xref: /openbmc/linux/drivers/s390/cio/vfio_ccw_drv.c (revision b4a6aaea)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * VFIO based Physical Subchannel device driver
4  *
5  * Copyright IBM Corp. 2017
6  * Copyright Red Hat, Inc. 2019
7  *
8  * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com>
9  *            Xiao Feng Ren <renxiaof@linux.vnet.ibm.com>
10  *            Cornelia Huck <cohuck@redhat.com>
11  */
12 
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/device.h>
16 #include <linux/slab.h>
17 #include <linux/uuid.h>
18 #include <linux/mdev.h>
19 
20 #include <asm/isc.h>
21 
22 #include "chp.h"
23 #include "ioasm.h"
24 #include "css.h"
25 #include "vfio_ccw_private.h"
26 
27 struct workqueue_struct *vfio_ccw_work_q;
28 static struct kmem_cache *vfio_ccw_io_region;
29 static struct kmem_cache *vfio_ccw_cmd_region;
30 static struct kmem_cache *vfio_ccw_schib_region;
31 static struct kmem_cache *vfio_ccw_crw_region;
32 
33 debug_info_t *vfio_ccw_debug_msg_id;
34 debug_info_t *vfio_ccw_debug_trace_id;
35 
36 /*
37  * Helpers
38  */
39 int vfio_ccw_sch_quiesce(struct subchannel *sch)
40 {
41 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
42 	DECLARE_COMPLETION_ONSTACK(completion);
43 	int iretry, ret = 0;
44 
45 	spin_lock_irq(sch->lock);
46 	if (!sch->schib.pmcw.ena)
47 		goto out_unlock;
48 	ret = cio_disable_subchannel(sch);
49 	if (ret != -EBUSY)
50 		goto out_unlock;
51 
52 	iretry = 255;
53 	do {
54 
55 		ret = cio_cancel_halt_clear(sch, &iretry);
56 
57 		if (ret == -EIO) {
58 			pr_err("vfio_ccw: could not quiesce subchannel 0.%x.%04x!\n",
59 			       sch->schid.ssid, sch->schid.sch_no);
60 			break;
61 		}
62 
63 		/*
64 		 * Flush all I/O and wait for
65 		 * cancel/halt/clear completion.
66 		 */
67 		private->completion = &completion;
68 		spin_unlock_irq(sch->lock);
69 
70 		if (ret == -EBUSY)
71 			wait_for_completion_timeout(&completion, 3*HZ);
72 
73 		private->completion = NULL;
74 		flush_workqueue(vfio_ccw_work_q);
75 		spin_lock_irq(sch->lock);
76 		ret = cio_disable_subchannel(sch);
77 	} while (ret == -EBUSY);
78 out_unlock:
79 	private->state = VFIO_CCW_STATE_NOT_OPER;
80 	spin_unlock_irq(sch->lock);
81 	return ret;
82 }
83 
84 static void vfio_ccw_sch_io_todo(struct work_struct *work)
85 {
86 	struct vfio_ccw_private *private;
87 	struct irb *irb;
88 	bool is_final;
89 	bool cp_is_finished = false;
90 
91 	private = container_of(work, struct vfio_ccw_private, io_work);
92 	irb = &private->irb;
93 
94 	is_final = !(scsw_actl(&irb->scsw) &
95 		     (SCSW_ACTL_DEVACT | SCSW_ACTL_SCHACT));
96 	if (scsw_is_solicited(&irb->scsw)) {
97 		cp_update_scsw(&private->cp, &irb->scsw);
98 		if (is_final && private->state == VFIO_CCW_STATE_CP_PENDING) {
99 			cp_free(&private->cp);
100 			cp_is_finished = true;
101 		}
102 	}
103 	mutex_lock(&private->io_mutex);
104 	memcpy(private->io_region->irb_area, irb, sizeof(*irb));
105 	mutex_unlock(&private->io_mutex);
106 
107 	/*
108 	 * Reset to IDLE only if processing of a channel program
109 	 * has finished. Do not overwrite a possible processing
110 	 * state if the final interrupt was for HSCH or CSCH.
111 	 */
112 	if (private->mdev && cp_is_finished)
113 		private->state = VFIO_CCW_STATE_IDLE;
114 
115 	if (private->io_trigger)
116 		eventfd_signal(private->io_trigger, 1);
117 }
118 
119 static void vfio_ccw_crw_todo(struct work_struct *work)
120 {
121 	struct vfio_ccw_private *private;
122 
123 	private = container_of(work, struct vfio_ccw_private, crw_work);
124 
125 	if (!list_empty(&private->crw) && private->crw_trigger)
126 		eventfd_signal(private->crw_trigger, 1);
127 }
128 
129 /*
130  * Css driver callbacks
131  */
132 static void vfio_ccw_sch_irq(struct subchannel *sch)
133 {
134 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
135 
136 	inc_irq_stat(IRQIO_CIO);
137 	vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_INTERRUPT);
138 }
139 
140 static struct vfio_ccw_private *vfio_ccw_alloc_private(struct subchannel *sch)
141 {
142 	struct vfio_ccw_private *private;
143 
144 	private = kzalloc(sizeof(*private), GFP_KERNEL);
145 	if (!private)
146 		return ERR_PTR(-ENOMEM);
147 
148 	private->sch = sch;
149 	mutex_init(&private->io_mutex);
150 	private->state = VFIO_CCW_STATE_NOT_OPER;
151 	INIT_LIST_HEAD(&private->crw);
152 	INIT_WORK(&private->io_work, vfio_ccw_sch_io_todo);
153 	INIT_WORK(&private->crw_work, vfio_ccw_crw_todo);
154 	atomic_set(&private->avail, 1);
155 
156 	private->cp.guest_cp = kcalloc(CCWCHAIN_LEN_MAX, sizeof(struct ccw1),
157 				       GFP_KERNEL);
158 	if (!private->cp.guest_cp)
159 		goto out_free_private;
160 
161 	private->io_region = kmem_cache_zalloc(vfio_ccw_io_region,
162 					       GFP_KERNEL | GFP_DMA);
163 	if (!private->io_region)
164 		goto out_free_cp;
165 
166 	private->cmd_region = kmem_cache_zalloc(vfio_ccw_cmd_region,
167 						GFP_KERNEL | GFP_DMA);
168 	if (!private->cmd_region)
169 		goto out_free_io;
170 
171 	private->schib_region = kmem_cache_zalloc(vfio_ccw_schib_region,
172 						  GFP_KERNEL | GFP_DMA);
173 
174 	if (!private->schib_region)
175 		goto out_free_cmd;
176 
177 	private->crw_region = kmem_cache_zalloc(vfio_ccw_crw_region,
178 						GFP_KERNEL | GFP_DMA);
179 
180 	if (!private->crw_region)
181 		goto out_free_schib;
182 	return private;
183 
184 out_free_schib:
185 	kmem_cache_free(vfio_ccw_schib_region, private->schib_region);
186 out_free_cmd:
187 	kmem_cache_free(vfio_ccw_cmd_region, private->cmd_region);
188 out_free_io:
189 	kmem_cache_free(vfio_ccw_io_region, private->io_region);
190 out_free_cp:
191 	kfree(private->cp.guest_cp);
192 out_free_private:
193 	mutex_destroy(&private->io_mutex);
194 	kfree(private);
195 	return ERR_PTR(-ENOMEM);
196 }
197 
198 static void vfio_ccw_free_private(struct vfio_ccw_private *private)
199 {
200 	struct vfio_ccw_crw *crw, *temp;
201 
202 	list_for_each_entry_safe(crw, temp, &private->crw, next) {
203 		list_del(&crw->next);
204 		kfree(crw);
205 	}
206 
207 	kmem_cache_free(vfio_ccw_crw_region, private->crw_region);
208 	kmem_cache_free(vfio_ccw_schib_region, private->schib_region);
209 	kmem_cache_free(vfio_ccw_cmd_region, private->cmd_region);
210 	kmem_cache_free(vfio_ccw_io_region, private->io_region);
211 	kfree(private->cp.guest_cp);
212 	mutex_destroy(&private->io_mutex);
213 	kfree(private);
214 }
215 
216 static int vfio_ccw_sch_probe(struct subchannel *sch)
217 {
218 	struct pmcw *pmcw = &sch->schib.pmcw;
219 	struct vfio_ccw_private *private;
220 	int ret = -ENOMEM;
221 
222 	if (pmcw->qf) {
223 		dev_warn(&sch->dev, "vfio: ccw: does not support QDIO: %s\n",
224 			 dev_name(&sch->dev));
225 		return -ENODEV;
226 	}
227 
228 	private = vfio_ccw_alloc_private(sch);
229 	if (IS_ERR(private))
230 		return PTR_ERR(private);
231 
232 	dev_set_drvdata(&sch->dev, private);
233 
234 	spin_lock_irq(sch->lock);
235 	sch->isc = VFIO_CCW_ISC;
236 	ret = cio_enable_subchannel(sch, (u32)(unsigned long)sch);
237 	spin_unlock_irq(sch->lock);
238 	if (ret)
239 		goto out_free;
240 
241 	private->state = VFIO_CCW_STATE_STANDBY;
242 
243 	ret = vfio_ccw_mdev_reg(sch);
244 	if (ret)
245 		goto out_disable;
246 
247 	if (dev_get_uevent_suppress(&sch->dev)) {
248 		dev_set_uevent_suppress(&sch->dev, 0);
249 		kobject_uevent(&sch->dev.kobj, KOBJ_ADD);
250 	}
251 
252 	VFIO_CCW_MSG_EVENT(4, "bound to subchannel %x.%x.%04x\n",
253 			   sch->schid.cssid, sch->schid.ssid,
254 			   sch->schid.sch_no);
255 	return 0;
256 
257 out_disable:
258 	cio_disable_subchannel(sch);
259 out_free:
260 	dev_set_drvdata(&sch->dev, NULL);
261 	vfio_ccw_free_private(private);
262 	return ret;
263 }
264 
265 static void vfio_ccw_sch_remove(struct subchannel *sch)
266 {
267 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
268 
269 	vfio_ccw_sch_quiesce(sch);
270 	vfio_ccw_mdev_unreg(sch);
271 
272 	dev_set_drvdata(&sch->dev, NULL);
273 
274 	vfio_ccw_free_private(private);
275 
276 	VFIO_CCW_MSG_EVENT(4, "unbound from subchannel %x.%x.%04x\n",
277 			   sch->schid.cssid, sch->schid.ssid,
278 			   sch->schid.sch_no);
279 }
280 
281 static void vfio_ccw_sch_shutdown(struct subchannel *sch)
282 {
283 	vfio_ccw_sch_quiesce(sch);
284 }
285 
286 /**
287  * vfio_ccw_sch_event - process subchannel event
288  * @sch: subchannel
289  * @process: non-zero if function is called in process context
290  *
291  * An unspecified event occurred for this subchannel. Adjust data according
292  * to the current operational state of the subchannel. Return zero when the
293  * event has been handled sufficiently or -EAGAIN when this function should
294  * be called again in process context.
295  */
296 static int vfio_ccw_sch_event(struct subchannel *sch, int process)
297 {
298 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
299 	unsigned long flags;
300 	int rc = -EAGAIN;
301 
302 	spin_lock_irqsave(sch->lock, flags);
303 	if (!device_is_registered(&sch->dev))
304 		goto out_unlock;
305 
306 	if (work_pending(&sch->todo_work))
307 		goto out_unlock;
308 
309 	if (cio_update_schib(sch)) {
310 		vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_NOT_OPER);
311 		rc = 0;
312 		goto out_unlock;
313 	}
314 
315 	private = dev_get_drvdata(&sch->dev);
316 	if (private->state == VFIO_CCW_STATE_NOT_OPER) {
317 		private->state = private->mdev ? VFIO_CCW_STATE_IDLE :
318 				 VFIO_CCW_STATE_STANDBY;
319 	}
320 	rc = 0;
321 
322 out_unlock:
323 	spin_unlock_irqrestore(sch->lock, flags);
324 
325 	return rc;
326 }
327 
328 static void vfio_ccw_queue_crw(struct vfio_ccw_private *private,
329 			       unsigned int rsc,
330 			       unsigned int erc,
331 			       unsigned int rsid)
332 {
333 	struct vfio_ccw_crw *crw;
334 
335 	/*
336 	 * If unable to allocate a CRW, just drop the event and
337 	 * carry on.  The guest will either see a later one or
338 	 * learn when it issues its own store subchannel.
339 	 */
340 	crw = kzalloc(sizeof(*crw), GFP_ATOMIC);
341 	if (!crw)
342 		return;
343 
344 	/*
345 	 * Build the CRW based on the inputs given to us.
346 	 */
347 	crw->crw.rsc = rsc;
348 	crw->crw.erc = erc;
349 	crw->crw.rsid = rsid;
350 
351 	list_add_tail(&crw->next, &private->crw);
352 	queue_work(vfio_ccw_work_q, &private->crw_work);
353 }
354 
355 static int vfio_ccw_chp_event(struct subchannel *sch,
356 			      struct chp_link *link, int event)
357 {
358 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
359 	int mask = chp_ssd_get_mask(&sch->ssd_info, link);
360 	int retry = 255;
361 
362 	if (!private || !mask)
363 		return 0;
364 
365 	trace_vfio_ccw_chp_event(private->sch->schid, mask, event);
366 	VFIO_CCW_MSG_EVENT(2, "%pUl (%x.%x.%04x): mask=0x%x event=%d\n",
367 			   mdev_uuid(private->mdev), sch->schid.cssid,
368 			   sch->schid.ssid, sch->schid.sch_no,
369 			   mask, event);
370 
371 	if (cio_update_schib(sch))
372 		return -ENODEV;
373 
374 	switch (event) {
375 	case CHP_VARY_OFF:
376 		/* Path logically turned off */
377 		sch->opm &= ~mask;
378 		sch->lpm &= ~mask;
379 		if (sch->schib.pmcw.lpum & mask)
380 			cio_cancel_halt_clear(sch, &retry);
381 		break;
382 	case CHP_OFFLINE:
383 		/* Path is gone */
384 		if (sch->schib.pmcw.lpum & mask)
385 			cio_cancel_halt_clear(sch, &retry);
386 		vfio_ccw_queue_crw(private, CRW_RSC_CPATH, CRW_ERC_PERRN,
387 				   link->chpid.id);
388 		break;
389 	case CHP_VARY_ON:
390 		/* Path logically turned on */
391 		sch->opm |= mask;
392 		sch->lpm |= mask;
393 		break;
394 	case CHP_ONLINE:
395 		/* Path became available */
396 		sch->lpm |= mask & sch->opm;
397 		vfio_ccw_queue_crw(private, CRW_RSC_CPATH, CRW_ERC_INIT,
398 				   link->chpid.id);
399 		break;
400 	}
401 
402 	return 0;
403 }
404 
405 static struct css_device_id vfio_ccw_sch_ids[] = {
406 	{ .match_flags = 0x1, .type = SUBCHANNEL_TYPE_IO, },
407 	{ /* end of list */ },
408 };
409 MODULE_DEVICE_TABLE(css, vfio_ccw_sch_ids);
410 
411 static struct css_driver vfio_ccw_sch_driver = {
412 	.drv = {
413 		.name = "vfio_ccw",
414 		.owner = THIS_MODULE,
415 	},
416 	.subchannel_type = vfio_ccw_sch_ids,
417 	.irq = vfio_ccw_sch_irq,
418 	.probe = vfio_ccw_sch_probe,
419 	.remove = vfio_ccw_sch_remove,
420 	.shutdown = vfio_ccw_sch_shutdown,
421 	.sch_event = vfio_ccw_sch_event,
422 	.chp_event = vfio_ccw_chp_event,
423 };
424 
425 static int __init vfio_ccw_debug_init(void)
426 {
427 	vfio_ccw_debug_msg_id = debug_register("vfio_ccw_msg", 16, 1,
428 					       11 * sizeof(long));
429 	if (!vfio_ccw_debug_msg_id)
430 		goto out_unregister;
431 	debug_register_view(vfio_ccw_debug_msg_id, &debug_sprintf_view);
432 	debug_set_level(vfio_ccw_debug_msg_id, 2);
433 	vfio_ccw_debug_trace_id = debug_register("vfio_ccw_trace", 16, 1, 16);
434 	if (!vfio_ccw_debug_trace_id)
435 		goto out_unregister;
436 	debug_register_view(vfio_ccw_debug_trace_id, &debug_hex_ascii_view);
437 	debug_set_level(vfio_ccw_debug_trace_id, 2);
438 	return 0;
439 
440 out_unregister:
441 	debug_unregister(vfio_ccw_debug_msg_id);
442 	debug_unregister(vfio_ccw_debug_trace_id);
443 	return -1;
444 }
445 
446 static void vfio_ccw_debug_exit(void)
447 {
448 	debug_unregister(vfio_ccw_debug_msg_id);
449 	debug_unregister(vfio_ccw_debug_trace_id);
450 }
451 
452 static void vfio_ccw_destroy_regions(void)
453 {
454 	kmem_cache_destroy(vfio_ccw_crw_region);
455 	kmem_cache_destroy(vfio_ccw_schib_region);
456 	kmem_cache_destroy(vfio_ccw_cmd_region);
457 	kmem_cache_destroy(vfio_ccw_io_region);
458 }
459 
460 static int __init vfio_ccw_sch_init(void)
461 {
462 	int ret;
463 
464 	ret = vfio_ccw_debug_init();
465 	if (ret)
466 		return ret;
467 
468 	vfio_ccw_work_q = create_singlethread_workqueue("vfio-ccw");
469 	if (!vfio_ccw_work_q) {
470 		ret = -ENOMEM;
471 		goto out_regions;
472 	}
473 
474 	vfio_ccw_io_region = kmem_cache_create_usercopy("vfio_ccw_io_region",
475 					sizeof(struct ccw_io_region), 0,
476 					SLAB_ACCOUNT, 0,
477 					sizeof(struct ccw_io_region), NULL);
478 	if (!vfio_ccw_io_region) {
479 		ret = -ENOMEM;
480 		goto out_regions;
481 	}
482 
483 	vfio_ccw_cmd_region = kmem_cache_create_usercopy("vfio_ccw_cmd_region",
484 					sizeof(struct ccw_cmd_region), 0,
485 					SLAB_ACCOUNT, 0,
486 					sizeof(struct ccw_cmd_region), NULL);
487 	if (!vfio_ccw_cmd_region) {
488 		ret = -ENOMEM;
489 		goto out_regions;
490 	}
491 
492 	vfio_ccw_schib_region = kmem_cache_create_usercopy("vfio_ccw_schib_region",
493 					sizeof(struct ccw_schib_region), 0,
494 					SLAB_ACCOUNT, 0,
495 					sizeof(struct ccw_schib_region), NULL);
496 
497 	if (!vfio_ccw_schib_region) {
498 		ret = -ENOMEM;
499 		goto out_regions;
500 	}
501 
502 	vfio_ccw_crw_region = kmem_cache_create_usercopy("vfio_ccw_crw_region",
503 					sizeof(struct ccw_crw_region), 0,
504 					SLAB_ACCOUNT, 0,
505 					sizeof(struct ccw_crw_region), NULL);
506 
507 	if (!vfio_ccw_crw_region) {
508 		ret = -ENOMEM;
509 		goto out_regions;
510 	}
511 
512 	ret = mdev_register_driver(&vfio_ccw_mdev_driver);
513 	if (ret)
514 		goto out_regions;
515 
516 	isc_register(VFIO_CCW_ISC);
517 	ret = css_driver_register(&vfio_ccw_sch_driver);
518 	if (ret) {
519 		isc_unregister(VFIO_CCW_ISC);
520 		goto out_driver;
521 	}
522 
523 	return ret;
524 
525 out_driver:
526 	mdev_unregister_driver(&vfio_ccw_mdev_driver);
527 out_regions:
528 	vfio_ccw_destroy_regions();
529 	destroy_workqueue(vfio_ccw_work_q);
530 	vfio_ccw_debug_exit();
531 	return ret;
532 }
533 
534 static void __exit vfio_ccw_sch_exit(void)
535 {
536 	css_driver_unregister(&vfio_ccw_sch_driver);
537 	mdev_unregister_driver(&vfio_ccw_mdev_driver);
538 	isc_unregister(VFIO_CCW_ISC);
539 	vfio_ccw_destroy_regions();
540 	destroy_workqueue(vfio_ccw_work_q);
541 	vfio_ccw_debug_exit();
542 }
543 module_init(vfio_ccw_sch_init);
544 module_exit(vfio_ccw_sch_exit);
545 
546 MODULE_LICENSE("GPL v2");
547