2
* Cell Broadband Engine Performance Monitor
4
* (C) Copyright IBM Corporation 2001,2006
7
* David Erb (djerb@us.ibm.com)
8
* Kevin Corry (kevcorry@us.ibm.com)
10
* This program is free software; you can redistribute it and/or modify
11
* it under the terms of the GNU General Public License as published by
12
* the Free Software Foundation; either version 2, or (at your option)
15
* This program is distributed in the hope that it will be useful,
16
* but WITHOUT ANY WARRANTY; without even the implied warranty of
17
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18
* GNU General Public License for more details.
20
* You should have received a copy of the GNU General Public License
21
* along with this program; if not, write to the Free Software
22
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25
#include <linux/interrupt.h>
26
#include <linux/types.h>
27
#include <linux/export.h>
29
#include <asm/irq_regs.h>
30
#include <asm/machdep.h>
34
#include <asm/cell-regs.h>
36
#include "interrupt.h"
39
* When writing to write-only mmio addresses, save a shadow copy. All of the
40
* registers are 32-bit, but stored in the upper-half of a 64-bit field in
44
#define WRITE_WO_MMIO(reg, x) \
47
struct cbe_pmd_regs __iomem *pmd_regs; \
48
struct cbe_pmd_shadow_regs *shadow_regs; \
49
pmd_regs = cbe_get_cpu_pmd_regs(cpu); \
50
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu); \
51
out_be64(&(pmd_regs->reg), (((u64)_x) << 32)); \
52
shadow_regs->reg = _x; \
55
#define READ_SHADOW_REG(val, reg) \
57
struct cbe_pmd_shadow_regs *shadow_regs; \
58
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu); \
59
(val) = shadow_regs->reg; \
62
#define READ_MMIO_UPPER32(val, reg) \
64
struct cbe_pmd_regs __iomem *pmd_regs; \
65
pmd_regs = cbe_get_cpu_pmd_regs(cpu); \
66
(val) = (u32)(in_be64(&pmd_regs->reg) >> 32); \
70
* Physical counter registers.
71
* Each physical counter can act as one 32-bit counter or two 16-bit counters.
74
u32 cbe_read_phys_ctr(u32 cpu, u32 phys_ctr)
76
u32 val_in_latch, val = 0;
78
if (phys_ctr < NR_PHYS_CTRS) {
79
READ_SHADOW_REG(val_in_latch, counter_value_in_latch);
81
/* Read the latch or the actual counter, whichever is newer. */
82
if (val_in_latch & (1 << phys_ctr)) {
83
READ_SHADOW_REG(val, pm_ctr[phys_ctr]);
85
READ_MMIO_UPPER32(val, pm_ctr[phys_ctr]);
91
EXPORT_SYMBOL_GPL(cbe_read_phys_ctr);
93
void cbe_write_phys_ctr(u32 cpu, u32 phys_ctr, u32 val)
95
struct cbe_pmd_shadow_regs *shadow_regs;
98
if (phys_ctr < NR_PHYS_CTRS) {
99
/* Writing to a counter only writes to a hardware latch.
100
* The new value is not propagated to the actual counter
101
* until the performance monitor is enabled.
103
WRITE_WO_MMIO(pm_ctr[phys_ctr], val);
105
pm_ctrl = cbe_read_pm(cpu, pm_control);
106
if (pm_ctrl & CBE_PM_ENABLE_PERF_MON) {
107
/* The counters are already active, so we need to
108
* rewrite the pm_control register to "re-enable"
111
cbe_write_pm(cpu, pm_control, pm_ctrl);
113
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu);
114
shadow_regs->counter_value_in_latch |= (1 << phys_ctr);
118
EXPORT_SYMBOL_GPL(cbe_write_phys_ctr);
121
* "Logical" counter registers.
122
* These will read/write 16-bits or 32-bits depending on the
123
* current size of the counter. Counters 4 - 7 are always 16-bit.
126
u32 cbe_read_ctr(u32 cpu, u32 ctr)
129
u32 phys_ctr = ctr & (NR_PHYS_CTRS - 1);
131
val = cbe_read_phys_ctr(cpu, phys_ctr);
133
if (cbe_get_ctr_size(cpu, phys_ctr) == 16)
134
val = (ctr < NR_PHYS_CTRS) ? (val >> 16) : (val & 0xffff);
138
EXPORT_SYMBOL_GPL(cbe_read_ctr);
140
void cbe_write_ctr(u32 cpu, u32 ctr, u32 val)
145
phys_ctr = ctr & (NR_PHYS_CTRS - 1);
147
if (cbe_get_ctr_size(cpu, phys_ctr) == 16) {
148
phys_val = cbe_read_phys_ctr(cpu, phys_ctr);
150
if (ctr < NR_PHYS_CTRS)
151
val = (val << 16) | (phys_val & 0xffff);
153
val = (val & 0xffff) | (phys_val & 0xffff0000);
156
cbe_write_phys_ctr(cpu, phys_ctr, val);
158
EXPORT_SYMBOL_GPL(cbe_write_ctr);
161
* Counter-control registers.
162
* Each "logical" counter has a corresponding control register.
165
u32 cbe_read_pm07_control(u32 cpu, u32 ctr)
167
u32 pm07_control = 0;
170
READ_SHADOW_REG(pm07_control, pm07_control[ctr]);
174
EXPORT_SYMBOL_GPL(cbe_read_pm07_control);
176
void cbe_write_pm07_control(u32 cpu, u32 ctr, u32 val)
179
WRITE_WO_MMIO(pm07_control[ctr], val);
181
EXPORT_SYMBOL_GPL(cbe_write_pm07_control);
184
* Other PMU control registers. Most of these are write-only.
187
u32 cbe_read_pm(u32 cpu, enum pm_reg_name reg)
193
READ_SHADOW_REG(val, group_control);
196
case debug_bus_control:
197
READ_SHADOW_REG(val, debug_bus_control);
201
READ_MMIO_UPPER32(val, trace_address);
205
READ_SHADOW_REG(val, ext_tr_timer);
209
READ_MMIO_UPPER32(val, pm_status);
213
READ_SHADOW_REG(val, pm_control);
217
READ_MMIO_UPPER32(val, pm_interval);
221
READ_SHADOW_REG(val, pm_start_stop);
227
EXPORT_SYMBOL_GPL(cbe_read_pm);
229
void cbe_write_pm(u32 cpu, enum pm_reg_name reg, u32 val)
233
WRITE_WO_MMIO(group_control, val);
236
case debug_bus_control:
237
WRITE_WO_MMIO(debug_bus_control, val);
241
WRITE_WO_MMIO(trace_address, val);
245
WRITE_WO_MMIO(ext_tr_timer, val);
249
WRITE_WO_MMIO(pm_status, val);
253
WRITE_WO_MMIO(pm_control, val);
257
WRITE_WO_MMIO(pm_interval, val);
261
WRITE_WO_MMIO(pm_start_stop, val);
265
EXPORT_SYMBOL_GPL(cbe_write_pm);
268
* Get/set the size of a physical counter to either 16 or 32 bits.
271
u32 cbe_get_ctr_size(u32 cpu, u32 phys_ctr)
273
u32 pm_ctrl, size = 0;
275
if (phys_ctr < NR_PHYS_CTRS) {
276
pm_ctrl = cbe_read_pm(cpu, pm_control);
277
size = (pm_ctrl & CBE_PM_16BIT_CTR(phys_ctr)) ? 16 : 32;
282
EXPORT_SYMBOL_GPL(cbe_get_ctr_size);
284
void cbe_set_ctr_size(u32 cpu, u32 phys_ctr, u32 ctr_size)
288
if (phys_ctr < NR_PHYS_CTRS) {
289
pm_ctrl = cbe_read_pm(cpu, pm_control);
292
pm_ctrl |= CBE_PM_16BIT_CTR(phys_ctr);
296
pm_ctrl &= ~CBE_PM_16BIT_CTR(phys_ctr);
299
cbe_write_pm(cpu, pm_control, pm_ctrl);
302
EXPORT_SYMBOL_GPL(cbe_set_ctr_size);
305
* Enable/disable the entire performance monitoring unit.
306
* When we enable the PMU, all pending writes to counters get committed.
309
void cbe_enable_pm(u32 cpu)
311
struct cbe_pmd_shadow_regs *shadow_regs;
314
shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu);
315
shadow_regs->counter_value_in_latch = 0;
317
pm_ctrl = cbe_read_pm(cpu, pm_control) | CBE_PM_ENABLE_PERF_MON;
318
cbe_write_pm(cpu, pm_control, pm_ctrl);
320
EXPORT_SYMBOL_GPL(cbe_enable_pm);
322
void cbe_disable_pm(u32 cpu)
325
pm_ctrl = cbe_read_pm(cpu, pm_control) & ~CBE_PM_ENABLE_PERF_MON;
326
cbe_write_pm(cpu, pm_control, pm_ctrl);
328
EXPORT_SYMBOL_GPL(cbe_disable_pm);
331
* Reading from the trace_buffer.
332
* The trace buffer is two 64-bit registers. Reading from
333
* the second half automatically increments the trace_address.
336
void cbe_read_trace_buffer(u32 cpu, u64 *buf)
338
struct cbe_pmd_regs __iomem *pmd_regs = cbe_get_cpu_pmd_regs(cpu);
340
*buf++ = in_be64(&pmd_regs->trace_buffer_0_63);
341
*buf++ = in_be64(&pmd_regs->trace_buffer_64_127);
343
EXPORT_SYMBOL_GPL(cbe_read_trace_buffer);
346
* Enabling/disabling interrupts for the entire performance monitoring unit.
349
u32 cbe_get_and_clear_pm_interrupts(u32 cpu)
351
/* Reading pm_status clears the interrupt bits. */
352
return cbe_read_pm(cpu, pm_status);
354
EXPORT_SYMBOL_GPL(cbe_get_and_clear_pm_interrupts);
356
void cbe_enable_pm_interrupts(u32 cpu, u32 thread, u32 mask)
358
/* Set which node and thread will handle the next interrupt. */
359
iic_set_interrupt_routing(cpu, thread, 0);
361
/* Enable the interrupt bits in the pm_status register. */
363
cbe_write_pm(cpu, pm_status, mask);
365
EXPORT_SYMBOL_GPL(cbe_enable_pm_interrupts);
367
void cbe_disable_pm_interrupts(u32 cpu)
369
cbe_get_and_clear_pm_interrupts(cpu);
370
cbe_write_pm(cpu, pm_status, 0);
372
EXPORT_SYMBOL_GPL(cbe_disable_pm_interrupts);
374
static irqreturn_t cbe_pm_irq(int irq, void *dev_id)
376
perf_irq(get_irq_regs());
380
static int __init cbe_init_pm_irq(void)
385
for_each_node(node) {
386
irq = irq_create_mapping(NULL, IIC_IRQ_IOEX_PMI |
387
(node << IIC_IRQ_NODE_SHIFT));
389
printk("ERROR: Unable to allocate irq for node %d\n",
394
rc = request_irq(irq, cbe_pm_irq,
395
0, "cbe-pmu-0", NULL);
397
printk("ERROR: Request for irq on node %d failed\n",
405
machine_arch_initcall(cell, cbe_init_pm_irq);
407
void cbe_sync_irq(int node)
411
irq = irq_find_mapping(NULL,
413
| (node << IIC_IRQ_NODE_SHIFT));
416
printk(KERN_WARNING "ERROR, unable to get existing irq %d " \
417
"for node %d\n", irq, node);
421
synchronize_irq(irq);
423
EXPORT_SYMBOL_GPL(cbe_sync_irq);