MIPS: Loongson-3: Add oprofile support
authorHuacai Chen <chenhc@lemote.com>
Tue, 4 Nov 2014 06:15:07 +0000 (14:15 +0800)
committerRalf Baechle <ralf@linux-mips.org>
Mon, 24 Nov 2014 06:45:01 +0000 (07:45 +0100)
Loongson-3 has two groups of performance counters, they are 4 sub-
registers of CP0's REG25. This patch add oprofile support.

REG25, sel 0: Perf Control of group 0;
REG25, sel 1: Perf Counter of group 0;
REG25, sel 2: Perf Control of group 1;
REG25, sel 3: Perf Counter of group 1.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
Cc: John Crispin <john@phrozen.org>
Cc: Steven J. Hill <Steven.Hill@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Fuxin Zhang <zhangfx@lemote.com>
Cc: Zhangjin Wu <wuzhangjin@gmail.com>
Patchwork: https://patchwork.linux-mips.org/patch/8328/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
arch/mips/oprofile/Makefile
arch/mips/oprofile/common.c
arch/mips/oprofile/op_model_loongson3.c [new file with mode: 0644]

index 9c0a6782c091dcc903f3728b21efc2d712c9c67e..070afdb297df161077fc6289632545fb12e6e995 100644 (file)
@@ -14,3 +14,4 @@ oprofile-$(CONFIG_CPU_R10000)         += op_model_mipsxx.o
 oprofile-$(CONFIG_CPU_SB1)             += op_model_mipsxx.o
 oprofile-$(CONFIG_CPU_XLR)             += op_model_mipsxx.o
 oprofile-$(CONFIG_CPU_LOONGSON2)       += op_model_loongson2.o
+oprofile-$(CONFIG_CPU_LOONGSON3)       += op_model_loongson3.o
index e7473244947831bf7c46a54db40d33e886642929..feb987981f55c7891c41f81a5827cab1f6c6b0db 100644 (file)
@@ -18,6 +18,7 @@
 
 extern struct op_mips_model op_model_mipsxx_ops __weak;
 extern struct op_mips_model op_model_loongson2_ops __weak;
+extern struct op_mips_model op_model_loongson3_ops __weak;
 
 static struct op_mips_model *model;
 
@@ -104,6 +105,9 @@ int __init oprofile_arch_init(struct oprofile_operations *ops)
        case CPU_LOONGSON2:
                lmodel = &op_model_loongson2_ops;
                break;
+       case CPU_LOONGSON3:
+               lmodel = &op_model_loongson3_ops;
+               break;
        };
 
        if (!lmodel)
diff --git a/arch/mips/oprofile/op_model_loongson3.c b/arch/mips/oprofile/op_model_loongson3.c
new file mode 100644 (file)
index 0000000..8bcf7fc
--- /dev/null
@@ -0,0 +1,220 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ */
+#include <linux/init.h>
+#include <linux/cpu.h>
+#include <linux/smp.h>
+#include <linux/proc_fs.h>
+#include <linux/oprofile.h>
+#include <linux/spinlock.h>
+#include <linux/interrupt.h>
+#include <asm/uaccess.h>
+#include <irq.h>
+#include <loongson.h>
+#include "op_impl.h"
+
+#define LOONGSON3_PERFCNT_OVERFLOW     (1ULL << 63)
+
+#define LOONGSON3_PERFCTRL_EXL         (1UL << 0)
+#define LOONGSON3_PERFCTRL_KERNEL      (1UL << 1)
+#define LOONGSON3_PERFCTRL_SUPERVISOR  (1UL << 2)
+#define LOONGSON3_PERFCTRL_USER                (1UL << 3)
+#define LOONGSON3_PERFCTRL_ENABLE      (1UL << 4)
+#define LOONGSON3_PERFCTRL_W           (1UL << 30)
+#define LOONGSON3_PERFCTRL_M           (1UL << 31)
+#define LOONGSON3_PERFCTRL_EVENT(idx, event) \
+       (((event) & (idx ? 0x0f : 0x3f)) << 5)
+
+/* Loongson-3 PerfCount performance counter1 register */
+#define read_c0_perflo1() __read_64bit_c0_register($25, 0)
+#define write_c0_perflo1(val) __write_64bit_c0_register($25, 0, val)
+#define read_c0_perfhi1() __read_64bit_c0_register($25, 1)
+#define write_c0_perfhi1(val) __write_64bit_c0_register($25, 1, val)
+
+/* Loongson-3 PerfCount performance counter2 register */
+#define read_c0_perflo2() __read_64bit_c0_register($25, 2)
+#define write_c0_perflo2(val) __write_64bit_c0_register($25, 2, val)
+#define read_c0_perfhi2() __read_64bit_c0_register($25, 3)
+#define write_c0_perfhi2(val) __write_64bit_c0_register($25, 3, val)
+
+static int (*save_perf_irq)(void);
+
+static struct loongson3_register_config {
+       unsigned int control1;
+       unsigned int control2;
+       unsigned long long reset_counter1;
+       unsigned long long reset_counter2;
+       int ctr1_enable, ctr2_enable;
+} reg;
+
+static void reset_counters(void *arg)
+{
+       write_c0_perfhi1(0);
+       write_c0_perfhi2(0);
+       write_c0_perflo1(0xc0000000);
+       write_c0_perflo2(0x40000000);
+}
+
+/* Compute all of the registers in preparation for enabling profiling. */
+static void loongson3_reg_setup(struct op_counter_config *ctr)
+{
+       unsigned int control1 = 0;
+       unsigned int control2 = 0;
+
+       reg.reset_counter1 = 0;
+       reg.reset_counter2 = 0;
+       /* Compute the performance counter control word. */
+       /* For now count kernel and user mode */
+       if (ctr[0].enabled) {
+               control1 |= LOONGSON3_PERFCTRL_EVENT(0, ctr[0].event) |
+                                       LOONGSON3_PERFCTRL_ENABLE;
+               if (ctr[0].kernel)
+                       control1 |= LOONGSON3_PERFCTRL_KERNEL;
+               if (ctr[0].user)
+                       control1 |= LOONGSON3_PERFCTRL_USER;
+               reg.reset_counter1 = 0x8000000000000000ULL - ctr[0].count;
+       }
+
+       if (ctr[1].enabled) {
+               control2 |= LOONGSON3_PERFCTRL_EVENT(1, ctr[1].event) |
+                                       LOONGSON3_PERFCTRL_ENABLE;
+               if (ctr[1].kernel)
+                       control2 |= LOONGSON3_PERFCTRL_KERNEL;
+               if (ctr[1].user)
+                       control2 |= LOONGSON3_PERFCTRL_USER;
+               reg.reset_counter2 = 0x8000000000000000ULL - ctr[1].count;
+       }
+
+       if (ctr[0].enabled)
+               control1 |= LOONGSON3_PERFCTRL_EXL;
+       if (ctr[1].enabled)
+               control2 |= LOONGSON3_PERFCTRL_EXL;
+
+       reg.control1 = control1;
+       reg.control2 = control2;
+       reg.ctr1_enable = ctr[0].enabled;
+       reg.ctr2_enable = ctr[1].enabled;
+}
+
+/* Program all of the registers in preparation for enabling profiling. */
+static void loongson3_cpu_setup(void *args)
+{
+       uint64_t perfcount1, perfcount2;
+
+       perfcount1 = reg.reset_counter1;
+       perfcount2 = reg.reset_counter2;
+       write_c0_perfhi1(perfcount1);
+       write_c0_perfhi2(perfcount2);
+}
+
+static void loongson3_cpu_start(void *args)
+{
+       /* Start all counters on current CPU */
+       reg.control1 |= (LOONGSON3_PERFCTRL_W|LOONGSON3_PERFCTRL_M);
+       reg.control2 |= (LOONGSON3_PERFCTRL_W|LOONGSON3_PERFCTRL_M);
+
+       if (reg.ctr1_enable)
+               write_c0_perflo1(reg.control1);
+       if (reg.ctr2_enable)
+               write_c0_perflo2(reg.control2);
+}
+
+static void loongson3_cpu_stop(void *args)
+{
+       /* Stop all counters on current CPU */
+       write_c0_perflo1(0xc0000000);
+       write_c0_perflo2(0x40000000);
+       memset(&reg, 0, sizeof(reg));
+}
+
+static int loongson3_perfcount_handler(void)
+{
+       unsigned long flags;
+       uint64_t counter1, counter2;
+       uint32_t cause, handled = IRQ_NONE;
+       struct pt_regs *regs = get_irq_regs();
+
+       cause = read_c0_cause();
+       if (!(cause & CAUSEF_PCI))
+               return handled;
+
+       counter1 = read_c0_perfhi1();
+       counter2 = read_c0_perfhi2();
+
+       local_irq_save(flags);
+
+       if (counter1 & LOONGSON3_PERFCNT_OVERFLOW) {
+               if (reg.ctr1_enable)
+                       oprofile_add_sample(regs, 0);
+               counter1 = reg.reset_counter1;
+       }
+       if (counter2 & LOONGSON3_PERFCNT_OVERFLOW) {
+               if (reg.ctr2_enable)
+                       oprofile_add_sample(regs, 1);
+               counter2 = reg.reset_counter2;
+       }
+
+       local_irq_restore(flags);
+
+       write_c0_perfhi1(counter1);
+       write_c0_perfhi2(counter2);
+
+       if (!(cause & CAUSEF_TI))
+               handled = IRQ_HANDLED;
+
+       return handled;
+}
+
+static int loongson3_cpu_callback(struct notifier_block *nfb,
+       unsigned long action, void *hcpu)
+{
+       switch (action) {
+       case CPU_STARTING:
+       case CPU_STARTING_FROZEN:
+               write_c0_perflo1(reg.control1);
+               write_c0_perflo2(reg.control2);
+               break;
+       case CPU_DYING:
+       case CPU_DYING_FROZEN:
+               write_c0_perflo1(0xc0000000);
+               write_c0_perflo2(0x40000000);
+               break;
+       }
+
+       return NOTIFY_OK;
+}
+
+static struct notifier_block loongson3_notifier_block = {
+       .notifier_call = loongson3_cpu_callback
+};
+
+static int __init loongson3_init(void)
+{
+       on_each_cpu(reset_counters, NULL, 1);
+       register_hotcpu_notifier(&loongson3_notifier_block);
+       save_perf_irq = perf_irq;
+       perf_irq = loongson3_perfcount_handler;
+
+       return 0;
+}
+
+static void loongson3_exit(void)
+{
+       on_each_cpu(reset_counters, NULL, 1);
+       unregister_hotcpu_notifier(&loongson3_notifier_block);
+       perf_irq = save_perf_irq;
+}
+
+struct op_mips_model op_model_loongson3_ops = {
+       .reg_setup      = loongson3_reg_setup,
+       .cpu_setup      = loongson3_cpu_setup,
+       .init           = loongson3_init,
+       .exit           = loongson3_exit,
+       .cpu_start      = loongson3_cpu_start,
+       .cpu_stop       = loongson3_cpu_stop,
+       .cpu_type       = "mips/loongson3",
+       .num_counters   = 2
+};