AMU: Add configuration helpers for aarch64

Add some AMU helper functions to allow configuring, reading and
writing of the Group 0 and Group 1 counters.  Documentation for these
helpers will come in a separate patch.

Change-Id: I656e070d2dae830c22414f694aa655341d4e2c40
Signed-off-by: Dimitris Papastamos <dimitris.papastamos@arm.com>

bl31/bl31.mk

@@ -51,7 +51,8 @@ BL31_SOURCES		+=	lib/extensions/spe/spe.c
 endif
 
 ifeq (${ENABLE_AMU},1)
-BL31_SOURCES		+=	lib/extensions/amu/aarch64/amu.c
+BL31_SOURCES		+=	lib/extensions/amu/aarch64/amu.c		\
+				lib/extensions/amu/aarch64/amu_helpers.S
 endif
 
 ifeq (${ENABLE_SVE_FOR_NS},1)

include/lib/aarch64/arch.h

@@ -656,4 +656,45 @@
 #define AMEVTYPER02_EL0		S3_3_C13_C6_2
 #define AMEVTYPER03_EL0		S3_3_C13_C6_3
 
+/* Activity Monitor Group 1 Event Counter Registers */
+#define AMEVCNTR10_EL0		S3_3_C13_C12_0
+#define AMEVCNTR11_EL0		S3_3_C13_C12_1
+#define AMEVCNTR12_EL0		S3_3_C13_C12_2
+#define AMEVCNTR13_EL0		S3_3_C13_C12_3
+#define AMEVCNTR14_EL0		S3_3_C13_C12_4
+#define AMEVCNTR15_EL0		S3_3_C13_C12_5
+#define AMEVCNTR16_EL0		S3_3_C13_C12_6
+#define AMEVCNTR17_EL0		S3_3_C13_C12_7
+#define AMEVCNTR18_EL0		S3_3_C13_C13_0
+#define AMEVCNTR19_EL0		S3_3_C13_C13_1
+#define AMEVCNTR1A_EL0		S3_3_C13_C13_2
+#define AMEVCNTR1B_EL0		S3_3_C13_C13_3
+#define AMEVCNTR1C_EL0		S3_3_C13_C13_4
+#define AMEVCNTR1D_EL0		S3_3_C13_C13_5
+#define AMEVCNTR1E_EL0		S3_3_C13_C13_6
+#define AMEVCNTR1F_EL0		S3_3_C13_C13_7
+
+/* Activity Monitor Group 1 Event Type Registers */
+#define AMEVTYPER10_EL0		S3_3_C13_C14_0
+#define AMEVTYPER11_EL0		S3_3_C13_C14_1
+#define AMEVTYPER12_EL0		S3_3_C13_C14_2
+#define AMEVTYPER13_EL0		S3_3_C13_C14_3
+#define AMEVTYPER14_EL0		S3_3_C13_C14_4
+#define AMEVTYPER15_EL0		S3_3_C13_C14_5
+#define AMEVTYPER16_EL0		S3_3_C13_C14_6
+#define AMEVTYPER17_EL0		S3_3_C13_C14_7
+#define AMEVTYPER18_EL0		S3_3_C13_C15_0
+#define AMEVTYPER19_EL0		S3_3_C13_C15_1
+#define AMEVTYPER1A_EL0		S3_3_C13_C15_2
+#define AMEVTYPER1B_EL0		S3_3_C13_C15_3
+#define AMEVTYPER1C_EL0		S3_3_C13_C15_4
+#define AMEVTYPER1D_EL0		S3_3_C13_C15_5
+#define AMEVTYPER1E_EL0		S3_3_C13_C15_6
+#define AMEVTYPER1F_EL0		S3_3_C13_C15_7
+
+/* AMCGCR_EL0 definitions */
+#define AMCGCR_EL0_CG1NC_SHIFT	U(8)
+#define AMCGCR_EL0_CG1NC_LENGTH	U(8)
+#define AMCGCR_EL0_CG1NC_MASK	U(0xff)
+
 #endif /* __ARCH_H__ */

include/lib/aarch64/arch_helpers.h

@@ -322,6 +322,7 @@ DEFINE_RENAME_SYSREG_WRITE_FUNC(icc_eoir0_el1, ICC_EOIR0_EL1)
 DEFINE_RENAME_SYSREG_WRITE_FUNC(icc_eoir1_el1, ICC_EOIR1_EL1)
 DEFINE_RENAME_SYSREG_WRITE_FUNC(icc_sgi0r_el1, ICC_SGI0R_EL1)
 
+DEFINE_RENAME_SYSREG_RW_FUNCS(amcgcr_el0, AMCGCR_EL0)
 DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenclr0_el0, AMCNTENCLR0_EL0)
 DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenset0_el0, AMCNTENSET0_EL0)
 DEFINE_RENAME_SYSREG_RW_FUNCS(amcntenclr1_el0, AMCNTENCLR1_EL0)

include/lib/extensions/amu.h

@@ -10,6 +10,7 @@
 #include <sys/cdefs.h> /* for CASSERT() */
 #include <cassert.h>
 #include <platform_def.h>
+#include <stdint.h>
 
 /* All group 0 counters */
 #define AMU_GROUP0_COUNTERS_MASK	0xf
@@ -29,6 +30,16 @@
 CASSERT(AMU_GROUP1_COUNTERS_MASK <= 0xffff, invalid_amu_group1_counters_mask);
 CASSERT(AMU_GROUP1_NR_COUNTERS <= 16, invalid_amu_group1_nr_counters);
 
+int amu_supported(void);
 void amu_enable(int el2_unused);
 
+/* Group 0 configuration helpers */
+uint64_t amu_group0_cnt_read(int idx);
+void amu_group0_cnt_write(int idx, uint64_t val);
+
+/* Group 1 configuration helpers */
+uint64_t amu_group1_cnt_read(int idx);
+void amu_group1_cnt_write(int idx, uint64_t val);
+void amu_group1_set_evtype(int idx, unsigned int val);
+
 #endif /* __AMU_H__ */

include/lib/extensions/amu_private.h

@@ -0,0 +1,19 @@
+/*
+ * Copyright (c) 2017, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef __AMU_PRIVATE_H__
+#define __AMU_PRIVATE_H__
+
+#include <stdint.h>
+
+uint64_t amu_group0_cnt_read_internal(int idx);
+void amu_group0_cnt_write_internal(int idx, uint64_t);
+
+uint64_t amu_group1_cnt_read_internal(int idx);
+void amu_group1_cnt_write_internal(int idx, uint64_t);
+void amu_group1_set_evtype_internal(int idx, unsigned int val);
+
+#endif /* __AMU_PRIVATE_H__ */

lib/extensions/amu/aarch32/amu.c

@@ -7,13 +7,18 @@
 #include <amu.h>
 #include <arch.h>
 #include <arch_helpers.h>
+#include <debug.h>
 
 void amu_enable(int el2_unused)
 {
 	uint64_t features;
 
 	features = read_id_pfr0() >> ID_PFR0_AMU_SHIFT;
-	if ((features & ID_PFR0_AMU_MASK) == 1) {
+	if ((features & ID_PFR0_AMU_MASK) != 1) {
+		WARN("Cannot enable AMU - not supported\n");
+		return;
+	}
+
 	if (el2_unused) {
 		uint64_t v;
 
@@ -29,4 +34,3 @@ void amu_enable(int el2_unused)
 	/* Enable group 0 counters */
 	write_amcntenset0(AMU_GROUP0_COUNTERS_MASK);
 }
-}

lib/extensions/amu/aarch64/amu.c

@@ -5,17 +5,35 @@
  */
 
 #include <amu.h>
+#include <amu_private.h>
 #include <arch.h>
 #include <arch_helpers.h>
+#include <assert.h>
+#include <debug.h>
 
-void amu_enable(int el2_unused)
+#define AMU_GROUP0_NR_COUNTERS	4
+
+int amu_supported(void)
 {
 	uint64_t features;
 
 	features = read_id_aa64pfr0_el1() >> ID_AA64PFR0_AMU_SHIFT;
-	if ((features & ID_AA64PFR0_AMU_MASK) == 1) {
+	return (features & ID_AA64PFR0_AMU_MASK) == 1;
+}
+
+/*
+ * Enable counters.  This function is meant to be invoked
+ * by the context management library before exiting from EL3.
+ */
+void amu_enable(int el2_unused)
+{
 	uint64_t v;
 
+	if (!amu_supported()) {
+		WARN("Cannot enable AMU - not supported\n");
+		return;
+	}
+
 	if (el2_unused) {
 		/*
 		 * CPTR_EL2.TAM: Set to zero so any accesses to
@@ -39,4 +57,54 @@ void amu_enable(int el2_unused)
 	/* Enable group 1 counters */
 	write_amcntenset1_el0(AMU_GROUP1_COUNTERS_MASK);
 }
+
+/* Read the group 0 counter identified by the given `idx`. */
+uint64_t amu_group0_cnt_read(int idx)
+{
+	assert(amu_supported());
+	assert(idx >= 0 && idx < AMU_GROUP0_NR_COUNTERS);
+
+	return amu_group0_cnt_read_internal(idx);
+}
+
+/* Write the group 0 counter identified by the given `idx` with `val`. */
+void amu_group0_cnt_write(int idx, uint64_t val)
+{
+	assert(amu_supported());
+	assert(idx >= 0 && idx < AMU_GROUP0_NR_COUNTERS);
+
+	amu_group0_cnt_write_internal(idx, val);
+	isb();
+}
+
+/* Read the group 1 counter identified by the given `idx`. */
+uint64_t amu_group1_cnt_read(int idx)
+{
+	assert(amu_supported());
+	assert(idx >= 0 && idx < AMU_GROUP1_NR_COUNTERS);
+
+	return amu_group1_cnt_read_internal(idx);
+}
+
+/* Write the group 1 counter identified by the given `idx` with `val`. */
+void amu_group1_cnt_write(int idx, uint64_t val)
+{
+	assert(amu_supported());
+	assert(idx >= 0 && idx < AMU_GROUP1_NR_COUNTERS);
+
+	amu_group1_cnt_write_internal(idx, val);
+	isb();
+}
+
+/*
+ * Program the event type register for the given `idx` with
+ * the event number `val`.
+ */
+void amu_group1_set_evtype(int idx, unsigned int val)
+{
+	assert(amu_supported());
+	assert (idx >= 0 && idx < AMU_GROUP1_NR_COUNTERS);
+
+	amu_group1_set_evtype_internal(idx, val);
+	isb();
 }

lib/extensions/amu/aarch64/amu_helpers.S

@@ -0,0 +1,281 @@
+/*
+ * Copyright (c) 2017, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <arch.h>
+#include <assert_macros.S>
+#include <asm_macros.S>
+
+	.globl	amu_group0_cnt_read_internal
+	.globl	amu_group0_cnt_write_internal
+	.globl	amu_group1_cnt_read_internal
+	.globl	amu_group1_cnt_write_internal
+	.globl	amu_group1_set_evtype_internal
+
+/*
+ * uint64_t amu_group0_cnt_read_internal(int idx);
+ *
+ * Given `idx`, read the corresponding AMU counter
+ * and return it in `x0`.
+ */
+func amu_group0_cnt_read_internal
+#if ENABLE_ASSERTIONS
+	/*
+	 * It can be dangerous to call this function with an
+	 * out of bounds index.  Ensure `idx` is valid.
+	 */
+	mov	x1, x0
+	lsr	x1, x1, #2
+	cmp	x1, #0
+	ASM_ASSERT(eq)
+#endif
+
+	/*
+	 * Given `idx` calculate address of mrs/ret instruction pair
+	 * in the table below.
+	 */
+	adr	x1, 1f
+	lsl	x0, x0, #3		/* each mrs/ret sequence is 8 bytes */
+	add	x1, x1, x0
+	br	x1
+
+1:
+	mrs	x0, AMEVCNTR00_EL0	/* index 0 */
+	ret
+	mrs	x0, AMEVCNTR01_EL0	/* index 1 */
+	ret
+	mrs	x0, AMEVCNTR02_EL0	/* index 2 */
+	ret
+	mrs	x0, AMEVCNTR03_EL0	/* index 3 */
+	ret
+endfunc amu_group0_cnt_read_internal
+
+/*
+ * void amu_group0_cnt_write_internal(int idx, uint64_t val);
+ *
+ * Given `idx`, write `val` to the corresponding AMU counter.
+ */
+func amu_group0_cnt_write_internal
+#if ENABLE_ASSERTIONS
+	/*
+	 * It can be dangerous to call this function with an
+	 * out of bounds index.  Ensure `idx` is valid.
+	 */
+	mov	x2, x0
+	lsr	x2, x2, #2
+	cmp	x2, #0
+	ASM_ASSERT(eq)
+#endif
+
+	/*
+	 * Given `idx` calculate address of mrs/ret instruction pair
+	 * in the table below.
+	 */
+	adr	x2, 1f
+	lsl	x0, x0, #3		/* each msr/ret sequence is 8 bytes */
+	add	x2, x2, x0
+	br	x2
+
+1:
+	msr	AMEVCNTR00_EL0, x1	/* index 0 */
+	ret
+	msr	AMEVCNTR01_EL0, x1	/* index 1 */
+	ret
+	msr	AMEVCNTR02_EL0, x1	/* index 2 */
+	ret
+	msr	AMEVCNTR03_EL0, x1	/* index 3 */
+	ret
+endfunc amu_group0_cnt_write_internal
+
+/*
+ * uint64_t amu_group1_cnt_read_internal(int idx);
+ *
+ * Given `idx`, read the corresponding AMU counter
+ * and return it in `x0`.
+ */
+func amu_group1_cnt_read_internal
+#if ENABLE_ASSERTIONS
+	/*
+	 * It can be dangerous to call this function with an
+	 * out of bounds index.  Ensure `idx` is valid.
+	 */
+	mov	x1, x0
+	lsr	x1, x1, #4
+	cmp	x1, #0
+	ASM_ASSERT(eq)
+#endif
+
+	/*
+	 * Given `idx` calculate address of mrs/ret instruction pair
+	 * in the table below.
+	 */
+	adr	x1, 1f
+	lsl	x0, x0, #3		/* each mrs/ret sequence is 8 bytes */
+	add	x1, x1, x0
+	br	x1
+
+1:
+	mrs	x0, AMEVCNTR10_EL0	/* index 0 */
+	ret
+	mrs	x0, AMEVCNTR11_EL0	/* index 1 */
+	ret
+	mrs	x0, AMEVCNTR12_EL0	/* index 2 */
+	ret
+	mrs	x0, AMEVCNTR13_EL0	/* index 3 */
+	ret
+	mrs	x0, AMEVCNTR14_EL0	/* index 4 */
+	ret
+	mrs	x0, AMEVCNTR15_EL0	/* index 5 */
+	ret
+	mrs	x0, AMEVCNTR16_EL0	/* index 6 */
+	ret
+	mrs	x0, AMEVCNTR17_EL0	/* index 7 */
+	ret
+	mrs	x0, AMEVCNTR18_EL0	/* index 8 */
+	ret
+	mrs	x0, AMEVCNTR19_EL0	/* index 9 */
+	ret
+	mrs	x0, AMEVCNTR1A_EL0	/* index 10 */
+	ret
+	mrs	x0, AMEVCNTR1B_EL0	/* index 11 */
+	ret
+	mrs	x0, AMEVCNTR1C_EL0	/* index 12 */
+	ret
+	mrs	x0, AMEVCNTR1D_EL0	/* index 13 */
+	ret
+	mrs	x0, AMEVCNTR1E_EL0	/* index 14 */
+	ret
+	mrs	x0, AMEVCNTR1F_EL0	/* index 15 */
+	ret
+endfunc amu_group1_cnt_read_internal
+
+/*
+ * void amu_group1_cnt_write_internal(int idx, uint64_t val);
+ *
+ * Given `idx`, write `val` to the corresponding AMU counter.
+ */
+func amu_group1_cnt_write_internal
+#if ENABLE_ASSERTIONS
+	/*
+	 * It can be dangerous to call this function with an
+	 * out of bounds index.  Ensure `idx` is valid.
+	 */
+	mov	x2, x0
+	lsr	x2, x2, #4
+	cmp	x2, #0
+	ASM_ASSERT(eq)
+#endif
+
+	/*
+	 * Given `idx` calculate address of mrs/ret instruction pair
+	 * in the table below.
+	 */
+	adr	x2, 1f
+	lsl	x0, x0, #3		/* each msr/ret sequence is 8 bytes */
+	add	x2, x2, x0
+	br	x2
+
+1:
+	msr	AMEVCNTR10_EL0, x1	/* index 0 */
+	ret
+	msr	AMEVCNTR11_EL0, x1	/* index 1 */
+	ret
+	msr	AMEVCNTR12_EL0, x1	/* index 2 */
+	ret
+	msr	AMEVCNTR13_EL0, x1	/* index 3 */
+	ret
+	msr	AMEVCNTR14_EL0, x1	/* index 4 */
+	ret
+	msr	AMEVCNTR15_EL0, x1	/* index 5 */
+	ret
+	msr	AMEVCNTR16_EL0, x1	/* index 6 */
+	ret
+	msr	AMEVCNTR17_EL0, x1	/* index 7 */
+	ret
+	msr	AMEVCNTR18_EL0, x1	/* index 8 */
+	ret
+	msr	AMEVCNTR19_EL0, x1	/* index 9 */
+	ret
+	msr	AMEVCNTR1A_EL0, x1	/* index 10 */
+	ret
+	msr	AMEVCNTR1B_EL0, x1	/* index 11 */
+	ret
+	msr	AMEVCNTR1C_EL0, x1	/* index 12 */
+	ret
+	msr	AMEVCNTR1D_EL0, x1	/* index 13 */
+	ret
+	msr	AMEVCNTR1E_EL0, x1	/* index 14 */
+	ret
+	msr	AMEVCNTR1F_EL0, x1	/* index 15 */
+	ret
+endfunc amu_group1_cnt_write_internal
+
+/*
+ * void amu_group1_set_evtype_internal(int idx, unsigned int val);
+ *
+ * Program the AMU event type register indexed by `idx`
+ * with the value `val`.
+ */
+func amu_group1_set_evtype_internal
+#if ENABLE_ASSERTIONS
+	/*
+	 * It can be dangerous to call this function with an
+	 * out of bounds index.  Ensure `idx` is valid.
+	 */
+	mov	x2, x0
+	lsr	x2, x2, #4
+	cmp	x2, #0
+	ASM_ASSERT(eq)
+
+	/* val should be between [0, 65535] */
+	mov	x2, x1
+	lsr	x2, x2, #16
+	cmp	x2, #0
+	ASM_ASSERT(eq)
+#endif
+
+	/*
+	 * Given `idx` calculate address of msr/ret instruction pair
+	 * in the table below.
+	 */
+	adr	x2, 1f
+	lsl	x0, x0, #3		/* each msr/ret sequence is 8 bytes */
+	add	x2, x2, x0
+	br	x2
+
+1:
+	msr	AMEVTYPER10_EL0, x1	/* index 0 */
+	ret
+	msr	AMEVTYPER11_EL0, x1	/* index 1 */
+	ret
+	msr	AMEVTYPER12_EL0, x1	/* index 2 */
+	ret
+	msr	AMEVTYPER13_EL0, x1	/* index 3 */
+	ret
+	msr	AMEVTYPER14_EL0, x1	/* index 4 */
+	ret
+	msr	AMEVTYPER15_EL0, x1	/* index 5 */
+	ret
+	msr	AMEVTYPER16_EL0, x1	/* index 6 */
+	ret
+	msr	AMEVTYPER17_EL0, x1	/* index 7 */
+	ret
+	msr	AMEVTYPER18_EL0, x1	/* index 8 */
+	ret
+	msr	AMEVTYPER19_EL0, x1	/* index 9 */
+	ret
+	msr	AMEVTYPER1A_EL0, x1	/* index 10 */
+	ret
+	msr	AMEVTYPER1B_EL0, x1	/* index 11 */
+	ret
+	msr	AMEVTYPER1C_EL0, x1	/* index 12 */
+	ret
+	msr	AMEVTYPER1D_EL0, x1	/* index 13 */
+	ret
+	msr	AMEVTYPER1E_EL0, x1	/* index 14 */
+	ret
+	msr	AMEVTYPER1F_EL0, x1	/* index 15 */
+	ret
+endfunc amu_group1_set_evtype_internal