1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019, Intel Corporation. */
3 
4 #include <net/xdp_sock_drv.h>
5 #include "ice_base.h"
6 #include "ice_lib.h"
7 #include "ice_dcb_lib.h"
8 #include "ice_sriov.h"
9 
10 /**
11  * __ice_vsi_get_qs_contig - Assign a contiguous chunk of queues to VSI
12  * @qs_cfg: gathered variables needed for PF->VSI queues assignment
13  *
14  * Return 0 on success and -ENOMEM in case of no left space in PF queue bitmap
15  */
16 static int __ice_vsi_get_qs_contig(struct ice_qs_cfg *qs_cfg)
17 {
18 	unsigned int offset, i;
19 
20 	mutex_lock(qs_cfg->qs_mutex);
21 	offset = bitmap_find_next_zero_area(qs_cfg->pf_map, qs_cfg->pf_map_size,
22 					    0, qs_cfg->q_count, 0);
23 	if (offset >= qs_cfg->pf_map_size) {
24 		mutex_unlock(qs_cfg->qs_mutex);
25 		return -ENOMEM;
26 	}
27 
28 	bitmap_set(qs_cfg->pf_map, offset, qs_cfg->q_count);
29 	for (i = 0; i < qs_cfg->q_count; i++)
30 		qs_cfg->vsi_map[i + qs_cfg->vsi_map_offset] = (u16)(i + offset);
31 	mutex_unlock(qs_cfg->qs_mutex);
32 
33 	return 0;
34 }
35 
36 /**
37  * __ice_vsi_get_qs_sc - Assign a scattered queues from PF to VSI
38  * @qs_cfg: gathered variables needed for pf->vsi queues assignment
39  *
40  * Return 0 on success and -ENOMEM in case of no left space in PF queue bitmap
41  */
42 static int __ice_vsi_get_qs_sc(struct ice_qs_cfg *qs_cfg)
43 {
44 	unsigned int i, index = 0;
45 
46 	mutex_lock(qs_cfg->qs_mutex);
47 	for (i = 0; i < qs_cfg->q_count; i++) {
48 		index = find_next_zero_bit(qs_cfg->pf_map,
49 					   qs_cfg->pf_map_size, index);
50 		if (index >= qs_cfg->pf_map_size)
51 			goto err_scatter;
52 		set_bit(index, qs_cfg->pf_map);
53 		qs_cfg->vsi_map[i + qs_cfg->vsi_map_offset] = (u16)index;
54 	}
55 	mutex_unlock(qs_cfg->qs_mutex);
56 
57 	return 0;
58 err_scatter:
59 	for (index = 0; index < i; index++) {
60 		clear_bit(qs_cfg->vsi_map[index], qs_cfg->pf_map);
61 		qs_cfg->vsi_map[index + qs_cfg->vsi_map_offset] = 0;
62 	}
63 	mutex_unlock(qs_cfg->qs_mutex);
64 
65 	return -ENOMEM;
66 }
67 
68 /**
69  * ice_pf_rxq_wait - Wait for a PF's Rx queue to be enabled or disabled
70  * @pf: the PF being configured
71  * @pf_q: the PF queue
72  * @ena: enable or disable state of the queue
73  *
74  * This routine will wait for the given Rx queue of the PF to reach the
75  * enabled or disabled state.
76  * Returns -ETIMEDOUT in case of failing to reach the requested state after
77  * multiple retries; else will return 0 in case of success.
78  */
79 static int ice_pf_rxq_wait(struct ice_pf *pf, int pf_q, bool ena)
80 {
81 	int i;
82 
83 	for (i = 0; i < ICE_Q_WAIT_MAX_RETRY; i++) {
84 		if (ena == !!(rd32(&pf->hw, QRX_CTRL(pf_q)) &
85 			      QRX_CTRL_QENA_STAT_M))
86 			return 0;
87 
88 		usleep_range(20, 40);
89 	}
90 
91 	return -ETIMEDOUT;
92 }
93 
94 /**
95  * ice_vsi_alloc_q_vector - Allocate memory for a single interrupt vector
96  * @vsi: the VSI being configured
97  * @v_idx: index of the vector in the VSI struct
98  *
99  * We allocate one q_vector and set default value for ITR setting associated
100  * with this q_vector. If allocation fails we return -ENOMEM.
101  */
102 static int ice_vsi_alloc_q_vector(struct ice_vsi *vsi, u16 v_idx)
103 {
104 	struct ice_pf *pf = vsi->back;
105 	struct ice_q_vector *q_vector;
106 	int err;
107 
108 	/* allocate q_vector */
109 	q_vector = kzalloc(sizeof(*q_vector), GFP_KERNEL);
110 	if (!q_vector)
111 		return -ENOMEM;
112 
113 	q_vector->vsi = vsi;
114 	q_vector->v_idx = v_idx;
115 	q_vector->tx.itr_setting = ICE_DFLT_TX_ITR;
116 	q_vector->rx.itr_setting = ICE_DFLT_RX_ITR;
117 	q_vector->tx.itr_mode = ITR_DYNAMIC;
118 	q_vector->rx.itr_mode = ITR_DYNAMIC;
119 	q_vector->tx.type = ICE_TX_CONTAINER;
120 	q_vector->rx.type = ICE_RX_CONTAINER;
121 	q_vector->irq.index = -ENOENT;
122 
123 	if (vsi->type == ICE_VSI_VF) {
124 		q_vector->reg_idx = ice_calc_vf_reg_idx(vsi->vf, q_vector);
125 		goto out;
126 	} else if (vsi->type == ICE_VSI_CTRL && vsi->vf) {
127 		struct ice_vsi *ctrl_vsi = ice_get_vf_ctrl_vsi(pf, vsi);
128 
129 		if (ctrl_vsi) {
130 			if (unlikely(!ctrl_vsi->q_vectors)) {
131 				err = -ENOENT;
132 				goto err_free_q_vector;
133 			}
134 
135 			q_vector->irq = ctrl_vsi->q_vectors[0]->irq;
136 			goto skip_alloc;
137 		}
138 	}
139 
140 	q_vector->irq = ice_alloc_irq(pf, vsi->irq_dyn_alloc);
141 	if (q_vector->irq.index < 0) {
142 		err = -ENOMEM;
143 		goto err_free_q_vector;
144 	}
145 
146 skip_alloc:
147 	q_vector->reg_idx = q_vector->irq.index;
148 
149 	/* only set affinity_mask if the CPU is online */
150 	if (cpu_online(v_idx))
151 		cpumask_set_cpu(v_idx, &q_vector->affinity_mask);
152 
153 	/* This will not be called in the driver load path because the netdev
154 	 * will not be created yet. All other cases with register the NAPI
155 	 * handler here (i.e. resume, reset/rebuild, etc.)
156 	 */
157 	if (vsi->netdev)
158 		netif_napi_add(vsi->netdev, &q_vector->napi, ice_napi_poll);
159 
160 out:
161 	/* tie q_vector and VSI together */
162 	vsi->q_vectors[v_idx] = q_vector;
163 
164 	return 0;
165 
166 err_free_q_vector:
167 	kfree(q_vector);
168 
169 	return err;
170 }
171 
172 /**
173  * ice_free_q_vector - Free memory allocated for a specific interrupt vector
174  * @vsi: VSI having the memory freed
175  * @v_idx: index of the vector to be freed
176  */
177 static void ice_free_q_vector(struct ice_vsi *vsi, int v_idx)
178 {
179 	struct ice_q_vector *q_vector;
180 	struct ice_pf *pf = vsi->back;
181 	struct ice_tx_ring *tx_ring;
182 	struct ice_rx_ring *rx_ring;
183 	struct device *dev;
184 
185 	dev = ice_pf_to_dev(pf);
186 	if (!vsi->q_vectors[v_idx]) {
187 		dev_dbg(dev, "Queue vector at index %d not found\n", v_idx);
188 		return;
189 	}
190 	q_vector = vsi->q_vectors[v_idx];
191 
192 	ice_for_each_tx_ring(tx_ring, q_vector->tx)
193 		tx_ring->q_vector = NULL;
194 	ice_for_each_rx_ring(rx_ring, q_vector->rx)
195 		rx_ring->q_vector = NULL;
196 
197 	/* only VSI with an associated netdev is set up with NAPI */
198 	if (vsi->netdev)
199 		netif_napi_del(&q_vector->napi);
200 
201 	/* release MSIX interrupt if q_vector had interrupt allocated */
202 	if (q_vector->irq.index < 0)
203 		goto free_q_vector;
204 
205 	/* only free last VF ctrl vsi interrupt */
206 	if (vsi->type == ICE_VSI_CTRL && vsi->vf &&
207 	    ice_get_vf_ctrl_vsi(pf, vsi))
208 		goto free_q_vector;
209 
210 	ice_free_irq(pf, q_vector->irq);
211 
212 free_q_vector:
213 	kfree(q_vector);
214 	vsi->q_vectors[v_idx] = NULL;
215 }
216 
217 /**
218  * ice_cfg_itr_gran - set the ITR granularity to 2 usecs if not already set
219  * @hw: board specific structure
220  */
221 static void ice_cfg_itr_gran(struct ice_hw *hw)
222 {
223 	u32 regval = rd32(hw, GLINT_CTL);
224 
225 	/* no need to update global register if ITR gran is already set */
226 	if (!(regval & GLINT_CTL_DIS_AUTOMASK_M) &&
227 	    (((regval & GLINT_CTL_ITR_GRAN_200_M) >>
228 	     GLINT_CTL_ITR_GRAN_200_S) == ICE_ITR_GRAN_US) &&
229 	    (((regval & GLINT_CTL_ITR_GRAN_100_M) >>
230 	     GLINT_CTL_ITR_GRAN_100_S) == ICE_ITR_GRAN_US) &&
231 	    (((regval & GLINT_CTL_ITR_GRAN_50_M) >>
232 	     GLINT_CTL_ITR_GRAN_50_S) == ICE_ITR_GRAN_US) &&
233 	    (((regval & GLINT_CTL_ITR_GRAN_25_M) >>
234 	      GLINT_CTL_ITR_GRAN_25_S) == ICE_ITR_GRAN_US))
235 		return;
236 
237 	regval = ((ICE_ITR_GRAN_US << GLINT_CTL_ITR_GRAN_200_S) &
238 		  GLINT_CTL_ITR_GRAN_200_M) |
239 		 ((ICE_ITR_GRAN_US << GLINT_CTL_ITR_GRAN_100_S) &
240 		  GLINT_CTL_ITR_GRAN_100_M) |
241 		 ((ICE_ITR_GRAN_US << GLINT_CTL_ITR_GRAN_50_S) &
242 		  GLINT_CTL_ITR_GRAN_50_M) |
243 		 ((ICE_ITR_GRAN_US << GLINT_CTL_ITR_GRAN_25_S) &
244 		  GLINT_CTL_ITR_GRAN_25_M);
245 	wr32(hw, GLINT_CTL, regval);
246 }
247 
248 /**
249  * ice_calc_txq_handle - calculate the queue handle
250  * @vsi: VSI that ring belongs to
251  * @ring: ring to get the absolute queue index
252  * @tc: traffic class number
253  */
254 static u16 ice_calc_txq_handle(struct ice_vsi *vsi, struct ice_tx_ring *ring, u8 tc)
255 {
256 	WARN_ONCE(ice_ring_is_xdp(ring) && tc, "XDP ring can't belong to TC other than 0\n");
257 
258 	if (ring->ch)
259 		return ring->q_index - ring->ch->base_q;
260 
261 	/* Idea here for calculation is that we subtract the number of queue
262 	 * count from TC that ring belongs to from it's absolute queue index
263 	 * and as a result we get the queue's index within TC.
264 	 */
265 	return ring->q_index - vsi->tc_cfg.tc_info[tc].qoffset;
266 }
267 
268 /**
269  * ice_eswitch_calc_txq_handle
270  * @ring: pointer to ring which unique index is needed
271  *
272  * To correctly work with many netdevs ring->q_index of Tx rings on switchdev
273  * VSI can repeat. Hardware ring setup requires unique q_index. Calculate it
274  * here by finding index in vsi->tx_rings of this ring.
275  *
276  * Return ICE_INVAL_Q_INDEX when index wasn't found. Should never happen,
277  * because VSI is get from ring->vsi, so it has to be present in this VSI.
278  */
279 static u16 ice_eswitch_calc_txq_handle(struct ice_tx_ring *ring)
280 {
281 	struct ice_vsi *vsi = ring->vsi;
282 	int i;
283 
284 	ice_for_each_txq(vsi, i) {
285 		if (vsi->tx_rings[i] == ring)
286 			return i;
287 	}
288 
289 	return ICE_INVAL_Q_INDEX;
290 }
291 
292 /**
293  * ice_cfg_xps_tx_ring - Configure XPS for a Tx ring
294  * @ring: The Tx ring to configure
295  *
296  * This enables/disables XPS for a given Tx descriptor ring
297  * based on the TCs enabled for the VSI that ring belongs to.
298  */
299 static void ice_cfg_xps_tx_ring(struct ice_tx_ring *ring)
300 {
301 	if (!ring->q_vector || !ring->netdev)
302 		return;
303 
304 	/* We only initialize XPS once, so as not to overwrite user settings */
305 	if (test_and_set_bit(ICE_TX_XPS_INIT_DONE, ring->xps_state))
306 		return;
307 
308 	netif_set_xps_queue(ring->netdev, &ring->q_vector->affinity_mask,
309 			    ring->q_index);
310 }
311 
312 /**
313  * ice_setup_tx_ctx - setup a struct ice_tlan_ctx instance
314  * @ring: The Tx ring to configure
315  * @tlan_ctx: Pointer to the Tx LAN queue context structure to be initialized
316  * @pf_q: queue index in the PF space
317  *
318  * Configure the Tx descriptor ring in TLAN context.
319  */
320 static void
321 ice_setup_tx_ctx(struct ice_tx_ring *ring, struct ice_tlan_ctx *tlan_ctx, u16 pf_q)
322 {
323 	struct ice_vsi *vsi = ring->vsi;
324 	struct ice_hw *hw = &vsi->back->hw;
325 
326 	tlan_ctx->base = ring->dma >> ICE_TLAN_CTX_BASE_S;
327 
328 	tlan_ctx->port_num = vsi->port_info->lport;
329 
330 	/* Transmit Queue Length */
331 	tlan_ctx->qlen = ring->count;
332 
333 	ice_set_cgd_num(tlan_ctx, ring->dcb_tc);
334 
335 	/* PF number */
336 	tlan_ctx->pf_num = hw->pf_id;
337 
338 	/* queue belongs to a specific VSI type
339 	 * VF / VM index should be programmed per vmvf_type setting:
340 	 * for vmvf_type = VF, it is VF number between 0-256
341 	 * for vmvf_type = VM, it is VM number between 0-767
342 	 * for PF or EMP this field should be set to zero
343 	 */
344 	switch (vsi->type) {
345 	case ICE_VSI_LB:
346 	case ICE_VSI_CTRL:
347 	case ICE_VSI_PF:
348 		if (ring->ch)
349 			tlan_ctx->vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_VMQ;
350 		else
351 			tlan_ctx->vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_PF;
352 		break;
353 	case ICE_VSI_VF:
354 		/* Firmware expects vmvf_num to be absolute VF ID */
355 		tlan_ctx->vmvf_num = hw->func_caps.vf_base_id + vsi->vf->vf_id;
356 		tlan_ctx->vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_VF;
357 		break;
358 	case ICE_VSI_SWITCHDEV_CTRL:
359 		tlan_ctx->vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_VMQ;
360 		break;
361 	default:
362 		return;
363 	}
364 
365 	/* make sure the context is associated with the right VSI */
366 	if (ring->ch)
367 		tlan_ctx->src_vsi = ring->ch->vsi_num;
368 	else
369 		tlan_ctx->src_vsi = ice_get_hw_vsi_num(hw, vsi->idx);
370 
371 	/* Restrict Tx timestamps to the PF VSI */
372 	switch (vsi->type) {
373 	case ICE_VSI_PF:
374 		tlan_ctx->tsyn_ena = 1;
375 		break;
376 	default:
377 		break;
378 	}
379 
380 	tlan_ctx->tso_ena = ICE_TX_LEGACY;
381 	tlan_ctx->tso_qnum = pf_q;
382 
383 	/* Legacy or Advanced Host Interface:
384 	 * 0: Advanced Host Interface
385 	 * 1: Legacy Host Interface
386 	 */
387 	tlan_ctx->legacy_int = ICE_TX_LEGACY;
388 }
389 
390 /**
391  * ice_rx_offset - Return expected offset into page to access data
392  * @rx_ring: Ring we are requesting offset of
393  *
394  * Returns the offset value for ring into the data buffer.
395  */
396 static unsigned int ice_rx_offset(struct ice_rx_ring *rx_ring)
397 {
398 	if (ice_ring_uses_build_skb(rx_ring))
399 		return ICE_SKB_PAD;
400 	return 0;
401 }
402 
403 /**
404  * ice_setup_rx_ctx - Configure a receive ring context
405  * @ring: The Rx ring to configure
406  *
407  * Configure the Rx descriptor ring in RLAN context.
408  */
409 static int ice_setup_rx_ctx(struct ice_rx_ring *ring)
410 {
411 	int chain_len = ICE_MAX_CHAINED_RX_BUFS;
412 	struct ice_vsi *vsi = ring->vsi;
413 	u32 rxdid = ICE_RXDID_FLEX_NIC;
414 	struct ice_rlan_ctx rlan_ctx;
415 	struct ice_hw *hw;
416 	u16 pf_q;
417 	int err;
418 
419 	hw = &vsi->back->hw;
420 
421 	/* what is Rx queue number in global space of 2K Rx queues */
422 	pf_q = vsi->rxq_map[ring->q_index];
423 
424 	/* clear the context structure first */
425 	memset(&rlan_ctx, 0, sizeof(rlan_ctx));
426 
427 	/* Receive Queue Base Address.
428 	 * Indicates the starting address of the descriptor queue defined in
429 	 * 128 Byte units.
430 	 */
431 	rlan_ctx.base = ring->dma >> ICE_RLAN_BASE_S;
432 
433 	rlan_ctx.qlen = ring->count;
434 
435 	/* Receive Packet Data Buffer Size.
436 	 * The Packet Data Buffer Size is defined in 128 byte units.
437 	 */
438 	rlan_ctx.dbuf = ring->rx_buf_len >> ICE_RLAN_CTX_DBUF_S;
439 
440 	/* use 32 byte descriptors */
441 	rlan_ctx.dsize = 1;
442 
443 	/* Strip the Ethernet CRC bytes before the packet is posted to host
444 	 * memory.
445 	 */
446 	rlan_ctx.crcstrip = !(ring->flags & ICE_RX_FLAGS_CRC_STRIP_DIS);
447 
448 	/* L2TSEL flag defines the reported L2 Tags in the receive descriptor
449 	 * and it needs to remain 1 for non-DVM capable configurations to not
450 	 * break backward compatibility for VF drivers. Setting this field to 0
451 	 * will cause the single/outer VLAN tag to be stripped to the L2TAG2_2ND
452 	 * field in the Rx descriptor. Setting it to 1 allows the VLAN tag to
453 	 * be stripped in L2TAG1 of the Rx descriptor, which is where VFs will
454 	 * check for the tag
455 	 */
456 	if (ice_is_dvm_ena(hw))
457 		if (vsi->type == ICE_VSI_VF &&
458 		    ice_vf_is_port_vlan_ena(vsi->vf))
459 			rlan_ctx.l2tsel = 1;
460 		else
461 			rlan_ctx.l2tsel = 0;
462 	else
463 		rlan_ctx.l2tsel = 1;
464 
465 	rlan_ctx.dtype = ICE_RX_DTYPE_NO_SPLIT;
466 	rlan_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_NO_SPLIT;
467 	rlan_ctx.hsplit_1 = ICE_RLAN_RX_HSPLIT_1_NO_SPLIT;
468 
469 	/* This controls whether VLAN is stripped from inner headers
470 	 * The VLAN in the inner L2 header is stripped to the receive
471 	 * descriptor if enabled by this flag.
472 	 */
473 	rlan_ctx.showiv = 0;
474 
475 	/* For AF_XDP ZC, we disallow packets to span on
476 	 * multiple buffers, thus letting us skip that
477 	 * handling in the fast-path.
478 	 */
479 	if (ring->xsk_pool)
480 		chain_len = 1;
481 	/* Max packet size for this queue - must not be set to a larger value
482 	 * than 5 x DBUF
483 	 */
484 	rlan_ctx.rxmax = min_t(u32, vsi->max_frame,
485 			       chain_len * ring->rx_buf_len);
486 
487 	/* Rx queue threshold in units of 64 */
488 	rlan_ctx.lrxqthresh = 1;
489 
490 	/* Enable Flexible Descriptors in the queue context which
491 	 * allows this driver to select a specific receive descriptor format
492 	 * increasing context priority to pick up profile ID; default is 0x01;
493 	 * setting to 0x03 to ensure profile is programming if prev context is
494 	 * of same priority
495 	 */
496 	if (vsi->type != ICE_VSI_VF)
497 		ice_write_qrxflxp_cntxt(hw, pf_q, rxdid, 0x3, true);
498 	else
499 		ice_write_qrxflxp_cntxt(hw, pf_q, ICE_RXDID_LEGACY_1, 0x3,
500 					false);
501 
502 	/* Absolute queue number out of 2K needs to be passed */
503 	err = ice_write_rxq_ctx(hw, &rlan_ctx, pf_q);
504 	if (err) {
505 		dev_err(ice_pf_to_dev(vsi->back), "Failed to set LAN Rx queue context for absolute Rx queue %d error: %d\n",
506 			pf_q, err);
507 		return -EIO;
508 	}
509 
510 	if (vsi->type == ICE_VSI_VF)
511 		return 0;
512 
513 	/* configure Rx buffer alignment */
514 	if (!vsi->netdev || test_bit(ICE_FLAG_LEGACY_RX, vsi->back->flags))
515 		ice_clear_ring_build_skb_ena(ring);
516 	else
517 		ice_set_ring_build_skb_ena(ring);
518 
519 	ring->rx_offset = ice_rx_offset(ring);
520 
521 	/* init queue specific tail register */
522 	ring->tail = hw->hw_addr + QRX_TAIL(pf_q);
523 	writel(0, ring->tail);
524 
525 	return 0;
526 }
527 
528 /**
529  * ice_vsi_cfg_rxq - Configure an Rx queue
530  * @ring: the ring being configured
531  *
532  * Return 0 on success and a negative value on error.
533  */
534 int ice_vsi_cfg_rxq(struct ice_rx_ring *ring)
535 {
536 	struct device *dev = ice_pf_to_dev(ring->vsi->back);
537 	u32 num_bufs = ICE_RX_DESC_UNUSED(ring);
538 	int err;
539 
540 	ring->rx_buf_len = ring->vsi->rx_buf_len;
541 
542 	if (ring->vsi->type == ICE_VSI_PF) {
543 		if (!xdp_rxq_info_is_reg(&ring->xdp_rxq))
544 			/* coverity[check_return] */
545 			__xdp_rxq_info_reg(&ring->xdp_rxq, ring->netdev,
546 					   ring->q_index,
547 					   ring->q_vector->napi.napi_id,
548 					   ring->vsi->rx_buf_len);
549 
550 		ring->xsk_pool = ice_xsk_pool(ring);
551 		if (ring->xsk_pool) {
552 			xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
553 
554 			ring->rx_buf_len =
555 				xsk_pool_get_rx_frame_size(ring->xsk_pool);
556 			err = xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
557 							 MEM_TYPE_XSK_BUFF_POOL,
558 							 NULL);
559 			if (err)
560 				return err;
561 			xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
562 
563 			dev_info(dev, "Registered XDP mem model MEM_TYPE_XSK_BUFF_POOL on Rx ring %d\n",
564 				 ring->q_index);
565 		} else {
566 			if (!xdp_rxq_info_is_reg(&ring->xdp_rxq))
567 				/* coverity[check_return] */
568 				__xdp_rxq_info_reg(&ring->xdp_rxq,
569 						   ring->netdev,
570 						   ring->q_index,
571 						   ring->q_vector->napi.napi_id,
572 						   ring->vsi->rx_buf_len);
573 
574 			err = xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
575 							 MEM_TYPE_PAGE_SHARED,
576 							 NULL);
577 			if (err)
578 				return err;
579 		}
580 	}
581 
582 	xdp_init_buff(&ring->xdp, ice_rx_pg_size(ring) / 2, &ring->xdp_rxq);
583 	ring->xdp.data = NULL;
584 	err = ice_setup_rx_ctx(ring);
585 	if (err) {
586 		dev_err(dev, "ice_setup_rx_ctx failed for RxQ %d, err %d\n",
587 			ring->q_index, err);
588 		return err;
589 	}
590 
591 	if (ring->xsk_pool) {
592 		bool ok;
593 
594 		if (!xsk_buff_can_alloc(ring->xsk_pool, num_bufs)) {
595 			dev_warn(dev, "XSK buffer pool does not provide enough addresses to fill %d buffers on Rx ring %d\n",
596 				 num_bufs, ring->q_index);
597 			dev_warn(dev, "Change Rx ring/fill queue size to avoid performance issues\n");
598 
599 			return 0;
600 		}
601 
602 		ok = ice_alloc_rx_bufs_zc(ring, num_bufs);
603 		if (!ok) {
604 			u16 pf_q = ring->vsi->rxq_map[ring->q_index];
605 
606 			dev_info(dev, "Failed to allocate some buffers on XSK buffer pool enabled Rx ring %d (pf_q %d)\n",
607 				 ring->q_index, pf_q);
608 		}
609 
610 		return 0;
611 	}
612 
613 	ice_alloc_rx_bufs(ring, num_bufs);
614 
615 	return 0;
616 }
617 
618 /**
619  * __ice_vsi_get_qs - helper function for assigning queues from PF to VSI
620  * @qs_cfg: gathered variables needed for pf->vsi queues assignment
621  *
622  * This function first tries to find contiguous space. If it is not successful,
623  * it tries with the scatter approach.
624  *
625  * Return 0 on success and -ENOMEM in case of no left space in PF queue bitmap
626  */
627 int __ice_vsi_get_qs(struct ice_qs_cfg *qs_cfg)
628 {
629 	int ret = 0;
630 
631 	ret = __ice_vsi_get_qs_contig(qs_cfg);
632 	if (ret) {
633 		/* contig failed, so try with scatter approach */
634 		qs_cfg->mapping_mode = ICE_VSI_MAP_SCATTER;
635 		qs_cfg->q_count = min_t(unsigned int, qs_cfg->q_count,
636 					qs_cfg->scatter_count);
637 		ret = __ice_vsi_get_qs_sc(qs_cfg);
638 	}
639 	return ret;
640 }
641 
642 /**
643  * ice_vsi_ctrl_one_rx_ring - start/stop VSI's Rx ring with no busy wait
644  * @vsi: the VSI being configured
645  * @ena: start or stop the Rx ring
646  * @rxq_idx: 0-based Rx queue index for the VSI passed in
647  * @wait: wait or don't wait for configuration to finish in hardware
648  *
649  * Return 0 on success and negative on error.
650  */
651 int
652 ice_vsi_ctrl_one_rx_ring(struct ice_vsi *vsi, bool ena, u16 rxq_idx, bool wait)
653 {
654 	int pf_q = vsi->rxq_map[rxq_idx];
655 	struct ice_pf *pf = vsi->back;
656 	struct ice_hw *hw = &pf->hw;
657 	u32 rx_reg;
658 
659 	rx_reg = rd32(hw, QRX_CTRL(pf_q));
660 
661 	/* Skip if the queue is already in the requested state */
662 	if (ena == !!(rx_reg & QRX_CTRL_QENA_STAT_M))
663 		return 0;
664 
665 	/* turn on/off the queue */
666 	if (ena)
667 		rx_reg |= QRX_CTRL_QENA_REQ_M;
668 	else
669 		rx_reg &= ~QRX_CTRL_QENA_REQ_M;
670 	wr32(hw, QRX_CTRL(pf_q), rx_reg);
671 
672 	if (!wait)
673 		return 0;
674 
675 	ice_flush(hw);
676 	return ice_pf_rxq_wait(pf, pf_q, ena);
677 }
678 
679 /**
680  * ice_vsi_wait_one_rx_ring - wait for a VSI's Rx ring to be stopped/started
681  * @vsi: the VSI being configured
682  * @ena: true/false to verify Rx ring has been enabled/disabled respectively
683  * @rxq_idx: 0-based Rx queue index for the VSI passed in
684  *
685  * This routine will wait for the given Rx queue of the VSI to reach the
686  * enabled or disabled state. Returns -ETIMEDOUT in case of failing to reach
687  * the requested state after multiple retries; else will return 0 in case of
688  * success.
689  */
690 int ice_vsi_wait_one_rx_ring(struct ice_vsi *vsi, bool ena, u16 rxq_idx)
691 {
692 	int pf_q = vsi->rxq_map[rxq_idx];
693 	struct ice_pf *pf = vsi->back;
694 
695 	return ice_pf_rxq_wait(pf, pf_q, ena);
696 }
697 
698 /**
699  * ice_vsi_alloc_q_vectors - Allocate memory for interrupt vectors
700  * @vsi: the VSI being configured
701  *
702  * We allocate one q_vector per queue interrupt. If allocation fails we
703  * return -ENOMEM.
704  */
705 int ice_vsi_alloc_q_vectors(struct ice_vsi *vsi)
706 {
707 	struct device *dev = ice_pf_to_dev(vsi->back);
708 	u16 v_idx;
709 	int err;
710 
711 	if (vsi->q_vectors[0]) {
712 		dev_dbg(dev, "VSI %d has existing q_vectors\n", vsi->vsi_num);
713 		return -EEXIST;
714 	}
715 
716 	for (v_idx = 0; v_idx < vsi->num_q_vectors; v_idx++) {
717 		err = ice_vsi_alloc_q_vector(vsi, v_idx);
718 		if (err)
719 			goto err_out;
720 	}
721 
722 	return 0;
723 
724 err_out:
725 	while (v_idx--)
726 		ice_free_q_vector(vsi, v_idx);
727 
728 	dev_err(dev, "Failed to allocate %d q_vector for VSI %d, ret=%d\n",
729 		vsi->num_q_vectors, vsi->vsi_num, err);
730 	vsi->num_q_vectors = 0;
731 	return err;
732 }
733 
734 /**
735  * ice_vsi_map_rings_to_vectors - Map VSI rings to interrupt vectors
736  * @vsi: the VSI being configured
737  *
738  * This function maps descriptor rings to the queue-specific vectors allotted
739  * through the MSI-X enabling code. On a constrained vector budget, we map Tx
740  * and Rx rings to the vector as "efficiently" as possible.
741  */
742 void ice_vsi_map_rings_to_vectors(struct ice_vsi *vsi)
743 {
744 	int q_vectors = vsi->num_q_vectors;
745 	u16 tx_rings_rem, rx_rings_rem;
746 	int v_id;
747 
748 	/* initially assigning remaining rings count to VSIs num queue value */
749 	tx_rings_rem = vsi->num_txq;
750 	rx_rings_rem = vsi->num_rxq;
751 
752 	for (v_id = 0; v_id < q_vectors; v_id++) {
753 		struct ice_q_vector *q_vector = vsi->q_vectors[v_id];
754 		u8 tx_rings_per_v, rx_rings_per_v;
755 		u16 q_id, q_base;
756 
757 		/* Tx rings mapping to vector */
758 		tx_rings_per_v = (u8)DIV_ROUND_UP(tx_rings_rem,
759 						  q_vectors - v_id);
760 		q_vector->num_ring_tx = tx_rings_per_v;
761 		q_vector->tx.tx_ring = NULL;
762 		q_vector->tx.itr_idx = ICE_TX_ITR;
763 		q_base = vsi->num_txq - tx_rings_rem;
764 
765 		for (q_id = q_base; q_id < (q_base + tx_rings_per_v); q_id++) {
766 			struct ice_tx_ring *tx_ring = vsi->tx_rings[q_id];
767 
768 			tx_ring->q_vector = q_vector;
769 			tx_ring->next = q_vector->tx.tx_ring;
770 			q_vector->tx.tx_ring = tx_ring;
771 		}
772 		tx_rings_rem -= tx_rings_per_v;
773 
774 		/* Rx rings mapping to vector */
775 		rx_rings_per_v = (u8)DIV_ROUND_UP(rx_rings_rem,
776 						  q_vectors - v_id);
777 		q_vector->num_ring_rx = rx_rings_per_v;
778 		q_vector->rx.rx_ring = NULL;
779 		q_vector->rx.itr_idx = ICE_RX_ITR;
780 		q_base = vsi->num_rxq - rx_rings_rem;
781 
782 		for (q_id = q_base; q_id < (q_base + rx_rings_per_v); q_id++) {
783 			struct ice_rx_ring *rx_ring = vsi->rx_rings[q_id];
784 
785 			rx_ring->q_vector = q_vector;
786 			rx_ring->next = q_vector->rx.rx_ring;
787 			q_vector->rx.rx_ring = rx_ring;
788 		}
789 		rx_rings_rem -= rx_rings_per_v;
790 	}
791 }
792 
793 /**
794  * ice_vsi_free_q_vectors - Free memory allocated for interrupt vectors
795  * @vsi: the VSI having memory freed
796  */
797 void ice_vsi_free_q_vectors(struct ice_vsi *vsi)
798 {
799 	int v_idx;
800 
801 	ice_for_each_q_vector(vsi, v_idx)
802 		ice_free_q_vector(vsi, v_idx);
803 }
804 
805 /**
806  * ice_vsi_cfg_txq - Configure single Tx queue
807  * @vsi: the VSI that queue belongs to
808  * @ring: Tx ring to be configured
809  * @qg_buf: queue group buffer
810  */
811 int
812 ice_vsi_cfg_txq(struct ice_vsi *vsi, struct ice_tx_ring *ring,
813 		struct ice_aqc_add_tx_qgrp *qg_buf)
814 {
815 	u8 buf_len = struct_size(qg_buf, txqs, 1);
816 	struct ice_tlan_ctx tlan_ctx = { 0 };
817 	struct ice_aqc_add_txqs_perq *txq;
818 	struct ice_channel *ch = ring->ch;
819 	struct ice_pf *pf = vsi->back;
820 	struct ice_hw *hw = &pf->hw;
821 	int status;
822 	u16 pf_q;
823 	u8 tc;
824 
825 	/* Configure XPS */
826 	ice_cfg_xps_tx_ring(ring);
827 
828 	pf_q = ring->reg_idx;
829 	ice_setup_tx_ctx(ring, &tlan_ctx, pf_q);
830 	/* copy context contents into the qg_buf */
831 	qg_buf->txqs[0].txq_id = cpu_to_le16(pf_q);
832 	ice_set_ctx(hw, (u8 *)&tlan_ctx, qg_buf->txqs[0].txq_ctx,
833 		    ice_tlan_ctx_info);
834 
835 	/* init queue specific tail reg. It is referred as
836 	 * transmit comm scheduler queue doorbell.
837 	 */
838 	ring->tail = hw->hw_addr + QTX_COMM_DBELL(pf_q);
839 
840 	if (IS_ENABLED(CONFIG_DCB))
841 		tc = ring->dcb_tc;
842 	else
843 		tc = 0;
844 
845 	/* Add unique software queue handle of the Tx queue per
846 	 * TC into the VSI Tx ring
847 	 */
848 	if (vsi->type == ICE_VSI_SWITCHDEV_CTRL) {
849 		ring->q_handle = ice_eswitch_calc_txq_handle(ring);
850 
851 		if (ring->q_handle == ICE_INVAL_Q_INDEX)
852 			return -ENODEV;
853 	} else {
854 		ring->q_handle = ice_calc_txq_handle(vsi, ring, tc);
855 	}
856 
857 	if (ch)
858 		status = ice_ena_vsi_txq(vsi->port_info, ch->ch_vsi->idx, 0,
859 					 ring->q_handle, 1, qg_buf, buf_len,
860 					 NULL);
861 	else
862 		status = ice_ena_vsi_txq(vsi->port_info, vsi->idx, tc,
863 					 ring->q_handle, 1, qg_buf, buf_len,
864 					 NULL);
865 	if (status) {
866 		dev_err(ice_pf_to_dev(pf), "Failed to set LAN Tx queue context, error: %d\n",
867 			status);
868 		return status;
869 	}
870 
871 	/* Add Tx Queue TEID into the VSI Tx ring from the
872 	 * response. This will complete configuring and
873 	 * enabling the queue.
874 	 */
875 	txq = &qg_buf->txqs[0];
876 	if (pf_q == le16_to_cpu(txq->txq_id))
877 		ring->txq_teid = le32_to_cpu(txq->q_teid);
878 
879 	return 0;
880 }
881 
882 /**
883  * ice_cfg_itr - configure the initial interrupt throttle values
884  * @hw: pointer to the HW structure
885  * @q_vector: interrupt vector that's being configured
886  *
887  * Configure interrupt throttling values for the ring containers that are
888  * associated with the interrupt vector passed in.
889  */
890 void ice_cfg_itr(struct ice_hw *hw, struct ice_q_vector *q_vector)
891 {
892 	ice_cfg_itr_gran(hw);
893 
894 	if (q_vector->num_ring_rx)
895 		ice_write_itr(&q_vector->rx, q_vector->rx.itr_setting);
896 
897 	if (q_vector->num_ring_tx)
898 		ice_write_itr(&q_vector->tx, q_vector->tx.itr_setting);
899 
900 	ice_write_intrl(q_vector, q_vector->intrl);
901 }
902 
903 /**
904  * ice_cfg_txq_interrupt - configure interrupt on Tx queue
905  * @vsi: the VSI being configured
906  * @txq: Tx queue being mapped to MSI-X vector
907  * @msix_idx: MSI-X vector index within the function
908  * @itr_idx: ITR index of the interrupt cause
909  *
910  * Configure interrupt on Tx queue by associating Tx queue to MSI-X vector
911  * within the function space.
912  */
913 void
914 ice_cfg_txq_interrupt(struct ice_vsi *vsi, u16 txq, u16 msix_idx, u16 itr_idx)
915 {
916 	struct ice_pf *pf = vsi->back;
917 	struct ice_hw *hw = &pf->hw;
918 	u32 val;
919 
920 	itr_idx = (itr_idx << QINT_TQCTL_ITR_INDX_S) & QINT_TQCTL_ITR_INDX_M;
921 
922 	val = QINT_TQCTL_CAUSE_ENA_M | itr_idx |
923 	      ((msix_idx << QINT_TQCTL_MSIX_INDX_S) & QINT_TQCTL_MSIX_INDX_M);
924 
925 	wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), val);
926 	if (ice_is_xdp_ena_vsi(vsi)) {
927 		u32 xdp_txq = txq + vsi->num_xdp_txq;
928 
929 		wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]),
930 		     val);
931 	}
932 	ice_flush(hw);
933 }
934 
935 /**
936  * ice_cfg_rxq_interrupt - configure interrupt on Rx queue
937  * @vsi: the VSI being configured
938  * @rxq: Rx queue being mapped to MSI-X vector
939  * @msix_idx: MSI-X vector index within the function
940  * @itr_idx: ITR index of the interrupt cause
941  *
942  * Configure interrupt on Rx queue by associating Rx queue to MSI-X vector
943  * within the function space.
944  */
945 void
946 ice_cfg_rxq_interrupt(struct ice_vsi *vsi, u16 rxq, u16 msix_idx, u16 itr_idx)
947 {
948 	struct ice_pf *pf = vsi->back;
949 	struct ice_hw *hw = &pf->hw;
950 	u32 val;
951 
952 	itr_idx = (itr_idx << QINT_RQCTL_ITR_INDX_S) & QINT_RQCTL_ITR_INDX_M;
953 
954 	val = QINT_RQCTL_CAUSE_ENA_M | itr_idx |
955 	      ((msix_idx << QINT_RQCTL_MSIX_INDX_S) & QINT_RQCTL_MSIX_INDX_M);
956 
957 	wr32(hw, QINT_RQCTL(vsi->rxq_map[rxq]), val);
958 
959 	ice_flush(hw);
960 }
961 
962 /**
963  * ice_trigger_sw_intr - trigger a software interrupt
964  * @hw: pointer to the HW structure
965  * @q_vector: interrupt vector to trigger the software interrupt for
966  */
967 void ice_trigger_sw_intr(struct ice_hw *hw, struct ice_q_vector *q_vector)
968 {
969 	wr32(hw, GLINT_DYN_CTL(q_vector->reg_idx),
970 	     (ICE_ITR_NONE << GLINT_DYN_CTL_ITR_INDX_S) |
971 	     GLINT_DYN_CTL_SWINT_TRIG_M |
972 	     GLINT_DYN_CTL_INTENA_M);
973 }
974 
975 /**
976  * ice_vsi_stop_tx_ring - Disable single Tx ring
977  * @vsi: the VSI being configured
978  * @rst_src: reset source
979  * @rel_vmvf_num: Relative ID of VF/VM
980  * @ring: Tx ring to be stopped
981  * @txq_meta: Meta data of Tx ring to be stopped
982  */
983 int
984 ice_vsi_stop_tx_ring(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src,
985 		     u16 rel_vmvf_num, struct ice_tx_ring *ring,
986 		     struct ice_txq_meta *txq_meta)
987 {
988 	struct ice_pf *pf = vsi->back;
989 	struct ice_q_vector *q_vector;
990 	struct ice_hw *hw = &pf->hw;
991 	int status;
992 	u32 val;
993 
994 	/* clear cause_ena bit for disabled queues */
995 	val = rd32(hw, QINT_TQCTL(ring->reg_idx));
996 	val &= ~QINT_TQCTL_CAUSE_ENA_M;
997 	wr32(hw, QINT_TQCTL(ring->reg_idx), val);
998 
999 	/* software is expected to wait for 100 ns */
1000 	ndelay(100);
1001 
1002 	/* trigger a software interrupt for the vector
1003 	 * associated to the queue to schedule NAPI handler
1004 	 */
1005 	q_vector = ring->q_vector;
1006 	if (q_vector && !(vsi->vf && ice_is_vf_disabled(vsi->vf)))
1007 		ice_trigger_sw_intr(hw, q_vector);
1008 
1009 	status = ice_dis_vsi_txq(vsi->port_info, txq_meta->vsi_idx,
1010 				 txq_meta->tc, 1, &txq_meta->q_handle,
1011 				 &txq_meta->q_id, &txq_meta->q_teid, rst_src,
1012 				 rel_vmvf_num, NULL);
1013 
1014 	/* if the disable queue command was exercised during an
1015 	 * active reset flow, -EBUSY is returned.
1016 	 * This is not an error as the reset operation disables
1017 	 * queues at the hardware level anyway.
1018 	 */
1019 	if (status == -EBUSY) {
1020 		dev_dbg(ice_pf_to_dev(vsi->back), "Reset in progress. LAN Tx queues already disabled\n");
1021 	} else if (status == -ENOENT) {
1022 		dev_dbg(ice_pf_to_dev(vsi->back), "LAN Tx queues do not exist, nothing to disable\n");
1023 	} else if (status) {
1024 		dev_dbg(ice_pf_to_dev(vsi->back), "Failed to disable LAN Tx queues, error: %d\n",
1025 			status);
1026 		return status;
1027 	}
1028 
1029 	return 0;
1030 }
1031 
1032 /**
1033  * ice_fill_txq_meta - Prepare the Tx queue's meta data
1034  * @vsi: VSI that ring belongs to
1035  * @ring: ring that txq_meta will be based on
1036  * @txq_meta: a helper struct that wraps Tx queue's information
1037  *
1038  * Set up a helper struct that will contain all the necessary fields that
1039  * are needed for stopping Tx queue
1040  */
1041 void
1042 ice_fill_txq_meta(struct ice_vsi *vsi, struct ice_tx_ring *ring,
1043 		  struct ice_txq_meta *txq_meta)
1044 {
1045 	struct ice_channel *ch = ring->ch;
1046 	u8 tc;
1047 
1048 	if (IS_ENABLED(CONFIG_DCB))
1049 		tc = ring->dcb_tc;
1050 	else
1051 		tc = 0;
1052 
1053 	txq_meta->q_id = ring->reg_idx;
1054 	txq_meta->q_teid = ring->txq_teid;
1055 	txq_meta->q_handle = ring->q_handle;
1056 	if (ch) {
1057 		txq_meta->vsi_idx = ch->ch_vsi->idx;
1058 		txq_meta->tc = 0;
1059 	} else {
1060 		txq_meta->vsi_idx = vsi->idx;
1061 		txq_meta->tc = tc;
1062 	}
1063 }
1064