mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / arch / x86 / power / hibernate_64.c
blob692a179b1ba3224b2f7266a46b83c87e6e6a657f
1 /*
2 * Hibernation support for x86-64
4 * Distribute under GPLv2
6 * Copyright (c) 2007 Rafael J. Wysocki <rjw@sisk.pl>
7 * Copyright (c) 2002 Pavel Machek <pavel@ucw.cz>
8 * Copyright (c) 2001 Patrick Mochel <mochel@osdl.org>
9 */
11 #include <linux/gfp.h>
12 #include <linux/smp.h>
13 #include <linux/suspend.h>
14 #include <linux/scatterlist.h>
15 #include <linux/kdebug.h>
16 #include <linux/cpu.h>
18 #include <crypto/hash.h>
20 #include <asm/e820/api.h>
21 #include <asm/init.h>
22 #include <asm/proto.h>
23 #include <asm/page.h>
24 #include <asm/pgtable.h>
25 #include <asm/mtrr.h>
26 #include <asm/sections.h>
27 #include <asm/suspend.h>
28 #include <asm/tlbflush.h>
30 /* Defined in hibernate_asm_64.S */
31 extern asmlinkage __visible int restore_image(void);
34 * Address to jump to in the last phase of restore in order to get to the image
35 * kernel's text (this value is passed in the image header).
37 unsigned long restore_jump_address __visible;
38 unsigned long jump_address_phys;
41 * Value of the cr3 register from before the hibernation (this value is passed
42 * in the image header).
44 unsigned long restore_cr3 __visible;
46 unsigned long temp_level4_pgt __visible;
48 unsigned long relocated_restore_code __visible;
50 static int set_up_temporary_text_mapping(pgd_t *pgd)
52 pmd_t *pmd;
53 pud_t *pud;
54 p4d_t *p4d;
57 * The new mapping only has to cover the page containing the image
58 * kernel's entry point (jump_address_phys), because the switch over to
59 * it is carried out by relocated code running from a page allocated
60 * specifically for this purpose and covered by the identity mapping, so
61 * the temporary kernel text mapping is only needed for the final jump.
62 * Moreover, in that mapping the virtual address of the image kernel's
63 * entry point must be the same as its virtual address in the image
64 * kernel (restore_jump_address), so the image kernel's
65 * restore_registers() code doesn't find itself in a different area of
66 * the virtual address space after switching over to the original page
67 * tables used by the image kernel.
70 if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
71 p4d = (p4d_t *)get_safe_page(GFP_ATOMIC);
72 if (!p4d)
73 return -ENOMEM;
76 pud = (pud_t *)get_safe_page(GFP_ATOMIC);
77 if (!pud)
78 return -ENOMEM;
80 pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
81 if (!pmd)
82 return -ENOMEM;
84 set_pmd(pmd + pmd_index(restore_jump_address),
85 __pmd((jump_address_phys & PMD_MASK) | __PAGE_KERNEL_LARGE_EXEC));
86 set_pud(pud + pud_index(restore_jump_address),
87 __pud(__pa(pmd) | _KERNPG_TABLE));
88 if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
89 set_p4d(p4d + p4d_index(restore_jump_address), __p4d(__pa(pud) | _KERNPG_TABLE));
90 set_pgd(pgd + pgd_index(restore_jump_address), __pgd(__pa(p4d) | _KERNPG_TABLE));
91 } else {
92 /* No p4d for 4-level paging: point the pgd to the pud page table */
93 set_pgd(pgd + pgd_index(restore_jump_address), __pgd(__pa(pud) | _KERNPG_TABLE));
96 return 0;
99 static void *alloc_pgt_page(void *context)
101 return (void *)get_safe_page(GFP_ATOMIC);
104 static int set_up_temporary_mappings(void)
106 struct x86_mapping_info info = {
107 .alloc_pgt_page = alloc_pgt_page,
108 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
109 .offset = __PAGE_OFFSET,
111 unsigned long mstart, mend;
112 pgd_t *pgd;
113 int result;
114 int i;
116 pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
117 if (!pgd)
118 return -ENOMEM;
120 /* Prepare a temporary mapping for the kernel text */
121 result = set_up_temporary_text_mapping(pgd);
122 if (result)
123 return result;
125 /* Set up the direct mapping from scratch */
126 for (i = 0; i < nr_pfn_mapped; i++) {
127 mstart = pfn_mapped[i].start << PAGE_SHIFT;
128 mend = pfn_mapped[i].end << PAGE_SHIFT;
130 result = kernel_ident_mapping_init(&info, pgd, mstart, mend);
131 if (result)
132 return result;
135 temp_level4_pgt = __pa(pgd);
136 return 0;
139 static int relocate_restore_code(void)
141 pgd_t *pgd;
142 p4d_t *p4d;
143 pud_t *pud;
144 pmd_t *pmd;
145 pte_t *pte;
147 relocated_restore_code = get_safe_page(GFP_ATOMIC);
148 if (!relocated_restore_code)
149 return -ENOMEM;
151 memcpy((void *)relocated_restore_code, core_restore_code, PAGE_SIZE);
153 /* Make the page containing the relocated code executable */
154 pgd = (pgd_t *)__va(read_cr3_pa()) +
155 pgd_index(relocated_restore_code);
156 p4d = p4d_offset(pgd, relocated_restore_code);
157 if (p4d_large(*p4d)) {
158 set_p4d(p4d, __p4d(p4d_val(*p4d) & ~_PAGE_NX));
159 goto out;
161 pud = pud_offset(p4d, relocated_restore_code);
162 if (pud_large(*pud)) {
163 set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX));
164 goto out;
166 pmd = pmd_offset(pud, relocated_restore_code);
167 if (pmd_large(*pmd)) {
168 set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX));
169 goto out;
171 pte = pte_offset_kernel(pmd, relocated_restore_code);
172 set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX));
173 out:
174 __flush_tlb_all();
175 return 0;
178 asmlinkage int swsusp_arch_resume(void)
180 int error;
182 /* We have got enough memory and from now on we cannot recover */
183 error = set_up_temporary_mappings();
184 if (error)
185 return error;
187 error = relocate_restore_code();
188 if (error)
189 return error;
191 restore_image();
192 return 0;
196 * pfn_is_nosave - check if given pfn is in the 'nosave' section
199 int pfn_is_nosave(unsigned long pfn)
201 unsigned long nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT;
202 unsigned long nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
203 return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
206 #define MD5_DIGEST_SIZE 16
208 struct restore_data_record {
209 unsigned long jump_address;
210 unsigned long jump_address_phys;
211 unsigned long cr3;
212 unsigned long magic;
213 u8 e820_digest[MD5_DIGEST_SIZE];
216 #define RESTORE_MAGIC 0x23456789ABCDEF01UL
218 #if IS_BUILTIN(CONFIG_CRYPTO_MD5)
220 * get_e820_md5 - calculate md5 according to given e820 table
222 * @table: the e820 table to be calculated
223 * @buf: the md5 result to be stored to
225 static int get_e820_md5(struct e820_table *table, void *buf)
227 struct scatterlist sg;
228 struct crypto_ahash *tfm;
229 int size;
230 int ret = 0;
232 tfm = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
233 if (IS_ERR(tfm))
234 return -ENOMEM;
237 AHASH_REQUEST_ON_STACK(req, tfm);
238 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry) * table->nr_entries;
239 ahash_request_set_tfm(req, tfm);
240 sg_init_one(&sg, (u8 *)table, size);
241 ahash_request_set_callback(req, 0, NULL, NULL);
242 ahash_request_set_crypt(req, &sg, buf, size);
244 if (crypto_ahash_digest(req))
245 ret = -EINVAL;
246 ahash_request_zero(req);
248 crypto_free_ahash(tfm);
250 return ret;
253 static int hibernation_e820_save(void *buf)
255 return get_e820_md5(e820_table_firmware, buf);
258 static bool hibernation_e820_mismatch(void *buf)
260 int ret;
261 u8 result[MD5_DIGEST_SIZE];
263 memset(result, 0, MD5_DIGEST_SIZE);
264 /* If there is no digest in suspend kernel, let it go. */
265 if (!memcmp(result, buf, MD5_DIGEST_SIZE))
266 return false;
268 ret = get_e820_md5(e820_table_firmware, result);
269 if (ret)
270 return true;
272 return memcmp(result, buf, MD5_DIGEST_SIZE) ? true : false;
274 #else
275 static int hibernation_e820_save(void *buf)
277 return 0;
280 static bool hibernation_e820_mismatch(void *buf)
282 /* If md5 is not builtin for restore kernel, let it go. */
283 return false;
285 #endif
288 * arch_hibernation_header_save - populate the architecture specific part
289 * of a hibernation image header
290 * @addr: address to save the data at
292 int arch_hibernation_header_save(void *addr, unsigned int max_size)
294 struct restore_data_record *rdr = addr;
296 if (max_size < sizeof(struct restore_data_record))
297 return -EOVERFLOW;
298 rdr->jump_address = (unsigned long)restore_registers;
299 rdr->jump_address_phys = __pa_symbol(restore_registers);
302 * The restore code fixes up CR3 and CR4 in the following sequence:
304 * [in hibernation asm]
305 * 1. CR3 <= temporary page tables
306 * 2. CR4 <= mmu_cr4_features (from the kernel that restores us)
307 * 3. CR3 <= rdr->cr3
308 * 4. CR4 <= mmu_cr4_features (from us, i.e. the image kernel)
309 * [in restore_processor_state()]
310 * 5. CR4 <= saved CR4
311 * 6. CR3 <= saved CR3
313 * Our mmu_cr4_features has CR4.PCIDE=0, and toggling
314 * CR4.PCIDE while CR3's PCID bits are nonzero is illegal, so
315 * rdr->cr3 needs to point to valid page tables but must not
316 * have any of the PCID bits set.
318 rdr->cr3 = restore_cr3 & ~CR3_PCID_MASK;
320 rdr->magic = RESTORE_MAGIC;
322 return hibernation_e820_save(rdr->e820_digest);
326 * arch_hibernation_header_restore - read the architecture specific data
327 * from the hibernation image header
328 * @addr: address to read the data from
330 int arch_hibernation_header_restore(void *addr)
332 struct restore_data_record *rdr = addr;
334 restore_jump_address = rdr->jump_address;
335 jump_address_phys = rdr->jump_address_phys;
336 restore_cr3 = rdr->cr3;
338 if (rdr->magic != RESTORE_MAGIC) {
339 pr_crit("Unrecognized hibernate image header format!\n");
340 return -EINVAL;
343 if (hibernation_e820_mismatch(rdr->e820_digest)) {
344 pr_crit("Hibernate inconsistent memory map detected!\n");
345 return -ENODEV;
348 return 0;
351 int arch_resume_nosmt(void)
353 int ret = 0;
355 * We reached this while coming out of hibernation. This means
356 * that SMT siblings are sleeping in hlt, as mwait is not safe
357 * against control transition during resume (see comment in
358 * hibernate_resume_nonboot_cpu_disable()).
360 * If the resumed kernel has SMT disabled, we have to take all the
361 * SMT siblings out of hlt, and offline them again so that they
362 * end up in mwait proper.
364 * Called with hotplug disabled.
366 cpu_hotplug_enable();
367 if (cpu_smt_control == CPU_SMT_DISABLED ||
368 cpu_smt_control == CPU_SMT_FORCE_DISABLED) {
369 enum cpuhp_smt_control old = cpu_smt_control;
371 ret = cpuhp_smt_enable();
372 if (ret)
373 goto out;
374 ret = cpuhp_smt_disable(old);
375 if (ret)
376 goto out;
378 out:
379 cpu_hotplug_disable();
380 return ret;