1// SPDX-License-Identifier: GPL-2.0-only
  2/*:
  3 * Hibernate support specific for ARM64
  4 *
  5 * Derived from work on ARM hibernation support by:
  6 *
  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 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
 11 */
 12#define pr_fmt(x) "hibernate: " x
 13#include <linux/cpu.h>
 14#include <linux/kvm_host.h>
 15#include <linux/pm.h>
 16#include <linux/sched.h>
 17#include <linux/suspend.h>
 18#include <linux/utsname.h>
 19
 20#include <asm/barrier.h>
 21#include <asm/cacheflush.h>
 22#include <asm/cputype.h>
 23#include <asm/daifflags.h>
 24#include <asm/irqflags.h>
 25#include <asm/kexec.h>
 26#include <asm/memory.h>
 27#include <asm/mmu_context.h>
 28#include <asm/mte.h>
 29#include <asm/sections.h>
 30#include <asm/smp.h>
 31#include <asm/smp_plat.h>
 32#include <asm/suspend.h>
 33#include <asm/sysreg.h>
 34#include <asm/trans_pgd.h>
 35#include <asm/virt.h>
 36
 37/*
 38 * Hibernate core relies on this value being 0 on resume, and marks it
 39 * __nosavedata assuming it will keep the resume kernel's '0' value. This
 40 * doesn't happen with either KASLR.
 41 *
 42 * defined as "__visible int in_suspend __nosavedata" in
 43 * kernel/power/hibernate.c
 44 */
 45extern int in_suspend;
 46
 47/* Do we need to reset el2? */
 48#define el2_reset_needed() (is_hyp_nvhe())
 49
 50/* hyp-stub vectors, used to restore el2 during resume from hibernate. */
 51extern char __hyp_stub_vectors[];
 52
 53/*
 54 * The logical cpu number we should resume on, initialised to a non-cpu
 55 * number.
 56 */
 57static int sleep_cpu = -EINVAL;
 58
 59/*
 60 * Values that may not change over hibernate/resume. We put the build number
 61 * and date in here so that we guarantee not to resume with a different
 62 * kernel.
 63 */
 64struct arch_hibernate_hdr_invariants {
 65	char		uts_version[__NEW_UTS_LEN + 1];
 66};
 67
 68/* These values need to be know across a hibernate/restore. */
 69static struct arch_hibernate_hdr {
 70	struct arch_hibernate_hdr_invariants invariants;
 71
 72	/* These are needed to find the relocated kernel if built with kaslr */
 73	phys_addr_t	ttbr1_el1;
 74	void		(*reenter_kernel)(void);
 75
 76	/*
 77	 * We need to know where the __hyp_stub_vectors are after restore to
 78	 * re-configure el2.
 79	 */
 80	phys_addr_t	__hyp_stub_vectors;
 81
 82	u64		sleep_cpu_mpidr;
 83} resume_hdr;
 84
 85static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
 86{
 87	memset(i, 0, sizeof(*i));
 88	memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
 89}
 90
 91int pfn_is_nosave(unsigned long pfn)
 92{
 93	unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
 94	unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
 95
 96	return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn)) ||
 97		crash_is_nosave(pfn);
 98}
 99
100void notrace save_processor_state(void)
101{
102}
103
104void notrace restore_processor_state(void)
105{
106}
107
108int arch_hibernation_header_save(void *addr, unsigned int max_size)
109{
110	struct arch_hibernate_hdr *hdr = addr;
111
112	if (max_size < sizeof(*hdr))
113		return -EOVERFLOW;
114
115	arch_hdr_invariants(&hdr->invariants);
116	hdr->ttbr1_el1		= __pa_symbol(swapper_pg_dir);
117	hdr->reenter_kernel	= _cpu_resume;
118
119	/* We can't use __hyp_get_vectors() because kvm may still be loaded */
120	if (el2_reset_needed())
121		hdr->__hyp_stub_vectors = __pa_symbol(__hyp_stub_vectors);
122	else
123		hdr->__hyp_stub_vectors = 0;
124
125	/* Save the mpidr of the cpu we called cpu_suspend() on... */
126	if (sleep_cpu < 0) {
127		pr_err("Failing to hibernate on an unknown CPU.\n");
128		return -ENODEV;
129	}
130	hdr->sleep_cpu_mpidr = cpu_logical_map(sleep_cpu);
131	pr_info("Hibernating on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
132		hdr->sleep_cpu_mpidr);
133
134	return 0;
135}
136EXPORT_SYMBOL(arch_hibernation_header_save);
137
138int arch_hibernation_header_restore(void *addr)
139{
140	int ret;
141	struct arch_hibernate_hdr_invariants invariants;
142	struct arch_hibernate_hdr *hdr = addr;
143
144	arch_hdr_invariants(&invariants);
145	if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
146		pr_crit("Hibernate image not generated by this kernel!\n");
147		return -EINVAL;
148	}
149
150	sleep_cpu = get_logical_index(hdr->sleep_cpu_mpidr);
151	pr_info("Hibernated on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
152		hdr->sleep_cpu_mpidr);
153	if (sleep_cpu < 0) {
154		pr_crit("Hibernated on a CPU not known to this kernel!\n");
155		sleep_cpu = -EINVAL;
156		return -EINVAL;
157	}
158
159	ret = bringup_hibernate_cpu(sleep_cpu);
160	if (ret) {
161		sleep_cpu = -EINVAL;
162		return ret;
163	}
164
165	resume_hdr = *hdr;
166
167	return 0;
168}
169EXPORT_SYMBOL(arch_hibernation_header_restore);
170
171static void *hibernate_page_alloc(void *arg)
172{
173	return (void *)get_safe_page((__force gfp_t)(unsigned long)arg);
174}
175
176/*
177 * Copies length bytes, starting at src_start into an new page,
178 * perform cache maintenance, then maps it at the specified address low
179 * address as executable.
180 *
181 * This is used by hibernate to copy the code it needs to execute when
182 * overwriting the kernel text. This function generates a new set of page
183 * tables, which it loads into ttbr0.
184 *
185 * Length is provided as we probably only want 4K of data, even on a 64K
186 * page system.
187 */
188static int create_safe_exec_page(void *src_start, size_t length,
189				 phys_addr_t *phys_dst_addr)
190{
191	struct trans_pgd_info trans_info = {
192		.trans_alloc_page	= hibernate_page_alloc,
193		.trans_alloc_arg	= (__force void *)GFP_ATOMIC,
194	};
195
196	void *page = (void *)get_safe_page(GFP_ATOMIC);
197	phys_addr_t trans_ttbr0;
198	unsigned long t0sz;
199	int rc;
200
201	if (!page)
202		return -ENOMEM;
203
204	memcpy(page, src_start, length);
205	caches_clean_inval_pou((unsigned long)page, (unsigned long)page + length);
206	rc = trans_pgd_idmap_page(&trans_info, &trans_ttbr0, &t0sz, page);
207	if (rc)
208		return rc;
209
210	cpu_install_ttbr0(trans_ttbr0, t0sz);
211	*phys_dst_addr = virt_to_phys(page);
212
213	return 0;
214}
215
216#ifdef CONFIG_ARM64_MTE
217
218static DEFINE_XARRAY(mte_pages);
219
220static int save_tags(struct page *page, unsigned long pfn)
221{
222	void *tag_storage, *ret;
223
224	tag_storage = mte_allocate_tag_storage();
225	if (!tag_storage)
226		return -ENOMEM;
227
228	mte_save_page_tags(page_address(page), tag_storage);
229
230	ret = xa_store(&mte_pages, pfn, tag_storage, GFP_KERNEL);
231	if (WARN(xa_is_err(ret), "Failed to store MTE tags")) {
232		mte_free_tag_storage(tag_storage);
233		return xa_err(ret);
234	} else if (WARN(ret, "swsusp: %s: Duplicate entry", __func__)) {
235		mte_free_tag_storage(ret);
236	}
237
238	return 0;
239}
240
241static void swsusp_mte_free_storage(void)
242{
243	XA_STATE(xa_state, &mte_pages, 0);
244	void *tags;
245
246	xa_lock(&mte_pages);
247	xas_for_each(&xa_state, tags, ULONG_MAX) {
248		mte_free_tag_storage(tags);
249	}
250	xa_unlock(&mte_pages);
251
252	xa_destroy(&mte_pages);
253}
254
255static int swsusp_mte_save_tags(void)
256{
257	struct zone *zone;
258	unsigned long pfn, max_zone_pfn;
259	int ret = 0;
260	int n = 0;
261
262	if (!system_supports_mte())
263		return 0;
264
265	for_each_populated_zone(zone) {
266		max_zone_pfn = zone_end_pfn(zone);
267		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
268			struct page *page = pfn_to_online_page(pfn);
269			struct folio *folio;
270
271			if (!page)
272				continue;
273			folio = page_folio(page);
274
275			if (folio_test_hugetlb(folio) &&
276			    !folio_test_hugetlb_mte_tagged(folio))
277				continue;
278
279			if (!page_mte_tagged(page))
280				continue;
281
282			ret = save_tags(page, pfn);
283			if (ret) {
284				swsusp_mte_free_storage();
285				goto out;
286			}
287
288			n++;
289		}
290	}
291	pr_info("Saved %d MTE pages\n", n);
292
293out:
294	return ret;
295}
296
297static void swsusp_mte_restore_tags(void)
298{
299	XA_STATE(xa_state, &mte_pages, 0);
300	int n = 0;
301	void *tags;
302
303	xa_lock(&mte_pages);
304	xas_for_each(&xa_state, tags, ULONG_MAX) {
305		unsigned long pfn = xa_state.xa_index;
306		struct page *page = pfn_to_online_page(pfn);
307
308		mte_restore_page_tags(page_address(page), tags);
309
310		mte_free_tag_storage(tags);
311		n++;
312	}
313	xa_unlock(&mte_pages);
314
315	pr_info("Restored %d MTE pages\n", n);
316
317	xa_destroy(&mte_pages);
318}
319
320#else	/* CONFIG_ARM64_MTE */
321
322static int swsusp_mte_save_tags(void)
323{
324	return 0;
325}
326
327static void swsusp_mte_restore_tags(void)
328{
329}
330
331#endif	/* CONFIG_ARM64_MTE */
332
333int swsusp_arch_suspend(void)
334{
335	int ret = 0;
336	unsigned long flags;
337	struct sleep_stack_data state;
338
339	if (cpus_are_stuck_in_kernel()) {
340		pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
341		return -EBUSY;
342	}
343
344	flags = local_daif_save();
345
346	if (__cpu_suspend_enter(&state)) {
347		/* make the crash dump kernel image visible/saveable */
348		crash_prepare_suspend();
349
350		ret = swsusp_mte_save_tags();
351		if (ret)
352			return ret;
353
354		sleep_cpu = smp_processor_id();
355		ret = swsusp_save();
356	} else {
357		/* Clean kernel core startup/idle code to PoC*/
358		dcache_clean_inval_poc((unsigned long)__mmuoff_data_start,
359				    (unsigned long)__mmuoff_data_end);
360		dcache_clean_inval_poc((unsigned long)__idmap_text_start,
361				    (unsigned long)__idmap_text_end);
362
363		/* Clean kvm setup code to PoC? */
364		if (el2_reset_needed()) {
365			dcache_clean_inval_poc(
366				(unsigned long)__hyp_idmap_text_start,
367				(unsigned long)__hyp_idmap_text_end);
368			dcache_clean_inval_poc((unsigned long)__hyp_text_start,
369					    (unsigned long)__hyp_text_end);
370		}
371
372		swsusp_mte_restore_tags();
373
374		/* make the crash dump kernel image protected again */
375		crash_post_resume();
376
377		/*
378		 * Tell the hibernation core that we've just restored
379		 * the memory
380		 */
381		in_suspend = 0;
382
383		sleep_cpu = -EINVAL;
384		__cpu_suspend_exit();
385
386		/*
387		 * Just in case the boot kernel did turn the SSBD
388		 * mitigation off behind our back, let's set the state
389		 * to what we expect it to be.
390		 */
391		spectre_v4_enable_mitigation(NULL);
392	}
393
394	local_daif_restore(flags);
395
396	return ret;
397}
398
399/*
400 * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
401 *
402 * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
403 * we don't need to free it here.
404 */
405int swsusp_arch_resume(void)
406{
407	int rc;
408	void *zero_page;
409	size_t exit_size;
410	pgd_t *tmp_pg_dir;
411	phys_addr_t el2_vectors;
412	void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
413					  void *, phys_addr_t, phys_addr_t);
414	struct trans_pgd_info trans_info = {
415		.trans_alloc_page	= hibernate_page_alloc,
416		.trans_alloc_arg	= (__force void *)GFP_ATOMIC,
417	};
418
419	/*
420	 * Restoring the memory image will overwrite the ttbr1 page tables.
421	 * Create a second copy of just the linear map, and use this when
422	 * restoring.
423	 */
424	rc = trans_pgd_create_copy(&trans_info, &tmp_pg_dir, PAGE_OFFSET,
425				   PAGE_END);
426	if (rc)
427		return rc;
428
429	/*
430	 * We need a zero page that is zero before & after resume in order
431	 * to break before make on the ttbr1 page tables.
432	 */
433	zero_page = (void *)get_safe_page(GFP_ATOMIC);
434	if (!zero_page) {
435		pr_err("Failed to allocate zero page.\n");
436		return -ENOMEM;
437	}
438
439	if (el2_reset_needed()) {
440		rc = trans_pgd_copy_el2_vectors(&trans_info, &el2_vectors);
441		if (rc) {
442			pr_err("Failed to setup el2 vectors\n");
443			return rc;
444		}
445	}
446
447	exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
448	/*
449	 * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
450	 * a new set of ttbr0 page tables and load them.
451	 */
452	rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
453				   (phys_addr_t *)&hibernate_exit);
454	if (rc) {
455		pr_err("Failed to create safe executable page for hibernate_exit code.\n");
456		return rc;
457	}
458
459	/*
460	 * KASLR will cause the el2 vectors to be in a different location in
461	 * the resumed kernel. Load hibernate's temporary copy into el2.
462	 *
463	 * We can skip this step if we booted at EL1, or are running with VHE.
464	 */
465	if (el2_reset_needed())
466		__hyp_set_vectors(el2_vectors);
467
468	hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
469		       resume_hdr.reenter_kernel, restore_pblist,
470		       resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
471
472	return 0;
473}
474
475int hibernate_resume_nonboot_cpu_disable(void)
476{
477	if (sleep_cpu < 0) {
478		pr_err("Failing to resume from hibernate on an unknown CPU.\n");
479		return -ENODEV;
480	}
481
482	return freeze_secondary_cpus(sleep_cpu);
483}