1 /******************************************************************************
2
3 Copyright (c) 2013-2018, Intel Corporation
4 All rights reserved.
5
6 Redistribution and use in source and binary forms, with or without
7 modification, are permitted provided that the following conditions are met:
8
9 1. Redistributions of source code must retain the above copyright notice,
10 this list of conditions and the following disclaimer.
11
12 2. Redistributions in binary form must reproduce the above copyright
13 notice, this list of conditions and the following disclaimer in the
14 documentation and/or other materials provided with the distribution.
15
16 3. Neither the name of the Intel Corporation nor the names of its
17 contributors may be used to endorse or promote products derived from
18 this software without specific prior written permission.
19
20 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
21 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
24 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 POSSIBILITY OF SUCH DAMAGE.
31
32 ******************************************************************************/
33
34 /*
35 ** IXL driver TX/RX Routines:
36 ** This was seperated to allow usage by
37 ** both the PF and VF drivers.
38 */
39
40 #ifndef IXL_STANDALONE_BUILD
41 #include "opt_inet.h"
42 #include "opt_inet6.h"
43 #include "opt_rss.h"
44 #endif
45
46 #include "ixl.h"
47
48 #ifdef RSS
49 #include <net/rss_config.h>
50 #endif
51
52 /* Local Prototypes */
53 static u8 ixl_rx_checksum(if_rxd_info_t ri, u32 status, u32 error, u8 ptype);
54
55 static int ixl_isc_txd_encap(void *arg, if_pkt_info_t pi);
56 static void ixl_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx);
57 static int ixl_isc_txd_credits_update_hwb(void *arg, uint16_t txqid, bool clear);
58 static int ixl_isc_txd_credits_update_dwb(void *arg, uint16_t txqid, bool clear);
59
60 static void ixl_isc_rxd_refill(void *arg, if_rxd_update_t iru);
61 static void ixl_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused,
62 qidx_t pidx);
63 static int ixl_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx,
64 qidx_t budget);
65 static int ixl_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri);
66
67 struct if_txrx ixl_txrx_hwb = {
68 ixl_isc_txd_encap,
69 ixl_isc_txd_flush,
70 ixl_isc_txd_credits_update_hwb,
71 ixl_isc_rxd_available,
72 ixl_isc_rxd_pkt_get,
73 ixl_isc_rxd_refill,
74 ixl_isc_rxd_flush,
75 NULL
76 };
77
78 struct if_txrx ixl_txrx_dwb = {
79 ixl_isc_txd_encap,
80 ixl_isc_txd_flush,
81 ixl_isc_txd_credits_update_dwb,
82 ixl_isc_rxd_available,
83 ixl_isc_rxd_pkt_get,
84 ixl_isc_rxd_refill,
85 ixl_isc_rxd_flush,
86 NULL
87 };
88
89 /*
90 * @key key is saved into this parameter
91 */
92 void
ixl_get_default_rss_key(u32 * key)93 ixl_get_default_rss_key(u32 *key)
94 {
95 MPASS(key != NULL);
96
97 u32 rss_seed[IXL_RSS_KEY_SIZE_REG] = {0x41b01687,
98 0x183cfd8c, 0xce880440, 0x580cbc3c,
99 0x35897377, 0x328b25e1, 0x4fa98922,
100 0xb7d90c14, 0xd5bad70d, 0xcd15a2c1,
101 0x0, 0x0, 0x0};
102
103 bcopy(rss_seed, key, IXL_RSS_KEY_SIZE);
104 }
105
106 /**
107 * i40e_vc_stat_str - convert virtchnl status err code to a string
108 * @hw: pointer to the HW structure
109 * @stat_err: the status error code to convert
110 **/
111 const char *
i40e_vc_stat_str(struct i40e_hw * hw,enum virtchnl_status_code stat_err)112 i40e_vc_stat_str(struct i40e_hw *hw, enum virtchnl_status_code stat_err)
113 {
114 switch (stat_err) {
115 case VIRTCHNL_STATUS_SUCCESS:
116 return "OK";
117 case VIRTCHNL_ERR_PARAM:
118 return "VIRTCHNL_ERR_PARAM";
119 case VIRTCHNL_STATUS_ERR_OPCODE_MISMATCH:
120 return "VIRTCHNL_STATUS_ERR_OPCODE_MISMATCH";
121 case VIRTCHNL_STATUS_ERR_CQP_COMPL_ERROR:
122 return "VIRTCHNL_STATUS_ERR_CQP_COMPL_ERROR";
123 case VIRTCHNL_STATUS_ERR_INVALID_VF_ID:
124 return "VIRTCHNL_STATUS_ERR_INVALID_VF_ID";
125 case VIRTCHNL_STATUS_NOT_SUPPORTED:
126 return "VIRTCHNL_STATUS_NOT_SUPPORTED";
127 }
128
129 snprintf(hw->err_str, sizeof(hw->err_str), "%d", stat_err);
130 return hw->err_str;
131 }
132
133 void
ixl_debug_core(device_t dev,u32 enabled_mask,u32 mask,char * fmt,...)134 ixl_debug_core(device_t dev, u32 enabled_mask, u32 mask, char *fmt, ...)
135 {
136 va_list args;
137
138 if (!(mask & enabled_mask))
139 return;
140
141 /* Re-implement device_printf() */
142 device_print_prettyname(dev);
143 va_start(args, fmt);
144 vprintf(fmt, args);
145 va_end(args);
146 }
147
148 static bool
ixl_is_tx_desc_done(struct tx_ring * txr,int idx)149 ixl_is_tx_desc_done(struct tx_ring *txr, int idx)
150 {
151 return (((txr->tx_base[idx].cmd_type_offset_bsz >> I40E_TXD_QW1_DTYPE_SHIFT)
152 & I40E_TXD_QW1_DTYPE_MASK) == I40E_TX_DESC_DTYPE_DESC_DONE);
153 }
154
155 static int
ixl_tso_detect_sparse(bus_dma_segment_t * segs,int nsegs,if_pkt_info_t pi)156 ixl_tso_detect_sparse(bus_dma_segment_t *segs, int nsegs, if_pkt_info_t pi)
157 {
158 int count, curseg, i, hlen, segsz, seglen, tsolen;
159
160 if (nsegs <= IXL_MAX_TX_SEGS-2)
161 return (0);
162 segsz = pi->ipi_tso_segsz;
163 curseg = count = 0;
164
165 hlen = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen;
166 tsolen = pi->ipi_len - hlen;
167
168 i = 0;
169 curseg = segs[0].ds_len;
170 while (hlen > 0) {
171 count++;
172 if (count > IXL_MAX_TX_SEGS - 2)
173 return (1);
174 if (curseg == 0) {
175 i++;
176 if (__predict_false(i == nsegs))
177 return (1);
178
179 curseg = segs[i].ds_len;
180 }
181 seglen = min(curseg, hlen);
182 curseg -= seglen;
183 hlen -= seglen;
184 // printf("H:seglen = %d, count=%d\n", seglen, count);
185 }
186 while (tsolen > 0) {
187 segsz = pi->ipi_tso_segsz;
188 while (segsz > 0 && tsolen != 0) {
189 count++;
190 if (count > IXL_MAX_TX_SEGS - 2) {
191 // printf("bad: count = %d\n", count);
192 return (1);
193 }
194 if (curseg == 0) {
195 i++;
196 if (__predict_false(i == nsegs)) {
197 // printf("bad: tsolen = %d", tsolen);
198 return (1);
199 }
200 curseg = segs[i].ds_len;
201 }
202 seglen = min(curseg, segsz);
203 segsz -= seglen;
204 curseg -= seglen;
205 tsolen -= seglen;
206 // printf("D:seglen = %d, count=%d\n", seglen, count);
207 }
208 count = 0;
209 }
210
211 return (0);
212 }
213
214 /*********************************************************************
215 *
216 * Setup descriptor for hw offloads
217 *
218 **********************************************************************/
219
220 static void
ixl_tx_setup_offload(struct ixl_tx_queue * que,if_pkt_info_t pi,u32 * cmd,u32 * off)221 ixl_tx_setup_offload(struct ixl_tx_queue *que,
222 if_pkt_info_t pi, u32 *cmd, u32 *off)
223 {
224 switch (pi->ipi_etype) {
225 #ifdef INET
226 case ETHERTYPE_IP:
227 if (pi->ipi_csum_flags & IXL_CSUM_IPV4)
228 *cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
229 else
230 *cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
231 break;
232 #endif
233 #ifdef INET6
234 case ETHERTYPE_IPV6:
235 *cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
236 break;
237 #endif
238 default:
239 break;
240 }
241
242 *off |= (pi->ipi_ehdrlen >> 1) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
243 *off |= (pi->ipi_ip_hlen >> 2) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
244
245 switch (pi->ipi_ipproto) {
246 case IPPROTO_TCP:
247 if (pi->ipi_csum_flags & IXL_CSUM_TCP) {
248 *cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
249 *off |= (pi->ipi_tcp_hlen >> 2) <<
250 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
251 /* Check for NO_HEAD MDD event */
252 MPASS(pi->ipi_tcp_hlen != 0);
253 }
254 break;
255 case IPPROTO_UDP:
256 if (pi->ipi_csum_flags & IXL_CSUM_UDP) {
257 *cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
258 *off |= (sizeof(struct udphdr) >> 2) <<
259 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
260 }
261 break;
262 case IPPROTO_SCTP:
263 if (pi->ipi_csum_flags & IXL_CSUM_SCTP) {
264 *cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
265 *off |= (sizeof(struct sctphdr) >> 2) <<
266 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
267 }
268 /* Fall Thru */
269 default:
270 break;
271 }
272 }
273
274 /**********************************************************************
275 *
276 * Setup context for hardware segmentation offload (TSO)
277 *
278 **********************************************************************/
279 static int
ixl_tso_setup(struct tx_ring * txr,if_pkt_info_t pi)280 ixl_tso_setup(struct tx_ring *txr, if_pkt_info_t pi)
281 {
282 if_softc_ctx_t scctx;
283 struct i40e_tx_context_desc *TXD;
284 u32 cmd, mss, type, tsolen;
285 int idx, total_hdr_len;
286 u64 type_cmd_tso_mss;
287
288 idx = pi->ipi_pidx;
289 TXD = (struct i40e_tx_context_desc *) &txr->tx_base[idx];
290 total_hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen;
291 tsolen = pi->ipi_len - total_hdr_len;
292 scctx = txr->que->vsi->shared;
293
294 type = I40E_TX_DESC_DTYPE_CONTEXT;
295 cmd = I40E_TX_CTX_DESC_TSO;
296 /*
297 * TSO MSS must not be less than 64; this prevents a
298 * BAD_LSO_MSS MDD event when the MSS is too small.
299 */
300 if (pi->ipi_tso_segsz < IXL_MIN_TSO_MSS) {
301 txr->mss_too_small++;
302 pi->ipi_tso_segsz = IXL_MIN_TSO_MSS;
303 }
304 mss = pi->ipi_tso_segsz;
305
306 /* Check for BAD_LS0_MSS MDD event (mss too large) */
307 MPASS(mss <= IXL_MAX_TSO_MSS);
308 /* Check for NO_HEAD MDD event (header lengths are 0) */
309 MPASS(pi->ipi_ehdrlen != 0);
310 MPASS(pi->ipi_ip_hlen != 0);
311 /* Partial check for BAD_LSO_LEN MDD event */
312 MPASS(tsolen != 0);
313 /* Partial check for WRONG_SIZE MDD event (during TSO) */
314 MPASS(total_hdr_len + mss <= IXL_MAX_FRAME);
315
316 type_cmd_tso_mss = ((u64)type << I40E_TXD_CTX_QW1_DTYPE_SHIFT) |
317 ((u64)cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
318 ((u64)tsolen << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
319 ((u64)mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
320 TXD->type_cmd_tso_mss = htole64(type_cmd_tso_mss);
321
322 TXD->tunneling_params = htole32(0);
323 txr->que->tso++;
324
325 return ((idx + 1) & (scctx->isc_ntxd[0]-1));
326 }
327
328 /*********************************************************************
329 *
330 * This routine maps the mbufs to tx descriptors, allowing the
331 * TX engine to transmit the packets.
332 * - return 0 on success, positive on failure
333 *
334 **********************************************************************/
335 #define IXL_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
336
337 static int
ixl_isc_txd_encap(void * arg,if_pkt_info_t pi)338 ixl_isc_txd_encap(void *arg, if_pkt_info_t pi)
339 {
340 struct ixl_vsi *vsi = arg;
341 if_softc_ctx_t scctx = vsi->shared;
342 struct ixl_tx_queue *que = &vsi->tx_queues[pi->ipi_qsidx];
343 struct tx_ring *txr = &que->txr;
344 int nsegs = pi->ipi_nsegs;
345 bus_dma_segment_t *segs = pi->ipi_segs;
346 struct i40e_tx_desc *txd = NULL;
347 int i, j, mask, pidx_last;
348 u32 cmd, off, tx_intr;
349
350 cmd = off = 0;
351 i = pi->ipi_pidx;
352
353 tx_intr = (pi->ipi_flags & IPI_TX_INTR);
354
355 /* Set up the TSO/CSUM offload */
356 if (pi->ipi_csum_flags & CSUM_OFFLOAD) {
357 /* Set up the TSO context descriptor if required */
358 if (pi->ipi_csum_flags & CSUM_TSO) {
359 /* Prevent MAX_BUFF MDD event (for TSO) */
360 if (ixl_tso_detect_sparse(segs, nsegs, pi))
361 return (EFBIG);
362 i = ixl_tso_setup(txr, pi);
363 }
364 ixl_tx_setup_offload(que, pi, &cmd, &off);
365 }
366 if (pi->ipi_mflags & M_VLANTAG)
367 cmd |= I40E_TX_DESC_CMD_IL2TAG1;
368
369 cmd |= I40E_TX_DESC_CMD_ICRC;
370 mask = scctx->isc_ntxd[0] - 1;
371 /* Check for WRONG_SIZE MDD event */
372 MPASS(pi->ipi_len >= IXL_MIN_FRAME);
373 #ifdef INVARIANTS
374 if (!(pi->ipi_csum_flags & CSUM_TSO))
375 MPASS(pi->ipi_len <= IXL_MAX_FRAME);
376 #endif
377 for (j = 0; j < nsegs; j++) {
378 bus_size_t seglen;
379
380 txd = &txr->tx_base[i];
381 seglen = segs[j].ds_len;
382
383 /* Check for ZERO_BSIZE MDD event */
384 MPASS(seglen != 0);
385
386 txd->buffer_addr = htole64(segs[j].ds_addr);
387 txd->cmd_type_offset_bsz =
388 htole64(I40E_TX_DESC_DTYPE_DATA
389 | ((u64)cmd << I40E_TXD_QW1_CMD_SHIFT)
390 | ((u64)off << I40E_TXD_QW1_OFFSET_SHIFT)
391 | ((u64)seglen << I40E_TXD_QW1_TX_BUF_SZ_SHIFT)
392 | ((u64)htole16(pi->ipi_vtag) << I40E_TXD_QW1_L2TAG1_SHIFT));
393
394 txr->tx_bytes += seglen;
395 pidx_last = i;
396 i = (i+1) & mask;
397 }
398 /* Set the last descriptor for report */
399 txd->cmd_type_offset_bsz |=
400 htole64(((u64)IXL_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT));
401 /* Add to report status array (if using TX interrupts) */
402 if (!vsi->enable_head_writeback && tx_intr) {
403 txr->tx_rsq[txr->tx_rs_pidx] = pidx_last;
404 txr->tx_rs_pidx = (txr->tx_rs_pidx+1) & mask;
405 MPASS(txr->tx_rs_pidx != txr->tx_rs_cidx);
406 }
407 pi->ipi_new_pidx = i;
408
409 ++txr->tx_packets;
410 return (0);
411 }
412
413 static void
ixl_isc_txd_flush(void * arg,uint16_t txqid,qidx_t pidx)414 ixl_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx)
415 {
416 struct ixl_vsi *vsi = arg;
417 struct tx_ring *txr = &vsi->tx_queues[txqid].txr;
418
419 /*
420 * Advance the Transmit Descriptor Tail (Tdt), this tells the
421 * hardware that this frame is available to transmit.
422 */
423 /* Check for ENDLESS_TX MDD event */
424 MPASS(pidx < vsi->shared->isc_ntxd[0]);
425 wr32(vsi->hw, txr->tail, pidx);
426 }
427
428
429 /*********************************************************************
430 *
431 * (Re)Initialize a queue transmit ring by clearing its memory.
432 *
433 **********************************************************************/
434 void
ixl_init_tx_ring(struct ixl_vsi * vsi,struct ixl_tx_queue * que)435 ixl_init_tx_ring(struct ixl_vsi *vsi, struct ixl_tx_queue *que)
436 {
437 struct tx_ring *txr = &que->txr;
438
439 /* Clear the old ring contents */
440 bzero((void *)txr->tx_base,
441 (sizeof(struct i40e_tx_desc)) *
442 (vsi->shared->isc_ntxd[0] + (vsi->enable_head_writeback ? 1 : 0)));
443
444 wr32(vsi->hw, txr->tail, 0);
445 }
446
447 /*
448 * ixl_get_tx_head - Retrieve the value from the
449 * location the HW records its HEAD index
450 */
451 static inline u32
ixl_get_tx_head(struct ixl_tx_queue * que)452 ixl_get_tx_head(struct ixl_tx_queue *que)
453 {
454 if_softc_ctx_t scctx = que->vsi->shared;
455 struct tx_ring *txr = &que->txr;
456 void *head = &txr->tx_base[scctx->isc_ntxd[0]];
457
458 return LE32_TO_CPU(*(volatile __le32 *)head);
459 }
460
461 static int
ixl_isc_txd_credits_update_hwb(void * arg,uint16_t qid,bool clear)462 ixl_isc_txd_credits_update_hwb(void *arg, uint16_t qid, bool clear)
463 {
464 struct ixl_vsi *vsi = arg;
465 if_softc_ctx_t scctx = vsi->shared;
466 struct ixl_tx_queue *que = &vsi->tx_queues[qid];
467 struct tx_ring *txr = &que->txr;
468 int head, credits;
469
470 /* Get the Head WB value */
471 head = ixl_get_tx_head(que);
472
473 credits = head - txr->tx_cidx_processed;
474 if (credits < 0)
475 credits += scctx->isc_ntxd[0];
476 if (clear)
477 txr->tx_cidx_processed = head;
478
479 return (credits);
480 }
481
482 static int
ixl_isc_txd_credits_update_dwb(void * arg,uint16_t txqid,bool clear)483 ixl_isc_txd_credits_update_dwb(void *arg, uint16_t txqid, bool clear)
484 {
485 struct ixl_vsi *vsi = arg;
486 struct ixl_tx_queue *tx_que = &vsi->tx_queues[txqid];
487 if_softc_ctx_t scctx = vsi->shared;
488 struct tx_ring *txr = &tx_que->txr;
489
490 qidx_t processed = 0;
491 qidx_t cur, prev, ntxd, rs_cidx;
492 int32_t delta;
493 bool is_done;
494
495 rs_cidx = txr->tx_rs_cidx;
496 #if 0
497 device_printf(iflib_get_dev(vsi->ctx), "%s: (q%d) rs_cidx %d, txr->tx_rs_pidx %d\n", __func__,
498 txr->me, rs_cidx, txr->tx_rs_pidx);
499 #endif
500 if (rs_cidx == txr->tx_rs_pidx)
501 return (0);
502 cur = txr->tx_rsq[rs_cidx];
503 MPASS(cur != QIDX_INVALID);
504 is_done = ixl_is_tx_desc_done(txr, cur);
505
506 if (!is_done)
507 return (0);
508
509 /* If clear is false just let caller know that there
510 * are descriptors to reclaim */
511 if (!clear)
512 return (1);
513
514 prev = txr->tx_cidx_processed;
515 ntxd = scctx->isc_ntxd[0];
516 do {
517 MPASS(prev != cur);
518 delta = (int32_t)cur - (int32_t)prev;
519 if (delta < 0)
520 delta += ntxd;
521 MPASS(delta > 0);
522 #if 0
523 device_printf(iflib_get_dev(vsi->ctx),
524 "%s: (q%d) cidx_processed=%u cur=%u clear=%d delta=%d\n",
525 __func__, txr->me, prev, cur, clear, delta);
526 #endif
527 processed += delta;
528 prev = cur;
529 rs_cidx = (rs_cidx + 1) & (ntxd-1);
530 if (rs_cidx == txr->tx_rs_pidx)
531 break;
532 cur = txr->tx_rsq[rs_cidx];
533 MPASS(cur != QIDX_INVALID);
534 is_done = ixl_is_tx_desc_done(txr, cur);
535 } while (is_done);
536
537 txr->tx_rs_cidx = rs_cidx;
538 txr->tx_cidx_processed = prev;
539
540 #if 0
541 device_printf(iflib_get_dev(vsi->ctx), "%s: (q%d) processed %d\n", __func__, txr->me, processed);
542 #endif
543 return (processed);
544 }
545
546 static void
ixl_isc_rxd_refill(void * arg,if_rxd_update_t iru)547 ixl_isc_rxd_refill(void *arg, if_rxd_update_t iru)
548 {
549 struct ixl_vsi *vsi = arg;
550 if_softc_ctx_t scctx = vsi->shared;
551 struct rx_ring *rxr = &((vsi->rx_queues[iru->iru_qsidx]).rxr);
552 uint64_t *paddrs;
553 uint32_t next_pidx, pidx;
554 uint16_t count;
555 int i;
556
557 paddrs = iru->iru_paddrs;
558 pidx = iru->iru_pidx;
559 count = iru->iru_count;
560
561 for (i = 0, next_pidx = pidx; i < count; i++) {
562 rxr->rx_base[next_pidx].read.pkt_addr = htole64(paddrs[i]);
563 if (++next_pidx == scctx->isc_nrxd[0])
564 next_pidx = 0;
565 }
566 }
567
568 static void
ixl_isc_rxd_flush(void * arg,uint16_t rxqid,uint8_t flid __unused,qidx_t pidx)569 ixl_isc_rxd_flush(void * arg, uint16_t rxqid, uint8_t flid __unused, qidx_t pidx)
570 {
571 struct ixl_vsi *vsi = arg;
572 struct rx_ring *rxr = &vsi->rx_queues[rxqid].rxr;
573
574 wr32(vsi->hw, rxr->tail, pidx);
575 }
576
577 static int
ixl_isc_rxd_available(void * arg,uint16_t rxqid,qidx_t idx,qidx_t budget)578 ixl_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget)
579 {
580 struct ixl_vsi *vsi = arg;
581 struct rx_ring *rxr = &vsi->rx_queues[rxqid].rxr;
582 union i40e_rx_desc *rxd;
583 u64 qword;
584 uint32_t status;
585 int cnt, i, nrxd;
586
587 nrxd = vsi->shared->isc_nrxd[0];
588
589 for (cnt = 0, i = idx; cnt < nrxd - 1 && cnt <= budget;) {
590 rxd = &rxr->rx_base[i];
591 qword = le64toh(rxd->wb.qword1.status_error_len);
592 status = (qword & I40E_RXD_QW1_STATUS_MASK)
593 >> I40E_RXD_QW1_STATUS_SHIFT;
594
595 if ((status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) == 0)
596 break;
597 if (++i == nrxd)
598 i = 0;
599 if (status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT))
600 cnt++;
601 }
602
603 return (cnt);
604 }
605
606 /*
607 ** i40e_ptype_to_hash: parse the packet type
608 ** to determine the appropriate hash.
609 */
610 static inline int
ixl_ptype_to_hash(u8 ptype)611 ixl_ptype_to_hash(u8 ptype)
612 {
613 struct i40e_rx_ptype_decoded decoded;
614
615 decoded = decode_rx_desc_ptype(ptype);
616
617 if (!decoded.known)
618 return M_HASHTYPE_OPAQUE;
619
620 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_L2)
621 return M_HASHTYPE_OPAQUE;
622
623 /* Note: anything that gets to this point is IP */
624 if (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6) {
625 switch (decoded.inner_prot) {
626 case I40E_RX_PTYPE_INNER_PROT_TCP:
627 return M_HASHTYPE_RSS_TCP_IPV6;
628 case I40E_RX_PTYPE_INNER_PROT_UDP:
629 return M_HASHTYPE_RSS_UDP_IPV6;
630 default:
631 return M_HASHTYPE_RSS_IPV6;
632 }
633 }
634 if (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4) {
635 switch (decoded.inner_prot) {
636 case I40E_RX_PTYPE_INNER_PROT_TCP:
637 return M_HASHTYPE_RSS_TCP_IPV4;
638 case I40E_RX_PTYPE_INNER_PROT_UDP:
639 return M_HASHTYPE_RSS_UDP_IPV4;
640 default:
641 return M_HASHTYPE_RSS_IPV4;
642 }
643 }
644 /* We should never get here!! */
645 return M_HASHTYPE_OPAQUE;
646 }
647
648 /*********************************************************************
649 *
650 * This routine executes in ithread context. It sends data which has been
651 * dma'ed into host memory to upper layer.
652 *
653 * Returns 0 upon success, errno on failure
654 *
655 *********************************************************************/
656 static int
ixl_isc_rxd_pkt_get(void * arg,if_rxd_info_t ri)657 ixl_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri)
658 {
659 struct ixl_vsi *vsi = arg;
660 if_softc_ctx_t scctx = vsi->shared;
661 struct ixl_rx_queue *que = &vsi->rx_queues[ri->iri_qsidx];
662 struct rx_ring *rxr = &que->rxr;
663 union i40e_rx_desc *cur;
664 u32 status, error;
665 u16 plen;
666 u64 qword;
667 u8 ptype;
668 bool eop;
669 int i, cidx;
670
671 cidx = ri->iri_cidx;
672 i = 0;
673 do {
674 /* 5 descriptor receive limit */
675 MPASS(i < IXL_MAX_RX_SEGS);
676
677 cur = &rxr->rx_base[cidx];
678 qword = le64toh(cur->wb.qword1.status_error_len);
679 status = (qword & I40E_RXD_QW1_STATUS_MASK)
680 >> I40E_RXD_QW1_STATUS_SHIFT;
681 error = (qword & I40E_RXD_QW1_ERROR_MASK)
682 >> I40E_RXD_QW1_ERROR_SHIFT;
683 plen = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK)
684 >> I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
685 ptype = (qword & I40E_RXD_QW1_PTYPE_MASK)
686 >> I40E_RXD_QW1_PTYPE_SHIFT;
687
688 /* we should never be called without a valid descriptor */
689 MPASS((status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) != 0);
690
691 ri->iri_len += plen;
692 rxr->rx_bytes += plen;
693
694 cur->wb.qword1.status_error_len = 0;
695 eop = (status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT));
696
697 /*
698 ** Make sure bad packets are discarded,
699 ** note that only EOP descriptor has valid
700 ** error results.
701 */
702 if (eop && (error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
703 rxr->desc_errs++;
704 return (EBADMSG);
705 }
706 ri->iri_frags[i].irf_flid = 0;
707 ri->iri_frags[i].irf_idx = cidx;
708 ri->iri_frags[i].irf_len = plen;
709 if (++cidx == vsi->shared->isc_nrxd[0])
710 cidx = 0;
711 i++;
712 } while (!eop);
713
714 /* capture data for dynamic ITR adjustment */
715 rxr->packets++;
716 rxr->rx_packets++;
717
718 if ((scctx->isc_capenable & IFCAP_RXCSUM) != 0)
719 rxr->csum_errs += ixl_rx_checksum(ri, status, error, ptype);
720 ri->iri_flowid = le32toh(cur->wb.qword0.hi_dword.rss);
721 ri->iri_rsstype = ixl_ptype_to_hash(ptype);
722 if (status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
723 ri->iri_vtag = le16toh(cur->wb.qword0.lo_dword.l2tag1);
724 ri->iri_flags |= M_VLANTAG;
725 }
726 ri->iri_nfrags = i;
727 return (0);
728 }
729
730 /*********************************************************************
731 *
732 * Verify that the hardware indicated that the checksum is valid.
733 * Inform the stack about the status of checksum so that stack
734 * doesn't spend time verifying the checksum.
735 *
736 *********************************************************************/
737 static u8
ixl_rx_checksum(if_rxd_info_t ri,u32 status,u32 error,u8 ptype)738 ixl_rx_checksum(if_rxd_info_t ri, u32 status, u32 error, u8 ptype)
739 {
740 struct i40e_rx_ptype_decoded decoded;
741
742 ri->iri_csum_flags = 0;
743
744 /* No L3 or L4 checksum was calculated */
745 if (!(status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
746 return (0);
747
748 decoded = decode_rx_desc_ptype(ptype);
749
750 /* IPv6 with extension headers likely have bad csum */
751 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
752 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6) {
753 if (status &
754 (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT)) {
755 ri->iri_csum_flags = 0;
756 return (1);
757 }
758 }
759
760 ri->iri_csum_flags |= CSUM_L3_CALC;
761
762 /* IPv4 checksum error */
763 if (error & (1 << I40E_RX_DESC_ERROR_IPE_SHIFT))
764 return (1);
765
766 ri->iri_csum_flags |= CSUM_L3_VALID;
767 ri->iri_csum_flags |= CSUM_L4_CALC;
768
769 /* L4 checksum error */
770 if (error & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT))
771 return (1);
772
773 ri->iri_csum_flags |= CSUM_L4_VALID;
774 ri->iri_csum_data |= htons(0xffff);
775
776 return (0);
777 }
778
779 /* Set Report Status queue fields to 0 */
780 void
ixl_init_tx_rsqs(struct ixl_vsi * vsi)781 ixl_init_tx_rsqs(struct ixl_vsi *vsi)
782 {
783 if_softc_ctx_t scctx = vsi->shared;
784 struct ixl_tx_queue *tx_que;
785 int i, j;
786
787 for (i = 0, tx_que = vsi->tx_queues; i < vsi->num_tx_queues; i++, tx_que++) {
788 struct tx_ring *txr = &tx_que->txr;
789
790 txr->tx_rs_cidx = txr->tx_rs_pidx;
791
792 /* Initialize the last processed descriptor to be the end of
793 * the ring, rather than the start, so that we avoid an
794 * off-by-one error when calculating how many descriptors are
795 * done in the credits_update function.
796 */
797 txr->tx_cidx_processed = scctx->isc_ntxd[0] - 1;
798
799 for (j = 0; j < scctx->isc_ntxd[0]; j++)
800 txr->tx_rsq[j] = QIDX_INVALID;
801 }
802 }
803
804 void
ixl_init_tx_cidx(struct ixl_vsi * vsi)805 ixl_init_tx_cidx(struct ixl_vsi *vsi)
806 {
807 if_softc_ctx_t scctx = vsi->shared;
808 struct ixl_tx_queue *tx_que;
809 int i;
810
811 for (i = 0, tx_que = vsi->tx_queues; i < vsi->num_tx_queues; i++, tx_que++) {
812 struct tx_ring *txr = &tx_que->txr;
813
814 txr->tx_cidx_processed = scctx->isc_ntxd[0] - 1;
815 }
816 }
817
818 /*
819 * Input: bitmap of enum virtchnl_link_speed
820 */
821 u64
ixl_max_vc_speed_to_value(u8 link_speeds)822 ixl_max_vc_speed_to_value(u8 link_speeds)
823 {
824 if (link_speeds & VIRTCHNL_LINK_SPEED_40GB)
825 return IF_Gbps(40);
826 if (link_speeds & VIRTCHNL_LINK_SPEED_25GB)
827 return IF_Gbps(25);
828 if (link_speeds & VIRTCHNL_LINK_SPEED_20GB)
829 return IF_Gbps(20);
830 if (link_speeds & VIRTCHNL_LINK_SPEED_10GB)
831 return IF_Gbps(10);
832 if (link_speeds & VIRTCHNL_LINK_SPEED_1GB)
833 return IF_Gbps(1);
834 if (link_speeds & VIRTCHNL_LINK_SPEED_100MB)
835 return IF_Mbps(100);
836 else
837 /* Minimum supported link speed */
838 return IF_Mbps(100);
839 }
840
841 void
ixl_add_vsi_sysctls(device_t dev,struct ixl_vsi * vsi,struct sysctl_ctx_list * ctx,const char * sysctl_name)842 ixl_add_vsi_sysctls(device_t dev, struct ixl_vsi *vsi,
843 struct sysctl_ctx_list *ctx, const char *sysctl_name)
844 {
845 struct sysctl_oid *tree;
846 struct sysctl_oid_list *child;
847 struct sysctl_oid_list *vsi_list;
848
849 tree = device_get_sysctl_tree(dev);
850 child = SYSCTL_CHILDREN(tree);
851 vsi->vsi_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, sysctl_name,
852 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "VSI Number");
853 vsi_list = SYSCTL_CHILDREN(vsi->vsi_node);
854
855 ixl_add_sysctls_eth_stats(ctx, vsi_list, &vsi->eth_stats);
856 }
857
858 void
ixl_add_sysctls_eth_stats(struct sysctl_ctx_list * ctx,struct sysctl_oid_list * child,struct i40e_eth_stats * eth_stats)859 ixl_add_sysctls_eth_stats(struct sysctl_ctx_list *ctx,
860 struct sysctl_oid_list *child,
861 struct i40e_eth_stats *eth_stats)
862 {
863 struct ixl_sysctl_info ctls[] =
864 {
865 {ð_stats->rx_bytes, "good_octets_rcvd", "Good Octets Received"},
866 {ð_stats->rx_unicast, "ucast_pkts_rcvd",
867 "Unicast Packets Received"},
868 {ð_stats->rx_multicast, "mcast_pkts_rcvd",
869 "Multicast Packets Received"},
870 {ð_stats->rx_broadcast, "bcast_pkts_rcvd",
871 "Broadcast Packets Received"},
872 {ð_stats->rx_discards, "rx_discards", "Discarded RX packets"},
873 {ð_stats->tx_bytes, "good_octets_txd", "Good Octets Transmitted"},
874 {ð_stats->tx_unicast, "ucast_pkts_txd", "Unicast Packets Transmitted"},
875 {ð_stats->tx_multicast, "mcast_pkts_txd",
876 "Multicast Packets Transmitted"},
877 {ð_stats->tx_broadcast, "bcast_pkts_txd",
878 "Broadcast Packets Transmitted"},
879 // end
880 {0,0,0}
881 };
882
883 struct ixl_sysctl_info *entry = ctls;
884 while (entry->stat != 0)
885 {
886 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, entry->name,
887 CTLFLAG_RD, entry->stat,
888 entry->description);
889 entry++;
890 }
891 }
892
893 void
ixl_vsi_add_queues_stats(struct ixl_vsi * vsi,struct sysctl_ctx_list * ctx)894 ixl_vsi_add_queues_stats(struct ixl_vsi *vsi, struct sysctl_ctx_list *ctx)
895 {
896 struct sysctl_oid_list *vsi_list, *queue_list;
897 struct sysctl_oid *queue_node;
898 char queue_namebuf[IXL_QUEUE_NAME_LEN];
899
900 struct ixl_rx_queue *rx_que;
901 struct ixl_tx_queue *tx_que;
902 struct tx_ring *txr;
903 struct rx_ring *rxr;
904
905 vsi_list = SYSCTL_CHILDREN(vsi->vsi_node);
906
907 /* Queue statistics */
908 for (int q = 0; q < vsi->num_rx_queues; q++) {
909 bzero(queue_namebuf, sizeof(queue_namebuf));
910 snprintf(queue_namebuf, sizeof(queue_namebuf), "rxq%02d", q);
911 queue_node = SYSCTL_ADD_NODE(ctx, vsi_list,
912 OID_AUTO, queue_namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE,
913 NULL, "RX Queue #");
914 queue_list = SYSCTL_CHILDREN(queue_node);
915
916 rx_que = &(vsi->rx_queues[q]);
917 rxr = &(rx_que->rxr);
918
919 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "irqs",
920 CTLFLAG_RD, &(rx_que->irqs),
921 "irqs on this queue (both Tx and Rx)");
922
923 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "packets",
924 CTLFLAG_RD, &(rxr->rx_packets),
925 "Queue Packets Received");
926 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "bytes",
927 CTLFLAG_RD, &(rxr->rx_bytes),
928 "Queue Bytes Received");
929 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "desc_err",
930 CTLFLAG_RD, &(rxr->desc_errs),
931 "Queue Rx Descriptor Errors");
932 SYSCTL_ADD_UINT(ctx, queue_list, OID_AUTO, "itr",
933 CTLFLAG_RD, &(rxr->itr), 0,
934 "Queue Rx ITR Interval");
935 }
936 for (int q = 0; q < vsi->num_tx_queues; q++) {
937 bzero(queue_namebuf, sizeof(queue_namebuf));
938 snprintf(queue_namebuf, sizeof(queue_namebuf), "txq%02d", q);
939 queue_node = SYSCTL_ADD_NODE(ctx, vsi_list,
940 OID_AUTO, queue_namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE,
941 NULL, "TX Queue #");
942 queue_list = SYSCTL_CHILDREN(queue_node);
943
944 tx_que = &(vsi->tx_queues[q]);
945 txr = &(tx_que->txr);
946
947 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "tso",
948 CTLFLAG_RD, &(tx_que->tso),
949 "TSO");
950 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "mss_too_small",
951 CTLFLAG_RD, &(txr->mss_too_small),
952 "TSO sends with an MSS less than 64");
953 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "packets",
954 CTLFLAG_RD, &(txr->tx_packets),
955 "Queue Packets Transmitted");
956 SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "bytes",
957 CTLFLAG_RD, &(txr->tx_bytes),
958 "Queue Bytes Transmitted");
959 SYSCTL_ADD_UINT(ctx, queue_list, OID_AUTO, "itr",
960 CTLFLAG_RD, &(txr->itr), 0,
961 "Queue Tx ITR Interval");
962 }
963 }
964