Release 4.14 arch/powerpc/oprofile/op_model_power4.c
/*
* Copyright (C) 2004 Anton Blanchard <anton@au.ibm.com>, IBM
* Added mmcra[slot] support:
* Copyright (C) 2006-2007 Will Schmidt <willschm@us.ibm.com>, IBM
*
* 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/oprofile.h>
#include <linux/smp.h>
#include <asm/firmware.h>
#include <asm/ptrace.h>
#include <asm/processor.h>
#include <asm/cputable.h>
#include <asm/rtas.h>
#include <asm/oprofile_impl.h>
#include <asm/reg.h>
#define dbg(args...)
#define OPROFILE_PM_PMCSEL_MSK 0xffULL
#define OPROFILE_PM_UNIT_SHIFT 60
#define OPROFILE_PM_UNIT_MSK 0xfULL
#define OPROFILE_MAX_PMC_NUM 3
#define OPROFILE_PMSEL_FIELD_WIDTH 8
#define OPROFILE_UNIT_FIELD_WIDTH 4
#define MMCRA_SIAR_VALID_MASK 0x10000000ULL
static unsigned long reset_value[OP_MAX_COUNTER];
static int oprofile_running;
static int use_slot_nums;
/* mmcr values are set in power4_reg_setup, used in power4_cpu_setup */
static u32 mmcr0_val;
static u64 mmcr1_val;
static u64 mmcra_val;
static u32 cntr_marked_events;
static int power7_marked_instr_event(u64 mmcr1)
{
u64 psel, unit;
int pmc, cntr_marked_events = 0;
/* Given the MMCR1 value, look at the field for each counter to
* determine if it is a marked event. Code based on the function
* power7_marked_instr_event() in file arch/powerpc/perf/power7-pmu.c.
*/
for (pmc = 0; pmc < 4; pmc++) {
psel = mmcr1 & (OPROFILE_PM_PMCSEL_MSK
<< (OPROFILE_MAX_PMC_NUM - pmc)
* OPROFILE_PMSEL_FIELD_WIDTH);
psel = (psel >> ((OPROFILE_MAX_PMC_NUM - pmc)
* OPROFILE_PMSEL_FIELD_WIDTH)) & ~1ULL;
unit = mmcr1 & (OPROFILE_PM_UNIT_MSK
<< (OPROFILE_PM_UNIT_SHIFT
- (pmc * OPROFILE_PMSEL_FIELD_WIDTH )));
unit = unit >> (OPROFILE_PM_UNIT_SHIFT
- (pmc * OPROFILE_PMSEL_FIELD_WIDTH));
switch (psel >> 4) {
case 2:
cntr_marked_events |= (pmc == 1 || pmc == 3) << pmc;
break;
case 3:
if (psel == 0x3c) {
cntr_marked_events |= (pmc == 0) << pmc;
break;
}
if (psel == 0x3e) {
cntr_marked_events |= (pmc != 1) << pmc;
break;
}
cntr_marked_events |= 1 << pmc;
break;
case 4:
case 5:
cntr_marked_events |= (unit == 0xd) << pmc;
break;
case 6:
if (psel == 0x64)
cntr_marked_events |= (pmc >= 2) << pmc;
break;
case 8:
cntr_marked_events |= (unit == 0xd) << pmc;
break;
}
}
return cntr_marked_events;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Carl E. Love | 232 | 100.00% | 2 | 100.00% |
| Total | 232 | 100.00% | 2 | 100.00% |
static int power4_reg_setup(struct op_counter_config *ctr,
struct op_system_config *sys,
int num_ctrs)
{
int i;
/*
* The performance counter event settings are given in the mmcr0,
* mmcr1 and mmcra values passed from the user in the
* op_system_config structure (sys variable).
*/
mmcr0_val = sys->mmcr0;
mmcr1_val = sys->mmcr1;
mmcra_val = sys->mmcra;
/* Power 7+ and newer architectures:
* Determine which counter events in the group (the group of events is
* specified by the bit settings in the MMCR1 register) are marked
* events for use in the interrupt handler. Do the calculation once
* before OProfile starts. Information is used in the interrupt
* handler. Starting with Power 7+ we only record the sample for
* marked events if the SIAR valid bit is set. For non marked events
* the sample is always recorded.
*/
if (pvr_version_is(PVR_POWER7p))
cntr_marked_events = power7_marked_instr_event(mmcr1_val);
else
cntr_marked_events = 0; /* For older processors, set the bit map
* to zero so the sample will always be
* be recorded.
*/
for (i = 0; i < cur_cpu_spec->num_pmcs; ++i)
reset_value[i] = 0x80000000UL - ctr[i].count;
/* setup user and kernel profiling */
if (sys->enable_kernel)
mmcr0_val &= ~MMCR0_KERNEL_DISABLE;
else
mmcr0_val |= MMCR0_KERNEL_DISABLE;
if (sys->enable_user)
mmcr0_val &= ~MMCR0_PROBLEM_DISABLE;
else
mmcr0_val |= MMCR0_PROBLEM_DISABLE;
if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p) ||
pvr_version_is(PVR_970) || pvr_version_is(PVR_970FX) ||
pvr_version_is(PVR_970MP) || pvr_version_is(PVR_970GX) ||
pvr_version_is(PVR_POWER5) || pvr_version_is(PVR_POWER5p))
use_slot_nums = 1;
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Anton Blanchard | 55 | 31.79% | 2 | 28.57% |
| Andrew Morton | 47 | 27.17% | 1 | 14.29% |
| Maynard Johnson | 30 | 17.34% | 1 | 14.29% |
| Carl E. Love | 21 | 12.14% | 1 | 14.29% |
| Michael Ellerman | 16 | 9.25% | 1 | 14.29% |
| Bob Nelson | 4 | 2.31% | 1 | 14.29% |
| Total | 173 | 100.00% | 7 | 100.00% |
extern void ppc_enable_pmcs(void);
/*
* Older CPUs require the MMCRA sample bit to be always set, but newer
* CPUs only want it set for some groups. Eventually we will remove all
* knowledge of this bit in the kernel, oprofile userspace should be
* setting it when required.
*
* In order to keep current installations working we force the bit for
* those older CPUs. Once everyone has updated their oprofile userspace we
* can remove this hack.
*/
static inline int mmcra_must_set_sample(void)
{
if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p) ||
pvr_version_is(PVR_970) || pvr_version_is(PVR_970FX) ||
pvr_version_is(PVR_970MP) || pvr_version_is(PVR_970GX))
return 1;
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Anton Blanchard | 32 | 68.09% | 1 | 33.33% |
| Michael Ellerman | 12 | 25.53% | 1 | 33.33% |
| Jake Moilanen | 3 | 6.38% | 1 | 33.33% |
| Total | 47 | 100.00% | 3 | 100.00% |
static int power4_cpu_setup(struct op_counter_config *ctr)
{
unsigned int mmcr0 = mmcr0_val;
unsigned long mmcra = mmcra_val;
ppc_enable_pmcs();
/* set the freeze bit */
mmcr0 |= MMCR0_FC;
mtspr(SPRN_MMCR0, mmcr0);
mmcr0 |= MMCR0_FCM1|MMCR0_PMXE|MMCR0_FCECE;
mmcr0 |= MMCR0_PMC1CE|MMCR0_PMCjCE;
mtspr(SPRN_MMCR0, mmcr0);
mtspr(SPRN_MMCR1, mmcr1_val);
if (mmcra_must_set_sample())
mmcra |= MMCRA_SAMPLE_ENABLE;
mtspr(SPRN_MMCRA, mmcra);
dbg("setup on cpu %d, mmcr0 %lx\n", smp_processor_id(),
mfspr(SPRN_MMCR0));
dbg("setup on cpu %d, mmcr1 %lx\n", smp_processor_id(),
mfspr(SPRN_MMCR1));
dbg("setup on cpu %d, mmcra %lx\n", smp_processor_id(),
mfspr(SPRN_MMCRA));
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 100 | 80.65% | 1 | 14.29% |
| Anton Blanchard | 14 | 11.29% | 2 | 28.57% |
| Bob Nelson | 4 | 3.23% | 1 | 14.29% |
| Andy Fleming | 3 | 2.42% | 1 | 14.29% |
| David Gibson | 2 | 1.61% | 1 | 14.29% |
| Benjamin Herrenschmidt | 1 | 0.81% | 1 | 14.29% |
| Total | 124 | 100.00% | 7 | 100.00% |
static int power4_start(struct op_counter_config *ctr)
{
int i;
unsigned int mmcr0;
/* set the PMM bit (see comment below) */
mtmsr(mfmsr() | MSR_PMM);
for (i = 0; i < cur_cpu_spec->num_pmcs; ++i) {
if (ctr[i].enabled) {
classic_ctr_write(i, reset_value[i]);
} else {
classic_ctr_write(i, 0);
}
}
mmcr0 = mfspr(SPRN_MMCR0);
/*
* We must clear the PMAO bit on some (GQ) chips. Just do it
* all the time
*/
mmcr0 &= ~MMCR0_PMAO;
/*
* now clear the freeze bit, counting will not start until we
* rfid from this excetion, because only at that point will
* the PMM bit be cleared
*/
mmcr0 &= ~MMCR0_FC;
mtspr(SPRN_MMCR0, mmcr0);
oprofile_running = 1;
dbg("start on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 108 | 91.53% | 2 | 33.33% |
| Anton Blanchard | 4 | 3.39% | 2 | 33.33% |
| Bob Nelson | 4 | 3.39% | 1 | 16.67% |
| Olof Johansson | 2 | 1.69% | 1 | 16.67% |
| Total | 118 | 100.00% | 6 | 100.00% |
static void power4_stop(void)
{
unsigned int mmcr0;
/* freeze counters */
mmcr0 = mfspr(SPRN_MMCR0);
mmcr0 |= MMCR0_FC;
mtspr(SPRN_MMCR0, mmcr0);
oprofile_running = 0;
dbg("stop on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
mb();
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 48 | 100.00% | 2 | 100.00% |
| Total | 48 | 100.00% | 2 | 100.00% |
/* Fake functions used by canonicalize_pc */
static void __used hypervisor_bucket(void)
{
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 7 | 87.50% | 1 | 50.00% |
| Adrian Bunk | 1 | 12.50% | 1 | 50.00% |
| Total | 8 | 100.00% | 2 | 100.00% |
static void __used rtas_bucket(void)
{
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 7 | 87.50% | 1 | 50.00% |
| Adrian Bunk | 1 | 12.50% | 1 | 50.00% |
| Total | 8 | 100.00% | 2 | 100.00% |
static void __used kernel_unknown_bucket(void)
{
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 7 | 87.50% | 1 | 50.00% |
| Adrian Bunk | 1 | 12.50% | 1 | 50.00% |
| Total | 8 | 100.00% | 2 | 100.00% |
/*
* On GQ and newer the MMCRA stores the HV and PR bits at the time
* the SIAR was sampled. We use that to work out if the SIAR was sampled in
* the hypervisor, our exception vectors or RTAS.
* If the MMCRA_SAMPLE_ENABLE bit is set, we can use the MMCRA[slot] bits
* to more accurately identify the address of the sampled instruction. The
* mmcra[slot] bits represent the slot number of a sampled instruction
* within an instruction group. The slot will contain a value between 1
* and 5 if MMCRA_SAMPLE_ENABLE is set, otherwise 0.
*/
static unsigned long get_pc(struct pt_regs *regs)
{
unsigned long pc = mfspr(SPRN_SIAR);
unsigned long mmcra;
unsigned long slot;
/* Can't do much about it */
if (!cur_cpu_spec->oprofile_mmcra_sihv)
return pc;
mmcra = mfspr(SPRN_MMCRA);
if (use_slot_nums && (mmcra & MMCRA_SAMPLE_ENABLE)) {
slot = ((mmcra & MMCRA_SLOT) >> MMCRA_SLOT_SHIFT);
if (slot > 1)
pc += 4 * (slot - 1);
}
/* Were we in the hypervisor? */
if (firmware_has_feature(FW_FEATURE_LPAR) &&
(mmcra & cur_cpu_spec->oprofile_mmcra_sihv))
/* function descriptor madness */
return *((unsigned long *)hypervisor_bucket);
/* We were in userspace, nothing to do */
if (mmcra & cur_cpu_spec->oprofile_mmcra_sipr)
return pc;
#ifdef CONFIG_PPC_RTAS
/* Were we in RTAS? */
if (pc >= rtas.base && pc < (rtas.base + rtas.size))
/* function descriptor madness */
return *((unsigned long *)rtas_bucket);
#endif
/* Were we in our exception vectors or SLB real mode miss handler? */
if (pc < 0x1000000UL)
return (unsigned long)__va(pc);
/* Not sure where we were */
if (!is_kernel_addr(pc))
/* function descriptor madness */
return *((unsigned long *)kernel_unknown_bucket);
return pc;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 122 | 59.22% | 1 | 11.11% |
| Will Schmidt | 40 | 19.42% | 1 | 11.11% |
| Anton Blanchard | 21 | 10.19% | 1 | 11.11% |
| Michael Neuling | 9 | 4.37% | 1 | 11.11% |
| Michael Ellerman | 8 | 3.88% | 2 | 22.22% |
| Maynard Johnson | 4 | 1.94% | 1 | 11.11% |
| Lucas De Marchi | 1 | 0.49% | 1 | 11.11% |
| Paul Mackerras | 1 | 0.49% | 1 | 11.11% |
| Total | 206 | 100.00% | 9 | 100.00% |
static int get_kernel(unsigned long pc, unsigned long mmcra)
{
int is_kernel;
if (!cur_cpu_spec->oprofile_mmcra_sihv) {
is_kernel = is_kernel_addr(pc);
} else {
is_kernel = ((mmcra & cur_cpu_spec->oprofile_mmcra_sipr) == 0);
}
return is_kernel;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 40 | 75.47% | 1 | 33.33% |
| Michael Neuling | 10 | 18.87% | 1 | 33.33% |
| Michael Ellerman | 3 | 5.66% | 1 | 33.33% |
| Total | 53 | 100.00% | 3 | 100.00% |
static bool pmc_overflow(unsigned long val)
{
if ((int)val < 0)
return true;
/*
* Events on POWER7 can roll back if a speculative event doesn't
* eventually complete. Unfortunately in some rare cases they will
* raise a performance monitor exception. We need to catch this to
* ensure we reset the PMC. In all cases the PMC will be 256 or less
* cycles from overflow.
*
* We only do this if the first pass fails to find any overflowing
* PMCs because a user might set a period of less than 256 and we
* don't want to mistakenly reset them.
*/
if (pvr_version_is(PVR_POWER7) && ((0x80000000 - val) <= 256))
return true;
return false;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Eric B Munson | 44 | 95.65% | 1 | 50.00% |
| Michael Ellerman | 2 | 4.35% | 1 | 50.00% |
| Total | 46 | 100.00% | 2 | 100.00% |
static void power4_handle_interrupt(struct pt_regs *regs,
struct op_counter_config *ctr)
{
unsigned long pc;
int is_kernel;
int val;
int i;
unsigned int mmcr0;
unsigned long mmcra;
bool siar_valid = false;
mmcra = mfspr(SPRN_MMCRA);
pc = get_pc(regs);
is_kernel = get_kernel(pc, mmcra);
/* set the PMM bit (see comment below) */
mtmsr(mfmsr() | MSR_PMM);
/* Check that the SIAR valid bit in MMCRA is set to 1. */
if ((mmcra & MMCRA_SIAR_VALID_MASK) == MMCRA_SIAR_VALID_MASK)
siar_valid = true;
for (i = 0; i < cur_cpu_spec->num_pmcs; ++i) {
val = classic_ctr_read(i);
if (pmc_overflow(val)) {
if (oprofile_running && ctr[i].enabled) {
/* Power 7+ and newer architectures:
* If the event is a marked event, then only
* save the sample if the SIAR valid bit is
* set. If the event is not marked, then
* always save the sample.
* Note, the Sample enable bit in the MMCRA
* register must be set to 1 if the group
* contains a marked event.
*/
if ((siar_valid &&
(cntr_marked_events & (1 << i)))
|| !(cntr_marked_events & (1 << i)))
oprofile_add_ext_sample(pc, regs, i,
is_kernel);
classic_ctr_write(i, reset_value[i]);
} else {
classic_ctr_write(i, 0);
}
}
}
mmcr0 = mfspr(SPRN_MMCR0);
/* reset the perfmon trigger */
mmcr0 |= MMCR0_PMXE;
/*
* We must clear the PMAO bit on some (GQ) chips. Just do it
* all the time
*/
mmcr0 &= ~MMCR0_PMAO;
/* Clear the appropriate bits in the MMCRA */
mmcra &= ~cur_cpu_spec->oprofile_mmcra_clear;
mtspr(SPRN_MMCRA, mmcra);
/*
* now clear the freeze bit, counting will not start until we
* rfid from this exception, because only at that point will
* the PMM bit be cleared
*/
mmcr0 &= ~MMCR0_FC;
mtspr(SPRN_MMCR0, mmcr0);
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 147 | 61.25% | 2 | 20.00% |
| Carl E. Love | 48 | 20.00% | 1 | 10.00% |
| Michael Neuling | 28 | 11.67% | 1 | 10.00% |
| Anton Blanchard | 7 | 2.92% | 3 | 30.00% |
| Brian Rogan | 4 | 1.67% | 1 | 10.00% |
| Eric B Munson | 3 | 1.25% | 1 | 10.00% |
| Olof Johansson | 3 | 1.25% | 1 | 10.00% |
| Total | 240 | 100.00% | 10 | 100.00% |
struct op_powerpc_model op_model_power4 = {
.reg_setup = power4_reg_setup,
.cpu_setup = power4_cpu_setup,
.start = power4_start,
.stop = power4_stop,
.handle_interrupt = power4_handle_interrupt,
};
Overall Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Andrew Morton | 707 | 48.83% | 2 | 6.90% |
| Carl E. Love | 333 | 23.00% | 2 | 6.90% |
| Anton Blanchard | 153 | 10.57% | 7 | 24.14% |
| Michael Neuling | 47 | 3.25% | 1 | 3.45% |
| Eric B Munson | 47 | 3.25% | 1 | 3.45% |
| Michael Ellerman | 44 | 3.04% | 3 | 10.34% |
| Will Schmidt | 42 | 2.90% | 1 | 3.45% |
| Maynard Johnson | 38 | 2.62% | 1 | 3.45% |
| Bob Nelson | 12 | 0.83% | 1 | 3.45% |
| Olof Johansson | 5 | 0.35% | 1 | 3.45% |
| Brian Rogan | 4 | 0.28% | 1 | 3.45% |
| Adrian Bunk | 3 | 0.21% | 1 | 3.45% |
| Andy Fleming | 3 | 0.21% | 1 | 3.45% |
| Jake Moilanen | 3 | 0.21% | 1 | 3.45% |
| David Gibson | 2 | 0.14% | 1 | 3.45% |
| Benjamin Herrenschmidt | 2 | 0.14% | 1 | 3.45% |
| Paul Mackerras | 1 | 0.07% | 1 | 3.45% |
| Stephen Rothwell | 1 | 0.07% | 1 | 3.45% |
| Lucas De Marchi | 1 | 0.07% | 1 | 3.45% |
| Total | 1448 | 100.00% | 29 | 100.00% |
Information contained on this website is for historical information purposes only and does not indicate or represent copyright ownership.