linux/arch/powerpc/kernel/power5+-pmu.c

686 lines
18 KiB
C
Raw Normal View History

/*
* Performance counter support for POWER5+/++ (not POWER5) processors.
*
* Copyright 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 10:02:48 +00:00
#include <linux/perf_event.h>
#include <linux/string.h>
#include <asm/reg.h>
#include <asm/cputable.h>
/*
* Bits in event code for POWER5+ (POWER5 GS) and POWER5++ (POWER5 GS DD3)
*/
#define PM_PMC_SH 20 /* PMC number (1-based) for direct events */
#define PM_PMC_MSK 0xf
#define PM_PMC_MSKS (PM_PMC_MSK << PM_PMC_SH)
#define PM_UNIT_SH 16 /* TTMMUX number and setting - unit select */
#define PM_UNIT_MSK 0xf
#define PM_BYTE_SH 12 /* Byte number of event bus to use */
#define PM_BYTE_MSK 7
#define PM_GRS_SH 8 /* Storage subsystem mux select */
#define PM_GRS_MSK 7
#define PM_BUSEVENT_MSK 0x80 /* Set if event uses event bus */
#define PM_PMCSEL_MSK 0x7f
/* Values in PM_UNIT field */
#define PM_FPU 0
#define PM_ISU0 1
#define PM_IFU 2
#define PM_ISU1 3
#define PM_IDU 4
#define PM_ISU0_ALT 6
#define PM_GRS 7
#define PM_LSU0 8
#define PM_LSU1 0xc
#define PM_LASTUNIT 0xc
/*
* Bits in MMCR1 for POWER5+
*/
#define MMCR1_TTM0SEL_SH 62
#define MMCR1_TTM1SEL_SH 60
#define MMCR1_TTM2SEL_SH 58
#define MMCR1_TTM3SEL_SH 56
#define MMCR1_TTMSEL_MSK 3
#define MMCR1_TD_CP_DBG0SEL_SH 54
#define MMCR1_TD_CP_DBG1SEL_SH 52
#define MMCR1_TD_CP_DBG2SEL_SH 50
#define MMCR1_TD_CP_DBG3SEL_SH 48
#define MMCR1_GRS_L2SEL_SH 46
#define MMCR1_GRS_L2SEL_MSK 3
#define MMCR1_GRS_L3SEL_SH 44
#define MMCR1_GRS_L3SEL_MSK 3
#define MMCR1_GRS_MCSEL_SH 41
#define MMCR1_GRS_MCSEL_MSK 7
#define MMCR1_GRS_FABSEL_SH 39
#define MMCR1_GRS_FABSEL_MSK 3
#define MMCR1_PMC1_ADDER_SEL_SH 35
#define MMCR1_PMC2_ADDER_SEL_SH 34
#define MMCR1_PMC3_ADDER_SEL_SH 33
#define MMCR1_PMC4_ADDER_SEL_SH 32
#define MMCR1_PMC1SEL_SH 25
#define MMCR1_PMC2SEL_SH 17
#define MMCR1_PMC3SEL_SH 9
#define MMCR1_PMC4SEL_SH 1
#define MMCR1_PMCSEL_SH(n) (MMCR1_PMC1SEL_SH - (n) * 8)
#define MMCR1_PMCSEL_MSK 0x7f
/*
* Layout of constraint bits:
* 6666555555555544444444443333333333222222222211111111110000000000
* 3210987654321098765432109876543210987654321098765432109876543210
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 12:38:51 +00:00
* [ ><><>< ><> <><>[ > < >< >< >< ><><><><><><>
* NC G0G1G2 G3 T0T1 UC B0 B1 B2 B3 P6P5P4P3P2P1
*
* NC - number of counters
* 51: NC error 0x0008_0000_0000_0000
* 48-50: number of events needing PMC1-4 0x0007_0000_0000_0000
*
* G0..G3 - GRS mux constraints
* 46-47: GRS_L2SEL value
* 44-45: GRS_L3SEL value
* 41-44: GRS_MCSEL value
* 39-40: GRS_FABSEL value
* Note that these match up with their bit positions in MMCR1
*
* T0 - TTM0 constraint
* 36-37: TTM0SEL value (0=FPU, 2=IFU, 3=ISU1) 0x30_0000_0000
*
* T1 - TTM1 constraint
* 34-35: TTM1SEL value (0=IDU, 3=GRS) 0x0c_0000_0000
*
* UC - unit constraint: can't have all three of FPU|IFU|ISU1, ISU0, IDU|GRS
* 33: UC3 error 0x02_0000_0000
* 32: FPU|IFU|ISU1 events needed 0x01_0000_0000
* 31: ISU0 events needed 0x01_8000_0000
* 30: IDU|GRS events needed 0x00_4000_0000
*
* B0
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 12:38:51 +00:00
* 24-27: Byte 0 event source 0x0f00_0000
* Encoding as for the event code
*
* B1, B2, B3
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 12:38:51 +00:00
* 20-23, 16-19, 12-15: Byte 1, 2, 3 event sources
*
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 12:38:51 +00:00
* P6
* 11: P6 error 0x800
* 10-11: Count of events needing PMC6
*
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 12:38:51 +00:00
* P1..P5
* 0-9: Count of events needing PMC1..PMC5
*/
static const int grsel_shift[8] = {
MMCR1_GRS_L2SEL_SH, MMCR1_GRS_L2SEL_SH, MMCR1_GRS_L2SEL_SH,
MMCR1_GRS_L3SEL_SH, MMCR1_GRS_L3SEL_SH, MMCR1_GRS_L3SEL_SH,
MMCR1_GRS_MCSEL_SH, MMCR1_GRS_FABSEL_SH
};
/* Masks and values for using events from the various units */
static unsigned long unit_cons[PM_LASTUNIT+1][2] = {
[PM_FPU] = { 0x3200000000ul, 0x0100000000ul },
[PM_ISU0] = { 0x0200000000ul, 0x0080000000ul },
[PM_ISU1] = { 0x3200000000ul, 0x3100000000ul },
[PM_IFU] = { 0x3200000000ul, 0x2100000000ul },
[PM_IDU] = { 0x0e00000000ul, 0x0040000000ul },
[PM_GRS] = { 0x0e00000000ul, 0x0c40000000ul },
};
static int power5p_get_constraint(u64 event, unsigned long *maskp,
unsigned long *valp)
{
int pmc, byte, unit, sh;
int bit, fmask;
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 12:38:51 +00:00
if (pmc > 6)
return -1;
sh = (pmc - 1) * 2;
mask |= 2 << sh;
value |= 1 << 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 12:38:51 +00:00
if (pmc >= 5 && !(event == 0x500009 || event == 0x600005))
return -1;
}
if (event & PM_BUSEVENT_MSK) {
unit = (event >> PM_UNIT_SH) & PM_UNIT_MSK;
if (unit > PM_LASTUNIT)
return -1;
if (unit == PM_ISU0_ALT)
unit = PM_ISU0;
mask |= unit_cons[unit][0];
value |= unit_cons[unit][1];
byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK;
if (byte >= 4) {
if (unit != PM_LSU1)
return -1;
/* Map LSU1 low word (bytes 4-7) to unit LSU1+1 */
++unit;
byte &= 3;
}
if (unit == PM_GRS) {
bit = event & 7;
fmask = (bit == 6)? 7: 3;
sh = grsel_shift[bit];
mask |= (unsigned long)fmask << sh;
value |= (unsigned long)((event >> PM_GRS_SH) & fmask)
<< sh;
}
/* Set byte lane select field */
mask |= 0xfUL << (24 - 4 * byte);
value |= (unsigned long)unit << (24 - 4 * byte);
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 12:38:51 +00:00
}
if (pmc < 5) {
/* need a counter from PMC1-4 set */
mask |= 0x8000000000000ul;
value |= 0x1000000000000ul;
}
*maskp = mask;
*valp = value;
return 0;
}
static int power5p_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 12:38:51 +00:00
{
int pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
return pmc == 5 || pmc == 6;
}
#define MAX_ALT 3 /* at most 3 alternatives for any event */
static const unsigned int event_alternatives[][MAX_ALT] = {
{ 0x100c0, 0x40001f }, /* PM_GCT_FULL_CYC */
{ 0x120e4, 0x400002 }, /* PM_GRP_DISP_REJECT */
{ 0x230e2, 0x323087 }, /* PM_BR_PRED_CR */
{ 0x230e3, 0x223087, 0x3230a0 }, /* PM_BR_PRED_TA */
{ 0x410c7, 0x441084 }, /* PM_THRD_L2MISS_BOTH_CYC */
{ 0x800c4, 0xc20e0 }, /* PM_DTLB_MISS */
{ 0xc50c6, 0xc60e0 }, /* PM_MRK_DTLB_MISS */
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 12:38:51 +00:00
{ 0x100005, 0x600005 }, /* PM_RUN_CYC */
{ 0x100009, 0x200009 }, /* PM_INST_CMPL */
{ 0x200015, 0x300015 }, /* PM_LSU_LMQ_SRQ_EMPTY_CYC */
{ 0x300009, 0x400009 }, /* PM_INST_DISP */
};
/*
* Scan the alternatives table for a match and return the
* index into the alternatives table if found, else -1.
*/
static int find_alternative(unsigned int event)
{
int i, j;
for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) {
if (event < event_alternatives[i][0])
break;
for (j = 0; j < MAX_ALT && event_alternatives[i][j]; ++j)
if (event == event_alternatives[i][j])
return i;
}
return -1;
}
static const unsigned char bytedecode_alternatives[4][4] = {
/* PMC 1 */ { 0x21, 0x23, 0x25, 0x27 },
/* PMC 2 */ { 0x07, 0x17, 0x0e, 0x1e },
/* PMC 3 */ { 0x20, 0x22, 0x24, 0x26 },
/* PMC 4 */ { 0x07, 0x17, 0x0e, 0x1e }
};
/*
* Some direct events for decodes of event bus byte 3 have alternative
* PMCSEL values on other counters. This returns the alternative
* event code for those that do, or -1 otherwise. This also handles
* alternative PCMSEL values for add events.
*/
static s64 find_alternative_bdecode(u64 event)
{
int pmc, altpmc, pp, j;
pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
if (pmc == 0 || pmc > 4)
return -1;
altpmc = 5 - pmc; /* 1 <-> 4, 2 <-> 3 */
pp = event & PM_PMCSEL_MSK;
for (j = 0; j < 4; ++j) {
if (bytedecode_alternatives[pmc - 1][j] == pp) {
return (event & ~(PM_PMC_MSKS | PM_PMCSEL_MSK)) |
(altpmc << PM_PMC_SH) |
bytedecode_alternatives[altpmc - 1][j];
}
}
/* new decode alternatives for power5+ */
if (pmc == 1 && (pp == 0x0d || pp == 0x0e))
return event + (2 << PM_PMC_SH) + (0x2e - 0x0d);
if (pmc == 3 && (pp == 0x2e || pp == 0x2f))
return event - (2 << PM_PMC_SH) - (0x2e - 0x0d);
/* alternative add event encodings */
if (pp == 0x10 || pp == 0x28)
return ((event ^ (0x10 ^ 0x28)) & ~PM_PMC_MSKS) |
(altpmc << PM_PMC_SH);
return -1;
}
static int power5p_get_alternatives(u64 event, unsigned int flags, u64 alt[])
{
int i, j, nalt = 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 12:38:51 +00:00
int nlim;
s64 ae;
alt[0] = event;
nalt = 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 12:38:51 +00:00
nlim = power5p_limited_pmc_event(event);
i = find_alternative(event);
if (i >= 0) {
for (j = 0; j < MAX_ALT; ++j) {
ae = event_alternatives[i][j];
if (ae && ae != event)
alt[nalt++] = ae;
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 12:38:51 +00:00
nlim += power5p_limited_pmc_event(ae);
}
} else {
ae = find_alternative_bdecode(event);
if (ae > 0)
alt[nalt++] = ae;
}
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 12:38:51 +00:00
if (flags & PPMU_ONLY_COUNT_RUN) {
/*
* We're only counting in RUN state,
* so PM_CYC is equivalent to PM_RUN_CYC
* and PM_INST_CMPL === PM_RUN_INST_CMPL.
* This doesn't include alternatives that don't provide
* any extra flexibility in assigning PMCs (e.g.
* 0x100005 for PM_RUN_CYC vs. 0xf for PM_CYC).
* Note that even with these additional alternatives
* we never end up with more than 3 alternatives for any event.
*/
j = nalt;
for (i = 0; i < nalt; ++i) {
switch (alt[i]) {
case 0xf: /* PM_CYC */
alt[j++] = 0x600005; /* PM_RUN_CYC */
++nlim;
break;
case 0x600005: /* PM_RUN_CYC */
alt[j++] = 0xf;
break;
case 0x100009: /* PM_INST_CMPL */
alt[j++] = 0x500009; /* PM_RUN_INST_CMPL */
++nlim;
break;
case 0x500009: /* PM_RUN_INST_CMPL */
alt[j++] = 0x100009; /* PM_INST_CMPL */
alt[j++] = 0x200009;
break;
}
}
nalt = j;
}
if (!(flags & PPMU_LIMITED_PMC_OK) && nlim) {
/* remove the limited PMC events */
j = 0;
for (i = 0; i < nalt; ++i) {
if (!power5p_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 (power5p_limited_pmc_event(alt[i])) {
alt[j] = alt[i];
++j;
}
}
nalt = j;
}
return nalt;
}
/*
* Map of which direct events on which PMCs are marked instruction events.
* Indexed by PMCSEL value, bit i (LE) set if PMC i is a marked event.
* Bit 0 is set if it is marked for all PMCs.
* The 0x80 bit indicates a byte decode PMCSEL value.
*/
static unsigned char direct_event_is_marked[0x28] = {
0, /* 00 */
0x1f, /* 01 PM_IOPS_CMPL */
0x2, /* 02 PM_MRK_GRP_DISP */
0xe, /* 03 PM_MRK_ST_CMPL, PM_MRK_ST_GPS, PM_MRK_ST_CMPL_INT */
0, /* 04 */
0x1c, /* 05 PM_MRK_BRU_FIN, PM_MRK_INST_FIN, PM_MRK_CRU_FIN */
0x80, /* 06 */
0x80, /* 07 */
0, 0, 0,/* 08 - 0a */
0x18, /* 0b PM_THRESH_TIMEO, PM_MRK_GRP_TIMEO */
0, /* 0c */
0x80, /* 0d */
0x80, /* 0e */
0, /* 0f */
0, /* 10 */
0x14, /* 11 PM_MRK_GRP_BR_REDIR, PM_MRK_GRP_IC_MISS */
0, /* 12 */
0x10, /* 13 PM_MRK_GRP_CMPL */
0x1f, /* 14 PM_GRP_MRK, PM_MRK_{FXU,FPU,LSU}_FIN */
0x2, /* 15 PM_MRK_GRP_ISSUED */
0x80, /* 16 */
0x80, /* 17 */
0, 0, 0, 0, 0,
0x80, /* 1d */
0x80, /* 1e */
0, /* 1f */
0x80, /* 20 */
0x80, /* 21 */
0x80, /* 22 */
0x80, /* 23 */
0x80, /* 24 */
0x80, /* 25 */
0x80, /* 26 */
0x80, /* 27 */
};
/*
* 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 power5p_marked_instr_event(u64 event)
{
int pmc, psel;
int bit, byte, unit;
u32 mask;
pmc = (event >> PM_PMC_SH) & PM_PMC_MSK;
psel = event & PM_PMCSEL_MSK;
if (pmc >= 5)
return 0;
bit = -1;
if (psel < sizeof(direct_event_is_marked)) {
if (direct_event_is_marked[psel] & (1 << pmc))
return 1;
if (direct_event_is_marked[psel] & 0x80)
bit = 4;
else if (psel == 0x08)
bit = pmc - 1;
else if (psel == 0x10)
bit = 4 - pmc;
else if (psel == 0x1b && (pmc == 1 || pmc == 3))
bit = 4;
} else if ((psel & 0x48) == 0x40) {
bit = psel & 7;
} else if (psel == 0x28) {
bit = pmc - 1;
} else if (pmc == 3 && (psel == 0x2e || psel == 0x2f)) {
bit = 4;
}
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;
if (unit == PM_LSU0) {
/* byte 1 bits 0-7, byte 2 bits 0,2-4,6 */
mask = 0x5dff00;
} else if (unit == PM_LSU1 && byte >= 4) {
byte -= 4;
/* byte 5 bits 6-7, byte 6 bits 0,4, byte 7 bits 0-4,6 */
mask = 0x5f11c000;
} else
return 0;
return (mask >> (byte * 8 + bit)) & 1;
}
static int power5p_compute_mmcr(u64 event[], int n_ev,
unsigned int hwc[], unsigned long mmcr[])
{
unsigned long mmcr1 = 0;
unsigned long mmcra = 0;
unsigned int pmc, unit, byte, psel;
unsigned int ttm;
int i, isbus, bit, grsel;
unsigned int pmc_inuse = 0;
unsigned char busbyte[4];
unsigned char unituse[16];
int ttmuse;
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 12:38:51 +00:00
if (n_ev > 6)
return -1;
/* First pass to count resource use */
memset(busbyte, 0, sizeof(busbyte));
memset(unituse, 0, sizeof(unituse));
for (i = 0; i < n_ev; ++i) {
pmc = (event[i] >> 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 12:38:51 +00:00
if (pmc > 6)
return -1;
if (pmc_inuse & (1 << (pmc - 1)))
return -1;
pmc_inuse |= 1 << (pmc - 1);
}
if (event[i] & PM_BUSEVENT_MSK) {
unit = (event[i] >> PM_UNIT_SH) & PM_UNIT_MSK;
byte = (event[i] >> PM_BYTE_SH) & PM_BYTE_MSK;
if (unit > PM_LASTUNIT)
return -1;
if (unit == PM_ISU0_ALT)
unit = PM_ISU0;
if (byte >= 4) {
if (unit != PM_LSU1)
return -1;
++unit;
byte &= 3;
}
if (busbyte[byte] && busbyte[byte] != unit)
return -1;
busbyte[byte] = unit;
unituse[unit] = 1;
}
}
/*
* Assign resources and set multiplexer selects.
*
* PM_ISU0 can go either on TTM0 or TTM1, but that's the only
* choice we have to deal with.
*/
if (unituse[PM_ISU0] &
(unituse[PM_FPU] | unituse[PM_IFU] | unituse[PM_ISU1])) {
unituse[PM_ISU0_ALT] = 1; /* move ISU to TTM1 */
unituse[PM_ISU0] = 0;
}
/* Set TTM[01]SEL fields. */
ttmuse = 0;
for (i = PM_FPU; i <= PM_ISU1; ++i) {
if (!unituse[i])
continue;
if (ttmuse++)
return -1;
mmcr1 |= (unsigned long)i << MMCR1_TTM0SEL_SH;
}
ttmuse = 0;
for (; i <= PM_GRS; ++i) {
if (!unituse[i])
continue;
if (ttmuse++)
return -1;
mmcr1 |= (unsigned long)(i & 3) << MMCR1_TTM1SEL_SH;
}
if (ttmuse > 1)
return -1;
/* Set byte lane select fields, TTM[23]SEL and GRS_*SEL. */
for (byte = 0; byte < 4; ++byte) {
unit = busbyte[byte];
if (!unit)
continue;
if (unit == PM_ISU0 && unituse[PM_ISU0_ALT]) {
/* get ISU0 through TTM1 rather than TTM0 */
unit = PM_ISU0_ALT;
} else if (unit == PM_LSU1 + 1) {
/* select lower word of LSU1 for this byte */
mmcr1 |= 1ul << (MMCR1_TTM3SEL_SH + 3 - byte);
}
ttm = unit >> 2;
mmcr1 |= (unsigned long)ttm
<< (MMCR1_TD_CP_DBG0SEL_SH - 2 * byte);
}
/* Second pass: assign PMCs, set PMCxSEL and PMCx_ADDER_SEL fields */
for (i = 0; i < n_ev; ++i) {
pmc = (event[i] >> PM_PMC_SH) & PM_PMC_MSK;
unit = (event[i] >> PM_UNIT_SH) & PM_UNIT_MSK;
byte = (event[i] >> PM_BYTE_SH) & PM_BYTE_MSK;
psel = event[i] & PM_PMCSEL_MSK;
isbus = event[i] & PM_BUSEVENT_MSK;
if (!pmc) {
/* Bus event or any-PMC direct event */
for (pmc = 0; pmc < 4; ++pmc) {
if (!(pmc_inuse & (1 << pmc)))
break;
}
if (pmc >= 4)
return -1;
pmc_inuse |= 1 << 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 12:38:51 +00:00
} else if (pmc <= 4) {
/* Direct event */
--pmc;
if (isbus && (byte & 2) &&
(psel == 8 || psel == 0x10 || psel == 0x28))
/* add events on higher-numbered bus */
mmcr1 |= 1ul << (MMCR1_PMC1_ADDER_SEL_SH - 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 12:38:51 +00:00
} else {
/* Instructions or run cycles on PMC5/6 */
--pmc;
}
if (isbus && unit == PM_GRS) {
bit = psel & 7;
grsel = (event[i] >> PM_GRS_SH) & PM_GRS_MSK;
mmcr1 |= (unsigned long)grsel << grsel_shift[bit];
}
if (power5p_marked_instr_event(event[i]))
mmcra |= MMCRA_SAMPLE_ENABLE;
if ((psel & 0x58) == 0x40 && (byte & 1) != ((pmc >> 1) & 1))
/* select alternate byte lane */
psel |= 0x10;
if (pmc <= 3)
mmcr1 |= psel << MMCR1_PMCSEL_SH(pmc);
hwc[i] = pmc;
}
/* Return MMCRx values */
mmcr[0] = 0;
if (pmc_inuse & 1)
mmcr[0] = MMCR0_PMC1CE;
if (pmc_inuse & 0x3e)
mmcr[0] |= MMCR0_PMCjCE;
mmcr[1] = mmcr1;
mmcr[2] = mmcra;
return 0;
}
static void power5p_disable_pmc(unsigned int pmc, unsigned long mmcr[])
{
if (pmc <= 3)
mmcr[1] &= ~(0x7fUL << MMCR1_PMCSEL_SH(pmc));
}
static int power5p_generic_events[] = {
[PERF_COUNT_HW_CPU_CYCLES] = 0xf,
[PERF_COUNT_HW_INSTRUCTIONS] = 0x100009,
[PERF_COUNT_HW_CACHE_REFERENCES] = 0x1c10a8, /* LD_REF_L1 */
[PERF_COUNT_HW_CACHE_MISSES] = 0x3c1088, /* LD_MISS_L1 */
[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x230e4, /* BR_ISSUED */
[PERF_COUNT_HW_BRANCH_MISSES] = 0x230e5, /* BR_MPRED_CR */
};
#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.
*/
static int power5p_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
[C(L1D)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0x1c10a8, 0x3c1088 },
[C(OP_WRITE)] = { 0x2c10a8, 0xc10c3 },
[C(OP_PREFETCH)] = { 0xc70e7, -1 },
},
[C(L1I)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0, 0 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { 0, 0 },
},
[C(LL)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0, 0 },
[C(OP_WRITE)] = { 0, 0 },
[C(OP_PREFETCH)] = { 0xc50c3, 0 },
},
[C(DTLB)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0xc20e4, 0x800c4 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
[C(ITLB)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0, 0x800c0 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
[C(BPU)] = { /* RESULT_ACCESS RESULT_MISS */
[C(OP_READ)] = { 0x230e4, 0x230e5 },
[C(OP_WRITE)] = { -1, -1 },
[C(OP_PREFETCH)] = { -1, -1 },
},
};
static struct power_pmu power5p_pmu = {
.name = "POWER5+/++",
.n_counter = 6,
.max_alternatives = MAX_ALT,
.add_fields = 0x7000000000055ul,
.test_adder = 0x3000040000000ul,
.compute_mmcr = power5p_compute_mmcr,
.get_constraint = power5p_get_constraint,
.get_alternatives = power5p_get_alternatives,
.disable_pmc = power5p_disable_pmc,
.limited_pmc_event = power5p_limited_pmc_event,
.flags = PPMU_LIMITED_PMC5_6,
.n_generic = ARRAY_SIZE(power5p_generic_events),
.generic_events = power5p_generic_events,
.cache_events = &power5p_cache_events,
};
static int init_power5p_pmu(void)
{
if (!cur_cpu_spec->oprofile_cpu_type ||
(strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power5+")
&& strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power5++")))
return -ENODEV;
return register_power_pmu(&power5p_pmu);
}
arch_initcall(init_power5p_pmu);