2 * Copyright (C) IBM Corporation, 2014, 2017
3 * Anton Blanchard, Rashmica Gupta.
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.
11 #define pr_fmt(fmt) "memtrace: " fmt
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>
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>
26 /* This enables us to keep track of the memory removed from each node. */
27 struct memtrace_entry {
36 static DEFINE_MUTEX(memtrace_mutex);
37 static u64 memtrace_size;
39 static struct memtrace_entry *memtrace_array;
40 static unsigned int memtrace_array_nr;
43 static ssize_t memtrace_read(struct file *filp, char __user *ubuf,
44 size_t count, loff_t *ppos)
46 struct memtrace_entry *ent = filp->private_data;
48 return simple_read_from_buffer(ubuf, count, ppos, ent->mem, ent->size);
51 static const struct file_operations memtrace_fops = {
52 .llseek = default_llseek,
53 .read = memtrace_read,
57 static int check_memblock_online(struct memory_block *mem, void *arg)
59 if (mem->state != MEM_ONLINE)
65 static int change_memblock_state(struct memory_block *mem, void *arg)
67 unsigned long state = (unsigned long)arg;
74 static void memtrace_clear_range(unsigned long start_pfn,
75 unsigned long nr_pages)
80 * As pages are offline, we cannot trust the memmap anymore. As HIGHMEM
81 * does not apply, avoid passing around "struct page" and use
82 * clear_page() instead directly.
84 for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++) {
85 if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
87 clear_page(__va(PFN_PHYS(pfn)));
91 /* called with device_hotplug_lock held */
92 static bool memtrace_offline_pages(u32 nid, u64 start_pfn, u64 nr_pages)
94 u64 end_pfn = start_pfn + nr_pages - 1;
96 if (walk_memory_range(start_pfn, end_pfn, NULL,
97 check_memblock_online))
100 walk_memory_range(start_pfn, end_pfn, (void *)MEM_GOING_OFFLINE,
101 change_memblock_state);
103 if (offline_pages(start_pfn, nr_pages)) {
104 walk_memory_range(start_pfn, end_pfn, (void *)MEM_ONLINE,
105 change_memblock_state);
109 walk_memory_range(start_pfn, end_pfn, (void *)MEM_OFFLINE,
110 change_memblock_state);
116 static u64 memtrace_alloc_node(u32 nid, u64 size)
118 u64 start_pfn, end_pfn, nr_pages, pfn;
120 u64 bytes = memory_block_size_bytes();
122 if (!node_spanned_pages(nid))
125 start_pfn = node_start_pfn(nid);
126 end_pfn = node_end_pfn(nid);
127 nr_pages = size >> PAGE_SHIFT;
129 /* Trace memory needs to be aligned to the size */
130 end_pfn = round_down(end_pfn - nr_pages, nr_pages);
132 lock_device_hotplug();
133 for (base_pfn = end_pfn; base_pfn > start_pfn; base_pfn -= nr_pages) {
134 if (memtrace_offline_pages(nid, base_pfn, nr_pages) == true) {
136 * Clear the range while we still have a linear
139 memtrace_clear_range(base_pfn, nr_pages);
141 * Remove memory in memory block size chunks so that
142 * iomem resources are always split to the same size and
143 * we never try to remove memory that spans two iomem
146 end_pfn = base_pfn + nr_pages;
147 for (pfn = base_pfn; pfn < end_pfn; pfn += bytes>> PAGE_SHIFT) {
148 __remove_memory(nid, pfn << PAGE_SHIFT, bytes);
150 unlock_device_hotplug();
151 return base_pfn << PAGE_SHIFT;
154 unlock_device_hotplug();
159 static int memtrace_init_regions_runtime(u64 size)
164 memtrace_array = kcalloc(num_online_nodes(),
165 sizeof(struct memtrace_entry), GFP_KERNEL);
166 if (!memtrace_array) {
167 pr_err("Failed to allocate memtrace_array\n");
171 for_each_online_node(nid) {
172 m = memtrace_alloc_node(nid, size);
175 * A node might not have any local memory, so warn but
179 pr_err("Failed to allocate trace memory on node %d\n", nid);
183 pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m);
185 memtrace_array[memtrace_array_nr].start = m;
186 memtrace_array[memtrace_array_nr].size = size;
187 memtrace_array[memtrace_array_nr].nid = nid;
194 static struct dentry *memtrace_debugfs_dir;
196 static int memtrace_init_debugfs(void)
201 for (i = 0; i < memtrace_array_nr; i++) {
203 struct memtrace_entry *ent = &memtrace_array[i];
205 ent->mem = ioremap(ent->start, ent->size);
206 /* Warn but continue on */
208 pr_err("Failed to map trace memory at 0x%llx\n",
214 snprintf(ent->name, 16, "%08x", ent->nid);
215 dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir);
217 pr_err("Failed to create debugfs directory for node %d\n",
223 debugfs_create_file("trace", 0400, dir, ent, &memtrace_fops);
224 debugfs_create_x64("start", 0400, dir, &ent->start);
225 debugfs_create_x64("size", 0400, dir, &ent->size);
231 static int online_mem_block(struct memory_block *mem, void *arg)
233 return device_online(&mem->dev);
237 * Iterate through the chunks of memory we have removed from the kernel
238 * and attempt to add them back to the kernel.
240 static int memtrace_online(void)
243 struct memtrace_entry *ent;
245 for (i = memtrace_array_nr - 1; i >= 0; i--) {
246 ent = &memtrace_array[i];
248 /* We have onlined this chunk previously */
252 /* Remove from io mappings */
258 if (add_memory(ent->nid, ent->start, ent->size)) {
259 pr_err("Failed to add trace memory to node %d\n",
266 * If kernel isn't compiled with the auto online option
267 * we need to online the memory ourselves.
269 if (!memhp_auto_online) {
270 lock_device_hotplug();
271 walk_memory_range(PFN_DOWN(ent->start),
272 PFN_UP(ent->start + ent->size - 1),
273 NULL, online_mem_block);
274 unlock_device_hotplug();
278 * Memory was added successfully so clean up references to it
279 * so on reentry we can tell that this chunk was added.
281 debugfs_remove_recursive(ent->dir);
282 pr_info("Added trace memory back to node %d\n", ent->nid);
283 ent->size = ent->start = ent->nid = -1;
288 /* If all chunks of memory were added successfully, reset globals */
289 kfree(memtrace_array);
290 memtrace_array = NULL;
292 memtrace_array_nr = 0;
296 static int memtrace_enable_set(void *data, u64 val)
302 * Don't attempt to do anything if size isn't aligned to a memory
303 * block or equal to zero.
305 bytes = memory_block_size_bytes();
306 if (val & (bytes - 1)) {
307 pr_err("Value must be aligned with 0x%llx\n", bytes);
311 mutex_lock(&memtrace_mutex);
313 /* Re-add/online previously removed/offlined memory */
315 if (memtrace_online())
324 /* Offline and remove memory */
325 if (memtrace_init_regions_runtime(val))
328 if (memtrace_init_debugfs())
334 mutex_unlock(&memtrace_mutex);
338 static int memtrace_enable_get(void *data, u64 *val)
340 *val = memtrace_size;
344 DEFINE_SIMPLE_ATTRIBUTE(memtrace_init_fops, memtrace_enable_get,
345 memtrace_enable_set, "0x%016llx\n");
347 static int memtrace_init(void)
349 memtrace_debugfs_dir = debugfs_create_dir("memtrace",
350 powerpc_debugfs_root);
351 if (!memtrace_debugfs_dir)
354 debugfs_create_file("enable", 0600, memtrace_debugfs_dir,
355 NULL, &memtrace_init_fops);
359 machine_device_initcall(powernv, memtrace_init);