1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * SolidRun DPU driver for control plane
4  *
5  * Copyright (C) 2022-2023 SolidRun
6  *
7  * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
8  *
9  */
10 #include <linux/iopoll.h>
11 
12 #include "snet_vdpa.h"
13 
14 /* SNET DPU device ID */
15 #define SNET_DEVICE_ID          0x1000
16 /* SNET signature */
17 #define SNET_SIGNATURE          0xD0D06363
18 /* Max. config version that we can work with */
19 #define SNET_CFG_VERSION        0x2
20 /* Queue align */
21 #define SNET_QUEUE_ALIGNMENT    PAGE_SIZE
22 /* Kick value to notify that new data is available */
23 #define SNET_KICK_VAL           0x1
24 #define SNET_CONFIG_OFF         0x0
25 /* How long we are willing to wait for a SNET device */
26 #define SNET_DETECT_TIMEOUT	5000000
27 /* How long should we wait for the DPU to read our config */
28 #define SNET_READ_CFG_TIMEOUT	3000000
29 /* Size of configs written to the DPU */
30 #define SNET_GENERAL_CFG_LEN	36
31 #define SNET_GENERAL_CFG_VQ_LEN	40
32 
33 static struct snet *vdpa_to_snet(struct vdpa_device *vdpa)
34 {
35 	return container_of(vdpa, struct snet, vdpa);
36 }
37 
38 static irqreturn_t snet_cfg_irq_hndlr(int irq, void *data)
39 {
40 	struct snet *snet = data;
41 	/* Call callback if any */
42 	if (likely(snet->cb.callback))
43 		return snet->cb.callback(snet->cb.private);
44 
45 	return IRQ_HANDLED;
46 }
47 
48 static irqreturn_t snet_vq_irq_hndlr(int irq, void *data)
49 {
50 	struct snet_vq *vq = data;
51 	/* Call callback if any */
52 	if (likely(vq->cb.callback))
53 		return vq->cb.callback(vq->cb.private);
54 
55 	return IRQ_HANDLED;
56 }
57 
58 static void snet_free_irqs(struct snet *snet)
59 {
60 	struct psnet *psnet = snet->psnet;
61 	struct pci_dev *pdev;
62 	u32 i;
63 
64 	/* Which Device allcoated the IRQs? */
65 	if (PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_IRQ_PF))
66 		pdev = snet->pdev->physfn;
67 	else
68 		pdev = snet->pdev;
69 
70 	/* Free config's IRQ */
71 	if (snet->cfg_irq != -1) {
72 		devm_free_irq(&pdev->dev, snet->cfg_irq, snet);
73 		snet->cfg_irq = -1;
74 	}
75 	/* Free VQ IRQs */
76 	for (i = 0; i < snet->cfg->vq_num; i++) {
77 		if (snet->vqs[i] && snet->vqs[i]->irq != -1) {
78 			devm_free_irq(&pdev->dev, snet->vqs[i]->irq, snet->vqs[i]);
79 			snet->vqs[i]->irq = -1;
80 		}
81 	}
82 
83 	/* IRQ vectors are freed when the pci remove callback is called */
84 }
85 
86 static int snet_set_vq_address(struct vdpa_device *vdev, u16 idx, u64 desc_area,
87 			       u64 driver_area, u64 device_area)
88 {
89 	struct snet *snet = vdpa_to_snet(vdev);
90 	/* save received parameters in vqueue sturct */
91 	snet->vqs[idx]->desc_area = desc_area;
92 	snet->vqs[idx]->driver_area = driver_area;
93 	snet->vqs[idx]->device_area = device_area;
94 
95 	return 0;
96 }
97 
98 static void snet_set_vq_num(struct vdpa_device *vdev, u16 idx, u32 num)
99 {
100 	struct snet *snet = vdpa_to_snet(vdev);
101 	/* save num in vqueue */
102 	snet->vqs[idx]->num = num;
103 }
104 
105 static void snet_kick_vq(struct vdpa_device *vdev, u16 idx)
106 {
107 	struct snet *snet = vdpa_to_snet(vdev);
108 	/* not ready - ignore */
109 	if (unlikely(!snet->vqs[idx]->ready))
110 		return;
111 
112 	iowrite32(SNET_KICK_VAL, snet->vqs[idx]->kick_ptr);
113 }
114 
115 static void snet_kick_vq_with_data(struct vdpa_device *vdev, u32 data)
116 {
117 	struct snet *snet = vdpa_to_snet(vdev);
118 	u16 idx = data & 0xFFFF;
119 
120 	/* not ready - ignore */
121 	if (unlikely(!snet->vqs[idx]->ready))
122 		return;
123 
124 	iowrite32((data & 0xFFFF0000) | SNET_KICK_VAL, snet->vqs[idx]->kick_ptr);
125 }
126 
127 static void snet_set_vq_cb(struct vdpa_device *vdev, u16 idx, struct vdpa_callback *cb)
128 {
129 	struct snet *snet = vdpa_to_snet(vdev);
130 
131 	snet->vqs[idx]->cb.callback = cb->callback;
132 	snet->vqs[idx]->cb.private = cb->private;
133 }
134 
135 static void snet_set_vq_ready(struct vdpa_device *vdev, u16 idx, bool ready)
136 {
137 	struct snet *snet = vdpa_to_snet(vdev);
138 
139 	snet->vqs[idx]->ready = ready;
140 }
141 
142 static bool snet_get_vq_ready(struct vdpa_device *vdev, u16 idx)
143 {
144 	struct snet *snet = vdpa_to_snet(vdev);
145 
146 	return snet->vqs[idx]->ready;
147 }
148 
149 static bool snet_vq_state_is_initial(struct snet *snet, const struct vdpa_vq_state *state)
150 {
151 	if (SNET_HAS_FEATURE(snet, VIRTIO_F_RING_PACKED)) {
152 		const struct vdpa_vq_state_packed *p = &state->packed;
153 
154 		if (p->last_avail_counter == 1 && p->last_used_counter == 1 &&
155 		    p->last_avail_idx == 0 && p->last_used_idx == 0)
156 			return true;
157 	} else {
158 		const struct vdpa_vq_state_split *s = &state->split;
159 
160 		if (s->avail_index == 0)
161 			return true;
162 	}
163 
164 	return false;
165 }
166 
167 static int snet_set_vq_state(struct vdpa_device *vdev, u16 idx, const struct vdpa_vq_state *state)
168 {
169 	struct snet *snet = vdpa_to_snet(vdev);
170 
171 	/* We can set any state for config version 2+ */
172 	if (SNET_CFG_VER(snet, 2)) {
173 		memcpy(&snet->vqs[idx]->vq_state, state, sizeof(*state));
174 		return 0;
175 	}
176 
177 	/* Older config - we can't set the VQ state.
178 	 * Return 0 only if this is the initial state we use in the DPU.
179 	 */
180 	if (snet_vq_state_is_initial(snet, state))
181 		return 0;
182 
183 	return -EOPNOTSUPP;
184 }
185 
186 static int snet_get_vq_state(struct vdpa_device *vdev, u16 idx, struct vdpa_vq_state *state)
187 {
188 	struct snet *snet = vdpa_to_snet(vdev);
189 
190 	return snet_read_vq_state(snet, idx, state);
191 }
192 
193 static int snet_get_vq_irq(struct vdpa_device *vdev, u16 idx)
194 {
195 	struct snet *snet = vdpa_to_snet(vdev);
196 
197 	return snet->vqs[idx]->irq;
198 }
199 
200 static u32 snet_get_vq_align(struct vdpa_device *vdev)
201 {
202 	return (u32)SNET_QUEUE_ALIGNMENT;
203 }
204 
205 static int snet_reset_dev(struct snet *snet)
206 {
207 	struct pci_dev *pdev = snet->pdev;
208 	int ret = 0;
209 	u32 i;
210 
211 	/* If status is 0, nothing to do */
212 	if (!snet->status)
213 		return 0;
214 
215 	/* If DPU started, destroy it */
216 	if (snet->status & VIRTIO_CONFIG_S_DRIVER_OK)
217 		ret = snet_destroy_dev(snet);
218 
219 	/* Clear VQs */
220 	for (i = 0; i < snet->cfg->vq_num; i++) {
221 		if (!snet->vqs[i])
222 			continue;
223 		snet->vqs[i]->cb.callback = NULL;
224 		snet->vqs[i]->cb.private = NULL;
225 		snet->vqs[i]->desc_area = 0;
226 		snet->vqs[i]->device_area = 0;
227 		snet->vqs[i]->driver_area = 0;
228 		snet->vqs[i]->ready = false;
229 	}
230 
231 	/* Clear config callback */
232 	snet->cb.callback = NULL;
233 	snet->cb.private = NULL;
234 	/* Free IRQs */
235 	snet_free_irqs(snet);
236 	/* Reset status */
237 	snet->status = 0;
238 	snet->dpu_ready = false;
239 
240 	if (ret)
241 		SNET_WARN(pdev, "Incomplete reset to SNET[%u] device, err: %d\n", snet->sid, ret);
242 	else
243 		SNET_DBG(pdev, "Reset SNET[%u] device\n", snet->sid);
244 
245 	return 0;
246 }
247 
248 static int snet_reset(struct vdpa_device *vdev)
249 {
250 	struct snet *snet = vdpa_to_snet(vdev);
251 
252 	return snet_reset_dev(snet);
253 }
254 
255 static size_t snet_get_config_size(struct vdpa_device *vdev)
256 {
257 	struct snet *snet = vdpa_to_snet(vdev);
258 
259 	return (size_t)snet->cfg->cfg_size;
260 }
261 
262 static u64 snet_get_features(struct vdpa_device *vdev)
263 {
264 	struct snet *snet = vdpa_to_snet(vdev);
265 
266 	return snet->cfg->features;
267 }
268 
269 static int snet_set_drv_features(struct vdpa_device *vdev, u64 features)
270 {
271 	struct snet *snet = vdpa_to_snet(vdev);
272 
273 	snet->negotiated_features = snet->cfg->features & features;
274 	return 0;
275 }
276 
277 static u64 snet_get_drv_features(struct vdpa_device *vdev)
278 {
279 	struct snet *snet = vdpa_to_snet(vdev);
280 
281 	return snet->negotiated_features;
282 }
283 
284 static u16 snet_get_vq_num_max(struct vdpa_device *vdev)
285 {
286 	struct snet *snet = vdpa_to_snet(vdev);
287 
288 	return (u16)snet->cfg->vq_size;
289 }
290 
291 static void snet_set_config_cb(struct vdpa_device *vdev, struct vdpa_callback *cb)
292 {
293 	struct snet *snet = vdpa_to_snet(vdev);
294 
295 	snet->cb.callback = cb->callback;
296 	snet->cb.private = cb->private;
297 }
298 
299 static u32 snet_get_device_id(struct vdpa_device *vdev)
300 {
301 	struct snet *snet = vdpa_to_snet(vdev);
302 
303 	return snet->cfg->virtio_id;
304 }
305 
306 static u32 snet_get_vendor_id(struct vdpa_device *vdev)
307 {
308 	return (u32)PCI_VENDOR_ID_SOLIDRUN;
309 }
310 
311 static u8 snet_get_status(struct vdpa_device *vdev)
312 {
313 	struct snet *snet = vdpa_to_snet(vdev);
314 
315 	return snet->status;
316 }
317 
318 static int snet_write_conf(struct snet *snet)
319 {
320 	u32 off, i, tmp;
321 	int ret;
322 
323 	/* No need to write the config twice */
324 	if (snet->dpu_ready)
325 		return true;
326 
327 	/* Snet data :
328 	 *
329 	 * General data: SNET_GENERAL_CFG_LEN bytes long
330 	 *  0             0x4       0x8        0xC               0x10      0x14        0x1C     0x24
331 	 *  | MAGIC NUMBER | CFG VER | SNET SID | NUMBER OF QUEUES | IRQ IDX | FEATURES |  RSVD  |
332 	 *
333 	 * For every VQ: SNET_GENERAL_CFG_VQ_LEN bytes long
334 	 * 0                          0x4        0x8
335 	 * |  VQ SID  AND  QUEUE SIZE | IRQ Index |
336 	 * |             DESC AREA                |
337 	 * |            DEVICE AREA               |
338 	 * |            DRIVER AREA               |
339 	 * |    VQ STATE (CFG 2+)     |   RSVD    |
340 	 *
341 	 * Magic number should be written last, this is the DPU indication that the data is ready
342 	 */
343 
344 	/* Init offset */
345 	off = snet->psnet->cfg.host_cfg_off;
346 
347 	/* Ignore magic number for now */
348 	off += 4;
349 	snet_write32(snet, off, snet->psnet->negotiated_cfg_ver);
350 	off += 4;
351 	snet_write32(snet, off, snet->sid);
352 	off += 4;
353 	snet_write32(snet, off, snet->cfg->vq_num);
354 	off += 4;
355 	snet_write32(snet, off, snet->cfg_irq_idx);
356 	off += 4;
357 	snet_write64(snet, off, snet->negotiated_features);
358 	off += 8;
359 	/* Ignore reserved */
360 	off += 8;
361 	/* Write VQs */
362 	for (i = 0 ; i < snet->cfg->vq_num ; i++) {
363 		tmp = (i << 16) | (snet->vqs[i]->num & 0xFFFF);
364 		snet_write32(snet, off, tmp);
365 		off += 4;
366 		snet_write32(snet, off, snet->vqs[i]->irq_idx);
367 		off += 4;
368 		snet_write64(snet, off, snet->vqs[i]->desc_area);
369 		off += 8;
370 		snet_write64(snet, off, snet->vqs[i]->device_area);
371 		off += 8;
372 		snet_write64(snet, off, snet->vqs[i]->driver_area);
373 		off += 8;
374 		/* Write VQ state if config version is 2+ */
375 		if (SNET_CFG_VER(snet, 2))
376 			snet_write32(snet, off, *(u32 *)&snet->vqs[i]->vq_state);
377 		off += 4;
378 
379 		/* Ignore reserved */
380 		off += 4;
381 	}
382 
383 	/* Write magic number - data is ready */
384 	snet_write32(snet, snet->psnet->cfg.host_cfg_off, SNET_SIGNATURE);
385 
386 	/* The DPU will ACK the config by clearing the signature */
387 	ret = readx_poll_timeout(ioread32, snet->bar + snet->psnet->cfg.host_cfg_off,
388 				 tmp, !tmp, 10, SNET_READ_CFG_TIMEOUT);
389 	if (ret) {
390 		SNET_ERR(snet->pdev, "Timeout waiting for the DPU to read the config\n");
391 		return false;
392 	}
393 
394 	/* set DPU flag */
395 	snet->dpu_ready = true;
396 
397 	return true;
398 }
399 
400 static int snet_request_irqs(struct pci_dev *pdev, struct snet *snet)
401 {
402 	int ret, i, irq;
403 
404 	/* Request config IRQ */
405 	irq = pci_irq_vector(pdev, snet->cfg_irq_idx);
406 	ret = devm_request_irq(&pdev->dev, irq, snet_cfg_irq_hndlr, 0,
407 			       snet->cfg_irq_name, snet);
408 	if (ret) {
409 		SNET_ERR(pdev, "Failed to request IRQ\n");
410 		return ret;
411 	}
412 	snet->cfg_irq = irq;
413 
414 	/* Request IRQ for every VQ */
415 	for (i = 0; i < snet->cfg->vq_num; i++) {
416 		irq = pci_irq_vector(pdev, snet->vqs[i]->irq_idx);
417 		ret = devm_request_irq(&pdev->dev, irq, snet_vq_irq_hndlr, 0,
418 				       snet->vqs[i]->irq_name, snet->vqs[i]);
419 		if (ret) {
420 			SNET_ERR(pdev, "Failed to request IRQ\n");
421 			return ret;
422 		}
423 		snet->vqs[i]->irq = irq;
424 	}
425 	return 0;
426 }
427 
428 static void snet_set_status(struct vdpa_device *vdev, u8 status)
429 {
430 	struct snet *snet = vdpa_to_snet(vdev);
431 	struct psnet *psnet = snet->psnet;
432 	struct pci_dev *pdev = snet->pdev;
433 	int ret;
434 	bool pf_irqs;
435 
436 	if (status == snet->status)
437 		return;
438 
439 	if ((status & VIRTIO_CONFIG_S_DRIVER_OK) &&
440 	    !(snet->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
441 		/* Request IRQs */
442 		pf_irqs = PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_IRQ_PF);
443 		ret = snet_request_irqs(pf_irqs ? pdev->physfn : pdev, snet);
444 		if (ret)
445 			goto set_err;
446 
447 		/* Write config to the DPU */
448 		if (snet_write_conf(snet)) {
449 			SNET_INFO(pdev, "Create SNET[%u] device\n", snet->sid);
450 		} else {
451 			snet_free_irqs(snet);
452 			goto set_err;
453 		}
454 	}
455 
456 	/* Save the new status */
457 	snet->status = status;
458 	return;
459 
460 set_err:
461 	snet->status |= VIRTIO_CONFIG_S_FAILED;
462 }
463 
464 static void snet_get_config(struct vdpa_device *vdev, unsigned int offset,
465 			    void *buf, unsigned int len)
466 {
467 	struct snet *snet = vdpa_to_snet(vdev);
468 	void __iomem *cfg_ptr = snet->cfg->virtio_cfg + offset;
469 	u8 *buf_ptr = buf;
470 	u32 i;
471 
472 	/* check for offset error */
473 	if (offset + len > snet->cfg->cfg_size)
474 		return;
475 
476 	/* Write into buffer */
477 	for (i = 0; i < len; i++)
478 		*buf_ptr++ = ioread8(cfg_ptr + i);
479 }
480 
481 static void snet_set_config(struct vdpa_device *vdev, unsigned int offset,
482 			    const void *buf, unsigned int len)
483 {
484 	struct snet *snet = vdpa_to_snet(vdev);
485 	void __iomem *cfg_ptr = snet->cfg->virtio_cfg + offset;
486 	const u8 *buf_ptr = buf;
487 	u32 i;
488 
489 	/* check for offset error */
490 	if (offset + len > snet->cfg->cfg_size)
491 		return;
492 
493 	/* Write into PCI BAR */
494 	for (i = 0; i < len; i++)
495 		iowrite8(*buf_ptr++, cfg_ptr + i);
496 }
497 
498 static int snet_suspend(struct vdpa_device *vdev)
499 {
500 	struct snet *snet = vdpa_to_snet(vdev);
501 	int ret;
502 
503 	ret = snet_suspend_dev(snet);
504 	if (ret)
505 		SNET_ERR(snet->pdev, "SNET[%u] suspend failed, err: %d\n", snet->sid, ret);
506 	else
507 		SNET_DBG(snet->pdev, "Suspend SNET[%u] device\n", snet->sid);
508 
509 	return ret;
510 }
511 
512 static const struct vdpa_config_ops snet_config_ops = {
513 	.set_vq_address         = snet_set_vq_address,
514 	.set_vq_num             = snet_set_vq_num,
515 	.kick_vq                = snet_kick_vq,
516 	.kick_vq_with_data	= snet_kick_vq_with_data,
517 	.set_vq_cb              = snet_set_vq_cb,
518 	.set_vq_ready           = snet_set_vq_ready,
519 	.get_vq_ready           = snet_get_vq_ready,
520 	.set_vq_state           = snet_set_vq_state,
521 	.get_vq_state           = snet_get_vq_state,
522 	.get_vq_irq		= snet_get_vq_irq,
523 	.get_vq_align           = snet_get_vq_align,
524 	.reset                  = snet_reset,
525 	.get_config_size        = snet_get_config_size,
526 	.get_device_features    = snet_get_features,
527 	.set_driver_features    = snet_set_drv_features,
528 	.get_driver_features    = snet_get_drv_features,
529 	.get_vq_num_min         = snet_get_vq_num_max,
530 	.get_vq_num_max         = snet_get_vq_num_max,
531 	.set_config_cb          = snet_set_config_cb,
532 	.get_device_id          = snet_get_device_id,
533 	.get_vendor_id          = snet_get_vendor_id,
534 	.get_status             = snet_get_status,
535 	.set_status             = snet_set_status,
536 	.get_config             = snet_get_config,
537 	.set_config             = snet_set_config,
538 	.suspend		= snet_suspend,
539 };
540 
541 static int psnet_open_pf_bar(struct pci_dev *pdev, struct psnet *psnet)
542 {
543 	char name[50];
544 	int ret, i, mask = 0;
545 	/* We don't know which BAR will be used to communicate..
546 	 * We will map every bar with len > 0.
547 	 *
548 	 * Later, we will discover the BAR and unmap all other BARs.
549 	 */
550 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
551 		if (pci_resource_len(pdev, i))
552 			mask |= (1 << i);
553 	}
554 
555 	/* No BAR can be used.. */
556 	if (!mask) {
557 		SNET_ERR(pdev, "Failed to find a PCI BAR\n");
558 		return -ENODEV;
559 	}
560 
561 	snprintf(name, sizeof(name), "psnet[%s]-bars", pci_name(pdev));
562 	ret = pcim_iomap_regions(pdev, mask, name);
563 	if (ret) {
564 		SNET_ERR(pdev, "Failed to request and map PCI BARs\n");
565 		return ret;
566 	}
567 
568 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
569 		if (mask & (1 << i))
570 			psnet->bars[i] = pcim_iomap_table(pdev)[i];
571 	}
572 
573 	return 0;
574 }
575 
576 static int snet_open_vf_bar(struct pci_dev *pdev, struct snet *snet)
577 {
578 	char name[50];
579 	int ret;
580 
581 	snprintf(name, sizeof(name), "snet[%s]-bar", pci_name(pdev));
582 	/* Request and map BAR */
583 	ret = pcim_iomap_regions(pdev, BIT(snet->psnet->cfg.vf_bar), name);
584 	if (ret) {
585 		SNET_ERR(pdev, "Failed to request and map PCI BAR for a VF\n");
586 		return ret;
587 	}
588 
589 	snet->bar = pcim_iomap_table(pdev)[snet->psnet->cfg.vf_bar];
590 
591 	return 0;
592 }
593 
594 static void snet_free_cfg(struct snet_cfg *cfg)
595 {
596 	u32 i;
597 
598 	if (!cfg->devs)
599 		return;
600 
601 	/* Free devices */
602 	for (i = 0; i < cfg->devices_num; i++) {
603 		if (!cfg->devs[i])
604 			break;
605 
606 		kfree(cfg->devs[i]);
607 	}
608 	/* Free pointers to devices */
609 	kfree(cfg->devs);
610 }
611 
612 /* Detect which BAR is used for communication with the device. */
613 static int psnet_detect_bar(struct psnet *psnet, u32 off)
614 {
615 	unsigned long exit_time;
616 	int i;
617 
618 	exit_time = jiffies + usecs_to_jiffies(SNET_DETECT_TIMEOUT);
619 
620 	/* SNET DPU will write SNET's signature when the config is ready. */
621 	while (time_before(jiffies, exit_time)) {
622 		for (i = 0; i < PCI_STD_NUM_BARS; i++) {
623 			/* Is this BAR mapped? */
624 			if (!psnet->bars[i])
625 				continue;
626 
627 			if (ioread32(psnet->bars[i] + off) == SNET_SIGNATURE)
628 				return i;
629 		}
630 		usleep_range(1000, 10000);
631 	}
632 
633 	return -ENODEV;
634 }
635 
636 static void psnet_unmap_unused_bars(struct pci_dev *pdev, struct psnet *psnet)
637 {
638 	int i, mask = 0;
639 
640 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
641 		if (psnet->bars[i] && i != psnet->barno)
642 			mask |= (1 << i);
643 	}
644 
645 	if (mask)
646 		pcim_iounmap_regions(pdev, mask);
647 }
648 
649 /* Read SNET config from PCI BAR */
650 static int psnet_read_cfg(struct pci_dev *pdev, struct psnet *psnet)
651 {
652 	struct snet_cfg *cfg = &psnet->cfg;
653 	u32 i, off;
654 	int barno;
655 
656 	/* Move to where the config starts */
657 	off = SNET_CONFIG_OFF;
658 
659 	/* Find BAR used for communication */
660 	barno = psnet_detect_bar(psnet, off);
661 	if (barno < 0) {
662 		SNET_ERR(pdev, "SNET config is not ready.\n");
663 		return barno;
664 	}
665 
666 	/* Save used BAR number and unmap all other BARs */
667 	psnet->barno = barno;
668 	SNET_DBG(pdev, "Using BAR number %d\n", barno);
669 
670 	psnet_unmap_unused_bars(pdev, psnet);
671 
672 	/* load config from BAR */
673 	cfg->key = psnet_read32(psnet, off);
674 	off += 4;
675 	cfg->cfg_size = psnet_read32(psnet, off);
676 	off += 4;
677 	cfg->cfg_ver = psnet_read32(psnet, off);
678 	off += 4;
679 	/* The negotiated config version is the lower one between this driver's config
680 	 * and the DPU's.
681 	 */
682 	psnet->negotiated_cfg_ver = min_t(u32, cfg->cfg_ver, SNET_CFG_VERSION);
683 	SNET_DBG(pdev, "SNET config version %u\n", psnet->negotiated_cfg_ver);
684 
685 	cfg->vf_num = psnet_read32(psnet, off);
686 	off += 4;
687 	cfg->vf_bar = psnet_read32(psnet, off);
688 	off += 4;
689 	cfg->host_cfg_off = psnet_read32(psnet, off);
690 	off += 4;
691 	cfg->max_size_host_cfg = psnet_read32(psnet, off);
692 	off += 4;
693 	cfg->virtio_cfg_off = psnet_read32(psnet, off);
694 	off += 4;
695 	cfg->kick_off = psnet_read32(psnet, off);
696 	off += 4;
697 	cfg->hwmon_off = psnet_read32(psnet, off);
698 	off += 4;
699 	cfg->ctrl_off = psnet_read32(psnet, off);
700 	off += 4;
701 	cfg->flags = psnet_read32(psnet, off);
702 	off += 4;
703 	/* Ignore Reserved */
704 	off += sizeof(cfg->rsvd);
705 
706 	cfg->devices_num = psnet_read32(psnet, off);
707 	off += 4;
708 	/* Allocate memory to hold pointer to the devices */
709 	cfg->devs = kcalloc(cfg->devices_num, sizeof(void *), GFP_KERNEL);
710 	if (!cfg->devs)
711 		return -ENOMEM;
712 
713 	/* Load device configuration from BAR */
714 	for (i = 0; i < cfg->devices_num; i++) {
715 		cfg->devs[i] = kzalloc(sizeof(*cfg->devs[i]), GFP_KERNEL);
716 		if (!cfg->devs[i]) {
717 			snet_free_cfg(cfg);
718 			return -ENOMEM;
719 		}
720 		/* Read device config */
721 		cfg->devs[i]->virtio_id = psnet_read32(psnet, off);
722 		off += 4;
723 		cfg->devs[i]->vq_num = psnet_read32(psnet, off);
724 		off += 4;
725 		cfg->devs[i]->vq_size = psnet_read32(psnet, off);
726 		off += 4;
727 		cfg->devs[i]->vfid = psnet_read32(psnet, off);
728 		off += 4;
729 		cfg->devs[i]->features = psnet_read64(psnet, off);
730 		off += 8;
731 		/* Ignore Reserved */
732 		off += sizeof(cfg->devs[i]->rsvd);
733 
734 		cfg->devs[i]->cfg_size = psnet_read32(psnet, off);
735 		off += 4;
736 
737 		/* Is the config witten to the DPU going to be too big? */
738 		if (SNET_GENERAL_CFG_LEN + SNET_GENERAL_CFG_VQ_LEN * cfg->devs[i]->vq_num >
739 		    cfg->max_size_host_cfg) {
740 			SNET_ERR(pdev, "Failed to read SNET config, the config is too big..\n");
741 			snet_free_cfg(cfg);
742 			return -EINVAL;
743 		}
744 	}
745 	return 0;
746 }
747 
748 static int psnet_alloc_irq_vector(struct pci_dev *pdev, struct psnet *psnet)
749 {
750 	int ret = 0;
751 	u32 i, irq_num = 0;
752 
753 	/* Let's count how many IRQs we need, 1 for every VQ + 1 for config change */
754 	for (i = 0; i < psnet->cfg.devices_num; i++)
755 		irq_num += psnet->cfg.devs[i]->vq_num + 1;
756 
757 	ret = pci_alloc_irq_vectors(pdev, irq_num, irq_num, PCI_IRQ_MSIX);
758 	if (ret != irq_num) {
759 		SNET_ERR(pdev, "Failed to allocate IRQ vectors\n");
760 		return ret;
761 	}
762 	SNET_DBG(pdev, "Allocated %u IRQ vectors from physical function\n", irq_num);
763 
764 	return 0;
765 }
766 
767 static int snet_alloc_irq_vector(struct pci_dev *pdev, struct snet_dev_cfg *snet_cfg)
768 {
769 	int ret = 0;
770 	u32 irq_num;
771 
772 	/* We want 1 IRQ for every VQ + 1 for config change events */
773 	irq_num = snet_cfg->vq_num + 1;
774 
775 	ret = pci_alloc_irq_vectors(pdev, irq_num, irq_num, PCI_IRQ_MSIX);
776 	if (ret <= 0) {
777 		SNET_ERR(pdev, "Failed to allocate IRQ vectors\n");
778 		return ret;
779 	}
780 
781 	return 0;
782 }
783 
784 static void snet_free_vqs(struct snet *snet)
785 {
786 	u32 i;
787 
788 	if (!snet->vqs)
789 		return;
790 
791 	for (i = 0 ; i < snet->cfg->vq_num ; i++) {
792 		if (!snet->vqs[i])
793 			break;
794 
795 		kfree(snet->vqs[i]);
796 	}
797 	kfree(snet->vqs);
798 }
799 
800 static int snet_build_vqs(struct snet *snet)
801 {
802 	u32 i;
803 	/* Allocate the VQ pointers array */
804 	snet->vqs = kcalloc(snet->cfg->vq_num, sizeof(void *), GFP_KERNEL);
805 	if (!snet->vqs)
806 		return -ENOMEM;
807 
808 	/* Allocate the VQs */
809 	for (i = 0; i < snet->cfg->vq_num; i++) {
810 		snet->vqs[i] = kzalloc(sizeof(*snet->vqs[i]), GFP_KERNEL);
811 		if (!snet->vqs[i]) {
812 			snet_free_vqs(snet);
813 			return -ENOMEM;
814 		}
815 		/* Reset IRQ num */
816 		snet->vqs[i]->irq = -1;
817 		/* VQ serial ID */
818 		snet->vqs[i]->sid = i;
819 		/* Kick address - every VQ gets 4B */
820 		snet->vqs[i]->kick_ptr = snet->bar + snet->psnet->cfg.kick_off +
821 					 snet->vqs[i]->sid * 4;
822 		/* Clear kick address for this VQ */
823 		iowrite32(0, snet->vqs[i]->kick_ptr);
824 	}
825 	return 0;
826 }
827 
828 static int psnet_get_next_irq_num(struct psnet *psnet)
829 {
830 	int irq;
831 
832 	spin_lock(&psnet->lock);
833 	irq = psnet->next_irq++;
834 	spin_unlock(&psnet->lock);
835 
836 	return irq;
837 }
838 
839 static void snet_reserve_irq_idx(struct pci_dev *pdev, struct snet *snet)
840 {
841 	struct psnet *psnet = snet->psnet;
842 	int  i;
843 
844 	/* one IRQ for every VQ, and one for config changes */
845 	snet->cfg_irq_idx = psnet_get_next_irq_num(psnet);
846 	snprintf(snet->cfg_irq_name, SNET_NAME_SIZE, "snet[%s]-cfg[%d]",
847 		 pci_name(pdev), snet->cfg_irq_idx);
848 
849 	for (i = 0; i < snet->cfg->vq_num; i++) {
850 		/* Get next free IRQ ID */
851 		snet->vqs[i]->irq_idx = psnet_get_next_irq_num(psnet);
852 		/* Write IRQ name */
853 		snprintf(snet->vqs[i]->irq_name, SNET_NAME_SIZE, "snet[%s]-vq[%d]",
854 			 pci_name(pdev), snet->vqs[i]->irq_idx);
855 	}
856 }
857 
858 /* Find a device config based on virtual function id */
859 static struct snet_dev_cfg *snet_find_dev_cfg(struct snet_cfg *cfg, u32 vfid)
860 {
861 	u32 i;
862 
863 	for (i = 0; i < cfg->devices_num; i++) {
864 		if (cfg->devs[i]->vfid == vfid)
865 			return cfg->devs[i];
866 	}
867 	/* Oppss.. no config found.. */
868 	return NULL;
869 }
870 
871 /* Probe function for a physical PCI function */
872 static int snet_vdpa_probe_pf(struct pci_dev *pdev)
873 {
874 	struct psnet *psnet;
875 	int ret = 0;
876 	bool pf_irqs = false;
877 
878 	ret = pcim_enable_device(pdev);
879 	if (ret) {
880 		SNET_ERR(pdev, "Failed to enable PCI device\n");
881 		return ret;
882 	}
883 
884 	/* Allocate a PCI physical function device */
885 	psnet = kzalloc(sizeof(*psnet), GFP_KERNEL);
886 	if (!psnet)
887 		return -ENOMEM;
888 
889 	/* Init PSNET spinlock */
890 	spin_lock_init(&psnet->lock);
891 
892 	pci_set_master(pdev);
893 	pci_set_drvdata(pdev, psnet);
894 
895 	/* Open SNET MAIN BAR */
896 	ret = psnet_open_pf_bar(pdev, psnet);
897 	if (ret)
898 		goto free_psnet;
899 
900 	/* Try to read SNET's config from PCI BAR */
901 	ret = psnet_read_cfg(pdev, psnet);
902 	if (ret)
903 		goto free_psnet;
904 
905 	/* If SNET_CFG_FLAG_IRQ_PF flag is set, we should use
906 	 * PF MSI-X vectors
907 	 */
908 	pf_irqs = PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_IRQ_PF);
909 
910 	if (pf_irqs) {
911 		ret = psnet_alloc_irq_vector(pdev, psnet);
912 		if (ret)
913 			goto free_cfg;
914 	}
915 
916 	SNET_DBG(pdev, "Enable %u virtual functions\n", psnet->cfg.vf_num);
917 	ret = pci_enable_sriov(pdev, psnet->cfg.vf_num);
918 	if (ret) {
919 		SNET_ERR(pdev, "Failed to enable SR-IOV\n");
920 		goto free_irq;
921 	}
922 
923 	/* Create HW monitor device */
924 	if (PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_HWMON)) {
925 #if IS_ENABLED(CONFIG_HWMON)
926 		psnet_create_hwmon(pdev);
927 #else
928 		SNET_WARN(pdev, "Can't start HWMON, CONFIG_HWMON is not enabled\n");
929 #endif
930 	}
931 
932 	return 0;
933 
934 free_irq:
935 	if (pf_irqs)
936 		pci_free_irq_vectors(pdev);
937 free_cfg:
938 	snet_free_cfg(&psnet->cfg);
939 free_psnet:
940 	kfree(psnet);
941 	return ret;
942 }
943 
944 /* Probe function for a virtual PCI function */
945 static int snet_vdpa_probe_vf(struct pci_dev *pdev)
946 {
947 	struct pci_dev *pdev_pf = pdev->physfn;
948 	struct psnet *psnet = pci_get_drvdata(pdev_pf);
949 	struct snet_dev_cfg *dev_cfg;
950 	struct snet *snet;
951 	u32 vfid;
952 	int ret;
953 	bool pf_irqs = false;
954 
955 	/* Get virtual function id.
956 	 * (the DPU counts the VFs from 1)
957 	 */
958 	ret = pci_iov_vf_id(pdev);
959 	if (ret < 0) {
960 		SNET_ERR(pdev, "Failed to find a VF id\n");
961 		return ret;
962 	}
963 	vfid = ret + 1;
964 
965 	/* Find the snet_dev_cfg based on vfid */
966 	dev_cfg = snet_find_dev_cfg(&psnet->cfg, vfid);
967 	if (!dev_cfg) {
968 		SNET_WARN(pdev, "Failed to find a VF config..\n");
969 		return -ENODEV;
970 	}
971 
972 	/* Which PCI device should allocate the IRQs?
973 	 * If the SNET_CFG_FLAG_IRQ_PF flag set, the PF device allocates the IRQs
974 	 */
975 	pf_irqs = PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_IRQ_PF);
976 
977 	ret = pcim_enable_device(pdev);
978 	if (ret) {
979 		SNET_ERR(pdev, "Failed to enable PCI VF device\n");
980 		return ret;
981 	}
982 
983 	/* Request for MSI-X IRQs */
984 	if (!pf_irqs) {
985 		ret = snet_alloc_irq_vector(pdev, dev_cfg);
986 		if (ret)
987 			return ret;
988 	}
989 
990 	/* Allocate vdpa device */
991 	snet = vdpa_alloc_device(struct snet, vdpa, &pdev->dev, &snet_config_ops, 1, 1, NULL,
992 				 false);
993 	if (!snet) {
994 		SNET_ERR(pdev, "Failed to allocate a vdpa device\n");
995 		ret = -ENOMEM;
996 		goto free_irqs;
997 	}
998 
999 	/* Init control mutex and spinlock */
1000 	mutex_init(&snet->ctrl_lock);
1001 	spin_lock_init(&snet->ctrl_spinlock);
1002 
1003 	/* Save pci device pointer */
1004 	snet->pdev = pdev;
1005 	snet->psnet = psnet;
1006 	snet->cfg = dev_cfg;
1007 	snet->dpu_ready = false;
1008 	snet->sid = vfid;
1009 	/* Reset IRQ value */
1010 	snet->cfg_irq = -1;
1011 
1012 	ret = snet_open_vf_bar(pdev, snet);
1013 	if (ret)
1014 		goto put_device;
1015 
1016 	/* Create a VirtIO config pointer */
1017 	snet->cfg->virtio_cfg = snet->bar + snet->psnet->cfg.virtio_cfg_off;
1018 
1019 	/* Clear control registers */
1020 	snet_ctrl_clear(snet);
1021 
1022 	pci_set_master(pdev);
1023 	pci_set_drvdata(pdev, snet);
1024 
1025 	ret = snet_build_vqs(snet);
1026 	if (ret)
1027 		goto put_device;
1028 
1029 	/* Reserve IRQ indexes,
1030 	 * The IRQs may be requested and freed multiple times,
1031 	 * but the indexes won't change.
1032 	 */
1033 	snet_reserve_irq_idx(pf_irqs ? pdev_pf : pdev, snet);
1034 
1035 	/*set DMA device*/
1036 	snet->vdpa.dma_dev = &pdev->dev;
1037 
1038 	/* Register VDPA device */
1039 	ret = vdpa_register_device(&snet->vdpa, snet->cfg->vq_num);
1040 	if (ret) {
1041 		SNET_ERR(pdev, "Failed to register vdpa device\n");
1042 		goto free_vqs;
1043 	}
1044 
1045 	return 0;
1046 
1047 free_vqs:
1048 	snet_free_vqs(snet);
1049 put_device:
1050 	put_device(&snet->vdpa.dev);
1051 free_irqs:
1052 	if (!pf_irqs)
1053 		pci_free_irq_vectors(pdev);
1054 	return ret;
1055 }
1056 
1057 static int snet_vdpa_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1058 {
1059 	if (pdev->is_virtfn)
1060 		return snet_vdpa_probe_vf(pdev);
1061 	else
1062 		return snet_vdpa_probe_pf(pdev);
1063 }
1064 
1065 static void snet_vdpa_remove_pf(struct pci_dev *pdev)
1066 {
1067 	struct psnet *psnet = pci_get_drvdata(pdev);
1068 
1069 	pci_disable_sriov(pdev);
1070 	/* If IRQs are allocated from the PF, we should free the IRQs */
1071 	if (PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_IRQ_PF))
1072 		pci_free_irq_vectors(pdev);
1073 
1074 	snet_free_cfg(&psnet->cfg);
1075 	kfree(psnet);
1076 }
1077 
1078 static void snet_vdpa_remove_vf(struct pci_dev *pdev)
1079 {
1080 	struct snet *snet = pci_get_drvdata(pdev);
1081 	struct psnet *psnet = snet->psnet;
1082 
1083 	vdpa_unregister_device(&snet->vdpa);
1084 	snet_free_vqs(snet);
1085 	/* If IRQs are allocated from the VF, we should free the IRQs */
1086 	if (!PSNET_FLAG_ON(psnet, SNET_CFG_FLAG_IRQ_PF))
1087 		pci_free_irq_vectors(pdev);
1088 }
1089 
1090 static void snet_vdpa_remove(struct pci_dev *pdev)
1091 {
1092 	if (pdev->is_virtfn)
1093 		snet_vdpa_remove_vf(pdev);
1094 	else
1095 		snet_vdpa_remove_pf(pdev);
1096 }
1097 
1098 static struct pci_device_id snet_driver_pci_ids[] = {
1099 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_SOLIDRUN, SNET_DEVICE_ID,
1100 			 PCI_VENDOR_ID_SOLIDRUN, SNET_DEVICE_ID) },
1101 	{ 0 },
1102 };
1103 
1104 MODULE_DEVICE_TABLE(pci, snet_driver_pci_ids);
1105 
1106 static struct pci_driver snet_vdpa_driver = {
1107 	.name		= "snet-vdpa-driver",
1108 	.id_table	= snet_driver_pci_ids,
1109 	.probe		= snet_vdpa_probe,
1110 	.remove		= snet_vdpa_remove,
1111 };
1112 
1113 module_pci_driver(snet_vdpa_driver);
1114 
1115 MODULE_AUTHOR("Alvaro Karsz <alvaro.karsz@solid-run.com>");
1116 MODULE_DESCRIPTION("SolidRun vDPA driver");
1117 MODULE_LICENSE("GPL v2");
1118