xref: /linux/drivers/s390/cio/vfio_ccw_drv.c (revision 164666fa66669d437bdcc8d5f1744a2aee73be41)
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 	VFIO_CCW_MSG_EVENT(4, "bound to subchannel %x.%x.%04x\n",
248 			   sch->schid.cssid, sch->schid.ssid,
249 			   sch->schid.sch_no);
250 	return 0;
251 
252 out_disable:
253 	cio_disable_subchannel(sch);
254 out_free:
255 	dev_set_drvdata(&sch->dev, NULL);
256 	vfio_ccw_free_private(private);
257 	return ret;
258 }
259 
260 static void vfio_ccw_sch_remove(struct subchannel *sch)
261 {
262 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
263 
264 	vfio_ccw_sch_quiesce(sch);
265 	vfio_ccw_mdev_unreg(sch);
266 
267 	dev_set_drvdata(&sch->dev, NULL);
268 
269 	vfio_ccw_free_private(private);
270 
271 	VFIO_CCW_MSG_EVENT(4, "unbound from subchannel %x.%x.%04x\n",
272 			   sch->schid.cssid, sch->schid.ssid,
273 			   sch->schid.sch_no);
274 }
275 
276 static void vfio_ccw_sch_shutdown(struct subchannel *sch)
277 {
278 	vfio_ccw_sch_quiesce(sch);
279 }
280 
281 /**
282  * vfio_ccw_sch_event - process subchannel event
283  * @sch: subchannel
284  * @process: non-zero if function is called in process context
285  *
286  * An unspecified event occurred for this subchannel. Adjust data according
287  * to the current operational state of the subchannel. Return zero when the
288  * event has been handled sufficiently or -EAGAIN when this function should
289  * be called again in process context.
290  */
291 static int vfio_ccw_sch_event(struct subchannel *sch, int process)
292 {
293 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
294 	unsigned long flags;
295 	int rc = -EAGAIN;
296 
297 	spin_lock_irqsave(sch->lock, flags);
298 	if (!device_is_registered(&sch->dev))
299 		goto out_unlock;
300 
301 	if (work_pending(&sch->todo_work))
302 		goto out_unlock;
303 
304 	if (cio_update_schib(sch)) {
305 		vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_NOT_OPER);
306 		rc = 0;
307 		goto out_unlock;
308 	}
309 
310 	private = dev_get_drvdata(&sch->dev);
311 	if (private->state == VFIO_CCW_STATE_NOT_OPER) {
312 		private->state = private->mdev ? VFIO_CCW_STATE_IDLE :
313 				 VFIO_CCW_STATE_STANDBY;
314 	}
315 	rc = 0;
316 
317 out_unlock:
318 	spin_unlock_irqrestore(sch->lock, flags);
319 
320 	return rc;
321 }
322 
323 static void vfio_ccw_queue_crw(struct vfio_ccw_private *private,
324 			       unsigned int rsc,
325 			       unsigned int erc,
326 			       unsigned int rsid)
327 {
328 	struct vfio_ccw_crw *crw;
329 
330 	/*
331 	 * If unable to allocate a CRW, just drop the event and
332 	 * carry on.  The guest will either see a later one or
333 	 * learn when it issues its own store subchannel.
334 	 */
335 	crw = kzalloc(sizeof(*crw), GFP_ATOMIC);
336 	if (!crw)
337 		return;
338 
339 	/*
340 	 * Build the CRW based on the inputs given to us.
341 	 */
342 	crw->crw.rsc = rsc;
343 	crw->crw.erc = erc;
344 	crw->crw.rsid = rsid;
345 
346 	list_add_tail(&crw->next, &private->crw);
347 	queue_work(vfio_ccw_work_q, &private->crw_work);
348 }
349 
350 static int vfio_ccw_chp_event(struct subchannel *sch,
351 			      struct chp_link *link, int event)
352 {
353 	struct vfio_ccw_private *private = dev_get_drvdata(&sch->dev);
354 	int mask = chp_ssd_get_mask(&sch->ssd_info, link);
355 	int retry = 255;
356 
357 	if (!private || !mask)
358 		return 0;
359 
360 	trace_vfio_ccw_chp_event(private->sch->schid, mask, event);
361 	VFIO_CCW_MSG_EVENT(2, "%pUl (%x.%x.%04x): mask=0x%x event=%d\n",
362 			   mdev_uuid(private->mdev), sch->schid.cssid,
363 			   sch->schid.ssid, sch->schid.sch_no,
364 			   mask, event);
365 
366 	if (cio_update_schib(sch))
367 		return -ENODEV;
368 
369 	switch (event) {
370 	case CHP_VARY_OFF:
371 		/* Path logically turned off */
372 		sch->opm &= ~mask;
373 		sch->lpm &= ~mask;
374 		if (sch->schib.pmcw.lpum & mask)
375 			cio_cancel_halt_clear(sch, &retry);
376 		break;
377 	case CHP_OFFLINE:
378 		/* Path is gone */
379 		if (sch->schib.pmcw.lpum & mask)
380 			cio_cancel_halt_clear(sch, &retry);
381 		vfio_ccw_queue_crw(private, CRW_RSC_CPATH, CRW_ERC_PERRN,
382 				   link->chpid.id);
383 		break;
384 	case CHP_VARY_ON:
385 		/* Path logically turned on */
386 		sch->opm |= mask;
387 		sch->lpm |= mask;
388 		break;
389 	case CHP_ONLINE:
390 		/* Path became available */
391 		sch->lpm |= mask & sch->opm;
392 		vfio_ccw_queue_crw(private, CRW_RSC_CPATH, CRW_ERC_INIT,
393 				   link->chpid.id);
394 		break;
395 	}
396 
397 	return 0;
398 }
399 
400 static struct css_device_id vfio_ccw_sch_ids[] = {
401 	{ .match_flags = 0x1, .type = SUBCHANNEL_TYPE_IO, },
402 	{ /* end of list */ },
403 };
404 MODULE_DEVICE_TABLE(css, vfio_ccw_sch_ids);
405 
406 static struct css_driver vfio_ccw_sch_driver = {
407 	.drv = {
408 		.name = "vfio_ccw",
409 		.owner = THIS_MODULE,
410 	},
411 	.subchannel_type = vfio_ccw_sch_ids,
412 	.irq = vfio_ccw_sch_irq,
413 	.probe = vfio_ccw_sch_probe,
414 	.remove = vfio_ccw_sch_remove,
415 	.shutdown = vfio_ccw_sch_shutdown,
416 	.sch_event = vfio_ccw_sch_event,
417 	.chp_event = vfio_ccw_chp_event,
418 };
419 
420 static int __init vfio_ccw_debug_init(void)
421 {
422 	vfio_ccw_debug_msg_id = debug_register("vfio_ccw_msg", 16, 1,
423 					       11 * sizeof(long));
424 	if (!vfio_ccw_debug_msg_id)
425 		goto out_unregister;
426 	debug_register_view(vfio_ccw_debug_msg_id, &debug_sprintf_view);
427 	debug_set_level(vfio_ccw_debug_msg_id, 2);
428 	vfio_ccw_debug_trace_id = debug_register("vfio_ccw_trace", 16, 1, 16);
429 	if (!vfio_ccw_debug_trace_id)
430 		goto out_unregister;
431 	debug_register_view(vfio_ccw_debug_trace_id, &debug_hex_ascii_view);
432 	debug_set_level(vfio_ccw_debug_trace_id, 2);
433 	return 0;
434 
435 out_unregister:
436 	debug_unregister(vfio_ccw_debug_msg_id);
437 	debug_unregister(vfio_ccw_debug_trace_id);
438 	return -1;
439 }
440 
441 static void vfio_ccw_debug_exit(void)
442 {
443 	debug_unregister(vfio_ccw_debug_msg_id);
444 	debug_unregister(vfio_ccw_debug_trace_id);
445 }
446 
447 static void vfio_ccw_destroy_regions(void)
448 {
449 	kmem_cache_destroy(vfio_ccw_crw_region);
450 	kmem_cache_destroy(vfio_ccw_schib_region);
451 	kmem_cache_destroy(vfio_ccw_cmd_region);
452 	kmem_cache_destroy(vfio_ccw_io_region);
453 }
454 
455 static int __init vfio_ccw_sch_init(void)
456 {
457 	int ret;
458 
459 	ret = vfio_ccw_debug_init();
460 	if (ret)
461 		return ret;
462 
463 	vfio_ccw_work_q = create_singlethread_workqueue("vfio-ccw");
464 	if (!vfio_ccw_work_q) {
465 		ret = -ENOMEM;
466 		goto out_regions;
467 	}
468 
469 	vfio_ccw_io_region = kmem_cache_create_usercopy("vfio_ccw_io_region",
470 					sizeof(struct ccw_io_region), 0,
471 					SLAB_ACCOUNT, 0,
472 					sizeof(struct ccw_io_region), NULL);
473 	if (!vfio_ccw_io_region) {
474 		ret = -ENOMEM;
475 		goto out_regions;
476 	}
477 
478 	vfio_ccw_cmd_region = kmem_cache_create_usercopy("vfio_ccw_cmd_region",
479 					sizeof(struct ccw_cmd_region), 0,
480 					SLAB_ACCOUNT, 0,
481 					sizeof(struct ccw_cmd_region), NULL);
482 	if (!vfio_ccw_cmd_region) {
483 		ret = -ENOMEM;
484 		goto out_regions;
485 	}
486 
487 	vfio_ccw_schib_region = kmem_cache_create_usercopy("vfio_ccw_schib_region",
488 					sizeof(struct ccw_schib_region), 0,
489 					SLAB_ACCOUNT, 0,
490 					sizeof(struct ccw_schib_region), NULL);
491 
492 	if (!vfio_ccw_schib_region) {
493 		ret = -ENOMEM;
494 		goto out_regions;
495 	}
496 
497 	vfio_ccw_crw_region = kmem_cache_create_usercopy("vfio_ccw_crw_region",
498 					sizeof(struct ccw_crw_region), 0,
499 					SLAB_ACCOUNT, 0,
500 					sizeof(struct ccw_crw_region), NULL);
501 
502 	if (!vfio_ccw_crw_region) {
503 		ret = -ENOMEM;
504 		goto out_regions;
505 	}
506 
507 	ret = mdev_register_driver(&vfio_ccw_mdev_driver);
508 	if (ret)
509 		goto out_regions;
510 
511 	isc_register(VFIO_CCW_ISC);
512 	ret = css_driver_register(&vfio_ccw_sch_driver);
513 	if (ret) {
514 		isc_unregister(VFIO_CCW_ISC);
515 		goto out_driver;
516 	}
517 
518 	return ret;
519 
520 out_driver:
521 	mdev_unregister_driver(&vfio_ccw_mdev_driver);
522 out_regions:
523 	vfio_ccw_destroy_regions();
524 	destroy_workqueue(vfio_ccw_work_q);
525 	vfio_ccw_debug_exit();
526 	return ret;
527 }
528 
529 static void __exit vfio_ccw_sch_exit(void)
530 {
531 	css_driver_unregister(&vfio_ccw_sch_driver);
532 	mdev_unregister_driver(&vfio_ccw_mdev_driver);
533 	isc_unregister(VFIO_CCW_ISC);
534 	vfio_ccw_destroy_regions();
535 	destroy_workqueue(vfio_ccw_work_q);
536 	vfio_ccw_debug_exit();
537 }
538 module_init(vfio_ccw_sch_init);
539 module_exit(vfio_ccw_sch_exit);
540 
541 MODULE_LICENSE("GPL v2");
542