cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

ice_base.c (28891B)


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