xref: /linux/kernel/liveupdate/kho_block.c (revision 0349ff2887059112ce06831ab29aec47a2a7285a)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Copyright (c) 2026, Google LLC.
5  * Pasha Tatashin <pasha.tatashin@soleen.com>
6  */
7 
8 /**
9  * DOC: KHO Serialization Blocks
10  *
11  * KHO provides a mechanism to preserve stateful data across a kexec handover
12  * by serializing it into memory blocks, and provides the common
13  * infrastructure for managing these blocks.
14  *
15  * Each block consists of a header (struct kho_block_header_ser) followed by an
16  * array of serialized entries. Multiple blocks are linked together via a
17  * physical pointer in the header, forming a linked list that can be easily
18  * traversed in both the current and the next kernel.
19  */
20 
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 
23 #include <linux/io.h>
24 #include <linux/kexec_handover.h>
25 #include <linux/kho/abi/block.h>
26 #include <linux/kho_block.h>
27 #include <linux/slab.h>
28 
29 /*
30  * Safeguard limit for the number of serialization blocks. This is used to
31  * prevent infinite loops and excessive memory allocation in case of memory
32  * corruption in the preserved state.
33  *
34  * With a 4KB page size, 10k blocks is about 40MB. For 32-byte entries
35  * (e.g. 4 u64s), each block holds up to 127 entries (accounting for the
36  * 16-byte header), allowing the block set to hold up to 1.27M entries.
37  */
38 #define KHO_MAX_BLOCKS 10000
39 
40 /**
41  * kho_block_set_init - Initialize a block set.
42  * @bs:         The block set to initialize.
43  * @entry_size: The size of each entry in the blocks.
44  */
45 void kho_block_set_init(struct kho_block_set *bs, size_t entry_size)
46 {
47 	*bs = (struct kho_block_set)KHO_BLOCK_SET_INIT(*bs, entry_size);
48 	WARN_ON_ONCE(!bs->count_per_block);
49 }
50 
51 /* Serialized entries start immediately after the block header */
52 static void *kho_block_entries(struct kho_block *block)
53 {
54 	return (void *)(block->ser + 1);
55 }
56 
57 /* Get the address of the serialized entry at the specified index */
58 static void *kho_block_entry(struct kho_block_set_it *it, u64 index)
59 {
60 	return kho_block_entries(it->block) + (index * it->bs->entry_size);
61 }
62 
63 /* Free serialized data */
64 static void kho_block_free_ser(struct kho_block_set *bs,
65 			       struct kho_block_header_ser *ser)
66 {
67 	if (bs->incoming)
68 		kho_restore_free(ser);
69 	else
70 		kho_unpreserve_free(ser);
71 }
72 
73 static struct kho_block_header_ser *kho_block_alloc_ser(struct kho_block_set *bs)
74 {
75 	WARN_ON_ONCE(bs->incoming);
76 	return kho_alloc_preserve(KHO_BLOCK_SIZE);
77 }
78 
79 static int kho_block_add(struct kho_block_set *bs,
80 			 struct kho_block_header_ser *ser)
81 {
82 	struct kho_block *block, *last;
83 
84 	if (bs->nblocks >= KHO_MAX_BLOCKS)
85 		return -ENOSPC;
86 
87 	block = kzalloc_obj(*block);
88 	if (!block)
89 		return -ENOMEM;
90 
91 	block->ser = ser;
92 	last = list_last_entry_or_null(&bs->blocks, struct kho_block, list);
93 	list_add_tail(&block->list, &bs->blocks);
94 	bs->nblocks++;
95 
96 	if (last)
97 		last->ser->next = virt_to_phys(ser);
98 	else
99 		bs->head_pa = virt_to_phys(ser);
100 
101 	return 0;
102 }
103 
104 static int kho_block_set_grow_one(struct kho_block_set *bs)
105 {
106 	struct kho_block_header_ser *ser;
107 	int err;
108 
109 	ser = kho_block_alloc_ser(bs);
110 	if (IS_ERR(ser))
111 		return PTR_ERR(ser);
112 
113 	err = kho_block_add(bs, ser);
114 	if (err) {
115 		kho_block_free_ser(bs, ser);
116 		return err;
117 	}
118 
119 	return 0;
120 }
121 
122 static void kho_block_set_shrink_one(struct kho_block_set *bs)
123 {
124 	struct kho_block *last, *new_last;
125 
126 	if (list_empty(&bs->blocks))
127 		return;
128 
129 	last = list_last_entry(&bs->blocks, struct kho_block, list);
130 	list_del(&last->list);
131 	bs->nblocks--;
132 	kho_block_free_ser(bs, last->ser);
133 	kfree(last);
134 
135 	new_last = list_last_entry_or_null(&bs->blocks, struct kho_block, list);
136 	if (new_last)
137 		new_last->ser->next = 0;
138 	else
139 		bs->head_pa = 0;
140 }
141 
142 /**
143  * kho_block_set_grow - Expand the block set to accommodate the target count.
144  * @bs:    The block set.
145  * @count: The target number of valid entries to accommodate.
146  *
147  * Dynamically preallocates and links preserved memory blocks if the target
148  * entry count exceeds the current total capacity of the set, ensuring they
149  * are available during serialization/deserialization.
150  *
151  * Context: Caller must hold a lock protecting the block set.
152  * Return: 0 on success, or a negative errno on failure.
153  */
154 int kho_block_set_grow(struct kho_block_set *bs, u64 count)
155 {
156 	long orig_nblocks = bs->nblocks;
157 	int err;
158 
159 	if (WARN_ON_ONCE(bs->incoming))
160 		return -EINVAL;
161 
162 	while (count > bs->nblocks * bs->count_per_block) {
163 		err = kho_block_set_grow_one(bs);
164 		if (err)
165 			goto err_shrink;
166 	}
167 
168 	return 0;
169 
170 err_shrink:
171 	while (bs->nblocks > orig_nblocks)
172 		kho_block_set_shrink_one(bs);
173 	return err;
174 }
175 
176 /**
177  * kho_block_set_shrink - Shrink the block set to accommodate the target count.
178  * @bs:              The block set.
179  * @count:           The target number of valid entries to accommodate.
180  *
181  * Releases and unallocates redundant preserved memory blocks. Checks if the
182  * last block in the set can be removed because the remaining entry count is
183  * fully accommodated by the preceding blocks.
184  *
185  * Note: It is the caller's responsibility to ensure that entries are removed
186  * in the reverse order of their insertion. Because shrinking destroys the last
187  * block in the set, removing entries in any other order would corrupt active
188  * data.
189  *
190  * Context: Caller must hold a lock protecting the block set.
191  */
192 void kho_block_set_shrink(struct kho_block_set *bs, u64 count)
193 {
194 	while (bs->nblocks > 0 && count <= (bs->nblocks - 1) * bs->count_per_block)
195 		kho_block_set_shrink_one(bs);
196 }
197 
198 /*
199  * kho_block_set_is_cyclic - Check for cycles in a linked list of blocks.
200  * Uses Floyd's cycle-finding algorithm to ensure sanity of the incoming list.
201  *
202  * Return: true if a cycle or corruption is detected, false otherwise.
203  */
204 static bool kho_block_set_is_cyclic(struct kho_block_set *bs)
205 {
206 	struct kho_block_header_ser *fast;
207 	struct kho_block_header_ser *slow;
208 	int count = 0;
209 
210 	fast = phys_to_virt(bs->head_pa);
211 	slow = fast;
212 
213 	while (fast) {
214 		if (count++ >= KHO_MAX_BLOCKS) {
215 			pr_err("Block set is corrupted\n");
216 			return true;
217 		}
218 
219 		if (!fast->next)
220 			break;
221 
222 		fast = phys_to_virt(fast->next);
223 		if (!fast->next)
224 			break;
225 
226 		fast = phys_to_virt(fast->next);
227 		slow = phys_to_virt(slow->next);
228 
229 		if (slow == fast) {
230 			pr_err("Block set is corrupted\n");
231 			return true;
232 		}
233 	}
234 
235 	return false;
236 }
237 
238 /**
239  * kho_block_set_restore - Restore a block set from a physical address.
240  * @bs:      The block set to restore.
241  * @head_pa: Physical address of the first block header.
242  *
243  * Restores a serialized block set from a given physical address. The caller is
244  * responsible for ensuring that the block set @bs has been allocated and
245  * initialized prior to calling this function.
246  *
247  * Return: 0 on success, or a negative errno on failure.
248  */
249 int kho_block_set_restore(struct kho_block_set *bs, u64 head_pa)
250 {
251 	struct kho_block_header_ser *ser;
252 	u64 next_pa = head_pa;
253 	int err;
254 
255 	/* Restored block sets use size from the previous kernel */
256 	bs->incoming = true;
257 	if (!head_pa)
258 		return 0;
259 
260 	bs->head_pa = head_pa;
261 	if (kho_block_set_is_cyclic(bs)) {
262 		bs->head_pa = 0;
263 		return -EINVAL;
264 	}
265 
266 	while (next_pa) {
267 		ser = phys_to_virt(next_pa);
268 		if (!ser->count || ser->count > bs->count_per_block) {
269 			pr_warn("Block contains invalid entry count: %llu\n",
270 				ser->count);
271 			err = -EINVAL;
272 			goto err_destroy;
273 		}
274 		err = kho_block_add(bs, ser);
275 		if (err)
276 			goto err_destroy;
277 		next_pa = ser->next;
278 	}
279 
280 	return 0;
281 
282 err_destroy:
283 	kho_block_set_destroy(bs);
284 
285 	/* Free the remaining un-restored blocks in the physical chain */
286 	while (next_pa) {
287 		struct kho_block_header_ser *next_ser = phys_to_virt(next_pa);
288 
289 		next_pa = next_ser->next;
290 		kho_block_free_ser(bs, next_ser);
291 	}
292 	return err;
293 }
294 
295 /**
296  * kho_block_set_destroy - Destroy all blocks in a block set.
297  * @bs:          The block set.
298  */
299 void kho_block_set_destroy(struct kho_block_set *bs)
300 {
301 	struct kho_block *block, *tmp;
302 
303 	list_for_each_entry_safe(block, tmp, &bs->blocks, list) {
304 		list_del(&block->list);
305 		kho_block_free_ser(bs, block->ser);
306 		kfree(block);
307 	}
308 	bs->nblocks = 0;
309 	bs->head_pa = 0;
310 }
311 
312 /**
313  * kho_block_set_clear - Clear all serialized data in a block set.
314  * @bs: The block set to clear.
315  */
316 void kho_block_set_clear(struct kho_block_set *bs)
317 {
318 	struct kho_block *block;
319 
320 	list_for_each_entry(block, &bs->blocks, list) {
321 		block->ser->count = 0;
322 		memset(block->ser + 1, 0, KHO_BLOCK_SIZE - sizeof(*block->ser));
323 	}
324 }
325 
326 /**
327  * kho_block_set_it_init - Initialize a block set iterator.
328  * @it:         The iterator to initialize.
329  * @bs:         The block set to iterate over.
330  */
331 void kho_block_set_it_init(struct kho_block_set_it *it, struct kho_block_set *bs)
332 {
333 	it->bs = bs;
334 	it->block = list_first_entry_or_null(&bs->blocks, struct kho_block, list);
335 	it->i = 0;
336 }
337 
338 /**
339  * kho_block_set_it_reserve_entry - Reserve and return the next available slot for writing.
340  * @it: The block iterator.
341  *
342  * Reserves a slot in the current block during state serialization to add a new
343  * entry, advancing the internal index. If the current block is full, it
344  * automatically moves to the next block in the set.
345  *
346  * Return: A pointer to the reserved entry slot, or NULL if the block set's
347  * capacity is fully exhausted.
348  */
349 void *kho_block_set_it_reserve_entry(struct kho_block_set_it *it)
350 {
351 	void *entry;
352 
353 	if (!it->block)
354 		return NULL;
355 
356 	if (it->i == it->bs->count_per_block) {
357 		if (list_is_last(&it->block->list, &it->bs->blocks))
358 			return NULL;
359 		it->block = list_next_entry(it->block, list);
360 		it->i = 0;
361 	}
362 
363 	entry = kho_block_entry(it, it->i++);
364 	it->block->ser->count = it->i;
365 	return entry;
366 }
367 
368 /**
369  * kho_block_set_it_read_entry - Read the next serialized entry from the block set.
370  * @it: The block iterator.
371  *
372  * Iterates through previously written entries during state deserialization,
373  * respecting the actual count stored in each block's header.
374  *
375  * Return: A pointer to the next serialized entry, or NULL if all serialized
376  * entries have been read.
377  */
378 void *kho_block_set_it_read_entry(struct kho_block_set_it *it)
379 {
380 	if (!it->block)
381 		return NULL;
382 
383 	if (it->i == it->block->ser->count) {
384 		if (list_is_last(&it->block->list, &it->bs->blocks))
385 			return NULL;
386 		it->block = list_next_entry(it->block, list);
387 		it->i = 0;
388 	}
389 
390 	return kho_block_entry(it, it->i++);
391 }
392 
393 /**
394  * kho_block_set_it_prev - Return the previous entry slot in the block set.
395  * @it: The block iterator.
396  *
397  * If the current index is at the start of a block, it automatically moves to
398  * the end of the previous block.
399  *
400  * Return: A pointer to the previous entry slot, or NULL if at the very
401  * beginning of the block set.
402  */
403 void *kho_block_set_it_prev(struct kho_block_set_it *it)
404 {
405 	if (!it->block)
406 		return NULL;
407 
408 	if (it->i == 0) {
409 		if (list_is_first(&it->block->list, &it->bs->blocks))
410 			return NULL;
411 		it->block = list_prev_entry(it->block, list);
412 		it->i = it->bs->count_per_block;
413 	}
414 
415 	return kho_block_entry(it, --it->i);
416 }
417