xref: /openbmc/linux/arch/riscv/kernel/hibernate.c (revision 2a598d0b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Hibernation support for RISCV
4  *
5  * Copyright (C) 2023 StarFive Technology Co., Ltd.
6  *
7  * Author: Jee Heng Sia <jeeheng.sia@starfivetech.com>
8  */
9 
10 #include <asm/barrier.h>
11 #include <asm/cacheflush.h>
12 #include <asm/mmu_context.h>
13 #include <asm/page.h>
14 #include <asm/pgalloc.h>
15 #include <asm/pgtable.h>
16 #include <asm/sections.h>
17 #include <asm/set_memory.h>
18 #include <asm/smp.h>
19 #include <asm/suspend.h>
20 
21 #include <linux/cpu.h>
22 #include <linux/memblock.h>
23 #include <linux/pm.h>
24 #include <linux/sched.h>
25 #include <linux/suspend.h>
26 #include <linux/utsname.h>
27 
28 /* The logical cpu number we should resume on, initialised to a non-cpu number. */
29 static int sleep_cpu = -EINVAL;
30 
31 /* Pointer to the temporary resume page table. */
32 static pgd_t *resume_pg_dir;
33 
34 /* CPU context to be saved. */
35 struct suspend_context *hibernate_cpu_context;
36 EXPORT_SYMBOL_GPL(hibernate_cpu_context);
37 
38 unsigned long relocated_restore_code;
39 EXPORT_SYMBOL_GPL(relocated_restore_code);
40 
41 /**
42  * struct arch_hibernate_hdr_invariants - container to store kernel build version.
43  * @uts_version: to save the build number and date so that we do not resume with
44  *		a different kernel.
45  */
46 struct arch_hibernate_hdr_invariants {
47 	char		uts_version[__NEW_UTS_LEN + 1];
48 };
49 
50 /**
51  * struct arch_hibernate_hdr - helper parameters that help us to restore the image.
52  * @invariants: container to store kernel build version.
53  * @hartid: to make sure same boot_cpu executes the hibernate/restore code.
54  * @saved_satp: original page table used by the hibernated image.
55  * @restore_cpu_addr: the kernel's image address to restore the CPU context.
56  */
57 static struct arch_hibernate_hdr {
58 	struct arch_hibernate_hdr_invariants invariants;
59 	unsigned long	hartid;
60 	unsigned long	saved_satp;
61 	unsigned long	restore_cpu_addr;
62 } resume_hdr;
63 
64 static void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
65 {
66 	memset(i, 0, sizeof(*i));
67 	memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
68 }
69 
70 /*
71  * Check if the given pfn is in the 'nosave' section.
72  */
73 int pfn_is_nosave(unsigned long pfn)
74 {
75 	unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
76 	unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
77 
78 	return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn));
79 }
80 
81 void notrace save_processor_state(void)
82 {
83 }
84 
85 void notrace restore_processor_state(void)
86 {
87 }
88 
89 /*
90  * Helper parameters need to be saved to the hibernation image header.
91  */
92 int arch_hibernation_header_save(void *addr, unsigned int max_size)
93 {
94 	struct arch_hibernate_hdr *hdr = addr;
95 
96 	if (max_size < sizeof(*hdr))
97 		return -EOVERFLOW;
98 
99 	arch_hdr_invariants(&hdr->invariants);
100 
101 	hdr->hartid = cpuid_to_hartid_map(sleep_cpu);
102 	hdr->saved_satp = csr_read(CSR_SATP);
103 	hdr->restore_cpu_addr = (unsigned long)__hibernate_cpu_resume;
104 
105 	return 0;
106 }
107 EXPORT_SYMBOL_GPL(arch_hibernation_header_save);
108 
109 /*
110  * Retrieve the helper parameters from the hibernation image header.
111  */
112 int arch_hibernation_header_restore(void *addr)
113 {
114 	struct arch_hibernate_hdr_invariants invariants;
115 	struct arch_hibernate_hdr *hdr = addr;
116 	int ret = 0;
117 
118 	arch_hdr_invariants(&invariants);
119 
120 	if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
121 		pr_crit("Hibernate image not generated by this kernel!\n");
122 		return -EINVAL;
123 	}
124 
125 	sleep_cpu = riscv_hartid_to_cpuid(hdr->hartid);
126 	if (sleep_cpu < 0) {
127 		pr_crit("Hibernated on a CPU not known to this kernel!\n");
128 		sleep_cpu = -EINVAL;
129 		return -EINVAL;
130 	}
131 
132 #ifdef CONFIG_SMP
133 	ret = bringup_hibernate_cpu(sleep_cpu);
134 	if (ret) {
135 		sleep_cpu = -EINVAL;
136 		return ret;
137 	}
138 #endif
139 	resume_hdr = *hdr;
140 
141 	return ret;
142 }
143 EXPORT_SYMBOL_GPL(arch_hibernation_header_restore);
144 
145 int swsusp_arch_suspend(void)
146 {
147 	int ret = 0;
148 
149 	if (__cpu_suspend_enter(hibernate_cpu_context)) {
150 		sleep_cpu = smp_processor_id();
151 		suspend_save_csrs(hibernate_cpu_context);
152 		ret = swsusp_save();
153 	} else {
154 		suspend_restore_csrs(hibernate_cpu_context);
155 		flush_tlb_all();
156 		flush_icache_all();
157 
158 		/*
159 		 * Tell the hibernation core that we've just restored the memory.
160 		 */
161 		in_suspend = 0;
162 		sleep_cpu = -EINVAL;
163 	}
164 
165 	return ret;
166 }
167 
168 static int temp_pgtable_map_pte(pmd_t *dst_pmdp, pmd_t *src_pmdp, unsigned long start,
169 				unsigned long end, pgprot_t prot)
170 {
171 	pte_t *src_ptep;
172 	pte_t *dst_ptep;
173 
174 	if (pmd_none(READ_ONCE(*dst_pmdp))) {
175 		dst_ptep = (pte_t *)get_safe_page(GFP_ATOMIC);
176 		if (!dst_ptep)
177 			return -ENOMEM;
178 
179 		pmd_populate_kernel(NULL, dst_pmdp, dst_ptep);
180 	}
181 
182 	dst_ptep = pte_offset_kernel(dst_pmdp, start);
183 	src_ptep = pte_offset_kernel(src_pmdp, start);
184 
185 	do {
186 		pte_t pte = READ_ONCE(*src_ptep);
187 
188 		if (pte_present(pte))
189 			set_pte(dst_ptep, __pte(pte_val(pte) | pgprot_val(prot)));
190 	} while (dst_ptep++, src_ptep++, start += PAGE_SIZE, start < end);
191 
192 	return 0;
193 }
194 
195 static int temp_pgtable_map_pmd(pud_t *dst_pudp, pud_t *src_pudp, unsigned long start,
196 				unsigned long end, pgprot_t prot)
197 {
198 	unsigned long next;
199 	unsigned long ret;
200 	pmd_t *src_pmdp;
201 	pmd_t *dst_pmdp;
202 
203 	if (pud_none(READ_ONCE(*dst_pudp))) {
204 		dst_pmdp = (pmd_t *)get_safe_page(GFP_ATOMIC);
205 		if (!dst_pmdp)
206 			return -ENOMEM;
207 
208 		pud_populate(NULL, dst_pudp, dst_pmdp);
209 	}
210 
211 	dst_pmdp = pmd_offset(dst_pudp, start);
212 	src_pmdp = pmd_offset(src_pudp, start);
213 
214 	do {
215 		pmd_t pmd = READ_ONCE(*src_pmdp);
216 
217 		next = pmd_addr_end(start, end);
218 
219 		if (pmd_none(pmd))
220 			continue;
221 
222 		if (pmd_leaf(pmd)) {
223 			set_pmd(dst_pmdp, __pmd(pmd_val(pmd) | pgprot_val(prot)));
224 		} else {
225 			ret = temp_pgtable_map_pte(dst_pmdp, src_pmdp, start, next, prot);
226 			if (ret)
227 				return -ENOMEM;
228 		}
229 	} while (dst_pmdp++, src_pmdp++, start = next, start != end);
230 
231 	return 0;
232 }
233 
234 static int temp_pgtable_map_pud(p4d_t *dst_p4dp, p4d_t *src_p4dp, unsigned long start,
235 				unsigned long end, pgprot_t prot)
236 {
237 	unsigned long next;
238 	unsigned long ret;
239 	pud_t *dst_pudp;
240 	pud_t *src_pudp;
241 
242 	if (p4d_none(READ_ONCE(*dst_p4dp))) {
243 		dst_pudp = (pud_t *)get_safe_page(GFP_ATOMIC);
244 		if (!dst_pudp)
245 			return -ENOMEM;
246 
247 		p4d_populate(NULL, dst_p4dp, dst_pudp);
248 	}
249 
250 	dst_pudp = pud_offset(dst_p4dp, start);
251 	src_pudp = pud_offset(src_p4dp, start);
252 
253 	do {
254 		pud_t pud = READ_ONCE(*src_pudp);
255 
256 		next = pud_addr_end(start, end);
257 
258 		if (pud_none(pud))
259 			continue;
260 
261 		if (pud_leaf(pud)) {
262 			set_pud(dst_pudp, __pud(pud_val(pud) | pgprot_val(prot)));
263 		} else {
264 			ret = temp_pgtable_map_pmd(dst_pudp, src_pudp, start, next, prot);
265 			if (ret)
266 				return -ENOMEM;
267 		}
268 	} while (dst_pudp++, src_pudp++, start = next, start != end);
269 
270 	return 0;
271 }
272 
273 static int temp_pgtable_map_p4d(pgd_t *dst_pgdp, pgd_t *src_pgdp, unsigned long start,
274 				unsigned long end, pgprot_t prot)
275 {
276 	unsigned long next;
277 	unsigned long ret;
278 	p4d_t *dst_p4dp;
279 	p4d_t *src_p4dp;
280 
281 	if (pgd_none(READ_ONCE(*dst_pgdp))) {
282 		dst_p4dp = (p4d_t *)get_safe_page(GFP_ATOMIC);
283 		if (!dst_p4dp)
284 			return -ENOMEM;
285 
286 		pgd_populate(NULL, dst_pgdp, dst_p4dp);
287 	}
288 
289 	dst_p4dp = p4d_offset(dst_pgdp, start);
290 	src_p4dp = p4d_offset(src_pgdp, start);
291 
292 	do {
293 		p4d_t p4d = READ_ONCE(*src_p4dp);
294 
295 		next = p4d_addr_end(start, end);
296 
297 		if (p4d_none(p4d))
298 			continue;
299 
300 		if (p4d_leaf(p4d)) {
301 			set_p4d(dst_p4dp, __p4d(p4d_val(p4d) | pgprot_val(prot)));
302 		} else {
303 			ret = temp_pgtable_map_pud(dst_p4dp, src_p4dp, start, next, prot);
304 			if (ret)
305 				return -ENOMEM;
306 		}
307 	} while (dst_p4dp++, src_p4dp++, start = next, start != end);
308 
309 	return 0;
310 }
311 
312 static int temp_pgtable_mapping(pgd_t *pgdp, unsigned long start, unsigned long end, pgprot_t prot)
313 {
314 	pgd_t *dst_pgdp = pgd_offset_pgd(pgdp, start);
315 	pgd_t *src_pgdp = pgd_offset_k(start);
316 	unsigned long next;
317 	unsigned long ret;
318 
319 	do {
320 		pgd_t pgd = READ_ONCE(*src_pgdp);
321 
322 		next = pgd_addr_end(start, end);
323 
324 		if (pgd_none(pgd))
325 			continue;
326 
327 		if (pgd_leaf(pgd)) {
328 			set_pgd(dst_pgdp, __pgd(pgd_val(pgd) | pgprot_val(prot)));
329 		} else {
330 			ret = temp_pgtable_map_p4d(dst_pgdp, src_pgdp, start, next, prot);
331 			if (ret)
332 				return -ENOMEM;
333 		}
334 	} while (dst_pgdp++, src_pgdp++, start = next, start != end);
335 
336 	return 0;
337 }
338 
339 static unsigned long relocate_restore_code(void)
340 {
341 	void *page = (void *)get_safe_page(GFP_ATOMIC);
342 
343 	if (!page)
344 		return -ENOMEM;
345 
346 	copy_page(page, hibernate_core_restore_code);
347 
348 	/* Make the page containing the relocated code executable. */
349 	set_memory_x((unsigned long)page, 1);
350 
351 	return (unsigned long)page;
352 }
353 
354 int swsusp_arch_resume(void)
355 {
356 	unsigned long end = (unsigned long)pfn_to_virt(max_low_pfn);
357 	unsigned long start = PAGE_OFFSET;
358 	int ret;
359 
360 	/*
361 	 * Memory allocated by get_safe_page() will be dealt with by the hibernation core,
362 	 * we don't need to free it here.
363 	 */
364 	resume_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
365 	if (!resume_pg_dir)
366 		return -ENOMEM;
367 
368 	/*
369 	 * Create a temporary page table and map the whole linear region as executable and
370 	 * writable.
371 	 */
372 	ret = temp_pgtable_mapping(resume_pg_dir, start, end, __pgprot(_PAGE_WRITE | _PAGE_EXEC));
373 	if (ret)
374 		return ret;
375 
376 	/* Move the restore code to a new page so that it doesn't get overwritten by itself. */
377 	relocated_restore_code = relocate_restore_code();
378 	if (relocated_restore_code == -ENOMEM)
379 		return -ENOMEM;
380 
381 	/*
382 	 * Map the __hibernate_cpu_resume() address to the temporary page table so that the
383 	 * restore code can jumps to it after finished restore the image. The next execution
384 	 * code doesn't find itself in a different address space after switching over to the
385 	 * original page table used by the hibernated image.
386 	 * The __hibernate_cpu_resume() mapping is unnecessary for RV32 since the kernel and
387 	 * linear addresses are identical, but different for RV64. To ensure consistency, we
388 	 * map it for both RV32 and RV64 kernels.
389 	 * Additionally, we should ensure that the page is writable before restoring the image.
390 	 */
391 	start = (unsigned long)resume_hdr.restore_cpu_addr;
392 	end = start + PAGE_SIZE;
393 
394 	ret = temp_pgtable_mapping(resume_pg_dir, start, end, __pgprot(_PAGE_WRITE));
395 	if (ret)
396 		return ret;
397 
398 	hibernate_restore_image(resume_hdr.saved_satp, (PFN_DOWN(__pa(resume_pg_dir)) | satp_mode),
399 				resume_hdr.restore_cpu_addr);
400 
401 	return 0;
402 }
403 
404 #ifdef CONFIG_PM_SLEEP_SMP
405 int hibernate_resume_nonboot_cpu_disable(void)
406 {
407 	if (sleep_cpu < 0) {
408 		pr_err("Failing to resume from hibernate on an unknown CPU\n");
409 		return -ENODEV;
410 	}
411 
412 	return freeze_secondary_cpus(sleep_cpu);
413 }
414 #endif
415 
416 static int __init riscv_hibernate_init(void)
417 {
418 	hibernate_cpu_context = kzalloc(sizeof(*hibernate_cpu_context), GFP_KERNEL);
419 
420 	if (WARN_ON(!hibernate_cpu_context))
421 		return -ENOMEM;
422 
423 	return 0;
424 }
425 
426 early_initcall(riscv_hibernate_init);
427