mm/memory_hotplug: make remove_memory() take the device_hotplug_lock
[sfrench/cifs-2.6.git] / arch / powerpc / platforms / powernv / memtrace.c
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>
23 #include <asm/machdep.h>
24 #include <asm/debugfs.h>
25
26 /* This enables us to keep track of the memory removed from each node. */
27 struct memtrace_entry {
28         void *mem;
29         u64 start;
30         u64 size;
31         u32 nid;
32         struct dentry *dir;
33         char name[16];
34 };
35
36 static u64 memtrace_size;
37
38 static struct memtrace_entry *memtrace_array;
39 static unsigned int memtrace_array_nr;
40
41
42 static ssize_t memtrace_read(struct file *filp, char __user *ubuf,
43                              size_t count, loff_t *ppos)
44 {
45         struct memtrace_entry *ent = filp->private_data;
46
47         return simple_read_from_buffer(ubuf, count, ppos, ent->mem, ent->size);
48 }
49
50 static const struct file_operations memtrace_fops = {
51         .llseek = default_llseek,
52         .read   = memtrace_read,
53         .open   = simple_open,
54 };
55
56 static int check_memblock_online(struct memory_block *mem, void *arg)
57 {
58         if (mem->state != MEM_ONLINE)
59                 return -1;
60
61         return 0;
62 }
63
64 static int change_memblock_state(struct memory_block *mem, void *arg)
65 {
66         unsigned long state = (unsigned long)arg;
67
68         mem->state = state;
69
70         return 0;
71 }
72
73 static bool memtrace_offline_pages(u32 nid, u64 start_pfn, u64 nr_pages)
74 {
75         u64 end_pfn = start_pfn + nr_pages - 1;
76
77         if (walk_memory_range(start_pfn, end_pfn, NULL,
78             check_memblock_online))
79                 return false;
80
81         walk_memory_range(start_pfn, end_pfn, (void *)MEM_GOING_OFFLINE,
82                           change_memblock_state);
83
84         if (offline_pages(start_pfn, nr_pages)) {
85                 walk_memory_range(start_pfn, end_pfn, (void *)MEM_ONLINE,
86                                   change_memblock_state);
87                 return false;
88         }
89
90         walk_memory_range(start_pfn, end_pfn, (void *)MEM_OFFLINE,
91                           change_memblock_state);
92
93
94         return true;
95 }
96
97 static u64 memtrace_alloc_node(u32 nid, u64 size)
98 {
99         u64 start_pfn, end_pfn, nr_pages, pfn;
100         u64 base_pfn;
101         u64 bytes = memory_block_size_bytes();
102
103         if (!node_spanned_pages(nid))
104                 return 0;
105
106         start_pfn = node_start_pfn(nid);
107         end_pfn = node_end_pfn(nid);
108         nr_pages = size >> PAGE_SHIFT;
109
110         /* Trace memory needs to be aligned to the size */
111         end_pfn = round_down(end_pfn - nr_pages, nr_pages);
112
113         for (base_pfn = end_pfn; base_pfn > start_pfn; base_pfn -= nr_pages) {
114                 if (memtrace_offline_pages(nid, base_pfn, nr_pages) == true) {
115                         /*
116                          * Remove memory in memory block size chunks so that
117                          * iomem resources are always split to the same size and
118                          * we never try to remove memory that spans two iomem
119                          * resources.
120                          */
121                         lock_device_hotplug();
122                         end_pfn = base_pfn + nr_pages;
123                         for (pfn = base_pfn; pfn < end_pfn; pfn += bytes>> PAGE_SHIFT) {
124                                 __remove_memory(nid, pfn << PAGE_SHIFT, bytes);
125                         }
126                         unlock_device_hotplug();
127                         return base_pfn << PAGE_SHIFT;
128                 }
129         }
130
131         return 0;
132 }
133
134 static int memtrace_init_regions_runtime(u64 size)
135 {
136         u32 nid;
137         u64 m;
138
139         memtrace_array = kcalloc(num_online_nodes(),
140                                 sizeof(struct memtrace_entry), GFP_KERNEL);
141         if (!memtrace_array) {
142                 pr_err("Failed to allocate memtrace_array\n");
143                 return -EINVAL;
144         }
145
146         for_each_online_node(nid) {
147                 m = memtrace_alloc_node(nid, size);
148
149                 /*
150                  * A node might not have any local memory, so warn but
151                  * continue on.
152                  */
153                 if (!m) {
154                         pr_err("Failed to allocate trace memory on node %d\n", nid);
155                         continue;
156                 }
157
158                 pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m);
159
160                 memtrace_array[memtrace_array_nr].start = m;
161                 memtrace_array[memtrace_array_nr].size = size;
162                 memtrace_array[memtrace_array_nr].nid = nid;
163                 memtrace_array_nr++;
164         }
165
166         return 0;
167 }
168
169 static struct dentry *memtrace_debugfs_dir;
170
171 static int memtrace_init_debugfs(void)
172 {
173         int ret = 0;
174         int i;
175
176         for (i = 0; i < memtrace_array_nr; i++) {
177                 struct dentry *dir;
178                 struct memtrace_entry *ent = &memtrace_array[i];
179
180                 ent->mem = ioremap(ent->start, ent->size);
181                 /* Warn but continue on */
182                 if (!ent->mem) {
183                         pr_err("Failed to map trace memory at 0x%llx\n",
184                                  ent->start);
185                         ret = -1;
186                         continue;
187                 }
188
189                 snprintf(ent->name, 16, "%08x", ent->nid);
190                 dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir);
191                 if (!dir) {
192                         pr_err("Failed to create debugfs directory for node %d\n",
193                                 ent->nid);
194                         return -1;
195                 }
196
197                 ent->dir = dir;
198                 debugfs_create_file("trace", 0400, dir, ent, &memtrace_fops);
199                 debugfs_create_x64("start", 0400, dir, &ent->start);
200                 debugfs_create_x64("size", 0400, dir, &ent->size);
201         }
202
203         return ret;
204 }
205
206 static int online_mem_block(struct memory_block *mem, void *arg)
207 {
208         return device_online(&mem->dev);
209 }
210
211 /*
212  * Iterate through the chunks of memory we have removed from the kernel
213  * and attempt to add them back to the kernel.
214  */
215 static int memtrace_online(void)
216 {
217         int i, ret = 0;
218         struct memtrace_entry *ent;
219
220         for (i = memtrace_array_nr - 1; i >= 0; i--) {
221                 ent = &memtrace_array[i];
222
223                 /* We have onlined this chunk previously */
224                 if (ent->nid == -1)
225                         continue;
226
227                 /* Remove from io mappings */
228                 if (ent->mem) {
229                         iounmap(ent->mem);
230                         ent->mem = 0;
231                 }
232
233                 if (add_memory(ent->nid, ent->start, ent->size)) {
234                         pr_err("Failed to add trace memory to node %d\n",
235                                 ent->nid);
236                         ret += 1;
237                         continue;
238                 }
239
240                 /*
241                  * If kernel isn't compiled with the auto online option
242                  * we need to online the memory ourselves.
243                  */
244                 if (!memhp_auto_online) {
245                         walk_memory_range(PFN_DOWN(ent->start),
246                                           PFN_UP(ent->start + ent->size - 1),
247                                           NULL, online_mem_block);
248                 }
249
250                 /*
251                  * Memory was added successfully so clean up references to it
252                  * so on reentry we can tell that this chunk was added.
253                  */
254                 debugfs_remove_recursive(ent->dir);
255                 pr_info("Added trace memory back to node %d\n", ent->nid);
256                 ent->size = ent->start = ent->nid = -1;
257         }
258         if (ret)
259                 return ret;
260
261         /* If all chunks of memory were added successfully, reset globals */
262         kfree(memtrace_array);
263         memtrace_array = NULL;
264         memtrace_size = 0;
265         memtrace_array_nr = 0;
266         return 0;
267 }
268
269 static int memtrace_enable_set(void *data, u64 val)
270 {
271         u64 bytes;
272
273         /*
274          * Don't attempt to do anything if size isn't aligned to a memory
275          * block or equal to zero.
276          */
277         bytes = memory_block_size_bytes();
278         if (val & (bytes - 1)) {
279                 pr_err("Value must be aligned with 0x%llx\n", bytes);
280                 return -EINVAL;
281         }
282
283         /* Re-add/online previously removed/offlined memory */
284         if (memtrace_size) {
285                 if (memtrace_online())
286                         return -EAGAIN;
287         }
288
289         if (!val)
290                 return 0;
291
292         /* Offline and remove memory */
293         if (memtrace_init_regions_runtime(val))
294                 return -EINVAL;
295
296         if (memtrace_init_debugfs())
297                 return -EINVAL;
298
299         memtrace_size = val;
300
301         return 0;
302 }
303
304 static int memtrace_enable_get(void *data, u64 *val)
305 {
306         *val = memtrace_size;
307         return 0;
308 }
309
310 DEFINE_SIMPLE_ATTRIBUTE(memtrace_init_fops, memtrace_enable_get,
311                                         memtrace_enable_set, "0x%016llx\n");
312
313 static int memtrace_init(void)
314 {
315         memtrace_debugfs_dir = debugfs_create_dir("memtrace",
316                                                   powerpc_debugfs_root);
317         if (!memtrace_debugfs_dir)
318                 return -1;
319
320         debugfs_create_file("enable", 0600, memtrace_debugfs_dir,
321                             NULL, &memtrace_init_fops);
322
323         return 0;
324 }
325 machine_device_initcall(powernv, memtrace_init);