Full support for Ginger Console
[linux-ginger.git] / arch / mips / sibyte / common / sb_tbprof.c
blob15ea778b5e6664c0f305681b3c9a0c3c2fad3f2e
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.
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.
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.
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>
22 #undef DEBUG
24 #include <linux/device.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
27 #include <linux/types.h>
28 #include <linux/init.h>
29 #include <linux/interrupt.h>
30 #include <linux/slab.h>
31 #include <linux/smp_lock.h>
32 #include <linux/vmalloc.h>
33 #include <linux/fs.h>
34 #include <linux/errno.h>
35 #include <linux/wait.h>
36 #include <asm/io.h>
37 #include <asm/sibyte/sb1250.h>
39 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
40 #include <asm/sibyte/bcm1480_regs.h>
41 #include <asm/sibyte/bcm1480_scd.h>
42 #include <asm/sibyte/bcm1480_int.h>
43 #elif defined(CONFIG_SIBYTE_SB1250) || defined(CONFIG_SIBYTE_BCM112X)
44 #include <asm/sibyte/sb1250_regs.h>
45 #include <asm/sibyte/sb1250_scd.h>
46 #include <asm/sibyte/sb1250_int.h>
47 #else
48 #error invalid SiByte UART configuation
49 #endif
51 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
52 #undef K_INT_TRACE_FREEZE
53 #define K_INT_TRACE_FREEZE K_BCM1480_INT_TRACE_FREEZE
54 #undef K_INT_PERF_CNT
55 #define K_INT_PERF_CNT K_BCM1480_INT_PERF_CNT
56 #endif
58 #include <asm/system.h>
59 #include <asm/uaccess.h>
61 #define SBPROF_TB_MAJOR 240
63 typedef u64 tb_sample_t[6*256];
65 enum open_status {
66 SB_CLOSED,
67 SB_OPENING,
68 SB_OPEN
71 struct sbprof_tb {
72 wait_queue_head_t tb_sync;
73 wait_queue_head_t tb_read;
74 struct mutex lock;
75 enum open_status open;
76 tb_sample_t *sbprof_tbbuf;
77 int next_tb_sample;
79 volatile int tb_enable;
80 volatile int tb_armed;
84 static struct sbprof_tb sbp;
86 #define MAX_SAMPLE_BYTES (24*1024*1024)
87 #define MAX_TBSAMPLE_BYTES (12*1024*1024)
89 #define MAX_SAMPLES (MAX_SAMPLE_BYTES/sizeof(u_int32_t))
90 #define TB_SAMPLE_SIZE (sizeof(tb_sample_t))
91 #define MAX_TB_SAMPLES (MAX_TBSAMPLE_BYTES/TB_SAMPLE_SIZE)
93 /* ioctls */
94 #define SBPROF_ZBSTART _IOW('s', 0, int)
95 #define SBPROF_ZBSTOP _IOW('s', 1, int)
96 #define SBPROF_ZBWAITFULL _IOW('s', 2, int)
99 * Routines for using 40-bit SCD cycle counter
101 * Client responsible for either handling interrupts or making sure
102 * the cycles counter never saturates, e.g., by doing
103 * zclk_timer_init(0) at least every 2^40 - 1 ZCLKs.
107 * Configures SCD counter 0 to count ZCLKs starting from val;
108 * Configures SCD counters1,2,3 to count nothing.
109 * Must not be called while gathering ZBbus profiles.
112 #define zclk_timer_init(val) \
113 __asm__ __volatile__ (".set push;" \
114 ".set mips64;" \
115 "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \
116 "sd %0, 0x10($8);" /* write val to counter0 */ \
117 "sd %1, 0($8);" /* config counter0 for zclks*/ \
118 ".set pop" \
119 : /* no outputs */ \
120 /* enable, counter0 */ \
121 : /* inputs */ "r"(val), "r" ((1ULL << 33) | 1ULL) \
122 : /* modifies */ "$8" )
125 /* Reads SCD counter 0 and puts result in value
126 unsigned long long val; */
127 #define zclk_get(val) \
128 __asm__ __volatile__ (".set push;" \
129 ".set mips64;" \
130 "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \
131 "ld %0, 0x10($8);" /* write val to counter0 */ \
132 ".set pop" \
133 : /* outputs */ "=r"(val) \
134 : /* inputs */ \
135 : /* modifies */ "$8" )
137 #define DEVNAME "sb_tbprof"
139 #define TB_FULL (sbp.next_tb_sample == MAX_TB_SAMPLES)
142 * Support for ZBbus sampling using the trace buffer
144 * We use the SCD performance counter interrupt, caused by a Zclk counter
145 * overflow, to trigger the start of tracing.
147 * We set the trace buffer to sample everything and freeze on
148 * overflow.
150 * We map the interrupt for trace_buffer_freeze to handle it on CPU 0.
154 static u64 tb_period;
156 static void arm_tb(void)
158 u64 scdperfcnt;
159 u64 next = (1ULL << 40) - tb_period;
160 u64 tb_options = M_SCD_TRACE_CFG_FREEZE_FULL;
163 * Generate an SCD_PERFCNT interrupt in TB_PERIOD Zclks to
164 * trigger start of trace. XXX vary sampling period
166 __raw_writeq(0, IOADDR(A_SCD_PERF_CNT_1));
167 scdperfcnt = __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG));
170 * Unfortunately, in Pass 2 we must clear all counters to knock down
171 * a previous interrupt request. This means that bus profiling
172 * requires ALL of the SCD perf counters.
174 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
175 __raw_writeq((scdperfcnt & ~M_SPC_CFG_SRC1) |
176 /* keep counters 0,2,3,4,5,6,7 as is */
177 V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
178 IOADDR(A_BCM1480_SCD_PERF_CNT_CFG0));
179 __raw_writeq(
180 M_SPC_CFG_ENABLE | /* enable counting */
181 M_SPC_CFG_CLEAR | /* clear all counters */
182 V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
183 IOADDR(A_BCM1480_SCD_PERF_CNT_CFG1));
184 #else
185 __raw_writeq((scdperfcnt & ~M_SPC_CFG_SRC1) |
186 /* keep counters 0,2,3 as is */
187 M_SPC_CFG_ENABLE | /* enable counting */
188 M_SPC_CFG_CLEAR | /* clear all counters */
189 V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
190 IOADDR(A_SCD_PERF_CNT_CFG));
191 #endif
192 __raw_writeq(next, IOADDR(A_SCD_PERF_CNT_1));
193 /* Reset the trace buffer */
194 __raw_writeq(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG));
195 #if 0 && defined(M_SCD_TRACE_CFG_FORCECNT)
196 /* XXXKW may want to expose control to the data-collector */
197 tb_options |= M_SCD_TRACE_CFG_FORCECNT;
198 #endif
199 __raw_writeq(tb_options, IOADDR(A_SCD_TRACE_CFG));
200 sbp.tb_armed = 1;
203 static irqreturn_t sbprof_tb_intr(int irq, void *dev_id)
205 int i;
207 pr_debug(DEVNAME ": tb_intr\n");
209 if (sbp.next_tb_sample < MAX_TB_SAMPLES) {
210 /* XXX should use XKPHYS to make writes bypass L2 */
211 u64 *p = sbp.sbprof_tbbuf[sbp.next_tb_sample++];
212 /* Read out trace */
213 __raw_writeq(M_SCD_TRACE_CFG_START_READ,
214 IOADDR(A_SCD_TRACE_CFG));
215 __asm__ __volatile__ ("sync" : : : "memory");
216 /* Loop runs backwards because bundles are read out in reverse order */
217 for (i = 256 * 6; i > 0; i -= 6) {
218 /* Subscripts decrease to put bundle in the order */
219 /* t0 lo, t0 hi, t1 lo, t1 hi, t2 lo, t2 hi */
220 p[i - 1] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
221 /* read t2 hi */
222 p[i - 2] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
223 /* read t2 lo */
224 p[i - 3] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
225 /* read t1 hi */
226 p[i - 4] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
227 /* read t1 lo */
228 p[i - 5] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
229 /* read t0 hi */
230 p[i - 6] = __raw_readq(IOADDR(A_SCD_TRACE_READ));
231 /* read t0 lo */
233 if (!sbp.tb_enable) {
234 pr_debug(DEVNAME ": tb_intr shutdown\n");
235 __raw_writeq(M_SCD_TRACE_CFG_RESET,
236 IOADDR(A_SCD_TRACE_CFG));
237 sbp.tb_armed = 0;
238 wake_up_interruptible(&sbp.tb_sync);
239 } else {
240 /* knock down current interrupt and get another one later */
241 arm_tb();
243 } else {
244 /* No more trace buffer samples */
245 pr_debug(DEVNAME ": tb_intr full\n");
246 __raw_writeq(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG));
247 sbp.tb_armed = 0;
248 if (!sbp.tb_enable)
249 wake_up_interruptible(&sbp.tb_sync);
250 wake_up_interruptible(&sbp.tb_read);
252 return IRQ_HANDLED;
255 static irqreturn_t sbprof_pc_intr(int irq, void *dev_id)
257 printk(DEVNAME ": unexpected pc_intr");
258 return IRQ_NONE;
262 * Requires: Already called zclk_timer_init with a value that won't
263 * saturate 40 bits. No subsequent use of SCD performance counters
264 * or trace buffer.
267 static int sbprof_zbprof_start(struct file *filp)
269 u64 scdperfcnt;
270 int err;
272 if (xchg(&sbp.tb_enable, 1))
273 return -EBUSY;
275 pr_debug(DEVNAME ": starting\n");
277 sbp.next_tb_sample = 0;
278 filp->f_pos = 0;
280 err = request_irq(K_INT_TRACE_FREEZE, sbprof_tb_intr, 0,
281 DEVNAME " trace freeze", &sbp);
282 if (err)
283 return -EBUSY;
285 /* Make sure there isn't a perf-cnt interrupt waiting */
286 scdperfcnt = __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG));
287 /* Disable and clear counters, override SRC_1 */
288 __raw_writeq((scdperfcnt & ~(M_SPC_CFG_SRC1 | M_SPC_CFG_ENABLE)) |
289 M_SPC_CFG_ENABLE | M_SPC_CFG_CLEAR | V_SPC_CFG_SRC1(1),
290 IOADDR(A_SCD_PERF_CNT_CFG));
293 * We grab this interrupt to prevent others from trying to use
294 * it, even though we don't want to service the interrupts
295 * (they only feed into the trace-on-interrupt mechanism)
297 if (request_irq(K_INT_PERF_CNT, sbprof_pc_intr, 0, DEVNAME " scd perfcnt", &sbp)) {
298 free_irq(K_INT_TRACE_FREEZE, &sbp);
299 return -EBUSY;
303 * I need the core to mask these, but the interrupt mapper to
304 * pass them through. I am exploiting my knowledge that
305 * cp0_status masks out IP[5]. krw
307 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
308 __raw_writeq(K_BCM1480_INT_MAP_I3,
309 IOADDR(A_BCM1480_IMR_REGISTER(0, R_BCM1480_IMR_INTERRUPT_MAP_BASE_L) +
310 ((K_BCM1480_INT_PERF_CNT & 0x3f) << 3)));
311 #else
312 __raw_writeq(K_INT_MAP_I3,
313 IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_MAP_BASE) +
314 (K_INT_PERF_CNT << 3)));
315 #endif
317 /* Initialize address traps */
318 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_0));
319 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_1));
320 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_2));
321 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_3));
323 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_0));
324 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_1));
325 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_2));
326 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_3));
328 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_0));
329 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_1));
330 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_2));
331 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_3));
333 /* Initialize Trace Event 0-7 */
334 /* when interrupt */
335 __raw_writeq(M_SCD_TREVT_INTERRUPT, IOADDR(A_SCD_TRACE_EVENT_0));
336 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_1));
337 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_2));
338 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_3));
339 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_4));
340 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_5));
341 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_6));
342 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_7));
344 /* Initialize Trace Sequence 0-7 */
345 /* Start on event 0 (interrupt) */
346 __raw_writeq(V_SCD_TRSEQ_FUNC_START | 0x0fff,
347 IOADDR(A_SCD_TRACE_SEQUENCE_0));
348 /* dsamp when d used | asamp when a used */
349 __raw_writeq(M_SCD_TRSEQ_ASAMPLE | M_SCD_TRSEQ_DSAMPLE |
350 K_SCD_TRSEQ_TRIGGER_ALL,
351 IOADDR(A_SCD_TRACE_SEQUENCE_1));
352 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_2));
353 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_3));
354 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_4));
355 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_5));
356 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_6));
357 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_7));
359 /* Now indicate the PERF_CNT interrupt as a trace-relevant interrupt */
360 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
361 __raw_writeq(1ULL << (K_BCM1480_INT_PERF_CNT & 0x3f),
362 IOADDR(A_BCM1480_IMR_REGISTER(0, R_BCM1480_IMR_INTERRUPT_TRACE_L)));
363 #else
364 __raw_writeq(1ULL << K_INT_PERF_CNT,
365 IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_TRACE)));
366 #endif
367 arm_tb();
369 pr_debug(DEVNAME ": done starting\n");
371 return 0;
374 static int sbprof_zbprof_stop(void)
376 int err = 0;
378 pr_debug(DEVNAME ": stopping\n");
380 if (sbp.tb_enable) {
382 * XXXKW there is a window here where the intr handler may run,
383 * see the disable, and do the wake_up before this sleep
384 * happens.
386 pr_debug(DEVNAME ": wait for disarm\n");
387 err = wait_event_interruptible(sbp.tb_sync, !sbp.tb_armed);
388 pr_debug(DEVNAME ": disarm complete, stat %d\n", err);
390 if (err)
391 return err;
393 sbp.tb_enable = 0;
394 free_irq(K_INT_TRACE_FREEZE, &sbp);
395 free_irq(K_INT_PERF_CNT, &sbp);
398 pr_debug(DEVNAME ": done stopping\n");
400 return err;
403 static int sbprof_tb_open(struct inode *inode, struct file *filp)
405 int minor;
407 minor = iminor(inode);
408 if (minor != 0)
409 return -ENODEV;
411 if (xchg(&sbp.open, SB_OPENING) != SB_CLOSED)
412 return -EBUSY;
414 memset(&sbp, 0, sizeof(struct sbprof_tb));
415 sbp.sbprof_tbbuf = vmalloc(MAX_TBSAMPLE_BYTES);
416 if (!sbp.sbprof_tbbuf) {
417 sbp.open = SB_CLOSED;
418 wmb();
419 return -ENOMEM;
422 memset(sbp.sbprof_tbbuf, 0, MAX_TBSAMPLE_BYTES);
423 init_waitqueue_head(&sbp.tb_sync);
424 init_waitqueue_head(&sbp.tb_read);
425 mutex_init(&sbp.lock);
427 sbp.open = SB_OPEN;
428 wmb();
430 return 0;
433 static int sbprof_tb_release(struct inode *inode, struct file *filp)
435 int minor;
437 minor = iminor(inode);
438 if (minor != 0 || sbp.open != SB_CLOSED)
439 return -ENODEV;
441 mutex_lock(&sbp.lock);
443 if (sbp.tb_armed || sbp.tb_enable)
444 sbprof_zbprof_stop();
446 vfree(sbp.sbprof_tbbuf);
447 sbp.open = SB_CLOSED;
448 wmb();
450 mutex_unlock(&sbp.lock);
452 return 0;
455 static ssize_t sbprof_tb_read(struct file *filp, char *buf,
456 size_t size, loff_t *offp)
458 int cur_sample, sample_off, cur_count, sample_left;
459 char *src;
460 int count = 0;
461 char *dest = buf;
462 long cur_off = *offp;
464 if (!access_ok(VERIFY_WRITE, buf, size))
465 return -EFAULT;
467 mutex_lock(&sbp.lock);
469 count = 0;
470 cur_sample = cur_off / TB_SAMPLE_SIZE;
471 sample_off = cur_off % TB_SAMPLE_SIZE;
472 sample_left = TB_SAMPLE_SIZE - sample_off;
474 while (size && (cur_sample < sbp.next_tb_sample)) {
475 int err;
477 cur_count = size < sample_left ? size : sample_left;
478 src = (char *)(((long)sbp.sbprof_tbbuf[cur_sample])+sample_off);
479 err = __copy_to_user(dest, src, cur_count);
480 if (err) {
481 *offp = cur_off + cur_count - err;
482 mutex_unlock(&sbp.lock);
483 return err;
485 pr_debug(DEVNAME ": read from sample %d, %d bytes\n",
486 cur_sample, cur_count);
487 size -= cur_count;
488 sample_left -= cur_count;
489 if (!sample_left) {
490 cur_sample++;
491 sample_off = 0;
492 sample_left = TB_SAMPLE_SIZE;
493 } else {
494 sample_off += cur_count;
496 cur_off += cur_count;
497 dest += cur_count;
498 count += cur_count;
500 *offp = cur_off;
501 mutex_unlock(&sbp.lock);
503 return count;
506 static long sbprof_tb_ioctl(struct file *filp,
507 unsigned int command,
508 unsigned long arg)
510 int err = 0;
512 switch (command) {
513 case SBPROF_ZBSTART:
514 mutex_lock(&sbp.lock);
515 err = sbprof_zbprof_start(filp);
516 mutex_unlock(&sbp.lock);
517 break;
519 case SBPROF_ZBSTOP:
520 mutex_lock(&sbp.lock);
521 err = sbprof_zbprof_stop();
522 mutex_unlock(&sbp.lock);
523 break;
525 case SBPROF_ZBWAITFULL: {
526 err = wait_event_interruptible(sbp.tb_read, TB_FULL);
527 if (err)
528 break;
530 err = put_user(TB_FULL, (int *) arg);
531 break;
534 default:
535 err = -EINVAL;
536 break;
539 return err;
542 static const struct file_operations sbprof_tb_fops = {
543 .owner = THIS_MODULE,
544 .open = sbprof_tb_open,
545 .release = sbprof_tb_release,
546 .read = sbprof_tb_read,
547 .unlocked_ioctl = sbprof_tb_ioctl,
548 .compat_ioctl = sbprof_tb_ioctl,
549 .mmap = NULL,
552 static struct class *tb_class;
553 static struct device *tb_dev;
555 static int __init sbprof_tb_init(void)
557 struct device *dev;
558 struct class *tbc;
559 int err;
561 if (register_chrdev(SBPROF_TB_MAJOR, DEVNAME, &sbprof_tb_fops)) {
562 printk(KERN_WARNING DEVNAME ": initialization failed (dev %d)\n",
563 SBPROF_TB_MAJOR);
564 return -EIO;
567 tbc = class_create(THIS_MODULE, "sb_tracebuffer");
568 if (IS_ERR(tbc)) {
569 err = PTR_ERR(tbc);
570 goto out_chrdev;
573 tb_class = tbc;
575 dev = device_create(tbc, NULL, MKDEV(SBPROF_TB_MAJOR, 0), NULL, "tb");
576 if (IS_ERR(dev)) {
577 err = PTR_ERR(dev);
578 goto out_class;
580 tb_dev = dev;
582 sbp.open = SB_CLOSED;
583 wmb();
584 tb_period = zbbus_mhz * 10000LL;
585 pr_info(DEVNAME ": initialized - tb_period = %lld\n",
586 (long long) tb_period);
587 return 0;
589 out_class:
590 class_destroy(tb_class);
591 out_chrdev:
592 unregister_chrdev(SBPROF_TB_MAJOR, DEVNAME);
594 return err;
597 static void __exit sbprof_tb_cleanup(void)
599 device_destroy(tb_class, MKDEV(SBPROF_TB_MAJOR, 0));
600 unregister_chrdev(SBPROF_TB_MAJOR, DEVNAME);
601 class_destroy(tb_class);
604 module_init(sbprof_tb_init);
605 module_exit(sbprof_tb_cleanup);
607 MODULE_ALIAS_CHARDEV_MAJOR(SBPROF_TB_MAJOR);
608 MODULE_AUTHOR("Ralf Baechle <ralf@linux-mips.org>");
609 MODULE_LICENSE("GPL");