
359	__flush_tlb_all();						\
360	local_irq_restore(flags);					\
361									\
362	__s;								\
363})
364
365efi_status_t efi_thunk_set_virtual_address_map(
366	void *phys_set_virtual_address_map,
367	unsigned long memory_map_size,
368	unsigned long descriptor_size,
369	u32 descriptor_version,
370	efi_memory_desc_t *virtual_map)
371{
372	efi_status_t status;
373	unsigned long flags;
374	u32 func;
375
376	efi_sync_low_kernel_mappings();
377	local_irq_save(flags);
378
379	efi_scratch.prev_cr3 = read_cr3();
380	write_cr3((unsigned long)efi_scratch.efi_pgt);
381	__flush_tlb_all();
382
383	func = (u32)(unsigned long)phys_set_virtual_address_map;
384	status = efi64_thunk(func, memory_map_size, descriptor_size,
385			     descriptor_version, virtual_map);
386
387	write_cr3(efi_scratch.prev_cr3);
388	__flush_tlb_all();
389	local_irq_restore(flags);
390
391	return status;
392}
393
394static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
395{
396	efi_status_t status;
397	u32 phys_tm, phys_tc;
398
399	spin_lock(&rtc_lock);
400
401	phys_tm = virt_to_phys(tm);
402	phys_tc = virt_to_phys(tc);
403
404	status = efi_thunk(get_time, phys_tm, phys_tc);
405
406	spin_unlock(&rtc_lock);
407
408	return status;
409}
410
411static efi_status_t efi_thunk_set_time(efi_time_t *tm)
412{
413	efi_status_t status;
414	u32 phys_tm;
415
416	spin_lock(&rtc_lock);
417
418	phys_tm = virt_to_phys(tm);
419
420	status = efi_thunk(set_time, phys_tm);
421
422	spin_unlock(&rtc_lock);
423
424	return status;
425}
426
427static efi_status_t
428efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
429			  efi_time_t *tm)
430{
431	efi_status_t status;
432	u32 phys_enabled, phys_pending, phys_tm;
433
434	spin_lock(&rtc_lock);
435
436	phys_enabled = virt_to_phys(enabled);
437	phys_pending = virt_to_phys(pending);
438	phys_tm = virt_to_phys(tm);
439
440	status = efi_thunk(get_wakeup_time, phys_enabled,
441			     phys_pending, phys_tm);
442
443	spin_unlock(&rtc_lock);
444
445	return status;
446}
447
448static efi_status_t
449efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
450{
451	efi_status_t status;
452	u32 phys_tm;
453
454	spin_lock(&rtc_lock);
455
456	phys_tm = virt_to_phys(tm);
457
458	status = efi_thunk(set_wakeup_time, enabled, phys_tm);
459
460	spin_unlock(&rtc_lock);
461
462	return status;
463}
464
465
466static efi_status_t
467efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
468		       u32 *attr, unsigned long *data_size, void *data)
469{
470	efi_status_t status;
471	u32 phys_name, phys_vendor, phys_attr;
472	u32 phys_data_size, phys_data;
473
474	phys_data_size = virt_to_phys(data_size);
475	phys_vendor = virt_to_phys(vendor);
476	phys_name = virt_to_phys(name);
477	phys_attr = virt_to_phys(attr);
478	phys_data = virt_to_phys(data);
479
480	status = efi_thunk(get_variable, phys_name, phys_vendor,
481			   phys_attr, phys_data_size, phys_data);
482
483	return status;
484}
485
486static efi_status_t
487efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
488		       u32 attr, unsigned long data_size, void *data)
489{
490	u32 phys_name, phys_vendor, phys_data;
491	efi_status_t status;
492
493	phys_name = virt_to_phys(name);
494	phys_vendor = virt_to_phys(vendor);
495	phys_data = virt_to_phys(data);
496
497	/* If data_size is > sizeof(u32) we've got problems */
498	status = efi_thunk(set_variable, phys_name, phys_vendor,
499			   attr, data_size, phys_data);
500
501	return status;
502}
503
504static efi_status_t
505efi_thunk_get_next_variable(unsigned long *name_size,
506			    efi_char16_t *name,
507			    efi_guid_t *vendor)
508{
509	efi_status_t status;
510	u32 phys_name_size, phys_name, phys_vendor;
511
512	phys_name_size = virt_to_phys(name_size);
513	phys_vendor = virt_to_phys(vendor);
514	phys_name = virt_to_phys(name);
515
516	status = efi_thunk(get_next_variable, phys_name_size,
517			   phys_name, phys_vendor);
518
519	return status;
520}
521
522static efi_status_t
523efi_thunk_get_next_high_mono_count(u32 *count)
524{
525	efi_status_t status;
526	u32 phys_count;
527
528	phys_count = virt_to_phys(count);
529	status = efi_thunk(get_next_high_mono_count, phys_count);
530
531	return status;
532}
533
534static void
535efi_thunk_reset_system(int reset_type, efi_status_t status,
536		       unsigned long data_size, efi_char16_t *data)
537{
538	u32 phys_data;
539
540	phys_data = virt_to_phys(data);
541
542	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
543}
544
545static efi_status_t
546efi_thunk_update_capsule(efi_capsule_header_t **capsules,
547			 unsigned long count, unsigned long sg_list)
548{
549	/*
550	 * To properly support this function we would need to repackage
551	 * 'capsules' because the firmware doesn't understand 64-bit
552	 * pointers.
553	 */
554	return EFI_UNSUPPORTED;
555}
556
557static efi_status_t
558efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
559			      u64 *remaining_space,
560			      u64 *max_variable_size)
561{
562	efi_status_t status;
563	u32 phys_storage, phys_remaining, phys_max;
564
565	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
566		return EFI_UNSUPPORTED;
567
568	phys_storage = virt_to_phys(storage_space);
569	phys_remaining = virt_to_phys(remaining_space);
570	phys_max = virt_to_phys(max_variable_size);
571
572	status = efi_thunk(query_variable_info, attr, phys_storage,
573			   phys_remaining, phys_max);
574
575	return status;
576}
577
578static efi_status_t
579efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
580			     unsigned long count, u64 *max_size,
581			     int *reset_type)
582{
583	/*
584	 * To properly support this function we would need to repackage
585	 * 'capsules' because the firmware doesn't understand 64-bit
586	 * pointers.
587	 */
588	return EFI_UNSUPPORTED;
589}
590
591void efi_thunk_runtime_setup(void)
592{
593	efi.get_time = efi_thunk_get_time;
594	efi.set_time = efi_thunk_set_time;
595	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
596	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
597	efi.get_variable = efi_thunk_get_variable;
598	efi.get_next_variable = efi_thunk_get_next_variable;
599	efi.set_variable = efi_thunk_set_variable;
600	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
601	efi.reset_system = efi_thunk_reset_system;
602	efi.query_variable_info = efi_thunk_query_variable_info;
603	efi.update_capsule = efi_thunk_update_capsule;
604	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
605}
606#endif /* CONFIG_EFI_MIXED */
607