perf/x86/rapl: Enable Apollo Lake RAPL support
Subsystems (2)
PATCH 1/2 perf/x86/rapl: Enable Apollo Lake RAPL support
This patch enables RAPL counters (energy consumption counters) support for Intel Apollo Lake (Goldmont) processors (Model 92):
RAPL of Goldmont, unlikes ESU increment of Silvermont/Airmont, it likes the Haswell microarchitecture in 1/2^ESU joules and supports power domains in PP0/PP1/PKG/RAM.
ESU and power domains refer to Intel Software Developers' Manual, Vol. 3C, Order No. 325384, Table 35-12.
Usage example:
$ perf list $ perf stat -a -e power/energy-cores/,power/energy-pkg/ sleep 10
--- arch/x86/events/intel/rapl.c | 2 ++ 1 file changed, 2 insertions(+)
arch/x86/events/intel/rapl.c +2 -0
@@ -765,6 +765,8 @@ static const struct x86_cpu_id rapl_cpu_match[] __initconst = {X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_MOBILE , skl_rapl_init),X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_DESKTOP , skl_rapl_init),X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_X , hsx_rapl_init), + + X86_RAPL_MODEL_MATCH(INTEL_FAM6_ATOM_GOLDMONT, hsw_rapl_init), {}, };
Signature · 1 line
2.6.6
PATCH 2/2 perf/x86/rapl: Enable Baytrail/Braswell RAPL support
This patch enables RAPL counters (energy consumption counters) support for Intel Baytrail and Braswell processors (Model 55 and 76):
The Silvermont/Airmont microarchitecture actually uses fixed energy status unit (ESU) in smallest unit of microjoule, this patch adds quirk for these Atom processors (BYT/BSW) to calculate energy increment in 2^ESU microjoules.
ESU and power domains refer to Intel Software Developers' Manual, Vol. 3C, Order No. 325384, Table 35-8.
Usage example:
$ perf list $ perf stat -a -e power/energy-cores/,power/energy-pkg/ sleep 10
This patch also enables multiple quirks.
--- arch/x86/events/intel/rapl.c | 78 ++++++++++++++++++++++++++++++++++++++------ 1 file changed, 68 insertions(+), 10 deletions(-)
arch/x86/events/intel/rapl.c +68 -10
@@ -110,6 +110,10 @@ static const char *const rapl_domain_names[NR_RAPL_DOMAINS] __initconst = {#define RAPL_IDX_KNL (1<<RAPL_IDX_PKG_NRG_STAT|\ 1<<RAPL_IDX_RAM_NRG_STAT) +/* Baytrail/Braswell clients have PP0, PKG */ +#define RAPL_IDX_BYT (1<<RAPL_IDX_PP0_NRG_STAT|\ + 1<<RAPL_IDX_PKG_NRG_STAT) +/* * event code: LSB 8 bits, passed in attr->config * any other bit is reserved
@@ -136,6 +140,11 @@ static struct perf_pmu_events_attr event_attr_##v = { \ .event_str = str, \ }; +enum rapl_quirk { +RAPL_HSX_QUIRK =1 , +RAPL_BYT_QUIRK , +}; +struct rapl_pmu {raw_spinlock_t lock;int n_active;
@@ -158,6 +167,7 @@ static struct rapl_pmus *rapl_pmus;static cpumask_t rapl_cpu_mask;static unsigned int rapl_cntr_mask;static u64 rapl_timer_ms; +static bool is_baytrail;static inline struct rapl_pmu *cpu_to_rapl_pmu (unsigned int cpu) {
@@ -177,6 +187,16 @@ static inline u64 rapl_scale(u64 v, int cfg)pr_warn ("Invalid domain %d, failed to scale data\n" , cfg);return v; } + +/* +* Some Atom series processors (BYT/BSW) use 2^ESU microjoules. +* +* TODO: this looks hacky, it's better to refactor scale-up mechanism +* to compromise the main stream processors and Atom ones. +*/ +if (is_baytrail) +return v << rapl_hw_unit[cfg -1 ]; +/* * scale delta to smallest unit (1/2^32) * users must then scale back: count * 1/(1e9*2^32) to get Joules
@@ -452,6 +472,14 @@ RAPL_EVENT_ATTR_STR(energy-ram.scale, rapl_ram_scale, "2.3283064365386962890RAPL_EVENT_ATTR_STR (energy-gpu.scale, rapl_gpu_scale,"2.3283064365386962890625e-10" );RAPL_EVENT_ATTR_STR (energy-psys.scale, rapl_psys_scale,"2.3283064365386962890625e-10" ); +/* +* Some Atom series processors (BYT/BSW) have fixed +* energy status unit (ESU) in smallest unit of microjoule, +* and its increment is in 2^ESU microjoules. +*/ +RAPL_EVENT_ATTR_STR (energy-cores.scale, rapl_byt_cores_scale,"1.0e-6" ); +RAPL_EVENT_ATTR_STR (energy-pkg.scale, rapl_byt_pkg_scale,"1.0e-6" ); +static struct attribute *rapl_events_srv_attr[] = {EVENT_PTR (rapl_cores),EVENT_PTR (rapl_pkg),
@@ -533,6 +561,18 @@ static struct attribute *rapl_events_knl_attr[] = { NULL, }; +static struct attribute *rapl_events_byt_attr[] = { +EVENT_PTR (rapl_cores), +EVENT_PTR (rapl_pkg), + +EVENT_PTR (rapl_cores_unit), +EVENT_PTR (rapl_pkg_unit), + +EVENT_PTR (rapl_byt_cores_scale), +EVENT_PTR (rapl_byt_pkg_scale), +NULL , +}; +static struct attribute_group rapl_pmu_events_group = { .name ="events" , .attrs =NULL ,/* patched at runtime */
@@ -617,7 +657,7 @@ static int rapl_cpu_prepare(unsigned int cpu)return 0 ; } -static int rapl_check_hw_unit (bool apply_quirk) +static int rapl_check_hw_unit (enum rapl_quirk apply_quirk) {u64 msr_rapl_power_unit_bits;int i;
@@ -634,10 +674,20 @@ static int rapl_check_hw_unit(bool apply_quirk)* "Intel Xeon Processor E5-1600 and E5-2600 v3 Product Families, V2 * of 2. Datasheet, September 2014, Reference Number: 330784-001 " */ -if (apply_quirk) +if (apply_quirk ==RAPL_HSX_QUIRK ) rapl_hw_unit[RAPL_IDX_RAM_NRG_STAT ] =16 ;/* +* Some Atom processors (BYT/BSW) have 2^ESU microjoules increment, +* refer to Software Developers' Manual, Vol. 3C, Order No. 325384, +* Table 35-8 of MSR_RAPL_POWER_UNIT +*/ +if (apply_quirk ==RAPL_BYT_QUIRK ) + is_baytrail = true; +else + is_baytrail = false; + +/* * Calculate the timer rate: * Use reference of 200W for scaling the timeout to avoid counter * overflows. 200W = 200 Joules/sec
@@ -702,47 +752,53 @@ static int __init init_rapl_pmus(void) { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&init }struct intel_rapl_init_fun { -bool apply_quirk; +enum rapl_quirk apply_quirk;int cntr_mask;struct attribute **attrs; };static const struct intel_rapl_init_fun snb_rapl_init__initconst = { - .apply_quirk = false, + .apply_quirk =0 , .cntr_mask =RAPL_IDX_CLN , .attrs = rapl_events_cln_attr, };static const struct intel_rapl_init_fun hsx_rapl_init__initconst = { - .apply_quirk = true, + .apply_quirk =RAPL_HSX_QUIRK , .cntr_mask =RAPL_IDX_SRV , .attrs = rapl_events_srv_attr, };static const struct intel_rapl_init_fun hsw_rapl_init__initconst = { - .apply_quirk = false, + .apply_quirk =0 , .cntr_mask =RAPL_IDX_HSW , .attrs = rapl_events_hsw_attr, };static const struct intel_rapl_init_fun snbep_rapl_init__initconst = { - .apply_quirk = false, + .apply_quirk =0 , .cntr_mask =RAPL_IDX_SRV , .attrs = rapl_events_srv_attr, };static const struct intel_rapl_init_fun knl_rapl_init__initconst = { - .apply_quirk = true, + .apply_quirk =RAPL_HSX_QUIRK , .cntr_mask =RAPL_IDX_KNL , .attrs = rapl_events_knl_attr, };static const struct intel_rapl_init_fun skl_rapl_init__initconst = { - .apply_quirk = false, + .apply_quirk =0 , .cntr_mask =RAPL_IDX_SKL_CLN , .attrs = rapl_events_skl_attr, }; +static const struct intel_rapl_init_fun byt_rapl_init__initconst = { + .apply_quirk =RAPL_BYT_QUIRK , + .cntr_mask =RAPL_IDX_BYT , + .attrs = rapl_events_byt_attr, +}; +static const struct x86_cpu_id rapl_cpu_match[]__initconst = {X86_RAPL_MODEL_MATCH (INTEL_FAM6_SANDYBRIDGE , snb_rapl_init),X86_RAPL_MODEL_MATCH (INTEL_FAM6_SANDYBRIDGE_X , snbep_rapl_init),
@@ -766,6 +822,8 @@ static const struct x86_cpu_id rapl_cpu_match[] __initconst = {X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_DESKTOP , skl_rapl_init),X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_X , hsx_rapl_init), +X86_RAPL_MODEL_MATCH (INTEL_FAM6_ATOM_SILVERMONT1 , byt_rapl_init), +X86_RAPL_MODEL_MATCH (INTEL_FAM6_ATOM_AIRMONT , byt_rapl_init), X86_RAPL_MODEL_MATCH(INTEL_FAM6_ATOM_GOLDMONT, hsw_rapl_init), {}, };
@@ -776,7 +834,7 @@ static int __init rapl_pmu_init(void) {const struct x86_cpu_id *id;struct intel_rapl_init_fun *rapl_init; -bool apply_quirk; +enum rapl_quirk apply_quirk;int ret; id =x86_match_cpu (rapl_cpu_match);
Signature · 1 line
2.6.6
Peter Zijlstra
maintainerRe: [PATCH 2/2] perf/x86/rapl: Enable Baytrail/Braswell RAPL support
Quoted 4 lines · Harry Pan
arch/x86/events/intel/rapl.c +0
-0@@ -177,6 +187,16 @@ static inline u64 rapl_scale(u64 v, int cfg)pr_warn ("Invalid domain %d, failed to scale data\n" , cfg);return v; }
arch/x86/events/intel/rapl.c static inline u64 rapl_scale(u64 v, int cfg) · line 199
Quoted 10 lines
arch/x86/events/intel/rapl.c +10
-0+ +/* +* Some Atom series processors (BYT/BSW) use 2^ESU microjoules. +* +* TODO: this looks hacky, it's better to refactor scale-up mechanism +* to compromise the main stream processors and Atom ones. +*/ +if (is_baytrail) +return v << rapl_hw_unit[cfg -1 ]; +
Can't you simply set rapl_hw_unit[] such that 32 - rapl_hw_unit[] ends up at the right number? Then you only get to much with values in rapl_check_hw_unit without runtime overhead later.
Quoted 11 lines
arch/x86/events/intel/rapl.c +2
-1+static int rapl_check_hw_unit (enum rapl_quirk apply_quirk) {u64 msr_rapl_power_unit_bits;int i;@@ -634,10 +674,20 @@ static int rapl_check_hw_unit(bool apply_quirk)* "Intel Xeon Processor E5-1600 and E5-2600 v3 Product Families, V2 * of 2. Datasheet, September 2014, Reference Number: 330784-001 " */ -if (apply_quirk) +if (apply_quirk ==RAPL_HSX_QUIRK ) rapl_hw_unit[RAPL_IDX_RAM_NRG_STAT ] =16 ;
arch/x86/events/intel/rapl.c static int rapl_check_hw_unit(bool apply_quirk) · line 688
Quoted 10 lines
arch/x86/events/intel/rapl.c +8
-0/* +* Some Atom processors (BYT/BSW) have 2^ESU microjoules increment, +* refer to Software Developers' Manual, Vol. 3C, Order No. 325384, +* Table 35-8 of MSR_RAPL_POWER_UNIT +*/ +if (apply_quirk ==RAPL_BYT_QUIRK ) + is_baytrail = true; +else + is_baytrail = false;
it was already false...
/*
* comment explaining quirk goes here...
*/
if (apply_quirk = RAPL_BYT_QUIRK) {
for (i = 0; i < NR_RAPL_DOMAINS; i++)
rapl_hw_unit[i] = 32 - rapl_hw_unit[i];
}and then you get to verify what to do with rapl_timer_ms.
arch/x86/events/intel/rapl.c static int __init init_rapl_pmus(void) · line 761
Quoted 3 lines
arch/x86/events/intel/rapl.c +1
-1static const struct intel_rapl_init_fun snb_rapl_init__initconst = { - .apply_quirk = false, + .apply_quirk =0 ,
Either leave it out (unmentioned members get initialized to 0) or add RAPL_NO_QUIRK or so.
Quoted 4 lines
arch/x86/events/intel/rapl.c +0
-0.cntr_mask = RAPL_IDX_CLN, .attrs = rapl_events_cln_attr, };
Re: [PATCH 2/2] perf/x86/rapl: Enable Baytrail/Braswell RAPL support
Hi Peter,
Totally agreed and uploaded patchset again.
https://lkml.org/lkml/2016/9/9/467
https://lkml.org/lkml/2016/9/9/468
One more thing, I did not refine rapl_advertise() description. Advice is welcome.
Sincerely,
Harry
Quoted 68 lines · Peter Zijlstra
Quoted 4 lines · Harry Pan
arch/x86/events/intel/rapl.c +0
-0@@ -177,6 +187,16 @@ static inline u64 rapl_scale(u64 v, int cfg)pr_warn ("Invalid domain %d, failed to scale data\n" , cfg);return v; }arch/x86/events/intel/rapl.c static inline u64 rapl_scale(u64 v, int cfg) · line 199
Quoted 10 lines
arch/x86/events/intel/rapl.c +10
-0+ +/* +* Some Atom series processors (BYT/BSW) use 2^ESU microjoules. +* +* TODO: this looks hacky, it's better to refactor scale-up mechanism +* to compromise the main stream processors and Atom ones. +*/ +if (is_baytrail) +return v << rapl_hw_unit[cfg -1 ]; +Can't you simply set rapl_hw_unit[] such that 32 - rapl_hw_unit[] ends up at the right number? Then you only get to much with values in
rapl_check_hw_unitwithout runtime overhead later.Quoted 11 lines
arch/x86/events/intel/rapl.c +2
-1+static int rapl_check_hw_unit (enum rapl_quirk apply_quirk) {u64 msr_rapl_power_unit_bits;int i;@@ -634,10 +674,20 @@ static int rapl_check_hw_unit(bool apply_quirk)* "Intel Xeon Processor E5-1600 and E5-2600 v3 Product Families, V2 * of 2. Datasheet, September 2014, Reference Number: 330784-001 " */ -if (apply_quirk) +if (apply_quirk ==RAPL_HSX_QUIRK ) rapl_hw_unit[RAPL_IDX_RAM_NRG_STAT ] =16 ;arch/x86/events/intel/rapl.c static int rapl_check_hw_unit(bool apply_quirk) · line 688
Quoted 10 lines
arch/x86/events/intel/rapl.c +8
-0/* +* Some Atom processors (BYT/BSW) have 2^ESU microjoules increment, +* refer to Software Developers' Manual, Vol. 3C, Order No. 325384, +* Table 35-8 of MSR_RAPL_POWER_UNIT +*/ +if (apply_quirk ==RAPL_BYT_QUIRK ) + is_baytrail = true; +else + is_baytrail = false;it was already false...
/* * comment explaining quirk goes here... */ if (apply_quirk = RAPL_BYT_QUIRK) { for (i = 0; i < NR_RAPL_DOMAINS; i++) rapl_hw_unit[i] = 32 - rapl_hw_unit[i]; }and then you get to verify what to do with
rapl_timer_ms.arch/x86/events/intel/rapl.c static int __init init_rapl_pmus(void) · line 761
Quoted 3 lines
arch/x86/events/intel/rapl.c +1
-1static const struct intel_rapl_init_fun snb_rapl_init__initconst = { - .apply_quirk = false, + .apply_quirk =0 ,Either leave it out (unmentioned members get initialized to 0) or add
RAPL_NO_QUIRKor so.Quoted 4 lines
arch/x86/events/intel/rapl.c +0
-0.cntr_mask = RAPL_IDX_CLN, .attrs = rapl_events_cln_attr, };
tip-bot for Harry Pan
merge botperf/x86/rapl: Enable Apollo Lake RAPL support
Commit-ID: 2668c6195685f4b6f281767d10b4f4f2e32c2305 Gitweb: http://git.kernel.org/tip/2668c6195685f4b6f281767d10b4f4f2e32c2305 Author: Harry Pan <[email protected]> AuthorDate: Thu, 8 Sep 2016 17:08:57 +0800 Committer: Ingo Molnar <[email protected]> CommitDate: Sat, 10 Sep 2016 11:18:52 +0200
perf/x86/rapl: Enable Apollo Lake RAPL support
This patch enables RAPL counters (energy consumption counters) support for Intel Apollo Lake (Goldmont) processors (Model 92):
RAPL of Goldmont, unlikes ESU increment of Silvermont/Airmont, it likes the Haswell microarchitecture in 1/2^ESU joules and supports power domains in PP0/PP1/PKG/RAM.
ESU and power domains refer to Intel Software Developers' Manual, Vol. 3C, Order No. 325384, Table 35-12.
Usage example:
$ perf list $ perf stat -a -e power/energy-cores/,power/energy-pkg/ sleep 10
--- arch/x86/events/intel/rapl.c | 2 ++ 1 file changed, 2 insertions(+)
arch/x86/events/intel/rapl.c +2 -0
@@ -767,6 +767,8 @@ static const struct x86_cpu_id rapl_cpu_match[] __initconst = {X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_MOBILE , skl_rapl_init),X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_DESKTOP , skl_rapl_init),X86_RAPL_MODEL_MATCH (INTEL_FAM6_SKYLAKE_X , hsx_rapl_init), + + X86_RAPL_MODEL_MATCH(INTEL_FAM6_ATOM_GOLDMONT, hsw_rapl_init), {}, };