Workaround for Cortex A78 erratum 1951500

Cortex A78 erratum 1951500 is a Cat B erratum that applies to revisions
r0p0, r1p0, and r1p1.  The workaround is to insert a DMB ST before
acquire atomic instructions without release semantics.  This workaround
works on revisions r1p0 and r1p1, in r0p0 there is no workaround.

SDEN can be found here:
https://documentation-service.arm.com/static/5fb66157ca04df4095c1cc2e

Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: I47610cee75af6a127ea65edc4d5cffc7e6a2d0a3
diff --git a/lib/cpus/aarch64/cortex_a78.S b/lib/cpus/aarch64/cortex_a78.S
index ef760ed..f61726b 100644
--- a/lib/cpus/aarch64/cortex_a78.S
+++ b/lib/cpus/aarch64/cortex_a78.S
@@ -72,6 +72,60 @@
 	b	cpu_rev_var_ls
 endfunc check_errata_1941498
 
+	/* --------------------------------------------------
+	 * Errata Workaround for A78 Erratum 1951500.
+	 * This applies to revisions r1p0 and r1p1 of A78.
+	 * The issue also exists in r0p0 but there is no fix
+	 * in that revision.
+	 * Inputs:
+	 * x0: variant[4:7] and revision[0:3] of current cpu.
+	 * Shall clobber: x0-x17
+	 * --------------------------------------------------
+	 */
+func errata_a78_1951500_wa
+	/* Compare x0 against revisions r1p0 - r1p1 */
+	mov	x17, x30
+	bl	check_errata_1951500
+	cbz	x0, 1f
+
+	msr	S3_6_c15_c8_0, xzr
+	ldr	x0, =0x10E3900002
+	msr	S3_6_c15_c8_2, x0
+	ldr	x0, =0x10FFF00083
+	msr	S3_6_c15_c8_3, x0
+	ldr	x0, =0x2001003FF
+	msr	S3_6_c15_c8_1, x0
+
+	mov	x0, #1
+	msr	S3_6_c15_c8_0, x0
+	ldr	x0, =0x10E3800082
+	msr	S3_6_c15_c8_2, x0
+	ldr	x0, =0x10FFF00083
+	msr	S3_6_c15_c8_3, x0
+	ldr	x0, =0x2001003FF
+	msr	S3_6_c15_c8_1, x0
+
+	mov	x0, #2
+	msr	S3_6_c15_c8_0, x0
+	ldr	x0, =0x10E3800200
+	msr	S3_6_c15_c8_2, x0
+	ldr	x0, =0x10FFF003E0
+	msr	S3_6_c15_c8_3, x0
+	ldr	x0, =0x2001003FF
+	msr	S3_6_c15_c8_1, x0
+
+	isb
+1:
+	ret	x17
+endfunc errata_a78_1951500_wa
+
+func check_errata_1951500
+	/* Applies to revisions r1p0 and r1p1. */
+	mov	x1, #CPU_REV(1, 0)
+	mov	x2, #CPU_REV(1, 1)
+	b	cpu_rev_var_range
+endfunc check_errata_1951500
+
 	/* -------------------------------------------------
 	 * The CPU Ops reset function for Cortex-A78
 	 * -------------------------------------------------
@@ -91,6 +145,11 @@
 	bl	errata_a78_1941498_wa
 #endif
 
+#if ERRATA_A78_1951500
+	mov	x0, x18
+	bl	errata_a78_1951500_wa
+#endif
+
 #if ENABLE_AMU
 	/* Make sure accesses from EL0/EL1 and EL2 are not trapped to EL3 */
 	mrs	x0, actlr_el3
@@ -147,6 +206,7 @@
 	 */
 	report_errata ERRATA_A78_1688305, cortex_a78, 1688305
 	report_errata ERRATA_A78_1941498, cortex_a78, 1941498
+	report_errata ERRATA_A78_1951500, cortex_a78, 1951500
 
 	ldp	x8, x30, [sp], #16
 	ret
diff --git a/lib/cpus/cpu-ops.mk b/lib/cpus/cpu-ops.mk
index b7dec0b..93b538c 100644
--- a/lib/cpus/cpu-ops.mk
+++ b/lib/cpus/cpu-ops.mk
@@ -298,6 +298,11 @@
 # to revisions r0p0, r1p0, and r1p1 of the A78 cpu.
 ERRATA_A78_1941498	?=0
 
+# Flag to apply erratum 1951500 workaround during reset. This erratum applies
+# to revisions r1p0 and r1p1 of the A78 cpu.  The issue is present in r0p0 as
+# well but there is no workaround for that revision.
+ERRATA_A78_1951500	?=0
+
 # Flag to apply T32 CLREX workaround during reset. This erratum applies
 # only to r0p0 and r1p0 of the Neoverse N1 cpu.
 ERRATA_N1_1043202	?=0
@@ -583,6 +588,10 @@
 $(eval $(call assert_boolean,ERRATA_A78_1941498))
 $(eval $(call add_define,ERRATA_A78_1941498))
 
+# Process ERRATA_A78_1951500 flag
+$(eval $(call assert_boolean,ERRATA_A78_1951500))
+$(eval $(call add_define,ERRATA_A78_1951500))
+
 # Process ERRATA_N1_1043202 flag
 $(eval $(call assert_boolean,ERRATA_N1_1043202))
 $(eval $(call add_define,ERRATA_N1_1043202))