xref: /linux/arch/powerpc/platforms/powernv/memtrace.c (revision 9c5968db9e625019a0ee5226c7eebef5519d366a)
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 
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 
48 static int memtrace_mmap(struct file *filp, struct vm_area_struct *vma)
49 {
50 	struct memtrace_entry *ent = filp->private_data;
51 
52 	if (ent->size < vma->vm_end - vma->vm_start)
53 		return -EINVAL;
54 
55 	if (vma->vm_pgoff << PAGE_SHIFT >= ent->size)
56 		return -EINVAL;
57 
58 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
59 	return remap_pfn_range(vma, vma->vm_start, PHYS_PFN(ent->start) + vma->vm_pgoff,
60 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
61 }
62 
63 static const struct file_operations memtrace_fops = {
64 	.llseek = default_llseek,
65 	.read	= memtrace_read,
66 	.open	= simple_open,
67 	.mmap   = memtrace_mmap,
68 };
69 
70 #define FLUSH_CHUNK_SIZE SZ_1G
71 /**
72  * flush_dcache_range_chunked(): Write any modified data cache blocks out to
73  * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
74  * Does not invalidate the corresponding instruction cache blocks.
75  *
76  * @start: the start address
77  * @stop: the stop address (exclusive)
78  * @chunk: the max size of the chunks
79  */
80 static void flush_dcache_range_chunked(unsigned long start, unsigned long stop,
81 				       unsigned long chunk)
82 {
83 	unsigned long i;
84 
85 	for (i = start; i < stop; i += chunk) {
86 		flush_dcache_range(i, min(stop, i + chunk));
87 		cond_resched();
88 	}
89 }
90 
91 static u64 memtrace_alloc_node(u32 nid, u64 size)
92 {
93 	const unsigned long nr_pages = PHYS_PFN(size);
94 	unsigned long pfn, start_pfn;
95 	struct page *page;
96 
97 	/*
98 	 * Trace memory needs to be aligned to the size, which is guaranteed
99 	 * by alloc_contig_pages().
100 	 */
101 	page = alloc_contig_pages(nr_pages, GFP_KERNEL | __GFP_THISNODE |
102 				  __GFP_NOWARN | __GFP_ZERO, nid, NULL);
103 	if (!page)
104 		return 0;
105 	start_pfn = page_to_pfn(page);
106 
107 	/*
108 	 * Before we go ahead and use this range as cache inhibited range
109 	 * flush the cache.
110 	 */
111 	flush_dcache_range_chunked((unsigned long)pfn_to_kaddr(start_pfn),
112 				   (unsigned long)pfn_to_kaddr(start_pfn + nr_pages),
113 				   FLUSH_CHUNK_SIZE);
114 
115 	/*
116 	 * Set pages PageOffline(), to indicate that nobody (e.g., hibernation,
117 	 * dumping, ...) should be touching these pages.
118 	 */
119 	for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++)
120 		__SetPageOffline(pfn_to_page(pfn));
121 
122 	arch_remove_linear_mapping(PFN_PHYS(start_pfn), size);
123 
124 	return PFN_PHYS(start_pfn);
125 }
126 
127 static int memtrace_init_regions_runtime(u64 size)
128 {
129 	u32 nid;
130 	u64 m;
131 
132 	memtrace_array = kcalloc(num_online_nodes(),
133 				sizeof(struct memtrace_entry), GFP_KERNEL);
134 	if (!memtrace_array) {
135 		pr_err("Failed to allocate memtrace_array\n");
136 		return -EINVAL;
137 	}
138 
139 	for_each_online_node(nid) {
140 		m = memtrace_alloc_node(nid, size);
141 
142 		/*
143 		 * A node might not have any local memory, so warn but
144 		 * continue on.
145 		 */
146 		if (!m) {
147 			pr_err("Failed to allocate trace memory on node %d\n", nid);
148 			continue;
149 		}
150 
151 		pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m);
152 
153 		memtrace_array[memtrace_array_nr].start = m;
154 		memtrace_array[memtrace_array_nr].size = size;
155 		memtrace_array[memtrace_array_nr].nid = nid;
156 		memtrace_array_nr++;
157 	}
158 
159 	return 0;
160 }
161 
162 static struct dentry *memtrace_debugfs_dir;
163 
164 static int memtrace_init_debugfs(void)
165 {
166 	int ret = 0;
167 	int i;
168 
169 	for (i = 0; i < memtrace_array_nr; i++) {
170 		struct dentry *dir;
171 		struct memtrace_entry *ent = &memtrace_array[i];
172 
173 		ent->mem = ioremap(ent->start, ent->size);
174 		/* Warn but continue on */
175 		if (!ent->mem) {
176 			pr_err("Failed to map trace memory at 0x%llx\n",
177 				 ent->start);
178 			ret = -1;
179 			continue;
180 		}
181 
182 		snprintf(ent->name, 16, "%08x", ent->nid);
183 		dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir);
184 
185 		ent->dir = dir;
186 		debugfs_create_file_unsafe("trace", 0600, dir, ent, &memtrace_fops);
187 		debugfs_create_x64("start", 0400, dir, &ent->start);
188 		debugfs_create_x64("size", 0400, dir, &ent->size);
189 	}
190 
191 	return ret;
192 }
193 
194 static int memtrace_free(int nid, u64 start, u64 size)
195 {
196 	struct mhp_params params = { .pgprot = PAGE_KERNEL };
197 	const unsigned long nr_pages = PHYS_PFN(size);
198 	const unsigned long start_pfn = PHYS_PFN(start);
199 	unsigned long pfn;
200 	int ret;
201 
202 	ret = arch_create_linear_mapping(nid, start, size, &params);
203 	if (ret)
204 		return ret;
205 
206 	for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++)
207 		__ClearPageOffline(pfn_to_page(pfn));
208 
209 	free_contig_range(start_pfn, nr_pages);
210 	return 0;
211 }
212 
213 /*
214  * Iterate through the chunks of memory we allocated and attempt to expose
215  * them back to the kernel.
216  */
217 static int memtrace_free_regions(void)
218 {
219 	int i, ret = 0;
220 	struct memtrace_entry *ent;
221 
222 	for (i = memtrace_array_nr - 1; i >= 0; i--) {
223 		ent = &memtrace_array[i];
224 
225 		/* We have freed this chunk previously */
226 		if (ent->nid == NUMA_NO_NODE)
227 			continue;
228 
229 		/* Remove from io mappings */
230 		if (ent->mem) {
231 			iounmap(ent->mem);
232 			ent->mem = 0;
233 		}
234 
235 		if (memtrace_free(ent->nid, ent->start, ent->size)) {
236 			pr_err("Failed to free trace memory on node %d\n",
237 				ent->nid);
238 			ret += 1;
239 			continue;
240 		}
241 
242 		/*
243 		 * Memory was freed successfully so clean up references to it
244 		 * so on reentry we can tell that this chunk was freed.
245 		 */
246 		debugfs_remove_recursive(ent->dir);
247 		pr_info("Freed trace memory back on node %d\n", ent->nid);
248 		ent->size = ent->start = ent->nid = NUMA_NO_NODE;
249 	}
250 	if (ret)
251 		return ret;
252 
253 	/* If all chunks of memory were freed successfully, reset globals */
254 	kfree(memtrace_array);
255 	memtrace_array = NULL;
256 	memtrace_size = 0;
257 	memtrace_array_nr = 0;
258 	return 0;
259 }
260 
261 static int memtrace_enable_set(void *data, u64 val)
262 {
263 	int rc = -EAGAIN;
264 	u64 bytes;
265 
266 	/*
267 	 * Don't attempt to do anything if size isn't aligned to a memory
268 	 * block or equal to zero.
269 	 */
270 	bytes = memory_block_size_bytes();
271 	if (val & (bytes - 1)) {
272 		pr_err("Value must be aligned with 0x%llx\n", bytes);
273 		return -EINVAL;
274 	}
275 
276 	mutex_lock(&memtrace_mutex);
277 
278 	/* Free all previously allocated memory. */
279 	if (memtrace_size && memtrace_free_regions())
280 		goto out_unlock;
281 
282 	if (!val) {
283 		rc = 0;
284 		goto out_unlock;
285 	}
286 
287 	/* Allocate memory. */
288 	if (memtrace_init_regions_runtime(val))
289 		goto out_unlock;
290 
291 	if (memtrace_init_debugfs())
292 		goto out_unlock;
293 
294 	memtrace_size = val;
295 	rc = 0;
296 out_unlock:
297 	mutex_unlock(&memtrace_mutex);
298 	return rc;
299 }
300 
301 static int memtrace_enable_get(void *data, u64 *val)
302 {
303 	*val = memtrace_size;
304 	return 0;
305 }
306 
307 DEFINE_SIMPLE_ATTRIBUTE(memtrace_init_fops, memtrace_enable_get,
308 					memtrace_enable_set, "0x%016llx\n");
309 
310 static int memtrace_init(void)
311 {
312 	memtrace_debugfs_dir = debugfs_create_dir("memtrace",
313 						  arch_debugfs_dir);
314 
315 	debugfs_create_file("enable", 0600, memtrace_debugfs_dir,
316 			    NULL, &memtrace_init_fops);
317 
318 	return 0;
319 }
320 machine_device_initcall(powernv, memtrace_init);
321