metag: Perf

Add Perf support for metag.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>

arch/metag/Kconfig

@@ -22,6 +22,7 @@ config METAG
 	select HAVE_MEMBLOCK
 	select HAVE_MEMBLOCK_NODE_MAP
 	select HAVE_MOD_ARCH_SPECIFIC
+	select HAVE_PERF_EVENTS
 	select HAVE_SYSCALL_TRACEPOINTS
 	select IRQ_DOMAIN
 	select MODULES_USE_ELF_RELA

arch/metag/include/asm/perf_event.h

+#ifndef __ASM_METAG_PERF_EVENT_H
+#define __ASM_METAG_PERF_EVENT_H
+
+#endif /* __ASM_METAG_PERF_EVENT_H */

arch/metag/kernel/Makefile

@@ -25,6 +25,8 @@ obj-y	+= topology.o
 obj-y	+= traps.o
 obj-y	+= user_gateway.o
 
+obj-$(CONFIG_PERF_EVENTS)		+= perf/
+
 obj-$(CONFIG_METAG_COREMEM)		+= coremem.o
 obj-$(CONFIG_DYNAMIC_FTRACE)		+= ftrace.o
 obj-$(CONFIG_FUNCTION_TRACER)		+= ftrace_stub.o

arch/metag/kernel/perf/Makefile

+# Makefile for performance event core
+
+obj-y += perf_event.o

arch/metag/kernel/perf/perf_event.h

+/*
+ * Meta performance counter support.
+ *  Copyright (C) 2012 Imagination Technologies Ltd
+ *
+ * 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.
+ */
+
+#ifndef METAG_PERF_EVENT_H_
+#define METAG_PERF_EVENT_H_
+
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/perf_event.h>
+
+/* For performance counter definitions */
+#include <asm/metag_mem.h>
+
+/*
+ * The Meta core has two performance counters, with 24-bit resolution. Newer
+ * cores generate an overflow interrupt on transition from 0xffffff to 0.
+ *
+ * Each counter consists of the counter id, hardware thread id, and the count
+ * itself; each counter can be assigned to multiple hardware threads at any
+ * one time, with the returned count being an aggregate of events. A small
+ * number of events are thread global, i.e. they count the aggregate of all
+ * threads' events, regardless of the thread selected.
+ *
+ * Newer cores can store an arbitrary 24-bit number in the counter, whereas
+ * older cores will clear the counter bits on write.
+ *
+ * We also have a pseudo-counter in the form of the thread active cycles
+ * counter (which, incidentally, is also bound to
+ */
+
+#define MAX_HWEVENTS		3
+#define MAX_PERIOD		((1UL << 24) - 1)
+#define METAG_INST_COUNTER	(MAX_HWEVENTS - 1)
+
+/**
+ * struct cpu_hw_events - a processor core's performance events
+ * @events:	an array of perf_events active for a given index.
+ * @used_mask:	a bitmap of in-use counters.
+ * @pmu_lock:	a perf counter lock
+ *
+ * This is a per-cpu/core structure that maintains a record of its
+ * performance counters' state.
+ */
+struct cpu_hw_events {
+	struct perf_event	*events[MAX_HWEVENTS];
+	unsigned long		used_mask[BITS_TO_LONGS(MAX_HWEVENTS)];
+	raw_spinlock_t		pmu_lock;
+};
+
+/**
+ * struct metag_pmu - the Meta PMU structure
+ * @pmu:		core pmu structure
+ * @name:		pmu name
+ * @version:		core version
+ * @handle_irq:		overflow interrupt handler
+ * @enable:		enable a counter
+ * @disable:		disable a counter
+ * @read:		read the value of a counter
+ * @write:		write a value to a counter
+ * @event_map:		kernel event to counter event id map
+ * @cache_events:	kernel cache counter to core cache counter map
+ * @max_period:		maximum value of the counter before overflow
+ * @max_events:		maximum number of counters available at any one time
+ * @active_events:	number of active counters
+ * @reserve_mutex:	counter reservation mutex
+ *
+ * This describes the main functionality and data used by the performance
+ * event core.
+ */
+struct metag_pmu {
+	struct pmu	pmu;
+	const char	*name;
+	u32		version;
+	irqreturn_t	(*handle_irq)(int irq_num, void *dev);
+	void		(*enable)(struct hw_perf_event *evt, int idx);
+	void		(*disable)(struct hw_perf_event *evt, int idx);
+	u64		(*read)(int idx);
+	void		(*write)(int idx, u32 val);
+	int		(*event_map)(int idx);
+	const int	(*cache_events)[PERF_COUNT_HW_CACHE_MAX]
+				[PERF_COUNT_HW_CACHE_OP_MAX]
+				[PERF_COUNT_HW_CACHE_RESULT_MAX];
+	u32		max_period;
+	int		max_events;
+	atomic_t	active_events;
+	struct mutex	reserve_mutex;
+};
+
+/* Convenience macros for accessing the perf counters */
+/* Define some convenience accessors */
+#define PERF_COUNT(x)	(PERF_COUNT0 + (sizeof(u64) * (x)))
+#define PERF_ICORE(x)	(PERF_ICORE0 + (sizeof(u64) * (x)))
+#define PERF_CHAN(x)	(PERF_CHAN0 + (sizeof(u64) * (x)))
+
+/* Cache index macros */
+#define C(x) PERF_COUNT_HW_CACHE_##x
+#define CACHE_OP_UNSUPPORTED	0xfffe
+#define CACHE_OP_NONSENSE	0xffff
+
+#endif

arch/metag/kernel/perf_callchain.c

+/*
+ * Perf callchain handling code.
+ *
+ *   Based on the ARM perf implementation.
+ */
+
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/perf_event.h>
+#include <linux/uaccess.h>
+#include <asm/ptrace.h>
+#include <asm/stacktrace.h>
+
+static bool is_valid_call(unsigned long calladdr)
+{
+	unsigned int callinsn;
+
+	/* Check the possible return address is aligned. */
+	if (!(calladdr & 0x3)) {
+		if (!get_user(callinsn, (unsigned int *)calladdr)) {
+			/* Check for CALLR or SWAP PC,D1RtP. */
+			if ((callinsn & 0xff000000) == 0xab000000 ||
+			    callinsn == 0xa3200aa0)
+				return true;
+		}
+	}
+	return false;
+}
+
+static struct metag_frame __user *
+user_backtrace(struct metag_frame __user *user_frame,
+	       struct perf_callchain_entry *entry)
+{
+	struct metag_frame frame;
+	unsigned long calladdr;
+
+	/* We cannot rely on having frame pointers in user code. */
+	while (1) {
+		/* Also check accessibility of one struct frame beyond */
+		if (!access_ok(VERIFY_READ, user_frame, sizeof(frame)))
+			return 0;
+		if (__copy_from_user_inatomic(&frame, user_frame,
+					      sizeof(frame)))
+			return 0;
+
+		--user_frame;
+
+		calladdr = frame.lr - 4;
+		if (is_valid_call(calladdr)) {
+			perf_callchain_store(entry, calladdr);
+			return user_frame;
+		}
+	}
+
+	return 0;
+}
+
+void
+perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
+{
+	unsigned long sp = regs->ctx.AX[0].U0;
+	struct metag_frame __user *frame;
+
+	frame = (struct metag_frame __user *)sp;
+
+	--frame;
+
+	while ((entry->nr < PERF_MAX_STACK_DEPTH) && frame)
+		frame = user_backtrace(frame, entry);
+}
+
+/*
+ * Gets called by walk_stackframe() for every stackframe. This will be called
+ * whist unwinding the stackframe and is like a subroutine return so we use
+ * the PC.
+ */
+static int
+callchain_trace(struct stackframe *fr,
+		void *data)
+{
+	struct perf_callchain_entry *entry = data;
+	perf_callchain_store(entry, fr->pc);
+	return 0;
+}
+
+void
+perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
+{
+	struct stackframe fr;
+
+	fr.fp = regs->ctx.AX[1].U0;
+	fr.sp = regs->ctx.AX[0].U0;
+	fr.lr = regs->ctx.DX[4].U1;
+	fr.pc = regs->ctx.CurrPC;
+	walk_stackframe(&fr, callchain_trace, entry);
+}