1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) IBM Corporation, 2014, 2017
4 * Anton Blanchard, Rashmica Gupta.
5 */
6
7 #define pr_fmt(fmt) "memtrace: " fmt
8
9 #include <linux/bitops.h>
10 #include <linux/string.h>
11 #include <linux/memblock.h>
12 #include <linux/init.h>
13 #include <linux/moduleparam.h>
14 #include <linux/fs.h>
15 #include <linux/debugfs.h>
16 #include <linux/slab.h>
17 #include <linux/memory.h>
18 #include <linux/memory_hotplug.h>
19 #include <linux/numa.h>
20 #include <asm/machdep.h>
21 #include <asm/cacheflush.h>
22
23 /* This enables us to keep track of the memory removed from each node. */
24 struct memtrace_entry {
25 void *mem;
26 u64 start;
27 u64 size;
28 u32 nid;
29 struct dentry *dir;
30 char name[16];
31 };
32
33 static DEFINE_MUTEX(memtrace_mutex);
34 static u64 memtrace_size;
35
36 static struct memtrace_entry *memtrace_array;
37 static unsigned int memtrace_array_nr;
38
39
memtrace_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)40 static ssize_t memtrace_read(struct file *filp, char __user *ubuf,
41 size_t count, loff_t *ppos)
42 {
43 struct memtrace_entry *ent = filp->private_data;
44
45 return simple_read_from_buffer(ubuf, count, ppos, ent->mem, ent->size);
46 }
47
memtrace_mmap(struct file * filp,struct vm_area_struct * vma)48 static int memtrace_mmap(struct file *filp, struct vm_area_struct *vma)
49 {
50 struct memtrace_entry *ent = filp->private_data;
51 unsigned long ent_nrpages = ent->size >> PAGE_SHIFT;
52 unsigned long vma_nrpages = vma_pages(vma);
53
54 /* The requested page offset should be within object's page count */
55 if (vma->vm_pgoff >= ent_nrpages)
56 return -EINVAL;
57
58 /* The requested mapping range should remain within the bounds */
59 if (vma_nrpages > ent_nrpages - vma->vm_pgoff)
60 return -EINVAL;
61
62 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
63 return remap_pfn_range(vma, vma->vm_start, PHYS_PFN(ent->start) + vma->vm_pgoff,
64 vma->vm_end - vma->vm_start, vma->vm_page_prot);
65 }
66
67 static const struct file_operations memtrace_fops = {
68 .llseek = default_llseek,
69 .read = memtrace_read,
70 .open = simple_open,
71 .mmap = memtrace_mmap,
72 };
73
74 #define FLUSH_CHUNK_SIZE SZ_1G
75 /**
76 * flush_dcache_range_chunked(): Write any modified data cache blocks out to
77 * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
78 * Does not invalidate the corresponding instruction cache blocks.
79 *
80 * @start: the start address
81 * @stop: the stop address (exclusive)
82 * @chunk: the max size of the chunks
83 */
flush_dcache_range_chunked(unsigned long start,unsigned long stop,unsigned long chunk)84 static void flush_dcache_range_chunked(unsigned long start, unsigned long stop,
85 unsigned long chunk)
86 {
87 unsigned long i;
88
89 for (i = start; i < stop; i += chunk) {
90 flush_dcache_range(i, min(stop, i + chunk));
91 cond_resched();
92 }
93 }
94
memtrace_alloc_node(u32 nid,u64 size)95 static u64 memtrace_alloc_node(u32 nid, u64 size)
96 {
97 const unsigned long nr_pages = PHYS_PFN(size);
98 unsigned long pfn, start_pfn;
99 struct page *page;
100
101 /*
102 * Trace memory needs to be aligned to the size, which is guaranteed
103 * by alloc_contig_pages().
104 */
105 page = alloc_contig_pages(nr_pages, GFP_KERNEL | __GFP_THISNODE |
106 __GFP_NOWARN | __GFP_ZERO, nid, NULL);
107 if (!page)
108 return 0;
109 start_pfn = page_to_pfn(page);
110
111 /*
112 * Before we go ahead and use this range as cache inhibited range
113 * flush the cache.
114 */
115 flush_dcache_range_chunked((unsigned long)pfn_to_kaddr(start_pfn),
116 (unsigned long)pfn_to_kaddr(start_pfn + nr_pages),
117 FLUSH_CHUNK_SIZE);
118
119 /*
120 * Set pages PageOffline(), to indicate that nobody (e.g., hibernation,
121 * dumping, ...) should be touching these pages.
122 */
123 for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++)
124 __SetPageOffline(pfn_to_page(pfn));
125
126 arch_remove_linear_mapping(PFN_PHYS(start_pfn), size);
127
128 return PFN_PHYS(start_pfn);
129 }
130
memtrace_init_regions_runtime(u64 size)131 static int memtrace_init_regions_runtime(u64 size)
132 {
133 u32 nid;
134 u64 m;
135
136 memtrace_array = kzalloc_objs(struct memtrace_entry, num_online_nodes());
137 if (!memtrace_array) {
138 pr_err("Failed to allocate memtrace_array\n");
139 return -EINVAL;
140 }
141
142 for_each_online_node(nid) {
143 m = memtrace_alloc_node(nid, size);
144
145 /*
146 * A node might not have any local memory, so warn but
147 * continue on.
148 */
149 if (!m) {
150 pr_err("Failed to allocate trace memory on node %d\n", nid);
151 continue;
152 }
153
154 pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m);
155
156 memtrace_array[memtrace_array_nr].start = m;
157 memtrace_array[memtrace_array_nr].size = size;
158 memtrace_array[memtrace_array_nr].nid = nid;
159 memtrace_array_nr++;
160 }
161
162 return 0;
163 }
164
165 static struct dentry *memtrace_debugfs_dir;
166
memtrace_init_debugfs(void)167 static int memtrace_init_debugfs(void)
168 {
169 int ret = 0;
170 int i;
171
172 for (i = 0; i < memtrace_array_nr; i++) {
173 struct dentry *dir;
174 struct memtrace_entry *ent = &memtrace_array[i];
175
176 ent->mem = ioremap(ent->start, ent->size);
177 /* Warn but continue on */
178 if (!ent->mem) {
179 pr_err("Failed to map trace memory at 0x%llx\n",
180 ent->start);
181 ret = -1;
182 continue;
183 }
184
185 snprintf(ent->name, 16, "%08x", ent->nid);
186 dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir);
187
188 ent->dir = dir;
189 debugfs_create_file_unsafe("trace", 0600, dir, ent, &memtrace_fops);
190 debugfs_create_x64("start", 0400, dir, &ent->start);
191 debugfs_create_x64("size", 0400, dir, &ent->size);
192 }
193
194 return ret;
195 }
196
memtrace_free(int nid,u64 start,u64 size)197 static int memtrace_free(int nid, u64 start, u64 size)
198 {
199 struct mhp_params params = { .pgprot = PAGE_KERNEL };
200 const unsigned long nr_pages = PHYS_PFN(size);
201 const unsigned long start_pfn = PHYS_PFN(start);
202 unsigned long pfn;
203 int ret;
204
205 ret = arch_create_linear_mapping(nid, start, size, ¶ms);
206 if (ret)
207 return ret;
208
209 for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++)
210 __ClearPageOffline(pfn_to_page(pfn));
211
212 free_contig_range(start_pfn, nr_pages);
213 return 0;
214 }
215
216 /*
217 * Iterate through the chunks of memory we allocated and attempt to expose
218 * them back to the kernel.
219 */
memtrace_free_regions(void)220 static int memtrace_free_regions(void)
221 {
222 int i, ret = 0;
223 struct memtrace_entry *ent;
224
225 for (i = memtrace_array_nr - 1; i >= 0; i--) {
226 ent = &memtrace_array[i];
227
228 /* We have freed this chunk previously */
229 if (ent->nid == NUMA_NO_NODE)
230 continue;
231
232 /* Remove from io mappings */
233 if (ent->mem) {
234 iounmap(ent->mem);
235 ent->mem = 0;
236 }
237
238 if (memtrace_free(ent->nid, ent->start, ent->size)) {
239 pr_err("Failed to free trace memory on node %d\n",
240 ent->nid);
241 ret += 1;
242 continue;
243 }
244
245 /*
246 * Memory was freed successfully so clean up references to it
247 * so on reentry we can tell that this chunk was freed.
248 */
249 debugfs_remove_recursive(ent->dir);
250 pr_info("Freed trace memory back on node %d\n", ent->nid);
251 ent->size = ent->start = ent->nid = NUMA_NO_NODE;
252 }
253 if (ret)
254 return ret;
255
256 /* If all chunks of memory were freed successfully, reset globals */
257 kfree(memtrace_array);
258 memtrace_array = NULL;
259 memtrace_size = 0;
260 memtrace_array_nr = 0;
261 return 0;
262 }
263
memtrace_enable_set(void * data,u64 val)264 static int memtrace_enable_set(void *data, u64 val)
265 {
266 int rc = -EAGAIN;
267 u64 bytes;
268
269 /*
270 * Don't attempt to do anything if size isn't aligned to a memory
271 * block or equal to zero.
272 */
273 bytes = memory_block_size_bytes();
274 if (val & (bytes - 1)) {
275 pr_err("Value must be aligned with 0x%llx\n", bytes);
276 return -EINVAL;
277 }
278
279 mutex_lock(&memtrace_mutex);
280
281 /* Free all previously allocated memory. */
282 if (memtrace_size && memtrace_free_regions())
283 goto out_unlock;
284
285 if (!val) {
286 rc = 0;
287 goto out_unlock;
288 }
289
290 /* Allocate memory. */
291 if (memtrace_init_regions_runtime(val))
292 goto out_unlock;
293
294 if (memtrace_init_debugfs())
295 goto out_unlock;
296
297 memtrace_size = val;
298 rc = 0;
299 out_unlock:
300 mutex_unlock(&memtrace_mutex);
301 return rc;
302 }
303
memtrace_enable_get(void * data,u64 * val)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
memtrace_init(void)313 static int memtrace_init(void)
314 {
315 memtrace_debugfs_dir = debugfs_create_dir("memtrace",
316 arch_debugfs_dir);
317
318 debugfs_create_file("enable", 0600, memtrace_debugfs_dir,
319 NULL, &memtrace_init_fops);
320
321 return 0;
322 }
323 machine_device_initcall(powernv, memtrace_init);
324