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> |
2f9ade3c JA |
8 | #include <sys/mman.h> |
9 | ||
10 | #include "fio.h" | |
a5e0ee11 O |
11 | #ifndef FIO_NO_HAVE_SHM_H |
12 | #include <sys/shm.h> | |
13 | #endif | |
2f9ade3c | 14 | |
1b79a070 | 15 | void fio_unpin_memory(struct thread_data *td) |
2f9ade3c | 16 | { |
1b79a070 JA |
17 | if (td->pinned_mem) { |
18 | dprint(FD_MEM, "unpinning %llu bytes\n", td->o.lockmem); | |
19 | if (munlock(td->pinned_mem, td->o.lockmem) < 0) | |
2f9ade3c | 20 | perror("munlock"); |
1b79a070 JA |
21 | munmap(td->pinned_mem, td->o.lockmem); |
22 | td->pinned_mem = NULL; | |
2f9ade3c JA |
23 | } |
24 | } | |
25 | ||
1b79a070 | 26 | int fio_pin_memory(struct thread_data *td) |
2f9ade3c JA |
27 | { |
28 | unsigned long long phys_mem; | |
29 | ||
1b79a070 | 30 | if (!td->o.lockmem) |
2f9ade3c JA |
31 | return 0; |
32 | ||
1b79a070 | 33 | dprint(FD_MEM, "pinning %llu bytes\n", td->o.lockmem); |
ee56ad50 | 34 | |
2f9ade3c JA |
35 | /* |
36 | * Don't allow mlock of more than real_mem-128MB | |
37 | */ | |
38 | phys_mem = os_phys_mem(); | |
39 | if (phys_mem) { | |
1b79a070 JA |
40 | if ((td->o.lockmem + 128 * 1024 * 1024) > phys_mem) { |
41 | td->o.lockmem = phys_mem - 128 * 1024 * 1024; | |
b22989b9 | 42 | log_info("fio: limiting mlocked memory to %lluMB\n", |
1b79a070 | 43 | td->o.lockmem >> 20); |
2f9ade3c JA |
44 | } |
45 | } | |
46 | ||
1b79a070 | 47 | td->pinned_mem = mmap(NULL, td->o.lockmem, PROT_READ | PROT_WRITE, |
a55820db | 48 | MAP_PRIVATE | OS_MAP_ANON, -1, 0); |
1b79a070 | 49 | if (td->pinned_mem == MAP_FAILED) { |
2f9ade3c | 50 | perror("malloc locked mem"); |
1b79a070 | 51 | td->pinned_mem = NULL; |
2f9ade3c JA |
52 | return 1; |
53 | } | |
1b79a070 | 54 | if (mlock(td->pinned_mem, td->o.lockmem) < 0) { |
2f9ade3c | 55 | perror("mlock"); |
1b79a070 JA |
56 | munmap(td->pinned_mem, td->o.lockmem); |
57 | td->pinned_mem = NULL; | |
2f9ade3c JA |
58 | return 1; |
59 | } | |
60 | ||
61 | return 0; | |
62 | } | |
63 | ||
829a602c | 64 | static int alloc_mem_shm(struct thread_data *td, unsigned int total_mem) |
2f9ade3c | 65 | { |
91e47529 | 66 | #ifndef CONFIG_NO_SHM |
03e20d68 | 67 | int flags = IPC_CREAT | S_IRUSR | S_IWUSR; |
b6f9676e | 68 | |
a1242a20 JA |
69 | if (td->o.mem_type == MEM_SHMHUGE) { |
70 | unsigned long mask = td->o.hugepage_size - 1; | |
71 | ||
b6f9676e | 72 | flags |= SHM_HUGETLB; |
a1242a20 JA |
73 | total_mem = (total_mem + mask) & ~mask; |
74 | } | |
b6f9676e | 75 | |
829a602c JA |
76 | td->shm_id = shmget(IPC_PRIVATE, total_mem, flags); |
77 | dprint(FD_MEM, "shmget %u, %d\n", total_mem, td->shm_id); | |
b6f9676e JA |
78 | if (td->shm_id < 0) { |
79 | td_verror(td, errno, "shmget"); | |
da7d79b0 | 80 | if (geteuid() != 0 && (errno == ENOMEM || errno == EPERM)) |
b6f9676e | 81 | log_err("fio: you may need to run this job as root\n"); |
886b878a | 82 | if (td->o.mem_type == MEM_SHMHUGE) { |
5ec10eaa JA |
83 | if (errno == EINVAL) { |
84 | log_err("fio: check that you have free huge" | |
85 | " pages and that hugepage-size is" | |
86 | " correct.\n"); | |
87 | } else if (errno == ENOSYS) { | |
88 | log_err("fio: your system does not appear to" | |
89 | " support huge pages.\n"); | |
90 | } else if (errno == ENOMEM) { | |
91 | log_err("fio: no huge pages available, do you" | |
92 | " need to alocate some? See HOWTO.\n"); | |
93 | } | |
d8602dd0 | 94 | } |
5ec10eaa | 95 | |
b6f9676e JA |
96 | return 1; |
97 | } | |
2f9ade3c | 98 | |
b6f9676e | 99 | td->orig_buffer = shmat(td->shm_id, NULL, 0); |
ee56ad50 | 100 | dprint(FD_MEM, "shmat %d, %p\n", td->shm_id, td->orig_buffer); |
b6f9676e JA |
101 | if (td->orig_buffer == (void *) -1) { |
102 | td_verror(td, errno, "shmat"); | |
103 | td->orig_buffer = NULL; | |
104 | return 1; | |
105 | } | |
106 | ||
107 | return 0; | |
91e47529 JA |
108 | #else |
109 | log_err("fio: shm not supported\n"); | |
110 | return 1; | |
111 | #endif | |
b6f9676e JA |
112 | } |
113 | ||
829a602c JA |
114 | static void free_mem_shm(struct thread_data *td) |
115 | { | |
91e47529 | 116 | #ifndef CONFIG_NO_SHM |
829a602c JA |
117 | struct shmid_ds sbuf; |
118 | ||
119 | dprint(FD_MEM, "shmdt/ctl %d %p\n", td->shm_id, td->orig_buffer); | |
120 | shmdt(td->orig_buffer); | |
121 | shmctl(td->shm_id, IPC_RMID, &sbuf); | |
91e47529 | 122 | #endif |
829a602c JA |
123 | } |
124 | ||
0f805c00 | 125 | static int alloc_mem_mmap(struct thread_data *td, size_t total_mem) |
b6f9676e | 126 | { |
d9759b1e | 127 | int flags = 0; |
b6f9676e | 128 | |
4a995dda | 129 | td->mmapfd = -1; |
b6f9676e | 130 | |
d6dc02fb JA |
131 | if (td->o.mem_type == MEM_MMAPHUGE) { |
132 | unsigned long mask = td->o.hugepage_size - 1; | |
133 | ||
d9759b1e | 134 | /* TODO: make sure the file is a real hugetlbfs file */ |
836fcc0f | 135 | if (!td->o.mmapfile) |
d9759b1e | 136 | flags |= MAP_HUGETLB; |
d6dc02fb JA |
137 | total_mem = (total_mem + mask) & ~mask; |
138 | } | |
139 | ||
83ea422a JA |
140 | if (td->o.mmapfile) { |
141 | td->mmapfd = open(td->o.mmapfile, O_RDWR|O_CREAT, 0644); | |
b6f9676e JA |
142 | |
143 | if (td->mmapfd < 0) { | |
144 | td_verror(td, errno, "open mmap file"); | |
2f9ade3c JA |
145 | td->orig_buffer = NULL; |
146 | return 1; | |
147 | } | |
d9759b1e | 148 | if (td->o.mem_type != MEM_MMAPHUGE && |
217b0f1d | 149 | td->o.mem_type != MEM_MMAPSHARED && |
d9759b1e | 150 | ftruncate(td->mmapfd, total_mem) < 0) { |
b6f9676e | 151 | td_verror(td, errno, "truncate mmap file"); |
2f9ade3c JA |
152 | td->orig_buffer = NULL; |
153 | return 1; | |
154 | } | |
217b0f1d LG |
155 | if (td->o.mem_type == MEM_MMAPHUGE || |
156 | td->o.mem_type == MEM_MMAPSHARED) | |
d9759b1e SL |
157 | flags |= MAP_SHARED; |
158 | else | |
159 | flags |= MAP_PRIVATE; | |
b6f9676e | 160 | } else |
d9759b1e | 161 | flags |= OS_MAP_ANON | MAP_PRIVATE; |
b6f9676e | 162 | |
829a602c JA |
163 | td->orig_buffer = mmap(NULL, total_mem, PROT_READ | PROT_WRITE, flags, |
164 | td->mmapfd, 0); | |
4b91ee8f JA |
165 | dprint(FD_MEM, "mmap %llu/%d %p\n", (unsigned long long) total_mem, |
166 | td->mmapfd, td->orig_buffer); | |
b6f9676e JA |
167 | if (td->orig_buffer == MAP_FAILED) { |
168 | td_verror(td, errno, "mmap"); | |
169 | td->orig_buffer = NULL; | |
9ce94349 | 170 | if (td->mmapfd != 1 && td->mmapfd != -1) { |
b6f9676e | 171 | close(td->mmapfd); |
b3493a7a JA |
172 | if (td->o.mmapfile) |
173 | unlink(td->o.mmapfile); | |
b6f9676e | 174 | } |
5ec10eaa | 175 | |
b6f9676e | 176 | return 1; |
2f9ade3c JA |
177 | } |
178 | ||
179 | return 0; | |
180 | } | |
181 | ||
0f805c00 | 182 | static void free_mem_mmap(struct thread_data *td, size_t total_mem) |
b6f9676e | 183 | { |
4b91ee8f JA |
184 | dprint(FD_MEM, "munmap %llu %p\n", (unsigned long long) total_mem, |
185 | td->orig_buffer); | |
829a602c | 186 | munmap(td->orig_buffer, td->orig_buffer_size); |
83ea422a | 187 | if (td->o.mmapfile) { |
4a995dda JA |
188 | if (td->mmapfd != -1) |
189 | close(td->mmapfd); | |
83ea422a JA |
190 | unlink(td->o.mmapfile); |
191 | free(td->o.mmapfile); | |
829a602c JA |
192 | } |
193 | } | |
d87612ac | 194 | |
0f805c00 | 195 | static int alloc_mem_malloc(struct thread_data *td, size_t total_mem) |
829a602c JA |
196 | { |
197 | td->orig_buffer = malloc(total_mem); | |
4b91ee8f JA |
198 | dprint(FD_MEM, "malloc %llu %p\n", (unsigned long long) total_mem, |
199 | td->orig_buffer); | |
5ec10eaa | 200 | |
829a602c JA |
201 | return td->orig_buffer == NULL; |
202 | } | |
b6f9676e | 203 | |
829a602c JA |
204 | static void free_mem_malloc(struct thread_data *td) |
205 | { | |
206 | dprint(FD_MEM, "free malloc mem %p\n", td->orig_buffer); | |
207 | free(td->orig_buffer); | |
b6f9676e JA |
208 | } |
209 | ||
210 | /* | |
03e20d68 | 211 | * Set up the buffer area we need for io. |
b6f9676e JA |
212 | */ |
213 | int allocate_io_mem(struct thread_data *td) | |
214 | { | |
0f805c00 | 215 | size_t total_mem; |
b6f9676e JA |
216 | int ret = 0; |
217 | ||
b4c5e1ac JA |
218 | if (td->io_ops->flags & FIO_NOIO) |
219 | return 0; | |
220 | ||
829a602c | 221 | total_mem = td->orig_buffer_size; |
d529ee19 | 222 | |
d01612f3 | 223 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
ca7e0ddb | 224 | (td->io_ops->flags & FIO_MEMALIGN)) { |
829a602c | 225 | total_mem += page_mask; |
d529ee19 JA |
226 | if (td->o.mem_align && td->o.mem_align > page_size) |
227 | total_mem += td->o.mem_align - page_size; | |
228 | } | |
829a602c | 229 | |
4b91ee8f | 230 | dprint(FD_MEM, "Alloc %llu for buffers\n", (unsigned long long) total_mem); |
0f805c00 | 231 | |
a6cb85e2 JA |
232 | /* |
233 | * If the IO engine has hooks to allocate/free memory, use those. But | |
234 | * error out if the user explicitly asked for something else. | |
235 | */ | |
236 | if (td->io_ops->iomem_alloc) { | |
237 | if (fio_option_is_set(&td->o, mem_type)) { | |
238 | log_err("fio: option 'mem/iomem' conflicts with specified IO engine\n"); | |
239 | ret = 1; | |
240 | } else | |
241 | ret = td->io_ops->iomem_alloc(td, total_mem); | |
242 | } else if (td->o.mem_type == MEM_MALLOC) | |
829a602c | 243 | ret = alloc_mem_malloc(td, total_mem); |
b6f9676e | 244 | else if (td->o.mem_type == MEM_SHM || td->o.mem_type == MEM_SHMHUGE) |
829a602c | 245 | ret = alloc_mem_shm(td, total_mem); |
217b0f1d LG |
246 | else if (td->o.mem_type == MEM_MMAP || td->o.mem_type == MEM_MMAPHUGE || |
247 | td->o.mem_type == MEM_MMAPSHARED) | |
829a602c | 248 | ret = alloc_mem_mmap(td, total_mem); |
b6f9676e JA |
249 | else { |
250 | log_err("fio: bad mem type: %d\n", td->o.mem_type); | |
251 | ret = 1; | |
252 | } | |
253 | ||
3deb3101 JA |
254 | if (ret) |
255 | td_verror(td, ENOMEM, "iomem allocation"); | |
256 | ||
b6f9676e JA |
257 | return ret; |
258 | } | |
259 | ||
2f9ade3c JA |
260 | void free_io_mem(struct thread_data *td) |
261 | { | |
829a602c JA |
262 | unsigned int total_mem; |
263 | ||
264 | total_mem = td->orig_buffer_size; | |
d01612f3 | 265 | if (td->o.odirect || td->o.oatomic) |
829a602c JA |
266 | total_mem += page_mask; |
267 | ||
a6cb85e2 JA |
268 | if (td->io_ops->iomem_alloc) { |
269 | if (td->io_ops->iomem_free) | |
270 | td->io_ops->iomem_free(td); | |
271 | } else if (td->o.mem_type == MEM_MALLOC) | |
829a602c JA |
272 | free_mem_malloc(td); |
273 | else if (td->o.mem_type == MEM_SHM || td->o.mem_type == MEM_SHMHUGE) | |
274 | free_mem_shm(td); | |
217b0f1d LG |
275 | else if (td->o.mem_type == MEM_MMAP || td->o.mem_type == MEM_MMAPHUGE || |
276 | td->o.mem_type == MEM_MMAPSHARED) | |
829a602c JA |
277 | free_mem_mmap(td, total_mem); |
278 | else | |
2dc1bbeb | 279 | log_err("Bad memory type %u\n", td->o.mem_type); |
2f9ade3c JA |
280 | |
281 | td->orig_buffer = NULL; | |
829a602c | 282 | td->orig_buffer_size = 0; |
2f9ade3c | 283 | } |