diff --git a/arch/xtensa/Kconfig.debug b/arch/xtensa/Kconfig.debug
index be5fb4c..a34010e 100644
--- a/arch/xtensa/Kconfig.debug
+++ b/arch/xtensa/Kconfig.debug
@@ -13,4 +13,15 @@
 	  Enabling this option improves the link time but increases the
 	  code size, and possibly execution time.
 
+config S32C1I_SELFTEST
+	bool "Perform S32C1I instruction self-test at boot"
+	default y
+	help
+	  Enable this option to test S32C1I instruction behavior at boot.
+	  Correct operation of this instruction requires some cooperation from hardware
+	  external to the processor (such as bus bridge, bus fabric, or memory controller).
+	  It is easy to make wrong hardware configuration, this test should catch it early.
+
+	  Say 'N' on stable hardware.
+
 endmenu
diff --git a/arch/xtensa/include/asm/regs.h b/arch/xtensa/include/asm/regs.h
index 8a8aa61..6aaf6d6 100644
--- a/arch/xtensa/include/asm/regs.h
+++ b/arch/xtensa/include/asm/regs.h
@@ -52,6 +52,10 @@
 #define EXCCAUSE_SPECULATION			7
 #define EXCCAUSE_PRIVILEGED			8
 #define EXCCAUSE_UNALIGNED			9
+#define EXCCAUSE_INSTR_DATA_ERROR		12
+#define EXCCAUSE_LOAD_STORE_DATA_ERROR		13
+#define EXCCAUSE_INSTR_ADDR_ERROR		14
+#define EXCCAUSE_LOAD_STORE_ADDR_ERROR		15
 #define EXCCAUSE_ITLB_MISS			16
 #define EXCCAUSE_ITLB_MULTIHIT			17
 #define EXCCAUSE_ITLB_PRIVILEGE			18
diff --git a/arch/xtensa/kernel/setup.c b/arch/xtensa/kernel/setup.c
index e53a94b..4521761 100644
--- a/arch/xtensa/kernel/setup.c
+++ b/arch/xtensa/kernel/setup.c
@@ -42,6 +42,7 @@
 #include <asm/page.h>
 #include <asm/setup.h>
 #include <asm/param.h>
+#include <asm/traps.h>
 
 #include <platform/hardware.h>
 
@@ -235,6 +236,123 @@
 extern char _DoubleExceptionVector_literal_start;
 extern char _DoubleExceptionVector_text_end;
 
+
+#ifdef CONFIG_S32C1I_SELFTEST
+#if XCHAL_HAVE_S32C1I
+
+static int __initdata rcw_word, rcw_probe_pc, rcw_exc;
+
+/*
+ * Basic atomic compare-and-swap, that records PC of S32C1I for probing.
+ *
+ * If *v == cmp, set *v = set.  Return previous *v.
+ */
+static inline int probed_compare_swap(int *v, int cmp, int set)
+{
+	int tmp;
+
+	__asm__ __volatile__(
+			"	movi	%1, 1f\n"
+			"	s32i	%1, %4, 0\n"
+			"	wsr	%2, scompare1\n"
+			"1:	s32c1i	%0, %3, 0\n"
+			: "=a" (set), "=&a" (tmp)
+			: "a" (cmp), "a" (v), "a" (&rcw_probe_pc), "0" (set)
+			: "memory"
+			);
+	return set;
+}
+
+/* Handle probed exception */
+
+void __init do_probed_exception(struct pt_regs *regs, unsigned long exccause)
+{
+	if (regs->pc == rcw_probe_pc) {	/* exception on s32c1i ? */
+		regs->pc += 3;		/* skip the s32c1i instruction */
+		rcw_exc = exccause;
+	} else {
+		do_unhandled(regs, exccause);
+	}
+}
+
+/* Simple test of S32C1I (soc bringup assist) */
+
+void __init check_s32c1i(void)
+{
+	int n, cause1, cause2;
+	void *handbus, *handdata, *handaddr; /* temporarily saved handlers */
+
+	rcw_probe_pc = 0;
+	handbus  = trap_set_handler(EXCCAUSE_LOAD_STORE_ERROR,
+			do_probed_exception);
+	handdata = trap_set_handler(EXCCAUSE_LOAD_STORE_DATA_ERROR,
+			do_probed_exception);
+	handaddr = trap_set_handler(EXCCAUSE_LOAD_STORE_ADDR_ERROR,
+			do_probed_exception);
+
+	/* First try an S32C1I that does not store: */
+	rcw_exc = 0;
+	rcw_word = 1;
+	n = probed_compare_swap(&rcw_word, 0, 2);
+	cause1 = rcw_exc;
+
+	/* took exception? */
+	if (cause1 != 0) {
+		/* unclean exception? */
+		if (n != 2 || rcw_word != 1)
+			panic("S32C1I exception error");
+	} else if (rcw_word != 1 || n != 1) {
+		panic("S32C1I compare error");
+	}
+
+	/* Then an S32C1I that stores: */
+	rcw_exc = 0;
+	rcw_word = 0x1234567;
+	n = probed_compare_swap(&rcw_word, 0x1234567, 0xabcde);
+	cause2 = rcw_exc;
+
+	if (cause2 != 0) {
+		/* unclean exception? */
+		if (n != 0xabcde || rcw_word != 0x1234567)
+			panic("S32C1I exception error (b)");
+	} else if (rcw_word != 0xabcde || n != 0x1234567) {
+		panic("S32C1I store error");
+	}
+
+	/* Verify consistency of exceptions: */
+	if (cause1 || cause2) {
+		pr_warn("S32C1I took exception %d, %d\n", cause1, cause2);
+		/* If emulation of S32C1I upon bus error gets implemented,
+		   we can get rid of this panic for single core (not SMP) */
+		panic("S32C1I exceptions not currently supported");
+	}
+	if (cause1 != cause2)
+		panic("inconsistent S32C1I exceptions");
+
+	trap_set_handler(EXCCAUSE_LOAD_STORE_ERROR, handbus);
+	trap_set_handler(EXCCAUSE_LOAD_STORE_DATA_ERROR, handdata);
+	trap_set_handler(EXCCAUSE_LOAD_STORE_ADDR_ERROR, handaddr);
+}
+
+#else /* XCHAL_HAVE_S32C1I */
+
+/* This condition should not occur with a commercially deployed processor.
+   Display reminder for early engr test or demo chips / FPGA bitstreams */
+void __init check_s32c1i(void)
+{
+	pr_warn("Processor configuration lacks atomic compare-and-swap support!\n");
+}
+
+#endif /* XCHAL_HAVE_S32C1I */
+#else /* CONFIG_S32C1I_SELFTEST */
+
+void __init check_s32c1i(void)
+{
+}
+
+#endif /* CONFIG_S32C1I_SELFTEST */
+
+
 void __init setup_arch(char **cmdline_p)
 {
 	extern int mem_reserve(unsigned long, unsigned long, int);
@@ -244,6 +362,8 @@
 	boot_command_line[COMMAND_LINE_SIZE-1] = '\0';
 	*cmdline_p = command_line;
 
+	check_s32c1i();
+
 	/* Reserve some memory regions */
 
 #ifdef CONFIG_BLK_DEV_INITRD