xref: /dpdk/drivers/net/virtio/virtio_pci.c (revision ebee5594a35e2af743ff03cde7d4bac67ac772f2)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 #include <stdint.h>
34 
35 #ifdef RTE_EXEC_ENV_LINUXAPP
36  #include <dirent.h>
37  #include <fcntl.h>
38 #endif
39 
40 #include "virtio_pci.h"
41 #include "virtio_logs.h"
42 #include "virtqueue.h"
43 
44 /*
45  * Following macros are derived from linux/pci_regs.h, however,
46  * we can't simply include that header here, as there is no such
47  * file for non-Linux platform.
48  */
49 #define PCI_CAPABILITY_LIST	0x34
50 #define PCI_CAP_ID_VNDR		0x09
51 
52 /*
53  * The remaining space is defined by each driver as the per-driver
54  * configuration space.
55  */
56 #define VIRTIO_PCI_CONFIG(hw) (((hw)->use_msix) ? 24 : 20)
57 
58 static inline int
59 check_vq_phys_addr_ok(struct virtqueue *vq)
60 {
61 	/* Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
62 	 * and only accepts 32 bit page frame number.
63 	 * Check if the allocated physical memory exceeds 16TB.
64 	 */
65 	if ((vq->vq_ring_mem + vq->vq_ring_size - 1) >>
66 			(VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
67 		PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
68 		return 0;
69 	}
70 
71 	return 1;
72 }
73 
74 /*
75  * Since we are in legacy mode:
76  * http://ozlabs.org/~rusty/virtio-spec/virtio-0.9.5.pdf
77  *
78  * "Note that this is possible because while the virtio header is PCI (i.e.
79  * little) endian, the device-specific region is encoded in the native endian of
80  * the guest (where such distinction is applicable)."
81  *
82  * For powerpc which supports both, qemu supposes that cpu is big endian and
83  * enforces this for the virtio-net stuff.
84  */
85 static void
86 legacy_read_dev_config(struct virtio_hw *hw, size_t offset,
87 		       void *dst, int length)
88 {
89 #ifdef RTE_ARCH_PPC_64
90 	int size;
91 
92 	while (length > 0) {
93 		if (length >= 4) {
94 			size = 4;
95 			rte_eal_pci_ioport_read(&hw->io, dst, size,
96 				VIRTIO_PCI_CONFIG(hw) + offset);
97 			*(uint32_t *)dst = rte_be_to_cpu_32(*(uint32_t *)dst);
98 		} else if (length >= 2) {
99 			size = 2;
100 			rte_eal_pci_ioport_read(&hw->io, dst, size,
101 				VIRTIO_PCI_CONFIG(hw) + offset);
102 			*(uint16_t *)dst = rte_be_to_cpu_16(*(uint16_t *)dst);
103 		} else {
104 			size = 1;
105 			rte_eal_pci_ioport_read(&hw->io, dst, size,
106 				VIRTIO_PCI_CONFIG(hw) + offset);
107 		}
108 
109 		dst = (char *)dst + size;
110 		offset += size;
111 		length -= size;
112 	}
113 #else
114 	rte_eal_pci_ioport_read(&hw->io, dst, length,
115 				VIRTIO_PCI_CONFIG(hw) + offset);
116 #endif
117 }
118 
119 static void
120 legacy_write_dev_config(struct virtio_hw *hw, size_t offset,
121 			const void *src, int length)
122 {
123 #ifdef RTE_ARCH_PPC_64
124 	union {
125 		uint32_t u32;
126 		uint16_t u16;
127 	} tmp;
128 	int size;
129 
130 	while (length > 0) {
131 		if (length >= 4) {
132 			size = 4;
133 			tmp.u32 = rte_cpu_to_be_32(*(const uint32_t *)src);
134 			rte_eal_pci_ioport_write(&hw->io, &tmp.u32, size,
135 				VIRTIO_PCI_CONFIG(hw) + offset);
136 		} else if (length >= 2) {
137 			size = 2;
138 			tmp.u16 = rte_cpu_to_be_16(*(const uint16_t *)src);
139 			rte_eal_pci_ioport_write(&hw->io, &tmp.u16, size,
140 				VIRTIO_PCI_CONFIG(hw) + offset);
141 		} else {
142 			size = 1;
143 			rte_eal_pci_ioport_write(&hw->io, src, size,
144 				VIRTIO_PCI_CONFIG(hw) + offset);
145 		}
146 
147 		src = (const char *)src + size;
148 		offset += size;
149 		length -= size;
150 	}
151 #else
152 	rte_eal_pci_ioport_write(&hw->io, src, length,
153 				 VIRTIO_PCI_CONFIG(hw) + offset);
154 #endif
155 }
156 
157 static uint64_t
158 legacy_get_features(struct virtio_hw *hw)
159 {
160 	uint32_t dst;
161 
162 	rte_eal_pci_ioport_read(&hw->io, &dst, 4, VIRTIO_PCI_HOST_FEATURES);
163 	return dst;
164 }
165 
166 static void
167 legacy_set_features(struct virtio_hw *hw, uint64_t features)
168 {
169 	if ((features >> 32) != 0) {
170 		PMD_DRV_LOG(ERR,
171 			"only 32 bit features are allowed for legacy virtio!");
172 		return;
173 	}
174 	rte_eal_pci_ioport_write(&hw->io, &features, 4,
175 				 VIRTIO_PCI_GUEST_FEATURES);
176 }
177 
178 static uint8_t
179 legacy_get_status(struct virtio_hw *hw)
180 {
181 	uint8_t dst;
182 
183 	rte_eal_pci_ioport_read(&hw->io, &dst, 1, VIRTIO_PCI_STATUS);
184 	return dst;
185 }
186 
187 static void
188 legacy_set_status(struct virtio_hw *hw, uint8_t status)
189 {
190 	rte_eal_pci_ioport_write(&hw->io, &status, 1, VIRTIO_PCI_STATUS);
191 }
192 
193 static void
194 legacy_reset(struct virtio_hw *hw)
195 {
196 	legacy_set_status(hw, VIRTIO_CONFIG_STATUS_RESET);
197 }
198 
199 static uint8_t
200 legacy_get_isr(struct virtio_hw *hw)
201 {
202 	uint8_t dst;
203 
204 	rte_eal_pci_ioport_read(&hw->io, &dst, 1, VIRTIO_PCI_ISR);
205 	return dst;
206 }
207 
208 /* Enable one vector (0) for Link State Intrerrupt */
209 static uint16_t
210 legacy_set_config_irq(struct virtio_hw *hw, uint16_t vec)
211 {
212 	uint16_t dst;
213 
214 	rte_eal_pci_ioport_write(&hw->io, &vec, 2, VIRTIO_MSI_CONFIG_VECTOR);
215 	rte_eal_pci_ioport_read(&hw->io, &dst, 2, VIRTIO_MSI_CONFIG_VECTOR);
216 	return dst;
217 }
218 
219 static uint16_t
220 legacy_get_queue_num(struct virtio_hw *hw, uint16_t queue_id)
221 {
222 	uint16_t dst;
223 
224 	rte_eal_pci_ioport_write(&hw->io, &queue_id, 2, VIRTIO_PCI_QUEUE_SEL);
225 	rte_eal_pci_ioport_read(&hw->io, &dst, 2, VIRTIO_PCI_QUEUE_NUM);
226 	return dst;
227 }
228 
229 static int
230 legacy_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
231 {
232 	uint32_t src;
233 
234 	if (!check_vq_phys_addr_ok(vq))
235 		return -1;
236 
237 	rte_eal_pci_ioport_write(&hw->io, &vq->vq_queue_index, 2,
238 			 VIRTIO_PCI_QUEUE_SEL);
239 	src = vq->vq_ring_mem >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
240 	rte_eal_pci_ioport_write(&hw->io, &src, 4, VIRTIO_PCI_QUEUE_PFN);
241 
242 	return 0;
243 }
244 
245 static void
246 legacy_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
247 {
248 	uint32_t src = 0;
249 
250 	rte_eal_pci_ioport_write(&hw->io, &vq->vq_queue_index, 2,
251 			 VIRTIO_PCI_QUEUE_SEL);
252 	rte_eal_pci_ioport_write(&hw->io, &src, 4, VIRTIO_PCI_QUEUE_PFN);
253 }
254 
255 static void
256 legacy_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
257 {
258 	rte_eal_pci_ioport_write(&hw->io, &vq->vq_queue_index, 2,
259 			 VIRTIO_PCI_QUEUE_NOTIFY);
260 }
261 
262 #ifdef RTE_EXEC_ENV_LINUXAPP
263 static int
264 legacy_virtio_has_msix(const struct rte_pci_addr *loc)
265 {
266 	DIR *d;
267 	char dirname[PATH_MAX];
268 
269 	snprintf(dirname, sizeof(dirname),
270 		     "%s/" PCI_PRI_FMT "/msi_irqs", pci_get_sysfs_path(),
271 		     loc->domain, loc->bus, loc->devid, loc->function);
272 
273 	d = opendir(dirname);
274 	if (d)
275 		closedir(d);
276 
277 	return d != NULL;
278 }
279 #else
280 static int
281 legacy_virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
282 {
283 	/* nic_uio does not enable interrupts, return 0 (false). */
284 	return 0;
285 }
286 #endif
287 
288 static int
289 legacy_virtio_resource_init(struct rte_pci_device *pci_dev,
290 			    struct virtio_hw *hw, uint32_t *dev_flags)
291 {
292 	if (rte_eal_pci_ioport_map(pci_dev, 0, &hw->io) < 0)
293 		return -1;
294 
295 	if (pci_dev->intr_handle.type != RTE_INTR_HANDLE_UNKNOWN)
296 		*dev_flags |= RTE_ETH_DEV_INTR_LSC;
297 	else
298 		*dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
299 
300 	return 0;
301 }
302 
303 static const struct virtio_pci_ops legacy_ops = {
304 	.read_dev_cfg	= legacy_read_dev_config,
305 	.write_dev_cfg	= legacy_write_dev_config,
306 	.reset		= legacy_reset,
307 	.get_status	= legacy_get_status,
308 	.set_status	= legacy_set_status,
309 	.get_features	= legacy_get_features,
310 	.set_features	= legacy_set_features,
311 	.get_isr	= legacy_get_isr,
312 	.set_config_irq	= legacy_set_config_irq,
313 	.get_queue_num	= legacy_get_queue_num,
314 	.setup_queue	= legacy_setup_queue,
315 	.del_queue	= legacy_del_queue,
316 	.notify_queue	= legacy_notify_queue,
317 };
318 
319 
320 static inline uint8_t
321 io_read8(uint8_t *addr)
322 {
323 	return *(volatile uint8_t *)addr;
324 }
325 
326 static inline void
327 io_write8(uint8_t val, uint8_t *addr)
328 {
329 	*(volatile uint8_t *)addr = val;
330 }
331 
332 static inline uint16_t
333 io_read16(uint16_t *addr)
334 {
335 	return *(volatile uint16_t *)addr;
336 }
337 
338 static inline void
339 io_write16(uint16_t val, uint16_t *addr)
340 {
341 	*(volatile uint16_t *)addr = val;
342 }
343 
344 static inline uint32_t
345 io_read32(uint32_t *addr)
346 {
347 	return *(volatile uint32_t *)addr;
348 }
349 
350 static inline void
351 io_write32(uint32_t val, uint32_t *addr)
352 {
353 	*(volatile uint32_t *)addr = val;
354 }
355 
356 static inline void
357 io_write64_twopart(uint64_t val, uint32_t *lo, uint32_t *hi)
358 {
359 	io_write32(val & ((1ULL << 32) - 1), lo);
360 	io_write32(val >> 32,		     hi);
361 }
362 
363 static void
364 modern_read_dev_config(struct virtio_hw *hw, size_t offset,
365 		       void *dst, int length)
366 {
367 	int i;
368 	uint8_t *p;
369 	uint8_t old_gen, new_gen;
370 
371 	do {
372 		old_gen = io_read8(&hw->common_cfg->config_generation);
373 
374 		p = dst;
375 		for (i = 0;  i < length; i++)
376 			*p++ = io_read8((uint8_t *)hw->dev_cfg + offset + i);
377 
378 		new_gen = io_read8(&hw->common_cfg->config_generation);
379 	} while (old_gen != new_gen);
380 }
381 
382 static void
383 modern_write_dev_config(struct virtio_hw *hw, size_t offset,
384 			const void *src, int length)
385 {
386 	int i;
387 	const uint8_t *p = src;
388 
389 	for (i = 0;  i < length; i++)
390 		io_write8(*p++, (uint8_t *)hw->dev_cfg + offset + i);
391 }
392 
393 static uint64_t
394 modern_get_features(struct virtio_hw *hw)
395 {
396 	uint32_t features_lo, features_hi;
397 
398 	io_write32(0, &hw->common_cfg->device_feature_select);
399 	features_lo = io_read32(&hw->common_cfg->device_feature);
400 
401 	io_write32(1, &hw->common_cfg->device_feature_select);
402 	features_hi = io_read32(&hw->common_cfg->device_feature);
403 
404 	return ((uint64_t)features_hi << 32) | features_lo;
405 }
406 
407 static void
408 modern_set_features(struct virtio_hw *hw, uint64_t features)
409 {
410 	io_write32(0, &hw->common_cfg->guest_feature_select);
411 	io_write32(features & ((1ULL << 32) - 1),
412 		&hw->common_cfg->guest_feature);
413 
414 	io_write32(1, &hw->common_cfg->guest_feature_select);
415 	io_write32(features >> 32,
416 		&hw->common_cfg->guest_feature);
417 }
418 
419 static uint8_t
420 modern_get_status(struct virtio_hw *hw)
421 {
422 	return io_read8(&hw->common_cfg->device_status);
423 }
424 
425 static void
426 modern_set_status(struct virtio_hw *hw, uint8_t status)
427 {
428 	io_write8(status, &hw->common_cfg->device_status);
429 }
430 
431 static void
432 modern_reset(struct virtio_hw *hw)
433 {
434 	modern_set_status(hw, VIRTIO_CONFIG_STATUS_RESET);
435 	modern_get_status(hw);
436 }
437 
438 static uint8_t
439 modern_get_isr(struct virtio_hw *hw)
440 {
441 	return io_read8(hw->isr);
442 }
443 
444 static uint16_t
445 modern_set_config_irq(struct virtio_hw *hw, uint16_t vec)
446 {
447 	io_write16(vec, &hw->common_cfg->msix_config);
448 	return io_read16(&hw->common_cfg->msix_config);
449 }
450 
451 static uint16_t
452 modern_get_queue_num(struct virtio_hw *hw, uint16_t queue_id)
453 {
454 	io_write16(queue_id, &hw->common_cfg->queue_select);
455 	return io_read16(&hw->common_cfg->queue_size);
456 }
457 
458 static int
459 modern_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
460 {
461 	uint64_t desc_addr, avail_addr, used_addr;
462 	uint16_t notify_off;
463 
464 	if (!check_vq_phys_addr_ok(vq))
465 		return -1;
466 
467 	desc_addr = vq->vq_ring_mem;
468 	avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
469 	used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
470 							 ring[vq->vq_nentries]),
471 				   VIRTIO_PCI_VRING_ALIGN);
472 
473 	io_write16(vq->vq_queue_index, &hw->common_cfg->queue_select);
474 
475 	io_write64_twopart(desc_addr, &hw->common_cfg->queue_desc_lo,
476 				      &hw->common_cfg->queue_desc_hi);
477 	io_write64_twopart(avail_addr, &hw->common_cfg->queue_avail_lo,
478 				       &hw->common_cfg->queue_avail_hi);
479 	io_write64_twopart(used_addr, &hw->common_cfg->queue_used_lo,
480 				      &hw->common_cfg->queue_used_hi);
481 
482 	notify_off = io_read16(&hw->common_cfg->queue_notify_off);
483 	vq->notify_addr = (void *)((uint8_t *)hw->notify_base +
484 				notify_off * hw->notify_off_multiplier);
485 
486 	io_write16(1, &hw->common_cfg->queue_enable);
487 
488 	PMD_INIT_LOG(DEBUG, "queue %u addresses:", vq->vq_queue_index);
489 	PMD_INIT_LOG(DEBUG, "\t desc_addr: %" PRIx64, desc_addr);
490 	PMD_INIT_LOG(DEBUG, "\t aval_addr: %" PRIx64, avail_addr);
491 	PMD_INIT_LOG(DEBUG, "\t used_addr: %" PRIx64, used_addr);
492 	PMD_INIT_LOG(DEBUG, "\t notify addr: %p (notify offset: %u)",
493 		vq->notify_addr, notify_off);
494 
495 	return 0;
496 }
497 
498 static void
499 modern_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
500 {
501 	io_write16(vq->vq_queue_index, &hw->common_cfg->queue_select);
502 
503 	io_write64_twopart(0, &hw->common_cfg->queue_desc_lo,
504 				  &hw->common_cfg->queue_desc_hi);
505 	io_write64_twopart(0, &hw->common_cfg->queue_avail_lo,
506 				  &hw->common_cfg->queue_avail_hi);
507 	io_write64_twopart(0, &hw->common_cfg->queue_used_lo,
508 				  &hw->common_cfg->queue_used_hi);
509 
510 	io_write16(0, &hw->common_cfg->queue_enable);
511 }
512 
513 static void
514 modern_notify_queue(struct virtio_hw *hw __rte_unused, struct virtqueue *vq)
515 {
516 	io_write16(1, vq->notify_addr);
517 }
518 
519 static const struct virtio_pci_ops modern_ops = {
520 	.read_dev_cfg	= modern_read_dev_config,
521 	.write_dev_cfg	= modern_write_dev_config,
522 	.reset		= modern_reset,
523 	.get_status	= modern_get_status,
524 	.set_status	= modern_set_status,
525 	.get_features	= modern_get_features,
526 	.set_features	= modern_set_features,
527 	.get_isr	= modern_get_isr,
528 	.set_config_irq	= modern_set_config_irq,
529 	.get_queue_num	= modern_get_queue_num,
530 	.setup_queue	= modern_setup_queue,
531 	.del_queue	= modern_del_queue,
532 	.notify_queue	= modern_notify_queue,
533 };
534 
535 
536 void
537 vtpci_read_dev_config(struct virtio_hw *hw, size_t offset,
538 		      void *dst, int length)
539 {
540 	hw->vtpci_ops->read_dev_cfg(hw, offset, dst, length);
541 }
542 
543 void
544 vtpci_write_dev_config(struct virtio_hw *hw, size_t offset,
545 		       const void *src, int length)
546 {
547 	hw->vtpci_ops->write_dev_cfg(hw, offset, src, length);
548 }
549 
550 uint64_t
551 vtpci_negotiate_features(struct virtio_hw *hw, uint64_t host_features)
552 {
553 	uint64_t features;
554 
555 	/*
556 	 * Limit negotiated features to what the driver, virtqueue, and
557 	 * host all support.
558 	 */
559 	features = host_features & hw->guest_features;
560 	hw->vtpci_ops->set_features(hw, features);
561 
562 	return features;
563 }
564 
565 void
566 vtpci_reset(struct virtio_hw *hw)
567 {
568 	hw->vtpci_ops->set_status(hw, VIRTIO_CONFIG_STATUS_RESET);
569 	/* flush status write */
570 	hw->vtpci_ops->get_status(hw);
571 }
572 
573 void
574 vtpci_reinit_complete(struct virtio_hw *hw)
575 {
576 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
577 }
578 
579 void
580 vtpci_set_status(struct virtio_hw *hw, uint8_t status)
581 {
582 	if (status != VIRTIO_CONFIG_STATUS_RESET)
583 		status |= hw->vtpci_ops->get_status(hw);
584 
585 	hw->vtpci_ops->set_status(hw, status);
586 }
587 
588 uint8_t
589 vtpci_get_status(struct virtio_hw *hw)
590 {
591 	return hw->vtpci_ops->get_status(hw);
592 }
593 
594 uint8_t
595 vtpci_isr(struct virtio_hw *hw)
596 {
597 	return hw->vtpci_ops->get_isr(hw);
598 }
599 
600 
601 /* Enable one vector (0) for Link State Intrerrupt */
602 uint16_t
603 vtpci_irq_config(struct virtio_hw *hw, uint16_t vec)
604 {
605 	return hw->vtpci_ops->set_config_irq(hw, vec);
606 }
607 
608 static void *
609 get_cfg_addr(struct rte_pci_device *dev, struct virtio_pci_cap *cap)
610 {
611 	uint8_t  bar    = cap->bar;
612 	uint32_t length = cap->length;
613 	uint32_t offset = cap->offset;
614 	uint8_t *base;
615 
616 	if (bar > 5) {
617 		PMD_INIT_LOG(ERR, "invalid bar: %u", bar);
618 		return NULL;
619 	}
620 
621 	if (offset + length < offset) {
622 		PMD_INIT_LOG(ERR, "offset(%u) + length(%u) overflows",
623 			offset, length);
624 		return NULL;
625 	}
626 
627 	if (offset + length > dev->mem_resource[bar].len) {
628 		PMD_INIT_LOG(ERR,
629 			"invalid cap: overflows bar space: %u > %" PRIu64,
630 			offset + length, dev->mem_resource[bar].len);
631 		return NULL;
632 	}
633 
634 	base = dev->mem_resource[bar].addr;
635 	if (base == NULL) {
636 		PMD_INIT_LOG(ERR, "bar %u base addr is NULL", bar);
637 		return NULL;
638 	}
639 
640 	return base + offset;
641 }
642 
643 static int
644 virtio_read_caps(struct rte_pci_device *dev, struct virtio_hw *hw)
645 {
646 	uint8_t pos;
647 	struct virtio_pci_cap cap;
648 	int ret;
649 
650 	if (rte_eal_pci_map_device(dev)) {
651 		PMD_INIT_LOG(DEBUG, "failed to map pci device!");
652 		return -1;
653 	}
654 
655 	ret = rte_eal_pci_read_config(dev, &pos, 1, PCI_CAPABILITY_LIST);
656 	if (ret < 0) {
657 		PMD_INIT_LOG(DEBUG, "failed to read pci capability list");
658 		return -1;
659 	}
660 
661 	while (pos) {
662 		ret = rte_eal_pci_read_config(dev, &cap, sizeof(cap), pos);
663 		if (ret < 0) {
664 			PMD_INIT_LOG(ERR,
665 				"failed to read pci cap at pos: %x", pos);
666 			break;
667 		}
668 
669 		if (cap.cap_vndr != PCI_CAP_ID_VNDR) {
670 			PMD_INIT_LOG(DEBUG,
671 				"[%2x] skipping non VNDR cap id: %02x",
672 				pos, cap.cap_vndr);
673 			goto next;
674 		}
675 
676 		PMD_INIT_LOG(DEBUG,
677 			"[%2x] cfg type: %u, bar: %u, offset: %04x, len: %u",
678 			pos, cap.cfg_type, cap.bar, cap.offset, cap.length);
679 
680 		switch (cap.cfg_type) {
681 		case VIRTIO_PCI_CAP_COMMON_CFG:
682 			hw->common_cfg = get_cfg_addr(dev, &cap);
683 			break;
684 		case VIRTIO_PCI_CAP_NOTIFY_CFG:
685 			rte_eal_pci_read_config(dev, &hw->notify_off_multiplier,
686 						4, pos + sizeof(cap));
687 			hw->notify_base = get_cfg_addr(dev, &cap);
688 			break;
689 		case VIRTIO_PCI_CAP_DEVICE_CFG:
690 			hw->dev_cfg = get_cfg_addr(dev, &cap);
691 			break;
692 		case VIRTIO_PCI_CAP_ISR_CFG:
693 			hw->isr = get_cfg_addr(dev, &cap);
694 			break;
695 		}
696 
697 next:
698 		pos = cap.cap_next;
699 	}
700 
701 	if (hw->common_cfg == NULL || hw->notify_base == NULL ||
702 	    hw->dev_cfg == NULL    || hw->isr == NULL) {
703 		PMD_INIT_LOG(INFO, "no modern virtio pci device found.");
704 		return -1;
705 	}
706 
707 	PMD_INIT_LOG(INFO, "found modern virtio pci device.");
708 
709 	PMD_INIT_LOG(DEBUG, "common cfg mapped at: %p", hw->common_cfg);
710 	PMD_INIT_LOG(DEBUG, "device cfg mapped at: %p", hw->dev_cfg);
711 	PMD_INIT_LOG(DEBUG, "isr cfg mapped at: %p", hw->isr);
712 	PMD_INIT_LOG(DEBUG, "notify base: %p, notify off multiplier: %u",
713 		hw->notify_base, hw->notify_off_multiplier);
714 
715 	return 0;
716 }
717 
718 /*
719  * Return -1:
720  *   if there is error mapping with VFIO/UIO.
721  *   if port map error when driver type is KDRV_NONE.
722  *   if whitelisted but driver type is KDRV_UNKNOWN.
723  * Return 1 if kernel driver is managing the device.
724  * Return 0 on success.
725  */
726 int
727 vtpci_init(struct rte_pci_device *dev, struct virtio_hw *hw,
728 	   uint32_t *dev_flags)
729 {
730 	hw->dev = dev;
731 
732 	/*
733 	 * Try if we can succeed reading virtio pci caps, which exists
734 	 * only on modern pci device. If failed, we fallback to legacy
735 	 * virtio handling.
736 	 */
737 	if (virtio_read_caps(dev, hw) == 0) {
738 		PMD_INIT_LOG(INFO, "modern virtio pci detected.");
739 		hw->vtpci_ops = &modern_ops;
740 		hw->modern    = 1;
741 		*dev_flags |= RTE_ETH_DEV_INTR_LSC;
742 		return 0;
743 	}
744 
745 	PMD_INIT_LOG(INFO, "trying with legacy virtio pci.");
746 	if (legacy_virtio_resource_init(dev, hw, dev_flags) < 0) {
747 		if (dev->kdrv == RTE_KDRV_UNKNOWN &&
748 		    (!dev->devargs ||
749 		     dev->devargs->type != RTE_DEVTYPE_WHITELISTED_PCI)) {
750 			PMD_INIT_LOG(INFO,
751 				"skip kernel managed virtio device.");
752 			return 1;
753 		}
754 		return -1;
755 	}
756 
757 	hw->vtpci_ops = &legacy_ops;
758 	hw->use_msix = legacy_virtio_has_msix(&dev->addr);
759 	hw->modern   = 0;
760 
761 	return 0;
762 }
763