1 /*-
2 * BSD LICENSE
3 *
4 * Copyright (c) 2015-2017 Amazon.com, Inc. or its affiliates.
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 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 *
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
23 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
24 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 *
30 * $FreeBSD: head/sys/dev/ena/ena.c 325593 2017-11-09 13:38:17Z mw $
31 */
32
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bus.h>
36 #include <sys/endian.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/rman.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/sysctl.h>
46 #include <sys/taskqueue.h>
47 #include <sys/time.h>
48 #include <sys/eventhandler.h>
49
50 #include <net/bpf.h>
51 #include <net/if.h>
52 #include <net/if_var.h>
53 #include <net/if_arp.h>
54 #include <net/if_dl.h>
55 #include <net/if_media.h>
56 #include <net/if_types.h>
57 #include <net/ifq_var.h>
58 #include <net/vlan/if_vlan_var.h>
59
60 #include <netinet/in_systm.h>
61 #include <netinet/in.h>
62 #include <netinet/if_ether.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip6.h>
65 #include <netinet/tcp.h>
66 #include <netinet/udp.h>
67
68 #include <bus/pci/pcivar.h>
69 #include <bus/pci/pcireg.h>
70
71 #include "ena.h"
72 #include "ena_sysctl.h"
73
74 /*********************************************************
75 * Function prototypes
76 *********************************************************/
77 static int ena_probe(device_t);
78 static void ena_intr_msix_mgmnt(void *);
79 static int ena_allocate_pci_resources(struct ena_adapter*);
80 static void ena_free_pci_resources(struct ena_adapter *);
81 static int ena_change_mtu(if_t, int);
82 #if 0 /* XXX swildner counters */
83 static inline void ena_alloc_counters(counter_u64_t *, int);
84 static inline void ena_free_counters(counter_u64_t *, int);
85 static inline void ena_reset_counters(counter_u64_t *, int);
86 #endif
87 static void ena_init_io_rings_common(struct ena_adapter *,
88 struct ena_ring *, uint16_t);
89 static void ena_init_io_rings(struct ena_adapter *);
90 static void ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
91 static void ena_free_all_io_rings_resources(struct ena_adapter *);
92 static int ena_setup_tx_dma_tag(struct ena_adapter *);
93 static int ena_free_tx_dma_tag(struct ena_adapter *);
94 static int ena_setup_rx_dma_tag(struct ena_adapter *);
95 static int ena_free_rx_dma_tag(struct ena_adapter *);
96 static int ena_setup_tx_resources(struct ena_adapter *, int);
97 static void ena_free_tx_resources(struct ena_adapter *, int);
98 static int ena_setup_all_tx_resources(struct ena_adapter *);
99 static void ena_free_all_tx_resources(struct ena_adapter *);
100 static inline int validate_rx_req_id(struct ena_ring *, uint16_t);
101 static int ena_setup_rx_resources(struct ena_adapter *, unsigned int);
102 static void ena_free_rx_resources(struct ena_adapter *, unsigned int);
103 static int ena_setup_all_rx_resources(struct ena_adapter *);
104 static void ena_free_all_rx_resources(struct ena_adapter *);
105 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
106 struct ena_rx_buffer *);
107 static void ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
108 struct ena_rx_buffer *);
109 static int ena_refill_rx_bufs(struct ena_ring *, uint32_t);
110 static void ena_free_rx_bufs(struct ena_adapter *, unsigned int);
111 static void ena_refill_all_rx_bufs(struct ena_adapter *);
112 static void ena_free_all_rx_bufs(struct ena_adapter *);
113 static void ena_free_tx_bufs(struct ena_adapter *, unsigned int);
114 static void ena_free_all_tx_bufs(struct ena_adapter *);
115 static void ena_destroy_all_tx_queues(struct ena_adapter *);
116 static void ena_destroy_all_rx_queues(struct ena_adapter *);
117 static void ena_destroy_all_io_queues(struct ena_adapter *);
118 static int ena_create_io_queues(struct ena_adapter *);
119 static int ena_tx_cleanup(struct ena_ring *);
120 static void ena_deferred_rx_cleanup(void *, int);
121 static int ena_rx_cleanup(struct ena_ring *);
122 static inline int validate_tx_req_id(struct ena_ring *, uint16_t);
123 static void ena_rx_hash_mbuf(struct ena_ring *, struct ena_com_rx_ctx *,
124 struct mbuf *);
125 static struct mbuf* ena_rx_mbuf(struct ena_ring *, struct ena_com_rx_buf_info *,
126 struct ena_com_rx_ctx *, uint16_t *);
127 static inline void ena_rx_checksum(struct ena_ring *, struct ena_com_rx_ctx *,
128 struct mbuf *);
129 static void ena_handle_msix(void *);
130 static int ena_enable_msix(struct ena_adapter *);
131 static void ena_setup_mgmnt_intr(struct ena_adapter *);
132 static void ena_setup_io_intr(struct ena_adapter *);
133 static int ena_request_mgmnt_irq(struct ena_adapter *);
134 static int ena_request_io_irq(struct ena_adapter *);
135 static void ena_free_mgmnt_irq(struct ena_adapter *);
136 static void ena_free_io_irq(struct ena_adapter *);
137 static void ena_free_irqs(struct ena_adapter*);
138 static void ena_disable_msix(struct ena_adapter *);
139 static void ena_unmask_all_io_irqs(struct ena_adapter *);
140 static int ena_rss_configure(struct ena_adapter *);
141 static int ena_up_complete(struct ena_adapter *);
142 static int ena_up(struct ena_adapter *);
143 static void ena_down(struct ena_adapter *);
144 #if 0 /* XXX swildner counters */
145 static uint64_t ena_get_counter(if_t, ift_counter);
146 #endif
147 static int ena_media_change(if_t);
148 static void ena_media_status(if_t, struct ifmediareq *);
149 static void ena_init(void *);
150 static int ena_ioctl(if_t, u_long, caddr_t, struct ucred *);
151 static int ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
152 static void ena_update_host_info(struct ena_admin_host_info *, if_t);
153 static void ena_update_hwassist(struct ena_adapter *);
154 static int ena_setup_ifnet(device_t, struct ena_adapter *,
155 struct ena_com_dev_get_features_ctx *);
156 static void ena_tx_csum(struct ena_com_tx_ctx *, struct mbuf *);
157 static int ena_check_and_collapse_mbuf(struct ena_ring *tx_ring,
158 struct mbuf **mbuf);
159 static int ena_xmit_mbuf(struct ena_ring *, struct mbuf **);
160 static void ena_start_xmit(struct ifnet *, struct ifaltq_subque *);
161 static int ena_calc_io_queue_num(struct ena_adapter *,
162 struct ena_com_dev_get_features_ctx *);
163 static int ena_calc_queue_size(struct ena_adapter *, uint16_t *,
164 uint16_t *, struct ena_com_dev_get_features_ctx *);
165 static int ena_rss_init_default(struct ena_adapter *);
166 static void ena_rss_init_default_deferred(void *);
167 static void ena_config_host_info(struct ena_com_dev *);
168 static int ena_attach(device_t);
169 static int ena_detach(device_t);
170 static int ena_device_init(struct ena_adapter *, device_t,
171 struct ena_com_dev_get_features_ctx *, int *);
172 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *,
173 int);
174 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
175 static void unimplemented_aenq_handler(void *,
176 struct ena_admin_aenq_entry *);
177 static void ena_timer_service(void *);
178
179 static char ena_version[] = DEVICE_NAME DRV_MODULE_NAME " v" DRV_MODULE_VERSION;
180
181 static SYSCTL_NODE(_hw, OID_AUTO, ena, CTLFLAG_RD, 0, "ENA driver parameters");
182
183 /*
184 * Logging level for changing verbosity of the output
185 */
186 int ena_log_level = ENA_ALERT | ENA_WARNING;
187 TUNABLE_INT("hw.ena.ena_log_level", &ena_log_level);
188 SYSCTL_INT(_hw_ena, OID_AUTO, log_level, CTLFLAG_RW,
189 &ena_log_level, 0, "Logging level indicating verbosity of the logs");
190
191 static ena_vendor_info_t ena_vendor_info_array[] = {
192 { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
193 { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_PF, 0},
194 { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
195 { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_VF, 0},
196 /* Last entry */
197 { 0, 0, 0 }
198 };
199
200 /*
201 * Contains pointers to event handlers, e.g. link state chage.
202 */
203 static struct ena_aenq_handlers aenq_handlers;
204
205 void
ena_dmamap_callback(void * arg,bus_dma_segment_t * segs,int nseg,int error)206 ena_dmamap_callback(void *arg, bus_dma_segment_t *segs, int nseg, int error)
207 {
208 if (error != 0)
209 return;
210 *(bus_addr_t *) arg = segs[0].ds_addr;
211 }
212
213 int
ena_dma_alloc(device_t dmadev,bus_size_t size,ena_mem_handle_t * dma,int mapflags)214 ena_dma_alloc(device_t dmadev, bus_size_t size,
215 ena_mem_handle_t *dma , int mapflags)
216 {
217 struct ena_adapter* adapter = device_get_softc(dmadev);
218 uint32_t maxsize;
219 uint64_t dma_space_addr;
220 int error;
221
222 maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
223
224 dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
225 if (unlikely(dma_space_addr == 0))
226 dma_space_addr = BUS_SPACE_MAXADDR;
227
228 error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
229 8, 0, /* alignment, bounds */
230 dma_space_addr, /* lowaddr of exclusion window */
231 BUS_SPACE_MAXADDR,/* highaddr of exclusion window */
232 maxsize, /* maxsize */
233 1, /* nsegments */
234 maxsize, /* maxsegsize */
235 BUS_DMA_ALLOCNOW, /* flags */
236 &dma->tag);
237 if (unlikely(error != 0)) {
238 ena_trace(ENA_ALERT, "bus_dma_tag_create failed: %d\n", error);
239 goto fail_tag;
240 }
241
242 error = bus_dmamem_alloc(dma->tag, (void**) &dma->vaddr,
243 BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
244 if (unlikely(error != 0)) {
245 ena_trace(ENA_ALERT, "bus_dmamem_alloc(%ju) failed: %d\n",
246 (uintmax_t)size, error);
247 goto fail_map_create;
248 }
249
250 dma->paddr = 0;
251 error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
252 size, ena_dmamap_callback, &dma->paddr, mapflags);
253 if (unlikely((error != 0) || (dma->paddr == 0))) {
254 ena_trace(ENA_ALERT, ": bus_dmamap_load failed: %d\n", error);
255 goto fail_map_load;
256 }
257
258 return (0);
259
260 fail_map_load:
261 bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
262 fail_map_create:
263 bus_dma_tag_destroy(dma->tag);
264 fail_tag:
265 dma->tag = NULL;
266
267 return (error);
268 }
269
270 static int
ena_allocate_pci_resources(struct ena_adapter * adapter)271 ena_allocate_pci_resources(struct ena_adapter* adapter)
272 {
273 device_t pdev = adapter->pdev;
274 int rid;
275
276 rid = PCIR_BAR(ENA_REG_BAR);
277 adapter->memory = NULL;
278 adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
279 &rid, RF_ACTIVE);
280 if (unlikely(adapter->registers == NULL)) {
281 device_printf(pdev, "Unable to allocate bus resource: "
282 "registers\n");
283 return (ENXIO);
284 }
285
286 return (0);
287 }
288
289 static void
ena_free_pci_resources(struct ena_adapter * adapter)290 ena_free_pci_resources(struct ena_adapter *adapter)
291 {
292 device_t pdev = adapter->pdev;
293
294 if (adapter->memory != NULL) {
295 bus_release_resource(pdev, SYS_RES_MEMORY,
296 PCIR_BAR(ENA_MEM_BAR), adapter->memory);
297 }
298
299 if (adapter->registers != NULL) {
300 bus_release_resource(pdev, SYS_RES_MEMORY,
301 PCIR_BAR(ENA_REG_BAR), adapter->registers);
302 }
303 }
304
305 static int
ena_probe(device_t dev)306 ena_probe(device_t dev)
307 {
308 ena_vendor_info_t *ent;
309 char adapter_name[60];
310 uint16_t pci_vendor_id = 0;
311 uint16_t pci_device_id = 0;
312
313 pci_vendor_id = pci_get_vendor(dev);
314 pci_device_id = pci_get_device(dev);
315
316 ent = ena_vendor_info_array;
317 while (ent->vendor_id != 0) {
318 if ((pci_vendor_id == ent->vendor_id) &&
319 (pci_device_id == ent->device_id)) {
320 ena_trace(ENA_DBG, "vendor=%x device=%x ",
321 pci_vendor_id, pci_device_id);
322
323 ksprintf(adapter_name, DEVICE_DESC);
324 device_set_desc_copy(dev, adapter_name);
325 return (BUS_PROBE_DEFAULT);
326 }
327
328 ent++;
329
330 }
331
332 return (ENXIO);
333 }
334
335 static int
ena_change_mtu(if_t ifp,int new_mtu)336 ena_change_mtu(if_t ifp, int new_mtu)
337 {
338 struct ena_adapter *adapter = ifp->if_softc;
339 int rc;
340
341 if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
342 device_printf(adapter->pdev, "Invalid MTU setting. "
343 "new_mtu: %d max mtu: %d min mtu: %d\n",
344 new_mtu, adapter->max_mtu, ENA_MIN_MTU);
345 return (EINVAL);
346 }
347
348 rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
349 if (likely(rc == 0)) {
350 ena_trace(ENA_DBG, "set MTU to %d\n", new_mtu);
351 ifp->if_mtu = new_mtu;
352 } else {
353 device_printf(adapter->pdev, "Failed to set MTU to %d\n",
354 new_mtu);
355 }
356
357 return (rc);
358 }
359
360 #if 0 /* XXX swildner counters */
361 static inline void
362 ena_alloc_counters(counter_u64_t *begin, int size)
363 {
364 counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
365
366 for (; begin < end; ++begin)
367 *begin = counter_u64_alloc(M_WAITOK);
368 }
369
370 static inline void
371 ena_free_counters(counter_u64_t *begin, int size)
372 {
373 counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
374
375 for (; begin < end; ++begin)
376 counter_u64_free(*begin);
377 }
378
379 static inline void
380 ena_reset_counters(counter_u64_t *begin, int size)
381 {
382 counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
383
384 for (; begin < end; ++begin)
385 counter_u64_zero(*begin);
386 }
387 #endif
388
389 static void
ena_init_io_rings_common(struct ena_adapter * adapter,struct ena_ring * ring,uint16_t qid)390 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
391 uint16_t qid)
392 {
393
394 ring->qid = qid;
395 ring->adapter = adapter;
396 ring->ena_dev = adapter->ena_dev;
397 }
398
399 static void
ena_init_io_rings(struct ena_adapter * adapter)400 ena_init_io_rings(struct ena_adapter *adapter)
401 {
402 struct ena_com_dev *ena_dev;
403 struct ena_ring *txr, *rxr;
404 struct ena_que *que;
405 int i;
406
407 ena_dev = adapter->ena_dev;
408
409 for (i = 0; i < adapter->num_queues; i++) {
410 txr = &adapter->tx_ring[i];
411 rxr = &adapter->rx_ring[i];
412
413 /* TX/RX common ring state */
414 ena_init_io_rings_common(adapter, txr, i);
415 ena_init_io_rings_common(adapter, rxr, i);
416
417 /* TX specific ring state */
418 txr->ring_size = adapter->tx_ring_size;
419 txr->tx_max_header_size = ena_dev->tx_max_header_size;
420 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
421 txr->smoothed_interval =
422 ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
423
424 #if 0 /* XXX swildner counters */
425 /* Alloc TX statistics. */
426 ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
427 sizeof(txr->tx_stats));
428 #endif
429
430 /* RX specific ring state */
431 rxr->ring_size = adapter->rx_ring_size;
432 rxr->smoothed_interval =
433 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
434
435 #if 0 /* XXX swildner counters */
436 /* Alloc RX statistics. */
437 ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
438 sizeof(rxr->rx_stats));
439 #endif
440
441 /* Initialize locks */
442 ksnprintf(txr->lock_name, nitems(txr->lock_name), "%s:tx(%d)",
443 device_get_nameunit(adapter->pdev), i);
444 ksnprintf(rxr->lock_name, nitems(rxr->lock_name), "%s:rx(%d)",
445 device_get_nameunit(adapter->pdev), i);
446
447 lockinit(&txr->ring_lock, txr->lock_name, 0, LK_CANRECURSE);
448 lockinit(&rxr->ring_lock, rxr->lock_name, 0, LK_CANRECURSE);
449
450 que = &adapter->que[i];
451 que->adapter = adapter;
452 que->id = i;
453 que->tx_ring = txr;
454 que->rx_ring = rxr;
455
456 txr->que = que;
457 rxr->que = que;
458
459 rxr->empty_rx_queue = 0;
460 }
461 }
462
463 static void
ena_free_io_ring_resources(struct ena_adapter * adapter,unsigned int qid)464 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
465 {
466 struct ena_ring *txr = &adapter->tx_ring[qid];
467 struct ena_ring *rxr = &adapter->rx_ring[qid];
468
469 #if 0 /* XXX swildner counters */
470 ena_free_counters((counter_u64_t *)&txr->tx_stats,
471 sizeof(txr->tx_stats));
472 ena_free_counters((counter_u64_t *)&rxr->rx_stats,
473 sizeof(rxr->rx_stats));
474 #endif
475
476 lockuninit(&txr->ring_lock);
477 lockuninit(&rxr->ring_lock);
478 }
479
480 static void
ena_free_all_io_rings_resources(struct ena_adapter * adapter)481 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
482 {
483 int i;
484
485 for (i = 0; i < adapter->num_queues; i++)
486 ena_free_io_ring_resources(adapter, i);
487
488 }
489
490 static int
ena_setup_tx_dma_tag(struct ena_adapter * adapter)491 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
492 {
493 int ret;
494
495 /* Create DMA tag for Tx buffers */
496 ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
497 1, 0, /* alignment, bounds */
498 ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window */
499 BUS_SPACE_MAXADDR, /* highaddr of excl window */
500 ENA_TSO_MAXSIZE, /* maxsize */
501 ENA_BUS_DMA_SEGS, /* nsegments */
502 ENA_TSO_MAXSIZE, /* maxsegsize */
503 0, /* flags */
504 &adapter->tx_buf_tag);
505
506 return (ret);
507 }
508
509 static int
ena_free_tx_dma_tag(struct ena_adapter * adapter)510 ena_free_tx_dma_tag(struct ena_adapter *adapter)
511 {
512 int ret;
513
514 ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
515
516 if (likely(ret == 0))
517 adapter->tx_buf_tag = NULL;
518
519 return (ret);
520 }
521
522 static int
ena_setup_rx_dma_tag(struct ena_adapter * adapter)523 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
524 {
525 int ret;
526
527 /* Create DMA tag for Rx buffers*/
528 ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent */
529 1, 0, /* alignment, bounds */
530 ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window */
531 BUS_SPACE_MAXADDR, /* highaddr of excl window */
532 MJUM16BYTES, /* maxsize */
533 adapter->max_rx_sgl_size, /* nsegments */
534 MJUM16BYTES, /* maxsegsize */
535 0, /* flags */
536 &adapter->rx_buf_tag);
537
538 return (ret);
539 }
540
541 static int
ena_free_rx_dma_tag(struct ena_adapter * adapter)542 ena_free_rx_dma_tag(struct ena_adapter *adapter)
543 {
544 int ret;
545
546 ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
547
548 if (likely(ret == 0))
549 adapter->rx_buf_tag = NULL;
550
551 return (ret);
552 }
553
554 /**
555 * ena_setup_tx_resources - allocate Tx resources (Descriptors)
556 * @adapter: network interface device structure
557 * @qid: queue index
558 *
559 * Returns 0 on success, otherwise on failure.
560 **/
561 static int
ena_setup_tx_resources(struct ena_adapter * adapter,int qid)562 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
563 {
564 struct ena_que *que = &adapter->que[qid];
565 struct ena_ring *tx_ring = que->tx_ring;
566 int size, i, err;
567 #ifdef RSS
568 cpuset_t cpu_mask;
569 #endif
570
571 size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
572
573 tx_ring->tx_buffer_info = kmalloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
574 if (unlikely(tx_ring->tx_buffer_info == NULL))
575 return (ENOMEM);
576
577 size = sizeof(uint16_t) * tx_ring->ring_size;
578 tx_ring->free_tx_ids = kmalloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
579 if (unlikely(tx_ring->free_tx_ids == NULL))
580 goto err_buf_info_free;
581
582 /* Req id stack for TX OOO completions */
583 for (i = 0; i < tx_ring->ring_size; i++)
584 tx_ring->free_tx_ids[i] = i;
585
586 #if 0 /* XXX swildner counters */
587 /* Reset TX statistics. */
588 ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
589 sizeof(tx_ring->tx_stats));
590 #endif
591
592 tx_ring->next_to_use = 0;
593 tx_ring->next_to_clean = 0;
594
595 /* ... and create the buffer DMA maps */
596 for (i = 0; i < tx_ring->ring_size; i++) {
597 err = bus_dmamap_create(adapter->tx_buf_tag, 0,
598 &tx_ring->tx_buffer_info[i].map);
599 if (unlikely(err != 0)) {
600 ena_trace(ENA_ALERT,
601 "Unable to create Tx DMA map for buffer %d\n", i);
602 goto err_buf_info_unmap;
603 }
604 }
605
606 return (0);
607
608 err_buf_info_unmap:
609 while (i--) {
610 bus_dmamap_destroy(adapter->tx_buf_tag,
611 tx_ring->tx_buffer_info[i].map);
612 }
613 kfree(tx_ring->free_tx_ids, M_DEVBUF);
614 tx_ring->free_tx_ids = NULL;
615 err_buf_info_free:
616 kfree(tx_ring->tx_buffer_info, M_DEVBUF);
617 tx_ring->tx_buffer_info = NULL;
618
619 return (ENOMEM);
620 }
621
622 /**
623 * ena_free_tx_resources - Free Tx Resources per Queue
624 * @adapter: network interface device structure
625 * @qid: queue index
626 *
627 * Free all transmit software resources
628 **/
629 static void
ena_free_tx_resources(struct ena_adapter * adapter,int qid)630 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
631 {
632 struct ena_ring *tx_ring = &adapter->tx_ring[qid];
633
634 ENA_RING_MTX_LOCK(tx_ring);
635
636 /* Free buffer DMA maps, */
637 for (int i = 0; i < tx_ring->ring_size; i++) {
638 m_freem(tx_ring->tx_buffer_info[i].mbuf);
639 tx_ring->tx_buffer_info[i].mbuf = NULL;
640 bus_dmamap_unload(adapter->tx_buf_tag,
641 tx_ring->tx_buffer_info[i].map);
642 bus_dmamap_destroy(adapter->tx_buf_tag,
643 tx_ring->tx_buffer_info[i].map);
644 }
645 ENA_RING_MTX_UNLOCK(tx_ring);
646
647 /* And free allocated memory. */
648 kfree(tx_ring->tx_buffer_info, M_DEVBUF);
649 tx_ring->tx_buffer_info = NULL;
650
651 kfree(tx_ring->free_tx_ids, M_DEVBUF);
652 tx_ring->free_tx_ids = NULL;
653 }
654
655 /**
656 * ena_setup_all_tx_resources - allocate all queues Tx resources
657 * @adapter: network interface device structure
658 *
659 * Returns 0 on success, otherwise on failure.
660 **/
661 static int
ena_setup_all_tx_resources(struct ena_adapter * adapter)662 ena_setup_all_tx_resources(struct ena_adapter *adapter)
663 {
664 int i, rc;
665
666 for (i = 0; i < adapter->num_queues; i++) {
667 rc = ena_setup_tx_resources(adapter, i);
668 if (rc != 0) {
669 device_printf(adapter->pdev,
670 "Allocation for Tx Queue %u failed\n", i);
671 goto err_setup_tx;
672 }
673 }
674
675 return (0);
676
677 err_setup_tx:
678 /* Rewind the index freeing the rings as we go */
679 while (i--)
680 ena_free_tx_resources(adapter, i);
681 return (rc);
682 }
683
684 /**
685 * ena_free_all_tx_resources - Free Tx Resources for All Queues
686 * @adapter: network interface device structure
687 *
688 * Free all transmit software resources
689 **/
690 static void
ena_free_all_tx_resources(struct ena_adapter * adapter)691 ena_free_all_tx_resources(struct ena_adapter *adapter)
692 {
693 int i;
694
695 for (i = 0; i < adapter->num_queues; i++)
696 ena_free_tx_resources(adapter, i);
697 }
698
699 static inline int
validate_rx_req_id(struct ena_ring * rx_ring,uint16_t req_id)700 validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
701 {
702 if (likely(req_id < rx_ring->ring_size))
703 return (0);
704
705 device_printf(rx_ring->adapter->pdev, "Invalid rx req_id: %hu\n",
706 req_id);
707 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
708 #if 0 /* XXX swildner counters */
709 counter_u64_add(rx_ring->rx_stats.bad_req_id, 1);
710 #endif
711
712 /* Trigger device reset */
713 rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
714 rx_ring->adapter->trigger_reset = true;
715
716 return (EFAULT);
717 }
718
719 /**
720 * ena_setup_rx_resources - allocate Rx resources (Descriptors)
721 * @adapter: network interface device structure
722 * @qid: queue index
723 *
724 * Returns 0 on success, otherwise on failure.
725 **/
726 static int
ena_setup_rx_resources(struct ena_adapter * adapter,unsigned int qid)727 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
728 {
729 struct ena_que *que = &adapter->que[qid];
730 struct ena_ring *rx_ring = que->rx_ring;
731 int size, err, i;
732 #ifdef RSS
733 cpuset_t cpu_mask;
734 #endif
735
736 size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
737
738 /*
739 * Alloc extra element so in rx path
740 * we can always prefetch rx_info + 1
741 */
742 size += sizeof(struct ena_rx_buffer);
743
744 rx_ring->rx_buffer_info = kmalloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
745
746 size = sizeof(uint16_t) * rx_ring->ring_size;
747 rx_ring->free_rx_ids = kmalloc(size, M_DEVBUF, M_WAITOK);
748
749 for (i = 0; i < rx_ring->ring_size; i++)
750 rx_ring->free_rx_ids[i] = i;
751
752 #if 0 /* XXX swildner counters */
753 /* Reset RX statistics. */
754 ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
755 sizeof(rx_ring->rx_stats));
756 #endif
757
758 rx_ring->next_to_clean = 0;
759 rx_ring->next_to_use = 0;
760
761 /* ... and create the buffer DMA maps */
762 for (i = 0; i < rx_ring->ring_size; i++) {
763 err = bus_dmamap_create(adapter->rx_buf_tag, 0,
764 &(rx_ring->rx_buffer_info[i].map));
765 if (err != 0) {
766 ena_trace(ENA_ALERT,
767 "Unable to create Rx DMA map for buffer %d\n", i);
768 goto err_buf_info_unmap;
769 }
770 }
771
772 #if 0 /* XXX LRO */
773 /* Create LRO for the ring */
774 if ((adapter->ifp->if_capenable & IFCAP_LRO) != 0) {
775 int err = tcp_lro_init(&rx_ring->lro);
776 if (err != 0) {
777 device_printf(adapter->pdev,
778 "LRO[%d] Initialization failed!\n", qid);
779 } else {
780 ena_trace(ENA_INFO,
781 "RX Soft LRO[%d] Initialized\n", qid);
782 rx_ring->lro.ifp = adapter->ifp;
783 }
784 }
785 #endif
786
787 /* Allocate taskqueues */
788 TASK_INIT(&rx_ring->cmpl_task, 0, ena_deferred_rx_cleanup, rx_ring);
789 rx_ring->cmpl_tq = taskqueue_create("ena RX completion", M_WAITOK,
790 taskqueue_thread_enqueue, &rx_ring->cmpl_tq);
791
792 /* RSS set cpu for thread */
793 #ifdef RSS
794 CPU_SETOF(que->cpu, &cpu_mask);
795 taskqueue_start_threads_cpuset(&rx_ring->cmpl_tq, 1, PI_NET, &cpu_mask,
796 "%s rx_ring cmpl (bucket %d)",
797 device_get_nameunit(adapter->pdev), que->cpu);
798 #else
799 taskqueue_start_threads(&rx_ring->cmpl_tq, 1, TDPRI_KERN_DAEMON, -1,
800 "%s rx_ring cmpl", device_get_nameunit(adapter->pdev));
801 #endif
802
803 return (0);
804
805 err_buf_info_unmap:
806 while (i--) {
807 bus_dmamap_destroy(adapter->rx_buf_tag,
808 rx_ring->rx_buffer_info[i].map);
809 }
810
811 kfree(rx_ring->free_rx_ids, M_DEVBUF);
812 rx_ring->free_rx_ids = NULL;
813 kfree(rx_ring->rx_buffer_info, M_DEVBUF);
814 rx_ring->rx_buffer_info = NULL;
815 return (ENOMEM);
816 }
817
818 /**
819 * ena_free_rx_resources - Free Rx Resources
820 * @adapter: network interface device structure
821 * @qid: queue index
822 *
823 * Free all receive software resources
824 **/
825 static void
ena_free_rx_resources(struct ena_adapter * adapter,unsigned int qid)826 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
827 {
828 struct ena_ring *rx_ring = &adapter->rx_ring[qid];
829
830 while (taskqueue_cancel(rx_ring->cmpl_tq, &rx_ring->cmpl_task, NULL) != 0)
831 taskqueue_drain(rx_ring->cmpl_tq, &rx_ring->cmpl_task);
832
833 taskqueue_free(rx_ring->cmpl_tq);
834
835 /* Free buffer DMA maps, */
836 for (int i = 0; i < rx_ring->ring_size; i++) {
837 m_freem(rx_ring->rx_buffer_info[i].mbuf);
838 rx_ring->rx_buffer_info[i].mbuf = NULL;
839 bus_dmamap_unload(adapter->rx_buf_tag,
840 rx_ring->rx_buffer_info[i].map);
841 bus_dmamap_destroy(adapter->rx_buf_tag,
842 rx_ring->rx_buffer_info[i].map);
843 }
844
845 #if 0 /* XXX LRO */
846 /* free LRO resources, */
847 tcp_lro_free(&rx_ring->lro);
848 #endif
849
850 /* free allocated memory */
851 kfree(rx_ring->rx_buffer_info, M_DEVBUF);
852 rx_ring->rx_buffer_info = NULL;
853
854 kfree(rx_ring->free_rx_ids, M_DEVBUF);
855 rx_ring->free_rx_ids = NULL;
856 }
857
858 /**
859 * ena_setup_all_rx_resources - allocate all queues Rx resources
860 * @adapter: network interface device structure
861 *
862 * Returns 0 on success, otherwise on failure.
863 **/
864 static int
ena_setup_all_rx_resources(struct ena_adapter * adapter)865 ena_setup_all_rx_resources(struct ena_adapter *adapter)
866 {
867 int i, rc = 0;
868
869 for (i = 0; i < adapter->num_queues; i++) {
870 rc = ena_setup_rx_resources(adapter, i);
871 if (rc != 0) {
872 device_printf(adapter->pdev,
873 "Allocation for Rx Queue %u failed\n", i);
874 goto err_setup_rx;
875 }
876 }
877 return (0);
878
879 err_setup_rx:
880 /* rewind the index freeing the rings as we go */
881 while (i--)
882 ena_free_rx_resources(adapter, i);
883 return (rc);
884 }
885
886 /**
887 * ena_free_all_rx_resources - Free Rx resources for all queues
888 * @adapter: network interface device structure
889 *
890 * Free all receive software resources
891 **/
892 static void
ena_free_all_rx_resources(struct ena_adapter * adapter)893 ena_free_all_rx_resources(struct ena_adapter *adapter)
894 {
895 int i;
896
897 for (i = 0; i < adapter->num_queues; i++)
898 ena_free_rx_resources(adapter, i);
899 }
900
901 static inline int
ena_alloc_rx_mbuf(struct ena_adapter * adapter,struct ena_ring * rx_ring,struct ena_rx_buffer * rx_info)902 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
903 struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
904 {
905 struct ena_com_buf *ena_buf;
906 bus_dma_segment_t segs[1];
907 int nsegs, error;
908 int mlen;
909
910 /* if previous allocated frag is not used */
911 if (unlikely(rx_info->mbuf != NULL))
912 return (0);
913
914 /* Get mbuf using UMA allocator */
915 rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE);
916
917 if (unlikely(rx_info->mbuf == NULL)) {
918 #if 0 /* XXX swildner counters */
919 counter_u64_add(rx_ring->rx_stats.mjum_alloc_fail, 1);
920 #endif
921 rx_info->mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
922 if (unlikely(rx_info->mbuf == NULL)) {
923 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
924 #if 0 /* XXX swildner counters */
925 counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
926 #endif
927 return (ENOMEM);
928 }
929 mlen = MCLBYTES;
930 } else {
931 mlen = MJUMPAGESIZE;
932 }
933 /* Set mbuf length*/
934 rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = mlen;
935
936 /* Map packets for DMA */
937 ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
938 "Using tag %p for buffers' DMA mapping, mbuf %p len: %d",
939 adapter->rx_buf_tag,rx_info->mbuf, rx_info->mbuf->m_len);
940 error = bus_dmamap_load_mbuf_segment(adapter->rx_buf_tag, rx_info->map,
941 rx_info->mbuf, segs, 1, &nsegs, BUS_DMA_NOWAIT);
942 if (unlikely((error != 0) || (nsegs != 1))) {
943 ena_trace(ENA_WARNING, "failed to map mbuf, error: %d, "
944 "nsegs: %d\n", error, nsegs);
945 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
946 #if 0 /* XXX swildner counters */
947 counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
948 #endif
949 goto exit;
950
951 }
952
953 bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
954
955 ena_buf = &rx_info->ena_buf;
956 ena_buf->paddr = segs[0].ds_addr;
957 ena_buf->len = mlen;
958
959 ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
960 "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
961 rx_info->mbuf, rx_info,ena_buf->len, (uintmax_t)ena_buf->paddr);
962
963 return (0);
964
965 exit:
966 m_freem(rx_info->mbuf);
967 rx_info->mbuf = NULL;
968 return (EFAULT);
969 }
970
971 static void
ena_free_rx_mbuf(struct ena_adapter * adapter,struct ena_ring * rx_ring,struct ena_rx_buffer * rx_info)972 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
973 struct ena_rx_buffer *rx_info)
974 {
975
976 if (rx_info->mbuf == NULL) {
977 ena_trace(ENA_WARNING, "Trying to free unallocated buffer\n");
978 return;
979 }
980
981 bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
982 m_freem(rx_info->mbuf);
983 rx_info->mbuf = NULL;
984 }
985
986 /**
987 * ena_refill_rx_bufs - Refills ring with descriptors
988 * @rx_ring: the ring which we want to feed with free descriptors
989 * @num: number of descriptors to refill
990 * Refills the ring with newly allocated DMA-mapped mbufs for receiving
991 **/
992 static int
ena_refill_rx_bufs(struct ena_ring * rx_ring,uint32_t num)993 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
994 {
995 struct ena_adapter *adapter = rx_ring->adapter;
996 uint16_t next_to_use, req_id;
997 uint32_t i;
998 int rc;
999
1000 ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC, "refill qid: %d",
1001 rx_ring->qid);
1002
1003 next_to_use = rx_ring->next_to_use;
1004
1005 for (i = 0; i < num; i++) {
1006 struct ena_rx_buffer *rx_info;
1007
1008 ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC,
1009 "RX buffer - next to use: %d", next_to_use);
1010
1011 req_id = rx_ring->free_rx_ids[next_to_use];
1012 rc = validate_rx_req_id(rx_ring, req_id);
1013 if (unlikely(rc != 0))
1014 break;
1015
1016 rx_info = &rx_ring->rx_buffer_info[req_id];
1017
1018 rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1019 if (unlikely(rc != 0)) {
1020 ena_trace(ENA_WARNING,
1021 "failed to alloc buffer for rx queue %d\n",
1022 rx_ring->qid);
1023 break;
1024 }
1025 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1026 &rx_info->ena_buf, req_id);
1027 if (unlikely(rc != 0)) {
1028 ena_trace(ENA_WARNING,
1029 "failed to add buffer for rx queue %d\n",
1030 rx_ring->qid);
1031 break;
1032 }
1033 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1034 rx_ring->ring_size);
1035 }
1036
1037 if (unlikely(i < num)) {
1038 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
1039 #if 0 /* XXX swildner counters */
1040 counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1041 #endif
1042 ena_trace(ENA_WARNING,
1043 "refilled rx qid %d with only %d mbufs (from %d)\n",
1044 rx_ring->qid, i, num);
1045 }
1046
1047 if (likely(i != 0)) {
1048 wmb();
1049 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1050 }
1051 rx_ring->next_to_use = next_to_use;
1052 return (i);
1053 }
1054
1055 static void
ena_free_rx_bufs(struct ena_adapter * adapter,unsigned int qid)1056 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1057 {
1058 struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1059 unsigned int i;
1060
1061 for (i = 0; i < rx_ring->ring_size; i++) {
1062 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1063
1064 if (rx_info->mbuf != NULL)
1065 ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1066 }
1067 }
1068
1069 /**
1070 * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1071 * @adapter: network interface device structure
1072 *
1073 */
1074 static void
ena_refill_all_rx_bufs(struct ena_adapter * adapter)1075 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1076 {
1077 struct ena_ring *rx_ring;
1078 int i, rc, bufs_num;
1079
1080 for (i = 0; i < adapter->num_queues; i++) {
1081 rx_ring = &adapter->rx_ring[i];
1082 bufs_num = rx_ring->ring_size - 1;
1083 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1084
1085 if (unlikely(rc != bufs_num))
1086 ena_trace(ENA_WARNING, "refilling Queue %d failed. "
1087 "Allocated %d buffers from: %d\n", i, rc, bufs_num);
1088 }
1089 }
1090
1091 static void
ena_free_all_rx_bufs(struct ena_adapter * adapter)1092 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1093 {
1094 int i;
1095
1096 for (i = 0; i < adapter->num_queues; i++)
1097 ena_free_rx_bufs(adapter, i);
1098 }
1099
1100 /**
1101 * ena_free_tx_bufs - Free Tx Buffers per Queue
1102 * @adapter: network interface device structure
1103 * @qid: queue index
1104 **/
1105 static void
ena_free_tx_bufs(struct ena_adapter * adapter,unsigned int qid)1106 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1107 {
1108 bool print_once = true;
1109 struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1110
1111 ENA_RING_MTX_LOCK(tx_ring);
1112 for (int i = 0; i < tx_ring->ring_size; i++) {
1113 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1114
1115 if (tx_info->mbuf == NULL)
1116 continue;
1117
1118 if (print_once) {
1119 device_printf(adapter->pdev,
1120 "free uncompleted tx mbuf qid %d idx 0x%x",
1121 qid, i);
1122 print_once = false;
1123 } else {
1124 ena_trace(ENA_DBG,
1125 "free uncompleted tx mbuf qid %d idx 0x%x",
1126 qid, i);
1127 }
1128
1129 bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
1130 m_free(tx_info->mbuf);
1131 tx_info->mbuf = NULL;
1132 }
1133 ENA_RING_MTX_UNLOCK(tx_ring);
1134 }
1135
1136 static void
ena_free_all_tx_bufs(struct ena_adapter * adapter)1137 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1138 {
1139
1140 for (int i = 0; i < adapter->num_queues; i++)
1141 ena_free_tx_bufs(adapter, i);
1142 }
1143
1144 static void
ena_destroy_all_tx_queues(struct ena_adapter * adapter)1145 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1146 {
1147 uint16_t ena_qid;
1148 int i;
1149
1150 for (i = 0; i < adapter->num_queues; i++) {
1151 ena_qid = ENA_IO_TXQ_IDX(i);
1152 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1153 }
1154 }
1155
1156 static void
ena_destroy_all_rx_queues(struct ena_adapter * adapter)1157 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1158 {
1159 uint16_t ena_qid;
1160 int i;
1161
1162 for (i = 0; i < adapter->num_queues; i++) {
1163 ena_qid = ENA_IO_RXQ_IDX(i);
1164 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1165 }
1166 }
1167
1168 static void
ena_destroy_all_io_queues(struct ena_adapter * adapter)1169 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1170 {
1171 ena_destroy_all_tx_queues(adapter);
1172 ena_destroy_all_rx_queues(adapter);
1173 }
1174
1175 static inline int
validate_tx_req_id(struct ena_ring * tx_ring,uint16_t req_id)1176 validate_tx_req_id(struct ena_ring *tx_ring, uint16_t req_id)
1177 {
1178 struct ena_adapter *adapter = tx_ring->adapter;
1179 struct ena_tx_buffer *tx_info = NULL;
1180
1181 if (likely(req_id < tx_ring->ring_size)) {
1182 tx_info = &tx_ring->tx_buffer_info[req_id];
1183 if (tx_info->mbuf != NULL)
1184 return (0);
1185 }
1186
1187 if (tx_info->mbuf == NULL)
1188 device_printf(adapter->pdev,
1189 "tx_info doesn't have valid mbuf\n");
1190 else
1191 device_printf(adapter->pdev, "Invalid req_id: %hu\n", req_id);
1192
1193 IFNET_STAT_INC(tx_ring->adapter->ifp, oerrors, 1);
1194 #if 0 /* XXX swildner counters */
1195 counter_u64_add(tx_ring->tx_stats.bad_req_id, 1);
1196 #endif
1197
1198 return (EFAULT);
1199 }
1200
1201 static int
ena_create_io_queues(struct ena_adapter * adapter)1202 ena_create_io_queues(struct ena_adapter *adapter)
1203 {
1204 struct ena_com_dev *ena_dev = adapter->ena_dev;
1205 struct ena_com_create_io_ctx ctx;
1206 struct ena_ring *ring;
1207 uint16_t ena_qid;
1208 uint32_t msix_vector;
1209 int rc, i;
1210
1211 /* Create TX queues */
1212 for (i = 0; i < adapter->num_queues; i++) {
1213 msix_vector = ENA_IO_IRQ_IDX(i);
1214 ena_qid = ENA_IO_TXQ_IDX(i);
1215 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1216 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1217 ctx.queue_size = adapter->tx_ring_size;
1218 ctx.msix_vector = msix_vector;
1219 ctx.qid = ena_qid;
1220 rc = ena_com_create_io_queue(ena_dev, &ctx);
1221 if (rc != 0) {
1222 device_printf(adapter->pdev,
1223 "Failed to create io TX queue #%d rc: %d\n", i, rc);
1224 goto err_tx;
1225 }
1226 ring = &adapter->tx_ring[i];
1227 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1228 &ring->ena_com_io_sq,
1229 &ring->ena_com_io_cq);
1230 if (rc != 0) {
1231 device_printf(adapter->pdev,
1232 "Failed to get TX queue handlers. TX queue num"
1233 " %d rc: %d\n", i, rc);
1234 ena_com_destroy_io_queue(ena_dev, ena_qid);
1235 goto err_tx;
1236 }
1237 }
1238
1239 /* Create RX queues */
1240 for (i = 0; i < adapter->num_queues; i++) {
1241 msix_vector = ENA_IO_IRQ_IDX(i);
1242 ena_qid = ENA_IO_RXQ_IDX(i);
1243 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1244 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1245 ctx.queue_size = adapter->rx_ring_size;
1246 ctx.msix_vector = msix_vector;
1247 ctx.qid = ena_qid;
1248 rc = ena_com_create_io_queue(ena_dev, &ctx);
1249 if (unlikely(rc != 0)) {
1250 device_printf(adapter->pdev,
1251 "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1252 goto err_rx;
1253 }
1254
1255 ring = &adapter->rx_ring[i];
1256 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1257 &ring->ena_com_io_sq,
1258 &ring->ena_com_io_cq);
1259 if (unlikely(rc != 0)) {
1260 device_printf(adapter->pdev,
1261 "Failed to get RX queue handlers. RX queue num"
1262 " %d rc: %d\n", i, rc);
1263 ena_com_destroy_io_queue(ena_dev, ena_qid);
1264 goto err_rx;
1265 }
1266 }
1267
1268 return (0);
1269
1270 err_rx:
1271 while (i--)
1272 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1273 i = adapter->num_queues;
1274 err_tx:
1275 while (i--)
1276 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1277
1278 return (ENXIO);
1279 }
1280
1281 /**
1282 * ena_tx_cleanup - clear sent packets and corresponding descriptors
1283 * @tx_ring: ring for which we want to clean packets
1284 *
1285 * Once packets are sent, we ask the device in a loop for no longer used
1286 * descriptors. We find the related mbuf chain in a map (index in an array)
1287 * and free it, then update ring state.
1288 * This is performed in "endless" loop, updating ring pointers every
1289 * TX_COMMIT. The first check of free descriptor is performed before the actual
1290 * loop, then repeated at the loop end.
1291 **/
1292 static int
ena_tx_cleanup(struct ena_ring * tx_ring)1293 ena_tx_cleanup(struct ena_ring *tx_ring)
1294 {
1295 struct ena_adapter *adapter;
1296 struct ena_com_io_cq* io_cq;
1297 uint16_t next_to_clean;
1298 uint16_t req_id;
1299 uint16_t ena_qid;
1300 unsigned int total_done = 0;
1301 int rc;
1302 int commit = TX_COMMIT;
1303 int budget = TX_BUDGET;
1304 int work_done;
1305
1306 adapter = tx_ring->que->adapter;
1307 ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
1308 io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1309 next_to_clean = tx_ring->next_to_clean;
1310
1311 do {
1312 struct ena_tx_buffer *tx_info;
1313 struct mbuf *mbuf;
1314
1315 rc = ena_com_tx_comp_req_id_get(io_cq, &req_id);
1316 if (unlikely(rc != 0))
1317 break;
1318
1319 rc = validate_tx_req_id(tx_ring, req_id);
1320 if (unlikely(rc != 0))
1321 break;
1322
1323 tx_info = &tx_ring->tx_buffer_info[req_id];
1324
1325 mbuf = tx_info->mbuf;
1326
1327 tx_info->mbuf = NULL;
1328 timevalclear(&tx_info->timestamp);
1329
1330 if (likely(tx_info->num_of_bufs != 0)) {
1331 /* Map is no longer required */
1332 bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
1333 }
1334
1335 ena_trace(ENA_DBG | ENA_TXPTH, "tx: q %d mbuf %p completed",
1336 tx_ring->qid, mbuf);
1337
1338 m_freem(mbuf);
1339
1340 total_done += tx_info->tx_descs;
1341
1342 tx_ring->free_tx_ids[next_to_clean] = req_id;
1343 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1344 tx_ring->ring_size);
1345
1346 if (unlikely(--commit == 0)) {
1347 commit = TX_COMMIT;
1348 /* update ring state every TX_COMMIT descriptor */
1349 tx_ring->next_to_clean = next_to_clean;
1350 ena_com_comp_ack(
1351 &adapter->ena_dev->io_sq_queues[ena_qid],
1352 total_done);
1353 ena_com_update_dev_comp_head(io_cq);
1354 total_done = 0;
1355 }
1356 } while (likely(--budget));
1357
1358 work_done = TX_BUDGET - budget;
1359
1360 ena_trace(ENA_DBG | ENA_TXPTH, "tx: q %d done. total pkts: %d",
1361 tx_ring->qid, work_done);
1362
1363 /* If there is still something to commit update ring state */
1364 if (likely(commit != TX_COMMIT)) {
1365 tx_ring->next_to_clean = next_to_clean;
1366 ena_com_comp_ack(&adapter->ena_dev->io_sq_queues[ena_qid],
1367 total_done);
1368 ena_com_update_dev_comp_head(io_cq);
1369 }
1370
1371 return (work_done);
1372 }
1373
1374 static void
ena_rx_hash_mbuf(struct ena_ring * rx_ring,struct ena_com_rx_ctx * ena_rx_ctx,struct mbuf * mbuf)1375 ena_rx_hash_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1376 struct mbuf *mbuf)
1377 {
1378 struct ena_adapter *adapter = rx_ring->adapter;
1379
1380 if (likely(adapter->rss_support)) {
1381 //mbuf->m_pkthdr.flowid = ena_rx_ctx->hash;
1382 m_sethash(mbuf, ena_rx_ctx->hash);
1383
1384 #if 0 /* XXX rsstype doesn't seem to be needed by the network stack, we will only supply the hash. */
1385 if (ena_rx_ctx->frag &&
1386 (ena_rx_ctx->l3_proto != ENA_ETH_IO_L3_PROTO_UNKNOWN)) {
1387 M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1388 return;
1389 }
1390
1391 switch (ena_rx_ctx->l3_proto) {
1392 case ENA_ETH_IO_L3_PROTO_IPV4:
1393 switch (ena_rx_ctx->l4_proto) {
1394 case ENA_ETH_IO_L4_PROTO_TCP:
1395 M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV4);
1396 break;
1397 case ENA_ETH_IO_L4_PROTO_UDP:
1398 M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV4);
1399 break;
1400 default:
1401 M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV4);
1402 }
1403 break;
1404 case ENA_ETH_IO_L3_PROTO_IPV6:
1405 switch (ena_rx_ctx->l4_proto) {
1406 case ENA_ETH_IO_L4_PROTO_TCP:
1407 M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV6);
1408 break;
1409 case ENA_ETH_IO_L4_PROTO_UDP:
1410 M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV6);
1411 break;
1412 default:
1413 M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV6);
1414 }
1415 break;
1416 case ENA_ETH_IO_L3_PROTO_UNKNOWN:
1417 M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1418 break;
1419 default:
1420 M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1421 }
1422 #endif
1423 } else {
1424 //mbuf->m_pkthdr.flowid = rx_ring->qid;
1425 //M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1426 m_sethash(mbuf, rx_ring->qid);
1427 }
1428 }
1429
1430 /**
1431 * ena_rx_mbuf - assemble mbuf from descriptors
1432 * @rx_ring: ring for which we want to clean packets
1433 * @ena_bufs: buffer info
1434 * @ena_rx_ctx: metadata for this packet(s)
1435 * @next_to_clean: ring pointer, will be updated only upon success
1436 *
1437 **/
1438 static struct mbuf*
ena_rx_mbuf(struct ena_ring * rx_ring,struct ena_com_rx_buf_info * ena_bufs,struct ena_com_rx_ctx * ena_rx_ctx,uint16_t * next_to_clean)1439 ena_rx_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_buf_info *ena_bufs,
1440 struct ena_com_rx_ctx *ena_rx_ctx, uint16_t *next_to_clean)
1441 {
1442 struct mbuf *mbuf;
1443 struct ena_rx_buffer *rx_info;
1444 struct ena_adapter *adapter;
1445 unsigned int descs = ena_rx_ctx->descs;
1446 uint16_t ntc, len, req_id, buf = 0;
1447
1448 ntc = *next_to_clean;
1449 adapter = rx_ring->adapter;
1450 rx_info = &rx_ring->rx_buffer_info[ntc];
1451
1452 if (unlikely(rx_info->mbuf == NULL)) {
1453 device_printf(adapter->pdev, "NULL mbuf in rx_info");
1454 return (NULL);
1455 }
1456
1457 len = ena_bufs[buf].len;
1458 req_id = ena_bufs[buf].req_id;
1459 rx_info = &rx_ring->rx_buffer_info[req_id];
1460
1461 ena_trace(ENA_DBG | ENA_RXPTH, "rx_info %p, mbuf %p, paddr %jx",
1462 rx_info, rx_info->mbuf, (uintmax_t)rx_info->ena_buf.paddr);
1463
1464 mbuf = rx_info->mbuf;
1465 mbuf->m_flags |= M_PKTHDR;
1466 mbuf->m_pkthdr.len = len;
1467 mbuf->m_len = len;
1468 mbuf->m_pkthdr.rcvif = rx_ring->que->adapter->ifp;
1469
1470 /* Fill mbuf with hash key and it's interpretation for optimization */
1471 ena_rx_hash_mbuf(rx_ring, ena_rx_ctx, mbuf);
1472
1473 ena_trace(ENA_DBG | ENA_RXPTH, "rx mbuf 0x%p, flags=0x%x, len: %d",
1474 mbuf, mbuf->m_flags, mbuf->m_pkthdr.len);
1475
1476 /* DMA address is not needed anymore, unmap it */
1477 bus_dmamap_unload(rx_ring->adapter->rx_buf_tag, rx_info->map);
1478
1479 rx_info->mbuf = NULL;
1480 rx_ring->free_rx_ids[ntc] = req_id;
1481 ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1482
1483 /*
1484 * While we have more than 1 descriptors for one rcvd packet, append
1485 * other mbufs to the main one
1486 */
1487 while (--descs) {
1488 ++buf;
1489 len = ena_bufs[buf].len;
1490 req_id = ena_bufs[buf].req_id;
1491 rx_info = &rx_ring->rx_buffer_info[req_id];
1492
1493 if (unlikely(rx_info->mbuf == NULL)) {
1494 device_printf(adapter->pdev, "NULL mbuf in rx_info");
1495 /*
1496 * If one of the required mbufs was not allocated yet,
1497 * we can break there.
1498 * All earlier used descriptors will be reallocated
1499 * later and not used mbufs can be reused.
1500 * The next_to_clean pointer will not be updated in case
1501 * of an error, so caller should advance it manually
1502 * in error handling routine to keep it up to date
1503 * with hw ring.
1504 */
1505 m_freem(mbuf);
1506 return (NULL);
1507 }
1508
1509 if (unlikely(m_append(mbuf, len, rx_info->mbuf->m_data) == 0)) {
1510 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
1511 #if 0 /* XXX swildner counters */
1512 counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1513 #endif
1514 ena_trace(ENA_WARNING, "Failed to append Rx mbuf %p",
1515 mbuf);
1516 }
1517
1518 ena_trace(ENA_DBG | ENA_RXPTH,
1519 "rx mbuf updated. len %d", mbuf->m_pkthdr.len);
1520
1521 /* Free already appended mbuf, it won't be useful anymore */
1522 bus_dmamap_unload(rx_ring->adapter->rx_buf_tag, rx_info->map);
1523 m_freem(rx_info->mbuf);
1524 rx_info->mbuf = NULL;
1525
1526 rx_ring->free_rx_ids[ntc] = req_id;
1527 ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1528 }
1529
1530 *next_to_clean = ntc;
1531
1532 return (mbuf);
1533 }
1534
1535 /**
1536 * ena_rx_checksum - indicate in mbuf if hw indicated a good cksum
1537 **/
1538 static inline void
ena_rx_checksum(struct ena_ring * rx_ring,struct ena_com_rx_ctx * ena_rx_ctx,struct mbuf * mbuf)1539 ena_rx_checksum(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1540 struct mbuf *mbuf)
1541 {
1542
1543 /* if IP and error */
1544 if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1545 ena_rx_ctx->l3_csum_err)) {
1546 /* ipv4 checksum error */
1547 mbuf->m_pkthdr.csum_flags = 0;
1548 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
1549 #if 0 /* XXX swildner counters */
1550 counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1551 #endif
1552 ena_trace(ENA_DBG, "RX IPv4 header checksum error");
1553 return;
1554 }
1555
1556 /* if TCP/UDP */
1557 if ((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1558 (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)) {
1559 if (ena_rx_ctx->l4_csum_err) {
1560 /* TCP/UDP checksum error */
1561 mbuf->m_pkthdr.csum_flags = 0;
1562 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
1563 #if 0 /* XXX swildner counters */
1564 counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1565 #endif
1566 ena_trace(ENA_DBG, "RX L4 checksum error");
1567 } else {
1568 mbuf->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
1569 mbuf->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1570 }
1571 }
1572 }
1573
1574 static void
ena_deferred_rx_cleanup(void * arg,int pending)1575 ena_deferred_rx_cleanup(void *arg, int pending)
1576 {
1577 struct ena_ring *rx_ring = arg;
1578 int budget = CLEAN_BUDGET;
1579
1580 ENA_RING_MTX_LOCK(rx_ring);
1581 /*
1582 * If deferred task was executed, perform cleanup of all awaiting
1583 * descs (or until given budget is depleted to avoid infinite loop).
1584 */
1585 while (likely(budget--)) {
1586 if (ena_rx_cleanup(rx_ring) == 0)
1587 break;
1588 }
1589 ENA_RING_MTX_UNLOCK(rx_ring);
1590 }
1591
1592 /**
1593 * ena_rx_cleanup - handle rx irq
1594 * @arg: ring for which irq is being handled
1595 **/
1596 static int
ena_rx_cleanup(struct ena_ring * rx_ring)1597 ena_rx_cleanup(struct ena_ring *rx_ring)
1598 {
1599 struct ena_adapter *adapter;
1600 struct mbuf *mbuf;
1601 struct ena_com_rx_ctx ena_rx_ctx;
1602 struct ena_com_io_cq* io_cq;
1603 struct ena_com_io_sq* io_sq;
1604 if_t ifp;
1605 uint16_t ena_qid;
1606 uint16_t next_to_clean;
1607 uint32_t refill_required;
1608 uint32_t refill_threshold;
1609 uint32_t do_if_input = 0;
1610 unsigned int qid;
1611 int rc, i;
1612 int budget = RX_BUDGET;
1613
1614 adapter = rx_ring->que->adapter;
1615 ifp = adapter->ifp;
1616 qid = rx_ring->que->id;
1617 ena_qid = ENA_IO_RXQ_IDX(qid);
1618 io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1619 io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
1620 next_to_clean = rx_ring->next_to_clean;
1621
1622 ena_trace(ENA_DBG, "rx: qid %d", qid);
1623
1624 do {
1625 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1626 ena_rx_ctx.max_bufs = adapter->max_rx_sgl_size;
1627 ena_rx_ctx.descs = 0;
1628 rc = ena_com_rx_pkt(io_cq, io_sq, &ena_rx_ctx);
1629
1630 if (unlikely(rc != 0))
1631 goto error;
1632
1633 if (unlikely(ena_rx_ctx.descs == 0))
1634 break;
1635
1636 ena_trace(ENA_DBG | ENA_RXPTH, "rx: q %d got packet from ena. "
1637 "descs #: %d l3 proto %d l4 proto %d hash: %x",
1638 rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1639 ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1640
1641 /* Receive mbuf from the ring */
1642 mbuf = ena_rx_mbuf(rx_ring, rx_ring->ena_bufs,
1643 &ena_rx_ctx, &next_to_clean);
1644
1645 /* Exit if we failed to retrieve a buffer */
1646 if (unlikely(mbuf == NULL)) {
1647 for (i = 0; i < ena_rx_ctx.descs; ++i) {
1648 rx_ring->free_rx_ids[next_to_clean] =
1649 rx_ring->ena_bufs[i].req_id;
1650 next_to_clean =
1651 ENA_RX_RING_IDX_NEXT(next_to_clean,
1652 rx_ring->ring_size);
1653
1654 }
1655 break;
1656 }
1657
1658 /*
1659 * XXX Removed IFCAP_RXCSUM_IPV6 check because DragonFly
1660 * does not seem to support it
1661 */
1662 if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) {
1663 ena_rx_checksum(rx_ring, &ena_rx_ctx, mbuf);
1664 }
1665
1666 #if 0 /* XXX swildner counters */
1667 counter_enter();
1668 counter_u64_add_protected(rx_ring->rx_stats.bytes,
1669 mbuf->m_pkthdr.len);
1670 counter_u64_add_protected(adapter->hw_stats.rx_bytes,
1671 mbuf->m_pkthdr.len);
1672 counter_exit();
1673 #endif
1674 /*
1675 * LRO is only for IP/TCP packets and TCP checksum of the packet
1676 * should be computed by hardware.
1677 */
1678 do_if_input = 1;
1679 #if 0 /* XXX LRO */
1680 if (((ifp->if_capenable & IFCAP_LRO) != 0) &&
1681 ((mbuf->m_pkthdr.csum_flags & CSUM_IP_VALID) != 0) &&
1682 (ena_rx_ctx.l4_proto == ENA_ETH_IO_L4_PROTO_TCP)) {
1683 /*
1684 * Send to the stack if:
1685 * - LRO not enabled, or
1686 * - no LRO resources, or
1687 * - lro enqueue fails
1688 */
1689 if ((rx_ring->lro.lro_cnt != 0) &&
1690 (tcp_lro_rx(&rx_ring->lro, mbuf, 0) == 0))
1691 do_if_input = 0;
1692 }
1693 #endif
1694 if (do_if_input != 0) {
1695 ena_trace(ENA_DBG | ENA_RXPTH,
1696 "calling if_input() with mbuf %p", mbuf);
1697 ENA_RING_MTX_UNLOCK(rx_ring);
1698 (*ifp->if_input)(ifp, mbuf, NULL, -1);
1699 ENA_RING_MTX_LOCK(rx_ring);
1700 }
1701
1702 IFNET_STAT_INC(ifp, ipackets, 1);
1703 #if 0 /* XXX swildner counters */
1704 counter_enter();
1705 counter_u64_add_protected(rx_ring->rx_stats.cnt, 1);
1706 counter_u64_add_protected(adapter->hw_stats.rx_packets, 1);
1707 counter_exit();
1708 #endif
1709 } while (--budget);
1710
1711 rx_ring->next_to_clean = next_to_clean;
1712
1713 refill_required = ena_com_free_desc(io_sq);
1714 refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1715
1716 if (refill_required > refill_threshold) {
1717 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1718 ena_refill_rx_bufs(rx_ring, refill_required);
1719 }
1720
1721 #if 0 /* XXX LRO */
1722 tcp_lro_flush_all(&rx_ring->lro);
1723 #endif
1724
1725 return (RX_BUDGET - budget);
1726
1727 error:
1728 IFNET_STAT_INC(rx_ring->adapter->ifp, ierrors, 1);
1729 #if 0 /* XXX swildner counters */
1730 counter_u64_add(rx_ring->rx_stats.bad_desc_num, 1);
1731 #endif
1732 return (RX_BUDGET - budget);
1733 }
1734
1735 /*********************************************************************
1736 *
1737 * MSIX & Interrupt Service routine
1738 *
1739 **********************************************************************/
1740
1741 /**
1742 * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1743 * @arg: interrupt number
1744 **/
1745 static void
ena_intr_msix_mgmnt(void * arg)1746 ena_intr_msix_mgmnt(void *arg)
1747 {
1748 struct ena_adapter *adapter = (struct ena_adapter *)arg;
1749
1750 ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1751 if (likely(adapter->running))
1752 ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1753 }
1754
1755 /**
1756 * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1757 * @arg: interrupt number
1758 **/
1759 static void
ena_handle_msix(void * arg)1760 ena_handle_msix(void *arg)
1761 {
1762 struct ena_que *que = arg;
1763 struct ena_adapter *adapter = que->adapter;
1764 if_t ifp = adapter->ifp;
1765 struct ena_ring *tx_ring;
1766 struct ena_ring *rx_ring;
1767 struct ena_com_io_cq* io_cq;
1768 struct ena_eth_io_intr_reg intr_reg;
1769 int qid, ena_qid;
1770 int txc, rxc, i;
1771
1772 if (unlikely((ifp->if_flags & IFF_RUNNING) == 0))
1773 return;
1774
1775 ena_trace(ENA_DBG, "MSI-X TX/RX routine");
1776
1777 tx_ring = que->tx_ring;
1778 rx_ring = que->rx_ring;
1779 qid = que->id;
1780 ena_qid = ENA_IO_TXQ_IDX(qid);
1781 io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1782
1783 for (i = 0; i < CLEAN_BUDGET; ++i) {
1784 /*
1785 * If lock cannot be acquired, then deferred cleanup task was
1786 * being executed and rx ring is being cleaned up in
1787 * another thread.
1788 */
1789 if (likely(ENA_RING_MTX_TRYLOCK(rx_ring) != 0)) {
1790 rxc = ena_rx_cleanup(rx_ring);
1791 ENA_RING_MTX_UNLOCK(rx_ring);
1792 } else {
1793 rxc = 0;
1794 }
1795
1796 /* Protection from calling ena_tx_cleanup from ena_start_xmit */
1797 ENA_RING_MTX_LOCK(tx_ring);
1798 txc = ena_tx_cleanup(tx_ring);
1799 ENA_RING_MTX_UNLOCK(tx_ring);
1800
1801 if (unlikely((ifp->if_flags & IFF_RUNNING) == 0))
1802 return;
1803
1804 if ((txc != TX_BUDGET) && (rxc != RX_BUDGET))
1805 break;
1806 }
1807
1808 /* Signal that work is done and unmask interrupt */
1809 ena_com_update_intr_reg(&intr_reg,
1810 RX_IRQ_INTERVAL,
1811 TX_IRQ_INTERVAL,
1812 true);
1813 ena_com_unmask_intr(io_cq, &intr_reg);
1814 }
1815
1816 static int
ena_enable_msix(struct ena_adapter * adapter)1817 ena_enable_msix(struct ena_adapter *adapter)
1818 {
1819 device_t dev = adapter->pdev;
1820 int msix_vecs;
1821 int error, i, rc = 0;
1822
1823 /* Reserved the max msix vectors we might need */
1824 msix_vecs = ENA_MAX_MSIX_VEC(adapter->num_queues);
1825
1826 adapter->msix_entries = kmalloc(msix_vecs * sizeof(struct msix_entry),
1827 M_DEVBUF, M_WAITOK | M_ZERO);
1828
1829 ena_trace(ENA_DBG, "trying to enable MSI-X, vectors: %d", msix_vecs);
1830
1831 for (i = 0; i < msix_vecs; i++) {
1832 adapter->msix_entries[i].entry = i;
1833 /* Vectors must start from 1 */
1834 adapter->msix_entries[i].vector = i + 1;
1835 }
1836
1837 error = pci_setup_msix(dev);
1838 if (error) {
1839 device_printf(dev, "pci_setup_msix() failed\n");
1840 goto err_msix_free;
1841 }
1842
1843 adapter->msix_vecs = msix_vecs;
1844 adapter->msix_enabled = true;
1845
1846 return (0);
1847
1848 err_msix_free:
1849 kfree(adapter->msix_entries, M_DEVBUF);
1850 adapter->msix_entries = NULL;
1851
1852 return (rc);
1853 }
1854
1855 static void
ena_setup_mgmnt_intr(struct ena_adapter * adapter)1856 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1857 {
1858
1859 ksnprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1860 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1861 device_get_nameunit(adapter->pdev));
1862 /*
1863 * Handler is NULL on purpose, it will be set
1864 * when mgmnt interrupt is acquired
1865 */
1866 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1867 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1868 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1869 adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1870 }
1871
1872 static void
ena_setup_io_intr(struct ena_adapter * adapter)1873 ena_setup_io_intr(struct ena_adapter *adapter)
1874 {
1875 static int last_bind_cpu = -1;
1876 int irq_idx;
1877
1878 for (int i = 0; i < adapter->num_queues; i++) {
1879 irq_idx = ENA_IO_IRQ_IDX(i);
1880
1881 ksnprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1882 "%s-TxRx-%d", device_get_nameunit(adapter->pdev), i);
1883 adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1884 adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1885 adapter->irq_tbl[irq_idx].vector =
1886 adapter->msix_entries[irq_idx].vector;
1887 ena_trace(ENA_INFO | ENA_IOQ, "ena_setup_io_intr vector: %d\n",
1888 adapter->msix_entries[irq_idx].vector);
1889 #ifdef RSS
1890 adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1891 rss_getcpu(i % rss_getnumbuckets());
1892 #else
1893 /*
1894 * We still want to bind rings to the corresponding cpu
1895 * using something similar to the RSS round-robin technique.
1896 *
1897 * XXX It seems that this can be removed since DragonFly has
1898 * native support for RSS. DragonFly also does not have
1899 * support for CPU_FIRST or CPU_NEXT.
1900 */
1901
1902 if (last_bind_cpu < 0)
1903 last_bind_cpu = (last_bind_cpu + 1) % ncpus;
1904 adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1905 last_bind_cpu;
1906 last_bind_cpu = (last_bind_cpu + 1) % ncpus;
1907 #endif
1908 }
1909 }
1910
1911 static int
ena_request_mgmnt_irq(struct ena_adapter * adapter)1912 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1913 {
1914 struct ena_irq *irq;
1915 unsigned long flags;
1916 int error, rc, rcc;
1917
1918 flags = RF_ACTIVE | RF_SHAREABLE;
1919
1920 irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1921
1922 error = pci_alloc_msix_vector(adapter->pdev, 0, &irq->vector, 0);
1923 if (error) {
1924 device_printf(adapter->pdev, "Could not initialize MGMNT MSI-X Vector on cpu0\n");
1925 return (ENXIO);
1926 }
1927
1928 irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1929 &irq->vector, flags);
1930
1931 if (unlikely(irq->res == NULL)) {
1932 device_printf(adapter->pdev, "could not allocate "
1933 "irq vector: %d\n", irq->vector);
1934 pci_release_msix_vector(adapter->pdev, irq->vector);
1935 return (ENXIO);
1936 }
1937
1938 rc = bus_activate_resource(adapter->pdev, SYS_RES_IRQ,
1939 irq->vector, irq->res);
1940 if (unlikely(rc != 0)) {
1941 device_printf(adapter->pdev, "could not activate "
1942 "irq vector: %d\n", irq->vector);
1943 goto err_res_free;
1944 }
1945
1946 rc = bus_setup_intr(adapter->pdev, irq->res,
1947 INTR_MPSAFE, ena_intr_msix_mgmnt,
1948 irq->data, &irq->cookie, NULL);
1949 if (unlikely(rc != 0)) {
1950 device_printf(adapter->pdev, "failed to register "
1951 "interrupt handler for irq %ju: %d\n",
1952 rman_get_start(irq->res), rc);
1953 goto err_res_free;
1954 }
1955 irq->requested = true;
1956
1957 return (rc);
1958
1959 err_res_free:
1960 ena_trace(ENA_INFO | ENA_ADMQ, "releasing resource for irq %d\n",
1961 irq->vector);
1962 rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1963 irq->vector, irq->res);
1964 pci_release_msix_vector(adapter->pdev, irq->vector);
1965 if (unlikely(rcc != 0))
1966 device_printf(adapter->pdev, "dev has no parent while "
1967 "releasing res for irq: %d\n", irq->vector);
1968 irq->res = NULL;
1969
1970 return (rc);
1971 }
1972
1973 static int
ena_request_io_irq(struct ena_adapter * adapter)1974 ena_request_io_irq(struct ena_adapter *adapter)
1975 {
1976 struct ena_irq *irq;
1977 unsigned long flags = 0;
1978 int rc = 0, i, rcc, error;
1979
1980 if (unlikely(adapter->msix_enabled == 0)) {
1981 device_printf(adapter->pdev,
1982 "failed to request I/O IRQ: MSI-X is not enabled\n");
1983 return (EINVAL);
1984 } else {
1985 flags = RF_ACTIVE | RF_SHAREABLE;
1986 }
1987
1988 for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1989 irq = &adapter->irq_tbl[i];
1990
1991 if (unlikely(irq->requested))
1992 continue;
1993
1994 error = pci_alloc_msix_vector(adapter->pdev, i, &irq->vector, irq->cpu);
1995 if (error) {
1996 device_printf(adapter->pdev, "Unable to allocated MSI-X %d on cpu%d\n", i, irq->cpu);
1997 goto err;
1998 }
1999
2000 irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
2001 &irq->vector, flags);
2002 if (unlikely(irq->res == NULL)) {
2003 device_printf(adapter->pdev, "could not allocate "
2004 "irq vector: %d\n", irq->vector);
2005 goto err;
2006 }
2007
2008
2009 /*
2010 * TODO: Might need to setup desc and use irq->name as the
2011 * value
2012 */
2013 rc = bus_setup_intr(adapter->pdev, irq->res,
2014 INTR_MPSAFE,
2015 irq->handler, irq->data, &irq->cookie, NULL);
2016 if (unlikely(rc != 0)) {
2017 device_printf(adapter->pdev, "failed to register "
2018 "interrupt handler for irq %ju: %d\n",
2019 rman_get_start(irq->res), rc);
2020 goto err;
2021 }
2022 irq->requested = true;
2023
2024 #ifdef RSS
2025 ena_trace(ENA_INFO, "queue %d - RSS bucket %d\n",
2026 i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2027 #else
2028 ena_trace(ENA_INFO, "queue %d - cpu %d\n",
2029 i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2030 #endif
2031 }
2032
2033 return (rc);
2034
2035 err:
2036
2037 for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
2038 irq = &adapter->irq_tbl[i];
2039 rcc = 0;
2040
2041 /* Once we entered err: section and irq->requested is true we
2042 free both intr and resources */
2043 if (irq->requested)
2044 rcc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
2045 if (unlikely(rcc != 0))
2046 device_printf(adapter->pdev, "could not release"
2047 " irq: %d, error: %d\n", irq->vector, rcc);
2048
2049 /* If we entred err: section without irq->requested set we know
2050 it was bus_alloc_resource_any() that needs cleanup, provided
2051 res is not NULL. In case res is NULL no work in needed in
2052 this iteration */
2053 rcc = 0;
2054 if (irq->res != NULL) {
2055 rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2056 irq->vector, irq->res);
2057 pci_release_msix_vector(adapter->pdev, irq->vector);
2058 }
2059 if (unlikely(rcc != 0))
2060 device_printf(adapter->pdev, "dev has no parent while "
2061 "releasing res for irq: %d\n", irq->vector);
2062 irq->requested = false;
2063 irq->res = NULL;
2064 }
2065
2066 return (rc);
2067 }
2068
2069 static void
ena_free_mgmnt_irq(struct ena_adapter * adapter)2070 ena_free_mgmnt_irq(struct ena_adapter *adapter)
2071 {
2072 struct ena_irq *irq;
2073 int rc;
2074
2075 irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2076 if (irq->requested) {
2077 ena_trace(ENA_INFO | ENA_ADMQ, "tear down irq: %d\n",
2078 irq->vector);
2079 rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
2080 if (unlikely(rc != 0))
2081 device_printf(adapter->pdev, "failed to tear "
2082 "down irq: %d\n", irq->vector);
2083 irq->requested = 0;
2084 }
2085
2086 if (irq->res != NULL) {
2087 ena_trace(ENA_INFO | ENA_ADMQ, "release resource irq: %d\n",
2088 irq->vector);
2089 rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2090 irq->vector, irq->res);
2091 pci_release_msix_vector(adapter->pdev, irq->vector);
2092 irq->res = NULL;
2093 if (unlikely(rc != 0))
2094 device_printf(adapter->pdev, "dev has no parent while "
2095 "releasing res for irq: %d\n", irq->vector);
2096 }
2097 }
2098
2099 static void
ena_free_io_irq(struct ena_adapter * adapter)2100 ena_free_io_irq(struct ena_adapter *adapter)
2101 {
2102 struct ena_irq *irq;
2103 int rc;
2104
2105 for (int i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
2106 irq = &adapter->irq_tbl[i];
2107 if (irq->requested) {
2108 ena_trace(ENA_INFO | ENA_IOQ, "tear down irq: %d\n",
2109 irq->vector);
2110 rc = bus_teardown_intr(adapter->pdev, irq->res,
2111 irq->cookie);
2112 if (unlikely(rc != 0)) {
2113 device_printf(adapter->pdev, "failed to tear "
2114 "down irq: %d\n", irq->vector);
2115 }
2116 irq->requested = 0;
2117 }
2118
2119 if (irq->res != NULL) {
2120 ena_trace(ENA_INFO | ENA_IOQ, "release resource irq: %d\n",
2121 irq->vector);
2122 rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2123 irq->vector, irq->res);
2124 pci_release_msix_vector(adapter->pdev, irq->vector);
2125 irq->res = NULL;
2126 if (unlikely(rc != 0)) {
2127 device_printf(adapter->pdev, "dev has no parent"
2128 " while releasing res for irq: %d\n",
2129 irq->vector);
2130 }
2131 }
2132 }
2133 }
2134
2135 static void
ena_free_irqs(struct ena_adapter * adapter)2136 ena_free_irqs(struct ena_adapter* adapter)
2137 {
2138
2139 ena_free_io_irq(adapter);
2140 ena_free_mgmnt_irq(adapter);
2141 ena_disable_msix(adapter);
2142 }
2143
2144 static void
ena_disable_msix(struct ena_adapter * adapter)2145 ena_disable_msix(struct ena_adapter *adapter)
2146 {
2147
2148 pci_release_msi(adapter->pdev);
2149
2150 adapter->msix_vecs = 0;
2151 kfree(adapter->msix_entries, M_DEVBUF);
2152 adapter->msix_entries = NULL;
2153 }
2154
2155 static void
ena_unmask_all_io_irqs(struct ena_adapter * adapter)2156 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
2157 {
2158 struct ena_com_io_cq* io_cq;
2159 struct ena_eth_io_intr_reg intr_reg;
2160 uint16_t ena_qid;
2161 int i;
2162
2163 /* Unmask interrupts for all queues */
2164 for (i = 0; i < adapter->num_queues; i++) {
2165 ena_qid = ENA_IO_TXQ_IDX(i);
2166 io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
2167 ena_com_update_intr_reg(&intr_reg, 0, 0, true);
2168 ena_com_unmask_intr(io_cq, &intr_reg);
2169 }
2170 }
2171
2172 /* Configure the Rx forwarding */
2173 static int
ena_rss_configure(struct ena_adapter * adapter)2174 ena_rss_configure(struct ena_adapter *adapter)
2175 {
2176 struct ena_com_dev *ena_dev = adapter->ena_dev;
2177 int rc;
2178
2179 /* Set indirect table */
2180 rc = ena_com_indirect_table_set(ena_dev);
2181 if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2182 return (rc);
2183
2184 /* Configure hash function (if supported) */
2185 rc = ena_com_set_hash_function(ena_dev);
2186 if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2187 return (rc);
2188
2189 /* Configure hash inputs (if supported) */
2190 rc = ena_com_set_hash_ctrl(ena_dev);
2191 if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2192 return (rc);
2193
2194 return (0);
2195 }
2196
2197 static int
ena_up_complete(struct ena_adapter * adapter)2198 ena_up_complete(struct ena_adapter *adapter)
2199 {
2200 int rc;
2201
2202 if (likely(adapter->rss_support)) {
2203 rc = ena_rss_configure(adapter);
2204 if (rc != 0)
2205 return (rc);
2206 }
2207
2208 rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
2209 if (unlikely(rc != 0))
2210 return (rc);
2211
2212 ena_refill_all_rx_bufs(adapter);
2213 #if 0 /* XXX swildner counters */
2214 ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
2215 sizeof(adapter->hw_stats));
2216 #endif
2217
2218 return (0);
2219 }
2220
2221 static int
ena_up(struct ena_adapter * adapter)2222 ena_up(struct ena_adapter *adapter)
2223 {
2224 int rc = 0;
2225
2226 if (unlikely(device_is_attached(adapter->pdev) == 0)) {
2227 device_printf(adapter->pdev, "device is not attached!\n");
2228 return (ENXIO);
2229 }
2230
2231 if (unlikely(!adapter->running)) {
2232 device_printf(adapter->pdev, "device is not running!\n");
2233 return (ENXIO);
2234 }
2235
2236 if (!adapter->up) {
2237 device_printf(adapter->pdev, "device is going UP\n");
2238
2239 /* setup interrupts for IO queues */
2240 ena_setup_io_intr(adapter);
2241 rc = ena_request_io_irq(adapter);
2242 if (unlikely(rc != 0)) {
2243 ena_trace(ENA_ALERT, "err_req_irq");
2244 goto err_req_irq;
2245 }
2246
2247 /* allocate transmit descriptors */
2248 rc = ena_setup_all_tx_resources(adapter);
2249 if (unlikely(rc != 0)) {
2250 ena_trace(ENA_ALERT, "err_setup_tx");
2251 goto err_setup_tx;
2252 }
2253
2254 /* allocate receive descriptors */
2255 rc = ena_setup_all_rx_resources(adapter);
2256 if (unlikely(rc != 0)) {
2257 ena_trace(ENA_ALERT, "err_setup_rx");
2258 goto err_setup_rx;
2259 }
2260
2261 /* create IO queues for Rx & Tx */
2262 rc = ena_create_io_queues(adapter);
2263 if (unlikely(rc != 0)) {
2264 ena_trace(ENA_ALERT,
2265 "create IO queues failed");
2266 goto err_io_que;
2267 }
2268
2269 if (unlikely(adapter->link_status)) {
2270 adapter->ifp->if_link_state = LINK_STATE_UP;
2271 if_link_state_change(adapter->ifp);
2272 }
2273
2274 rc = ena_up_complete(adapter);
2275 if (unlikely(rc != 0))
2276 goto err_up_complete;
2277
2278 #if 0 /* XXX swildner counters */
2279 counter_u64_add(adapter->dev_stats.interface_up, 1);
2280 #endif
2281
2282 ena_update_hwassist(adapter);
2283
2284 adapter->ifp->if_flags |= IFF_RUNNING;
2285 ifq_clr_oactive(&adapter->ifp->if_snd);
2286
2287 callout_reset(&adapter->timer_service, hz,
2288 ena_timer_service, (void *)adapter);
2289
2290 adapter->up = true;
2291
2292 ena_unmask_all_io_irqs(adapter);
2293 }
2294
2295 return (0);
2296
2297 err_up_complete:
2298 ena_destroy_all_io_queues(adapter);
2299 err_io_que:
2300 ena_free_all_rx_resources(adapter);
2301 err_setup_rx:
2302 ena_free_all_tx_resources(adapter);
2303 err_setup_tx:
2304 ena_free_io_irq(adapter);
2305 err_req_irq:
2306 return (rc);
2307 }
2308
2309 #if 0 /* XXX swildner counters */
2310 static uint64_t
2311 ena_get_counter(if_t ifp, ift_counter cnt)
2312 {
2313 struct ena_adapter *adapter;
2314 struct ena_hw_stats *stats;
2315
2316 adapter = ifp->if_softc;
2317 stats = &adapter->hw_stats;
2318
2319 switch (cnt) {
2320 case IFCOUNTER_IPACKETS:
2321 return (counter_u64_fetch(stats->rx_packets));
2322 case IFCOUNTER_OPACKETS:
2323 return (counter_u64_fetch(stats->tx_packets));
2324 case IFCOUNTER_IBYTES:
2325 return (counter_u64_fetch(stats->rx_bytes));
2326 case IFCOUNTER_OBYTES:
2327 return (counter_u64_fetch(stats->tx_bytes));
2328 case IFCOUNTER_IQDROPS:
2329 return (counter_u64_fetch(stats->rx_drops));
2330 default:
2331 return (if_get_counter_default(ifp, cnt));
2332 }
2333 }
2334 #endif
2335
2336 static int
ena_media_change(if_t ifp)2337 ena_media_change(if_t ifp)
2338 {
2339 /* Media Change is not supported by firmware */
2340 return (0);
2341 }
2342
2343 static void
ena_media_status(if_t ifp,struct ifmediareq * ifmr)2344 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
2345 {
2346 struct ena_adapter *adapter = ifp->if_softc;
2347 ena_trace(ENA_DBG, "enter");
2348
2349 lockmgr(&adapter->global_lock, LK_EXCLUSIVE);
2350
2351 ifmr->ifm_status = IFM_AVALID;
2352 ifmr->ifm_active = IFM_ETHER;
2353
2354 if (!adapter->link_status) {
2355 lockmgr(&adapter->global_lock, LK_RELEASE);
2356 ena_trace(ENA_INFO, "link_status = false");
2357 return;
2358 }
2359
2360 ifmr->ifm_status |= IFM_ACTIVE;
2361 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
2362
2363 lockmgr(&adapter->global_lock, LK_RELEASE);
2364 }
2365
2366 static void
ena_init(void * arg)2367 ena_init(void *arg)
2368 {
2369 struct ena_adapter *adapter = (struct ena_adapter *)arg;
2370
2371 if (!adapter->up) {
2372 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
2373 ena_up(adapter);
2374 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
2375 }
2376 }
2377
2378 static int
ena_ioctl(if_t ifp,u_long command,caddr_t data,struct ucred * cred)2379 ena_ioctl(if_t ifp, u_long command, caddr_t data, struct ucred *cred)
2380 {
2381 struct ena_adapter *adapter;
2382 struct ifreq *ifr;
2383 int rc;
2384
2385 adapter = ifp->if_softc;
2386 ifr = (struct ifreq *)data;
2387
2388 /*
2389 * Acquiring lock to prevent from running up and down routines parallel.
2390 */
2391 rc = 0;
2392 switch (command) {
2393 case SIOCSIFMTU:
2394 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
2395 ena_down(adapter);
2396
2397 ena_change_mtu(ifp, ifr->ifr_mtu);
2398
2399 rc = ena_up(adapter);
2400 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
2401 break;
2402
2403 case SIOCSIFFLAGS:
2404 if ((ifp->if_flags & IFF_UP) != 0) {
2405 if ((ifp->if_flags & IFF_RUNNING) != 0) {
2406 if ((ifp->if_flags & (IFF_PROMISC |
2407 IFF_ALLMULTI)) != 0) {
2408 device_printf(adapter->pdev,
2409 "ioctl promisc/allmulti\n");
2410 }
2411 } else {
2412 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
2413 rc = ena_up(adapter);
2414 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
2415 }
2416 } else {
2417 if ((ifp->if_flags & IFF_RUNNING) != 0) {
2418 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
2419 ena_down(adapter);
2420 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
2421 }
2422 }
2423 break;
2424
2425 case SIOCADDMULTI:
2426 case SIOCDELMULTI:
2427 break;
2428
2429 case SIOCSIFMEDIA:
2430 case SIOCGIFMEDIA:
2431 rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
2432 break;
2433
2434 case SIOCSIFCAP:
2435 {
2436 int reinit = 0;
2437
2438 if (ifr->ifr_reqcap != ifp->if_capenable) {
2439 ifp->if_capenable = ifr->ifr_reqcap;
2440 reinit = 1;
2441 }
2442
2443 if ((reinit != 0) &&
2444 ((ifp->if_flags & IFF_RUNNING) != 0)) {
2445 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
2446 ena_down(adapter);
2447 rc = ena_up(adapter);
2448 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
2449 }
2450 }
2451
2452 break;
2453 default:
2454 rc = ether_ioctl(ifp, command, data);
2455 break;
2456 }
2457
2458 return (rc);
2459 }
2460
2461 static int
ena_get_dev_offloads(struct ena_com_dev_get_features_ctx * feat)2462 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2463 {
2464 int caps = 0;
2465
2466 if ((feat->offload.tx &
2467 (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2468 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2469 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2470 caps |= IFCAP_TXCSUM;
2471
2472 if ((feat->offload.tx &
2473 (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2474 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2475 caps |= IFCAP_TXCSUM;
2476
2477 if ((feat->offload.tx &
2478 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2479 caps |= IFCAP_TSO4;
2480
2481 if ((feat->offload.tx &
2482 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2483 caps |= IFCAP_TSO6;
2484
2485 if ((feat->offload.rx_supported &
2486 (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2487 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2488 caps |= IFCAP_RXCSUM;
2489
2490 #if 0
2491 if ((feat->offload.rx_supported &
2492 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2493 caps |= IFCAP_RXCSUM_IPV6;
2494 #endif
2495 #if 0 /* XXX LRO */
2496 caps |= IFCAP_LRO;
2497 #endif
2498 caps |= IFCAP_JUMBO_MTU;
2499
2500 return (caps);
2501 }
2502
2503 static void
ena_update_host_info(struct ena_admin_host_info * host_info,if_t ifp)2504 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2505 {
2506
2507 host_info->supported_network_features[0] =
2508 (uint32_t)ifp->if_capabilities;
2509 }
2510
2511 static void
ena_update_hwassist(struct ena_adapter * adapter)2512 ena_update_hwassist(struct ena_adapter *adapter)
2513 {
2514 if_t ifp = adapter->ifp;
2515 uint32_t feat = adapter->tx_offload_cap;
2516 int cap = ifp->if_capenable;
2517 int flags = 0;
2518
2519 ifp->if_hwassist = 0;
2520
2521 if ((cap & IFCAP_TXCSUM) != 0) {
2522 if ((feat &
2523 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2524 flags |= CSUM_IP;
2525 if ((feat &
2526 (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2527 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2528 flags |= CSUM_UDP | CSUM_TCP;
2529 }
2530
2531 #if 0
2532 if ((cap & IFCAP_TXCSUM_IPV6) != 0)
2533 flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2534 #endif
2535
2536 if ((cap & IFCAP_TSO4) != 0 || (cap & IFCAP_TSO6) != 0)
2537 flags |= CSUM_TSO;
2538
2539 ifp->if_hwassist |= flags;
2540 }
2541
2542 static int
ena_setup_ifnet(device_t pdev,struct ena_adapter * adapter,struct ena_com_dev_get_features_ctx * feat)2543 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2544 struct ena_com_dev_get_features_ctx *feat)
2545 {
2546 if_t ifp;
2547 int caps = 0;
2548
2549 ifp = adapter->ifp = if_alloc(IFT_ETHER);
2550 if (unlikely(ifp == NULL)) {
2551 ena_trace(ENA_ALERT, "can not allocate ifnet structure\n");
2552 return (ENXIO);
2553 }
2554 if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2555 ifp->if_softc = adapter;
2556
2557 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2558
2559 ifp->if_init = ena_init;
2560 ifp->if_start = ena_start_xmit;
2561 ifp->if_ioctl = ena_ioctl;
2562 #if 0 /* XXX swildner counter */
2563 if_setgetcounterfn(ifp, ena_get_counter);
2564 #endif
2565
2566 ifq_set_maxlen(&ifp->if_snd, adapter->tx_ring_size);
2567 ifq_set_ready(&ifp->if_snd);
2568 ifp->if_mtu = ETHERMTU;
2569 ifp->if_baudrate = 0;
2570 /* Zeroize capabilities... */
2571 ifp->if_capabilities = 0;
2572 ifp->if_capenable = 0;
2573 /* check hardware support */
2574 caps = ena_get_dev_offloads(feat);
2575 /* ... and set them */
2576 //if_setcapabilitiesbit(ifp, caps, 0);
2577 ((struct ifnet *)ifp)->if_capabilities |= caps;
2578 ((struct ifnet *)ifp)->if_capabilities &= ~0;
2579
2580 /* TSO parameters */
2581 //ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2582 // (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2583 ifp->if_tsolen = adapter->max_tx_sgl_size - 1;
2584 //ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2585
2586 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
2587 ifp->if_capenable= ifp->if_capabilities;
2588
2589 /*
2590 * Specify the media types supported by this adapter and register
2591 * callbacks to update media and link information
2592 */
2593 ifmedia_init(&adapter->media, IFM_IMASK,
2594 ena_media_change, ena_media_status);
2595 ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2596 ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2597
2598 ether_ifattach(ifp, adapter->mac_addr, NULL);
2599
2600 return (0);
2601 }
2602
2603 static void
ena_down(struct ena_adapter * adapter)2604 ena_down(struct ena_adapter *adapter)
2605 {
2606 int rc;
2607
2608 if (adapter->up) {
2609 device_printf(adapter->pdev, "device is going DOWN\n");
2610
2611 callout_drain(&adapter->timer_service);
2612
2613 adapter->up = false;
2614 ifq_set_oactive(&adapter->ifp->if_snd);
2615 adapter->ifp->if_flags &= ~IFF_RUNNING;
2616
2617 ena_free_io_irq(adapter);
2618
2619 if (adapter->trigger_reset) {
2620 rc = ena_com_dev_reset(adapter->ena_dev,
2621 adapter->reset_reason);
2622 if (unlikely(rc != 0))
2623 device_printf(adapter->pdev,
2624 "Device reset failed\n");
2625 }
2626
2627 ena_destroy_all_io_queues(adapter);
2628
2629 ena_free_all_tx_bufs(adapter);
2630 ena_free_all_rx_bufs(adapter);
2631 ena_free_all_tx_resources(adapter);
2632 ena_free_all_rx_resources(adapter);
2633
2634 #if 0 /* XXX swildner counters */
2635 counter_u64_add(adapter->dev_stats.interface_down, 1);
2636 #endif
2637 }
2638 }
2639
2640 static void
ena_tx_csum(struct ena_com_tx_ctx * ena_tx_ctx,struct mbuf * mbuf)2641 ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct mbuf *mbuf)
2642 {
2643 struct ena_com_tx_meta *ena_meta;
2644 struct ether_vlan_header *eh;
2645 u32 mss;
2646 bool offload;
2647 uint16_t etype;
2648 int ehdrlen;
2649 struct ip *ip;
2650 int iphlen;
2651 struct tcphdr *th;
2652
2653 offload = false;
2654 ena_meta = &ena_tx_ctx->ena_meta;
2655 mss = mbuf->m_pkthdr.tso_segsz;
2656
2657 if (mss != 0)
2658 offload = true;
2659
2660 if ((mbuf->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2661 offload = true;
2662
2663 if ((mbuf->m_pkthdr.csum_flags & CSUM_OFFLOAD) != 0)
2664 offload = true;
2665
2666 if (!offload) {
2667 ena_tx_ctx->meta_valid = 0;
2668 return;
2669 }
2670
2671 /* Determine where frame payload starts. */
2672 eh = mtod(mbuf, struct ether_vlan_header *);
2673 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2674 etype = ntohs(eh->evl_proto);
2675 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
2676 } else {
2677 etype = ntohs(eh->evl_encap_proto);
2678 ehdrlen = ETHER_HDR_LEN;
2679 }
2680
2681 ip = (struct ip *)(mbuf->m_data + ehdrlen);
2682 iphlen = ip->ip_hl << 2;
2683 th = (struct tcphdr *)((caddr_t)ip + iphlen);
2684
2685 if ((mbuf->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2686 ena_tx_ctx->l3_csum_enable = 1;
2687 }
2688 if ((mbuf->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2689 ena_tx_ctx->tso_enable = 1;
2690 ena_meta->l4_hdr_len = (th->th_off);
2691 }
2692
2693 switch (etype) {
2694 case ETHERTYPE_IP:
2695 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2696 if ((ip->ip_off & htons(IP_DF)) != 0)
2697 ena_tx_ctx->df = 1;
2698 break;
2699 case ETHERTYPE_IPV6:
2700 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2701
2702 default:
2703 break;
2704 }
2705
2706 if (ip->ip_p == IPPROTO_TCP) {
2707 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2708 if ((mbuf->m_pkthdr.csum_flags &
2709 CSUM_TCP) != 0)
2710 ena_tx_ctx->l4_csum_enable = 1;
2711 else
2712 ena_tx_ctx->l4_csum_enable = 0;
2713 } else if (ip->ip_p == IPPROTO_UDP) {
2714 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2715 if ((mbuf->m_pkthdr.csum_flags &
2716 CSUM_UDP) != 0)
2717 ena_tx_ctx->l4_csum_enable = 1;
2718 else
2719 ena_tx_ctx->l4_csum_enable = 0;
2720 } else {
2721 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
2722 ena_tx_ctx->l4_csum_enable = 0;
2723 }
2724
2725 ena_meta->mss = mss;
2726 ena_meta->l3_hdr_len = iphlen;
2727 ena_meta->l3_hdr_offset = ehdrlen;
2728 ena_tx_ctx->meta_valid = 1;
2729 }
2730
2731 static int
ena_check_and_collapse_mbuf(struct ena_ring * tx_ring,struct mbuf ** mbuf)2732 ena_check_and_collapse_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2733 {
2734 struct ena_adapter *adapter;
2735 struct mbuf *collapsed_mbuf;
2736 int num_frags;
2737
2738 adapter = tx_ring->adapter;
2739 num_frags = ena_mbuf_count(*mbuf);
2740
2741 /* One segment must be reserved for configuration descriptor. */
2742 if (num_frags < adapter->max_tx_sgl_size)
2743 return (0);
2744 #if 0 /* XXX swildner counters */
2745 counter_u64_add(tx_ring->tx_stats.collapse, 1);
2746 #endif
2747
2748 collapsed_mbuf = m_defrag(*mbuf, M_NOWAIT);
2749 if (unlikely(collapsed_mbuf == NULL)) {
2750 IFNET_STAT_INC(tx_ring->adapter->ifp, oerrors, 1);
2751 #if 0 /* XXX swildner counters */
2752 counter_u64_add(tx_ring->tx_stats.collapse_err, 1);
2753 #endif
2754 return (ENOMEM);
2755 }
2756
2757 /* If mbuf was collapsed succesfully, original mbuf is released. */
2758 *mbuf = collapsed_mbuf;
2759
2760 return (0);
2761 }
2762
2763 static int
ena_xmit_mbuf(struct ena_ring * tx_ring,struct mbuf ** mbuf)2764 ena_xmit_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2765 {
2766 struct ena_adapter *adapter;
2767 struct ena_tx_buffer *tx_info;
2768 struct ena_com_tx_ctx ena_tx_ctx;
2769 struct ena_com_dev *ena_dev;
2770 struct ena_com_buf *ena_buf;
2771 struct ena_com_io_sq* io_sq;
2772 bus_dma_segment_t segs[ENA_BUS_DMA_SEGS];
2773 void *push_hdr;
2774 uint16_t next_to_use;
2775 uint16_t req_id;
2776 uint16_t push_len;
2777 uint16_t ena_qid;
2778 uint32_t len, nsegs, header_len;
2779 int i, rc;
2780 int nb_hw_desc;
2781
2782 ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2783 adapter = tx_ring->que->adapter;
2784 ena_dev = adapter->ena_dev;
2785 io_sq = &ena_dev->io_sq_queues[ena_qid];
2786
2787 //tx_ring is just used to grab the adapter
2788 rc = ena_check_and_collapse_mbuf(tx_ring, mbuf);
2789 if (unlikely(rc != 0)) {
2790 ena_trace(ENA_WARNING,
2791 "Failed to collapse mbuf! err: %d", rc);
2792 return (rc);
2793 }
2794
2795 next_to_use = tx_ring->next_to_use;
2796 req_id = tx_ring->free_tx_ids[next_to_use];
2797 tx_info = &tx_ring->tx_buffer_info[req_id];
2798
2799 tx_info->mbuf = *mbuf;
2800 tx_info->num_of_bufs = 0;
2801
2802 ena_buf = tx_info->bufs;
2803 len = (*mbuf)->m_len;
2804
2805 ena_trace(ENA_DBG | ENA_TXPTH, "Tx: %d bytes", (*mbuf)->m_pkthdr.len);
2806
2807 push_len = 0;
2808 header_len = min_t(uint32_t, len, tx_ring->tx_max_header_size);
2809 push_hdr = NULL;
2810
2811 rc = bus_dmamap_load_mbuf_segment(adapter->tx_buf_tag, tx_info->map,
2812 *mbuf, segs, adapter->max_tx_sgl_size, &nsegs, BUS_DMA_NOWAIT);
2813
2814 if (unlikely((rc != 0) || (nsegs == 0))) {
2815 ena_trace(ENA_WARNING,
2816 "dmamap load failed! err: %d nsegs: %d", rc, nsegs);
2817 IFNET_STAT_INC(tx_ring->adapter->ifp, oerrors, 1);
2818 #if 0 /* XXX swildner counters */
2819 counter_u64_add(tx_ring->tx_stats.dma_mapping_err, 1);
2820 #endif
2821 tx_info->mbuf = NULL;
2822 if (rc == ENOMEM)
2823 return (ENA_COM_NO_MEM);
2824 else
2825 return (ENA_COM_INVAL);
2826 }
2827
2828 for (i = 0; i < nsegs; i++) {
2829 ena_buf->len = segs[i].ds_len;
2830 ena_buf->paddr = segs[i].ds_addr;
2831 ena_buf++;
2832 }
2833 tx_info->num_of_bufs = nsegs;
2834
2835 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2836 ena_tx_ctx.ena_bufs = tx_info->bufs;
2837 ena_tx_ctx.push_header = push_hdr;
2838 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2839 ena_tx_ctx.req_id = req_id;
2840 ena_tx_ctx.header_len = header_len;
2841
2842 /* Set flags and meta data */
2843 ena_tx_csum(&ena_tx_ctx, *mbuf);
2844 /* Prepare the packet's descriptors and send them to device */
2845 rc = ena_com_prepare_tx(io_sq, &ena_tx_ctx, &nb_hw_desc);
2846 if (unlikely(rc != 0)) {
2847 device_printf(adapter->pdev, "failed to prepare tx bufs\n");
2848 IFNET_STAT_INC(tx_ring->adapter->ifp, oerrors, 1);
2849 #if 0 /* XXX swildner counters */
2850 counter_u64_add(tx_ring->tx_stats.prepare_ctx_err, 1);
2851 #endif
2852 goto dma_error;
2853 }
2854
2855 IFNET_STAT_INC(tx_ring->adapter->ifp, opackets, 1);
2856 #if 0 /* XXX swildner counters */
2857 counter_enter();
2858 counter_u64_add_protected(tx_ring->tx_stats.cnt, 1);
2859 counter_u64_add_protected(tx_ring->tx_stats.bytes,
2860 (*mbuf)->m_pkthdr.len);
2861
2862 counter_u64_add_protected(adapter->hw_stats.tx_packets, 1);
2863 counter_u64_add_protected(adapter->hw_stats.tx_bytes,
2864 (*mbuf)->m_pkthdr.len);
2865 counter_exit();
2866 #endif
2867
2868 tx_info->tx_descs = nb_hw_desc;
2869 getmicrouptime(&tx_info->timestamp);
2870 tx_info->print_once = true;
2871
2872 tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2873 tx_ring->ring_size);
2874
2875 bus_dmamap_sync(adapter->tx_buf_tag, tx_info->map,
2876 BUS_DMASYNC_PREWRITE);
2877
2878 return (0);
2879
2880 dma_error:
2881 tx_info->mbuf = NULL;
2882 bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
2883
2884 return (rc);
2885 }
2886
2887 static void
ena_start_xmit(struct ifnet * ifp,struct ifaltq_subque * ifsq)2888 ena_start_xmit(struct ifnet *ifp, struct ifaltq_subque *ifsq)
2889 {
2890 /*
2891 * TODO: Might need to initialize an ena_ring with the
2892 * ifaltq_subque in it
2893 */
2894 struct ena_adapter *adapter = ifp->if_softc;
2895 struct ena_com_io_sq *io_sq;
2896 struct ena_ring *tx_ring;
2897 int ena_qid;
2898 int acum_pkts = 0;
2899 int ret = 0;
2900
2901 if (unlikely((adapter->ifp->if_flags & IFF_RUNNING) == 0) ||
2902 ifsq_is_oactive(ifsq)) {
2903 return;
2904 }
2905
2906 /* Check is link_active and some other shit. If it is, purge. */
2907
2908 #if 0
2909 if (unlikely(!adapter->link_status))
2910 return;
2911 #endif
2912
2913 io_sq = NULL;
2914 tx_ring = NULL;
2915
2916 while (!ifsq_is_empty(ifsq)) {
2917 struct mbuf *m_head;
2918 int i;
2919
2920 //Grab head from mbuf list
2921 m_head = ifsq_dequeue(ifsq);
2922 if (m_head == NULL)
2923 break;
2924
2925 //pick the associated tx_ring based on hash
2926 i = m_head->m_pkthdr.hash % adapter->num_queues;
2927
2928 tx_ring = &adapter->tx_ring[i];
2929 ENA_RING_MTX_LOCK(tx_ring);
2930 ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2931 io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
2932
2933 if (unlikely(!ena_com_sq_have_enough_space(io_sq, ENA_TX_CLEANUP_THRESHOLD)))
2934 ena_tx_cleanup(tx_ring);
2935
2936 if (unlikely((ret = ena_xmit_mbuf(tx_ring, &m_head)) != 0)) {
2937 if (ret == ENA_COM_NO_MEM) {
2938 /* XXX put mbuf back on queue */
2939 } else if (ret == ENA_COM_NO_SPACE) {
2940 /* XXX put mbuf back on queue */
2941 } else {
2942 m_freem(m_head);
2943 /* XXX advance mbuf queue aka move it forward? */
2944 }
2945 ENA_RING_MTX_UNLOCK(tx_ring);
2946 break;
2947 }
2948
2949 //advance mbuf queue, might already be handled by dequeue
2950
2951 #if 0
2952 // dillon - wtf is this doing here?
2953 // NOT SURE WHAT TO DO WITH THIS CODE
2954 if (unlikely((adapter->ifp->if_flags & IFF_RUNNING) == 0))
2955 return; // break here, not return. tx_ring locked
2956 #endif
2957
2958 acum_pkts++;
2959
2960 ENA_RING_MTX_UNLOCK(tx_ring);
2961 BPF_MTAP(adapter->ifp, m_head);
2962
2963 if (unlikely(acum_pkts == DB_THRESHOLD)) {
2964 acum_pkts = 0;
2965 wmb();
2966 /* Trigger the dma engine */
2967 ena_com_write_sq_doorbell(io_sq);
2968 #if 0 /* XXX swildner counters */
2969 counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2970 #endif
2971 }
2972
2973 }
2974
2975 if (likely(acum_pkts != 0)) {
2976 wmb();
2977 /* Trigger the dma engine */
2978 ena_com_write_sq_doorbell(io_sq);
2979 #if 0 /* XXX swildner counters */
2980 counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2981 #endif
2982 }
2983
2984 if (io_sq &&
2985 !ena_com_sq_have_enough_space(io_sq, ENA_TX_CLEANUP_THRESHOLD)) {
2986 ENA_RING_MTX_LOCK(tx_ring);
2987 ena_tx_cleanup(tx_ring);
2988 ENA_RING_MTX_UNLOCK(tx_ring);
2989 }
2990 }
2991
2992 static int
ena_calc_io_queue_num(struct ena_adapter * adapter,struct ena_com_dev_get_features_ctx * get_feat_ctx)2993 ena_calc_io_queue_num(struct ena_adapter *adapter,
2994 struct ena_com_dev_get_features_ctx *get_feat_ctx)
2995 {
2996 int io_sq_num, io_cq_num, io_queue_num;
2997
2998 io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2999 io_cq_num = get_feat_ctx->max_queues.max_cq_num;
3000
3001 io_queue_num = min_t(int, ncpus, ENA_MAX_NUM_IO_QUEUES);
3002 io_queue_num = min_t(int, io_queue_num, io_sq_num);
3003 io_queue_num = min_t(int, io_queue_num, io_cq_num);
3004 /* 1 IRQ for for mgmnt and 1 IRQ for each TX/RX pair */
3005 io_queue_num = min_t(int, io_queue_num,
3006 pci_msix_count(adapter->pdev) - 1);
3007 #ifdef RSS
3008 io_queue_num = min_t(int, io_queue_num, rss_getnumbuckets());
3009 #endif
3010
3011 return (io_queue_num);
3012 }
3013
3014 static int
ena_calc_queue_size(struct ena_adapter * adapter,uint16_t * max_tx_sgl_size,uint16_t * max_rx_sgl_size,struct ena_com_dev_get_features_ctx * feat)3015 ena_calc_queue_size(struct ena_adapter *adapter, uint16_t *max_tx_sgl_size,
3016 uint16_t *max_rx_sgl_size, struct ena_com_dev_get_features_ctx *feat)
3017 {
3018 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
3019 uint32_t v;
3020 uint32_t q;
3021
3022 queue_size = min_t(uint32_t, queue_size,
3023 feat->max_queues.max_cq_depth);
3024 queue_size = min_t(uint32_t, queue_size,
3025 feat->max_queues.max_sq_depth);
3026
3027 /* round down to the nearest power of 2 */
3028 v = queue_size;
3029 while (v != 0) {
3030 if (powerof2(queue_size) != 0)
3031 break;
3032 v /= 2;
3033 q = rounddown2(queue_size, v);
3034 if (q != 0) {
3035 queue_size = q;
3036 break;
3037 }
3038 }
3039
3040 if (unlikely(queue_size == 0)) {
3041 device_printf(adapter->pdev, "Invalid queue size\n");
3042 return (ENA_COM_FAULT);
3043 }
3044
3045 *max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3046 feat->max_queues.max_packet_tx_descs);
3047 *max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3048 feat->max_queues.max_packet_rx_descs);
3049
3050 return (queue_size);
3051 }
3052
3053 static int
ena_rss_init_default(struct ena_adapter * adapter)3054 ena_rss_init_default(struct ena_adapter *adapter)
3055 {
3056 struct ena_com_dev *ena_dev = adapter->ena_dev;
3057 device_t dev = adapter->pdev;
3058 int qid, rc, i;
3059
3060 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3061 if (unlikely(rc != 0)) {
3062 device_printf(dev, "Cannot init indirect table\n");
3063 return (rc);
3064 }
3065
3066 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3067 #ifdef RSS
3068 qid = rss_get_indirection_to_bucket(i);
3069 qid = qid % adapter->num_queues;
3070 #else
3071 qid = i % adapter->num_queues;
3072 #endif
3073 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3074 ENA_IO_RXQ_IDX(qid));
3075 if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3076 device_printf(dev, "Cannot fill indirect table\n");
3077 goto err_rss_destroy;
3078 }
3079 }
3080
3081 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3082 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3083 if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3084 device_printf(dev, "Cannot fill hash function\n");
3085 goto err_rss_destroy;
3086 }
3087
3088 rc = ena_com_set_default_hash_ctrl(ena_dev);
3089 if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3090 device_printf(dev, "Cannot fill hash control\n");
3091 goto err_rss_destroy;
3092 }
3093
3094 return (0);
3095
3096 err_rss_destroy:
3097 ena_com_rss_destroy(ena_dev);
3098 return (rc);
3099 }
3100
3101 static void
ena_rss_init_default_deferred(void * arg)3102 ena_rss_init_default_deferred(void *arg)
3103 {
3104 struct ena_adapter *adapter;
3105 devclass_t dc;
3106 int max;
3107 int rc;
3108
3109 dc = devclass_find("ena");
3110 if (unlikely(dc == NULL)) {
3111 ena_trace(ENA_ALERT, "No devclass ena\n");
3112 return;
3113 }
3114
3115 max = devclass_get_maxunit(dc);
3116 while (max-- >= 0) {
3117 adapter = devclass_get_softc(dc, max);
3118 if (adapter != NULL) {
3119 rc = ena_rss_init_default(adapter);
3120 adapter->rss_support = true;
3121 if (unlikely(rc != 0)) {
3122 device_printf(adapter->pdev,
3123 "WARNING: RSS was not properly initialized,"
3124 " it will affect bandwidth\n");
3125 adapter->rss_support = false;
3126 }
3127 }
3128 }
3129 }
3130 SYSINIT(ena_rss_init, SI_SUB_KICK_SCHEDULER, SI_ORDER_SECOND, ena_rss_init_default_deferred, NULL);
3131
3132 static void
ena_config_host_info(struct ena_com_dev * ena_dev)3133 ena_config_host_info(struct ena_com_dev *ena_dev)
3134 {
3135 struct ena_admin_host_info *host_info;
3136 int rc;
3137
3138 /* Allocate only the host info */
3139 rc = ena_com_allocate_host_info(ena_dev);
3140 if (unlikely(rc != 0)) {
3141 ena_trace(ENA_ALERT, "Cannot allocate host info\n");
3142 return;
3143 }
3144
3145 host_info = ena_dev->host_attr.host_info;
3146
3147 host_info->os_type = ENA_ADMIN_OS_FREEBSD;
3148 host_info->kernel_ver = osreldate;
3149
3150 ksprintf(host_info->kernel_ver_str, "%d", osreldate);
3151 host_info->os_dist = 0;
3152 strncpy(host_info->os_dist_str, osrelease,
3153 sizeof(host_info->os_dist_str) - 1);
3154
3155 host_info->driver_version =
3156 (DRV_MODULE_VER_MAJOR) |
3157 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3158 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
3159
3160 rc = ena_com_set_host_attributes(ena_dev);
3161 if (unlikely(rc != 0)) {
3162 if (rc == EOPNOTSUPP)
3163 ena_trace(ENA_WARNING, "Cannot set host attributes\n");
3164 else
3165 ena_trace(ENA_ALERT, "Cannot set host attributes\n");
3166
3167 goto err;
3168 }
3169
3170 return;
3171
3172 err:
3173 ena_com_delete_host_info(ena_dev);
3174 }
3175
3176 static int
ena_device_init(struct ena_adapter * adapter,device_t pdev,struct ena_com_dev_get_features_ctx * get_feat_ctx,int * wd_active)3177 ena_device_init(struct ena_adapter *adapter, device_t pdev,
3178 struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
3179 {
3180 struct ena_com_dev* ena_dev = adapter->ena_dev;
3181 bool readless_supported;
3182 uint32_t aenq_groups;
3183 int dma_width;
3184 int rc;
3185
3186 rc = ena_com_mmio_reg_read_request_init(ena_dev);
3187 if (unlikely(rc != 0)) {
3188 device_printf(pdev, "failed to init mmio read less\n");
3189 return (rc);
3190 }
3191
3192 /*
3193 * The PCIe configuration space revision id indicate if mmio reg
3194 * read is disabled
3195 */
3196 readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
3197 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3198
3199 rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3200 if (unlikely(rc != 0)) {
3201 device_printf(pdev, "Can not reset device\n");
3202 goto err_mmio_read_less;
3203 }
3204
3205 rc = ena_com_validate_version(ena_dev);
3206 if (unlikely(rc != 0)) {
3207 device_printf(pdev, "device version is too low\n");
3208 goto err_mmio_read_less;
3209 }
3210
3211 dma_width = ena_com_get_dma_width(ena_dev);
3212 if (unlikely(dma_width < 0)) {
3213 device_printf(pdev, "Invalid dma width value %d", dma_width);
3214 rc = dma_width;
3215 goto err_mmio_read_less;
3216 }
3217 adapter->dma_width = dma_width;
3218
3219 /* ENA admin level init */
3220 rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
3221 if (unlikely(rc != 0)) {
3222 device_printf(pdev,
3223 "Can not initialize ena admin queue with device\n");
3224 goto err_mmio_read_less;
3225 }
3226
3227 /*
3228 * To enable the msix interrupts the driver needs to know the number
3229 * of queues. So the driver uses polling mode to retrieve this
3230 * information
3231 */
3232 ena_com_set_admin_polling_mode(ena_dev, true);
3233
3234 ena_config_host_info(ena_dev);
3235
3236 /* Get Device Attributes */
3237 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3238 if (unlikely(rc != 0)) {
3239 device_printf(pdev,
3240 "Cannot get attribute for ena device rc: %d\n", rc);
3241 goto err_admin_init;
3242 }
3243
3244 aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) | BIT(ENA_ADMIN_KEEP_ALIVE);
3245
3246 aenq_groups &= get_feat_ctx->aenq.supported_groups;
3247 rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3248 if (unlikely(rc != 0)) {
3249 device_printf(pdev, "Cannot configure aenq groups rc: %d\n", rc);
3250 goto err_admin_init;
3251 }
3252
3253 *wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3254
3255 return (0);
3256
3257 err_admin_init:
3258 ena_com_delete_host_info(ena_dev);
3259 ena_com_admin_destroy(ena_dev);
3260 err_mmio_read_less:
3261 ena_com_mmio_reg_read_request_destroy(ena_dev);
3262
3263 return (rc);
3264 }
3265
ena_enable_msix_and_set_admin_interrupts(struct ena_adapter * adapter,int io_vectors)3266 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
3267 int io_vectors)
3268 {
3269 struct ena_com_dev *ena_dev = adapter->ena_dev;
3270 int rc;
3271
3272 rc = ena_enable_msix(adapter);
3273 if (unlikely(rc != 0)) {
3274 device_printf(adapter->pdev, "Error with MSI-X enablement\n");
3275 return (rc);
3276 }
3277
3278 ena_setup_mgmnt_intr(adapter);
3279
3280 rc = ena_request_mgmnt_irq(adapter);
3281 if (unlikely(rc != 0)) {
3282 device_printf(adapter->pdev, "Cannot setup mgmnt queue intr\n");
3283 goto err_disable_msix;
3284 }
3285
3286 pci_enable_msix(adapter->pdev);
3287
3288 ena_com_set_admin_polling_mode(ena_dev, false);
3289
3290 ena_com_admin_aenq_enable(ena_dev);
3291
3292 return (0);
3293
3294 err_disable_msix:
3295 ena_disable_msix(adapter);
3296
3297 return (rc);
3298 }
3299
3300 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
ena_keep_alive_wd(void * adapter_data,struct ena_admin_aenq_entry * aenq_e)3301 static void ena_keep_alive_wd(void *adapter_data,
3302 struct ena_admin_aenq_entry *aenq_e)
3303 {
3304 struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3305 struct ena_admin_aenq_keep_alive_desc *desc;
3306 struct timeval time;
3307 uint64_t rx_drops;
3308
3309 desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3310
3311 rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
3312 IFNET_STAT_INC(adapter->ifp, iqdrops, 1);
3313 #if 0 /* XXX swildner counters */
3314 counter_u64_zero(adapter->hw_stats.rx_drops);
3315 counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
3316 #endif
3317
3318 getmicrouptime(&time);
3319 atomic_store_rel_64(&adapter->keep_alive_timestamp.tv_sec, time.tv_sec);
3320 }
3321
3322 /* Check for keep alive expiration */
check_for_missing_keep_alive(struct ena_adapter * adapter)3323 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3324 {
3325 struct timeval timestamp, time;
3326
3327 if (adapter->wd_active == 0)
3328 return;
3329
3330 if (likely(adapter->keep_alive_timeout == 0))
3331 return;
3332
3333 timestamp.tv_sec = atomic_load_acq_64(&adapter->keep_alive_timestamp.tv_sec);
3334 getmicrouptime(&time);
3335 timevalsub(&time, ×tamp);
3336 if (unlikely(time.tv_sec > adapter->keep_alive_timeout)) {
3337 device_printf(adapter->pdev,
3338 "Keep alive watchdog timeout.\n");
3339 #if 0 /* XXX swildner counters */
3340 counter_u64_add(adapter->dev_stats.wd_expired, 1);
3341 #endif
3342 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3343 adapter->trigger_reset = true;
3344 }
3345 }
3346
3347 /* Check if admin queue is enabled */
check_for_admin_com_state(struct ena_adapter * adapter)3348 static void check_for_admin_com_state(struct ena_adapter *adapter)
3349 {
3350 if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) ==
3351 false)) {
3352 device_printf(adapter->pdev,
3353 "ENA admin queue is not in running state!\n");
3354 #if 0 /* XXX swildner counters */
3355 counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
3356 #endif
3357 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3358 adapter->trigger_reset = true;
3359 }
3360 }
3361
3362 static int
check_missing_comp_in_queue(struct ena_adapter * adapter,struct ena_ring * tx_ring)3363 check_missing_comp_in_queue(struct ena_adapter *adapter,
3364 struct ena_ring *tx_ring)
3365 {
3366 struct timeval curtime, time;
3367 struct ena_tx_buffer *tx_buf;
3368 uint32_t missed_tx = 0;
3369 int i;
3370
3371 getmicrouptime(&curtime);
3372
3373 for (i = 0; i < tx_ring->ring_size; i++) {
3374 tx_buf = &tx_ring->tx_buffer_info[i];
3375
3376 if (timevalisset(&tx_buf->timestamp) == 0)
3377 continue;
3378
3379 time = curtime;
3380 timevalsub(&time, &tx_buf->timestamp);
3381
3382 /* Check again if packet is still waiting */
3383 //WATCH: Might not be exactly comparable
3384 if (unlikely(time.tv_sec > adapter->missing_tx_timeout)) {
3385
3386 if (!tx_buf->print_once)
3387 ena_trace(ENA_WARNING, "Found a Tx that wasn't "
3388 "completed on time, qid %d, index %d.\n",
3389 tx_ring->qid, i);
3390
3391 tx_buf->print_once = true;
3392 missed_tx++;
3393 #if 0 /* XXX swildner counters */
3394 counter_u64_add(tx_ring->tx_stats.missing_tx_comp, 1);
3395 #endif
3396
3397 if (unlikely(missed_tx >
3398 adapter->missing_tx_threshold)) {
3399 device_printf(adapter->pdev,
3400 "The number of lost tx completion "
3401 "is above the threshold (%d > %d). "
3402 "Reset the device\n",
3403 missed_tx, adapter->missing_tx_threshold);
3404 adapter->reset_reason =
3405 ENA_REGS_RESET_MISS_TX_CMPL;
3406 adapter->trigger_reset = true;
3407 return (EIO);
3408 }
3409 }
3410 }
3411
3412 return (0);
3413 }
3414
3415 /*
3416 * Check for TX which were not completed on time.
3417 * Timeout is defined by "missing_tx_timeout".
3418 * Reset will be performed if number of incompleted
3419 * transactions exceeds "missing_tx_threshold".
3420 */
3421 static void
check_for_missing_tx_completions(struct ena_adapter * adapter)3422 check_for_missing_tx_completions(struct ena_adapter *adapter)
3423 {
3424 struct ena_ring *tx_ring;
3425 int i, budget, rc;
3426
3427 /* Make sure the driver doesn't turn the device in other process */
3428 rmb();
3429
3430 if (!adapter->up)
3431 return;
3432
3433 if (adapter->trigger_reset)
3434 return;
3435
3436 if (adapter->missing_tx_timeout == 0)
3437 return;
3438
3439 budget = adapter->missing_tx_max_queues;
3440
3441 for (i = adapter->next_monitored_tx_qid; i < adapter->num_queues; i++) {
3442 tx_ring = &adapter->tx_ring[i];
3443
3444 rc = check_missing_comp_in_queue(adapter, tx_ring);
3445 if (unlikely(rc != 0))
3446 return;
3447
3448 budget--;
3449 if (budget == 0) {
3450 i++;
3451 break;
3452 }
3453 }
3454
3455 adapter->next_monitored_tx_qid = i % adapter->num_queues;
3456 }
3457
3458 /* trigger deferred rx cleanup after 2 consecutive detections */
3459 #define EMPTY_RX_REFILL 2
3460 /* For the rare case where the device runs out of Rx descriptors and the
3461 * msix handler failed to refill new Rx descriptors (due to a lack of memory
3462 * for example).
3463 * This case will lead to a deadlock:
3464 * The device won't send interrupts since all the new Rx packets will be dropped
3465 * The msix handler won't allocate new Rx descriptors so the device won't be
3466 * able to send new packets.
3467 *
3468 * When such a situation is detected - execute rx cleanup task in another thread
3469 */
3470 static void
check_for_empty_rx_ring(struct ena_adapter * adapter)3471 check_for_empty_rx_ring(struct ena_adapter *adapter)
3472 {
3473 struct ena_ring *rx_ring;
3474 int i, refill_required;
3475
3476 if (!adapter->up)
3477 return;
3478
3479 if (adapter->trigger_reset)
3480 return;
3481
3482 for (i = 0; i < adapter->num_queues; i++) {
3483 rx_ring = &adapter->rx_ring[i];
3484
3485 refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
3486 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3487 rx_ring->empty_rx_queue++;
3488
3489 if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3490 #if 0 /* XXX swildner counters */
3491 counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3492 1);
3493 #endif
3494
3495 device_printf(adapter->pdev,
3496 "trigger refill for ring %d\n", i);
3497
3498 taskqueue_enqueue(rx_ring->cmpl_tq,
3499 &rx_ring->cmpl_task);
3500 rx_ring->empty_rx_queue = 0;
3501 }
3502 } else {
3503 rx_ring->empty_rx_queue = 0;
3504 }
3505 }
3506 }
3507
3508 static void
ena_timer_service(void * data)3509 ena_timer_service(void *data)
3510 {
3511 struct ena_adapter *adapter = (struct ena_adapter *)data;
3512 struct ena_admin_host_info *host_info =
3513 adapter->ena_dev->host_attr.host_info;
3514
3515 check_for_missing_keep_alive(adapter);
3516
3517 check_for_admin_com_state(adapter);
3518
3519 check_for_missing_tx_completions(adapter);
3520
3521 check_for_empty_rx_ring(adapter);
3522
3523 if (host_info != NULL)
3524 ena_update_host_info(host_info, adapter->ifp);
3525
3526 if (unlikely(adapter->trigger_reset)) {
3527 device_printf(adapter->pdev, "Trigger reset is on\n");
3528 taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
3529 return;
3530 }
3531
3532 /*
3533 * Schedule another timeout one second from now.
3534 */
3535 /* XXX swildner callout_schedule_sbt(&adapter->timer_service, SBT_1S, SBT_1S, 0); */
3536 callout_reset(&adapter->timer_service, hz, ena_timer_service,
3537 (void *)adapter);
3538 }
3539
3540 static void
ena_reset_task(void * arg,int pending)3541 ena_reset_task(void *arg, int pending)
3542 {
3543 struct ena_com_dev_get_features_ctx get_feat_ctx;
3544 struct ena_adapter *adapter = (struct ena_adapter *)arg;
3545 struct ena_com_dev *ena_dev = adapter->ena_dev;
3546 bool dev_up;
3547 int rc;
3548
3549 if (unlikely(!adapter->trigger_reset)) {
3550 device_printf(adapter->pdev,
3551 "device reset scheduled but trigger_reset is off\n");
3552 return;
3553 }
3554
3555 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
3556
3557 callout_drain(&adapter->timer_service);
3558
3559 dev_up = adapter->up;
3560
3561 ena_com_set_admin_running_state(ena_dev, false);
3562 ena_down(adapter);
3563 ena_free_mgmnt_irq(adapter);
3564 ena_disable_msix(adapter);
3565 ena_com_abort_admin_commands(ena_dev);
3566 ena_com_wait_for_abort_completion(ena_dev);
3567 ena_com_admin_destroy(ena_dev);
3568 ena_com_mmio_reg_read_request_destroy(ena_dev);
3569
3570 adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3571 adapter->trigger_reset = false;
3572
3573 /* Finished destroy part. Restart the device */
3574 rc = ena_device_init(adapter, adapter->pdev, &get_feat_ctx,
3575 &adapter->wd_active);
3576 if (unlikely(rc != 0)) {
3577 device_printf(adapter->pdev,
3578 "ENA device init failed! (err: %d)\n", rc);
3579 goto err_dev_free;
3580 }
3581
3582 rc = ena_enable_msix_and_set_admin_interrupts(adapter,
3583 adapter->num_queues);
3584 if (unlikely(rc != 0)) {
3585 device_printf(adapter->pdev, "Enable MSI-X failed\n");
3586 goto err_com_free;
3587 }
3588
3589 /* If the interface was up before the reset bring it up */
3590 if (dev_up) {
3591 rc = ena_up(adapter);
3592 if (unlikely(rc != 0)) {
3593 device_printf(adapter->pdev,
3594 "Failed to create I/O queues\n");
3595 goto err_msix_free;
3596 }
3597 }
3598
3599 callout_reset(&adapter->timer_service, hz,
3600 ena_timer_service, (void *)adapter);
3601
3602 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
3603
3604 return;
3605
3606 err_msix_free:
3607 ena_free_mgmnt_irq(adapter);
3608 ena_disable_msix(adapter);
3609 err_com_free:
3610 ena_com_admin_destroy(ena_dev);
3611 err_dev_free:
3612 device_printf(adapter->pdev, "ENA reset failed!\n");
3613 adapter->running = false;
3614 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
3615 }
3616
3617 /**
3618 * ena_attach - Device Initialization Routine
3619 * @pdev: device information struct
3620 *
3621 * Returns 0 on success, otherwise on failure.
3622 *
3623 * ena_attach initializes an adapter identified by a device structure.
3624 * The OS initialization, configuring of the adapter private structure,
3625 * and a hardware reset occur.
3626 **/
3627 static int
ena_attach(device_t pdev)3628 ena_attach(device_t pdev)
3629 {
3630 struct ena_com_dev_get_features_ctx get_feat_ctx;
3631 static int version_printed;
3632 struct ena_adapter *adapter;
3633 struct ena_com_dev *ena_dev = NULL;
3634 uint16_t tx_sgl_size = 0;
3635 uint16_t rx_sgl_size = 0;
3636 int io_queue_num;
3637 int queue_size;
3638 int rc;
3639 adapter = device_get_softc(pdev);
3640 adapter->pdev = pdev;
3641
3642 lockinit(&adapter->global_lock, "ENA global mtx", 0, LK_CANRECURSE);
3643 lockinit(&adapter->ioctl_lock, "ENA ioctl sx", 0, LK_CANRECURSE);
3644
3645 /* Set up the timer service */
3646 callout_init_lk(&adapter->timer_service, &adapter->global_lock);
3647 adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3648 adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3649 adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3650 adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3651
3652 if (version_printed++ == 0)
3653 device_printf(pdev, "%s\n", ena_version);
3654
3655 rc = ena_allocate_pci_resources(adapter);
3656 if (unlikely(rc != 0)) {
3657 device_printf(pdev, "PCI resource allocation failed!\n");
3658 ena_free_pci_resources(adapter);
3659 return (rc);
3660 }
3661
3662 /* Allocate memory for ena_dev structure */
3663 ena_dev = kmalloc(sizeof(struct ena_com_dev), M_DEVBUF,
3664 M_WAITOK | M_ZERO);
3665
3666 adapter->ena_dev = ena_dev;
3667 ena_dev->dmadev = pdev;
3668 ena_dev->bus = kmalloc(sizeof(struct ena_bus), M_DEVBUF,
3669 M_WAITOK | M_ZERO);
3670
3671 /* Store register resources */
3672 ((struct ena_bus*)(ena_dev->bus))->reg_bar_t =
3673 rman_get_bustag(adapter->registers);
3674 ((struct ena_bus*)(ena_dev->bus))->reg_bar_h =
3675 rman_get_bushandle(adapter->registers);
3676
3677 if (unlikely(((struct ena_bus*)(ena_dev->bus))->reg_bar_h == 0)) {
3678 device_printf(pdev, "failed to pmap registers bar\n");
3679 rc = ENXIO;
3680 goto err_bus_free;
3681 }
3682
3683 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3684
3685 /* Device initialization */
3686 rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3687 if (unlikely(rc != 0)) {
3688 device_printf(pdev, "ENA device init failed! (err: %d)\n", rc);
3689 rc = ENXIO;
3690 goto err_bus_free;
3691 }
3692
3693 getmicrouptime(&adapter->keep_alive_timestamp);
3694
3695 adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3696
3697 /* Set for sure that interface is not up */
3698 adapter->up = false;
3699
3700 memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3701 ETHER_ADDR_LEN);
3702
3703 /* calculate IO queue number to create */
3704 io_queue_num = ena_calc_io_queue_num(adapter, &get_feat_ctx);
3705
3706 ENA_ASSERT(io_queue_num > 0, "Invalid queue number: %d\n",
3707 io_queue_num);
3708 adapter->num_queues = io_queue_num;
3709
3710 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3711
3712 /* calculatre ring sizes */
3713 queue_size = ena_calc_queue_size(adapter,&tx_sgl_size,
3714 &rx_sgl_size, &get_feat_ctx);
3715 if (unlikely((queue_size <= 0) || (io_queue_num <= 0))) {
3716 rc = ENA_COM_FAULT;
3717 goto err_com_free;
3718 }
3719
3720 adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3721
3722 adapter->tx_ring_size = queue_size;
3723 adapter->rx_ring_size = queue_size;
3724
3725 adapter->max_tx_sgl_size = tx_sgl_size;
3726 adapter->max_rx_sgl_size = rx_sgl_size;
3727
3728 /* set up dma tags for rx and tx buffers */
3729 rc = ena_setup_tx_dma_tag(adapter);
3730 if (unlikely(rc != 0)) {
3731 device_printf(pdev, "Failed to create TX DMA tag\n");
3732 goto err_com_free;
3733 }
3734
3735 rc = ena_setup_rx_dma_tag(adapter);
3736 if (unlikely(rc != 0)) {
3737 device_printf(pdev, "Failed to create RX DMA tag\n");
3738 goto err_tx_tag_free;
3739 }
3740
3741 /* initialize rings basic information */
3742 device_printf(pdev, "initialize %d io queues\n", io_queue_num);
3743 ena_init_io_rings(adapter);
3744
3745 /* setup network interface */
3746 rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3747 if (unlikely(rc != 0)) {
3748 device_printf(pdev, "Error with network interface setup\n");
3749 goto err_io_free;
3750 }
3751
3752 rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3753 if (unlikely(rc != 0)) {
3754 device_printf(pdev,
3755 "Failed to enable and set the admin interrupts\n");
3756 goto err_ifp_free;
3757 }
3758
3759 /* Initialize reset task queue */
3760 TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3761 adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3762 M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3763 taskqueue_start_threads(&adapter->reset_tq, 1, TDPRI_KERN_DAEMON, -1,
3764 "%s rstq", device_get_nameunit(adapter->pdev));
3765
3766 /* Initialize statistics */
3767 #if 0 /* XXX swildner counters */
3768 ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3769 sizeof(struct ena_stats_dev));
3770 ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3771 sizeof(struct ena_hw_stats));
3772 #endif
3773 ena_sysctl_add_nodes(adapter);
3774
3775 /* Tell the stack that the interface is not active */
3776 ifq_set_oactive(&adapter->ifp->if_snd);
3777 adapter->ifp->if_flags &= ~IFF_RUNNING;
3778
3779 adapter->running = true;
3780 return (0);
3781
3782 err_ifp_free:
3783 if_detach(adapter->ifp);
3784 if_free(adapter->ifp);
3785 err_io_free:
3786 ena_free_all_io_rings_resources(adapter);
3787 ena_free_rx_dma_tag(adapter);
3788 err_tx_tag_free:
3789 ena_free_tx_dma_tag(adapter);
3790 err_com_free:
3791 ena_com_admin_destroy(ena_dev);
3792 ena_com_delete_host_info(ena_dev);
3793 ena_com_mmio_reg_read_request_destroy(ena_dev);
3794 err_bus_free:
3795 kfree(ena_dev->bus, M_DEVBUF);
3796 kfree(ena_dev, M_DEVBUF);
3797 ena_free_pci_resources(adapter);
3798
3799 return (rc);
3800 }
3801
3802 /**
3803 * ena_detach - Device Removal Routine
3804 * @pdev: device information struct
3805 *
3806 * ena_detach is called by the device subsystem to alert the driver
3807 * that it should release a PCI device.
3808 **/
3809 static int
ena_detach(device_t pdev)3810 ena_detach(device_t pdev)
3811 {
3812 struct ena_adapter *adapter = device_get_softc(pdev);
3813 struct ena_com_dev *ena_dev = adapter->ena_dev;
3814 int rc;
3815
3816 /* Make sure VLANS are not using driver */
3817 if (adapter->ifp->if_vlantrunks != NULL) {
3818 device_printf(adapter->pdev ,"VLAN is in use, detach first\n");
3819 return (EBUSY);
3820 }
3821
3822 /* Free reset task and callout */
3823 callout_drain(&adapter->timer_service);
3824 while (taskqueue_cancel(adapter->reset_tq, &adapter->reset_task, NULL))
3825 taskqueue_drain(adapter->reset_tq, &adapter->reset_task);
3826 taskqueue_free(adapter->reset_tq);
3827
3828 lockmgr(&adapter->ioctl_lock, LK_EXCLUSIVE);
3829 ena_down(adapter);
3830 lockmgr(&adapter->ioctl_lock, LK_RELEASE);
3831
3832 if (adapter->ifp != NULL) {
3833 ether_ifdetach(adapter->ifp);
3834 if_free(adapter->ifp);
3835 }
3836
3837 ena_free_all_io_rings_resources(adapter);
3838
3839 #if 0 /* XXX swildner counters */
3840 ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3841 sizeof(struct ena_hw_stats));
3842 ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3843 sizeof(struct ena_stats_dev));
3844 #endif
3845
3846 if (likely(adapter->rss_support))
3847 ena_com_rss_destroy(ena_dev);
3848
3849 rc = ena_free_rx_dma_tag(adapter);
3850 if (unlikely(rc != 0))
3851 device_printf(adapter->pdev,
3852 "Unmapped RX DMA tag associations\n");
3853
3854 rc = ena_free_tx_dma_tag(adapter);
3855 if (unlikely(rc != 0))
3856 device_printf(adapter->pdev,
3857 "Unmapped TX DMA tag associations\n");
3858
3859 /* Reset the device only if the device is running. */
3860 if (adapter->running)
3861 ena_com_dev_reset(ena_dev, adapter->reset_reason);
3862
3863 ena_com_delete_host_info(ena_dev);
3864
3865 ena_free_irqs(adapter);
3866
3867 ena_com_abort_admin_commands(ena_dev);
3868
3869 ena_com_wait_for_abort_completion(ena_dev);
3870
3871 ena_com_admin_destroy(ena_dev);
3872
3873 ena_com_mmio_reg_read_request_destroy(ena_dev);
3874
3875 ena_free_pci_resources(adapter);
3876
3877 lockuninit(&adapter->global_lock);
3878 lockuninit(&adapter->ioctl_lock);
3879
3880 if (ena_dev->bus != NULL)
3881 kfree(ena_dev->bus, M_DEVBUF);
3882
3883 if (ena_dev != NULL)
3884 kfree(ena_dev, M_DEVBUF);
3885
3886 return (bus_generic_detach(pdev));
3887 }
3888
3889 /******************************************************************************
3890 ******************************** AENQ Handlers *******************************
3891 *****************************************************************************/
3892 /**
3893 * ena_update_on_link_change:
3894 * Notify the network interface about the change in link status
3895 **/
3896 static void
ena_update_on_link_change(void * adapter_data,struct ena_admin_aenq_entry * aenq_e)3897 ena_update_on_link_change(void *adapter_data,
3898 struct ena_admin_aenq_entry *aenq_e)
3899 {
3900 struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3901 struct ena_admin_aenq_link_change_desc *aenq_desc;
3902 int status;
3903 if_t ifp;
3904
3905 aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3906 ifp = adapter->ifp;
3907 status = aenq_desc->flags &
3908 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3909
3910 if (status != 0) {
3911 device_printf(adapter->pdev, "link is UP\n");
3912 ifp->if_link_state = LINK_STATE_UP;
3913 if_link_state_change(ifp);
3914 } else if (status == 0) {
3915 device_printf(adapter->pdev, "link is DOWN\n");
3916 ifp->if_link_state = LINK_STATE_DOWN;
3917 if_link_state_change(ifp);
3918 } else {
3919 device_printf(adapter->pdev, "invalid value recvd\n");
3920 BUG();
3921 }
3922
3923 adapter->link_status = status;
3924 }
3925
3926 /**
3927 * This handler will called for unknown event group or unimplemented handlers
3928 **/
3929 static void
unimplemented_aenq_handler(void * data,struct ena_admin_aenq_entry * aenq_e)3930 unimplemented_aenq_handler(void *data,
3931 struct ena_admin_aenq_entry *aenq_e)
3932 {
3933 return;
3934 }
3935
3936 static struct ena_aenq_handlers aenq_handlers = {
3937 .handlers = {
3938 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3939 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3940 },
3941 .unimplemented_handler = unimplemented_aenq_handler
3942 };
3943
3944 /*********************************************************************
3945 * FreeBSD Device Interface Entry Points
3946 *********************************************************************/
3947
3948 static device_method_t ena_methods[] = {
3949 /* Device interface */
3950 DEVMETHOD(device_probe, ena_probe),
3951 DEVMETHOD(device_attach, ena_attach),
3952 DEVMETHOD(device_detach, ena_detach),
3953 DEVMETHOD_END
3954 };
3955
3956 static driver_t ena_driver = {
3957 "ena", ena_methods, sizeof(struct ena_adapter),
3958 };
3959
3960 devclass_t ena_devclass;
3961 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, NULL, NULL);
3962 MODULE_DEPEND(ena, pci, 1, 1, 1);
3963 MODULE_DEPEND(ena, ether, 1, 1, 1);
3964
3965 /*********************************************************************/
3966