Commit | Line | Data |
---|---|---|
9d5171a8 RG |
1 | /* |
2 | * Copyright (C) IBM Corporation, 2014, 2017 | |
3 | * Anton Blanchard, Rashmica Gupta. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | */ | |
10 | ||
11 | #define pr_fmt(fmt) "memtrace: " fmt | |
12 | ||
13 | #include <linux/bitops.h> | |
14 | #include <linux/string.h> | |
15 | #include <linux/memblock.h> | |
16 | #include <linux/init.h> | |
17 | #include <linux/moduleparam.h> | |
18 | #include <linux/fs.h> | |
19 | #include <linux/debugfs.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/memory.h> | |
22 | #include <linux/memory_hotplug.h> | |
98fa15f3 | 23 | #include <linux/numa.h> |
9d5171a8 RG |
24 | #include <asm/machdep.h> |
25 | #include <asm/debugfs.h> | |
26 | ||
27 | /* This enables us to keep track of the memory removed from each node. */ | |
28 | struct memtrace_entry { | |
29 | void *mem; | |
30 | u64 start; | |
31 | u64 size; | |
32 | u32 nid; | |
33 | struct dentry *dir; | |
34 | char name[16]; | |
35 | }; | |
36 | ||
37 | static u64 memtrace_size; | |
38 | ||
39 | static struct memtrace_entry *memtrace_array; | |
40 | static unsigned int memtrace_array_nr; | |
41 | ||
42 | ||
43 | static ssize_t memtrace_read(struct file *filp, char __user *ubuf, | |
44 | size_t count, loff_t *ppos) | |
45 | { | |
46 | struct memtrace_entry *ent = filp->private_data; | |
47 | ||
48 | return simple_read_from_buffer(ubuf, count, ppos, ent->mem, ent->size); | |
49 | } | |
50 | ||
9d5171a8 RG |
51 | static const struct file_operations memtrace_fops = { |
52 | .llseek = default_llseek, | |
53 | .read = memtrace_read, | |
9d5171a8 RG |
54 | .open = simple_open, |
55 | }; | |
56 | ||
9d5171a8 RG |
57 | static int check_memblock_online(struct memory_block *mem, void *arg) |
58 | { | |
59 | if (mem->state != MEM_ONLINE) | |
60 | return -1; | |
61 | ||
62 | return 0; | |
63 | } | |
64 | ||
65 | static int change_memblock_state(struct memory_block *mem, void *arg) | |
66 | { | |
67 | unsigned long state = (unsigned long)arg; | |
68 | ||
69 | mem->state = state; | |
70 | ||
71 | return 0; | |
72 | } | |
73 | ||
56668487 | 74 | /* called with device_hotplug_lock held */ |
9d5171a8 RG |
75 | static bool memtrace_offline_pages(u32 nid, u64 start_pfn, u64 nr_pages) |
76 | { | |
77 | u64 end_pfn = start_pfn + nr_pages - 1; | |
78 | ||
79 | if (walk_memory_range(start_pfn, end_pfn, NULL, | |
80 | check_memblock_online)) | |
81 | return false; | |
82 | ||
83 | walk_memory_range(start_pfn, end_pfn, (void *)MEM_GOING_OFFLINE, | |
84 | change_memblock_state); | |
85 | ||
86 | if (offline_pages(start_pfn, nr_pages)) { | |
87 | walk_memory_range(start_pfn, end_pfn, (void *)MEM_ONLINE, | |
88 | change_memblock_state); | |
89 | return false; | |
90 | } | |
91 | ||
92 | walk_memory_range(start_pfn, end_pfn, (void *)MEM_OFFLINE, | |
93 | change_memblock_state); | |
94 | ||
9d5171a8 RG |
95 | |
96 | return true; | |
97 | } | |
98 | ||
99 | static u64 memtrace_alloc_node(u32 nid, u64 size) | |
100 | { | |
3f7daf3d | 101 | u64 start_pfn, end_pfn, nr_pages, pfn; |
9d5171a8 | 102 | u64 base_pfn; |
3f7daf3d | 103 | u64 bytes = memory_block_size_bytes(); |
9d5171a8 | 104 | |
8ccb442d | 105 | if (!node_spanned_pages(nid)) |
9d5171a8 RG |
106 | return 0; |
107 | ||
108 | start_pfn = node_start_pfn(nid); | |
109 | end_pfn = node_end_pfn(nid); | |
110 | nr_pages = size >> PAGE_SHIFT; | |
111 | ||
112 | /* Trace memory needs to be aligned to the size */ | |
113 | end_pfn = round_down(end_pfn - nr_pages, nr_pages); | |
114 | ||
56668487 | 115 | lock_device_hotplug(); |
9d5171a8 | 116 | for (base_pfn = end_pfn; base_pfn > start_pfn; base_pfn -= nr_pages) { |
3f7daf3d RG |
117 | if (memtrace_offline_pages(nid, base_pfn, nr_pages) == true) { |
118 | /* | |
119 | * Remove memory in memory block size chunks so that | |
120 | * iomem resources are always split to the same size and | |
121 | * we never try to remove memory that spans two iomem | |
122 | * resources. | |
123 | */ | |
3f7daf3d RG |
124 | end_pfn = base_pfn + nr_pages; |
125 | for (pfn = base_pfn; pfn < end_pfn; pfn += bytes>> PAGE_SHIFT) { | |
d15e5926 | 126 | __remove_memory(nid, pfn << PAGE_SHIFT, bytes); |
3f7daf3d RG |
127 | } |
128 | unlock_device_hotplug(); | |
9d5171a8 | 129 | return base_pfn << PAGE_SHIFT; |
3f7daf3d | 130 | } |
9d5171a8 | 131 | } |
56668487 | 132 | unlock_device_hotplug(); |
9d5171a8 RG |
133 | |
134 | return 0; | |
135 | } | |
136 | ||
137 | static int memtrace_init_regions_runtime(u64 size) | |
138 | { | |
139 | u32 nid; | |
140 | u64 m; | |
141 | ||
142 | memtrace_array = kcalloc(num_online_nodes(), | |
143 | sizeof(struct memtrace_entry), GFP_KERNEL); | |
144 | if (!memtrace_array) { | |
145 | pr_err("Failed to allocate memtrace_array\n"); | |
146 | return -EINVAL; | |
147 | } | |
148 | ||
149 | for_each_online_node(nid) { | |
150 | m = memtrace_alloc_node(nid, size); | |
151 | ||
152 | /* | |
153 | * A node might not have any local memory, so warn but | |
154 | * continue on. | |
155 | */ | |
156 | if (!m) { | |
157 | pr_err("Failed to allocate trace memory on node %d\n", nid); | |
158 | continue; | |
159 | } | |
160 | ||
161 | pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m); | |
162 | ||
163 | memtrace_array[memtrace_array_nr].start = m; | |
164 | memtrace_array[memtrace_array_nr].size = size; | |
165 | memtrace_array[memtrace_array_nr].nid = nid; | |
166 | memtrace_array_nr++; | |
167 | } | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
172 | static struct dentry *memtrace_debugfs_dir; | |
173 | ||
174 | static int memtrace_init_debugfs(void) | |
175 | { | |
176 | int ret = 0; | |
177 | int i; | |
178 | ||
179 | for (i = 0; i < memtrace_array_nr; i++) { | |
180 | struct dentry *dir; | |
181 | struct memtrace_entry *ent = &memtrace_array[i]; | |
182 | ||
183 | ent->mem = ioremap(ent->start, ent->size); | |
184 | /* Warn but continue on */ | |
185 | if (!ent->mem) { | |
186 | pr_err("Failed to map trace memory at 0x%llx\n", | |
187 | ent->start); | |
188 | ret = -1; | |
189 | continue; | |
190 | } | |
191 | ||
192 | snprintf(ent->name, 16, "%08x", ent->nid); | |
193 | dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir); | |
d3da701d RG |
194 | if (!dir) { |
195 | pr_err("Failed to create debugfs directory for node %d\n", | |
196 | ent->nid); | |
9d5171a8 | 197 | return -1; |
d3da701d | 198 | } |
9d5171a8 RG |
199 | |
200 | ent->dir = dir; | |
201 | debugfs_create_file("trace", 0400, dir, ent, &memtrace_fops); | |
202 | debugfs_create_x64("start", 0400, dir, &ent->start); | |
203 | debugfs_create_x64("size", 0400, dir, &ent->size); | |
204 | } | |
205 | ||
206 | return ret; | |
207 | } | |
208 | ||
d3da701d RG |
209 | static int online_mem_block(struct memory_block *mem, void *arg) |
210 | { | |
211 | return device_online(&mem->dev); | |
212 | } | |
213 | ||
214 | /* | |
215 | * Iterate through the chunks of memory we have removed from the kernel | |
216 | * and attempt to add them back to the kernel. | |
217 | */ | |
218 | static int memtrace_online(void) | |
219 | { | |
220 | int i, ret = 0; | |
221 | struct memtrace_entry *ent; | |
222 | ||
223 | for (i = memtrace_array_nr - 1; i >= 0; i--) { | |
224 | ent = &memtrace_array[i]; | |
225 | ||
226 | /* We have onlined this chunk previously */ | |
98fa15f3 | 227 | if (ent->nid == NUMA_NO_NODE) |
d3da701d RG |
228 | continue; |
229 | ||
230 | /* Remove from io mappings */ | |
231 | if (ent->mem) { | |
232 | iounmap(ent->mem); | |
233 | ent->mem = 0; | |
234 | } | |
235 | ||
236 | if (add_memory(ent->nid, ent->start, ent->size)) { | |
237 | pr_err("Failed to add trace memory to node %d\n", | |
238 | ent->nid); | |
239 | ret += 1; | |
240 | continue; | |
241 | } | |
242 | ||
243 | /* | |
244 | * If kernel isn't compiled with the auto online option | |
245 | * we need to online the memory ourselves. | |
246 | */ | |
247 | if (!memhp_auto_online) { | |
cec16805 | 248 | lock_device_hotplug(); |
d3da701d RG |
249 | walk_memory_range(PFN_DOWN(ent->start), |
250 | PFN_UP(ent->start + ent->size - 1), | |
251 | NULL, online_mem_block); | |
cec16805 | 252 | unlock_device_hotplug(); |
d3da701d RG |
253 | } |
254 | ||
255 | /* | |
256 | * Memory was added successfully so clean up references to it | |
257 | * so on reentry we can tell that this chunk was added. | |
258 | */ | |
259 | debugfs_remove_recursive(ent->dir); | |
260 | pr_info("Added trace memory back to node %d\n", ent->nid); | |
98fa15f3 | 261 | ent->size = ent->start = ent->nid = NUMA_NO_NODE; |
d3da701d RG |
262 | } |
263 | if (ret) | |
264 | return ret; | |
265 | ||
266 | /* If all chunks of memory were added successfully, reset globals */ | |
267 | kfree(memtrace_array); | |
268 | memtrace_array = NULL; | |
269 | memtrace_size = 0; | |
270 | memtrace_array_nr = 0; | |
271 | return 0; | |
272 | } | |
273 | ||
9d5171a8 RG |
274 | static int memtrace_enable_set(void *data, u64 val) |
275 | { | |
d3da701d RG |
276 | u64 bytes; |
277 | ||
278 | /* | |
279 | * Don't attempt to do anything if size isn't aligned to a memory | |
280 | * block or equal to zero. | |
281 | */ | |
282 | bytes = memory_block_size_bytes(); | |
283 | if (val & (bytes - 1)) { | |
284 | pr_err("Value must be aligned with 0x%llx\n", bytes); | |
9d5171a8 | 285 | return -EINVAL; |
d3da701d | 286 | } |
9d5171a8 | 287 | |
d3da701d RG |
288 | /* Re-add/online previously removed/offlined memory */ |
289 | if (memtrace_size) { | |
290 | if (memtrace_online()) | |
291 | return -EAGAIN; | |
292 | } | |
9d5171a8 | 293 | |
d3da701d RG |
294 | if (!val) |
295 | return 0; | |
9d5171a8 | 296 | |
d3da701d | 297 | /* Offline and remove memory */ |
9d5171a8 RG |
298 | if (memtrace_init_regions_runtime(val)) |
299 | return -EINVAL; | |
300 | ||
301 | if (memtrace_init_debugfs()) | |
302 | return -EINVAL; | |
303 | ||
304 | memtrace_size = val; | |
305 | ||
306 | return 0; | |
307 | } | |
308 | ||
309 | static int memtrace_enable_get(void *data, u64 *val) | |
310 | { | |
311 | *val = memtrace_size; | |
312 | return 0; | |
313 | } | |
314 | ||
315 | DEFINE_SIMPLE_ATTRIBUTE(memtrace_init_fops, memtrace_enable_get, | |
316 | memtrace_enable_set, "0x%016llx\n"); | |
317 | ||
318 | static int memtrace_init(void) | |
319 | { | |
320 | memtrace_debugfs_dir = debugfs_create_dir("memtrace", | |
321 | powerpc_debugfs_root); | |
322 | if (!memtrace_debugfs_dir) | |
323 | return -1; | |
324 | ||
325 | debugfs_create_file("enable", 0600, memtrace_debugfs_dir, | |
326 | NULL, &memtrace_init_fops); | |
327 | ||
328 | return 0; | |
329 | } | |
330 | machine_device_initcall(powernv, memtrace_init); |