blob: 4952872d6f0ae1c05c58bd7d046fe24029aef5ac [file] [log] [blame]
Heinz Graalfsec6a3df2011-01-21 10:06:52 +00001/**
2 * arch/s390/oprofile/hwsampler.c
3 *
4 * Copyright IBM Corp. 2010
5 * Author: Heinz Graalfs <graalfs@de.ibm.com>
6 */
7
8#include <linux/kernel.h>
9#include <linux/module.h>
10#include <linux/smp.h>
11#include <linux/errno.h>
12#include <linux/workqueue.h>
13#include <linux/interrupt.h>
14#include <linux/notifier.h>
15#include <linux/cpu.h>
16#include <linux/semaphore.h>
17#include <linux/oom.h>
18#include <linux/oprofile.h>
19
20#include <asm/lowcore.h>
21#include <asm/s390_ext.h>
22
23#include "hwsampler.h"
24
25#define MAX_NUM_SDB 511
26#define MIN_NUM_SDB 1
27
28#define ALERT_REQ_MASK 0x4000000000000000ul
29#define BUFFER_FULL_MASK 0x8000000000000000ul
30
31#define EI_IEA (1 << 31) /* invalid entry address */
32#define EI_ISE (1 << 30) /* incorrect SDBT entry */
33#define EI_PRA (1 << 29) /* program request alert */
34#define EI_SACA (1 << 23) /* sampler authorization change alert */
35#define EI_LSDA (1 << 22) /* loss of sample data alert */
36
37DECLARE_PER_CPU(struct hws_cpu_buffer, sampler_cpu_buffer);
38
39struct hws_execute_parms {
40 void *buffer;
41 signed int rc;
42};
43
44DEFINE_PER_CPU(struct hws_cpu_buffer, sampler_cpu_buffer);
45EXPORT_PER_CPU_SYMBOL(sampler_cpu_buffer);
46
47static DEFINE_MUTEX(hws_sem);
48static DEFINE_MUTEX(hws_sem_oom);
49
50static unsigned char hws_flush_all;
51static unsigned int hws_oom;
52static struct workqueue_struct *hws_wq;
53
54static unsigned int hws_state;
55enum {
56 HWS_INIT = 1,
57 HWS_DEALLOCATED,
58 HWS_STOPPED,
59 HWS_STARTED,
60 HWS_STOPPING };
61
62/* set to 1 if called by kernel during memory allocation */
63static unsigned char oom_killer_was_active;
64/* size of SDBT and SDB as of allocate API */
65static unsigned long num_sdbt = 100;
66static unsigned long num_sdb = 511;
67/* sampling interval (machine cycles) */
68static unsigned long interval;
69
70static unsigned long min_sampler_rate;
71static unsigned long max_sampler_rate;
72
73static int ssctl(void *buffer)
74{
75 int cc;
76
77 /* set in order to detect a program check */
78 cc = 1;
79
80 asm volatile(
81 "0: .insn s,0xB2870000,0(%1)\n"
82 "1: ipm %0\n"
83 " srl %0,28\n"
84 "2:\n"
85 EX_TABLE(0b, 2b) EX_TABLE(1b, 2b)
86 : "+d" (cc), "+a" (buffer)
87 : "m" (*((struct hws_ssctl_request_block *)buffer))
88 : "cc", "memory");
89
90 return cc ? -EINVAL : 0 ;
91}
92
93static int qsi(void *buffer)
94{
95 int cc;
96 cc = 1;
97
98 asm volatile(
99 "0: .insn s,0xB2860000,0(%1)\n"
100 "1: lhi %0,0\n"
101 "2:\n"
102 EX_TABLE(0b, 2b) EX_TABLE(1b, 2b)
103 : "=d" (cc), "+a" (buffer)
104 : "m" (*((struct hws_qsi_info_block *)buffer))
105 : "cc", "memory");
106
107 return cc ? -EINVAL : 0;
108}
109
110static void execute_qsi(void *parms)
111{
112 struct hws_execute_parms *ep = parms;
113
114 ep->rc = qsi(ep->buffer);
115}
116
117static void execute_ssctl(void *parms)
118{
119 struct hws_execute_parms *ep = parms;
120
121 ep->rc = ssctl(ep->buffer);
122}
123
124static int smp_ctl_ssctl_stop(int cpu)
125{
126 int rc;
127 struct hws_execute_parms ep;
128 struct hws_cpu_buffer *cb;
129
130 cb = &per_cpu(sampler_cpu_buffer, cpu);
131
132 cb->ssctl.es = 0;
133 cb->ssctl.cs = 0;
134
135 ep.buffer = &cb->ssctl;
136 smp_call_function_single(cpu, execute_ssctl, &ep, 1);
137 rc = ep.rc;
138 if (rc) {
139 printk(KERN_ERR "hwsampler: CPU %d CPUMF SSCTL failed.\n", cpu);
140 dump_stack();
141 }
142
143 ep.buffer = &cb->qsi;
144 smp_call_function_single(cpu, execute_qsi, &ep, 1);
145
146 if (cb->qsi.es || cb->qsi.cs) {
147 printk(KERN_EMERG "CPUMF sampling did not stop properly.\n");
148 dump_stack();
149 }
150
151 return rc;
152}
153
154static int smp_ctl_ssctl_deactivate(int cpu)
155{
156 int rc;
157 struct hws_execute_parms ep;
158 struct hws_cpu_buffer *cb;
159
160 cb = &per_cpu(sampler_cpu_buffer, cpu);
161
162 cb->ssctl.es = 1;
163 cb->ssctl.cs = 0;
164
165 ep.buffer = &cb->ssctl;
166 smp_call_function_single(cpu, execute_ssctl, &ep, 1);
167 rc = ep.rc;
168 if (rc)
169 printk(KERN_ERR "hwsampler: CPU %d CPUMF SSCTL failed.\n", cpu);
170
171 ep.buffer = &cb->qsi;
172 smp_call_function_single(cpu, execute_qsi, &ep, 1);
173
174 if (cb->qsi.cs)
175 printk(KERN_EMERG "CPUMF sampling was not set inactive.\n");
176
177 return rc;
178}
179
180static int smp_ctl_ssctl_enable_activate(int cpu, unsigned long interval)
181{
182 int rc;
183 struct hws_execute_parms ep;
184 struct hws_cpu_buffer *cb;
185
186 cb = &per_cpu(sampler_cpu_buffer, cpu);
187
188 cb->ssctl.h = 1;
189 cb->ssctl.tear = cb->first_sdbt;
190 cb->ssctl.dear = *(unsigned long *) cb->first_sdbt;
191 cb->ssctl.interval = interval;
192 cb->ssctl.es = 1;
193 cb->ssctl.cs = 1;
194
195 ep.buffer = &cb->ssctl;
196 smp_call_function_single(cpu, execute_ssctl, &ep, 1);
197 rc = ep.rc;
198 if (rc)
199 printk(KERN_ERR "hwsampler: CPU %d CPUMF SSCTL failed.\n", cpu);
200
201 ep.buffer = &cb->qsi;
202 smp_call_function_single(cpu, execute_qsi, &ep, 1);
203 if (ep.rc)
204 printk(KERN_ERR "hwsampler: CPU %d CPUMF QSI failed.\n", cpu);
205
206 return rc;
207}
208
209static int smp_ctl_qsi(int cpu)
210{
211 struct hws_execute_parms ep;
212 struct hws_cpu_buffer *cb;
213
214 cb = &per_cpu(sampler_cpu_buffer, cpu);
215
216 ep.buffer = &cb->qsi;
217 smp_call_function_single(cpu, execute_qsi, &ep, 1);
218
219 return ep.rc;
220}
221
222static inline unsigned long *trailer_entry_ptr(unsigned long v)
223{
224 void *ret;
225
226 ret = (void *)v;
227 ret += PAGE_SIZE;
228 ret -= sizeof(struct hws_trailer_entry);
229
230 return (unsigned long *) ret;
231}
232
233/* prototypes for external interrupt handler and worker */
234static void hws_ext_handler(unsigned int ext_int_code,
235 unsigned int param32, unsigned long param64);
236
237static void worker(struct work_struct *work);
238
239static void add_samples_to_oprofile(unsigned cpu, unsigned long *,
240 unsigned long *dear);
241
242static void init_all_cpu_buffers(void)
243{
244 int cpu;
245 struct hws_cpu_buffer *cb;
246
247 for_each_online_cpu(cpu) {
248 cb = &per_cpu(sampler_cpu_buffer, cpu);
249 memset(cb, 0, sizeof(struct hws_cpu_buffer));
250 }
251}
252
253static int is_link_entry(unsigned long *s)
254{
255 return *s & 0x1ul ? 1 : 0;
256}
257
258static unsigned long *get_next_sdbt(unsigned long *s)
259{
260 return (unsigned long *) (*s & ~0x1ul);
261}
262
263static int prepare_cpu_buffers(void)
264{
265 int cpu;
266 int rc;
267 struct hws_cpu_buffer *cb;
268
269 rc = 0;
270 for_each_online_cpu(cpu) {
271 cb = &per_cpu(sampler_cpu_buffer, cpu);
272 atomic_set(&cb->ext_params, 0);
273 cb->worker_entry = 0;
274 cb->sample_overflow = 0;
275 cb->req_alert = 0;
276 cb->incorrect_sdbt_entry = 0;
277 cb->invalid_entry_address = 0;
278 cb->loss_of_sample_data = 0;
279 cb->sample_auth_change_alert = 0;
280 cb->finish = 0;
281 cb->oom = 0;
282 cb->stop_mode = 0;
283 }
284
285 return rc;
286}
287
288/*
289 * allocate_sdbt() - allocate sampler memory
290 * @cpu: the cpu for which sampler memory is allocated
291 *
292 * A 4K page is allocated for each requested SDBT.
293 * A maximum of 511 4K pages are allocated for the SDBs in each of the SDBTs.
294 * Set ALERT_REQ mask in each SDBs trailer.
295 * Returns zero if successful, <0 otherwise.
296 */
297static int allocate_sdbt(int cpu)
298{
299 int j, k, rc;
300 unsigned long *sdbt;
301 unsigned long sdb;
302 unsigned long *tail;
303 unsigned long *trailer;
304 struct hws_cpu_buffer *cb;
305
306 cb = &per_cpu(sampler_cpu_buffer, cpu);
307
308 if (cb->first_sdbt)
309 return -EINVAL;
310
311 sdbt = NULL;
312 tail = sdbt;
313
314 for (j = 0; j < num_sdbt; j++) {
315 sdbt = (unsigned long *)get_zeroed_page(GFP_KERNEL);
316
317 mutex_lock(&hws_sem_oom);
318 /* OOM killer might have been activated */
319 barrier();
320 if (oom_killer_was_active || !sdbt) {
321 if (sdbt)
322 free_page((unsigned long)sdbt);
323
324 goto allocate_sdbt_error;
325 }
326 if (cb->first_sdbt == 0)
327 cb->first_sdbt = (unsigned long)sdbt;
328
329 /* link current page to tail of chain */
330 if (tail)
331 *tail = (unsigned long)(void *)sdbt + 1;
332
333 mutex_unlock(&hws_sem_oom);
334
335 for (k = 0; k < num_sdb; k++) {
336 /* get and set SDB page */
337 sdb = get_zeroed_page(GFP_KERNEL);
338
339 mutex_lock(&hws_sem_oom);
340 /* OOM killer might have been activated */
341 barrier();
342 if (oom_killer_was_active || !sdb) {
343 if (sdb)
344 free_page(sdb);
345
346 goto allocate_sdbt_error;
347 }
348 *sdbt = sdb;
349 trailer = trailer_entry_ptr(*sdbt);
350 *trailer = ALERT_REQ_MASK;
351 sdbt++;
352 mutex_unlock(&hws_sem_oom);
353 }
354 tail = sdbt;
355 }
356 mutex_lock(&hws_sem_oom);
357 if (oom_killer_was_active)
358 goto allocate_sdbt_error;
359
360 rc = 0;
361 if (tail)
362 *tail = (unsigned long)
363 ((void *)cb->first_sdbt) + 1;
364
365allocate_sdbt_exit:
366 mutex_unlock(&hws_sem_oom);
367 return rc;
368
369allocate_sdbt_error:
370 rc = -ENOMEM;
371 goto allocate_sdbt_exit;
372}
373
374/*
375 * deallocate_sdbt() - deallocate all sampler memory
376 *
377 * For each online CPU all SDBT trees are deallocated.
378 * Returns the number of freed pages.
379 */
380static int deallocate_sdbt(void)
381{
382 int cpu;
383 int counter;
384
385 counter = 0;
386
387 for_each_online_cpu(cpu) {
388 unsigned long start;
389 unsigned long sdbt;
390 unsigned long *curr;
391 struct hws_cpu_buffer *cb;
392
393 cb = &per_cpu(sampler_cpu_buffer, cpu);
394
395 if (!cb->first_sdbt)
396 continue;
397
398 sdbt = cb->first_sdbt;
399 curr = (unsigned long *) sdbt;
400 start = sdbt;
401
402 /* we'll free the SDBT after all SDBs are processed... */
403 while (1) {
404 if (!*curr || !sdbt)
405 break;
406
407 /* watch for link entry reset if found */
408 if (is_link_entry(curr)) {
409 curr = get_next_sdbt(curr);
410 if (sdbt)
411 free_page(sdbt);
412
413 /* we are done if we reach the start */
414 if ((unsigned long) curr == start)
415 break;
416 else
417 sdbt = (unsigned long) curr;
418 } else {
419 /* process SDB pointer */
420 if (*curr) {
421 free_page(*curr);
422 curr++;
423 }
424 }
425 counter++;
426 }
427 cb->first_sdbt = 0;
428 }
429 return counter;
430}
431
432static int start_sampling(int cpu)
433{
434 int rc;
435 struct hws_cpu_buffer *cb;
436
437 cb = &per_cpu(sampler_cpu_buffer, cpu);
438 rc = smp_ctl_ssctl_enable_activate(cpu, interval);
439 if (rc) {
440 printk(KERN_INFO "hwsampler: CPU %d ssctl failed.\n", cpu);
441 goto start_exit;
442 }
443
444 rc = -EINVAL;
445 if (!cb->qsi.es) {
446 printk(KERN_INFO "hwsampler: CPU %d ssctl not enabled.\n", cpu);
447 goto start_exit;
448 }
449
450 if (!cb->qsi.cs) {
451 printk(KERN_INFO "hwsampler: CPU %d ssctl not active.\n", cpu);
452 goto start_exit;
453 }
454
455 printk(KERN_INFO
456 "hwsampler: CPU %d, CPUMF Sampling started, interval %lu.\n",
457 cpu, interval);
458
459 rc = 0;
460
461start_exit:
462 return rc;
463}
464
465static int stop_sampling(int cpu)
466{
467 unsigned long v;
468 int rc;
469 struct hws_cpu_buffer *cb;
470
471 rc = smp_ctl_qsi(cpu);
472 WARN_ON(rc);
473
474 cb = &per_cpu(sampler_cpu_buffer, cpu);
475 if (!rc && !cb->qsi.es)
476 printk(KERN_INFO "hwsampler: CPU %d, already stopped.\n", cpu);
477
478 rc = smp_ctl_ssctl_stop(cpu);
479 if (rc) {
480 printk(KERN_INFO "hwsampler: CPU %d, ssctl stop error %d.\n",
481 cpu, rc);
482 goto stop_exit;
483 }
484
485 printk(KERN_INFO "hwsampler: CPU %d, CPUMF Sampling stopped.\n", cpu);
486
487stop_exit:
488 v = cb->req_alert;
489 if (v)
490 printk(KERN_ERR "hwsampler: CPU %d CPUMF Request alert,"
491 " count=%lu.\n", cpu, v);
492
493 v = cb->loss_of_sample_data;
494 if (v)
495 printk(KERN_ERR "hwsampler: CPU %d CPUMF Loss of sample data,"
496 " count=%lu.\n", cpu, v);
497
498 v = cb->invalid_entry_address;
499 if (v)
500 printk(KERN_ERR "hwsampler: CPU %d CPUMF Invalid entry address,"
501 " count=%lu.\n", cpu, v);
502
503 v = cb->incorrect_sdbt_entry;
504 if (v)
505 printk(KERN_ERR
506 "hwsampler: CPU %d CPUMF Incorrect SDBT address,"
507 " count=%lu.\n", cpu, v);
508
509 v = cb->sample_auth_change_alert;
510 if (v)
511 printk(KERN_ERR
512 "hwsampler: CPU %d CPUMF Sample authorization change,"
513 " count=%lu.\n", cpu, v);
514
515 return rc;
516}
517
518static int check_hardware_prerequisites(void)
519{
Jan Glauber65a94b12011-04-04 09:43:29 +0200520 if (!test_facility(68))
Heinz Graalfsec6a3df2011-01-21 10:06:52 +0000521 return -EOPNOTSUPP;
Heinz Graalfsec6a3df2011-01-21 10:06:52 +0000522 return 0;
523}
524/*
525 * hws_oom_callback() - the OOM callback function
526 *
527 * In case the callback is invoked during memory allocation for the
528 * hw sampler, all obtained memory is deallocated and a flag is set
529 * so main sampler memory allocation can exit with a failure code.
530 * In case the callback is invoked during sampling the hw sampler
531 * is deactivated for all CPUs.
532 */
533static int hws_oom_callback(struct notifier_block *nfb,
534 unsigned long dummy, void *parm)
535{
536 unsigned long *freed;
537 int cpu;
538 struct hws_cpu_buffer *cb;
539
540 freed = parm;
541
542 mutex_lock(&hws_sem_oom);
543
544 if (hws_state == HWS_DEALLOCATED) {
545 /* during memory allocation */
546 if (oom_killer_was_active == 0) {
547 oom_killer_was_active = 1;
548 *freed += deallocate_sdbt();
549 }
550 } else {
551 int i;
552 cpu = get_cpu();
553 cb = &per_cpu(sampler_cpu_buffer, cpu);
554
555 if (!cb->oom) {
556 for_each_online_cpu(i) {
557 smp_ctl_ssctl_deactivate(i);
558 cb->oom = 1;
559 }
560 cb->finish = 1;
561
562 printk(KERN_INFO
563 "hwsampler: CPU %d, OOM notify during CPUMF Sampling.\n",
564 cpu);
565 }
566 }
567
568 mutex_unlock(&hws_sem_oom);
569
570 return NOTIFY_OK;
571}
572
573static struct notifier_block hws_oom_notifier = {
574 .notifier_call = hws_oom_callback
575};
576
Robert Richter7bb2e262011-02-14 19:08:33 +0100577static int hws_cpu_callback(struct notifier_block *nfb,
Heinz Graalfsec6a3df2011-01-21 10:06:52 +0000578 unsigned long action, void *hcpu)
579{
580 /* We do not have sampler space available for all possible CPUs.
581 All CPUs should be online when hw sampling is activated. */
582 return NOTIFY_BAD;
583}
584
585static struct notifier_block hws_cpu_notifier = {
586 .notifier_call = hws_cpu_callback
587};
588
589/**
590 * hwsampler_deactivate() - set hardware sampling temporarily inactive
591 * @cpu: specifies the CPU to be set inactive.
592 *
593 * Returns 0 on success, !0 on failure.
594 */
595int hwsampler_deactivate(unsigned int cpu)
596{
597 /*
598 * Deactivate hw sampling temporarily and flush the buffer
599 * by pushing all the pending samples to oprofile buffer.
600 *
601 * This function can be called under one of the following conditions:
602 * Memory unmap, task is exiting.
603 */
604 int rc;
605 struct hws_cpu_buffer *cb;
606
607 rc = 0;
608 mutex_lock(&hws_sem);
609
610 cb = &per_cpu(sampler_cpu_buffer, cpu);
611 if (hws_state == HWS_STARTED) {
612 rc = smp_ctl_qsi(cpu);
613 WARN_ON(rc);
614 if (cb->qsi.cs) {
615 rc = smp_ctl_ssctl_deactivate(cpu);
616 if (rc) {
617 printk(KERN_INFO
618 "hwsampler: CPU %d, CPUMF Deactivation failed.\n", cpu);
619 cb->finish = 1;
620 hws_state = HWS_STOPPING;
621 } else {
622 hws_flush_all = 1;
623 /* Add work to queue to read pending samples.*/
624 queue_work_on(cpu, hws_wq, &cb->worker);
625 }
626 }
627 }
628 mutex_unlock(&hws_sem);
629
630 if (hws_wq)
631 flush_workqueue(hws_wq);
632
633 return rc;
634}
635
636/**
637 * hwsampler_activate() - activate/resume hardware sampling which was deactivated
638 * @cpu: specifies the CPU to be set active.
639 *
640 * Returns 0 on success, !0 on failure.
641 */
642int hwsampler_activate(unsigned int cpu)
643{
644 /*
645 * Re-activate hw sampling. This should be called in pair with
646 * hwsampler_deactivate().
647 */
648 int rc;
649 struct hws_cpu_buffer *cb;
650
651 rc = 0;
652 mutex_lock(&hws_sem);
653
654 cb = &per_cpu(sampler_cpu_buffer, cpu);
655 if (hws_state == HWS_STARTED) {
656 rc = smp_ctl_qsi(cpu);
657 WARN_ON(rc);
658 if (!cb->qsi.cs) {
659 hws_flush_all = 0;
660 rc = smp_ctl_ssctl_enable_activate(cpu, interval);
661 if (rc) {
662 printk(KERN_ERR
663 "CPU %d, CPUMF activate sampling failed.\n",
664 cpu);
665 }
666 }
667 }
668
669 mutex_unlock(&hws_sem);
670
671 return rc;
672}
673
674static void hws_ext_handler(unsigned int ext_int_code,
675 unsigned int param32, unsigned long param64)
676{
677 int cpu;
678 struct hws_cpu_buffer *cb;
679
680 cpu = smp_processor_id();
681 cb = &per_cpu(sampler_cpu_buffer, cpu);
682
683 atomic_xchg(
684 &cb->ext_params,
685 atomic_read(&cb->ext_params)
686 | S390_lowcore.ext_params);
687
688 if (hws_wq)
689 queue_work(hws_wq, &cb->worker);
690}
691
692static int check_qsi_on_setup(void)
693{
694 int rc;
695 unsigned int cpu;
696 struct hws_cpu_buffer *cb;
697
698 for_each_online_cpu(cpu) {
699 cb = &per_cpu(sampler_cpu_buffer, cpu);
700 rc = smp_ctl_qsi(cpu);
701 WARN_ON(rc);
702 if (rc)
703 return -EOPNOTSUPP;
704
705 if (!cb->qsi.as) {
706 printk(KERN_INFO "hwsampler: CPUMF sampling is not authorized.\n");
707 return -EINVAL;
708 }
709
710 if (cb->qsi.es) {
711 printk(KERN_WARNING "hwsampler: CPUMF is still enabled.\n");
712 rc = smp_ctl_ssctl_stop(cpu);
713 if (rc)
714 return -EINVAL;
715
716 printk(KERN_INFO
717 "CPU %d, CPUMF Sampling stopped now.\n", cpu);
718 }
719 }
720 return 0;
721}
722
723static int check_qsi_on_start(void)
724{
725 unsigned int cpu;
726 int rc;
727 struct hws_cpu_buffer *cb;
728
729 for_each_online_cpu(cpu) {
730 cb = &per_cpu(sampler_cpu_buffer, cpu);
731 rc = smp_ctl_qsi(cpu);
732 WARN_ON(rc);
733
734 if (!cb->qsi.as)
735 return -EINVAL;
736
737 if (cb->qsi.es)
738 return -EINVAL;
739
740 if (cb->qsi.cs)
741 return -EINVAL;
742 }
743 return 0;
744}
745
746static void worker_on_start(unsigned int cpu)
747{
748 struct hws_cpu_buffer *cb;
749
750 cb = &per_cpu(sampler_cpu_buffer, cpu);
751 cb->worker_entry = cb->first_sdbt;
752}
753
754static int worker_check_error(unsigned int cpu, int ext_params)
755{
756 int rc;
757 unsigned long *sdbt;
758 struct hws_cpu_buffer *cb;
759
760 rc = 0;
761 cb = &per_cpu(sampler_cpu_buffer, cpu);
762 sdbt = (unsigned long *) cb->worker_entry;
763
764 if (!sdbt || !*sdbt)
765 return -EINVAL;
766
767 if (ext_params & EI_IEA)
768 cb->req_alert++;
769
770 if (ext_params & EI_LSDA)
771 cb->loss_of_sample_data++;
772
773 if (ext_params & EI_IEA) {
774 cb->invalid_entry_address++;
775 rc = -EINVAL;
776 }
777
778 if (ext_params & EI_ISE) {
779 cb->incorrect_sdbt_entry++;
780 rc = -EINVAL;
781 }
782
783 if (ext_params & EI_SACA) {
784 cb->sample_auth_change_alert++;
785 rc = -EINVAL;
786 }
787
788 return rc;
789}
790
791static void worker_on_finish(unsigned int cpu)
792{
793 int rc, i;
794 struct hws_cpu_buffer *cb;
795
796 cb = &per_cpu(sampler_cpu_buffer, cpu);
797
798 if (cb->finish) {
799 rc = smp_ctl_qsi(cpu);
800 WARN_ON(rc);
801 if (cb->qsi.es) {
802 printk(KERN_INFO
803 "hwsampler: CPU %d, CPUMF Stop/Deactivate sampling.\n",
804 cpu);
805 rc = smp_ctl_ssctl_stop(cpu);
806 if (rc)
807 printk(KERN_INFO
808 "hwsampler: CPU %d, CPUMF Deactivation failed.\n",
809 cpu);
810
811 for_each_online_cpu(i) {
812 if (i == cpu)
813 continue;
814 if (!cb->finish) {
815 cb->finish = 1;
816 queue_work_on(i, hws_wq,
817 &cb->worker);
818 }
819 }
820 }
821 }
822}
823
824static void worker_on_interrupt(unsigned int cpu)
825{
826 unsigned long *sdbt;
827 unsigned char done;
828 struct hws_cpu_buffer *cb;
829
830 cb = &per_cpu(sampler_cpu_buffer, cpu);
831
832 sdbt = (unsigned long *) cb->worker_entry;
833
834 done = 0;
835 /* do not proceed if stop was entered,
836 * forget the buffers not yet processed */
837 while (!done && !cb->stop_mode) {
838 unsigned long *trailer;
839 struct hws_trailer_entry *te;
840 unsigned long *dear = 0;
841
842 trailer = trailer_entry_ptr(*sdbt);
843 /* leave loop if no more work to do */
844 if (!(*trailer & BUFFER_FULL_MASK)) {
845 done = 1;
846 if (!hws_flush_all)
847 continue;
848 }
849
850 te = (struct hws_trailer_entry *)trailer;
851 cb->sample_overflow += te->overflow;
852
853 add_samples_to_oprofile(cpu, sdbt, dear);
854
855 /* reset trailer */
856 xchg((unsigned char *) te, 0x40);
857
858 /* advance to next sdb slot in current sdbt */
859 sdbt++;
860 /* in case link bit is set use address w/o link bit */
861 if (is_link_entry(sdbt))
862 sdbt = get_next_sdbt(sdbt);
863
864 cb->worker_entry = (unsigned long)sdbt;
865 }
866}
867
868static void add_samples_to_oprofile(unsigned int cpu, unsigned long *sdbt,
869 unsigned long *dear)
870{
871 struct hws_data_entry *sample_data_ptr;
872 unsigned long *trailer;
873
874 trailer = trailer_entry_ptr(*sdbt);
875 if (dear) {
876 if (dear > trailer)
877 return;
878 trailer = dear;
879 }
880
881 sample_data_ptr = (struct hws_data_entry *)(*sdbt);
882
883 while ((unsigned long *)sample_data_ptr < trailer) {
884 struct pt_regs *regs = NULL;
885 struct task_struct *tsk = NULL;
886
887 /*
888 * Check sampling mode, 1 indicates basic (=customer) sampling
889 * mode.
890 */
891 if (sample_data_ptr->def != 1) {
892 /* sample slot is not yet written */
893 break;
894 } else {
895 /* make sure we don't use it twice,
896 * the next time the sampler will set it again */
897 sample_data_ptr->def = 0;
898 }
899
900 /* Get pt_regs. */
901 if (sample_data_ptr->P == 1) {
902 /* userspace sample */
903 unsigned int pid = sample_data_ptr->prim_asn;
904 rcu_read_lock();
905 tsk = pid_task(find_vpid(pid), PIDTYPE_PID);
906 if (tsk)
907 regs = task_pt_regs(tsk);
908 rcu_read_unlock();
909 } else {
910 /* kernelspace sample */
911 regs = task_pt_regs(current);
912 }
913
914 mutex_lock(&hws_sem);
915 oprofile_add_ext_hw_sample(sample_data_ptr->ia, regs, 0,
916 !sample_data_ptr->P, tsk);
917 mutex_unlock(&hws_sem);
918
919 sample_data_ptr++;
920 }
921}
922
923static void worker(struct work_struct *work)
924{
925 unsigned int cpu;
926 int ext_params;
927 struct hws_cpu_buffer *cb;
928
929 cb = container_of(work, struct hws_cpu_buffer, worker);
930 cpu = smp_processor_id();
931 ext_params = atomic_xchg(&cb->ext_params, 0);
932
933 if (!cb->worker_entry)
934 worker_on_start(cpu);
935
936 if (worker_check_error(cpu, ext_params))
937 return;
938
939 if (!cb->finish)
940 worker_on_interrupt(cpu);
941
942 if (cb->finish)
943 worker_on_finish(cpu);
944}
945
946/**
947 * hwsampler_allocate() - allocate memory for the hardware sampler
948 * @sdbt: number of SDBTs per online CPU (must be > 0)
949 * @sdb: number of SDBs per SDBT (minimum 1, maximum 511)
950 *
951 * Returns 0 on success, !0 on failure.
952 */
953int hwsampler_allocate(unsigned long sdbt, unsigned long sdb)
954{
955 int cpu, rc;
956 mutex_lock(&hws_sem);
957
958 rc = -EINVAL;
959 if (hws_state != HWS_DEALLOCATED)
960 goto allocate_exit;
961
962 if (sdbt < 1)
963 goto allocate_exit;
964
965 if (sdb > MAX_NUM_SDB || sdb < MIN_NUM_SDB)
966 goto allocate_exit;
967
968 num_sdbt = sdbt;
969 num_sdb = sdb;
970
971 oom_killer_was_active = 0;
972 register_oom_notifier(&hws_oom_notifier);
973
974 for_each_online_cpu(cpu) {
975 if (allocate_sdbt(cpu)) {
976 unregister_oom_notifier(&hws_oom_notifier);
977 goto allocate_error;
978 }
979 }
980 unregister_oom_notifier(&hws_oom_notifier);
981 if (oom_killer_was_active)
982 goto allocate_error;
983
984 hws_state = HWS_STOPPED;
985 rc = 0;
986
987allocate_exit:
988 mutex_unlock(&hws_sem);
989 return rc;
990
991allocate_error:
992 rc = -ENOMEM;
993 printk(KERN_ERR "hwsampler: CPUMF Memory allocation failed.\n");
994 goto allocate_exit;
995}
996
997/**
998 * hwsampler_deallocate() - deallocate hardware sampler memory
999 *
1000 * Returns 0 on success, !0 on failure.
1001 */
1002int hwsampler_deallocate()
1003{
1004 int rc;
1005
1006 mutex_lock(&hws_sem);
1007
1008 rc = -EINVAL;
1009 if (hws_state != HWS_STOPPED)
1010 goto deallocate_exit;
1011
1012 smp_ctl_clear_bit(0, 5); /* set bit 58 CR0 off */
1013 deallocate_sdbt();
1014
1015 hws_state = HWS_DEALLOCATED;
1016 rc = 0;
1017
1018deallocate_exit:
1019 mutex_unlock(&hws_sem);
1020
1021 return rc;
1022}
1023
1024long hwsampler_query_min_interval(void)
1025{
1026 if (min_sampler_rate)
1027 return min_sampler_rate;
1028 else
1029 return -EINVAL;
1030}
1031
1032long hwsampler_query_max_interval(void)
1033{
1034 if (max_sampler_rate)
1035 return max_sampler_rate;
1036 else
1037 return -EINVAL;
1038}
1039
1040unsigned long hwsampler_get_sample_overflow_count(unsigned int cpu)
1041{
1042 struct hws_cpu_buffer *cb;
1043
1044 cb = &per_cpu(sampler_cpu_buffer, cpu);
1045
1046 return cb->sample_overflow;
1047}
1048
1049int hwsampler_setup()
1050{
1051 int rc;
1052 int cpu;
1053 struct hws_cpu_buffer *cb;
1054
1055 mutex_lock(&hws_sem);
1056
1057 rc = -EINVAL;
1058 if (hws_state)
1059 goto setup_exit;
1060
1061 hws_state = HWS_INIT;
1062
1063 init_all_cpu_buffers();
1064
1065 rc = check_hardware_prerequisites();
1066 if (rc)
1067 goto setup_exit;
1068
1069 rc = check_qsi_on_setup();
1070 if (rc)
1071 goto setup_exit;
1072
1073 rc = -EINVAL;
1074 hws_wq = create_workqueue("hwsampler");
1075 if (!hws_wq)
1076 goto setup_exit;
1077
1078 register_cpu_notifier(&hws_cpu_notifier);
1079
1080 for_each_online_cpu(cpu) {
1081 cb = &per_cpu(sampler_cpu_buffer, cpu);
1082 INIT_WORK(&cb->worker, worker);
1083 rc = smp_ctl_qsi(cpu);
1084 WARN_ON(rc);
1085 if (min_sampler_rate != cb->qsi.min_sampl_rate) {
1086 if (min_sampler_rate) {
1087 printk(KERN_WARNING
1088 "hwsampler: different min sampler rate values.\n");
1089 if (min_sampler_rate < cb->qsi.min_sampl_rate)
1090 min_sampler_rate =
1091 cb->qsi.min_sampl_rate;
1092 } else
1093 min_sampler_rate = cb->qsi.min_sampl_rate;
1094 }
1095 if (max_sampler_rate != cb->qsi.max_sampl_rate) {
1096 if (max_sampler_rate) {
1097 printk(KERN_WARNING
1098 "hwsampler: different max sampler rate values.\n");
1099 if (max_sampler_rate > cb->qsi.max_sampl_rate)
1100 max_sampler_rate =
1101 cb->qsi.max_sampl_rate;
1102 } else
1103 max_sampler_rate = cb->qsi.max_sampl_rate;
1104 }
1105 }
1106 register_external_interrupt(0x1407, hws_ext_handler);
1107
1108 hws_state = HWS_DEALLOCATED;
1109 rc = 0;
1110
1111setup_exit:
1112 mutex_unlock(&hws_sem);
1113 return rc;
1114}
1115
1116int hwsampler_shutdown()
1117{
1118 int rc;
1119
1120 mutex_lock(&hws_sem);
1121
1122 rc = -EINVAL;
1123 if (hws_state == HWS_DEALLOCATED || hws_state == HWS_STOPPED) {
1124 mutex_unlock(&hws_sem);
1125
1126 if (hws_wq)
1127 flush_workqueue(hws_wq);
1128
1129 mutex_lock(&hws_sem);
1130
1131 if (hws_state == HWS_STOPPED) {
1132 smp_ctl_clear_bit(0, 5); /* set bit 58 CR0 off */
1133 deallocate_sdbt();
1134 }
1135 if (hws_wq) {
1136 destroy_workqueue(hws_wq);
1137 hws_wq = NULL;
1138 }
1139
1140 unregister_external_interrupt(0x1407, hws_ext_handler);
1141 hws_state = HWS_INIT;
1142 rc = 0;
1143 }
1144 mutex_unlock(&hws_sem);
1145
1146 unregister_cpu_notifier(&hws_cpu_notifier);
1147
1148 return rc;
1149}
1150
1151/**
1152 * hwsampler_start_all() - start hardware sampling on all online CPUs
1153 * @rate: specifies the used interval when samples are taken
1154 *
1155 * Returns 0 on success, !0 on failure.
1156 */
1157int hwsampler_start_all(unsigned long rate)
1158{
1159 int rc, cpu;
1160
1161 mutex_lock(&hws_sem);
1162
1163 hws_oom = 0;
1164
1165 rc = -EINVAL;
1166 if (hws_state != HWS_STOPPED)
1167 goto start_all_exit;
1168
1169 interval = rate;
1170
1171 /* fail if rate is not valid */
1172 if (interval < min_sampler_rate || interval > max_sampler_rate)
1173 goto start_all_exit;
1174
1175 rc = check_qsi_on_start();
1176 if (rc)
1177 goto start_all_exit;
1178
1179 rc = prepare_cpu_buffers();
1180 if (rc)
1181 goto start_all_exit;
1182
1183 for_each_online_cpu(cpu) {
1184 rc = start_sampling(cpu);
1185 if (rc)
1186 break;
1187 }
1188 if (rc) {
1189 for_each_online_cpu(cpu) {
1190 stop_sampling(cpu);
1191 }
1192 goto start_all_exit;
1193 }
1194 hws_state = HWS_STARTED;
1195 rc = 0;
1196
1197start_all_exit:
1198 mutex_unlock(&hws_sem);
1199
1200 if (rc)
1201 return rc;
1202
1203 register_oom_notifier(&hws_oom_notifier);
1204 hws_oom = 1;
1205 hws_flush_all = 0;
1206 /* now let them in, 1407 CPUMF external interrupts */
1207 smp_ctl_set_bit(0, 5); /* set CR0 bit 58 */
1208
1209 return 0;
1210}
1211
1212/**
1213 * hwsampler_stop_all() - stop hardware sampling on all online CPUs
1214 *
1215 * Returns 0 on success, !0 on failure.
1216 */
1217int hwsampler_stop_all()
1218{
1219 int tmp_rc, rc, cpu;
1220 struct hws_cpu_buffer *cb;
1221
1222 mutex_lock(&hws_sem);
1223
1224 rc = 0;
1225 if (hws_state == HWS_INIT) {
1226 mutex_unlock(&hws_sem);
1227 return rc;
1228 }
1229 hws_state = HWS_STOPPING;
1230 mutex_unlock(&hws_sem);
1231
1232 for_each_online_cpu(cpu) {
1233 cb = &per_cpu(sampler_cpu_buffer, cpu);
1234 cb->stop_mode = 1;
1235 tmp_rc = stop_sampling(cpu);
1236 if (tmp_rc)
1237 rc = tmp_rc;
1238 }
1239
1240 if (hws_wq)
1241 flush_workqueue(hws_wq);
1242
1243 mutex_lock(&hws_sem);
1244 if (hws_oom) {
1245 unregister_oom_notifier(&hws_oom_notifier);
1246 hws_oom = 0;
1247 }
1248 hws_state = HWS_STOPPED;
1249 mutex_unlock(&hws_sem);
1250
1251 return rc;
1252}