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/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");
169 if (!cpu_online(sleep_cpu)) {
170 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
171 ret = cpu_up(sleep_cpu);
173 pr_err("Failed to bring hibernate-CPU up!\n");
183 EXPORT_SYMBOL(arch_hibernation_header_restore);
186 * Copies length bytes, starting at src_start into an new page,
187 * perform cache maintentance, then maps it at the specified address low
188 * address as executable.
190 * This is used by hibernate to copy the code it needs to execute when
191 * overwriting the kernel text. This function generates a new set of page
192 * tables, which it loads into ttbr0.
194 * Length is provided as we probably only want 4K of data, even on a 64K
197 static int create_safe_exec_page(void *src_start, size_t length,
198 unsigned long dst_addr,
199 phys_addr_t *phys_dst_addr,
200 void *(*allocator)(gfp_t mask),
208 unsigned long dst = (unsigned long)allocator(mask);
215 memcpy((void *)dst, src_start, length);
216 flush_icache_range(dst, dst + length);
218 pgd = pgd_offset_raw(allocator(mask), dst_addr);
219 if (pgd_none(*pgd)) {
220 pud = allocator(mask);
225 pgd_populate(&init_mm, pgd, pud);
228 pud = pud_offset(pgd, dst_addr);
229 if (pud_none(*pud)) {
230 pmd = allocator(mask);
235 pud_populate(&init_mm, pud, pmd);
238 pmd = pmd_offset(pud, dst_addr);
239 if (pmd_none(*pmd)) {
240 pte = allocator(mask);
245 pmd_populate_kernel(&init_mm, pmd, pte);
248 pte = pte_offset_kernel(pmd, dst_addr);
249 set_pte(pte, __pte(virt_to_phys((void *)dst) |
250 pgprot_val(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 /* make the crash dump kernel image visible/saveable */
292 crash_prepare_suspend();
294 sleep_cpu = smp_processor_id();
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);
301 /* Clean kvm setup code to PoC? */
302 if (el2_reset_needed())
303 dcache_clean_range(__hyp_idmap_text_start, __hyp_idmap_text_end);
305 /* make the crash dump kernel image protected again */
309 * Tell the hibernation core that we've just restored
315 __cpu_suspend_exit();
318 local_dbg_restore(flags);
323 static void _copy_pte(pte_t *dst_pte, pte_t *src_pte, unsigned long addr)
325 pte_t pte = *src_pte;
327 if (pte_valid(pte)) {
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.
333 set_pte(dst_pte, pte_clear_rdonly(pte));
334 } else if (debug_pagealloc_enabled() && !pte_none(pte)) {
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.
341 * Before marking this entry valid, check the pfn should
344 BUG_ON(!pfn_valid(pte_pfn(pte)));
346 set_pte(dst_pte, pte_mkpresent(pte_clear_rdonly(pte)));
350 static int copy_pte(pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long start,
355 unsigned long addr = start;
357 dst_pte = (pte_t *)get_safe_page(GFP_ATOMIC);
360 pmd_populate_kernel(&init_mm, dst_pmd, dst_pte);
361 dst_pte = pte_offset_kernel(dst_pmd, start);
363 src_pte = pte_offset_kernel(src_pmd, start);
365 _copy_pte(dst_pte, src_pte, addr);
366 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
371 static int copy_pmd(pud_t *dst_pud, pud_t *src_pud, unsigned long start,
377 unsigned long addr = start;
379 if (pud_none(*dst_pud)) {
380 dst_pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
383 pud_populate(&init_mm, dst_pud, dst_pmd);
385 dst_pmd = pmd_offset(dst_pud, start);
387 src_pmd = pmd_offset(src_pud, start);
389 next = pmd_addr_end(addr, end);
390 if (pmd_none(*src_pmd))
392 if (pmd_table(*src_pmd)) {
393 if (copy_pte(dst_pmd, src_pmd, addr, next))
397 __pmd(pmd_val(*src_pmd) & ~PMD_SECT_RDONLY));
399 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
404 static int copy_pud(pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long start,
410 unsigned long addr = start;
412 if (pgd_none(*dst_pgd)) {
413 dst_pud = (pud_t *)get_safe_page(GFP_ATOMIC);
416 pgd_populate(&init_mm, dst_pgd, dst_pud);
418 dst_pud = pud_offset(dst_pgd, start);
420 src_pud = pud_offset(src_pgd, start);
422 next = pud_addr_end(addr, end);
423 if (pud_none(*src_pud))
425 if (pud_table(*(src_pud))) {
426 if (copy_pmd(dst_pud, src_pud, addr, next))
430 __pud(pud_val(*src_pud) & ~PMD_SECT_RDONLY));
432 } while (dst_pud++, src_pud++, addr = next, addr != end);
437 static int copy_page_tables(pgd_t *dst_pgd, unsigned long start,
441 unsigned long addr = start;
442 pgd_t *src_pgd = pgd_offset_k(start);
444 dst_pgd = pgd_offset_raw(dst_pgd, start);
446 next = pgd_addr_end(addr, end);
447 if (pgd_none(*src_pgd))
449 if (copy_pud(dst_pgd, src_pgd, addr, next))
451 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
457 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
459 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
460 * we don't need to free it here.
462 int swsusp_arch_resume(void)
468 phys_addr_t phys_hibernate_exit;
469 void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
470 void *, phys_addr_t, phys_addr_t);
473 * Restoring the memory image will overwrite the ttbr1 page tables.
474 * Create a second copy of just the linear map, and use this when
477 tmp_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
479 pr_err("Failed to allocate memory for temporary page tables.\n");
483 rc = copy_page_tables(tmp_pg_dir, PAGE_OFFSET, 0);
488 * We need a zero page that is zero before & after resume in order to
489 * to break before make on the ttbr1 page tables.
491 zero_page = (void *)get_safe_page(GFP_ATOMIC);
493 pr_err("Failed to allocate zero page.\n");
499 * Locate the exit code in the bottom-but-one page, so that *NULL
500 * still has disastrous affects.
502 hibernate_exit = (void *)PAGE_SIZE;
503 exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
505 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
506 * a new set of ttbr0 page tables and load them.
508 rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
509 (unsigned long)hibernate_exit,
510 &phys_hibernate_exit,
511 (void *)get_safe_page, GFP_ATOMIC);
513 pr_err("Failed to create safe executable page for hibernate_exit code.\n");
518 * The hibernate exit text contains a set of el2 vectors, that will
519 * be executed at el2 with the mmu off in order to reload hyp-stub.
521 __flush_dcache_area(hibernate_exit, exit_size);
524 * KASLR will cause the el2 vectors to be in a different location in
525 * the resumed kernel. Load hibernate's temporary copy into el2.
527 * We can skip this step if we booted at EL1, or are running with VHE.
529 if (el2_reset_needed()) {
530 phys_addr_t el2_vectors = phys_hibernate_exit; /* base */
531 el2_vectors += hibernate_el2_vectors -
532 __hibernate_exit_text_start; /* offset */
534 __hyp_set_vectors(el2_vectors);
537 hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
538 resume_hdr.reenter_kernel, restore_pblist,
539 resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
545 int hibernate_resume_nonboot_cpu_disable(void)
548 pr_err("Failing to resume from hibernate on an unknown CPU.\n");
552 return freeze_secondary_cpus(sleep_cpu);