Refactor Statistical Profiling Extensions implementation

Factor out SPE operations in a separate file.  Use the publish
subscribe framework to drain the SPE buffers before entering secure
world.  Additionally, enable SPE before entering normal world.

A side effect of this change is that the profiling buffers are now
only drained when a transition from normal world to secure world
happens.  Previously they were drained also on return from secure
world, which is unnecessary as SPE is not supported in S-EL1.

Change-Id: I17582c689b4b525770dbb6db098b3a0b5777b70a
Signed-off-by: Dimitris Papastamos <dimitris.papastamos@arm.com>

Makefile

@@ -189,6 +189,7 @@ INCLUDES		+=	-Iinclude/bl1				\
 				-Iinclude/lib/cpus/${ARCH}		\
 				-Iinclude/lib/el3_runtime		\
 				-Iinclude/lib/el3_runtime/${ARCH}	\
+				-Iinclude/lib/extensions		\
 				-Iinclude/lib/pmf			\
 				-Iinclude/lib/psci			\
 				-Iinclude/lib/xlat_tables		\

bl31/bl31.mk

@@ -46,6 +46,10 @@ BL31_SOURCES		+=	services/std_svc/sdei/sdei_event.c	\
 				services/std_svc/sdei/sdei_state.c
 endif
 
+ifeq (${ENABLE_SPE_FOR_LOWER_ELS},1)
+BL31_SOURCES		+=	lib/extensions/spe/spe.c
+endif
+
 BL31_LINKERFILE		:=	bl31/bl31.ld.S
 
 # Flag used to indicate if Crash reporting via console should be included

include/common/aarch64/el3_common_macros.S

@@ -95,10 +95,6 @@
 	 * MDCR_EL3.SPD32: Set to 0b10 to disable AArch32 Secure self-hosted
 	 *  privileged debug from S-EL1.
 	 *
-	 * MDCR_EL3.NSPB (ARM v8.2): SPE enabled in non-secure state and
-	 * disabled in secure state. Accesses to SPE registers at SEL1 generate
-	 * trap exceptions to EL3.
-	 *
 	 * MDCR_EL3.TDOSA: Set to zero so that EL2 and EL2 System register
 	 *  access to the powerdown debug registers do not trap to EL3.
 	 *
@@ -112,19 +108,6 @@
 	 */
 	mov_imm	x0, ((MDCR_EL3_RESET_VAL | MDCR_SDD_BIT | MDCR_SPD32(MDCR_SPD32_DISABLE)) \
 			& ~(MDCR_TDOSA_BIT | MDCR_TDA_BIT | MDCR_TPM_BIT))
-
-#if ENABLE_SPE_FOR_LOWER_ELS
-	/* Detect if SPE is implemented */
-	mrs	x1, id_aa64dfr0_el1
-	ubfx	x1, x1, #ID_AA64DFR0_PMS_SHIFT, #ID_AA64DFR0_PMS_LENGTH
-	cmp	x1, #0x1
-	b.ne	1f
-
-	/* Enable SPE for use by normal world */
-	orr	x0, x0, #MDCR_NSPB(MDCR_NSPB_EL1)
-1:
-#endif
-
 	msr	mdcr_el3, x0
 
 	/* ---------------------------------------------------------------------

include/lib/aarch64/arch.h

@@ -604,4 +604,9 @@
 #define PAR_ADDR_SHIFT	12
 #define PAR_ADDR_MASK	(BIT(40) - 1) /* 40-bits-wide page address */
 
+/*******************************************************************************
+ * Definitions for system register interface to SPE
+ ******************************************************************************/
+#define PMBLIMITR_EL1		S3_0_C9_C10_0
+
 #endif /* __ARCH_H__ */

include/lib/aarch64/arch_helpers.h

@@ -197,6 +197,7 @@ DEFINE_SYSOP_TYPE_FUNC(dmb, sy)
 DEFINE_SYSOP_TYPE_FUNC(dmb, st)
 DEFINE_SYSOP_TYPE_FUNC(dmb, ld)
 DEFINE_SYSOP_TYPE_FUNC(dsb, ish)
+DEFINE_SYSOP_TYPE_FUNC(dsb, nsh)
 DEFINE_SYSOP_TYPE_FUNC(dsb, ishst)
 DEFINE_SYSOP_TYPE_FUNC(dmb, ish)
 DEFINE_SYSOP_TYPE_FUNC(dmb, ishst)
@@ -301,6 +302,7 @@ DEFINE_SYSREG_READ_FUNC(isr_el1)
 DEFINE_SYSREG_READ_FUNC(ctr_el0)
 
 DEFINE_SYSREG_RW_FUNCS(mdcr_el2)
+DEFINE_SYSREG_RW_FUNCS(mdcr_el3)
 DEFINE_SYSREG_RW_FUNCS(hstr_el2)
 DEFINE_SYSREG_RW_FUNCS(cnthp_ctl_el2)
 DEFINE_SYSREG_RW_FUNCS(pmcr_el0)
@@ -320,6 +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(pmblimitr_el1, PMBLIMITR_EL1)
 
 #define IS_IN_EL(x) \
 	(GET_EL(read_CurrentEl()) == MODE_EL##x)

include/lib/el3_runtime/aarch64/context.h

@@ -313,7 +313,6 @@ CASSERT(CTX_EL3STATE_OFFSET == __builtin_offsetof(cpu_context_t, el3state_ctx),
  * Function prototypes
  ******************************************************************************/
 void el1_sysregs_context_save(el1_sys_regs_t *regs);
-void el1_sysregs_context_save_post_ops(void);
 void el1_sysregs_context_restore(el1_sys_regs_t *regs);
 #if CTX_INCLUDE_FPREGS
 void fpregs_context_save(fp_regs_t *regs);

include/lib/extensions/spe.h

@@ -0,0 +1,13 @@
+/*
+ * Copyright (c) 2017, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef __SPE_H__
+#define __SPE_H__
+
+void spe_enable(int el2_unused);
+void spe_disable(void);
+
+#endif /* __SPE_H__ */

include/plat/arm/common/plat_arm.h

@@ -227,7 +227,4 @@ int arm_execution_state_switch(unsigned int smc_fid,
 		uint32_t cookie_lo,
 		void *handle);
 
-/* Disable Statistical Profiling Extensions helper */
-void arm_disable_spe(void);
-
 #endif /* __PLAT_ARM_H__ */

lib/el3_runtime/aarch64/context.S

@@ -9,7 +9,6 @@
 #include <context.h>
 
 	.global	el1_sysregs_context_save
-	.global el1_sysregs_context_save_post_ops
 	.global	el1_sysregs_context_restore
 #if CTX_INCLUDE_FPREGS
 	.global	fpregs_context_save
@@ -109,36 +108,6 @@ func el1_sysregs_context_save
 	ret
 endfunc el1_sysregs_context_save
 
-/* -----------------------------------------------------
- * The following function strictly follows the AArch64
- * PCS to use x9-x17 (temporary caller-saved registers)
- * to do post operations after saving the EL1 system
- * register context.
- * -----------------------------------------------------
- */
-func el1_sysregs_context_save_post_ops
-#if ENABLE_SPE_FOR_LOWER_ELS
-	/* Detect if SPE is implemented */
-	mrs	x9, id_aa64dfr0_el1
-	ubfx	x9, x9, #ID_AA64DFR0_PMS_SHIFT, #ID_AA64DFR0_PMS_LENGTH
-	cmp	x9, #0x1
-	b.ne	1f
-
-	/*
-	 * Before switching from normal world to secure world
-	 * the profiling buffers need to be drained out to memory.  This is
-	 * required to avoid an invalid memory access when TTBR is switched
-	 * for entry to SEL1.
-	 */
-	.arch	armv8.2-a+profile
-	psb	csync
-	dsb	nsh
-	.arch	armv8-a
-1:
-#endif
-	ret
-endfunc el1_sysregs_context_save_post_ops
-
 /* -----------------------------------------------------
  * The following function strictly follows the AArch64
  * PCS to use x9-x17 (temporary caller-saved registers)

lib/el3_runtime/aarch64/context_mgmt.c

@@ -15,6 +15,7 @@
 #include <platform_def.h>
 #include <pubsub_events.h>
 #include <smcc_helpers.h>
+#include <spe.h>
 #include <string.h>
 #include <utils.h>
 
@@ -216,6 +217,9 @@ static void cm_init_context_common(cpu_context_t *ctx, const entry_point_info_t
 static void enable_extensions_nonsecure(int el2_unused)
 {
 #if IMAGE_BL31
+#if ENABLE_SPE_FOR_LOWER_ELS
+	spe_enable(el2_unused);
+#endif
 #endif
 }
 
@@ -354,13 +358,6 @@ void cm_prepare_el3_exit(uint32_t security_state)
 			 * relying on hw. Some fields are architecturally
 			 * UNKNOWN on reset.
 			 *
-			 * MDCR_EL2.TPMS (ARM v8.2): Do not trap statistical
-			 * profiling controls to EL2.
-			 *
-			 * MDCR_EL2.E2PB (ARM v8.2): SPE enabled in non-secure
-			 * state. Accesses to profiling buffer controls at
-			 * non-secure EL1 are not trapped to EL2.
-			 *
 			 * MDCR_EL2.TDRA: Set to zero so that Non-secure EL0 and
 			 *  EL1 System register accesses to the Debug ROM
 			 *  registers are not trapped to EL2.
@@ -397,22 +394,6 @@ void cm_prepare_el3_exit(uint32_t security_state)
 					| MDCR_EL2_HPME_BIT | MDCR_EL2_TPM_BIT
 					| MDCR_EL2_TPMCR_BIT));
 
-#if ENABLE_SPE_FOR_LOWER_ELS
-			uint64_t id_aa64dfr0_el1;
-
-			/* Detect if SPE is implemented */
-			id_aa64dfr0_el1 = read_id_aa64dfr0_el1() >>
-				ID_AA64DFR0_PMS_SHIFT;
-			if ((id_aa64dfr0_el1 & ID_AA64DFR0_PMS_MASK) == 1) {
-				/*
-				 * Make sure traps to EL2 are not generated if
-				 * EL2 is implemented but not used.
-				 */
-				mdcr_el2 &= ~MDCR_EL2_TPMS;
-				mdcr_el2 |= MDCR_EL2_E2PB(MDCR_EL2_E2PB_EL1);
-			}
-#endif
-
 			write_mdcr_el2(mdcr_el2);
 
 			/*
@@ -454,7 +435,6 @@ void cm_el1_sysregs_context_save(uint32_t security_state)
 	assert(ctx);
 
 	el1_sysregs_context_save(get_sysregs_ctx(ctx));
-	el1_sysregs_context_save_post_ops();
 
 #if IMAGE_BL31
 	if (security_state == SECURE)

lib/extensions/spe/spe.c

@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2017, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <arch.h>
+#include <arch_helpers.h>
+#include <pubsub.h>
+
+/*
+ * The assembler does not yet understand the psb csync mnemonic
+ * so use the equivalent hint instruction.
+ */
+#define psb_csync()	asm volatile("hint #17")
+
+void spe_enable(int el2_unused)
+{
+	uint64_t features;
+
+	features = read_id_aa64dfr0_el1() >> ID_AA64DFR0_PMS_SHIFT;
+	if ((features & ID_AA64DFR0_PMS_MASK) == 1) {
+		uint64_t v;
+
+		if (el2_unused) {
+			/*
+			 * MDCR_EL2.TPMS (ARM v8.2): Do not trap statistical
+			 * profiling controls to EL2.
+			 *
+			 * MDCR_EL2.E2PB (ARM v8.2): SPE enabled in Non-secure
+			 * state. Accesses to profiling buffer controls at
+			 * Non-secure EL1 are not trapped to EL2.
+			 */
+			v = read_mdcr_el2();
+			v &= ~MDCR_EL2_TPMS;
+			v |= MDCR_EL2_E2PB(MDCR_EL2_E2PB_EL1);
+			write_mdcr_el2(v);
+		}
+
+		/*
+		 * MDCR_EL2.NSPB (ARM v8.2): SPE enabled in Non-secure state
+		 * and disabled in secure state. Accesses to SPE registers at
+		 * S-EL1 generate trap exceptions to EL3.
+		 */
+		v = read_mdcr_el3();
+		v |= MDCR_NSPB(MDCR_NSPB_EL1);
+		write_mdcr_el3(v);
+	}
+}
+
+void spe_disable(void)
+{
+	uint64_t features;
+
+	features = read_id_aa64dfr0_el1() >> ID_AA64DFR0_PMS_SHIFT;
+	if ((features & ID_AA64DFR0_PMS_MASK) == 1) {
+		uint64_t v;
+
+		/* Drain buffered data */
+		psb_csync();
+		dsbnsh();
+
+		/* Disable profiling buffer */
+		v = read_pmblimitr_el1();
+		v &= ~(1ULL << 0);
+		write_pmblimitr_el1(v);
+		isb();
+	}
+}
+
+static void *spe_drain_buffers_hook(const void *arg)
+{
+	uint64_t features;
+
+	features = read_id_aa64dfr0_el1() >> ID_AA64DFR0_PMS_SHIFT;
+	if ((features & ID_AA64DFR0_PMS_MASK) == 1) {
+		/* Drain buffered data */
+		psb_csync();
+		dsbnsh();
+	}
+
+	return 0;
+}
+
+SUBSCRIBE_TO_EVENT(cm_entering_secure_world, spe_drain_buffers_hook);

plat/arm/board/fvp/fvp_pm.c

@@ -14,6 +14,7 @@
 #include <plat_arm.h>
 #include <platform.h>
 #include <psci.h>
+#include <spe.h>
 #include <v2m_def.h>
 #include "drivers/pwrc/fvp_pwrc.h"
 #include "fvp_def.h"
@@ -57,7 +58,7 @@ static void fvp_cluster_pwrdwn_common(void)
 	 * On power down we need to disable statistical profiling extensions
 	 * before exiting coherency.
 	 */
-	arm_disable_spe();
+	spe_disable();
 #endif
 
 	/* Disable coherency if this cluster is to be turned off */

plat/arm/common/aarch64/arm_helpers.S

@@ -12,7 +12,6 @@
 	.globl	plat_crash_console_putc
 	.globl	plat_crash_console_flush
 	.globl	platform_mem_init
-	.globl	arm_disable_spe
 
 
 	/* -----------------------------------------------------
@@ -88,34 +87,6 @@ func platform_mem_init
 	ret
 endfunc platform_mem_init
 
-	/* -----------------------------------------------------
-	 * void arm_disable_spe (void);
-	 * -----------------------------------------------------
-	 */
-#if ENABLE_SPE_FOR_LOWER_ELS
-func arm_disable_spe
-	/* Detect if SPE is implemented */
-	mrs	x0, id_aa64dfr0_el1
-	ubfx	x0, x0, #ID_AA64DFR0_PMS_SHIFT, #ID_AA64DFR0_PMS_LENGTH
-	cmp	x0, #0x1
-	b.ne	1f
-
-	/* Drain buffered data */
-	.arch	armv8.2-a+profile
-	psb	csync
-	dsb	nsh
-
-	/* Disable Profiling Buffer */
-	mrs	x0, pmblimitr_el1
-	bic	x0, x0, #1
-	msr	pmblimitr_el1, x0
-	isb
-	.arch	armv8-a
-1:
-	ret
-endfunc arm_disable_spe
-#endif
-
 /*
  * Need to use coherent stack when ARM Cryptocell is used to autheticate images
  * since Cryptocell uses DMA to transfer data and it is not coherent with the