sh: perf events: Add preliminary support for SH-4A counters.

This adds in preliminary support for the SH-4A performance counters.
Presently only the first 2 counters are supported, as these are the ones
of the most interest to the perf tool and end users. Counter chaining is
not presently handled, so these are simply implemented as 32-bit
counters.

This also establishes a perf event support framework for other hardware
counters, which the existing SH-4 oprofile code will migrate over to as
the SH-4A support evolves.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

5 files changed, 576 insertions, 2 deletions

diff --git a/arch/sh/include/asm/perf_event.h b/arch/sh/include/asm/perf_event.h
index 11a302297ab..3d0c9f36d15 100644
--- a/arch/sh/include/asm/perf_event.h
+++ b/arch/sh/include/asm/perf_event.h
@@ -1,8 +1,35 @@
 #ifndef __ASM_SH_PERF_EVENT_H
 #define __ASM_SH_PERF_EVENT_H
 
-/* SH only supports software events through this interface. */
-static inline void set_perf_event_pending(void) {}
+struct hw_perf_event;
+
+#define MAX_HWEVENTS    2
+
+struct sh_pmu {
+        const char      *name;
+        unsigned int    num_events;
+        void            (*disable_all)(void);
+        void            (*enable_all)(void);
+        void            (*enable)(struct hw_perf_event *, int);
+        void            (*disable)(struct hw_perf_event *, int);
+        u64             (*read)(int);
+        int             (*event_map)(int);
+        unsigned int    max_events;
+        unsigned long   raw_event_mask;
+        const int       (*cache_events)[PERF_COUNT_HW_CACHE_MAX]
+                                       [PERF_COUNT_HW_CACHE_OP_MAX]
+                                       [PERF_COUNT_HW_CACHE_RESULT_MAX];
+};
+
+/* arch/sh/kernel/perf_event.c */
+extern int register_sh_pmu(struct sh_pmu *);
+extern int reserve_pmc_hardware(void);
+extern void release_pmc_hardware(void);
+
+static inline void set_perf_event_pending(void)
+{
+        /* Nothing to see here, move along. */
+}
 
 #define PERF_EVENT_INDEX_OFFSET 0
 
diff --git a/arch/sh/kernel/Makefile b/arch/sh/kernel/Makefile
index 097ae5ceb0e..0a67bafce42 100644
--- a/arch/sh/kernel/Makefile
+++ b/arch/sh/kernel/Makefile
@@ -39,6 +39,7 @@ obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o
 obj-$(CONFIG_DUMP_CODE)         += disassemble.o
 obj-$(CONFIG_HIBERNATION)       += swsusp.o
 obj-$(CONFIG_DWARF_UNWINDER)    += dwarf.o
+obj-$(CONFIG_PERF_EVENTS)       += perf_event.o
 
 obj-$(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST)     += localtimer.o
 
diff --git a/arch/sh/kernel/cpu/sh4a/Makefile b/arch/sh/kernel/cpu/sh4a/Makefile
index 490d5dc9e37..33bab477d2e 100644
--- a/arch/sh/kernel/cpu/sh4a/Makefile
+++ b/arch/sh/kernel/cpu/sh4a/Makefile
@@ -44,3 +44,4 @@ pinmux-$(CONFIG_CPU_SUBTYPE_SH7786)	:= pinmux-sh7786.o
 obj-y                           += $(clock-y)
 obj-$(CONFIG_SMP)               += $(smp-y)
 obj-$(CONFIG_GENERIC_GPIO)      += $(pinmux-y)
+obj-$(CONFIG_PERF_EVENTS)       += perf_event.o
diff --git a/arch/sh/kernel/cpu/sh4a/perf_event.c b/arch/sh/kernel/cpu/sh4a/perf_event.c
new file mode 100644
index 00000000000..d0938345799
--- /dev/null
+++ b/arch/sh/kernel/cpu/sh4a/perf_event.c
@@ -0,0 +1,231 @@
+/*
+ * Performance events support for SH-4A performance counters
+ *
+ *  Copyright (C) 2009  Paul Mundt
+ *
+ * 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/kernel.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/irq.h>
+#include <linux/perf_event.h>
+#include <asm/processor.h>
+
+#define PPC_CCBR(idx)   (0xff200800 + (sizeof(u32) * idx))
+#define PPC_PMCTR(idx)  (0xfc100000 + (sizeof(u32) * idx))
+
+#define CCBR_CIT_MASK   (0x7ff << 6)
+#define CCBR_DUC        (1 << 3)
+#define CCBR_CMDS       (1 << 1)
+#define CCBR_PPCE       (1 << 0)
+
+#define PPC_PMCAT       0xfc100080
+
+#define PMCAT_OVF3      (1 << 27)
+#define PMCAT_CNN3      (1 << 26)
+#define PMCAT_CLR3      (1 << 25)
+#define PMCAT_OVF2      (1 << 19)
+#define PMCAT_CLR2      (1 << 17)
+#define PMCAT_OVF1      (1 << 11)
+#define PMCAT_CNN1      (1 << 10)
+#define PMCAT_CLR1      (1 << 9)
+#define PMCAT_OVF0      (1 << 3)
+#define PMCAT_CLR0      (1 << 1)
+
+static struct sh_pmu sh4a_pmu;
+
+/*
+ * Special reserved bits used by hardware emulators, read values will
+ * vary, but writes must always be 0.
+ */
+#define PMCAT_EMU_CLR_MASK      ((1 << 24) | (1 << 16) | (1 << 8) | (1 << 0))
+
+static const int sh4a_general_events[] = {
+        [PERF_COUNT_HW_CPU_CYCLES]              = 0x0000,
+        [PERF_COUNT_HW_INSTRUCTIONS]            = 0x0202,
+        [PERF_COUNT_HW_CACHE_REFERENCES]        = 0x0029,       /* I-cache */
+        [PERF_COUNT_HW_CACHE_MISSES]            = 0x002a,       /* I-cache */
+        [PERF_COUNT_HW_BRANCH_INSTRUCTIONS]     = 0x0204,
+        [PERF_COUNT_HW_BRANCH_MISSES]           = -1,
+        [PERF_COUNT_HW_BUS_CYCLES]              = -1,
+};
+
+#define C(x)    PERF_COUNT_HW_CACHE_##x
+
+static const int sh4a_cache_events
+                        [PERF_COUNT_HW_CACHE_MAX]
+                        [PERF_COUNT_HW_CACHE_OP_MAX]
+                        [PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+        [ C(L1D) ] = {
+                [ C(OP_READ) ] = {
+                        [ C(RESULT_ACCESS) ] = 0x0031,
+                        [ C(RESULT_MISS)   ] = 0x0032,
+                },
+                [ C(OP_WRITE) ] = {
+                        [ C(RESULT_ACCESS) ] = 0x0039,
+                        [ C(RESULT_MISS)   ] = 0x003a,
+                },
+                [ C(OP_PREFETCH) ] = {
+                        [ C(RESULT_ACCESS) ] = 0,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+        },
+
+        [ C(L1I) ] = {
+                [ C(OP_READ) ] = {
+                        [ C(RESULT_ACCESS) ] = 0x0029,
+                        [ C(RESULT_MISS)   ] = 0x002a,
+                },
+                [ C(OP_WRITE) ] = {
+                        [ C(RESULT_ACCESS) ] = -1,
+                        [ C(RESULT_MISS)   ] = -1,
+                },
+                [ C(OP_PREFETCH) ] = {
+                        [ C(RESULT_ACCESS) ] = 0,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+        },
+
+        [ C(LL) ] = {
+                [ C(OP_READ) ] = {
+                        [ C(RESULT_ACCESS) ] = 0x0030,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+                [ C(OP_WRITE) ] = {
+                        [ C(RESULT_ACCESS) ] = 0x0038,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+                [ C(OP_PREFETCH) ] = {
+                        [ C(RESULT_ACCESS) ] = 0,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+        },
+
+        [ C(DTLB) ] = {
+                [ C(OP_READ) ] = {
+                        [ C(RESULT_ACCESS) ] = 0x0222,
+                        [ C(RESULT_MISS)   ] = 0x0220,
+                },
+                [ C(OP_WRITE) ] = {
+                        [ C(RESULT_ACCESS) ] = 0,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+                [ C(OP_PREFETCH) ] = {
+                        [ C(RESULT_ACCESS) ] = 0,
+                        [ C(RESULT_MISS)   ] = 0,
+                },
+        },
+
+        [ C(ITLB) ] = {
+                [ C(OP_READ) ] = {
+                        [ C(RESULT_ACCESS) ] = 0,
+                        [ C(RESULT_MISS)   ] = 0x02a0,
+                },
+                [ C(OP_WRITE) ] = {
+                        [ C(RESULT_ACCESS) ] = -1,
+                        [ C(RESULT_MISS)   ] = -1,
+                },
+                [ C(OP_PREFETCH) ] = {
+                        [ C(RESULT_ACCESS) ] = -1,
+                        [ C(RESULT_MISS)   ] = -1,
+                },
+        },
+
+        [ C(BPU) ] = {
+                [ C(OP_READ) ] = {
+                        [ C(RESULT_ACCESS) ] = -1,
+                        [ C(RESULT_MISS)   ] = -1,
+                },
+                [ C(OP_WRITE) ] = {
+                        [ C(RESULT_ACCESS) ] = -1,
+                        [ C(RESULT_MISS)   ] = -1,
+                },
+                [ C(OP_PREFETCH) ] = {
+                        [ C(RESULT_ACCESS) ] = -1,
+                        [ C(RESULT_MISS)   ] = -1,
+                },
+        },
+};
+
+static int sh4a_event_map(int event)
+{
+        return sh4a_general_events[event];
+}
+
+static u64 sh4a_pmu_read(int idx)
+{
+        return __raw_readl(PPC_PMCTR(idx));
+}
+
+static void sh4a_pmu_disable(struct hw_perf_event *hwc, int idx)
+{
+        unsigned int tmp;
+
+        tmp = __raw_readl(PPC_CCBR(idx));
+        tmp &= ~(CCBR_CIT_MASK | CCBR_DUC);
+        __raw_writel(tmp, PPC_CCBR(idx));
+}
+
+static void sh4a_pmu_enable(struct hw_perf_event *hwc, int idx)
+{
+        unsigned int tmp;
+
+        tmp = __raw_readl(PPC_PMCAT);
+        tmp &= ~PMCAT_EMU_CLR_MASK;
+        tmp |= idx ? PMCAT_CLR1 : PMCAT_CLR0;
+        __raw_writel(tmp, PPC_PMCAT);
+
+        tmp = __raw_readl(PPC_CCBR(idx));
+        tmp |= (hwc->config << 6) | CCBR_CMDS | CCBR_PPCE;
+        __raw_writel(tmp, PPC_CCBR(idx));
+
+        __raw_writel(__raw_readl(PPC_CCBR(idx)) | CCBR_DUC, PPC_CCBR(idx));
+}
+
+static void sh4a_pmu_disable_all(void)
+{
+        int i;
+
+        for (i = 0; i < sh4a_pmu.num_events; i++)
+                __raw_writel(__raw_readl(PPC_CCBR(i)) & ~CCBR_DUC, PPC_CCBR(i));
+}
+
+static void sh4a_pmu_enable_all(void)
+{
+        int i;
+
+        for (i = 0; i < sh4a_pmu.num_events; i++)
+                __raw_writel(__raw_readl(PPC_CCBR(i)) | CCBR_DUC, PPC_CCBR(i));
+}
+
+static struct sh_pmu sh4a_pmu = {
+        .name           = "SH-4A",
+        .num_events     = 2,
+        .event_map      = sh4a_event_map,
+        .max_events     = ARRAY_SIZE(sh4a_general_events),
+        .raw_event_mask = 0x3ff,
+        .cache_events   = &sh4a_cache_events,
+        .read           = sh4a_pmu_read,
+        .disable        = sh4a_pmu_disable,
+        .enable         = sh4a_pmu_enable,
+        .disable_all    = sh4a_pmu_disable_all,
+        .enable_all     = sh4a_pmu_enable_all,
+};
+
+static int __init sh4a_pmu_init(void)
+{
+        /*
+         * Make sure this CPU actually has perf counters.
+         */
+        if (!(boot_cpu_data.flags & CPU_HAS_PERF_COUNTER)) {
+                pr_notice("HW perf events unsupported, software events only.\n");
+                return -ENODEV;
+        }
+
+        return register_sh_pmu(&sh4a_pmu);
+}
+arch_initcall(sh4a_pmu_init);
diff --git a/arch/sh/kernel/perf_event.c b/arch/sh/kernel/perf_event.c
new file mode 100644
index 00000000000..d1510702f20
--- /dev/null
+++ b/arch/sh/kernel/perf_event.c
@@ -0,0 +1,314 @@
+/*
+ * Performance event support framework for SuperH hardware counters.
+ *
+ *  Copyright (C) 2009  Paul Mundt
+ *
+ * Heavily based on the x86 and PowerPC implementations.
+ *
+ * x86:
+ *  Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
+ *  Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar
+ *  Copyright (C) 2009 Jaswinder Singh Rajput
+ *  Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter
+ *  Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
+ *  Copyright (C) 2009 Intel Corporation, <markus.t.metzger@intel.com>
+ *
+ * ppc:
+ *  Copyright 2008-2009 Paul Mackerras, IBM Corporation.
+ *
+ * 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/kernel.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/irq.h>
+#include <linux/perf_event.h>
+#include <asm/processor.h>
+
+struct cpu_hw_events {
+        struct perf_event       *events[MAX_HWEVENTS];
+        unsigned long           used_mask[BITS_TO_LONGS(MAX_HWEVENTS)];
+        unsigned long           active_mask[BITS_TO_LONGS(MAX_HWEVENTS)];
+};
+
+DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
+
+static struct sh_pmu *sh_pmu __read_mostly;
+
+/* Number of perf_events counting hardware events */
+static atomic_t num_events;
+/* Used to avoid races in calling reserve/release_pmc_hardware */
+static DEFINE_MUTEX(pmc_reserve_mutex);
+
+/*
+ * Stub these out for now, do something more profound later.
+ */
+int reserve_pmc_hardware(void)
+{
+        return 0;
+}
+
+void release_pmc_hardware(void)
+{
+}
+
+static inline int sh_pmu_initialized(void)
+{
+        return !!sh_pmu;
+}
+
+/*
+ * Release the PMU if this is the last perf_event.
+ */
+static void hw_perf_event_destroy(struct perf_event *event)
+{
+        if (!atomic_add_unless(&num_events, -1, 1)) {
+                mutex_lock(&pmc_reserve_mutex);
+                if (atomic_dec_return(&num_events) == 0)
+                        release_pmc_hardware();
+                mutex_unlock(&pmc_reserve_mutex);
+        }
+}
+
+static int hw_perf_cache_event(int config, int *evp)
+{
+        unsigned long type, op, result;
+        int ev;
+
+        if (!sh_pmu->cache_events)
+                return -EINVAL;
+
+        /* unpack config */
+        type = config & 0xff;
+        op = (config >> 8) & 0xff;
+        result = (config >> 16) & 0xff;
+
+        if (type >= PERF_COUNT_HW_CACHE_MAX ||
+            op >= PERF_COUNT_HW_CACHE_OP_MAX ||
+            result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
+                return -EINVAL;
+
+        ev = (*sh_pmu->cache_events)[type][op][result];
+        if (ev == 0)
+                return -EOPNOTSUPP;
+        if (ev == -1)
+                return -EINVAL;
+        *evp = ev;
+        return 0;
+}
+
+static int __hw_perf_event_init(struct perf_event *event)
+{
+        struct perf_event_attr *attr = &event->attr;
+        struct hw_perf_event *hwc = &event->hw;
+        int config;
+        int err;
+
+        if (!sh_pmu_initialized())
+                return -ENODEV;
+
+        /*
+         * All of the on-chip counters are "limited", in that they have
+         * no interrupts, and are therefore unable to do sampling without
+         * further work and timer assistance.
+         */
+        if (hwc->sample_period)
+                return -EINVAL;
+
+        /*
+         * See if we need to reserve the counter.
+         *
+         * If no events are currently in use, then we have to take a
+         * mutex to ensure that we don't race with another task doing
+         * reserve_pmc_hardware or release_pmc_hardware.
+         */
+        err = 0;
+        if (!atomic_inc_not_zero(&num_events)) {
+                mutex_lock(&pmc_reserve_mutex);
+                if (atomic_read(&num_events) == 0 &&
+                    reserve_pmc_hardware())
+                        err = -EBUSY;
+                else
+                        atomic_inc(&num_events);
+                mutex_unlock(&pmc_reserve_mutex);
+        }
+
+        if (err)
+                return err;
+
+        event->destroy = hw_perf_event_destroy;
+
+        switch (attr->type) {
+        case PERF_TYPE_RAW:
+                config = attr->config & sh_pmu->raw_event_mask;
+                break;
+        case PERF_TYPE_HW_CACHE:
+                err = hw_perf_cache_event(attr->config, &config);
+                if (err)
+                        return err;
+                break;
+        case PERF_TYPE_HARDWARE:
+                if (attr->config >= sh_pmu->max_events)
+                        return -EINVAL;
+
+                config = sh_pmu->event_map(attr->config);
+                break;
+        default:
+                return -EINVAL;
+        }
+
+        if (config == -1)
+                return -EINVAL;
+
+        hwc->config |= config;
+
+        return 0;
+}
+
+static void sh_perf_event_update(struct perf_event *event,
+                                   struct hw_perf_event *hwc, int idx)
+{
+        u64 prev_raw_count, new_raw_count;
+        s64 delta;
+        int shift = 0;
+
+        /*
+         * Depending on the counter configuration, they may or may not
+         * be chained, in which case the previous counter value can be
+         * updated underneath us if the lower-half overflows.
+         *
+         * Our tactic to handle this is to first atomically read and
+         * exchange a new raw count - then add that new-prev delta
+         * count to the generic counter atomically.
+         *
+         * As there is no interrupt associated with the overflow events,
+         * this is the simplest approach for maintaining consistency.
+         */
+again:
+        prev_raw_count = atomic64_read(&hwc->prev_count);
+        new_raw_count = sh_pmu->read(idx);
+
+        if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count,
+                             new_raw_count) != prev_raw_count)
+                goto again;
+
+        /*
+         * Now we have the new raw value and have updated the prev
+         * timestamp already. We can now calculate the elapsed delta
+         * (counter-)time and add that to the generic counter.
+         *
+         * Careful, not all hw sign-extends above the physical width
+         * of the count.
+         */
+        delta = (new_raw_count << shift) - (prev_raw_count << shift);
+        delta >>= shift;
+
+        atomic64_add(delta, &event->count);
+}
+
+static void sh_pmu_disable(struct perf_event *event)
+{
+        struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+        struct hw_perf_event *hwc = &event->hw;
+        int idx = hwc->idx;
+
+        clear_bit(idx, cpuc->active_mask);
+        sh_pmu->disable(hwc, idx);
+
+        barrier();
+
+        sh_perf_event_update(event, &event->hw, idx);
+
+        cpuc->events[idx] = NULL;
+        clear_bit(idx, cpuc->used_mask);
+
+        perf_event_update_userpage(event);
+}
+
+static int sh_pmu_enable(struct perf_event *event)
+{
+        struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+        struct hw_perf_event *hwc = &event->hw;
+        int idx = hwc->idx;
+
+        if (test_and_set_bit(idx, cpuc->used_mask)) {
+                idx = find_first_zero_bit(cpuc->used_mask, sh_pmu->num_events);
+                if (idx == sh_pmu->num_events)
+                        return -EAGAIN;
+
+                set_bit(idx, cpuc->used_mask);
+                hwc->idx = idx;
+        }
+
+        sh_pmu->disable(hwc, idx);
+
+        cpuc->events[idx] = event;
+        set_bit(idx, cpuc->active_mask);
+
+        sh_pmu->enable(hwc, idx);
+
+        perf_event_update_userpage(event);
+
+        return 0;
+}
+
+static void sh_pmu_read(struct perf_event *event)
+{
+        sh_perf_event_update(event, &event->hw, event->hw.idx);
+}
+
+static const struct pmu pmu = {
+        .enable         = sh_pmu_enable,
+        .disable        = sh_pmu_disable,
+        .read           = sh_pmu_read,
+};
+
+const struct pmu *hw_perf_event_init(struct perf_event *event)
+{
+        int err = __hw_perf_event_init(event);
+        if (unlikely(err)) {
+                if (event->destroy)
+                        event->destroy(event);
+                return ERR_PTR(err);
+        }
+
+        return &pmu;
+}
+
+void hw_perf_event_setup(int cpu)
+{
+        struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu);
+
+        memset(cpuhw, 0, sizeof(struct cpu_hw_events));
+}
+
+void hw_perf_enable(void)
+{
+        if (!sh_pmu_initialized())
+                return;
+
+        sh_pmu->enable_all();
+}
+
+void hw_perf_disable(void)
+{
+        if (!sh_pmu_initialized())
+                return;
+
+        sh_pmu->disable_all();
+}
+
+int register_sh_pmu(struct sh_pmu *pmu)
+{
+        if (sh_pmu)
+                return -EBUSY;
+        sh_pmu = pmu;
+
+        pr_info("Performance Events: %s support registered\n", pmu->name);
+
+        WARN_ON(pmu->num_events >= MAX_HWEVENTS);
+
+        return 0;
+}