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 *
pds_vfio_get_lm_file(const struct file_operations * fops,int flags,u64 size)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(sizeof(*lm_file), GFP_KERNEL);
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_array(npages, sizeof(*pages), GFP_KERNEL);
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
pds_vfio_put_lm_file(struct pds_vfio_lm_file * lm_file)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
pds_vfio_put_save_file(struct pds_vfio_pci_device * pds_vfio)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
pds_vfio_put_restore_file(struct pds_vfio_pci_device * pds_vfio)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
pds_vfio_get_file_page(struct pds_vfio_lm_file * lm_file,unsigned long offset)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 nth_page(sg_page(sg),
155 (offset - cur_offset) / PAGE_SIZE);
156 }
157 cur_offset += sg->length;
158 }
159
160 return NULL;
161 }
162
pds_vfio_release_file(struct inode * inode,struct file * filp)163 static int pds_vfio_release_file(struct inode *inode, struct file *filp)
164 {
165 struct pds_vfio_lm_file *lm_file = filp->private_data;
166
167 mutex_lock(&lm_file->lock);
168 lm_file->filep->f_pos = 0;
169 lm_file->size = 0;
170 mutex_unlock(&lm_file->lock);
171 mutex_destroy(&lm_file->lock);
172 kfree(lm_file);
173
174 return 0;
175 }
176
pds_vfio_save_read(struct file * filp,char __user * buf,size_t len,loff_t * pos)177 static ssize_t pds_vfio_save_read(struct file *filp, char __user *buf,
178 size_t len, loff_t *pos)
179 {
180 struct pds_vfio_lm_file *lm_file = filp->private_data;
181 ssize_t done = 0;
182
183 if (pos)
184 return -ESPIPE;
185 pos = &filp->f_pos;
186
187 mutex_lock(&lm_file->lock);
188
189 if (lm_file->disabled) {
190 done = -ENODEV;
191 goto out_unlock;
192 }
193
194 if (*pos > lm_file->size) {
195 done = -EINVAL;
196 goto out_unlock;
197 }
198
199 len = min_t(size_t, lm_file->size - *pos, len);
200 while (len) {
201 size_t page_offset;
202 struct page *page;
203 size_t page_len;
204 u8 *from_buff;
205 int err;
206
207 page_offset = (*pos) % PAGE_SIZE;
208 page = pds_vfio_get_file_page(lm_file, *pos - page_offset);
209 if (!page) {
210 if (done == 0)
211 done = -EINVAL;
212 goto out_unlock;
213 }
214
215 page_len = min_t(size_t, len, PAGE_SIZE - page_offset);
216 from_buff = kmap_local_page(page);
217 err = copy_to_user(buf, from_buff + page_offset, page_len);
218 kunmap_local(from_buff);
219 if (err) {
220 done = -EFAULT;
221 goto out_unlock;
222 }
223 *pos += page_len;
224 len -= page_len;
225 done += page_len;
226 buf += page_len;
227 }
228
229 out_unlock:
230 mutex_unlock(&lm_file->lock);
231 return done;
232 }
233
234 static const struct file_operations pds_vfio_save_fops = {
235 .owner = THIS_MODULE,
236 .read = pds_vfio_save_read,
237 .release = pds_vfio_release_file,
238 };
239
pds_vfio_get_save_file(struct pds_vfio_pci_device * pds_vfio)240 static int pds_vfio_get_save_file(struct pds_vfio_pci_device *pds_vfio)
241 {
242 struct device *dev = &pds_vfio->vfio_coredev.pdev->dev;
243 struct pds_vfio_lm_file *lm_file;
244 u64 size;
245 int err;
246
247 /* Get live migration state size in this state */
248 err = pds_vfio_get_lm_state_size_cmd(pds_vfio, &size);
249 if (err) {
250 dev_err(dev, "failed to get save status: %pe\n", ERR_PTR(err));
251 return err;
252 }
253
254 dev_dbg(dev, "save status, size = %lld\n", size);
255
256 if (!size) {
257 dev_err(dev, "invalid state size\n");
258 return -EIO;
259 }
260
261 lm_file = pds_vfio_get_lm_file(&pds_vfio_save_fops, O_RDONLY, size);
262 if (!lm_file) {
263 dev_err(dev, "failed to create save file\n");
264 return -ENOENT;
265 }
266
267 dev_dbg(dev, "size = %lld, alloc_size = %lld, npages = %lld\n",
268 lm_file->size, lm_file->alloc_size, lm_file->npages);
269
270 pds_vfio->save_file = lm_file;
271
272 return 0;
273 }
274
pds_vfio_restore_write(struct file * filp,const char __user * buf,size_t len,loff_t * pos)275 static ssize_t pds_vfio_restore_write(struct file *filp, const char __user *buf,
276 size_t len, loff_t *pos)
277 {
278 struct pds_vfio_lm_file *lm_file = filp->private_data;
279 loff_t requested_length;
280 ssize_t done = 0;
281
282 if (pos)
283 return -ESPIPE;
284
285 pos = &filp->f_pos;
286
287 if (*pos < 0 ||
288 check_add_overflow((loff_t)len, *pos, &requested_length))
289 return -EINVAL;
290
291 mutex_lock(&lm_file->lock);
292
293 if (lm_file->disabled) {
294 done = -ENODEV;
295 goto out_unlock;
296 }
297
298 while (len) {
299 size_t page_offset;
300 struct page *page;
301 size_t page_len;
302 u8 *to_buff;
303 int err;
304
305 page_offset = (*pos) % PAGE_SIZE;
306 page = pds_vfio_get_file_page(lm_file, *pos - page_offset);
307 if (!page) {
308 if (done == 0)
309 done = -EINVAL;
310 goto out_unlock;
311 }
312
313 page_len = min_t(size_t, len, PAGE_SIZE - page_offset);
314 to_buff = kmap_local_page(page);
315 err = copy_from_user(to_buff + page_offset, buf, page_len);
316 kunmap_local(to_buff);
317 if (err) {
318 done = -EFAULT;
319 goto out_unlock;
320 }
321 *pos += page_len;
322 len -= page_len;
323 done += page_len;
324 buf += page_len;
325 lm_file->size += page_len;
326 }
327 out_unlock:
328 mutex_unlock(&lm_file->lock);
329 return done;
330 }
331
332 static const struct file_operations pds_vfio_restore_fops = {
333 .owner = THIS_MODULE,
334 .write = pds_vfio_restore_write,
335 .release = pds_vfio_release_file,
336 };
337
pds_vfio_get_restore_file(struct pds_vfio_pci_device * pds_vfio)338 static int pds_vfio_get_restore_file(struct pds_vfio_pci_device *pds_vfio)
339 {
340 struct device *dev = &pds_vfio->vfio_coredev.pdev->dev;
341 struct pds_vfio_lm_file *lm_file;
342 u64 size;
343
344 size = sizeof(union pds_lm_dev_state);
345 dev_dbg(dev, "restore status, size = %lld\n", size);
346
347 if (!size) {
348 dev_err(dev, "invalid state size");
349 return -EIO;
350 }
351
352 lm_file = pds_vfio_get_lm_file(&pds_vfio_restore_fops, O_WRONLY, size);
353 if (!lm_file) {
354 dev_err(dev, "failed to create restore file");
355 return -ENOENT;
356 }
357 pds_vfio->restore_file = lm_file;
358
359 return 0;
360 }
361
362 struct file *
pds_vfio_step_device_state_locked(struct pds_vfio_pci_device * pds_vfio,enum vfio_device_mig_state next)363 pds_vfio_step_device_state_locked(struct pds_vfio_pci_device *pds_vfio,
364 enum vfio_device_mig_state next)
365 {
366 enum vfio_device_mig_state cur = pds_vfio->state;
367 int err;
368
369 if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_STOP_COPY) {
370 err = pds_vfio_get_save_file(pds_vfio);
371 if (err)
372 return ERR_PTR(err);
373
374 err = pds_vfio_get_lm_state_cmd(pds_vfio);
375 if (err) {
376 pds_vfio_put_save_file(pds_vfio);
377 return ERR_PTR(err);
378 }
379
380 return pds_vfio->save_file->filep;
381 }
382
383 if (cur == VFIO_DEVICE_STATE_STOP_COPY && next == VFIO_DEVICE_STATE_STOP) {
384 pds_vfio_put_save_file(pds_vfio);
385 pds_vfio_dirty_disable(pds_vfio, true);
386 return NULL;
387 }
388
389 if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RESUMING) {
390 err = pds_vfio_get_restore_file(pds_vfio);
391 if (err)
392 return ERR_PTR(err);
393
394 return pds_vfio->restore_file->filep;
395 }
396
397 if (cur == VFIO_DEVICE_STATE_RESUMING && next == VFIO_DEVICE_STATE_STOP) {
398 err = pds_vfio_set_lm_state_cmd(pds_vfio);
399 if (err)
400 return ERR_PTR(err);
401
402 pds_vfio_put_restore_file(pds_vfio);
403 return NULL;
404 }
405
406 if (cur == VFIO_DEVICE_STATE_RUNNING && next == VFIO_DEVICE_STATE_RUNNING_P2P) {
407 pds_vfio_send_host_vf_lm_status_cmd(pds_vfio,
408 PDS_LM_STA_IN_PROGRESS);
409 err = pds_vfio_suspend_device_cmd(pds_vfio,
410 PDS_LM_SUSPEND_RESUME_TYPE_P2P);
411 if (err)
412 return ERR_PTR(err);
413
414 return NULL;
415 }
416
417 if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_RUNNING) {
418 err = pds_vfio_resume_device_cmd(pds_vfio,
419 PDS_LM_SUSPEND_RESUME_TYPE_FULL);
420 if (err)
421 return ERR_PTR(err);
422
423 pds_vfio_send_host_vf_lm_status_cmd(pds_vfio, PDS_LM_STA_NONE);
424 return NULL;
425 }
426
427 if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RUNNING_P2P) {
428 err = pds_vfio_resume_device_cmd(pds_vfio,
429 PDS_LM_SUSPEND_RESUME_TYPE_P2P);
430 if (err)
431 return ERR_PTR(err);
432
433 return NULL;
434 }
435
436 if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_STOP) {
437 err = pds_vfio_suspend_device_cmd(pds_vfio,
438 PDS_LM_SUSPEND_RESUME_TYPE_FULL);
439 if (err)
440 return ERR_PTR(err);
441 return NULL;
442 }
443
444 return ERR_PTR(-EINVAL);
445 }
446