3 * Copyright IBM Corp. 2002, 2011
4 * Author(s): Thomas Spatzier (tspat@de.ibm.com)
5 * Author(s): Mahesh Salgaonkar (mahesh@linux.vnet.ibm.com)
6 * Author(s): Heinz Graalfs (graalfs@linux.vnet.ibm.com)
7 * Author(s): Andreas Krebbel (krebbel@linux.vnet.ibm.com)
9 * @remark Copyright 2002-2011 OProfile authors
12 #include <linux/oprofile.h>
13 #include <linux/perf_event.h>
14 #include <linux/init.h>
15 #include <linux/errno.h>
17 #include <linux/module.h>
18 #include <asm/processor.h>
20 #include "../../../drivers/oprofile/oprof.h"
22 extern void s390_backtrace(struct pt_regs
* const regs
, unsigned int depth
);
24 #include "hwsampler.h"
25 #include "op_counter.h"
27 #define DEFAULT_INTERVAL 4127518
29 #define DEFAULT_SDBT_BLOCKS 1
30 #define DEFAULT_SDB_BLOCKS 511
32 static unsigned long oprofile_hw_interval
= DEFAULT_INTERVAL
;
33 static unsigned long oprofile_min_interval
;
34 static unsigned long oprofile_max_interval
;
36 static unsigned long oprofile_sdbt_blocks
= DEFAULT_SDBT_BLOCKS
;
37 static unsigned long oprofile_sdb_blocks
= DEFAULT_SDB_BLOCKS
;
39 static int hwsampler_enabled
;
40 static int hwsampler_running
; /* start_mutex must be held to change */
41 static int hwsampler_available
;
43 static struct oprofile_operations timer_ops
;
45 struct op_counter_config counter_config
;
47 enum __force_cpu_type
{
48 reserved
= 0, /* do not force */
51 static int force_cpu_type
;
53 static int set_cpu_type(const char *str
, struct kernel_param
*kp
)
55 if (!strcmp(str
, "timer")) {
56 force_cpu_type
= timer
;
57 printk(KERN_INFO
"oprofile: forcing timer to be returned "
65 module_param_call(cpu_type
, set_cpu_type
, NULL
, NULL
, 0);
66 MODULE_PARM_DESC(cpu_type
, "Force legacy basic mode sampling"
67 "(report cpu_type \"timer\"");
69 static int __oprofile_hwsampler_start(void)
73 retval
= hwsampler_allocate(oprofile_sdbt_blocks
, oprofile_sdb_blocks
);
77 retval
= hwsampler_start_all(oprofile_hw_interval
);
79 hwsampler_deallocate();
84 static int oprofile_hwsampler_start(void)
88 hwsampler_running
= hwsampler_enabled
;
90 if (!hwsampler_running
)
91 return timer_ops
.start();
93 retval
= perf_reserve_sampling();
97 retval
= __oprofile_hwsampler_start();
99 perf_release_sampling();
104 static void oprofile_hwsampler_stop(void)
106 if (!hwsampler_running
) {
111 hwsampler_stop_all();
112 hwsampler_deallocate();
113 perf_release_sampling();
119 * /dev/oprofile/0/enabled
120 * /dev/oprofile/hwsampling/hwsampler (cpu_type = timer)
123 static ssize_t
hwsampler_read(struct file
*file
, char __user
*buf
,
124 size_t count
, loff_t
*offset
)
126 return oprofilefs_ulong_to_user(hwsampler_enabled
, buf
, count
, offset
);
129 static ssize_t
hwsampler_write(struct file
*file
, char const __user
*buf
,
130 size_t count
, loff_t
*offset
)
138 retval
= oprofilefs_ulong_from_user(&val
, buf
, count
);
142 if (val
!= 0 && val
!= 1)
145 if (oprofile_started
)
147 * save to do without locking as we set
148 * hwsampler_running in start() when start_mutex is
153 hwsampler_enabled
= val
;
158 static const struct file_operations hwsampler_fops
= {
159 .read
= hwsampler_read
,
160 .write
= hwsampler_write
,
165 * /dev/oprofile/0/count
166 * /dev/oprofile/hwsampling/hw_interval (cpu_type = timer)
168 * Make sure that the value is within the hardware range.
171 static ssize_t
hw_interval_read(struct file
*file
, char __user
*buf
,
172 size_t count
, loff_t
*offset
)
174 return oprofilefs_ulong_to_user(oprofile_hw_interval
, buf
,
178 static ssize_t
hw_interval_write(struct file
*file
, char const __user
*buf
,
179 size_t count
, loff_t
*offset
)
186 retval
= oprofilefs_ulong_from_user(&val
, buf
, count
);
189 if (val
< oprofile_min_interval
)
190 oprofile_hw_interval
= oprofile_min_interval
;
191 else if (val
> oprofile_max_interval
)
192 oprofile_hw_interval
= oprofile_max_interval
;
194 oprofile_hw_interval
= val
;
199 static const struct file_operations hw_interval_fops
= {
200 .read
= hw_interval_read
,
201 .write
= hw_interval_write
,
206 * /dev/oprofile/0/event
207 * Only a single event with number 0 is supported with this counter.
209 * /dev/oprofile/0/unit_mask
210 * This is a dummy file needed by the user space tools.
211 * No value other than 0 is accepted or returned.
214 static ssize_t
hwsampler_zero_read(struct file
*file
, char __user
*buf
,
215 size_t count
, loff_t
*offset
)
217 return oprofilefs_ulong_to_user(0, buf
, count
, offset
);
220 static ssize_t
hwsampler_zero_write(struct file
*file
, char const __user
*buf
,
221 size_t count
, loff_t
*offset
)
229 retval
= oprofilefs_ulong_from_user(&val
, buf
, count
);
237 static const struct file_operations zero_fops
= {
238 .read
= hwsampler_zero_read
,
239 .write
= hwsampler_zero_write
,
242 /* /dev/oprofile/0/kernel file ops. */
244 static ssize_t
hwsampler_kernel_read(struct file
*file
, char __user
*buf
,
245 size_t count
, loff_t
*offset
)
247 return oprofilefs_ulong_to_user(counter_config
.kernel
,
251 static ssize_t
hwsampler_kernel_write(struct file
*file
, char const __user
*buf
,
252 size_t count
, loff_t
*offset
)
260 retval
= oprofilefs_ulong_from_user(&val
, buf
, count
);
264 if (val
!= 0 && val
!= 1)
267 counter_config
.kernel
= val
;
272 static const struct file_operations kernel_fops
= {
273 .read
= hwsampler_kernel_read
,
274 .write
= hwsampler_kernel_write
,
277 /* /dev/oprofile/0/user file ops. */
279 static ssize_t
hwsampler_user_read(struct file
*file
, char __user
*buf
,
280 size_t count
, loff_t
*offset
)
282 return oprofilefs_ulong_to_user(counter_config
.user
,
286 static ssize_t
hwsampler_user_write(struct file
*file
, char const __user
*buf
,
287 size_t count
, loff_t
*offset
)
295 retval
= oprofilefs_ulong_from_user(&val
, buf
, count
);
299 if (val
!= 0 && val
!= 1)
302 counter_config
.user
= val
;
307 static const struct file_operations user_fops
= {
308 .read
= hwsampler_user_read
,
309 .write
= hwsampler_user_write
,
314 * File ops used for: /dev/oprofile/timer/enabled
315 * The value always has to be the inverted value of hwsampler_enabled. So
316 * no separate variable is created. That way we do not need locking.
319 static ssize_t
timer_enabled_read(struct file
*file
, char __user
*buf
,
320 size_t count
, loff_t
*offset
)
322 return oprofilefs_ulong_to_user(!hwsampler_enabled
, buf
, count
, offset
);
325 static ssize_t
timer_enabled_write(struct file
*file
, char const __user
*buf
,
326 size_t count
, loff_t
*offset
)
334 retval
= oprofilefs_ulong_from_user(&val
, buf
, count
);
338 if (val
!= 0 && val
!= 1)
341 /* Timer cannot be disabled without having hardware sampling. */
342 if (val
== 0 && !hwsampler_available
)
345 if (oprofile_started
)
347 * save to do without locking as we set
348 * hwsampler_running in start() when start_mutex is
353 hwsampler_enabled
= !val
;
358 static const struct file_operations timer_enabled_fops
= {
359 .read
= timer_enabled_read
,
360 .write
= timer_enabled_write
,
364 static int oprofile_create_hwsampling_files(struct dentry
*root
)
368 dir
= oprofilefs_mkdir(root
, "timer");
372 oprofilefs_create_file(dir
, "enabled", &timer_enabled_fops
);
374 if (!hwsampler_available
)
377 /* reinitialize default values */
378 hwsampler_enabled
= 1;
379 counter_config
.kernel
= 1;
380 counter_config
.user
= 1;
382 if (!force_cpu_type
) {
384 * Create the counter file system. A single virtual
385 * counter is created which can be used to
386 * enable/disable hardware sampling dynamically from
387 * user space. The user space will configure a single
388 * counter with a single event. The value of 'event'
389 * and 'unit_mask' are not evaluated by the kernel code
390 * and can only be set to 0.
393 dir
= oprofilefs_mkdir(root
, "0");
397 oprofilefs_create_file(dir
, "enabled", &hwsampler_fops
);
398 oprofilefs_create_file(dir
, "event", &zero_fops
);
399 oprofilefs_create_file(dir
, "count", &hw_interval_fops
);
400 oprofilefs_create_file(dir
, "unit_mask", &zero_fops
);
401 oprofilefs_create_file(dir
, "kernel", &kernel_fops
);
402 oprofilefs_create_file(dir
, "user", &user_fops
);
403 oprofilefs_create_ulong(dir
, "hw_sdbt_blocks",
404 &oprofile_sdbt_blocks
);
408 * Hardware sampling can be used but the cpu_type is
409 * forced to timer in order to deal with legacy user
410 * space tools. The /dev/oprofile/hwsampling fs is
411 * provided in that case.
413 dir
= oprofilefs_mkdir(root
, "hwsampling");
417 oprofilefs_create_file(dir
, "hwsampler",
419 oprofilefs_create_file(dir
, "hw_interval",
421 oprofilefs_create_ro_ulong(dir
, "hw_min_interval",
422 &oprofile_min_interval
);
423 oprofilefs_create_ro_ulong(dir
, "hw_max_interval",
424 &oprofile_max_interval
);
425 oprofilefs_create_ulong(dir
, "hw_sdbt_blocks",
426 &oprofile_sdbt_blocks
);
431 static int oprofile_hwsampler_init(struct oprofile_operations
*ops
)
434 * Initialize the timer mode infrastructure as well in order
435 * to be able to switch back dynamically. oprofile_timer_init
436 * is not supposed to fail.
438 if (oprofile_timer_init(ops
))
441 memcpy(&timer_ops
, ops
, sizeof(timer_ops
));
442 ops
->create_files
= oprofile_create_hwsampling_files
;
445 * If the user space tools do not support newer cpu types,
446 * the force_cpu_type module parameter
447 * can be used to always return \"timer\" as cpu type.
449 if (force_cpu_type
!= timer
) {
454 switch (id
.machine
) {
455 case 0x2097: case 0x2098: ops
->cpu_type
= "s390/z10"; break;
456 case 0x2817: case 0x2818: ops
->cpu_type
= "s390/z196"; break;
457 case 0x2827: case 0x2828: ops
->cpu_type
= "s390/zEC12"; break;
458 default: return -ENODEV
;
462 if (hwsampler_setup())
466 * Query the range for the sampling interval from the
469 oprofile_min_interval
= hwsampler_query_min_interval();
470 if (oprofile_min_interval
== 0)
472 oprofile_max_interval
= hwsampler_query_max_interval();
473 if (oprofile_max_interval
== 0)
476 /* The initial value should be sane */
477 if (oprofile_hw_interval
< oprofile_min_interval
)
478 oprofile_hw_interval
= oprofile_min_interval
;
479 if (oprofile_hw_interval
> oprofile_max_interval
)
480 oprofile_hw_interval
= oprofile_max_interval
;
482 printk(KERN_INFO
"oprofile: System z hardware sampling "
483 "facility found.\n");
485 ops
->start
= oprofile_hwsampler_start
;
486 ops
->stop
= oprofile_hwsampler_stop
;
491 static void oprofile_hwsampler_exit(void)
493 hwsampler_shutdown();
496 int __init
oprofile_arch_init(struct oprofile_operations
*ops
)
498 ops
->backtrace
= s390_backtrace
;
501 * -ENODEV is not reported to the caller. The module itself
502 * will use the timer mode sampling as fallback and this is
505 hwsampler_available
= oprofile_hwsampler_init(ops
) == 0;
510 void oprofile_arch_exit(void)
512 oprofile_hwsampler_exit();