Merge tag 'kvm-s390-next-6.3-1' of https://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into HEAD

* Two more V!=R patches
* The last part of the cmpxchg patches
* A few fixes

2 files changed, 527 insertions, 148 deletions

diff --git a/tools/testing/selftests/kvm/Makefile b/tools/testing/selftests/kvm/Makefile
index 844417601618..84a627c43795 100644
--- a/tools/testing/selftests/kvm/Makefile
+++ b/tools/testing/selftests/kvm/Makefile
@@ -201,6 +201,9 @@ CFLAGS += -Wall -Wstrict-prototypes -Wuninitialized -O2 -g -std=gnu99 \
 	-I$(LINUX_TOOL_ARCH_INCLUDE) -I$(LINUX_HDR_PATH) -Iinclude \
 	-I$(<D) -Iinclude/$(ARCH_DIR) -I ../rseq -I.. $(EXTRA_CFLAGS) \
 	$(KHDR_INCLUDES)
+ifeq ($(ARCH),s390)
+	CFLAGS += -march=z10
+endif
 
 no-pie-option := $(call try-run, echo 'int main(void) { return 0; }' | \
         $(CC) -Werror $(CFLAGS) -no-pie -x c - -o "$$TMP", -no-pie)
diff --git a/tools/testing/selftests/kvm/s390x/memop.c b/tools/testing/selftests/kvm/s390x/memop.c
index 3fd81e58f40c..8e4b94d7b8dd 100644
--- a/tools/testing/selftests/kvm/s390x/memop.c
+++ b/tools/testing/selftests/kvm/s390x/memop.c
@@ -9,6 +9,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/ioctl.h>
+#include <pthread.h>
 
 #include <linux/bits.h>
 
@@ -26,6 +27,7 @@ enum mop_target {
 enum mop_access_mode {
 	READ,
 	WRITE,
+	CMPXCHG,
 };
 
 struct mop_desc {
@@ -44,57 +46,67 @@ struct mop_desc {
 	enum mop_access_mode mode;
 	void *buf;
 	uint32_t sida_offset;
+	void *old;
+	uint8_t old_value[16];
+	bool *cmpxchg_success;
 	uint8_t ar;
 	uint8_t key;
 };
 
-static struct kvm_s390_mem_op ksmo_from_desc(struct mop_desc desc)
+const uint8_t NO_KEY = 0xff;
+
+static struct kvm_s390_mem_op ksmo_from_desc(struct mop_desc *desc)
 {
 	struct kvm_s390_mem_op ksmo = {
-		.gaddr = (uintptr_t)desc.gaddr,
-		.size = desc.size,
-		.buf = ((uintptr_t)desc.buf),
+		.gaddr = (uintptr_t)desc->gaddr,
+		.size = desc->size,
+		.buf = ((uintptr_t)desc->buf),
 		.reserved = "ignored_ignored_ignored_ignored"
 	};
 
-	switch (desc.target) {
+	switch (desc->target) {
 	case LOGICAL:
-		if (desc.mode == READ)
+		if (desc->mode == READ)
 			ksmo.op = KVM_S390_MEMOP_LOGICAL_READ;
-		if (desc.mode == WRITE)
+		if (desc->mode == WRITE)
 			ksmo.op = KVM_S390_MEMOP_LOGICAL_WRITE;
 		break;
 	case SIDA:
-		if (desc.mode == READ)
+		if (desc->mode == READ)
 			ksmo.op = KVM_S390_MEMOP_SIDA_READ;
-		if (desc.mode == WRITE)
+		if (desc->mode == WRITE)
 			ksmo.op = KVM_S390_MEMOP_SIDA_WRITE;
 		break;
 	case ABSOLUTE:
-		if (desc.mode == READ)
+		if (desc->mode == READ)
 			ksmo.op = KVM_S390_MEMOP_ABSOLUTE_READ;
-		if (desc.mode == WRITE)
+		if (desc->mode == WRITE)
 			ksmo.op = KVM_S390_MEMOP_ABSOLUTE_WRITE;
+		if (desc->mode == CMPXCHG) {
+			ksmo.op = KVM_S390_MEMOP_ABSOLUTE_CMPXCHG;
+			ksmo.old_addr = (uint64_t)desc->old;
+			memcpy(desc->old_value, desc->old, desc->size);
+		}
 		break;
 	case INVALID:
 		ksmo.op = -1;
 	}
-	if (desc.f_check)
+	if (desc->f_check)
 		ksmo.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
-	if (desc.f_inject)
+	if (desc->f_inject)
 		ksmo.flags |= KVM_S390_MEMOP_F_INJECT_EXCEPTION;
-	if (desc._set_flags)
-		ksmo.flags = desc.set_flags;
-	if (desc.f_key) {
+	if (desc->_set_flags)
+		ksmo.flags = desc->set_flags;
+	if (desc->f_key && desc->key != NO_KEY) {
 		ksmo.flags |= KVM_S390_MEMOP_F_SKEY_PROTECTION;
-		ksmo.key = desc.key;
+		ksmo.key = desc->key;
 	}
-	if (desc._ar)
-		ksmo.ar = desc.ar;
+	if (desc->_ar)
+		ksmo.ar = desc->ar;
 	else
 		ksmo.ar = 0;
-	if (desc._sida_offset)
-		ksmo.sida_offset = desc.sida_offset;
+	if (desc->_sida_offset)
+		ksmo.sida_offset = desc->sida_offset;
 
 	return ksmo;
 }
@@ -133,9 +145,13 @@ static void print_memop(struct kvm_vcpu *vcpu, const struct kvm_s390_mem_op *ksm
 	case KVM_S390_MEMOP_ABSOLUTE_WRITE:
 		printf("ABSOLUTE, WRITE, ");
 		break;
+	case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG:
+		printf("ABSOLUTE, CMPXCHG, ");
+		break;
 	}
-	printf("gaddr=%llu, size=%u, buf=%llu, ar=%u, key=%u",
-	       ksmo->gaddr, ksmo->size, ksmo->buf, ksmo->ar, ksmo->key);
+	printf("gaddr=%llu, size=%u, buf=%llu, ar=%u, key=%u, old_addr=%llx",
+	       ksmo->gaddr, ksmo->size, ksmo->buf, ksmo->ar, ksmo->key,
+	       ksmo->old_addr);
 	if (ksmo->flags & KVM_S390_MEMOP_F_CHECK_ONLY)
 		printf(", CHECK_ONLY");
 	if (ksmo->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION)
@@ -145,24 +161,30 @@ static void print_memop(struct kvm_vcpu *vcpu, const struct kvm_s390_mem_op *ksm
 	puts(")");
 }
 
-static void memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo)
+static int err_memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo,
+			   struct mop_desc *desc)
 {
 	struct kvm_vcpu *vcpu = info.vcpu;
 
 	if (!vcpu)
-		vm_ioctl(info.vm, KVM_S390_MEM_OP, ksmo);
+		return __vm_ioctl(info.vm, KVM_S390_MEM_OP, ksmo);
 	else
-		vcpu_ioctl(vcpu, KVM_S390_MEM_OP, ksmo);
+		return __vcpu_ioctl(vcpu, KVM_S390_MEM_OP, ksmo);
 }
 
-static int err_memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo)
+static void memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo,
+			struct mop_desc *desc)
 {
-	struct kvm_vcpu *vcpu = info.vcpu;
+	int r;
 
-	if (!vcpu)
-		return __vm_ioctl(info.vm, KVM_S390_MEM_OP, ksmo);
-	else
-		return __vcpu_ioctl(vcpu, KVM_S390_MEM_OP, ksmo);
+	r = err_memop_ioctl(info, ksmo, desc);
+	if (ksmo->op == KVM_S390_MEMOP_ABSOLUTE_CMPXCHG) {
+		if (desc->cmpxchg_success) {
+			int diff = memcmp(desc->old_value, desc->old, desc->size);
+			*desc->cmpxchg_success = !diff;
+		}
+	}
+	TEST_ASSERT(!r, __KVM_IOCTL_ERROR("KVM_S390_MEM_OP", r));
 }
 
 #define MEMOP(err, info_p, mop_target_p, access_mode_p, buf_p, size_p, ...)	\
@@ -183,9 +205,9 @@ static int err_memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo)
 		else								\
 			__desc.gaddr = __desc.gaddr_v;				\
 	}									\
-	__ksmo = ksmo_from_desc(__desc);					\
+	__ksmo = ksmo_from_desc(&__desc);					\
 	print_memop(__info.vcpu, &__ksmo);					\
-	err##memop_ioctl(__info, &__ksmo);					\
+	err##memop_ioctl(__info, &__ksmo, &__desc);				\
 })
 
 #define MOP(...) MEMOP(, __VA_ARGS__)
@@ -199,6 +221,8 @@ static int err_memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo)
 #define AR(a) ._ar = 1, .ar = (a)
 #define KEY(a) .f_key = 1, .key = (a)
 #define INJECT .f_inject = 1
+#define CMPXCHG_OLD(o) .old = (o)
+#define CMPXCHG_SUCCESS(s) .cmpxchg_success = (s)
 
 #define CHECK_N_DO(f, ...) ({ f(__VA_ARGS__, CHECK_ONLY); f(__VA_ARGS__); })
 
@@ -208,8 +232,8 @@ static int err_memop_ioctl(struct test_info info, struct kvm_s390_mem_op *ksmo)
 #define CR0_FETCH_PROTECTION_OVERRIDE	(1UL << (63 - 38))
 #define CR0_STORAGE_PROTECTION_OVERRIDE	(1UL << (63 - 39))
 
-static uint8_t mem1[65536];
-static uint8_t mem2[65536];
+static uint8_t __aligned(PAGE_SIZE) mem1[65536];
+static uint8_t __aligned(PAGE_SIZE) mem2[65536];
 
 struct test_default {
 	struct kvm_vm *kvm_vm;
@@ -241,6 +265,8 @@ enum stage {
 	STAGE_SKEYS_SET,
 	/* Guest copied memory (locations up to test case) */
 	STAGE_COPIED,
+	/* End of guest code reached */
+	STAGE_DONE,
 };
 
 #define HOST_SYNC(info_p, stage)					\
@@ -252,6 +278,9 @@ enum stage {
 									\
 	vcpu_run(__vcpu);						\
 	get_ucall(__vcpu, &uc);						\
+	if (uc.cmd == UCALL_ABORT) {					\
+		REPORT_GUEST_ASSERT_2(uc, "hints: %lu, %lu");		\
+	}								\
 	ASSERT_EQ(uc.cmd, UCALL_SYNC);					\
 	ASSERT_EQ(uc.args[1], __stage);					\
 })									\
@@ -268,34 +297,66 @@ static void prepare_mem12(void)
 #define ASSERT_MEM_EQ(p1, p2, size) \
 	TEST_ASSERT(!memcmp(p1, p2, size), "Memory contents do not match!")
 
-#define DEFAULT_WRITE_READ(copy_cpu, mop_cpu, mop_target_p, size, ...)		\
-({										\
-	struct test_info __copy_cpu = (copy_cpu), __mop_cpu = (mop_cpu);	\
-	enum mop_target __target = (mop_target_p);				\
-	uint32_t __size = (size);						\
-										\
-	prepare_mem12();							\
-	CHECK_N_DO(MOP, __mop_cpu, __target, WRITE, mem1, __size,		\
-			GADDR_V(mem1), ##__VA_ARGS__);				\
-	HOST_SYNC(__copy_cpu, STAGE_COPIED);					\
-	CHECK_N_DO(MOP, __mop_cpu, __target, READ, mem2, __size,		\
-			GADDR_V(mem2), ##__VA_ARGS__);				\
-	ASSERT_MEM_EQ(mem1, mem2, __size);					\
-})
+static void default_write_read(struct test_info copy_cpu, struct test_info mop_cpu,
+			       enum mop_target mop_target, uint32_t size, uint8_t key)
+{
+	prepare_mem12();
+	CHECK_N_DO(MOP, mop_cpu, mop_target, WRITE, mem1, size,
+		   GADDR_V(mem1), KEY(key));
+	HOST_SYNC(copy_cpu, STAGE_COPIED);
+	CHECK_N_DO(MOP, mop_cpu, mop_target, READ, mem2, size,
+		   GADDR_V(mem2), KEY(key));
+	ASSERT_MEM_EQ(mem1, mem2, size);
+}
 
-#define DEFAULT_READ(copy_cpu, mop_cpu, mop_target_p, size, ...)		\
-({										\
-	struct test_info __copy_cpu = (copy_cpu), __mop_cpu = (mop_cpu);	\
-	enum mop_target __target = (mop_target_p);				\
-	uint32_t __size = (size);						\
-										\
-	prepare_mem12();							\
-	CHECK_N_DO(MOP, __mop_cpu, __target, WRITE, mem1, __size,		\
-			GADDR_V(mem1));						\
-	HOST_SYNC(__copy_cpu, STAGE_COPIED);					\
-	CHECK_N_DO(MOP, __mop_cpu, __target, READ, mem2, __size, ##__VA_ARGS__);\
-	ASSERT_MEM_EQ(mem1, mem2, __size);					\
-})
+static void default_read(struct test_info copy_cpu, struct test_info mop_cpu,
+			 enum mop_target mop_target, uint32_t size, uint8_t key)
+{
+	prepare_mem12();
+	CHECK_N_DO(MOP, mop_cpu, mop_target, WRITE, mem1, size, GADDR_V(mem1));
+	HOST_SYNC(copy_cpu, STAGE_COPIED);
+	CHECK_N_DO(MOP, mop_cpu, mop_target, READ, mem2, size,
+		   GADDR_V(mem2), KEY(key));
+	ASSERT_MEM_EQ(mem1, mem2, size);
+}
+
+static void default_cmpxchg(struct test_default *test, uint8_t key)
+{
+	for (int size = 1; size <= 16; size *= 2) {
+		for (int offset = 0; offset < 16; offset += size) {
+			uint8_t __aligned(16) new[16] = {};
+			uint8_t __aligned(16) old[16];
+			bool succ;
+
+			prepare_mem12();
+			default_write_read(test->vcpu, test->vcpu, LOGICAL, 16, NO_KEY);
+
+			memcpy(&old, mem1, 16);
+			MOP(test->vm, ABSOLUTE, CMPXCHG, new + offset,
+			    size, GADDR_V(mem1 + offset),
+			    CMPXCHG_OLD(old + offset),
+			    CMPXCHG_SUCCESS(&succ), KEY(key));
+			HOST_SYNC(test->vcpu, STAGE_COPIED);
+			MOP(test->vm, ABSOLUTE, READ, mem2, 16, GADDR_V(mem2));
+			TEST_ASSERT(succ, "exchange of values should succeed");
+			memcpy(mem1 + offset, new + offset, size);
+			ASSERT_MEM_EQ(mem1, mem2, 16);
+
+			memcpy(&old, mem1, 16);
+			new[offset]++;
+			old[offset]++;
+			MOP(test->vm, ABSOLUTE, CMPXCHG, new + offset,
+			    size, GADDR_V(mem1 + offset),
+			    CMPXCHG_OLD(old + offset),
+			    CMPXCHG_SUCCESS(&succ), KEY(key));
+			HOST_SYNC(test->vcpu, STAGE_COPIED);
+			MOP(test->vm, ABSOLUTE, READ, mem2, 16, GADDR_V(mem2));
+			TEST_ASSERT(!succ, "exchange of values should not succeed");
+			ASSERT_MEM_EQ(mem1, mem2, 16);
+			ASSERT_MEM_EQ(&old, mem1, 16);
+		}
+	}
+}
 
 static void guest_copy(void)
 {
@@ -310,7 +371,7 @@ static void test_copy(void)
 
 	HOST_SYNC(t.vcpu, STAGE_INITED);
 
-	DEFAULT_WRITE_READ(t.vcpu, t.vcpu, LOGICAL, t.size);
+	default_write_read(t.vcpu, t.vcpu, LOGICAL, t.size, NO_KEY);
 
 	kvm_vm_free(t.kvm_vm);
 }
@@ -357,26 +418,268 @@ static void test_copy_key(void)
 	HOST_SYNC(t.vcpu, STAGE_SKEYS_SET);
 
 	/* vm, no key */
-	DEFAULT_WRITE_READ(t.vcpu, t.vm, ABSOLUTE, t.size);
+	default_write_read(t.vcpu, t.vm, ABSOLUTE, t.size, NO_KEY);
 
 	/* vm/vcpu, machting key or key 0 */
-	DEFAULT_WRITE_READ(t.vcpu, t.vcpu, LOGICAL, t.size, KEY(0));
-	DEFAULT_WRITE_READ(t.vcpu, t.vcpu, LOGICAL, t.size, KEY(9));
-	DEFAULT_WRITE_READ(t.vcpu, t.vm, ABSOLUTE, t.size, KEY(0));
-	DEFAULT_WRITE_READ(t.vcpu, t.vm, ABSOLUTE, t.size, KEY(9));
+	default_write_read(t.vcpu, t.vcpu, LOGICAL, t.size, 0);
+	default_write_read(t.vcpu, t.vcpu, LOGICAL, t.size, 9);
+	default_write_read(t.vcpu, t.vm, ABSOLUTE, t.size, 0);
+	default_write_read(t.vcpu, t.vm, ABSOLUTE, t.size, 9);
 	/*
 	 * There used to be different code paths for key handling depending on
 	 * if the region crossed a page boundary.
 	 * There currently are not, but the more tests the merrier.
 	 */
-	DEFAULT_WRITE_READ(t.vcpu, t.vcpu, LOGICAL, 1, KEY(0));
-	DEFAULT_WRITE_READ(t.vcpu, t.vcpu, LOGICAL, 1, KEY(9));
-	DEFAULT_WRITE_READ(t.vcpu, t.vm, ABSOLUTE, 1, KEY(0));
-	DEFAULT_WRITE_READ(t.vcpu, t.vm, ABSOLUTE, 1, KEY(9));
+	default_write_read(t.vcpu, t.vcpu, LOGICAL, 1, 0);
+	default_write_read(t.vcpu, t.vcpu, LOGICAL, 1, 9);
+	default_write_read(t.vcpu, t.vm, ABSOLUTE, 1, 0);
+	default_write_read(t.vcpu, t.vm, ABSOLUTE, 1, 9);
 
 	/* vm/vcpu, mismatching keys on read, but no fetch protection */
-	DEFAULT_READ(t.vcpu, t.vcpu, LOGICAL, t.size, GADDR_V(mem2), KEY(2));
-	DEFAULT_READ(t.vcpu, t.vm, ABSOLUTE, t.size, GADDR_V(mem1), KEY(2));
+	default_read(t.vcpu, t.vcpu, LOGICAL, t.size, 2);
+	default_read(t.vcpu, t.vm, ABSOLUTE, t.size, 2);
+
+	kvm_vm_free(t.kvm_vm);
+}
+
+static void test_cmpxchg_key(void)
+{
+	struct test_default t = test_default_init(guest_copy_key);
+
+	HOST_SYNC(t.vcpu, STAGE_SKEYS_SET);
+
+	default_cmpxchg(&t, NO_KEY);
+	default_cmpxchg(&t, 0);
+	default_cmpxchg(&t, 9);
+
+	kvm_vm_free(t.kvm_vm);
+}
+
+static __uint128_t cut_to_size(int size, __uint128_t val)
+{
+	switch (size) {
+	case 1:
+		return (uint8_t)val;
+	case 2:
+		return (uint16_t)val;
+	case 4:
+		return (uint32_t)val;
+	case 8:
+		return (uint64_t)val;
+	case 16:
+		return val;
+	}
+	GUEST_ASSERT_1(false, "Invalid size");
+	return 0;
+}
+
+static bool popcount_eq(__uint128_t a, __uint128_t b)
+{
+	unsigned int count_a, count_b;
+
+	count_a = __builtin_popcountl((uint64_t)(a >> 64)) +
+		  __builtin_popcountl((uint64_t)a);
+	count_b = __builtin_popcountl((uint64_t)(b >> 64)) +
+		  __builtin_popcountl((uint64_t)b);
+	return count_a == count_b;
+}
+
+static __uint128_t rotate(int size, __uint128_t val, int amount)
+{
+	unsigned int bits = size * 8;
+
+	amount = (amount + bits) % bits;
+	val = cut_to_size(size, val);
+	return (val << (bits - amount)) | (val >> amount);
+}
+
+const unsigned int max_block = 16;
+
+static void choose_block(bool guest, int i, int *size, int *offset)
+{
+	unsigned int rand;
+
+	rand = i;
+	if (guest) {
+		rand = rand * 19 + 11;
+		*size = 1 << ((rand % 3) + 2);
+		rand = rand * 19 + 11;
+		*offset = (rand % max_block) & ~(*size - 1);
+	} else {
+		rand = rand * 17 + 5;
+		*size = 1 << (rand % 5);
+		rand = rand * 17 + 5;
+		*offset = (rand % max_block) & ~(*size - 1);
+	}
+}
+
+static __uint128_t permutate_bits(bool guest, int i, int size, __uint128_t old)
+{
+	unsigned int rand;
+	int amount;
+	bool swap;
+
+	rand = i;
+	rand = rand * 3 + 1;
+	if (guest)
+		rand = rand * 3 + 1;
+	swap = rand % 2 == 0;
+	if (swap) {
+		int i, j;
+		__uint128_t new;
+		uint8_t byte0, byte1;
+
+		rand = rand * 3 + 1;
+		i = rand % size;
+		rand = rand * 3 + 1;
+		j = rand % size;
+		if (i == j)
+			return old;
+		new = rotate(16, old, i * 8);
+		byte0 = new & 0xff;
+		new &= ~0xff;
+		new = rotate(16, new, -i * 8);
+		new = rotate(16, new, j * 8);
+		byte1 = new & 0xff;
+		new = (new & ~0xff) | byte0;
+		new = rotate(16, new, -j * 8);
+		new = rotate(16, new, i * 8);
+		new = new | byte1;
+		new = rotate(16, new, -i * 8);
+		return new;
+	}
+	rand = rand * 3 + 1;
+	amount = rand % (size * 8);
+	return rotate(size, old, amount);
+}
+
+static bool _cmpxchg(int size, void *target, __uint128_t *old_addr, __uint128_t new)
+{
+	bool ret;
+
+	switch (size) {
+	case 4: {
+			uint32_t old = *old_addr;
+
+			asm volatile ("cs %[old],%[new],%[address]"
+			    : [old] "+d" (old),
+			      [address] "+Q" (*(uint32_t *)(target))
+			    : [new] "d" ((uint32_t)new)
+			    : "cc"
+			);
+			ret = old == (uint32_t)*old_addr;
+			*old_addr = old;
+			return ret;
+		}
+	case 8: {
+			uint64_t old = *old_addr;
+
+			asm volatile ("csg %[old],%[new],%[address]"
+			    : [old] "+d" (old),
+			      [address] "+Q" (*(uint64_t *)(target))
+			    : [new] "d" ((uint64_t)new)
+			    : "cc"
+			);
+			ret = old == (uint64_t)*old_addr;
+			*old_addr = old;
+			return ret;
+		}
+	case 16: {
+			__uint128_t old = *old_addr;
+
+			asm volatile ("cdsg %[old],%[new],%[address]"
+			    : [old] "+d" (old),
+			      [address] "+Q" (*(__uint128_t *)(target))
+			    : [new] "d" (new)
+			    : "cc"
+			);
+			ret = old == *old_addr;
+			*old_addr = old;
+			return ret;
+		}
+	}
+	GUEST_ASSERT_1(false, "Invalid size");
+	return 0;
+}
+
+const unsigned int cmpxchg_iter_outer = 100, cmpxchg_iter_inner = 10000;
+
+static void guest_cmpxchg_key(void)
+{
+	int size, offset;
+	__uint128_t old, new;
+
+	set_storage_key_range(mem1, max_block, 0x10);
+	set_storage_key_range(mem2, max_block, 0x10);
+	GUEST_SYNC(STAGE_SKEYS_SET);
+
+	for (int i = 0; i < cmpxchg_iter_outer; i++) {
+		do {
+			old = 1;
+		} while (!_cmpxchg(16, mem1, &old, 0));
+		for (int j = 0; j < cmpxchg_iter_inner; j++) {
+			choose_block(true, i + j, &size, &offset);
+			do {
+				new = permutate_bits(true, i + j, size, old);
+			} while (!_cmpxchg(size, mem2 + offset, &old, new));
+		}
+	}
+
+	GUEST_SYNC(STAGE_DONE);
+}
+
+static void *run_guest(void *data)
+{
+	struct test_info *info = data;
+
+	HOST_SYNC(*info, STAGE_DONE);
+	return NULL;
+}
+
+static char *quad_to_char(__uint128_t *quad, int size)
+{
+	return ((char *)quad) + (sizeof(*quad) - size);
+}
+
+static void test_cmpxchg_key_concurrent(void)
+{
+	struct test_default t = test_default_init(guest_cmpxchg_key);
+	int size, offset;
+	__uint128_t old, new;
+	bool success;
+	pthread_t thread;
+
+	HOST_SYNC(t.vcpu, STAGE_SKEYS_SET);
+	prepare_mem12();
+	MOP(t.vcpu, LOGICAL, WRITE, mem1, max_block, GADDR_V(mem2));
+	pthread_create(&thread, NULL, run_guest, &t.vcpu);
+
+	for (int i = 0; i < cmpxchg_iter_outer; i++) {
+		do {
+			old = 0;
+			new = 1;
+			MOP(t.vm, ABSOLUTE, CMPXCHG, &new,
+			    sizeof(new), GADDR_V(mem1),
+			    CMPXCHG_OLD(&old),
+			    CMPXCHG_SUCCESS(&success), KEY(1));
+		} while (!success);
+		for (int j = 0; j < cmpxchg_iter_inner; j++) {
+			choose_block(false, i + j, &size, &offset);
+			do {
+				new = permutate_bits(false, i + j, size, old);
+				MOP(t.vm, ABSOLUTE, CMPXCHG, quad_to_char(&new, size),
+				    size, GADDR_V(mem2 + offset),
+				    CMPXCHG_OLD(quad_to_char(&old, size)),
+				    CMPXCHG_SUCCESS(&success), KEY(1));
+			} while (!success);
+		}
+	}
+
+	pthread_join(thread, NULL);
+
+	MOP(t.vcpu, LOGICAL, READ, mem2, max_block, GADDR_V(mem2));
+	TEST_ASSERT(popcount_eq(*(__uint128_t *)mem1, *(__uint128_t *)mem2),
+		    "Must retain number of set bits");
 
 	kvm_vm_free(t.kvm_vm);
 }
@@ -409,7 +712,7 @@ static void test_copy_key_storage_prot_override(void)
 	HOST_SYNC(t.vcpu, STAGE_SKEYS_SET);
 
 	/* vcpu, mismatching keys, storage protection override in effect */
-	DEFAULT_WRITE_READ(t.vcpu, t.vcpu, LOGICAL, t.size, KEY(2));
+	default_write_read(t.vcpu, t.vcpu, LOGICAL, t.size, 2);
 
 	kvm_vm_free(t.kvm_vm);
 }
@@ -422,8 +725,8 @@ static void test_copy_key_fetch_prot(void)
 	HOST_SYNC(t.vcpu, STAGE_SKEYS_SET);
 
 	/* vm/vcpu, matching key, fetch protection in effect */
-	DEFAULT_READ(t.vcpu, t.vcpu, LOGICAL, t.size, GADDR_V(mem2), KEY(9));
-	DEFAULT_READ(t.vcpu, t.vm, ABSOLUTE, t.size, GADDR_V(mem2), KEY(9));
+	default_read(t.vcpu, t.vcpu, LOGICAL, t.size, 9);
+	default_read(t.vcpu, t.vm, ABSOLUTE, t.size, 9);
 
 	kvm_vm_free(t.kvm_vm);
 }
@@ -454,9 +757,27 @@ static void test_errors_key(void)
 
 	/* vm/vcpu, mismatching keys, fetch protection in effect */
 	CHECK_N_DO(ERR_PROT_MOP, t.vcpu, LOGICAL, WRITE, mem1, t.size, GADDR_V(mem1), KEY(2));
-	CHECK_N_DO(ERR_PROT_MOP, t.vcpu, LOGICAL, READ, mem2, t.size, GADDR_V(mem2), KEY(2));
+	CHECK_N_DO(ERR_PROT_MOP, t.vcpu, LOGICAL, READ, mem2, t.size, GADDR_V(mem1), KEY(2));
 	CHECK_N_DO(ERR_PROT_MOP, t.vm, ABSOLUTE, WRITE, mem1, t.size, GADDR_V(mem1), KEY(2));
-	CHECK_N_DO(ERR_PROT_MOP, t.vm, ABSOLUTE, READ, mem2, t.size, GADDR_V(mem2), KEY(2));
+	CHECK_N_DO(ERR_PROT_MOP, t.vm, ABSOLUTE, READ, mem2, t.size, GADDR_V(mem1), KEY(2));
+
+	kvm_vm_free(t.kvm_vm);
+}
+
+static void test_errors_cmpxchg_key(void)
+{
+	struct test_default t = test_default_init(guest_copy_key_fetch_prot);
+	int i;
+
+	HOST_SYNC(t.vcpu, STAGE_INITED);
+	HOST_SYNC(t.vcpu, STAGE_SKEYS_SET);
+
+	for (i = 1; i <= 16; i *= 2) {
+		__uint128_t old = 0;
+
+		ERR_PROT_MOP(t.vm, ABSOLUTE, CMPXCHG, mem2, i, GADDR_V(mem2),
+			     CMPXCHG_OLD(&old), KEY(2));
+	}
 
 	kvm_vm_free(t.kvm_vm);
 }
@@ -518,7 +839,7 @@ static void guest_copy_key_fetch_prot_override(void)
 	GUEST_SYNC(STAGE_INITED);
 	set_storage_key_range(0, PAGE_SIZE, 0x18);
 	set_storage_key_range((void *)last_page_addr, PAGE_SIZE, 0x0);
-	asm volatile ("sske %[key],%[addr]\n" :: [addr] "r"(0), [key] "r"(0x18) : "cc");
+	asm volatile ("sske %[key],%[addr]\n" :: [addr] "r"(0L), [key] "r"(0x18) : "cc");
 	GUEST_SYNC(STAGE_SKEYS_SET);
 
 	for (;;) {
@@ -606,7 +927,7 @@ static void test_errors_key_fetch_prot_override_enabled(void)
 
 	/*
 	 * vcpu, mismatching keys on fetch,
-	 * fetch protection override does not apply because memory range acceeded
+	 * fetch protection override does not apply because memory range exceeded
 	 */
 	CHECK_N_DO(ERR_PROT_MOP, t.vcpu, LOGICAL, READ, mem2, 2048 + 1, GADDR_V(0), KEY(2));
 	CHECK_N_DO(ERR_PROT_MOP, t.vcpu, LOGICAL, READ, mem2, PAGE_SIZE + 2048 + 1,
@@ -645,7 +966,9 @@ static void _test_errors_common(struct test_info info, enum mop_target target, i
 
 	/* Bad guest address: */
 	rv = ERR_MOP(info, target, WRITE, mem1, size, GADDR((void *)~0xfffUL), CHECK_ONLY);
-	TEST_ASSERT(rv > 0, "ioctl does not report bad guest memory access");
+	TEST_ASSERT(rv > 0, "ioctl does not report bad guest memory address with CHECK_ONLY");
+	rv = ERR_MOP(info, target, WRITE, mem1, size, GADDR((void *)~0xfffUL));
+	TEST_ASSERT(rv > 0, "ioctl does not report bad guest memory address on write");
 
 	/* Bad host address: */
 	rv = ERR_MOP(info, target, WRITE, 0, size, GADDR_V(mem1));
@@ -694,85 +1017,138 @@ static void test_errors(void)
 	kvm_vm_free(t.kvm_vm);
 }
 
-struct testdef {
-	const char *name;
-	void (*test)(void);
-	int extension;
-} testlist[] = {
-	{
-		.name = "simple copy",
-		.test = test_copy,
-	},
-	{
-		.name = "generic error checks",
-		.test = test_errors,
-	},
-	{
-		.name = "copy with storage keys",
-		.test = test_copy_key,
-		.extension = 1,
-	},
-	{
-		.name = "copy with key storage protection override",
-		.test = test_copy_key_storage_prot_override,
-		.extension = 1,
-	},
-	{
-		.name = "copy with key fetch protection",
-		.test = test_copy_key_fetch_prot,
-		.extension = 1,
-	},
-	{
-		.name = "copy with key fetch protection override",
-		.test = test_copy_key_fetch_prot_override,
-		.extension = 1,
-	},
-	{
-		.name = "error checks with key",
-		.test = test_errors_key,
-		.extension = 1,
-	},
-	{
-		.name = "termination",
-		.test = test_termination,
-		.extension = 1,
-	},
-	{
-		.name = "error checks with key storage protection override",
-		.test = test_errors_key_storage_prot_override,
-		.extension = 1,
-	},
-	{
-		.name = "error checks without key fetch prot override",
-		.test = test_errors_key_fetch_prot_override_not_enabled,
-		.extension = 1,
-	},
-	{
-		.name = "error checks with key fetch prot override",
-		.test = test_errors_key_fetch_prot_override_enabled,
-		.extension = 1,
-	},
-};
+static void test_errors_cmpxchg(void)
+{
+	struct test_default t = test_default_init(guest_idle);
+	__uint128_t old;
+	int rv, i, power = 1;
+
+	HOST_SYNC(t.vcpu, STAGE_INITED);
+
+	for (i = 0; i < 32; i++) {
+		if (i == power) {
+			power *= 2;
+			continue;
+		}
+		rv = ERR_MOP(t.vm, ABSOLUTE, CMPXCHG, mem1, i, GADDR_V(mem1),
+			     CMPXCHG_OLD(&old));
+		TEST_ASSERT(rv == -1 && errno == EINVAL,
+			    "ioctl allows bad size for cmpxchg");
+	}
+	for (i = 1; i <= 16; i *= 2) {
+		rv = ERR_MOP(t.vm, ABSOLUTE, CMPXCHG, mem1, i, GADDR((void *)~0xfffUL),
+			     CMPXCHG_OLD(&old));
+		TEST_ASSERT(rv > 0, "ioctl allows bad guest address for cmpxchg");
+	}
+	for (i = 2; i <= 16; i *= 2) {
+		rv = ERR_MOP(t.vm, ABSOLUTE, CMPXCHG, mem1, i, GADDR_V(mem1 + 1),
+			     CMPXCHG_OLD(&old));
+		TEST_ASSERT(rv == -1 && errno == EINVAL,
+			    "ioctl allows bad alignment for cmpxchg");
+	}
+
+	kvm_vm_free(t.kvm_vm);
+}
 
 int main(int argc, char *argv[])
 {
 	int extension_cap, idx;
 
 	TEST_REQUIRE(kvm_has_cap(KVM_CAP_S390_MEM_OP));
+	extension_cap = kvm_check_cap(KVM_CAP_S390_MEM_OP_EXTENSION);
 
-	ksft_print_header();
+	struct testdef {
+		const char *name;
+		void (*test)(void);
+		bool requirements_met;
+	} testlist[] = {
+		{
+			.name = "simple copy",
+			.test = test_copy,
+			.requirements_met = true,
+		},
+		{
+			.name = "generic error checks",
+			.test = test_errors,
+			.requirements_met = true,
+		},
+		{
+			.name = "copy with storage keys",
+			.test = test_copy_key,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "cmpxchg with storage keys",
+			.test = test_cmpxchg_key,
+			.requirements_met = extension_cap & 0x2,
+		},
+		{
+			.name = "concurrently cmpxchg with storage keys",
+			.test = test_cmpxchg_key_concurrent,
+			.requirements_met = extension_cap & 0x2,
+		},
+		{
+			.name = "copy with key storage protection override",
+			.test = test_copy_key_storage_prot_override,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "copy with key fetch protection",
+			.test = test_copy_key_fetch_prot,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "copy with key fetch protection override",
+			.test = test_copy_key_fetch_prot_override,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "error checks with key",
+			.test = test_errors_key,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "error checks for cmpxchg with key",
+			.test = test_errors_cmpxchg_key,
+			.requirements_met = extension_cap & 0x2,
+		},
+		{
+			.name = "error checks for cmpxchg",
+			.test = test_errors_cmpxchg,
+			.requirements_met = extension_cap & 0x2,
+		},
+		{
+			.name = "termination",
+			.test = test_termination,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "error checks with key storage protection override",
+			.test = test_errors_key_storage_prot_override,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "error checks without key fetch prot override",
+			.test = test_errors_key_fetch_prot_override_not_enabled,
+			.requirements_met = extension_cap > 0,
+		},
+		{
+			.name = "error checks with key fetch prot override",
+			.test = test_errors_key_fetch_prot_override_enabled,
+			.requirements_met = extension_cap > 0,
+		},
+	};
 
+	ksft_print_header();
 	ksft_set_plan(ARRAY_SIZE(testlist));
 
-	extension_cap = kvm_check_cap(KVM_CAP_S390_MEM_OP_EXTENSION);
 	for (idx = 0; idx < ARRAY_SIZE(testlist); idx++) {
-		if (extension_cap >= testlist[idx].extension) {
+		if (testlist[idx].requirements_met) {
 			testlist[idx].test();
 			ksft_test_result_pass("%s\n", testlist[idx].name);
 		} else {
-			ksft_test_result_skip("%s - extension level %d not supported\n",
-					      testlist[idx].name,
-					      testlist[idx].extension);
+			ksft_test_result_skip("%s - requirements not met (kernel has extension cap %#x)\n",
+					      testlist[idx].name, extension_cap);
 		}
 	}