pci.c revision f41244f4
1/*
2 * Copyright (c) 2018 Cisco and/or its affiliates.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
6 *
7 *     http://www.apache.org/licenses/LICENSE-2.0
8 *
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
14 */
15
16#include <fcntl.h>
17#include <sys/ioctl.h>
18
19#include <vppinfra/types.h>
20#include <vlib/vlib.h>
21#include <vlib/pci/pci.h>
22#include <vnet/ethernet/ethernet.h>
23#include <vnet/ip/ip4_packet.h>
24#include <vnet/ip/ip6_packet.h>
25#include <vnet/devices/virtio/virtio.h>
26#include <vnet/devices/virtio/pci.h>
27
28#define PCI_VENDOR_ID_VIRTIO				0x1af4
29#define PCI_DEVICE_ID_VIRTIO_NIC			0x1000
30/* Doesn't support modern device */
31#define PCI_DEVICE_ID_VIRTIO_NIC_MODERN			0x1041
32
33#define PCI_CAPABILITY_LIST     0x34
34#define PCI_CAP_ID_VNDR         0x09
35#define PCI_CAP_ID_MSIX         0x11
36
37#define PCI_MSIX_ENABLE 0x8000
38
39#define PCI_CONFIG_SIZE(vif) ((vif->msix_enabled == VIRTIO_MSIX_ENABLED) ? \
40  24 : 20)
41
42static pci_device_id_t virtio_pci_device_ids[] = {
43  {
44   .vendor_id = PCI_VENDOR_ID_VIRTIO,
45   .device_id = PCI_DEVICE_ID_VIRTIO_NIC},
46  {
47   .vendor_id = PCI_VENDOR_ID_VIRTIO,
48   .device_id = PCI_DEVICE_ID_VIRTIO_NIC_MODERN},
49  {0},
50};
51
52static void
53virtio_pci_legacy_read_config (vlib_main_t * vm, virtio_if_t * vif, void *dst,
54			       int len, u32 addr)
55{
56  u32 size = 0;
57  vlib_pci_dev_handle_t h = vif->pci_dev_handle;
58
59  while (len > 0)
60    {
61      if (len >= 4)
62	{
63	  size = 4;
64	  vlib_pci_read_io_u32 (vm, h, PCI_CONFIG_SIZE (vif) + addr, dst);
65	}
66      else if (len >= 2)
67	{
68	  size = 2;
69	  vlib_pci_read_io_u16 (vm, h, PCI_CONFIG_SIZE (vif) + addr, dst);
70	}
71      else
72	{
73	  size = 1;
74	  vlib_pci_read_io_u8 (vm, h, PCI_CONFIG_SIZE (vif) + addr, dst);
75	}
76      dst = (u8 *) dst + size;
77      addr += size;
78      len -= size;
79    }
80}
81
82static void
83virtio_pci_legacy_write_config (vlib_main_t * vm, virtio_if_t * vif,
84				void *src, int len, u32 addr)
85{
86  u32 size = 0;
87  vlib_pci_dev_handle_t h = vif->pci_dev_handle;
88
89  while (len > 0)
90    {
91      if (len >= 4)
92	{
93	  size = 4;
94	  vlib_pci_write_io_u32 (vm, h, PCI_CONFIG_SIZE (vif) + addr, src);
95	}
96      else if (len >= 2)
97	{
98	  size = 2;
99	  vlib_pci_write_io_u16 (vm, h, PCI_CONFIG_SIZE (vif) + addr, src);
100	}
101      else
102	{
103	  size = 1;
104	  vlib_pci_write_io_u8 (vm, h, PCI_CONFIG_SIZE (vif) + addr, src);
105	}
106      src = (u8 *) src + size;
107      addr += size;
108      len -= size;
109    }
110}
111
112static u64
113virtio_pci_legacy_get_features (vlib_main_t * vm, virtio_if_t * vif)
114{
115  u32 features;
116  vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_HOST_FEATURES,
117			&features);
118  return features;
119}
120
121static u32
122virtio_pci_legacy_set_features (vlib_main_t * vm, virtio_if_t * vif,
123				u64 features)
124{
125  if ((features >> 32) != 0)
126    {
127      clib_warning ("only 32 bit features are allowed for legacy virtio!");
128    }
129  u32 feature = 0, guest_features = (u32) features;
130  vlib_pci_write_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_GUEST_FEATURES,
131			 &guest_features);
132  vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_GUEST_FEATURES,
133			&feature);
134  return feature;
135}
136
137static u8
138virtio_pci_legacy_get_status (vlib_main_t * vm, virtio_if_t * vif)
139{
140  u8 status = 0;
141  vlib_pci_read_io_u8 (vm, vif->pci_dev_handle, VIRTIO_PCI_STATUS, &status);
142  return status;
143}
144
145static void
146virtio_pci_legacy_set_status (vlib_main_t * vm, virtio_if_t * vif, u8 status)
147{
148  if (status != VIRTIO_CONFIG_STATUS_RESET)
149    status |= virtio_pci_legacy_get_status (vm, vif);
150  vlib_pci_write_io_u8 (vm, vif->pci_dev_handle, VIRTIO_PCI_STATUS, &status);
151}
152
153static u8
154virtio_pci_legacy_reset (vlib_main_t * vm, virtio_if_t * vif)
155{
156  virtio_pci_legacy_set_status (vm, vif, VIRTIO_CONFIG_STATUS_RESET);
157  return virtio_pci_legacy_get_status (vm, vif);
158}
159
160static u8
161virtio_pci_legacy_get_isr (vlib_main_t * vm, virtio_if_t * vif)
162{
163  u8 isr = 0;
164  vlib_pci_read_io_u8 (vm, vif->pci_dev_handle, VIRTIO_PCI_ISR, &isr);
165  return isr;
166}
167
168static u16
169virtio_pci_legacy_get_queue_num (vlib_main_t * vm, virtio_if_t * vif,
170				 u16 queue_id)
171{
172  u16 queue_num = 0;
173  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_SEL,
174			 &queue_id);
175  vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_NUM,
176			&queue_num);
177  return queue_num;
178}
179
180static int
181virtio_pci_legacy_setup_queue (vlib_main_t * vm, virtio_if_t * vif,
182			       u16 queue_id, void *p)
183{
184  u64 addr = vlib_physmem_get_pa (vm, p) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
185  u32 addr2 = 0;
186  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_SEL,
187			 &queue_id);
188  vlib_pci_write_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_PFN,
189			 (u32 *) & addr);
190  vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_PFN,
191			&addr2);
192  if ((u32) addr == addr2)
193    return 0;
194  return 1;
195}
196
197static void
198virtio_pci_legacy_del_queue (vlib_main_t * vm, virtio_if_t * vif,
199			     u16 queue_id)
200{
201  u32 src = 0;
202  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_SEL,
203			 &queue_id);
204  vlib_pci_write_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_PFN, &src);
205}
206
207inline void
208virtio_pci_legacy_notify_queue (vlib_main_t * vm, virtio_if_t * vif,
209				u16 queue_id)
210{
211  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_NOTIFY,
212			 &queue_id);
213}
214
215/* Enable one vector (0) for Link State Intrerrupt */
216static u16
217virtio_pci_legacy_set_config_irq (vlib_main_t * vm, virtio_if_t * vif,
218				  u16 vec)
219{
220  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_MSI_CONFIG_VECTOR,
221			 &vec);
222  vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_MSI_CONFIG_VECTOR,
223			&vec);
224  return vec;
225}
226
227static u16
228virtio_pci_legacy_set_queue_irq (vlib_main_t * vm, virtio_if_t * vif, u16 vec,
229				 u16 queue_id)
230{
231  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_SEL,
232			 &queue_id);
233  vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_MSI_QUEUE_VECTOR,
234			 &vec);
235  vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_MSI_QUEUE_VECTOR,
236			&vec);
237  return vec;
238}
239
240static u32
241virtio_pci_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hw,
242			u32 flags)
243{
244  return 0;
245}
246
247static clib_error_t *
248virtio_pci_get_max_virtqueue_pairs (vlib_main_t * vm, virtio_if_t * vif)
249{
250  virtio_net_config_t config;
251  clib_error_t *error = 0;
252  u16 max_queue_pairs = 1;
253
254  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_MQ))
255    {
256      virtio_pci_legacy_read_config (vm, vif, &config.max_virtqueue_pairs,
257				     sizeof (config.max_virtqueue_pairs),
258				     STRUCT_OFFSET_OF (virtio_net_config_t,
259						       max_virtqueue_pairs));
260      max_queue_pairs = config.max_virtqueue_pairs;
261    }
262
263  virtio_log_debug (vif, "max queue pair is %x", max_queue_pairs);
264  if (max_queue_pairs < 1 || max_queue_pairs > 0x8000)
265    return clib_error_return (error, "max queue pair is %x", max_queue_pairs);
266
267  vif->max_queue_pairs = max_queue_pairs;
268  return error;
269}
270
271static void
272virtio_pci_set_mac (vlib_main_t * vm, virtio_if_t * vif)
273{
274  virtio_pci_legacy_write_config (vm, vif, vif->mac_addr,
275				  sizeof (vif->mac_addr), 0);
276}
277
278static u32
279virtio_pci_get_mac (vlib_main_t * vm, virtio_if_t * vif)
280{
281  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_MAC))
282    {
283      virtio_pci_legacy_read_config (vm, vif, vif->mac_addr,
284				     sizeof (vif->mac_addr), 0);
285      return 0;
286    }
287  return 1;
288}
289
290static u16
291virtio_pci_is_link_up (vlib_main_t * vm, virtio_if_t * vif)
292{
293  /*
294   * Minimal driver: assumes link is up
295   */
296  u16 status = 1;
297  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_STATUS))
298    virtio_pci_legacy_read_config (vm, vif, &status, sizeof (status),	/* mac */
299				   STRUCT_OFFSET_OF (virtio_net_config_t,
300						     status));
301  return status;
302}
303
304static void
305virtio_pci_irq_0_handler (vlib_main_t * vm, vlib_pci_dev_handle_t h, u16 line)
306{
307  vnet_main_t *vnm = vnet_get_main ();
308  virtio_main_t *vim = &virtio_main;
309  uword pd = vlib_pci_get_private_data (vm, h);
310  virtio_if_t *vif = pool_elt_at_index (vim->interfaces, pd);
311  u16 qid = line;
312
313  vnet_device_input_set_interrupt_pending (vnm, vif->hw_if_index, qid);
314}
315
316static void
317virtio_pci_irq_1_handler (vlib_main_t * vm, vlib_pci_dev_handle_t h, u16 line)
318{
319  vnet_main_t *vnm = vnet_get_main ();
320  virtio_main_t *vim = &virtio_main;
321  uword pd = vlib_pci_get_private_data (vm, h);
322  virtio_if_t *vif = pool_elt_at_index (vim->interfaces, pd);
323
324  if (virtio_pci_is_link_up (vm, vif) & VIRTIO_NET_S_LINK_UP)
325    {
326      vif->flags |= VIRTIO_IF_FLAG_ADMIN_UP;
327      vnet_hw_interface_set_flags (vnm, vif->hw_if_index,
328				   VNET_HW_INTERFACE_FLAG_LINK_UP);
329    }
330  else
331    {
332      vif->flags &= ~VIRTIO_IF_FLAG_ADMIN_UP;
333      vnet_hw_interface_set_flags (vnm, vif->hw_if_index, 0);
334    }
335}
336
337static void
338virtio_pci_irq_handler (vlib_main_t * vm, vlib_pci_dev_handle_t h)
339{
340  virtio_main_t *vim = &virtio_main;
341  uword pd = vlib_pci_get_private_data (vm, h);
342  virtio_if_t *vif = pool_elt_at_index (vim->interfaces, pd);
343  u8 isr = 0;
344  u16 line = 0;
345
346  isr = virtio_pci_legacy_get_isr (vm, vif);
347
348  /*
349   * If the lower bit is set: look through the used rings of
350   * all virtqueues for the device, to see if any progress has
351   * been made by the device which requires servicing.
352   */
353  if (isr & VIRTIO_PCI_ISR_INTR)
354    virtio_pci_irq_0_handler (vm, h, line);
355
356  if (isr & VIRTIO_PCI_ISR_CONFIG)
357    virtio_pci_irq_1_handler (vm, h, line);
358}
359
360inline void
361device_status (vlib_main_t * vm, virtio_if_t * vif)
362{
363  struct status_struct
364  {
365    u8 bit;
366    char *str;
367  };
368  struct status_struct *status_entry;
369  static struct status_struct status_array[] = {
370#define _(s,b) { .str = #s, .bit = b, },
371    foreach_virtio_config_status_flags
372#undef _
373    {.str = NULL}
374  };
375
376  vlib_cli_output (vm, "  status 0x%x", vif->status);
377
378  status_entry = (struct status_struct *) &status_array;
379  while (status_entry->str)
380    {
381      if (vif->status & status_entry->bit)
382	vlib_cli_output (vm, "    %s (%x)", status_entry->str,
383			 status_entry->bit);
384      status_entry++;
385    }
386}
387
388inline void
389debug_device_config_space (vlib_main_t * vm, virtio_if_t * vif)
390{
391  u32 data_u32;
392  u16 data_u16;
393  u8 data_u8;
394  vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_HOST_FEATURES,
395			&data_u32);
396  vlib_cli_output (vm, "remote features 0x%lx", data_u32);
397  vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_GUEST_FEATURES,
398			&data_u32);
399  vlib_cli_output (vm, "guest features 0x%lx", data_u32);
400  vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_PFN,
401			&data_u32);
402  vlib_cli_output (vm, "queue address 0x%lx", data_u32);
403  vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_NUM,
404			&data_u16);
405  vlib_cli_output (vm, "queue size 0x%x", data_u16);
406  vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_SEL,
407			&data_u16);
408  vlib_cli_output (vm, "queue select 0x%x", data_u16);
409  vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_NOTIFY,
410			&data_u16);
411  vlib_cli_output (vm, "queue notify 0x%x", data_u16);
412  vlib_pci_read_io_u8 (vm, vif->pci_dev_handle, VIRTIO_PCI_STATUS, &data_u8);
413  vlib_cli_output (vm, "status 0x%x", data_u8);
414  vlib_pci_read_io_u8 (vm, vif->pci_dev_handle, VIRTIO_PCI_ISR, &data_u8);
415  vlib_cli_output (vm, "isr 0x%x", data_u8);
416
417  if (vif->msix_enabled == VIRTIO_MSIX_ENABLED)
418    {
419      vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_MSI_CONFIG_VECTOR,
420			    &data_u16);
421      vlib_cli_output (vm, "config vector 0x%x", data_u16);
422      u16 queue_id = 0;
423      vlib_pci_write_io_u16 (vm, vif->pci_dev_handle, VIRTIO_PCI_QUEUE_SEL,
424			     &queue_id);
425      vlib_pci_read_io_u16 (vm, vif->pci_dev_handle, VIRTIO_MSI_QUEUE_VECTOR,
426			    &data_u16);
427      vlib_cli_output (vm, "queue vector for queue (0) 0x%x", data_u16);
428    }
429
430  u8 mac[6];
431  virtio_pci_legacy_read_config (vm, vif, mac, sizeof (mac), 0);
432  vlib_cli_output (vm, "mac %U", format_ethernet_address, mac);
433  virtio_pci_legacy_read_config (vm, vif, &data_u16, sizeof (u16),	/* offset to status */
434				 6);
435  vlib_cli_output (vm, "link up/down status 0x%x", data_u16);
436  virtio_pci_legacy_read_config (vm, vif, &data_u16, sizeof (u16),
437				 /* offset to max_virtqueue */ 8);
438  vlib_cli_output (vm, "num of virtqueue 0x%x", data_u16);
439  virtio_pci_legacy_read_config (vm, vif, &data_u16, sizeof (u16),	/* offset to mtu */
440				 10);
441  vlib_cli_output (vm, "mtu 0x%x", data_u16);
442
443  u32 i = PCI_CONFIG_SIZE (vif) + 12, a = 4;
444  i += a;
445  i &= ~a;
446  for (; i < 64; i += 4)
447    {
448      u32 data = 0;
449      vlib_pci_read_io_u32 (vm, vif->pci_dev_handle, i, &data);
450      vlib_cli_output (vm, "0x%lx", data);
451    }
452}
453
454struct virtio_ctrl_msg
455{
456  struct virtio_net_ctrl_hdr ctrl;
457  virtio_net_ctrl_ack status;
458  u8 data[1024];
459};
460
461static int
462virtio_pci_send_ctrl_msg (vlib_main_t * vm, virtio_if_t * vif,
463			  struct virtio_ctrl_msg *data, u32 len)
464{
465  virtio_vring_t *vring = vif->cxq_vring;
466  virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
467  struct virtio_ctrl_msg result;
468  u32 buffer_index;
469  vlib_buffer_t *b;
470  u16 used, next, avail;
471  u16 sz = vring->size;
472  u16 mask = sz - 1;
473
474  used = vring->desc_in_use;
475  next = vring->desc_next;
476  avail = vring->avail->idx;
477  struct vring_desc *d = &vring->desc[next];
478
479  if (vlib_buffer_alloc (vm, &buffer_index, 1))
480    b = vlib_get_buffer (vm, buffer_index);
481  else
482    return VIRTIO_NET_ERR;
483  /*
484   * current_data may not be initialized with 0 and may contain
485   * previous offset.
486   */
487  b->current_data = 0;
488  clib_memcpy (vlib_buffer_get_current (b), data,
489	       sizeof (struct virtio_ctrl_msg));
490  d->flags = VRING_DESC_F_NEXT;
491  d->addr = vlib_buffer_get_current_pa (vm, b);
492  d->len = sizeof (struct virtio_net_ctrl_hdr);
493  vring->avail->ring[avail & mask] = next;
494  avail++;
495  next = (next + 1) & mask;
496  d->next = next;
497  used++;
498
499  d = &vring->desc[next];
500  d->flags = VRING_DESC_F_NEXT;
501  d->addr = vlib_buffer_get_current_pa (vm, b) +
502    STRUCT_OFFSET_OF (struct virtio_ctrl_msg, data);
503  d->len = len;
504  next = (next + 1) & mask;
505  d->next = next;
506  used++;
507
508  d = &vring->desc[next];
509  d->flags = VRING_DESC_F_WRITE;
510  d->addr = vlib_buffer_get_current_pa (vm, b) +
511    STRUCT_OFFSET_OF (struct virtio_ctrl_msg, status);
512  d->len = sizeof (data->status);
513  next = (next + 1) & mask;
514  used++;
515
516  CLIB_MEMORY_STORE_BARRIER ();
517  vring->avail->idx = avail;
518  vring->desc_next = next;
519  vring->desc_in_use = used;
520
521  if ((vring->used->flags & VIRTIO_RING_FLAG_MASK_INT) == 0)
522    {
523      virtio_kick (vm, vring, vif);
524    }
525
526  u16 last = vring->last_used_idx, n_left = 0;
527  n_left = vring->used->idx - last;
528
529  while (n_left)
530    {
531      struct vring_used_elem *e = &vring->used->ring[last & mask];
532      u16 slot = e->id;
533
534      d = &vring->desc[slot];
535      while (d->flags & VRING_DESC_F_NEXT)
536	{
537	  used--;
538	  slot = d->next;
539	  d = &vring->desc[slot];
540	}
541      used--;
542      last++;
543      n_left--;
544    }
545  vring->desc_in_use = used;
546  vring->last_used_idx = last;
547
548  CLIB_MEMORY_BARRIER ();
549  clib_memcpy (&result, vlib_buffer_get_current (b),
550	       sizeof (struct virtio_ctrl_msg));
551  virtio_log_debug (vif, "ctrl-queue: status %u", result.status);
552  status = result.status;
553  vlib_buffer_free (vm, &buffer_index, 1);
554  return status;
555}
556
557static int
558virtio_pci_enable_gso (vlib_main_t * vm, virtio_if_t * vif)
559{
560  struct virtio_ctrl_msg gso_hdr;
561  virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
562
563  gso_hdr.ctrl.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS;
564  gso_hdr.ctrl.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET;
565  gso_hdr.status = VIRTIO_NET_ERR;
566  u64 offloads = VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_CSUM)
567    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_TSO4)
568    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_TSO6)
569    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_UFO);
570  clib_memcpy (gso_hdr.data, &offloads, sizeof (offloads));
571
572  status = virtio_pci_send_ctrl_msg (vm, vif, &gso_hdr, sizeof (offloads));
573  virtio_log_debug (vif, "enable gso");
574  return status;
575}
576
577static int
578virtio_pci_enable_multiqueue (vlib_main_t * vm, virtio_if_t * vif,
579			      u16 num_queues)
580{
581  struct virtio_ctrl_msg mq_hdr;
582  virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
583
584  mq_hdr.ctrl.class = VIRTIO_NET_CTRL_MQ;
585  mq_hdr.ctrl.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
586  mq_hdr.status = VIRTIO_NET_ERR;
587  clib_memcpy (mq_hdr.data, &num_queues, sizeof (num_queues));
588
589  status = virtio_pci_send_ctrl_msg (vm, vif, &mq_hdr, sizeof (num_queues));
590  virtio_log_debug (vif, "multi-queue enable %u queues", num_queues);
591  return status;
592}
593
594static u8
595virtio_pci_queue_size_valid (u16 qsz)
596{
597  if (qsz < 64 || qsz > 4096)
598    return 0;
599  if ((qsz % 64) != 0)
600    return 0;
601  return 1;
602}
603
604clib_error_t *
605virtio_pci_control_vring_init (vlib_main_t * vm, virtio_if_t * vif,
606			       u16 queue_num)
607{
608  clib_error_t *error = 0;
609  u16 queue_size = 0;
610  virtio_vring_t *vring;
611  struct vring vr;
612  u32 i = 0;
613  void *ptr = NULL;
614
615  queue_size = virtio_pci_legacy_get_queue_num (vm, vif, queue_num);
616  if (!virtio_pci_queue_size_valid (queue_size))
617    clib_warning ("queue size is not valid");
618
619  if (!is_pow2 (queue_size))
620    return clib_error_return (0, "ring size must be power of 2");
621
622  if (queue_size > 32768)
623    return clib_error_return (0, "ring size must be 32768 or lower");
624
625  if (queue_size == 0)
626    queue_size = 256;
627
628  vec_validate_aligned (vif->cxq_vring, 0, CLIB_CACHE_LINE_BYTES);
629  vring = vec_elt_at_index (vif->cxq_vring, 0);
630  i = vring_size (queue_size, VIRTIO_PCI_VRING_ALIGN);
631  i = round_pow2 (i, VIRTIO_PCI_VRING_ALIGN);
632  ptr =
633    vlib_physmem_alloc_aligned_on_numa (vm, i, VIRTIO_PCI_VRING_ALIGN,
634					vif->numa_node);
635  if (!ptr)
636    return vlib_physmem_last_error (vm);
637  clib_memset (ptr, 0, i);
638  vring_init (&vr, queue_size, ptr, VIRTIO_PCI_VRING_ALIGN);
639  vring->desc = vr.desc;
640  vring->avail = vr.avail;
641  vring->used = vr.used;
642  vring->queue_id = queue_num;
643  vring->avail->flags = VIRTIO_RING_FLAG_MASK_INT;
644
645  ASSERT (vring->buffers == 0);
646
647  vring->size = queue_size;
648  virtio_log_debug (vif, "control-queue: number %u, size %u", queue_num,
649		    queue_size);
650  virtio_pci_legacy_setup_queue (vm, vif, queue_num, ptr);
651  vring->kick_fd = -1;
652
653  return error;
654}
655
656clib_error_t *
657virtio_pci_vring_init (vlib_main_t * vm, virtio_if_t * vif, u16 queue_num)
658{
659  clib_error_t *error = 0;
660  vlib_thread_main_t *vtm = vlib_get_thread_main ();
661  u16 queue_size = 0;
662  virtio_vring_t *vring;
663  struct vring vr;
664  u32 i = 0;
665  void *ptr = NULL;
666
667  queue_size = virtio_pci_legacy_get_queue_num (vm, vif, queue_num);
668  if (!virtio_pci_queue_size_valid (queue_size))
669    clib_warning ("queue size is not valid");
670
671  if (!is_pow2 (queue_size))
672    return clib_error_return (0, "ring size must be power of 2");
673
674  if (queue_size > 32768)
675    return clib_error_return (0, "ring size must be 32768 or lower");
676
677  if (queue_size == 0)
678    queue_size = 256;
679
680  if (queue_num % 2)
681    {
682      if (TX_QUEUE_ACCESS (queue_num) > vtm->n_vlib_mains)
683	return error;
684      vec_validate_aligned (vif->txq_vrings, TX_QUEUE_ACCESS (queue_num),
685			    CLIB_CACHE_LINE_BYTES);
686      vring = vec_elt_at_index (vif->txq_vrings, TX_QUEUE_ACCESS (queue_num));
687      clib_spinlock_init (&vring->lockp);
688    }
689  else
690    {
691      vec_validate_aligned (vif->rxq_vrings, RX_QUEUE_ACCESS (queue_num),
692			    CLIB_CACHE_LINE_BYTES);
693      vring = vec_elt_at_index (vif->rxq_vrings, RX_QUEUE_ACCESS (queue_num));
694    }
695  i = vring_size (queue_size, VIRTIO_PCI_VRING_ALIGN);
696  i = round_pow2 (i, VIRTIO_PCI_VRING_ALIGN);
697  ptr =
698    vlib_physmem_alloc_aligned_on_numa (vm, i, VIRTIO_PCI_VRING_ALIGN,
699					vif->numa_node);
700  if (!ptr)
701    return vlib_physmem_last_error (vm);
702  clib_memset (ptr, 0, i);
703  vring_init (&vr, queue_size, ptr, VIRTIO_PCI_VRING_ALIGN);
704  vring->desc = vr.desc;
705  vring->avail = vr.avail;
706  vring->used = vr.used;
707  vring->queue_id = queue_num;
708  vring->avail->flags = VIRTIO_RING_FLAG_MASK_INT;
709
710  ASSERT (vring->buffers == 0);
711  vec_validate_aligned (vring->buffers, queue_size, CLIB_CACHE_LINE_BYTES);
712  if (queue_num % 2)
713    {
714      virtio_log_debug (vif, "tx-queue: number %u, size %u", queue_num,
715			queue_size);
716    }
717  else
718    {
719      virtio_log_debug (vif, "rx-queue: number %u, size %u", queue_num,
720			queue_size);
721    }
722  vring->size = queue_size;
723  if (virtio_pci_legacy_setup_queue (vm, vif, queue_num, ptr))
724    return clib_error_return (0, "error in queue address setup");
725
726  vring->kick_fd = -1;
727  return error;
728}
729
730static void
731virtio_negotiate_features (vlib_main_t * vm, virtio_if_t * vif,
732			   u64 req_features)
733{
734  /*
735   * if features are not requested
736   * default: all supported features
737   */
738  u64 supported_features = VIRTIO_FEATURE (VIRTIO_NET_F_CSUM)
739    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_CSUM)
740    | VIRTIO_FEATURE (VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
741    | VIRTIO_FEATURE (VIRTIO_NET_F_MTU)
742    | VIRTIO_FEATURE (VIRTIO_NET_F_MAC)
743    | VIRTIO_FEATURE (VIRTIO_NET_F_GSO)
744    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_TSO4)
745    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_TSO6)
746    | VIRTIO_FEATURE (VIRTIO_NET_F_GUEST_UFO)
747    | VIRTIO_FEATURE (VIRTIO_NET_F_HOST_TSO4)
748    | VIRTIO_FEATURE (VIRTIO_NET_F_HOST_TSO6)
749    | VIRTIO_FEATURE (VIRTIO_NET_F_HOST_UFO)
750    | VIRTIO_FEATURE (VIRTIO_NET_F_MRG_RXBUF)
751    | VIRTIO_FEATURE (VIRTIO_NET_F_STATUS)
752    | VIRTIO_FEATURE (VIRTIO_NET_F_CTRL_VQ)
753    | VIRTIO_FEATURE (VIRTIO_NET_F_MQ)
754    | VIRTIO_FEATURE (VIRTIO_F_NOTIFY_ON_EMPTY)
755    | VIRTIO_FEATURE (VIRTIO_F_ANY_LAYOUT)
756    | VIRTIO_FEATURE (VIRTIO_RING_F_INDIRECT_DESC);
757
758  if (req_features == 0)
759    {
760      req_features = supported_features;
761    }
762
763  vif->features = req_features & vif->remote_features & supported_features;
764
765  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_MTU))
766    {
767      virtio_net_config_t config;
768      virtio_pci_legacy_read_config (vm, vif, &config.mtu,
769				     sizeof (config.mtu),
770				     STRUCT_OFFSET_OF (virtio_net_config_t,
771						       mtu));
772      if (config.mtu < 64)
773	vif->features &= ~VIRTIO_FEATURE (VIRTIO_NET_F_MTU);
774    }
775
776  vif->features = virtio_pci_legacy_set_features (vm, vif, vif->features);
777}
778
779void
780virtio_pci_read_device_feature (vlib_main_t * vm, virtio_if_t * vif)
781{
782  vif->remote_features = virtio_pci_legacy_get_features (vm, vif);
783}
784
785int
786virtio_pci_reset_device (vlib_main_t * vm, virtio_if_t * vif)
787{
788  u8 status = 0;
789
790  /*
791   * Reset the device
792   */
793  status = virtio_pci_legacy_reset (vm, vif);
794
795  /*
796   * Set the Acknowledge status bit
797   */
798  virtio_pci_legacy_set_status (vm, vif, VIRTIO_CONFIG_STATUS_ACK);
799
800  /*
801   * Set the Driver status bit
802   */
803  virtio_pci_legacy_set_status (vm, vif, VIRTIO_CONFIG_STATUS_DRIVER);
804
805  /*
806   * Read the status and verify it
807   */
808  status = virtio_pci_legacy_get_status (vm, vif);
809  if (!
810      ((status & VIRTIO_CONFIG_STATUS_ACK)
811       && (status & VIRTIO_CONFIG_STATUS_DRIVER)))
812    return -1;
813  vif->status = status;
814
815  return 0;
816}
817
818clib_error_t *
819virtio_pci_read_caps (vlib_main_t * vm, virtio_if_t * vif)
820{
821  clib_error_t *error = 0;
822  struct virtio_pci_cap cap;
823  u8 pos, common_cfg = 0, notify_base = 0, dev_cfg = 0, isr = 0, pci_cfg = 0;
824  vlib_pci_dev_handle_t h = vif->pci_dev_handle;
825
826  if ((error = vlib_pci_read_config_u8 (vm, h, PCI_CAPABILITY_LIST, &pos)))
827    {
828      virtio_log_error (vif, "error in reading capabilty list position");
829      clib_error_return (error, "error in reading capabilty list position");
830    }
831  while (pos)
832    {
833      if ((error =
834	   vlib_pci_read_write_config (vm, h, VLIB_READ, pos, &cap,
835				       sizeof (cap))))
836	{
837	  virtio_log_error (vif, "%s [%2x]",
838			    "error in reading the capability at", pos);
839	  clib_error_return (error,
840			     "error in reading the capability at [%2x]", pos);
841	}
842
843      if (cap.cap_vndr == PCI_CAP_ID_MSIX)
844	{
845	  u16 flags, table_size, table_size_mask = 0x07FF;
846
847	  if ((error =
848	       vlib_pci_read_write_config (vm, h, VLIB_READ, pos + 2, &flags,
849					   sizeof (flags))))
850	    clib_error_return (error,
851			       "error in reading the capability at [%2x]",
852			       pos + 2);
853
854	  table_size = flags & table_size_mask;
855	  virtio_log_debug (vif, "flags:0x%x %s 0x%x", flags,
856			    "msix interrupt vector table-size", table_size);
857
858	  if (flags & PCI_MSIX_ENABLE)
859	    {
860	      virtio_log_debug (vif, "msix interrupt enabled");
861	      vif->msix_enabled = VIRTIO_MSIX_ENABLED;
862	    }
863	  else
864	    {
865	      virtio_log_debug (vif, "msix interrupt disabled");
866	      vif->msix_enabled = VIRTIO_MSIX_DISABLED;
867	    }
868	}
869
870      if (cap.cap_vndr != PCI_CAP_ID_VNDR)
871	{
872	  virtio_log_debug (vif, "[%2x] %s %2x ", pos,
873			    "skipping non VNDR cap id:", cap.cap_vndr);
874	  goto next;
875	}
876
877      virtio_log_debug (vif,
878			"[%4x] cfg type: %u, bar: %u, offset: %04x, len: %u",
879			pos, cap.cfg_type, cap.bar, cap.offset, cap.length);
880      switch (cap.cfg_type)
881	{
882	case VIRTIO_PCI_CAP_COMMON_CFG:
883	  common_cfg = 1;
884	  break;
885	case VIRTIO_PCI_CAP_NOTIFY_CFG:
886	  notify_base = 1;
887	  break;
888	case VIRTIO_PCI_CAP_DEVICE_CFG:
889	  dev_cfg = 1;
890	  break;
891	case VIRTIO_PCI_CAP_ISR_CFG:
892	  isr = 1;
893	  break;
894	case VIRTIO_PCI_CAP_PCI_CFG:
895	  if (cap.bar == 0)
896	    pci_cfg = 1;
897	  break;
898	}
899    next:
900      pos = cap.cap_next;
901    }
902
903  if (common_cfg == 0 || notify_base == 0 || dev_cfg == 0 || isr == 0)
904    {
905      virtio_log_debug (vif, "legacy virtio pci device found");
906      return error;
907    }
908
909  if (!pci_cfg)
910    clib_error_return (error, "modern virtio pci device found");
911
912  virtio_log_debug (vif, "transitional virtio pci device found");
913  return error;
914}
915
916static clib_error_t *
917virtio_pci_device_init (vlib_main_t * vm, virtio_if_t * vif,
918			virtio_pci_create_if_args_t * args)
919{
920  clib_error_t *error = 0;
921  u8 status = 0;
922
923  if ((error = virtio_pci_read_caps (vm, vif)))
924    clib_error_return (error, "Device is not supported");
925
926  if (virtio_pci_reset_device (vm, vif) < 0)
927    {
928      virtio_log_error (vif, "Failed to reset the device");
929      clib_error_return (error, "Failed to reset the device");
930    }
931  /*
932   * read device features and negotiate (user) requested features
933   */
934  virtio_pci_read_device_feature (vm, vif);
935  virtio_negotiate_features (vm, vif, args->features);
936
937  /*
938   * After FEATURE_OK, driver should not accept new feature bits
939   */
940  virtio_pci_legacy_set_status (vm, vif, VIRTIO_CONFIG_STATUS_FEATURES_OK);
941  status = virtio_pci_legacy_get_status (vm, vif);
942  if (!(status & VIRTIO_CONFIG_STATUS_FEATURES_OK))
943    {
944      virtio_log_error (vif,
945			"error encountered: Device doesn't support requested features");
946      clib_error_return (error, "Device doesn't support requested features");
947    }
948  vif->status = status;
949
950  /*
951   * get or set the mac address
952   */
953  if (virtio_pci_get_mac (vm, vif))
954    {
955      f64 now = vlib_time_now (vm);
956      u32 rnd;
957      rnd = (u32) (now * 1e6);
958      rnd = random_u32 (&rnd);
959
960      memcpy (vif->mac_addr + 2, &rnd, sizeof (rnd));
961      vif->mac_addr[0] = 2;
962      vif->mac_addr[1] = 0xfe;
963      virtio_pci_set_mac (vm, vif);
964    }
965
966  virtio_set_net_hdr_size (vif);
967
968  /*
969   * Initialize the virtqueues
970   */
971  if ((error = virtio_pci_get_max_virtqueue_pairs (vm, vif)))
972    goto err;
973
974  for (int i = 0; i < vif->max_queue_pairs; i++)
975    {
976      if ((error = virtio_pci_vring_init (vm, vif, RX_QUEUE (i))))
977	{
978	  virtio_log_warning (vif, "%s (%u) %s", "error in rxq-queue",
979			      RX_QUEUE (i), "initialization");
980	}
981      else
982	{
983	  vif->num_rxqs++;
984	}
985
986      if ((error = virtio_pci_vring_init (vm, vif, TX_QUEUE (i))))
987	{
988	  virtio_log_warning (vif, "%s (%u) %s", "error in txq-queue",
989			      TX_QUEUE (i), "initialization");
990	}
991      else
992	{
993	  vif->num_txqs++;
994	}
995    }
996
997  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_CTRL_VQ))
998    {
999      if ((error =
1000	   virtio_pci_control_vring_init (vm, vif, vif->max_queue_pairs * 2)))
1001	{
1002	  virtio_log_warning (vif, "%s (%u) %s", "error in control-queue",
1003			      vif->max_queue_pairs * 2, "initialization");
1004	  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_MQ))
1005	    vif->features &= ~VIRTIO_FEATURE (VIRTIO_NET_F_MQ);
1006	}
1007    }
1008  else
1009    {
1010      virtio_log_debug (vif, "control queue is not available");
1011      vif->cxq_vring = NULL;
1012    }
1013
1014  /*
1015   * set the msix interrupts
1016   */
1017  if (vif->msix_enabled == VIRTIO_MSIX_ENABLED)
1018    {
1019      if (virtio_pci_legacy_set_config_irq (vm, vif, 1) ==
1020	  VIRTIO_MSI_NO_VECTOR)
1021	virtio_log_warning (vif, "config vector 1 is not set");
1022      if (virtio_pci_legacy_set_queue_irq (vm, vif, 0, 0) ==
1023	  VIRTIO_MSI_NO_VECTOR)
1024	virtio_log_warning (vif, "queue vector 0 is not set");
1025    }
1026
1027  /*
1028   * set the driver status OK
1029   */
1030  virtio_pci_legacy_set_status (vm, vif, VIRTIO_CONFIG_STATUS_DRIVER_OK);
1031  vif->status = virtio_pci_legacy_get_status (vm, vif);
1032err:
1033  return error;
1034}
1035
1036void
1037virtio_pci_create_if (vlib_main_t * vm, virtio_pci_create_if_args_t * args)
1038{
1039  vnet_main_t *vnm = vnet_get_main ();
1040  virtio_main_t *vim = &virtio_main;
1041  virtio_if_t *vif;
1042  vlib_pci_dev_handle_t h;
1043  clib_error_t *error = 0;
1044
1045  /* *INDENT-OFF* */
1046  pool_foreach (vif, vim->interfaces, ({
1047    if (vif->pci_addr.as_u32 == args->addr)
1048      {
1049	args->rv = VNET_API_ERROR_INVALID_VALUE;
1050	args->error =
1051	  clib_error_return (error, "PCI address in use");
1052	  vlib_log (VLIB_LOG_LEVEL_ERR, vim->log_default, "%U: %s",
1053                format_vlib_pci_addr, &args->addr,
1054                " PCI address in use");
1055	return;
1056      }
1057  }));
1058  /* *INDENT-ON* */
1059
1060  pool_get (vim->interfaces, vif);
1061  vif->dev_instance = vif - vim->interfaces;
1062  vif->per_interface_next_index = ~0;
1063  vif->pci_addr.as_u32 = args->addr;
1064
1065  if ((error =
1066       vlib_pci_device_open (vm, (vlib_pci_addr_t *) & vif->pci_addr,
1067			     virtio_pci_device_ids, &h)))
1068    {
1069      pool_put (vim->interfaces, vif);
1070      args->rv = VNET_API_ERROR_INVALID_INTERFACE;
1071      args->error =
1072	clib_error_return (error, "pci-addr %U", format_vlib_pci_addr,
1073			   &vif->pci_addr);
1074      vlib_log (VLIB_LOG_LEVEL_ERR, vim->log_default, "%U: %s",
1075		format_vlib_pci_addr, &vif->pci_addr,
1076		"error encountered on pci device open");
1077      return;
1078    }
1079  vif->pci_dev_handle = h;
1080  vlib_pci_set_private_data (vm, h, vif->dev_instance);
1081  vif->numa_node = vlib_pci_get_numa_node (vm, h);
1082  vif->type = VIRTIO_IF_TYPE_PCI;
1083
1084  if ((error = vlib_pci_bus_master_enable (vm, h)))
1085    {
1086      virtio_log_error (vif, "error encountered on pci bus master enable");
1087      goto error;
1088    }
1089
1090  if ((error = vlib_pci_io_region (vm, h, 0)))
1091    {
1092      virtio_log_error (vif, "error encountered on pci io region");
1093      goto error;
1094    }
1095
1096  if (vlib_pci_get_num_msix_interrupts (vm, h) > 1)
1097    {
1098      if ((error = vlib_pci_register_msix_handler (vm, h, 0, 1,
1099						   &virtio_pci_irq_0_handler)))
1100	{
1101	  virtio_log_error (vif,
1102			    "error encountered on pci register msix handler 0");
1103	  goto error;
1104	}
1105      if ((error = vlib_pci_register_msix_handler (vm, h, 1, 1,
1106						   &virtio_pci_irq_1_handler)))
1107	{
1108	  virtio_log_error (vif,
1109			    "error encountered on pci register msix handler 1");
1110	  goto error;
1111	}
1112
1113      if ((error = vlib_pci_enable_msix_irq (vm, h, 0, 2)))
1114	{
1115	  virtio_log_error (vif, "error encountered on pci enable msix irq");
1116	  goto error;
1117	}
1118      vif->support_int_mode = 1;
1119      virtio_log_debug (vif, "device supports msix interrupts");
1120    }
1121  else if (vlib_pci_get_num_msix_interrupts (vm, h) == 1)
1122    {
1123      /*
1124       * if msix table-size is 1, fall back to intX.
1125       */
1126      if ((error =
1127	   vlib_pci_register_intx_handler (vm, h, &virtio_pci_irq_handler)))
1128	{
1129	  virtio_log_error (vif,
1130			    "error encountered on pci register interrupt handler");
1131	  goto error;
1132	}
1133      vif->support_int_mode = 1;
1134      virtio_log_debug (vif, "pci register interrupt handler");
1135    }
1136  else
1137    {
1138      /*
1139       * WARN: intX is showing some weird behaviour.
1140       * Please don't use interrupt mode with UIO driver.
1141       */
1142      vif->support_int_mode = 0;
1143      virtio_log_debug (vif, "driver is configured in poll mode only");
1144    }
1145
1146  if ((error = vlib_pci_intr_enable (vm, h)))
1147    {
1148      virtio_log_error (vif, "error encountered on pci interrupt enable");
1149      goto error;
1150    }
1151
1152  if ((error = virtio_pci_device_init (vm, vif, args)))
1153    {
1154      virtio_log_error (vif, "error encountered on device init");
1155      goto error;
1156    }
1157
1158  /* create interface */
1159  error = ethernet_register_interface (vnm, virtio_device_class.index,
1160				       vif->dev_instance, vif->mac_addr,
1161				       &vif->hw_if_index,
1162				       virtio_pci_flag_change);
1163
1164  if (error)
1165    {
1166      virtio_log_error (vif,
1167			"error encountered on ethernet register interface");
1168      goto error;
1169    }
1170
1171  vnet_sw_interface_t *sw = vnet_get_hw_sw_interface (vnm, vif->hw_if_index);
1172  vif->sw_if_index = sw->sw_if_index;
1173  args->sw_if_index = sw->sw_if_index;
1174
1175  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, vif->hw_if_index);
1176  hw->flags |= VNET_HW_INTERFACE_FLAG_SUPPORTS_INT_MODE;
1177  vnet_hw_interface_set_input_node (vnm, vif->hw_if_index,
1178				    virtio_input_node.index);
1179  u32 i = 0;
1180  vec_foreach_index (i, vif->rxq_vrings)
1181  {
1182    vnet_hw_interface_assign_rx_thread (vnm, vif->hw_if_index, i, ~0);
1183    virtio_vring_set_numa_node (vm, vif, RX_QUEUE (i));
1184    /* Set default rx mode to POLLING */
1185    vnet_hw_interface_set_rx_mode (vnm, vif->hw_if_index, i,
1186				   VNET_HW_INTERFACE_RX_MODE_POLLING);
1187  }
1188  if (virtio_pci_is_link_up (vm, vif) & VIRTIO_NET_S_LINK_UP)
1189    {
1190      vif->flags |= VIRTIO_IF_FLAG_ADMIN_UP;
1191      vnet_hw_interface_set_flags (vnm, vif->hw_if_index,
1192				   VNET_HW_INTERFACE_FLAG_LINK_UP);
1193    }
1194  else
1195    vnet_hw_interface_set_flags (vnm, vif->hw_if_index, 0);
1196
1197  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_CTRL_VQ))
1198    {
1199      if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) &&
1200	  args->gso_enabled)
1201	{
1202	  if (virtio_pci_enable_gso (vm, vif))
1203	    {
1204	      virtio_log_warning (vif, "gso is not enabled");
1205	    }
1206	  else
1207	    {
1208	      vif->gso_enabled = 1;
1209	      hw->flags |= VNET_HW_INTERFACE_FLAG_SUPPORTS_GSO;
1210	      vnm->interface_main.gso_interface_count++;
1211	    }
1212	}
1213      if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_MQ))
1214	{
1215	  if (virtio_pci_enable_multiqueue (vm, vif, vif->max_queue_pairs))
1216	    virtio_log_warning (vif, "multiqueue is not set");
1217	}
1218    }
1219  return;
1220
1221
1222error:
1223  virtio_pci_delete_if (vm, vif);
1224  args->rv = VNET_API_ERROR_INVALID_INTERFACE;
1225  args->error = error;
1226}
1227
1228int
1229virtio_pci_delete_if (vlib_main_t * vm, virtio_if_t * vif)
1230{
1231  vnet_main_t *vnm = vnet_get_main ();
1232  virtio_main_t *vim = &virtio_main;
1233  u32 i = 0;
1234
1235  if (vif->type != VIRTIO_IF_TYPE_PCI)
1236    return VNET_API_ERROR_INVALID_INTERFACE;
1237
1238  vlib_pci_intr_disable (vm, vif->pci_dev_handle);
1239
1240  for (i = 0; i < vif->max_queue_pairs; i++)
1241    {
1242      virtio_pci_legacy_del_queue (vm, vif, RX_QUEUE (i));
1243      virtio_pci_legacy_del_queue (vm, vif, TX_QUEUE (i));
1244    }
1245
1246  if (vif->features & VIRTIO_FEATURE (VIRTIO_NET_F_CTRL_VQ))
1247    virtio_pci_legacy_del_queue (vm, vif, vif->max_queue_pairs * 2);
1248
1249  virtio_pci_legacy_reset (vm, vif);
1250
1251  if (vif->gso_enabled)
1252    vnm->interface_main.gso_interface_count--;
1253
1254  if (vif->hw_if_index)
1255    {
1256      vnet_hw_interface_set_flags (vnm, vif->hw_if_index, 0);
1257      vec_foreach_index (i, vif->rxq_vrings)
1258      {
1259	vnet_hw_interface_unassign_rx_thread (vnm, vif->hw_if_index, i);
1260      }
1261      ethernet_delete_interface (vnm, vif->hw_if_index);
1262    }
1263
1264  vlib_pci_device_close (vm, vif->pci_dev_handle);
1265
1266  vec_foreach_index (i, vif->rxq_vrings)
1267  {
1268    virtio_vring_t *vring = vec_elt_at_index (vif->rxq_vrings, i);
1269    if (vring->kick_fd != -1)
1270      close (vring->kick_fd);
1271    if (vring->used)
1272      {
1273	virtio_free_rx_buffers (vm, vring);
1274      }
1275    vec_free (vring->buffers);
1276    vlib_physmem_free (vm, vring->desc);
1277  }
1278
1279  vec_foreach_index (i, vif->txq_vrings)
1280  {
1281    virtio_vring_t *vring = vec_elt_at_index (vif->txq_vrings, i);
1282    if (vring->kick_fd != -1)
1283      close (vring->kick_fd);
1284    if (vring->used)
1285      {
1286	virtio_free_used_desc (vm, vring);
1287      }
1288    vec_free (vring->buffers);
1289    clib_spinlock_free (&vring->lockp);
1290    vlib_physmem_free (vm, vring->desc);
1291  }
1292
1293  if (vif->cxq_vring != NULL)
1294    {
1295      u16 last = vif->cxq_vring->last_used_idx;
1296      u16 n_left = vif->cxq_vring->used->idx - last;
1297      while (n_left)
1298	{
1299	  last++;
1300	  n_left--;
1301	}
1302
1303      vif->cxq_vring->last_used_idx = last;
1304      vlib_physmem_free (vm, vif->cxq_vring->desc);
1305    }
1306
1307  vec_free (vif->rxq_vrings);
1308  vec_free (vif->txq_vrings);
1309  vec_free (vif->cxq_vring);
1310
1311  clib_error_free (vif->error);
1312  memset (vif, 0, sizeof (*vif));
1313  pool_put (vim->interfaces, vif);
1314
1315  return 0;
1316}
1317
1318/*
1319 * fd.io coding-style-patch-verification: ON
1320 *
1321 * Local Variables:
1322 * eval: (c-set-style "gnu")
1323 * End:
1324 */
1325