Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
awilliam
GitHub Repository: awilliam/linux-vfio
Path: blob/master/arch/s390/oprofile/hwsampler.c
10817 views
1
/**
2
* arch/s390/oprofile/hwsampler.c
3
*
4
* Copyright IBM Corp. 2010
5
* Author: Heinz Graalfs <[email protected]>
6
*/
7
8
#include <linux/kernel_stat.h>
9
#include <linux/kernel.h>
10
#include <linux/module.h>
11
#include <linux/smp.h>
12
#include <linux/errno.h>
13
#include <linux/workqueue.h>
14
#include <linux/interrupt.h>
15
#include <linux/notifier.h>
16
#include <linux/cpu.h>
17
#include <linux/semaphore.h>
18
#include <linux/oom.h>
19
#include <linux/oprofile.h>
20
21
#include <asm/lowcore.h>
22
#include <asm/irq.h>
23
24
#include "hwsampler.h"
25
26
#define MAX_NUM_SDB 511
27
#define MIN_NUM_SDB 1
28
29
#define ALERT_REQ_MASK 0x4000000000000000ul
30
#define BUFFER_FULL_MASK 0x8000000000000000ul
31
32
#define EI_IEA (1 << 31) /* invalid entry address */
33
#define EI_ISE (1 << 30) /* incorrect SDBT entry */
34
#define EI_PRA (1 << 29) /* program request alert */
35
#define EI_SACA (1 << 23) /* sampler authorization change alert */
36
#define EI_LSDA (1 << 22) /* loss of sample data alert */
37
38
DECLARE_PER_CPU(struct hws_cpu_buffer, sampler_cpu_buffer);
39
40
struct hws_execute_parms {
41
void *buffer;
42
signed int rc;
43
};
44
45
DEFINE_PER_CPU(struct hws_cpu_buffer, sampler_cpu_buffer);
46
EXPORT_PER_CPU_SYMBOL(sampler_cpu_buffer);
47
48
static DEFINE_MUTEX(hws_sem);
49
static DEFINE_MUTEX(hws_sem_oom);
50
51
static unsigned char hws_flush_all;
52
static unsigned int hws_oom;
53
static struct workqueue_struct *hws_wq;
54
55
static unsigned int hws_state;
56
enum {
57
HWS_INIT = 1,
58
HWS_DEALLOCATED,
59
HWS_STOPPED,
60
HWS_STARTED,
61
HWS_STOPPING };
62
63
/* set to 1 if called by kernel during memory allocation */
64
static unsigned char oom_killer_was_active;
65
/* size of SDBT and SDB as of allocate API */
66
static unsigned long num_sdbt = 100;
67
static unsigned long num_sdb = 511;
68
/* sampling interval (machine cycles) */
69
static unsigned long interval;
70
71
static unsigned long min_sampler_rate;
72
static unsigned long max_sampler_rate;
73
74
static int ssctl(void *buffer)
75
{
76
int cc;
77
78
/* set in order to detect a program check */
79
cc = 1;
80
81
asm volatile(
82
"0: .insn s,0xB2870000,0(%1)\n"
83
"1: ipm %0\n"
84
" srl %0,28\n"
85
"2:\n"
86
EX_TABLE(0b, 2b) EX_TABLE(1b, 2b)
87
: "+d" (cc), "+a" (buffer)
88
: "m" (*((struct hws_ssctl_request_block *)buffer))
89
: "cc", "memory");
90
91
return cc ? -EINVAL : 0 ;
92
}
93
94
static int qsi(void *buffer)
95
{
96
int cc;
97
cc = 1;
98
99
asm volatile(
100
"0: .insn s,0xB2860000,0(%1)\n"
101
"1: lhi %0,0\n"
102
"2:\n"
103
EX_TABLE(0b, 2b) EX_TABLE(1b, 2b)
104
: "=d" (cc), "+a" (buffer)
105
: "m" (*((struct hws_qsi_info_block *)buffer))
106
: "cc", "memory");
107
108
return cc ? -EINVAL : 0;
109
}
110
111
static void execute_qsi(void *parms)
112
{
113
struct hws_execute_parms *ep = parms;
114
115
ep->rc = qsi(ep->buffer);
116
}
117
118
static void execute_ssctl(void *parms)
119
{
120
struct hws_execute_parms *ep = parms;
121
122
ep->rc = ssctl(ep->buffer);
123
}
124
125
static int smp_ctl_ssctl_stop(int cpu)
126
{
127
int rc;
128
struct hws_execute_parms ep;
129
struct hws_cpu_buffer *cb;
130
131
cb = &per_cpu(sampler_cpu_buffer, cpu);
132
133
cb->ssctl.es = 0;
134
cb->ssctl.cs = 0;
135
136
ep.buffer = &cb->ssctl;
137
smp_call_function_single(cpu, execute_ssctl, &ep, 1);
138
rc = ep.rc;
139
if (rc) {
140
printk(KERN_ERR "hwsampler: CPU %d CPUMF SSCTL failed.\n", cpu);
141
dump_stack();
142
}
143
144
ep.buffer = &cb->qsi;
145
smp_call_function_single(cpu, execute_qsi, &ep, 1);
146
147
if (cb->qsi.es || cb->qsi.cs) {
148
printk(KERN_EMERG "CPUMF sampling did not stop properly.\n");
149
dump_stack();
150
}
151
152
return rc;
153
}
154
155
static int smp_ctl_ssctl_deactivate(int cpu)
156
{
157
int rc;
158
struct hws_execute_parms ep;
159
struct hws_cpu_buffer *cb;
160
161
cb = &per_cpu(sampler_cpu_buffer, cpu);
162
163
cb->ssctl.es = 1;
164
cb->ssctl.cs = 0;
165
166
ep.buffer = &cb->ssctl;
167
smp_call_function_single(cpu, execute_ssctl, &ep, 1);
168
rc = ep.rc;
169
if (rc)
170
printk(KERN_ERR "hwsampler: CPU %d CPUMF SSCTL failed.\n", cpu);
171
172
ep.buffer = &cb->qsi;
173
smp_call_function_single(cpu, execute_qsi, &ep, 1);
174
175
if (cb->qsi.cs)
176
printk(KERN_EMERG "CPUMF sampling was not set inactive.\n");
177
178
return rc;
179
}
180
181
static int smp_ctl_ssctl_enable_activate(int cpu, unsigned long interval)
182
{
183
int rc;
184
struct hws_execute_parms ep;
185
struct hws_cpu_buffer *cb;
186
187
cb = &per_cpu(sampler_cpu_buffer, cpu);
188
189
cb->ssctl.h = 1;
190
cb->ssctl.tear = cb->first_sdbt;
191
cb->ssctl.dear = *(unsigned long *) cb->first_sdbt;
192
cb->ssctl.interval = interval;
193
cb->ssctl.es = 1;
194
cb->ssctl.cs = 1;
195
196
ep.buffer = &cb->ssctl;
197
smp_call_function_single(cpu, execute_ssctl, &ep, 1);
198
rc = ep.rc;
199
if (rc)
200
printk(KERN_ERR "hwsampler: CPU %d CPUMF SSCTL failed.\n", cpu);
201
202
ep.buffer = &cb->qsi;
203
smp_call_function_single(cpu, execute_qsi, &ep, 1);
204
if (ep.rc)
205
printk(KERN_ERR "hwsampler: CPU %d CPUMF QSI failed.\n", cpu);
206
207
return rc;
208
}
209
210
static int smp_ctl_qsi(int cpu)
211
{
212
struct hws_execute_parms ep;
213
struct hws_cpu_buffer *cb;
214
215
cb = &per_cpu(sampler_cpu_buffer, cpu);
216
217
ep.buffer = &cb->qsi;
218
smp_call_function_single(cpu, execute_qsi, &ep, 1);
219
220
return ep.rc;
221
}
222
223
static inline unsigned long *trailer_entry_ptr(unsigned long v)
224
{
225
void *ret;
226
227
ret = (void *)v;
228
ret += PAGE_SIZE;
229
ret -= sizeof(struct hws_trailer_entry);
230
231
return (unsigned long *) ret;
232
}
233
234
/* prototypes for external interrupt handler and worker */
235
static void hws_ext_handler(unsigned int ext_int_code,
236
unsigned int param32, unsigned long param64);
237
238
static void worker(struct work_struct *work);
239
240
static void add_samples_to_oprofile(unsigned cpu, unsigned long *,
241
unsigned long *dear);
242
243
static void init_all_cpu_buffers(void)
244
{
245
int cpu;
246
struct hws_cpu_buffer *cb;
247
248
for_each_online_cpu(cpu) {
249
cb = &per_cpu(sampler_cpu_buffer, cpu);
250
memset(cb, 0, sizeof(struct hws_cpu_buffer));
251
}
252
}
253
254
static int is_link_entry(unsigned long *s)
255
{
256
return *s & 0x1ul ? 1 : 0;
257
}
258
259
static unsigned long *get_next_sdbt(unsigned long *s)
260
{
261
return (unsigned long *) (*s & ~0x1ul);
262
}
263
264
static int prepare_cpu_buffers(void)
265
{
266
int cpu;
267
int rc;
268
struct hws_cpu_buffer *cb;
269
270
rc = 0;
271
for_each_online_cpu(cpu) {
272
cb = &per_cpu(sampler_cpu_buffer, cpu);
273
atomic_set(&cb->ext_params, 0);
274
cb->worker_entry = 0;
275
cb->sample_overflow = 0;
276
cb->req_alert = 0;
277
cb->incorrect_sdbt_entry = 0;
278
cb->invalid_entry_address = 0;
279
cb->loss_of_sample_data = 0;
280
cb->sample_auth_change_alert = 0;
281
cb->finish = 0;
282
cb->oom = 0;
283
cb->stop_mode = 0;
284
}
285
286
return rc;
287
}
288
289
/*
290
* allocate_sdbt() - allocate sampler memory
291
* @cpu: the cpu for which sampler memory is allocated
292
*
293
* A 4K page is allocated for each requested SDBT.
294
* A maximum of 511 4K pages are allocated for the SDBs in each of the SDBTs.
295
* Set ALERT_REQ mask in each SDBs trailer.
296
* Returns zero if successful, <0 otherwise.
297
*/
298
static int allocate_sdbt(int cpu)
299
{
300
int j, k, rc;
301
unsigned long *sdbt;
302
unsigned long sdb;
303
unsigned long *tail;
304
unsigned long *trailer;
305
struct hws_cpu_buffer *cb;
306
307
cb = &per_cpu(sampler_cpu_buffer, cpu);
308
309
if (cb->first_sdbt)
310
return -EINVAL;
311
312
sdbt = NULL;
313
tail = sdbt;
314
315
for (j = 0; j < num_sdbt; j++) {
316
sdbt = (unsigned long *)get_zeroed_page(GFP_KERNEL);
317
318
mutex_lock(&hws_sem_oom);
319
/* OOM killer might have been activated */
320
barrier();
321
if (oom_killer_was_active || !sdbt) {
322
if (sdbt)
323
free_page((unsigned long)sdbt);
324
325
goto allocate_sdbt_error;
326
}
327
if (cb->first_sdbt == 0)
328
cb->first_sdbt = (unsigned long)sdbt;
329
330
/* link current page to tail of chain */
331
if (tail)
332
*tail = (unsigned long)(void *)sdbt + 1;
333
334
mutex_unlock(&hws_sem_oom);
335
336
for (k = 0; k < num_sdb; k++) {
337
/* get and set SDB page */
338
sdb = get_zeroed_page(GFP_KERNEL);
339
340
mutex_lock(&hws_sem_oom);
341
/* OOM killer might have been activated */
342
barrier();
343
if (oom_killer_was_active || !sdb) {
344
if (sdb)
345
free_page(sdb);
346
347
goto allocate_sdbt_error;
348
}
349
*sdbt = sdb;
350
trailer = trailer_entry_ptr(*sdbt);
351
*trailer = ALERT_REQ_MASK;
352
sdbt++;
353
mutex_unlock(&hws_sem_oom);
354
}
355
tail = sdbt;
356
}
357
mutex_lock(&hws_sem_oom);
358
if (oom_killer_was_active)
359
goto allocate_sdbt_error;
360
361
rc = 0;
362
if (tail)
363
*tail = (unsigned long)
364
((void *)cb->first_sdbt) + 1;
365
366
allocate_sdbt_exit:
367
mutex_unlock(&hws_sem_oom);
368
return rc;
369
370
allocate_sdbt_error:
371
rc = -ENOMEM;
372
goto allocate_sdbt_exit;
373
}
374
375
/*
376
* deallocate_sdbt() - deallocate all sampler memory
377
*
378
* For each online CPU all SDBT trees are deallocated.
379
* Returns the number of freed pages.
380
*/
381
static int deallocate_sdbt(void)
382
{
383
int cpu;
384
int counter;
385
386
counter = 0;
387
388
for_each_online_cpu(cpu) {
389
unsigned long start;
390
unsigned long sdbt;
391
unsigned long *curr;
392
struct hws_cpu_buffer *cb;
393
394
cb = &per_cpu(sampler_cpu_buffer, cpu);
395
396
if (!cb->first_sdbt)
397
continue;
398
399
sdbt = cb->first_sdbt;
400
curr = (unsigned long *) sdbt;
401
start = sdbt;
402
403
/* we'll free the SDBT after all SDBs are processed... */
404
while (1) {
405
if (!*curr || !sdbt)
406
break;
407
408
/* watch for link entry reset if found */
409
if (is_link_entry(curr)) {
410
curr = get_next_sdbt(curr);
411
if (sdbt)
412
free_page(sdbt);
413
414
/* we are done if we reach the start */
415
if ((unsigned long) curr == start)
416
break;
417
else
418
sdbt = (unsigned long) curr;
419
} else {
420
/* process SDB pointer */
421
if (*curr) {
422
free_page(*curr);
423
curr++;
424
}
425
}
426
counter++;
427
}
428
cb->first_sdbt = 0;
429
}
430
return counter;
431
}
432
433
static int start_sampling(int cpu)
434
{
435
int rc;
436
struct hws_cpu_buffer *cb;
437
438
cb = &per_cpu(sampler_cpu_buffer, cpu);
439
rc = smp_ctl_ssctl_enable_activate(cpu, interval);
440
if (rc) {
441
printk(KERN_INFO "hwsampler: CPU %d ssctl failed.\n", cpu);
442
goto start_exit;
443
}
444
445
rc = -EINVAL;
446
if (!cb->qsi.es) {
447
printk(KERN_INFO "hwsampler: CPU %d ssctl not enabled.\n", cpu);
448
goto start_exit;
449
}
450
451
if (!cb->qsi.cs) {
452
printk(KERN_INFO "hwsampler: CPU %d ssctl not active.\n", cpu);
453
goto start_exit;
454
}
455
456
printk(KERN_INFO
457
"hwsampler: CPU %d, CPUMF Sampling started, interval %lu.\n",
458
cpu, interval);
459
460
rc = 0;
461
462
start_exit:
463
return rc;
464
}
465
466
static int stop_sampling(int cpu)
467
{
468
unsigned long v;
469
int rc;
470
struct hws_cpu_buffer *cb;
471
472
rc = smp_ctl_qsi(cpu);
473
WARN_ON(rc);
474
475
cb = &per_cpu(sampler_cpu_buffer, cpu);
476
if (!rc && !cb->qsi.es)
477
printk(KERN_INFO "hwsampler: CPU %d, already stopped.\n", cpu);
478
479
rc = smp_ctl_ssctl_stop(cpu);
480
if (rc) {
481
printk(KERN_INFO "hwsampler: CPU %d, ssctl stop error %d.\n",
482
cpu, rc);
483
goto stop_exit;
484
}
485
486
printk(KERN_INFO "hwsampler: CPU %d, CPUMF Sampling stopped.\n", cpu);
487
488
stop_exit:
489
v = cb->req_alert;
490
if (v)
491
printk(KERN_ERR "hwsampler: CPU %d CPUMF Request alert,"
492
" count=%lu.\n", cpu, v);
493
494
v = cb->loss_of_sample_data;
495
if (v)
496
printk(KERN_ERR "hwsampler: CPU %d CPUMF Loss of sample data,"
497
" count=%lu.\n", cpu, v);
498
499
v = cb->invalid_entry_address;
500
if (v)
501
printk(KERN_ERR "hwsampler: CPU %d CPUMF Invalid entry address,"
502
" count=%lu.\n", cpu, v);
503
504
v = cb->incorrect_sdbt_entry;
505
if (v)
506
printk(KERN_ERR
507
"hwsampler: CPU %d CPUMF Incorrect SDBT address,"
508
" count=%lu.\n", cpu, v);
509
510
v = cb->sample_auth_change_alert;
511
if (v)
512
printk(KERN_ERR
513
"hwsampler: CPU %d CPUMF Sample authorization change,"
514
" count=%lu.\n", cpu, v);
515
516
return rc;
517
}
518
519
static int check_hardware_prerequisites(void)
520
{
521
if (!test_facility(68))
522
return -EOPNOTSUPP;
523
return 0;
524
}
525
/*
526
* hws_oom_callback() - the OOM callback function
527
*
528
* In case the callback is invoked during memory allocation for the
529
* hw sampler, all obtained memory is deallocated and a flag is set
530
* so main sampler memory allocation can exit with a failure code.
531
* In case the callback is invoked during sampling the hw sampler
532
* is deactivated for all CPUs.
533
*/
534
static int hws_oom_callback(struct notifier_block *nfb,
535
unsigned long dummy, void *parm)
536
{
537
unsigned long *freed;
538
int cpu;
539
struct hws_cpu_buffer *cb;
540
541
freed = parm;
542
543
mutex_lock(&hws_sem_oom);
544
545
if (hws_state == HWS_DEALLOCATED) {
546
/* during memory allocation */
547
if (oom_killer_was_active == 0) {
548
oom_killer_was_active = 1;
549
*freed += deallocate_sdbt();
550
}
551
} else {
552
int i;
553
cpu = get_cpu();
554
cb = &per_cpu(sampler_cpu_buffer, cpu);
555
556
if (!cb->oom) {
557
for_each_online_cpu(i) {
558
smp_ctl_ssctl_deactivate(i);
559
cb->oom = 1;
560
}
561
cb->finish = 1;
562
563
printk(KERN_INFO
564
"hwsampler: CPU %d, OOM notify during CPUMF Sampling.\n",
565
cpu);
566
}
567
}
568
569
mutex_unlock(&hws_sem_oom);
570
571
return NOTIFY_OK;
572
}
573
574
static struct notifier_block hws_oom_notifier = {
575
.notifier_call = hws_oom_callback
576
};
577
578
static int hws_cpu_callback(struct notifier_block *nfb,
579
unsigned long action, void *hcpu)
580
{
581
/* We do not have sampler space available for all possible CPUs.
582
All CPUs should be online when hw sampling is activated. */
583
return (hws_state <= HWS_DEALLOCATED) ? NOTIFY_OK : NOTIFY_BAD;
584
}
585
586
static struct notifier_block hws_cpu_notifier = {
587
.notifier_call = hws_cpu_callback
588
};
589
590
/**
591
* hwsampler_deactivate() - set hardware sampling temporarily inactive
592
* @cpu: specifies the CPU to be set inactive.
593
*
594
* Returns 0 on success, !0 on failure.
595
*/
596
int hwsampler_deactivate(unsigned int cpu)
597
{
598
/*
599
* Deactivate hw sampling temporarily and flush the buffer
600
* by pushing all the pending samples to oprofile buffer.
601
*
602
* This function can be called under one of the following conditions:
603
* Memory unmap, task is exiting.
604
*/
605
int rc;
606
struct hws_cpu_buffer *cb;
607
608
rc = 0;
609
mutex_lock(&hws_sem);
610
611
cb = &per_cpu(sampler_cpu_buffer, cpu);
612
if (hws_state == HWS_STARTED) {
613
rc = smp_ctl_qsi(cpu);
614
WARN_ON(rc);
615
if (cb->qsi.cs) {
616
rc = smp_ctl_ssctl_deactivate(cpu);
617
if (rc) {
618
printk(KERN_INFO
619
"hwsampler: CPU %d, CPUMF Deactivation failed.\n", cpu);
620
cb->finish = 1;
621
hws_state = HWS_STOPPING;
622
} else {
623
hws_flush_all = 1;
624
/* Add work to queue to read pending samples.*/
625
queue_work_on(cpu, hws_wq, &cb->worker);
626
}
627
}
628
}
629
mutex_unlock(&hws_sem);
630
631
if (hws_wq)
632
flush_workqueue(hws_wq);
633
634
return rc;
635
}
636
637
/**
638
* hwsampler_activate() - activate/resume hardware sampling which was deactivated
639
* @cpu: specifies the CPU to be set active.
640
*
641
* Returns 0 on success, !0 on failure.
642
*/
643
int hwsampler_activate(unsigned int cpu)
644
{
645
/*
646
* Re-activate hw sampling. This should be called in pair with
647
* hwsampler_deactivate().
648
*/
649
int rc;
650
struct hws_cpu_buffer *cb;
651
652
rc = 0;
653
mutex_lock(&hws_sem);
654
655
cb = &per_cpu(sampler_cpu_buffer, cpu);
656
if (hws_state == HWS_STARTED) {
657
rc = smp_ctl_qsi(cpu);
658
WARN_ON(rc);
659
if (!cb->qsi.cs) {
660
hws_flush_all = 0;
661
rc = smp_ctl_ssctl_enable_activate(cpu, interval);
662
if (rc) {
663
printk(KERN_ERR
664
"CPU %d, CPUMF activate sampling failed.\n",
665
cpu);
666
}
667
}
668
}
669
670
mutex_unlock(&hws_sem);
671
672
return rc;
673
}
674
675
static void hws_ext_handler(unsigned int ext_int_code,
676
unsigned int param32, unsigned long param64)
677
{
678
struct hws_cpu_buffer *cb;
679
680
kstat_cpu(smp_processor_id()).irqs[EXTINT_CPM]++;
681
cb = &__get_cpu_var(sampler_cpu_buffer);
682
atomic_xchg(&cb->ext_params, atomic_read(&cb->ext_params) | param32);
683
if (hws_wq)
684
queue_work(hws_wq, &cb->worker);
685
}
686
687
static int check_qsi_on_setup(void)
688
{
689
int rc;
690
unsigned int cpu;
691
struct hws_cpu_buffer *cb;
692
693
for_each_online_cpu(cpu) {
694
cb = &per_cpu(sampler_cpu_buffer, cpu);
695
rc = smp_ctl_qsi(cpu);
696
WARN_ON(rc);
697
if (rc)
698
return -EOPNOTSUPP;
699
700
if (!cb->qsi.as) {
701
printk(KERN_INFO "hwsampler: CPUMF sampling is not authorized.\n");
702
return -EINVAL;
703
}
704
705
if (cb->qsi.es) {
706
printk(KERN_WARNING "hwsampler: CPUMF is still enabled.\n");
707
rc = smp_ctl_ssctl_stop(cpu);
708
if (rc)
709
return -EINVAL;
710
711
printk(KERN_INFO
712
"CPU %d, CPUMF Sampling stopped now.\n", cpu);
713
}
714
}
715
return 0;
716
}
717
718
static int check_qsi_on_start(void)
719
{
720
unsigned int cpu;
721
int rc;
722
struct hws_cpu_buffer *cb;
723
724
for_each_online_cpu(cpu) {
725
cb = &per_cpu(sampler_cpu_buffer, cpu);
726
rc = smp_ctl_qsi(cpu);
727
WARN_ON(rc);
728
729
if (!cb->qsi.as)
730
return -EINVAL;
731
732
if (cb->qsi.es)
733
return -EINVAL;
734
735
if (cb->qsi.cs)
736
return -EINVAL;
737
}
738
return 0;
739
}
740
741
static void worker_on_start(unsigned int cpu)
742
{
743
struct hws_cpu_buffer *cb;
744
745
cb = &per_cpu(sampler_cpu_buffer, cpu);
746
cb->worker_entry = cb->first_sdbt;
747
}
748
749
static int worker_check_error(unsigned int cpu, int ext_params)
750
{
751
int rc;
752
unsigned long *sdbt;
753
struct hws_cpu_buffer *cb;
754
755
rc = 0;
756
cb = &per_cpu(sampler_cpu_buffer, cpu);
757
sdbt = (unsigned long *) cb->worker_entry;
758
759
if (!sdbt || !*sdbt)
760
return -EINVAL;
761
762
if (ext_params & EI_PRA)
763
cb->req_alert++;
764
765
if (ext_params & EI_LSDA)
766
cb->loss_of_sample_data++;
767
768
if (ext_params & EI_IEA) {
769
cb->invalid_entry_address++;
770
rc = -EINVAL;
771
}
772
773
if (ext_params & EI_ISE) {
774
cb->incorrect_sdbt_entry++;
775
rc = -EINVAL;
776
}
777
778
if (ext_params & EI_SACA) {
779
cb->sample_auth_change_alert++;
780
rc = -EINVAL;
781
}
782
783
return rc;
784
}
785
786
static void worker_on_finish(unsigned int cpu)
787
{
788
int rc, i;
789
struct hws_cpu_buffer *cb;
790
791
cb = &per_cpu(sampler_cpu_buffer, cpu);
792
793
if (cb->finish) {
794
rc = smp_ctl_qsi(cpu);
795
WARN_ON(rc);
796
if (cb->qsi.es) {
797
printk(KERN_INFO
798
"hwsampler: CPU %d, CPUMF Stop/Deactivate sampling.\n",
799
cpu);
800
rc = smp_ctl_ssctl_stop(cpu);
801
if (rc)
802
printk(KERN_INFO
803
"hwsampler: CPU %d, CPUMF Deactivation failed.\n",
804
cpu);
805
806
for_each_online_cpu(i) {
807
if (i == cpu)
808
continue;
809
if (!cb->finish) {
810
cb->finish = 1;
811
queue_work_on(i, hws_wq,
812
&cb->worker);
813
}
814
}
815
}
816
}
817
}
818
819
static void worker_on_interrupt(unsigned int cpu)
820
{
821
unsigned long *sdbt;
822
unsigned char done;
823
struct hws_cpu_buffer *cb;
824
825
cb = &per_cpu(sampler_cpu_buffer, cpu);
826
827
sdbt = (unsigned long *) cb->worker_entry;
828
829
done = 0;
830
/* do not proceed if stop was entered,
831
* forget the buffers not yet processed */
832
while (!done && !cb->stop_mode) {
833
unsigned long *trailer;
834
struct hws_trailer_entry *te;
835
unsigned long *dear = 0;
836
837
trailer = trailer_entry_ptr(*sdbt);
838
/* leave loop if no more work to do */
839
if (!(*trailer & BUFFER_FULL_MASK)) {
840
done = 1;
841
if (!hws_flush_all)
842
continue;
843
}
844
845
te = (struct hws_trailer_entry *)trailer;
846
cb->sample_overflow += te->overflow;
847
848
add_samples_to_oprofile(cpu, sdbt, dear);
849
850
/* reset trailer */
851
xchg((unsigned char *) te, 0x40);
852
853
/* advance to next sdb slot in current sdbt */
854
sdbt++;
855
/* in case link bit is set use address w/o link bit */
856
if (is_link_entry(sdbt))
857
sdbt = get_next_sdbt(sdbt);
858
859
cb->worker_entry = (unsigned long)sdbt;
860
}
861
}
862
863
static void add_samples_to_oprofile(unsigned int cpu, unsigned long *sdbt,
864
unsigned long *dear)
865
{
866
struct hws_data_entry *sample_data_ptr;
867
unsigned long *trailer;
868
869
trailer = trailer_entry_ptr(*sdbt);
870
if (dear) {
871
if (dear > trailer)
872
return;
873
trailer = dear;
874
}
875
876
sample_data_ptr = (struct hws_data_entry *)(*sdbt);
877
878
while ((unsigned long *)sample_data_ptr < trailer) {
879
struct pt_regs *regs = NULL;
880
struct task_struct *tsk = NULL;
881
882
/*
883
* Check sampling mode, 1 indicates basic (=customer) sampling
884
* mode.
885
*/
886
if (sample_data_ptr->def != 1) {
887
/* sample slot is not yet written */
888
break;
889
} else {
890
/* make sure we don't use it twice,
891
* the next time the sampler will set it again */
892
sample_data_ptr->def = 0;
893
}
894
895
/* Get pt_regs. */
896
if (sample_data_ptr->P == 1) {
897
/* userspace sample */
898
unsigned int pid = sample_data_ptr->prim_asn;
899
rcu_read_lock();
900
tsk = pid_task(find_vpid(pid), PIDTYPE_PID);
901
if (tsk)
902
regs = task_pt_regs(tsk);
903
rcu_read_unlock();
904
} else {
905
/* kernelspace sample */
906
regs = task_pt_regs(current);
907
}
908
909
mutex_lock(&hws_sem);
910
oprofile_add_ext_hw_sample(sample_data_ptr->ia, regs, 0,
911
!sample_data_ptr->P, tsk);
912
mutex_unlock(&hws_sem);
913
914
sample_data_ptr++;
915
}
916
}
917
918
static void worker(struct work_struct *work)
919
{
920
unsigned int cpu;
921
int ext_params;
922
struct hws_cpu_buffer *cb;
923
924
cb = container_of(work, struct hws_cpu_buffer, worker);
925
cpu = smp_processor_id();
926
ext_params = atomic_xchg(&cb->ext_params, 0);
927
928
if (!cb->worker_entry)
929
worker_on_start(cpu);
930
931
if (worker_check_error(cpu, ext_params))
932
return;
933
934
if (!cb->finish)
935
worker_on_interrupt(cpu);
936
937
if (cb->finish)
938
worker_on_finish(cpu);
939
}
940
941
/**
942
* hwsampler_allocate() - allocate memory for the hardware sampler
943
* @sdbt: number of SDBTs per online CPU (must be > 0)
944
* @sdb: number of SDBs per SDBT (minimum 1, maximum 511)
945
*
946
* Returns 0 on success, !0 on failure.
947
*/
948
int hwsampler_allocate(unsigned long sdbt, unsigned long sdb)
949
{
950
int cpu, rc;
951
mutex_lock(&hws_sem);
952
953
rc = -EINVAL;
954
if (hws_state != HWS_DEALLOCATED)
955
goto allocate_exit;
956
957
if (sdbt < 1)
958
goto allocate_exit;
959
960
if (sdb > MAX_NUM_SDB || sdb < MIN_NUM_SDB)
961
goto allocate_exit;
962
963
num_sdbt = sdbt;
964
num_sdb = sdb;
965
966
oom_killer_was_active = 0;
967
register_oom_notifier(&hws_oom_notifier);
968
969
for_each_online_cpu(cpu) {
970
if (allocate_sdbt(cpu)) {
971
unregister_oom_notifier(&hws_oom_notifier);
972
goto allocate_error;
973
}
974
}
975
unregister_oom_notifier(&hws_oom_notifier);
976
if (oom_killer_was_active)
977
goto allocate_error;
978
979
hws_state = HWS_STOPPED;
980
rc = 0;
981
982
allocate_exit:
983
mutex_unlock(&hws_sem);
984
return rc;
985
986
allocate_error:
987
rc = -ENOMEM;
988
printk(KERN_ERR "hwsampler: CPUMF Memory allocation failed.\n");
989
goto allocate_exit;
990
}
991
992
/**
993
* hwsampler_deallocate() - deallocate hardware sampler memory
994
*
995
* Returns 0 on success, !0 on failure.
996
*/
997
int hwsampler_deallocate()
998
{
999
int rc;
1000
1001
mutex_lock(&hws_sem);
1002
1003
rc = -EINVAL;
1004
if (hws_state != HWS_STOPPED)
1005
goto deallocate_exit;
1006
1007
ctl_clear_bit(0, 5); /* set bit 58 CR0 off */
1008
deallocate_sdbt();
1009
1010
hws_state = HWS_DEALLOCATED;
1011
rc = 0;
1012
1013
deallocate_exit:
1014
mutex_unlock(&hws_sem);
1015
1016
return rc;
1017
}
1018
1019
unsigned long hwsampler_query_min_interval(void)
1020
{
1021
return min_sampler_rate;
1022
}
1023
1024
unsigned long hwsampler_query_max_interval(void)
1025
{
1026
return max_sampler_rate;
1027
}
1028
1029
unsigned long hwsampler_get_sample_overflow_count(unsigned int cpu)
1030
{
1031
struct hws_cpu_buffer *cb;
1032
1033
cb = &per_cpu(sampler_cpu_buffer, cpu);
1034
1035
return cb->sample_overflow;
1036
}
1037
1038
int hwsampler_setup()
1039
{
1040
int rc;
1041
int cpu;
1042
struct hws_cpu_buffer *cb;
1043
1044
mutex_lock(&hws_sem);
1045
1046
rc = -EINVAL;
1047
if (hws_state)
1048
goto setup_exit;
1049
1050
hws_state = HWS_INIT;
1051
1052
init_all_cpu_buffers();
1053
1054
rc = check_hardware_prerequisites();
1055
if (rc)
1056
goto setup_exit;
1057
1058
rc = check_qsi_on_setup();
1059
if (rc)
1060
goto setup_exit;
1061
1062
rc = -EINVAL;
1063
hws_wq = create_workqueue("hwsampler");
1064
if (!hws_wq)
1065
goto setup_exit;
1066
1067
register_cpu_notifier(&hws_cpu_notifier);
1068
1069
for_each_online_cpu(cpu) {
1070
cb = &per_cpu(sampler_cpu_buffer, cpu);
1071
INIT_WORK(&cb->worker, worker);
1072
rc = smp_ctl_qsi(cpu);
1073
WARN_ON(rc);
1074
if (min_sampler_rate != cb->qsi.min_sampl_rate) {
1075
if (min_sampler_rate) {
1076
printk(KERN_WARNING
1077
"hwsampler: different min sampler rate values.\n");
1078
if (min_sampler_rate < cb->qsi.min_sampl_rate)
1079
min_sampler_rate =
1080
cb->qsi.min_sampl_rate;
1081
} else
1082
min_sampler_rate = cb->qsi.min_sampl_rate;
1083
}
1084
if (max_sampler_rate != cb->qsi.max_sampl_rate) {
1085
if (max_sampler_rate) {
1086
printk(KERN_WARNING
1087
"hwsampler: different max sampler rate values.\n");
1088
if (max_sampler_rate > cb->qsi.max_sampl_rate)
1089
max_sampler_rate =
1090
cb->qsi.max_sampl_rate;
1091
} else
1092
max_sampler_rate = cb->qsi.max_sampl_rate;
1093
}
1094
}
1095
register_external_interrupt(0x1407, hws_ext_handler);
1096
1097
hws_state = HWS_DEALLOCATED;
1098
rc = 0;
1099
1100
setup_exit:
1101
mutex_unlock(&hws_sem);
1102
return rc;
1103
}
1104
1105
int hwsampler_shutdown()
1106
{
1107
int rc;
1108
1109
mutex_lock(&hws_sem);
1110
1111
rc = -EINVAL;
1112
if (hws_state == HWS_DEALLOCATED || hws_state == HWS_STOPPED) {
1113
mutex_unlock(&hws_sem);
1114
1115
if (hws_wq)
1116
flush_workqueue(hws_wq);
1117
1118
mutex_lock(&hws_sem);
1119
1120
if (hws_state == HWS_STOPPED) {
1121
ctl_clear_bit(0, 5); /* set bit 58 CR0 off */
1122
deallocate_sdbt();
1123
}
1124
if (hws_wq) {
1125
destroy_workqueue(hws_wq);
1126
hws_wq = NULL;
1127
}
1128
1129
unregister_external_interrupt(0x1407, hws_ext_handler);
1130
hws_state = HWS_INIT;
1131
rc = 0;
1132
}
1133
mutex_unlock(&hws_sem);
1134
1135
unregister_cpu_notifier(&hws_cpu_notifier);
1136
1137
return rc;
1138
}
1139
1140
/**
1141
* hwsampler_start_all() - start hardware sampling on all online CPUs
1142
* @rate: specifies the used interval when samples are taken
1143
*
1144
* Returns 0 on success, !0 on failure.
1145
*/
1146
int hwsampler_start_all(unsigned long rate)
1147
{
1148
int rc, cpu;
1149
1150
mutex_lock(&hws_sem);
1151
1152
hws_oom = 0;
1153
1154
rc = -EINVAL;
1155
if (hws_state != HWS_STOPPED)
1156
goto start_all_exit;
1157
1158
interval = rate;
1159
1160
/* fail if rate is not valid */
1161
if (interval < min_sampler_rate || interval > max_sampler_rate)
1162
goto start_all_exit;
1163
1164
rc = check_qsi_on_start();
1165
if (rc)
1166
goto start_all_exit;
1167
1168
rc = prepare_cpu_buffers();
1169
if (rc)
1170
goto start_all_exit;
1171
1172
for_each_online_cpu(cpu) {
1173
rc = start_sampling(cpu);
1174
if (rc)
1175
break;
1176
}
1177
if (rc) {
1178
for_each_online_cpu(cpu) {
1179
stop_sampling(cpu);
1180
}
1181
goto start_all_exit;
1182
}
1183
hws_state = HWS_STARTED;
1184
rc = 0;
1185
1186
start_all_exit:
1187
mutex_unlock(&hws_sem);
1188
1189
if (rc)
1190
return rc;
1191
1192
register_oom_notifier(&hws_oom_notifier);
1193
hws_oom = 1;
1194
hws_flush_all = 0;
1195
/* now let them in, 1407 CPUMF external interrupts */
1196
ctl_set_bit(0, 5); /* set CR0 bit 58 */
1197
1198
return 0;
1199
}
1200
1201
/**
1202
* hwsampler_stop_all() - stop hardware sampling on all online CPUs
1203
*
1204
* Returns 0 on success, !0 on failure.
1205
*/
1206
int hwsampler_stop_all()
1207
{
1208
int tmp_rc, rc, cpu;
1209
struct hws_cpu_buffer *cb;
1210
1211
mutex_lock(&hws_sem);
1212
1213
rc = 0;
1214
if (hws_state == HWS_INIT) {
1215
mutex_unlock(&hws_sem);
1216
return rc;
1217
}
1218
hws_state = HWS_STOPPING;
1219
mutex_unlock(&hws_sem);
1220
1221
for_each_online_cpu(cpu) {
1222
cb = &per_cpu(sampler_cpu_buffer, cpu);
1223
cb->stop_mode = 1;
1224
tmp_rc = stop_sampling(cpu);
1225
if (tmp_rc)
1226
rc = tmp_rc;
1227
}
1228
1229
if (hws_wq)
1230
flush_workqueue(hws_wq);
1231
1232
mutex_lock(&hws_sem);
1233
if (hws_oom) {
1234
unregister_oom_notifier(&hws_oom_notifier);
1235
hws_oom = 0;
1236
}
1237
hws_state = HWS_STOPPED;
1238
mutex_unlock(&hws_sem);
1239
1240
return rc;
1241
}
1242
1243