openkylin
/
linux
mirror of https://gitee.com/openkylin/linux.git
9
0
Fork 0
Code Issues Projects Releases Wiki Activity
linux/arch/x86/kernel/cpu/perf_event_p4.c

834 lines
24 KiB
C
Raw Normal View History

perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* Netburst Perfomance Events (P4, old Xeon)
*
* Copyright (C) 2010 Parallels, Inc., Cyrill Gorcunov <gorcunov@openvz.org>
* Copyright (C) 2010 Intel Corporation, Lin Ming <ming.m.lin@intel.com>
*
* For licencing details see kernel-base/COPYING
*/
#ifdef CONFIG_CPU_SUP_INTEL
#include <asm/perf_event_p4.h>
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
#define P4_CNTR_LIMIT 3
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* array indices: 0,1 - HT threads, used with HT enabled cpu
*/
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct p4_event_bind {
unsigned int opcode; /* Event code and ESCR selector */
unsigned int escr_msr[2]; /* ESCR MSR for this event */
unsigned char cntr[2][P4_CNTR_LIMIT]; /* counter index (offset), -1 on abscence */
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct p4_cache_event_bind {
unsigned int metric_pebs;
unsigned int metric_vert;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
#define P4_GEN_CACHE_EVENT_BIND(name) \
[P4_CACHE__##name] = { \
.metric_pebs = P4_PEBS__##name, \
.metric_vert = P4_VERT__##name, \
}
static struct p4_cache_event_bind p4_cache_event_bind_map[] = {
P4_GEN_CACHE_EVENT_BIND(1stl_cache_load_miss_retired),
P4_GEN_CACHE_EVENT_BIND(2ndl_cache_load_miss_retired),
P4_GEN_CACHE_EVENT_BIND(dtlb_load_miss_retired),
P4_GEN_CACHE_EVENT_BIND(dtlb_store_miss_retired),
};
/*
* Note that we don't use CCCR1 here, there is an
* exception for P4_BSQ_ALLOCATION but we just have
* no workaround
*
* consider this binding as resources which particular
* event may borrow, it doesn't contain EventMask,
* Tags and friends -- they are left to a caller
*/
static struct p4_event_bind p4_event_bind_map[] = {
[P4_EVENT_TC_DELIVER_MODE] = {
.opcode = P4_OPCODE(P4_EVENT_TC_DELIVER_MODE),
.escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_BPU_FETCH_REQUEST] = {
.opcode = P4_OPCODE(P4_EVENT_BPU_FETCH_REQUEST),
.escr_msr = { MSR_P4_BPU_ESCR0, MSR_P4_BPU_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_ITLB_REFERENCE] = {
.opcode = P4_OPCODE(P4_EVENT_ITLB_REFERENCE),
.escr_msr = { MSR_P4_ITLB_ESCR0, MSR_P4_ITLB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_MEMORY_CANCEL] = {
.opcode = P4_OPCODE(P4_EVENT_MEMORY_CANCEL),
.escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_MEMORY_COMPLETE] = {
.opcode = P4_OPCODE(P4_EVENT_MEMORY_COMPLETE),
.escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_LOAD_PORT_REPLAY] = {
.opcode = P4_OPCODE(P4_EVENT_LOAD_PORT_REPLAY),
.escr_msr = { MSR_P4_SAAT_ESCR0, MSR_P4_SAAT_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_STORE_PORT_REPLAY] = {
.opcode = P4_OPCODE(P4_EVENT_STORE_PORT_REPLAY),
.escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_MOB_LOAD_REPLAY] = {
.opcode = P4_OPCODE(P4_EVENT_MOB_LOAD_REPLAY),
.escr_msr = { MSR_P4_MOB_ESCR0, MSR_P4_MOB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_PAGE_WALK_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_PAGE_WALK_TYPE),
.escr_msr = { MSR_P4_PMH_ESCR0, MSR_P4_PMH_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_BSQ_CACHE_REFERENCE] = {
.opcode = P4_OPCODE(P4_EVENT_BSQ_CACHE_REFERENCE),
.escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_IOQ_ALLOCATION] = {
.opcode = P4_OPCODE(P4_EVENT_IOQ_ALLOCATION),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_IOQ_ACTIVE_ENTRIES] = { /* shared ESCR */
.opcode = P4_OPCODE(P4_EVENT_IOQ_ACTIVE_ENTRIES),
.escr_msr = { MSR_P4_FSB_ESCR1, MSR_P4_FSB_ESCR1 },
.cntr = { {2, -1, -1}, {3, -1, -1} },
},
[P4_EVENT_FSB_DATA_ACTIVITY] = {
.opcode = P4_OPCODE(P4_EVENT_FSB_DATA_ACTIVITY),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_BSQ_ALLOCATION] = { /* shared ESCR, broken CCCR1 */
.opcode = P4_OPCODE(P4_EVENT_BSQ_ALLOCATION),
.escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR0 },
.cntr = { {0, -1, -1}, {1, -1, -1} },
},
[P4_EVENT_BSQ_ACTIVE_ENTRIES] = { /* shared ESCR */
.opcode = P4_OPCODE(P4_EVENT_BSQ_ACTIVE_ENTRIES),
.escr_msr = { MSR_P4_BSU_ESCR1 , MSR_P4_BSU_ESCR1 },
.cntr = { {2, -1, -1}, {3, -1, -1} },
},
[P4_EVENT_SSE_INPUT_ASSIST] = {
.opcode = P4_OPCODE(P4_EVENT_SSE_INPUT_ASSIST),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_PACKED_SP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_PACKED_SP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_PACKED_DP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_PACKED_DP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_SCALAR_SP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_SCALAR_SP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_SCALAR_DP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_SCALAR_DP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_64BIT_MMX_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_64BIT_MMX_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_128BIT_MMX_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_128BIT_MMX_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_X87_FP_UOP] = {
.opcode = P4_OPCODE(P4_EVENT_X87_FP_UOP),
.escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_TC_MISC] = {
.opcode = P4_OPCODE(P4_EVENT_TC_MISC),
.escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_GLOBAL_POWER_EVENTS] = {
.opcode = P4_OPCODE(P4_EVENT_GLOBAL_POWER_EVENTS),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_TC_MS_XFER] = {
.opcode = P4_OPCODE(P4_EVENT_TC_MS_XFER),
.escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_UOP_QUEUE_WRITES] = {
.opcode = P4_OPCODE(P4_EVENT_UOP_QUEUE_WRITES),
.escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE),
.escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR0 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_RETIRED_BRANCH_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_RETIRED_BRANCH_TYPE),
.escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR1 },
.cntr = { {4, 5, -1}, {6, 7, -1} },
},
[P4_EVENT_RESOURCE_STALL] = {
.opcode = P4_OPCODE(P4_EVENT_RESOURCE_STALL),
.escr_msr = { MSR_P4_ALF_ESCR0, MSR_P4_ALF_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_WC_BUFFER] = {
.opcode = P4_OPCODE(P4_EVENT_WC_BUFFER),
.escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 },
.cntr = { {8, 9, -1}, {10, 11, -1} },
},
[P4_EVENT_B2B_CYCLES] = {
.opcode = P4_OPCODE(P4_EVENT_B2B_CYCLES),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_BNR] = {
.opcode = P4_OPCODE(P4_EVENT_BNR),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_SNOOP] = {
.opcode = P4_OPCODE(P4_EVENT_SNOOP),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_RESPONSE] = {
.opcode = P4_OPCODE(P4_EVENT_RESPONSE),
.escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
.cntr = { {0, -1, -1}, {2, -1, -1} },
},
[P4_EVENT_FRONT_END_EVENT] = {
.opcode = P4_OPCODE(P4_EVENT_FRONT_END_EVENT),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_EXECUTION_EVENT] = {
.opcode = P4_OPCODE(P4_EVENT_EXECUTION_EVENT),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_REPLAY_EVENT] = {
.opcode = P4_OPCODE(P4_EVENT_REPLAY_EVENT),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_INSTR_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_INSTR_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_UOPS_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_UOPS_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_UOP_TYPE] = {
.opcode = P4_OPCODE(P4_EVENT_UOP_TYPE),
.escr_msr = { MSR_P4_RAT_ESCR0, MSR_P4_RAT_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_BRANCH_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_BRANCH_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_MISPRED_BRANCH_RETIRED] = {
.opcode = P4_OPCODE(P4_EVENT_MISPRED_BRANCH_RETIRED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_X87_ASSIST] = {
.opcode = P4_OPCODE(P4_EVENT_X87_ASSIST),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_MACHINE_CLEAR] = {
.opcode = P4_OPCODE(P4_EVENT_MACHINE_CLEAR),
.escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
[P4_EVENT_INSTR_COMPLETED] = {
.opcode = P4_OPCODE(P4_EVENT_INSTR_COMPLETED),
.escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
.cntr = { {12, 13, 16}, {14, 15, 17} },
},
};
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
#define P4_GEN_CACHE_EVENT(event, bit, cache_event) \
p4_config_pack_escr(P4_ESCR_EVENT(event) | \
P4_ESCR_EMASK_BIT(event, bit)) | \
p4_config_pack_cccr(cache_event | \
P4_CCCR_ESEL(P4_OPCODE_ESEL(P4_OPCODE(event))))
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
perf, x86: Fix __initconst vs const All variables that have __initconst should also be const. Suggested-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-29 19:09:53 +08:00
static __initconst const u64 p4_hw_cache_event_ids
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] =
{
[ C(L1D ) ] = {
[ C(OP_READ) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
P4_CACHE__1stl_cache_load_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
},
[ C(LL ) ] = {
[ C(OP_READ) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
P4_CACHE__2ndl_cache_load_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
},
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
[ C(DTLB) ] = {
[ C(OP_READ) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
P4_CACHE__dtlb_load_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
[ C(OP_WRITE) ] = {
[ C(RESULT_ACCESS) ] = 0x0,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_REPLAY_EVENT, NBOGUS,
P4_CACHE__dtlb_store_miss_retired),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
},
[ C(ITLB) ] = {
[ C(OP_READ) ] = {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
[ C(RESULT_ACCESS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_ITLB_REFERENCE, HIT,
P4_CACHE__itlb_reference_hit),
[ C(RESULT_MISS) ] = P4_GEN_CACHE_EVENT(P4_EVENT_ITLB_REFERENCE, MISS,
P4_CACHE__itlb_reference_miss),
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
},
[ C(OP_WRITE) ] = {
[ C(RESULT_ACCESS) ] = -1,
[ C(RESULT_MISS) ] = -1,
},
[ C(OP_PREFETCH) ] = {
[ C(RESULT_ACCESS) ] = -1,
[ C(RESULT_MISS) ] = -1,
},
},
};
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
static u64 p4_general_events[PERF_COUNT_HW_MAX] = {
/* non-halted CPU clocks */
[PERF_COUNT_HW_CPU_CYCLES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_GLOBAL_POWER_EVENTS) |
P4_ESCR_EMASK_BIT(P4_EVENT_GLOBAL_POWER_EVENTS, RUNNING)),
/*
* retired instructions
* in a sake of simplicity we don't use the FSB tagging
*/
[PERF_COUNT_HW_INSTRUCTIONS] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_INSTR_RETIRED) |
P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, NBOGUSNTAG) |
P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, BOGUSNTAG)),
/* cache hits */
[PERF_COUNT_HW_CACHE_REFERENCES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_BSQ_CACHE_REFERENCE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITM) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITM)),
/* cache misses */
[PERF_COUNT_HW_CACHE_MISSES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_BSQ_CACHE_REFERENCE) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_MISS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_MISS) |
P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, WR_2ndL_MISS)),
/* branch instructions retired */
[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_RETIRED_BRANCH_TYPE) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, CONDITIONAL) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, CALL) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, RETURN) |
P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, INDIRECT)),
/* mispredicted branches retired */
[PERF_COUNT_HW_BRANCH_MISSES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_MISPRED_BRANCH_RETIRED) |
P4_ESCR_EMASK_BIT(P4_EVENT_MISPRED_BRANCH_RETIRED, NBOGUS)),
/* bus ready clocks (cpu is driving #DRDY_DRV\#DRDY_OWN): */
[PERF_COUNT_HW_BUS_CYCLES] =
p4_config_pack_escr(P4_ESCR_EVENT(P4_EVENT_FSB_DATA_ACTIVITY) |
P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_DRV) |
P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_OWN)) |
p4_config_pack_cccr(P4_CCCR_EDGE | P4_CCCR_COMPARE),
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
};
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
static struct p4_event_bind *p4_config_get_bind(u64 config)
{
unsigned int evnt = p4_config_unpack_event(config);
struct p4_event_bind *bind = NULL;
if (evnt < ARRAY_SIZE(p4_event_bind_map))
bind = &p4_event_bind_map[evnt];
return bind;
}
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
static u64 p4_pmu_event_map(int hw_event)
{
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct p4_event_bind *bind;
unsigned int esel;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
u64 config;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (hw_event > ARRAY_SIZE(p4_general_events)) {
printk_once(KERN_ERR "P4 PMU: Bad index: %i\n", hw_event);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
return 0;
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
config = p4_general_events[hw_event];
bind = p4_config_get_bind(config);
esel = P4_OPCODE_ESEL(bind->opcode);
config |= p4_config_pack_cccr(P4_CCCR_ESEL(esel));
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
return config;
}
perf, x86: Fix up the ANY flag stuff Stephane noticed that the ANY flag was in generic arch code, and Cyrill reported that it broke the P4 code. Solve this by merging x86_pmu::raw_event into x86_pmu::hw_config and provide intel_pmu and amd_pmu specific versions of this callback. The intel_pmu one deals with the ANY flag, the amd_pmu adds the few extra event bits AMD64 has. Reported-by: Stephane Eranian <eranian@google.com> Reported-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Robert Richter <robert.richter@amd.com> Acked-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269968113.5258.442.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 23:00:06 +08:00
static int p4_hw_config(struct perf_event *event)
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
{
int cpu = raw_smp_processor_id();
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
u32 escr, cccr;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* the reason we use cpu that early is that: if we get scheduled
* first time on the same cpu -- we will not need swap thread
* specific flags in config (and will save some cpu cycles)
*/
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
cccr = p4_default_cccr_conf(cpu);
perf, x86: Fix up the ANY flag stuff Stephane noticed that the ANY flag was in generic arch code, and Cyrill reported that it broke the P4 code. Solve this by merging x86_pmu::raw_event into x86_pmu::hw_config and provide intel_pmu and amd_pmu specific versions of this callback. The intel_pmu one deals with the ANY flag, the amd_pmu adds the few extra event bits AMD64 has. Reported-by: Stephane Eranian <eranian@google.com> Reported-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Robert Richter <robert.richter@amd.com> Acked-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269968113.5258.442.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 23:00:06 +08:00
escr = p4_default_escr_conf(cpu, event->attr.exclude_kernel,
event->attr.exclude_user);
event->hw.config = p4_config_pack_escr(escr) |
p4_config_pack_cccr(cccr);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
if (p4_ht_active() && p4_ht_thread(cpu))
perf, x86: Fix up the ANY flag stuff Stephane noticed that the ANY flag was in generic arch code, and Cyrill reported that it broke the P4 code. Solve this by merging x86_pmu::raw_event into x86_pmu::hw_config and provide intel_pmu and amd_pmu specific versions of this callback. The intel_pmu one deals with the ANY flag, the amd_pmu adds the few extra event bits AMD64 has. Reported-by: Stephane Eranian <eranian@google.com> Reported-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Robert Richter <robert.richter@amd.com> Acked-by: Cyrill Gorcunov <gorcunov@gmail.com> Acked-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269968113.5258.442.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 23:00:06 +08:00
event->hw.config = p4_set_ht_bit(event->hw.config);
x86, perf: P4 PMU -- configure predefined events If an event is not RAW we should not exit p4_hw_config early but call x86_setup_perfctr as well. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Lin Ming <ming.m.lin@intel.com> Cc: Robert Richter <robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-05-08 19:39:52 +08:00
if (event->attr.type == PERF_TYPE_RAW) {
/*
* We don't control raw events so it's up to the caller
* to pass sane values (and we don't count the thread number
* on HT machine but allow HT-compatible specifics to be
* passed on)
*
* XXX: HT wide things should check perf_paranoid_cpu() &&
* CAP_SYS_ADMIN
*/
event->hw.config |= event->attr.config &
(p4_config_pack_escr(P4_ESCR_MASK_HT) |
p4_config_pack_cccr(P4_CCCR_MASK_HT));
}
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
perf, x86: Call x86_setup_perfctr() from .hw_config() The perfctr setup calls are in the corresponding .hw_config() functions now. This makes it possible to introduce config functions for other pmu events that are not perfctr specific. Also, all of a sudden the code looks much nicer. Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1271190201-25705-4-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-04-14 04:23:12 +08:00
return x86_setup_perfctr(event);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
static inline void p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc)
{
unsigned long dummy;
rdmsrl(hwc->config_base + hwc->idx, dummy);
if (dummy & P4_CCCR_OVF) {
(void)checking_wrmsrl(hwc->config_base + hwc->idx,
((u64)dummy) & ~P4_CCCR_OVF);
}
}
static inline void p4_pmu_disable_event(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
/*
* If event gets disabled while counter is in overflowed
* state we need to clear P4_CCCR_OVF, otherwise interrupt get
* asserted again and again
*/
(void)checking_wrmsrl(hwc->config_base + hwc->idx,
(u64)(p4_config_unpack_cccr(hwc->config)) &
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
~P4_CCCR_ENABLE & ~P4_CCCR_OVF & ~P4_CCCR_RESERVED);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
static void p4_pmu_disable_all(void)
{
struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
int idx;
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
for (idx = 0; idx < x86_pmu.num_counters; idx++) {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
struct perf_event *event = cpuc->events[idx];
if (!test_bit(idx, cpuc->active_mask))
continue;
p4_pmu_disable_event(event);
}
}
static void p4_pmu_enable_event(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
int thread = p4_ht_config_thread(hwc->config);
u64 escr_conf = p4_config_unpack_escr(p4_clear_ht_bit(hwc->config));
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
unsigned int idx = p4_config_unpack_event(hwc->config);
unsigned int idx_cache = p4_config_unpack_cache_event(hwc->config);
struct p4_event_bind *bind;
struct p4_cache_event_bind *bind_cache;
u64 escr_addr, cccr;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
bind = &p4_event_bind_map[idx];
escr_addr = (u64)bind->escr_msr[thread];
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
/*
* - we dont support cascaded counters yet
* - and counter 1 is broken (erratum)
*/
WARN_ON_ONCE(p4_is_event_cascaded(hwc->config));
WARN_ON_ONCE(hwc->idx == 1);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
/* we need a real Event value */
escr_conf &= ~P4_ESCR_EVENT_MASK;
escr_conf |= P4_ESCR_EVENT(P4_OPCODE_EVNT(bind->opcode));
cccr = p4_config_unpack_cccr(hwc->config);
/*
* it could be Cache event so that we need to
* set metrics into additional MSRs
*/
BUILD_BUG_ON(P4_CACHE__MAX > P4_CCCR_CACHE_OPS_MASK);
if (idx_cache > P4_CACHE__NONE &&
idx_cache < ARRAY_SIZE(p4_cache_event_bind_map)) {
bind_cache = &p4_cache_event_bind_map[idx_cache];
(void)checking_wrmsrl(MSR_IA32_PEBS_ENABLE, (u64)bind_cache->metric_pebs);
(void)checking_wrmsrl(MSR_P4_PEBS_MATRIX_VERT, (u64)bind_cache->metric_vert);
}
(void)checking_wrmsrl(escr_addr, escr_conf);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
(void)checking_wrmsrl(hwc->config_base + hwc->idx,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
(cccr & ~P4_CCCR_RESERVED) | P4_CCCR_ENABLE);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
perf, x86: Add Nehelem PMU programming errata workaround Implement the workaround for Intel Errata AAK100 and AAP53. Also, remove the Core-i7 name for Nehalem events since there are also Westmere based i7 chips. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> LKML-Reference: <1269608924.12097.147.camel@laptop> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-26 21:08:44 +08:00
static void p4_pmu_enable_all(int added)
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
{
struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
int idx;
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
for (idx = 0; idx < x86_pmu.num_counters; idx++) {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
struct perf_event *event = cpuc->events[idx];
if (!test_bit(idx, cpuc->active_mask))
continue;
p4_pmu_enable_event(event);
}
}
static int p4_pmu_handle_irq(struct pt_regs *regs)
{
struct perf_sample_data data;
struct cpu_hw_events *cpuc;
struct perf_event *event;
struct hw_perf_event *hwc;
int idx, handled = 0;
u64 val;
data.addr = 0;
data.raw = NULL;
cpuc = &__get_cpu_var(cpu_hw_events);
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
for (idx = 0; idx < x86_pmu.num_counters; idx++) {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (!test_bit(idx, cpuc->active_mask))
continue;
event = cpuc->events[idx];
hwc = &event->hw;
WARN_ON_ONCE(hwc->idx != idx);
/*
* FIXME: Redundant call, actually not needed
* but just to check if we're screwed
*/
p4_pmu_clear_cccr_ovf(hwc);
val = x86_perf_event_update(event);
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
if (val & (1ULL << (x86_pmu.cntval_bits - 1)))
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
continue;
/*
* event overflow
*/
handled = 1;
data.period = event->hw.last_period;
if (!x86_perf_event_set_period(event))
continue;
if (perf_event_overflow(event, 1, &data, regs))
p4_pmu_disable_event(event);
}
if (handled) {
/* p4 quirk: unmask it again */
apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED);
inc_irq_stat(apic_perf_irqs);
}
return handled;
}
/*
* swap thread specific fields according to a thread
* we are going to run on
*/
static void p4_pmu_swap_config_ts(struct hw_perf_event *hwc, int cpu)
{
u32 escr, cccr;
/*
* we either lucky and continue on same cpu or no HT support
*/
if (!p4_should_swap_ts(hwc->config, cpu))
return;
/*
* the event is migrated from an another logical
* cpu, so we need to swap thread specific flags
*/
escr = p4_config_unpack_escr(hwc->config);
cccr = p4_config_unpack_cccr(hwc->config);
if (p4_ht_thread(cpu)) {
cccr &= ~P4_CCCR_OVF_PMI_T0;
cccr |= P4_CCCR_OVF_PMI_T1;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T0_OS) {
escr &= ~P4_ESCR_T0_OS;
escr |= P4_ESCR_T1_OS;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T0_USR) {
escr &= ~P4_ESCR_T0_USR;
escr |= P4_ESCR_T1_USR;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
hwc->config = p4_config_pack_escr(escr);
hwc->config |= p4_config_pack_cccr(cccr);
hwc->config |= P4_CONFIG_HT;
} else {
cccr &= ~P4_CCCR_OVF_PMI_T1;
cccr |= P4_CCCR_OVF_PMI_T0;
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T1_OS) {
escr &= ~P4_ESCR_T1_OS;
escr |= P4_ESCR_T0_OS;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (escr & P4_ESCR_T1_USR) {
escr &= ~P4_ESCR_T1_USR;
escr |= P4_ESCR_T0_USR;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
}
hwc->config = p4_config_pack_escr(escr);
hwc->config |= p4_config_pack_cccr(cccr);
hwc->config &= ~P4_CONFIG_HT;
}
}
/* ESCRs are not sequential in memory so we need a map */
x86, perf: Fix few cosmetic dabs for P4 pmu (comments and constantify) - A few ESCR have escaped fixing at previous attempt. - p4_escr_map is read only, make it const. Nothing serious. Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Lin Ming <ming.m.lin@intel.com> LKML-Reference: <20100318211256.GH5062@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-19 05:12:56 +08:00
static const unsigned int p4_escr_map[ARCH_P4_TOTAL_ESCR] = {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
MSR_P4_ALF_ESCR0, /* 0 */
MSR_P4_ALF_ESCR1, /* 1 */
MSR_P4_BPU_ESCR0, /* 2 */
MSR_P4_BPU_ESCR1, /* 3 */
MSR_P4_BSU_ESCR0, /* 4 */
MSR_P4_BSU_ESCR1, /* 5 */
MSR_P4_CRU_ESCR0, /* 6 */
MSR_P4_CRU_ESCR1, /* 7 */
MSR_P4_CRU_ESCR2, /* 8 */
MSR_P4_CRU_ESCR3, /* 9 */
MSR_P4_CRU_ESCR4, /* 10 */
MSR_P4_CRU_ESCR5, /* 11 */
MSR_P4_DAC_ESCR0, /* 12 */
MSR_P4_DAC_ESCR1, /* 13 */
MSR_P4_FIRM_ESCR0, /* 14 */
MSR_P4_FIRM_ESCR1, /* 15 */
MSR_P4_FLAME_ESCR0, /* 16 */
MSR_P4_FLAME_ESCR1, /* 17 */
MSR_P4_FSB_ESCR0, /* 18 */
MSR_P4_FSB_ESCR1, /* 19 */
MSR_P4_IQ_ESCR0, /* 20 */
MSR_P4_IQ_ESCR1, /* 21 */
MSR_P4_IS_ESCR0, /* 22 */
MSR_P4_IS_ESCR1, /* 23 */
MSR_P4_ITLB_ESCR0, /* 24 */
MSR_P4_ITLB_ESCR1, /* 25 */
MSR_P4_IX_ESCR0, /* 26 */
MSR_P4_IX_ESCR1, /* 27 */
MSR_P4_MOB_ESCR0, /* 28 */
MSR_P4_MOB_ESCR1, /* 29 */
MSR_P4_MS_ESCR0, /* 30 */
MSR_P4_MS_ESCR1, /* 31 */
MSR_P4_PMH_ESCR0, /* 32 */
MSR_P4_PMH_ESCR1, /* 33 */
MSR_P4_RAT_ESCR0, /* 34 */
MSR_P4_RAT_ESCR1, /* 35 */
MSR_P4_SAAT_ESCR0, /* 36 */
MSR_P4_SAAT_ESCR1, /* 37 */
MSR_P4_SSU_ESCR0, /* 38 */
MSR_P4_SSU_ESCR1, /* 39 */
MSR_P4_TBPU_ESCR0, /* 40 */
MSR_P4_TBPU_ESCR1, /* 41 */
MSR_P4_TC_ESCR0, /* 42 */
MSR_P4_TC_ESCR1, /* 43 */
MSR_P4_U2L_ESCR0, /* 44 */
MSR_P4_U2L_ESCR1, /* 45 */
};
static int p4_get_escr_idx(unsigned int addr)
{
unsigned int i;
for (i = 0; i < ARRAY_SIZE(p4_escr_map); i++) {
if (addr == p4_escr_map[i])
return i;
}
return -1;
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
static int p4_next_cntr(int thread, unsigned long *used_mask,
struct p4_event_bind *bind)
{
int i = 0, j;
for (i = 0; i < P4_CNTR_LIMIT; i++) {
j = bind->cntr[thread][i++];
if (j == -1 || !test_bit(j, used_mask))
return j;
}
return -1;
}
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
static int p4_pmu_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
{
unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
int cpu = raw_smp_processor_id();
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
struct hw_perf_event *hwc;
struct p4_event_bind *bind;
unsigned int i, thread, num;
int cntr_idx, escr_idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
bitmap_zero(used_mask, X86_PMC_IDX_MAX);
bitmap_zero(escr_mask, ARCH_P4_TOTAL_ESCR);
for (i = 0, num = n; i < n; i++, num--) {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
hwc = &cpuc->event_list[i]->hw;
thread = p4_ht_thread(cpu);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
bind = p4_config_get_bind(hwc->config);
escr_idx = p4_get_escr_idx(bind->escr_msr[thread]);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (hwc->idx != -1 && !p4_should_swap_ts(hwc->config, cpu)) {
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
cntr_idx = hwc->idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
if (assign)
assign[i] = hwc->idx;
goto reserve;
}
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
cntr_idx = p4_next_cntr(thread, used_mask, bind);
if (cntr_idx == -1 || test_bit(escr_idx, escr_mask))
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
goto done;
p4_pmu_swap_config_ts(hwc, cpu);
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
if (assign)
assign[i] = cntr_idx;
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
reserve:
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
set_bit(cntr_idx, used_mask);
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
set_bit(escr_idx, escr_mask);
}
done:
return num ? -ENOSPC : 0;
}
perf, x86: Fix __initconst vs const All variables that have __initconst should also be const. Suggested-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-29 19:09:53 +08:00
static __initconst const struct x86_pmu p4_pmu = {
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.name = "Netburst P4/Xeon",
.handle_irq = p4_pmu_handle_irq,
.disable_all = p4_pmu_disable_all,
.enable_all = p4_pmu_enable_all,
.enable = p4_pmu_enable_event,
.disable = p4_pmu_disable_event,
.eventsel = MSR_P4_BPU_CCCR0,
.perfctr = MSR_P4_BPU_PERFCTR0,
.event_map = p4_pmu_event_map,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
.max_events = ARRAY_SIZE(p4_general_events),
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.get_event_constraints = x86_get_event_constraints,
/*
* IF HT disabled we may need to use all
* ARCH_P4_MAX_CCCR counters simulaneously
* though leave it restricted at moment assuming
* HT is on
*/
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
.num_counters = ARCH_P4_MAX_CCCR,
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.apic = 1,
perf, x86: Undo some some *_counter* -> *_event* renames The big rename: cdd6c48 perf: Do the big rename: Performance Counters -> Performance Events accidentally renamed some members of stucts that were named after registers in the spec. To avoid confusion this patch reverts some changes. The related specs are MSR descriptions in AMD's BKDGs and the ARCHITECTURAL PERFORMANCE MONITORING section in the Intel 64 and IA-32 Architectures Software Developer's Manuals. This patch does: $ sed -i -e 's:num_events:num_counters:g' \ arch/x86/include/asm/perf_event.h \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c \ arch/x86/oprofile/op_model_ppro.c $ sed -i -e 's:event_bits:cntval_bits:g' -e 's:event_mask:cntval_mask:g' \ arch/x86/kernel/cpu/perf_event_amd.c \ arch/x86/kernel/cpu/perf_event.c \ arch/x86/kernel/cpu/perf_event_intel.c \ arch/x86/kernel/cpu/perf_event_p6.c \ arch/x86/kernel/cpu/perf_event_p4.c Signed-off-by: Robert Richter <robert.richter@amd.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1269880612-25800-2-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-30 00:36:50 +08:00
.cntval_bits = 40,
.cntval_mask = (1ULL << 40) - 1,
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
.max_period = (1ULL << 39) - 1,
.hw_config = p4_hw_config,
.schedule_events = p4_pmu_schedule_events,
};
static __init int p4_pmu_init(void)
{
unsigned int low, high;
/* If we get stripped -- indexig fails */
BUILD_BUG_ON(ARCH_P4_MAX_CCCR > X86_PMC_MAX_GENERIC);
rdmsr(MSR_IA32_MISC_ENABLE, low, high);
if (!(low & (1 << 7))) {
pr_cont("unsupported Netburst CPU model %d ",
boot_cpu_data.x86_model);
return -ENODEV;
}
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
memcpy(hw_cache_event_ids, p4_hw_cache_event_ids,
x86, perf: Add raw events support for the P4 PMU The adding of raw event support lead to complete code refactoring. I hope is became more readable then it was. The list of changes: 1) The 64bit config field is enough to hold all information we need to track event details. To achieve it we used *own* enum for events selection in ESCR register and map this key into proper value at moment of event enabling. For the same reason we use 12LSB bits in CCCR register -- to track which exactly cache trace event was requested. And we cear this bits at real 'write' moment. 2) There is no per-cpu area reserved for P4 PMU anymore. We don't need it. All is held by config. 3) Now we may use any available counter, ie we try to grab any possible counter. v2: - Lin Ming reported the lack of ESCR selector in CCCR for cache events v3: - Don't loose cache event codes at config unpacking procedure, we may need it one day so no obscure hack behind our back, better to clear reserved bits explicitly when needed (thanks Ming for pointing out) - Lin Ming fixed misplaced opcodes in cache events Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Tested-by: Lin Ming <ming.m.lin@intel.com> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1269403766.3409.6.camel@minggr.sh.intel.com> [ v4: did a few whitespace fixlets ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-24 12:09:26 +08:00
sizeof(hw_cache_event_ids));
perf, x86: Add cache events for the Pentium-4 PMU Move the HT bit setting code from p4_pmu_event_map to p4_hw_config. So the cache events can get HT bit set correctly. Tested on my P4 desktop, below 6 cache events work: L1-dcache-load-misses LLC-load-misses dTLB-load-misses dTLB-store-misses iTLB-loads iTLB-load-misses Signed-off-by: Lin Ming <ming.m.lin@intel.com> Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <1268908392.13901.128.camel@minggr.sh.intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-18 18:33:12 +08:00
perf, x86: Implement initial P4 PMU driver The netburst PMU is way different from the "architectural perfomance monitoring" specification that current CPUs use. P4 uses a tuple of ESCR+CCCR+COUNTER MSR registers to handle perfomance monitoring events. A few implementational details: 1) We need a separate x86_pmu::hw_config helper in struct x86_pmu since register bit-fields are quite different from P6, Core and later cpu series. 2) For the same reason is a x86_pmu::schedule_events helper introduced. 3) hw_perf_event::config consists of packed ESCR+CCCR values. It's allowed since in reality both registers only use a half of their size. Of course before making a real write into a particular MSR we need to unpack the value and extend it to a proper size. 4) The tuple of packed ESCR+CCCR in hw_perf_event::config doesn't describe the memory address of ESCR MSR register so that we need to keep a mapping between these tuples used and available ESCR (various P4 events may use same ESCRs but not simultaneously), for this sake every active event has a per-cpu map of hw_perf_event::idx <--> ESCR addresses. 5) Since hw_perf_event::idx is an offset to counter/control register we need to lift X86_PMC_MAX_GENERIC up, otherwise kernel strips it down to 8 registers and event armed may never be turned off (ie the bit in active_mask is set but the loop never reaches this index to check), thanks to Peter Zijlstra Restrictions: - No cascaded counters support (do we ever need them?) - No dependent events support (so PERF_COUNT_HW_INSTRUCTIONS doesn't work for now) - There are events with same counters which can't work simultaneously (need to use intersected ones due to broken counter 1) - No PERF_COUNT_HW_CACHE_ events yet Todo: - Implement dependent events - Need proper hashing for event opcodes (no linear search, good for debugging stage but not in real loads) - Some events counted during a clock cycle -- need to set threshold for them and count every clock cycle just to get summary statistics (ie to behave the same way as other PMUs do) - Need to swicth to use event_constraints - To support RAW events we need to encode a global list of P4 events into p4_templates - Cache events need to be added Event support status matrix: Event status ----------------------------- cycles works cache-references works cache-misses works branch-misses works bus-cycles partially (does not work on 64bit cpu with HT enabled) instruction doesnt work (needs dependent event [mop tagging]) branches doesnt work Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by: Lin Ming <ming.m.lin@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: Robert Richter <robert.richter@amd.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <20100311165439.GB5129@lenovo> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 00:54:39 +08:00
pr_cont("Netburst events, ");
x86_pmu = p4_pmu;
return 0;
}
#endif /* CONFIG_CPU_SUP_INTEL */