~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/arm64/kernel/hibernate.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/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>
 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 '' 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  */
 51 extern int in_suspend;
 52 
 53 /* Find a symbols alias in the linear map */
 54 #define LMADDR(x)       phys_to_virt(virt_to_phys(x))
 55 
 56 /* Do we need to reset el2? */
 57 #define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
 58 
 59 /* temporary el2 vectors in the __hibernate_exit_text section. */
 60 extern char hibernate_el2_vectors[];
 61 
 62 /* hyp-stub vectors, used to restore el2 during resume from hibernate. */
 63 extern char __hyp_stub_vectors[];
 64 
 65 /*
 66  * The logical cpu number we should resume on, initialised to a non-cpu
 67  * number.
 68  */
 69 static int sleep_cpu = -EINVAL;
 70 
 71 /*
 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
 74  * kernel.
 75  */
 76 struct arch_hibernate_hdr_invariants {
 77         char            uts_version[__NEW_UTS_LEN + 1];
 78 };
 79 
 80 /* These values need to be know across a hibernate/restore. */
 81 static struct arch_hibernate_hdr {
 82         struct arch_hibernate_hdr_invariants invariants;
 83 
 84         /* These are needed to find the relocated kernel if built with kaslr */
 85         phys_addr_t     ttbr1_el1;
 86         void            (*reenter_kernel)(void);
 87 
 88         /*
 89          * We need to know where the __hyp_stub_vectors are after restore to
 90          * re-configure el2.
 91          */
 92         phys_addr_t     __hyp_stub_vectors;
 93 
 94         u64             sleep_cpu_mpidr;
 95 } resume_hdr;
 96 
 97 static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
 98 {
 99         memset(i, 0, sizeof(*i));
100         memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
101 }
102 
103 int pfn_is_nosave(unsigned long pfn)
104 {
105         unsigned long nosave_begin_pfn = virt_to_pfn(&__nosave_begin);
106         unsigned long nosave_end_pfn = virt_to_pfn(&__nosave_end - 1);
107 
108         return (pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn);
109 }
110 
111 void notrace save_processor_state(void)
112 {
113         WARN_ON(num_online_cpus() != 1);
114 }
115 
116 void notrace restore_processor_state(void)
117 {
118 }
119 
120 int 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          = virt_to_phys(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 = virt_to_phys(__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 }
148 EXPORT_SYMBOL(arch_hibernation_header_save);
149 
150 int 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 }
184 EXPORT_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  */
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),
202                                  gfp_t mask)
203 {
204         int rc = 0;
205         pgd_t *pgd;
206         pud_t *pud;
207         pmd_t *pmd;
208         pte_t *pte;
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         pgd = pgd_offset_raw(allocator(mask), dst_addr);
220         if (pgd_none(*pgd)) {
221                 pud = allocator(mask);
222                 if (!pud) {
223                         rc = -ENOMEM;
224                         goto out;
225                 }
226                 pgd_populate(&init_mm, pgd, pud);
227         }
228 
229         pud = pud_offset(pgd, dst_addr);
230         if (pud_none(*pud)) {
231                 pmd = allocator(mask);
232                 if (!pmd) {
233                         rc = -ENOMEM;
234                         goto out;
235                 }
236                 pud_populate(&init_mm, pud, pmd);
237         }
238 
239         pmd = pmd_offset(pud, dst_addr);
240         if (pmd_none(*pmd)) {
241                 pte = allocator(mask);
242                 if (!pte) {
243                         rc = -ENOMEM;
244                         goto out;
245                 }
246                 pmd_populate_kernel(&init_mm, pmd, pte);
247         }
248 
249         pte = pte_offset_kernel(pmd, dst_addr);
250         set_pte(pte, __pte(virt_to_phys((void *)dst) |
251                          pgprot_val(PAGE_KERNEL_EXEC)));
252 
253         /*
254          * Load our new page tables. A strict BBM approach requires that we
255          * ensure that TLBs are free of any entries that may overlap with the
256          * global mappings we are about to install.
257          *
258          * For a real hibernate/resume cycle TTBR0 currently points to a zero
259          * page, but TLBs may contain stale ASID-tagged entries (e.g. for EFI
260          * runtime services), while for a userspace-driven test_resume cycle it
261          * points to userspace page tables (and we must point it at a zero page
262          * ourselves). Elsewhere we only (un)install the idmap with preemption
263          * disabled, so T0SZ should be as required regardless.
264          */
265         cpu_set_reserved_ttbr0();
266         local_flush_tlb_all();
267         write_sysreg(virt_to_phys(pgd), ttbr0_el1);
268         isb();
269 
270         *phys_dst_addr = virt_to_phys((void *)dst);
271 
272 out:
273         return rc;
274 }
275 
276 #define dcache_clean_range(start, end)  __flush_dcache_area(start, (end - start))
277 
278 int swsusp_arch_suspend(void)
279 {
280         int ret = 0;
281         unsigned long flags;
282         struct sleep_stack_data state;
283 
284         if (cpus_are_stuck_in_kernel()) {
285                 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
286                 return -EBUSY;
287         }
288 
289         local_dbg_save(flags);
290 
291         if (__cpu_suspend_enter(&state)) {
292                 sleep_cpu = smp_processor_id();
293                 ret = swsusp_save();
294         } else {
295                 /* Clean kernel core startup/idle code to PoC*/
296                 dcache_clean_range(__mmuoff_data_start, __mmuoff_data_end);
297                 dcache_clean_range(__idmap_text_start, __idmap_text_end);
298 
299                 /* Clean kvm setup code to PoC? */
300                 if (el2_reset_needed())
301                         dcache_clean_range(__hyp_idmap_text_start, __hyp_idmap_text_end);
302 
303                 /*
304                  * Tell the hibernation core that we've just restored
305                  * the memory
306                  */
307                 in_suspend = 0;
308 
309                 sleep_cpu = -EINVAL;
310                 __cpu_suspend_exit();
311         }
312 
313         local_dbg_restore(flags);
314 
315         return ret;
316 }
317 
318 static void _copy_pte(pte_t *dst_pte, pte_t *src_pte, unsigned long addr)
319 {
320         pte_t pte = *src_pte;
321 
322         if (pte_valid(pte)) {
323                 /*
324                  * Resume will overwrite areas that may be marked
325                  * read only (code, rodata). Clear the RDONLY bit from
326                  * the temporary mappings we use during restore.
327                  */
328                 set_pte(dst_pte, pte_clear_rdonly(pte));
329         } else if (debug_pagealloc_enabled() && !pte_none(pte)) {
330                 /*
331                  * debug_pagealloc will removed the PTE_VALID bit if
332                  * the page isn't in use by the resume kernel. It may have
333                  * been in use by the original kernel, in which case we need
334                  * to put it back in our copy to do the restore.
335                  *
336                  * Before marking this entry valid, check the pfn should
337                  * be mapped.
338                  */
339                 BUG_ON(!pfn_valid(pte_pfn(pte)));
340 
341                 set_pte(dst_pte, pte_mkpresent(pte_clear_rdonly(pte)));
342         }
343 }
344 
345 static int copy_pte(pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long start,
346                     unsigned long end)
347 {
348         pte_t *src_pte;
349         pte_t *dst_pte;
350         unsigned long addr = start;
351 
352         dst_pte = (pte_t *)get_safe_page(GFP_ATOMIC);
353         if (!dst_pte)
354                 return -ENOMEM;
355         pmd_populate_kernel(&init_mm, dst_pmd, dst_pte);
356         dst_pte = pte_offset_kernel(dst_pmd, start);
357 
358         src_pte = pte_offset_kernel(src_pmd, start);
359         do {
360                 _copy_pte(dst_pte, src_pte, addr);
361         } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
362 
363         return 0;
364 }
365 
366 static int copy_pmd(pud_t *dst_pud, pud_t *src_pud, unsigned long start,
367                     unsigned long end)
368 {
369         pmd_t *src_pmd;
370         pmd_t *dst_pmd;
371         unsigned long next;
372         unsigned long addr = start;
373 
374         if (pud_none(*dst_pud)) {
375                 dst_pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
376                 if (!dst_pmd)
377                         return -ENOMEM;
378                 pud_populate(&init_mm, dst_pud, dst_pmd);
379         }
380         dst_pmd = pmd_offset(dst_pud, start);
381 
382         src_pmd = pmd_offset(src_pud, start);
383         do {
384                 next = pmd_addr_end(addr, end);
385                 if (pmd_none(*src_pmd))
386                         continue;
387                 if (pmd_table(*src_pmd)) {
388                         if (copy_pte(dst_pmd, src_pmd, addr, next))
389                                 return -ENOMEM;
390                 } else {
391                         set_pmd(dst_pmd,
392                                 __pmd(pmd_val(*src_pmd) & ~PMD_SECT_RDONLY));
393                 }
394         } while (dst_pmd++, src_pmd++, addr = next, addr != end);
395 
396         return 0;
397 }
398 
399 static int copy_pud(pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long start,
400                     unsigned long end)
401 {
402         pud_t *dst_pud;
403         pud_t *src_pud;
404         unsigned long next;
405         unsigned long addr = start;
406 
407         if (pgd_none(*dst_pgd)) {
408                 dst_pud = (pud_t *)get_safe_page(GFP_ATOMIC);
409                 if (!dst_pud)
410                         return -ENOMEM;
411                 pgd_populate(&init_mm, dst_pgd, dst_pud);
412         }
413         dst_pud = pud_offset(dst_pgd, start);
414 
415         src_pud = pud_offset(src_pgd, start);
416         do {
417                 next = pud_addr_end(addr, end);
418                 if (pud_none(*src_pud))
419                         continue;
420                 if (pud_table(*(src_pud))) {
421                         if (copy_pmd(dst_pud, src_pud, addr, next))
422                                 return -ENOMEM;
423                 } else {
424                         set_pud(dst_pud,
425                                 __pud(pud_val(*src_pud) & ~PMD_SECT_RDONLY));
426                 }
427         } while (dst_pud++, src_pud++, addr = next, addr != end);
428 
429         return 0;
430 }
431 
432 static int copy_page_tables(pgd_t *dst_pgd, unsigned long start,
433                             unsigned long end)
434 {
435         unsigned long next;
436         unsigned long addr = start;
437         pgd_t *src_pgd = pgd_offset_k(start);
438 
439         dst_pgd = pgd_offset_raw(dst_pgd, start);
440         do {
441                 next = pgd_addr_end(addr, end);
442                 if (pgd_none(*src_pgd))
443                         continue;
444                 if (copy_pud(dst_pgd, src_pgd, addr, next))
445                         return -ENOMEM;
446         } while (dst_pgd++, src_pgd++, addr = next, addr != end);
447 
448         return 0;
449 }
450 
451 /*
452  * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
453  *
454  * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
455  * we don't need to free it here.
456  */
457 int swsusp_arch_resume(void)
458 {
459         int rc = 0;
460         void *zero_page;
461         size_t exit_size;
462         pgd_t *tmp_pg_dir;
463         void *lm_restore_pblist;
464         phys_addr_t phys_hibernate_exit;
465         void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
466                                           void *, phys_addr_t, phys_addr_t);
467 
468         /*
469          * Restoring the memory image will overwrite the ttbr1 page tables.
470          * Create a second copy of just the linear map, and use this when
471          * restoring.
472          */
473         tmp_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
474         if (!tmp_pg_dir) {
475                 pr_err("Failed to allocate memory for temporary page tables.");
476                 rc = -ENOMEM;
477                 goto out;
478         }
479         rc = copy_page_tables(tmp_pg_dir, PAGE_OFFSET, 0);
480         if (rc)
481                 goto out;
482 
483         /*
484          * Since we only copied the linear map, we need to find restore_pblist's
485          * linear map address.
486          */
487         lm_restore_pblist = LMADDR(restore_pblist);
488 
489         /*
490          * We need a zero page that is zero before & after resume in order to
491          * to break before make on the ttbr1 page tables.
492          */
493         zero_page = (void *)get_safe_page(GFP_ATOMIC);
494         if (!zero_page) {
495                 pr_err("Failed to allocate zero page.");
496                 rc = -ENOMEM;
497                 goto out;
498         }
499 
500         /*
501          * Locate the exit code in the bottom-but-one page, so that *NULL
502          * still has disastrous affects.
503          */
504         hibernate_exit = (void *)PAGE_SIZE;
505         exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
506         /*
507          * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
508          * a new set of ttbr0 page tables and load them.
509          */
510         rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
511                                    (unsigned long)hibernate_exit,
512                                    &phys_hibernate_exit,
513                                    (void *)get_safe_page, GFP_ATOMIC);
514         if (rc) {
515                 pr_err("Failed to create safe executable page for hibernate_exit code.");
516                 goto out;
517         }
518 
519         /*
520          * The hibernate exit text contains a set of el2 vectors, that will
521          * be executed at el2 with the mmu off in order to reload hyp-stub.
522          */
523         __flush_dcache_area(hibernate_exit, exit_size);
524 
525         /*
526          * KASLR will cause the el2 vectors to be in a different location in
527          * the resumed kernel. Load hibernate's temporary copy into el2.
528          *
529          * We can skip this step if we booted at EL1, or are running with VHE.
530          */
531         if (el2_reset_needed()) {
532                 phys_addr_t el2_vectors = phys_hibernate_exit;  /* base */
533                 el2_vectors += hibernate_el2_vectors -
534                                __hibernate_exit_text_start;     /* offset */
535 
536                 __hyp_set_vectors(el2_vectors);
537         }
538 
539         hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
540                        resume_hdr.reenter_kernel, lm_restore_pblist,
541                        resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
542 
543 out:
544         return rc;
545 }
546 
547 int hibernate_resume_nonboot_cpu_disable(void)
548 {
549         if (sleep_cpu < 0) {
550                 pr_err("Failing to resume from hibernate on an unknown CPU.\n");
551                 return -ENODEV;
552         }
553 
554         return freeze_secondary_cpus(sleep_cpu);
555 }
556 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp