path: root/tools/testing/selftests/kvm/memslot_perf_test.c

// SPDX-License-Identifier: GPL-2.0
/*
 * A memslot-related performance benchmark.
 *
 * Copyright (C) 2021 Oracle and/or its affiliates.
 *
 * Basic guest setup / host vCPU thread code lifted from set_memory_region_test.
 */
#include <pthread.h>
#include <sched.h>
#include <semaphore.h>
#include <stdatomic.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mman.h>
#include <time.h>
#include <unistd.h>

#include <linux/compiler.h>
#include <linux/sizes.h>

#include <test_util.h>
#include <kvm_util.h>
#include <processor.h>

#define MEM_EXTRA_SIZE		SZ_64K

#define MEM_SIZE		(SZ_512M + MEM_EXTRA_SIZE)
#define MEM_GPA			SZ_256M
#define MEM_AUX_GPA		MEM_GPA
#define MEM_SYNC_GPA		MEM_AUX_GPA
#define MEM_TEST_GPA		(MEM_AUX_GPA + MEM_EXTRA_SIZE)
#define MEM_TEST_SIZE		(MEM_SIZE - MEM_EXTRA_SIZE)

/*
 * 32 MiB is max size that gets well over 100 iterations on 509 slots.
 * Considering that each slot needs to have at least one page up to
 * 8194 slots in use can then be tested (although with slightly
 * limited resolution).
 */
#define MEM_SIZE_MAP		(SZ_32M + MEM_EXTRA_SIZE)
#define MEM_TEST_MAP_SIZE	(MEM_SIZE_MAP - MEM_EXTRA_SIZE)

/*
 * 128 MiB is min size that fills 32k slots with at least one page in each
 * while at the same time gets 100+ iterations in such test
 *
 * 2 MiB chunk size like a typical huge page
 */
#define MEM_TEST_UNMAP_SIZE		SZ_128M
#define MEM_TEST_UNMAP_CHUNK_SIZE	SZ_2M

/*
 * For the move active test the middle of the test area is placed on
 * a memslot boundary: half lies in the memslot being moved, half in
 * other memslot(s).
 *
 * We have different number of memory slots, excluding the reserved
 * memory slot 0, on various architectures and configurations. The
 * memory size in this test is calculated by picking the maximal
 * last memory slot's memory size, with alignment to the largest
 * supported page size (64KB). In this way, the selected memory
 * size for this test is compatible with test_memslot_move_prepare().
 *
 * architecture   slots    memory-per-slot    memory-on-last-slot
 * --------------------------------------------------------------
 * x86-4KB        32763    16KB               160KB
 * arm64-4KB      32766    16KB               112KB
 * arm64-16KB     32766    16KB               112KB
 * arm64-64KB     8192     64KB               128KB
 */
#define MEM_TEST_MOVE_SIZE		(3 * SZ_64K)
#define MEM_TEST_MOVE_GPA_DEST		(MEM_GPA + MEM_SIZE)
static_assert(MEM_TEST_MOVE_SIZE <= MEM_TEST_SIZE,
	      "invalid move test region size");

#define MEM_TEST_VAL_1 0x1122334455667788
#define MEM_TEST_VAL_2 0x99AABBCCDDEEFF00

struct vm_data {
	struct kvm_vm *vm;
	struct kvm_vcpu *vcpu;
	pthread_t vcpu_thread;
	uint32_t nslots;
	uint64_t npages;
	uint64_t pages_per_slot;
	void **hva_slots;
	bool mmio_ok;
	uint64_t mmio_gpa_min;
	uint64_t mmio_gpa_max;
};

struct sync_area {
	uint32_t    guest_page_size;
	atomic_bool start_flag;
	atomic_bool exit_flag;
	atomic_bool sync_flag;
	void *move_area_ptr;
};

/*
 * Technically, we need also for the atomic bool to be address-free, which
 * is recommended, but not strictly required, by C11 for lockless
 * implementations.
 * However, in practice both GCC and Clang fulfill this requirement on
 * all KVM-supported platforms.
 */
static_assert(ATOMIC_BOOL_LOCK_FREE == 2, "atomic bool is not lockless");

static sem_t vcpu_ready;

static bool map_unmap_verify;
#ifdef __x86_64__
static bool disable_slot_zap_quirk;
#endif

static bool verbose;
#define pr_info_v(...)				\
	do {					\
		if (verbose)			\
			pr_info(__VA_ARGS__);	\
	} while (0)

static void check_mmio_access(struct vm_data *data, struct kvm_run *run)
{
	TEST_ASSERT(data->mmio_ok, "Unexpected mmio exit");
	TEST_ASSERT(run->mmio.is_write, "Unexpected mmio read");
	TEST_ASSERT(run->mmio.len == 8,
		    "Unexpected exit mmio size = %u", run->mmio.len);
	TEST_ASSERT(run->mmio.phys_addr >= data->mmio_gpa_min &&
		    run->mmio.phys_addr <= data->mmio_gpa_max,
		    "Unexpected exit mmio address = 0x%llx",
		    run->mmio.phys_addr);
}

static void *vcpu_worker(void *__data)
{
	struct vm_data *data = __data;
	struct kvm_vcpu *vcpu = data->vcpu;
	struct kvm_run *run = vcpu->run;
	struct ucall uc;

	while (1) {
		vcpu_run(vcpu);

		switch (get_ucall(vcpu, &uc)) {
		case UCALL_SYNC:
			TEST_ASSERT(uc.args[1] == 0,
				"Unexpected sync ucall, got %lx",
				(ulong)uc.args[1]);
			sem_post(&vcpu_ready);
			continue;
		case UCALL_NONE:
			if (run->exit_reason == KVM_EXIT_MMIO)
				check_mmio_access(data, run);
			else
				goto done;
			break;
		case UCALL_ABORT:
			REPORT_GUEST_ASSERT(uc);
			break;
		case UCALL_DONE:
			goto done;
		default:
			TEST_FAIL("Unknown ucall %lu", uc.cmd);
		}
	}

done:
	return NULL;
}

static void wait_for_vcpu(void)
{
	struct timespec ts;

	TEST_ASSERT(!clock_gettime(CLOCK_REALTIME, &ts),
		    "clock_gettime() failed: %d", errno);

	ts.tv_sec += 2;
	TEST_ASSERT(!sem_timedwait(&vcpu_ready, &ts),
		    "sem_timedwait() failed: %d", errno);
}

static void *vm_gpa2hva(struct vm_data *data, uint64_t gpa, uint64_t *rempages)
{
	uint64_t gpage, pgoffs;
	uint32_t slot, slotoffs;
	void