nds32: Perf Support

index 8e84fc385b94..19b2841219ad 100644
@@ -692,8 +692,8 @@
 #define PFM_CTL_offKU1		13	/* Enable user mode event counting for PFMC1 */
 #define PFM_CTL_offKU2		14	/* Enable user mode event counting for PFMC2 */
 #define PFM_CTL_offSEL0		15	/* The event selection for PFMC0 */
-#define PFM_CTL_offSEL1		21	/* The event selection for PFMC1 */
-#define PFM_CTL_offSEL2		27	/* The event selection for PFMC2 */
+#define PFM_CTL_offSEL1		16	/* The event selection for PFMC1 */
+#define PFM_CTL_offSEL2		22	/* The event selection for PFMC2 */
 /* bit 28:31 reserved */
 
 #define PFM_CTL_mskEN0		( 0x01  << PFM_CTL_offEN0 )

new file mode 100644
index 000000000000..fcdff02acc14
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (C) 2008-2018 Andes Technology Corporation */
+
+#ifndef __ASM_PERF_EVENT_H
+#define __ASM_PERF_EVENT_H
+
+/*
+ * This file is request by Perf,
+ * please refer to tools/perf/design.txt for more details
+ */
+struct pt_regs;
+unsigned long perf_instruction_pointer(struct pt_regs *regs);
+unsigned long perf_misc_flags(struct pt_regs *regs);
+#define perf_misc_flags(regs)   perf_misc_flags(regs)
+
+#endif

386 lines not on this page · raw message

new file mode 100644
index 000000000000..6bf7c777bda4
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (C) 2008-2018 Andes Technology Corporation */
+
+#ifndef __ASM_STACKTRACE_H
+#define __ASM_STACKTRACE_H
+
+/* Kernel callchain */
+struct stackframe {
+	unsigned long fp;
+	unsigned long sp;
+	unsigned long lp;
+};
+
+/*
+ * struct frame_tail: User callchain
+ * IMPORTANT:
+ * This struct is used for call-stack walking,
+ * the order and types matters.
+ * Do not use array, it only stores sizeof(pointer)
+ *
+ * The details can refer to arch/arm/kernel/perf_event.c
+ */
+struct frame_tail {
+	unsigned long stack_fp;
+	unsigned long stack_lp;
+};
+
+/* For User callchain with optimize for size */
+struct frame_tail_opt_size {
+	unsigned long stack_r6;
+	unsigned long stack_fp;
+	unsigned long stack_gp;
+	unsigned long stack_lp;
+};
+
+extern void
+get_real_ret_addr(unsigned long *addr, struct task_struct *tsk, int *graph);
+
+#endif /* __ASM_STACKTRACE_H */

1,223 lines not on this page · raw message

index b740534b152c..68d5f2a27f38 100644
@@ -9,6 +9,7 @@
 #include <linux/init.h>
 #include <linux/hardirq.h>
 #include <linux/uaccess.h>
+#include <linux/perf_event.h>
 
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>

@@ -169,8 +170,6 @@ void do_page_fault(unsigned long entry, unsigned long addr,
 			mask = VM_EXEC;
 		else {
 			mask = VM_READ | VM_WRITE;
-			if (vma->vm_flags & VM_WRITE)
-				flags |= FAULT_FLAG_WRITE;
 		}
 	} else if (entry == ENTRY_TLB_MISC) {
 		switch (error_code & ITYPE_mskETYPE) {

@@ -231,11 +230,17 @@ void do_page_fault(unsigned long entry, unsigned long addr,
 	 * attempt. If we go through a retry, it is extremely likely that the
 	 * page will be found in page cache at that point.
 	 */
+	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
-		if (fault & VM_FAULT_MAJOR)
+		if (fault & VM_FAULT_MAJOR) {
 			tsk->maj_flt++;
-		else
+			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
+				      1, regs, addr);
+		} else {
 			tsk->min_flt++;
+			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
+				      1, regs, addr);
+		}
 		if (fault & VM_FAULT_RETRY) {
 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
 			flags |= FAULT_FLAG_TRIED;

index 391d942536e5..5531becf0ba7 100644
@@ -23,6 +23,8 @@
 #include "../../arch/ia64/include/asm/barrier.h"
 #elif defined(__xtensa__)
 #include "../../arch/xtensa/include/asm/barrier.h"
+#elif defined(__nds32__)
+#include "../../arch/nds32/include/asm/barrier.h"
 #else
 #include <asm-generic/barrier.h>
 #endif

new file mode 100644
index 000000000000..ef9dbdbe7968
@@ -0,0 +1,29 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (C) 2005-2017 Andes Technology Corporation
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <api/fs/fs.h>
+#include "header.h"
+
+#define STR_LEN 1024
+
+char *get_cpuid_str(struct perf_pmu *pmu)
+{
+	/* In nds32, we only have one cpu */
+	char *buf = NULL;
+	struct cpu_map *cpus;
+	const char *sysfs = sysfs__mountpoint();
+
+	if (!sysfs || !pmu || !pmu->cpus)
+		return NULL;
+
+	buf = malloc(STR_LEN);
+	if (!buf)
+		return NULL;
+
+	cpus = cpu_map__get(pmu->cpus);
+	sprintf(buf, "0x%x", cpus->nr - 1);
+	cpu_map__put(cpus);
+	return buf;
+}

index a6e723d0fdbc..5e00ce54d0ff 100644
@@ -1193,6 +1193,305 @@ static int __init register_pmu_driver(void)
 
 device_initcall(register_pmu_driver);
 
+/*
+ * References: arch/nds32/kernel/traps.c:__dump()
+ * You will need to know the NDS ABI first.
+ */
+static int unwind_frame_kernel(struct stackframe *frame)
+{
+	int graph = 0;
+#ifdef CONFIG_FRAME_POINTER
+	/* 0x3 means misalignment */
+	if (!kstack_end((void *)frame->fp) &&
+	    !((unsigned long)frame->fp & 0x3) &&
+	    ((unsigned long)frame->fp >= TASK_SIZE)) {
+		/*
+		 *	The array index is based on the ABI, the below graph
+		 *	illustrate the reasons.
+		 *	Function call procedure: "smw" and "lmw" will always
+		 *	update SP and FP for you automatically.
+		 *
+		 *	Stack                                 Relative Address
+		 *	|  |                                          0
+		 *	----
+		 *	|LP| <-- SP(before smw)  <-- FP(after smw)   -1
+		 *	----
+		 *	|FP|                                         -2
+		 *	----
+		 *	|  | <-- SP(after smw)                       -3
+		 */
+		frame->lp = ((unsigned long *)frame->fp)[-1];
+		frame->fp = ((unsigned long *)frame->fp)[FP_OFFSET];
+		/* make sure CONFIG_FUNCTION_GRAPH_TRACER is turned on */
+		if (__kernel_text_address(frame->lp))
+			frame->lp = ftrace_graph_ret_addr
+						(NULL, &graph, frame->lp, NULL);
+
+		return 0;
+	} else {
+		return -EPERM;
+	}
+#else
+	/*
+	 * You can refer to arch/nds32/kernel/traps.c:__dump()
+	 * Treat "sp" as "fp", but the "sp" is one frame ahead of "fp".
+	 * And, the "sp" is not always correct.
+	 *
+	 *   Stack                                 Relative Address
+	 *   |  |                                          0
+	 *   ----
+	 *   |LP| <-- SP(before smw)                      -1
+	 *   ----
+	 *   |  | <-- SP(after smw)                       -2
+	 *   ----
+	 */
+	if (!kstack_end((void *)frame->sp)) {
+		frame->lp = ((unsigned long *)frame->sp)[1];
+		/* TODO: How to deal with the value in first
+		 * "sp" is not correct?
+		 */
+		if (__kernel_text_address(frame->lp))
+			frame->lp = ftrace_graph_ret_addr
+						(tsk, &graph, frame->lp, NULL);
+
+		frame->sp = ((unsigned long *)frame->sp) + 1;
+
+		return 0;
+	} else {
+		return -EPERM;
+	}
+#endif
+}
+
+static void notrace
+walk_stackframe(struct stackframe *frame,
+		int (*fn_record)(struct stackframe *, void *),
+		void *data)
+{
+	while (1) {
+		int ret;
+
+		if (fn_record(frame, data))
+			break;
+
+		ret = unwind_frame_kernel(frame);
+		if (ret < 0)
+			break;
+	}
+}
+
+/*
+ * 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_ctx *entry = data;
+
+	perf_callchain_store(entry, fr->lp);
+	return 0;
+}
+
+/*
+ * Get the return address for a single stackframe and return a pointer to the
+ * next frame tail.
+ */
+static unsigned long
+user_backtrace(struct perf_callchain_entry_ctx *entry, unsigned long fp)
+{
+	struct frame_tail buftail;
+	unsigned long lp = 0;
+	unsigned long *user_frame_tail =
+		(unsigned long *)(fp - (unsigned long)sizeof(buftail));
+
+	/* Check accessibility of one struct frame_tail beyond */
+	if (!access_ok(VERIFY_READ, user_frame_tail, sizeof(buftail)))
+		return 0;
+	if (__copy_from_user_inatomic
+		(&buftail, user_frame_tail, sizeof(buftail)))
+		return 0;
+
+	/*
+	 * Refer to unwind_frame_kernel() for more illurstration
+	 */
+	lp = buftail.stack_lp;  /* ((unsigned long *)fp)[-1] */
+	fp = buftail.stack_fp;  /* ((unsigned long *)fp)[FP_OFFSET] */
+	perf_callchain_store(entry, lp);
+	return fp;
+}
+
+static unsigned long
+user_backtrace_opt_size(struct perf_callchain_entry_ctx *entry,
+			unsigned long fp)
+{
+	struct frame_tail_opt_size buftail;
+	unsigned long lp = 0;
+
+	unsigned long *user_frame_tail =
+		(unsigned long *)(fp - (unsigned long)sizeof(buftail));
+
+	/* Check accessibility of one struct frame_tail beyond */
+	if (!access_ok(VERIFY_READ, user_frame_tail, sizeof(buftail)))
+		return 0;
+	if (__copy_from_user_inatomic
+		(&buftail, user_frame_tail, sizeof(buftail)))
+		return 0;
+
+	/*
+	 * Refer to unwind_frame_kernel() for more illurstration
+	 */
+	lp = buftail.stack_lp;  /* ((unsigned long *)fp)[-1] */
+	fp = buftail.stack_fp;  /* ((unsigned long *)fp)[FP_OFFSET] */
+
+	perf_callchain_store(entry, lp);
+	return fp;
+}
+
+/*
+ * This will be called when the target is in user mode
+ * This function will only be called when we use
+ * "PERF_SAMPLE_CALLCHAIN" in
+ * kernel/events/core.c:perf_prepare_sample()
+ *
+ * How to trigger perf_callchain_[user/kernel] :
+ * $ perf record -e cpu-clock --call-graph fp ./program
+ * $ perf report --call-graph
+ */
+unsigned long leaf_fp;
+void
+perf_callchain_user(struct perf_callchain_entry_ctx *entry,
+		    struct pt_regs *regs)
+{
+	unsigned long fp = 0;
+	unsigned long gp = 0;
+	unsigned long lp = 0;
+	unsigned long sp = 0;
+	unsigned long *user_frame_tail;
+
+	leaf_fp = 0;
+
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		/* We don't support guest os callchain now */
+		return;
+	}
+
+	perf_callchain_store(entry, regs->ipc);
+	fp = regs->fp;
+	gp = regs->gp;
+	lp = regs->lp;
+	sp = regs->sp;
+	if (entry->nr < PERF_MAX_STACK_DEPTH &&
+	    (unsigned long)fp && !((unsigned long)fp & 0x7) && fp > sp) {
+		user_frame_tail =
+			(unsigned long *)(fp - (unsigned long)sizeof(fp));
+
+		if (!access_ok(VERIFY_READ, user_frame_tail, sizeof(fp)))
+			return;
+
+		if (__copy_from_user_inatomic
+			(&leaf_fp, user_frame_tail, sizeof(fp)))
+			return;
+
+		if (leaf_fp == lp) {
+			/*
+			 * Maybe this is non leaf function
+			 * with optimize for size,
+			 * or maybe this is the function
+			 * with optimize for size
+			 */
+			struct frame_tail buftail;
+
+			user_frame_tail =
+				(unsigned long *)(fp -
+					(unsigned long)sizeof(buftail));
+
+			if (!access_ok
+				(VERIFY_READ, user_frame_tail, sizeof(buftail)))
+				return;
+
+			if (__copy_from_user_inatomic
+				(&buftail, user_frame_tail, sizeof(buftail)))
+				return;
+
+			if (buftail.stack_fp == gp) {
+				/* non leaf function with optimize
+				 * for size condition
+				 */
+				struct frame_tail_opt_size buftail_opt_size;
+
+				user_frame_tail =
+					(unsigned long *)(fp - (unsigned long)
+						sizeof(buftail_opt_size));
+
+				if (!access_ok(VERIFY_READ, user_frame_tail,
+					       sizeof(buftail_opt_size)))
+					return;
+
+				if (__copy_from_user_inatomic
+				   (&buftail_opt_size, user_frame_tail,
+				   sizeof(buftail_opt_size)))
+					return;
+
+				perf_callchain_store(entry, lp);
+				fp = buftail_opt_size.stack_fp;
+
+				while ((entry->nr < PERF_MAX_STACK_DEPTH) &&
+				       (unsigned long)fp &&
+						!((unsigned long)fp & 0x7) &&
+						fp > sp) {
+					sp = fp;
+					fp = user_backtrace_opt_size(entry, fp);
+				}
+
+			} else {
+				/* this is the function
+				 * without optimize for size
+				 */
+				fp = buftail.stack_fp;
+				perf_callchain_store(entry, lp);
+				while ((entry->nr < PERF_MAX_STACK_DEPTH) &&
+				       (unsigned long)fp &&
+						!((unsigned long)fp & 0x7) &&
+						fp > sp) {
+					sp = fp;
+					fp = user_backtrace(entry, fp);
+				}
+			}
+		} else {
+			/* this is leaf function */
+			fp = leaf_fp;
+			perf_callchain_store(entry, lp);
+
+			/* previous function callcahin  */
+			while ((entry->nr < PERF_MAX_STACK_DEPTH) &&
+			       (unsigned long)fp &&
+				   !((unsigned long)fp & 0x7) && fp > sp) {
+				sp = fp;
+				fp = user_backtrace(entry, fp);
+			}
+		}
+		return;
+	}
+}
+
+/* This will be called when the target is in kernel mode */
+void
+perf_callchain_kernel(struct perf_callchain_entry_ctx *entry,
+		      struct pt_regs *regs)
+{
+	struct stackframe fr;
+
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		/* We don't support guest os callchain now */
+		return;
+	}
+	fr.fp = regs->fp;
+	fr.lp = regs->lp;
+	fr.sp = regs->sp;
+	walk_stackframe(&fr, callchain_trace, entry);
+}
+
 unsigned long perf_instruction_pointer(struct pt_regs *regs)
 {
 	/* However, NDS32 does not support virtualization */