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>
28 #include <asm/irq_regs.h>
29 #include <asm/machdep.h>
33 #include <asm/cell-regs.h>
35 #include "interrupt.h"
38 * When writing to write-only mmio addresses, save a shadow copy. All of the
39 * registers are 32-bit, but stored in the upper-half of a 64-bit field in
43 #define WRITE_WO_MMIO(reg, x) \
46 struct cbe_pmd_regs __iomem *pmd_regs; \
47 struct cbe_pmd_shadow_regs *shadow_regs; \
48 pmd_regs = cbe_get_cpu_pmd_regs(cpu); \
49 shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu); \
50 out_be64(&(pmd_regs->reg), (((u64)_x) << 32)); \
51 shadow_regs->reg = _x; \
54 #define READ_SHADOW_REG(val, reg) \
56 struct cbe_pmd_shadow_regs *shadow_regs; \
57 shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu); \
58 (val) = shadow_regs->reg; \
61 #define READ_MMIO_UPPER32(val, reg) \
63 struct cbe_pmd_regs __iomem *pmd_regs; \
64 pmd_regs = cbe_get_cpu_pmd_regs(cpu); \
65 (val) = (u32)(in_be64(&pmd_regs->reg) >> 32); \
69 * Physical counter registers.
70 * Each physical counter can act as one 32-bit counter or two 16-bit counters.
73 u32 cbe_read_phys_ctr(u32 cpu, u32 phys_ctr)
75 u32 val_in_latch, val = 0;
77 if (phys_ctr < NR_PHYS_CTRS) {
78 READ_SHADOW_REG(val_in_latch, counter_value_in_latch);
80 /* Read the latch or the actual counter, whichever is newer. */
81 if (val_in_latch & (1 << phys_ctr)) {
82 READ_SHADOW_REG(val, pm_ctr[phys_ctr]);
84 READ_MMIO_UPPER32(val, pm_ctr[phys_ctr]);
90 EXPORT_SYMBOL_GPL(cbe_read_phys_ctr);
92 void cbe_write_phys_ctr(u32 cpu, u32 phys_ctr, u32 val)
94 struct cbe_pmd_shadow_regs *shadow_regs;
97 if (phys_ctr < NR_PHYS_CTRS) {
98 /* Writing to a counter only writes to a hardware latch.
99 * The new value is not propagated to the actual counter
100 * until the performance monitor is enabled.
102 WRITE_WO_MMIO(pm_ctr[phys_ctr], val);
104 pm_ctrl = cbe_read_pm(cpu, pm_control);
105 if (pm_ctrl & CBE_PM_ENABLE_PERF_MON) {
106 /* The counters are already active, so we need to
107 * rewrite the pm_control register to "re-enable"
110 cbe_write_pm(cpu, pm_control, pm_ctrl);
112 shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu);
113 shadow_regs->counter_value_in_latch |= (1 << phys_ctr);
117 EXPORT_SYMBOL_GPL(cbe_write_phys_ctr);
120 * "Logical" counter registers.
121 * These will read/write 16-bits or 32-bits depending on the
122 * current size of the counter. Counters 4 - 7 are always 16-bit.
125 u32 cbe_read_ctr(u32 cpu, u32 ctr)
128 u32 phys_ctr = ctr & (NR_PHYS_CTRS - 1);
130 val = cbe_read_phys_ctr(cpu, phys_ctr);
132 if (cbe_get_ctr_size(cpu, phys_ctr) == 16)
133 val = (ctr < NR_PHYS_CTRS) ? (val >> 16) : (val & 0xffff);
137 EXPORT_SYMBOL_GPL(cbe_read_ctr);
139 void cbe_write_ctr(u32 cpu, u32 ctr, u32 val)
144 phys_ctr = ctr & (NR_PHYS_CTRS - 1);
146 if (cbe_get_ctr_size(cpu, phys_ctr) == 16) {
147 phys_val = cbe_read_phys_ctr(cpu, phys_ctr);
149 if (ctr < NR_PHYS_CTRS)
150 val = (val << 16) | (phys_val & 0xffff);
152 val = (val & 0xffff) | (phys_val & 0xffff0000);
155 cbe_write_phys_ctr(cpu, phys_ctr, val);
157 EXPORT_SYMBOL_GPL(cbe_write_ctr);
160 * Counter-control registers.
161 * Each "logical" counter has a corresponding control register.
164 u32 cbe_read_pm07_control(u32 cpu, u32 ctr)
166 u32 pm07_control = 0;
169 READ_SHADOW_REG(pm07_control, pm07_control[ctr]);
173 EXPORT_SYMBOL_GPL(cbe_read_pm07_control);
175 void cbe_write_pm07_control(u32 cpu, u32 ctr, u32 val)
178 WRITE_WO_MMIO(pm07_control[ctr], val);
180 EXPORT_SYMBOL_GPL(cbe_write_pm07_control);
183 * Other PMU control registers. Most of these are write-only.
186 u32 cbe_read_pm(u32 cpu, enum pm_reg_name reg)
192 READ_SHADOW_REG(val, group_control);
195 case debug_bus_control:
196 READ_SHADOW_REG(val, debug_bus_control);
200 READ_MMIO_UPPER32(val, trace_address);
204 READ_SHADOW_REG(val, ext_tr_timer);
208 READ_MMIO_UPPER32(val, pm_status);
212 READ_SHADOW_REG(val, pm_control);
216 READ_MMIO_UPPER32(val, pm_interval);
220 READ_SHADOW_REG(val, pm_start_stop);
226 EXPORT_SYMBOL_GPL(cbe_read_pm);
228 void cbe_write_pm(u32 cpu, enum pm_reg_name reg, u32 val)
232 WRITE_WO_MMIO(group_control, val);
235 case debug_bus_control:
236 WRITE_WO_MMIO(debug_bus_control, val);
240 WRITE_WO_MMIO(trace_address, val);
244 WRITE_WO_MMIO(ext_tr_timer, val);
248 WRITE_WO_MMIO(pm_status, val);
252 WRITE_WO_MMIO(pm_control, val);
256 WRITE_WO_MMIO(pm_interval, val);
260 WRITE_WO_MMIO(pm_start_stop, val);
264 EXPORT_SYMBOL_GPL(cbe_write_pm);
267 * Get/set the size of a physical counter to either 16 or 32 bits.
270 u32 cbe_get_ctr_size(u32 cpu, u32 phys_ctr)
272 u32 pm_ctrl, size = 0;
274 if (phys_ctr < NR_PHYS_CTRS) {
275 pm_ctrl = cbe_read_pm(cpu, pm_control);
276 size = (pm_ctrl & CBE_PM_16BIT_CTR(phys_ctr)) ? 16 : 32;
281 EXPORT_SYMBOL_GPL(cbe_get_ctr_size);
283 void cbe_set_ctr_size(u32 cpu, u32 phys_ctr, u32 ctr_size)
287 if (phys_ctr < NR_PHYS_CTRS) {
288 pm_ctrl = cbe_read_pm(cpu, pm_control);
291 pm_ctrl |= CBE_PM_16BIT_CTR(phys_ctr);
295 pm_ctrl &= ~CBE_PM_16BIT_CTR(phys_ctr);
298 cbe_write_pm(cpu, pm_control, pm_ctrl);
301 EXPORT_SYMBOL_GPL(cbe_set_ctr_size);
304 * Enable/disable the entire performance monitoring unit.
305 * When we enable the PMU, all pending writes to counters get committed.
308 void cbe_enable_pm(u32 cpu)
310 struct cbe_pmd_shadow_regs *shadow_regs;
313 shadow_regs = cbe_get_cpu_pmd_shadow_regs(cpu);
314 shadow_regs->counter_value_in_latch = 0;
316 pm_ctrl = cbe_read_pm(cpu, pm_control) | CBE_PM_ENABLE_PERF_MON;
317 cbe_write_pm(cpu, pm_control, pm_ctrl);
319 EXPORT_SYMBOL_GPL(cbe_enable_pm);
321 void cbe_disable_pm(u32 cpu)
324 pm_ctrl = cbe_read_pm(cpu, pm_control) & ~CBE_PM_ENABLE_PERF_MON;
325 cbe_write_pm(cpu, pm_control, pm_ctrl);
327 EXPORT_SYMBOL_GPL(cbe_disable_pm);
330 * Reading from the trace_buffer.
331 * The trace buffer is two 64-bit registers. Reading from
332 * the second half automatically increments the trace_address.
335 void cbe_read_trace_buffer(u32 cpu, u64 *buf)
337 struct cbe_pmd_regs __iomem *pmd_regs = cbe_get_cpu_pmd_regs(cpu);
339 *buf++ = in_be64(&pmd_regs->trace_buffer_0_63);
340 *buf++ = in_be64(&pmd_regs->trace_buffer_64_127);
342 EXPORT_SYMBOL_GPL(cbe_read_trace_buffer);
345 * Enabling/disabling interrupts for the entire performance monitoring unit.
348 u32 cbe_get_and_clear_pm_interrupts(u32 cpu)
350 /* Reading pm_status clears the interrupt bits. */
351 return cbe_read_pm(cpu, pm_status);
353 EXPORT_SYMBOL_GPL(cbe_get_and_clear_pm_interrupts);
355 void cbe_enable_pm_interrupts(u32 cpu, u32 thread, u32 mask)
357 /* Set which node and thread will handle the next interrupt. */
358 iic_set_interrupt_routing(cpu, thread, 0);
360 /* Enable the interrupt bits in the pm_status register. */
362 cbe_write_pm(cpu, pm_status, mask);
364 EXPORT_SYMBOL_GPL(cbe_enable_pm_interrupts);
366 void cbe_disable_pm_interrupts(u32 cpu)
368 cbe_get_and_clear_pm_interrupts(cpu);
369 cbe_write_pm(cpu, pm_status, 0);
371 EXPORT_SYMBOL_GPL(cbe_disable_pm_interrupts);
373 static irqreturn_t cbe_pm_irq(int irq, void *dev_id)
375 perf_irq(get_irq_regs());
379 static int __init cbe_init_pm_irq(void)
384 for_each_node(node) {
385 irq = irq_create_mapping(NULL, IIC_IRQ_IOEX_PMI |
386 (node << IIC_IRQ_NODE_SHIFT));
388 printk("ERROR: Unable to allocate irq for node %d\n",
393 rc = request_irq(irq, cbe_pm_irq,
394 IRQF_DISABLED, "cbe-pmu-0", NULL);
396 printk("ERROR: Request for irq on node %d failed\n",
404 machine_arch_initcall(cell, cbe_init_pm_irq);
406 void cbe_sync_irq(int node)
410 irq = irq_find_mapping(NULL,
412 | (node << IIC_IRQ_NODE_SHIFT));
415 printk(KERN_WARNING "ERROR, unable to get existing irq %d " \
416 "for node %d\n", irq, node);
420 synchronize_irq(irq);
422 EXPORT_SYMBOL_GPL(cbe_sync_irq);