[fio.git] / memory.c

/*
 * Memory helpers
 */
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/shm.h>
#include <sys/mman.h>

#include "fio.h"

void fio_unpin_memory(struct thread_data *td)
{
	if (td->pinned_mem) {
		dprint(FD_MEM, "unpinning %llu bytes\n", td->o.lockmem);
		if (munlock(td->pinned_mem, td->o.lockmem) < 0)
			perror("munlock");
		munmap(td->pinned_mem, td->o.lockmem);
		td->pinned_mem = NULL;
	}
}

int fio_pin_memory(struct thread_data *td)
{
	unsigned long long phys_mem;

	if (!td->o.lockmem)
		return 0;

	dprint(FD_MEM, "pinning %llu bytes\n", td->o.lockmem);

	/*
	 * Don't allow mlock of more than real_mem-128MB
	 */
	phys_mem = os_phys_mem();
	if (phys_mem) {
		if ((td->o.lockmem + 128 * 1024 * 1024) > phys_mem) {
			td->o.lockmem = phys_mem - 128 * 1024 * 1024;
			log_info("fio: limiting mlocked memory to %lluMB\n",
							td->o.lockmem >> 20);
		}
	}

	td->pinned_mem = mmap(NULL, td->o.lockmem, PROT_READ | PROT_WRITE,
				MAP_PRIVATE | OS_MAP_ANON, -1, 0);
	if (td->pinned_mem == MAP_FAILED) {
		perror("malloc locked mem");
		td->pinned_mem = NULL;
		return 1;
	}
	if (mlock(td->pinned_mem, td->o.lockmem) < 0) {
		perror("mlock");
		munmap(td->pinned_mem, td->o.lockmem);
		td->pinned_mem = NULL;
		return 1;
	}

	return 0;
}

static int alloc_mem_shm(struct thread_data *td, unsigned int total_mem)
{
	int flags = IPC_CREAT | S_IRUSR | S_IWUSR;

	if (td->o.mem_type == MEM_SHMHUGE) {
		unsigned long mask = td->o.hugepage_size - 1;

		flags |= SHM_HUGETLB;
		total_mem = (total_mem + mask) & ~mask;
	}

	td->shm_id = shmget(IPC_PRIVATE, total_mem, flags);
	dprint(FD_MEM, "shmget %u, %d\n", total_mem, td->shm_id);
	if (td->shm_id < 0) {
		td_verror(td, errno, "shmget");
		if (geteuid() != 0 && (errno == ENOMEM || errno == EPERM))
			log_err("fio: you may need to run this job as root\n");
		if (td->o.mem_type == MEM_SHMHUGE) {
			if (errno == EINVAL) {
				log_err("fio: check that you have free huge"
					" pages and that hugepage-size is"
					" correct.\n");
			} else if (errno == ENOSYS) {
				log_err("fio: your system does not appear to"
					" support huge pages.\n");
			} else if (errno == ENOMEM) {
				log_err("fio: no huge pages available, do you"
					" need to alocate some? See HOWTO.\n");
			}
		}

		return 1;
	}

	td->orig_buffer = shmat(td->shm_id, NULL, 0);
	dprint(FD_MEM, "shmat %d, %p\n", td->shm_id, td->orig_buffer);
	if (td->orig_buffer == (void *) -1) {
		td_verror(td, errno, "shmat");
		td->orig_buffer = NULL;
		return 1;
	}

	return 0;
}

static void free_mem_shm(struct thread_data *td)
{
	struct shmid_ds sbuf;

	dprint(FD_MEM, "shmdt/ctl %d %p\n", td->shm_id, td->orig_buffer);
	shmdt(td->orig_buffer);
	shmctl(td->shm_id, IPC_RMID, &sbuf);
}

static int alloc_mem_mmap(struct thread_data *td, size_t total_mem)
{
	int flags = MAP_PRIVATE;

	td->mmapfd = 1;

	if (td->o.mmapfile) {
		td->mmapfd = open(td->o.mmapfile, O_RDWR|O_CREAT, 0644);

		if (td->mmapfd < 0) {
			td_verror(td, errno, "open mmap file");
			td->orig_buffer = NULL;
			return 1;
		}
		if (ftruncate(td->mmapfd, total_mem) < 0) {
			td_verror(td, errno, "truncate mmap file");
			td->orig_buffer = NULL;
			return 1;
		}
	} else
		flags |= OS_MAP_ANON;

	td->orig_buffer = mmap(NULL, total_mem, PROT_READ | PROT_WRITE, flags,
				td->mmapfd, 0);
	dprint(FD_MEM, "mmap %u/%d %p\n", total_mem, td->mmapfd,
						td->orig_buffer);
	if (td->orig_buffer == MAP_FAILED) {
		td_verror(td, errno, "mmap");
		td->orig_buffer = NULL;
		if (td->mmapfd) {
			close(td->mmapfd);
			unlink(td->o.mmapfile);
		}

		return 1;
	}

	return 0;
}

static void free_mem_mmap(struct thread_data *td, size_t total_mem)
{
	dprint(FD_MEM, "munmap %u %p\n", total_mem, td->orig_buffer);
	munmap(td->orig_buffer, td->orig_buffer_size);
	if (td->o.mmapfile) {
		close(td->mmapfd);
		unlink(td->o.mmapfile);
		free(td->o.mmapfile);
	}
}

static int alloc_mem_malloc(struct thread_data *td, size_t total_mem)
{
	td->orig_buffer = malloc(total_mem);
	dprint(FD_MEM, "malloc %u %p\n", total_mem, td->orig_buffer);

	return td->orig_buffer == NULL;
}

static void free_mem_malloc(struct thread_data *td)
{
	dprint(FD_MEM, "free malloc mem %p\n", td->orig_buffer);
	free(td->orig_buffer);
}

/*
 * Set up the buffer area we need for io.
 */
int allocate_io_mem(struct thread_data *td)
{
	size_t total_mem;
	int ret = 0;

	if (td->io_ops->flags & FIO_NOIO)
		return 0;

	total_mem = td->orig_buffer_size;

	if (td->o.odirect || td->o.mem_align ||
	    (td->io_ops->flags & FIO_MEMALIGN)) {
		total_mem += page_mask;
		if (td->o.mem_align && td->o.mem_align > page_size)
			total_mem += td->o.mem_align - page_size;
	}

	dprint(FD_MEM, "Alloc %lu for buffers\n", (size_t) total_mem);

	if (td->o.mem_type == MEM_MALLOC)
		ret = alloc_mem_malloc(td, total_mem);
	else if (td->o.mem_type == MEM_SHM || td->o.mem_type == MEM_SHMHUGE)
		ret = alloc_mem_shm(td, total_mem);
	else if (td->o.mem_type == MEM_MMAP || td->o.mem_type == MEM_MMAPHUGE)
		ret = alloc_mem_mmap(td, total_mem);
	else {
		log_err("fio: bad mem type: %d\n", td->o.mem_type);
		ret = 1;
	}

	if (ret)
		td_verror(td, ENOMEM, "iomem allocation");

	return ret;
}

void free_io_mem(struct thread_data *td)
{
	unsigned int total_mem;

	total_mem = td->orig_buffer_size;
	if (td->o.odirect)
		total_mem += page_mask;

	if (td->o.mem_type == MEM_MALLOC)
		free_mem_malloc(td);
	else if (td->o.mem_type == MEM_SHM || td->o.mem_type == MEM_SHMHUGE)
		free_mem_shm(td);
	else if (td->o.mem_type == MEM_MMAP || td->o.mem_type == MEM_MMAPHUGE)
		free_mem_mmap(td, total_mem);
	else
		log_err("Bad memory type %u\n", td->o.mem_type);

	td->orig_buffer = NULL;
	td->orig_buffer_size = 0;
}
Commit	Line	Data
2f9ade3c JA	1	/*
	2	* Memory helpers
	3	*/
5921e80c JA	4	#include <sys/types.h>
	5	#include <sys/stat.h>
	6	#include <fcntl.h>
2f9ade3c JA	7	#include <unistd.h>
	8	#include <sys/shm.h>
	9	#include <sys/mman.h>
	10
	11	#include "fio.h"
2f9ade3c	12
1b79a070	13	void fio_unpin_memory(struct thread_data *td)
2f9ade3c	14	{
1b79a070 JA	15	if (td->pinned_mem) {
	16	dprint(FD_MEM, "unpinning %llu bytes\n", td->o.lockmem);
	17	if (munlock(td->pinned_mem, td->o.lockmem) < 0)
2f9ade3c	18	perror("munlock");
1b79a070 JA	19	munmap(td->pinned_mem, td->o.lockmem);
1b79a070 JA	20	td->pinned_mem = NULL;
2f9ade3c JA	21	}
	22	}
	23
1b79a070	24	int fio_pin_memory(struct thread_data *td)
2f9ade3c JA	25	{
	26	unsigned long long phys_mem;
	27
1b79a070	28	if (!td->o.lockmem)
2f9ade3c JA	29	return 0;
2f9ade3c JA	30
1b79a070	31	dprint(FD_MEM, "pinning %llu bytes\n", td->o.lockmem);
ee56ad50	32
2f9ade3c JA	33	/*
	34	* Don't allow mlock of more than real_mem-128MB
	35	*/
	36	phys_mem = os_phys_mem();
	37	if (phys_mem) {
1b79a070 JA	38	if ((td->o.lockmem + 128 * 1024 * 1024) > phys_mem) {
1b79a070 JA	39	td->o.lockmem = phys_mem - 128 * 1024 * 1024;
b22989b9	40	log_info("fio: limiting mlocked memory to %lluMB\n",
1b79a070	41	td->o.lockmem >> 20);
2f9ade3c JA	42	}
	43	}
	44
1b79a070	45	td->pinned_mem = mmap(NULL, td->o.lockmem, PROT_READ \| PROT_WRITE,
a55820db	46	MAP_PRIVATE \| OS_MAP_ANON, -1, 0);
1b79a070	47	if (td->pinned_mem == MAP_FAILED) {
2f9ade3c	48	perror("malloc locked mem");
1b79a070	49	td->pinned_mem = NULL;
2f9ade3c JA	50	return 1;
2f9ade3c JA	51	}
1b79a070	52	if (mlock(td->pinned_mem, td->o.lockmem) < 0) {
2f9ade3c	53	perror("mlock");
1b79a070 JA	54	munmap(td->pinned_mem, td->o.lockmem);
1b79a070 JA	55	td->pinned_mem = NULL;
2f9ade3c JA	56	return 1;
	57	}
	58
	59	return 0;
	60	}
	61
829a602c	62	static int alloc_mem_shm(struct thread_data *td, unsigned int total_mem)
2f9ade3c	63	{
03e20d68	64	int flags = IPC_CREAT \| S_IRUSR \| S_IWUSR;
b6f9676e	65
a1242a20 JA	66	if (td->o.mem_type == MEM_SHMHUGE) {
	67	unsigned long mask = td->o.hugepage_size - 1;
	68
b6f9676e	69	flags \|= SHM_HUGETLB;
a1242a20 JA	70	total_mem = (total_mem + mask) & ~mask;
a1242a20 JA	71	}
b6f9676e	72
829a602c JA	73	td->shm_id = shmget(IPC_PRIVATE, total_mem, flags);
829a602c JA	74	dprint(FD_MEM, "shmget %u, %d\n", total_mem, td->shm_id);
b6f9676e JA	75	if (td->shm_id < 0) {
b6f9676e JA	76	td_verror(td, errno, "shmget");
da7d79b0	77	if (geteuid() != 0 && (errno == ENOMEM \|\| errno == EPERM))
b6f9676e	78	log_err("fio: you may need to run this job as root\n");
886b878a	79	if (td->o.mem_type == MEM_SHMHUGE) {
5ec10eaa JA	80	if (errno == EINVAL) {
	81	log_err("fio: check that you have free huge"
	82	" pages and that hugepage-size is"
	83	" correct.\n");
	84	} else if (errno == ENOSYS) {
	85	log_err("fio: your system does not appear to"
	86	" support huge pages.\n");
	87	} else if (errno == ENOMEM) {
	88	log_err("fio: no huge pages available, do you"
	89	" need to alocate some? See HOWTO.\n");
	90	}
d8602dd0	91	}
5ec10eaa	92
b6f9676e JA	93	return 1;
b6f9676e JA	94	}
2f9ade3c	95
b6f9676e	96	td->orig_buffer = shmat(td->shm_id, NULL, 0);
ee56ad50	97	dprint(FD_MEM, "shmat %d, %p\n", td->shm_id, td->orig_buffer);
b6f9676e JA	98	if (td->orig_buffer == (void *) -1) {
	99	td_verror(td, errno, "shmat");
	100	td->orig_buffer = NULL;
	101	return 1;
	102	}
	103
	104	return 0;
	105	}
	106
829a602c JA	107	static void free_mem_shm(struct thread_data *td)
	108	{
	109	struct shmid_ds sbuf;
	110
	111	dprint(FD_MEM, "shmdt/ctl %d %p\n", td->shm_id, td->orig_buffer);
	112	shmdt(td->orig_buffer);
	113	shmctl(td->shm_id, IPC_RMID, &sbuf);
	114	}
	115
0f805c00	116	static int alloc_mem_mmap(struct thread_data *td, size_t total_mem)
b6f9676e JA	117	{
	118	int flags = MAP_PRIVATE;
	119
a55820db	120	td->mmapfd = 1;
b6f9676e	121
83ea422a JA	122	if (td->o.mmapfile) {
83ea422a JA	123	td->mmapfd = open(td->o.mmapfile, O_RDWR\|O_CREAT, 0644);
b6f9676e JA	124
	125	if (td->mmapfd < 0) {
	126	td_verror(td, errno, "open mmap file");
2f9ade3c JA	127	td->orig_buffer = NULL;
	128	return 1;
	129	}
829a602c	130	if (ftruncate(td->mmapfd, total_mem) < 0) {
b6f9676e	131	td_verror(td, errno, "truncate mmap file");
2f9ade3c JA	132	td->orig_buffer = NULL;
	133	return 1;
	134	}
b6f9676e JA	135	} else
	136	flags \|= OS_MAP_ANON;
	137
829a602c JA	138	td->orig_buffer = mmap(NULL, total_mem, PROT_READ \| PROT_WRITE, flags,
	139	td->mmapfd, 0);
	140	dprint(FD_MEM, "mmap %u/%d %p\n", total_mem, td->mmapfd,
	141	td->orig_buffer);
b6f9676e JA	142	if (td->orig_buffer == MAP_FAILED) {
	143	td_verror(td, errno, "mmap");
	144	td->orig_buffer = NULL;
	145	if (td->mmapfd) {
	146	close(td->mmapfd);
83ea422a	147	unlink(td->o.mmapfile);
b6f9676e	148	}
5ec10eaa	149
b6f9676e	150	return 1;
2f9ade3c JA	151	}
	152
	153	return 0;
	154	}
	155
0f805c00	156	static void free_mem_mmap(struct thread_data *td, size_t total_mem)
b6f9676e	157	{
829a602c JA	158	dprint(FD_MEM, "munmap %u %p\n", total_mem, td->orig_buffer);
829a602c JA	159	munmap(td->orig_buffer, td->orig_buffer_size);
83ea422a	160	if (td->o.mmapfile) {
829a602c	161	close(td->mmapfd);
83ea422a JA	162	unlink(td->o.mmapfile);
83ea422a JA	163	free(td->o.mmapfile);
829a602c JA	164	}
829a602c JA	165	}
d87612ac	166
0f805c00	167	static int alloc_mem_malloc(struct thread_data *td, size_t total_mem)
829a602c JA	168	{
	169	td->orig_buffer = malloc(total_mem);
	170	dprint(FD_MEM, "malloc %u %p\n", total_mem, td->orig_buffer);
5ec10eaa	171
829a602c JA	172	return td->orig_buffer == NULL;
829a602c JA	173	}
b6f9676e	174
829a602c JA	175	static void free_mem_malloc(struct thread_data *td)
	176	{
	177	dprint(FD_MEM, "free malloc mem %p\n", td->orig_buffer);
	178	free(td->orig_buffer);
b6f9676e JA	179	}
	180
	181	/*
03e20d68	182	* Set up the buffer area we need for io.
b6f9676e JA	183	*/
	184	int allocate_io_mem(struct thread_data *td)
	185	{
0f805c00	186	size_t total_mem;
b6f9676e JA	187	int ret = 0;
b6f9676e JA	188
b4c5e1ac JA	189	if (td->io_ops->flags & FIO_NOIO)
	190	return 0;
	191
829a602c	192	total_mem = td->orig_buffer_size;
d529ee19	193
ca7e0ddb JA	194	if (td->o.odirect \|\| td->o.mem_align \|\|
ca7e0ddb JA	195	(td->io_ops->flags & FIO_MEMALIGN)) {
829a602c	196	total_mem += page_mask;
d529ee19 JA	197	if (td->o.mem_align && td->o.mem_align > page_size)
	198	total_mem += td->o.mem_align - page_size;
	199	}
829a602c	200
0f805c00 JA	201	dprint(FD_MEM, "Alloc %lu for buffers\n", (size_t) total_mem);
0f805c00 JA	202
b6f9676e	203	if (td->o.mem_type == MEM_MALLOC)
829a602c	204	ret = alloc_mem_malloc(td, total_mem);
b6f9676e	205	else if (td->o.mem_type == MEM_SHM \|\| td->o.mem_type == MEM_SHMHUGE)
829a602c	206	ret = alloc_mem_shm(td, total_mem);
b6f9676e	207	else if (td->o.mem_type == MEM_MMAP \|\| td->o.mem_type == MEM_MMAPHUGE)
829a602c	208	ret = alloc_mem_mmap(td, total_mem);
b6f9676e JA	209	else {
	210	log_err("fio: bad mem type: %d\n", td->o.mem_type);
	211	ret = 1;
	212	}
	213
3deb3101 JA	214	if (ret)
	215	td_verror(td, ENOMEM, "iomem allocation");
	216
b6f9676e JA	217	return ret;
	218	}
	219
2f9ade3c JA	220	void free_io_mem(struct thread_data *td)
2f9ade3c JA	221	{
829a602c JA	222	unsigned int total_mem;
	223
	224	total_mem = td->orig_buffer_size;
	225	if (td->o.odirect)
	226	total_mem += page_mask;
	227
	228	if (td->o.mem_type == MEM_MALLOC)
	229	free_mem_malloc(td);
	230	else if (td->o.mem_type == MEM_SHM \|\| td->o.mem_type == MEM_SHMHUGE)
	231	free_mem_shm(td);
	232	else if (td->o.mem_type == MEM_MMAP \|\| td->o.mem_type == MEM_MMAPHUGE)
	233	free_mem_mmap(td, total_mem);
	234	else
2dc1bbeb	235	log_err("Bad memory type %u\n", td->o.mem_type);
2f9ade3c JA	236
2f9ade3c JA	237	td->orig_buffer = NULL;
829a602c	238	td->orig_buffer_size = 0;
2f9ade3c	239	}