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 u64 memtrace_size;
38 static struct memtrace_entry *memtrace_array;
39 static unsigned int memtrace_array_nr;
42 static ssize_t memtrace_read(struct file *filp, char __user *ubuf,
43 size_t count, loff_t *ppos)
45 struct memtrace_entry *ent = filp->private_data;
47 return simple_read_from_buffer(ubuf, count, ppos, ent->mem, ent->size);
50 static const struct file_operations memtrace_fops = {
51 .llseek = default_llseek,
52 .read = memtrace_read,
56 static int check_memblock_online(struct memory_block *mem, void *arg)
58 if (mem->state != MEM_ONLINE)
64 static int change_memblock_state(struct memory_block *mem, void *arg)
66 unsigned long state = (unsigned long)arg;
73 /* called with device_hotplug_lock held */
74 static bool memtrace_offline_pages(u32 nid, u64 start_pfn, u64 nr_pages)
76 u64 end_pfn = start_pfn + nr_pages - 1;
78 if (walk_memory_range(start_pfn, end_pfn, NULL,
79 check_memblock_online))
82 walk_memory_range(start_pfn, end_pfn, (void *)MEM_GOING_OFFLINE,
83 change_memblock_state);
85 if (offline_pages(start_pfn, nr_pages)) {
86 walk_memory_range(start_pfn, end_pfn, (void *)MEM_ONLINE,
87 change_memblock_state);
91 walk_memory_range(start_pfn, end_pfn, (void *)MEM_OFFLINE,
92 change_memblock_state);
98 static u64 memtrace_alloc_node(u32 nid, u64 size)
100 u64 start_pfn, end_pfn, nr_pages, pfn;
102 u64 bytes = memory_block_size_bytes();
104 if (!node_spanned_pages(nid))
107 start_pfn = node_start_pfn(nid);
108 end_pfn = node_end_pfn(nid);
109 nr_pages = size >> PAGE_SHIFT;
111 /* Trace memory needs to be aligned to the size */
112 end_pfn = round_down(end_pfn - nr_pages, nr_pages);
114 lock_device_hotplug();
115 for (base_pfn = end_pfn; base_pfn > start_pfn; base_pfn -= nr_pages) {
116 if (memtrace_offline_pages(nid, base_pfn, nr_pages) == true) {
118 * Remove memory in memory block size chunks so that
119 * iomem resources are always split to the same size and
120 * we never try to remove memory that spans two iomem
123 end_pfn = base_pfn + nr_pages;
124 for (pfn = base_pfn; pfn < end_pfn; pfn += bytes>> PAGE_SHIFT) {
125 __remove_memory(nid, pfn << PAGE_SHIFT, bytes);
127 unlock_device_hotplug();
128 return base_pfn << PAGE_SHIFT;
131 unlock_device_hotplug();
136 static int memtrace_init_regions_runtime(u64 size)
141 memtrace_array = kcalloc(num_online_nodes(),
142 sizeof(struct memtrace_entry), GFP_KERNEL);
143 if (!memtrace_array) {
144 pr_err("Failed to allocate memtrace_array\n");
148 for_each_online_node(nid) {
149 m = memtrace_alloc_node(nid, size);
152 * A node might not have any local memory, so warn but
156 pr_err("Failed to allocate trace memory on node %d\n", nid);
160 pr_info("Allocated trace memory on node %d at 0x%016llx\n", nid, m);
162 memtrace_array[memtrace_array_nr].start = m;
163 memtrace_array[memtrace_array_nr].size = size;
164 memtrace_array[memtrace_array_nr].nid = nid;
171 static struct dentry *memtrace_debugfs_dir;
173 static int memtrace_init_debugfs(void)
178 for (i = 0; i < memtrace_array_nr; i++) {
180 struct memtrace_entry *ent = &memtrace_array[i];
182 ent->mem = ioremap(ent->start, ent->size);
183 /* Warn but continue on */
185 pr_err("Failed to map trace memory at 0x%llx\n",
191 snprintf(ent->name, 16, "%08x", ent->nid);
192 dir = debugfs_create_dir(ent->name, memtrace_debugfs_dir);
194 pr_err("Failed to create debugfs directory for node %d\n",
200 debugfs_create_file("trace", 0400, dir, ent, &memtrace_fops);
201 debugfs_create_x64("start", 0400, dir, &ent->start);
202 debugfs_create_x64("size", 0400, dir, &ent->size);
208 static int online_mem_block(struct memory_block *mem, void *arg)
210 return device_online(&mem->dev);
214 * Iterate through the chunks of memory we have removed from the kernel
215 * and attempt to add them back to the kernel.
217 static int memtrace_online(void)
220 struct memtrace_entry *ent;
222 for (i = memtrace_array_nr - 1; i >= 0; i--) {
223 ent = &memtrace_array[i];
225 /* We have onlined this chunk previously */
229 /* Remove from io mappings */
235 if (add_memory(ent->nid, ent->start, ent->size)) {
236 pr_err("Failed to add trace memory to node %d\n",
243 * If kernel isn't compiled with the auto online option
244 * we need to online the memory ourselves.
246 if (!memhp_auto_online) {
247 lock_device_hotplug();
248 walk_memory_range(PFN_DOWN(ent->start),
249 PFN_UP(ent->start + ent->size - 1),
250 NULL, online_mem_block);
251 unlock_device_hotplug();
255 * Memory was added successfully so clean up references to it
256 * so on reentry we can tell that this chunk was added.
258 debugfs_remove_recursive(ent->dir);
259 pr_info("Added trace memory back to node %d\n", ent->nid);
260 ent->size = ent->start = ent->nid = -1;
265 /* If all chunks of memory were added successfully, reset globals */
266 kfree(memtrace_array);
267 memtrace_array = NULL;
269 memtrace_array_nr = 0;
273 static int memtrace_enable_set(void *data, u64 val)
278 * Don't attempt to do anything if size isn't aligned to a memory
279 * block or equal to zero.
281 bytes = memory_block_size_bytes();
282 if (val & (bytes - 1)) {
283 pr_err("Value must be aligned with 0x%llx\n", bytes);
287 /* Re-add/online previously removed/offlined memory */
289 if (memtrace_online())
296 /* Offline and remove memory */
297 if (memtrace_init_regions_runtime(val))
300 if (memtrace_init_debugfs())
308 static int memtrace_enable_get(void *data, u64 *val)
310 *val = memtrace_size;
314 DEFINE_SIMPLE_ATTRIBUTE(memtrace_init_fops, memtrace_enable_get,
315 memtrace_enable_set, "0x%016llx\n");
317 static int memtrace_init(void)
319 memtrace_debugfs_dir = debugfs_create_dir("memtrace",
320 powerpc_debugfs_root);
321 if (!memtrace_debugfs_dir)
324 debugfs_create_file("enable", 0600, memtrace_debugfs_dir,
325 NULL, &memtrace_init_fops);
329 machine_device_initcall(powernv, memtrace_init);