xref: /openbmc/linux/drivers/dma/idxd/device.c (revision f7af616c)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2019 Intel Corporation. All rights rsvd. */
3 #include <linux/init.h>
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/pci.h>
7 #include <linux/io-64-nonatomic-lo-hi.h>
8 #include <linux/dmaengine.h>
9 #include <linux/irq.h>
10 #include <linux/msi.h>
11 #include <uapi/linux/idxd.h>
12 #include "../dmaengine.h"
13 #include "idxd.h"
14 #include "registers.h"
15 
16 static void idxd_cmd_exec(struct idxd_device *idxd, int cmd_code, u32 operand,
17 			  u32 *status);
18 
19 /* Interrupt control bits */
20 void idxd_mask_msix_vector(struct idxd_device *idxd, int vec_id)
21 {
22 	struct irq_data *data = irq_get_irq_data(idxd->irq_entries[vec_id].vector);
23 
24 	pci_msi_mask_irq(data);
25 }
26 
27 void idxd_mask_msix_vectors(struct idxd_device *idxd)
28 {
29 	struct pci_dev *pdev = idxd->pdev;
30 	int msixcnt = pci_msix_vec_count(pdev);
31 	int i;
32 
33 	for (i = 0; i < msixcnt; i++)
34 		idxd_mask_msix_vector(idxd, i);
35 }
36 
37 void idxd_unmask_msix_vector(struct idxd_device *idxd, int vec_id)
38 {
39 	struct irq_data *data = irq_get_irq_data(idxd->irq_entries[vec_id].vector);
40 
41 	pci_msi_unmask_irq(data);
42 }
43 
44 void idxd_unmask_error_interrupts(struct idxd_device *idxd)
45 {
46 	union genctrl_reg genctrl;
47 
48 	genctrl.bits = ioread32(idxd->reg_base + IDXD_GENCTRL_OFFSET);
49 	genctrl.softerr_int_en = 1;
50 	genctrl.halt_int_en = 1;
51 	iowrite32(genctrl.bits, idxd->reg_base + IDXD_GENCTRL_OFFSET);
52 }
53 
54 void idxd_mask_error_interrupts(struct idxd_device *idxd)
55 {
56 	union genctrl_reg genctrl;
57 
58 	genctrl.bits = ioread32(idxd->reg_base + IDXD_GENCTRL_OFFSET);
59 	genctrl.softerr_int_en = 0;
60 	genctrl.halt_int_en = 0;
61 	iowrite32(genctrl.bits, idxd->reg_base + IDXD_GENCTRL_OFFSET);
62 }
63 
64 static void free_hw_descs(struct idxd_wq *wq)
65 {
66 	int i;
67 
68 	for (i = 0; i < wq->num_descs; i++)
69 		kfree(wq->hw_descs[i]);
70 
71 	kfree(wq->hw_descs);
72 }
73 
74 static int alloc_hw_descs(struct idxd_wq *wq, int num)
75 {
76 	struct device *dev = &wq->idxd->pdev->dev;
77 	int i;
78 	int node = dev_to_node(dev);
79 
80 	wq->hw_descs = kcalloc_node(num, sizeof(struct dsa_hw_desc *),
81 				    GFP_KERNEL, node);
82 	if (!wq->hw_descs)
83 		return -ENOMEM;
84 
85 	for (i = 0; i < num; i++) {
86 		wq->hw_descs[i] = kzalloc_node(sizeof(*wq->hw_descs[i]),
87 					       GFP_KERNEL, node);
88 		if (!wq->hw_descs[i]) {
89 			free_hw_descs(wq);
90 			return -ENOMEM;
91 		}
92 	}
93 
94 	return 0;
95 }
96 
97 static void free_descs(struct idxd_wq *wq)
98 {
99 	int i;
100 
101 	for (i = 0; i < wq->num_descs; i++)
102 		kfree(wq->descs[i]);
103 
104 	kfree(wq->descs);
105 }
106 
107 static int alloc_descs(struct idxd_wq *wq, int num)
108 {
109 	struct device *dev = &wq->idxd->pdev->dev;
110 	int i;
111 	int node = dev_to_node(dev);
112 
113 	wq->descs = kcalloc_node(num, sizeof(struct idxd_desc *),
114 				 GFP_KERNEL, node);
115 	if (!wq->descs)
116 		return -ENOMEM;
117 
118 	for (i = 0; i < num; i++) {
119 		wq->descs[i] = kzalloc_node(sizeof(*wq->descs[i]),
120 					    GFP_KERNEL, node);
121 		if (!wq->descs[i]) {
122 			free_descs(wq);
123 			return -ENOMEM;
124 		}
125 	}
126 
127 	return 0;
128 }
129 
130 /* WQ control bits */
131 int idxd_wq_alloc_resources(struct idxd_wq *wq)
132 {
133 	struct idxd_device *idxd = wq->idxd;
134 	struct device *dev = &idxd->pdev->dev;
135 	int rc, num_descs, i;
136 	int align;
137 	u64 tmp;
138 
139 	if (wq->type != IDXD_WQT_KERNEL)
140 		return 0;
141 
142 	wq->num_descs = wq->size;
143 	num_descs = wq->size;
144 
145 	rc = alloc_hw_descs(wq, num_descs);
146 	if (rc < 0)
147 		return rc;
148 
149 	align = idxd->data->align;
150 	wq->compls_size = num_descs * idxd->data->compl_size + align;
151 	wq->compls_raw = dma_alloc_coherent(dev, wq->compls_size,
152 					    &wq->compls_addr_raw, GFP_KERNEL);
153 	if (!wq->compls_raw) {
154 		rc = -ENOMEM;
155 		goto fail_alloc_compls;
156 	}
157 
158 	/* Adjust alignment */
159 	wq->compls_addr = (wq->compls_addr_raw + (align - 1)) & ~(align - 1);
160 	tmp = (u64)wq->compls_raw;
161 	tmp = (tmp + (align - 1)) & ~(align - 1);
162 	wq->compls = (struct dsa_completion_record *)tmp;
163 
164 	rc = alloc_descs(wq, num_descs);
165 	if (rc < 0)
166 		goto fail_alloc_descs;
167 
168 	rc = sbitmap_queue_init_node(&wq->sbq, num_descs, -1, false, GFP_KERNEL,
169 				     dev_to_node(dev));
170 	if (rc < 0)
171 		goto fail_sbitmap_init;
172 
173 	for (i = 0; i < num_descs; i++) {
174 		struct idxd_desc *desc = wq->descs[i];
175 
176 		desc->hw = wq->hw_descs[i];
177 		if (idxd->data->type == IDXD_TYPE_DSA)
178 			desc->completion = &wq->compls[i];
179 		else if (idxd->data->type == IDXD_TYPE_IAX)
180 			desc->iax_completion = &wq->iax_compls[i];
181 		desc->compl_dma = wq->compls_addr + idxd->data->compl_size * i;
182 		desc->id = i;
183 		desc->wq = wq;
184 		desc->cpu = -1;
185 	}
186 
187 	return 0;
188 
189  fail_sbitmap_init:
190 	free_descs(wq);
191  fail_alloc_descs:
192 	dma_free_coherent(dev, wq->compls_size, wq->compls_raw,
193 			  wq->compls_addr_raw);
194  fail_alloc_compls:
195 	free_hw_descs(wq);
196 	return rc;
197 }
198 
199 void idxd_wq_free_resources(struct idxd_wq *wq)
200 {
201 	struct device *dev = &wq->idxd->pdev->dev;
202 
203 	if (wq->type != IDXD_WQT_KERNEL)
204 		return;
205 
206 	free_hw_descs(wq);
207 	free_descs(wq);
208 	dma_free_coherent(dev, wq->compls_size, wq->compls_raw,
209 			  wq->compls_addr_raw);
210 	sbitmap_queue_free(&wq->sbq);
211 }
212 
213 int idxd_wq_enable(struct idxd_wq *wq)
214 {
215 	struct idxd_device *idxd = wq->idxd;
216 	struct device *dev = &idxd->pdev->dev;
217 	u32 status;
218 
219 	if (wq->state == IDXD_WQ_ENABLED) {
220 		dev_dbg(dev, "WQ %d already enabled\n", wq->id);
221 		return -ENXIO;
222 	}
223 
224 	idxd_cmd_exec(idxd, IDXD_CMD_ENABLE_WQ, wq->id, &status);
225 
226 	if (status != IDXD_CMDSTS_SUCCESS &&
227 	    status != IDXD_CMDSTS_ERR_WQ_ENABLED) {
228 		dev_dbg(dev, "WQ enable failed: %#x\n", status);
229 		return -ENXIO;
230 	}
231 
232 	wq->state = IDXD_WQ_ENABLED;
233 	dev_dbg(dev, "WQ %d enabled\n", wq->id);
234 	return 0;
235 }
236 
237 int idxd_wq_disable(struct idxd_wq *wq)
238 {
239 	struct idxd_device *idxd = wq->idxd;
240 	struct device *dev = &idxd->pdev->dev;
241 	u32 status, operand;
242 
243 	dev_dbg(dev, "Disabling WQ %d\n", wq->id);
244 
245 	if (wq->state != IDXD_WQ_ENABLED) {
246 		dev_dbg(dev, "WQ %d in wrong state: %d\n", wq->id, wq->state);
247 		return 0;
248 	}
249 
250 	operand = BIT(wq->id % 16) | ((wq->id / 16) << 16);
251 	idxd_cmd_exec(idxd, IDXD_CMD_DISABLE_WQ, operand, &status);
252 
253 	if (status != IDXD_CMDSTS_SUCCESS) {
254 		dev_dbg(dev, "WQ disable failed: %#x\n", status);
255 		return -ENXIO;
256 	}
257 
258 	wq->state = IDXD_WQ_DISABLED;
259 	dev_dbg(dev, "WQ %d disabled\n", wq->id);
260 	return 0;
261 }
262 
263 void idxd_wq_drain(struct idxd_wq *wq)
264 {
265 	struct idxd_device *idxd = wq->idxd;
266 	struct device *dev = &idxd->pdev->dev;
267 	u32 operand;
268 
269 	if (wq->state != IDXD_WQ_ENABLED) {
270 		dev_dbg(dev, "WQ %d in wrong state: %d\n", wq->id, wq->state);
271 		return;
272 	}
273 
274 	dev_dbg(dev, "Draining WQ %d\n", wq->id);
275 	operand = BIT(wq->id % 16) | ((wq->id / 16) << 16);
276 	idxd_cmd_exec(idxd, IDXD_CMD_DRAIN_WQ, operand, NULL);
277 }
278 
279 void idxd_wq_reset(struct idxd_wq *wq)
280 {
281 	struct idxd_device *idxd = wq->idxd;
282 	struct device *dev = &idxd->pdev->dev;
283 	u32 operand;
284 
285 	if (wq->state != IDXD_WQ_ENABLED) {
286 		dev_dbg(dev, "WQ %d in wrong state: %d\n", wq->id, wq->state);
287 		return;
288 	}
289 
290 	operand = BIT(wq->id % 16) | ((wq->id / 16) << 16);
291 	idxd_cmd_exec(idxd, IDXD_CMD_RESET_WQ, operand, NULL);
292 	wq->state = IDXD_WQ_DISABLED;
293 }
294 
295 int idxd_wq_map_portal(struct idxd_wq *wq)
296 {
297 	struct idxd_device *idxd = wq->idxd;
298 	struct pci_dev *pdev = idxd->pdev;
299 	struct device *dev = &pdev->dev;
300 	resource_size_t start;
301 
302 	start = pci_resource_start(pdev, IDXD_WQ_BAR);
303 	start += idxd_get_wq_portal_full_offset(wq->id, IDXD_PORTAL_LIMITED);
304 
305 	wq->portal = devm_ioremap(dev, start, IDXD_PORTAL_SIZE);
306 	if (!wq->portal)
307 		return -ENOMEM;
308 
309 	return 0;
310 }
311 
312 void idxd_wq_unmap_portal(struct idxd_wq *wq)
313 {
314 	struct device *dev = &wq->idxd->pdev->dev;
315 
316 	devm_iounmap(dev, wq->portal);
317 	wq->portal = NULL;
318 }
319 
320 void idxd_wqs_unmap_portal(struct idxd_device *idxd)
321 {
322 	int i;
323 
324 	for (i = 0; i < idxd->max_wqs; i++) {
325 		struct idxd_wq *wq = idxd->wqs[i];
326 
327 		if (wq->portal)
328 			idxd_wq_unmap_portal(wq);
329 	}
330 }
331 
332 int idxd_wq_set_pasid(struct idxd_wq *wq, int pasid)
333 {
334 	struct idxd_device *idxd = wq->idxd;
335 	int rc;
336 	union wqcfg wqcfg;
337 	unsigned int offset;
338 	unsigned long flags;
339 
340 	rc = idxd_wq_disable(wq);
341 	if (rc < 0)
342 		return rc;
343 
344 	offset = WQCFG_OFFSET(idxd, wq->id, WQCFG_PASID_IDX);
345 	spin_lock_irqsave(&idxd->dev_lock, flags);
346 	wqcfg.bits[WQCFG_PASID_IDX] = ioread32(idxd->reg_base + offset);
347 	wqcfg.pasid_en = 1;
348 	wqcfg.pasid = pasid;
349 	iowrite32(wqcfg.bits[WQCFG_PASID_IDX], idxd->reg_base + offset);
350 	spin_unlock_irqrestore(&idxd->dev_lock, flags);
351 
352 	rc = idxd_wq_enable(wq);
353 	if (rc < 0)
354 		return rc;
355 
356 	return 0;
357 }
358 
359 int idxd_wq_disable_pasid(struct idxd_wq *wq)
360 {
361 	struct idxd_device *idxd = wq->idxd;
362 	int rc;
363 	union wqcfg wqcfg;
364 	unsigned int offset;
365 	unsigned long flags;
366 
367 	rc = idxd_wq_disable(wq);
368 	if (rc < 0)
369 		return rc;
370 
371 	offset = WQCFG_OFFSET(idxd, wq->id, WQCFG_PASID_IDX);
372 	spin_lock_irqsave(&idxd->dev_lock, flags);
373 	wqcfg.bits[WQCFG_PASID_IDX] = ioread32(idxd->reg_base + offset);
374 	wqcfg.pasid_en = 0;
375 	wqcfg.pasid = 0;
376 	iowrite32(wqcfg.bits[WQCFG_PASID_IDX], idxd->reg_base + offset);
377 	spin_unlock_irqrestore(&idxd->dev_lock, flags);
378 
379 	rc = idxd_wq_enable(wq);
380 	if (rc < 0)
381 		return rc;
382 
383 	return 0;
384 }
385 
386 void idxd_wq_disable_cleanup(struct idxd_wq *wq)
387 {
388 	struct idxd_device *idxd = wq->idxd;
389 
390 	lockdep_assert_held(&idxd->dev_lock);
391 	memset(wq->wqcfg, 0, idxd->wqcfg_size);
392 	wq->type = IDXD_WQT_NONE;
393 	wq->size = 0;
394 	wq->group = NULL;
395 	wq->threshold = 0;
396 	wq->priority = 0;
397 	wq->ats_dis = 0;
398 	clear_bit(WQ_FLAG_DEDICATED, &wq->flags);
399 	memset(wq->name, 0, WQ_NAME_SIZE);
400 }
401 
402 static void idxd_wq_ref_release(struct percpu_ref *ref)
403 {
404 	struct idxd_wq *wq = container_of(ref, struct idxd_wq, wq_active);
405 
406 	complete(&wq->wq_dead);
407 }
408 
409 int idxd_wq_init_percpu_ref(struct idxd_wq *wq)
410 {
411 	int rc;
412 
413 	memset(&wq->wq_active, 0, sizeof(wq->wq_active));
414 	rc = percpu_ref_init(&wq->wq_active, idxd_wq_ref_release, 0, GFP_KERNEL);
415 	if (rc < 0)
416 		return rc;
417 	reinit_completion(&wq->wq_dead);
418 	return 0;
419 }
420 
421 void idxd_wq_quiesce(struct idxd_wq *wq)
422 {
423 	percpu_ref_kill(&wq->wq_active);
424 	wait_for_completion(&wq->wq_dead);
425 	percpu_ref_exit(&wq->wq_active);
426 }
427 
428 /* Device control bits */
429 static inline bool idxd_is_enabled(struct idxd_device *idxd)
430 {
431 	union gensts_reg gensts;
432 
433 	gensts.bits = ioread32(idxd->reg_base + IDXD_GENSTATS_OFFSET);
434 
435 	if (gensts.state == IDXD_DEVICE_STATE_ENABLED)
436 		return true;
437 	return false;
438 }
439 
440 static inline bool idxd_device_is_halted(struct idxd_device *idxd)
441 {
442 	union gensts_reg gensts;
443 
444 	gensts.bits = ioread32(idxd->reg_base + IDXD_GENSTATS_OFFSET);
445 
446 	return (gensts.state == IDXD_DEVICE_STATE_HALT);
447 }
448 
449 /*
450  * This is function is only used for reset during probe and will
451  * poll for completion. Once the device is setup with interrupts,
452  * all commands will be done via interrupt completion.
453  */
454 int idxd_device_init_reset(struct idxd_device *idxd)
455 {
456 	struct device *dev = &idxd->pdev->dev;
457 	union idxd_command_reg cmd;
458 	unsigned long flags;
459 
460 	if (idxd_device_is_halted(idxd)) {
461 		dev_warn(&idxd->pdev->dev, "Device is HALTED!\n");
462 		return -ENXIO;
463 	}
464 
465 	memset(&cmd, 0, sizeof(cmd));
466 	cmd.cmd = IDXD_CMD_RESET_DEVICE;
467 	dev_dbg(dev, "%s: sending reset for init.\n", __func__);
468 	spin_lock_irqsave(&idxd->cmd_lock, flags);
469 	iowrite32(cmd.bits, idxd->reg_base + IDXD_CMD_OFFSET);
470 
471 	while (ioread32(idxd->reg_base + IDXD_CMDSTS_OFFSET) &
472 	       IDXD_CMDSTS_ACTIVE)
473 		cpu_relax();
474 	spin_unlock_irqrestore(&idxd->cmd_lock, flags);
475 	return 0;
476 }
477 
478 static void idxd_cmd_exec(struct idxd_device *idxd, int cmd_code, u32 operand,
479 			  u32 *status)
480 {
481 	union idxd_command_reg cmd;
482 	DECLARE_COMPLETION_ONSTACK(done);
483 	unsigned long flags;
484 
485 	if (idxd_device_is_halted(idxd)) {
486 		dev_warn(&idxd->pdev->dev, "Device is HALTED!\n");
487 		if (status)
488 			*status = IDXD_CMDSTS_HW_ERR;
489 		return;
490 	}
491 
492 	memset(&cmd, 0, sizeof(cmd));
493 	cmd.cmd = cmd_code;
494 	cmd.operand = operand;
495 	cmd.int_req = 1;
496 
497 	spin_lock_irqsave(&idxd->cmd_lock, flags);
498 	wait_event_lock_irq(idxd->cmd_waitq,
499 			    !test_bit(IDXD_FLAG_CMD_RUNNING, &idxd->flags),
500 			    idxd->cmd_lock);
501 
502 	dev_dbg(&idxd->pdev->dev, "%s: sending cmd: %#x op: %#x\n",
503 		__func__, cmd_code, operand);
504 
505 	idxd->cmd_status = 0;
506 	__set_bit(IDXD_FLAG_CMD_RUNNING, &idxd->flags);
507 	idxd->cmd_done = &done;
508 	iowrite32(cmd.bits, idxd->reg_base + IDXD_CMD_OFFSET);
509 
510 	/*
511 	 * After command submitted, release lock and go to sleep until
512 	 * the command completes via interrupt.
513 	 */
514 	spin_unlock_irqrestore(&idxd->cmd_lock, flags);
515 	wait_for_completion(&done);
516 	spin_lock_irqsave(&idxd->cmd_lock, flags);
517 	if (status) {
518 		*status = ioread32(idxd->reg_base + IDXD_CMDSTS_OFFSET);
519 		idxd->cmd_status = *status & GENMASK(7, 0);
520 	}
521 
522 	__clear_bit(IDXD_FLAG_CMD_RUNNING, &idxd->flags);
523 	/* Wake up other pending commands */
524 	wake_up(&idxd->cmd_waitq);
525 	spin_unlock_irqrestore(&idxd->cmd_lock, flags);
526 }
527 
528 int idxd_device_enable(struct idxd_device *idxd)
529 {
530 	struct device *dev = &idxd->pdev->dev;
531 	u32 status;
532 
533 	if (idxd_is_enabled(idxd)) {
534 		dev_dbg(dev, "Device already enabled\n");
535 		return -ENXIO;
536 	}
537 
538 	idxd_cmd_exec(idxd, IDXD_CMD_ENABLE_DEVICE, 0, &status);
539 
540 	/* If the command is successful or if the device was enabled */
541 	if (status != IDXD_CMDSTS_SUCCESS &&
542 	    status != IDXD_CMDSTS_ERR_DEV_ENABLED) {
543 		dev_dbg(dev, "%s: err_code: %#x\n", __func__, status);
544 		return -ENXIO;
545 	}
546 
547 	idxd->state = IDXD_DEV_ENABLED;
548 	return 0;
549 }
550 
551 void idxd_device_wqs_clear_state(struct idxd_device *idxd)
552 {
553 	int i;
554 
555 	lockdep_assert_held(&idxd->dev_lock);
556 
557 	for (i = 0; i < idxd->max_wqs; i++) {
558 		struct idxd_wq *wq = idxd->wqs[i];
559 
560 		if (wq->state == IDXD_WQ_ENABLED) {
561 			idxd_wq_disable_cleanup(wq);
562 			wq->state = IDXD_WQ_DISABLED;
563 		}
564 	}
565 }
566 
567 int idxd_device_disable(struct idxd_device *idxd)
568 {
569 	struct device *dev = &idxd->pdev->dev;
570 	u32 status;
571 	unsigned long flags;
572 
573 	if (!idxd_is_enabled(idxd)) {
574 		dev_dbg(dev, "Device is not enabled\n");
575 		return 0;
576 	}
577 
578 	idxd_cmd_exec(idxd, IDXD_CMD_DISABLE_DEVICE, 0, &status);
579 
580 	/* If the command is successful or if the device was disabled */
581 	if (status != IDXD_CMDSTS_SUCCESS &&
582 	    !(status & IDXD_CMDSTS_ERR_DIS_DEV_EN)) {
583 		dev_dbg(dev, "%s: err_code: %#x\n", __func__, status);
584 		return -ENXIO;
585 	}
586 
587 	spin_lock_irqsave(&idxd->dev_lock, flags);
588 	idxd_device_wqs_clear_state(idxd);
589 	idxd->state = IDXD_DEV_CONF_READY;
590 	spin_unlock_irqrestore(&idxd->dev_lock, flags);
591 	return 0;
592 }
593 
594 void idxd_device_reset(struct idxd_device *idxd)
595 {
596 	unsigned long flags;
597 
598 	idxd_cmd_exec(idxd, IDXD_CMD_RESET_DEVICE, 0, NULL);
599 	spin_lock_irqsave(&idxd->dev_lock, flags);
600 	idxd_device_wqs_clear_state(idxd);
601 	idxd->state = IDXD_DEV_CONF_READY;
602 	spin_unlock_irqrestore(&idxd->dev_lock, flags);
603 }
604 
605 void idxd_device_drain_pasid(struct idxd_device *idxd, int pasid)
606 {
607 	struct device *dev = &idxd->pdev->dev;
608 	u32 operand;
609 
610 	operand = pasid;
611 	dev_dbg(dev, "cmd: %u operand: %#x\n", IDXD_CMD_DRAIN_PASID, operand);
612 	idxd_cmd_exec(idxd, IDXD_CMD_DRAIN_PASID, operand, NULL);
613 	dev_dbg(dev, "pasid %d drained\n", pasid);
614 }
615 
616 int idxd_device_request_int_handle(struct idxd_device *idxd, int idx, int *handle,
617 				   enum idxd_interrupt_type irq_type)
618 {
619 	struct device *dev = &idxd->pdev->dev;
620 	u32 operand, status;
621 
622 	if (!(idxd->hw.cmd_cap & BIT(IDXD_CMD_REQUEST_INT_HANDLE)))
623 		return -EOPNOTSUPP;
624 
625 	dev_dbg(dev, "get int handle, idx %d\n", idx);
626 
627 	operand = idx & GENMASK(15, 0);
628 	if (irq_type == IDXD_IRQ_IMS)
629 		operand |= CMD_INT_HANDLE_IMS;
630 
631 	dev_dbg(dev, "cmd: %u operand: %#x\n", IDXD_CMD_REQUEST_INT_HANDLE, operand);
632 
633 	idxd_cmd_exec(idxd, IDXD_CMD_REQUEST_INT_HANDLE, operand, &status);
634 
635 	if ((status & IDXD_CMDSTS_ERR_MASK) != IDXD_CMDSTS_SUCCESS) {
636 		dev_dbg(dev, "request int handle failed: %#x\n", status);
637 		return -ENXIO;
638 	}
639 
640 	*handle = (status >> IDXD_CMDSTS_RES_SHIFT) & GENMASK(15, 0);
641 
642 	dev_dbg(dev, "int handle acquired: %u\n", *handle);
643 	return 0;
644 }
645 
646 int idxd_device_release_int_handle(struct idxd_device *idxd, int handle,
647 				   enum idxd_interrupt_type irq_type)
648 {
649 	struct device *dev = &idxd->pdev->dev;
650 	u32 operand, status;
651 	union idxd_command_reg cmd;
652 	unsigned long flags;
653 
654 	if (!(idxd->hw.cmd_cap & BIT(IDXD_CMD_RELEASE_INT_HANDLE)))
655 		return -EOPNOTSUPP;
656 
657 	dev_dbg(dev, "release int handle, handle %d\n", handle);
658 
659 	memset(&cmd, 0, sizeof(cmd));
660 	operand = handle & GENMASK(15, 0);
661 
662 	if (irq_type == IDXD_IRQ_IMS)
663 		operand |= CMD_INT_HANDLE_IMS;
664 
665 	cmd.cmd = IDXD_CMD_RELEASE_INT_HANDLE;
666 	cmd.operand = operand;
667 
668 	dev_dbg(dev, "cmd: %u operand: %#x\n", IDXD_CMD_RELEASE_INT_HANDLE, operand);
669 
670 	spin_lock_irqsave(&idxd->cmd_lock, flags);
671 	iowrite32(cmd.bits, idxd->reg_base + IDXD_CMD_OFFSET);
672 
673 	while (ioread32(idxd->reg_base + IDXD_CMDSTS_OFFSET) & IDXD_CMDSTS_ACTIVE)
674 		cpu_relax();
675 	status = ioread32(idxd->reg_base + IDXD_CMDSTS_OFFSET);
676 	spin_unlock_irqrestore(&idxd->cmd_lock, flags);
677 
678 	if ((status & IDXD_CMDSTS_ERR_MASK) != IDXD_CMDSTS_SUCCESS) {
679 		dev_dbg(dev, "release int handle failed: %#x\n", status);
680 		return -ENXIO;
681 	}
682 
683 	dev_dbg(dev, "int handle released.\n");
684 	return 0;
685 }
686 
687 /* Device configuration bits */
688 void idxd_msix_perm_setup(struct idxd_device *idxd)
689 {
690 	union msix_perm mperm;
691 	int i, msixcnt;
692 
693 	msixcnt = pci_msix_vec_count(idxd->pdev);
694 	if (msixcnt < 0)
695 		return;
696 
697 	mperm.bits = 0;
698 	mperm.pasid = idxd->pasid;
699 	mperm.pasid_en = device_pasid_enabled(idxd);
700 	for (i = 1; i < msixcnt; i++)
701 		iowrite32(mperm.bits, idxd->reg_base + idxd->msix_perm_offset + i * 8);
702 }
703 
704 void idxd_msix_perm_clear(struct idxd_device *idxd)
705 {
706 	union msix_perm mperm;
707 	int i, msixcnt;
708 
709 	msixcnt = pci_msix_vec_count(idxd->pdev);
710 	if (msixcnt < 0)
711 		return;
712 
713 	mperm.bits = 0;
714 	for (i = 1; i < msixcnt; i++)
715 		iowrite32(mperm.bits, idxd->reg_base + idxd->msix_perm_offset + i * 8);
716 }
717 
718 static void idxd_group_config_write(struct idxd_group *group)
719 {
720 	struct idxd_device *idxd = group->idxd;
721 	struct device *dev = &idxd->pdev->dev;
722 	int i;
723 	u32 grpcfg_offset;
724 
725 	dev_dbg(dev, "Writing group %d cfg registers\n", group->id);
726 
727 	/* setup GRPWQCFG */
728 	for (i = 0; i < GRPWQCFG_STRIDES; i++) {
729 		grpcfg_offset = GRPWQCFG_OFFSET(idxd, group->id, i);
730 		iowrite64(group->grpcfg.wqs[i], idxd->reg_base + grpcfg_offset);
731 		dev_dbg(dev, "GRPCFG wq[%d:%d: %#x]: %#llx\n",
732 			group->id, i, grpcfg_offset,
733 			ioread64(idxd->reg_base + grpcfg_offset));
734 	}
735 
736 	/* setup GRPENGCFG */
737 	grpcfg_offset = GRPENGCFG_OFFSET(idxd, group->id);
738 	iowrite64(group->grpcfg.engines, idxd->reg_base + grpcfg_offset);
739 	dev_dbg(dev, "GRPCFG engs[%d: %#x]: %#llx\n", group->id,
740 		grpcfg_offset, ioread64(idxd->reg_base + grpcfg_offset));
741 
742 	/* setup GRPFLAGS */
743 	grpcfg_offset = GRPFLGCFG_OFFSET(idxd, group->id);
744 	iowrite32(group->grpcfg.flags.bits, idxd->reg_base + grpcfg_offset);
745 	dev_dbg(dev, "GRPFLAGS flags[%d: %#x]: %#x\n",
746 		group->id, grpcfg_offset,
747 		ioread32(idxd->reg_base + grpcfg_offset));
748 }
749 
750 static int idxd_groups_config_write(struct idxd_device *idxd)
751 
752 {
753 	union gencfg_reg reg;
754 	int i;
755 	struct device *dev = &idxd->pdev->dev;
756 
757 	/* Setup bandwidth token limit */
758 	if (idxd->token_limit) {
759 		reg.bits = ioread32(idxd->reg_base + IDXD_GENCFG_OFFSET);
760 		reg.token_limit = idxd->token_limit;
761 		iowrite32(reg.bits, idxd->reg_base + IDXD_GENCFG_OFFSET);
762 	}
763 
764 	dev_dbg(dev, "GENCFG(%#x): %#x\n", IDXD_GENCFG_OFFSET,
765 		ioread32(idxd->reg_base + IDXD_GENCFG_OFFSET));
766 
767 	for (i = 0; i < idxd->max_groups; i++) {
768 		struct idxd_group *group = idxd->groups[i];
769 
770 		idxd_group_config_write(group);
771 	}
772 
773 	return 0;
774 }
775 
776 static int idxd_wq_config_write(struct idxd_wq *wq)
777 {
778 	struct idxd_device *idxd = wq->idxd;
779 	struct device *dev = &idxd->pdev->dev;
780 	u32 wq_offset;
781 	int i;
782 
783 	if (!wq->group)
784 		return 0;
785 
786 	/*
787 	 * Instead of memset the entire shadow copy of WQCFG, copy from the hardware after
788 	 * wq reset. This will copy back the sticky values that are present on some devices.
789 	 */
790 	for (i = 0; i < WQCFG_STRIDES(idxd); i++) {
791 		wq_offset = WQCFG_OFFSET(idxd, wq->id, i);
792 		wq->wqcfg->bits[i] = ioread32(idxd->reg_base + wq_offset);
793 	}
794 
795 	/* byte 0-3 */
796 	wq->wqcfg->wq_size = wq->size;
797 
798 	if (wq->size == 0) {
799 		dev_warn(dev, "Incorrect work queue size: 0\n");
800 		return -EINVAL;
801 	}
802 
803 	/* bytes 4-7 */
804 	wq->wqcfg->wq_thresh = wq->threshold;
805 
806 	/* byte 8-11 */
807 	wq->wqcfg->priv = !!(wq->type == IDXD_WQT_KERNEL);
808 	if (wq_dedicated(wq))
809 		wq->wqcfg->mode = 1;
810 
811 	if (device_pasid_enabled(idxd)) {
812 		wq->wqcfg->pasid_en = 1;
813 		if (wq->type == IDXD_WQT_KERNEL && wq_dedicated(wq))
814 			wq->wqcfg->pasid = idxd->pasid;
815 	}
816 
817 	wq->wqcfg->priority = wq->priority;
818 
819 	if (idxd->hw.gen_cap.block_on_fault &&
820 	    test_bit(WQ_FLAG_BLOCK_ON_FAULT, &wq->flags))
821 		wq->wqcfg->bof = 1;
822 
823 	if (idxd->hw.wq_cap.wq_ats_support)
824 		wq->wqcfg->wq_ats_disable = wq->ats_dis;
825 
826 	/* bytes 12-15 */
827 	wq->wqcfg->max_xfer_shift = ilog2(wq->max_xfer_bytes);
828 	wq->wqcfg->max_batch_shift = ilog2(wq->max_batch_size);
829 
830 	dev_dbg(dev, "WQ %d CFGs\n", wq->id);
831 	for (i = 0; i < WQCFG_STRIDES(idxd); i++) {
832 		wq_offset = WQCFG_OFFSET(idxd, wq->id, i);
833 		iowrite32(wq->wqcfg->bits[i], idxd->reg_base + wq_offset);
834 		dev_dbg(dev, "WQ[%d][%d][%#x]: %#x\n",
835 			wq->id, i, wq_offset,
836 			ioread32(idxd->reg_base + wq_offset));
837 	}
838 
839 	return 0;
840 }
841 
842 static int idxd_wqs_config_write(struct idxd_device *idxd)
843 {
844 	int i, rc;
845 
846 	for (i = 0; i < idxd->max_wqs; i++) {
847 		struct idxd_wq *wq = idxd->wqs[i];
848 
849 		rc = idxd_wq_config_write(wq);
850 		if (rc < 0)
851 			return rc;
852 	}
853 
854 	return 0;
855 }
856 
857 static void idxd_group_flags_setup(struct idxd_device *idxd)
858 {
859 	int i;
860 
861 	/* TC-A 0 and TC-B 1 should be defaults */
862 	for (i = 0; i < idxd->max_groups; i++) {
863 		struct idxd_group *group = idxd->groups[i];
864 
865 		if (group->tc_a == -1)
866 			group->tc_a = group->grpcfg.flags.tc_a = 0;
867 		else
868 			group->grpcfg.flags.tc_a = group->tc_a;
869 		if (group->tc_b == -1)
870 			group->tc_b = group->grpcfg.flags.tc_b = 1;
871 		else
872 			group->grpcfg.flags.tc_b = group->tc_b;
873 		group->grpcfg.flags.use_token_limit = group->use_token_limit;
874 		group->grpcfg.flags.tokens_reserved = group->tokens_reserved;
875 		if (group->tokens_allowed)
876 			group->grpcfg.flags.tokens_allowed =
877 				group->tokens_allowed;
878 		else
879 			group->grpcfg.flags.tokens_allowed = idxd->max_tokens;
880 	}
881 }
882 
883 static int idxd_engines_setup(struct idxd_device *idxd)
884 {
885 	int i, engines = 0;
886 	struct idxd_engine *eng;
887 	struct idxd_group *group;
888 
889 	for (i = 0; i < idxd->max_groups; i++) {
890 		group = idxd->groups[i];
891 		group->grpcfg.engines = 0;
892 	}
893 
894 	for (i = 0; i < idxd->max_engines; i++) {
895 		eng = idxd->engines[i];
896 		group = eng->group;
897 
898 		if (!group)
899 			continue;
900 
901 		group->grpcfg.engines |= BIT(eng->id);
902 		engines++;
903 	}
904 
905 	if (!engines)
906 		return -EINVAL;
907 
908 	return 0;
909 }
910 
911 static int idxd_wqs_setup(struct idxd_device *idxd)
912 {
913 	struct idxd_wq *wq;
914 	struct idxd_group *group;
915 	int i, j, configured = 0;
916 	struct device *dev = &idxd->pdev->dev;
917 
918 	for (i = 0; i < idxd->max_groups; i++) {
919 		group = idxd->groups[i];
920 		for (j = 0; j < 4; j++)
921 			group->grpcfg.wqs[j] = 0;
922 	}
923 
924 	for (i = 0; i < idxd->max_wqs; i++) {
925 		wq = idxd->wqs[i];
926 		group = wq->group;
927 
928 		if (!wq->group)
929 			continue;
930 		if (!wq->size)
931 			continue;
932 
933 		if (wq_shared(wq) && !device_swq_supported(idxd)) {
934 			dev_warn(dev, "No shared wq support but configured.\n");
935 			return -EINVAL;
936 		}
937 
938 		group->grpcfg.wqs[wq->id / 64] |= BIT(wq->id % 64);
939 		configured++;
940 	}
941 
942 	if (configured == 0)
943 		return -EINVAL;
944 
945 	return 0;
946 }
947 
948 int idxd_device_config(struct idxd_device *idxd)
949 {
950 	int rc;
951 
952 	lockdep_assert_held(&idxd->dev_lock);
953 	rc = idxd_wqs_setup(idxd);
954 	if (rc < 0)
955 		return rc;
956 
957 	rc = idxd_engines_setup(idxd);
958 	if (rc < 0)
959 		return rc;
960 
961 	idxd_group_flags_setup(idxd);
962 
963 	rc = idxd_wqs_config_write(idxd);
964 	if (rc < 0)
965 		return rc;
966 
967 	rc = idxd_groups_config_write(idxd);
968 	if (rc < 0)
969 		return rc;
970 
971 	return 0;
972 }
973 
974 static int idxd_wq_load_config(struct idxd_wq *wq)
975 {
976 	struct idxd_device *idxd = wq->idxd;
977 	struct device *dev = &idxd->pdev->dev;
978 	int wqcfg_offset;
979 	int i;
980 
981 	wqcfg_offset = WQCFG_OFFSET(idxd, wq->id, 0);
982 	memcpy_fromio(wq->wqcfg, idxd->reg_base + wqcfg_offset, idxd->wqcfg_size);
983 
984 	wq->size = wq->wqcfg->wq_size;
985 	wq->threshold = wq->wqcfg->wq_thresh;
986 	if (wq->wqcfg->priv)
987 		wq->type = IDXD_WQT_KERNEL;
988 
989 	/* The driver does not support shared WQ mode in read-only config yet */
990 	if (wq->wqcfg->mode == 0 || wq->wqcfg->pasid_en)
991 		return -EOPNOTSUPP;
992 
993 	set_bit(WQ_FLAG_DEDICATED, &wq->flags);
994 
995 	wq->priority = wq->wqcfg->priority;
996 
997 	for (i = 0; i < WQCFG_STRIDES(idxd); i++) {
998 		wqcfg_offset = WQCFG_OFFSET(idxd, wq->id, i);
999 		dev_dbg(dev, "WQ[%d][%d][%#x]: %#x\n", wq->id, i, wqcfg_offset, wq->wqcfg->bits[i]);
1000 	}
1001 
1002 	return 0;
1003 }
1004 
1005 static void idxd_group_load_config(struct idxd_group *group)
1006 {
1007 	struct idxd_device *idxd = group->idxd;
1008 	struct device *dev = &idxd->pdev->dev;
1009 	int i, j, grpcfg_offset;
1010 
1011 	/*
1012 	 * Load WQS bit fields
1013 	 * Iterate through all 256 bits 64 bits at a time
1014 	 */
1015 	for (i = 0; i < GRPWQCFG_STRIDES; i++) {
1016 		struct idxd_wq *wq;
1017 
1018 		grpcfg_offset = GRPWQCFG_OFFSET(idxd, group->id, i);
1019 		group->grpcfg.wqs[i] = ioread64(idxd->reg_base + grpcfg_offset);
1020 		dev_dbg(dev, "GRPCFG wq[%d:%d: %#x]: %#llx\n",
1021 			group->id, i, grpcfg_offset, group->grpcfg.wqs[i]);
1022 
1023 		if (i * 64 >= idxd->max_wqs)
1024 			break;
1025 
1026 		/* Iterate through all 64 bits and check for wq set */
1027 		for (j = 0; j < 64; j++) {
1028 			int id = i * 64 + j;
1029 
1030 			/* No need to check beyond max wqs */
1031 			if (id >= idxd->max_wqs)
1032 				break;
1033 
1034 			/* Set group assignment for wq if wq bit is set */
1035 			if (group->grpcfg.wqs[i] & BIT(j)) {
1036 				wq = idxd->wqs[id];
1037 				wq->group = group;
1038 			}
1039 		}
1040 	}
1041 
1042 	grpcfg_offset = GRPENGCFG_OFFSET(idxd, group->id);
1043 	group->grpcfg.engines = ioread64(idxd->reg_base + grpcfg_offset);
1044 	dev_dbg(dev, "GRPCFG engs[%d: %#x]: %#llx\n", group->id,
1045 		grpcfg_offset, group->grpcfg.engines);
1046 
1047 	/* Iterate through all 64 bits to check engines set */
1048 	for (i = 0; i < 64; i++) {
1049 		if (i >= idxd->max_engines)
1050 			break;
1051 
1052 		if (group->grpcfg.engines & BIT(i)) {
1053 			struct idxd_engine *engine = idxd->engines[i];
1054 
1055 			engine->group = group;
1056 		}
1057 	}
1058 
1059 	grpcfg_offset = GRPFLGCFG_OFFSET(idxd, group->id);
1060 	group->grpcfg.flags.bits = ioread32(idxd->reg_base + grpcfg_offset);
1061 	dev_dbg(dev, "GRPFLAGS flags[%d: %#x]: %#x\n",
1062 		group->id, grpcfg_offset, group->grpcfg.flags.bits);
1063 }
1064 
1065 int idxd_device_load_config(struct idxd_device *idxd)
1066 {
1067 	union gencfg_reg reg;
1068 	int i, rc;
1069 
1070 	reg.bits = ioread32(idxd->reg_base + IDXD_GENCFG_OFFSET);
1071 	idxd->token_limit = reg.token_limit;
1072 
1073 	for (i = 0; i < idxd->max_groups; i++) {
1074 		struct idxd_group *group = idxd->groups[i];
1075 
1076 		idxd_group_load_config(group);
1077 	}
1078 
1079 	for (i = 0; i < idxd->max_wqs; i++) {
1080 		struct idxd_wq *wq = idxd->wqs[i];
1081 
1082 		rc = idxd_wq_load_config(wq);
1083 		if (rc < 0)
1084 			return rc;
1085 	}
1086 
1087 	return 0;
1088 }
1089