1 /* 2 * Virtual Processor Dispatch Trace Log 3 * 4 * (C) Copyright IBM Corporation 2009 5 * 6 * Author: Jeremy Kerr <jk@ozlabs.org> 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2, or (at your option) 11 * any later version. 12 * 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public License 19 * along with this program; if not, write to the Free Software 20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 21 */ 22 23 #include <linux/init.h> 24 #include <linux/slab.h> 25 #include <linux/debugfs.h> 26 #include <linux/spinlock.h> 27 #include <asm/smp.h> 28 #include <asm/uaccess.h> 29 #include <asm/firmware.h> 30 #include <asm/lppaca.h> 31 #include <asm/debug.h> 32 33 #include "plpar_wrappers.h" 34 35 struct dtl { 36 struct dtl_entry *buf; 37 struct dentry *file; 38 int cpu; 39 int buf_entries; 40 u64 last_idx; 41 spinlock_t lock; 42 }; 43 static DEFINE_PER_CPU(struct dtl, cpu_dtl); 44 45 /* 46 * Dispatch trace log event mask: 47 * 0x7: 0x1: voluntary virtual processor waits 48 * 0x2: time-slice preempts 49 * 0x4: virtual partition memory page faults 50 */ 51 static u8 dtl_event_mask = 0x7; 52 53 54 /* 55 * Size of per-cpu log buffers. Firmware requires that the buffer does 56 * not cross a 4k boundary. 57 */ 58 static int dtl_buf_entries = N_DISPATCH_LOG; 59 60 #ifdef CONFIG_VIRT_CPU_ACCOUNTING 61 struct dtl_ring { 62 u64 write_index; 63 struct dtl_entry *write_ptr; 64 struct dtl_entry *buf; 65 struct dtl_entry *buf_end; 66 u8 saved_dtl_mask; 67 }; 68 69 static DEFINE_PER_CPU(struct dtl_ring, dtl_rings); 70 71 static atomic_t dtl_count; 72 73 /* 74 * The cpu accounting code controls the DTL ring buffer, and we get 75 * given entries as they are processed. 76 */ 77 static void consume_dtle(struct dtl_entry *dtle, u64 index) 78 { 79 struct dtl_ring *dtlr = &__get_cpu_var(dtl_rings); 80 struct dtl_entry *wp = dtlr->write_ptr; 81 struct lppaca *vpa = local_paca->lppaca_ptr; 82 83 if (!wp) 84 return; 85 86 *wp = *dtle; 87 barrier(); 88 89 /* check for hypervisor ring buffer overflow, ignore this entry if so */ 90 if (index + N_DISPATCH_LOG < vpa->dtl_idx) 91 return; 92 93 ++wp; 94 if (wp == dtlr->buf_end) 95 wp = dtlr->buf; 96 dtlr->write_ptr = wp; 97 98 /* incrementing write_index makes the new entry visible */ 99 smp_wmb(); 100 ++dtlr->write_index; 101 } 102 103 static int dtl_start(struct dtl *dtl) 104 { 105 struct dtl_ring *dtlr = &per_cpu(dtl_rings, dtl->cpu); 106 107 dtlr->buf = dtl->buf; 108 dtlr->buf_end = dtl->buf + dtl->buf_entries; 109 dtlr->write_index = 0; 110 111 /* setting write_ptr enables logging into our buffer */ 112 smp_wmb(); 113 dtlr->write_ptr = dtl->buf; 114 115 /* enable event logging */ 116 dtlr->saved_dtl_mask = lppaca_of(dtl->cpu).dtl_enable_mask; 117 lppaca_of(dtl->cpu).dtl_enable_mask |= dtl_event_mask; 118 119 dtl_consumer = consume_dtle; 120 atomic_inc(&dtl_count); 121 return 0; 122 } 123 124 static void dtl_stop(struct dtl *dtl) 125 { 126 struct dtl_ring *dtlr = &per_cpu(dtl_rings, dtl->cpu); 127 128 dtlr->write_ptr = NULL; 129 smp_wmb(); 130 131 dtlr->buf = NULL; 132 133 /* restore dtl_enable_mask */ 134 lppaca_of(dtl->cpu).dtl_enable_mask = dtlr->saved_dtl_mask; 135 136 if (atomic_dec_and_test(&dtl_count)) 137 dtl_consumer = NULL; 138 } 139 140 static u64 dtl_current_index(struct dtl *dtl) 141 { 142 return per_cpu(dtl_rings, dtl->cpu).write_index; 143 } 144 145 #else /* CONFIG_VIRT_CPU_ACCOUNTING */ 146 147 static int dtl_start(struct dtl *dtl) 148 { 149 unsigned long addr; 150 int ret, hwcpu; 151 152 /* Register our dtl buffer with the hypervisor. The HV expects the 153 * buffer size to be passed in the second word of the buffer */ 154 ((u32 *)dtl->buf)[1] = DISPATCH_LOG_BYTES; 155 156 hwcpu = get_hard_smp_processor_id(dtl->cpu); 157 addr = __pa(dtl->buf); 158 ret = register_dtl(hwcpu, addr); 159 if (ret) { 160 printk(KERN_WARNING "%s: DTL registration for cpu %d (hw %d) " 161 "failed with %d\n", __func__, dtl->cpu, hwcpu, ret); 162 return -EIO; 163 } 164 165 /* set our initial buffer indices */ 166 lppaca_of(dtl->cpu).dtl_idx = 0; 167 168 /* ensure that our updates to the lppaca fields have occurred before 169 * we actually enable the logging */ 170 smp_wmb(); 171 172 /* enable event logging */ 173 lppaca_of(dtl->cpu).dtl_enable_mask = dtl_event_mask; 174 175 return 0; 176 } 177 178 static void dtl_stop(struct dtl *dtl) 179 { 180 int hwcpu = get_hard_smp_processor_id(dtl->cpu); 181 182 lppaca_of(dtl->cpu).dtl_enable_mask = 0x0; 183 184 unregister_dtl(hwcpu); 185 } 186 187 static u64 dtl_current_index(struct dtl *dtl) 188 { 189 return lppaca_of(dtl->cpu).dtl_idx; 190 } 191 #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ 192 193 static int dtl_enable(struct dtl *dtl) 194 { 195 long int n_entries; 196 long int rc; 197 struct dtl_entry *buf = NULL; 198 199 if (!dtl_cache) 200 return -ENOMEM; 201 202 /* only allow one reader */ 203 if (dtl->buf) 204 return -EBUSY; 205 206 n_entries = dtl_buf_entries; 207 buf = kmem_cache_alloc_node(dtl_cache, GFP_KERNEL, cpu_to_node(dtl->cpu)); 208 if (!buf) { 209 printk(KERN_WARNING "%s: buffer alloc failed for cpu %d\n", 210 __func__, dtl->cpu); 211 return -ENOMEM; 212 } 213 214 spin_lock(&dtl->lock); 215 rc = -EBUSY; 216 if (!dtl->buf) { 217 /* store the original allocation size for use during read */ 218 dtl->buf_entries = n_entries; 219 dtl->buf = buf; 220 dtl->last_idx = 0; 221 rc = dtl_start(dtl); 222 if (rc) 223 dtl->buf = NULL; 224 } 225 spin_unlock(&dtl->lock); 226 227 if (rc) 228 kmem_cache_free(dtl_cache, buf); 229 return rc; 230 } 231 232 static void dtl_disable(struct dtl *dtl) 233 { 234 spin_lock(&dtl->lock); 235 dtl_stop(dtl); 236 kmem_cache_free(dtl_cache, dtl->buf); 237 dtl->buf = NULL; 238 dtl->buf_entries = 0; 239 spin_unlock(&dtl->lock); 240 } 241 242 /* file interface */ 243 244 static int dtl_file_open(struct inode *inode, struct file *filp) 245 { 246 struct dtl *dtl = inode->i_private; 247 int rc; 248 249 rc = dtl_enable(dtl); 250 if (rc) 251 return rc; 252 253 filp->private_data = dtl; 254 return 0; 255 } 256 257 static int dtl_file_release(struct inode *inode, struct file *filp) 258 { 259 struct dtl *dtl = inode->i_private; 260 dtl_disable(dtl); 261 return 0; 262 } 263 264 static ssize_t dtl_file_read(struct file *filp, char __user *buf, size_t len, 265 loff_t *pos) 266 { 267 long int rc, n_read, n_req, read_size; 268 struct dtl *dtl; 269 u64 cur_idx, last_idx, i; 270 271 if ((len % sizeof(struct dtl_entry)) != 0) 272 return -EINVAL; 273 274 dtl = filp->private_data; 275 276 /* requested number of entries to read */ 277 n_req = len / sizeof(struct dtl_entry); 278 279 /* actual number of entries read */ 280 n_read = 0; 281 282 spin_lock(&dtl->lock); 283 284 cur_idx = dtl_current_index(dtl); 285 last_idx = dtl->last_idx; 286 287 if (last_idx + dtl->buf_entries <= cur_idx) 288 last_idx = cur_idx - dtl->buf_entries + 1; 289 290 if (last_idx + n_req > cur_idx) 291 n_req = cur_idx - last_idx; 292 293 if (n_req > 0) 294 dtl->last_idx = last_idx + n_req; 295 296 spin_unlock(&dtl->lock); 297 298 if (n_req <= 0) 299 return 0; 300 301 i = last_idx % dtl->buf_entries; 302 303 /* read the tail of the buffer if we've wrapped */ 304 if (i + n_req > dtl->buf_entries) { 305 read_size = dtl->buf_entries - i; 306 307 rc = copy_to_user(buf, &dtl->buf[i], 308 read_size * sizeof(struct dtl_entry)); 309 if (rc) 310 return -EFAULT; 311 312 i = 0; 313 n_req -= read_size; 314 n_read += read_size; 315 buf += read_size * sizeof(struct dtl_entry); 316 } 317 318 /* .. and now the head */ 319 rc = copy_to_user(buf, &dtl->buf[i], n_req * sizeof(struct dtl_entry)); 320 if (rc) 321 return -EFAULT; 322 323 n_read += n_req; 324 325 return n_read * sizeof(struct dtl_entry); 326 } 327 328 static const struct file_operations dtl_fops = { 329 .open = dtl_file_open, 330 .release = dtl_file_release, 331 .read = dtl_file_read, 332 .llseek = no_llseek, 333 }; 334 335 static struct dentry *dtl_dir; 336 337 static int dtl_setup_file(struct dtl *dtl) 338 { 339 char name[10]; 340 341 sprintf(name, "cpu-%d", dtl->cpu); 342 343 dtl->file = debugfs_create_file(name, 0400, dtl_dir, dtl, &dtl_fops); 344 if (!dtl->file) 345 return -ENOMEM; 346 347 return 0; 348 } 349 350 static int dtl_init(void) 351 { 352 struct dentry *event_mask_file, *buf_entries_file; 353 int rc, i; 354 355 if (!firmware_has_feature(FW_FEATURE_SPLPAR)) 356 return -ENODEV; 357 358 /* set up common debugfs structure */ 359 360 rc = -ENOMEM; 361 dtl_dir = debugfs_create_dir("dtl", powerpc_debugfs_root); 362 if (!dtl_dir) { 363 printk(KERN_WARNING "%s: can't create dtl root dir\n", 364 __func__); 365 goto err; 366 } 367 368 event_mask_file = debugfs_create_x8("dtl_event_mask", 0600, 369 dtl_dir, &dtl_event_mask); 370 buf_entries_file = debugfs_create_u32("dtl_buf_entries", 0400, 371 dtl_dir, &dtl_buf_entries); 372 373 if (!event_mask_file || !buf_entries_file) { 374 printk(KERN_WARNING "%s: can't create dtl files\n", __func__); 375 goto err_remove_dir; 376 } 377 378 /* set up the per-cpu log structures */ 379 for_each_possible_cpu(i) { 380 struct dtl *dtl = &per_cpu(cpu_dtl, i); 381 spin_lock_init(&dtl->lock); 382 dtl->cpu = i; 383 384 rc = dtl_setup_file(dtl); 385 if (rc) 386 goto err_remove_dir; 387 } 388 389 return 0; 390 391 err_remove_dir: 392 debugfs_remove_recursive(dtl_dir); 393 err: 394 return rc; 395 } 396 arch_initcall(dtl_init); 397