staging/fsl-dpaa2/ethernet/dpaa2-eth.c

   1 // SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
   2 /* Copyright 2014-2016 Freescale Semiconductor Inc.
   3  * Copyright 2016-2017 NXP
   4  */
   5 #include <linux/init.h>
   6 #include <linux/module.h>
   7 #include <linux/platform_device.h>
   8 #include <linux/etherdevice.h>
   9 #include <linux/of_net.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/msi.h>
  12 #include <linux/kthread.h>
  13 #include <linux/iommu.h>
  14 #include <linux/net_tstamp.h>
  15 #include <linux/fsl/mc.h>
  16
  17 #include <net/sock.h>
  18
  19 #include "dpaa2-eth.h"
  20
  21 /* CREATE_TRACE_POINTS only needs to be defined once. Other dpa files
  22  * using trace events only need to #include <trace/events/sched.h>
  23  */
  24 #define CREATE_TRACE_POINTS
  25 #include "dpaa2-eth-trace.h"
  26
  27 MODULE_LICENSE("Dual BSD/GPL");
  28 MODULE_AUTHOR("Freescale Semiconductor, Inc");
  29 MODULE_DESCRIPTION("Freescale DPAA2 Ethernet Driver");
  30
  31 static void *dpaa2_iova_to_virt(struct iommu_domain *domain,
  32                                 dma_addr_t iova_addr)
  33 {
  34         phys_addr_t phys_addr;
  35
  36         phys_addr = domain ? iommu_iova_to_phys(domain, iova_addr) : iova_addr;
  37
  38         return phys_to_virt(phys_addr);
  39 }
  40
  41 static void validate_rx_csum(struct dpaa2_eth_priv *priv,
  42                              u32 fd_status,
  43                              struct sk_buff *skb)
  44 {
  45         skb_checksum_none_assert(skb);
  46
  47         /* HW checksum validation is disabled, nothing to do here */
  48         if (!(priv->net_dev->features & NETIF_F_RXCSUM))
  49                 return;
  50
  51         /* Read checksum validation bits */
  52         if (!((fd_status & DPAA2_FAS_L3CV) &&
  53               (fd_status & DPAA2_FAS_L4CV)))
  54                 return;
  55
  56         /* Inform the stack there's no need to compute L3/L4 csum anymore */
  57         skb->ip_summed = CHECKSUM_UNNECESSARY;
  58 }
  59
  60 /* Free a received FD.
  61  * Not to be used for Tx conf FDs or on any other paths.
  62  */
  63 static void free_rx_fd(struct dpaa2_eth_priv *priv,
  64                        const struct dpaa2_fd *fd,
  65                        void *vaddr)
  66 {
  67         struct device *dev = priv->net_dev->dev.parent;
  68         dma_addr_t addr = dpaa2_fd_get_addr(fd);
  69         u8 fd_format = dpaa2_fd_get_format(fd);
  70         struct dpaa2_sg_entry *sgt;
  71         void *sg_vaddr;
  72         int i;
  73
  74         /* If single buffer frame, just free the data buffer */
  75         if (fd_format == dpaa2_fd_single)
  76                 goto free_buf;
  77         else if (fd_format != dpaa2_fd_sg)
  78                 /* We don't support any other format */
  79                 return;
  80
  81         /* For S/G frames, we first need to free all SG entries
  82          * except the first one, which was taken care of already
  83          */
  84         sgt = vaddr + dpaa2_fd_get_offset(fd);
  85         for (i = 1; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
  86                 addr = dpaa2_sg_get_addr(&sgt[i]);
  87                 sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
  88                 dma_unmap_single(dev, addr, DPAA2_ETH_RX_BUF_SIZE,
  89                                  DMA_FROM_DEVICE);
  90
  91                 skb_free_frag(sg_vaddr);
  92                 if (dpaa2_sg_is_final(&sgt[i]))
  93                         break;
  94         }
  95
  96 free_buf:
  97         skb_free_frag(vaddr);
  98 }
  99
 100 /* Build a linear skb based on a single-buffer frame descriptor */
 101 static struct sk_buff *build_linear_skb(struct dpaa2_eth_priv *priv,
 102                                         struct dpaa2_eth_channel *ch,
 103                                         const struct dpaa2_fd *fd,
 104                                         void *fd_vaddr)
 105 {
 106         struct sk_buff *skb = NULL;
 107         u16 fd_offset = dpaa2_fd_get_offset(fd);
 108         u32 fd_length = dpaa2_fd_get_len(fd);
 109
 110         ch->buf_count--;
 111
 112         skb = build_skb(fd_vaddr, DPAA2_ETH_SKB_SIZE);
 113         if (unlikely(!skb))
 114                 return NULL;
 115
 116         skb_reserve(skb, fd_offset);
 117         skb_put(skb, fd_length);
 118
 119         return skb;
 120 }
 121
 122 /* Build a non linear (fragmented) skb based on a S/G table */
 123 static struct sk_buff *build_frag_skb(struct dpaa2_eth_priv *priv,
 124                                       struct dpaa2_eth_channel *ch,
 125                                       struct dpaa2_sg_entry *sgt)
 126 {
 127         struct sk_buff *skb = NULL;
 128         struct device *dev = priv->net_dev->dev.parent;
 129         void *sg_vaddr;
 130         dma_addr_t sg_addr;
 131         u16 sg_offset;
 132         u32 sg_length;
 133         struct page *page, *head_page;
 134         int page_offset;
 135         int i;
 136
 137         for (i = 0; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
 138                 struct dpaa2_sg_entry *sge = &sgt[i];
 139
 140                 /* NOTE: We only support SG entries in dpaa2_sg_single format,
 141                  * but this is the only format we may receive from HW anyway
 142                  */
 143
 144                 /* Get the address and length from the S/G entry */
 145                 sg_addr = dpaa2_sg_get_addr(sge);
 146                 sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, sg_addr);
 147                 dma_unmap_single(dev, sg_addr, DPAA2_ETH_RX_BUF_SIZE,
 148                                  DMA_FROM_DEVICE);
 149
 150                 sg_length = dpaa2_sg_get_len(sge);
 151
 152                 if (i == 0) {
 153                         /* We build the skb around the first data buffer */
 154                         skb = build_skb(sg_vaddr, DPAA2_ETH_SKB_SIZE);
 155                         if (unlikely(!skb)) {
 156                                 /* Free the first SG entry now, since we already
 157                                  * unmapped it and obtained the virtual address
 158                                  */
 159                                 skb_free_frag(sg_vaddr);
 160
 161                                 /* We still need to subtract the buffers used
 162                                  * by this FD from our software counter
 163                                  */
 164                                 while (!dpaa2_sg_is_final(&sgt[i]) &&
 165                                        i < DPAA2_ETH_MAX_SG_ENTRIES)
 166                                         i++;
 167                                 break;
 168                         }
 169
 170                         sg_offset = dpaa2_sg_get_offset(sge);
 171                         skb_reserve(skb, sg_offset);
 172                         skb_put(skb, sg_length);
 173                 } else {
 174                         /* Rest of the data buffers are stored as skb frags */
 175                         page = virt_to_page(sg_vaddr);
 176                         head_page = virt_to_head_page(sg_vaddr);
 177
 178                         /* Offset in page (which may be compound).
 179                          * Data in subsequent SG entries is stored from the
 180                          * beginning of the buffer, so we don't need to add the
 181                          * sg_offset.
 182                          */
 183                         page_offset = ((unsigned long)sg_vaddr &
 184                                 (PAGE_SIZE - 1)) +
 185                                 (page_address(page) - page_address(head_page));
 186
 187                         skb_add_rx_frag(skb, i - 1, head_page, page_offset,
 188                                         sg_length, DPAA2_ETH_RX_BUF_SIZE);
 189                 }
 190
 191                 if (dpaa2_sg_is_final(sge))
 192                         break;
 193         }
 194
 195         WARN_ONCE(i == DPAA2_ETH_MAX_SG_ENTRIES, "Final bit not set in SGT");
 196
 197         /* Count all data buffers + SG table buffer */
 198         ch->buf_count -= i + 2;
 199
 200         return skb;
 201 }
 202
 203 /* Main Rx frame processing routine */
 204 static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
 205                          struct dpaa2_eth_channel *ch,
 206                          const struct dpaa2_fd *fd,
 207                          struct napi_struct *napi,
 208                          u16 queue_id)
 209 {
 210         dma_addr_t addr = dpaa2_fd_get_addr(fd);
 211         u8 fd_format = dpaa2_fd_get_format(fd);
 212         void *vaddr;
 213         struct sk_buff *skb;
 214         struct rtnl_link_stats64 *percpu_stats;
 215         struct dpaa2_eth_drv_stats *percpu_extras;
 216         struct device *dev = priv->net_dev->dev.parent;
 217         struct dpaa2_fas *fas;
 218         void *buf_data;
 219         u32 status = 0;
 220
 221         /* Tracing point */
 222         trace_dpaa2_rx_fd(priv->net_dev, fd);
 223
 224         vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
 225         dma_unmap_single(dev, addr, DPAA2_ETH_RX_BUF_SIZE, DMA_FROM_DEVICE);
 226
 227         fas = dpaa2_get_fas(vaddr, false);
 228         prefetch(fas);
 229         buf_data = vaddr + dpaa2_fd_get_offset(fd);
 230         prefetch(buf_data);
 231
 232         percpu_stats = this_cpu_ptr(priv->percpu_stats);
 233         percpu_extras = this_cpu_ptr(priv->percpu_extras);
 234
 235         if (fd_format == dpaa2_fd_single) {
 236                 skb = build_linear_skb(priv, ch, fd, vaddr);
 237         } else if (fd_format == dpaa2_fd_sg) {
 238                 skb = build_frag_skb(priv, ch, buf_data);
 239                 skb_free_frag(vaddr);
 240                 percpu_extras->rx_sg_frames++;
 241                 percpu_extras->rx_sg_bytes += dpaa2_fd_get_len(fd);
 242         } else {
 243                 /* We don't support any other format */
 244                 goto err_frame_format;
 245         }
 246
 247         if (unlikely(!skb))
 248                 goto err_build_skb;
 249
 250         prefetch(skb->data);
 251
 252         /* Get the timestamp value */
 253         if (priv->rx_tstamp) {
 254                 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 255                 __le64 *ts = dpaa2_get_ts(vaddr, false);
 256                 u64 ns;
 257
 258                 memset(shhwtstamps, 0, sizeof(*shhwtstamps));
 259
 260                 ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
 261                 shhwtstamps->hwtstamp = ns_to_ktime(ns);
 262         }
 263
 264         /* Check if we need to validate the L4 csum */
 265         if (likely(dpaa2_fd_get_frc(fd) & DPAA2_FD_FRC_FASV)) {
 266                 status = le32_to_cpu(fas->status);
 267                 validate_rx_csum(priv, status, skb);
 268         }
 269
 270         skb->protocol = eth_type_trans(skb, priv->net_dev);
 271         skb_record_rx_queue(skb, queue_id);
 272
 273         percpu_stats->rx_packets++;
 274         percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
 275
 276         napi_gro_receive(napi, skb);
 277
 278         return;
 279
 280 err_build_skb:
 281         free_rx_fd(priv, fd, vaddr);
 282 err_frame_format:
 283         percpu_stats->rx_dropped++;
 284 }
 285
 286 /* Consume all frames pull-dequeued into the store. This is the simplest way to
 287  * make sure we don't accidentally issue another volatile dequeue which would
 288  * overwrite (leak) frames already in the store.
 289  *
 290  * Observance of NAPI budget is not our concern, leaving that to the caller.
 291  */
 292 static int consume_frames(struct dpaa2_eth_channel *ch)
 293 {
 294         struct dpaa2_eth_priv *priv = ch->priv;
 295         struct dpaa2_eth_fq *fq;
 296         struct dpaa2_dq *dq;
 297         const struct dpaa2_fd *fd;
 298         int cleaned = 0;
 299         int is_last;
 300
 301         do {
 302                 dq = dpaa2_io_store_next(ch->store, &is_last);
 303                 if (unlikely(!dq)) {
 304                         /* If we're here, we *must* have placed a
 305                          * volatile dequeue comnmand, so keep reading through
 306                          * the store until we get some sort of valid response
 307                          * token (either a valid frame or an "empty dequeue")
 308                          */
 309                         continue;
 310                 }
 311
 312                 fd = dpaa2_dq_fd(dq);
 313                 fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
 314                 fq->stats.frames++;
 315
 316                 fq->consume(priv, ch, fd, &ch->napi, fq->flowid);
 317                 cleaned++;
 318         } while (!is_last);
 319
 320         return cleaned;
 321 }
 322
 323 /* Configure the egress frame annotation for timestamp update */
 324 static void enable_tx_tstamp(struct dpaa2_fd *fd, void *buf_start)
 325 {
 326         struct dpaa2_faead *faead;
 327         u32 ctrl, frc;
 328
 329         /* Mark the egress frame annotation area as valid */
 330         frc = dpaa2_fd_get_frc(fd);
 331         dpaa2_fd_set_frc(fd, frc | DPAA2_FD_FRC_FAEADV);
 332
 333         /* Set hardware annotation size */
 334         ctrl = dpaa2_fd_get_ctrl(fd);
 335         dpaa2_fd_set_ctrl(fd, ctrl | DPAA2_FD_CTRL_ASAL);
 336
 337         /* enable UPD (update prepanded data) bit in FAEAD field of
 338          * hardware frame annotation area
 339          */
 340         ctrl = DPAA2_FAEAD_A2V | DPAA2_FAEAD_UPDV | DPAA2_FAEAD_UPD;
 341         faead = dpaa2_get_faead(buf_start, true);
 342         faead->ctrl = cpu_to_le32(ctrl);
 343 }
 344
 345 /* Create a frame descriptor based on a fragmented skb */
 346 static int build_sg_fd(struct dpaa2_eth_priv *priv,
 347                        struct sk_buff *skb,
 348                        struct dpaa2_fd *fd)
 349 {
 350         struct device *dev = priv->net_dev->dev.parent;
 351         void *sgt_buf = NULL;
 352         dma_addr_t addr;
 353         int nr_frags = skb_shinfo(skb)->nr_frags;
 354         struct dpaa2_sg_entry *sgt;
 355         int i, err;
 356         int sgt_buf_size;
 357         struct scatterlist *scl, *crt_scl;
 358         int num_sg;
 359         int num_dma_bufs;
 360         struct dpaa2_eth_swa *swa;
 361
 362         /* Create and map scatterlist.
 363          * We don't advertise NETIF_F_FRAGLIST, so skb_to_sgvec() will not have
 364          * to go beyond nr_frags+1.
 365          * Note: We don't support chained scatterlists
 366          */
 367         if (unlikely(PAGE_SIZE / sizeof(struct scatterlist) < nr_frags + 1))
 368                 return -EINVAL;
 369
 370         scl = kcalloc(nr_frags + 1, sizeof(struct scatterlist), GFP_ATOMIC);
 371         if (unlikely(!scl))
 372                 return -ENOMEM;
 373
 374         sg_init_table(scl, nr_frags + 1);
 375         num_sg = skb_to_sgvec(skb, scl, 0, skb->len);
 376         num_dma_bufs = dma_map_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
 377         if (unlikely(!num_dma_bufs)) {
 378                 err = -ENOMEM;
 379                 goto dma_map_sg_failed;
 380         }
 381
 382         /* Prepare the HW SGT structure */
 383         sgt_buf_size = priv->tx_data_offset +
 384                        sizeof(struct dpaa2_sg_entry) *  num_dma_bufs;
 385         sgt_buf = netdev_alloc_frag(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN);
 386         if (unlikely(!sgt_buf)) {
 387                 err = -ENOMEM;
 388                 goto sgt_buf_alloc_failed;
 389         }
 390         sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
 391         memset(sgt_buf, 0, sgt_buf_size);
 392
 393         sgt = (struct dpaa2_sg_entry *)(sgt_buf + priv->tx_data_offset);
 394
 395         /* Fill in the HW SGT structure.
 396          *
 397          * sgt_buf is zeroed out, so the following fields are implicit
 398          * in all sgt entries:
 399          *   - offset is 0
 400          *   - format is 'dpaa2_sg_single'
 401          */
 402         for_each_sg(scl, crt_scl, num_dma_bufs, i) {
 403                 dpaa2_sg_set_addr(&sgt[i], sg_dma_address(crt_scl));
 404                 dpaa2_sg_set_len(&sgt[i], sg_dma_len(crt_scl));
 405         }
 406         dpaa2_sg_set_final(&sgt[i - 1], true);
 407
 408         /* Store the skb backpointer in the SGT buffer.
 409          * Fit the scatterlist and the number of buffers alongside the
 410          * skb backpointer in the software annotation area. We'll need
 411          * all of them on Tx Conf.
 412          */
 413         swa = (struct dpaa2_eth_swa *)sgt_buf;
 414         swa->skb = skb;
 415         swa->scl = scl;
 416         swa->num_sg = num_sg;
 417         swa->sgt_size = sgt_buf_size;
 418
 419         /* Separately map the SGT buffer */
 420         addr = dma_map_single(dev, sgt_buf, sgt_buf_size, DMA_BIDIRECTIONAL);
 421         if (unlikely(dma_mapping_error(dev, addr))) {
 422                 err = -ENOMEM;
 423                 goto dma_map_single_failed;
 424         }
 425         dpaa2_fd_set_offset(fd, priv->tx_data_offset);
 426         dpaa2_fd_set_format(fd, dpaa2_fd_sg);
 427         dpaa2_fd_set_addr(fd, addr);
 428         dpaa2_fd_set_len(fd, skb->len);
 429         dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA | FD_CTRL_PTV1);
 430
 431         if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
 432                 enable_tx_tstamp(fd, sgt_buf);
 433
 434         return 0;
 435
 436 dma_map_single_failed:
 437         skb_free_frag(sgt_buf);
 438 sgt_buf_alloc_failed:
 439         dma_unmap_sg(dev, scl, num_sg, DMA_BIDIRECTIONAL);
 440 dma_map_sg_failed:
 441         kfree(scl);
 442         return err;
 443 }
 444
 445 /* Create a frame descriptor based on a linear skb */
 446 static int build_single_fd(struct dpaa2_eth_priv *priv,
 447                            struct sk_buff *skb,
 448                            struct dpaa2_fd *fd)
 449 {
 450         struct device *dev = priv->net_dev->dev.parent;
 451         u8 *buffer_start, *aligned_start;
 452         struct sk_buff **skbh;
 453         dma_addr_t addr;
 454
 455         buffer_start = skb->data - dpaa2_eth_needed_headroom(priv, skb);
 456
 457         /* If there's enough room to align the FD address, do it.
 458          * It will help hardware optimize accesses.
 459          */
 460         aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
 461                                   DPAA2_ETH_TX_BUF_ALIGN);
 462         if (aligned_start >= skb->head)
 463                 buffer_start = aligned_start;
 464
 465         /* Store a backpointer to the skb at the beginning of the buffer
 466          * (in the private data area) such that we can release it
 467          * on Tx confirm
 468          */
 469         skbh = (struct sk_buff **)buffer_start;
 470         *skbh = skb;
 471
 472         addr = dma_map_single(dev, buffer_start,
 473                               skb_tail_pointer(skb) - buffer_start,
 474                               DMA_BIDIRECTIONAL);
 475         if (unlikely(dma_mapping_error(dev, addr)))
 476                 return -ENOMEM;
 477
 478         dpaa2_fd_set_addr(fd, addr);
 479         dpaa2_fd_set_offset(fd, (u16)(skb->data - buffer_start));
 480         dpaa2_fd_set_len(fd, skb->len);
 481         dpaa2_fd_set_format(fd, dpaa2_fd_single);
 482         dpaa2_fd_set_ctrl(fd, FD_CTRL_PTA | FD_CTRL_PTV1);
 483
 484         if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
 485                 enable_tx_tstamp(fd, buffer_start);
 486
 487         return 0;
 488 }
 489
 490 /* FD freeing routine on the Tx path
 491  *
 492  * DMA-unmap and free FD and possibly SGT buffer allocated on Tx. The skb
 493  * back-pointed to is also freed.
 494  * This can be called either from dpaa2_eth_tx_conf() or on the error path of
 495  * dpaa2_eth_tx().
 496  */
 497 static void free_tx_fd(const struct dpaa2_eth_priv *priv,
 498                        const struct dpaa2_fd *fd)
 499 {
 500         struct device *dev = priv->net_dev->dev.parent;
 501         dma_addr_t fd_addr;
 502         struct sk_buff **skbh, *skb;
 503         unsigned char *buffer_start;
 504         struct dpaa2_eth_swa *swa;
 505         u8 fd_format = dpaa2_fd_get_format(fd);
 506
 507         fd_addr = dpaa2_fd_get_addr(fd);
 508         skbh = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
 509
 510         if (fd_format == dpaa2_fd_single) {
 511                 skb = *skbh;
 512                 buffer_start = (unsigned char *)skbh;
 513                 /* Accessing the skb buffer is safe before dma unmap, because
 514                  * we didn't map the actual skb shell.
 515                  */
 516                 dma_unmap_single(dev, fd_addr,
 517                                  skb_tail_pointer(skb) - buffer_start,
 518                                  DMA_BIDIRECTIONAL);
 519         } else if (fd_format == dpaa2_fd_sg) {
 520                 swa = (struct dpaa2_eth_swa *)skbh;
 521                 skb = swa->skb;
 522
 523                 /* Unmap the scatterlist */
 524                 dma_unmap_sg(dev, swa->scl, swa->num_sg, DMA_BIDIRECTIONAL);
 525                 kfree(swa->scl);
 526
 527                 /* Unmap the SGT buffer */
 528                 dma_unmap_single(dev, fd_addr, swa->sgt_size,
 529                                  DMA_BIDIRECTIONAL);
 530         } else {
 531                 netdev_dbg(priv->net_dev, "Invalid FD format\n");
 532                 return;
 533         }
 534
 535         /* Get the timestamp value */
 536         if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
 537                 struct skb_shared_hwtstamps shhwtstamps;
 538                 __le64 *ts = dpaa2_get_ts(skbh, true);
 539                 u64 ns;
 540
 541                 memset(&shhwtstamps, 0, sizeof(shhwtstamps));
 542
 543                 ns = DPAA2_PTP_CLK_PERIOD_NS * le64_to_cpup(ts);
 544                 shhwtstamps.hwtstamp = ns_to_ktime(ns);
 545                 skb_tstamp_tx(skb, &shhwtstamps);
 546         }
 547
 548         /* Free SGT buffer allocated on tx */
 549         if (fd_format != dpaa2_fd_single)
 550                 skb_free_frag(skbh);
 551
 552         /* Move on with skb release */
 553         dev_kfree_skb(skb);
 554 }
 555
 556 static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 557 {
 558         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
 559         struct dpaa2_fd fd;
 560         struct rtnl_link_stats64 *percpu_stats;
 561         struct dpaa2_eth_drv_stats *percpu_extras;
 562         struct dpaa2_eth_fq *fq;
 563         u16 queue_mapping;
 564         unsigned int needed_headroom;
 565         int err, i;
 566
 567         percpu_stats = this_cpu_ptr(priv->percpu_stats);
 568         percpu_extras = this_cpu_ptr(priv->percpu_extras);
 569
 570         needed_headroom = dpaa2_eth_needed_headroom(priv, skb);
 571         if (skb_headroom(skb) < needed_headroom) {
 572                 struct sk_buff *ns;
 573
 574                 ns = skb_realloc_headroom(skb, needed_headroom);
 575                 if (unlikely(!ns)) {
 576                         percpu_stats->tx_dropped++;
 577                         goto err_alloc_headroom;
 578                 }
 579                 percpu_extras->tx_reallocs++;
 580
 581                 if (skb->sk)
 582                         skb_set_owner_w(ns, skb->sk);
 583
 584                 dev_kfree_skb(skb);
 585                 skb = ns;
 586         }
 587
 588         /* We'll be holding a back-reference to the skb until Tx Confirmation;
 589          * we don't want that overwritten by a concurrent Tx with a cloned skb.
 590          */
 591         skb = skb_unshare(skb, GFP_ATOMIC);
 592         if (unlikely(!skb)) {
 593                 /* skb_unshare() has already freed the skb */
 594                 percpu_stats->tx_dropped++;
 595                 return NETDEV_TX_OK;
 596         }
 597
 598         /* Setup the FD fields */
 599         memset(&fd, 0, sizeof(fd));
 600
 601         if (skb_is_nonlinear(skb)) {
 602                 err = build_sg_fd(priv, skb, &fd);
 603                 percpu_extras->tx_sg_frames++;
 604                 percpu_extras->tx_sg_bytes += skb->len;
 605         } else {
 606                 err = build_single_fd(priv, skb, &fd);
 607         }
 608
 609         if (unlikely(err)) {
 610                 percpu_stats->tx_dropped++;
 611                 goto err_build_fd;
 612         }
 613
 614         /* Tracing point */
 615         trace_dpaa2_tx_fd(net_dev, &fd);
 616
 617         /* TxConf FQ selection relies on queue id from the stack.
 618          * In case of a forwarded frame from another DPNI interface, we choose
 619          * a queue affined to the same core that processed the Rx frame
 620          */
 621         queue_mapping = skb_get_queue_mapping(skb);
 622         fq = &priv->fq[queue_mapping];
 623         for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
 624                 err = dpaa2_io_service_enqueue_qd(fq->channel->dpio,
 625                                                   priv->tx_qdid, 0,
 626                                                   fq->tx_qdbin, &fd);
 627                 if (err != -EBUSY)
 628                         break;
 629         }
 630         percpu_extras->tx_portal_busy += i;
 631         if (unlikely(err < 0)) {
 632                 percpu_stats->tx_errors++;
 633                 /* Clean up everything, including freeing the skb */
 634                 free_tx_fd(priv, &fd);
 635         } else {
 636                 percpu_stats->tx_packets++;
 637                 percpu_stats->tx_bytes += dpaa2_fd_get_len(&fd);
 638         }
 639
 640         return NETDEV_TX_OK;
 641
 642 err_build_fd:
 643 err_alloc_headroom:
 644         dev_kfree_skb(skb);
 645
 646         return NETDEV_TX_OK;
 647 }
 648
 649 /* Tx confirmation frame processing routine */
 650 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
 651                               struct dpaa2_eth_channel *ch,
 652                               const struct dpaa2_fd *fd,
 653                               struct napi_struct *napi __always_unused,
 654                               u16 queue_id __always_unused)
 655 {
 656         struct rtnl_link_stats64 *percpu_stats;
 657         struct dpaa2_eth_drv_stats *percpu_extras;
 658         u32 fd_errors;
 659
 660         /* Tracing point */
 661         trace_dpaa2_tx_conf_fd(priv->net_dev, fd);
 662
 663         percpu_extras = this_cpu_ptr(priv->percpu_extras);
 664         percpu_extras->tx_conf_frames++;
 665         percpu_extras->tx_conf_bytes += dpaa2_fd_get_len(fd);
 666
 667         /* Check frame errors in the FD field */
 668         fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
 669         free_tx_fd(priv, fd);
 670
 671         if (likely(!fd_errors))
 672                 return;
 673
 674         if (net_ratelimit())
 675                 netdev_dbg(priv->net_dev, "TX frame FD error: 0x%08x\n",
 676                            fd_errors);
 677
 678         percpu_stats = this_cpu_ptr(priv->percpu_stats);
 679         /* Tx-conf logically pertains to the egress path. */
 680         percpu_stats->tx_errors++;
 681 }
 682
 683 static int set_rx_csum(struct dpaa2_eth_priv *priv, bool enable)
 684 {
 685         int err;
 686
 687         err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
 688                                DPNI_OFF_RX_L3_CSUM, enable);
 689         if (err) {
 690                 netdev_err(priv->net_dev,
 691                            "dpni_set_offload(RX_L3_CSUM) failed\n");
 692                 return err;
 693         }
 694
 695         err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
 696                                DPNI_OFF_RX_L4_CSUM, enable);
 697         if (err) {
 698                 netdev_err(priv->net_dev,
 699                            "dpni_set_offload(RX_L4_CSUM) failed\n");
 700                 return err;
 701         }
 702
 703         return 0;
 704 }
 705
 706 static int set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
 707 {
 708         int err;
 709
 710         err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
 711                                DPNI_OFF_TX_L3_CSUM, enable);
 712         if (err) {
 713                 netdev_err(priv->net_dev, "dpni_set_offload(TX_L3_CSUM) failed\n");
 714                 return err;
 715         }
 716
 717         err = dpni_set_offload(priv->mc_io, 0, priv->mc_token,
 718                                DPNI_OFF_TX_L4_CSUM, enable);
 719         if (err) {
 720                 netdev_err(priv->net_dev, "dpni_set_offload(TX_L4_CSUM) failed\n");
 721                 return err;
 722         }
 723
 724         return 0;
 725 }
 726
 727 /* Free buffers acquired from the buffer pool or which were meant to
 728  * be released in the pool
 729  */
 730 static void free_bufs(struct dpaa2_eth_priv *priv, u64 *buf_array, int count)
 731 {
 732         struct device *dev = priv->net_dev->dev.parent;
 733         void *vaddr;
 734         int i;
 735
 736         for (i = 0; i < count; i++) {
 737                 vaddr = dpaa2_iova_to_virt(priv->iommu_domain, buf_array[i]);
 738                 dma_unmap_single(dev, buf_array[i], DPAA2_ETH_RX_BUF_SIZE,
 739                                  DMA_FROM_DEVICE);
 740                 skb_free_frag(vaddr);
 741         }
 742 }
 743
 744 /* Perform a single release command to add buffers
 745  * to the specified buffer pool
 746  */
 747 static int add_bufs(struct dpaa2_eth_priv *priv,
 748                     struct dpaa2_eth_channel *ch, u16 bpid)
 749 {
 750         struct device *dev = priv->net_dev->dev.parent;
 751         u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
 752         void *buf;
 753         dma_addr_t addr;
 754         int i, err;
 755
 756         for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
 757                 /* Allocate buffer visible to WRIOP + skb shared info +
 758                  * alignment padding
 759                  */
 760                 buf = napi_alloc_frag(dpaa2_eth_buf_raw_size(priv));
 761                 if (unlikely(!buf))
 762                         goto err_alloc;
 763
 764                 buf = PTR_ALIGN(buf, priv->rx_buf_align);
 765
 766                 addr = dma_map_single(dev, buf, DPAA2_ETH_RX_BUF_SIZE,
 767                                       DMA_FROM_DEVICE);
 768                 if (unlikely(dma_mapping_error(dev, addr)))
 769                         goto err_map;
 770
 771                 buf_array[i] = addr;
 772
 773                 /* tracing point */
 774                 trace_dpaa2_eth_buf_seed(priv->net_dev,
 775                                          buf, dpaa2_eth_buf_raw_size(priv),
 776                                          addr, DPAA2_ETH_RX_BUF_SIZE,
 777                                          bpid);
 778         }
 779
 780 release_bufs:
 781         /* In case the portal is busy, retry until successful */
 782         while ((err = dpaa2_io_service_release(ch->dpio, bpid,
 783                                                buf_array, i)) == -EBUSY)
 784                 cpu_relax();
 785
 786         /* If release command failed, clean up and bail out;
 787          * not much else we can do about it
 788          */
 789         if (err) {
 790                 free_bufs(priv, buf_array, i);
 791                 return 0;
 792         }
 793
 794         return i;
 795
 796 err_map:
 797         skb_free_frag(buf);
 798 err_alloc:
 799         /* If we managed to allocate at least some buffers,
 800          * release them to hardware
 801          */
 802         if (i)
 803                 goto release_bufs;
 804
 805         return 0;
 806 }
 807
 808 static int seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
 809 {
 810         int i, j;
 811         int new_count;
 812
 813         /* This is the lazy seeding of Rx buffer pools.
 814          * dpaa2_add_bufs() is also used on the Rx hotpath and calls
 815          * napi_alloc_frag(). The trouble with that is that it in turn ends up
 816          * calling this_cpu_ptr(), which mandates execution in atomic context.
 817          * Rather than splitting up the code, do a one-off preempt disable.
 818          */
 819         preempt_disable();
 820         for (j = 0; j < priv->num_channels; j++) {
 821                 for (i = 0; i < DPAA2_ETH_NUM_BUFS;
 822                      i += DPAA2_ETH_BUFS_PER_CMD) {
 823                         new_count = add_bufs(priv, priv->channel[j], bpid);
 824                         priv->channel[j]->buf_count += new_count;
 825
 826                         if (new_count < DPAA2_ETH_BUFS_PER_CMD) {
 827                                 preempt_enable();
 828                                 return -ENOMEM;
 829                         }
 830                 }
 831         }
 832         preempt_enable();
 833
 834         return 0;
 835 }
 836
 837 /**
 838  * Drain the specified number of buffers from the DPNI's private buffer pool.
 839  * @count must not exceeed DPAA2_ETH_BUFS_PER_CMD
 840  */
 841 static void drain_bufs(struct dpaa2_eth_priv *priv, int count)
 842 {
 843         u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
 844         int ret;
 845
 846         do {
 847                 ret = dpaa2_io_service_acquire(NULL, priv->bpid,
 848                                                buf_array, count);
 849                 if (ret < 0) {
 850                         netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
 851                         return;
 852                 }
 853                 free_bufs(priv, buf_array, ret);
 854         } while (ret);
 855 }
 856
 857 static void drain_pool(struct dpaa2_eth_priv *priv)
 858 {
 859         int i;
 860
 861         drain_bufs(priv, DPAA2_ETH_BUFS_PER_CMD);
 862         drain_bufs(priv, 1);
 863
 864         for (i = 0; i < priv->num_channels; i++)
 865                 priv->channel[i]->buf_count = 0;
 866 }
 867
 868 /* Function is called from softirq context only, so we don't need to guard
 869  * the access to percpu count
 870  */
 871 static int refill_pool(struct dpaa2_eth_priv *priv,
 872                        struct dpaa2_eth_channel *ch,
 873                        u16 bpid)
 874 {
 875         int new_count;
 876
 877         if (likely(ch->buf_count >= DPAA2_ETH_REFILL_THRESH))
 878                 return 0;
 879
 880         do {
 881                 new_count = add_bufs(priv, ch, bpid);
 882                 if (unlikely(!new_count)) {
 883                         /* Out of memory; abort for now, we'll try later on */
 884                         break;
 885                 }
 886                 ch->buf_count += new_count;
 887         } while (ch->buf_count < DPAA2_ETH_NUM_BUFS);
 888
 889         if (unlikely(ch->buf_count < DPAA2_ETH_NUM_BUFS))
 890                 return -ENOMEM;
 891
 892         return 0;
 893 }
 894
 895 static int pull_channel(struct dpaa2_eth_channel *ch)
 896 {
 897         int err;
 898         int dequeues = -1;
 899
 900         /* Retry while portal is busy */
 901         do {
 902                 err = dpaa2_io_service_pull_channel(ch->dpio, ch->ch_id,
 903                                                     ch->store);
 904                 dequeues++;
 905                 cpu_relax();
 906         } while (err == -EBUSY);
 907
 908         ch->stats.dequeue_portal_busy += dequeues;
 909         if (unlikely(err))
 910                 ch->stats.pull_err++;
 911
 912         return err;
 913 }
 914
 915 /* NAPI poll routine
 916  *
 917  * Frames are dequeued from the QMan channel associated with this NAPI context.
 918  * Rx, Tx confirmation and (if configured) Rx error frames all count
 919  * towards the NAPI budget.
 920  */
 921 static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 922 {
 923         struct dpaa2_eth_channel *ch;
 924         int cleaned = 0, store_cleaned;
 925         struct dpaa2_eth_priv *priv;
 926         int err;
 927
 928         ch = container_of(napi, struct dpaa2_eth_channel, napi);
 929         priv = ch->priv;
 930
 931         while (cleaned < budget) {
 932                 err = pull_channel(ch);
 933                 if (unlikely(err))
 934                         break;
 935
 936                 /* Refill pool if appropriate */
 937                 refill_pool(priv, ch, priv->bpid);
 938
 939                 store_cleaned = consume_frames(ch);
 940                 cleaned += store_cleaned;
 941
 942                 /* If we have enough budget left for a full store,
 943                  * try a new pull dequeue, otherwise we're done here
 944                  */
 945                 if (store_cleaned == 0 ||
 946                     cleaned > budget - DPAA2_ETH_STORE_SIZE)
 947                         break;
 948         }
 949
 950         if (cleaned < budget && napi_complete_done(napi, cleaned)) {
 951                 /* Re-enable data available notifications */
 952                 do {
 953                         err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
 954                         cpu_relax();
 955                 } while (err == -EBUSY);
 956                 WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
 957                           ch->nctx.desired_cpu);
 958         }
 959
 960         ch->stats.frames += cleaned;
 961
 962         return cleaned;
 963 }
 964
 965 static void enable_ch_napi(struct dpaa2_eth_priv *priv)
 966 {
 967         struct dpaa2_eth_channel *ch;
 968         int i;
 969
 970         for (i = 0; i < priv->num_channels; i++) {
 971                 ch = priv->channel[i];
 972                 napi_enable(&ch->napi);
 973         }
 974 }
 975
 976 static void disable_ch_napi(struct dpaa2_eth_priv *priv)
 977 {
 978         struct dpaa2_eth_channel *ch;
 979         int i;
 980
 981         for (i = 0; i < priv->num_channels; i++) {
 982                 ch = priv->channel[i];
 983                 napi_disable(&ch->napi);
 984         }
 985 }
 986
 987 static int link_state_update(struct dpaa2_eth_priv *priv)
 988 {
 989         struct dpni_link_state state;
 990         int err;
 991
 992         err = dpni_get_link_state(priv->mc_io, 0, priv->mc_token, &state);
 993         if (unlikely(err)) {
 994                 netdev_err(priv->net_dev,
 995                            "dpni_get_link_state() failed\n");
 996                 return err;
 997         }
 998
 999         /* Chech link state; speed / duplex changes are not treated yet */
1000         if (priv->link_state.up == state.up)
1001                 return 0;
1002
1003         priv->link_state = state;
1004         if (state.up) {
1005                 netif_carrier_on(priv->net_dev);
1006                 netif_tx_start_all_queues(priv->net_dev);
1007         } else {
1008                 netif_tx_stop_all_queues(priv->net_dev);
1009                 netif_carrier_off(priv->net_dev);
1010         }
1011
1012         netdev_info(priv->net_dev, "Link Event: state %s\n",
1013                     state.up ? "up" : "down");
1014
1015         return 0;
1016 }
1017
1018 static int dpaa2_eth_open(struct net_device *net_dev)
1019 {
1020         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1021         int err;
1022
1023         err = seed_pool(priv, priv->bpid);
1024         if (err) {
1025                 /* Not much to do; the buffer pool, though not filled up,
1026                  * may still contain some buffers which would enable us
1027                  * to limp on.
1028                  */
1029                 netdev_err(net_dev, "Buffer seeding failed for DPBP %d (bpid=%d)\n",
1030                            priv->dpbp_dev->obj_desc.id, priv->bpid);
1031         }
1032
1033         /* We'll only start the txqs when the link is actually ready; make sure
1034          * we don't race against the link up notification, which may come
1035          * immediately after dpni_enable();
1036          */
1037         netif_tx_stop_all_queues(net_dev);
1038         enable_ch_napi(priv);
1039         /* Also, explicitly set carrier off, otherwise netif_carrier_ok() will
1040          * return true and cause 'ip link show' to report the LOWER_UP flag,
1041          * even though the link notification wasn't even received.
1042          */
1043         netif_carrier_off(net_dev);
1044
1045         err = dpni_enable(priv->mc_io, 0, priv->mc_token);
1046         if (err < 0) {
1047                 netdev_err(net_dev, "dpni_enable() failed\n");
1048                 goto enable_err;
1049         }
1050
1051         /* If the DPMAC object has already processed the link up interrupt,
1052          * we have to learn the link state ourselves.
1053          */
1054         err = link_state_update(priv);
1055         if (err < 0) {
1056                 netdev_err(net_dev, "Can't update link state\n");
1057                 goto link_state_err;
1058         }
1059
1060         return 0;
1061
1062 link_state_err:
1063 enable_err:
1064         disable_ch_napi(priv);
1065         drain_pool(priv);
1066         return err;
1067 }
1068
1069 /* The DPIO store must be empty when we call this,
1070  * at the end of every NAPI cycle.
1071  */
1072 static u32 drain_channel(struct dpaa2_eth_priv *priv,
1073                          struct dpaa2_eth_channel *ch)
1074 {
1075         u32 drained = 0, total = 0;
1076
1077         do {
1078                 pull_channel(ch);
1079                 drained = consume_frames(ch);
1080                 total += drained;
1081         } while (drained);
1082
1083         return total;
1084 }
1085
1086 static u32 drain_ingress_frames(struct dpaa2_eth_priv *priv)
1087 {
1088         struct dpaa2_eth_channel *ch;
1089         int i;
1090         u32 drained = 0;
1091
1092         for (i = 0; i < priv->num_channels; i++) {
1093                 ch = priv->channel[i];
1094                 drained += drain_channel(priv, ch);
1095         }
1096
1097         return drained;
1098 }
1099
1100 static int dpaa2_eth_stop(struct net_device *net_dev)
1101 {
1102         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1103         int dpni_enabled;
1104         int retries = 10;
1105         u32 drained;
1106
1107         netif_tx_stop_all_queues(net_dev);
1108         netif_carrier_off(net_dev);
1109
1110         /* Loop while dpni_disable() attempts to drain the egress FQs
1111          * and confirm them back to us.
1112          */
1113         do {
1114                 dpni_disable(priv->mc_io, 0, priv->mc_token);
1115                 dpni_is_enabled(priv->mc_io, 0, priv->mc_token, &dpni_enabled);
1116                 if (dpni_enabled)
1117                         /* Allow the hardware some slack */
1118                         msleep(100);
1119         } while (dpni_enabled && --retries);
1120         if (!retries) {
1121                 netdev_warn(net_dev, "Retry count exceeded disabling DPNI\n");
1122                 /* Must go on and disable NAPI nonetheless, so we don't crash at
1123                  * the next "ifconfig up"
1124                  */
1125         }
1126
1127         /* Wait for NAPI to complete on every core and disable it.
1128          * In particular, this will also prevent NAPI from being rescheduled if
1129          * a new CDAN is serviced, effectively discarding the CDAN. We therefore
1130          * don't even need to disarm the channels, except perhaps for the case
1131          * of a huge coalescing value.
1132          */
1133         disable_ch_napi(priv);
1134
1135          /* Manually drain the Rx and TxConf queues */
1136         drained = drain_ingress_frames(priv);
1137         if (drained)
1138                 netdev_dbg(net_dev, "Drained %d frames.\n", drained);
1139
1140         /* Empty the buffer pool */
1141         drain_pool(priv);
1142
1143         return 0;
1144 }
1145
1146 static int dpaa2_eth_init(struct net_device *net_dev)
1147 {
1148         u64 supported = 0;
1149         u64 not_supported = 0;
1150         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1151         u32 options = priv->dpni_attrs.options;
1152
1153         /* Capabilities listing */
1154         supported |= IFF_LIVE_ADDR_CHANGE;
1155
1156         if (options & DPNI_OPT_NO_MAC_FILTER)
1157                 not_supported |= IFF_UNICAST_FLT;
1158         else
1159                 supported |= IFF_UNICAST_FLT;
1160
1161         net_dev->priv_flags |= supported;
1162         net_dev->priv_flags &= ~not_supported;
1163
1164         /* Features */
1165         net_dev->features = NETIF_F_RXCSUM |
1166                             NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1167                             NETIF_F_SG | NETIF_F_HIGHDMA |
1168                             NETIF_F_LLTX;
1169         net_dev->hw_features = net_dev->features;
1170
1171         return 0;
1172 }
1173
1174 static int dpaa2_eth_set_addr(struct net_device *net_dev, void *addr)
1175 {
1176         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1177         struct device *dev = net_dev->dev.parent;
1178         int err;
1179
1180         err = eth_mac_addr(net_dev, addr);
1181         if (err < 0) {
1182                 dev_err(dev, "eth_mac_addr() failed (%d)\n", err);
1183                 return err;
1184         }
1185
1186         err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
1187                                         net_dev->dev_addr);
1188         if (err) {
1189                 dev_err(dev, "dpni_set_primary_mac_addr() failed (%d)\n", err);
1190                 return err;
1191         }
1192
1193         return 0;
1194 }
1195
1196 /** Fill in counters maintained by the GPP driver. These may be different from
1197  * the hardware counters obtained by ethtool.
1198  */
1199 static void dpaa2_eth_get_stats(struct net_device *net_dev,
1200                                 struct rtnl_link_stats64 *stats)
1201 {
1202         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1203         struct rtnl_link_stats64 *percpu_stats;
1204         u64 *cpustats;
1205         u64 *netstats = (u64 *)stats;
1206         int i, j;
1207         int num = sizeof(struct rtnl_link_stats64) / sizeof(u64);
1208
1209         for_each_possible_cpu(i) {
1210                 percpu_stats = per_cpu_ptr(priv->percpu_stats, i);
1211                 cpustats = (u64 *)percpu_stats;
1212                 for (j = 0; j < num; j++)
1213                         netstats[j] += cpustats[j];
1214         }
1215 }
1216
1217 /* Copy mac unicast addresses from @net_dev to @priv.
1218  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1219  */
1220 static void add_uc_hw_addr(const struct net_device *net_dev,
1221                            struct dpaa2_eth_priv *priv)
1222 {
1223         struct netdev_hw_addr *ha;
1224         int err;
1225
1226         netdev_for_each_uc_addr(ha, net_dev) {
1227                 err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1228                                         ha->addr);
1229                 if (err)
1230                         netdev_warn(priv->net_dev,
1231                                     "Could not add ucast MAC %pM to the filtering table (err %d)\n",
1232                                     ha->addr, err);
1233         }
1234 }
1235
1236 /* Copy mac multicast addresses from @net_dev to @priv
1237  * Its sole purpose is to make dpaa2_eth_set_rx_mode() more readable.
1238  */
1239 static void add_mc_hw_addr(const struct net_device *net_dev,
1240                            struct dpaa2_eth_priv *priv)
1241 {
1242         struct netdev_hw_addr *ha;
1243         int err;
1244
1245         netdev_for_each_mc_addr(ha, net_dev) {
1246                 err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token,
1247                                         ha->addr);
1248                 if (err)
1249                         netdev_warn(priv->net_dev,
1250                                     "Could not add mcast MAC %pM to the filtering table (err %d)\n",
1251                                     ha->addr, err);
1252         }
1253 }
1254
1255 static void dpaa2_eth_set_rx_mode(struct net_device *net_dev)
1256 {
1257         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1258         int uc_count = netdev_uc_count(net_dev);
1259         int mc_count = netdev_mc_count(net_dev);
1260         u8 max_mac = priv->dpni_attrs.mac_filter_entries;
1261         u32 options = priv->dpni_attrs.options;
1262         u16 mc_token = priv->mc_token;
1263         struct fsl_mc_io *mc_io = priv->mc_io;
1264         int err;
1265
1266         /* Basic sanity checks; these probably indicate a misconfiguration */
1267         if (options & DPNI_OPT_NO_MAC_FILTER && max_mac != 0)
1268                 netdev_info(net_dev,
1269                             "mac_filter_entries=%d, DPNI_OPT_NO_MAC_FILTER option must be disabled\n",
1270                             max_mac);
1271
1272         /* Force promiscuous if the uc or mc counts exceed our capabilities. */
1273         if (uc_count > max_mac) {
1274                 netdev_info(net_dev,
1275                             "Unicast addr count reached %d, max allowed is %d; forcing promisc\n",
1276                             uc_count, max_mac);
1277                 goto force_promisc;
1278         }
1279         if (mc_count + uc_count > max_mac) {
1280                 netdev_info(net_dev,
1281                             "Unicast + multicast addr count reached %d, max allowed is %d; forcing promisc\n",
1282                             uc_count + mc_count, max_mac);
1283                 goto force_mc_promisc;
1284         }
1285
1286         /* Adjust promisc settings due to flag combinations */
1287         if (net_dev->flags & IFF_PROMISC)
1288                 goto force_promisc;
1289         if (net_dev->flags & IFF_ALLMULTI) {
1290                 /* First, rebuild unicast filtering table. This should be done
1291                  * in promisc mode, in order to avoid frame loss while we
1292                  * progressively add entries to the table.
1293                  * We don't know whether we had been in promisc already, and
1294                  * making an MC call to find out is expensive; so set uc promisc
1295                  * nonetheless.
1296                  */
1297                 err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1298                 if (err)
1299                         netdev_warn(net_dev, "Can't set uc promisc\n");
1300
1301                 /* Actual uc table reconstruction. */
1302                 err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 0);
1303                 if (err)
1304                         netdev_warn(net_dev, "Can't clear uc filters\n");
1305                 add_uc_hw_addr(net_dev, priv);
1306
1307                 /* Finally, clear uc promisc and set mc promisc as requested. */
1308                 err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
1309                 if (err)
1310                         netdev_warn(net_dev, "Can't clear uc promisc\n");
1311                 goto force_mc_promisc;
1312         }
1313
1314         /* Neither unicast, nor multicast promisc will be on... eventually.
1315          * For now, rebuild mac filtering tables while forcing both of them on.
1316          */
1317         err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1318         if (err)
1319                 netdev_warn(net_dev, "Can't set uc promisc (%d)\n", err);
1320         err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
1321         if (err)
1322                 netdev_warn(net_dev, "Can't set mc promisc (%d)\n", err);
1323
1324         /* Actual mac filtering tables reconstruction */
1325         err = dpni_clear_mac_filters(mc_io, 0, mc_token, 1, 1);
1326         if (err)
1327                 netdev_warn(net_dev, "Can't clear mac filters\n");
1328         add_mc_hw_addr(net_dev, priv);
1329         add_uc_hw_addr(net_dev, priv);
1330
1331         /* Now we can clear both ucast and mcast promisc, without risking
1332          * to drop legitimate frames anymore.
1333          */
1334         err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 0);
1335         if (err)
1336                 netdev_warn(net_dev, "Can't clear ucast promisc\n");
1337         err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 0);
1338         if (err)
1339                 netdev_warn(net_dev, "Can't clear mcast promisc\n");
1340
1341         return;
1342
1343 force_promisc:
1344         err = dpni_set_unicast_promisc(mc_io, 0, mc_token, 1);
1345         if (err)
1346                 netdev_warn(net_dev, "Can't set ucast promisc\n");
1347 force_mc_promisc:
1348         err = dpni_set_multicast_promisc(mc_io, 0, mc_token, 1);
1349         if (err)
1350                 netdev_warn(net_dev, "Can't set mcast promisc\n");
1351 }
1352
1353 static int dpaa2_eth_set_features(struct net_device *net_dev,
1354                                   netdev_features_t features)
1355 {
1356         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
1357         netdev_features_t changed = features ^ net_dev->features;
1358         bool enable;
1359         int err;
1360
1361         if (changed & NETIF_F_RXCSUM) {
1362                 enable = !!(features & NETIF_F_RXCSUM);
1363                 err = set_rx_csum(priv, enable);
1364                 if (err)
1365                         return err;
1366         }
1367
1368         if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
1369                 enable = !!(features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
1370                 err = set_tx_csum(priv, enable);
1371                 if (err)
1372                         return err;
1373         }
1374
1375         return 0;
1376 }
1377
1378 static int dpaa2_eth_ts_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1379 {
1380         struct dpaa2_eth_priv *priv = netdev_priv(dev);
1381         struct hwtstamp_config config;
1382
1383         if (copy_from_user(&config, rq->ifr_data, sizeof(config)))
1384                 return -EFAULT;
1385
1386         switch (config.tx_type) {
1387         case HWTSTAMP_TX_OFF:
1388                 priv->tx_tstamp = false;
1389                 break;
1390         case HWTSTAMP_TX_ON:
1391                 priv->tx_tstamp = true;
1392                 break;
1393         default:
1394                 return -ERANGE;
1395         }
1396
1397         if (config.rx_filter == HWTSTAMP_FILTER_NONE) {
1398                 priv->rx_tstamp = false;
1399         } else {
1400                 priv->rx_tstamp = true;
1401                 /* TS is set for all frame types, not only those requested */
1402                 config.rx_filter = HWTSTAMP_FILTER_ALL;
1403         }
1404
1405         return copy_to_user(rq->ifr_data, &config, sizeof(config)) ?
1406                         -EFAULT : 0;
1407 }
1408
1409 static int dpaa2_eth_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1410 {
1411         if (cmd == SIOCSHWTSTAMP)
1412                 return dpaa2_eth_ts_ioctl(dev, rq, cmd);
1413
1414         return -EINVAL;
1415 }
1416
1417 static const struct net_device_ops dpaa2_eth_ops = {
1418         .ndo_open = dpaa2_eth_open,
1419         .ndo_start_xmit = dpaa2_eth_tx,
1420         .ndo_stop = dpaa2_eth_stop,
1421         .ndo_init = dpaa2_eth_init,
1422         .ndo_set_mac_address = dpaa2_eth_set_addr,
1423         .ndo_get_stats64 = dpaa2_eth_get_stats,
1424         .ndo_set_rx_mode = dpaa2_eth_set_rx_mode,
1425         .ndo_set_features = dpaa2_eth_set_features,
1426         .ndo_do_ioctl = dpaa2_eth_ioctl,
1427 };
1428
1429 static void cdan_cb(struct dpaa2_io_notification_ctx *ctx)
1430 {
1431         struct dpaa2_eth_channel *ch;
1432
1433         ch = container_of(ctx, struct dpaa2_eth_channel, nctx);
1434
1435         /* Update NAPI statistics */
1436         ch->stats.cdan++;
1437
1438         napi_schedule_irqoff(&ch->napi);
1439 }
1440
1441 /* Allocate and configure a DPCON object */
1442 static struct fsl_mc_device *setup_dpcon(struct dpaa2_eth_priv *priv)
1443 {
1444         struct fsl_mc_device *dpcon;
1445         struct device *dev = priv->net_dev->dev.parent;
1446         struct dpcon_attr attrs;
1447         int err;
1448
1449         err = fsl_mc_object_allocate(to_fsl_mc_device(dev),
1450                                      FSL_MC_POOL_DPCON, &dpcon);
1451         if (err) {
1452                 dev_info(dev, "Not enough DPCONs, will go on as-is\n");
1453                 return NULL;
1454         }
1455
1456         err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
1457         if (err) {
1458                 dev_err(dev, "dpcon_open() failed\n");
1459                 goto free;
1460         }
1461
1462         err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
1463         if (err) {
1464                 dev_err(dev, "dpcon_reset() failed\n");
1465                 goto close;
1466         }
1467
1468         err = dpcon_get_attributes(priv->mc_io, 0, dpcon->mc_handle, &attrs);
1469         if (err) {
1470                 dev_err(dev, "dpcon_get_attributes() failed\n");
1471                 goto close;
1472         }
1473
1474         err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
1475         if (err) {
1476                 dev_err(dev, "dpcon_enable() failed\n");
1477                 goto close;
1478         }
1479
1480         return dpcon;
1481
1482 close:
1483         dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
1484 free:
1485         fsl_mc_object_free(dpcon);
1486
1487         return NULL;
1488 }
1489
1490 static void free_dpcon(struct dpaa2_eth_priv *priv,
1491                        struct fsl_mc_device *dpcon)
1492 {
1493         dpcon_disable(priv->mc_io, 0, dpcon->mc_handle);
1494         dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
1495         fsl_mc_object_free(dpcon);
1496 }
1497
1498 static struct dpaa2_eth_channel *
1499 alloc_channel(struct dpaa2_eth_priv *priv)
1500 {
1501         struct dpaa2_eth_channel *channel;
1502         struct dpcon_attr attr;
1503         struct device *dev = priv->net_dev->dev.parent;
1504         int err;
1505
1506         channel = kzalloc(sizeof(*channel), GFP_KERNEL);
1507         if (!channel)
1508                 return NULL;
1509
1510         channel->dpcon = setup_dpcon(priv);
1511         if (!channel->dpcon)
1512                 goto err_setup;
1513
1514         err = dpcon_get_attributes(priv->mc_io, 0, channel->dpcon->mc_handle,
1515                                    &attr);
1516         if (err) {
1517                 dev_err(dev, "dpcon_get_attributes() failed\n");
1518                 goto err_get_attr;
1519         }
1520
1521         channel->dpcon_id = attr.id;
1522         channel->ch_id = attr.qbman_ch_id;
1523         channel->priv = priv;
1524
1525         return channel;
1526
1527 err_get_attr:
1528         free_dpcon(priv, channel->dpcon);
1529 err_setup:
1530         kfree(channel);
1531         return NULL;
1532 }
1533
1534 static void free_channel(struct dpaa2_eth_priv *priv,
1535                          struct dpaa2_eth_channel *channel)
1536 {
1537         free_dpcon(priv, channel->dpcon);
1538         kfree(channel);
1539 }
1540
1541 /* DPIO setup: allocate and configure QBMan channels, setup core affinity
1542  * and register data availability notifications
1543  */
1544 static int setup_dpio(struct dpaa2_eth_priv *priv)
1545 {
1546         struct dpaa2_io_notification_ctx *nctx;
1547         struct dpaa2_eth_channel *channel;
1548         struct dpcon_notification_cfg dpcon_notif_cfg;
1549         struct device *dev = priv->net_dev->dev.parent;
1550         int i, err;
1551
1552         /* We want the ability to spread ingress traffic (RX, TX conf) to as
1553          * many cores as possible, so we need one channel for each core
1554          * (unless there's fewer queues than cores, in which case the extra
1555          * channels would be wasted).
1556          * Allocate one channel per core and register it to the core's
1557          * affine DPIO. If not enough channels are available for all cores
1558          * or if some cores don't have an affine DPIO, there will be no
1559          * ingress frame processing on those cores.
1560          */
1561         cpumask_clear(&priv->dpio_cpumask);
1562         for_each_online_cpu(i) {
1563                 /* Try to allocate a channel */
1564                 channel = alloc_channel(priv);
1565                 if (!channel) {
1566                         dev_info(dev,
1567                                  "No affine channel for cpu %d and above\n", i);
1568                         err = -ENODEV;
1569                         goto err_alloc_ch;
1570                 }
1571
1572                 priv->channel[priv->num_channels] = channel;
1573
1574                 nctx = &channel->nctx;
1575                 nctx->is_cdan = 1;
1576                 nctx->cb = cdan_cb;
1577                 nctx->id = channel->ch_id;
1578                 nctx->desired_cpu = i;
1579
1580                 /* Register the new context */
1581                 channel->dpio = dpaa2_io_service_select(i);
1582                 err = dpaa2_io_service_register(channel->dpio, nctx);
1583                 if (err) {
1584                         dev_dbg(dev, "No affine DPIO for cpu %d\n", i);
1585                         /* If no affine DPIO for this core, there's probably
1586                          * none available for next cores either. Signal we want
1587                          * to retry later, in case the DPIO devices weren't
1588                          * probed yet.
1589                          */
1590                         err = -EPROBE_DEFER;
1591                         goto err_service_reg;
1592                 }
1593
1594                 /* Register DPCON notification with MC */
1595                 dpcon_notif_cfg.dpio_id = nctx->dpio_id;
1596                 dpcon_notif_cfg.priority = 0;
1597                 dpcon_notif_cfg.user_ctx = nctx->qman64;
1598                 err = dpcon_set_notification(priv->mc_io, 0,
1599                                              channel->dpcon->mc_handle,
1600                                              &dpcon_notif_cfg);
1601                 if (err) {
1602                         dev_err(dev, "dpcon_set_notification failed()\n");
1603                         goto err_set_cdan;
1604                 }
1605
1606                 /* If we managed to allocate a channel and also found an affine
1607                  * DPIO for this core, add it to the final mask
1608                  */
1609                 cpumask_set_cpu(i, &priv->dpio_cpumask);
1610                 priv->num_channels++;
1611
1612                 /* Stop if we already have enough channels to accommodate all
1613                  * RX and TX conf queues
1614                  */
1615                 if (priv->num_channels == dpaa2_eth_queue_count(priv))
1616                         break;
1617         }
1618
1619         return 0;
1620
1621 err_set_cdan:
1622         dpaa2_io_service_deregister(channel->dpio, nctx);
1623 err_service_reg:
1624         free_channel(priv, channel);
1625 err_alloc_ch:
1626         if (cpumask_empty(&priv->dpio_cpumask)) {
1627                 dev_err(dev, "No cpu with an affine DPIO/DPCON\n");
1628                 return err;
1629         }
1630
1631         dev_info(dev, "Cores %*pbl available for processing ingress traffic\n",
1632                  cpumask_pr_args(&priv->dpio_cpumask));
1633
1634         return 0;
1635 }
1636
1637 static void free_dpio(struct dpaa2_eth_priv *priv)
1638 {
1639         int i;
1640         struct dpaa2_eth_channel *ch;
1641
1642         /* deregister CDAN notifications and free channels */
1643         for (i = 0; i < priv->num_channels; i++) {
1644                 ch = priv->channel[i];
1645                 dpaa2_io_service_deregister(ch->dpio, &ch->nctx);
1646                 free_channel(priv, ch);
1647         }
1648 }
1649
1650 static struct dpaa2_eth_channel *get_affine_channel(struct dpaa2_eth_priv *priv,
1651                                                     int cpu)
1652 {
1653         struct device *dev = priv->net_dev->dev.parent;
1654         int i;
1655
1656         for (i = 0; i < priv->num_channels; i++)
1657                 if (priv->channel[i]->nctx.desired_cpu == cpu)
1658                         return priv->channel[i];
1659
1660         /* We should never get here. Issue a warning and return
1661          * the first channel, because it's still better than nothing
1662          */
1663         dev_warn(dev, "No affine channel found for cpu %d\n", cpu);
1664
1665         return priv->channel[0];
1666 }
1667
1668 static void set_fq_affinity(struct dpaa2_eth_priv *priv)
1669 {
1670         struct device *dev = priv->net_dev->dev.parent;
1671         struct cpumask xps_mask;
1672         struct dpaa2_eth_fq *fq;
1673         int rx_cpu, txc_cpu;
1674         int i, err;
1675
1676         /* For each FQ, pick one channel/CPU to deliver frames to.
1677          * This may well change at runtime, either through irqbalance or
1678          * through direct user intervention.
1679          */
1680         rx_cpu = txc_cpu = cpumask_first(&priv->dpio_cpumask);
1681
1682         for (i = 0; i < priv->num_fqs; i++) {
1683                 fq = &priv->fq[i];
1684                 switch (fq->type) {
1685                 case DPAA2_RX_FQ:
1686                         fq->target_cpu = rx_cpu;
1687                         rx_cpu = cpumask_next(rx_cpu, &priv->dpio_cpumask);
1688                         if (rx_cpu >= nr_cpu_ids)
1689                                 rx_cpu = cpumask_first(&priv->dpio_cpumask);
1690                         break;
1691                 case DPAA2_TX_CONF_FQ:
1692                         fq->target_cpu = txc_cpu;
1693
1694                         /* Tell the stack to affine to txc_cpu the Tx queue
1695                          * associated with the confirmation one
1696                          */
1697                         cpumask_clear(&xps_mask);
1698                         cpumask_set_cpu(txc_cpu, &xps_mask);
1699                         err = netif_set_xps_queue(priv->net_dev, &xps_mask,
1700                                                   fq->flowid);
1701                         if (err)
1702                                 dev_err(dev, "Error setting XPS queue\n");
1703
1704                         txc_cpu = cpumask_next(txc_cpu, &priv->dpio_cpumask);
1705                         if (txc_cpu >= nr_cpu_ids)
1706                                 txc_cpu = cpumask_first(&priv->dpio_cpumask);
1707                         break;
1708                 default:
1709                         dev_err(dev, "Unknown FQ type: %d\n", fq->type);
1710                 }
1711                 fq->channel = get_affine_channel(priv, fq->target_cpu);
1712         }
1713 }
1714
1715 static void setup_fqs(struct dpaa2_eth_priv *priv)
1716 {
1717         int i;
1718
1719         /* We have one TxConf FQ per Tx flow.
1720          * The number of Tx and Rx queues is the same.
1721          * Tx queues come first in the fq array.
1722          */
1723         for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
1724                 priv->fq[priv->num_fqs].type = DPAA2_TX_CONF_FQ;
1725                 priv->fq[priv->num_fqs].consume = dpaa2_eth_tx_conf;
1726                 priv->fq[priv->num_fqs++].flowid = (u16)i;
1727         }
1728
1729         for (i = 0; i < dpaa2_eth_queue_count(priv); i++) {
1730                 priv->fq[priv->num_fqs].type = DPAA2_RX_FQ;
1731                 priv->fq[priv->num_fqs].consume = dpaa2_eth_rx;
1732                 priv->fq[priv->num_fqs++].flowid = (u16)i;
1733         }
1734
1735         /* For each FQ, decide on which core to process incoming frames */
1736         set_fq_affinity(priv);
1737 }
1738
1739 /* Allocate and configure one buffer pool for each interface */
1740 static int setup_dpbp(struct dpaa2_eth_priv *priv)
1741 {
1742         int err;
1743         struct fsl_mc_device *dpbp_dev;
1744         struct device *dev = priv->net_dev->dev.parent;
1745         struct dpbp_attr dpbp_attrs;
1746
1747         err = fsl_mc_object_allocate(to_fsl_mc_device(dev), FSL_MC_POOL_DPBP,
1748                                      &dpbp_dev);
1749         if (err) {
1750                 dev_err(dev, "DPBP device allocation failed\n");
1751                 return err;
1752         }
1753
1754         priv->dpbp_dev = dpbp_dev;
1755
1756         err = dpbp_open(priv->mc_io, 0, priv->dpbp_dev->obj_desc.id,
1757                         &dpbp_dev->mc_handle);
1758         if (err) {
1759                 dev_err(dev, "dpbp_open() failed\n");
1760                 goto err_open;
1761         }
1762
1763         err = dpbp_reset(priv->mc_io, 0, dpbp_dev->mc_handle);
1764         if (err) {
1765                 dev_err(dev, "dpbp_reset() failed\n");
1766                 goto err_reset;
1767         }
1768
1769         err = dpbp_enable(priv->mc_io, 0, dpbp_dev->mc_handle);
1770         if (err) {
1771                 dev_err(dev, "dpbp_enable() failed\n");
1772                 goto err_enable;
1773         }
1774
1775         err = dpbp_get_attributes(priv->mc_io, 0, dpbp_dev->mc_handle,
1776                                   &dpbp_attrs);
1777         if (err) {
1778                 dev_err(dev, "dpbp_get_attributes() failed\n");
1779                 goto err_get_attr;
1780         }
1781         priv->bpid = dpbp_attrs.bpid;
1782
1783         return 0;
1784
1785 err_get_attr:
1786         dpbp_disable(priv->mc_io, 0, dpbp_dev->mc_handle);
1787 err_enable:
1788 err_reset:
1789         dpbp_close(priv->mc_io, 0, dpbp_dev->mc_handle);
1790 err_open:
1791         fsl_mc_object_free(dpbp_dev);
1792
1793         return err;
1794 }
1795
1796 static void free_dpbp(struct dpaa2_eth_priv *priv)
1797 {
1798         drain_pool(priv);
1799         dpbp_disable(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
1800         dpbp_close(priv->mc_io, 0, priv->dpbp_dev->mc_handle);
1801         fsl_mc_object_free(priv->dpbp_dev);
1802 }
1803
1804 static int set_buffer_layout(struct dpaa2_eth_priv *priv)
1805 {
1806         struct device *dev = priv->net_dev->dev.parent;
1807         struct dpni_buffer_layout buf_layout = {0};
1808         int err;
1809
1810         /* We need to check for WRIOP version 1.0.0, but depending on the MC
1811          * version, this number is not always provided correctly on rev1.
1812          * We need to check for both alternatives in this situation.
1813          */
1814         if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
1815             priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
1816                 priv->rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
1817         else
1818                 priv->rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
1819
1820         /* tx buffer */
1821         buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
1822         buf_layout.pass_timestamp = true;
1823         buf_layout.options = DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
1824                              DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
1825         err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
1826                                      DPNI_QUEUE_TX, &buf_layout);
1827         if (err) {
1828                 dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
1829                 return err;
1830         }
1831
1832         /* tx-confirm buffer */
1833         buf_layout.options = DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
1834         err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
1835                                      DPNI_QUEUE_TX_CONFIRM, &buf_layout);
1836         if (err) {
1837                 dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
1838                 return err;
1839         }
1840
1841         /* Now that we've set our tx buffer layout, retrieve the minimum
1842          * required tx data offset.
1843          */
1844         err = dpni_get_tx_data_offset(priv->mc_io, 0, priv->mc_token,
1845                                       &priv->tx_data_offset);
1846         if (err) {
1847                 dev_err(dev, "dpni_get_tx_data_offset() failed\n");
1848                 return err;
1849         }
1850
1851         if ((priv->tx_data_offset % 64) != 0)
1852                 dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
1853                          priv->tx_data_offset);
1854
1855         /* rx buffer */
1856         buf_layout.pass_frame_status = true;
1857         buf_layout.pass_parser_result = true;
1858         buf_layout.data_align = priv->rx_buf_align;
1859         buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
1860         buf_layout.private_data_size = 0;
1861         buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
1862                              DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
1863                              DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
1864                              DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM |
1865                              DPNI_BUF_LAYOUT_OPT_TIMESTAMP;
1866         err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
1867                                      DPNI_QUEUE_RX, &buf_layout);
1868         if (err) {
1869                 dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
1870                 return err;
1871         }
1872
1873         return 0;
1874 }
1875
1876 /* Configure the DPNI object this interface is associated with */
1877 static int setup_dpni(struct fsl_mc_device *ls_dev)
1878 {
1879         struct device *dev = &ls_dev->dev;
1880         struct dpaa2_eth_priv *priv;
1881         struct net_device *net_dev;
1882         int err;
1883
1884         net_dev = dev_get_drvdata(dev);
1885         priv = netdev_priv(net_dev);
1886
1887         /* get a handle for the DPNI object */
1888         err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
1889         if (err) {
1890                 dev_err(dev, "dpni_open() failed\n");
1891                 return err;
1892         }
1893
1894         /* Check if we can work with this DPNI object */
1895         err = dpni_get_api_version(priv->mc_io, 0, &priv->dpni_ver_major,
1896                                    &priv->dpni_ver_minor);
1897         if (err) {
1898                 dev_err(dev, "dpni_get_api_version() failed\n");
1899                 goto close;
1900         }
1901         if (dpaa2_eth_cmp_dpni_ver(priv, DPNI_VER_MAJOR, DPNI_VER_MINOR) < 0) {
1902                 dev_err(dev, "DPNI version %u.%u not supported, need >= %u.%u\n",
1903                         priv->dpni_ver_major, priv->dpni_ver_minor,
1904                         DPNI_VER_MAJOR, DPNI_VER_MINOR);
1905                 err = -ENOTSUPP;
1906                 goto close;
1907         }
1908
1909         ls_dev->mc_io = priv->mc_io;
1910         ls_dev->mc_handle = priv->mc_token;
1911
1912         err = dpni_reset(priv->mc_io, 0, priv->mc_token);
1913         if (err) {
1914                 dev_err(dev, "dpni_reset() failed\n");
1915                 goto close;
1916         }
1917
1918         err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
1919                                   &priv->dpni_attrs);
1920         if (err) {
1921                 dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
1922                 goto close;
1923         }
1924
1925         err = set_buffer_layout(priv);
1926         if (err)
1927                 goto close;
1928
1929         return 0;
1930
1931 close:
1932         dpni_close(priv->mc_io, 0, priv->mc_token);
1933
1934         return err;
1935 }
1936
1937 static void free_dpni(struct dpaa2_eth_priv *priv)
1938 {
1939         int err;
1940
1941         err = dpni_reset(priv->mc_io, 0, priv->mc_token);
1942         if (err)
1943                 netdev_warn(priv->net_dev, "dpni_reset() failed (err %d)\n",
1944                             err);
1945
1946         dpni_close(priv->mc_io, 0, priv->mc_token);
1947 }
1948
1949 static int setup_rx_flow(struct dpaa2_eth_priv *priv,
1950                          struct dpaa2_eth_fq *fq)
1951 {
1952         struct device *dev = priv->net_dev->dev.parent;
1953         struct dpni_queue queue;
1954         struct dpni_queue_id qid;
1955         struct dpni_taildrop td;
1956         int err;
1957
1958         err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
1959                              DPNI_QUEUE_RX, 0, fq->flowid, &queue, &qid);
1960         if (err) {
1961                 dev_err(dev, "dpni_get_queue(RX) failed\n");
1962                 return err;
1963         }
1964
1965         fq->fqid = qid.fqid;
1966
1967         queue.destination.id = fq->channel->dpcon_id;
1968         queue.destination.type = DPNI_DEST_DPCON;
1969         queue.destination.priority = 1;
1970         queue.user_context = (u64)(uintptr_t)fq;
1971         err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
1972                              DPNI_QUEUE_RX, 0, fq->flowid,
1973                              DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
1974                              &queue);
1975         if (err) {
1976                 dev_err(dev, "dpni_set_queue(RX) failed\n");
1977                 return err;
1978         }
1979
1980         td.enable = 1;
1981         td.threshold = DPAA2_ETH_TAILDROP_THRESH;
1982         err = dpni_set_taildrop(priv->mc_io, 0, priv->mc_token, DPNI_CP_QUEUE,
1983                                 DPNI_QUEUE_RX, 0, fq->flowid, &td);
1984         if (err) {
1985                 dev_err(dev, "dpni_set_threshold() failed\n");
1986                 return err;
1987         }
1988
1989         return 0;
1990 }
1991
1992 static int setup_tx_flow(struct dpaa2_eth_priv *priv,
1993                          struct dpaa2_eth_fq *fq)
1994 {
1995         struct device *dev = priv->net_dev->dev.parent;
1996         struct dpni_queue queue;
1997         struct dpni_queue_id qid;
1998         int err;
1999
2000         err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
2001                              DPNI_QUEUE_TX, 0, fq->flowid, &queue, &qid);
2002         if (err) {
2003                 dev_err(dev, "dpni_get_queue(TX) failed\n");
2004                 return err;
2005         }
2006
2007         fq->tx_qdbin = qid.qdbin;
2008
2009         err = dpni_get_queue(priv->mc_io, 0, priv->mc_token,
2010                              DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
2011                              &queue, &qid);
2012         if (err) {
2013                 dev_err(dev, "dpni_get_queue(TX_CONF) failed\n");
2014                 return err;
2015         }
2016
2017         fq->fqid = qid.fqid;
2018
2019         queue.destination.id = fq->channel->dpcon_id;
2020         queue.destination.type = DPNI_DEST_DPCON;
2021         queue.destination.priority = 0;
2022         queue.user_context = (u64)(uintptr_t)fq;
2023         err = dpni_set_queue(priv->mc_io, 0, priv->mc_token,
2024                              DPNI_QUEUE_TX_CONFIRM, 0, fq->flowid,
2025                              DPNI_QUEUE_OPT_USER_CTX | DPNI_QUEUE_OPT_DEST,
2026                              &queue);
2027         if (err) {
2028                 dev_err(dev, "dpni_set_queue(TX_CONF) failed\n");
2029                 return err;
2030         }
2031
2032         return 0;
2033 }
2034
2035 /* Hash key is a 5-tuple: IPsrc, IPdst, IPnextproto, L4src, L4dst */
2036 static const struct dpaa2_eth_hash_fields hash_fields[] = {
2037         {
2038                 /* IP header */
2039                 .rxnfc_field = RXH_IP_SRC,
2040                 .cls_prot = NET_PROT_IP,
2041                 .cls_field = NH_FLD_IP_SRC,
2042                 .size = 4,
2043         }, {
2044                 .rxnfc_field = RXH_IP_DST,
2045                 .cls_prot = NET_PROT_IP,
2046                 .cls_field = NH_FLD_IP_DST,
2047                 .size = 4,
2048         }, {
2049                 .rxnfc_field = RXH_L3_PROTO,
2050                 .cls_prot = NET_PROT_IP,
2051                 .cls_field = NH_FLD_IP_PROTO,
2052                 .size = 1,
2053         }, {
2054                 /* Using UDP ports, this is functionally equivalent to raw
2055                  * byte pairs from L4 header.
2056                  */
2057                 .rxnfc_field = RXH_L4_B_0_1,
2058                 .cls_prot = NET_PROT_UDP,
2059                 .cls_field = NH_FLD_UDP_PORT_SRC,
2060                 .size = 2,
2061         }, {
2062                 .rxnfc_field = RXH_L4_B_2_3,
2063                 .cls_prot = NET_PROT_UDP,
2064                 .cls_field = NH_FLD_UDP_PORT_DST,
2065                 .size = 2,
2066         },
2067 };
2068
2069 /* Set RX hash options
2070  * flags is a combination of RXH_ bits
2071  */
2072 static int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags)
2073 {
2074         struct device *dev = net_dev->dev.parent;
2075         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2076         struct dpkg_profile_cfg cls_cfg;
2077         struct dpni_rx_tc_dist_cfg dist_cfg;
2078         u8 *dma_mem;
2079         int i;
2080         int err = 0;
2081
2082         if (!dpaa2_eth_hash_enabled(priv)) {
2083                 dev_dbg(dev, "Hashing support is not enabled\n");
2084                 return 0;
2085         }
2086
2087         memset(&cls_cfg, 0, sizeof(cls_cfg));
2088
2089         for (i = 0; i < ARRAY_SIZE(hash_fields); i++) {
2090                 struct dpkg_extract *key =
2091                         &cls_cfg.extracts[cls_cfg.num_extracts];
2092
2093                 if (!(flags & hash_fields[i].rxnfc_field))
2094                         continue;
2095
2096                 if (cls_cfg.num_extracts >= DPKG_MAX_NUM_OF_EXTRACTS) {
2097                         dev_err(dev, "error adding key extraction rule, too many rules?\n");
2098                         return -E2BIG;
2099                 }
2100
2101                 key->type = DPKG_EXTRACT_FROM_HDR;
2102                 key->extract.from_hdr.prot = hash_fields[i].cls_prot;
2103                 key->extract.from_hdr.type = DPKG_FULL_FIELD;
2104                 key->extract.from_hdr.field = hash_fields[i].cls_field;
2105                 cls_cfg.num_extracts++;
2106
2107                 priv->rx_hash_fields |= hash_fields[i].rxnfc_field;
2108         }
2109
2110         dma_mem = kzalloc(DPAA2_CLASSIFIER_DMA_SIZE, GFP_KERNEL);
2111         if (!dma_mem)
2112                 return -ENOMEM;
2113
2114         err = dpni_prepare_key_cfg(&cls_cfg, dma_mem);
2115         if (err) {
2116                 dev_err(dev, "dpni_prepare_key_cfg error %d\n", err);
2117                 goto err_prep_key;
2118         }
2119
2120         memset(&dist_cfg, 0, sizeof(dist_cfg));
2121
2122         /* Prepare for setting the rx dist */
2123         dist_cfg.key_cfg_iova = dma_map_single(dev, dma_mem,
2124                                                DPAA2_CLASSIFIER_DMA_SIZE,
2125                                                DMA_TO_DEVICE);
2126         if (dma_mapping_error(dev, dist_cfg.key_cfg_iova)) {
2127                 dev_err(dev, "DMA mapping failed\n");
2128                 err = -ENOMEM;
2129                 goto err_dma_map;
2130         }
2131
2132         dist_cfg.dist_size = dpaa2_eth_queue_count(priv);
2133         dist_cfg.dist_mode = DPNI_DIST_MODE_HASH;
2134
2135         err = dpni_set_rx_tc_dist(priv->mc_io, 0, priv->mc_token, 0, &dist_cfg);
2136         dma_unmap_single(dev, dist_cfg.key_cfg_iova,
2137                          DPAA2_CLASSIFIER_DMA_SIZE, DMA_TO_DEVICE);
2138         if (err)
2139                 dev_err(dev, "dpni_set_rx_tc_dist() error %d\n", err);
2140
2141 err_dma_map:
2142 err_prep_key:
2143         kfree(dma_mem);
2144         return err;
2145 }
2146
2147 /* Bind the DPNI to its needed objects and resources: buffer pool, DPIOs,
2148  * frame queues and channels
2149  */
2150 static int bind_dpni(struct dpaa2_eth_priv *priv)
2151 {
2152         struct net_device *net_dev = priv->net_dev;
2153         struct device *dev = net_dev->dev.parent;
2154         struct dpni_pools_cfg pools_params;
2155         struct dpni_error_cfg err_cfg;
2156         int err = 0;
2157         int i;
2158
2159         pools_params.num_dpbp = 1;
2160         pools_params.pools[0].dpbp_id = priv->dpbp_dev->obj_desc.id;
2161         pools_params.pools[0].backup_pool = 0;
2162         pools_params.pools[0].buffer_size = DPAA2_ETH_RX_BUF_SIZE;
2163         err = dpni_set_pools(priv->mc_io, 0, priv->mc_token, &pools_params);
2164         if (err) {
2165                 dev_err(dev, "dpni_set_pools() failed\n");
2166                 return err;
2167         }
2168
2169         /* have the interface implicitly distribute traffic based on
2170          * the default hash key
2171          */
2172         err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_DEFAULT);
2173         if (err)
2174                 dev_err(dev, "Failed to configure hashing\n");
2175
2176         /* Configure handling of error frames */
2177         err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
2178         err_cfg.set_frame_annotation = 1;
2179         err_cfg.error_action = DPNI_ERROR_ACTION_DISCARD;
2180         err = dpni_set_errors_behavior(priv->mc_io, 0, priv->mc_token,
2181                                        &err_cfg);
2182         if (err) {
2183                 dev_err(dev, "dpni_set_errors_behavior failed\n");
2184                 return err;
2185         }
2186
2187         /* Configure Rx and Tx conf queues to generate CDANs */
2188         for (i = 0; i < priv->num_fqs; i++) {
2189                 switch (priv->fq[i].type) {
2190                 case DPAA2_RX_FQ:
2191                         err = setup_rx_flow(priv, &priv->fq[i]);
2192                         break;
2193                 case DPAA2_TX_CONF_FQ:
2194                         err = setup_tx_flow(priv, &priv->fq[i]);
2195                         break;
2196                 default:
2197                         dev_err(dev, "Invalid FQ type %d\n", priv->fq[i].type);
2198                         return -EINVAL;
2199                 }
2200                 if (err)
2201                         return err;
2202         }
2203
2204         err = dpni_get_qdid(priv->mc_io, 0, priv->mc_token,
2205                             DPNI_QUEUE_TX, &priv->tx_qdid);
2206         if (err) {
2207                 dev_err(dev, "dpni_get_qdid() failed\n");
2208                 return err;
2209         }
2210
2211         return 0;
2212 }
2213
2214 /* Allocate rings for storing incoming frame descriptors */
2215 static int alloc_rings(struct dpaa2_eth_priv *priv)
2216 {
2217         struct net_device *net_dev = priv->net_dev;
2218         struct device *dev = net_dev->dev.parent;
2219         int i;
2220
2221         for (i = 0; i < priv->num_channels; i++) {
2222                 priv->channel[i]->store =
2223                         dpaa2_io_store_create(DPAA2_ETH_STORE_SIZE, dev);
2224                 if (!priv->channel[i]->store) {
2225                         netdev_err(net_dev, "dpaa2_io_store_create() failed\n");
2226                         goto err_ring;
2227                 }
2228         }
2229
2230         return 0;
2231
2232 err_ring:
2233         for (i = 0; i < priv->num_channels; i++) {
2234                 if (!priv->channel[i]->store)
2235                         break;
2236                 dpaa2_io_store_destroy(priv->channel[i]->store);
2237         }
2238
2239         return -ENOMEM;
2240 }
2241
2242 static void free_rings(struct dpaa2_eth_priv *priv)
2243 {
2244         int i;
2245
2246         for (i = 0; i < priv->num_channels; i++)
2247                 dpaa2_io_store_destroy(priv->channel[i]->store);
2248 }
2249
2250 static int set_mac_addr(struct dpaa2_eth_priv *priv)
2251 {
2252         struct net_device *net_dev = priv->net_dev;
2253         struct device *dev = net_dev->dev.parent;
2254         u8 mac_addr[ETH_ALEN], dpni_mac_addr[ETH_ALEN];
2255         int err;
2256
2257         /* Get firmware address, if any */
2258         err = dpni_get_port_mac_addr(priv->mc_io, 0, priv->mc_token, mac_addr);
2259         if (err) {
2260                 dev_err(dev, "dpni_get_port_mac_addr() failed\n");
2261                 return err;
2262         }
2263
2264         /* Get DPNI attributes address, if any */
2265         err = dpni_get_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
2266                                         dpni_mac_addr);
2267         if (err) {
2268                 dev_err(dev, "dpni_get_primary_mac_addr() failed\n");
2269                 return err;
2270         }
2271
2272         /* First check if firmware has any address configured by bootloader */
2273         if (!is_zero_ether_addr(mac_addr)) {
2274                 /* If the DPMAC addr != DPNI addr, update it */
2275                 if (!ether_addr_equal(mac_addr, dpni_mac_addr)) {
2276                         err = dpni_set_primary_mac_addr(priv->mc_io, 0,
2277                                                         priv->mc_token,
2278                                                         mac_addr);
2279                         if (err) {
2280                                 dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
2281                                 return err;
2282                         }
2283                 }
2284                 memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
2285         } else if (is_zero_ether_addr(dpni_mac_addr)) {
2286                 /* No MAC address configured, fill in net_dev->dev_addr
2287                  * with a random one
2288                  */
2289                 eth_hw_addr_random(net_dev);
2290                 dev_dbg_once(dev, "device(s) have all-zero hwaddr, replaced with random\n");
2291
2292                 err = dpni_set_primary_mac_addr(priv->mc_io, 0, priv->mc_token,
2293                                                 net_dev->dev_addr);
2294                 if (err) {
2295                         dev_err(dev, "dpni_set_primary_mac_addr() failed\n");
2296                         return err;
2297                 }
2298
2299                 /* Override NET_ADDR_RANDOM set by eth_hw_addr_random(); for all
2300                  * practical purposes, this will be our "permanent" mac address,
2301                  * at least until the next reboot. This move will also permit
2302                  * register_netdevice() to properly fill up net_dev->perm_addr.
2303                  */
2304                 net_dev->addr_assign_type = NET_ADDR_PERM;
2305         } else {
2306                 /* NET_ADDR_PERM is default, all we have to do is
2307                  * fill in the device addr.
2308                  */
2309                 memcpy(net_dev->dev_addr, dpni_mac_addr, net_dev->addr_len);
2310         }
2311
2312         return 0;
2313 }
2314
2315 static int netdev_init(struct net_device *net_dev)
2316 {
2317         struct device *dev = net_dev->dev.parent;
2318         struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
2319         u8 bcast_addr[ETH_ALEN];
2320         u8 num_queues;
2321         int err;
2322
2323         net_dev->netdev_ops = &dpaa2_eth_ops;
2324
2325         err = set_mac_addr(priv);
2326         if (err)
2327                 return err;
2328
2329         /* Explicitly add the broadcast address to the MAC filtering table */
2330         eth_broadcast_addr(bcast_addr);
2331         err = dpni_add_mac_addr(priv->mc_io, 0, priv->mc_token, bcast_addr);
2332         if (err) {
2333                 dev_err(dev, "dpni_add_mac_addr() failed\n");
2334                 return err;
2335         }
2336
2337         /* Set MTU upper limit; lower limit is 68B (default value) */
2338         net_dev->max_mtu = DPAA2_ETH_MAX_MTU;
2339         err = dpni_set_max_frame_length(priv->mc_io, 0, priv->mc_token,
2340                                         DPAA2_ETH_MFL);
2341         if (err) {
2342                 dev_err(dev, "dpni_set_max_frame_length() failed\n");
2343                 return err;
2344         }
2345
2346         /* Set actual number of queues in the net device */
2347         num_queues = dpaa2_eth_queue_count(priv);
2348         err = netif_set_real_num_tx_queues(net_dev, num_queues);
2349         if (err) {
2350                 dev_err(dev, "netif_set_real_num_tx_queues() failed\n");
2351                 return err;
2352         }
2353         err = netif_set_real_num_rx_queues(net_dev, num_queues);
2354         if (err) {
2355                 dev_err(dev, "netif_set_real_num_rx_queues() failed\n");
2356                 return err;
2357         }
2358
2359         /* Our .ndo_init will be called herein */
2360         err = register_netdev(net_dev);
2361         if (err < 0) {
2362                 dev_err(dev, "register_netdev() failed\n");
2363                 return err;
2364         }
2365
2366         return 0;
2367 }
2368
2369 static int poll_link_state(void *arg)
2370 {
2371         struct dpaa2_eth_priv *priv = (struct dpaa2_eth_priv *)arg;
2372         int err;
2373
2374         while (!kthread_should_stop()) {
2375                 err = link_state_update(priv);
2376                 if (unlikely(err))
2377                         return err;
2378
2379                 msleep(DPAA2_ETH_LINK_STATE_REFRESH);
2380         }
2381
2382         return 0;
2383 }
2384
2385 static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
2386 {
2387         u32 status = ~0;
2388         struct device *dev = (struct device *)arg;
2389         struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
2390         struct net_device *net_dev = dev_get_drvdata(dev);
2391         int err;
2392
2393         err = dpni_get_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
2394                                   DPNI_IRQ_INDEX, &status);
2395         if (unlikely(err)) {
2396                 netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
2397                 return IRQ_HANDLED;
2398         }
2399
2400         if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
2401                 link_state_update(netdev_priv(net_dev));
2402
2403         return IRQ_HANDLED;
2404 }
2405
2406 static int setup_irqs(struct fsl_mc_device *ls_dev)
2407 {
2408         int err = 0;
2409         struct fsl_mc_device_irq *irq;
2410
2411         err = fsl_mc_allocate_irqs(ls_dev);
2412         if (err) {
2413                 dev_err(&ls_dev->dev, "MC irqs allocation failed\n");
2414                 return err;
2415         }
2416
2417         irq = ls_dev->irqs[0];
2418         err = devm_request_threaded_irq(&ls_dev->dev, irq->msi_desc->irq,
2419                                         NULL, dpni_irq0_handler_thread,
2420                                         IRQF_NO_SUSPEND | IRQF_ONESHOT,
2421                                         dev_name(&ls_dev->dev), &ls_dev->dev);
2422         if (err < 0) {
2423                 dev_err(&ls_dev->dev, "devm_request_threaded_irq(): %d\n", err);
2424                 goto free_mc_irq;
2425         }
2426
2427         err = dpni_set_irq_mask(ls_dev->mc_io, 0, ls_dev->mc_handle,
2428                                 DPNI_IRQ_INDEX, DPNI_IRQ_EVENT_LINK_CHANGED);
2429         if (err < 0) {
2430                 dev_err(&ls_dev->dev, "dpni_set_irq_mask(): %d\n", err);
2431                 goto free_irq;
2432         }
2433
2434         err = dpni_set_irq_enable(ls_dev->mc_io, 0, ls_dev->mc_handle,
2435                                   DPNI_IRQ_INDEX, 1);
2436         if (err < 0) {
2437                 dev_err(&ls_dev->dev, "dpni_set_irq_enable(): %d\n", err);
2438                 goto free_irq;
2439         }
2440
2441         return 0;
2442
2443 free_irq:
2444         devm_free_irq(&ls_dev->dev, irq->msi_desc->irq, &ls_dev->dev);
2445 free_mc_irq:
2446         fsl_mc_free_irqs(ls_dev);
2447
2448         return err;
2449 }
2450
2451 static void add_ch_napi(struct dpaa2_eth_priv *priv)
2452 {
2453         int i;
2454         struct dpaa2_eth_channel *ch;
2455
2456         for (i = 0; i < priv->num_channels; i++) {
2457                 ch = priv->channel[i];
2458                 /* NAPI weight *MUST* be a multiple of DPAA2_ETH_STORE_SIZE */
2459                 netif_napi_add(priv->net_dev, &ch->napi, dpaa2_eth_poll,
2460                                NAPI_POLL_WEIGHT);
2461         }
2462 }
2463
2464 static void del_ch_napi(struct dpaa2_eth_priv *priv)
2465 {
2466         int i;
2467         struct dpaa2_eth_channel *ch;
2468
2469         for (i = 0; i < priv->num_channels; i++) {
2470                 ch = priv->channel[i];
2471                 netif_napi_del(&ch->napi);
2472         }
2473 }
2474
2475 static int dpaa2_eth_probe(struct fsl_mc_device *dpni_dev)
2476 {
2477         struct device *dev;
2478         struct net_device *net_dev = NULL;
2479         struct dpaa2_eth_priv *priv = NULL;
2480         int err = 0;
2481
2482         dev = &dpni_dev->dev;
2483
2484         /* Net device */
2485         net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA2_ETH_MAX_TX_QUEUES);
2486         if (!net_dev) {
2487                 dev_err(dev, "alloc_etherdev_mq() failed\n");
2488                 return -ENOMEM;
2489         }
2490
2491         SET_NETDEV_DEV(net_dev, dev);
2492         dev_set_drvdata(dev, net_dev);
2493
2494         priv = netdev_priv(net_dev);
2495         priv->net_dev = net_dev;
2496
2497         priv->iommu_domain = iommu_get_domain_for_dev(dev);
2498
2499         /* Obtain a MC portal */
2500         err = fsl_mc_portal_allocate(dpni_dev, FSL_MC_IO_ATOMIC_CONTEXT_PORTAL,
2501                                      &priv->mc_io);
2502         if (err) {
2503                 if (err == -ENXIO)
2504                         err = -EPROBE_DEFER;
2505                 else
2506                         dev_err(dev, "MC portal allocation failed\n");
2507                 goto err_portal_alloc;
2508         }
2509
2510         /* MC objects initialization and configuration */
2511         err = setup_dpni(dpni_dev);
2512         if (err)
2513                 goto err_dpni_setup;
2514
2515         err = setup_dpio(priv);
2516         if (err)
2517                 goto err_dpio_setup;
2518
2519         setup_fqs(priv);
2520
2521         err = setup_dpbp(priv);
2522         if (err)
2523                 goto err_dpbp_setup;
2524
2525         err = bind_dpni(priv);
2526         if (err)
2527                 goto err_bind;
2528
2529         /* Add a NAPI context for each channel */
2530         add_ch_napi(priv);
2531
2532         /* Percpu statistics */
2533         priv->percpu_stats = alloc_percpu(*priv->percpu_stats);
2534         if (!priv->percpu_stats) {
2535                 dev_err(dev, "alloc_percpu(percpu_stats) failed\n");
2536                 err = -ENOMEM;
2537                 goto err_alloc_percpu_stats;
2538         }
2539         priv->percpu_extras = alloc_percpu(*priv->percpu_extras);
2540         if (!priv->percpu_extras) {
2541                 dev_err(dev, "alloc_percpu(percpu_extras) failed\n");
2542                 err = -ENOMEM;
2543                 goto err_alloc_percpu_extras;
2544         }
2545
2546         err = netdev_init(net_dev);
2547         if (err)
2548                 goto err_netdev_init;
2549
2550         /* Configure checksum offload based on current interface flags */
2551         err = set_rx_csum(priv, !!(net_dev->features & NETIF_F_RXCSUM));
2552         if (err)
2553                 goto err_csum;
2554
2555         err = set_tx_csum(priv, !!(net_dev->features &
2556                                    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)));
2557         if (err)
2558                 goto err_csum;
2559
2560         err = alloc_rings(priv);
2561         if (err)
2562                 goto err_alloc_rings;
2563
2564         net_dev->ethtool_ops = &dpaa2_ethtool_ops;
2565
2566         err = setup_irqs(dpni_dev);
2567         if (err) {
2568                 netdev_warn(net_dev, "Failed to set link interrupt, fall back to polling\n");
2569                 priv->poll_thread = kthread_run(poll_link_state, priv,
2570                                                 "%s_poll_link", net_dev->name);
2571                 if (IS_ERR(priv->poll_thread)) {
2572                         netdev_err(net_dev, "Error starting polling thread\n");
2573                         goto err_poll_thread;
2574                 }
2575                 priv->do_link_poll = true;
2576         }
2577
2578         dev_info(dev, "Probed interface %s\n", net_dev->name);
2579         return 0;
2580
2581 err_poll_thread:
2582         free_rings(priv);
2583 err_alloc_rings:
2584 err_csum:
2585         unregister_netdev(net_dev);
2586 err_netdev_init:
2587         free_percpu(priv->percpu_extras);
2588 err_alloc_percpu_extras:
2589         free_percpu(priv->percpu_stats);
2590 err_alloc_percpu_stats:
2591         del_ch_napi(priv);
2592 err_bind:
2593         free_dpbp(priv);
2594 err_dpbp_setup:
2595         free_dpio(priv);
2596 err_dpio_setup:
2597         free_dpni(priv);
2598 err_dpni_setup:
2599         fsl_mc_portal_free(priv->mc_io);
2600 err_portal_alloc:
2601         dev_set_drvdata(dev, NULL);
2602         free_netdev(net_dev);
2603
2604         return err;
2605 }
2606
2607 static int dpaa2_eth_remove(struct fsl_mc_device *ls_dev)
2608 {
2609         struct device *dev;
2610         struct net_device *net_dev;
2611         struct dpaa2_eth_priv *priv;
2612
2613         dev = &ls_dev->dev;
2614         net_dev = dev_get_drvdata(dev);
2615         priv = netdev_priv(net_dev);
2616
2617         unregister_netdev(net_dev);
2618
2619         if (priv->do_link_poll)
2620                 kthread_stop(priv->poll_thread);
2621         else
2622                 fsl_mc_free_irqs(ls_dev);
2623
2624         free_rings(priv);
2625         free_percpu(priv->percpu_stats);
2626         free_percpu(priv->percpu_extras);
2627
2628         del_ch_napi(priv);
2629         free_dpbp(priv);
2630         free_dpio(priv);
2631         free_dpni(priv);
2632
2633         fsl_mc_portal_free(priv->mc_io);
2634
2635         free_netdev(net_dev);
2636
2637         dev_dbg(net_dev->dev.parent, "Removed interface %s\n", net_dev->name);
2638
2639         return 0;
2640 }
2641
2642 static const struct fsl_mc_device_id dpaa2_eth_match_id_table[] = {
2643         {
2644                 .vendor = FSL_MC_VENDOR_FREESCALE,
2645                 .obj_type = "dpni",
2646         },
2647         { .vendor = 0x0 }
2648 };
2649 MODULE_DEVICE_TABLE(fslmc, dpaa2_eth_match_id_table);
2650
2651 static struct fsl_mc_driver dpaa2_eth_driver = {
2652         .driver = {
2653                 .name = KBUILD_MODNAME,
2654                 .owner = THIS_MODULE,
2655         },
2656         .probe = dpaa2_eth_probe,
2657         .remove = dpaa2_eth_remove,
2658         .match_id_table = dpaa2_eth_match_id_table
2659 };
2660
2661 module_fsl_mc_driver(dpaa2_eth_driver);