[PATCH 09/22] KVM: selftests: Verify KVM correctly handles mprotect(PROT_READ)

[Sean Christopherson <seanjc@google.com>]

Add two phases to mmu_stress_test to verify that KVM correctly handles
guest memory that was writable, and then made read-only in the primary MMU,
and then made writable again.

Add bonus coverage for x86 to verify that all of guest memory was marked
read-only.  Making forward progress (without making memory writable)
requires arch specific code to skip over the faulting instruction, but the
test can at least verify each vCPU's starting page was made read-only.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 tools/testing/selftests/kvm/mmu_stress_test.c | 87 ++++++++++++++++++-
 1 file changed, 84 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/kvm/mmu_stress_test.c b/tools/testing/selftests/kvm/mmu_stress_test.c
index 50c3a17418c4..98f9a4660269 100644
--- a/tools/testing/selftests/kvm/mmu_stress_test.c
+++ b/tools/testing/selftests/kvm/mmu_stress_test.c
@@ -16,6 +16,8 @@
 #include "guest_modes.h"
 #include "processor.h"
 
+static bool mprotect_ro_done;
+
 static void guest_code(uint64_t start_gpa, uint64_t end_gpa, uint64_t stride)
 {
 	uint64_t gpa;
@@ -31,6 +33,33 @@ static void guest_code(uint64_t start_gpa, uint64_t end_gpa, uint64_t stride)
 		*((volatile uint64_t *)gpa);
 	GUEST_SYNC(2);
 
+	/*
+	 * Write to the region while mprotect(PROT_READ) is underway.  Keep
+	 * looping until the memory is guaranteed to be read-only, otherwise
+	 * vCPUs may complete their writes and advance to the next stage
+	 * prematurely.
+	 */
+	do {
+		for (gpa = start_gpa; gpa < end_gpa; gpa += stride)
+#ifdef __x86_64__
+			asm volatile(".byte 0xc6,0x40,0x0,0x0" :: "a" (gpa) : "memory");
+#else
+			vcpu_arch_put_guest(*((volatile uint64_t *)gpa), gpa);
+#endif
+	} while (!READ_ONCE(mprotect_ro_done));
+
+	/*
+	 * Only x86 can explicitly sync, as other architectures will be stuck
+	 * on the write fault.
+	 */
+#ifdef __x86_64__
+	GUEST_SYNC(3);
+#endif
+
+	for (gpa = start_gpa; gpa < end_gpa; gpa += stride)
+		vcpu_arch_put_guest(*((volatile uint64_t *)gpa), gpa);
+	GUEST_SYNC(4);
+
 	GUEST_ASSERT(0);
 }
 
@@ -78,6 +107,7 @@ static void *vcpu_worker(void *data)
 	struct vcpu_info *info = data;
 	struct kvm_vcpu *vcpu = info->vcpu;
 	struct kvm_vm *vm = vcpu->vm;
+	int r;
 
 	vcpu_args_set(vcpu, 3, info->start_gpa, info->end_gpa, vm->page_size);
 
@@ -100,6 +130,49 @@ static void *vcpu_worker(void *data)
 
 	/* Stage 2, read all of guest memory, which is now read-only. */
 	run_vcpu(vcpu, 2);
+
+	/*
+	 * Stage 3, write guest memory and verify KVM returns -EFAULT for once
+	 * the mprotect(PROT_READ) lands.  Only architectures that support
+	 * validating *all* of guest memory sync for this stage, as vCPUs will
+	 * be stuck on the faulting instruction for other architectures.  Go to
+	 * stage 3 without a rendezvous
+	 */
+	do {
+		r = _vcpu_run(vcpu);
+	} while (!r);
+	TEST_ASSERT(r == -1 && errno == EFAULT,
+		    "Expected EFAULT on write to RO memory, got r = %d, errno = %d", r, errno);
+
+#ifdef __x86_64__
+	/*
+	 * Verify *all* writes from the guest hit EFAULT due to the VMA now
+	 * being read-only.  x86-only at this time as skipping the instruction
+	 * that hits the EFAULT requires advancing the program counter, which
+	 * is arch specific and currently relies on hand-coded assembly.
+	 */
+	vcpu->run->kvm_valid_regs = KVM_SYNC_X86_REGS;
+	for (;;) {
+		r = _vcpu_run(vcpu);
+		if (!r)
+			break;
+		TEST_ASSERT_EQ(errno, EFAULT);
+		WRITE_ONCE(vcpu->run->kvm_dirty_regs, KVM_SYNC_X86_REGS);
+		vcpu->run->s.regs.regs.rip += 4;
+	}
+	assert_sync_stage(vcpu, 3);
+#endif
+	rendezvous_with_boss();
+
+	/*
+	 * Stage 4.  Run to completion, waiting for mprotect(PROT_WRITE) to
+	 * make the memory writable again.
+	 */
+	do {
+		r = _vcpu_run(vcpu);
+	} while (r && errno == EFAULT);
+	TEST_ASSERT_EQ(r, 0);
+	assert_sync_stage(vcpu, 4);
 	rendezvous_with_boss();
 
 	return NULL;
@@ -182,7 +255,7 @@ int main(int argc, char *argv[])
 	const uint64_t start_gpa = SZ_4G;
 	const int first_slot = 1;
 
-	struct timespec time_start, time_run1, time_reset, time_run2, time_ro;
+	struct timespec time_start, time_run1, time_reset, time_run2, time_ro, time_rw;
 	uint64_t max_gpa, gpa, slot_size, max_mem, i;
 	int max_slots, slot, opt, fd;
 	bool hugepages = false;
@@ -287,19 +360,27 @@ int main(int argc, char *argv[])
 	rendezvous_with_vcpus(&time_run2, "run 2");
 
 	mprotect(mem, slot_size, PROT_READ);
+	usleep(10);
+	mprotect_ro_done = true;
+	sync_global_to_guest(vm, mprotect_ro_done);
+
 	rendezvous_with_vcpus(&time_ro, "mprotect RO");
+	mprotect(mem, slot_size, PROT_READ | PROT_WRITE);
+	rendezvous_with_vcpus(&time_rw, "mprotect RW");
 
+	time_rw    = timespec_sub(time_rw,     time_ro);
 	time_ro    = timespec_sub(time_ro,     time_run2);
 	time_run2  = timespec_sub(time_run2,   time_reset);
 	time_reset = timespec_sub(time_reset,  time_run1);
 	time_run1  = timespec_sub(time_run1,   time_start);
 
 	pr_info("run1 = %ld.%.9lds, reset = %ld.%.9lds, run2 = %ld.%.9lds, "
-		"ro = %ld.%.9lds\n",
+		"ro = %ld.%.9lds, rw = %ld.%.9lds\n",
 		time_run1.tv_sec, time_run1.tv_nsec,
 		time_reset.tv_sec, time_reset.tv_nsec,
 		time_run2.tv_sec, time_run2.tv_nsec,
-		time_ro.tv_sec, time_ro.tv_nsec);
+		time_ro.tv_sec, time_ro.tv_nsec,
+		time_rw.tv_sec, time_rw.tv_nsec);
 
 	/*
 	 * Delete even numbered slots (arbitrary) and unmap the first half of
-- 
2.46.0.76.ge559c4bf1a-goog