1
linux/arch/powerpc/platforms/cell/pmu.c

427 lines
9.5 KiB
C
Raw Normal View History

/*
* Cell Broadband Engine Performance Monitor
*
* (C) Copyright IBM Corporation 2001,2006
*
* Author:
* David Erb (djerb@us.ibm.com)
* Kevin Corry (kevcorry@us.ibm.com)
*
* 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, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
#include <linux/interrupt.h>
#include <linux/types.h>
#include <asm/io.h>
[POWERPC] cell: Add oprofile support Add PPU event-based and cycle-based profiling support to Oprofile for Cell. Oprofile is expected to collect data on all CPUs simultaneously. However, there is one set of performance counters per node. There are two hardware threads or virtual CPUs on each node. Hence, OProfile must multiplex in time the performance counter collection on the two virtual CPUs. The multiplexing of the performance counters is done by a virtual counter routine. Initially, the counters are configured to collect data on the even CPUs in the system, one CPU per node. In order to capture the PC for the virtual CPU when the performance counter interrupt occurs (the specified number of events between samples has occurred), the even processors are configured to handle the performance counter interrupts for their node. The virtual counter routine is called via a kernel timer after the virtual sample time. The routine stops the counters, saves the current counts, loads the last counts for the other virtual CPU on the node, sets interrupts to be handled by the other virtual CPU and restarts the counters, the virtual timer routine is scheduled to run again. The virtual sample time is kept relatively small to make sure sampling occurs on both CPUs on the node with a relatively small granularity. Whenever the counters overflow, the performance counter interrupt is called to collect the PC for the CPU where data is being collected. The oprofile driver relies on a firmware RTAS call to setup the debug bus to route the desired signals to the performance counter hardware to be counted. The RTAS call must set the routing registers appropriately in each of the islands to pass the signals down the debug bus as well as routing the signals from a particular island onto the bus. There is a second firmware RTAS call to reset the debug bus to the non pass thru state when the counters are not in use. Signed-off-by: Carl Love <carll@us.ibm.com> Signed-off-by: Maynard Johnson <mpjohn@us.ibm.com> Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-11-20 10:45:16 -07:00
#include <asm/irq_regs.h>
#include <asm/machdep.h>
#include <asm/pmc.h>
#include <asm/reg.h>
#include <asm/spu.h>
#include "cbe_regs.h"
#include "interrupt.h"
/*
* When writing to write-only mmio addresses, save a shadow copy. All of the
* registers are 32-bit, but stored in the upper-half of a 64-bit field in
* pmd_regs.
*/
#define WRITE_WO_MMIO(reg, x) \
do { \
u32 _x = (x); \
struct cbe_pmd_regs __iomem *pmd_regs; \
struct cbe_pmd_shadow_regs *shadow_regs; \
pmd_regs = cbe_get_cpu_pmd_regs(cpu); \
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu); \
out_be64(&(pmd_regs->reg), (((u64)_x) << 32)); \
shadow_regs->reg = _x; \
} while (0)
#define READ_SHADOW_REG(val, reg) \
do { \
struct cbe_pmd_shadow_regs *shadow_regs; \
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu); \
(val) = shadow_regs->reg; \
} while (0)
#define READ_MMIO_UPPER32(val, reg) \
do { \
struct cbe_pmd_regs __iomem *pmd_regs; \
pmd_regs = cbe_get_cpu_pmd_regs(cpu); \
(val) = (u32)(in_be64(&pmd_regs->reg) >> 32); \
} while (0)
/*
* Physical counter registers.
* Each physical counter can act as one 32-bit counter or two 16-bit counters.
*/
u32 cbe_read_phys_ctr(u32 cpu, u32 phys_ctr)
{
u32 val_in_latch, val = 0;
if (phys_ctr < NR_PHYS_CTRS) {
READ_SHADOW_REG(val_in_latch, counter_value_in_latch);
/* Read the latch or the actual counter, whichever is newer. */
if (val_in_latch & (1 << phys_ctr)) {
READ_SHADOW_REG(val, pm_ctr[phys_ctr]);
} else {
READ_MMIO_UPPER32(val, pm_ctr[phys_ctr]);
}
}
return val;
}
EXPORT_SYMBOL_GPL(cbe_read_phys_ctr);
void cbe_write_phys_ctr(u32 cpu, u32 phys_ctr, u32 val)
{
struct cbe_pmd_shadow_regs *shadow_regs;
u32 pm_ctrl;
if (phys_ctr < NR_PHYS_CTRS) {
/* Writing to a counter only writes to a hardware latch.
* The new value is not propagated to the actual counter
* until the performance monitor is enabled.
*/
WRITE_WO_MMIO(pm_ctr[phys_ctr], val);
pm_ctrl = cbe_read_pm(cpu, pm_control);
if (pm_ctrl & CBE_PM_ENABLE_PERF_MON) {
/* The counters are already active, so we need to
* rewrite the pm_control register to "re-enable"
* the PMU.
*/
cbe_write_pm(cpu, pm_control, pm_ctrl);
} else {
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu);
shadow_regs->counter_value_in_latch |= (1 << phys_ctr);
}
}
}
EXPORT_SYMBOL_GPL(cbe_write_phys_ctr);
/*
* "Logical" counter registers.
* These will read/write 16-bits or 32-bits depending on the
* current size of the counter. Counters 4 - 7 are always 16-bit.
*/
u32 cbe_read_ctr(u32 cpu, u32 ctr)
{
u32 val;
u32 phys_ctr = ctr & (NR_PHYS_CTRS - 1);
val = cbe_read_phys_ctr(cpu, phys_ctr);
if (cbe_get_ctr_size(cpu, phys_ctr) == 16)
val = (ctr < NR_PHYS_CTRS) ? (val >> 16) : (val & 0xffff);
return val;
}
EXPORT_SYMBOL_GPL(cbe_read_ctr);
void cbe_write_ctr(u32 cpu, u32 ctr, u32 val)
{
u32 phys_ctr;
u32 phys_val;
phys_ctr = ctr & (NR_PHYS_CTRS - 1);
if (cbe_get_ctr_size(cpu, phys_ctr) == 16) {
phys_val = cbe_read_phys_ctr(cpu, phys_ctr);
if (ctr < NR_PHYS_CTRS)
val = (val << 16) | (phys_val & 0xffff);
else
val = (val & 0xffff) | (phys_val & 0xffff0000);
}
cbe_write_phys_ctr(cpu, phys_ctr, val);
}
EXPORT_SYMBOL_GPL(cbe_write_ctr);
/*
* Counter-control registers.
* Each "logical" counter has a corresponding control register.
*/
u32 cbe_read_pm07_control(u32 cpu, u32 ctr)
{
u32 pm07_control = 0;
if (ctr < NR_CTRS)
READ_SHADOW_REG(pm07_control, pm07_control[ctr]);
return pm07_control;
}
EXPORT_SYMBOL_GPL(cbe_read_pm07_control);
void cbe_write_pm07_control(u32 cpu, u32 ctr, u32 val)
{
if (ctr < NR_CTRS)
WRITE_WO_MMIO(pm07_control[ctr], val);
}
EXPORT_SYMBOL_GPL(cbe_write_pm07_control);
/*
* Other PMU control registers. Most of these are write-only.
*/
u32 cbe_read_pm(u32 cpu, enum pm_reg_name reg)
{
u32 val = 0;
switch (reg) {
case group_control:
READ_SHADOW_REG(val, group_control);
break;
case debug_bus_control:
READ_SHADOW_REG(val, debug_bus_control);
break;
case trace_address:
READ_MMIO_UPPER32(val, trace_address);
break;
case ext_tr_timer:
READ_SHADOW_REG(val, ext_tr_timer);
break;
case pm_status:
READ_MMIO_UPPER32(val, pm_status);
break;
case pm_control:
READ_SHADOW_REG(val, pm_control);
break;
case pm_interval:
READ_SHADOW_REG(val, pm_interval);
break;
case pm_start_stop:
READ_SHADOW_REG(val, pm_start_stop);
break;
}
return val;
}
EXPORT_SYMBOL_GPL(cbe_read_pm);
void cbe_write_pm(u32 cpu, enum pm_reg_name reg, u32 val)
{
switch (reg) {
case group_control:
WRITE_WO_MMIO(group_control, val);
break;
case debug_bus_control:
WRITE_WO_MMIO(debug_bus_control, val);
break;
case trace_address:
WRITE_WO_MMIO(trace_address, val);
break;
case ext_tr_timer:
WRITE_WO_MMIO(ext_tr_timer, val);
break;
case pm_status:
WRITE_WO_MMIO(pm_status, val);
break;
case pm_control:
WRITE_WO_MMIO(pm_control, val);
break;
case pm_interval:
WRITE_WO_MMIO(pm_interval, val);
break;
case pm_start_stop:
WRITE_WO_MMIO(pm_start_stop, val);
break;
}
}
EXPORT_SYMBOL_GPL(cbe_write_pm);
/*
* Get/set the size of a physical counter to either 16 or 32 bits.
*/
u32 cbe_get_ctr_size(u32 cpu, u32 phys_ctr)
{
u32 pm_ctrl, size = 0;
if (phys_ctr < NR_PHYS_CTRS) {
pm_ctrl = cbe_read_pm(cpu, pm_control);
size = (pm_ctrl & CBE_PM_16BIT_CTR(phys_ctr)) ? 16 : 32;
}
return size;
}
EXPORT_SYMBOL_GPL(cbe_get_ctr_size);
void cbe_set_ctr_size(u32 cpu, u32 phys_ctr, u32 ctr_size)
{
u32 pm_ctrl;
if (phys_ctr < NR_PHYS_CTRS) {
pm_ctrl = cbe_read_pm(cpu, pm_control);
switch (ctr_size) {
case 16:
pm_ctrl |= CBE_PM_16BIT_CTR(phys_ctr);
break;
case 32:
pm_ctrl &= ~CBE_PM_16BIT_CTR(phys_ctr);
break;
}
cbe_write_pm(cpu, pm_control, pm_ctrl);
}
}
EXPORT_SYMBOL_GPL(cbe_set_ctr_size);
/*
* Enable/disable the entire performance monitoring unit.
* When we enable the PMU, all pending writes to counters get committed.
*/
void cbe_enable_pm(u32 cpu)
{
struct cbe_pmd_shadow_regs *shadow_regs;
u32 pm_ctrl;
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu);
shadow_regs->counter_value_in_latch = 0;
pm_ctrl = cbe_read_pm(cpu, pm_control) | CBE_PM_ENABLE_PERF_MON;
cbe_write_pm(cpu, pm_control, pm_ctrl);
}
EXPORT_SYMBOL_GPL(cbe_enable_pm);
void cbe_disable_pm(u32 cpu)
{
u32 pm_ctrl;
pm_ctrl = cbe_read_pm(cpu, pm_control) & ~CBE_PM_ENABLE_PERF_MON;
cbe_write_pm(cpu, pm_control, pm_ctrl);
}
EXPORT_SYMBOL_GPL(cbe_disable_pm);
/*
* Reading from the trace_buffer.
* The trace buffer is two 64-bit registers. Reading from
* the second half automatically increments the trace_address.
*/
void cbe_read_trace_buffer(u32 cpu, u64 *buf)
{
struct cbe_pmd_regs __iomem *pmd_regs = cbe_get_cpu_pmd_regs(cpu);
*buf++ = in_be64(&pmd_regs->trace_buffer_0_63);
*buf++ = in_be64(&pmd_regs->trace_buffer_64_127);
}
EXPORT_SYMBOL_GPL(cbe_read_trace_buffer);
/*
* Enabling/disabling interrupts for the entire performance monitoring unit.
*/
u32 cbe_get_and_clear_pm_interrupts(u32 cpu)
{
/* Reading pm_status clears the interrupt bits. */
return cbe_read_pm(cpu, pm_status);
}
EXPORT_SYMBOL_GPL(cbe_get_and_clear_pm_interrupts);
void cbe_enable_pm_interrupts(u32 cpu, u32 thread, u32 mask)
{
/* Set which node and thread will handle the next interrupt. */
iic_set_interrupt_routing(cpu, thread, 0);
/* Enable the interrupt bits in the pm_status register. */
if (mask)
cbe_write_pm(cpu, pm_status, mask);
}
EXPORT_SYMBOL_GPL(cbe_enable_pm_interrupts);
void cbe_disable_pm_interrupts(u32 cpu)
{
cbe_get_and_clear_pm_interrupts(cpu);
cbe_write_pm(cpu, pm_status, 0);
}
EXPORT_SYMBOL_GPL(cbe_disable_pm_interrupts);
[POWERPC] cell: Add oprofile support Add PPU event-based and cycle-based profiling support to Oprofile for Cell. Oprofile is expected to collect data on all CPUs simultaneously. However, there is one set of performance counters per node. There are two hardware threads or virtual CPUs on each node. Hence, OProfile must multiplex in time the performance counter collection on the two virtual CPUs. The multiplexing of the performance counters is done by a virtual counter routine. Initially, the counters are configured to collect data on the even CPUs in the system, one CPU per node. In order to capture the PC for the virtual CPU when the performance counter interrupt occurs (the specified number of events between samples has occurred), the even processors are configured to handle the performance counter interrupts for their node. The virtual counter routine is called via a kernel timer after the virtual sample time. The routine stops the counters, saves the current counts, loads the last counts for the other virtual CPU on the node, sets interrupts to be handled by the other virtual CPU and restarts the counters, the virtual timer routine is scheduled to run again. The virtual sample time is kept relatively small to make sure sampling occurs on both CPUs on the node with a relatively small granularity. Whenever the counters overflow, the performance counter interrupt is called to collect the PC for the CPU where data is being collected. The oprofile driver relies on a firmware RTAS call to setup the debug bus to route the desired signals to the performance counter hardware to be counted. The RTAS call must set the routing registers appropriately in each of the islands to pass the signals down the debug bus as well as routing the signals from a particular island onto the bus. There is a second firmware RTAS call to reset the debug bus to the non pass thru state when the counters are not in use. Signed-off-by: Carl Love <carll@us.ibm.com> Signed-off-by: Maynard Johnson <mpjohn@us.ibm.com> Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-11-20 10:45:16 -07:00
static irqreturn_t cbe_pm_irq(int irq, void *dev_id)
{
[POWERPC] cell: Add oprofile support Add PPU event-based and cycle-based profiling support to Oprofile for Cell. Oprofile is expected to collect data on all CPUs simultaneously. However, there is one set of performance counters per node. There are two hardware threads or virtual CPUs on each node. Hence, OProfile must multiplex in time the performance counter collection on the two virtual CPUs. The multiplexing of the performance counters is done by a virtual counter routine. Initially, the counters are configured to collect data on the even CPUs in the system, one CPU per node. In order to capture the PC for the virtual CPU when the performance counter interrupt occurs (the specified number of events between samples has occurred), the even processors are configured to handle the performance counter interrupts for their node. The virtual counter routine is called via a kernel timer after the virtual sample time. The routine stops the counters, saves the current counts, loads the last counts for the other virtual CPU on the node, sets interrupts to be handled by the other virtual CPU and restarts the counters, the virtual timer routine is scheduled to run again. The virtual sample time is kept relatively small to make sure sampling occurs on both CPUs on the node with a relatively small granularity. Whenever the counters overflow, the performance counter interrupt is called to collect the PC for the CPU where data is being collected. The oprofile driver relies on a firmware RTAS call to setup the debug bus to route the desired signals to the performance counter hardware to be counted. The RTAS call must set the routing registers appropriately in each of the islands to pass the signals down the debug bus as well as routing the signals from a particular island onto the bus. There is a second firmware RTAS call to reset the debug bus to the non pass thru state when the counters are not in use. Signed-off-by: Carl Love <carll@us.ibm.com> Signed-off-by: Maynard Johnson <mpjohn@us.ibm.com> Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-11-20 10:45:16 -07:00
perf_irq(get_irq_regs());
return IRQ_HANDLED;
}
static int __init cbe_init_pm_irq(void)
{
unsigned int irq;
int rc, node;
if (!machine_is(cell))
return 0;
for_each_node(node) {
irq = irq_create_mapping(NULL, IIC_IRQ_IOEX_PMI |
(node << IIC_IRQ_NODE_SHIFT));
if (irq == NO_IRQ) {
printk("ERROR: Unable to allocate irq for node %d\n",
node);
return -EINVAL;
}
rc = request_irq(irq, cbe_pm_irq,
IRQF_DISABLED, "cbe-pmu-0", NULL);
if (rc) {
printk("ERROR: Request for irq on node %d failed\n",
node);
return rc;
}
}
return 0;
}
arch_initcall(cbe_init_pm_irq);
[POWERPC] cell: Add oprofile support Add PPU event-based and cycle-based profiling support to Oprofile for Cell. Oprofile is expected to collect data on all CPUs simultaneously. However, there is one set of performance counters per node. There are two hardware threads or virtual CPUs on each node. Hence, OProfile must multiplex in time the performance counter collection on the two virtual CPUs. The multiplexing of the performance counters is done by a virtual counter routine. Initially, the counters are configured to collect data on the even CPUs in the system, one CPU per node. In order to capture the PC for the virtual CPU when the performance counter interrupt occurs (the specified number of events between samples has occurred), the even processors are configured to handle the performance counter interrupts for their node. The virtual counter routine is called via a kernel timer after the virtual sample time. The routine stops the counters, saves the current counts, loads the last counts for the other virtual CPU on the node, sets interrupts to be handled by the other virtual CPU and restarts the counters, the virtual timer routine is scheduled to run again. The virtual sample time is kept relatively small to make sure sampling occurs on both CPUs on the node with a relatively small granularity. Whenever the counters overflow, the performance counter interrupt is called to collect the PC for the CPU where data is being collected. The oprofile driver relies on a firmware RTAS call to setup the debug bus to route the desired signals to the performance counter hardware to be counted. The RTAS call must set the routing registers appropriately in each of the islands to pass the signals down the debug bus as well as routing the signals from a particular island onto the bus. There is a second firmware RTAS call to reset the debug bus to the non pass thru state when the counters are not in use. Signed-off-by: Carl Love <carll@us.ibm.com> Signed-off-by: Maynard Johnson <mpjohn@us.ibm.com> Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-11-20 10:45:16 -07:00
void cbe_sync_irq(int node)
{
unsigned int irq;
irq = irq_find_mapping(NULL,
IIC_IRQ_IOEX_PMI
| (node << IIC_IRQ_NODE_SHIFT));
if (irq == NO_IRQ) {
printk(KERN_WARNING "ERROR, unable to get existing irq %d " \
"for node %d\n", irq, node);
return;
}
synchronize_irq(irq);
}
EXPORT_SYMBOL_GPL(cbe_sync_irq);