Loading...
1/*:
2 * Hibernate support specific for ARM64
3 *
4 * Derived from work on ARM hibernation support by:
5 *
6 * Ubuntu project, hibernation support for mach-dove
7 * Copyright (C) 2010 Nokia Corporation (Hiroshi Doyu)
8 * Copyright (C) 2010 Texas Instruments, Inc. (Teerth Reddy et al.)
9 * https://lkml.org/lkml/2010/6/18/4
10 * https://lists.linux-foundation.org/pipermail/linux-pm/2010-June/027422.html
11 * https://patchwork.kernel.org/patch/96442/
12 *
13 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
14 *
15 * License terms: GNU General Public License (GPL) version 2
16 */
17#define pr_fmt(x) "hibernate: " x
18#include <linux/cpu.h>
19#include <linux/kvm_host.h>
20#include <linux/mm.h>
21#include <linux/pm.h>
22#include <linux/sched.h>
23#include <linux/suspend.h>
24#include <linux/utsname.h>
25#include <linux/version.h>
26
27#include <asm/barrier.h>
28#include <asm/cacheflush.h>
29#include <asm/cputype.h>
30#include <asm/daifflags.h>
31#include <asm/irqflags.h>
32#include <asm/kexec.h>
33#include <asm/memory.h>
34#include <asm/mmu_context.h>
35#include <asm/pgalloc.h>
36#include <asm/pgtable.h>
37#include <asm/pgtable-hwdef.h>
38#include <asm/sections.h>
39#include <asm/smp.h>
40#include <asm/smp_plat.h>
41#include <asm/suspend.h>
42#include <asm/sysreg.h>
43#include <asm/virt.h>
44
45/*
46 * Hibernate core relies on this value being 0 on resume, and marks it
47 * __nosavedata assuming it will keep the resume kernel's '0' value. This
48 * doesn't happen with either KASLR.
49 *
50 * defined as "__visible int in_suspend __nosavedata" in
51 * kernel/power/hibernate.c
52 */
53extern int in_suspend;
54
55/* Do we need to reset el2? */
56#define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
57
58/* temporary el2 vectors in the __hibernate_exit_text section. */
59extern char hibernate_el2_vectors[];
60
61/* hyp-stub vectors, used to restore el2 during resume from hibernate. */
62extern char __hyp_stub_vectors[];
63
64/*
65 * The logical cpu number we should resume on, initialised to a non-cpu
66 * number.
67 */
68static int sleep_cpu = -EINVAL;
69
70/*
71 * Values that may not change over hibernate/resume. We put the build number
72 * and date in here so that we guarantee not to resume with a different
73 * kernel.
74 */
75struct arch_hibernate_hdr_invariants {
76 char uts_version[__NEW_UTS_LEN + 1];
77};
78
79/* These values need to be know across a hibernate/restore. */
80static struct arch_hibernate_hdr {
81 struct arch_hibernate_hdr_invariants invariants;
82
83 /* These are needed to find the relocated kernel if built with kaslr */
84 phys_addr_t ttbr1_el1;
85 void (*reenter_kernel)(void);
86
87 /*
88 * We need to know where the __hyp_stub_vectors are after restore to
89 * re-configure el2.
90 */
91 phys_addr_t __hyp_stub_vectors;
92
93 u64 sleep_cpu_mpidr;
94} resume_hdr;
95
96static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
97{
98 memset(i, 0, sizeof(*i));
99 memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
100}
101
102int pfn_is_nosave(unsigned long pfn)
103{
104 unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
105 unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
106
107 return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn)) ||
108 crash_is_nosave(pfn);
109}
110
111void notrace save_processor_state(void)
112{
113 WARN_ON(num_online_cpus() != 1);
114}
115
116void notrace restore_processor_state(void)
117{
118}
119
120int arch_hibernation_header_save(void *addr, unsigned int max_size)
121{
122 struct arch_hibernate_hdr *hdr = addr;
123
124 if (max_size < sizeof(*hdr))
125 return -EOVERFLOW;
126
127 arch_hdr_invariants(&hdr->invariants);
128 hdr->ttbr1_el1 = __pa_symbol(swapper_pg_dir);
129 hdr->reenter_kernel = _cpu_resume;
130
131 /* We can't use __hyp_get_vectors() because kvm may still be loaded */
132 if (el2_reset_needed())
133 hdr->__hyp_stub_vectors = __pa_symbol(__hyp_stub_vectors);
134 else
135 hdr->__hyp_stub_vectors = 0;
136
137 /* Save the mpidr of the cpu we called cpu_suspend() on... */
138 if (sleep_cpu < 0) {
139 pr_err("Failing to hibernate on an unknown CPU.\n");
140 return -ENODEV;
141 }
142 hdr->sleep_cpu_mpidr = cpu_logical_map(sleep_cpu);
143 pr_info("Hibernating on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
144 hdr->sleep_cpu_mpidr);
145
146 return 0;
147}
148EXPORT_SYMBOL(arch_hibernation_header_save);
149
150int arch_hibernation_header_restore(void *addr)
151{
152 int ret;
153 struct arch_hibernate_hdr_invariants invariants;
154 struct arch_hibernate_hdr *hdr = addr;
155
156 arch_hdr_invariants(&invariants);
157 if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
158 pr_crit("Hibernate image not generated by this kernel!\n");
159 return -EINVAL;
160 }
161
162 sleep_cpu = get_logical_index(hdr->sleep_cpu_mpidr);
163 pr_info("Hibernated on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
164 hdr->sleep_cpu_mpidr);
165 if (sleep_cpu < 0) {
166 pr_crit("Hibernated on a CPU not known to this kernel!\n");
167 sleep_cpu = -EINVAL;
168 return -EINVAL;
169 }
170 if (!cpu_online(sleep_cpu)) {
171 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
172 ret = cpu_up(sleep_cpu);
173 if (ret) {
174 pr_err("Failed to bring hibernate-CPU up!\n");
175 sleep_cpu = -EINVAL;
176 return ret;
177 }
178 }
179
180 resume_hdr = *hdr;
181
182 return 0;
183}
184EXPORT_SYMBOL(arch_hibernation_header_restore);
185
186/*
187 * Copies length bytes, starting at src_start into an new page,
188 * perform cache maintentance, then maps it at the specified address low
189 * address as executable.
190 *
191 * This is used by hibernate to copy the code it needs to execute when
192 * overwriting the kernel text. This function generates a new set of page
193 * tables, which it loads into ttbr0.
194 *
195 * Length is provided as we probably only want 4K of data, even on a 64K
196 * page system.
197 */
198static int create_safe_exec_page(void *src_start, size_t length,
199 unsigned long dst_addr,
200 phys_addr_t *phys_dst_addr,
201 void *(*allocator)(gfp_t mask),
202 gfp_t mask)
203{
204 int rc = 0;
205 pgd_t *pgdp;
206 pud_t *pudp;
207 pmd_t *pmdp;
208 pte_t *ptep;
209 unsigned long dst = (unsigned long)allocator(mask);
210
211 if (!dst) {
212 rc = -ENOMEM;
213 goto out;
214 }
215
216 memcpy((void *)dst, src_start, length);
217 flush_icache_range(dst, dst + length);
218
219 pgdp = pgd_offset_raw(allocator(mask), dst_addr);
220 if (pgd_none(READ_ONCE(*pgdp))) {
221 pudp = allocator(mask);
222 if (!pudp) {
223 rc = -ENOMEM;
224 goto out;
225 }
226 pgd_populate(&init_mm, pgdp, pudp);
227 }
228
229 pudp = pud_offset(pgdp, dst_addr);
230 if (pud_none(READ_ONCE(*pudp))) {
231 pmdp = allocator(mask);
232 if (!pmdp) {
233 rc = -ENOMEM;
234 goto out;
235 }
236 pud_populate(&init_mm, pudp, pmdp);
237 }
238
239 pmdp = pmd_offset(pudp, dst_addr);
240 if (pmd_none(READ_ONCE(*pmdp))) {
241 ptep = allocator(mask);
242 if (!ptep) {
243 rc = -ENOMEM;
244 goto out;
245 }
246 pmd_populate_kernel(&init_mm, pmdp, ptep);
247 }
248
249 ptep = pte_offset_kernel(pmdp, dst_addr);
250 set_pte(ptep, pfn_pte(virt_to_pfn(dst), PAGE_KERNEL_EXEC));
251
252 /*
253 * Load our new page tables. A strict BBM approach requires that we
254 * ensure that TLBs are free of any entries that may overlap with the
255 * global mappings we are about to install.
256 *
257 * For a real hibernate/resume cycle TTBR0 currently points to a zero
258 * page, but TLBs may contain stale ASID-tagged entries (e.g. for EFI
259 * runtime services), while for a userspace-driven test_resume cycle it
260 * points to userspace page tables (and we must point it at a zero page
261 * ourselves). Elsewhere we only (un)install the idmap with preemption
262 * disabled, so T0SZ should be as required regardless.
263 */
264 cpu_set_reserved_ttbr0();
265 local_flush_tlb_all();
266 write_sysreg(phys_to_ttbr(virt_to_phys(pgdp)), ttbr0_el1);
267 isb();
268
269 *phys_dst_addr = virt_to_phys((void *)dst);
270
271out:
272 return rc;
273}
274
275#define dcache_clean_range(start, end) __flush_dcache_area(start, (end - start))
276
277int swsusp_arch_suspend(void)
278{
279 int ret = 0;
280 unsigned long flags;
281 struct sleep_stack_data state;
282
283 if (cpus_are_stuck_in_kernel()) {
284 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
285 return -EBUSY;
286 }
287
288 flags = local_daif_save();
289
290 if (__cpu_suspend_enter(&state)) {
291 /* make the crash dump kernel image visible/saveable */
292 crash_prepare_suspend();
293
294 sleep_cpu = smp_processor_id();
295 ret = swsusp_save();
296 } else {
297 /* Clean kernel core startup/idle code to PoC*/
298 dcache_clean_range(__mmuoff_data_start, __mmuoff_data_end);
299 dcache_clean_range(__idmap_text_start, __idmap_text_end);
300
301 /* Clean kvm setup code to PoC? */
302 if (el2_reset_needed())
303 dcache_clean_range(__hyp_idmap_text_start, __hyp_idmap_text_end);
304
305 /* make the crash dump kernel image protected again */
306 crash_post_resume();
307
308 /*
309 * Tell the hibernation core that we've just restored
310 * the memory
311 */
312 in_suspend = 0;
313
314 sleep_cpu = -EINVAL;
315 __cpu_suspend_exit();
316 }
317
318 local_daif_restore(flags);
319
320 return ret;
321}
322
323static void _copy_pte(pte_t *dst_ptep, pte_t *src_ptep, unsigned long addr)
324{
325 pte_t pte = READ_ONCE(*src_ptep);
326
327 if (pte_valid(pte)) {
328 /*
329 * Resume will overwrite areas that may be marked
330 * read only (code, rodata). Clear the RDONLY bit from
331 * the temporary mappings we use during restore.
332 */
333 set_pte(dst_ptep, pte_mkwrite(pte));
334 } else if (debug_pagealloc_enabled() && !pte_none(pte)) {
335 /*
336 * debug_pagealloc will removed the PTE_VALID bit if
337 * the page isn't in use by the resume kernel. It may have
338 * been in use by the original kernel, in which case we need
339 * to put it back in our copy to do the restore.
340 *
341 * Before marking this entry valid, check the pfn should
342 * be mapped.
343 */
344 BUG_ON(!pfn_valid(pte_pfn(pte)));
345
346 set_pte(dst_ptep, pte_mkpresent(pte_mkwrite(pte)));
347 }
348}
349
350static int copy_pte(pmd_t *dst_pmdp, pmd_t *src_pmdp, unsigned long start,
351 unsigned long end)
352{
353 pte_t *src_ptep;
354 pte_t *dst_ptep;
355 unsigned long addr = start;
356
357 dst_ptep = (pte_t *)get_safe_page(GFP_ATOMIC);
358 if (!dst_ptep)
359 return -ENOMEM;
360 pmd_populate_kernel(&init_mm, dst_pmdp, dst_ptep);
361 dst_ptep = pte_offset_kernel(dst_pmdp, start);
362
363 src_ptep = pte_offset_kernel(src_pmdp, start);
364 do {
365 _copy_pte(dst_ptep, src_ptep, addr);
366 } while (dst_ptep++, src_ptep++, addr += PAGE_SIZE, addr != end);
367
368 return 0;
369}
370
371static int copy_pmd(pud_t *dst_pudp, pud_t *src_pudp, unsigned long start,
372 unsigned long end)
373{
374 pmd_t *src_pmdp;
375 pmd_t *dst_pmdp;
376 unsigned long next;
377 unsigned long addr = start;
378
379 if (pud_none(READ_ONCE(*dst_pudp))) {
380 dst_pmdp = (pmd_t *)get_safe_page(GFP_ATOMIC);
381 if (!dst_pmdp)
382 return -ENOMEM;
383 pud_populate(&init_mm, dst_pudp, dst_pmdp);
384 }
385 dst_pmdp = pmd_offset(dst_pudp, start);
386
387 src_pmdp = pmd_offset(src_pudp, start);
388 do {
389 pmd_t pmd = READ_ONCE(*src_pmdp);
390
391 next = pmd_addr_end(addr, end);
392 if (pmd_none(pmd))
393 continue;
394 if (pmd_table(pmd)) {
395 if (copy_pte(dst_pmdp, src_pmdp, addr, next))
396 return -ENOMEM;
397 } else {
398 set_pmd(dst_pmdp,
399 __pmd(pmd_val(pmd) & ~PMD_SECT_RDONLY));
400 }
401 } while (dst_pmdp++, src_pmdp++, addr = next, addr != end);
402
403 return 0;
404}
405
406static int copy_pud(pgd_t *dst_pgdp, pgd_t *src_pgdp, unsigned long start,
407 unsigned long end)
408{
409 pud_t *dst_pudp;
410 pud_t *src_pudp;
411 unsigned long next;
412 unsigned long addr = start;
413
414 if (pgd_none(READ_ONCE(*dst_pgdp))) {
415 dst_pudp = (pud_t *)get_safe_page(GFP_ATOMIC);
416 if (!dst_pudp)
417 return -ENOMEM;
418 pgd_populate(&init_mm, dst_pgdp, dst_pudp);
419 }
420 dst_pudp = pud_offset(dst_pgdp, start);
421
422 src_pudp = pud_offset(src_pgdp, start);
423 do {
424 pud_t pud = READ_ONCE(*src_pudp);
425
426 next = pud_addr_end(addr, end);
427 if (pud_none(pud))
428 continue;
429 if (pud_table(pud)) {
430 if (copy_pmd(dst_pudp, src_pudp, addr, next))
431 return -ENOMEM;
432 } else {
433 set_pud(dst_pudp,
434 __pud(pud_val(pud) & ~PMD_SECT_RDONLY));
435 }
436 } while (dst_pudp++, src_pudp++, addr = next, addr != end);
437
438 return 0;
439}
440
441static int copy_page_tables(pgd_t *dst_pgdp, unsigned long start,
442 unsigned long end)
443{
444 unsigned long next;
445 unsigned long addr = start;
446 pgd_t *src_pgdp = pgd_offset_k(start);
447
448 dst_pgdp = pgd_offset_raw(dst_pgdp, start);
449 do {
450 next = pgd_addr_end(addr, end);
451 if (pgd_none(READ_ONCE(*src_pgdp)))
452 continue;
453 if (copy_pud(dst_pgdp, src_pgdp, addr, next))
454 return -ENOMEM;
455 } while (dst_pgdp++, src_pgdp++, addr = next, addr != end);
456
457 return 0;
458}
459
460/*
461 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
462 *
463 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
464 * we don't need to free it here.
465 */
466int swsusp_arch_resume(void)
467{
468 int rc = 0;
469 void *zero_page;
470 size_t exit_size;
471 pgd_t *tmp_pg_dir;
472 phys_addr_t phys_hibernate_exit;
473 void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
474 void *, phys_addr_t, phys_addr_t);
475
476 /*
477 * Restoring the memory image will overwrite the ttbr1 page tables.
478 * Create a second copy of just the linear map, and use this when
479 * restoring.
480 */
481 tmp_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
482 if (!tmp_pg_dir) {
483 pr_err("Failed to allocate memory for temporary page tables.\n");
484 rc = -ENOMEM;
485 goto out;
486 }
487 rc = copy_page_tables(tmp_pg_dir, PAGE_OFFSET, 0);
488 if (rc)
489 goto out;
490
491 /*
492 * We need a zero page that is zero before & after resume in order to
493 * to break before make on the ttbr1 page tables.
494 */
495 zero_page = (void *)get_safe_page(GFP_ATOMIC);
496 if (!zero_page) {
497 pr_err("Failed to allocate zero page.\n");
498 rc = -ENOMEM;
499 goto out;
500 }
501
502 /*
503 * Locate the exit code in the bottom-but-one page, so that *NULL
504 * still has disastrous affects.
505 */
506 hibernate_exit = (void *)PAGE_SIZE;
507 exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
508 /*
509 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
510 * a new set of ttbr0 page tables and load them.
511 */
512 rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
513 (unsigned long)hibernate_exit,
514 &phys_hibernate_exit,
515 (void *)get_safe_page, GFP_ATOMIC);
516 if (rc) {
517 pr_err("Failed to create safe executable page for hibernate_exit code.\n");
518 goto out;
519 }
520
521 /*
522 * The hibernate exit text contains a set of el2 vectors, that will
523 * be executed at el2 with the mmu off in order to reload hyp-stub.
524 */
525 __flush_dcache_area(hibernate_exit, exit_size);
526
527 /*
528 * KASLR will cause the el2 vectors to be in a different location in
529 * the resumed kernel. Load hibernate's temporary copy into el2.
530 *
531 * We can skip this step if we booted at EL1, or are running with VHE.
532 */
533 if (el2_reset_needed()) {
534 phys_addr_t el2_vectors = phys_hibernate_exit; /* base */
535 el2_vectors += hibernate_el2_vectors -
536 __hibernate_exit_text_start; /* offset */
537
538 __hyp_set_vectors(el2_vectors);
539 }
540
541 hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
542 resume_hdr.reenter_kernel, restore_pblist,
543 resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
544
545out:
546 return rc;
547}
548
549int hibernate_resume_nonboot_cpu_disable(void)
550{
551 if (sleep_cpu < 0) {
552 pr_err("Failing to resume from hibernate on an unknown CPU.\n");
553 return -ENODEV;
554 }
555
556 return freeze_secondary_cpus(sleep_cpu);
557}
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/pm.h>
20#include <linux/sched.h>
21#include <linux/suspend.h>
22#include <linux/utsname.h>
23
24#include <asm/barrier.h>
25#include <asm/cacheflush.h>
26#include <asm/cputype.h>
27#include <asm/daifflags.h>
28#include <asm/irqflags.h>
29#include <asm/kexec.h>
30#include <asm/memory.h>
31#include <asm/mmu_context.h>
32#include <asm/mte.h>
33#include <asm/sections.h>
34#include <asm/smp.h>
35#include <asm/smp_plat.h>
36#include <asm/suspend.h>
37#include <asm/sysreg.h>
38#include <asm/trans_pgd.h>
39#include <asm/virt.h>
40
41/*
42 * Hibernate core relies on this value being 0 on resume, and marks it
43 * __nosavedata assuming it will keep the resume kernel's '0' value. This
44 * doesn't happen with either KASLR.
45 *
46 * defined as "__visible int in_suspend __nosavedata" in
47 * kernel/power/hibernate.c
48 */
49extern int in_suspend;
50
51/* Do we need to reset el2? */
52#define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
53
54/* temporary el2 vectors in the __hibernate_exit_text section. */
55extern char hibernate_el2_vectors[];
56
57/* hyp-stub vectors, used to restore el2 during resume from hibernate. */
58extern char __hyp_stub_vectors[];
59
60/*
61 * The logical cpu number we should resume on, initialised to a non-cpu
62 * number.
63 */
64static int sleep_cpu = -EINVAL;
65
66/*
67 * Values that may not change over hibernate/resume. We put the build number
68 * and date in here so that we guarantee not to resume with a different
69 * kernel.
70 */
71struct arch_hibernate_hdr_invariants {
72 char uts_version[__NEW_UTS_LEN + 1];
73};
74
75/* These values need to be know across a hibernate/restore. */
76static struct arch_hibernate_hdr {
77 struct arch_hibernate_hdr_invariants invariants;
78
79 /* These are needed to find the relocated kernel if built with kaslr */
80 phys_addr_t ttbr1_el1;
81 void (*reenter_kernel)(void);
82
83 /*
84 * We need to know where the __hyp_stub_vectors are after restore to
85 * re-configure el2.
86 */
87 phys_addr_t __hyp_stub_vectors;
88
89 u64 sleep_cpu_mpidr;
90} resume_hdr;
91
92static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
93{
94 memset(i, 0, sizeof(*i));
95 memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
96}
97
98int pfn_is_nosave(unsigned long pfn)
99{
100 unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
101 unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
102
103 return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn)) ||
104 crash_is_nosave(pfn);
105}
106
107void notrace save_processor_state(void)
108{
109 WARN_ON(num_online_cpus() != 1);
110}
111
112void notrace restore_processor_state(void)
113{
114}
115
116int arch_hibernation_header_save(void *addr, unsigned int max_size)
117{
118 struct arch_hibernate_hdr *hdr = addr;
119
120 if (max_size < sizeof(*hdr))
121 return -EOVERFLOW;
122
123 arch_hdr_invariants(&hdr->invariants);
124 hdr->ttbr1_el1 = __pa_symbol(swapper_pg_dir);
125 hdr->reenter_kernel = _cpu_resume;
126
127 /* We can't use __hyp_get_vectors() because kvm may still be loaded */
128 if (el2_reset_needed())
129 hdr->__hyp_stub_vectors = __pa_symbol(__hyp_stub_vectors);
130 else
131 hdr->__hyp_stub_vectors = 0;
132
133 /* Save the mpidr of the cpu we called cpu_suspend() on... */
134 if (sleep_cpu < 0) {
135 pr_err("Failing to hibernate on an unknown CPU.\n");
136 return -ENODEV;
137 }
138 hdr->sleep_cpu_mpidr = cpu_logical_map(sleep_cpu);
139 pr_info("Hibernating on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
140 hdr->sleep_cpu_mpidr);
141
142 return 0;
143}
144EXPORT_SYMBOL(arch_hibernation_header_save);
145
146int arch_hibernation_header_restore(void *addr)
147{
148 int ret;
149 struct arch_hibernate_hdr_invariants invariants;
150 struct arch_hibernate_hdr *hdr = addr;
151
152 arch_hdr_invariants(&invariants);
153 if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
154 pr_crit("Hibernate image not generated by this kernel!\n");
155 return -EINVAL;
156 }
157
158 sleep_cpu = get_logical_index(hdr->sleep_cpu_mpidr);
159 pr_info("Hibernated on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
160 hdr->sleep_cpu_mpidr);
161 if (sleep_cpu < 0) {
162 pr_crit("Hibernated on a CPU not known to this kernel!\n");
163 sleep_cpu = -EINVAL;
164 return -EINVAL;
165 }
166
167 ret = bringup_hibernate_cpu(sleep_cpu);
168 if (ret) {
169 sleep_cpu = -EINVAL;
170 return ret;
171 }
172
173 resume_hdr = *hdr;
174
175 return 0;
176}
177EXPORT_SYMBOL(arch_hibernation_header_restore);
178
179static void *hibernate_page_alloc(void *arg)
180{
181 return (void *)get_safe_page((__force gfp_t)(unsigned long)arg);
182}
183
184/*
185 * Copies length bytes, starting at src_start into an new page,
186 * perform cache maintenance, then maps it at the specified address low
187 * address as executable.
188 *
189 * This is used by hibernate to copy the code it needs to execute when
190 * overwriting the kernel text. This function generates a new set of page
191 * tables, which it loads into ttbr0.
192 *
193 * Length is provided as we probably only want 4K of data, even on a 64K
194 * page system.
195 */
196static int create_safe_exec_page(void *src_start, size_t length,
197 phys_addr_t *phys_dst_addr)
198{
199 struct trans_pgd_info trans_info = {
200 .trans_alloc_page = hibernate_page_alloc,
201 .trans_alloc_arg = (__force void *)GFP_ATOMIC,
202 };
203
204 void *page = (void *)get_safe_page(GFP_ATOMIC);
205 phys_addr_t trans_ttbr0;
206 unsigned long t0sz;
207 int rc;
208
209 if (!page)
210 return -ENOMEM;
211
212 memcpy(page, src_start, length);
213 caches_clean_inval_pou((unsigned long)page, (unsigned long)page + length);
214 rc = trans_pgd_idmap_page(&trans_info, &trans_ttbr0, &t0sz, page);
215 if (rc)
216 return rc;
217
218 /*
219 * Load our new page tables. A strict BBM approach requires that we
220 * ensure that TLBs are free of any entries that may overlap with the
221 * global mappings we are about to install.
222 *
223 * For a real hibernate/resume cycle TTBR0 currently points to a zero
224 * page, but TLBs may contain stale ASID-tagged entries (e.g. for EFI
225 * runtime services), while for a userspace-driven test_resume cycle it
226 * points to userspace page tables (and we must point it at a zero page
227 * ourselves).
228 *
229 * We change T0SZ as part of installing the idmap. This is undone by
230 * cpu_uninstall_idmap() in __cpu_suspend_exit().
231 */
232 cpu_set_reserved_ttbr0();
233 local_flush_tlb_all();
234 __cpu_set_tcr_t0sz(t0sz);
235 write_sysreg(trans_ttbr0, ttbr0_el1);
236 isb();
237
238 *phys_dst_addr = virt_to_phys(page);
239
240 return 0;
241}
242
243#ifdef CONFIG_ARM64_MTE
244
245static DEFINE_XARRAY(mte_pages);
246
247static int save_tags(struct page *page, unsigned long pfn)
248{
249 void *tag_storage, *ret;
250
251 tag_storage = mte_allocate_tag_storage();
252 if (!tag_storage)
253 return -ENOMEM;
254
255 mte_save_page_tags(page_address(page), tag_storage);
256
257 ret = xa_store(&mte_pages, pfn, tag_storage, GFP_KERNEL);
258 if (WARN(xa_is_err(ret), "Failed to store MTE tags")) {
259 mte_free_tag_storage(tag_storage);
260 return xa_err(ret);
261 } else if (WARN(ret, "swsusp: %s: Duplicate entry", __func__)) {
262 mte_free_tag_storage(ret);
263 }
264
265 return 0;
266}
267
268static void swsusp_mte_free_storage(void)
269{
270 XA_STATE(xa_state, &mte_pages, 0);
271 void *tags;
272
273 xa_lock(&mte_pages);
274 xas_for_each(&xa_state, tags, ULONG_MAX) {
275 mte_free_tag_storage(tags);
276 }
277 xa_unlock(&mte_pages);
278
279 xa_destroy(&mte_pages);
280}
281
282static int swsusp_mte_save_tags(void)
283{
284 struct zone *zone;
285 unsigned long pfn, max_zone_pfn;
286 int ret = 0;
287 int n = 0;
288
289 if (!system_supports_mte())
290 return 0;
291
292 for_each_populated_zone(zone) {
293 max_zone_pfn = zone_end_pfn(zone);
294 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
295 struct page *page = pfn_to_online_page(pfn);
296
297 if (!page)
298 continue;
299
300 if (!test_bit(PG_mte_tagged, &page->flags))
301 continue;
302
303 ret = save_tags(page, pfn);
304 if (ret) {
305 swsusp_mte_free_storage();
306 goto out;
307 }
308
309 n++;
310 }
311 }
312 pr_info("Saved %d MTE pages\n", n);
313
314out:
315 return ret;
316}
317
318static void swsusp_mte_restore_tags(void)
319{
320 XA_STATE(xa_state, &mte_pages, 0);
321 int n = 0;
322 void *tags;
323
324 xa_lock(&mte_pages);
325 xas_for_each(&xa_state, tags, ULONG_MAX) {
326 unsigned long pfn = xa_state.xa_index;
327 struct page *page = pfn_to_online_page(pfn);
328
329 /*
330 * It is not required to invoke page_kasan_tag_reset(page)
331 * at this point since the tags stored in page->flags are
332 * already restored.
333 */
334 mte_restore_page_tags(page_address(page), tags);
335
336 mte_free_tag_storage(tags);
337 n++;
338 }
339 xa_unlock(&mte_pages);
340
341 pr_info("Restored %d MTE pages\n", n);
342
343 xa_destroy(&mte_pages);
344}
345
346#else /* CONFIG_ARM64_MTE */
347
348static int swsusp_mte_save_tags(void)
349{
350 return 0;
351}
352
353static void swsusp_mte_restore_tags(void)
354{
355}
356
357#endif /* CONFIG_ARM64_MTE */
358
359int swsusp_arch_suspend(void)
360{
361 int ret = 0;
362 unsigned long flags;
363 struct sleep_stack_data state;
364
365 if (cpus_are_stuck_in_kernel()) {
366 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
367 return -EBUSY;
368 }
369
370 flags = local_daif_save();
371
372 if (__cpu_suspend_enter(&state)) {
373 /* make the crash dump kernel image visible/saveable */
374 crash_prepare_suspend();
375
376 ret = swsusp_mte_save_tags();
377 if (ret)
378 return ret;
379
380 sleep_cpu = smp_processor_id();
381 ret = swsusp_save();
382 } else {
383 /* Clean kernel core startup/idle code to PoC*/
384 dcache_clean_inval_poc((unsigned long)__mmuoff_data_start,
385 (unsigned long)__mmuoff_data_end);
386 dcache_clean_inval_poc((unsigned long)__idmap_text_start,
387 (unsigned long)__idmap_text_end);
388
389 /* Clean kvm setup code to PoC? */
390 if (el2_reset_needed()) {
391 dcache_clean_inval_poc(
392 (unsigned long)__hyp_idmap_text_start,
393 (unsigned long)__hyp_idmap_text_end);
394 dcache_clean_inval_poc((unsigned long)__hyp_text_start,
395 (unsigned long)__hyp_text_end);
396 }
397
398 swsusp_mte_restore_tags();
399
400 /* make the crash dump kernel image protected again */
401 crash_post_resume();
402
403 /*
404 * Tell the hibernation core that we've just restored
405 * the memory
406 */
407 in_suspend = 0;
408
409 sleep_cpu = -EINVAL;
410 __cpu_suspend_exit();
411
412 /*
413 * Just in case the boot kernel did turn the SSBD
414 * mitigation off behind our back, let's set the state
415 * to what we expect it to be.
416 */
417 spectre_v4_enable_mitigation(NULL);
418 }
419
420 local_daif_restore(flags);
421
422 return ret;
423}
424
425/*
426 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
427 *
428 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
429 * we don't need to free it here.
430 */
431int swsusp_arch_resume(void)
432{
433 int rc;
434 void *zero_page;
435 size_t exit_size;
436 pgd_t *tmp_pg_dir;
437 void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
438 void *, phys_addr_t, phys_addr_t);
439 struct trans_pgd_info trans_info = {
440 .trans_alloc_page = hibernate_page_alloc,
441 .trans_alloc_arg = (void *)GFP_ATOMIC,
442 };
443
444 /*
445 * Restoring the memory image will overwrite the ttbr1 page tables.
446 * Create a second copy of just the linear map, and use this when
447 * restoring.
448 */
449 rc = trans_pgd_create_copy(&trans_info, &tmp_pg_dir, PAGE_OFFSET,
450 PAGE_END);
451 if (rc)
452 return rc;
453
454 /*
455 * We need a zero page that is zero before & after resume in order to
456 * to break before make on the ttbr1 page tables.
457 */
458 zero_page = (void *)get_safe_page(GFP_ATOMIC);
459 if (!zero_page) {
460 pr_err("Failed to allocate zero page.\n");
461 return -ENOMEM;
462 }
463
464 exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
465 /*
466 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
467 * a new set of ttbr0 page tables and load them.
468 */
469 rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
470 (phys_addr_t *)&hibernate_exit);
471 if (rc) {
472 pr_err("Failed to create safe executable page for hibernate_exit code.\n");
473 return rc;
474 }
475
476 /*
477 * The hibernate exit text contains a set of el2 vectors, that will
478 * be executed at el2 with the mmu off in order to reload hyp-stub.
479 */
480 dcache_clean_inval_poc((unsigned long)hibernate_exit,
481 (unsigned long)hibernate_exit + exit_size);
482
483 /*
484 * KASLR will cause the el2 vectors to be in a different location in
485 * the resumed kernel. Load hibernate's temporary copy into el2.
486 *
487 * We can skip this step if we booted at EL1, or are running with VHE.
488 */
489 if (el2_reset_needed()) {
490 phys_addr_t el2_vectors = (phys_addr_t)hibernate_exit;
491 el2_vectors += hibernate_el2_vectors -
492 __hibernate_exit_text_start; /* offset */
493
494 __hyp_set_vectors(el2_vectors);
495 }
496
497 hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
498 resume_hdr.reenter_kernel, restore_pblist,
499 resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
500
501 return 0;
502}
503
504int hibernate_resume_nonboot_cpu_disable(void)
505{
506 if (sleep_cpu < 0) {
507 pr_err("Failing to resume from hibernate on an unknown CPU.\n");
508 return -ENODEV;
509 }
510
511 return freeze_secondary_cpus(sleep_cpu);
512}