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>
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/sched.h>
31 #include <linux/vmalloc.h>
33 #include <linux/errno.h>
34 #include <linux/wait.h>
36 #include <asm/sibyte/sb1250.h>
38 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
39 #include <asm/sibyte/bcm1480_regs.h>
40 #include <asm/sibyte/bcm1480_scd.h>
41 #include <asm/sibyte/bcm1480_int.h>
42 #elif defined(CONFIG_SIBYTE_SB1250) || defined(CONFIG_SIBYTE_BCM112X)
43 #include <asm/sibyte/sb1250_regs.h>
44 #include <asm/sibyte/sb1250_scd.h>
45 #include <asm/sibyte/sb1250_int.h>
47 #error invalid SiByte UART configuration
50 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
51 #undef K_INT_TRACE_FREEZE
52 #define K_INT_TRACE_FREEZE K_BCM1480_INT_TRACE_FREEZE
54 #define K_INT_PERF_CNT K_BCM1480_INT_PERF_CNT
57 #include <asm/uaccess.h>
59 #define SBPROF_TB_MAJOR 240
61 typedef u64 tb_sample_t
[6*256];
70 wait_queue_head_t tb_sync
;
71 wait_queue_head_t tb_read
;
73 enum open_status open
;
74 tb_sample_t
*sbprof_tbbuf
;
77 volatile int tb_enable
;
78 volatile int tb_armed
;
82 static struct sbprof_tb sbp
;
84 #define MAX_SAMPLE_BYTES (24*1024*1024)
85 #define MAX_TBSAMPLE_BYTES (12*1024*1024)
87 #define MAX_SAMPLES (MAX_SAMPLE_BYTES/sizeof(u_int32_t))
88 #define TB_SAMPLE_SIZE (sizeof(tb_sample_t))
89 #define MAX_TB_SAMPLES (MAX_TBSAMPLE_BYTES/TB_SAMPLE_SIZE)
92 #define SBPROF_ZBSTART _IOW('s', 0, int)
93 #define SBPROF_ZBSTOP _IOW('s', 1, int)
94 #define SBPROF_ZBWAITFULL _IOW('s', 2, int)
97 * Routines for using 40-bit SCD cycle counter
99 * Client responsible for either handling interrupts or making sure
100 * the cycles counter never saturates, e.g., by doing
101 * zclk_timer_init(0) at least every 2^40 - 1 ZCLKs.
105 * Configures SCD counter 0 to count ZCLKs starting from val;
106 * Configures SCD counters1,2,3 to count nothing.
107 * Must not be called while gathering ZBbus profiles.
110 #define zclk_timer_init(val) \
111 __asm__ __volatile__ (".set push;" \
113 "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \
114 "sd %0, 0x10($8);" /* write val to counter0 */ \
115 "sd %1, 0($8);" /* config counter0 for zclks*/ \
118 /* enable, counter0 */ \
119 : /* inputs */ "r"(val), "r" ((1ULL << 33) | 1ULL) \
120 : /* modifies */ "$8" )
123 /* Reads SCD counter 0 and puts result in value
124 unsigned long long val; */
125 #define zclk_get(val) \
126 __asm__ __volatile__ (".set push;" \
128 "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \
129 "ld %0, 0x10($8);" /* write val to counter0 */ \
131 : /* outputs */ "=r"(val) \
133 : /* modifies */ "$8" )
135 #define DEVNAME "sb_tbprof"
137 #define TB_FULL (sbp.next_tb_sample == MAX_TB_SAMPLES)
140 * Support for ZBbus sampling using the trace buffer
142 * We use the SCD performance counter interrupt, caused by a Zclk counter
143 * overflow, to trigger the start of tracing.
145 * We set the trace buffer to sample everything and freeze on
148 * We map the interrupt for trace_buffer_freeze to handle it on CPU 0.
152 static u64 tb_period
;
154 static void arm_tb(void)
157 u64 next
= (1ULL << 40) - tb_period
;
158 u64 tb_options
= M_SCD_TRACE_CFG_FREEZE_FULL
;
161 * Generate an SCD_PERFCNT interrupt in TB_PERIOD Zclks to
162 * trigger start of trace. XXX vary sampling period
164 __raw_writeq(0, IOADDR(A_SCD_PERF_CNT_1
));
165 scdperfcnt
= __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG
));
168 * Unfortunately, in Pass 2 we must clear all counters to knock down
169 * a previous interrupt request. This means that bus profiling
170 * requires ALL of the SCD perf counters.
172 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
173 __raw_writeq((scdperfcnt
& ~M_SPC_CFG_SRC1
) |
174 /* keep counters 0,2,3,4,5,6,7 as is */
175 V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
176 IOADDR(A_BCM1480_SCD_PERF_CNT_CFG0
));
178 M_SPC_CFG_ENABLE
| /* enable counting */
179 M_SPC_CFG_CLEAR
| /* clear all counters */
180 V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
181 IOADDR(A_BCM1480_SCD_PERF_CNT_CFG1
));
183 __raw_writeq((scdperfcnt
& ~M_SPC_CFG_SRC1
) |
184 /* keep counters 0,2,3 as is */
185 M_SPC_CFG_ENABLE
| /* enable counting */
186 M_SPC_CFG_CLEAR
| /* clear all counters */
187 V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */
188 IOADDR(A_SCD_PERF_CNT_CFG
));
190 __raw_writeq(next
, IOADDR(A_SCD_PERF_CNT_1
));
191 /* Reset the trace buffer */
192 __raw_writeq(M_SCD_TRACE_CFG_RESET
, IOADDR(A_SCD_TRACE_CFG
));
193 #if 0 && defined(M_SCD_TRACE_CFG_FORCECNT)
194 /* XXXKW may want to expose control to the data-collector */
195 tb_options
|= M_SCD_TRACE_CFG_FORCECNT
;
197 __raw_writeq(tb_options
, IOADDR(A_SCD_TRACE_CFG
));
201 static irqreturn_t
sbprof_tb_intr(int irq
, void *dev_id
)
205 pr_debug(DEVNAME
": tb_intr\n");
207 if (sbp
.next_tb_sample
< MAX_TB_SAMPLES
) {
208 /* XXX should use XKPHYS to make writes bypass L2 */
209 u64
*p
= sbp
.sbprof_tbbuf
[sbp
.next_tb_sample
++];
211 __raw_writeq(M_SCD_TRACE_CFG_START_READ
,
212 IOADDR(A_SCD_TRACE_CFG
));
213 __asm__
__volatile__ ("sync" : : : "memory");
214 /* Loop runs backwards because bundles are read out in reverse order */
215 for (i
= 256 * 6; i
> 0; i
-= 6) {
216 /* Subscripts decrease to put bundle in the order */
217 /* t0 lo, t0 hi, t1 lo, t1 hi, t2 lo, t2 hi */
218 p
[i
- 1] = __raw_readq(IOADDR(A_SCD_TRACE_READ
));
220 p
[i
- 2] = __raw_readq(IOADDR(A_SCD_TRACE_READ
));
222 p
[i
- 3] = __raw_readq(IOADDR(A_SCD_TRACE_READ
));
224 p
[i
- 4] = __raw_readq(IOADDR(A_SCD_TRACE_READ
));
226 p
[i
- 5] = __raw_readq(IOADDR(A_SCD_TRACE_READ
));
228 p
[i
- 6] = __raw_readq(IOADDR(A_SCD_TRACE_READ
));
231 if (!sbp
.tb_enable
) {
232 pr_debug(DEVNAME
": tb_intr shutdown\n");
233 __raw_writeq(M_SCD_TRACE_CFG_RESET
,
234 IOADDR(A_SCD_TRACE_CFG
));
236 wake_up_interruptible(&sbp
.tb_sync
);
238 /* knock down current interrupt and get another one later */
242 /* No more trace buffer samples */
243 pr_debug(DEVNAME
": tb_intr full\n");
244 __raw_writeq(M_SCD_TRACE_CFG_RESET
, IOADDR(A_SCD_TRACE_CFG
));
247 wake_up_interruptible(&sbp
.tb_sync
);
248 wake_up_interruptible(&sbp
.tb_read
);
253 static irqreturn_t
sbprof_pc_intr(int irq
, void *dev_id
)
255 printk(DEVNAME
": unexpected pc_intr");
260 * Requires: Already called zclk_timer_init with a value that won't
261 * saturate 40 bits. No subsequent use of SCD performance counters
265 static int sbprof_zbprof_start(struct file
*filp
)
270 if (xchg(&sbp
.tb_enable
, 1))
273 pr_debug(DEVNAME
": starting\n");
275 sbp
.next_tb_sample
= 0;
278 err
= request_irq(K_INT_TRACE_FREEZE
, sbprof_tb_intr
, 0,
279 DEVNAME
" trace freeze", &sbp
);
283 /* Make sure there isn't a perf-cnt interrupt waiting */
284 scdperfcnt
= __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG
));
285 /* Disable and clear counters, override SRC_1 */
286 __raw_writeq((scdperfcnt
& ~(M_SPC_CFG_SRC1
| M_SPC_CFG_ENABLE
)) |
287 M_SPC_CFG_ENABLE
| M_SPC_CFG_CLEAR
| V_SPC_CFG_SRC1(1),
288 IOADDR(A_SCD_PERF_CNT_CFG
));
291 * We grab this interrupt to prevent others from trying to use
292 * it, even though we don't want to service the interrupts
293 * (they only feed into the trace-on-interrupt mechanism)
295 if (request_irq(K_INT_PERF_CNT
, sbprof_pc_intr
, 0, DEVNAME
" scd perfcnt", &sbp
)) {
296 free_irq(K_INT_TRACE_FREEZE
, &sbp
);
301 * I need the core to mask these, but the interrupt mapper to
302 * pass them through. I am exploiting my knowledge that
303 * cp0_status masks out IP[5]. krw
305 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
306 __raw_writeq(K_BCM1480_INT_MAP_I3
,
307 IOADDR(A_BCM1480_IMR_REGISTER(0, R_BCM1480_IMR_INTERRUPT_MAP_BASE_L
) +
308 ((K_BCM1480_INT_PERF_CNT
& 0x3f) << 3)));
310 __raw_writeq(K_INT_MAP_I3
,
311 IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_MAP_BASE
) +
312 (K_INT_PERF_CNT
<< 3)));
315 /* Initialize address traps */
316 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_0
));
317 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_1
));
318 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_2
));
319 __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_3
));
321 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_0
));
322 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_1
));
323 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_2
));
324 __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_3
));
326 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_0
));
327 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_1
));
328 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_2
));
329 __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_3
));
331 /* Initialize Trace Event 0-7 */
333 __raw_writeq(M_SCD_TREVT_INTERRUPT
, IOADDR(A_SCD_TRACE_EVENT_0
));
334 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_1
));
335 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_2
));
336 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_3
));
337 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_4
));
338 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_5
));
339 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_6
));
340 __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_7
));
342 /* Initialize Trace Sequence 0-7 */
343 /* Start on event 0 (interrupt) */
344 __raw_writeq(V_SCD_TRSEQ_FUNC_START
| 0x0fff,
345 IOADDR(A_SCD_TRACE_SEQUENCE_0
));
346 /* dsamp when d used | asamp when a used */
347 __raw_writeq(M_SCD_TRSEQ_ASAMPLE
| M_SCD_TRSEQ_DSAMPLE
|
348 K_SCD_TRSEQ_TRIGGER_ALL
,
349 IOADDR(A_SCD_TRACE_SEQUENCE_1
));
350 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_2
));
351 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_3
));
352 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_4
));
353 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_5
));
354 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_6
));
355 __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_7
));
357 /* Now indicate the PERF_CNT interrupt as a trace-relevant interrupt */
358 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
359 __raw_writeq(1ULL << (K_BCM1480_INT_PERF_CNT
& 0x3f),
360 IOADDR(A_BCM1480_IMR_REGISTER(0, R_BCM1480_IMR_INTERRUPT_TRACE_L
)));
362 __raw_writeq(1ULL << K_INT_PERF_CNT
,
363 IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_TRACE
)));
367 pr_debug(DEVNAME
": done starting\n");
372 static int sbprof_zbprof_stop(void)
376 pr_debug(DEVNAME
": stopping\n");
380 * XXXKW there is a window here where the intr handler may run,
381 * see the disable, and do the wake_up before this sleep
384 pr_debug(DEVNAME
": wait for disarm\n");
385 err
= wait_event_interruptible(sbp
.tb_sync
, !sbp
.tb_armed
);
386 pr_debug(DEVNAME
": disarm complete, stat %d\n", err
);
392 free_irq(K_INT_TRACE_FREEZE
, &sbp
);
393 free_irq(K_INT_PERF_CNT
, &sbp
);
396 pr_debug(DEVNAME
": done stopping\n");
401 static int sbprof_tb_open(struct inode
*inode
, struct file
*filp
)
405 minor
= iminor(inode
);
409 if (xchg(&sbp
.open
, SB_OPENING
) != SB_CLOSED
)
412 memset(&sbp
, 0, sizeof(struct sbprof_tb
));
413 sbp
.sbprof_tbbuf
= vzalloc(MAX_TBSAMPLE_BYTES
);
414 if (!sbp
.sbprof_tbbuf
) {
415 sbp
.open
= SB_CLOSED
;
420 init_waitqueue_head(&sbp
.tb_sync
);
421 init_waitqueue_head(&sbp
.tb_read
);
422 mutex_init(&sbp
.lock
);
430 static int sbprof_tb_release(struct inode
*inode
, struct file
*filp
)
434 minor
= iminor(inode
);
435 if (minor
!= 0 || sbp
.open
!= SB_CLOSED
)
438 mutex_lock(&sbp
.lock
);
440 if (sbp
.tb_armed
|| sbp
.tb_enable
)
441 sbprof_zbprof_stop();
443 vfree(sbp
.sbprof_tbbuf
);
444 sbp
.open
= SB_CLOSED
;
447 mutex_unlock(&sbp
.lock
);
452 static ssize_t
sbprof_tb_read(struct file
*filp
, char *buf
,
453 size_t size
, loff_t
*offp
)
455 int cur_sample
, sample_off
, cur_count
, sample_left
;
459 long cur_off
= *offp
;
461 if (!access_ok(VERIFY_WRITE
, buf
, size
))
464 mutex_lock(&sbp
.lock
);
467 cur_sample
= cur_off
/ TB_SAMPLE_SIZE
;
468 sample_off
= cur_off
% TB_SAMPLE_SIZE
;
469 sample_left
= TB_SAMPLE_SIZE
- sample_off
;
471 while (size
&& (cur_sample
< sbp
.next_tb_sample
)) {
474 cur_count
= size
< sample_left
? size
: sample_left
;
475 src
= (char *)(((long)sbp
.sbprof_tbbuf
[cur_sample
])+sample_off
);
476 err
= __copy_to_user(dest
, src
, cur_count
);
478 *offp
= cur_off
+ cur_count
- err
;
479 mutex_unlock(&sbp
.lock
);
482 pr_debug(DEVNAME
": read from sample %d, %d bytes\n",
483 cur_sample
, cur_count
);
485 sample_left
-= cur_count
;
489 sample_left
= TB_SAMPLE_SIZE
;
491 sample_off
+= cur_count
;
493 cur_off
+= cur_count
;
498 mutex_unlock(&sbp
.lock
);
503 static long sbprof_tb_ioctl(struct file
*filp
,
504 unsigned int command
,
511 mutex_lock(&sbp
.lock
);
512 err
= sbprof_zbprof_start(filp
);
513 mutex_unlock(&sbp
.lock
);
517 mutex_lock(&sbp
.lock
);
518 err
= sbprof_zbprof_stop();
519 mutex_unlock(&sbp
.lock
);
522 case SBPROF_ZBWAITFULL
: {
523 err
= wait_event_interruptible(sbp
.tb_read
, TB_FULL
);
527 err
= put_user(TB_FULL
, (int *) arg
);
539 static const struct file_operations sbprof_tb_fops
= {
540 .owner
= THIS_MODULE
,
541 .open
= sbprof_tb_open
,
542 .release
= sbprof_tb_release
,
543 .read
= sbprof_tb_read
,
544 .unlocked_ioctl
= sbprof_tb_ioctl
,
545 .compat_ioctl
= sbprof_tb_ioctl
,
547 .llseek
= default_llseek
,
550 static struct class *tb_class
;
551 static struct device
*tb_dev
;
553 static int __init
sbprof_tb_init(void)
559 if (register_chrdev(SBPROF_TB_MAJOR
, DEVNAME
, &sbprof_tb_fops
)) {
560 printk(KERN_WARNING DEVNAME
": initialization failed (dev %d)\n",
565 tbc
= class_create(THIS_MODULE
, "sb_tracebuffer");
573 dev
= device_create(tbc
, NULL
, MKDEV(SBPROF_TB_MAJOR
, 0), NULL
, "tb");
580 sbp
.open
= SB_CLOSED
;
582 tb_period
= zbbus_mhz
* 10000LL;
583 pr_info(DEVNAME
": initialized - tb_period = %lld\n",
584 (long long) tb_period
);
588 class_destroy(tb_class
);
590 unregister_chrdev(SBPROF_TB_MAJOR
, DEVNAME
);
595 static void __exit
sbprof_tb_cleanup(void)
597 device_destroy(tb_class
, MKDEV(SBPROF_TB_MAJOR
, 0));
598 unregister_chrdev(SBPROF_TB_MAJOR
, DEVNAME
);
599 class_destroy(tb_class
);
602 module_init(sbprof_tb_init
);
603 module_exit(sbprof_tb_cleanup
);
605 MODULE_ALIAS_CHARDEV_MAJOR(SBPROF_TB_MAJOR
);
606 MODULE_AUTHOR("Ralf Baechle <ralf@linux-mips.org>");
607 MODULE_LICENSE("GPL");