aa3bf63e创建于 2025年7月4日历史提交
/*
 * Copyright (C) 2013-2021 Canonical, Ltd.
 * Copyright (C) 2022-2025 Colin Ian King.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version 2
 * of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 *
 */
#include "stress-ng.h"
#include "core-attribute.h"
#include "core-cpu-cache.h"
#include "core-madvise.h"
#include "core-pragma.h"

#define DEBUG_TAG_INFO		(0)

static const stress_help_t help[] = {
	{ NULL,	"l1cache N",	 	"start N CPU level 1 cache thrashing workers" },
	{ NULL, "l1cache-line-size N",	"specify level 1 cache line size" },
	{ NULL,	"l1cache-method M",	"l1 cache thrashing method: forward, reverse, random" },
	{ NULL,	"l1cache-mlock",	"attempt to mlock memory" },
	{ NULL, "l1cache-sets N",	"specify level 1 cache sets" },
	{ NULL, "l1cache-size N",	"specify level 1 cache size" },
	{ NULL,	"l1cache-ways N",	"only fill specified number of cache ways" },
	{ NULL,	NULL,			NULL }
};


#if DEBUG_TAG_INFO
/*
 *  stress_l1cache_ln2()
 *	calculate log base 2
 */
static uint32_t PURE stress_l1cache_ln2(const uint32_t val)
{
	int ln2 = 0;

	while (val >>= 1)
		++ln2;

	return ln2;
}
#endif

/*
 *  stress_l1cache_info_check()
 *	sanity check cache information
 */
static int stress_l1cache_info_check(
	stress_args_t *args,
	const uint32_t ways,
	const uint32_t size,
	const uint32_t sets,
	const uint32_t line_size)
{
	const uint64_t sz = size ? size :
		(uint64_t)ways * (uint64_t)sets * (uint64_t)line_size;

	if (stress_instance_zero(args)) {
		char szstr[64];

		stress_uint64_to_str(szstr, sizeof(szstr), sz, 0, true);

		pr_inf("%s: l1cache: size: %s, sets: %" PRIu32
			", ways: %" PRIu32 ", line size: %" PRIu32 " bytes\n",
			args->name, szstr, sets, ways, line_size);
#if DEBUG_TAG_INFO
		pr_inf("%s: l1cache: offset size: %" PRIu32 ", index size: %" PRIu32 "\n",
			args->name, stress_l1cache_ln2(line_size),
			stress_l1cache_ln2(sets));
#endif
	}

	if (size == 0) {
		pr_inf("%s: invalid cache size of 0\n", args->name);
		return EXIT_FAILURE;
	}
	if (sets == 0) {
		pr_inf("%s: invalid 0 number of sets\n", args->name);
		return EXIT_FAILURE;
	}
	if (ways == 0) {
		pr_inf("%s: invalid 0 number of ways\n", args->name);
		return EXIT_FAILURE;
	}
	if (line_size == 0) {
		pr_inf("%s: invalid cache line size of 0\n", args->name);
		return EXIT_FAILURE;
	}
	if (sz != size) {
		pr_inf("%s: cache size %" PRIu32 " not equal to ways %" PRIu32
			" * sets %" PRIu32 " * line size %" PRIu32 "\n",
		args->name, size, ways, sets, line_size);
		return EXIT_FAILURE;
	}

	return EXIT_SUCCESS;
}

static int stress_l1cache_info_ok(
	stress_args_t *args,
	uint32_t *ways,
	uint32_t *size,
	uint32_t *sets,
	uint32_t *line_size)
{
	int ret;
#if defined(__linux__)
	stress_cpu_cache_cpus_t *cpu_caches;
	const stress_cpu_cache_t *cache = NULL;
#endif

	if ((*ways > 0) && (*size > 0) && (*sets > 0) && (*line_size > 0)) {
		return stress_l1cache_info_check(args, *ways, *size, *sets, *line_size);
	}
	if ((*ways > 0) && (*size > 0) && (*sets > 0) && (*line_size == 0)) {
		*line_size = *size / (*ways * *sets);
		return stress_l1cache_info_check(args, *ways, *size, *sets, *line_size);
	}
	if ((*ways > 0) && (*size > 0) && (*sets == 0) && (*line_size > 0)) {
		*sets = *size / (*ways * *line_size);
		return stress_l1cache_info_check(args, *ways, *size, *sets, *line_size);
	}
	if ((*ways > 0) && (*size == 0) && (*sets > 0) && (*line_size > 0)) {
		*size = *sets * *ways * *line_size;
		return stress_l1cache_info_check(args, *ways, *size, *sets, *line_size);
	}
	if ((*ways == 0) && (*size > 0) && (*sets > 0) && (*line_size > 0)) {
		*ways = *size / (*sets * *line_size);
		return stress_l1cache_info_check(args, *ways, *size, *sets, *line_size);
	}

#if defined(__linux__)
	/*
	 *  User didn't provide cache info, try and figure it
	 *  out
	 */
	cpu_caches = stress_cpu_cache_get_all_details();
	if (!cpu_caches)
		goto bad_cache;

	if (stress_cpu_cache_get_max_level(cpu_caches) < 1)
		goto bad_cache_free;

	cache = stress_cpu_cache_get(cpu_caches, 1);
	if (!cache)
		goto bad_cache_free;

	if (*size == 0)
		*size = (uint32_t)cache->size;
	if (*line_size == 0)
		*line_size = cache->line_size;
	if (*ways == 0)
		*ways = cache->ways;
	if ((*sets == 0) && (*ways > 0) && (*line_size > 0))
		*sets = *size / (*ways * *line_size);

	stress_free_cpu_caches(cpu_caches);
#endif
	if ((*size == 0) && (*line_size == 0) && (*ways == 0) && (*sets == 0))
		goto bad_cache;
	ret = stress_l1cache_info_check(args, *ways, *size, *sets, *line_size);
	if (ret != EXIT_SUCCESS)
		goto bad_cache;
	return ret;

#if defined(__linux__)
bad_cache_free:
	stress_free_cpu_caches(cpu_caches);
#endif
bad_cache:
	pr_inf_skip("%s: skipping stressor, cannot determine "
		"cache level 1 information from kernel\n",
		args->name);

	return EXIT_NO_RESOURCE;
}

static int OPTIMIZE3 stress_l1cache_forward(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size)
{
	register int i;
	static uint32_t set;

	const uint32_t set_offset = set * l1cache_set_size;
	uint8_t * const cache_start = cache_aligned + set_offset;
	const uint8_t * const cache_end = cache_start + (l1cache_size << 1);

	(void)args;

	/*
	 * cycle around 2 x cache size to force evictions
	 * in cache set steps to hit every cache line in
	 * the set, reads then writes
	 */
	for (i = 0; i < 1000000; i++) {
		register volatile uint8_t *ptr;

PRAGMA_UNROLL_N(2)
		for (ptr = cache_start; ptr < cache_end; ptr += l1cache_set_size)
			*(ptr);

PRAGMA_UNROLL_N(2)
		for (ptr = cache_start; ptr < cache_end; ptr += l1cache_set_size)
			*(ptr) = (uint8_t)set;

		set++;
		if (UNLIKELY(set >= l1cache_sets))
			set = 0;
	}
	return EXIT_SUCCESS;
}

static int OPTIMIZE3 stress_l1cache_forward_and_verify(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size)
{
	register int i;
	static uint32_t set;

	const uint32_t set_offset = set * l1cache_set_size;
	uint8_t * const cache_start = cache_aligned + set_offset;
	const uint8_t * const cache_end = cache_start + (l1cache_size << 1);

	/*
	 * cycle around 2 x cache size to force evictions
	 * in cache set steps to hit every cache line in
	 * the set, reads then writes
	 */
	for (i = 0; i < 1000000; i++) {
		register volatile uint8_t *ptr;

PRAGMA_UNROLL_N(2)
		for (ptr = cache_start; ptr < cache_end; ptr += l1cache_set_size)
			(void)*(ptr);

PRAGMA_UNROLL_N(2)
		for (ptr = cache_start; ptr < cache_end; ptr += l1cache_set_size)
			*(ptr) = (uint8_t)set;

		for (ptr = cache_start; ptr < cache_end; ptr += l1cache_set_size) {
			if (UNLIKELY(*ptr != (uint8_t)set)) {
				pr_fail("%s: cache value mismatch at offset %zd, 0x%2.2x vs 0x%2.2x\n",
					args->name, (size_t)(ptr - cache_start),
					*ptr, (uint8_t)set);
				return EXIT_FAILURE;
			}
		}
		set++;
		if (UNLIKELY(set >= l1cache_sets))
			set = 0;
	}
	return EXIT_SUCCESS;
}

static int OPTIMIZE3 stress_l1cache_reverse(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size)
{
	register int i;
	static uint32_t set;

	const uint32_t set_offset = set * l1cache_set_size;
	uint8_t * const cache_start = cache_aligned + set_offset;
	uint8_t * const cache_end = cache_start + (l1cache_size << 1);

	(void)args;

	/*
	 * cycle around 2 x cache size to force evictions
	 * in cache set steps to hit every cache line in
	 * the set, reads then writes
	 */
	for (i = 0; i < 1000000; i++) {
		register volatile uint8_t *ptr;

PRAGMA_UNROLL_N(2)
		for (ptr = cache_end - l1cache_set_size + 1; ptr >= cache_start; ptr -= l1cache_set_size)
			*(ptr);

PRAGMA_UNROLL_N(2)
		for (ptr = cache_end - l1cache_set_size + 1; ptr >= cache_start; ptr -= l1cache_set_size)
			*(ptr) = (uint8_t)set;

		set++;
		if (UNLIKELY(set >= l1cache_sets))
			set = 0;
	}
	return EXIT_SUCCESS;
}

static int OPTIMIZE3 stress_l1cache_reverse_and_verify(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size)
{
	register int i;
	static uint32_t set;

	const uint32_t set_offset = set * l1cache_set_size;
	uint8_t * const cache_start = cache_aligned + set_offset;
	uint8_t * const cache_end = cache_start + (l1cache_size << 1);

	(void)args;

	/*
	 * cycle around 2 x cache size to force evictions
	 * in cache set steps to hit every cache line in
	 * the set, reads then writes
	 */
	for (i = 0; i < 1000000; i++) {
		register volatile uint8_t *ptr;

PRAGMA_UNROLL_N(2)
		for (ptr = cache_end - l1cache_set_size + 1; ptr >= cache_start; ptr -= l1cache_set_size)
			*(ptr);

PRAGMA_UNROLL_N(2)
		for (ptr = cache_end - l1cache_set_size + 1; ptr >= cache_start; ptr -= l1cache_set_size)
			*(ptr) = (uint8_t)set;

		for (ptr = cache_end - l1cache_set_size + 1; ptr >= cache_start; ptr -= l1cache_set_size) {
			if (UNLIKELY(*ptr != (uint8_t)set)) {
				pr_fail("%s: cache value mismatch at offset %zd, 0x%2.2x vs 0x%2.2x\n",
					args->name, (size_t)(ptr - cache_start),
					*ptr, (uint8_t)set);
				return EXIT_FAILURE;
			}
		}
		set++;
		if (UNLIKELY(set >= l1cache_sets))
			set = 0;
	}
	return EXIT_SUCCESS;
}

static int OPTIMIZE3 stress_l1cache_random(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size)
{
	register int i;
	static uint32_t set;
	uint32_t w, z;

	const uint32_t set_offset = set * l1cache_set_size;
	uint8_t * const cache_start = cache_aligned + set_offset;
	const uint8_t * const cache_end = cache_start + (l1cache_size << 1);
	const size_t cache_size = cache_end - cache_start;
	const size_t loops = cache_size / l1cache_set_size;
	register volatile uint8_t *ptr = (volatile uint8_t *)cache_start;

	(void)args;
	stress_mwc_get_seed(&w, &z);

	/*
	 * cycle around 2 x cache size to force evictions
	 * in cache set steps to hit every cache line in
	 * the set, reads then writes
	 */
	for (i = 0; i < 1000000; i++) {
		register size_t j;

		stress_mwc_set_seed(w, z);
PRAGMA_UNROLL_N(2)
		for (j = 0; j < loops; j++)
			(void)*(ptr + stress_mwc32modn((uint32_t)cache_size));

		stress_mwc_set_seed(w, z);
PRAGMA_UNROLL_N(2)
		for (j = 0; j < loops; j++)
			*(ptr + stress_mwc32modn((uint32_t)cache_size)) = (uint8_t)set;

	}
	set++;
	if (UNLIKELY(set >= l1cache_sets))
		set = 0;
	return EXIT_SUCCESS;
}

static int OPTIMIZE3 stress_l1cache_random_and_verify(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size)
{
	register int i;
	static uint32_t set;
	uint32_t w, z;

	const uint32_t set_offset = set * l1cache_set_size;
	uint8_t * const cache_start = cache_aligned + set_offset;
	const uint8_t * const cache_end = cache_start + (l1cache_size << 1);
	const size_t cache_size = cache_end - cache_start;
	const size_t loops = cache_size / l1cache_set_size;
	register volatile uint8_t *ptr = (volatile uint8_t *)cache_start;

	(void)args;
	stress_mwc_get_seed(&w, &z);

	/*
	 * cycle around 2 x cache size to force evictions
	 * in cache set steps to hit every cache line in
	 * the set, reads then writes
	 */
	for (i = 0; i < 1000000; i++) {
		register size_t j;

		stress_mwc_set_seed(w, z);
PRAGMA_UNROLL_N(2)
		for (j = 0; j < loops; j++)
			(void)*(ptr + stress_mwc32modn((uint32_t)cache_size));

		stress_mwc_set_seed(w, z);
PRAGMA_UNROLL_N(2)
		for (j = 0; j < loops; j++)
			*(ptr + stress_mwc32modn((uint32_t)cache_size)) = (uint8_t)set;

		stress_mwc_set_seed(w, z);
		for (j = 0; j < loops; j++) {
			register const size_t idx = stress_mwc32modn((uint32_t)cache_size);

			if (UNLIKELY(*(ptr + idx) != (uint8_t)set)) {
				pr_fail("%s: cache value mismatch at offset %zd, 0x%2.2x vs 0x%2.2x\n",
					args->name, (size_t)(ptr - cache_start),
					*(ptr + idx), (uint8_t)set);
				return EXIT_FAILURE;
			}
		}

	}
	set++;
	if (UNLIKELY(set >= l1cache_sets))
		set = 0;
	return EXIT_SUCCESS;
}

typedef int (*l1cache_func_t)(
	stress_args_t *args,
	uint8_t *cache_aligned,
	const uint32_t l1cache_size,
	const uint32_t l1cache_sets,
	const uint32_t l1cache_set_size);

typedef struct {
	const char *name;
	const l1cache_func_t	func[2];
} stress_l1cache_method_t;

static const stress_l1cache_method_t stress_l1cache_methods[] = {
	{ "forward",	{ stress_l1cache_forward,	stress_l1cache_forward_and_verify } },
	{ "random",	{ stress_l1cache_random,	stress_l1cache_random_and_verify } },
	{ "reverse",	{ stress_l1cache_reverse,	stress_l1cache_reverse_and_verify } },
};

static const char *stress_l1cache_method(const size_t i)
{
	return (i < SIZEOF_ARRAY(stress_l1cache_methods)) ? stress_l1cache_methods[i].name : NULL;
}

static const stress_opt_t opts[] = {
	{ OPT_l1cache_sets,      "l1cache-sets",      TYPE_ID_UINT32, 1, 65536, NULL },
	{ OPT_l1cache_size,      "l1cache-size",      TYPE_ID_UINT32, 1, INT_MAX, NULL },
	{ OPT_l1cache_line_size, "l1cache-line-size", TYPE_ID_UINT32, 1, INT_MAX, NULL },
	{ OPT_l1cache_method,    "l1cache-method",    TYPE_ID_SIZE_T_METHOD, 0, 0, stress_l1cache_method },
	{ OPT_l1cache_mlock,     "l1cache-mlock",     TYPE_ID_BOOL, 0, 1, NULL },
	{ OPT_l1cache_ways,      "l1cache-ways",      TYPE_ID_UINT32, 1, 65536, NULL },
	END_OPT,
};

static int stress_l1cache(stress_args_t *args)
{
	int ret, rc = EXIT_SUCCESS;
	uint32_t l1cache_ways = 0;
	uint32_t l1cache_size = 0;
	uint32_t l1cache_sets = 0;
	uint32_t l1cache_line_size = 0;
	uint32_t l1cache_set_size;
	uint8_t *cache, *cache_aligned;
	uintptr_t addr;
	uint32_t padding;
	size_t l1cache_method = 0;	/* Default forward */
	const size_t verify = (g_opt_flags & OPT_FLAGS_VERIFY) ? 1 : 0;
	l1cache_func_t stress_l1cache_func;
	bool l1cache_mlock = false;

	(void)stress_get_setting("l1cache-ways", &l1cache_ways);
	(void)stress_get_setting("l1cache-size", &l1cache_size);
	(void)stress_get_setting("l1cache-sets", &l1cache_sets);
	(void)stress_get_setting("l1cache-line-size", &l1cache_line_size);
	(void)stress_get_setting("l1cache-method", &l1cache_method);
	(void)stress_get_setting("l1cache-mlock", &l1cache_mlock);

	stress_l1cache_func = stress_l1cache_methods[l1cache_method].func[verify];

	ret = stress_l1cache_info_ok(args, &l1cache_ways, &l1cache_size,
				     &l1cache_sets, &l1cache_line_size);
	if (ret != EXIT_SUCCESS)
		return ret;

	cache = (uint8_t *)stress_mmap_populate(NULL, l1cache_size << 2,
				PROT_READ | PROT_WRITE,
				MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
	if (cache == MAP_FAILED) {
		pr_inf_skip("%s: cannot mmap cache %" PRIu32" byte test buffer%s, "
			"skipping stressor, errno=%d (%s)\n",
			args->name, l1cache_size << 2,
			stress_get_memfree_str(), errno, strerror(errno));
		return EXIT_NO_RESOURCE;
	}
	stress_set_vma_anon_name(cache, l1cache_size, "l1cache");
	(void)stress_madvise_mergeable(cache, l1cache_size << 2);

	if (l1cache_mlock)
		(void)shim_mlock(cache, l1cache_size << 2);

	/* Get cache aligned buffer */
	l1cache_set_size = l1cache_ways * l1cache_line_size;
	if (l1cache_set_size == 0) {
		pr_inf("%s: invalid level 1 cache set size is zero\n", args->name);
		return EXIT_NO_RESOURCE;
	}
	addr = l1cache_size + (uintptr_t)cache;
	padding = (l1cache_set_size - (addr % l1cache_set_size)) % l1cache_set_size;
	cache_aligned = (uint8_t *)(addr + padding);

	if ((cache_aligned < cache) || (cache_aligned > cache + (l1cache_size << 1))) {
		pr_inf("%s: aligned cache address is out of range\n", args->name);
		(void)munmap((void *)cache, l1cache_size << 2);
		return EXIT_NO_RESOURCE;
	}

	stress_set_proc_state(args->name, STRESS_STATE_SYNC_WAIT);
	stress_sync_start_wait(args);
	stress_set_proc_state(args->name, STRESS_STATE_RUN);

	do {
		if (UNLIKELY(stress_l1cache_func(args, cache_aligned, l1cache_size, l1cache_sets, l1cache_set_size) == EXIT_FAILURE)) {
			rc = EXIT_FAILURE;
			break;
		}

		stress_bogo_add(args, l1cache_sets);
	} while (stress_continue(args));

	stress_set_proc_state(args->name, STRESS_STATE_DEINIT);

	(void)munmap((void *)cache, l1cache_size << 2);

	return rc;
}

const stressor_info_t stress_l1cache_info = {
	.stressor = stress_l1cache,
	.classifier = CLASS_CPU_CACHE,
	.opts = opts,
	.verify = VERIFY_OPTIONAL,
	.help = help
};