2 * Hibernate support specific for ARM64
4 * Derived from work on ARM hibernation support by:
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/
13 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
15 * License terms: GNU General Public License (GPL) version 2
17 #define pr_fmt(x) "hibernate: " x
18 #include <linux/cpu.h>
19 #include <linux/kvm_host.h>
22 #include <linux/sched.h>
23 #include <linux/suspend.h>
24 #include <linux/utsname.h>
25 #include <linux/version.h>
27 #include <asm/barrier.h>
28 #include <asm/cacheflush.h>
29 #include <asm/cputype.h>
30 #include <asm/irqflags.h>
31 #include <asm/memory.h>
32 #include <asm/mmu_context.h>
33 #include <asm/pgalloc.h>
34 #include <asm/pgtable.h>
35 #include <asm/pgtable-hwdef.h>
36 #include <asm/sections.h>
38 #include <asm/smp_plat.h>
39 #include <asm/suspend.h>
40 #include <asm/sysreg.h>
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.
48 * defined as "__visible int in_suspend __nosavedata" in
49 * kernel/power/hibernate.c
51 extern int in_suspend;
53 /* Find a symbols alias in the linear map */
54 #define LMADDR(x) phys_to_virt(virt_to_phys(x))
56 /* Do we need to reset el2? */
57 #define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
59 /* temporary el2 vectors in the __hibernate_exit_text section. */
60 extern char hibernate_el2_vectors[];
62 /* hyp-stub vectors, used to restore el2 during resume from hibernate. */
63 extern char __hyp_stub_vectors[];
66 * The logical cpu number we should resume on, initialised to a non-cpu
69 static int sleep_cpu = -EINVAL;
72 * Values that may not change over hibernate/resume. We put the build number
73 * and date in here so that we guarantee not to resume with a different
76 struct arch_hibernate_hdr_invariants {
77 char uts_version[__NEW_UTS_LEN + 1];
80 /* These values need to be know across a hibernate/restore. */
81 static struct arch_hibernate_hdr {
82 struct arch_hibernate_hdr_invariants invariants;
84 /* These are needed to find the relocated kernel if built with kaslr */
85 phys_addr_t ttbr1_el1;
86 void (*reenter_kernel)(void);
89 * We need to know where the __hyp_stub_vectors are after restore to
92 phys_addr_t __hyp_stub_vectors;
97 static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
99 memset(i, 0, sizeof(*i));
100 memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
103 int pfn_is_nosave(unsigned long pfn)
105 unsigned long nosave_begin_pfn = virt_to_pfn(&__nosave_begin);
106 unsigned long nosave_end_pfn = virt_to_pfn(&__nosave_end - 1);
108 return (pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn);
111 void notrace save_processor_state(void)
113 WARN_ON(num_online_cpus() != 1);
116 void notrace restore_processor_state(void)
120 int arch_hibernation_header_save(void *addr, unsigned int max_size)
122 struct arch_hibernate_hdr *hdr = addr;
124 if (max_size < sizeof(*hdr))
127 arch_hdr_invariants(&hdr->invariants);
128 hdr->ttbr1_el1 = virt_to_phys(swapper_pg_dir);
129 hdr->reenter_kernel = _cpu_resume;
131 /* We can't use __hyp_get_vectors() because kvm may still be loaded */
132 if (el2_reset_needed())
133 hdr->__hyp_stub_vectors = virt_to_phys(__hyp_stub_vectors);
135 hdr->__hyp_stub_vectors = 0;
137 /* Save the mpidr of the cpu we called cpu_suspend() on... */
139 pr_err("Failing to hibernate on an unkown CPU.\n");
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);
148 EXPORT_SYMBOL(arch_hibernation_header_save);
150 int arch_hibernation_header_restore(void *addr)
153 struct arch_hibernate_hdr_invariants invariants;
154 struct arch_hibernate_hdr *hdr = addr;
156 arch_hdr_invariants(&invariants);
157 if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
158 pr_crit("Hibernate image not generated by this kernel!\n");
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);
166 pr_crit("Hibernated on a CPU not known to this kernel!\n");
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);
174 pr_err("Failed to bring hibernate-CPU up!\n");
184 EXPORT_SYMBOL(arch_hibernation_header_restore);
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.
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.
195 * Length is provided as we probably only want 4K of data, even on a 64K
198 static 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),
209 unsigned long dst = (unsigned long)allocator(mask);
216 memcpy((void *)dst, src_start, length);
217 flush_icache_range(dst, dst + length);
219 pgd = pgd_offset_raw(allocator(mask), dst_addr);
220 if (pgd_none(*pgd)) {
221 pud = allocator(mask);
226 pgd_populate(&init_mm, pgd, pud);
229 pud = pud_offset(pgd, dst_addr);
230 if (pud_none(*pud)) {
231 pmd = allocator(mask);
236 pud_populate(&init_mm, pud, pmd);
239 pmd = pmd_offset(pud, dst_addr);
240 if (pmd_none(*pmd)) {
241 pte = allocator(mask);
246 pmd_populate_kernel(&init_mm, pmd, pte);
249 pte = pte_offset_kernel(pmd, dst_addr);
250 set_pte(pte, pfn_pte(virt_to_pfn(dst), PAGE_KERNEL_EXEC));
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.
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.
264 cpu_set_reserved_ttbr0();
265 local_flush_tlb_all();
266 write_sysreg(virt_to_phys(pgd), ttbr0_el1);
269 *phys_dst_addr = virt_to_phys((void *)dst);
275 #define dcache_clean_range(start, end) __flush_dcache_area(start, (end - start))
277 int swsusp_arch_suspend(void)
281 struct sleep_stack_data state;
283 if (cpus_are_stuck_in_kernel()) {
284 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
288 local_dbg_save(flags);
290 if (__cpu_suspend_enter(&state)) {
291 sleep_cpu = smp_processor_id();
294 /* Clean kernel core startup/idle code to PoC*/
295 dcache_clean_range(__mmuoff_data_start, __mmuoff_data_end);
296 dcache_clean_range(__idmap_text_start, __idmap_text_end);
298 /* Clean kvm setup code to PoC? */
299 if (el2_reset_needed()) {
300 dcache_clean_range(__hyp_idmap_text_start, __hyp_idmap_text_end);
301 dcache_clean_range(__hyp_text_start, __hyp_text_end);
305 * Tell the hibernation core that we've just restored
311 __cpu_suspend_exit();
314 * Just in case the boot kernel did turn the SSBD
315 * mitigation off behind our back, let's set the state
316 * to what we expect it to be.
318 switch (arm64_get_ssbd_state()) {
319 case ARM64_SSBD_FORCE_ENABLE:
320 case ARM64_SSBD_KERNEL:
321 arm64_set_ssbd_mitigation(true);
325 local_dbg_restore(flags);
330 static void _copy_pte(pte_t *dst_pte, pte_t *src_pte, unsigned long addr)
332 pte_t pte = *src_pte;
334 if (pte_valid(pte)) {
336 * Resume will overwrite areas that may be marked
337 * read only (code, rodata). Clear the RDONLY bit from
338 * the temporary mappings we use during restore.
340 set_pte(dst_pte, pte_clear_rdonly(pte));
341 } else if (debug_pagealloc_enabled() && !pte_none(pte)) {
343 * debug_pagealloc will removed the PTE_VALID bit if
344 * the page isn't in use by the resume kernel. It may have
345 * been in use by the original kernel, in which case we need
346 * to put it back in our copy to do the restore.
348 * Before marking this entry valid, check the pfn should
351 BUG_ON(!pfn_valid(pte_pfn(pte)));
353 set_pte(dst_pte, pte_mkpresent(pte_clear_rdonly(pte)));
357 static int copy_pte(pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long start,
362 unsigned long addr = start;
364 dst_pte = (pte_t *)get_safe_page(GFP_ATOMIC);
367 pmd_populate_kernel(&init_mm, dst_pmd, dst_pte);
368 dst_pte = pte_offset_kernel(dst_pmd, start);
370 src_pte = pte_offset_kernel(src_pmd, start);
372 _copy_pte(dst_pte, src_pte, addr);
373 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
378 static int copy_pmd(pud_t *dst_pud, pud_t *src_pud, unsigned long start,
384 unsigned long addr = start;
386 if (pud_none(*dst_pud)) {
387 dst_pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
390 pud_populate(&init_mm, dst_pud, dst_pmd);
392 dst_pmd = pmd_offset(dst_pud, start);
394 src_pmd = pmd_offset(src_pud, start);
396 next = pmd_addr_end(addr, end);
397 if (pmd_none(*src_pmd))
399 if (pmd_table(*src_pmd)) {
400 if (copy_pte(dst_pmd, src_pmd, addr, next))
404 __pmd(pmd_val(*src_pmd) & ~PMD_SECT_RDONLY));
406 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
411 static int copy_pud(pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long start,
417 unsigned long addr = start;
419 if (pgd_none(*dst_pgd)) {
420 dst_pud = (pud_t *)get_safe_page(GFP_ATOMIC);
423 pgd_populate(&init_mm, dst_pgd, dst_pud);
425 dst_pud = pud_offset(dst_pgd, start);
427 src_pud = pud_offset(src_pgd, start);
429 next = pud_addr_end(addr, end);
430 if (pud_none(*src_pud))
432 if (pud_table(*(src_pud))) {
433 if (copy_pmd(dst_pud, src_pud, addr, next))
437 __pud(pud_val(*src_pud) & ~PMD_SECT_RDONLY));
439 } while (dst_pud++, src_pud++, addr = next, addr != end);
444 static int copy_page_tables(pgd_t *dst_pgd, unsigned long start,
448 unsigned long addr = start;
449 pgd_t *src_pgd = pgd_offset_k(start);
451 dst_pgd = pgd_offset_raw(dst_pgd, start);
453 next = pgd_addr_end(addr, end);
454 if (pgd_none(*src_pgd))
456 if (copy_pud(dst_pgd, src_pgd, addr, next))
458 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
464 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
466 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
467 * we don't need to free it here.
469 int swsusp_arch_resume(void)
475 void *lm_restore_pblist;
476 phys_addr_t phys_hibernate_exit;
477 void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
478 void *, phys_addr_t, phys_addr_t);
481 * Restoring the memory image will overwrite the ttbr1 page tables.
482 * Create a second copy of just the linear map, and use this when
485 tmp_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
487 pr_err("Failed to allocate memory for temporary page tables.");
491 rc = copy_page_tables(tmp_pg_dir, PAGE_OFFSET, 0);
496 * Since we only copied the linear map, we need to find restore_pblist's
497 * linear map address.
499 lm_restore_pblist = LMADDR(restore_pblist);
502 * We need a zero page that is zero before & after resume in order to
503 * to break before make on the ttbr1 page tables.
505 zero_page = (void *)get_safe_page(GFP_ATOMIC);
507 pr_err("Failed to allocate zero page.");
513 * Locate the exit code in the bottom-but-one page, so that *NULL
514 * still has disastrous affects.
516 hibernate_exit = (void *)PAGE_SIZE;
517 exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
519 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
520 * a new set of ttbr0 page tables and load them.
522 rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
523 (unsigned long)hibernate_exit,
524 &phys_hibernate_exit,
525 (void *)get_safe_page, GFP_ATOMIC);
527 pr_err("Failed to create safe executable page for hibernate_exit code.");
532 * The hibernate exit text contains a set of el2 vectors, that will
533 * be executed at el2 with the mmu off in order to reload hyp-stub.
535 __flush_dcache_area(hibernate_exit, exit_size);
538 * KASLR will cause the el2 vectors to be in a different location in
539 * the resumed kernel. Load hibernate's temporary copy into el2.
541 * We can skip this step if we booted at EL1, or are running with VHE.
543 if (el2_reset_needed()) {
544 phys_addr_t el2_vectors = phys_hibernate_exit; /* base */
545 el2_vectors += hibernate_el2_vectors -
546 __hibernate_exit_text_start; /* offset */
548 __hyp_set_vectors(el2_vectors);
551 hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
552 resume_hdr.reenter_kernel, lm_restore_pblist,
553 resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
559 int hibernate_resume_nonboot_cpu_disable(void)
562 pr_err("Failing to resume from hibernate on an unkown CPU.\n");
566 return freeze_secondary_cpus(sleep_cpu);