Neoverse N1: Forces cacheable atomic to near

This patch forces all cacheable atomic instructions to be near, which
improves performance in highly contended parallelized use-cases.

Change-Id: I93fac62847f4af8d5eaaf3b52318c30893e947d3
Signed-off-by: Louis Mayencourt <louis.mayencourt@arm.com>

include/lib/cpus/aarch64/neoverse_n1.h

@@ -13,10 +13,9 @@
 #define NEOVERSE_N1_MIDR		U(0x410fd0c0)
 
 /*******************************************************************************
- * CPU Extended Control register specific definitions.
+ * CPU Power Control register specific definitions.
  ******************************************************************************/
 #define NEOVERSE_N1_CPUPWRCTLR_EL1	S3_0_C15_C2_7
-#define NEOVERSE_N1_CPUECTLR_EL1	S3_0_C15_C1_4
 
 /* Definitions of register field mask in NEOVERSE_N1_CPUPWRCTLR_EL1 */
 #define NEOVERSE_N1_CORE_PWRDN_EN_MASK	U(0x1)
@@ -26,6 +25,18 @@
 #define NEOVERSE_N1_AMU_NR_COUNTERS	U(5)
 #define NEOVERSE_N1_AMU_GROUP0_MASK	U(0x1f)
 
+/*******************************************************************************
+ * CPU Extended Control register specific definitions.
+ ******************************************************************************/
+#define NEOVERSE_N1_CPUECTLR_EL1	S3_0_C15_C1_4
+
+/*******************************************************************************
+ * CPU Auxiliary Control register specific definitions.
+ ******************************************************************************/
+#define NEOVERSE_N1_CPUACTLR2_EL1	S3_0_C15_C1_1
+
+#define NEOVERSE_N1_CPUACTLR2_EL1_BIT_2	(ULL(1) << 2)
+
 /* Instruction patching registers */
 #define CPUPSELR_EL3	S3_6_C15_C8_0
 #define CPUPCR_EL3	S3_6_C15_C8_1

lib/cpus/aarch64/neoverse_n1.S

@@ -50,6 +50,12 @@ func neoverse_n1_reset_func
 	/* Disables speculative loads */
 	msr	SSBS, xzr
 
+	/* Forces all cacheable atomic instructions to be near */
+	mrs	x0, NEOVERSE_N1_CPUACTLR2_EL1
+	orr	x0, x0, #NEOVERSE_N1_CPUACTLR2_EL1_BIT_2
+	msr	NEOVERSE_N1_CPUACTLR2_EL1, x0
+	isb
+
 	bl	cpu_get_rev_var
 	mov	x18, x0