forked from torvalds/linux
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathop_model_cell.c
1713 lines (1458 loc) · 49.2 KB
/
op_model_cell.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Cell Broadband Engine OProfile Support
*
* (C) Copyright IBM Corporation 2006
*
* Author: David Erb ([email protected])
* Modifications:
* Carl Love <[email protected]>
* Maynard Johnson <[email protected]>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/cpufreq.h>
#include <linux/delay.h>
#include <linux/jiffies.h>
#include <linux/kthread.h>
#include <linux/oprofile.h>
#include <linux/percpu.h>
#include <linux/smp.h>
#include <linux/spinlock.h>
#include <linux/timer.h>
#include <asm/cell-pmu.h>
#include <asm/cputable.h>
#include <asm/firmware.h>
#include <asm/io.h>
#include <asm/oprofile_impl.h>
#include <asm/processor.h>
#include <asm/prom.h>
#include <asm/ptrace.h>
#include <asm/reg.h>
#include <asm/rtas.h>
#include <asm/cell-regs.h>
#include "../platforms/cell/interrupt.h"
#include "cell/pr_util.h"
#define PPU_PROFILING 0
#define SPU_PROFILING_CYCLES 1
#define SPU_PROFILING_EVENTS 2
#define SPU_EVENT_NUM_START 4100
#define SPU_EVENT_NUM_STOP 4399
#define SPU_PROFILE_EVENT_ADDR 4363 /* spu, address trace, decimal */
#define SPU_PROFILE_EVENT_ADDR_MASK_A 0x146 /* sub unit set to zero */
#define SPU_PROFILE_EVENT_ADDR_MASK_B 0x186 /* sub unit set to zero */
#define NUM_SPUS_PER_NODE 8
#define SPU_CYCLES_EVENT_NUM 2 /* event number for SPU_CYCLES */
#define PPU_CYCLES_EVENT_NUM 1 /* event number for CYCLES */
#define PPU_CYCLES_GRP_NUM 1 /* special group number for identifying
* PPU_CYCLES event
*/
#define CBE_COUNT_ALL_CYCLES 0x42800000 /* PPU cycle event specifier */
#define NUM_THREADS 2 /* number of physical threads in
* physical processor
*/
#define NUM_DEBUG_BUS_WORDS 4
#define NUM_INPUT_BUS_WORDS 2
#define MAX_SPU_COUNT 0xFFFFFF /* maximum 24 bit LFSR value */
/* Minimum HW interval timer setting to send value to trace buffer is 10 cycle.
* To configure counter to send value every N cycles set counter to
* 2^32 - 1 - N.
*/
#define NUM_INTERVAL_CYC 0xFFFFFFFF - 10
/*
* spu_cycle_reset is the number of cycles between samples.
* This variable is used for SPU profiling and should ONLY be set
* at the beginning of cell_reg_setup; otherwise, it's read-only.
*/
static unsigned int spu_cycle_reset;
static unsigned int profiling_mode;
static int spu_evnt_phys_spu_indx;
struct pmc_cntrl_data {
unsigned long vcntr;
unsigned long evnts;
unsigned long masks;
unsigned long enabled;
};
/*
* ibm,cbe-perftools rtas parameters
*/
struct pm_signal {
u16 cpu; /* Processor to modify */
u16 sub_unit; /* hw subunit this applies to (if applicable)*/
short int signal_group; /* Signal Group to Enable/Disable */
u8 bus_word; /* Enable/Disable on this Trace/Trigger/Event
* Bus Word(s) (bitmask)
*/
u8 bit; /* Trigger/Event bit (if applicable) */
};
/*
* rtas call arguments
*/
enum {
SUBFUNC_RESET = 1,
SUBFUNC_ACTIVATE = 2,
SUBFUNC_DEACTIVATE = 3,
PASSTHRU_IGNORE = 0,
PASSTHRU_ENABLE = 1,
PASSTHRU_DISABLE = 2,
};
struct pm_cntrl {
u16 enable;
u16 stop_at_max;
u16 trace_mode;
u16 freeze;
u16 count_mode;
u16 spu_addr_trace;
u8 trace_buf_ovflw;
};
static struct {
u32 group_control;
u32 debug_bus_control;
struct pm_cntrl pm_cntrl;
u32 pm07_cntrl[NR_PHYS_CTRS];
} pm_regs;
#define GET_SUB_UNIT(x) ((x & 0x0000f000) >> 12)
#define GET_BUS_WORD(x) ((x & 0x000000f0) >> 4)
#define GET_BUS_TYPE(x) ((x & 0x00000300) >> 8)
#define GET_POLARITY(x) ((x & 0x00000002) >> 1)
#define GET_COUNT_CYCLES(x) (x & 0x00000001)
#define GET_INPUT_CONTROL(x) ((x & 0x00000004) >> 2)
static DEFINE_PER_CPU(unsigned long[NR_PHYS_CTRS], pmc_values);
static unsigned long spu_pm_cnt[MAX_NUMNODES * NUM_SPUS_PER_NODE];
static struct pmc_cntrl_data pmc_cntrl[NUM_THREADS][NR_PHYS_CTRS];
/*
* The CELL profiling code makes rtas calls to setup the debug bus to
* route the performance signals. Additionally, SPU profiling requires
* a second rtas call to setup the hardware to capture the SPU PCs.
* The EIO error value is returned if the token lookups or the rtas
* call fail. The EIO error number is the best choice of the existing
* error numbers. The probability of rtas related error is very low. But
* by returning EIO and printing additional information to dmsg the user
* will know that OProfile did not start and dmesg will tell them why.
* OProfile does not support returning errors on Stop. Not a huge issue
* since failure to reset the debug bus or stop the SPU PC collection is
* not a fatel issue. Chances are if the Stop failed, Start doesn't work
* either.
*/
/*
* Interpetation of hdw_thread:
* 0 - even virtual cpus 0, 2, 4,...
* 1 - odd virtual cpus 1, 3, 5, ...
*
* FIXME: this is strictly wrong, we need to clean this up in a number
* of places. It works for now. -arnd
*/
static u32 hdw_thread;
static u32 virt_cntr_inter_mask;
static struct timer_list timer_virt_cntr;
static struct timer_list timer_spu_event_swap;
/*
* pm_signal needs to be global since it is initialized in
* cell_reg_setup at the time when the necessary information
* is available.
*/
static struct pm_signal pm_signal[NR_PHYS_CTRS];
static int pm_rtas_token; /* token for debug bus setup call */
static int spu_rtas_token; /* token for SPU cycle profiling */
static u32 reset_value[NR_PHYS_CTRS];
static int num_counters;
static int oprofile_running;
static DEFINE_SPINLOCK(cntr_lock);
static u32 ctr_enabled;
static unsigned char input_bus[NUM_INPUT_BUS_WORDS];
/*
* Firmware interface functions
*/
static int
rtas_ibm_cbe_perftools(int subfunc, int passthru,
void *address, unsigned long length)
{
u64 paddr = __pa(address);
return rtas_call(pm_rtas_token, 5, 1, NULL, subfunc,
passthru, paddr >> 32, paddr & 0xffffffff, length);
}
static void pm_rtas_reset_signals(u32 node)
{
int ret;
struct pm_signal pm_signal_local;
/*
* The debug bus is being set to the passthru disable state.
* However, the FW still expects at least one legal signal routing
* entry or it will return an error on the arguments. If we don't
* supply a valid entry, we must ignore all return values. Ignoring
* all return values means we might miss an error we should be
* concerned about.
*/
/* fw expects physical cpu #. */
pm_signal_local.cpu = node;
pm_signal_local.signal_group = 21;
pm_signal_local.bus_word = 1;
pm_signal_local.sub_unit = 0;
pm_signal_local.bit = 0;
ret = rtas_ibm_cbe_perftools(SUBFUNC_RESET, PASSTHRU_DISABLE,
&pm_signal_local,
sizeof(struct pm_signal));
if (unlikely(ret))
/*
* Not a fatal error. For Oprofile stop, the oprofile
* functions do not support returning an error for
* failure to stop OProfile.
*/
printk(KERN_WARNING "%s: rtas returned: %d\n",
__func__, ret);
}
static int pm_rtas_activate_signals(u32 node, u32 count)
{
int ret;
int i, j;
struct pm_signal pm_signal_local[NR_PHYS_CTRS];
/*
* There is no debug setup required for the cycles event.
* Note that only events in the same group can be used.
* Otherwise, there will be conflicts in correctly routing
* the signals on the debug bus. It is the responsibility
* of the OProfile user tool to check the events are in
* the same group.
*/
i = 0;
for (j = 0; j < count; j++) {
if (pm_signal[j].signal_group != PPU_CYCLES_GRP_NUM) {
/* fw expects physical cpu # */
pm_signal_local[i].cpu = node;
pm_signal_local[i].signal_group
= pm_signal[j].signal_group;
pm_signal_local[i].bus_word = pm_signal[j].bus_word;
pm_signal_local[i].sub_unit = pm_signal[j].sub_unit;
pm_signal_local[i].bit = pm_signal[j].bit;
i++;
}
}
if (i != 0) {
ret = rtas_ibm_cbe_perftools(SUBFUNC_ACTIVATE, PASSTHRU_ENABLE,
pm_signal_local,
i * sizeof(struct pm_signal));
if (unlikely(ret)) {
printk(KERN_WARNING "%s: rtas returned: %d\n",
__func__, ret);
return -EIO;
}
}
return 0;
}
/*
* PM Signal functions
*/
static void set_pm_event(u32 ctr, int event, u32 unit_mask)
{
struct pm_signal *p;
u32 signal_bit;
u32 bus_word, bus_type, count_cycles, polarity, input_control;
int j, i;
if (event == PPU_CYCLES_EVENT_NUM) {
/* Special Event: Count all cpu cycles */
pm_regs.pm07_cntrl[ctr] = CBE_COUNT_ALL_CYCLES;
p = &(pm_signal[ctr]);
p->signal_group = PPU_CYCLES_GRP_NUM;
p->bus_word = 1;
p->sub_unit = 0;
p->bit = 0;
goto out;
} else {
pm_regs.pm07_cntrl[ctr] = 0;
}
bus_word = GET_BUS_WORD(unit_mask);
bus_type = GET_BUS_TYPE(unit_mask);
count_cycles = GET_COUNT_CYCLES(unit_mask);
polarity = GET_POLARITY(unit_mask);
input_control = GET_INPUT_CONTROL(unit_mask);
signal_bit = (event % 100);
p = &(pm_signal[ctr]);
p->signal_group = event / 100;
p->bus_word = bus_word;
p->sub_unit = GET_SUB_UNIT(unit_mask);
pm_regs.pm07_cntrl[ctr] = 0;
pm_regs.pm07_cntrl[ctr] |= PM07_CTR_COUNT_CYCLES(count_cycles);
pm_regs.pm07_cntrl[ctr] |= PM07_CTR_POLARITY(polarity);
pm_regs.pm07_cntrl[ctr] |= PM07_CTR_INPUT_CONTROL(input_control);
/*
* Some of the islands signal selection is based on 64 bit words.
* The debug bus words are 32 bits, the input words to the performance
* counters are defined as 32 bits. Need to convert the 64 bit island
* specification to the appropriate 32 input bit and bus word for the
* performance counter event selection. See the CELL Performance
* monitoring signals manual and the Perf cntr hardware descriptions
* for the details.
*/
if (input_control == 0) {
if (signal_bit > 31) {
signal_bit -= 32;
if (bus_word == 0x3)
bus_word = 0x2;
else if (bus_word == 0xc)
bus_word = 0x8;
}
if ((bus_type == 0) && p->signal_group >= 60)
bus_type = 2;
if ((bus_type == 1) && p->signal_group >= 50)
bus_type = 0;
pm_regs.pm07_cntrl[ctr] |= PM07_CTR_INPUT_MUX(signal_bit);
} else {
pm_regs.pm07_cntrl[ctr] = 0;
p->bit = signal_bit;
}
for (i = 0; i < NUM_DEBUG_BUS_WORDS; i++) {
if (bus_word & (1 << i)) {
pm_regs.debug_bus_control |=
(bus_type << (30 - (2 * i)));
for (j = 0; j < NUM_INPUT_BUS_WORDS; j++) {
if (input_bus[j] == 0xff) {
input_bus[j] = i;
pm_regs.group_control |=
(i << (30 - (2 * j)));
break;
}
}
}
}
out:
;
}
static void write_pm_cntrl(int cpu)
{
/*
* Oprofile will use 32 bit counters, set bits 7:10 to 0
* pmregs.pm_cntrl is a global
*/
u32 val = 0;
if (pm_regs.pm_cntrl.enable == 1)
val |= CBE_PM_ENABLE_PERF_MON;
if (pm_regs.pm_cntrl.stop_at_max == 1)
val |= CBE_PM_STOP_AT_MAX;
if (pm_regs.pm_cntrl.trace_mode != 0)
val |= CBE_PM_TRACE_MODE_SET(pm_regs.pm_cntrl.trace_mode);
if (pm_regs.pm_cntrl.trace_buf_ovflw == 1)
val |= CBE_PM_TRACE_BUF_OVFLW(pm_regs.pm_cntrl.trace_buf_ovflw);
if (pm_regs.pm_cntrl.freeze == 1)
val |= CBE_PM_FREEZE_ALL_CTRS;
val |= CBE_PM_SPU_ADDR_TRACE_SET(pm_regs.pm_cntrl.spu_addr_trace);
/*
* Routine set_count_mode must be called previously to set
* the count mode based on the user selection of user and kernel.
*/
val |= CBE_PM_COUNT_MODE_SET(pm_regs.pm_cntrl.count_mode);
cbe_write_pm(cpu, pm_control, val);
}
static inline void
set_count_mode(u32 kernel, u32 user)
{
/*
* The user must specify user and kernel if they want them. If
* neither is specified, OProfile will count in hypervisor mode.
* pm_regs.pm_cntrl is a global
*/
if (kernel) {
if (user)
pm_regs.pm_cntrl.count_mode = CBE_COUNT_ALL_MODES;
else
pm_regs.pm_cntrl.count_mode =
CBE_COUNT_SUPERVISOR_MODE;
} else {
if (user)
pm_regs.pm_cntrl.count_mode = CBE_COUNT_PROBLEM_MODE;
else
pm_regs.pm_cntrl.count_mode =
CBE_COUNT_HYPERVISOR_MODE;
}
}
static inline void enable_ctr(u32 cpu, u32 ctr, u32 *pm07_cntrl)
{
pm07_cntrl[ctr] |= CBE_PM_CTR_ENABLE;
cbe_write_pm07_control(cpu, ctr, pm07_cntrl[ctr]);
}
/*
* Oprofile is expected to collect data on all CPUs simultaneously.
* However, there is one set of performance counters per node. There are
* two hardware threads or virtual CPUs on each node. Hence, OProfile must
* multiplex in time the performance counter collection on the two virtual
* CPUs. The multiplexing of the performance counters is done by this
* virtual counter routine.
*
* The pmc_values used below is defined as 'per-cpu' but its use is
* more akin to 'per-node'. We need to store two sets of counter
* values per node -- one for the previous run and one for the next.
* The per-cpu[NR_PHYS_CTRS] gives us the storage we need. Each odd/even
* pair of per-cpu arrays is used for storing the previous and next
* pmc values for a given node.
* NOTE: We use the per-cpu variable to improve cache performance.
*
* This routine will alternate loading the virtual counters for
* virtual CPUs
*/
static void cell_virtual_cntr(struct timer_list *unused)
{
int i, prev_hdw_thread, next_hdw_thread;
u32 cpu;
unsigned long flags;
/*
* Make sure that the interrupt_hander and the virt counter are
* not both playing with the counters on the same node.
*/
spin_lock_irqsave(&cntr_lock, flags);
prev_hdw_thread = hdw_thread;
/* switch the cpu handling the interrupts */
hdw_thread = 1 ^ hdw_thread;
next_hdw_thread = hdw_thread;
pm_regs.group_control = 0;
pm_regs.debug_bus_control = 0;
for (i = 0; i < NUM_INPUT_BUS_WORDS; i++)
input_bus[i] = 0xff;
/*
* There are some per thread events. Must do the
* set event, for the thread that is being started
*/
for (i = 0; i < num_counters; i++)
set_pm_event(i,
pmc_cntrl[next_hdw_thread][i].evnts,
pmc_cntrl[next_hdw_thread][i].masks);
/*
* The following is done only once per each node, but
* we need cpu #, not node #, to pass to the cbe_xxx functions.
*/
for_each_online_cpu(cpu) {
if (cbe_get_hw_thread_id(cpu))
continue;
/*
* stop counters, save counter values, restore counts
* for previous thread
*/
cbe_disable_pm(cpu);
cbe_disable_pm_interrupts(cpu);
for (i = 0; i < num_counters; i++) {
per_cpu(pmc_values, cpu + prev_hdw_thread)[i]
= cbe_read_ctr(cpu, i);
if (per_cpu(pmc_values, cpu + next_hdw_thread)[i]
== 0xFFFFFFFF)
/* If the cntr value is 0xffffffff, we must
* reset that to 0xfffffff0 when the current
* thread is restarted. This will generate a
* new interrupt and make sure that we never
* restore the counters to the max value. If
* the counters were restored to the max value,
* they do not increment and no interrupts are
* generated. Hence no more samples will be
* collected on that cpu.
*/
cbe_write_ctr(cpu, i, 0xFFFFFFF0);
else
cbe_write_ctr(cpu, i,
per_cpu(pmc_values,
cpu +
next_hdw_thread)[i]);
}
/*
* Switch to the other thread. Change the interrupt
* and control regs to be scheduled on the CPU
* corresponding to the thread to execute.
*/
for (i = 0; i < num_counters; i++) {
if (pmc_cntrl[next_hdw_thread][i].enabled) {
/*
* There are some per thread events.
* Must do the set event, enable_cntr
* for each cpu.
*/
enable_ctr(cpu, i,
pm_regs.pm07_cntrl);
} else {
cbe_write_pm07_control(cpu, i, 0);
}
}
/* Enable interrupts on the CPU thread that is starting */
cbe_enable_pm_interrupts(cpu, next_hdw_thread,
virt_cntr_inter_mask);
cbe_enable_pm(cpu);
}
spin_unlock_irqrestore(&cntr_lock, flags);
mod_timer(&timer_virt_cntr, jiffies + HZ / 10);
}
static void start_virt_cntrs(void)
{
timer_setup(&timer_virt_cntr, cell_virtual_cntr, 0);
timer_virt_cntr.expires = jiffies + HZ / 10;
add_timer(&timer_virt_cntr);
}
static int cell_reg_setup_spu_cycles(struct op_counter_config *ctr,
struct op_system_config *sys, int num_ctrs)
{
spu_cycle_reset = ctr[0].count;
/*
* Each node will need to make the rtas call to start
* and stop SPU profiling. Get the token once and store it.
*/
spu_rtas_token = rtas_token("ibm,cbe-spu-perftools");
if (unlikely(spu_rtas_token == RTAS_UNKNOWN_SERVICE)) {
printk(KERN_ERR
"%s: rtas token ibm,cbe-spu-perftools unknown\n",
__func__);
return -EIO;
}
return 0;
}
/* Unfortunately, the hardware will only support event profiling
* on one SPU per node at a time. Therefore, we must time slice
* the profiling across all SPUs in the node. Note, we do this
* in parallel for each node. The following routine is called
* periodically based on kernel timer to switch which SPU is
* being monitored in a round robbin fashion.
*/
static void spu_evnt_swap(struct timer_list *unused)
{
int node;
int cur_phys_spu, nxt_phys_spu, cur_spu_evnt_phys_spu_indx;
unsigned long flags;
int cpu;
int ret;
u32 interrupt_mask;
/* enable interrupts on cntr 0 */
interrupt_mask = CBE_PM_CTR_OVERFLOW_INTR(0);
hdw_thread = 0;
/* Make sure spu event interrupt handler and spu event swap
* don't access the counters simultaneously.
*/
spin_lock_irqsave(&cntr_lock, flags);
cur_spu_evnt_phys_spu_indx = spu_evnt_phys_spu_indx;
if (++(spu_evnt_phys_spu_indx) == NUM_SPUS_PER_NODE)
spu_evnt_phys_spu_indx = 0;
pm_signal[0].sub_unit = spu_evnt_phys_spu_indx;
pm_signal[1].sub_unit = spu_evnt_phys_spu_indx;
pm_signal[2].sub_unit = spu_evnt_phys_spu_indx;
/* switch the SPU being profiled on each node */
for_each_online_cpu(cpu) {
if (cbe_get_hw_thread_id(cpu))
continue;
node = cbe_cpu_to_node(cpu);
cur_phys_spu = (node * NUM_SPUS_PER_NODE)
+ cur_spu_evnt_phys_spu_indx;
nxt_phys_spu = (node * NUM_SPUS_PER_NODE)
+ spu_evnt_phys_spu_indx;
/*
* stop counters, save counter values, restore counts
* for previous physical SPU
*/
cbe_disable_pm(cpu);
cbe_disable_pm_interrupts(cpu);
spu_pm_cnt[cur_phys_spu]
= cbe_read_ctr(cpu, 0);
/* restore previous count for the next spu to sample */
/* NOTE, hardware issue, counter will not start if the
* counter value is at max (0xFFFFFFFF).
*/
if (spu_pm_cnt[nxt_phys_spu] >= 0xFFFFFFFF)
cbe_write_ctr(cpu, 0, 0xFFFFFFF0);
else
cbe_write_ctr(cpu, 0, spu_pm_cnt[nxt_phys_spu]);
pm_rtas_reset_signals(cbe_cpu_to_node(cpu));
/* setup the debug bus measure the one event and
* the two events to route the next SPU's PC on
* the debug bus
*/
ret = pm_rtas_activate_signals(cbe_cpu_to_node(cpu), 3);
if (ret)
printk(KERN_ERR "%s: pm_rtas_activate_signals failed, "
"SPU event swap\n", __func__);
/* clear the trace buffer, don't want to take PC for
* previous SPU*/
cbe_write_pm(cpu, trace_address, 0);
enable_ctr(cpu, 0, pm_regs.pm07_cntrl);
/* Enable interrupts on the CPU thread that is starting */
cbe_enable_pm_interrupts(cpu, hdw_thread,
interrupt_mask);
cbe_enable_pm(cpu);
}
spin_unlock_irqrestore(&cntr_lock, flags);
/* swap approximately every 0.1 seconds */
mod_timer(&timer_spu_event_swap, jiffies + HZ / 25);
}
static void start_spu_event_swap(void)
{
timer_setup(&timer_spu_event_swap, spu_evnt_swap, 0);
timer_spu_event_swap.expires = jiffies + HZ / 25;
add_timer(&timer_spu_event_swap);
}
static int cell_reg_setup_spu_events(struct op_counter_config *ctr,
struct op_system_config *sys, int num_ctrs)
{
int i;
/* routine is called once for all nodes */
spu_evnt_phys_spu_indx = 0;
/*
* For all events except PPU CYCLEs, each node will need to make
* the rtas cbe-perftools call to setup and reset the debug bus.
* Make the token lookup call once and store it in the global
* variable pm_rtas_token.
*/
pm_rtas_token = rtas_token("ibm,cbe-perftools");
if (unlikely(pm_rtas_token == RTAS_UNKNOWN_SERVICE)) {
printk(KERN_ERR
"%s: rtas token ibm,cbe-perftools unknown\n",
__func__);
return -EIO;
}
/* setup the pm_control register settings,
* settings will be written per node by the
* cell_cpu_setup() function.
*/
pm_regs.pm_cntrl.trace_buf_ovflw = 1;
/* Use the occurrence trace mode to have SPU PC saved
* to the trace buffer. Occurrence data in trace buffer
* is not used. Bit 2 must be set to store SPU addresses.
*/
pm_regs.pm_cntrl.trace_mode = 2;
pm_regs.pm_cntrl.spu_addr_trace = 0x1; /* using debug bus
event 2 & 3 */
/* setup the debug bus event array with the SPU PC routing events.
* Note, pm_signal[0] will be filled in by set_pm_event() call below.
*/
pm_signal[1].signal_group = SPU_PROFILE_EVENT_ADDR / 100;
pm_signal[1].bus_word = GET_BUS_WORD(SPU_PROFILE_EVENT_ADDR_MASK_A);
pm_signal[1].bit = SPU_PROFILE_EVENT_ADDR % 100;
pm_signal[1].sub_unit = spu_evnt_phys_spu_indx;
pm_signal[2].signal_group = SPU_PROFILE_EVENT_ADDR / 100;
pm_signal[2].bus_word = GET_BUS_WORD(SPU_PROFILE_EVENT_ADDR_MASK_B);
pm_signal[2].bit = SPU_PROFILE_EVENT_ADDR % 100;
pm_signal[2].sub_unit = spu_evnt_phys_spu_indx;
/* Set the user selected spu event to profile on,
* note, only one SPU profiling event is supported
*/
num_counters = 1; /* Only support one SPU event at a time */
set_pm_event(0, ctr[0].event, ctr[0].unit_mask);
reset_value[0] = 0xFFFFFFFF - ctr[0].count;
/* global, used by cell_cpu_setup */
ctr_enabled |= 1;
/* Initialize the count for each SPU to the reset value */
for (i=0; i < MAX_NUMNODES * NUM_SPUS_PER_NODE; i++)
spu_pm_cnt[i] = reset_value[0];
return 0;
}
static int cell_reg_setup_ppu(struct op_counter_config *ctr,
struct op_system_config *sys, int num_ctrs)
{
/* routine is called once for all nodes */
int i, j, cpu;
num_counters = num_ctrs;
if (unlikely(num_ctrs > NR_PHYS_CTRS)) {
printk(KERN_ERR
"%s: Oprofile, number of specified events " \
"exceeds number of physical counters\n",
__func__);
return -EIO;
}
set_count_mode(sys->enable_kernel, sys->enable_user);
/* Setup the thread 0 events */
for (i = 0; i < num_ctrs; ++i) {
pmc_cntrl[0][i].evnts = ctr[i].event;
pmc_cntrl[0][i].masks = ctr[i].unit_mask;
pmc_cntrl[0][i].enabled = ctr[i].enabled;
pmc_cntrl[0][i].vcntr = i;
for_each_possible_cpu(j)
per_cpu(pmc_values, j)[i] = 0;
}
/*
* Setup the thread 1 events, map the thread 0 event to the
* equivalent thread 1 event.
*/
for (i = 0; i < num_ctrs; ++i) {
if ((ctr[i].event >= 2100) && (ctr[i].event <= 2111))
pmc_cntrl[1][i].evnts = ctr[i].event + 19;
else if (ctr[i].event == 2203)
pmc_cntrl[1][i].evnts = ctr[i].event;
else if ((ctr[i].event >= 2200) && (ctr[i].event <= 2215))
pmc_cntrl[1][i].evnts = ctr[i].event + 16;
else
pmc_cntrl[1][i].evnts = ctr[i].event;
pmc_cntrl[1][i].masks = ctr[i].unit_mask;
pmc_cntrl[1][i].enabled = ctr[i].enabled;
pmc_cntrl[1][i].vcntr = i;
}
for (i = 0; i < NUM_INPUT_BUS_WORDS; i++)
input_bus[i] = 0xff;
/*
* Our counters count up, and "count" refers to
* how much before the next interrupt, and we interrupt
* on overflow. So we calculate the starting value
* which will give us "count" until overflow.
* Then we set the events on the enabled counters.
*/
for (i = 0; i < num_counters; ++i) {
/* start with virtual counter set 0 */
if (pmc_cntrl[0][i].enabled) {
/* Using 32bit counters, reset max - count */
reset_value[i] = 0xFFFFFFFF - ctr[i].count;
set_pm_event(i,
pmc_cntrl[0][i].evnts,
pmc_cntrl[0][i].masks);
/* global, used by cell_cpu_setup */
ctr_enabled |= (1 << i);
}
}
/* initialize the previous counts for the virtual cntrs */
for_each_online_cpu(cpu)
for (i = 0; i < num_counters; ++i) {
per_cpu(pmc_values, cpu)[i] = reset_value[i];
}
return 0;
}
/* This function is called once for all cpus combined */
static int cell_reg_setup(struct op_counter_config *ctr,
struct op_system_config *sys, int num_ctrs)
{
int ret=0;
spu_cycle_reset = 0;
/* initialize the spu_arr_trace value, will be reset if
* doing spu event profiling.
*/
pm_regs.group_control = 0;
pm_regs.debug_bus_control = 0;
pm_regs.pm_cntrl.stop_at_max = 1;
pm_regs.pm_cntrl.trace_mode = 0;
pm_regs.pm_cntrl.freeze = 1;
pm_regs.pm_cntrl.trace_buf_ovflw = 0;
pm_regs.pm_cntrl.spu_addr_trace = 0;
/*
* For all events except PPU CYCLEs, each node will need to make
* the rtas cbe-perftools call to setup and reset the debug bus.
* Make the token lookup call once and store it in the global
* variable pm_rtas_token.
*/
pm_rtas_token = rtas_token("ibm,cbe-perftools");
if (unlikely(pm_rtas_token == RTAS_UNKNOWN_SERVICE)) {
printk(KERN_ERR
"%s: rtas token ibm,cbe-perftools unknown\n",
__func__);
return -EIO;
}
if (ctr[0].event == SPU_CYCLES_EVENT_NUM) {
profiling_mode = SPU_PROFILING_CYCLES;
ret = cell_reg_setup_spu_cycles(ctr, sys, num_ctrs);
} else if ((ctr[0].event >= SPU_EVENT_NUM_START) &&
(ctr[0].event <= SPU_EVENT_NUM_STOP)) {
profiling_mode = SPU_PROFILING_EVENTS;
spu_cycle_reset = ctr[0].count;
/* for SPU event profiling, need to setup the
* pm_signal array with the events to route the
* SPU PC before making the FW call. Note, only
* one SPU event for profiling can be specified
* at a time.
*/
cell_reg_setup_spu_events(ctr, sys, num_ctrs);
} else {
profiling_mode = PPU_PROFILING;
ret = cell_reg_setup_ppu(ctr, sys, num_ctrs);
}
return ret;
}
/* This function is called once for each cpu */
static int cell_cpu_setup(struct op_counter_config *cntr)
{
u32 cpu = smp_processor_id();
u32 num_enabled = 0;
int i;
int ret;
/* Cycle based SPU profiling does not use the performance
* counters. The trace array is configured to collect
* the data.
*/
if (profiling_mode == SPU_PROFILING_CYCLES)
return 0;
/* There is one performance monitor per processor chip (i.e. node),
* so we only need to perform this function once per node.
*/
if (cbe_get_hw_thread_id(cpu))
return 0;
/* Stop all counters */
cbe_disable_pm(cpu);
cbe_disable_pm_interrupts(cpu);
cbe_write_pm(cpu, pm_start_stop, 0);
cbe_write_pm(cpu, group_control, pm_regs.group_control);
cbe_write_pm(cpu, debug_bus_control, pm_regs.debug_bus_control);
write_pm_cntrl(cpu);
for (i = 0; i < num_counters; ++i) {
if (ctr_enabled & (1 << i)) {
pm_signal[num_enabled].cpu = cbe_cpu_to_node(cpu);
num_enabled++;
}
}
/*
* The pm_rtas_activate_signals will return -EIO if the FW
* call failed.
*/
if (profiling_mode == SPU_PROFILING_EVENTS) {
/* For SPU event profiling also need to setup the
* pm interval timer
*/
ret = pm_rtas_activate_signals(cbe_cpu_to_node(cpu),
num_enabled+2);
/* store PC from debug bus to Trace buffer as often
* as possible (every 10 cycles)
*/
cbe_write_pm(cpu, pm_interval, NUM_INTERVAL_CYC);
return ret;
} else
return pm_rtas_activate_signals(cbe_cpu_to_node(cpu),
num_enabled);
}
#define ENTRIES 303
#define MAXLFSR 0xFFFFFF
/* precomputed table of 24 bit LFSR values */
static int initial_lfsr[] = {
8221349, 12579195, 5379618, 10097839, 7512963, 7519310, 3955098, 10753424,
15507573, 7458917, 285419, 2641121, 9780088, 3915503, 6668768, 1548716,
4885000, 8774424, 9650099, 2044357, 2304411, 9326253, 10332526, 4421547,
3440748, 10179459, 13332843, 10375561, 1313462, 8375100, 5198480, 6071392,
9341783, 1526887, 3985002, 1439429, 13923762, 7010104, 11969769, 4547026,
2040072, 4025602, 3437678, 7939992, 11444177, 4496094, 9803157, 10745556,
3671780, 4257846, 5662259, 13196905, 3237343, 12077182, 16222879, 7587769,
14706824, 2184640, 12591135, 10420257, 7406075, 3648978, 11042541, 15906893,
11914928, 4732944, 10695697, 12928164, 11980531, 4430912, 11939291, 2917017,
6119256, 4172004, 9373765, 8410071, 14788383, 5047459, 5474428, 1737756,
15967514, 13351758, 6691285, 8034329, 2856544, 14394753, 11310160, 12149558,
7487528, 7542781, 15668898, 12525138, 12790975, 3707933, 9106617, 1965401,
16219109, 12801644, 2443203, 4909502, 8762329, 3120803, 6360315, 9309720,
15164599, 10844842, 4456529, 6667610, 14924259, 884312, 6234963, 3326042,
15973422, 13919464, 5272099, 6414643, 3909029, 2764324, 5237926, 4774955,
10445906, 4955302, 5203726, 10798229, 11443419, 2303395, 333836, 9646934,
3464726, 4159182, 568492, 995747, 10318756, 13299332, 4836017, 8237783,
3878992, 2581665, 11394667, 5672745, 14412947, 3159169, 9094251, 16467278,
8671392, 15230076, 4843545, 7009238, 15504095, 1494895, 9627886, 14485051,
8304291, 252817, 12421642, 16085736, 4774072, 2456177, 4160695, 15409741,
4902868, 5793091, 13162925, 16039714, 782255, 11347835, 14884586, 366972,
16308990, 11913488, 13390465, 2958444, 10340278, 1177858, 1319431, 10426302,
2868597, 126119, 5784857, 5245324, 10903900, 16436004, 3389013, 1742384,
14674502, 10279218, 8536112, 10364279, 6877778, 14051163, 1025130, 6072469,
1988305, 8354440, 8216060, 16342977, 13112639, 3976679, 5913576, 8816697,
6879995, 14043764, 3339515, 9364420, 15808858, 12261651, 2141560, 5636398,
10345425, 10414756, 781725, 6155650, 4746914, 5078683, 7469001, 6799140,
10156444, 9667150, 10116470, 4133858, 2121972, 1124204, 1003577, 1611214,
14304602, 16221850, 13878465, 13577744, 3629235, 8772583, 10881308, 2410386,
7300044, 5378855, 9301235, 12755149, 4977682, 8083074, 10327581, 6395087,
9155434, 15501696, 7514362, 14520507, 15808945, 3244584, 4741962, 9658130,
14336147, 8654727, 7969093, 15759799, 14029445, 5038459, 9894848, 8659300,
13699287, 8834306, 10712885, 14753895, 10410465, 3373251, 309501, 9561475,
5526688, 14647426, 14209836, 5339224, 207299, 14069911, 8722990, 2290950,
3258216, 12505185, 6007317, 9218111, 14661019, 10537428, 11731949, 9027003,
6641507, 9490160, 200241, 9720425, 16277895, 10816638, 1554761, 10431375,
7467528, 6790302, 3429078, 14633753, 14428997, 11463204, 3576212, 2003426,
6123687, 820520, 9992513, 15784513, 5778891, 6428165, 8388607
};
/*
* The hardware uses an LFSR counting sequence to determine when to capture
* the SPU PCs. An LFSR sequence is like a puesdo random number sequence
* where each number occurs once in the sequence but the sequence is not in