aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/cpu
diff options
context:
space:
mode:
authorPaul Mundt <lethal@linux-sh.org>2012-08-01 00:49:13 -0400
committerPaul Mundt <lethal@linux-sh.org>2012-08-01 00:49:13 -0400
commit91ba548cfd5cc8ee93b9435527efb8fa4caf5c5e (patch)
treec96ed92413044a28d17783e84a8824bfd2437af1 /arch/x86/kernel/cpu
parentb9ccfda293ee6fca9a89a1584f0900e0627b975e (diff)
parent4dc4c51675c137c30838425ecc8d471ff5eb138b (diff)
Merge branch 'sh/dmaengine' into sh-latest
Diffstat (limited to 'arch/x86/kernel/cpu')
-rw-r--r--arch/x86/kernel/cpu/Makefile2
-rw-r--r--arch/x86/kernel/cpu/common.c31
-rw-r--r--arch/x86/kernel/cpu/cpu.h9
-rw-r--r--arch/x86/kernel/cpu/intel.c176
-rw-r--r--arch/x86/kernel/cpu/perf_event.h2
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel.c92
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_uncore.c1316
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_uncore.h207
-rw-r--r--arch/x86/kernel/cpu/sched.c55
9 files changed, 1689 insertions, 201 deletions
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index bac4c3804cc7..d30a6a9a0121 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -14,7 +14,7 @@ CFLAGS_common.o := $(nostackp)
14 14
15obj-y := intel_cacheinfo.o scattered.o topology.o 15obj-y := intel_cacheinfo.o scattered.o topology.o
16obj-y += proc.o capflags.o powerflags.o common.o 16obj-y += proc.o capflags.o powerflags.o common.o
17obj-y += vmware.o hypervisor.o sched.o mshyperv.o 17obj-y += vmware.o hypervisor.o mshyperv.o
18obj-y += rdrand.o 18obj-y += rdrand.o
19obj-y += match.o 19obj-y += match.o
20 20
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 5bbc082c47ad..46d8786d655e 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -452,6 +452,35 @@ void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
452 c->x86_cache_size = l2size; 452 c->x86_cache_size = l2size;
453} 453}
454 454
455u16 __read_mostly tlb_lli_4k[NR_INFO];
456u16 __read_mostly tlb_lli_2m[NR_INFO];
457u16 __read_mostly tlb_lli_4m[NR_INFO];
458u16 __read_mostly tlb_lld_4k[NR_INFO];
459u16 __read_mostly tlb_lld_2m[NR_INFO];
460u16 __read_mostly tlb_lld_4m[NR_INFO];
461
462/*
463 * tlb_flushall_shift shows the balance point in replacing cr3 write
464 * with multiple 'invlpg'. It will do this replacement when
465 * flush_tlb_lines <= active_lines/2^tlb_flushall_shift.
466 * If tlb_flushall_shift is -1, means the replacement will be disabled.
467 */
468s8 __read_mostly tlb_flushall_shift = -1;
469
470void __cpuinit cpu_detect_tlb(struct cpuinfo_x86 *c)
471{
472 if (this_cpu->c_detect_tlb)
473 this_cpu->c_detect_tlb(c);
474
475 printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \
476 "Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \
477 "tlb_flushall_shift is 0x%x\n",
478 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
479 tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
480 tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
481 tlb_flushall_shift);
482}
483
455void __cpuinit detect_ht(struct cpuinfo_x86 *c) 484void __cpuinit detect_ht(struct cpuinfo_x86 *c)
456{ 485{
457#ifdef CONFIG_X86_HT 486#ifdef CONFIG_X86_HT
@@ -911,6 +940,8 @@ void __init identify_boot_cpu(void)
911#else 940#else
912 vgetcpu_set_mode(); 941 vgetcpu_set_mode();
913#endif 942#endif
943 if (boot_cpu_data.cpuid_level >= 2)
944 cpu_detect_tlb(&boot_cpu_data);
914} 945}
915 946
916void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) 947void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
diff --git a/arch/x86/kernel/cpu/cpu.h b/arch/x86/kernel/cpu/cpu.h
index 8bacc7826fb3..4041c24ae7db 100644
--- a/arch/x86/kernel/cpu/cpu.h
+++ b/arch/x86/kernel/cpu/cpu.h
@@ -20,10 +20,19 @@ struct cpu_dev {
20 void (*c_bsp_init)(struct cpuinfo_x86 *); 20 void (*c_bsp_init)(struct cpuinfo_x86 *);
21 void (*c_init)(struct cpuinfo_x86 *); 21 void (*c_init)(struct cpuinfo_x86 *);
22 void (*c_identify)(struct cpuinfo_x86 *); 22 void (*c_identify)(struct cpuinfo_x86 *);
23 void (*c_detect_tlb)(struct cpuinfo_x86 *);
23 unsigned int (*c_size_cache)(struct cpuinfo_x86 *, unsigned int); 24 unsigned int (*c_size_cache)(struct cpuinfo_x86 *, unsigned int);
24 int c_x86_vendor; 25 int c_x86_vendor;
25}; 26};
26 27
28struct _tlb_table {
29 unsigned char descriptor;
30 char tlb_type;
31 unsigned int entries;
32 /* unsigned int ways; */
33 char info[128];
34};
35
27#define cpu_dev_register(cpu_devX) \ 36#define cpu_dev_register(cpu_devX) \
28 static const struct cpu_dev *const __cpu_dev_##cpu_devX __used \ 37 static const struct cpu_dev *const __cpu_dev_##cpu_devX __used \
29 __attribute__((__section__(".x86_cpu_dev.init"))) = \ 38 __attribute__((__section__(".x86_cpu_dev.init"))) = \
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index 3e6ff6cbf42a..0a4ce2980a5a 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -491,6 +491,181 @@ static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned i
491} 491}
492#endif 492#endif
493 493
494#define TLB_INST_4K 0x01
495#define TLB_INST_4M 0x02
496#define TLB_INST_2M_4M 0x03
497
498#define TLB_INST_ALL 0x05
499#define TLB_INST_1G 0x06
500
501#define TLB_DATA_4K 0x11
502#define TLB_DATA_4M 0x12
503#define TLB_DATA_2M_4M 0x13
504#define TLB_DATA_4K_4M 0x14
505
506#define TLB_DATA_1G 0x16
507
508#define TLB_DATA0_4K 0x21
509#define TLB_DATA0_4M 0x22
510#define TLB_DATA0_2M_4M 0x23
511
512#define STLB_4K 0x41
513
514static const struct _tlb_table intel_tlb_table[] __cpuinitconst = {
515 { 0x01, TLB_INST_4K, 32, " TLB_INST 4 KByte pages, 4-way set associative" },
516 { 0x02, TLB_INST_4M, 2, " TLB_INST 4 MByte pages, full associative" },
517 { 0x03, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way set associative" },
518 { 0x04, TLB_DATA_4M, 8, " TLB_DATA 4 MByte pages, 4-way set associative" },
519 { 0x05, TLB_DATA_4M, 32, " TLB_DATA 4 MByte pages, 4-way set associative" },
520 { 0x0b, TLB_INST_4M, 4, " TLB_INST 4 MByte pages, 4-way set associative" },
521 { 0x4f, TLB_INST_4K, 32, " TLB_INST 4 KByte pages */" },
522 { 0x50, TLB_INST_ALL, 64, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
523 { 0x51, TLB_INST_ALL, 128, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
524 { 0x52, TLB_INST_ALL, 256, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
525 { 0x55, TLB_INST_2M_4M, 7, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
526 { 0x56, TLB_DATA0_4M, 16, " TLB_DATA0 4 MByte pages, 4-way set associative" },
527 { 0x57, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, 4-way associative" },
528 { 0x59, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, fully associative" },
529 { 0x5a, TLB_DATA0_2M_4M, 32, " TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
530 { 0x5b, TLB_DATA_4K_4M, 64, " TLB_DATA 4 KByte and 4 MByte pages" },
531 { 0x5c, TLB_DATA_4K_4M, 128, " TLB_DATA 4 KByte and 4 MByte pages" },
532 { 0x5d, TLB_DATA_4K_4M, 256, " TLB_DATA 4 KByte and 4 MByte pages" },
533 { 0xb0, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 4-way set associative" },
534 { 0xb1, TLB_INST_2M_4M, 4, " TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
535 { 0xb2, TLB_INST_4K, 64, " TLB_INST 4KByte pages, 4-way set associative" },
536 { 0xb3, TLB_DATA_4K, 128, " TLB_DATA 4 KByte pages, 4-way set associative" },
537 { 0xb4, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 4-way associative" },
538 { 0xba, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way associative" },
539 { 0xc0, TLB_DATA_4K_4M, 8, " TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
540 { 0xca, STLB_4K, 512, " STLB 4 KByte pages, 4-way associative" },
541 { 0x00, 0, 0 }
542};
543
544static void __cpuinit intel_tlb_lookup(const unsigned char desc)
545{
546 unsigned char k;
547 if (desc == 0)
548 return;
549
550 /* look up this descriptor in the table */
551 for (k = 0; intel_tlb_table[k].descriptor != desc && \
552 intel_tlb_table[k].descriptor != 0; k++)
553 ;
554
555 if (intel_tlb_table[k].tlb_type == 0)
556 return;
557
558 switch (intel_tlb_table[k].tlb_type) {
559 case STLB_4K:
560 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
561 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
562 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
563 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
564 break;
565 case TLB_INST_ALL:
566 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
567 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
568 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
569 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
570 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
571 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
572 break;
573 case TLB_INST_4K:
574 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
575 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
576 break;
577 case TLB_INST_4M:
578 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
579 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
580 break;
581 case TLB_INST_2M_4M:
582 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
583 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
584 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
585 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
586 break;
587 case TLB_DATA_4K:
588 case TLB_DATA0_4K:
589 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
590 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
591 break;
592 case TLB_DATA_4M:
593 case TLB_DATA0_4M:
594 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
595 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
596 break;
597 case TLB_DATA_2M_4M:
598 case TLB_DATA0_2M_4M:
599 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
600 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
601 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
602 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
603 break;
604 case TLB_DATA_4K_4M:
605 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
606 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
607 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
608 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
609 break;
610 }
611}
612
613static void __cpuinit intel_tlb_flushall_shift_set(struct cpuinfo_x86 *c)
614{
615 if (!cpu_has_invlpg) {
616 tlb_flushall_shift = -1;
617 return;
618 }
619 switch ((c->x86 << 8) + c->x86_model) {
620 case 0x60f: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
621 case 0x616: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
622 case 0x617: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
623 case 0x61d: /* six-core 45 nm xeon "Dunnington" */
624 tlb_flushall_shift = -1;
625 break;
626 case 0x61a: /* 45 nm nehalem, "Bloomfield" */
627 case 0x61e: /* 45 nm nehalem, "Lynnfield" */
628 case 0x625: /* 32 nm nehalem, "Clarkdale" */
629 case 0x62c: /* 32 nm nehalem, "Gulftown" */
630 case 0x62e: /* 45 nm nehalem-ex, "Beckton" */
631 case 0x62f: /* 32 nm Xeon E7 */
632 tlb_flushall_shift = 6;
633 break;
634 case 0x62a: /* SandyBridge */
635 case 0x62d: /* SandyBridge, "Romely-EP" */
636 tlb_flushall_shift = 5;
637 break;
638 case 0x63a: /* Ivybridge */
639 tlb_flushall_shift = 1;
640 break;
641 default:
642 tlb_flushall_shift = 6;
643 }
644}
645
646static void __cpuinit intel_detect_tlb(struct cpuinfo_x86 *c)
647{
648 int i, j, n;
649 unsigned int regs[4];
650 unsigned char *desc = (unsigned char *)regs;
651 /* Number of times to iterate */
652 n = cpuid_eax(2) & 0xFF;
653
654 for (i = 0 ; i < n ; i++) {
655 cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
656
657 /* If bit 31 is set, this is an unknown format */
658 for (j = 0 ; j < 3 ; j++)
659 if (regs[j] & (1 << 31))
660 regs[j] = 0;
661
662 /* Byte 0 is level count, not a descriptor */
663 for (j = 1 ; j < 16 ; j++)
664 intel_tlb_lookup(desc[j]);
665 }
666 intel_tlb_flushall_shift_set(c);
667}
668
494static const struct cpu_dev __cpuinitconst intel_cpu_dev = { 669static const struct cpu_dev __cpuinitconst intel_cpu_dev = {
495 .c_vendor = "Intel", 670 .c_vendor = "Intel",
496 .c_ident = { "GenuineIntel" }, 671 .c_ident = { "GenuineIntel" },
@@ -546,6 +721,7 @@ static const struct cpu_dev __cpuinitconst intel_cpu_dev = {
546 }, 721 },
547 .c_size_cache = intel_size_cache, 722 .c_size_cache = intel_size_cache,
548#endif 723#endif
724 .c_detect_tlb = intel_detect_tlb,
549 .c_early_init = early_init_intel, 725 .c_early_init = early_init_intel,
550 .c_init = init_intel, 726 .c_init = init_intel,
551 .c_x86_vendor = X86_VENDOR_INTEL, 727 .c_x86_vendor = X86_VENDOR_INTEL,
diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
index a15df4be151f..821d53b696d1 100644
--- a/arch/x86/kernel/cpu/perf_event.h
+++ b/arch/x86/kernel/cpu/perf_event.h
@@ -374,7 +374,7 @@ struct x86_pmu {
374 /* 374 /*
375 * Intel DebugStore bits 375 * Intel DebugStore bits
376 */ 376 */
377 int bts :1, 377 unsigned int bts :1,
378 bts_active :1, 378 bts_active :1,
379 pebs :1, 379 pebs :1,
380 pebs_active :1, 380 pebs_active :1,
diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index 7a8b9d0abcaa..382366977d4c 100644
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -138,6 +138,84 @@ static u64 intel_pmu_event_map(int hw_event)
138 return intel_perfmon_event_map[hw_event]; 138 return intel_perfmon_event_map[hw_event];
139} 139}
140 140
141#define SNB_DMND_DATA_RD (1ULL << 0)
142#define SNB_DMND_RFO (1ULL << 1)
143#define SNB_DMND_IFETCH (1ULL << 2)
144#define SNB_DMND_WB (1ULL << 3)
145#define SNB_PF_DATA_RD (1ULL << 4)
146#define SNB_PF_RFO (1ULL << 5)
147#define SNB_PF_IFETCH (1ULL << 6)
148#define SNB_LLC_DATA_RD (1ULL << 7)
149#define SNB_LLC_RFO (1ULL << 8)
150#define SNB_LLC_IFETCH (1ULL << 9)
151#define SNB_BUS_LOCKS (1ULL << 10)
152#define SNB_STRM_ST (1ULL << 11)
153#define SNB_OTHER (1ULL << 15)
154#define SNB_RESP_ANY (1ULL << 16)
155#define SNB_NO_SUPP (1ULL << 17)
156#define SNB_LLC_HITM (1ULL << 18)
157#define SNB_LLC_HITE (1ULL << 19)
158#define SNB_LLC_HITS (1ULL << 20)
159#define SNB_LLC_HITF (1ULL << 21)
160#define SNB_LOCAL (1ULL << 22)
161#define SNB_REMOTE (0xffULL << 23)
162#define SNB_SNP_NONE (1ULL << 31)
163#define SNB_SNP_NOT_NEEDED (1ULL << 32)
164#define SNB_SNP_MISS (1ULL << 33)
165#define SNB_NO_FWD (1ULL << 34)
166#define SNB_SNP_FWD (1ULL << 35)
167#define SNB_HITM (1ULL << 36)
168#define SNB_NON_DRAM (1ULL << 37)
169
170#define SNB_DMND_READ (SNB_DMND_DATA_RD|SNB_LLC_DATA_RD)
171#define SNB_DMND_WRITE (SNB_DMND_RFO|SNB_LLC_RFO)
172#define SNB_DMND_PREFETCH (SNB_PF_DATA_RD|SNB_PF_RFO)
173
174#define SNB_SNP_ANY (SNB_SNP_NONE|SNB_SNP_NOT_NEEDED| \
175 SNB_SNP_MISS|SNB_NO_FWD|SNB_SNP_FWD| \
176 SNB_HITM)
177
178#define SNB_DRAM_ANY (SNB_LOCAL|SNB_REMOTE|SNB_SNP_ANY)
179#define SNB_DRAM_REMOTE (SNB_REMOTE|SNB_SNP_ANY)
180
181#define SNB_L3_ACCESS SNB_RESP_ANY
182#define SNB_L3_MISS (SNB_DRAM_ANY|SNB_NON_DRAM)
183
184static __initconst const u64 snb_hw_cache_extra_regs
185 [PERF_COUNT_HW_CACHE_MAX]
186 [PERF_COUNT_HW_CACHE_OP_MAX]
187 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
188{
189 [ C(LL ) ] = {
190 [ C(OP_READ) ] = {
191 [ C(RESULT_ACCESS) ] = SNB_DMND_READ|SNB_L3_ACCESS,
192 [ C(RESULT_MISS) ] = SNB_DMND_READ|SNB_L3_MISS,
193 },
194 [ C(OP_WRITE) ] = {
195 [ C(RESULT_ACCESS) ] = SNB_DMND_WRITE|SNB_L3_ACCESS,
196 [ C(RESULT_MISS) ] = SNB_DMND_WRITE|SNB_L3_MISS,
197 },
198 [ C(OP_PREFETCH) ] = {
199 [ C(RESULT_ACCESS) ] = SNB_DMND_PREFETCH|SNB_L3_ACCESS,
200 [ C(RESULT_MISS) ] = SNB_DMND_PREFETCH|SNB_L3_MISS,
201 },
202 },
203 [ C(NODE) ] = {
204 [ C(OP_READ) ] = {
205 [ C(RESULT_ACCESS) ] = SNB_DMND_READ|SNB_DRAM_ANY,
206 [ C(RESULT_MISS) ] = SNB_DMND_READ|SNB_DRAM_REMOTE,
207 },
208 [ C(OP_WRITE) ] = {
209 [ C(RESULT_ACCESS) ] = SNB_DMND_WRITE|SNB_DRAM_ANY,
210 [ C(RESULT_MISS) ] = SNB_DMND_WRITE|SNB_DRAM_REMOTE,
211 },
212 [ C(OP_PREFETCH) ] = {
213 [ C(RESULT_ACCESS) ] = SNB_DMND_PREFETCH|SNB_DRAM_ANY,
214 [ C(RESULT_MISS) ] = SNB_DMND_PREFETCH|SNB_DRAM_REMOTE,
215 },
216 },
217};
218
141static __initconst const u64 snb_hw_cache_event_ids 219static __initconst const u64 snb_hw_cache_event_ids
142 [PERF_COUNT_HW_CACHE_MAX] 220 [PERF_COUNT_HW_CACHE_MAX]
143 [PERF_COUNT_HW_CACHE_OP_MAX] 221 [PERF_COUNT_HW_CACHE_OP_MAX]
@@ -235,16 +313,16 @@ static __initconst const u64 snb_hw_cache_event_ids
235 }, 313 },
236 [ C(NODE) ] = { 314 [ C(NODE) ] = {
237 [ C(OP_READ) ] = { 315 [ C(OP_READ) ] = {
238 [ C(RESULT_ACCESS) ] = -1, 316 [ C(RESULT_ACCESS) ] = 0x01b7,
239 [ C(RESULT_MISS) ] = -1, 317 [ C(RESULT_MISS) ] = 0x01b7,
240 }, 318 },
241 [ C(OP_WRITE) ] = { 319 [ C(OP_WRITE) ] = {
242 [ C(RESULT_ACCESS) ] = -1, 320 [ C(RESULT_ACCESS) ] = 0x01b7,
243 [ C(RESULT_MISS) ] = -1, 321 [ C(RESULT_MISS) ] = 0x01b7,
244 }, 322 },
245 [ C(OP_PREFETCH) ] = { 323 [ C(OP_PREFETCH) ] = {
246 [ C(RESULT_ACCESS) ] = -1, 324 [ C(RESULT_ACCESS) ] = 0x01b7,
247 [ C(RESULT_MISS) ] = -1, 325 [ C(RESULT_MISS) ] = 0x01b7,
248 }, 326 },
249 }, 327 },
250 328
@@ -1964,6 +2042,8 @@ __init int intel_pmu_init(void)
1964 case 58: /* IvyBridge */ 2042 case 58: /* IvyBridge */
1965 memcpy(hw_cache_event_ids, snb_hw_cache_event_ids, 2043 memcpy(hw_cache_event_ids, snb_hw_cache_event_ids,
1966 sizeof(hw_cache_event_ids)); 2044 sizeof(hw_cache_event_ids));
2045 memcpy(hw_cache_extra_regs, snb_hw_cache_extra_regs,
2046 sizeof(hw_cache_extra_regs));
1967 2047
1968 intel_pmu_lbr_init_snb(); 2048 intel_pmu_lbr_init_snb();
1969 2049
diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c
index 19faffc60886..7563fda9f033 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c
@@ -18,6 +18,7 @@ static struct event_constraint constraint_empty =
18 EVENT_CONSTRAINT(0, 0, 0); 18 EVENT_CONSTRAINT(0, 0, 0);
19 19
20DEFINE_UNCORE_FORMAT_ATTR(event, event, "config:0-7"); 20DEFINE_UNCORE_FORMAT_ATTR(event, event, "config:0-7");
21DEFINE_UNCORE_FORMAT_ATTR(event_ext, event, "config:0-7,21");
21DEFINE_UNCORE_FORMAT_ATTR(umask, umask, "config:8-15"); 22DEFINE_UNCORE_FORMAT_ATTR(umask, umask, "config:8-15");
22DEFINE_UNCORE_FORMAT_ATTR(edge, edge, "config:18"); 23DEFINE_UNCORE_FORMAT_ATTR(edge, edge, "config:18");
23DEFINE_UNCORE_FORMAT_ATTR(tid_en, tid_en, "config:19"); 24DEFINE_UNCORE_FORMAT_ATTR(tid_en, tid_en, "config:19");
@@ -33,10 +34,81 @@ DEFINE_UNCORE_FORMAT_ATTR(filter_tid, filter_tid, "config1:0-4");
33DEFINE_UNCORE_FORMAT_ATTR(filter_nid, filter_nid, "config1:10-17"); 34DEFINE_UNCORE_FORMAT_ATTR(filter_nid, filter_nid, "config1:10-17");
34DEFINE_UNCORE_FORMAT_ATTR(filter_state, filter_state, "config1:18-22"); 35DEFINE_UNCORE_FORMAT_ATTR(filter_state, filter_state, "config1:18-22");
35DEFINE_UNCORE_FORMAT_ATTR(filter_opc, filter_opc, "config1:23-31"); 36DEFINE_UNCORE_FORMAT_ATTR(filter_opc, filter_opc, "config1:23-31");
36DEFINE_UNCORE_FORMAT_ATTR(filter_brand0, filter_brand0, "config1:0-7"); 37DEFINE_UNCORE_FORMAT_ATTR(filter_band0, filter_band0, "config1:0-7");
37DEFINE_UNCORE_FORMAT_ATTR(filter_brand1, filter_brand1, "config1:8-15"); 38DEFINE_UNCORE_FORMAT_ATTR(filter_band1, filter_band1, "config1:8-15");
38DEFINE_UNCORE_FORMAT_ATTR(filter_brand2, filter_brand2, "config1:16-23"); 39DEFINE_UNCORE_FORMAT_ATTR(filter_band2, filter_band2, "config1:16-23");
39DEFINE_UNCORE_FORMAT_ATTR(filter_brand3, filter_brand3, "config1:24-31"); 40DEFINE_UNCORE_FORMAT_ATTR(filter_band3, filter_band3, "config1:24-31");
41
42static u64 uncore_msr_read_counter(struct intel_uncore_box *box, struct perf_event *event)
43{
44 u64 count;
45
46 rdmsrl(event->hw.event_base, count);
47
48 return count;
49}
50
51/*
52 * generic get constraint function for shared match/mask registers.
53 */
54static struct event_constraint *
55uncore_get_constraint(struct intel_uncore_box *box, struct perf_event *event)
56{
57 struct intel_uncore_extra_reg *er;
58 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
59 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
60 unsigned long flags;
61 bool ok = false;
62
63 /*
64 * reg->alloc can be set due to existing state, so for fake box we
65 * need to ignore this, otherwise we might fail to allocate proper
66 * fake state for this extra reg constraint.
67 */
68 if (reg1->idx == EXTRA_REG_NONE ||
69 (!uncore_box_is_fake(box) && reg1->alloc))
70 return NULL;
71
72 er = &box->shared_regs[reg1->idx];
73 raw_spin_lock_irqsave(&er->lock, flags);
74 if (!atomic_read(&er->ref) ||
75 (er->config1 == reg1->config && er->config2 == reg2->config)) {
76 atomic_inc(&er->ref);
77 er->config1 = reg1->config;
78 er->config2 = reg2->config;
79 ok = true;
80 }
81 raw_spin_unlock_irqrestore(&er->lock, flags);
82
83 if (ok) {
84 if (!uncore_box_is_fake(box))
85 reg1->alloc = 1;
86 return NULL;
87 }
88
89 return &constraint_empty;
90}
91
92static void uncore_put_constraint(struct intel_uncore_box *box, struct perf_event *event)
93{
94 struct intel_uncore_extra_reg *er;
95 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
96
97 /*
98 * Only put constraint if extra reg was actually allocated. Also
99 * takes care of event which do not use an extra shared reg.
100 *
101 * Also, if this is a fake box we shouldn't touch any event state
102 * (reg->alloc) and we don't care about leaving inconsistent box
103 * state either since it will be thrown out.
104 */
105 if (uncore_box_is_fake(box) || !reg1->alloc)
106 return;
107
108 er = &box->shared_regs[reg1->idx];
109 atomic_dec(&er->ref);
110 reg1->alloc = 0;
111}
40 112
41/* Sandy Bridge-EP uncore support */ 113/* Sandy Bridge-EP uncore support */
42static struct intel_uncore_type snbep_uncore_cbox; 114static struct intel_uncore_type snbep_uncore_cbox;
@@ -64,18 +136,15 @@ static void snbep_uncore_pci_enable_box(struct intel_uncore_box *box)
64 pci_write_config_dword(pdev, box_ctl, config); 136 pci_write_config_dword(pdev, box_ctl, config);
65} 137}
66 138
67static void snbep_uncore_pci_enable_event(struct intel_uncore_box *box, 139static void snbep_uncore_pci_enable_event(struct intel_uncore_box *box, struct perf_event *event)
68 struct perf_event *event)
69{ 140{
70 struct pci_dev *pdev = box->pci_dev; 141 struct pci_dev *pdev = box->pci_dev;
71 struct hw_perf_event *hwc = &event->hw; 142 struct hw_perf_event *hwc = &event->hw;
72 143
73 pci_write_config_dword(pdev, hwc->config_base, hwc->config | 144 pci_write_config_dword(pdev, hwc->config_base, hwc->config | SNBEP_PMON_CTL_EN);
74 SNBEP_PMON_CTL_EN);
75} 145}
76 146
77static void snbep_uncore_pci_disable_event(struct intel_uncore_box *box, 147static void snbep_uncore_pci_disable_event(struct intel_uncore_box *box, struct perf_event *event)
78 struct perf_event *event)
79{ 148{
80 struct pci_dev *pdev = box->pci_dev; 149 struct pci_dev *pdev = box->pci_dev;
81 struct hw_perf_event *hwc = &event->hw; 150 struct hw_perf_event *hwc = &event->hw;
@@ -83,8 +152,7 @@ static void snbep_uncore_pci_disable_event(struct intel_uncore_box *box,
83 pci_write_config_dword(pdev, hwc->config_base, hwc->config); 152 pci_write_config_dword(pdev, hwc->config_base, hwc->config);
84} 153}
85 154
86static u64 snbep_uncore_pci_read_counter(struct intel_uncore_box *box, 155static u64 snbep_uncore_pci_read_counter(struct intel_uncore_box *box, struct perf_event *event)
87 struct perf_event *event)
88{ 156{
89 struct pci_dev *pdev = box->pci_dev; 157 struct pci_dev *pdev = box->pci_dev;
90 struct hw_perf_event *hwc = &event->hw; 158 struct hw_perf_event *hwc = &event->hw;
@@ -92,14 +160,15 @@ static u64 snbep_uncore_pci_read_counter(struct intel_uncore_box *box,
92 160
93 pci_read_config_dword(pdev, hwc->event_base, (u32 *)&count); 161 pci_read_config_dword(pdev, hwc->event_base, (u32 *)&count);
94 pci_read_config_dword(pdev, hwc->event_base + 4, (u32 *)&count + 1); 162 pci_read_config_dword(pdev, hwc->event_base + 4, (u32 *)&count + 1);
163
95 return count; 164 return count;
96} 165}
97 166
98static void snbep_uncore_pci_init_box(struct intel_uncore_box *box) 167static void snbep_uncore_pci_init_box(struct intel_uncore_box *box)
99{ 168{
100 struct pci_dev *pdev = box->pci_dev; 169 struct pci_dev *pdev = box->pci_dev;
101 pci_write_config_dword(pdev, SNBEP_PCI_PMON_BOX_CTL, 170
102 SNBEP_PMON_BOX_CTL_INT); 171 pci_write_config_dword(pdev, SNBEP_PCI_PMON_BOX_CTL, SNBEP_PMON_BOX_CTL_INT);
103} 172}
104 173
105static void snbep_uncore_msr_disable_box(struct intel_uncore_box *box) 174static void snbep_uncore_msr_disable_box(struct intel_uncore_box *box)
@@ -112,7 +181,6 @@ static void snbep_uncore_msr_disable_box(struct intel_uncore_box *box)
112 rdmsrl(msr, config); 181 rdmsrl(msr, config);
113 config |= SNBEP_PMON_BOX_CTL_FRZ; 182 config |= SNBEP_PMON_BOX_CTL_FRZ;
114 wrmsrl(msr, config); 183 wrmsrl(msr, config);
115 return;
116 } 184 }
117} 185}
118 186
@@ -126,12 +194,10 @@ static void snbep_uncore_msr_enable_box(struct intel_uncore_box *box)
126 rdmsrl(msr, config); 194 rdmsrl(msr, config);
127 config &= ~SNBEP_PMON_BOX_CTL_FRZ; 195 config &= ~SNBEP_PMON_BOX_CTL_FRZ;
128 wrmsrl(msr, config); 196 wrmsrl(msr, config);
129 return;
130 } 197 }
131} 198}
132 199
133static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, 200static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
134 struct perf_event *event)
135{ 201{
136 struct hw_perf_event *hwc = &event->hw; 202 struct hw_perf_event *hwc = &event->hw;
137 struct hw_perf_event_extra *reg1 = &hwc->extra_reg; 203 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
@@ -150,68 +216,15 @@ static void snbep_uncore_msr_disable_event(struct intel_uncore_box *box,
150 wrmsrl(hwc->config_base, hwc->config); 216 wrmsrl(hwc->config_base, hwc->config);
151} 217}
152 218
153static u64 snbep_uncore_msr_read_counter(struct intel_uncore_box *box,
154 struct perf_event *event)
155{
156 struct hw_perf_event *hwc = &event->hw;
157 u64 count;
158
159 rdmsrl(hwc->event_base, count);
160 return count;
161}
162
163static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) 219static void snbep_uncore_msr_init_box(struct intel_uncore_box *box)
164{ 220{
165 unsigned msr = uncore_msr_box_ctl(box); 221 unsigned msr = uncore_msr_box_ctl(box);
222
166 if (msr) 223 if (msr)
167 wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); 224 wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT);
168} 225}
169 226
170static struct event_constraint * 227static int snbep_uncore_hw_config(struct intel_uncore_box *box, struct perf_event *event)
171snbep_uncore_get_constraint(struct intel_uncore_box *box,
172 struct perf_event *event)
173{
174 struct intel_uncore_extra_reg *er;
175 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
176 unsigned long flags;
177 bool ok = false;
178
179 if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc))
180 return NULL;
181
182 er = &box->shared_regs[reg1->idx];
183 raw_spin_lock_irqsave(&er->lock, flags);
184 if (!atomic_read(&er->ref) || er->config1 == reg1->config) {
185 atomic_inc(&er->ref);
186 er->config1 = reg1->config;
187 ok = true;
188 }
189 raw_spin_unlock_irqrestore(&er->lock, flags);
190
191 if (ok) {
192 if (box->phys_id >= 0)
193 reg1->alloc = 1;
194 return NULL;
195 }
196 return &constraint_empty;
197}
198
199static void snbep_uncore_put_constraint(struct intel_uncore_box *box,
200 struct perf_event *event)
201{
202 struct intel_uncore_extra_reg *er;
203 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
204
205 if (box->phys_id < 0 || !reg1->alloc)
206 return;
207
208 er = &box->shared_regs[reg1->idx];
209 atomic_dec(&er->ref);
210 reg1->alloc = 0;
211}
212
213static int snbep_uncore_hw_config(struct intel_uncore_box *box,
214 struct perf_event *event)
215{ 228{
216 struct hw_perf_event *hwc = &event->hw; 229 struct hw_perf_event *hwc = &event->hw;
217 struct hw_perf_event_extra *reg1 = &hwc->extra_reg; 230 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
@@ -221,14 +234,16 @@ static int snbep_uncore_hw_config(struct intel_uncore_box *box,
221 SNBEP_CBO_MSR_OFFSET * box->pmu->pmu_idx; 234 SNBEP_CBO_MSR_OFFSET * box->pmu->pmu_idx;
222 reg1->config = event->attr.config1 & 235 reg1->config = event->attr.config1 &
223 SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK; 236 SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK;
224 } else if (box->pmu->type == &snbep_uncore_pcu) {
225 reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER;
226 reg1->config = event->attr.config1 &
227 SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK;
228 } else { 237 } else {
229 return 0; 238 if (box->pmu->type == &snbep_uncore_pcu) {
239 reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER;
240 reg1->config = event->attr.config1 & SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK;
241 } else {
242 return 0;
243 }
230 } 244 }
231 reg1->idx = 0; 245 reg1->idx = 0;
246
232 return 0; 247 return 0;
233} 248}
234 249
@@ -272,10 +287,19 @@ static struct attribute *snbep_uncore_pcu_formats_attr[] = {
272 &format_attr_thresh5.attr, 287 &format_attr_thresh5.attr,
273 &format_attr_occ_invert.attr, 288 &format_attr_occ_invert.attr,
274 &format_attr_occ_edge.attr, 289 &format_attr_occ_edge.attr,
275 &format_attr_filter_brand0.attr, 290 &format_attr_filter_band0.attr,
276 &format_attr_filter_brand1.attr, 291 &format_attr_filter_band1.attr,
277 &format_attr_filter_brand2.attr, 292 &format_attr_filter_band2.attr,
278 &format_attr_filter_brand3.attr, 293 &format_attr_filter_band3.attr,
294 NULL,
295};
296
297static struct attribute *snbep_uncore_qpi_formats_attr[] = {
298 &format_attr_event_ext.attr,
299 &format_attr_umask.attr,
300 &format_attr_edge.attr,
301 &format_attr_inv.attr,
302 &format_attr_thresh8.attr,
279 NULL, 303 NULL,
280}; 304};
281 305
@@ -314,15 +338,20 @@ static struct attribute_group snbep_uncore_pcu_format_group = {
314 .attrs = snbep_uncore_pcu_formats_attr, 338 .attrs = snbep_uncore_pcu_formats_attr,
315}; 339};
316 340
341static struct attribute_group snbep_uncore_qpi_format_group = {
342 .name = "format",
343 .attrs = snbep_uncore_qpi_formats_attr,
344};
345
317static struct intel_uncore_ops snbep_uncore_msr_ops = { 346static struct intel_uncore_ops snbep_uncore_msr_ops = {
318 .init_box = snbep_uncore_msr_init_box, 347 .init_box = snbep_uncore_msr_init_box,
319 .disable_box = snbep_uncore_msr_disable_box, 348 .disable_box = snbep_uncore_msr_disable_box,
320 .enable_box = snbep_uncore_msr_enable_box, 349 .enable_box = snbep_uncore_msr_enable_box,
321 .disable_event = snbep_uncore_msr_disable_event, 350 .disable_event = snbep_uncore_msr_disable_event,
322 .enable_event = snbep_uncore_msr_enable_event, 351 .enable_event = snbep_uncore_msr_enable_event,
323 .read_counter = snbep_uncore_msr_read_counter, 352 .read_counter = uncore_msr_read_counter,
324 .get_constraint = snbep_uncore_get_constraint, 353 .get_constraint = uncore_get_constraint,
325 .put_constraint = snbep_uncore_put_constraint, 354 .put_constraint = uncore_put_constraint,
326 .hw_config = snbep_uncore_hw_config, 355 .hw_config = snbep_uncore_hw_config,
327}; 356};
328 357
@@ -485,8 +514,13 @@ static struct intel_uncore_type snbep_uncore_qpi = {
485 .num_counters = 4, 514 .num_counters = 4,
486 .num_boxes = 2, 515 .num_boxes = 2,
487 .perf_ctr_bits = 48, 516 .perf_ctr_bits = 48,
517 .perf_ctr = SNBEP_PCI_PMON_CTR0,
518 .event_ctl = SNBEP_PCI_PMON_CTL0,
519 .event_mask = SNBEP_QPI_PCI_PMON_RAW_EVENT_MASK,
520 .box_ctl = SNBEP_PCI_PMON_BOX_CTL,
521 .ops = &snbep_uncore_pci_ops,
488 .event_descs = snbep_uncore_qpi_events, 522 .event_descs = snbep_uncore_qpi_events,
489 SNBEP_UNCORE_PCI_COMMON_INIT(), 523 .format_group = &snbep_uncore_qpi_format_group,
490}; 524};
491 525
492 526
@@ -603,10 +637,8 @@ static void snbep_pci2phy_map_init(void)
603} 637}
604/* end of Sandy Bridge-EP uncore support */ 638/* end of Sandy Bridge-EP uncore support */
605 639
606
607/* Sandy Bridge uncore support */ 640/* Sandy Bridge uncore support */
608static void snb_uncore_msr_enable_event(struct intel_uncore_box *box, 641static void snb_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
609 struct perf_event *event)
610{ 642{
611 struct hw_perf_event *hwc = &event->hw; 643 struct hw_perf_event *hwc = &event->hw;
612 644
@@ -616,20 +648,11 @@ static void snb_uncore_msr_enable_event(struct intel_uncore_box *box,
616 wrmsrl(hwc->config_base, SNB_UNC_CTL_EN); 648 wrmsrl(hwc->config_base, SNB_UNC_CTL_EN);
617} 649}
618 650
619static void snb_uncore_msr_disable_event(struct intel_uncore_box *box, 651static void snb_uncore_msr_disable_event(struct intel_uncore_box *box, struct perf_event *event)
620 struct perf_event *event)
621{ 652{
622 wrmsrl(event->hw.config_base, 0); 653 wrmsrl(event->hw.config_base, 0);
623} 654}
624 655
625static u64 snb_uncore_msr_read_counter(struct intel_uncore_box *box,
626 struct perf_event *event)
627{
628 u64 count;
629 rdmsrl(event->hw.event_base, count);
630 return count;
631}
632
633static void snb_uncore_msr_init_box(struct intel_uncore_box *box) 656static void snb_uncore_msr_init_box(struct intel_uncore_box *box)
634{ 657{
635 if (box->pmu->pmu_idx == 0) { 658 if (box->pmu->pmu_idx == 0) {
@@ -648,15 +671,15 @@ static struct attribute *snb_uncore_formats_attr[] = {
648}; 671};
649 672
650static struct attribute_group snb_uncore_format_group = { 673static struct attribute_group snb_uncore_format_group = {
651 .name = "format", 674 .name = "format",
652 .attrs = snb_uncore_formats_attr, 675 .attrs = snb_uncore_formats_attr,
653}; 676};
654 677
655static struct intel_uncore_ops snb_uncore_msr_ops = { 678static struct intel_uncore_ops snb_uncore_msr_ops = {
656 .init_box = snb_uncore_msr_init_box, 679 .init_box = snb_uncore_msr_init_box,
657 .disable_event = snb_uncore_msr_disable_event, 680 .disable_event = snb_uncore_msr_disable_event,
658 .enable_event = snb_uncore_msr_enable_event, 681 .enable_event = snb_uncore_msr_enable_event,
659 .read_counter = snb_uncore_msr_read_counter, 682 .read_counter = uncore_msr_read_counter,
660}; 683};
661 684
662static struct event_constraint snb_uncore_cbox_constraints[] = { 685static struct event_constraint snb_uncore_cbox_constraints[] = {
@@ -697,12 +720,10 @@ static void nhm_uncore_msr_disable_box(struct intel_uncore_box *box)
697 720
698static void nhm_uncore_msr_enable_box(struct intel_uncore_box *box) 721static void nhm_uncore_msr_enable_box(struct intel_uncore_box *box)
699{ 722{
700 wrmsrl(NHM_UNC_PERF_GLOBAL_CTL, 723 wrmsrl(NHM_UNC_PERF_GLOBAL_CTL, NHM_UNC_GLOBAL_CTL_EN_PC_ALL | NHM_UNC_GLOBAL_CTL_EN_FC);
701 NHM_UNC_GLOBAL_CTL_EN_PC_ALL | NHM_UNC_GLOBAL_CTL_EN_FC);
702} 724}
703 725
704static void nhm_uncore_msr_enable_event(struct intel_uncore_box *box, 726static void nhm_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
705 struct perf_event *event)
706{ 727{
707 struct hw_perf_event *hwc = &event->hw; 728 struct hw_perf_event *hwc = &event->hw;
708 729
@@ -744,7 +765,7 @@ static struct intel_uncore_ops nhm_uncore_msr_ops = {
744 .enable_box = nhm_uncore_msr_enable_box, 765 .enable_box = nhm_uncore_msr_enable_box,
745 .disable_event = snb_uncore_msr_disable_event, 766 .disable_event = snb_uncore_msr_disable_event,
746 .enable_event = nhm_uncore_msr_enable_event, 767 .enable_event = nhm_uncore_msr_enable_event,
747 .read_counter = snb_uncore_msr_read_counter, 768 .read_counter = uncore_msr_read_counter,
748}; 769};
749 770
750static struct intel_uncore_type nhm_uncore = { 771static struct intel_uncore_type nhm_uncore = {
@@ -769,8 +790,1041 @@ static struct intel_uncore_type *nhm_msr_uncores[] = {
769}; 790};
770/* end of Nehalem uncore support */ 791/* end of Nehalem uncore support */
771 792
772static void uncore_assign_hw_event(struct intel_uncore_box *box, 793/* Nehalem-EX uncore support */
773 struct perf_event *event, int idx) 794#define __BITS_VALUE(x, i, n) ((typeof(x))(((x) >> ((i) * (n))) & \
795 ((1ULL << (n)) - 1)))
796
797DEFINE_UNCORE_FORMAT_ATTR(event5, event, "config:1-5");
798DEFINE_UNCORE_FORMAT_ATTR(counter, counter, "config:6-7");
799DEFINE_UNCORE_FORMAT_ATTR(mm_cfg, mm_cfg, "config:63");
800DEFINE_UNCORE_FORMAT_ATTR(match, match, "config1:0-63");
801DEFINE_UNCORE_FORMAT_ATTR(mask, mask, "config2:0-63");
802
803static void nhmex_uncore_msr_init_box(struct intel_uncore_box *box)
804{
805 wrmsrl(NHMEX_U_MSR_PMON_GLOBAL_CTL, NHMEX_U_PMON_GLOBAL_EN_ALL);
806}
807
808static void nhmex_uncore_msr_disable_box(struct intel_uncore_box *box)
809{
810 unsigned msr = uncore_msr_box_ctl(box);
811 u64 config;
812
813 if (msr) {
814 rdmsrl(msr, config);
815 config &= ~((1ULL << uncore_num_counters(box)) - 1);
816 /* WBox has a fixed counter */
817 if (uncore_msr_fixed_ctl(box))
818 config &= ~NHMEX_W_PMON_GLOBAL_FIXED_EN;
819 wrmsrl(msr, config);
820 }
821}
822
823static void nhmex_uncore_msr_enable_box(struct intel_uncore_box *box)
824{
825 unsigned msr = uncore_msr_box_ctl(box);
826 u64 config;
827
828 if (msr) {
829 rdmsrl(msr, config);
830 config |= (1ULL << uncore_num_counters(box)) - 1;
831 /* WBox has a fixed counter */
832 if (uncore_msr_fixed_ctl(box))
833 config |= NHMEX_W_PMON_GLOBAL_FIXED_EN;
834 wrmsrl(msr, config);
835 }
836}
837
838static void nhmex_uncore_msr_disable_event(struct intel_uncore_box *box, struct perf_event *event)
839{
840 wrmsrl(event->hw.config_base, 0);
841}
842
843static void nhmex_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
844{
845 struct hw_perf_event *hwc = &event->hw;
846
847 if (hwc->idx >= UNCORE_PMC_IDX_FIXED)
848 wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0);
849 else if (box->pmu->type->event_mask & NHMEX_PMON_CTL_EN_BIT0)
850 wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22);
851 else
852 wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0);
853}
854
855#define NHMEX_UNCORE_OPS_COMMON_INIT() \
856 .init_box = nhmex_uncore_msr_init_box, \
857 .disable_box = nhmex_uncore_msr_disable_box, \
858 .enable_box = nhmex_uncore_msr_enable_box, \
859 .disable_event = nhmex_uncore_msr_disable_event, \
860 .read_counter = uncore_msr_read_counter
861
862static struct intel_uncore_ops nhmex_uncore_ops = {
863 NHMEX_UNCORE_OPS_COMMON_INIT(),
864 .enable_event = nhmex_uncore_msr_enable_event,
865};
866
867static struct attribute *nhmex_uncore_ubox_formats_attr[] = {
868 &format_attr_event.attr,
869 &format_attr_edge.attr,
870 NULL,
871};
872
873static struct attribute_group nhmex_uncore_ubox_format_group = {
874 .name = "format",
875 .attrs = nhmex_uncore_ubox_formats_attr,
876};
877
878static struct intel_uncore_type nhmex_uncore_ubox = {
879 .name = "ubox",
880 .num_counters = 1,
881 .num_boxes = 1,
882 .perf_ctr_bits = 48,
883 .event_ctl = NHMEX_U_MSR_PMON_EV_SEL,
884 .perf_ctr = NHMEX_U_MSR_PMON_CTR,
885 .event_mask = NHMEX_U_PMON_RAW_EVENT_MASK,
886 .box_ctl = NHMEX_U_MSR_PMON_GLOBAL_CTL,
887 .ops = &nhmex_uncore_ops,
888 .format_group = &nhmex_uncore_ubox_format_group
889};
890
891static struct attribute *nhmex_uncore_cbox_formats_attr[] = {
892 &format_attr_event.attr,
893 &format_attr_umask.attr,
894 &format_attr_edge.attr,
895 &format_attr_inv.attr,
896 &format_attr_thresh8.attr,
897 NULL,
898};
899
900static struct attribute_group nhmex_uncore_cbox_format_group = {
901 .name = "format",
902 .attrs = nhmex_uncore_cbox_formats_attr,
903};
904
905static struct intel_uncore_type nhmex_uncore_cbox = {
906 .name = "cbox",
907 .num_counters = 6,
908 .num_boxes = 8,
909 .perf_ctr_bits = 48,
910 .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0,
911 .perf_ctr = NHMEX_C0_MSR_PMON_CTR0,
912 .event_mask = NHMEX_PMON_RAW_EVENT_MASK,
913 .box_ctl = NHMEX_C0_MSR_PMON_GLOBAL_CTL,
914 .msr_offset = NHMEX_C_MSR_OFFSET,
915 .pair_ctr_ctl = 1,
916 .ops = &nhmex_uncore_ops,
917 .format_group = &nhmex_uncore_cbox_format_group
918};
919
920static struct uncore_event_desc nhmex_uncore_wbox_events[] = {
921 INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0"),
922 { /* end: all zeroes */ },
923};
924
925static struct intel_uncore_type nhmex_uncore_wbox = {
926 .name = "wbox",
927 .num_counters = 4,
928 .num_boxes = 1,
929 .perf_ctr_bits = 48,
930 .event_ctl = NHMEX_W_MSR_PMON_CNT0,
931 .perf_ctr = NHMEX_W_MSR_PMON_EVT_SEL0,
932 .fixed_ctr = NHMEX_W_MSR_PMON_FIXED_CTR,
933 .fixed_ctl = NHMEX_W_MSR_PMON_FIXED_CTL,
934 .event_mask = NHMEX_PMON_RAW_EVENT_MASK,
935 .box_ctl = NHMEX_W_MSR_GLOBAL_CTL,
936 .pair_ctr_ctl = 1,
937 .event_descs = nhmex_uncore_wbox_events,
938 .ops = &nhmex_uncore_ops,
939 .format_group = &nhmex_uncore_cbox_format_group
940};
941
942static int nhmex_bbox_hw_config(struct intel_uncore_box *box, struct perf_event *event)
943{
944 struct hw_perf_event *hwc = &event->hw;
945 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
946 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
947 int ctr, ev_sel;
948
949 ctr = (hwc->config & NHMEX_B_PMON_CTR_MASK) >>
950 NHMEX_B_PMON_CTR_SHIFT;
951 ev_sel = (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK) >>
952 NHMEX_B_PMON_CTL_EV_SEL_SHIFT;
953
954 /* events that do not use the match/mask registers */
955 if ((ctr == 0 && ev_sel > 0x3) || (ctr == 1 && ev_sel > 0x6) ||
956 (ctr == 2 && ev_sel != 0x4) || ctr == 3)
957 return 0;
958
959 if (box->pmu->pmu_idx == 0)
960 reg1->reg = NHMEX_B0_MSR_MATCH;
961 else
962 reg1->reg = NHMEX_B1_MSR_MATCH;
963 reg1->idx = 0;
964 reg1->config = event->attr.config1;
965 reg2->config = event->attr.config2;
966 return 0;
967}
968
969static void nhmex_bbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
970{
971 struct hw_perf_event *hwc = &event->hw;
972 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
973 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
974
975 if (reg1->idx != EXTRA_REG_NONE) {
976 wrmsrl(reg1->reg, reg1->config);
977 wrmsrl(reg1->reg + 1, reg2->config);
978 }
979 wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 |
980 (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK));
981}
982
983/*
984 * The Bbox has 4 counters, but each counter monitors different events.
985 * Use bits 6-7 in the event config to select counter.
986 */
987static struct event_constraint nhmex_uncore_bbox_constraints[] = {
988 EVENT_CONSTRAINT(0 , 1, 0xc0),
989 EVENT_CONSTRAINT(0x40, 2, 0xc0),
990 EVENT_CONSTRAINT(0x80, 4, 0xc0),
991 EVENT_CONSTRAINT(0xc0, 8, 0xc0),
992 EVENT_CONSTRAINT_END,
993};
994
995static struct attribute *nhmex_uncore_bbox_formats_attr[] = {
996 &format_attr_event5.attr,
997 &format_attr_counter.attr,
998 &format_attr_match.attr,
999 &format_attr_mask.attr,
1000 NULL,
1001};
1002
1003static struct attribute_group nhmex_uncore_bbox_format_group = {
1004 .name = "format",
1005 .attrs = nhmex_uncore_bbox_formats_attr,
1006};
1007
1008static struct intel_uncore_ops nhmex_uncore_bbox_ops = {
1009 NHMEX_UNCORE_OPS_COMMON_INIT(),
1010 .enable_event = nhmex_bbox_msr_enable_event,
1011 .hw_config = nhmex_bbox_hw_config,
1012 .get_constraint = uncore_get_constraint,
1013 .put_constraint = uncore_put_constraint,
1014};
1015
1016static struct intel_uncore_type nhmex_uncore_bbox = {
1017 .name = "bbox",
1018 .num_counters = 4,
1019 .num_boxes = 2,
1020 .perf_ctr_bits = 48,
1021 .event_ctl = NHMEX_B0_MSR_PMON_CTL0,
1022 .perf_ctr = NHMEX_B0_MSR_PMON_CTR0,
1023 .event_mask = NHMEX_B_PMON_RAW_EVENT_MASK,
1024 .box_ctl = NHMEX_B0_MSR_PMON_GLOBAL_CTL,
1025 .msr_offset = NHMEX_B_MSR_OFFSET,
1026 .pair_ctr_ctl = 1,
1027 .num_shared_regs = 1,
1028 .constraints = nhmex_uncore_bbox_constraints,
1029 .ops = &nhmex_uncore_bbox_ops,
1030 .format_group = &nhmex_uncore_bbox_format_group
1031};
1032
1033static int nhmex_sbox_hw_config(struct intel_uncore_box *box, struct perf_event *event)
1034{
1035 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1036 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
1037
1038 if (event->attr.config & NHMEX_S_PMON_MM_CFG_EN) {
1039 reg1->config = event->attr.config1;
1040 reg2->config = event->attr.config2;
1041 } else {
1042 reg1->config = ~0ULL;
1043 reg2->config = ~0ULL;
1044 }
1045
1046 if (box->pmu->pmu_idx == 0)
1047 reg1->reg = NHMEX_S0_MSR_MM_CFG;
1048 else
1049 reg1->reg = NHMEX_S1_MSR_MM_CFG;
1050
1051 reg1->idx = 0;
1052
1053 return 0;
1054}
1055
1056static void nhmex_sbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
1057{
1058 struct hw_perf_event *hwc = &event->hw;
1059 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1060 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1061
1062 wrmsrl(reg1->reg, 0);
1063 if (reg1->config != ~0ULL || reg2->config != ~0ULL) {
1064 wrmsrl(reg1->reg + 1, reg1->config);
1065 wrmsrl(reg1->reg + 2, reg2->config);
1066 wrmsrl(reg1->reg, NHMEX_S_PMON_MM_CFG_EN);
1067 }
1068 wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22);
1069}
1070
1071static struct attribute *nhmex_uncore_sbox_formats_attr[] = {
1072 &format_attr_event.attr,
1073 &format_attr_umask.attr,
1074 &format_attr_edge.attr,
1075 &format_attr_inv.attr,
1076 &format_attr_thresh8.attr,
1077 &format_attr_mm_cfg.attr,
1078 &format_attr_match.attr,
1079 &format_attr_mask.attr,
1080 NULL,
1081};
1082
1083static struct attribute_group nhmex_uncore_sbox_format_group = {
1084 .name = "format",
1085 .attrs = nhmex_uncore_sbox_formats_attr,
1086};
1087
1088static struct intel_uncore_ops nhmex_uncore_sbox_ops = {
1089 NHMEX_UNCORE_OPS_COMMON_INIT(),
1090 .enable_event = nhmex_sbox_msr_enable_event,
1091 .hw_config = nhmex_sbox_hw_config,
1092 .get_constraint = uncore_get_constraint,
1093 .put_constraint = uncore_put_constraint,
1094};
1095
1096static struct intel_uncore_type nhmex_uncore_sbox = {
1097 .name = "sbox",
1098 .num_counters = 4,
1099 .num_boxes = 2,
1100 .perf_ctr_bits = 48,
1101 .event_ctl = NHMEX_S0_MSR_PMON_CTL0,
1102 .perf_ctr = NHMEX_S0_MSR_PMON_CTR0,
1103 .event_mask = NHMEX_PMON_RAW_EVENT_MASK,
1104 .box_ctl = NHMEX_S0_MSR_PMON_GLOBAL_CTL,
1105 .msr_offset = NHMEX_S_MSR_OFFSET,
1106 .pair_ctr_ctl = 1,
1107 .num_shared_regs = 1,
1108 .ops = &nhmex_uncore_sbox_ops,
1109 .format_group = &nhmex_uncore_sbox_format_group
1110};
1111
1112enum {
1113 EXTRA_REG_NHMEX_M_FILTER,
1114 EXTRA_REG_NHMEX_M_DSP,
1115 EXTRA_REG_NHMEX_M_ISS,
1116 EXTRA_REG_NHMEX_M_MAP,
1117 EXTRA_REG_NHMEX_M_MSC_THR,
1118 EXTRA_REG_NHMEX_M_PGT,
1119 EXTRA_REG_NHMEX_M_PLD,
1120 EXTRA_REG_NHMEX_M_ZDP_CTL_FVC,
1121};
1122
1123static struct extra_reg nhmex_uncore_mbox_extra_regs[] = {
1124 MBOX_INC_SEL_EXTAR_REG(0x0, DSP),
1125 MBOX_INC_SEL_EXTAR_REG(0x4, MSC_THR),
1126 MBOX_INC_SEL_EXTAR_REG(0x5, MSC_THR),
1127 MBOX_INC_SEL_EXTAR_REG(0x9, ISS),
1128 /* event 0xa uses two extra registers */
1129 MBOX_INC_SEL_EXTAR_REG(0xa, ISS),
1130 MBOX_INC_SEL_EXTAR_REG(0xa, PLD),
1131 MBOX_INC_SEL_EXTAR_REG(0xb, PLD),
1132 /* events 0xd ~ 0x10 use the same extra register */
1133 MBOX_INC_SEL_EXTAR_REG(0xd, ZDP_CTL_FVC),
1134 MBOX_INC_SEL_EXTAR_REG(0xe, ZDP_CTL_FVC),
1135 MBOX_INC_SEL_EXTAR_REG(0xf, ZDP_CTL_FVC),
1136 MBOX_INC_SEL_EXTAR_REG(0x10, ZDP_CTL_FVC),
1137 MBOX_INC_SEL_EXTAR_REG(0x16, PGT),
1138 MBOX_SET_FLAG_SEL_EXTRA_REG(0x0, DSP),
1139 MBOX_SET_FLAG_SEL_EXTRA_REG(0x1, ISS),
1140 MBOX_SET_FLAG_SEL_EXTRA_REG(0x5, PGT),
1141 MBOX_SET_FLAG_SEL_EXTRA_REG(0x6, MAP),
1142 EVENT_EXTRA_END
1143};
1144
1145static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, int idx, u64 config)
1146{
1147 struct intel_uncore_extra_reg *er;
1148 unsigned long flags;
1149 bool ret = false;
1150 u64 mask;
1151
1152 if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) {
1153 er = &box->shared_regs[idx];
1154 raw_spin_lock_irqsave(&er->lock, flags);
1155 if (!atomic_read(&er->ref) || er->config == config) {
1156 atomic_inc(&er->ref);
1157 er->config = config;
1158 ret = true;
1159 }
1160 raw_spin_unlock_irqrestore(&er->lock, flags);
1161
1162 return ret;
1163 }
1164 /*
1165 * The ZDP_CTL_FVC MSR has 4 fields which are used to control
1166 * events 0xd ~ 0x10. Besides these 4 fields, there are additional
1167 * fields which are shared.
1168 */
1169 idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC;
1170 if (WARN_ON_ONCE(idx >= 4))
1171 return false;
1172
1173 /* mask of the shared fields */
1174 mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK;
1175 er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC];
1176
1177 raw_spin_lock_irqsave(&er->lock, flags);
1178 /* add mask of the non-shared field if it's in use */
1179 if (__BITS_VALUE(atomic_read(&er->ref), idx, 8))
1180 mask |= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1181
1182 if (!atomic_read(&er->ref) || !((er->config ^ config) & mask)) {
1183 atomic_add(1 << (idx * 8), &er->ref);
1184 mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK |
1185 NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1186 er->config &= ~mask;
1187 er->config |= (config & mask);
1188 ret = true;
1189 }
1190 raw_spin_unlock_irqrestore(&er->lock, flags);
1191
1192 return ret;
1193}
1194
1195static void nhmex_mbox_put_shared_reg(struct intel_uncore_box *box, int idx)
1196{
1197 struct intel_uncore_extra_reg *er;
1198
1199 if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) {
1200 er = &box->shared_regs[idx];
1201 atomic_dec(&er->ref);
1202 return;
1203 }
1204
1205 idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC;
1206 er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC];
1207 atomic_sub(1 << (idx * 8), &er->ref);
1208}
1209
1210u64 nhmex_mbox_alter_er(struct perf_event *event, int new_idx, bool modify)
1211{
1212 struct hw_perf_event *hwc = &event->hw;
1213 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1214 int idx, orig_idx = __BITS_VALUE(reg1->idx, 0, 8);
1215 u64 config = reg1->config;
1216
1217 /* get the non-shared control bits and shift them */
1218 idx = orig_idx - EXTRA_REG_NHMEX_M_ZDP_CTL_FVC;
1219 config &= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx);
1220 if (new_idx > orig_idx) {
1221 idx = new_idx - orig_idx;
1222 config <<= 3 * idx;
1223 } else {
1224 idx = orig_idx - new_idx;
1225 config >>= 3 * idx;
1226 }
1227
1228 /* add the shared control bits back */
1229 config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config;
1230 if (modify) {
1231 /* adjust the main event selector */
1232 if (new_idx > orig_idx)
1233 hwc->config += idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT;
1234 else
1235 hwc->config -= idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT;
1236 reg1->config = config;
1237 reg1->idx = ~0xff | new_idx;
1238 }
1239 return config;
1240}
1241
1242static struct event_constraint *
1243nhmex_mbox_get_constraint(struct intel_uncore_box *box, struct perf_event *event)
1244{
1245 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1246 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
1247 int i, idx[2], alloc = 0;
1248 u64 config1 = reg1->config;
1249
1250 idx[0] = __BITS_VALUE(reg1->idx, 0, 8);
1251 idx[1] = __BITS_VALUE(reg1->idx, 1, 8);
1252again:
1253 for (i = 0; i < 2; i++) {
1254 if (!uncore_box_is_fake(box) && (reg1->alloc & (0x1 << i)))
1255 idx[i] = 0xff;
1256
1257 if (idx[i] == 0xff)
1258 continue;
1259
1260 if (!nhmex_mbox_get_shared_reg(box, idx[i],
1261 __BITS_VALUE(config1, i, 32)))
1262 goto fail;
1263 alloc |= (0x1 << i);
1264 }
1265
1266 /* for the match/mask registers */
1267 if ((uncore_box_is_fake(box) || !reg2->alloc) &&
1268 !nhmex_mbox_get_shared_reg(box, reg2->idx, reg2->config))
1269 goto fail;
1270
1271 /*
1272 * If it's a fake box -- as per validate_{group,event}() we
1273 * shouldn't touch event state and we can avoid doing so
1274 * since both will only call get_event_constraints() once
1275 * on each event, this avoids the need for reg->alloc.
1276 */
1277 if (!uncore_box_is_fake(box)) {
1278 if (idx[0] != 0xff && idx[0] != __BITS_VALUE(reg1->idx, 0, 8))
1279 nhmex_mbox_alter_er(event, idx[0], true);
1280 reg1->alloc |= alloc;
1281 reg2->alloc = 1;
1282 }
1283 return NULL;
1284fail:
1285 if (idx[0] != 0xff && !(alloc & 0x1) &&
1286 idx[0] >= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) {
1287 /*
1288 * events 0xd ~ 0x10 are functional identical, but are
1289 * controlled by different fields in the ZDP_CTL_FVC
1290 * register. If we failed to take one field, try the
1291 * rest 3 choices.
1292 */
1293 BUG_ON(__BITS_VALUE(reg1->idx, 1, 8) != 0xff);
1294 idx[0] -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC;
1295 idx[0] = (idx[0] + 1) % 4;
1296 idx[0] += EXTRA_REG_NHMEX_M_ZDP_CTL_FVC;
1297 if (idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) {
1298 config1 = nhmex_mbox_alter_er(event, idx[0], false);
1299 goto again;
1300 }
1301 }
1302
1303 if (alloc & 0x1)
1304 nhmex_mbox_put_shared_reg(box, idx[0]);
1305 if (alloc & 0x2)
1306 nhmex_mbox_put_shared_reg(box, idx[1]);
1307 return &constraint_empty;
1308}
1309
1310static void nhmex_mbox_put_constraint(struct intel_uncore_box *box, struct perf_event *event)
1311{
1312 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1313 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
1314
1315 if (uncore_box_is_fake(box))
1316 return;
1317
1318 if (reg1->alloc & 0x1)
1319 nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 0, 8));
1320 if (reg1->alloc & 0x2)
1321 nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 1, 8));
1322 reg1->alloc = 0;
1323
1324 if (reg2->alloc) {
1325 nhmex_mbox_put_shared_reg(box, reg2->idx);
1326 reg2->alloc = 0;
1327 }
1328}
1329
1330static int nhmex_mbox_extra_reg_idx(struct extra_reg *er)
1331{
1332 if (er->idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC)
1333 return er->idx;
1334 return er->idx + (er->event >> NHMEX_M_PMON_CTL_INC_SEL_SHIFT) - 0xd;
1335}
1336
1337static int nhmex_mbox_hw_config(struct intel_uncore_box *box, struct perf_event *event)
1338{
1339 struct intel_uncore_type *type = box->pmu->type;
1340 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1341 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
1342 struct extra_reg *er;
1343 unsigned msr;
1344 int reg_idx = 0;
1345
1346 if (WARN_ON_ONCE(reg1->idx != -1))
1347 return -EINVAL;
1348 /*
1349 * The mbox events may require 2 extra MSRs at the most. But only
1350 * the lower 32 bits in these MSRs are significant, so we can use
1351 * config1 to pass two MSRs' config.
1352 */
1353 for (er = nhmex_uncore_mbox_extra_regs; er->msr; er++) {
1354 if (er->event != (event->hw.config & er->config_mask))
1355 continue;
1356 if (event->attr.config1 & ~er->valid_mask)
1357 return -EINVAL;
1358 if (er->idx == __BITS_VALUE(reg1->idx, 0, 8) ||
1359 er->idx == __BITS_VALUE(reg1->idx, 1, 8))
1360 continue;
1361 if (WARN_ON_ONCE(reg_idx >= 2))
1362 return -EINVAL;
1363
1364 msr = er->msr + type->msr_offset * box->pmu->pmu_idx;
1365 if (WARN_ON_ONCE(msr >= 0xffff || er->idx >= 0xff))
1366 return -EINVAL;
1367
1368 /* always use the 32~63 bits to pass the PLD config */
1369 if (er->idx == EXTRA_REG_NHMEX_M_PLD)
1370 reg_idx = 1;
1371
1372 reg1->idx &= ~(0xff << (reg_idx * 8));
1373 reg1->reg &= ~(0xffff << (reg_idx * 16));
1374 reg1->idx |= nhmex_mbox_extra_reg_idx(er) << (reg_idx * 8);
1375 reg1->reg |= msr << (reg_idx * 16);
1376 reg1->config = event->attr.config1;
1377 reg_idx++;
1378 }
1379 /* use config2 to pass the filter config */
1380 reg2->idx = EXTRA_REG_NHMEX_M_FILTER;
1381 if (event->attr.config2 & NHMEX_M_PMON_MM_CFG_EN)
1382 reg2->config = event->attr.config2;
1383 else
1384 reg2->config = ~0ULL;
1385 if (box->pmu->pmu_idx == 0)
1386 reg2->reg = NHMEX_M0_MSR_PMU_MM_CFG;
1387 else
1388 reg2->reg = NHMEX_M1_MSR_PMU_MM_CFG;
1389
1390 return 0;
1391}
1392
1393static u64 nhmex_mbox_shared_reg_config(struct intel_uncore_box *box, int idx)
1394{
1395 struct intel_uncore_extra_reg *er;
1396 unsigned long flags;
1397 u64 config;
1398
1399 if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC)
1400 return box->shared_regs[idx].config;
1401
1402 er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC];
1403 raw_spin_lock_irqsave(&er->lock, flags);
1404 config = er->config;
1405 raw_spin_unlock_irqrestore(&er->lock, flags);
1406 return config;
1407}
1408
1409static void nhmex_mbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
1410{
1411 struct hw_perf_event *hwc = &event->hw;
1412 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1413 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1414 int idx;
1415
1416 idx = __BITS_VALUE(reg1->idx, 0, 8);
1417 if (idx != 0xff)
1418 wrmsrl(__BITS_VALUE(reg1->reg, 0, 16),
1419 nhmex_mbox_shared_reg_config(box, idx));
1420 idx = __BITS_VALUE(reg1->idx, 1, 8);
1421 if (idx != 0xff)
1422 wrmsrl(__BITS_VALUE(reg1->reg, 1, 16),
1423 nhmex_mbox_shared_reg_config(box, idx));
1424
1425 wrmsrl(reg2->reg, 0);
1426 if (reg2->config != ~0ULL) {
1427 wrmsrl(reg2->reg + 1,
1428 reg2->config & NHMEX_M_PMON_ADDR_MATCH_MASK);
1429 wrmsrl(reg2->reg + 2, NHMEX_M_PMON_ADDR_MASK_MASK &
1430 (reg2->config >> NHMEX_M_PMON_ADDR_MASK_SHIFT));
1431 wrmsrl(reg2->reg, NHMEX_M_PMON_MM_CFG_EN);
1432 }
1433
1434 wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0);
1435}
1436
1437DEFINE_UNCORE_FORMAT_ATTR(count_mode, count_mode, "config:2-3");
1438DEFINE_UNCORE_FORMAT_ATTR(storage_mode, storage_mode, "config:4-5");
1439DEFINE_UNCORE_FORMAT_ATTR(wrap_mode, wrap_mode, "config:6");
1440DEFINE_UNCORE_FORMAT_ATTR(flag_mode, flag_mode, "config:7");
1441DEFINE_UNCORE_FORMAT_ATTR(inc_sel, inc_sel, "config:9-13");
1442DEFINE_UNCORE_FORMAT_ATTR(set_flag_sel, set_flag_sel, "config:19-21");
1443DEFINE_UNCORE_FORMAT_ATTR(filter_cfg, filter_cfg, "config2:63");
1444DEFINE_UNCORE_FORMAT_ATTR(filter_match, filter_match, "config2:0-33");
1445DEFINE_UNCORE_FORMAT_ATTR(filter_mask, filter_mask, "config2:34-61");
1446DEFINE_UNCORE_FORMAT_ATTR(dsp, dsp, "config1:0-31");
1447DEFINE_UNCORE_FORMAT_ATTR(thr, thr, "config1:0-31");
1448DEFINE_UNCORE_FORMAT_ATTR(fvc, fvc, "config1:0-31");
1449DEFINE_UNCORE_FORMAT_ATTR(pgt, pgt, "config1:0-31");
1450DEFINE_UNCORE_FORMAT_ATTR(map, map, "config1:0-31");
1451DEFINE_UNCORE_FORMAT_ATTR(iss, iss, "config1:0-31");
1452DEFINE_UNCORE_FORMAT_ATTR(pld, pld, "config1:32-63");
1453
1454static struct attribute *nhmex_uncore_mbox_formats_attr[] = {
1455 &format_attr_count_mode.attr,
1456 &format_attr_storage_mode.attr,
1457 &format_attr_wrap_mode.attr,
1458 &format_attr_flag_mode.attr,
1459 &format_attr_inc_sel.attr,
1460 &format_attr_set_flag_sel.attr,
1461 &format_attr_filter_cfg.attr,
1462 &format_attr_filter_match.attr,
1463 &format_attr_filter_mask.attr,
1464 &format_attr_dsp.attr,
1465 &format_attr_thr.attr,
1466 &format_attr_fvc.attr,
1467 &format_attr_pgt.attr,
1468 &format_attr_map.attr,
1469 &format_attr_iss.attr,
1470 &format_attr_pld.attr,
1471 NULL,
1472};
1473
1474static struct attribute_group nhmex_uncore_mbox_format_group = {
1475 .name = "format",
1476 .attrs = nhmex_uncore_mbox_formats_attr,
1477};
1478
1479static struct uncore_event_desc nhmex_uncore_mbox_events[] = {
1480 INTEL_UNCORE_EVENT_DESC(bbox_cmds_read, "inc_sel=0xd,fvc=0x2800"),
1481 INTEL_UNCORE_EVENT_DESC(bbox_cmds_write, "inc_sel=0xd,fvc=0x2820"),
1482 { /* end: all zeroes */ },
1483};
1484
1485static struct intel_uncore_ops nhmex_uncore_mbox_ops = {
1486 NHMEX_UNCORE_OPS_COMMON_INIT(),
1487 .enable_event = nhmex_mbox_msr_enable_event,
1488 .hw_config = nhmex_mbox_hw_config,
1489 .get_constraint = nhmex_mbox_get_constraint,
1490 .put_constraint = nhmex_mbox_put_constraint,
1491};
1492
1493static struct intel_uncore_type nhmex_uncore_mbox = {
1494 .name = "mbox",
1495 .num_counters = 6,
1496 .num_boxes = 2,
1497 .perf_ctr_bits = 48,
1498 .event_ctl = NHMEX_M0_MSR_PMU_CTL0,
1499 .perf_ctr = NHMEX_M0_MSR_PMU_CNT0,
1500 .event_mask = NHMEX_M_PMON_RAW_EVENT_MASK,
1501 .box_ctl = NHMEX_M0_MSR_GLOBAL_CTL,
1502 .msr_offset = NHMEX_M_MSR_OFFSET,
1503 .pair_ctr_ctl = 1,
1504 .num_shared_regs = 8,
1505 .event_descs = nhmex_uncore_mbox_events,
1506 .ops = &nhmex_uncore_mbox_ops,
1507 .format_group = &nhmex_uncore_mbox_format_group,
1508};
1509
1510void nhmex_rbox_alter_er(struct intel_uncore_box *box, struct perf_event *event)
1511{
1512 struct hw_perf_event *hwc = &event->hw;
1513 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1514 int port;
1515
1516 /* adjust the main event selector */
1517 if (reg1->idx % 2) {
1518 reg1->idx--;
1519 hwc->config -= 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT;
1520 } else {
1521 reg1->idx++;
1522 hwc->config += 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT;
1523 }
1524
1525 /* adjust address or config of extra register */
1526 port = reg1->idx / 6 + box->pmu->pmu_idx * 4;
1527 switch (reg1->idx % 6) {
1528 case 0:
1529 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port);
1530 break;
1531 case 1:
1532 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port);
1533 break;
1534 case 2:
1535 /* the 8~15 bits to the 0~7 bits */
1536 reg1->config >>= 8;
1537 break;
1538 case 3:
1539 /* the 0~7 bits to the 8~15 bits */
1540 reg1->config <<= 8;
1541 break;
1542 case 4:
1543 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port);
1544 break;
1545 case 5:
1546 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port);
1547 break;
1548 };
1549}
1550
1551/*
1552 * Each rbox has 4 event set which monitor PQI port 0~3 or 4~7.
1553 * An event set consists of 6 events, the 3rd and 4th events in
1554 * an event set use the same extra register. So an event set uses
1555 * 5 extra registers.
1556 */
1557static struct event_constraint *
1558nhmex_rbox_get_constraint(struct intel_uncore_box *box, struct perf_event *event)
1559{
1560 struct hw_perf_event *hwc = &event->hw;
1561 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1562 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1563 struct intel_uncore_extra_reg *er;
1564 unsigned long flags;
1565 int idx, er_idx;
1566 u64 config1;
1567 bool ok = false;
1568
1569 if (!uncore_box_is_fake(box) && reg1->alloc)
1570 return NULL;
1571
1572 idx = reg1->idx % 6;
1573 config1 = reg1->config;
1574again:
1575 er_idx = idx;
1576 /* the 3rd and 4th events use the same extra register */
1577 if (er_idx > 2)
1578 er_idx--;
1579 er_idx += (reg1->idx / 6) * 5;
1580
1581 er = &box->shared_regs[er_idx];
1582 raw_spin_lock_irqsave(&er->lock, flags);
1583 if (idx < 2) {
1584 if (!atomic_read(&er->ref) || er->config == reg1->config) {
1585 atomic_inc(&er->ref);
1586 er->config = reg1->config;
1587 ok = true;
1588 }
1589 } else if (idx == 2 || idx == 3) {
1590 /*
1591 * these two events use different fields in a extra register,
1592 * the 0~7 bits and the 8~15 bits respectively.
1593 */
1594 u64 mask = 0xff << ((idx - 2) * 8);
1595 if (!__BITS_VALUE(atomic_read(&er->ref), idx - 2, 8) ||
1596 !((er->config ^ config1) & mask)) {
1597 atomic_add(1 << ((idx - 2) * 8), &er->ref);
1598 er->config &= ~mask;
1599 er->config |= config1 & mask;
1600 ok = true;
1601 }
1602 } else {
1603 if (!atomic_read(&er->ref) ||
1604 (er->config == (hwc->config >> 32) &&
1605 er->config1 == reg1->config &&
1606 er->config2 == reg2->config)) {
1607 atomic_inc(&er->ref);
1608 er->config = (hwc->config >> 32);
1609 er->config1 = reg1->config;
1610 er->config2 = reg2->config;
1611 ok = true;
1612 }
1613 }
1614 raw_spin_unlock_irqrestore(&er->lock, flags);
1615
1616 if (!ok) {
1617 /*
1618 * The Rbox events are always in pairs. The paired
1619 * events are functional identical, but use different
1620 * extra registers. If we failed to take an extra
1621 * register, try the alternative.
1622 */
1623 if (idx % 2)
1624 idx--;
1625 else
1626 idx++;
1627 if (idx != reg1->idx % 6) {
1628 if (idx == 2)
1629 config1 >>= 8;
1630 else if (idx == 3)
1631 config1 <<= 8;
1632 goto again;
1633 }
1634 } else {
1635 if (!uncore_box_is_fake(box)) {
1636 if (idx != reg1->idx % 6)
1637 nhmex_rbox_alter_er(box, event);
1638 reg1->alloc = 1;
1639 }
1640 return NULL;
1641 }
1642 return &constraint_empty;
1643}
1644
1645static void nhmex_rbox_put_constraint(struct intel_uncore_box *box, struct perf_event *event)
1646{
1647 struct intel_uncore_extra_reg *er;
1648 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1649 int idx, er_idx;
1650
1651 if (uncore_box_is_fake(box) || !reg1->alloc)
1652 return;
1653
1654 idx = reg1->idx % 6;
1655 er_idx = idx;
1656 if (er_idx > 2)
1657 er_idx--;
1658 er_idx += (reg1->idx / 6) * 5;
1659
1660 er = &box->shared_regs[er_idx];
1661 if (idx == 2 || idx == 3)
1662 atomic_sub(1 << ((idx - 2) * 8), &er->ref);
1663 else
1664 atomic_dec(&er->ref);
1665
1666 reg1->alloc = 0;
1667}
1668
1669static int nhmex_rbox_hw_config(struct intel_uncore_box *box, struct perf_event *event)
1670{
1671 struct hw_perf_event *hwc = &event->hw;
1672 struct hw_perf_event_extra *reg1 = &event->hw.extra_reg;
1673 struct hw_perf_event_extra *reg2 = &event->hw.branch_reg;
1674 int port, idx;
1675
1676 idx = (event->hw.config & NHMEX_R_PMON_CTL_EV_SEL_MASK) >>
1677 NHMEX_R_PMON_CTL_EV_SEL_SHIFT;
1678 if (idx >= 0x18)
1679 return -EINVAL;
1680
1681 reg1->idx = idx;
1682 reg1->config = event->attr.config1;
1683
1684 port = idx / 6 + box->pmu->pmu_idx * 4;
1685 idx %= 6;
1686 switch (idx) {
1687 case 0:
1688 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port);
1689 break;
1690 case 1:
1691 reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port);
1692 break;
1693 case 2:
1694 case 3:
1695 reg1->reg = NHMEX_R_MSR_PORTN_QLX_CFG(port);
1696 break;
1697 case 4:
1698 case 5:
1699 if (idx == 4)
1700 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port);
1701 else
1702 reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port);
1703 reg2->config = event->attr.config2;
1704 hwc->config |= event->attr.config & (~0ULL << 32);
1705 break;
1706 };
1707 return 0;
1708}
1709
1710static u64 nhmex_rbox_shared_reg_config(struct intel_uncore_box *box, int idx)
1711{
1712 struct intel_uncore_extra_reg *er;
1713 unsigned long flags;
1714 u64 config;
1715
1716 er = &box->shared_regs[idx];
1717
1718 raw_spin_lock_irqsave(&er->lock, flags);
1719 config = er->config;
1720 raw_spin_unlock_irqrestore(&er->lock, flags);
1721
1722 return config;
1723}
1724
1725static void nhmex_rbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event)
1726{
1727 struct hw_perf_event *hwc = &event->hw;
1728 struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
1729 struct hw_perf_event_extra *reg2 = &hwc->branch_reg;
1730 int idx, er_idx;
1731
1732 idx = reg1->idx % 6;
1733 er_idx = idx;
1734 if (er_idx > 2)
1735 er_idx--;
1736 er_idx += (reg1->idx / 6) * 5;
1737
1738 switch (idx) {
1739 case 0:
1740 case 1:
1741 wrmsrl(reg1->reg, reg1->config);
1742 break;
1743 case 2:
1744 case 3:
1745 wrmsrl(reg1->reg, nhmex_rbox_shared_reg_config(box, er_idx));
1746 break;
1747 case 4:
1748 case 5:
1749 wrmsrl(reg1->reg, reg1->config);
1750 wrmsrl(reg1->reg + 1, hwc->config >> 32);
1751 wrmsrl(reg1->reg + 2, reg2->config);
1752 break;
1753 };
1754
1755 wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 |
1756 (hwc->config & NHMEX_R_PMON_CTL_EV_SEL_MASK));
1757}
1758
1759DEFINE_UNCORE_FORMAT_ATTR(xbr_match, xbr_match, "config:32-63");
1760DEFINE_UNCORE_FORMAT_ATTR(xbr_mm_cfg, xbr_mm_cfg, "config1:0-63");
1761DEFINE_UNCORE_FORMAT_ATTR(xbr_mask, xbr_mask, "config2:0-63");
1762DEFINE_UNCORE_FORMAT_ATTR(qlx_cfg, qlx_cfg, "config1:0-15");
1763DEFINE_UNCORE_FORMAT_ATTR(iperf_cfg, iperf_cfg, "config1:0-31");
1764
1765static struct attribute *nhmex_uncore_rbox_formats_attr[] = {
1766 &format_attr_event5.attr,
1767 &format_attr_xbr_mm_cfg.attr,
1768 &format_attr_xbr_match.attr,
1769 &format_attr_xbr_mask.attr,
1770 &format_attr_qlx_cfg.attr,
1771 &format_attr_iperf_cfg.attr,
1772 NULL,
1773};
1774
1775static struct attribute_group nhmex_uncore_rbox_format_group = {
1776 .name = "format",
1777 .attrs = nhmex_uncore_rbox_formats_attr,
1778};
1779
1780static struct uncore_event_desc nhmex_uncore_rbox_events[] = {
1781 INTEL_UNCORE_EVENT_DESC(qpi0_flit_send, "event=0x0,iperf_cfg=0x80000000"),
1782 INTEL_UNCORE_EVENT_DESC(qpi1_filt_send, "event=0x6,iperf_cfg=0x80000000"),
1783 INTEL_UNCORE_EVENT_DESC(qpi0_idle_filt, "event=0x0,iperf_cfg=0x40000000"),
1784 INTEL_UNCORE_EVENT_DESC(qpi1_idle_filt, "event=0x6,iperf_cfg=0x40000000"),
1785 INTEL_UNCORE_EVENT_DESC(qpi0_date_response, "event=0x0,iperf_cfg=0xc4"),
1786 INTEL_UNCORE_EVENT_DESC(qpi1_date_response, "event=0x6,iperf_cfg=0xc4"),
1787 { /* end: all zeroes */ },
1788};
1789
1790static struct intel_uncore_ops nhmex_uncore_rbox_ops = {
1791 NHMEX_UNCORE_OPS_COMMON_INIT(),
1792 .enable_event = nhmex_rbox_msr_enable_event,
1793 .hw_config = nhmex_rbox_hw_config,
1794 .get_constraint = nhmex_rbox_get_constraint,
1795 .put_constraint = nhmex_rbox_put_constraint,
1796};
1797
1798static struct intel_uncore_type nhmex_uncore_rbox = {
1799 .name = "rbox",
1800 .num_counters = 8,
1801 .num_boxes = 2,
1802 .perf_ctr_bits = 48,
1803 .event_ctl = NHMEX_R_MSR_PMON_CTL0,
1804 .perf_ctr = NHMEX_R_MSR_PMON_CNT0,
1805 .event_mask = NHMEX_R_PMON_RAW_EVENT_MASK,
1806 .box_ctl = NHMEX_R_MSR_GLOBAL_CTL,
1807 .msr_offset = NHMEX_R_MSR_OFFSET,
1808 .pair_ctr_ctl = 1,
1809 .num_shared_regs = 20,
1810 .event_descs = nhmex_uncore_rbox_events,
1811 .ops = &nhmex_uncore_rbox_ops,
1812 .format_group = &nhmex_uncore_rbox_format_group
1813};
1814
1815static struct intel_uncore_type *nhmex_msr_uncores[] = {
1816 &nhmex_uncore_ubox,
1817 &nhmex_uncore_cbox,
1818 &nhmex_uncore_bbox,
1819 &nhmex_uncore_sbox,
1820 &nhmex_uncore_mbox,
1821 &nhmex_uncore_rbox,
1822 &nhmex_uncore_wbox,
1823 NULL,
1824};
1825/* end of Nehalem-EX uncore support */
1826
1827static void uncore_assign_hw_event(struct intel_uncore_box *box, struct perf_event *event, int idx)
774{ 1828{
775 struct hw_perf_event *hwc = &event->hw; 1829 struct hw_perf_event *hwc = &event->hw;
776 1830
@@ -787,8 +1841,7 @@ static void uncore_assign_hw_event(struct intel_uncore_box *box,
787 hwc->event_base = uncore_perf_ctr(box, hwc->idx); 1841 hwc->event_base = uncore_perf_ctr(box, hwc->idx);
788} 1842}
789 1843
790static void uncore_perf_event_update(struct intel_uncore_box *box, 1844static void uncore_perf_event_update(struct intel_uncore_box *box, struct perf_event *event)
791 struct perf_event *event)
792{ 1845{
793 u64 prev_count, new_count, delta; 1846 u64 prev_count, new_count, delta;
794 int shift; 1847 int shift;
@@ -858,14 +1911,12 @@ static void uncore_pmu_init_hrtimer(struct intel_uncore_box *box)
858 box->hrtimer.function = uncore_pmu_hrtimer; 1911 box->hrtimer.function = uncore_pmu_hrtimer;
859} 1912}
860 1913
861struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, 1914struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, int cpu)
862 int cpu)
863{ 1915{
864 struct intel_uncore_box *box; 1916 struct intel_uncore_box *box;
865 int i, size; 1917 int i, size;
866 1918
867 size = sizeof(*box) + type->num_shared_regs * 1919 size = sizeof(*box) + type->num_shared_regs * sizeof(struct intel_uncore_extra_reg);
868 sizeof(struct intel_uncore_extra_reg);
869 1920
870 box = kmalloc_node(size, GFP_KERNEL | __GFP_ZERO, cpu_to_node(cpu)); 1921 box = kmalloc_node(size, GFP_KERNEL | __GFP_ZERO, cpu_to_node(cpu));
871 if (!box) 1922 if (!box)
@@ -915,12 +1966,11 @@ static struct intel_uncore_box *uncore_event_to_box(struct perf_event *event)
915 * perf core schedules event on the basis of cpu, uncore events are 1966 * perf core schedules event on the basis of cpu, uncore events are
916 * collected by one of the cpus inside a physical package. 1967 * collected by one of the cpus inside a physical package.
917 */ 1968 */
918 return uncore_pmu_to_box(uncore_event_to_pmu(event), 1969 return uncore_pmu_to_box(uncore_event_to_pmu(event), smp_processor_id());
919 smp_processor_id());
920} 1970}
921 1971
922static int uncore_collect_events(struct intel_uncore_box *box, 1972static int
923 struct perf_event *leader, bool dogrp) 1973uncore_collect_events(struct intel_uncore_box *box, struct perf_event *leader, bool dogrp)
924{ 1974{
925 struct perf_event *event; 1975 struct perf_event *event;
926 int n, max_count; 1976 int n, max_count;
@@ -952,8 +2002,7 @@ static int uncore_collect_events(struct intel_uncore_box *box,
952} 2002}
953 2003
954static struct event_constraint * 2004static struct event_constraint *
955uncore_get_event_constraint(struct intel_uncore_box *box, 2005uncore_get_event_constraint(struct intel_uncore_box *box, struct perf_event *event)
956 struct perf_event *event)
957{ 2006{
958 struct intel_uncore_type *type = box->pmu->type; 2007 struct intel_uncore_type *type = box->pmu->type;
959 struct event_constraint *c; 2008 struct event_constraint *c;
@@ -977,15 +2026,13 @@ uncore_get_event_constraint(struct intel_uncore_box *box,
977 return &type->unconstrainted; 2026 return &type->unconstrainted;
978} 2027}
979 2028
980static void uncore_put_event_constraint(struct intel_uncore_box *box, 2029static void uncore_put_event_constraint(struct intel_uncore_box *box, struct perf_event *event)
981 struct perf_event *event)
982{ 2030{
983 if (box->pmu->type->ops->put_constraint) 2031 if (box->pmu->type->ops->put_constraint)
984 box->pmu->type->ops->put_constraint(box, event); 2032 box->pmu->type->ops->put_constraint(box, event);
985} 2033}
986 2034
987static int uncore_assign_events(struct intel_uncore_box *box, 2035static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int n)
988 int assign[], int n)
989{ 2036{
990 unsigned long used_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)]; 2037 unsigned long used_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)];
991 struct event_constraint *c, *constraints[UNCORE_PMC_IDX_MAX]; 2038 struct event_constraint *c, *constraints[UNCORE_PMC_IDX_MAX];
@@ -1407,8 +2454,7 @@ static bool pcidrv_registered;
1407/* 2454/*
1408 * add a pci uncore device 2455 * add a pci uncore device
1409 */ 2456 */
1410static int __devinit uncore_pci_add(struct intel_uncore_type *type, 2457static int __devinit uncore_pci_add(struct intel_uncore_type *type, struct pci_dev *pdev)
1411 struct pci_dev *pdev)
1412{ 2458{
1413 struct intel_uncore_pmu *pmu; 2459 struct intel_uncore_pmu *pmu;
1414 struct intel_uncore_box *box; 2460 struct intel_uncore_box *box;
@@ -1485,6 +2531,7 @@ static int __devinit uncore_pci_probe(struct pci_dev *pdev,
1485 struct intel_uncore_type *type; 2531 struct intel_uncore_type *type;
1486 2532
1487 type = (struct intel_uncore_type *)id->driver_data; 2533 type = (struct intel_uncore_type *)id->driver_data;
2534
1488 return uncore_pci_add(type, pdev); 2535 return uncore_pci_add(type, pdev);
1489} 2536}
1490 2537
@@ -1612,8 +2659,8 @@ static int __cpuinit uncore_cpu_prepare(int cpu, int phys_id)
1612 return 0; 2659 return 0;
1613} 2660}
1614 2661
1615static void __cpuinit uncore_change_context(struct intel_uncore_type **uncores, 2662static void __cpuinit
1616 int old_cpu, int new_cpu) 2663uncore_change_context(struct intel_uncore_type **uncores, int old_cpu, int new_cpu)
1617{ 2664{
1618 struct intel_uncore_type *type; 2665 struct intel_uncore_type *type;
1619 struct intel_uncore_pmu *pmu; 2666 struct intel_uncore_pmu *pmu;
@@ -1694,8 +2741,8 @@ static void __cpuinit uncore_event_init_cpu(int cpu)
1694 uncore_change_context(pci_uncores, -1, cpu); 2741 uncore_change_context(pci_uncores, -1, cpu);
1695} 2742}
1696 2743
1697static int __cpuinit uncore_cpu_notifier(struct notifier_block *self, 2744static int
1698 unsigned long action, void *hcpu) 2745 __cpuinit uncore_cpu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
1699{ 2746{
1700 unsigned int cpu = (long)hcpu; 2747 unsigned int cpu = (long)hcpu;
1701 2748
@@ -1732,12 +2779,12 @@ static int __cpuinit uncore_cpu_notifier(struct notifier_block *self,
1732} 2779}
1733 2780
1734static struct notifier_block uncore_cpu_nb __cpuinitdata = { 2781static struct notifier_block uncore_cpu_nb __cpuinitdata = {
1735 .notifier_call = uncore_cpu_notifier, 2782 .notifier_call = uncore_cpu_notifier,
1736 /* 2783 /*
1737 * to migrate uncore events, our notifier should be executed 2784 * to migrate uncore events, our notifier should be executed
1738 * before perf core's notifier. 2785 * before perf core's notifier.
1739 */ 2786 */
1740 .priority = CPU_PRI_PERF + 1, 2787 .priority = CPU_PRI_PERF + 1,
1741}; 2788};
1742 2789
1743static void __init uncore_cpu_setup(void *dummy) 2790static void __init uncore_cpu_setup(void *dummy)
@@ -1767,6 +2814,9 @@ static int __init uncore_cpu_init(void)
1767 snbep_uncore_cbox.num_boxes = max_cores; 2814 snbep_uncore_cbox.num_boxes = max_cores;
1768 msr_uncores = snbep_msr_uncores; 2815 msr_uncores = snbep_msr_uncores;
1769 break; 2816 break;
2817 case 46:
2818 msr_uncores = nhmex_msr_uncores;
2819 break;
1770 default: 2820 default:
1771 return 0; 2821 return 0;
1772 } 2822 }
diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
index b13e9ea81def..f3851892e077 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h
+++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
@@ -5,8 +5,6 @@
5#include "perf_event.h" 5#include "perf_event.h"
6 6
7#define UNCORE_PMU_NAME_LEN 32 7#define UNCORE_PMU_NAME_LEN 32
8#define UNCORE_BOX_HASH_SIZE 8
9
10#define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC) 8#define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC)
11 9
12#define UNCORE_FIXED_EVENT 0xff 10#define UNCORE_FIXED_EVENT 0xff
@@ -115,6 +113,10 @@
115 SNBEP_PCU_MSR_PMON_CTL_OCC_INVERT | \ 113 SNBEP_PCU_MSR_PMON_CTL_OCC_INVERT | \
116 SNBEP_PCU_MSR_PMON_CTL_OCC_EDGE_DET) 114 SNBEP_PCU_MSR_PMON_CTL_OCC_EDGE_DET)
117 115
116#define SNBEP_QPI_PCI_PMON_RAW_EVENT_MASK \
117 (SNBEP_PMON_RAW_EVENT_MASK | \
118 SNBEP_PMON_CTL_EV_SEL_EXT)
119
118/* SNB-EP pci control register */ 120/* SNB-EP pci control register */
119#define SNBEP_PCI_PMON_BOX_CTL 0xf4 121#define SNBEP_PCI_PMON_BOX_CTL 0xf4
120#define SNBEP_PCI_PMON_CTL0 0xd8 122#define SNBEP_PCI_PMON_CTL0 0xd8
@@ -158,6 +160,193 @@
158#define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc 160#define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc
159#define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd 161#define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd
160 162
163/* NHM-EX event control */
164#define NHMEX_PMON_CTL_EV_SEL_MASK 0x000000ff
165#define NHMEX_PMON_CTL_UMASK_MASK 0x0000ff00
166#define NHMEX_PMON_CTL_EN_BIT0 (1 << 0)
167#define NHMEX_PMON_CTL_EDGE_DET (1 << 18)
168#define NHMEX_PMON_CTL_PMI_EN (1 << 20)
169#define NHMEX_PMON_CTL_EN_BIT22 (1 << 22)
170#define NHMEX_PMON_CTL_INVERT (1 << 23)
171#define NHMEX_PMON_CTL_TRESH_MASK 0xff000000
172#define NHMEX_PMON_RAW_EVENT_MASK (NHMEX_PMON_CTL_EV_SEL_MASK | \
173 NHMEX_PMON_CTL_UMASK_MASK | \
174 NHMEX_PMON_CTL_EDGE_DET | \
175 NHMEX_PMON_CTL_INVERT | \
176 NHMEX_PMON_CTL_TRESH_MASK)
177
178/* NHM-EX Ubox */
179#define NHMEX_U_MSR_PMON_GLOBAL_CTL 0xc00
180#define NHMEX_U_MSR_PMON_CTR 0xc11
181#define NHMEX_U_MSR_PMON_EV_SEL 0xc10
182
183#define NHMEX_U_PMON_GLOBAL_EN (1 << 0)
184#define NHMEX_U_PMON_GLOBAL_PMI_CORE_SEL 0x0000001e
185#define NHMEX_U_PMON_GLOBAL_EN_ALL (1 << 28)
186#define NHMEX_U_PMON_GLOBAL_RST_ALL (1 << 29)
187#define NHMEX_U_PMON_GLOBAL_FRZ_ALL (1 << 31)
188
189#define NHMEX_U_PMON_RAW_EVENT_MASK \
190 (NHMEX_PMON_CTL_EV_SEL_MASK | \
191 NHMEX_PMON_CTL_EDGE_DET)
192
193/* NHM-EX Cbox */
194#define NHMEX_C0_MSR_PMON_GLOBAL_CTL 0xd00
195#define NHMEX_C0_MSR_PMON_CTR0 0xd11
196#define NHMEX_C0_MSR_PMON_EV_SEL0 0xd10
197#define NHMEX_C_MSR_OFFSET 0x20
198
199/* NHM-EX Bbox */
200#define NHMEX_B0_MSR_PMON_GLOBAL_CTL 0xc20
201#define NHMEX_B0_MSR_PMON_CTR0 0xc31
202#define NHMEX_B0_MSR_PMON_CTL0 0xc30
203#define NHMEX_B_MSR_OFFSET 0x40
204#define NHMEX_B0_MSR_MATCH 0xe45
205#define NHMEX_B0_MSR_MASK 0xe46
206#define NHMEX_B1_MSR_MATCH 0xe4d
207#define NHMEX_B1_MSR_MASK 0xe4e
208
209#define NHMEX_B_PMON_CTL_EN (1 << 0)
210#define NHMEX_B_PMON_CTL_EV_SEL_SHIFT 1
211#define NHMEX_B_PMON_CTL_EV_SEL_MASK \
212 (0x1f << NHMEX_B_PMON_CTL_EV_SEL_SHIFT)
213#define NHMEX_B_PMON_CTR_SHIFT 6
214#define NHMEX_B_PMON_CTR_MASK \
215 (0x3 << NHMEX_B_PMON_CTR_SHIFT)
216#define NHMEX_B_PMON_RAW_EVENT_MASK \
217 (NHMEX_B_PMON_CTL_EV_SEL_MASK | \
218 NHMEX_B_PMON_CTR_MASK)
219
220/* NHM-EX Sbox */
221#define NHMEX_S0_MSR_PMON_GLOBAL_CTL 0xc40
222#define NHMEX_S0_MSR_PMON_CTR0 0xc51
223#define NHMEX_S0_MSR_PMON_CTL0 0xc50
224#define NHMEX_S_MSR_OFFSET 0x80
225#define NHMEX_S0_MSR_MM_CFG 0xe48
226#define NHMEX_S0_MSR_MATCH 0xe49
227#define NHMEX_S0_MSR_MASK 0xe4a
228#define NHMEX_S1_MSR_MM_CFG 0xe58
229#define NHMEX_S1_MSR_MATCH 0xe59
230#define NHMEX_S1_MSR_MASK 0xe5a
231
232#define NHMEX_S_PMON_MM_CFG_EN (0x1ULL << 63)
233
234/* NHM-EX Mbox */
235#define NHMEX_M0_MSR_GLOBAL_CTL 0xca0
236#define NHMEX_M0_MSR_PMU_DSP 0xca5
237#define NHMEX_M0_MSR_PMU_ISS 0xca6
238#define NHMEX_M0_MSR_PMU_MAP 0xca7
239#define NHMEX_M0_MSR_PMU_MSC_THR 0xca8
240#define NHMEX_M0_MSR_PMU_PGT 0xca9
241#define NHMEX_M0_MSR_PMU_PLD 0xcaa
242#define NHMEX_M0_MSR_PMU_ZDP_CTL_FVC 0xcab
243#define NHMEX_M0_MSR_PMU_CTL0 0xcb0
244#define NHMEX_M0_MSR_PMU_CNT0 0xcb1
245#define NHMEX_M_MSR_OFFSET 0x40
246#define NHMEX_M0_MSR_PMU_MM_CFG 0xe54
247#define NHMEX_M1_MSR_PMU_MM_CFG 0xe5c
248
249#define NHMEX_M_PMON_MM_CFG_EN (1ULL << 63)
250#define NHMEX_M_PMON_ADDR_MATCH_MASK 0x3ffffffffULL
251#define NHMEX_M_PMON_ADDR_MASK_MASK 0x7ffffffULL
252#define NHMEX_M_PMON_ADDR_MASK_SHIFT 34
253
254#define NHMEX_M_PMON_CTL_EN (1 << 0)
255#define NHMEX_M_PMON_CTL_PMI_EN (1 << 1)
256#define NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT 2
257#define NHMEX_M_PMON_CTL_COUNT_MODE_MASK \
258 (0x3 << NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT)
259#define NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT 4
260#define NHMEX_M_PMON_CTL_STORAGE_MODE_MASK \
261 (0x3 << NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT)
262#define NHMEX_M_PMON_CTL_WRAP_MODE (1 << 6)
263#define NHMEX_M_PMON_CTL_FLAG_MODE (1 << 7)
264#define NHMEX_M_PMON_CTL_INC_SEL_SHIFT 9
265#define NHMEX_M_PMON_CTL_INC_SEL_MASK \
266 (0x1f << NHMEX_M_PMON_CTL_INC_SEL_SHIFT)
267#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT 19
268#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK \
269 (0x7 << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT)
270#define NHMEX_M_PMON_RAW_EVENT_MASK \
271 (NHMEX_M_PMON_CTL_COUNT_MODE_MASK | \
272 NHMEX_M_PMON_CTL_STORAGE_MODE_MASK | \
273 NHMEX_M_PMON_CTL_WRAP_MODE | \
274 NHMEX_M_PMON_CTL_FLAG_MODE | \
275 NHMEX_M_PMON_CTL_INC_SEL_MASK | \
276 NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK)
277
278
279#define NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK 0x1f
280#define NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK (0x7 << 5)
281#define NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK (0x7 << 8)
282#define NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR (1 << 23)
283#define NHMEX_M_PMON_ZDP_CTL_FVC_MASK \
284 (NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK | \
285 NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK | \
286 NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK | \
287 NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR)
288#define NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (11 + 3 * (n)))
289
290/*
291 * use the 9~13 bits to select event If the 7th bit is not set,
292 * otherwise use the 19~21 bits to select event.
293 */
294#define MBOX_INC_SEL(x) ((x) << NHMEX_M_PMON_CTL_INC_SEL_SHIFT)
295#define MBOX_SET_FLAG_SEL(x) (((x) << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) | \
296 NHMEX_M_PMON_CTL_FLAG_MODE)
297#define MBOX_INC_SEL_MASK (NHMEX_M_PMON_CTL_INC_SEL_MASK | \
298 NHMEX_M_PMON_CTL_FLAG_MODE)
299#define MBOX_SET_FLAG_SEL_MASK (NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK | \
300 NHMEX_M_PMON_CTL_FLAG_MODE)
301#define MBOX_INC_SEL_EXTAR_REG(c, r) \
302 EVENT_EXTRA_REG(MBOX_INC_SEL(c), NHMEX_M0_MSR_PMU_##r, \
303 MBOX_INC_SEL_MASK, (u64)-1, NHMEX_M_##r)
304#define MBOX_SET_FLAG_SEL_EXTRA_REG(c, r) \
305 EVENT_EXTRA_REG(MBOX_SET_FLAG_SEL(c), NHMEX_M0_MSR_PMU_##r, \
306 MBOX_SET_FLAG_SEL_MASK, \
307 (u64)-1, NHMEX_M_##r)
308
309/* NHM-EX Rbox */
310#define NHMEX_R_MSR_GLOBAL_CTL 0xe00
311#define NHMEX_R_MSR_PMON_CTL0 0xe10
312#define NHMEX_R_MSR_PMON_CNT0 0xe11
313#define NHMEX_R_MSR_OFFSET 0x20
314
315#define NHMEX_R_MSR_PORTN_QLX_CFG(n) \
316 ((n) < 4 ? (0xe0c + (n)) : (0xe2c + (n) - 4))
317#define NHMEX_R_MSR_PORTN_IPERF_CFG0(n) (0xe04 + (n))
318#define NHMEX_R_MSR_PORTN_IPERF_CFG1(n) (0xe24 + (n))
319#define NHMEX_R_MSR_PORTN_XBR_OFFSET(n) \
320 (((n) < 4 ? 0 : 0x10) + (n) * 4)
321#define NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) \
322 (0xe60 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n))
323#define NHMEX_R_MSR_PORTN_XBR_SET1_MATCH(n) \
324 (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 1)
325#define NHMEX_R_MSR_PORTN_XBR_SET1_MASK(n) \
326 (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 2)
327#define NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) \
328 (0xe70 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n))
329#define NHMEX_R_MSR_PORTN_XBR_SET2_MATCH(n) \
330 (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 1)
331#define NHMEX_R_MSR_PORTN_XBR_SET2_MASK(n) \
332 (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 2)
333
334#define NHMEX_R_PMON_CTL_EN (1 << 0)
335#define NHMEX_R_PMON_CTL_EV_SEL_SHIFT 1
336#define NHMEX_R_PMON_CTL_EV_SEL_MASK \
337 (0x1f << NHMEX_R_PMON_CTL_EV_SEL_SHIFT)
338#define NHMEX_R_PMON_CTL_PMI_EN (1 << 6)
339#define NHMEX_R_PMON_RAW_EVENT_MASK NHMEX_R_PMON_CTL_EV_SEL_MASK
340
341/* NHM-EX Wbox */
342#define NHMEX_W_MSR_GLOBAL_CTL 0xc80
343#define NHMEX_W_MSR_PMON_CNT0 0xc90
344#define NHMEX_W_MSR_PMON_EVT_SEL0 0xc91
345#define NHMEX_W_MSR_PMON_FIXED_CTR 0x394
346#define NHMEX_W_MSR_PMON_FIXED_CTL 0x395
347
348#define NHMEX_W_PMON_GLOBAL_FIXED_EN (1ULL << 31)
349
161struct intel_uncore_ops; 350struct intel_uncore_ops;
162struct intel_uncore_pmu; 351struct intel_uncore_pmu;
163struct intel_uncore_box; 352struct intel_uncore_box;
@@ -178,6 +367,7 @@ struct intel_uncore_type {
178 unsigned msr_offset; 367 unsigned msr_offset;
179 unsigned num_shared_regs:8; 368 unsigned num_shared_regs:8;
180 unsigned single_fixed:1; 369 unsigned single_fixed:1;
370 unsigned pair_ctr_ctl:1;
181 struct event_constraint unconstrainted; 371 struct event_constraint unconstrainted;
182 struct event_constraint *constraints; 372 struct event_constraint *constraints;
183 struct intel_uncore_pmu *pmus; 373 struct intel_uncore_pmu *pmus;
@@ -213,7 +403,7 @@ struct intel_uncore_pmu {
213 403
214struct intel_uncore_extra_reg { 404struct intel_uncore_extra_reg {
215 raw_spinlock_t lock; 405 raw_spinlock_t lock;
216 u64 config1; 406 u64 config, config1, config2;
217 atomic_t ref; 407 atomic_t ref;
218}; 408};
219 409
@@ -323,14 +513,16 @@ unsigned uncore_msr_fixed_ctr(struct intel_uncore_box *box)
323static inline 513static inline
324unsigned uncore_msr_event_ctl(struct intel_uncore_box *box, int idx) 514unsigned uncore_msr_event_ctl(struct intel_uncore_box *box, int idx)
325{ 515{
326 return idx + box->pmu->type->event_ctl + 516 return box->pmu->type->event_ctl +
517 (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) +
327 box->pmu->type->msr_offset * box->pmu->pmu_idx; 518 box->pmu->type->msr_offset * box->pmu->pmu_idx;
328} 519}
329 520
330static inline 521static inline
331unsigned uncore_msr_perf_ctr(struct intel_uncore_box *box, int idx) 522unsigned uncore_msr_perf_ctr(struct intel_uncore_box *box, int idx)
332{ 523{
333 return idx + box->pmu->type->perf_ctr + 524 return box->pmu->type->perf_ctr +
525 (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) +
334 box->pmu->type->msr_offset * box->pmu->pmu_idx; 526 box->pmu->type->msr_offset * box->pmu->pmu_idx;
335} 527}
336 528
@@ -422,3 +614,8 @@ static inline void uncore_box_init(struct intel_uncore_box *box)
422 box->pmu->type->ops->init_box(box); 614 box->pmu->type->ops->init_box(box);
423 } 615 }
424} 616}
617
618static inline bool uncore_box_is_fake(struct intel_uncore_box *box)
619{
620 return (box->phys_id < 0);
621}
diff --git a/arch/x86/kernel/cpu/sched.c b/arch/x86/kernel/cpu/sched.c
deleted file mode 100644
index a640ae5ad201..000000000000
--- a/arch/x86/kernel/cpu/sched.c
+++ /dev/null
@@ -1,55 +0,0 @@
1#include <linux/sched.h>
2#include <linux/math64.h>
3#include <linux/percpu.h>
4#include <linux/irqflags.h>
5
6#include <asm/cpufeature.h>
7#include <asm/processor.h>
8
9#ifdef CONFIG_SMP
10
11static DEFINE_PER_CPU(struct aperfmperf, old_perf_sched);
12
13static unsigned long scale_aperfmperf(void)
14{
15 struct aperfmperf val, *old = &__get_cpu_var(old_perf_sched);
16 unsigned long ratio, flags;
17
18 local_irq_save(flags);
19 get_aperfmperf(&val);
20 local_irq_restore(flags);
21
22 ratio = calc_aperfmperf_ratio(old, &val);
23 *old = val;
24
25 return ratio;
26}
27
28unsigned long arch_scale_freq_power(struct sched_domain *sd, int cpu)
29{
30 /*
31 * do aperf/mperf on the cpu level because it includes things
32 * like turbo mode, which are relevant to full cores.
33 */
34 if (boot_cpu_has(X86_FEATURE_APERFMPERF))
35 return scale_aperfmperf();
36
37 /*
38 * maybe have something cpufreq here
39 */
40
41 return default_scale_freq_power(sd, cpu);
42}
43
44unsigned long arch_scale_smt_power(struct sched_domain *sd, int cpu)
45{
46 /*
47 * aperf/mperf already includes the smt gain
48 */
49 if (boot_cpu_has(X86_FEATURE_APERFMPERF))
50 return SCHED_LOAD_SCALE;
51
52 return default_scale_smt_power(sd, cpu);
53}
54
55#endif
generated by cgit v1.2.2 (git 2.25.0) at 2026-05-27 16:53:31 -0400