bcm1250_tbprof.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571
  1. /*
  2. * This program is free software; you can redistribute it and/or
  3. * modify it under the terms of the GNU General Public License
  4. * as published by the Free Software Foundation; either version 2
  5. * of the License, or (at your option) any later version.
  6. *
  7. * This program is distributed in the hope that it will be useful,
  8. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  9. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  10. * GNU General Public License for more details.
  11. *
  12. * You should have received a copy of the GNU General Public License
  13. * along with this program; if not, write to the Free Software
  14. * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  15. *
  16. * Copyright (C) 2001, 2002, 2003 Broadcom Corporation
  17. * Copyright (C) 2007 Ralf Baechle <ralf@linux-mips.org>
  18. * Copyright (C) 2007 MIPS Technologies, Inc.
  19. * written by Ralf Baechle <ralf@linux-mips.org>
  20. */
  21. #undef DEBUG
  22. #include <linux/device.h>
  23. #include <linux/module.h>
  24. #include <linux/kernel.h>
  25. #include <linux/types.h>
  26. #include <linux/init.h>
  27. #include <linux/interrupt.h>
  28. #include <linux/slab.h>
  29. #include <linux/vmalloc.h>
  30. #include <linux/fs.h>
  31. #include <linux/errno.h>
  32. #include <linux/types.h>
  33. #include <linux/wait.h>
  34. #include <asm/io.h>
  35. #include <asm/sibyte/sb1250.h>
  36. #include <asm/sibyte/sb1250_regs.h>
  37. #include <asm/sibyte/sb1250_scd.h>
  38. #include <asm/sibyte/sb1250_int.h>
  39. #include <asm/system.h>
  40. #include <asm/uaccess.h>
  41. #define SBPROF_TB_MAJOR 240
  42. typedef u64 tb_sample_t[6*256];
  43. enum open_status {
  44. SB_CLOSED,
  45. SB_OPENING,
  46. SB_OPEN
  47. };
  48. struct sbprof_tb {
  49. wait_queue_head_t tb_sync;
  50. wait_queue_head_t tb_read;
  51. struct mutex lock;
  52. enum open_status open;
  53. tb_sample_t *sbprof_tbbuf;
  54. int next_tb_sample;
  55. volatile int tb_enable;
  56. volatile int tb_armed;
  57. };
  58. static struct sbprof_tb sbp;
  59. #define MAX_SAMPLE_BYTES (24*1024*1024)
  60. #define MAX_TBSAMPLE_BYTES (12*1024*1024)
  61. #define MAX_SAMPLES (MAX_SAMPLE_BYTES/sizeof(u_int32_t))
  62. #define TB_SAMPLE_SIZE (sizeof(tb_sample_t))
  63. #define MAX_TB_SAMPLES (MAX_TBSAMPLE_BYTES/TB_SAMPLE_SIZE)
  64. /* ioctls */
  65. #define SBPROF_ZBSTART _IOW('s', 0, int)
  66. #define SBPROF_ZBSTOP _IOW('s', 1, int)
  67. #define SBPROF_ZBWAITFULL _IOW('s', 2, int)
  68. /*
  69. * Routines for using 40-bit SCD cycle counter
  70. *
  71. * Client responsible for either handling interrupts or making sure
  72. * the cycles counter never saturates, e.g., by doing
  73. * zclk_timer_init(0) at least every 2^40 - 1 ZCLKs.
  74. */
  75. /*
  76. * Configures SCD counter 0 to count ZCLKs starting from val;
  77. * Configures SCD counters1,2,3 to count nothing.
  78. * Must not be called while gathering ZBbus profiles.
  79. */
  80. #define zclk_timer_init(val) \
  81. __asm__ __volatile__ (".set push;" \
  82. ".set mips64;" \
  83. "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \
  84. "sd %0, 0x10($8);" /* write val to counter0 */ \
  85. "sd %1, 0($8);" /* config counter0 for zclks*/ \
  86. ".set pop" \
  87. : /* no outputs */ \
  88. /* enable, counter0 */ \
  89. : /* inputs */ "r"(val), "r" ((1ULL << 33) | 1ULL) \
  90. : /* modifies */ "$8" )
  91. /* Reads SCD counter 0 and puts result in value
  92. unsigned long long val; */
  93. #define zclk_get(val) \
  94. __asm__ __volatile__ (".set push;" \
  95. ".set mips64;" \
  96. "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \
  97. "ld %0, 0x10($8);" /* write val to counter0 */ \
  98. ".set pop" \
  99. : /* outputs */ "=r"(val) \
  100. : /* inputs */ \
  101. : /* modifies */ "$8" )
  102. #define DEVNAME "bcm1250_tbprof"
  103. #define TB_FULL (sbp.next_tb_sample == MAX_TB_SAMPLES)
  104. /*
  105. * Support for ZBbus sampling using the trace buffer
  106. *
  107. * We use the SCD performance counter interrupt, caused by a Zclk counter
  108. * overflow, to trigger the start of tracing.
  109. *
  110. * We set the trace buffer to sample everything and freeze on
  111. * overflow.
  112. *
  113. * We map the interrupt for trace_buffer_freeze to handle it on CPU 0.
  114. */
  115. static u64 tb_period;
  116. static void arm_tb(void)
  117. {
  118. u64 scdperfcnt;
  119. u64 next = (1ULL << 40) - tb_period;
  120. u64 tb_options = M_SCD_TRACE_CFG_FREEZE_FULL;
  121. /*
  122. * Generate an SCD_PERFCNT interrupt in TB_PERIOD Zclks to trigger
  123. *start of trace. XXX vary sampling period
  124. */
  125. __raw_writeq(0, IOADDR(A_SCD_PERF_CNT_1));
  126. scdperfcnt = __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG));
  127. /*
  128. * Unfortunately, in Pass 2 we must clear all counters to knock down a
  129. * previous interrupt request. This means that bus profiling requires
  130. * ALL of the SCD perf counters.
  131. */
  132. __raw_writeq((scdperfcnt & ~M_SPC_CFG_SRC1) |
  133. /* keep counters 0,2,3 as is */
  134. M_SPC_CFG_ENABLE | /* enable counting */
  135. M_SPC_CFG_CLEAR | /* clear all counters */
  136. V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
  137. IOADDR(A_SCD_PERF_CNT_CFG));
  138. __raw_writeq(next, IOADDR(A_SCD_PERF_CNT_1));
  139. /* Reset the trace buffer */
  140. __raw_writeq(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG));
  141. #if 0 && defined(M_SCD_TRACE_CFG_FORCECNT)
  142. /* XXXKW may want to expose control to the data-collector */
  143. tb_options |= M_SCD_TRACE_CFG_FORCECNT;
  144. #endif
  145. __raw_writeq(tb_options, IOADDR(A_SCD_TRACE_CFG));
  146. sbp.tb_armed = 1;
  147. }
  148. static irqreturn_t sbprof_tb_intr(int irq, void *dev_id)
  149. {
  150. int i;
  151. pr_debug(DEVNAME ": tb_intr\n");
  152. if (sbp.next_tb_sample < MAX_TB_SAMPLES) {
  153. /* XXX should use XKPHYS to make writes bypass L2 */
  154. u64 *p = sbp.sbprof_tbbuf[sbp.next_tb_sample++];
  155. /* Read out trace */
  156. __raw_writeq(M_SCD_TRACE_CFG_START_READ,
  157. IOADDR(A_SCD_TRACE_CFG));
  158. __asm__ __volatile__ ("sync" : : : "memory");
  159. /* Loop runs backwards because bundles are read out in reverse order */
  160. for (i = 256 * 6; i > 0; i -= 6) {
  161. /* Subscripts decrease to put bundle in the order */
  162. /* t0 lo, t0 hi, t1 lo, t1 hi, t2 lo, t2 hi */
  163. p[i - 1] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
  164. /* read t2 hi */
  165. p[i - 2] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
  166. /* read t2 lo */
  167. p[i - 3] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
  168. /* read t1 hi */
  169. p[i - 4] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
  170. /* read t1 lo */
  171. p[i - 5] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
  172. /* read t0 hi */
  173. p[i - 6] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
  174. /* read t0 lo */
  175. }
  176. if (!sbp.tb_enable) {
  177. pr_debug(DEVNAME ": tb_intr shutdown\n");
  178. __raw_writeq(M_SCD_TRACE_CFG_RESET,
  179. IOADDR(A_SCD_TRACE_CFG));
  180. sbp.tb_armed = 0;
  181. wake_up(&sbp.tb_sync);
  182. } else {
  183. arm_tb(); /* knock down current interrupt and get another one later */
  184. }
  185. } else {
  186. /* No more trace buffer samples */
  187. pr_debug(DEVNAME ": tb_intr full\n");
  188. __raw_writeq(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG));
  189. sbp.tb_armed = 0;
  190. if (!sbp.tb_enable) {
  191. wake_up(&sbp.tb_sync);
  192. }
  193. wake_up(&sbp.tb_read);
  194. }
  195. return IRQ_HANDLED;
  196. }
  197. static irqreturn_t sbprof_pc_intr(int irq, void *dev_id)
  198. {
  199. printk(DEVNAME ": unexpected pc_intr");
  200. return IRQ_NONE;
  201. }
  202. /*
  203. * Requires: Already called zclk_timer_init with a value that won't
  204. * saturate 40 bits. No subsequent use of SCD performance counters
  205. * or trace buffer.
  206. */
  207. static int sbprof_zbprof_start(struct file *filp)
  208. {
  209. u64 scdperfcnt;
  210. int err;
  211. if (xchg(&sbp.tb_enable, 1))
  212. return -EBUSY;
  213. pr_debug(DEVNAME ": starting\n");
  214. sbp.next_tb_sample = 0;
  215. filp->f_pos = 0;
  216. err = request_irq(K_INT_TRACE_FREEZE, sbprof_tb_intr, 0,
  217. DEVNAME " trace freeze", &sbp);
  218. if (err)
  219. return -EBUSY;
  220. /* Make sure there isn't a perf-cnt interrupt waiting */
  221. scdperfcnt = __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG));
  222. /* Disable and clear counters, override SRC_1 */
  223. __raw_writeq((scdperfcnt & ~(M_SPC_CFG_SRC1 | M_SPC_CFG_ENABLE)) |
  224. M_SPC_CFG_ENABLE | M_SPC_CFG_CLEAR | V_SPC_CFG_SRC1(1),
  225. IOADDR(A_SCD_PERF_CNT_CFG));
  226. /*
  227. * We grab this interrupt to prevent others from trying to use it, even
  228. * though we don't want to service the interrupts (they only feed into
  229. * the trace-on-interrupt mechanism)
  230. */
  231. err = request_irq(K_INT_PERF_CNT, sbprof_pc_intr, 0,
  232. DEVNAME " scd perfcnt", &sbp);
  233. if (err)
  234. goto out_free_irq;
  235. /*
  236. * I need the core to mask these, but the interrupt mapper to pass them
  237. * through. I am exploiting my knowledge that cp0_status masks out
  238. * IP[5]. krw
  239. */
  240. __raw_writeq(K_INT_MAP_I3,
  241. IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_MAP_BASE) +
  242. (K_INT_PERF_CNT << 3)));
  243. /* Initialize address traps */
  244. __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_0));
  245. __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_1));
  246. __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_2));
  247. __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_3));
  248. __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_0));
  249. __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_1));
  250. __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_2));
  251. __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_3));
  252. __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_0));
  253. __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_1));
  254. __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_2));
  255. __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_3));
  256. /* Initialize Trace Event 0-7 */
  257. /* when interrupt */
  258. __raw_writeq(M_SCD_TREVT_INTERRUPT, IOADDR(A_SCD_TRACE_EVENT_0));
  259. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_1));
  260. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_2));
  261. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_3));
  262. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_4));
  263. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_5));
  264. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_6));
  265. __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_7));
  266. /* Initialize Trace Sequence 0-7 */
  267. /* Start on event 0 (interrupt) */
  268. __raw_writeq(V_SCD_TRSEQ_FUNC_START | 0x0fff,
  269. IOADDR(A_SCD_TRACE_SEQUENCE_0));
  270. /* dsamp when d used | asamp when a used */
  271. __raw_writeq(M_SCD_TRSEQ_ASAMPLE | M_SCD_TRSEQ_DSAMPLE |
  272. K_SCD_TRSEQ_TRIGGER_ALL,
  273. IOADDR(A_SCD_TRACE_SEQUENCE_1));
  274. __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_2));
  275. __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_3));
  276. __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_4));
  277. __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_5));
  278. __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_6));
  279. __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_7));
  280. /* Now indicate the PERF_CNT interrupt as a trace-relevant interrupt */
  281. __raw_writeq(1ULL << K_INT_PERF_CNT,
  282. IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_TRACE)));
  283. arm_tb();
  284. pr_debug(DEVNAME ": done starting\n");
  285. return 0;
  286. out_free_irq:
  287. free_irq(K_INT_TRACE_FREEZE, &sbp);
  288. return err;
  289. }
  290. static int sbprof_zbprof_stop(void)
  291. {
  292. int err;
  293. pr_debug(DEVNAME ": stopping\n");
  294. if (sbp.tb_enable) {
  295. /*
  296. * XXXKW there is a window here where the intr handler may run,
  297. * see the disable, and do the wake_up before this sleep
  298. * happens.
  299. */
  300. pr_debug(DEVNAME ": wait for disarm\n");
  301. err = wait_event_interruptible(sbp.tb_sync, !sbp.tb_armed);
  302. pr_debug(DEVNAME ": disarm complete, stat %d\n", err);
  303. if (err)
  304. return err;
  305. sbp.tb_enable = 0;
  306. free_irq(K_INT_TRACE_FREEZE, &sbp);
  307. free_irq(K_INT_PERF_CNT, &sbp);
  308. }
  309. pr_debug(DEVNAME ": done stopping\n");
  310. return 0;
  311. }
  312. static int sbprof_tb_open(struct inode *inode, struct file *filp)
  313. {
  314. int minor;
  315. minor = iminor(inode);
  316. if (minor != 0)
  317. return -ENODEV;
  318. if (xchg(&sbp.open, SB_OPENING) != SB_CLOSED)
  319. return -EBUSY;
  320. memset(&sbp, 0, sizeof(struct sbprof_tb));
  321. sbp.sbprof_tbbuf = vmalloc(MAX_TBSAMPLE_BYTES);
  322. if (!sbp.sbprof_tbbuf)
  323. return -ENOMEM;
  324. memset(sbp.sbprof_tbbuf, 0, MAX_TBSAMPLE_BYTES);
  325. init_waitqueue_head(&sbp.tb_sync);
  326. init_waitqueue_head(&sbp.tb_read);
  327. mutex_init(&sbp.lock);
  328. sbp.open = SB_OPEN;
  329. return 0;
  330. }
  331. static int sbprof_tb_release(struct inode *inode, struct file *filp)
  332. {
  333. int minor = iminor(inode);
  334. if (minor != 0 || !sbp.open)
  335. return -ENODEV;
  336. mutex_lock(&sbp.lock);
  337. if (sbp.tb_armed || sbp.tb_enable)
  338. sbprof_zbprof_stop();
  339. vfree(sbp.sbprof_tbbuf);
  340. sbp.open = 0;
  341. mutex_unlock(&sbp.lock);
  342. return 0;
  343. }
  344. static ssize_t sbprof_tb_read(struct file *filp, char *buf,
  345. size_t size, loff_t *offp)
  346. {
  347. int cur_sample, sample_off, cur_count, sample_left;
  348. long cur_off = *offp;
  349. char *dest = buf;
  350. int count = 0;
  351. char *src;
  352. if (!access_ok(VERIFY_WRITE, buf, size))
  353. return -EFAULT;
  354. mutex_lock(&sbp.lock);
  355. count = 0;
  356. cur_sample = cur_off / TB_SAMPLE_SIZE;
  357. sample_off = cur_off % TB_SAMPLE_SIZE;
  358. sample_left = TB_SAMPLE_SIZE - sample_off;
  359. while (size && (cur_sample < sbp.next_tb_sample)) {
  360. int err;
  361. cur_count = size < sample_left ? size : sample_left;
  362. src = (char *)(((long)sbp.sbprof_tbbuf[cur_sample])+sample_off);
  363. err = __copy_to_user(dest, src, cur_count);
  364. if (err) {
  365. *offp = cur_off + cur_count - err;
  366. mutex_unlock(&sbp.lock);
  367. return err;
  368. }
  369. pr_debug(DEVNAME ": read from sample %d, %d bytes\n",
  370. cur_sample, cur_count);
  371. size -= cur_count;
  372. sample_left -= cur_count;
  373. if (!sample_left) {
  374. cur_sample++;
  375. sample_off = 0;
  376. sample_left = TB_SAMPLE_SIZE;
  377. } else {
  378. sample_off += cur_count;
  379. }
  380. cur_off += cur_count;
  381. dest += cur_count;
  382. count += cur_count;
  383. }
  384. *offp = cur_off;
  385. mutex_unlock(&sbp.lock);
  386. return count;
  387. }
  388. static long sbprof_tb_ioctl(struct file *filp, unsigned int command,
  389. unsigned long arg)
  390. {
  391. int error = 0;
  392. switch (command) {
  393. case SBPROF_ZBSTART:
  394. mutex_lock(&sbp.lock);
  395. error = sbprof_zbprof_start(filp);
  396. mutex_unlock(&sbp.lock);
  397. break;
  398. case SBPROF_ZBSTOP:
  399. mutex_lock(&sbp.lock);
  400. error = sbprof_zbprof_stop();
  401. mutex_unlock(&sbp.lock);
  402. break;
  403. case SBPROF_ZBWAITFULL:
  404. error = wait_event_interruptible(sbp.tb_read, TB_FULL);
  405. if (error)
  406. break;
  407. error = put_user(TB_FULL, (int *) arg);
  408. break;
  409. default:
  410. error = -EINVAL;
  411. break;
  412. }
  413. return error;
  414. }
  415. static const struct file_operations sbprof_tb_fops = {
  416. .owner = THIS_MODULE,
  417. .open = sbprof_tb_open,
  418. .release = sbprof_tb_release,
  419. .read = sbprof_tb_read,
  420. .unlocked_ioctl = sbprof_tb_ioctl,
  421. .compat_ioctl = sbprof_tb_ioctl,
  422. .mmap = NULL,
  423. };
  424. static struct class *tb_class;
  425. static struct device *tb_dev;
  426. static int __init sbprof_tb_init(void)
  427. {
  428. struct device *dev;
  429. struct class *tbc;
  430. int err;
  431. if (register_chrdev(SBPROF_TB_MAJOR, DEVNAME, &sbprof_tb_fops)) {
  432. printk(KERN_WARNING DEVNAME ": initialization failed (dev %d)\n",
  433. SBPROF_TB_MAJOR);
  434. return -EIO;
  435. }
  436. tbc = class_create(THIS_MODULE, "sb_tracebuffer");
  437. if (IS_ERR(tbc)) {
  438. err = PTR_ERR(tbc);
  439. goto out_chrdev;
  440. }
  441. tb_class = tbc;
  442. dev = device_create(tbc, NULL, MKDEV(SBPROF_TB_MAJOR, 0), "tb");
  443. if (IS_ERR(dev)) {
  444. err = PTR_ERR(dev);
  445. goto out_class;
  446. }
  447. tb_dev = dev;
  448. sbp.open = 0;
  449. tb_period = zbbus_mhz * 10000LL;
  450. pr_info(DEVNAME ": initialized - tb_period = %lld\n", tb_period);
  451. return 0;
  452. out_class:
  453. class_destroy(tb_class);
  454. out_chrdev:
  455. unregister_chrdev(SBPROF_TB_MAJOR, DEVNAME);
  456. return err;
  457. }
  458. static void __exit sbprof_tb_cleanup(void)
  459. {
  460. device_destroy(tb_class, MKDEV(SBPROF_TB_MAJOR, 0));
  461. unregister_chrdev(SBPROF_TB_MAJOR, DEVNAME);
  462. class_destroy(tb_class);
  463. }
  464. module_init(sbprof_tb_init);
  465. module_exit(sbprof_tb_cleanup);
  466. MODULE_ALIAS_CHARDEV_MAJOR(SBPROF_TB_MAJOR);
  467. MODULE_AUTHOR("Ralf Baechle <ralf@linux-mips.org>");
  468. MODULE_LICENSE("GPL");