memtrace.c

  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. */
 24struct memtrace_entry {
 25	void *mem;
 26	u64 start;
 27	u64 size;
 28	u32 nid;
 29	struct dentry *dir;
 30	char name[16];
 31};
 32
 33static DEFINE_MUTEX(memtrace_mutex);
 34static u64 memtrace_size;
 35
 36static struct memtrace_entry *memtrace_array;
 37static unsigned int memtrace_array_nr;
 38
 39
 40static 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
 48static 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
 67static 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 */
 84static 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
 95static 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
131static int memtrace_init_regions_runtime(u64 size)
132{
133	u32 nid;
134	u64 m;
135
136	memtrace_array = kcalloc(num_online_nodes(),
137				sizeof(struct memtrace_entry), GFP_KERNEL);
138	if (!memtrace_array) {
139		pr_err("Failed to allocate memtrace_array\n");
140		return -EINVAL;
141	}
142
143	for_each_online_node(nid) {
144		m = memtrace_alloc_node(nid, size);
145
146		/*
147		 * A node might not have any local memory, so warn but
148		 * continue on.
149		 */
150		if (!m) {
151			pr_err("Failed to allocate trace memory on node %d\n", nid);
152			continue;
153		}
154
155		pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m);
156
157		memtrace_array[memtrace_array_nr].start = m;
158		memtrace_array[memtrace_array_nr].size = size;
159		memtrace_array[memtrace_array_nr].nid = nid;
160		memtrace_array_nr++;
161	}
162
163	return 0;
164}
165
166static struct dentry *memtrace_debugfs_dir;
167
168static int memtrace_init_debugfs(void)
169{
170	int ret = 0;
171	int i;
172
173	for (i = 0; i < memtrace_array_nr; i++) {
174		struct dentry *dir;
175		struct memtrace_entry *ent = &memtrace_array[i];
176
177		ent->mem = ioremap(ent->start, ent->size);
178		/* Warn but continue on */
179		if (!ent->mem) {
180			pr_err("Failed to map trace memory at 0x%llx\n",
181				 ent->start);
182			ret = -1;
183			continue;
184		}
185
186		snprintf(ent->name, 16, "%08x", ent->nid);
187		dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir);
188
189		ent->dir = dir;
190		debugfs_create_file_unsafe("trace", 0600, dir, ent, &memtrace_fops);
191		debugfs_create_x64("start", 0400, dir, &ent->start);
192		debugfs_create_x64("size", 0400, dir, &ent->size);
193	}
194
195	return ret;
196}
197
198static int memtrace_free(int nid, u64 start, u64 size)
199{
200	struct mhp_params params = { .pgprot = PAGE_KERNEL };
201	const unsigned long nr_pages = PHYS_PFN(size);
202	const unsigned long start_pfn = PHYS_PFN(start);
203	unsigned long pfn;
204	int ret;
205
206	ret = arch_create_linear_mapping(nid, start, size, &params);
207	if (ret)
208		return ret;
209
210	for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++)
211		__ClearPageOffline(pfn_to_page(pfn));
212
213	free_contig_range(start_pfn, nr_pages);
214	return 0;
215}
216
217/*
218 * Iterate through the chunks of memory we allocated and attempt to expose
219 * them back to the kernel.
220 */
221static int memtrace_free_regions(void)
222{
223	int i, ret = 0;
224	struct memtrace_entry *ent;
225
226	for (i = memtrace_array_nr - 1; i >= 0; i--) {
227		ent = &memtrace_array[i];
228
229		/* We have freed this chunk previously */
230		if (ent->nid == NUMA_NO_NODE)
231			continue;
232
233		/* Remove from io mappings */
234		if (ent->mem) {
235			iounmap(ent->mem);
236			ent->mem = 0;
237		}
238
239		if (memtrace_free(ent->nid, ent->start, ent->size)) {
240			pr_err("Failed to free trace memory on node %d\n",
241				ent->nid);
242			ret += 1;
243			continue;
244		}
245
246		/*
247		 * Memory was freed successfully so clean up references to it
248		 * so on reentry we can tell that this chunk was freed.
249		 */
250		debugfs_remove_recursive(ent->dir);
251		pr_info("Freed trace memory back on node %d\n", ent->nid);
252		ent->size = ent->start = ent->nid = NUMA_NO_NODE;
253	}
254	if (ret)
255		return ret;
256
257	/* If all chunks of memory were freed successfully, reset globals */
258	kfree(memtrace_array);
259	memtrace_array = NULL;
260	memtrace_size = 0;
261	memtrace_array_nr = 0;
262	return 0;
263}
264
265static int memtrace_enable_set(void *data, u64 val)
266{
267	int rc = -EAGAIN;
268	u64 bytes;
269
270	/*
271	 * Don't attempt to do anything if size isn't aligned to a memory
272	 * block or equal to zero.
273	 */
274	bytes = memory_block_size_bytes();
275	if (val & (bytes - 1)) {
276		pr_err("Value must be aligned with 0x%llx\n", bytes);
277		return -EINVAL;
278	}
279
280	mutex_lock(&memtrace_mutex);
281
282	/* Free all previously allocated memory. */
283	if (memtrace_size && memtrace_free_regions())
284		goto out_unlock;
285
286	if (!val) {
287		rc = 0;
288		goto out_unlock;
289	}
290
291	/* Allocate memory. */
292	if (memtrace_init_regions_runtime(val))
293		goto out_unlock;
294
295	if (memtrace_init_debugfs())
296		goto out_unlock;
297
298	memtrace_size = val;
299	rc = 0;
300out_unlock:
301	mutex_unlock(&memtrace_mutex);
302	return rc;
303}
304
305static int memtrace_enable_get(void *data, u64 *val)
306{
307	*val = memtrace_size;
308	return 0;
309}
310
311DEFINE_SIMPLE_ATTRIBUTE(memtrace_init_fops, memtrace_enable_get,
312					memtrace_enable_set, "0x%016llx\n");
313
314static int memtrace_init(void)
315{
316	memtrace_debugfs_dir = debugfs_create_dir("memtrace",
317						  arch_debugfs_dir);
318
319	debugfs_create_file("enable", 0600, memtrace_debugfs_dir,
320			    NULL, &memtrace_init_fops);
321
322	return 0;
323}
324machine_device_initcall(powernv, memtrace_init);