1 // SPDX-License-Identifier: GPL-2.0-only
3 * Hibernate support specific for ARM64
5 * Derived from work on ARM hibernation support by:
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/
14 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
16 #define pr_fmt(x) "hibernate: " x
17 #include <linux/cpu.h>
18 #include <linux/kvm_host.h>
21 #include <linux/sched.h>
22 #include <linux/suspend.h>
23 #include <linux/utsname.h>
24 #include <linux/version.h>
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.h>
36 #include <asm/pgtable-hwdef.h>
37 #include <asm/sections.h>
39 #include <asm/smp_plat.h>
40 #include <asm/suspend.h>
41 #include <asm/sysreg.h>
45 * Hibernate core relies on this value being 0 on resume, and marks it
46 * __nosavedata assuming it will keep the resume kernel's '0' value. This
47 * doesn't happen with either KASLR.
49 * defined as "__visible int in_suspend __nosavedata" in
50 * kernel/power/hibernate.c
52 extern int in_suspend;
54 /* Do we need to reset el2? */
55 #define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
57 /* temporary el2 vectors in the __hibernate_exit_text section. */
58 extern char hibernate_el2_vectors[];
60 /* hyp-stub vectors, used to restore el2 during resume from hibernate. */
61 extern char __hyp_stub_vectors[];
64 * The logical cpu number we should resume on, initialised to a non-cpu
67 static int sleep_cpu = -EINVAL;
70 * Values that may not change over hibernate/resume. We put the build number
71 * and date in here so that we guarantee not to resume with a different
74 struct arch_hibernate_hdr_invariants {
75 char uts_version[__NEW_UTS_LEN + 1];
78 /* These values need to be know across a hibernate/restore. */
79 static struct arch_hibernate_hdr {
80 struct arch_hibernate_hdr_invariants invariants;
82 /* These are needed to find the relocated kernel if built with kaslr */
83 phys_addr_t ttbr1_el1;
84 void (*reenter_kernel)(void);
87 * We need to know where the __hyp_stub_vectors are after restore to
90 phys_addr_t __hyp_stub_vectors;
95 static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
97 memset(i, 0, sizeof(*i));
98 memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
101 int pfn_is_nosave(unsigned long pfn)
103 unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
104 unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
106 return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn)) ||
107 crash_is_nosave(pfn);
110 void notrace save_processor_state(void)
112 WARN_ON(num_online_cpus() != 1);
115 void notrace restore_processor_state(void)
119 int arch_hibernation_header_save(void *addr, unsigned int max_size)
121 struct arch_hibernate_hdr *hdr = addr;
123 if (max_size < sizeof(*hdr))
126 arch_hdr_invariants(&hdr->invariants);
127 hdr->ttbr1_el1 = __pa_symbol(swapper_pg_dir);
128 hdr->reenter_kernel = _cpu_resume;
130 /* We can't use __hyp_get_vectors() because kvm may still be loaded */
131 if (el2_reset_needed())
132 hdr->__hyp_stub_vectors = __pa_symbol(__hyp_stub_vectors);
134 hdr->__hyp_stub_vectors = 0;
136 /* Save the mpidr of the cpu we called cpu_suspend() on... */
138 pr_err("Failing to hibernate on an unknown CPU.\n");
141 hdr->sleep_cpu_mpidr = cpu_logical_map(sleep_cpu);
142 pr_info("Hibernating on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
143 hdr->sleep_cpu_mpidr);
147 EXPORT_SYMBOL(arch_hibernation_header_save);
149 int arch_hibernation_header_restore(void *addr)
152 struct arch_hibernate_hdr_invariants invariants;
153 struct arch_hibernate_hdr *hdr = addr;
155 arch_hdr_invariants(&invariants);
156 if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
157 pr_crit("Hibernate image not generated by this kernel!\n");
161 sleep_cpu = get_logical_index(hdr->sleep_cpu_mpidr);
162 pr_info("Hibernated on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
163 hdr->sleep_cpu_mpidr);
165 pr_crit("Hibernated on a CPU not known to this kernel!\n");
170 ret = bringup_hibernate_cpu(sleep_cpu);
180 EXPORT_SYMBOL(arch_hibernation_header_restore);
182 static int trans_pgd_map_page(pgd_t *trans_pgd, void *page,
183 unsigned long dst_addr,
191 pgdp = pgd_offset_raw(trans_pgd, dst_addr);
192 if (pgd_none(READ_ONCE(*pgdp))) {
193 pudp = (void *)get_safe_page(GFP_ATOMIC);
196 pgd_populate(&init_mm, pgdp, pudp);
199 pudp = pud_offset(pgdp, dst_addr);
200 if (pud_none(READ_ONCE(*pudp))) {
201 pmdp = (void *)get_safe_page(GFP_ATOMIC);
204 pud_populate(&init_mm, pudp, pmdp);
207 pmdp = pmd_offset(pudp, dst_addr);
208 if (pmd_none(READ_ONCE(*pmdp))) {
209 ptep = (void *)get_safe_page(GFP_ATOMIC);
212 pmd_populate_kernel(&init_mm, pmdp, ptep);
215 ptep = pte_offset_kernel(pmdp, dst_addr);
216 set_pte(ptep, pfn_pte(virt_to_pfn(page), PAGE_KERNEL_EXEC));
222 * Copies length bytes, starting at src_start into an new page,
223 * perform cache maintenance, then maps it at the specified address low
224 * address as executable.
226 * This is used by hibernate to copy the code it needs to execute when
227 * overwriting the kernel text. This function generates a new set of page
228 * tables, which it loads into ttbr0.
230 * Length is provided as we probably only want 4K of data, even on a 64K
233 static int create_safe_exec_page(void *src_start, size_t length,
234 unsigned long dst_addr,
235 phys_addr_t *phys_dst_addr)
237 void *page = (void *)get_safe_page(GFP_ATOMIC);
244 memcpy(page, src_start, length);
245 __flush_icache_range((unsigned long)page, (unsigned long)page + length);
247 trans_pgd = (void *)get_safe_page(GFP_ATOMIC);
251 rc = trans_pgd_map_page(trans_pgd, page, dst_addr,
257 * Load our new page tables. A strict BBM approach requires that we
258 * ensure that TLBs are free of any entries that may overlap with the
259 * global mappings we are about to install.
261 * For a real hibernate/resume cycle TTBR0 currently points to a zero
262 * page, but TLBs may contain stale ASID-tagged entries (e.g. for EFI
263 * runtime services), while for a userspace-driven test_resume cycle it
264 * points to userspace page tables (and we must point it at a zero page
265 * ourselves). Elsewhere we only (un)install the idmap with preemption
266 * disabled, so T0SZ should be as required regardless.
268 cpu_set_reserved_ttbr0();
269 local_flush_tlb_all();
270 write_sysreg(phys_to_ttbr(virt_to_phys(trans_pgd)), ttbr0_el1);
273 *phys_dst_addr = virt_to_phys(page);
278 #define dcache_clean_range(start, end) __flush_dcache_area(start, (end - start))
280 int swsusp_arch_suspend(void)
284 struct sleep_stack_data state;
286 if (cpus_are_stuck_in_kernel()) {
287 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
291 flags = local_daif_save();
293 if (__cpu_suspend_enter(&state)) {
294 /* make the crash dump kernel image visible/saveable */
295 crash_prepare_suspend();
297 sleep_cpu = smp_processor_id();
300 /* Clean kernel core startup/idle code to PoC*/
301 dcache_clean_range(__mmuoff_data_start, __mmuoff_data_end);
302 dcache_clean_range(__idmap_text_start, __idmap_text_end);
304 /* Clean kvm setup code to PoC? */
305 if (el2_reset_needed()) {
306 dcache_clean_range(__hyp_idmap_text_start, __hyp_idmap_text_end);
307 dcache_clean_range(__hyp_text_start, __hyp_text_end);
310 /* make the crash dump kernel image protected again */
314 * Tell the hibernation core that we've just restored
320 __cpu_suspend_exit();
323 * Just in case the boot kernel did turn the SSBD
324 * mitigation off behind our back, let's set the state
325 * to what we expect it to be.
327 switch (arm64_get_ssbd_state()) {
328 case ARM64_SSBD_FORCE_ENABLE:
329 case ARM64_SSBD_KERNEL:
330 arm64_set_ssbd_mitigation(true);
334 local_daif_restore(flags);
339 static void _copy_pte(pte_t *dst_ptep, pte_t *src_ptep, unsigned long addr)
341 pte_t pte = READ_ONCE(*src_ptep);
343 if (pte_valid(pte)) {
345 * Resume will overwrite areas that may be marked
346 * read only (code, rodata). Clear the RDONLY bit from
347 * the temporary mappings we use during restore.
349 set_pte(dst_ptep, pte_mkwrite(pte));
350 } else if (debug_pagealloc_enabled() && !pte_none(pte)) {
352 * debug_pagealloc will removed the PTE_VALID bit if
353 * the page isn't in use by the resume kernel. It may have
354 * been in use by the original kernel, in which case we need
355 * to put it back in our copy to do the restore.
357 * Before marking this entry valid, check the pfn should
360 BUG_ON(!pfn_valid(pte_pfn(pte)));
362 set_pte(dst_ptep, pte_mkpresent(pte_mkwrite(pte)));
366 static int copy_pte(pmd_t *dst_pmdp, pmd_t *src_pmdp, unsigned long start,
371 unsigned long addr = start;
373 dst_ptep = (pte_t *)get_safe_page(GFP_ATOMIC);
376 pmd_populate_kernel(&init_mm, dst_pmdp, dst_ptep);
377 dst_ptep = pte_offset_kernel(dst_pmdp, start);
379 src_ptep = pte_offset_kernel(src_pmdp, start);
381 _copy_pte(dst_ptep, src_ptep, addr);
382 } while (dst_ptep++, src_ptep++, addr += PAGE_SIZE, addr != end);
387 static int copy_pmd(pud_t *dst_pudp, pud_t *src_pudp, unsigned long start,
393 unsigned long addr = start;
395 if (pud_none(READ_ONCE(*dst_pudp))) {
396 dst_pmdp = (pmd_t *)get_safe_page(GFP_ATOMIC);
399 pud_populate(&init_mm, dst_pudp, dst_pmdp);
401 dst_pmdp = pmd_offset(dst_pudp, start);
403 src_pmdp = pmd_offset(src_pudp, start);
405 pmd_t pmd = READ_ONCE(*src_pmdp);
407 next = pmd_addr_end(addr, end);
410 if (pmd_table(pmd)) {
411 if (copy_pte(dst_pmdp, src_pmdp, addr, next))
415 __pmd(pmd_val(pmd) & ~PMD_SECT_RDONLY));
417 } while (dst_pmdp++, src_pmdp++, addr = next, addr != end);
422 static int copy_pud(pgd_t *dst_pgdp, pgd_t *src_pgdp, unsigned long start,
428 unsigned long addr = start;
430 if (pgd_none(READ_ONCE(*dst_pgdp))) {
431 dst_pudp = (pud_t *)get_safe_page(GFP_ATOMIC);
434 pgd_populate(&init_mm, dst_pgdp, dst_pudp);
436 dst_pudp = pud_offset(dst_pgdp, start);
438 src_pudp = pud_offset(src_pgdp, start);
440 pud_t pud = READ_ONCE(*src_pudp);
442 next = pud_addr_end(addr, end);
445 if (pud_table(pud)) {
446 if (copy_pmd(dst_pudp, src_pudp, addr, next))
450 __pud(pud_val(pud) & ~PUD_SECT_RDONLY));
452 } while (dst_pudp++, src_pudp++, addr = next, addr != end);
457 static int copy_page_tables(pgd_t *dst_pgdp, unsigned long start,
461 unsigned long addr = start;
462 pgd_t *src_pgdp = pgd_offset_k(start);
464 dst_pgdp = pgd_offset_raw(dst_pgdp, start);
466 next = pgd_addr_end(addr, end);
467 if (pgd_none(READ_ONCE(*src_pgdp)))
469 if (copy_pud(dst_pgdp, src_pgdp, addr, next))
471 } while (dst_pgdp++, src_pgdp++, addr = next, addr != end);
476 static int trans_pgd_create_copy(pgd_t **dst_pgdp, unsigned long start,
480 pgd_t *trans_pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
483 pr_err("Failed to allocate memory for temporary page tables.\n");
487 rc = copy_page_tables(trans_pgd, start, end);
489 *dst_pgdp = trans_pgd;
495 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
497 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
498 * we don't need to free it here.
500 int swsusp_arch_resume(void)
506 phys_addr_t phys_hibernate_exit;
507 void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
508 void *, phys_addr_t, phys_addr_t);
511 * Restoring the memory image will overwrite the ttbr1 page tables.
512 * Create a second copy of just the linear map, and use this when
515 rc = trans_pgd_create_copy(&tmp_pg_dir, PAGE_OFFSET, PAGE_END);
520 * We need a zero page that is zero before & after resume in order to
521 * to break before make on the ttbr1 page tables.
523 zero_page = (void *)get_safe_page(GFP_ATOMIC);
525 pr_err("Failed to allocate zero page.\n");
530 * Locate the exit code in the bottom-but-one page, so that *NULL
531 * still has disastrous affects.
533 hibernate_exit = (void *)PAGE_SIZE;
534 exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
536 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
537 * a new set of ttbr0 page tables and load them.
539 rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
540 (unsigned long)hibernate_exit,
541 &phys_hibernate_exit);
543 pr_err("Failed to create safe executable page for hibernate_exit code.\n");
548 * The hibernate exit text contains a set of el2 vectors, that will
549 * be executed at el2 with the mmu off in order to reload hyp-stub.
551 __flush_dcache_area(hibernate_exit, exit_size);
554 * KASLR will cause the el2 vectors to be in a different location in
555 * the resumed kernel. Load hibernate's temporary copy into el2.
557 * We can skip this step if we booted at EL1, or are running with VHE.
559 if (el2_reset_needed()) {
560 phys_addr_t el2_vectors = phys_hibernate_exit; /* base */
561 el2_vectors += hibernate_el2_vectors -
562 __hibernate_exit_text_start; /* offset */
564 __hyp_set_vectors(el2_vectors);
567 hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
568 resume_hdr.reenter_kernel, restore_pblist,
569 resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
574 int hibernate_resume_nonboot_cpu_disable(void)
577 pr_err("Failing to resume from hibernate on an unknown CPU.\n");
581 return freeze_secondary_cpus(sleep_cpu);