[fio.git] / memory.c

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

#include "fio.h"

static void *pinned_mem;

void fio_unpin_memory(void)
{
	if (pinned_mem) {
		dprint(FD_MEM, "unpinning %llu bytes\n", mlock_size);
		if (munlock(pinned_mem, mlock_size) < 0)
			perror("munlock");
		munmap(pinned_mem, mlock_size);
		pinned_mem = NULL;
	}
}

int fio_pin_memory(void)
{
	unsigned long long phys_mem;

	if (!mlock_size)
		return 0;

	dprint(FD_MEM, "pinning %llu bytes\n", mlock_size);

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

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

	td->mmapfd = 1;

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

		/* TODO: make sure the file is a real hugetlbfs file */
		if (!td->mmapfile)
			flags |= MAP_HUGETLB;
		total_mem = (total_mem + mask) & ~mask;
	}

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

		if (td->mmapfd < 0) {
			td_verror(td, errno, "open mmap file");
			td->orig_buffer = NULL;
			return 1;
		}
		if (td->o.mem_type != MEM_MMAPHUGE &&
		    ftruncate(td->mmapfd, total_mem) < 0) {
			td_verror(td, errno, "truncate mmap file");
			td->orig_buffer = NULL;
			return 1;
		}
		if (td->o.mem_type == MEM_MMAPHUGE)
			flags |= MAP_SHARED;
		else
			flags |= MAP_PRIVATE;
	} else
		flags |= OS_MAP_ANON | MAP_PRIVATE;

	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->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->mmapfile) {
		close(td->mmapfd);
		unlink(td->mmapfile);
		free(td->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	7	#include <unistd.h>
ec5c6b12	8	#ifndef FIO_NO_HAVE_SHM_H
2f9ade3c	9	#include <sys/shm.h>
ec5c6b12	10	#endif
2f9ade3c JA	11	#include <sys/mman.h>
	12
	13	#include "fio.h"
2f9ade3c	14
2f9ade3c JA	15	static void *pinned_mem;
	16
	17	void fio_unpin_memory(void)
	18	{
	19	if (pinned_mem) {
ee56ad50	20	dprint(FD_MEM, "unpinning %llu bytes\n", mlock_size);
2f9ade3c JA	21	if (munlock(pinned_mem, mlock_size) < 0)
	22	perror("munlock");
	23	munmap(pinned_mem, mlock_size);
	24	pinned_mem = NULL;
	25	}
	26	}
	27
	28	int fio_pin_memory(void)
	29	{
	30	unsigned long long phys_mem;
	31
	32	if (!mlock_size)
	33	return 0;
	34
ee56ad50 JA	35	dprint(FD_MEM, "pinning %llu bytes\n", mlock_size);
ee56ad50 JA	36
2f9ade3c JA	37	/*
	38	* Don't allow mlock of more than real_mem-128MB
	39	*/
	40	phys_mem = os_phys_mem();
	41	if (phys_mem) {
	42	if ((mlock_size + 128 * 1024 * 1024) > phys_mem) {
	43	mlock_size = phys_mem - 128 * 1024 * 1024;
b22989b9	44	log_info("fio: limiting mlocked memory to %lluMB\n",
5ec10eaa	45	mlock_size >> 20);
2f9ade3c JA	46	}
	47	}
	48
5ec10eaa	49	pinned_mem = mmap(NULL, mlock_size, PROT_READ \| PROT_WRITE,
a55820db	50	MAP_PRIVATE \| OS_MAP_ANON, -1, 0);
2f9ade3c JA	51	if (pinned_mem == MAP_FAILED) {
	52	perror("malloc locked mem");
	53	pinned_mem = NULL;
	54	return 1;
	55	}
	56	if (mlock(pinned_mem, mlock_size) < 0) {
	57	perror("mlock");
	58	munmap(pinned_mem, mlock_size);
	59	pinned_mem = NULL;
	60	return 1;
	61	}
	62
	63	return 0;
	64	}
	65
829a602c	66	static int alloc_mem_shm(struct thread_data *td, unsigned int total_mem)
2f9ade3c	67	{
03e20d68	68	int flags = IPC_CREAT \| S_IRUSR \| S_IWUSR;
b6f9676e	69
a1242a20 JA	70	if (td->o.mem_type == MEM_SHMHUGE) {
	71	unsigned long mask = td->o.hugepage_size - 1;
	72
b6f9676e	73	flags \|= SHM_HUGETLB;
a1242a20 JA	74	total_mem = (total_mem + mask) & ~mask;
a1242a20 JA	75	}
b6f9676e	76
829a602c JA	77	td->shm_id = shmget(IPC_PRIVATE, total_mem, flags);
829a602c JA	78	dprint(FD_MEM, "shmget %u, %d\n", total_mem, td->shm_id);
b6f9676e JA	79	if (td->shm_id < 0) {
b6f9676e JA	80	td_verror(td, errno, "shmget");
da7d79b0	81	if (geteuid() != 0 && (errno == ENOMEM \|\| errno == EPERM))
b6f9676e	82	log_err("fio: you may need to run this job as root\n");
886b878a	83	if (td->o.mem_type == MEM_SHMHUGE) {
5ec10eaa JA	84	if (errno == EINVAL) {
	85	log_err("fio: check that you have free huge"
	86	" pages and that hugepage-size is"
	87	" correct.\n");
	88	} else if (errno == ENOSYS) {
	89	log_err("fio: your system does not appear to"
	90	" support huge pages.\n");
	91	} else if (errno == ENOMEM) {
	92	log_err("fio: no huge pages available, do you"
	93	" need to alocate some? See HOWTO.\n");
	94	}
d8602dd0	95	}
5ec10eaa	96
b6f9676e JA	97	return 1;
b6f9676e JA	98	}
2f9ade3c	99
b6f9676e	100	td->orig_buffer = shmat(td->shm_id, NULL, 0);
ee56ad50	101	dprint(FD_MEM, "shmat %d, %p\n", td->shm_id, td->orig_buffer);
b6f9676e JA	102	if (td->orig_buffer == (void *) -1) {
	103	td_verror(td, errno, "shmat");
	104	td->orig_buffer = NULL;
	105	return 1;
	106	}
	107
	108	return 0;
	109	}
	110
829a602c JA	111	static void free_mem_shm(struct thread_data *td)
	112	{
	113	struct shmid_ds sbuf;
	114
	115	dprint(FD_MEM, "shmdt/ctl %d %p\n", td->shm_id, td->orig_buffer);
	116	shmdt(td->orig_buffer);
	117	shmctl(td->shm_id, IPC_RMID, &sbuf);
	118	}
	119
0f805c00	120	static int alloc_mem_mmap(struct thread_data *td, size_t total_mem)
b6f9676e	121	{
d9759b1e	122	int flags = 0;
b6f9676e	123
a55820db	124	td->mmapfd = 1;
b6f9676e	125
d6dc02fb JA	126	if (td->o.mem_type == MEM_MMAPHUGE) {
	127	unsigned long mask = td->o.hugepage_size - 1;
	128
d9759b1e SL	129	/* TODO: make sure the file is a real hugetlbfs file */
	130	if (!td->mmapfile)
	131	flags \|= MAP_HUGETLB;
d6dc02fb JA	132	total_mem = (total_mem + mask) & ~mask;
	133	}
	134
b6f9676e JA	135	if (td->mmapfile) {
	136	td->mmapfd = open(td->mmapfile, O_RDWR\|O_CREAT, 0644);
	137
	138	if (td->mmapfd < 0) {
	139	td_verror(td, errno, "open mmap file");
2f9ade3c JA	140	td->orig_buffer = NULL;
	141	return 1;
	142	}
d9759b1e SL	143	if (td->o.mem_type != MEM_MMAPHUGE &&
d9759b1e SL	144	ftruncate(td->mmapfd, total_mem) < 0) {
b6f9676e	145	td_verror(td, errno, "truncate mmap file");
2f9ade3c JA	146	td->orig_buffer = NULL;
	147	return 1;
	148	}
d9759b1e SL	149	if (td->o.mem_type == MEM_MMAPHUGE)
	150	flags \|= MAP_SHARED;
	151	else
	152	flags \|= MAP_PRIVATE;
b6f9676e	153	} else
d9759b1e	154	flags \|= OS_MAP_ANON \| MAP_PRIVATE;
b6f9676e	155
829a602c JA	156	td->orig_buffer = mmap(NULL, total_mem, PROT_READ \| PROT_WRITE, flags,
	157	td->mmapfd, 0);
	158	dprint(FD_MEM, "mmap %u/%d %p\n", total_mem, td->mmapfd,
	159	td->orig_buffer);
b6f9676e JA	160	if (td->orig_buffer == MAP_FAILED) {
	161	td_verror(td, errno, "mmap");
	162	td->orig_buffer = NULL;
	163	if (td->mmapfd) {
	164	close(td->mmapfd);
	165	unlink(td->mmapfile);
	166	}
5ec10eaa	167
b6f9676e	168	return 1;
2f9ade3c JA	169	}
	170
	171	return 0;
	172	}
	173
0f805c00	174	static void free_mem_mmap(struct thread_data *td, size_t total_mem)
b6f9676e	175	{
829a602c JA	176	dprint(FD_MEM, "munmap %u %p\n", total_mem, td->orig_buffer);
	177	munmap(td->orig_buffer, td->orig_buffer_size);
	178	if (td->mmapfile) {
	179	close(td->mmapfd);
	180	unlink(td->mmapfile);
	181	free(td->mmapfile);
	182	}
	183	}
d87612ac	184
0f805c00	185	static int alloc_mem_malloc(struct thread_data *td, size_t total_mem)
829a602c JA	186	{
	187	td->orig_buffer = malloc(total_mem);
	188	dprint(FD_MEM, "malloc %u %p\n", total_mem, td->orig_buffer);
5ec10eaa	189
829a602c JA	190	return td->orig_buffer == NULL;
829a602c JA	191	}
b6f9676e	192
829a602c JA	193	static void free_mem_malloc(struct thread_data *td)
	194	{
	195	dprint(FD_MEM, "free malloc mem %p\n", td->orig_buffer);
	196	free(td->orig_buffer);
b6f9676e JA	197	}
	198
	199	/*
03e20d68	200	* Set up the buffer area we need for io.
b6f9676e JA	201	*/
	202	int allocate_io_mem(struct thread_data *td)
	203	{
0f805c00	204	size_t total_mem;
b6f9676e JA	205	int ret = 0;
b6f9676e JA	206
b4c5e1ac JA	207	if (td->io_ops->flags & FIO_NOIO)
	208	return 0;
	209
829a602c	210	total_mem = td->orig_buffer_size;
d529ee19	211
ca7e0ddb JA	212	if (td->o.odirect \|\| td->o.mem_align \|\|
ca7e0ddb JA	213	(td->io_ops->flags & FIO_MEMALIGN)) {
829a602c	214	total_mem += page_mask;
d529ee19 JA	215	if (td->o.mem_align && td->o.mem_align > page_size)
	216	total_mem += td->o.mem_align - page_size;
	217	}
829a602c	218
0f805c00 JA	219	dprint(FD_MEM, "Alloc %lu for buffers\n", (size_t) total_mem);
0f805c00 JA	220
b6f9676e	221	if (td->o.mem_type == MEM_MALLOC)
829a602c	222	ret = alloc_mem_malloc(td, total_mem);
b6f9676e	223	else if (td->o.mem_type == MEM_SHM \|\| td->o.mem_type == MEM_SHMHUGE)
829a602c	224	ret = alloc_mem_shm(td, total_mem);
b6f9676e	225	else if (td->o.mem_type == MEM_MMAP \|\| td->o.mem_type == MEM_MMAPHUGE)
829a602c	226	ret = alloc_mem_mmap(td, total_mem);
b6f9676e JA	227	else {
	228	log_err("fio: bad mem type: %d\n", td->o.mem_type);
	229	ret = 1;
	230	}
	231
3deb3101 JA	232	if (ret)
	233	td_verror(td, ENOMEM, "iomem allocation");
	234
b6f9676e JA	235	return ret;
	236	}
	237
2f9ade3c JA	238	void free_io_mem(struct thread_data *td)
2f9ade3c JA	239	{
829a602c JA	240	unsigned int total_mem;
	241
	242	total_mem = td->orig_buffer_size;
	243	if (td->o.odirect)
	244	total_mem += page_mask;
	245
	246	if (td->o.mem_type == MEM_MALLOC)
	247	free_mem_malloc(td);
	248	else if (td->o.mem_type == MEM_SHM \|\| td->o.mem_type == MEM_SHMHUGE)
	249	free_mem_shm(td);
	250	else if (td->o.mem_type == MEM_MMAP \|\| td->o.mem_type == MEM_MMAPHUGE)
	251	free_mem_mmap(td, total_mem);
	252	else
2dc1bbeb	253	log_err("Bad memory type %u\n", td->o.mem_type);
2f9ade3c JA	254
2f9ade3c JA	255	td->orig_buffer = NULL;
829a602c	256	td->orig_buffer_size = 0;
2f9ade3c	257	}