| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* Copyright(c) 2023 Advanced Micro Devices, Inc. */ |
| 3 |
|
| 4 |
#include <linux/anon_inodes.h> |
| 5 |
#include <linux/file.h> |
| 6 |
#include <linux/fs.h> |
| 7 |
#include <linux/highmem.h> |
| 8 |
#include <linux/vfio.h> |
| 9 |
#include <linux/vfio_pci_core.h> |
| 10 |
|
| 11 |
#include "vfio_dev.h" |
| 12 |
#include "cmds.h" |
| 13 |
|
| 14 |
static struct pds_vfio_lm_file * |
| 15 |
pds_vfio_get_lm_file(const struct file_operations *fops, int flags, u64 size) |
| 16 |
{ |
| 17 |
struct pds_vfio_lm_file *lm_file = NULL; |
| 18 |
unsigned long long npages; |
| 19 |
struct page **pages; |
| 20 |
void *page_mem; |
| 21 |
const void *p; |
| 22 |
|
| 23 |
if (!size) |
| 24 |
return NULL; |
| 25 |
|
| 26 |
/* Alloc file structure */ |
| 27 |
lm_file = kzalloc_obj(*lm_file); |
| 28 |
if (!lm_file) |
| 29 |
return NULL; |
| 30 |
|
| 31 |
/* Create file */ |
| 32 |
lm_file->filep = |
| 33 |
anon_inode_getfile("pds_vfio_lm", fops, lm_file, flags); |
| 34 |
if (IS_ERR(lm_file->filep)) |
| 35 |
goto out_free_file; |
| 36 |
|
| 37 |
stream_open(lm_file->filep->f_inode, lm_file->filep); |
| 38 |
mutex_init(&lm_file->lock); |
| 39 |
|
| 40 |
/* prevent file from being released before we are done with it */ |
| 41 |
get_file(lm_file->filep); |
| 42 |
|
| 43 |
/* Allocate memory for file pages */ |
| 44 |
npages = DIV_ROUND_UP_ULL(size, PAGE_SIZE); |
| 45 |
pages = kmalloc_objs(*pages, npages); |
| 46 |
if (!pages) |
| 47 |
goto out_put_file; |
| 48 |
|
| 49 |
page_mem = kvzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL); |
| 50 |
if (!page_mem) |
| 51 |
goto out_free_pages_array; |
| 52 |
|
| 53 |
p = page_mem - offset_in_page(page_mem); |
| 54 |
for (unsigned long long i = 0; i < npages; i++) { |
| 55 |
if (is_vmalloc_addr(p)) |
| 56 |
pages[i] = vmalloc_to_page(p); |
| 57 |
else |
| 58 |
pages[i] = kmap_to_page((void *)p); |
| 59 |
if (!pages[i]) |
| 60 |
goto out_free_page_mem; |
| 61 |
|
| 62 |
p += PAGE_SIZE; |
| 63 |
} |
| 64 |
|
| 65 |
/* Create scatterlist of file pages to use for DMA mapping later */ |
| 66 |
if (sg_alloc_table_from_pages(&lm_file->sg_table, pages, npages, 0, |
| 67 |
size, GFP_KERNEL)) |
| 68 |
goto out_free_page_mem; |
| 69 |
|
| 70 |
lm_file->size = size; |
| 71 |
lm_file->pages = pages; |
| 72 |
lm_file->npages = npages; |
| 73 |
lm_file->page_mem = page_mem; |
| 74 |
lm_file->alloc_size = npages * PAGE_SIZE; |
| 75 |
|
| 76 |
return lm_file; |
| 77 |
|
| 78 |
out_free_page_mem: |
| 79 |
kvfree(page_mem); |
| 80 |
out_free_pages_array: |
| 81 |
kfree(pages); |
| 82 |
out_put_file: |
| 83 |
fput(lm_file->filep); |
| 84 |
mutex_destroy(&lm_file->lock); |
| 85 |
out_free_file: |
| 86 |
kfree(lm_file); |
| 87 |
|
| 88 |
return NULL; |
| 89 |
} |
| 90 |
|
| 91 |
static void pds_vfio_put_lm_file(struct pds_vfio_lm_file *lm_file) |
| 92 |
{ |
| 93 |
mutex_lock(&lm_file->lock); |
| 94 |
|
| 95 |
lm_file->disabled = true; |
| 96 |
lm_file->size = 0; |
| 97 |
lm_file->alloc_size = 0; |
| 98 |
lm_file->filep->f_pos = 0; |
| 99 |
|
| 100 |
/* Free scatter list of file pages */ |
| 101 |
sg_free_table(&lm_file->sg_table); |
| 102 |
|
| 103 |
kvfree(lm_file->page_mem); |
| 104 |
lm_file->page_mem = NULL; |
| 105 |
kfree(lm_file->pages); |
| 106 |
lm_file->pages = NULL; |
| 107 |
|
| 108 |
mutex_unlock(&lm_file->lock); |
| 109 |
|
| 110 |
/* allow file to be released since we are done with it */ |
| 111 |
fput(lm_file->filep); |
| 112 |
} |
| 113 |
|
| 114 |
void pds_vfio_put_save_file(struct pds_vfio_pci_device *pds_vfio) |
| 115 |
{ |
| 116 |
if (!pds_vfio->save_file) |
| 117 |
return; |
| 118 |
|
| 119 |
pds_vfio_put_lm_file(pds_vfio->save_file); |
| 120 |
pds_vfio->save_file = NULL; |
| 121 |
} |
| 122 |
|
| 123 |
void pds_vfio_put_restore_file(struct pds_vfio_pci_device *pds_vfio) |
| 124 |
{ |
| 125 |
if (!pds_vfio->restore_file) |
| 126 |
return; |
| 127 |
|
| 128 |
pds_vfio_put_lm_file(pds_vfio->restore_file); |
| 129 |
pds_vfio->restore_file = NULL; |
| 130 |
} |
| 131 |
|
| 132 |
static struct page *pds_vfio_get_file_page(struct pds_vfio_lm_file *lm_file, |
| 133 |
unsigned long offset) |
| 134 |
{ |
| 135 |
unsigned long cur_offset = 0; |
| 136 |
struct scatterlist *sg; |
| 137 |
unsigned int i; |
| 138 |
|
| 139 |
/* All accesses are sequential */ |
| 140 |
if (offset < lm_file->last_offset || !lm_file->last_offset_sg) { |
| 141 |
lm_file->last_offset = 0; |
| 142 |
lm_file->last_offset_sg = lm_file->sg_table.sgl; |
| 143 |
lm_file->sg_last_entry = 0; |
| 144 |
} |
| 145 |
|
| 146 |
cur_offset = lm_file->last_offset; |
| 147 |
|
| 148 |
for_each_sg(lm_file->last_offset_sg, sg, |
| 149 |
lm_file->sg_table.orig_nents - lm_file->sg_last_entry, i) { |
| 150 |
if (offset < sg->length + cur_offset) { |
| 151 |
lm_file->last_offset_sg = sg; |
| 152 |
lm_file->sg_last_entry += i; |
| 153 |
lm_file->last_offset = cur_offset; |
| 154 |
return sg_page(sg) + (offset - cur_offset) / PAGE_SIZE; |
| 155 |
} |
| 156 |
cur_offset += sg->length; |
| 157 |
} |
| 158 |
|
| 159 |
return NULL; |
| 160 |
} |
| 161 |
|
| 162 |
static int pds_vfio_release_file(struct inode *inode, struct file *filp) |
| 163 |
{ |
| 164 |
struct pds_vfio_lm_file *lm_file = filp->private_data; |
| 165 |
|
| 166 |
mutex_lock(&lm_file->lock); |
| 167 |
lm_file->filep->f_pos = 0; |
| 168 |
lm_file->size = 0; |
| 169 |
mutex_unlock(&lm_file->lock); |
| 170 |
mutex_destroy(&lm_file->lock); |
| 171 |
kfree(lm_file); |
| 172 |
|
| 173 |
return 0; |
| 174 |
} |
| 175 |
|
| 176 |
static ssize_t pds_vfio_save_read(struct file *filp, char __user *buf, |
| 177 |
size_t len, loff_t *pos) |
| 178 |
{ |
| 179 |
struct pds_vfio_lm_file *lm_file = filp->private_data; |
| 180 |
ssize_t done = 0; |
| 181 |
|
| 182 |
if (pos) |
| 183 |
return -ESPIPE; |
| 184 |
pos = &filp->f_pos; |
| 185 |
|
| 186 |
mutex_lock(&lm_file->lock); |
| 187 |
|
| 188 |
if (lm_file->disabled) { |
| 189 |
done = -ENODEV; |
| 190 |
goto out_unlock; |
| 191 |
} |
| 192 |
|
| 193 |
if (*pos > lm_file->size) { |
| 194 |
done = -EINVAL; |
| 195 |
goto out_unlock; |
| 196 |
} |
| 197 |
|
| 198 |
len = min_t(size_t, lm_file->size - *pos, len); |
| 199 |
while (len) { |
| 200 |
size_t page_offset; |
| 201 |
struct page *page; |
| 202 |
size_t page_len; |
| 203 |
u8 *from_buff; |
| 204 |
int err; |
| 205 |
|
| 206 |
page_offset = (*pos) % PAGE_SIZE; |
| 207 |
page = pds_vfio_get_file_page(lm_file, *pos - page_offset); |
| 208 |
if (!page) { |
| 209 |
if (done == 0) |
| 210 |
done = -EINVAL; |
| 211 |
goto out_unlock; |
| 212 |
} |
| 213 |
|
| 214 |
page_len = min_t(size_t, len, PAGE_SIZE - page_offset); |
| 215 |
from_buff = kmap_local_page(page); |
| 216 |
err = copy_to_user(buf, from_buff + page_offset, page_len); |
| 217 |
kunmap_local(from_buff); |
| 218 |
if (err) { |
| 219 |
done = -EFAULT; |
| 220 |
goto out_unlock; |
| 221 |
} |
| 222 |
*pos += page_len; |
| 223 |
len -= page_len; |
| 224 |
done += page_len; |
| 225 |
buf += page_len; |
| 226 |
} |
| 227 |
|
| 228 |
out_unlock: |
| 229 |
mutex_unlock(&lm_file->lock); |
| 230 |
return done; |
| 231 |
} |
| 232 |
|
| 233 |
static const struct file_operations pds_vfio_save_fops = { |
| 234 |
.owner = THIS_MODULE, |
| 235 |
.read = pds_vfio_save_read, |
| 236 |
.release = pds_vfio_release_file, |
| 237 |
}; |
| 238 |
|
| 239 |
static int pds_vfio_get_save_file(struct pds_vfio_pci_device *pds_vfio) |
| 240 |
{ |
| 241 |
struct device *dev = &pds_vfio->vfio_coredev.pdev->dev; |
| 242 |
struct pds_vfio_lm_file *lm_file; |
| 243 |
u64 size; |
| 244 |
int err; |
| 245 |
|
| 246 |
/* Get live migration state size in this state */ |
| 247 |
err = pds_vfio_get_lm_state_size_cmd(pds_vfio, &size); |
| 248 |
if (err) { |
| 249 |
dev_err(dev, "failed to get save status: %pe\n", ERR_PTR(err)); |
| 250 |
return err; |
| 251 |
} |
| 252 |
|
| 253 |
dev_dbg(dev, "save status, size = %lld\n", size); |
| 254 |
|
| 255 |
if (!size) { |
| 256 |
dev_err(dev, "invalid state size\n"); |
| 257 |
return -EIO; |
| 258 |
} |
| 259 |
|
| 260 |
lm_file = pds_vfio_get_lm_file(&pds_vfio_save_fops, O_RDONLY, size); |
| 261 |
if (!lm_file) { |
| 262 |
dev_err(dev, "failed to create save file\n"); |
| 263 |
return -ENOENT; |
| 264 |
} |
| 265 |
|
| 266 |
dev_dbg(dev, "size = %lld, alloc_size = %lld, npages = %lld\n", |
| 267 |
lm_file->size, lm_file->alloc_size, lm_file->npages); |
| 268 |
|
| 269 |
pds_vfio->save_file = lm_file; |
| 270 |
|
| 271 |
return 0; |
| 272 |
} |
| 273 |
|
| 274 |
static ssize_t pds_vfio_restore_write(struct file *filp, const char __user *buf, |
| 275 |
size_t len, loff_t *pos) |
| 276 |
{ |
| 277 |
struct pds_vfio_lm_file *lm_file = filp->private_data; |
| 278 |
loff_t requested_length; |
| 279 |
ssize_t done = 0; |
| 280 |
|
| 281 |
if (pos) |
| 282 |
return -ESPIPE; |
| 283 |
|
| 284 |
pos = &filp->f_pos; |
| 285 |
|
| 286 |
if (*pos < 0 || |
| 287 |
check_add_overflow((loff_t)len, *pos, &requested_length)) |
| 288 |
return -EINVAL; |
| 289 |
|
| 290 |
mutex_lock(&lm_file->lock); |
| 291 |
|
| 292 |
if (lm_file->disabled) { |
| 293 |
done = -ENODEV; |
| 294 |
goto out_unlock; |
| 295 |
} |
| 296 |
|
| 297 |
while (len) { |
| 298 |
size_t page_offset; |
| 299 |
struct page *page; |
| 300 |
size_t page_len; |
| 301 |
u8 *to_buff; |
| 302 |
int err; |
| 303 |
|
| 304 |
page_offset = (*pos) % PAGE_SIZE; |
| 305 |
page = pds_vfio_get_file_page(lm_file, *pos - page_offset); |
| 306 |
if (!page) { |
| 307 |
if (done == 0) |
| 308 |
done = -EINVAL; |
| 309 |
goto out_unlock; |
| 310 |
} |
| 311 |
|
| 312 |
page_len = min_t(size_t, len, PAGE_SIZE - page_offset); |
| 313 |
to_buff = kmap_local_page(page); |
| 314 |
err = copy_from_user(to_buff + page_offset, buf, page_len); |
| 315 |
kunmap_local(to_buff); |
| 316 |
if (err) { |
| 317 |
done = -EFAULT; |
| 318 |
goto out_unlock; |
| 319 |
} |
| 320 |
*pos += page_len; |
| 321 |
len -= page_len; |
| 322 |
done += page_len; |
| 323 |
buf += page_len; |
| 324 |
lm_file->size += page_len; |
| 325 |
} |
| 326 |
out_unlock: |
| 327 |
mutex_unlock(&lm_file->lock); |
| 328 |
return done; |
| 329 |
} |
| 330 |
|
| 331 |
static const struct file_operations pds_vfio_restore_fops = { |
| 332 |
.owner = THIS_MODULE, |
| 333 |
.write = pds_vfio_restore_write, |
| 334 |
.release = pds_vfio_release_file, |
| 335 |
}; |
| 336 |
|
| 337 |
static int pds_vfio_get_restore_file(struct pds_vfio_pci_device *pds_vfio) |
| 338 |
{ |
| 339 |
struct device *dev = &pds_vfio->vfio_coredev.pdev->dev; |
| 340 |
struct pds_vfio_lm_file *lm_file; |
| 341 |
u64 size; |
| 342 |
|
| 343 |
size = sizeof(union pds_lm_dev_state); |
| 344 |
dev_dbg(dev, "restore status, size = %lld\n", size); |
| 345 |
|
| 346 |
if (!size) { |
| 347 |
dev_err(dev, "invalid state size"); |
| 348 |
return -EIO; |
| 349 |
} |
| 350 |
|
| 351 |
lm_file = pds_vfio_get_lm_file(&pds_vfio_restore_fops, O_WRONLY, size); |
| 352 |
if (!lm_file) { |
| 353 |
dev_err(dev, "failed to create restore file"); |
| 354 |
return -ENOENT; |
| 355 |
} |
| 356 |
pds_vfio->restore_file = lm_file; |
| 357 |
|
| 358 |
return 0; |
| 359 |
} |
| 360 |
|
| 361 |
struct file * |
| 362 |
pds_vfio_step_device_state_locked(struct pds_vfio_pci_device *pds_vfio, |
| 363 |
enum vfio_device_mig_state next) |
| 364 |
{ |
| 365 |
enum vfio_device_mig_state cur = pds_vfio->state; |
| 366 |
int err; |
| 367 |
|
| 368 |
if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_STOP_COPY) { |
| 369 |
err = pds_vfio_get_save_file(pds_vfio); |
| 370 |
if (err) |
| 371 |
return ERR_PTR(err); |
| 372 |
|
| 373 |
err = pds_vfio_get_lm_state_cmd(pds_vfio); |
| 374 |
if (err) { |
| 375 |
pds_vfio_put_save_file(pds_vfio); |
| 376 |
return ERR_PTR(err); |
| 377 |
} |
| 378 |
|
| 379 |
return pds_vfio->save_file->filep; |
| 380 |
} |
| 381 |
|
| 382 |
if (cur == VFIO_DEVICE_STATE_STOP_COPY && next == VFIO_DEVICE_STATE_STOP) { |
| 383 |
pds_vfio_put_save_file(pds_vfio); |
| 384 |
pds_vfio_dirty_disable(pds_vfio, true); |
| 385 |
return NULL; |
| 386 |
} |
| 387 |
|
| 388 |
if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RESUMING) { |
| 389 |
err = pds_vfio_get_restore_file(pds_vfio); |
| 390 |
if (err) |
| 391 |
return ERR_PTR(err); |
| 392 |
|
| 393 |
return pds_vfio->restore_file->filep; |
| 394 |
} |
| 395 |
|
| 396 |
if (cur == VFIO_DEVICE_STATE_RESUMING && next == VFIO_DEVICE_STATE_STOP) { |
| 397 |
err = pds_vfio_set_lm_state_cmd(pds_vfio); |
| 398 |
if (err) |
| 399 |
return ERR_PTR(err); |
| 400 |
|
| 401 |
pds_vfio_put_restore_file(pds_vfio); |
| 402 |
return NULL; |
| 403 |
} |
| 404 |
|
| 405 |
if (cur == VFIO_DEVICE_STATE_RUNNING && next == VFIO_DEVICE_STATE_RUNNING_P2P) { |
| 406 |
pds_vfio_send_host_vf_lm_status_cmd(pds_vfio, |
| 407 |
PDS_LM_STA_IN_PROGRESS); |
| 408 |
err = pds_vfio_suspend_device_cmd(pds_vfio, |
| 409 |
PDS_LM_SUSPEND_RESUME_TYPE_P2P); |
| 410 |
if (err) |
| 411 |
return ERR_PTR(err); |
| 412 |
|
| 413 |
return NULL; |
| 414 |
} |
| 415 |
|
| 416 |
if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_RUNNING) { |
| 417 |
err = pds_vfio_resume_device_cmd(pds_vfio, |
| 418 |
PDS_LM_SUSPEND_RESUME_TYPE_FULL); |
| 419 |
if (err) |
| 420 |
return ERR_PTR(err); |
| 421 |
|
| 422 |
pds_vfio_send_host_vf_lm_status_cmd(pds_vfio, PDS_LM_STA_NONE); |
| 423 |
return NULL; |
| 424 |
} |
| 425 |
|
| 426 |
if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RUNNING_P2P) { |
| 427 |
err = pds_vfio_resume_device_cmd(pds_vfio, |
| 428 |
PDS_LM_SUSPEND_RESUME_TYPE_P2P); |
| 429 |
if (err) |
| 430 |
return ERR_PTR(err); |
| 431 |
|
| 432 |
return NULL; |
| 433 |
} |
| 434 |
|
| 435 |
if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_STOP) { |
| 436 |
err = pds_vfio_suspend_device_cmd(pds_vfio, |
| 437 |
PDS_LM_SUSPEND_RESUME_TYPE_FULL); |
| 438 |
if (err) |
| 439 |
return ERR_PTR(err); |
| 440 |
return NULL; |
| 441 |
} |
| 442 |
|
| 443 |
return ERR_PTR(-EINVAL); |
| 444 |
} |