xref: /openbmc/linux/arch/arm64/kernel/hibernate.c (revision 0edbfea5)
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/kvm_host.h>
19 #include <linux/mm.h>
20 #include <linux/notifier.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/irqflags.h>
30 #include <asm/memory.h>
31 #include <asm/mmu_context.h>
32 #include <asm/pgalloc.h>
33 #include <asm/pgtable.h>
34 #include <asm/pgtable-hwdef.h>
35 #include <asm/sections.h>
36 #include <asm/suspend.h>
37 #include <asm/virt.h>
38 
39 /*
40  * Hibernate core relies on this value being 0 on resume, and marks it
41  * __nosavedata assuming it will keep the resume kernel's '0' value. This
42  * doesn't happen with either KASLR.
43  *
44  * defined as "__visible int in_suspend __nosavedata" in
45  * kernel/power/hibernate.c
46  */
47 extern int in_suspend;
48 
49 /* Find a symbols alias in the linear map */
50 #define LMADDR(x)	phys_to_virt(virt_to_phys(x))
51 
52 /* Do we need to reset el2? */
53 #define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
54 
55 /*
56  * Start/end of the hibernate exit code, this must be copied to a 'safe'
57  * location in memory, and executed from there.
58  */
59 extern char __hibernate_exit_text_start[], __hibernate_exit_text_end[];
60 
61 /* temporary el2 vectors in the __hibernate_exit_text section. */
62 extern char hibernate_el2_vectors[];
63 
64 /* hyp-stub vectors, used to restore el2 during resume from hibernate. */
65 extern char __hyp_stub_vectors[];
66 
67 /*
68  * Values that may not change over hibernate/resume. We put the build number
69  * and date in here so that we guarantee not to resume with a different
70  * kernel.
71  */
72 struct arch_hibernate_hdr_invariants {
73 	char		uts_version[__NEW_UTS_LEN + 1];
74 };
75 
76 /* These values need to be know across a hibernate/restore. */
77 static struct arch_hibernate_hdr {
78 	struct arch_hibernate_hdr_invariants invariants;
79 
80 	/* These are needed to find the relocated kernel if built with kaslr */
81 	phys_addr_t	ttbr1_el1;
82 	void		(*reenter_kernel)(void);
83 
84 	/*
85 	 * We need to know where the __hyp_stub_vectors are after restore to
86 	 * re-configure el2.
87 	 */
88 	phys_addr_t	__hyp_stub_vectors;
89 } resume_hdr;
90 
91 static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
92 {
93 	memset(i, 0, sizeof(*i));
94 	memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
95 }
96 
97 int pfn_is_nosave(unsigned long pfn)
98 {
99 	unsigned long nosave_begin_pfn = virt_to_pfn(&__nosave_begin);
100 	unsigned long nosave_end_pfn = virt_to_pfn(&__nosave_end - 1);
101 
102 	return (pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn);
103 }
104 
105 void notrace save_processor_state(void)
106 {
107 	WARN_ON(num_online_cpus() != 1);
108 }
109 
110 void notrace restore_processor_state(void)
111 {
112 }
113 
114 int arch_hibernation_header_save(void *addr, unsigned int max_size)
115 {
116 	struct arch_hibernate_hdr *hdr = addr;
117 
118 	if (max_size < sizeof(*hdr))
119 		return -EOVERFLOW;
120 
121 	arch_hdr_invariants(&hdr->invariants);
122 	hdr->ttbr1_el1		= virt_to_phys(swapper_pg_dir);
123 	hdr->reenter_kernel	= _cpu_resume;
124 
125 	/* We can't use __hyp_get_vectors() because kvm may still be loaded */
126 	if (el2_reset_needed())
127 		hdr->__hyp_stub_vectors = virt_to_phys(__hyp_stub_vectors);
128 	else
129 		hdr->__hyp_stub_vectors = 0;
130 
131 	return 0;
132 }
133 EXPORT_SYMBOL(arch_hibernation_header_save);
134 
135 int arch_hibernation_header_restore(void *addr)
136 {
137 	struct arch_hibernate_hdr_invariants invariants;
138 	struct arch_hibernate_hdr *hdr = addr;
139 
140 	arch_hdr_invariants(&invariants);
141 	if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
142 		pr_crit("Hibernate image not generated by this kernel!\n");
143 		return -EINVAL;
144 	}
145 
146 	resume_hdr = *hdr;
147 
148 	return 0;
149 }
150 EXPORT_SYMBOL(arch_hibernation_header_restore);
151 
152 /*
153  * Copies length bytes, starting at src_start into an new page,
154  * perform cache maintentance, then maps it at the specified address low
155  * address as executable.
156  *
157  * This is used by hibernate to copy the code it needs to execute when
158  * overwriting the kernel text. This function generates a new set of page
159  * tables, which it loads into ttbr0.
160  *
161  * Length is provided as we probably only want 4K of data, even on a 64K
162  * page system.
163  */
164 static int create_safe_exec_page(void *src_start, size_t length,
165 				 unsigned long dst_addr,
166 				 phys_addr_t *phys_dst_addr,
167 				 void *(*allocator)(gfp_t mask),
168 				 gfp_t mask)
169 {
170 	int rc = 0;
171 	pgd_t *pgd;
172 	pud_t *pud;
173 	pmd_t *pmd;
174 	pte_t *pte;
175 	unsigned long dst = (unsigned long)allocator(mask);
176 
177 	if (!dst) {
178 		rc = -ENOMEM;
179 		goto out;
180 	}
181 
182 	memcpy((void *)dst, src_start, length);
183 	flush_icache_range(dst, dst + length);
184 
185 	pgd = pgd_offset_raw(allocator(mask), dst_addr);
186 	if (pgd_none(*pgd)) {
187 		pud = allocator(mask);
188 		if (!pud) {
189 			rc = -ENOMEM;
190 			goto out;
191 		}
192 		pgd_populate(&init_mm, pgd, pud);
193 	}
194 
195 	pud = pud_offset(pgd, dst_addr);
196 	if (pud_none(*pud)) {
197 		pmd = allocator(mask);
198 		if (!pmd) {
199 			rc = -ENOMEM;
200 			goto out;
201 		}
202 		pud_populate(&init_mm, pud, pmd);
203 	}
204 
205 	pmd = pmd_offset(pud, dst_addr);
206 	if (pmd_none(*pmd)) {
207 		pte = allocator(mask);
208 		if (!pte) {
209 			rc = -ENOMEM;
210 			goto out;
211 		}
212 		pmd_populate_kernel(&init_mm, pmd, pte);
213 	}
214 
215 	pte = pte_offset_kernel(pmd, dst_addr);
216 	set_pte(pte, __pte(virt_to_phys((void *)dst) |
217 			 pgprot_val(PAGE_KERNEL_EXEC)));
218 
219 	/* Load our new page tables */
220 	asm volatile("msr	ttbr0_el1, %0;"
221 		     "isb;"
222 		     "tlbi	vmalle1is;"
223 		     "dsb	ish;"
224 		     "isb" : : "r"(virt_to_phys(pgd)));
225 
226 	*phys_dst_addr = virt_to_phys((void *)dst);
227 
228 out:
229 	return rc;
230 }
231 
232 
233 int swsusp_arch_suspend(void)
234 {
235 	int ret = 0;
236 	unsigned long flags;
237 	struct sleep_stack_data state;
238 
239 	local_dbg_save(flags);
240 
241 	if (__cpu_suspend_enter(&state)) {
242 		ret = swsusp_save();
243 	} else {
244 		/* Clean kernel to PoC for secondary core startup */
245 		__flush_dcache_area(LMADDR(KERNEL_START), KERNEL_END - KERNEL_START);
246 
247 		/*
248 		 * Tell the hibernation core that we've just restored
249 		 * the memory
250 		 */
251 		in_suspend = 0;
252 
253 		__cpu_suspend_exit();
254 	}
255 
256 	local_dbg_restore(flags);
257 
258 	return ret;
259 }
260 
261 static int copy_pte(pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long start,
262 		    unsigned long end)
263 {
264 	pte_t *src_pte;
265 	pte_t *dst_pte;
266 	unsigned long addr = start;
267 
268 	dst_pte = (pte_t *)get_safe_page(GFP_ATOMIC);
269 	if (!dst_pte)
270 		return -ENOMEM;
271 	pmd_populate_kernel(&init_mm, dst_pmd, dst_pte);
272 	dst_pte = pte_offset_kernel(dst_pmd, start);
273 
274 	src_pte = pte_offset_kernel(src_pmd, start);
275 	do {
276 		if (!pte_none(*src_pte))
277 			/*
278 			 * Resume will overwrite areas that may be marked
279 			 * read only (code, rodata). Clear the RDONLY bit from
280 			 * the temporary mappings we use during restore.
281 			 */
282 			set_pte(dst_pte, __pte(pte_val(*src_pte) & ~PTE_RDONLY));
283 	} while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
284 
285 	return 0;
286 }
287 
288 static int copy_pmd(pud_t *dst_pud, pud_t *src_pud, unsigned long start,
289 		    unsigned long end)
290 {
291 	pmd_t *src_pmd;
292 	pmd_t *dst_pmd;
293 	unsigned long next;
294 	unsigned long addr = start;
295 
296 	if (pud_none(*dst_pud)) {
297 		dst_pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
298 		if (!dst_pmd)
299 			return -ENOMEM;
300 		pud_populate(&init_mm, dst_pud, dst_pmd);
301 	}
302 	dst_pmd = pmd_offset(dst_pud, start);
303 
304 	src_pmd = pmd_offset(src_pud, start);
305 	do {
306 		next = pmd_addr_end(addr, end);
307 		if (pmd_none(*src_pmd))
308 			continue;
309 		if (pmd_table(*src_pmd)) {
310 			if (copy_pte(dst_pmd, src_pmd, addr, next))
311 				return -ENOMEM;
312 		} else {
313 			set_pmd(dst_pmd,
314 				__pmd(pmd_val(*src_pmd) & ~PMD_SECT_RDONLY));
315 		}
316 	} while (dst_pmd++, src_pmd++, addr = next, addr != end);
317 
318 	return 0;
319 }
320 
321 static int copy_pud(pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long start,
322 		    unsigned long end)
323 {
324 	pud_t *dst_pud;
325 	pud_t *src_pud;
326 	unsigned long next;
327 	unsigned long addr = start;
328 
329 	if (pgd_none(*dst_pgd)) {
330 		dst_pud = (pud_t *)get_safe_page(GFP_ATOMIC);
331 		if (!dst_pud)
332 			return -ENOMEM;
333 		pgd_populate(&init_mm, dst_pgd, dst_pud);
334 	}
335 	dst_pud = pud_offset(dst_pgd, start);
336 
337 	src_pud = pud_offset(src_pgd, start);
338 	do {
339 		next = pud_addr_end(addr, end);
340 		if (pud_none(*src_pud))
341 			continue;
342 		if (pud_table(*(src_pud))) {
343 			if (copy_pmd(dst_pud, src_pud, addr, next))
344 				return -ENOMEM;
345 		} else {
346 			set_pud(dst_pud,
347 				__pud(pud_val(*src_pud) & ~PMD_SECT_RDONLY));
348 		}
349 	} while (dst_pud++, src_pud++, addr = next, addr != end);
350 
351 	return 0;
352 }
353 
354 static int copy_page_tables(pgd_t *dst_pgd, unsigned long start,
355 			    unsigned long end)
356 {
357 	unsigned long next;
358 	unsigned long addr = start;
359 	pgd_t *src_pgd = pgd_offset_k(start);
360 
361 	dst_pgd = pgd_offset_raw(dst_pgd, start);
362 	do {
363 		next = pgd_addr_end(addr, end);
364 		if (pgd_none(*src_pgd))
365 			continue;
366 		if (copy_pud(dst_pgd, src_pgd, addr, next))
367 			return -ENOMEM;
368 	} while (dst_pgd++, src_pgd++, addr = next, addr != end);
369 
370 	return 0;
371 }
372 
373 /*
374  * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
375  *
376  * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
377  * we don't need to free it here.
378  */
379 int swsusp_arch_resume(void)
380 {
381 	int rc = 0;
382 	void *zero_page;
383 	size_t exit_size;
384 	pgd_t *tmp_pg_dir;
385 	void *lm_restore_pblist;
386 	phys_addr_t phys_hibernate_exit;
387 	void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
388 					  void *, phys_addr_t, phys_addr_t);
389 
390 	/*
391 	 * Locate the exit code in the bottom-but-one page, so that *NULL
392 	 * still has disastrous affects.
393 	 */
394 	hibernate_exit = (void *)PAGE_SIZE;
395 	exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
396 	/*
397 	 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
398 	 * a new set of ttbr0 page tables and load them.
399 	 */
400 	rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
401 				   (unsigned long)hibernate_exit,
402 				   &phys_hibernate_exit,
403 				   (void *)get_safe_page, GFP_ATOMIC);
404 	if (rc) {
405 		pr_err("Failed to create safe executable page for hibernate_exit code.");
406 		goto out;
407 	}
408 
409 	/*
410 	 * The hibernate exit text contains a set of el2 vectors, that will
411 	 * be executed at el2 with the mmu off in order to reload hyp-stub.
412 	 */
413 	__flush_dcache_area(hibernate_exit, exit_size);
414 
415 	/*
416 	 * Restoring the memory image will overwrite the ttbr1 page tables.
417 	 * Create a second copy of just the linear map, and use this when
418 	 * restoring.
419 	 */
420 	tmp_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
421 	if (!tmp_pg_dir) {
422 		pr_err("Failed to allocate memory for temporary page tables.");
423 		rc = -ENOMEM;
424 		goto out;
425 	}
426 	rc = copy_page_tables(tmp_pg_dir, PAGE_OFFSET, 0);
427 	if (rc)
428 		goto out;
429 
430 	/*
431 	 * Since we only copied the linear map, we need to find restore_pblist's
432 	 * linear map address.
433 	 */
434 	lm_restore_pblist = LMADDR(restore_pblist);
435 
436 	/*
437 	 * KASLR will cause the el2 vectors to be in a different location in
438 	 * the resumed kernel. Load hibernate's temporary copy into el2.
439 	 *
440 	 * We can skip this step if we booted at EL1, or are running with VHE.
441 	 */
442 	if (el2_reset_needed()) {
443 		phys_addr_t el2_vectors = phys_hibernate_exit;  /* base */
444 		el2_vectors += hibernate_el2_vectors -
445 			       __hibernate_exit_text_start;     /* offset */
446 
447 		__hyp_set_vectors(el2_vectors);
448 	}
449 
450 	/*
451 	 * We need a zero page that is zero before & after resume in order to
452 	 * to break before make on the ttbr1 page tables.
453 	 */
454 	zero_page = (void *)get_safe_page(GFP_ATOMIC);
455 
456 	hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
457 		       resume_hdr.reenter_kernel, lm_restore_pblist,
458 		       resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
459 
460 out:
461 	return rc;
462 }
463 
464 static int check_boot_cpu_online_pm_callback(struct notifier_block *nb,
465 					     unsigned long action, void *ptr)
466 {
467 	if (action == PM_HIBERNATION_PREPARE &&
468 	     cpumask_first(cpu_online_mask) != 0) {
469 		pr_warn("CPU0 is offline.\n");
470 		return notifier_from_errno(-ENODEV);
471 	}
472 
473 	return NOTIFY_OK;
474 }
475 
476 static int __init check_boot_cpu_online_init(void)
477 {
478 	/*
479 	 * Set this pm_notifier callback with a lower priority than
480 	 * cpu_hotplug_pm_callback, so that cpu_hotplug_pm_callback will be
481 	 * called earlier to disable cpu hotplug before the cpu online check.
482 	 */
483 	pm_notifier(check_boot_cpu_online_pm_callback, -INT_MAX);
484 
485 	return 0;
486 }
487 core_initcall(check_boot_cpu_online_init);
488