alistair23-linux/arch/powerpc/perf/power6-pmu.c

553 lines
16 KiB
C
Raw Normal View History

/*
* Performance counter support for POWER6 processors.
*
* Copyright 2008-2009 Paul Mackerras, IBM Corporation.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/kernel.h>
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 04:02:48 -06:00
#include <linux/perf_event.h>
#include <linux/string.h>
#include <asm/reg.h>
#include <asm/cputable.h>
/*
* Bits in event code for POWER6
*/
#define PM_PMC_SH 20 /* PMC number (1-based) for direct events */
#define PM_PMC_MSK 0x7
#define PM_PMC_MSKS (PM_PMC_MSK << PM_PMC_SH)
#define PM_UNIT_SH 16 /* Unit event comes (TTMxSEL encoding) */
#define PM_UNIT_MSK 0xf
#define PM_UNIT_MSKS (PM_UNIT_MSK << PM_UNIT_SH)
#define PM_LLAV 0x8000 /* Load lookahead match value */
#define PM_LLA 0x4000 /* Load lookahead match enable */
#define PM_BYTE_SH 12 /* Byte of event bus to use */
#define PM_BYTE_MSK 3
#define PM_SUBUNIT_SH 8 /* Subunit event comes from (NEST_SEL enc.) */
#define PM_SUBUNIT_MSK 7
#define PM_SUBUNIT_MSKS (PM_SUBUNIT_MSK << PM_SUBUNIT_SH)
#define PM_PMCSEL_MSK 0xff /* PMCxSEL value */
#define PM_BUSEVENT_MSK 0xf3700
/*
* Bits in MMCR1 for POWER6
*/
#define MMCR1_TTM0SEL_SH 60
#define MMCR1_TTMSEL_SH(n) (MMCR1_TTM0SEL_SH - (n) * 4)
#define MMCR1_TTMSEL_MSK 0xf
#define MMCR1_TTMSEL(m, n) (((m) >> MMCR1_TTMSEL_SH(n)) & MMCR1_TTMSEL_MSK)
#define MMCR1_NESTSEL_SH 45
#define MMCR1_NESTSEL_MSK 0x7
#define MMCR1_NESTSEL(m) (((m) >> MMCR1_NESTSEL_SH) & MMCR1_NESTSEL_MSK)
#define MMCR1_PMC1_LLA (1ul << 44)
#define MMCR1_PMC1_LLA_VALUE (1ul << 39)
#define MMCR1_PMC1_ADDR_SEL (1ul << 35)
#define MMCR1_PMC1SEL_SH 24
#define MMCR1_PMCSEL_SH(n) (MMCR1_PMC1SEL_SH - (n) * 8)
#define MMCR1_PMCSEL_MSK 0xff
/*
* Map of which direct events on which PMCs are marked instruction events.
* Indexed by PMCSEL value >> 1.
* Bottom 4 bits are a map of which PMCs are interesting,
* top 4 bits say what sort of event:
* 0 = direct marked event,
* 1 = byte decode event,
* 4 = add/and event (PMC1 -> bits 0 & 4),
* 5 = add/and event (PMC1 -> bits 1 & 5),
* 6 = add/and event (PMC1 -> bits 2 & 6),
* 7 = add/and event (PMC1 -> bits 3 & 7).
*/
static unsigned char direct_event_is_marked[0x60 >> 1] = {
0, /* 00 */
0, /* 02 */
0, /* 04 */
0x07, /* 06 PM_MRK_ST_CMPL, PM_MRK_ST_GPS, PM_MRK_ST_CMPL_INT */
0x04, /* 08 PM_MRK_DFU_FIN */
0x06, /* 0a PM_MRK_IFU_FIN, PM_MRK_INST_FIN */
0, /* 0c */
0, /* 0e */
0x02, /* 10 PM_MRK_INST_DISP */
0x08, /* 12 PM_MRK_LSU_DERAT_MISS */
0, /* 14 */
0, /* 16 */
0x0c, /* 18 PM_THRESH_TIMEO, PM_MRK_INST_FIN */
0x0f, /* 1a PM_MRK_INST_DISP, PM_MRK_{FXU,FPU,LSU}_FIN */
0x01, /* 1c PM_MRK_INST_ISSUED */
0, /* 1e */
0, /* 20 */
0, /* 22 */
0, /* 24 */
0, /* 26 */
0x15, /* 28 PM_MRK_DATA_FROM_L2MISS, PM_MRK_DATA_FROM_L3MISS */
0, /* 2a */
0, /* 2c */
0, /* 2e */
0x4f, /* 30 */
0x7f, /* 32 */
0x4f, /* 34 */
0x5f, /* 36 */
0x6f, /* 38 */
0x4f, /* 3a */
0, /* 3c */
0x08, /* 3e PM_MRK_INST_TIMEO */
0x1f, /* 40 */
0x1f, /* 42 */
0x1f, /* 44 */
0x1f, /* 46 */
0x1f, /* 48 */
0x1f, /* 4a */
0x1f, /* 4c */
0x1f, /* 4e */
0, /* 50 */
0x05, /* 52 PM_MRK_BR_TAKEN, PM_MRK_BR_MPRED */
0x1c, /* 54 PM_MRK_PTEG_FROM_L3MISS, PM_MRK_PTEG_FROM_L2MISS */
0x02, /* 56 PM_MRK_LD_MISS_L1 */
0, /* 58 */
0, /* 5a */
0, /* 5c */
0, /* 5e */
};
/*
* Masks showing for each unit which bits are marked events.
* These masks are in LE order, i.e. 0x00000001 is byte 0, bit 0.
*/
static u32 marked_bus_events[16] = {
0x01000000, /* direct events set 1: byte 3 bit 0 */
0x00010000, /* direct events set 2: byte 2 bit 0 */
0, 0, 0, 0, /* IDU, IFU, nest: nothing */
0x00000088, /* VMX set 1: byte 0 bits 3, 7 */
0x000000c0, /* VMX set 2: byte 0 bits 4-7 */
0x04010000, /* LSU set 1: byte 2 bit 0, byte 3 bit 2 */
0xff010000u, /* LSU set 2: byte 2 bit 0, all of byte 3 */
0, /* LSU set 3 */
0x00000010, /* VMX set 3: byte 0 bit 4 */
0, /* BFP set 1 */
0x00000022, /* BFP set 2: byte 0 bits 1, 5 */
0, 0
};
/*
* Returns 1 if event counts things relating to marked instructions
* and thus needs the MMCRA_SAMPLE_ENABLE bit set, or 0 if not.
*/
static int power6_marked_instr_event(u64 event)
{
int pmc, psel, ptype;
int bit, byte, unit;
u32 mask;
pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
psel = (event & PM_PMCSEL_MSK) >> 1; /* drop edge/level bit */
if (pmc >= 5)
return 0;
bit = -1;
if (psel < sizeof(direct_event_is_marked)) {
ptype = direct_event_is_marked[psel];
if (pmc == 0 || !(ptype & (1 << (pmc - 1))))
return 0;
ptype >>= 4;
if (ptype == 0)
return 1;
if (ptype == 1)
bit = 0;
else
bit = ptype ^ (pmc - 1);
} else if ((psel & 0x48) == 0x40)
bit = psel & 7;
if (!(event & PM_BUSEVENT_MSK) || bit == -1)
return 0;
byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK;
unit = (event >> PM_UNIT_SH) & PM_UNIT_MSK;
mask = marked_bus_events[unit];
return (mask >> (byte * 8 + bit)) & 1;
}
/*
* Assign PMC numbers and compute MMCR1 value for a set of events
*/
static int p6_compute_mmcr(u64 event[], int n_ev,
unsigned int hwc[], unsigned long mmcr[], struct perf_event *pevents[])
{
unsigned long mmcr1 = 0;
unsigned long mmcra = MMCRA_SDAR_DCACHE_MISS | MMCRA_SDAR_ERAT_MISS;
int i;
unsigned int pmc, ev, b, u, s, psel;
unsigned int ttmset = 0;
unsigned int pmc_inuse = 0;
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (n_ev > 6)
return -1;
for (i = 0; i < n_ev; ++i) {
pmc = (event[i] >> PM_PMC_SH) & PM_PMC_MSK;
if (pmc) {
if (pmc_inuse & (1 << (pmc - 1)))
return -1; /* collision! */
pmc_inuse |= 1 << (pmc - 1);
}
}
for (i = 0; i < n_ev; ++i) {
ev = event[i];
pmc = (ev >> PM_PMC_SH) & PM_PMC_MSK;
if (pmc) {
--pmc;
} else {
/* can go on any PMC; find a free one */
for (pmc = 0; pmc < 4; ++pmc)
if (!(pmc_inuse & (1 << pmc)))
break;
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (pmc >= 4)
return -1;
pmc_inuse |= 1 << pmc;
}
hwc[i] = pmc;
psel = ev & PM_PMCSEL_MSK;
if (ev & PM_BUSEVENT_MSK) {
/* this event uses the event bus */
b = (ev >> PM_BYTE_SH) & PM_BYTE_MSK;
u = (ev >> PM_UNIT_SH) & PM_UNIT_MSK;
/* check for conflict on this byte of event bus */
if ((ttmset & (1 << b)) && MMCR1_TTMSEL(mmcr1, b) != u)
return -1;
mmcr1 |= (unsigned long)u << MMCR1_TTMSEL_SH(b);
ttmset |= 1 << b;
if (u == 5) {
/* Nest events have a further mux */
s = (ev >> PM_SUBUNIT_SH) & PM_SUBUNIT_MSK;
if ((ttmset & 0x10) &&
MMCR1_NESTSEL(mmcr1) != s)
return -1;
ttmset |= 0x10;
mmcr1 |= (unsigned long)s << MMCR1_NESTSEL_SH;
}
if (0x30 <= psel && psel <= 0x3d) {
/* these need the PMCx_ADDR_SEL bits */
if (b >= 2)
mmcr1 |= MMCR1_PMC1_ADDR_SEL >> pmc;
}
/* bus select values are different for PMC3/4 */
if (pmc >= 2 && (psel & 0x90) == 0x80)
psel ^= 0x20;
}
if (ev & PM_LLA) {
mmcr1 |= MMCR1_PMC1_LLA >> pmc;
if (ev & PM_LLAV)
mmcr1 |= MMCR1_PMC1_LLA_VALUE >> pmc;
}
if (power6_marked_instr_event(event[i]))
mmcra |= MMCRA_SAMPLE_ENABLE;
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (pmc < 4)
mmcr1 |= (unsigned long)psel << MMCR1_PMCSEL_SH(pmc);
}
mmcr[0] = 0;
if (pmc_inuse & 1)
mmcr[0] = MMCR0_PMC1CE;
if (pmc_inuse & 0xe)
mmcr[0] |= MMCR0_PMCjCE;
mmcr[1] = mmcr1;
mmcr[2] = mmcra;
return 0;
}
/*
* Layout of constraint bits:
*
* 0-1 add field: number of uses of PMC1 (max 1)
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
* 2-3, 4-5, 6-7, 8-9, 10-11: ditto for PMC2, 3, 4, 5, 6
* 12-15 add field: number of uses of PMC1-4 (max 4)
* 16-19 select field: unit on byte 0 of event bus
* 20-23, 24-27, 28-31 ditto for bytes 1, 2, 3
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
* 32-34 select field: nest (subunit) event selector
*/
static int p6_get_constraint(u64 event, unsigned long *maskp,
unsigned long *valp)
{
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
int pmc, byte, sh, subunit;
unsigned long mask = 0, value = 0;
pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
if (pmc) {
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (pmc > 4 && !(event == 0x500009 || event == 0x600005))
return -1;
sh = (pmc - 1) * 2;
mask |= 2 << sh;
value |= 1 << sh;
}
if (event & PM_BUSEVENT_MSK) {
byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK;
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
sh = byte * 4 + (16 - PM_UNIT_SH);
mask |= PM_UNIT_MSKS << sh;
value |= (unsigned long)(event & PM_UNIT_MSKS) << sh;
if ((event & PM_UNIT_MSKS) == (5 << PM_UNIT_SH)) {
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
subunit = (event >> PM_SUBUNIT_SH) & PM_SUBUNIT_MSK;
mask |= (unsigned long)PM_SUBUNIT_MSK << 32;
value |= (unsigned long)subunit << 32;
}
}
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (pmc <= 4) {
mask |= 0x8000; /* add field for count of PMC1-4 uses */
value |= 0x1000;
}
*maskp = mask;
*valp = value;
return 0;
}
static int p6_limited_pmc_event(u64 event)
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
{
int pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
return pmc == 5 || pmc == 6;
}
#define MAX_ALT 4 /* at most 4 alternatives for any event */
static const unsigned int event_alternatives[][MAX_ALT] = {
{ 0x0130e8, 0x2000f6, 0x3000fc }, /* PM_PTEG_RELOAD_VALID */
{ 0x080080, 0x10000d, 0x30000c, 0x4000f0 }, /* PM_LD_MISS_L1 */
{ 0x080088, 0x200054, 0x3000f0 }, /* PM_ST_MISS_L1 */
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
{ 0x10000a, 0x2000f4, 0x600005 }, /* PM_RUN_CYC */
{ 0x10000b, 0x2000f5 }, /* PM_RUN_COUNT */
{ 0x10000e, 0x400010 }, /* PM_PURR */
{ 0x100010, 0x4000f8 }, /* PM_FLUSH */
{ 0x10001a, 0x200010 }, /* PM_MRK_INST_DISP */
{ 0x100026, 0x3000f8 }, /* PM_TB_BIT_TRANS */
{ 0x100054, 0x2000f0 }, /* PM_ST_FIN */
{ 0x100056, 0x2000fc }, /* PM_L1_ICACHE_MISS */
{ 0x1000f0, 0x40000a }, /* PM_INST_IMC_MATCH_CMPL */
{ 0x1000f8, 0x200008 }, /* PM_GCT_EMPTY_CYC */
{ 0x1000fc, 0x400006 }, /* PM_LSU_DERAT_MISS_CYC */
{ 0x20000e, 0x400007 }, /* PM_LSU_DERAT_MISS */
{ 0x200012, 0x300012 }, /* PM_INST_DISP */
{ 0x2000f2, 0x3000f2 }, /* PM_INST_DISP */
{ 0x2000f8, 0x300010 }, /* PM_EXT_INT */
{ 0x2000fe, 0x300056 }, /* PM_DATA_FROM_L2MISS */
{ 0x2d0030, 0x30001a }, /* PM_MRK_FPU_FIN */
{ 0x30000a, 0x400018 }, /* PM_MRK_INST_FIN */
{ 0x3000f6, 0x40000e }, /* PM_L1_DCACHE_RELOAD_VALID */
{ 0x3000fe, 0x400056 }, /* PM_DATA_FROM_L3MISS */
};
/*
* This could be made more efficient with a binary search on
* a presorted list, if necessary
*/
static int find_alternatives_list(u64 event)
{
int i, j;
unsigned int alt;
for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) {
if (event < event_alternatives[i][0])
return -1;
for (j = 0; j < MAX_ALT; ++j) {
alt = event_alternatives[i][j];
if (!alt || event < alt)
break;
if (event == alt)
return i;
}
}
return -1;
}
static int p6_get_alternatives(u64 event, unsigned int flags, u64 alt[])
{
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
int i, j, nlim;
unsigned int psel, pmc;
unsigned int nalt = 1;
u64 aevent;
alt[0] = event;
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
nlim = p6_limited_pmc_event(event);
/* check the alternatives table */
i = find_alternatives_list(event);
if (i >= 0) {
/* copy out alternatives from list */
for (j = 0; j < MAX_ALT; ++j) {
aevent = event_alternatives[i][j];
if (!aevent)
break;
if (aevent != event)
alt[nalt++] = aevent;
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
nlim += p6_limited_pmc_event(aevent);
}
} else {
/* Check for alternative ways of computing sum events */
/* PMCSEL 0x32 counter N == PMCSEL 0x34 counter 5-N */
psel = event & (PM_PMCSEL_MSK & ~1); /* ignore edge bit */
pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
if (pmc && (psel == 0x32 || psel == 0x34))
alt[nalt++] = ((event ^ 0x6) & ~PM_PMC_MSKS) |
((5 - pmc) << PM_PMC_SH);
/* PMCSEL 0x38 counter N == PMCSEL 0x3a counter N+/-2 */
if (pmc && (psel == 0x38 || psel == 0x3a))
alt[nalt++] = ((event ^ 0x2) & ~PM_PMC_MSKS) |
((pmc > 2? pmc - 2: pmc + 2) << PM_PMC_SH);
}
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (flags & PPMU_ONLY_COUNT_RUN) {
/*
* We're only counting in RUN state,
* so PM_CYC is equivalent to PM_RUN_CYC,
* PM_INST_CMPL === PM_RUN_INST_CMPL, PM_PURR === PM_RUN_PURR.
* This doesn't include alternatives that don't provide
* any extra flexibility in assigning PMCs (e.g.
* 0x10000a for PM_RUN_CYC vs. 0x1e for PM_CYC).
* Note that even with these additional alternatives
* we never end up with more than 4 alternatives for any event.
*/
j = nalt;
for (i = 0; i < nalt; ++i) {
switch (alt[i]) {
case 0x1e: /* PM_CYC */
alt[j++] = 0x600005; /* PM_RUN_CYC */
++nlim;
break;
case 0x10000a: /* PM_RUN_CYC */
alt[j++] = 0x1e; /* PM_CYC */
break;
case 2: /* PM_INST_CMPL */
alt[j++] = 0x500009; /* PM_RUN_INST_CMPL */
++nlim;
break;
case 0x500009: /* PM_RUN_INST_CMPL */
alt[j++] = 2; /* PM_INST_CMPL */
break;
case 0x10000e: /* PM_PURR */
alt[j++] = 0x4000f4; /* PM_RUN_PURR */
break;
case 0x4000f4: /* PM_RUN_PURR */
alt[j++] = 0x10000e; /* PM_PURR */
break;
}
}
nalt = j;
}
if (!(flags & PPMU_LIMITED_PMC_OK) && nlim) {
/* remove the limited PMC events */
j = 0;
for (i = 0; i < nalt; ++i) {
if (!p6_limited_pmc_event(alt[i])) {
alt[j] = alt[i];
++j;
}
}
nalt = j;
} else if ((flags & PPMU_LIMITED_PMC_REQD) && nlim < nalt) {
/* remove all but the limited PMC events */
j = 0;
for (i = 0; i < nalt; ++i) {
if (p6_limited_pmc_event(alt[i])) {
alt[j] = alt[i];
++j;
}
}
nalt = j;
}
return nalt;
}
static void p6_disable_pmc(unsigned int pmc, unsigned long mmcr[])
{
/* Set PMCxSEL to 0 to disable PMCx */
perf_counter: powerpc: allow use of limited-function counters POWER5+ and POWER6 have two hardware counters with limited functionality: PMC5 counts instructions completed in run state and PMC6 counts cycles in run state. (Run state is the state when a hardware RUN bit is 1; the idle task clears RUN while waiting for work to do and sets it when there is work to do.) These counters can't be written to by the kernel, can't generate interrupts, and don't obey the freeze conditions. That means we can only use them for per-task counters (where we know we'll always be in run state; we can't put a per-task counter on an idle task), and only if we don't want interrupts and we do want to count in all processor modes. Obviously some counters can't go on a limited hardware counter, but there are also situations where we can only put a counter on a limited hardware counter - if there are already counters on that exclude some processor modes and we want to put on a per-task cycle or instruction counter that doesn't exclude any processor mode, it could go on if it can use a limited hardware counter. To keep track of these constraints, this adds a flags argument to the processor-specific get_alternatives() functions, with three bits defined: one to say that we can accept alternative event codes that go on limited counters, one to say we only want alternatives on limited counters, and one to say that this is a per-task counter and therefore events that are gated by run state are equivalent to those that aren't (e.g. a "cycles" event is equivalent to a "cycles in run state" event). These flags are computed for each counter and stored in the counter->hw.counter_base field (slightly wonky name for what it does, but it was an existing unused field). Since the limited counters don't freeze when we freeze the other counters, we need some special handling to avoid getting skew between things counted on the limited counters and those counted on normal counters. To minimize this skew, if we are using any limited counters, we read PMC5 and PMC6 immediately after setting and clearing the freeze bit. This is done in a single asm in the new write_mmcr0() function. The code here is specific to PMC5 and PMC6 being the limited hardware counters. Being more general (e.g. having a bitmap of limited hardware counter numbers) would have meant more complex code to read the limited counters when freezing and unfreezing the normal counters, with conditional branches, which would have increased the skew. Since it isn't necessary for the code to be more general at this stage, it isn't. This also extends the back-ends for POWER5+ and POWER6 to be able to handle up to 6 counters rather than the 4 they previously handled. Signed-off-by: Paul Mackerras <paulus@samba.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> LKML-Reference: <18936.19035.163066.892208@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-04-29 06:38:51 -06:00
if (pmc <= 3)
mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SH(pmc));
}
static int power6_generic_events[] = {
[PERF_COUNT_HW_CPU_CYCLES] = 0x1e,
[PERF_COUNT_HW_INSTRUCTIONS] = 2,
[PERF_COUNT_HW_CACHE_REFERENCES] = 0x280030, /* LD_REF_L1 */
[PERF_COUNT_HW_CACHE_MISSES] = 0x30000c, /* LD_MISS_L1 */
[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x410a0, /* BR_PRED */
[PERF_COUNT_HW_BRANCH_MISSES] = 0x400052, /* BR_MPRED */
};
#define C(x) PERF_COUNT_HW_CACHE_##x
/*
* Table of generalized cache-related events.
* 0 means not supported, -1 means nonsensical, other values
* are event codes.
* The "DTLB" and "ITLB" events relate to the DERAT and IERAT.
*/
static int power6_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
[C(L1D)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0x280030, 0x80080 },
[C(OP_WRITE)] = { 0x180032, 0x80088 },
[C(OP_PREFETCH)] = { 0x810a4, 0 },
},
[C(L1I)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0, 0x100056 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { 0x4008c, 0 },
},
[C(LL)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0x150730, 0x250532 },
[C(OP_WRITE)] = { 0x250432, 0x150432 },
[C(OP_PREFETCH)] = { 0x810a6, 0 },
},
[C(DTLB)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0, 0x20000e },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
[C(ITLB)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0, 0x420ce },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
[C(BPU)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0x430e6, 0x400052 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
[C(NODE)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { -1, -1 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
};
static struct power_pmu power6_pmu = {
.name = "POWER6",
.n_counter = 6,
.max_alternatives = MAX_ALT,
.add_fields = 0x1555,
.test_adder = 0x3000,
.compute_mmcr = p6_compute_mmcr,
.get_constraint = p6_get_constraint,
.get_alternatives = p6_get_alternatives,
.disable_pmc = p6_disable_pmc,
.limited_pmc_event = p6_limited_pmc_event,
.flags = PPMU_LIMITED_PMC5_6 | PPMU_ALT_SIPR,
.n_generic = ARRAY_SIZE(power6_generic_events),
.generic_events = power6_generic_events,
.cache_events = &power6_cache_events,
};
static int __init init_power6_pmu(void)
{
if (!cur_cpu_spec->oprofile_cpu_type ||
strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power6"))
return -ENODEV;
return register_power_pmu(&power6_pmu);
}
early_initcall(init_power6_pmu);