1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright 2014 IBM Corp.
4 */
5
6 #include <linux/pci.h>
7 #include <linux/slab.h>
8 #include <linux/file.h>
9 #include <misc/cxl.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/pseudo_fs.h>
13 #include <linux/sched/mm.h>
14 #include <linux/mmu_context.h>
15
16 #include "cxl.h"
17
18 /*
19 * Since we want to track memory mappings to be able to force-unmap
20 * when the AFU is no longer reachable, we need an inode. For devices
21 * opened through the cxl user API, this is not a problem, but a
22 * userland process can also get a cxl fd through the cxl_get_fd()
23 * API, which is used by the cxlflash driver.
24 *
25 * Therefore we implement our own simple pseudo-filesystem and inode
26 * allocator. We don't use the anonymous inode, as we need the
27 * meta-data associated with it (address_space) and it is shared by
28 * other drivers/processes, so it could lead to cxl unmapping VMAs
29 * from random processes.
30 */
31
32 #define CXL_PSEUDO_FS_MAGIC 0x1697697f
33
34 static int cxl_fs_cnt;
35 static struct vfsmount *cxl_vfs_mount;
36
cxl_fs_init_fs_context(struct fs_context * fc)37 static int cxl_fs_init_fs_context(struct fs_context *fc)
38 {
39 return init_pseudo(fc, CXL_PSEUDO_FS_MAGIC) ? 0 : -ENOMEM;
40 }
41
42 static struct file_system_type cxl_fs_type = {
43 .name = "cxl",
44 .owner = THIS_MODULE,
45 .init_fs_context = cxl_fs_init_fs_context,
46 .kill_sb = kill_anon_super,
47 };
48
49
cxl_release_mapping(struct cxl_context * ctx)50 void cxl_release_mapping(struct cxl_context *ctx)
51 {
52 if (ctx->kernelapi && ctx->mapping)
53 simple_release_fs(&cxl_vfs_mount, &cxl_fs_cnt);
54 }
55
cxl_getfile(const char * name,const struct file_operations * fops,void * priv,int flags)56 static struct file *cxl_getfile(const char *name,
57 const struct file_operations *fops,
58 void *priv, int flags)
59 {
60 struct file *file;
61 struct inode *inode;
62 int rc;
63
64 /* strongly inspired by anon_inode_getfile() */
65
66 if (fops->owner && !try_module_get(fops->owner))
67 return ERR_PTR(-ENOENT);
68
69 rc = simple_pin_fs(&cxl_fs_type, &cxl_vfs_mount, &cxl_fs_cnt);
70 if (rc < 0) {
71 pr_err("Cannot mount cxl pseudo filesystem: %d\n", rc);
72 file = ERR_PTR(rc);
73 goto err_module;
74 }
75
76 inode = alloc_anon_inode(cxl_vfs_mount->mnt_sb);
77 if (IS_ERR(inode)) {
78 file = ERR_CAST(inode);
79 goto err_fs;
80 }
81
82 file = alloc_file_pseudo(inode, cxl_vfs_mount, name,
83 flags & (O_ACCMODE | O_NONBLOCK), fops);
84 if (IS_ERR(file))
85 goto err_inode;
86
87 file->private_data = priv;
88
89 return file;
90
91 err_inode:
92 iput(inode);
93 err_fs:
94 simple_release_fs(&cxl_vfs_mount, &cxl_fs_cnt);
95 err_module:
96 module_put(fops->owner);
97 return file;
98 }
99
cxl_dev_context_init(struct pci_dev * dev)100 struct cxl_context *cxl_dev_context_init(struct pci_dev *dev)
101 {
102 struct cxl_afu *afu;
103 struct cxl_context *ctx;
104 int rc;
105
106 afu = cxl_pci_to_afu(dev);
107 if (IS_ERR(afu))
108 return ERR_CAST(afu);
109
110 ctx = cxl_context_alloc();
111 if (!ctx)
112 return ERR_PTR(-ENOMEM);
113
114 ctx->kernelapi = true;
115
116 /* Make it a slave context. We can promote it later? */
117 rc = cxl_context_init(ctx, afu, false);
118 if (rc)
119 goto err_ctx;
120
121 return ctx;
122
123 err_ctx:
124 kfree(ctx);
125 return ERR_PTR(rc);
126 }
127 EXPORT_SYMBOL_GPL(cxl_dev_context_init);
128
cxl_get_context(struct pci_dev * dev)129 struct cxl_context *cxl_get_context(struct pci_dev *dev)
130 {
131 return dev->dev.archdata.cxl_ctx;
132 }
133 EXPORT_SYMBOL_GPL(cxl_get_context);
134
cxl_release_context(struct cxl_context * ctx)135 int cxl_release_context(struct cxl_context *ctx)
136 {
137 if (ctx->status >= STARTED)
138 return -EBUSY;
139
140 cxl_context_free(ctx);
141
142 return 0;
143 }
144 EXPORT_SYMBOL_GPL(cxl_release_context);
145
cxl_find_afu_irq(struct cxl_context * ctx,int num)146 static irq_hw_number_t cxl_find_afu_irq(struct cxl_context *ctx, int num)
147 {
148 __u16 range;
149 int r;
150
151 for (r = 0; r < CXL_IRQ_RANGES; r++) {
152 range = ctx->irqs.range[r];
153 if (num < range) {
154 return ctx->irqs.offset[r] + num;
155 }
156 num -= range;
157 }
158 return 0;
159 }
160
161
cxl_set_priv(struct cxl_context * ctx,void * priv)162 int cxl_set_priv(struct cxl_context *ctx, void *priv)
163 {
164 if (!ctx)
165 return -EINVAL;
166
167 ctx->priv = priv;
168
169 return 0;
170 }
171 EXPORT_SYMBOL_GPL(cxl_set_priv);
172
cxl_get_priv(struct cxl_context * ctx)173 void *cxl_get_priv(struct cxl_context *ctx)
174 {
175 if (!ctx)
176 return ERR_PTR(-EINVAL);
177
178 return ctx->priv;
179 }
180 EXPORT_SYMBOL_GPL(cxl_get_priv);
181
cxl_allocate_afu_irqs(struct cxl_context * ctx,int num)182 int cxl_allocate_afu_irqs(struct cxl_context *ctx, int num)
183 {
184 int res;
185 irq_hw_number_t hwirq;
186
187 if (num == 0)
188 num = ctx->afu->pp_irqs;
189 res = afu_allocate_irqs(ctx, num);
190 if (res)
191 return res;
192
193 if (!cpu_has_feature(CPU_FTR_HVMODE)) {
194 /* In a guest, the PSL interrupt is not multiplexed. It was
195 * allocated above, and we need to set its handler
196 */
197 hwirq = cxl_find_afu_irq(ctx, 0);
198 if (hwirq)
199 cxl_map_irq(ctx->afu->adapter, hwirq, cxl_ops->psl_interrupt, ctx, "psl");
200 }
201
202 if (ctx->status == STARTED) {
203 if (cxl_ops->update_ivtes)
204 cxl_ops->update_ivtes(ctx);
205 else WARN(1, "BUG: cxl_allocate_afu_irqs must be called prior to starting the context on this platform\n");
206 }
207
208 return res;
209 }
210 EXPORT_SYMBOL_GPL(cxl_allocate_afu_irqs);
211
cxl_free_afu_irqs(struct cxl_context * ctx)212 void cxl_free_afu_irqs(struct cxl_context *ctx)
213 {
214 irq_hw_number_t hwirq;
215 unsigned int virq;
216
217 if (!cpu_has_feature(CPU_FTR_HVMODE)) {
218 hwirq = cxl_find_afu_irq(ctx, 0);
219 if (hwirq) {
220 virq = irq_find_mapping(NULL, hwirq);
221 if (virq)
222 cxl_unmap_irq(virq, ctx);
223 }
224 }
225 afu_irq_name_free(ctx);
226 cxl_ops->release_irq_ranges(&ctx->irqs, ctx->afu->adapter);
227 }
228 EXPORT_SYMBOL_GPL(cxl_free_afu_irqs);
229
cxl_map_afu_irq(struct cxl_context * ctx,int num,irq_handler_t handler,void * cookie,char * name)230 int cxl_map_afu_irq(struct cxl_context *ctx, int num,
231 irq_handler_t handler, void *cookie, char *name)
232 {
233 irq_hw_number_t hwirq;
234
235 /*
236 * Find interrupt we are to register.
237 */
238 hwirq = cxl_find_afu_irq(ctx, num);
239 if (!hwirq)
240 return -ENOENT;
241
242 return cxl_map_irq(ctx->afu->adapter, hwirq, handler, cookie, name);
243 }
244 EXPORT_SYMBOL_GPL(cxl_map_afu_irq);
245
cxl_unmap_afu_irq(struct cxl_context * ctx,int num,void * cookie)246 void cxl_unmap_afu_irq(struct cxl_context *ctx, int num, void *cookie)
247 {
248 irq_hw_number_t hwirq;
249 unsigned int virq;
250
251 hwirq = cxl_find_afu_irq(ctx, num);
252 if (!hwirq)
253 return;
254
255 virq = irq_find_mapping(NULL, hwirq);
256 if (virq)
257 cxl_unmap_irq(virq, cookie);
258 }
259 EXPORT_SYMBOL_GPL(cxl_unmap_afu_irq);
260
261 /*
262 * Start a context
263 * Code here similar to afu_ioctl_start_work().
264 */
cxl_start_context(struct cxl_context * ctx,u64 wed,struct task_struct * task)265 int cxl_start_context(struct cxl_context *ctx, u64 wed,
266 struct task_struct *task)
267 {
268 int rc = 0;
269 bool kernel = true;
270
271 pr_devel("%s: pe: %i\n", __func__, ctx->pe);
272
273 mutex_lock(&ctx->status_mutex);
274 if (ctx->status == STARTED)
275 goto out; /* already started */
276
277 /*
278 * Increment the mapped context count for adapter. This also checks
279 * if adapter_context_lock is taken.
280 */
281 rc = cxl_adapter_context_get(ctx->afu->adapter);
282 if (rc)
283 goto out;
284
285 if (task) {
286 ctx->pid = get_task_pid(task, PIDTYPE_PID);
287 kernel = false;
288
289 /* acquire a reference to the task's mm */
290 ctx->mm = get_task_mm(current);
291
292 /* ensure this mm_struct can't be freed */
293 cxl_context_mm_count_get(ctx);
294
295 if (ctx->mm) {
296 /* decrement the use count from above */
297 mmput(ctx->mm);
298 /* make TLBIs for this context global */
299 mm_context_add_copro(ctx->mm);
300 }
301 }
302
303 /*
304 * Increment driver use count. Enables global TLBIs for hash
305 * and callbacks to handle the segment table
306 */
307 cxl_ctx_get();
308
309 /* See the comment in afu_ioctl_start_work() */
310 smp_mb();
311
312 if ((rc = cxl_ops->attach_process(ctx, kernel, wed, 0))) {
313 put_pid(ctx->pid);
314 ctx->pid = NULL;
315 cxl_adapter_context_put(ctx->afu->adapter);
316 cxl_ctx_put();
317 if (task) {
318 cxl_context_mm_count_put(ctx);
319 if (ctx->mm)
320 mm_context_remove_copro(ctx->mm);
321 }
322 goto out;
323 }
324
325 ctx->status = STARTED;
326 out:
327 mutex_unlock(&ctx->status_mutex);
328 return rc;
329 }
330 EXPORT_SYMBOL_GPL(cxl_start_context);
331
cxl_process_element(struct cxl_context * ctx)332 int cxl_process_element(struct cxl_context *ctx)
333 {
334 return ctx->external_pe;
335 }
336 EXPORT_SYMBOL_GPL(cxl_process_element);
337
338 /* Stop a context. Returns 0 on success, otherwise -Errno */
cxl_stop_context(struct cxl_context * ctx)339 int cxl_stop_context(struct cxl_context *ctx)
340 {
341 return __detach_context(ctx);
342 }
343 EXPORT_SYMBOL_GPL(cxl_stop_context);
344
cxl_set_master(struct cxl_context * ctx)345 void cxl_set_master(struct cxl_context *ctx)
346 {
347 ctx->master = true;
348 }
349 EXPORT_SYMBOL_GPL(cxl_set_master);
350
351 /* wrappers around afu_* file ops which are EXPORTED */
cxl_fd_open(struct inode * inode,struct file * file)352 int cxl_fd_open(struct inode *inode, struct file *file)
353 {
354 return afu_open(inode, file);
355 }
356 EXPORT_SYMBOL_GPL(cxl_fd_open);
cxl_fd_release(struct inode * inode,struct file * file)357 int cxl_fd_release(struct inode *inode, struct file *file)
358 {
359 return afu_release(inode, file);
360 }
361 EXPORT_SYMBOL_GPL(cxl_fd_release);
cxl_fd_ioctl(struct file * file,unsigned int cmd,unsigned long arg)362 long cxl_fd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
363 {
364 return afu_ioctl(file, cmd, arg);
365 }
366 EXPORT_SYMBOL_GPL(cxl_fd_ioctl);
cxl_fd_mmap(struct file * file,struct vm_area_struct * vm)367 int cxl_fd_mmap(struct file *file, struct vm_area_struct *vm)
368 {
369 return afu_mmap(file, vm);
370 }
371 EXPORT_SYMBOL_GPL(cxl_fd_mmap);
cxl_fd_poll(struct file * file,struct poll_table_struct * poll)372 __poll_t cxl_fd_poll(struct file *file, struct poll_table_struct *poll)
373 {
374 return afu_poll(file, poll);
375 }
376 EXPORT_SYMBOL_GPL(cxl_fd_poll);
cxl_fd_read(struct file * file,char __user * buf,size_t count,loff_t * off)377 ssize_t cxl_fd_read(struct file *file, char __user *buf, size_t count,
378 loff_t *off)
379 {
380 return afu_read(file, buf, count, off);
381 }
382 EXPORT_SYMBOL_GPL(cxl_fd_read);
383
384 #define PATCH_FOPS(NAME) if (!fops->NAME) fops->NAME = afu_fops.NAME
385
386 /* Get a struct file and fd for a context and attach the ops */
cxl_get_fd(struct cxl_context * ctx,struct file_operations * fops,int * fd)387 struct file *cxl_get_fd(struct cxl_context *ctx, struct file_operations *fops,
388 int *fd)
389 {
390 struct file *file;
391 int rc, flags, fdtmp;
392 char *name = NULL;
393
394 /* only allow one per context */
395 if (ctx->mapping)
396 return ERR_PTR(-EEXIST);
397
398 flags = O_RDWR | O_CLOEXEC;
399
400 /* This code is similar to anon_inode_getfd() */
401 rc = get_unused_fd_flags(flags);
402 if (rc < 0)
403 return ERR_PTR(rc);
404 fdtmp = rc;
405
406 /*
407 * Patch the file ops. Needs to be careful that this is rentrant safe.
408 */
409 if (fops) {
410 PATCH_FOPS(open);
411 PATCH_FOPS(poll);
412 PATCH_FOPS(read);
413 PATCH_FOPS(release);
414 PATCH_FOPS(unlocked_ioctl);
415 PATCH_FOPS(compat_ioctl);
416 PATCH_FOPS(mmap);
417 } else /* use default ops */
418 fops = (struct file_operations *)&afu_fops;
419
420 name = kasprintf(GFP_KERNEL, "cxl:%d", ctx->pe);
421 file = cxl_getfile(name, fops, ctx, flags);
422 kfree(name);
423 if (IS_ERR(file))
424 goto err_fd;
425
426 cxl_context_set_mapping(ctx, file->f_mapping);
427 *fd = fdtmp;
428 return file;
429
430 err_fd:
431 put_unused_fd(fdtmp);
432 return NULL;
433 }
434 EXPORT_SYMBOL_GPL(cxl_get_fd);
435
cxl_fops_get_context(struct file * file)436 struct cxl_context *cxl_fops_get_context(struct file *file)
437 {
438 return file->private_data;
439 }
440 EXPORT_SYMBOL_GPL(cxl_fops_get_context);
441
cxl_set_driver_ops(struct cxl_context * ctx,struct cxl_afu_driver_ops * ops)442 void cxl_set_driver_ops(struct cxl_context *ctx,
443 struct cxl_afu_driver_ops *ops)
444 {
445 WARN_ON(!ops->fetch_event || !ops->event_delivered);
446 atomic_set(&ctx->afu_driver_events, 0);
447 ctx->afu_driver_ops = ops;
448 }
449 EXPORT_SYMBOL_GPL(cxl_set_driver_ops);
450
cxl_context_events_pending(struct cxl_context * ctx,unsigned int new_events)451 void cxl_context_events_pending(struct cxl_context *ctx,
452 unsigned int new_events)
453 {
454 atomic_add(new_events, &ctx->afu_driver_events);
455 wake_up_all(&ctx->wq);
456 }
457 EXPORT_SYMBOL_GPL(cxl_context_events_pending);
458
cxl_start_work(struct cxl_context * ctx,struct cxl_ioctl_start_work * work)459 int cxl_start_work(struct cxl_context *ctx,
460 struct cxl_ioctl_start_work *work)
461 {
462 int rc;
463
464 /* code taken from afu_ioctl_start_work */
465 if (!(work->flags & CXL_START_WORK_NUM_IRQS))
466 work->num_interrupts = ctx->afu->pp_irqs;
467 else if ((work->num_interrupts < ctx->afu->pp_irqs) ||
468 (work->num_interrupts > ctx->afu->irqs_max)) {
469 return -EINVAL;
470 }
471
472 rc = afu_register_irqs(ctx, work->num_interrupts);
473 if (rc)
474 return rc;
475
476 rc = cxl_start_context(ctx, work->work_element_descriptor, current);
477 if (rc < 0) {
478 afu_release_irqs(ctx, ctx);
479 return rc;
480 }
481
482 return 0;
483 }
484 EXPORT_SYMBOL_GPL(cxl_start_work);
485
cxl_psa_map(struct cxl_context * ctx)486 void __iomem *cxl_psa_map(struct cxl_context *ctx)
487 {
488 if (ctx->status != STARTED)
489 return NULL;
490
491 pr_devel("%s: psn_phys%llx size:%llx\n",
492 __func__, ctx->psn_phys, ctx->psn_size);
493 return ioremap(ctx->psn_phys, ctx->psn_size);
494 }
495 EXPORT_SYMBOL_GPL(cxl_psa_map);
496
cxl_psa_unmap(void __iomem * addr)497 void cxl_psa_unmap(void __iomem *addr)
498 {
499 iounmap(addr);
500 }
501 EXPORT_SYMBOL_GPL(cxl_psa_unmap);
502
cxl_afu_reset(struct cxl_context * ctx)503 int cxl_afu_reset(struct cxl_context *ctx)
504 {
505 struct cxl_afu *afu = ctx->afu;
506 int rc;
507
508 rc = cxl_ops->afu_reset(afu);
509 if (rc)
510 return rc;
511
512 return cxl_ops->afu_check_and_enable(afu);
513 }
514 EXPORT_SYMBOL_GPL(cxl_afu_reset);
515
cxl_perst_reloads_same_image(struct cxl_afu * afu,bool perst_reloads_same_image)516 void cxl_perst_reloads_same_image(struct cxl_afu *afu,
517 bool perst_reloads_same_image)
518 {
519 afu->adapter->perst_same_image = perst_reloads_same_image;
520 }
521 EXPORT_SYMBOL_GPL(cxl_perst_reloads_same_image);
522
cxl_read_adapter_vpd(struct pci_dev * dev,void * buf,size_t count)523 ssize_t cxl_read_adapter_vpd(struct pci_dev *dev, void *buf, size_t count)
524 {
525 struct cxl_afu *afu = cxl_pci_to_afu(dev);
526 if (IS_ERR(afu))
527 return -ENODEV;
528
529 return cxl_ops->read_adapter_vpd(afu->adapter, buf, count);
530 }
531 EXPORT_SYMBOL_GPL(cxl_read_adapter_vpd);
532