KVM: riscv: selftests: Prove Zawrs support works

The proof is that the test will run much, much faster with
the --zawrs command line option given than without.

Signed-off-by: Andrew Jones <ajones@ventanamicro.com>

tools/testing/selftests/kvm/Makefile

@@ -189,6 +189,7 @@ TEST_GEN_PROGS_s390x += rseq_test
 TEST_GEN_PROGS_s390x += set_memory_region_test
 TEST_GEN_PROGS_s390x += kvm_binary_stats_test
 
+TEST_GEN_PROGS_riscv += riscv/zawrs
 TEST_GEN_PROGS_riscv += arch_timer
 TEST_GEN_PROGS_riscv += demand_paging_test
 TEST_GEN_PROGS_riscv += dirty_log_test

tools/testing/selftests/kvm/riscv/zawrs.c

@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <sched.h>
+#include <pthread.h>
+#include "kvm_util.h"
+#include "test_util.h"
+
+static uint32_t count;
+static bool use_zawrs;
+
+static void wrs_nto(void)
+{
+	if (!use_zawrs)
+		return;
+
+	asm volatile(
+	".option push\n"
+	".option norvc\n"
+	".option norelax\n"
+	".long 0x00d00073\n"
+	".option pop\n"
+	);
+}
+
+static void guest_code(void)
+{
+	uint32_t cpu = guest_get_vcpuid();
+	uint32_t tmp = 0;
+
+	while (tmp < 4000) {
+		tmp = READ_ONCE(count);
+		if ((tmp & 1) == cpu)
+			WRITE_ONCE(count, tmp + 1);
+		wrs_nto();
+	}
+
+	GUEST_DONE();
+}
+
+static void *run_test(void *arg)
+{
+	struct kvm_vcpu *vcpu = arg;
+
+	vcpu_run(vcpu);
+
+	assert(vcpu->run->exit_reason == UCALL_EXIT_REASON);
+	assert(get_ucall(vcpu, NULL) == UCALL_DONE);
+
+	return NULL;
+}
+
+int main(int ac, char **av)
+{
+	struct kvm_vcpu *vcpus[2];
+	struct kvm_vm *vm;
+	pthread_t threads[2];
+	pthread_attr_t attr;
+	cpu_set_t cpuset;
+	int i, ret;
+
+	vm = vm_create_with_vcpus(2, guest_code, vcpus);
+
+	if (ac > 1 && !strcmp(av[1], "--zawrs")) {
+		use_zawrs = true;
+		sync_global_to_guest(vm, use_zawrs);
+	}
+
+	/* Force both vcpus onto cpu0 */
+	CPU_ZERO(&cpuset);
+	CPU_SET(0, &cpuset);
+	pthread_attr_init(&attr);
+	pthread_attr_setaffinity_np(&attr, sizeof(cpuset), &cpuset);
+
+	for (i = 0; i < 2; ++i) {
+		ret = pthread_create(&threads[i], &attr, run_test, vcpus[i]);
+		assert(ret == 0);
+	}
+	for (i = 0; i < 2; ++i)
+		pthread_join(threads[i], NULL);
+
+	kvm_vm_free(vm);
+	return 0;
+}