Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*:
3 * Hibernate support specific for ARM64
4 *
5 * Derived from work on ARM hibernation support by:
6 *
7 * Ubuntu project, hibernation support for mach-dove
8 * Copyright (C) 2010 Nokia Corporation (Hiroshi Doyu)
9 * Copyright (C) 2010 Texas Instruments, Inc. (Teerth Reddy et al.)
10 * https://lkml.org/lkml/2010/6/18/4
11 * https://lists.linux-foundation.org/pipermail/linux-pm/2010-June/027422.html
12 * https://patchwork.kernel.org/patch/96442/
13 *
14 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
15 */
16#define pr_fmt(x) "hibernate: " x
17#include <linux/cpu.h>
18#include <linux/kvm_host.h>
19#include <linux/mm.h>
20#include <linux/pm.h>
21#include <linux/sched.h>
22#include <linux/suspend.h>
23#include <linux/utsname.h>
24#include <linux/version.h>
25
26#include <asm/barrier.h>
27#include <asm/cacheflush.h>
28#include <asm/cputype.h>
29#include <asm/daifflags.h>
30#include <asm/irqflags.h>
31#include <asm/kexec.h>
32#include <asm/memory.h>
33#include <asm/mmu_context.h>
34#include <asm/pgalloc.h>
35#include <asm/pgtable-hwdef.h>
36#include <asm/sections.h>
37#include <asm/smp.h>
38#include <asm/smp_plat.h>
39#include <asm/suspend.h>
40#include <asm/sysreg.h>
41#include <asm/virt.h>
42
43/*
44 * Hibernate core relies on this value being 0 on resume, and marks it
45 * __nosavedata assuming it will keep the resume kernel's '0' value. This
46 * doesn't happen with either KASLR.
47 *
48 * defined as "__visible int in_suspend __nosavedata" in
49 * kernel/power/hibernate.c
50 */
51extern int in_suspend;
52
53/* Do we need to reset el2? */
54#define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
55
56/* temporary el2 vectors in the __hibernate_exit_text section. */
57extern char hibernate_el2_vectors[];
58
59/* hyp-stub vectors, used to restore el2 during resume from hibernate. */
60extern char __hyp_stub_vectors[];
61
62/*
63 * The logical cpu number we should resume on, initialised to a non-cpu
64 * number.
65 */
66static int sleep_cpu = -EINVAL;
67
68/*
69 * Values that may not change over hibernate/resume. We put the build number
70 * and date in here so that we guarantee not to resume with a different
71 * kernel.
72 */
73struct arch_hibernate_hdr_invariants {
74 char uts_version[__NEW_UTS_LEN + 1];
75};
76
77/* These values need to be know across a hibernate/restore. */
78static struct arch_hibernate_hdr {
79 struct arch_hibernate_hdr_invariants invariants;
80
81 /* These are needed to find the relocated kernel if built with kaslr */
82 phys_addr_t ttbr1_el1;
83 void (*reenter_kernel)(void);
84
85 /*
86 * We need to know where the __hyp_stub_vectors are after restore to
87 * re-configure el2.
88 */
89 phys_addr_t __hyp_stub_vectors;
90
91 u64 sleep_cpu_mpidr;
92} resume_hdr;
93
94static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
95{
96 memset(i, 0, sizeof(*i));
97 memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
98}
99
100int pfn_is_nosave(unsigned long pfn)
101{
102 unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
103 unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
104
105 return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn)) ||
106 crash_is_nosave(pfn);
107}
108
109void notrace save_processor_state(void)
110{
111 WARN_ON(num_online_cpus() != 1);
112}
113
114void notrace restore_processor_state(void)
115{
116}
117
118int arch_hibernation_header_save(void *addr, unsigned int max_size)
119{
120 struct arch_hibernate_hdr *hdr = addr;
121
122 if (max_size < sizeof(*hdr))
123 return -EOVERFLOW;
124
125 arch_hdr_invariants(&hdr->invariants);
126 hdr->ttbr1_el1 = __pa_symbol(swapper_pg_dir);
127 hdr->reenter_kernel = _cpu_resume;
128
129 /* We can't use __hyp_get_vectors() because kvm may still be loaded */
130 if (el2_reset_needed())
131 hdr->__hyp_stub_vectors = __pa_symbol(__hyp_stub_vectors);
132 else
133 hdr->__hyp_stub_vectors = 0;
134
135 /* Save the mpidr of the cpu we called cpu_suspend() on... */
136 if (sleep_cpu < 0) {
137 pr_err("Failing to hibernate on an unknown CPU.\n");
138 return -ENODEV;
139 }
140 hdr->sleep_cpu_mpidr = cpu_logical_map(sleep_cpu);
141 pr_info("Hibernating on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
142 hdr->sleep_cpu_mpidr);
143
144 return 0;
145}
146EXPORT_SYMBOL(arch_hibernation_header_save);
147
148int arch_hibernation_header_restore(void *addr)
149{
150 int ret;
151 struct arch_hibernate_hdr_invariants invariants;
152 struct arch_hibernate_hdr *hdr = addr;
153
154 arch_hdr_invariants(&invariants);
155 if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
156 pr_crit("Hibernate image not generated by this kernel!\n");
157 return -EINVAL;
158 }
159
160 sleep_cpu = get_logical_index(hdr->sleep_cpu_mpidr);
161 pr_info("Hibernated on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
162 hdr->sleep_cpu_mpidr);
163 if (sleep_cpu < 0) {
164 pr_crit("Hibernated on a CPU not known to this kernel!\n");
165 sleep_cpu = -EINVAL;
166 return -EINVAL;
167 }
168
169 ret = bringup_hibernate_cpu(sleep_cpu);
170 if (ret) {
171 sleep_cpu = -EINVAL;
172 return ret;
173 }
174
175 resume_hdr = *hdr;
176
177 return 0;
178}
179EXPORT_SYMBOL(arch_hibernation_header_restore);
180
181static int trans_pgd_map_page(pgd_t *trans_pgd, void *page,
182 unsigned long dst_addr,
183 pgprot_t pgprot)
184{
185 pgd_t *pgdp;
186 p4d_t *p4dp;
187 pud_t *pudp;
188 pmd_t *pmdp;
189 pte_t *ptep;
190
191 pgdp = pgd_offset_pgd(trans_pgd, dst_addr);
192 if (pgd_none(READ_ONCE(*pgdp))) {
193 pudp = (void *)get_safe_page(GFP_ATOMIC);
194 if (!pudp)
195 return -ENOMEM;
196 pgd_populate(&init_mm, pgdp, pudp);
197 }
198
199 p4dp = p4d_offset(pgdp, dst_addr);
200 if (p4d_none(READ_ONCE(*p4dp))) {
201 pudp = (void *)get_safe_page(GFP_ATOMIC);
202 if (!pudp)
203 return -ENOMEM;
204 p4d_populate(&init_mm, p4dp, pudp);
205 }
206
207 pudp = pud_offset(p4dp, dst_addr);
208 if (pud_none(READ_ONCE(*pudp))) {
209 pmdp = (void *)get_safe_page(GFP_ATOMIC);
210 if (!pmdp)
211 return -ENOMEM;
212 pud_populate(&init_mm, pudp, pmdp);
213 }
214
215 pmdp = pmd_offset(pudp, dst_addr);
216 if (pmd_none(READ_ONCE(*pmdp))) {
217 ptep = (void *)get_safe_page(GFP_ATOMIC);
218 if (!ptep)
219 return -ENOMEM;
220 pmd_populate_kernel(&init_mm, pmdp, ptep);
221 }
222
223 ptep = pte_offset_kernel(pmdp, dst_addr);
224 set_pte(ptep, pfn_pte(virt_to_pfn(page), PAGE_KERNEL_EXEC));
225
226 return 0;
227}
228
229/*
230 * Copies length bytes, starting at src_start into an new page,
231 * perform cache maintenance, then maps it at the specified address low
232 * address as executable.
233 *
234 * This is used by hibernate to copy the code it needs to execute when
235 * overwriting the kernel text. This function generates a new set of page
236 * tables, which it loads into ttbr0.
237 *
238 * Length is provided as we probably only want 4K of data, even on a 64K
239 * page system.
240 */
241static int create_safe_exec_page(void *src_start, size_t length,
242 unsigned long dst_addr,
243 phys_addr_t *phys_dst_addr)
244{
245 void *page = (void *)get_safe_page(GFP_ATOMIC);
246 pgd_t *trans_pgd;
247 int rc;
248
249 if (!page)
250 return -ENOMEM;
251
252 memcpy(page, src_start, length);
253 __flush_icache_range((unsigned long)page, (unsigned long)page + length);
254
255 trans_pgd = (void *)get_safe_page(GFP_ATOMIC);
256 if (!trans_pgd)
257 return -ENOMEM;
258
259 rc = trans_pgd_map_page(trans_pgd, page, dst_addr,
260 PAGE_KERNEL_EXEC);
261 if (rc)
262 return rc;
263
264 /*
265 * Load our new page tables. A strict BBM approach requires that we
266 * ensure that TLBs are free of any entries that may overlap with the
267 * global mappings we are about to install.
268 *
269 * For a real hibernate/resume cycle TTBR0 currently points to a zero
270 * page, but TLBs may contain stale ASID-tagged entries (e.g. for EFI
271 * runtime services), while for a userspace-driven test_resume cycle it
272 * points to userspace page tables (and we must point it at a zero page
273 * ourselves). Elsewhere we only (un)install the idmap with preemption
274 * disabled, so T0SZ should be as required regardless.
275 */
276 cpu_set_reserved_ttbr0();
277 local_flush_tlb_all();
278 write_sysreg(phys_to_ttbr(virt_to_phys(trans_pgd)), ttbr0_el1);
279 isb();
280
281 *phys_dst_addr = virt_to_phys(page);
282
283 return 0;
284}
285
286#define dcache_clean_range(start, end) __flush_dcache_area(start, (end - start))
287
288int swsusp_arch_suspend(void)
289{
290 int ret = 0;
291 unsigned long flags;
292 struct sleep_stack_data state;
293
294 if (cpus_are_stuck_in_kernel()) {
295 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
296 return -EBUSY;
297 }
298
299 flags = local_daif_save();
300
301 if (__cpu_suspend_enter(&state)) {
302 /* make the crash dump kernel image visible/saveable */
303 crash_prepare_suspend();
304
305 sleep_cpu = smp_processor_id();
306 ret = swsusp_save();
307 } else {
308 /* Clean kernel core startup/idle code to PoC*/
309 dcache_clean_range(__mmuoff_data_start, __mmuoff_data_end);
310 dcache_clean_range(__idmap_text_start, __idmap_text_end);
311
312 /* Clean kvm setup code to PoC? */
313 if (el2_reset_needed()) {
314 dcache_clean_range(__hyp_idmap_text_start, __hyp_idmap_text_end);
315 dcache_clean_range(__hyp_text_start, __hyp_text_end);
316 }
317
318 /* make the crash dump kernel image protected again */
319 crash_post_resume();
320
321 /*
322 * Tell the hibernation core that we've just restored
323 * the memory
324 */
325 in_suspend = 0;
326
327 sleep_cpu = -EINVAL;
328 __cpu_suspend_exit();
329
330 /*
331 * Just in case the boot kernel did turn the SSBD
332 * mitigation off behind our back, let's set the state
333 * to what we expect it to be.
334 */
335 switch (arm64_get_ssbd_state()) {
336 case ARM64_SSBD_FORCE_ENABLE:
337 case ARM64_SSBD_KERNEL:
338 arm64_set_ssbd_mitigation(true);
339 }
340 }
341
342 local_daif_restore(flags);
343
344 return ret;
345}
346
347static void _copy_pte(pte_t *dst_ptep, pte_t *src_ptep, unsigned long addr)
348{
349 pte_t pte = READ_ONCE(*src_ptep);
350
351 if (pte_valid(pte)) {
352 /*
353 * Resume will overwrite areas that may be marked
354 * read only (code, rodata). Clear the RDONLY bit from
355 * the temporary mappings we use during restore.
356 */
357 set_pte(dst_ptep, pte_mkwrite(pte));
358 } else if (debug_pagealloc_enabled() && !pte_none(pte)) {
359 /*
360 * debug_pagealloc will removed the PTE_VALID bit if
361 * the page isn't in use by the resume kernel. It may have
362 * been in use by the original kernel, in which case we need
363 * to put it back in our copy to do the restore.
364 *
365 * Before marking this entry valid, check the pfn should
366 * be mapped.
367 */
368 BUG_ON(!pfn_valid(pte_pfn(pte)));
369
370 set_pte(dst_ptep, pte_mkpresent(pte_mkwrite(pte)));
371 }
372}
373
374static int copy_pte(pmd_t *dst_pmdp, pmd_t *src_pmdp, unsigned long start,
375 unsigned long end)
376{
377 pte_t *src_ptep;
378 pte_t *dst_ptep;
379 unsigned long addr = start;
380
381 dst_ptep = (pte_t *)get_safe_page(GFP_ATOMIC);
382 if (!dst_ptep)
383 return -ENOMEM;
384 pmd_populate_kernel(&init_mm, dst_pmdp, dst_ptep);
385 dst_ptep = pte_offset_kernel(dst_pmdp, start);
386
387 src_ptep = pte_offset_kernel(src_pmdp, start);
388 do {
389 _copy_pte(dst_ptep, src_ptep, addr);
390 } while (dst_ptep++, src_ptep++, addr += PAGE_SIZE, addr != end);
391
392 return 0;
393}
394
395static int copy_pmd(pud_t *dst_pudp, pud_t *src_pudp, unsigned long start,
396 unsigned long end)
397{
398 pmd_t *src_pmdp;
399 pmd_t *dst_pmdp;
400 unsigned long next;
401 unsigned long addr = start;
402
403 if (pud_none(READ_ONCE(*dst_pudp))) {
404 dst_pmdp = (pmd_t *)get_safe_page(GFP_ATOMIC);
405 if (!dst_pmdp)
406 return -ENOMEM;
407 pud_populate(&init_mm, dst_pudp, dst_pmdp);
408 }
409 dst_pmdp = pmd_offset(dst_pudp, start);
410
411 src_pmdp = pmd_offset(src_pudp, start);
412 do {
413 pmd_t pmd = READ_ONCE(*src_pmdp);
414
415 next = pmd_addr_end(addr, end);
416 if (pmd_none(pmd))
417 continue;
418 if (pmd_table(pmd)) {
419 if (copy_pte(dst_pmdp, src_pmdp, addr, next))
420 return -ENOMEM;
421 } else {
422 set_pmd(dst_pmdp,
423 __pmd(pmd_val(pmd) & ~PMD_SECT_RDONLY));
424 }
425 } while (dst_pmdp++, src_pmdp++, addr = next, addr != end);
426
427 return 0;
428}
429
430static int copy_pud(p4d_t *dst_p4dp, p4d_t *src_p4dp, unsigned long start,
431 unsigned long end)
432{
433 pud_t *dst_pudp;
434 pud_t *src_pudp;
435 unsigned long next;
436 unsigned long addr = start;
437
438 if (p4d_none(READ_ONCE(*dst_p4dp))) {
439 dst_pudp = (pud_t *)get_safe_page(GFP_ATOMIC);
440 if (!dst_pudp)
441 return -ENOMEM;
442 p4d_populate(&init_mm, dst_p4dp, dst_pudp);
443 }
444 dst_pudp = pud_offset(dst_p4dp, start);
445
446 src_pudp = pud_offset(src_p4dp, start);
447 do {
448 pud_t pud = READ_ONCE(*src_pudp);
449
450 next = pud_addr_end(addr, end);
451 if (pud_none(pud))
452 continue;
453 if (pud_table(pud)) {
454 if (copy_pmd(dst_pudp, src_pudp, addr, next))
455 return -ENOMEM;
456 } else {
457 set_pud(dst_pudp,
458 __pud(pud_val(pud) & ~PUD_SECT_RDONLY));
459 }
460 } while (dst_pudp++, src_pudp++, addr = next, addr != end);
461
462 return 0;
463}
464
465static int copy_p4d(pgd_t *dst_pgdp, pgd_t *src_pgdp, unsigned long start,
466 unsigned long end)
467{
468 p4d_t *dst_p4dp;
469 p4d_t *src_p4dp;
470 unsigned long next;
471 unsigned long addr = start;
472
473 dst_p4dp = p4d_offset(dst_pgdp, start);
474 src_p4dp = p4d_offset(src_pgdp, start);
475 do {
476 next = p4d_addr_end(addr, end);
477 if (p4d_none(READ_ONCE(*src_p4dp)))
478 continue;
479 if (copy_pud(dst_p4dp, src_p4dp, addr, next))
480 return -ENOMEM;
481 } while (dst_p4dp++, src_p4dp++, addr = next, addr != end);
482
483 return 0;
484}
485
486static int copy_page_tables(pgd_t *dst_pgdp, unsigned long start,
487 unsigned long end)
488{
489 unsigned long next;
490 unsigned long addr = start;
491 pgd_t *src_pgdp = pgd_offset_k(start);
492
493 dst_pgdp = pgd_offset_pgd(dst_pgdp, start);
494 do {
495 next = pgd_addr_end(addr, end);
496 if (pgd_none(READ_ONCE(*src_pgdp)))
497 continue;
498 if (copy_p4d(dst_pgdp, src_pgdp, addr, next))
499 return -ENOMEM;
500 } while (dst_pgdp++, src_pgdp++, addr = next, addr != end);
501
502 return 0;
503}
504
505static int trans_pgd_create_copy(pgd_t **dst_pgdp, unsigned long start,
506 unsigned long end)
507{
508 int rc;
509 pgd_t *trans_pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
510
511 if (!trans_pgd) {
512 pr_err("Failed to allocate memory for temporary page tables.\n");
513 return -ENOMEM;
514 }
515
516 rc = copy_page_tables(trans_pgd, start, end);
517 if (!rc)
518 *dst_pgdp = trans_pgd;
519
520 return rc;
521}
522
523/*
524 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
525 *
526 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
527 * we don't need to free it here.
528 */
529int swsusp_arch_resume(void)
530{
531 int rc;
532 void *zero_page;
533 size_t exit_size;
534 pgd_t *tmp_pg_dir;
535 phys_addr_t phys_hibernate_exit;
536 void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
537 void *, phys_addr_t, phys_addr_t);
538
539 /*
540 * Restoring the memory image will overwrite the ttbr1 page tables.
541 * Create a second copy of just the linear map, and use this when
542 * restoring.
543 */
544 rc = trans_pgd_create_copy(&tmp_pg_dir, PAGE_OFFSET, PAGE_END);
545 if (rc)
546 return rc;
547
548 /*
549 * We need a zero page that is zero before & after resume in order to
550 * to break before make on the ttbr1 page tables.
551 */
552 zero_page = (void *)get_safe_page(GFP_ATOMIC);
553 if (!zero_page) {
554 pr_err("Failed to allocate zero page.\n");
555 return -ENOMEM;
556 }
557
558 /*
559 * Locate the exit code in the bottom-but-one page, so that *NULL
560 * still has disastrous affects.
561 */
562 hibernate_exit = (void *)PAGE_SIZE;
563 exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
564 /*
565 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
566 * a new set of ttbr0 page tables and load them.
567 */
568 rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
569 (unsigned long)hibernate_exit,
570 &phys_hibernate_exit);
571 if (rc) {
572 pr_err("Failed to create safe executable page for hibernate_exit code.\n");
573 return rc;
574 }
575
576 /*
577 * The hibernate exit text contains a set of el2 vectors, that will
578 * be executed at el2 with the mmu off in order to reload hyp-stub.
579 */
580 __flush_dcache_area(hibernate_exit, exit_size);
581
582 /*
583 * KASLR will cause the el2 vectors to be in a different location in
584 * the resumed kernel. Load hibernate's temporary copy into el2.
585 *
586 * We can skip this step if we booted at EL1, or are running with VHE.
587 */
588 if (el2_reset_needed()) {
589 phys_addr_t el2_vectors = phys_hibernate_exit; /* base */
590 el2_vectors += hibernate_el2_vectors -
591 __hibernate_exit_text_start; /* offset */
592
593 __hyp_set_vectors(el2_vectors);
594 }
595
596 hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
597 resume_hdr.reenter_kernel, restore_pblist,
598 resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
599
600 return 0;
601}
602
603int hibernate_resume_nonboot_cpu_disable(void)
604{
605 if (sleep_cpu < 0) {
606 pr_err("Failing to resume from hibernate on an unknown CPU.\n");
607 return -ENODEV;
608 }
609
610 return freeze_secondary_cpus(sleep_cpu);
611}