linux/drivers/net/ethernet/sfc/farch.c
<<
>>
Prefs
   1/****************************************************************************
   2 * Driver for Solarflare network controllers and boards
   3 * Copyright 2005-2006 Fen Systems Ltd.
   4 * Copyright 2006-2013 Solarflare Communications Inc.
   5 *
   6 * This program is free software; you can redistribute it and/or modify it
   7 * under the terms of the GNU General Public License version 2 as published
   8 * by the Free Software Foundation, incorporated herein by reference.
   9 */
  10
  11#include <linux/bitops.h>
  12#include <linux/delay.h>
  13#include <linux/interrupt.h>
  14#include <linux/pci.h>
  15#include <linux/module.h>
  16#include <linux/seq_file.h>
  17#include <linux/crc32.h>
  18#include "net_driver.h"
  19#include "bitfield.h"
  20#include "efx.h"
  21#include "nic.h"
  22#include "farch_regs.h"
  23#include "sriov.h"
  24#include "siena_sriov.h"
  25#include "io.h"
  26#include "workarounds.h"
  27
  28/* Falcon-architecture (SFC4000 and SFC9000-family) support */
  29
  30/**************************************************************************
  31 *
  32 * Configurable values
  33 *
  34 **************************************************************************
  35 */
  36
  37/* This is set to 16 for a good reason.  In summary, if larger than
  38 * 16, the descriptor cache holds more than a default socket
  39 * buffer's worth of packets (for UDP we can only have at most one
  40 * socket buffer's worth outstanding).  This combined with the fact
  41 * that we only get 1 TX event per descriptor cache means the NIC
  42 * goes idle.
  43 */
  44#define TX_DC_ENTRIES 16
  45#define TX_DC_ENTRIES_ORDER 1
  46
  47#define RX_DC_ENTRIES 64
  48#define RX_DC_ENTRIES_ORDER 3
  49
  50/* If EFX_MAX_INT_ERRORS internal errors occur within
  51 * EFX_INT_ERROR_EXPIRE seconds, we consider the NIC broken and
  52 * disable it.
  53 */
  54#define EFX_INT_ERROR_EXPIRE 3600
  55#define EFX_MAX_INT_ERRORS 5
  56
  57/* Depth of RX flush request fifo */
  58#define EFX_RX_FLUSH_COUNT 4
  59
  60/* Driver generated events */
  61#define _EFX_CHANNEL_MAGIC_TEST         0x000101
  62#define _EFX_CHANNEL_MAGIC_FILL         0x000102
  63#define _EFX_CHANNEL_MAGIC_RX_DRAIN     0x000103
  64#define _EFX_CHANNEL_MAGIC_TX_DRAIN     0x000104
  65
  66#define _EFX_CHANNEL_MAGIC(_code, _data)        ((_code) << 8 | (_data))
  67#define _EFX_CHANNEL_MAGIC_CODE(_magic)         ((_magic) >> 8)
  68
  69#define EFX_CHANNEL_MAGIC_TEST(_channel)                                \
  70        _EFX_CHANNEL_MAGIC(_EFX_CHANNEL_MAGIC_TEST, (_channel)->channel)
  71#define EFX_CHANNEL_MAGIC_FILL(_rx_queue)                               \
  72        _EFX_CHANNEL_MAGIC(_EFX_CHANNEL_MAGIC_FILL,                     \
  73                           efx_rx_queue_index(_rx_queue))
  74#define EFX_CHANNEL_MAGIC_RX_DRAIN(_rx_queue)                           \
  75        _EFX_CHANNEL_MAGIC(_EFX_CHANNEL_MAGIC_RX_DRAIN,                 \
  76                           efx_rx_queue_index(_rx_queue))
  77#define EFX_CHANNEL_MAGIC_TX_DRAIN(_tx_queue)                           \
  78        _EFX_CHANNEL_MAGIC(_EFX_CHANNEL_MAGIC_TX_DRAIN,                 \
  79                           (_tx_queue)->queue)
  80
  81static void efx_farch_magic_event(struct efx_channel *channel, u32 magic);
  82
  83/**************************************************************************
  84 *
  85 * Hardware access
  86 *
  87 **************************************************************************/
  88
  89static inline void efx_write_buf_tbl(struct efx_nic *efx, efx_qword_t *value,
  90                                     unsigned int index)
  91{
  92        efx_sram_writeq(efx, efx->membase + efx->type->buf_tbl_base,
  93                        value, index);
  94}
  95
  96static bool efx_masked_compare_oword(const efx_oword_t *a, const efx_oword_t *b,
  97                                     const efx_oword_t *mask)
  98{
  99        return ((a->u64[0] ^ b->u64[0]) & mask->u64[0]) ||
 100                ((a->u64[1] ^ b->u64[1]) & mask->u64[1]);
 101}
 102
 103int efx_farch_test_registers(struct efx_nic *efx,
 104                             const struct efx_farch_register_test *regs,
 105                             size_t n_regs)
 106{
 107        unsigned address = 0, i, j;
 108        efx_oword_t mask, imask, original, reg, buf;
 109
 110        for (i = 0; i < n_regs; ++i) {
 111                address = regs[i].address;
 112                mask = imask = regs[i].mask;
 113                EFX_INVERT_OWORD(imask);
 114
 115                efx_reado(efx, &original, address);
 116
 117                /* bit sweep on and off */
 118                for (j = 0; j < 128; j++) {
 119                        if (!EFX_EXTRACT_OWORD32(mask, j, j))
 120                                continue;
 121
 122                        /* Test this testable bit can be set in isolation */
 123                        EFX_AND_OWORD(reg, original, mask);
 124                        EFX_SET_OWORD32(reg, j, j, 1);
 125
 126                        efx_writeo(efx, &reg, address);
 127                        efx_reado(efx, &buf, address);
 128
 129                        if (efx_masked_compare_oword(&reg, &buf, &mask))
 130                                goto fail;
 131
 132                        /* Test this testable bit can be cleared in isolation */
 133                        EFX_OR_OWORD(reg, original, mask);
 134                        EFX_SET_OWORD32(reg, j, j, 0);
 135
 136                        efx_writeo(efx, &reg, address);
 137                        efx_reado(efx, &buf, address);
 138
 139                        if (efx_masked_compare_oword(&reg, &buf, &mask))
 140                                goto fail;
 141                }
 142
 143                efx_writeo(efx, &original, address);
 144        }
 145
 146        return 0;
 147
 148fail:
 149        netif_err(efx, hw, efx->net_dev,
 150                  "wrote "EFX_OWORD_FMT" read "EFX_OWORD_FMT
 151                  " at address 0x%x mask "EFX_OWORD_FMT"\n", EFX_OWORD_VAL(reg),
 152                  EFX_OWORD_VAL(buf), address, EFX_OWORD_VAL(mask));
 153        return -EIO;
 154}
 155
 156/**************************************************************************
 157 *
 158 * Special buffer handling
 159 * Special buffers are used for event queues and the TX and RX
 160 * descriptor rings.
 161 *
 162 *************************************************************************/
 163
 164/*
 165 * Initialise a special buffer
 166 *
 167 * This will define a buffer (previously allocated via
 168 * efx_alloc_special_buffer()) in the buffer table, allowing
 169 * it to be used for event queues, descriptor rings etc.
 170 */
 171static void
 172efx_init_special_buffer(struct efx_nic *efx, struct efx_special_buffer *buffer)
 173{
 174        efx_qword_t buf_desc;
 175        unsigned int index;
 176        dma_addr_t dma_addr;
 177        int i;
 178
 179        EFX_BUG_ON_PARANOID(!buffer->buf.addr);
 180
 181        /* Write buffer descriptors to NIC */
 182        for (i = 0; i < buffer->entries; i++) {
 183                index = buffer->index + i;
 184                dma_addr = buffer->buf.dma_addr + (i * EFX_BUF_SIZE);
 185                netif_dbg(efx, probe, efx->net_dev,
 186                          "mapping special buffer %d at %llx\n",
 187                          index, (unsigned long long)dma_addr);
 188                EFX_POPULATE_QWORD_3(buf_desc,
 189                                     FRF_AZ_BUF_ADR_REGION, 0,
 190                                     FRF_AZ_BUF_ADR_FBUF, dma_addr >> 12,
 191                                     FRF_AZ_BUF_OWNER_ID_FBUF, 0);
 192                efx_write_buf_tbl(efx, &buf_desc, index);
 193        }
 194}
 195
 196/* Unmaps a buffer and clears the buffer table entries */
 197static void
 198efx_fini_special_buffer(struct efx_nic *efx, struct efx_special_buffer *buffer)
 199{
 200        efx_oword_t buf_tbl_upd;
 201        unsigned int start = buffer->index;
 202        unsigned int end = (buffer->index + buffer->entries - 1);
 203
 204        if (!buffer->entries)
 205                return;
 206
 207        netif_dbg(efx, hw, efx->net_dev, "unmapping special buffers %d-%d\n",
 208                  buffer->index, buffer->index + buffer->entries - 1);
 209
 210        EFX_POPULATE_OWORD_4(buf_tbl_upd,
 211                             FRF_AZ_BUF_UPD_CMD, 0,
 212                             FRF_AZ_BUF_CLR_CMD, 1,
 213                             FRF_AZ_BUF_CLR_END_ID, end,
 214                             FRF_AZ_BUF_CLR_START_ID, start);
 215        efx_writeo(efx, &buf_tbl_upd, FR_AZ_BUF_TBL_UPD);
 216}
 217
 218/*
 219 * Allocate a new special buffer
 220 *
 221 * This allocates memory for a new buffer, clears it and allocates a
 222 * new buffer ID range.  It does not write into the buffer table.
 223 *
 224 * This call will allocate 4KB buffers, since 8KB buffers can't be
 225 * used for event queues and descriptor rings.
 226 */
 227static int efx_alloc_special_buffer(struct efx_nic *efx,
 228                                    struct efx_special_buffer *buffer,
 229                                    unsigned int len)
 230{
 231#ifdef CONFIG_SFC_SRIOV
 232        struct siena_nic_data *nic_data = efx->nic_data;
 233#endif
 234        len = ALIGN(len, EFX_BUF_SIZE);
 235
 236        if (efx_nic_alloc_buffer(efx, &buffer->buf, len, GFP_KERNEL))
 237                return -ENOMEM;
 238        buffer->entries = len / EFX_BUF_SIZE;
 239        BUG_ON(buffer->buf.dma_addr & (EFX_BUF_SIZE - 1));
 240
 241        /* Select new buffer ID */
 242        buffer->index = efx->next_buffer_table;
 243        efx->next_buffer_table += buffer->entries;
 244#ifdef CONFIG_SFC_SRIOV
 245        BUG_ON(efx_siena_sriov_enabled(efx) &&
 246               nic_data->vf_buftbl_base < efx->next_buffer_table);
 247#endif
 248
 249        netif_dbg(efx, probe, efx->net_dev,
 250                  "allocating special buffers %d-%d at %llx+%x "
 251                  "(virt %p phys %llx)\n", buffer->index,
 252                  buffer->index + buffer->entries - 1,
 253                  (u64)buffer->buf.dma_addr, len,
 254                  buffer->buf.addr, (u64)virt_to_phys(buffer->buf.addr));
 255
 256        return 0;
 257}
 258
 259static void
 260efx_free_special_buffer(struct efx_nic *efx, struct efx_special_buffer *buffer)
 261{
 262        if (!buffer->buf.addr)
 263                return;
 264
 265        netif_dbg(efx, hw, efx->net_dev,
 266                  "deallocating special buffers %d-%d at %llx+%x "
 267                  "(virt %p phys %llx)\n", buffer->index,
 268                  buffer->index + buffer->entries - 1,
 269                  (u64)buffer->buf.dma_addr, buffer->buf.len,
 270                  buffer->buf.addr, (u64)virt_to_phys(buffer->buf.addr));
 271
 272        efx_nic_free_buffer(efx, &buffer->buf);
 273        buffer->entries = 0;
 274}
 275
 276/**************************************************************************
 277 *
 278 * TX path
 279 *
 280 **************************************************************************/
 281
 282/* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
 283static inline void efx_farch_notify_tx_desc(struct efx_tx_queue *tx_queue)
 284{
 285        unsigned write_ptr;
 286        efx_dword_t reg;
 287
 288        write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
 289        EFX_POPULATE_DWORD_1(reg, FRF_AZ_TX_DESC_WPTR_DWORD, write_ptr);
 290        efx_writed_page(tx_queue->efx, &reg,
 291                        FR_AZ_TX_DESC_UPD_DWORD_P0, tx_queue->queue);
 292}
 293
 294/* Write pointer and first descriptor for TX descriptor ring */
 295static inline void efx_farch_push_tx_desc(struct efx_tx_queue *tx_queue,
 296                                          const efx_qword_t *txd)
 297{
 298        unsigned write_ptr;
 299        efx_oword_t reg;
 300
 301        BUILD_BUG_ON(FRF_AZ_TX_DESC_LBN != 0);
 302        BUILD_BUG_ON(FR_AA_TX_DESC_UPD_KER != FR_BZ_TX_DESC_UPD_P0);
 303
 304        write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
 305        EFX_POPULATE_OWORD_2(reg, FRF_AZ_TX_DESC_PUSH_CMD, true,
 306                             FRF_AZ_TX_DESC_WPTR, write_ptr);
 307        reg.qword[0] = *txd;
 308        efx_writeo_page(tx_queue->efx, &reg,
 309                        FR_BZ_TX_DESC_UPD_P0, tx_queue->queue);
 310}
 311
 312
 313/* For each entry inserted into the software descriptor ring, create a
 314 * descriptor in the hardware TX descriptor ring (in host memory), and
 315 * write a doorbell.
 316 */
 317void efx_farch_tx_write(struct efx_tx_queue *tx_queue)
 318{
 319        struct efx_tx_buffer *buffer;
 320        efx_qword_t *txd;
 321        unsigned write_ptr;
 322        unsigned old_write_count = tx_queue->write_count;
 323
 324        tx_queue->xmit_more_available = false;
 325        if (unlikely(tx_queue->write_count == tx_queue->insert_count))
 326                return;
 327
 328        do {
 329                write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
 330                buffer = &tx_queue->buffer[write_ptr];
 331                txd = efx_tx_desc(tx_queue, write_ptr);
 332                ++tx_queue->write_count;
 333
 334                EFX_BUG_ON_PARANOID(buffer->flags & EFX_TX_BUF_OPTION);
 335
 336                /* Create TX descriptor ring entry */
 337                BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
 338                EFX_POPULATE_QWORD_4(*txd,
 339                                     FSF_AZ_TX_KER_CONT,
 340                                     buffer->flags & EFX_TX_BUF_CONT,
 341                                     FSF_AZ_TX_KER_BYTE_COUNT, buffer->len,
 342                                     FSF_AZ_TX_KER_BUF_REGION, 0,
 343                                     FSF_AZ_TX_KER_BUF_ADDR, buffer->dma_addr);
 344        } while (tx_queue->write_count != tx_queue->insert_count);
 345
 346        wmb(); /* Ensure descriptors are written before they are fetched */
 347
 348        if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
 349                txd = efx_tx_desc(tx_queue,
 350                                  old_write_count & tx_queue->ptr_mask);
 351                efx_farch_push_tx_desc(tx_queue, txd);
 352                ++tx_queue->pushes;
 353        } else {
 354                efx_farch_notify_tx_desc(tx_queue);
 355        }
 356}
 357
 358/* Allocate hardware resources for a TX queue */
 359int efx_farch_tx_probe(struct efx_tx_queue *tx_queue)
 360{
 361        struct efx_nic *efx = tx_queue->efx;
 362        unsigned entries;
 363
 364        entries = tx_queue->ptr_mask + 1;
 365        return efx_alloc_special_buffer(efx, &tx_queue->txd,
 366                                        entries * sizeof(efx_qword_t));
 367}
 368
 369void efx_farch_tx_init(struct efx_tx_queue *tx_queue)
 370{
 371        struct efx_nic *efx = tx_queue->efx;
 372        efx_oword_t reg;
 373
 374        /* Pin TX descriptor ring */
 375        efx_init_special_buffer(efx, &tx_queue->txd);
 376
 377        /* Push TX descriptor ring to card */
 378        EFX_POPULATE_OWORD_10(reg,
 379                              FRF_AZ_TX_DESCQ_EN, 1,
 380                              FRF_AZ_TX_ISCSI_DDIG_EN, 0,
 381                              FRF_AZ_TX_ISCSI_HDIG_EN, 0,
 382                              FRF_AZ_TX_DESCQ_BUF_BASE_ID, tx_queue->txd.index,
 383                              FRF_AZ_TX_DESCQ_EVQ_ID,
 384                              tx_queue->channel->channel,
 385                              FRF_AZ_TX_DESCQ_OWNER_ID, 0,
 386                              FRF_AZ_TX_DESCQ_LABEL, tx_queue->queue,
 387                              FRF_AZ_TX_DESCQ_SIZE,
 388                              __ffs(tx_queue->txd.entries),
 389                              FRF_AZ_TX_DESCQ_TYPE, 0,
 390                              FRF_BZ_TX_NON_IP_DROP_DIS, 1);
 391
 392        if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
 393                int csum = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
 394                EFX_SET_OWORD_FIELD(reg, FRF_BZ_TX_IP_CHKSM_DIS, !csum);
 395                EFX_SET_OWORD_FIELD(reg, FRF_BZ_TX_TCP_CHKSM_DIS,
 396                                    !csum);
 397        }
 398
 399        efx_writeo_table(efx, &reg, efx->type->txd_ptr_tbl_base,
 400                         tx_queue->queue);
 401
 402        if (efx_nic_rev(efx) < EFX_REV_FALCON_B0) {
 403                /* Only 128 bits in this register */
 404                BUILD_BUG_ON(EFX_MAX_TX_QUEUES > 128);
 405
 406                efx_reado(efx, &reg, FR_AA_TX_CHKSM_CFG);
 407                if (tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD)
 408                        __clear_bit_le(tx_queue->queue, &reg);
 409                else
 410                        __set_bit_le(tx_queue->queue, &reg);
 411                efx_writeo(efx, &reg, FR_AA_TX_CHKSM_CFG);
 412        }
 413
 414        if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
 415                EFX_POPULATE_OWORD_1(reg,
 416                                     FRF_BZ_TX_PACE,
 417                                     (tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI) ?
 418                                     FFE_BZ_TX_PACE_OFF :
 419                                     FFE_BZ_TX_PACE_RESERVED);
 420                efx_writeo_table(efx, &reg, FR_BZ_TX_PACE_TBL,
 421                                 tx_queue->queue);
 422        }
 423}
 424
 425static void efx_farch_flush_tx_queue(struct efx_tx_queue *tx_queue)
 426{
 427        struct efx_nic *efx = tx_queue->efx;
 428        efx_oword_t tx_flush_descq;
 429
 430        WARN_ON(atomic_read(&tx_queue->flush_outstanding));
 431        atomic_set(&tx_queue->flush_outstanding, 1);
 432
 433        EFX_POPULATE_OWORD_2(tx_flush_descq,
 434                             FRF_AZ_TX_FLUSH_DESCQ_CMD, 1,
 435                             FRF_AZ_TX_FLUSH_DESCQ, tx_queue->queue);
 436        efx_writeo(efx, &tx_flush_descq, FR_AZ_TX_FLUSH_DESCQ);
 437}
 438
 439void efx_farch_tx_fini(struct efx_tx_queue *tx_queue)
 440{
 441        struct efx_nic *efx = tx_queue->efx;
 442        efx_oword_t tx_desc_ptr;
 443
 444        /* Remove TX descriptor ring from card */
 445        EFX_ZERO_OWORD(tx_desc_ptr);
 446        efx_writeo_table(efx, &tx_desc_ptr, efx->type->txd_ptr_tbl_base,
 447                         tx_queue->queue);
 448
 449        /* Unpin TX descriptor ring */
 450        efx_fini_special_buffer(efx, &tx_queue->txd);
 451}
 452
 453/* Free buffers backing TX queue */
 454void efx_farch_tx_remove(struct efx_tx_queue *tx_queue)
 455{
 456        efx_free_special_buffer(tx_queue->efx, &tx_queue->txd);
 457}
 458
 459/**************************************************************************
 460 *
 461 * RX path
 462 *
 463 **************************************************************************/
 464
 465/* This creates an entry in the RX descriptor queue */
 466static inline void
 467efx_farch_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned index)
 468{
 469        struct efx_rx_buffer *rx_buf;
 470        efx_qword_t *rxd;
 471
 472        rxd = efx_rx_desc(rx_queue, index);
 473        rx_buf = efx_rx_buffer(rx_queue, index);
 474        EFX_POPULATE_QWORD_3(*rxd,
 475                             FSF_AZ_RX_KER_BUF_SIZE,
 476                             rx_buf->len -
 477                             rx_queue->efx->type->rx_buffer_padding,
 478                             FSF_AZ_RX_KER_BUF_REGION, 0,
 479                             FSF_AZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
 480}
 481
 482/* This writes to the RX_DESC_WPTR register for the specified receive
 483 * descriptor ring.
 484 */
 485void efx_farch_rx_write(struct efx_rx_queue *rx_queue)
 486{
 487        struct efx_nic *efx = rx_queue->efx;
 488        efx_dword_t reg;
 489        unsigned write_ptr;
 490
 491        while (rx_queue->notified_count != rx_queue->added_count) {
 492                efx_farch_build_rx_desc(
 493                        rx_queue,
 494                        rx_queue->notified_count & rx_queue->ptr_mask);
 495                ++rx_queue->notified_count;
 496        }
 497
 498        wmb();
 499        write_ptr = rx_queue->added_count & rx_queue->ptr_mask;
 500        EFX_POPULATE_DWORD_1(reg, FRF_AZ_RX_DESC_WPTR_DWORD, write_ptr);
 501        efx_writed_page(efx, &reg, FR_AZ_RX_DESC_UPD_DWORD_P0,
 502                        efx_rx_queue_index(rx_queue));
 503}
 504
 505int efx_farch_rx_probe(struct efx_rx_queue *rx_queue)
 506{
 507        struct efx_nic *efx = rx_queue->efx;
 508        unsigned entries;
 509
 510        entries = rx_queue->ptr_mask + 1;
 511        return efx_alloc_special_buffer(efx, &rx_queue->rxd,
 512                                        entries * sizeof(efx_qword_t));
 513}
 514
 515void efx_farch_rx_init(struct efx_rx_queue *rx_queue)
 516{
 517        efx_oword_t rx_desc_ptr;
 518        struct efx_nic *efx = rx_queue->efx;
 519        bool is_b0 = efx_nic_rev(efx) >= EFX_REV_FALCON_B0;
 520        bool iscsi_digest_en = is_b0;
 521        bool jumbo_en;
 522
 523        /* For kernel-mode queues in Falcon A1, the JUMBO flag enables
 524         * DMA to continue after a PCIe page boundary (and scattering
 525         * is not possible).  In Falcon B0 and Siena, it enables
 526         * scatter.
 527         */
 528        jumbo_en = !is_b0 || efx->rx_scatter;
 529
 530        netif_dbg(efx, hw, efx->net_dev,
 531                  "RX queue %d ring in special buffers %d-%d\n",
 532                  efx_rx_queue_index(rx_queue), rx_queue->rxd.index,
 533                  rx_queue->rxd.index + rx_queue->rxd.entries - 1);
 534
 535        rx_queue->scatter_n = 0;
 536
 537        /* Pin RX descriptor ring */
 538        efx_init_special_buffer(efx, &rx_queue->rxd);
 539
 540        /* Push RX descriptor ring to card */
 541        EFX_POPULATE_OWORD_10(rx_desc_ptr,
 542                              FRF_AZ_RX_ISCSI_DDIG_EN, iscsi_digest_en,
 543                              FRF_AZ_RX_ISCSI_HDIG_EN, iscsi_digest_en,
 544                              FRF_AZ_RX_DESCQ_BUF_BASE_ID, rx_queue->rxd.index,
 545                              FRF_AZ_RX_DESCQ_EVQ_ID,
 546                              efx_rx_queue_channel(rx_queue)->channel,
 547                              FRF_AZ_RX_DESCQ_OWNER_ID, 0,
 548                              FRF_AZ_RX_DESCQ_LABEL,
 549                              efx_rx_queue_index(rx_queue),
 550                              FRF_AZ_RX_DESCQ_SIZE,
 551                              __ffs(rx_queue->rxd.entries),
 552                              FRF_AZ_RX_DESCQ_TYPE, 0 /* kernel queue */ ,
 553                              FRF_AZ_RX_DESCQ_JUMBO, jumbo_en,
 554                              FRF_AZ_RX_DESCQ_EN, 1);
 555        efx_writeo_table(efx, &rx_desc_ptr, efx->type->rxd_ptr_tbl_base,
 556                         efx_rx_queue_index(rx_queue));
 557}
 558
 559static void efx_farch_flush_rx_queue(struct efx_rx_queue *rx_queue)
 560{
 561        struct efx_nic *efx = rx_queue->efx;
 562        efx_oword_t rx_flush_descq;
 563
 564        EFX_POPULATE_OWORD_2(rx_flush_descq,
 565                             FRF_AZ_RX_FLUSH_DESCQ_CMD, 1,
 566                             FRF_AZ_RX_FLUSH_DESCQ,
 567                             efx_rx_queue_index(rx_queue));
 568        efx_writeo(efx, &rx_flush_descq, FR_AZ_RX_FLUSH_DESCQ);
 569}
 570
 571void efx_farch_rx_fini(struct efx_rx_queue *rx_queue)
 572{
 573        efx_oword_t rx_desc_ptr;
 574        struct efx_nic *efx = rx_queue->efx;
 575
 576        /* Remove RX descriptor ring from card */
 577        EFX_ZERO_OWORD(rx_desc_ptr);
 578        efx_writeo_table(efx, &rx_desc_ptr, efx->type->rxd_ptr_tbl_base,
 579                         efx_rx_queue_index(rx_queue));
 580
 581        /* Unpin RX descriptor ring */
 582        efx_fini_special_buffer(efx, &rx_queue->rxd);
 583}
 584
 585/* Free buffers backing RX queue */
 586void efx_farch_rx_remove(struct efx_rx_queue *rx_queue)
 587{
 588        efx_free_special_buffer(rx_queue->efx, &rx_queue->rxd);
 589}
 590
 591/**************************************************************************
 592 *
 593 * Flush handling
 594 *
 595 **************************************************************************/
 596
 597/* efx_farch_flush_queues() must be woken up when all flushes are completed,
 598 * or more RX flushes can be kicked off.
 599 */
 600static bool efx_farch_flush_wake(struct efx_nic *efx)
 601{
 602        /* Ensure that all updates are visible to efx_farch_flush_queues() */
 603        smp_mb();
 604
 605        return (atomic_read(&efx->active_queues) == 0 ||
 606                (atomic_read(&efx->rxq_flush_outstanding) < EFX_RX_FLUSH_COUNT
 607                 && atomic_read(&efx->rxq_flush_pending) > 0));
 608}
 609
 610static bool efx_check_tx_flush_complete(struct efx_nic *efx)
 611{
 612        bool i = true;
 613        efx_oword_t txd_ptr_tbl;
 614        struct efx_channel *channel;
 615        struct efx_tx_queue *tx_queue;
 616
 617        efx_for_each_channel(channel, efx) {
 618                efx_for_each_channel_tx_queue(tx_queue, channel) {
 619                        efx_reado_table(efx, &txd_ptr_tbl,
 620                                        FR_BZ_TX_DESC_PTR_TBL, tx_queue->queue);
 621                        if (EFX_OWORD_FIELD(txd_ptr_tbl,
 622                                            FRF_AZ_TX_DESCQ_FLUSH) ||
 623                            EFX_OWORD_FIELD(txd_ptr_tbl,
 624                                            FRF_AZ_TX_DESCQ_EN)) {
 625                                netif_dbg(efx, hw, efx->net_dev,
 626                                          "flush did not complete on TXQ %d\n",
 627                                          tx_queue->queue);
 628                                i = false;
 629                        } else if (atomic_cmpxchg(&tx_queue->flush_outstanding,
 630                                                  1, 0)) {
 631                                /* The flush is complete, but we didn't
 632                                 * receive a flush completion event
 633                                 */
 634                                netif_dbg(efx, hw, efx->net_dev,
 635                                          "flush complete on TXQ %d, so drain "
 636                                          "the queue\n", tx_queue->queue);
 637                                /* Don't need to increment active_queues as it
 638                                 * has already been incremented for the queues
 639                                 * which did not drain
 640                                 */
 641                                efx_farch_magic_event(channel,
 642                                                      EFX_CHANNEL_MAGIC_TX_DRAIN(
 643                                                              tx_queue));
 644                        }
 645                }
 646        }
 647
 648        return i;
 649}
 650
 651/* Flush all the transmit queues, and continue flushing receive queues until
 652 * they're all flushed. Wait for the DRAIN events to be received so that there
 653 * are no more RX and TX events left on any channel. */
 654static int efx_farch_do_flush(struct efx_nic *efx)
 655{
 656        unsigned timeout = msecs_to_jiffies(5000); /* 5s for all flushes and drains */
 657        struct efx_channel *channel;
 658        struct efx_rx_queue *rx_queue;
 659        struct efx_tx_queue *tx_queue;
 660        int rc = 0;
 661
 662        efx_for_each_channel(channel, efx) {
 663                efx_for_each_channel_tx_queue(tx_queue, channel) {
 664                        efx_farch_flush_tx_queue(tx_queue);
 665                }
 666                efx_for_each_channel_rx_queue(rx_queue, channel) {
 667                        rx_queue->flush_pending = true;
 668                        atomic_inc(&efx->rxq_flush_pending);
 669                }
 670        }
 671
 672        while (timeout && atomic_read(&efx->active_queues) > 0) {
 673                /* If SRIOV is enabled, then offload receive queue flushing to
 674                 * the firmware (though we will still have to poll for
 675                 * completion). If that fails, fall back to the old scheme.
 676                 */
 677                if (efx_siena_sriov_enabled(efx)) {
 678                        rc = efx_mcdi_flush_rxqs(efx);
 679                        if (!rc)
 680                                goto wait;
 681                }
 682
 683                /* The hardware supports four concurrent rx flushes, each of
 684                 * which may need to be retried if there is an outstanding
 685                 * descriptor fetch
 686                 */
 687                efx_for_each_channel(channel, efx) {
 688                        efx_for_each_channel_rx_queue(rx_queue, channel) {
 689                                if (atomic_read(&efx->rxq_flush_outstanding) >=
 690                                    EFX_RX_FLUSH_COUNT)
 691                                        break;
 692
 693                                if (rx_queue->flush_pending) {
 694                                        rx_queue->flush_pending = false;
 695                                        atomic_dec(&efx->rxq_flush_pending);
 696                                        atomic_inc(&efx->rxq_flush_outstanding);
 697                                        efx_farch_flush_rx_queue(rx_queue);
 698                                }
 699                        }
 700                }
 701
 702        wait:
 703                timeout = wait_event_timeout(efx->flush_wq,
 704                                             efx_farch_flush_wake(efx),
 705                                             timeout);
 706        }
 707
 708        if (atomic_read(&efx->active_queues) &&
 709            !efx_check_tx_flush_complete(efx)) {
 710                netif_err(efx, hw, efx->net_dev, "failed to flush %d queues "
 711                          "(rx %d+%d)\n", atomic_read(&efx->active_queues),
 712                          atomic_read(&efx->rxq_flush_outstanding),
 713                          atomic_read(&efx->rxq_flush_pending));
 714                rc = -ETIMEDOUT;
 715
 716                atomic_set(&efx->active_queues, 0);
 717                atomic_set(&efx->rxq_flush_pending, 0);
 718                atomic_set(&efx->rxq_flush_outstanding, 0);
 719        }
 720
 721        return rc;
 722}
 723
 724int efx_farch_fini_dmaq(struct efx_nic *efx)
 725{
 726        struct efx_channel *channel;
 727        struct efx_tx_queue *tx_queue;
 728        struct efx_rx_queue *rx_queue;
 729        int rc = 0;
 730
 731        /* Do not attempt to write to the NIC during EEH recovery */
 732        if (efx->state != STATE_RECOVERY) {
 733                /* Only perform flush if DMA is enabled */
 734                if (efx->pci_dev->is_busmaster) {
 735                        efx->type->prepare_flush(efx);
 736                        rc = efx_farch_do_flush(efx);
 737                        efx->type->finish_flush(efx);
 738                }
 739
 740                efx_for_each_channel(channel, efx) {
 741                        efx_for_each_channel_rx_queue(rx_queue, channel)
 742                                efx_farch_rx_fini(rx_queue);
 743                        efx_for_each_channel_tx_queue(tx_queue, channel)
 744                                efx_farch_tx_fini(tx_queue);
 745                }
 746        }
 747
 748        return rc;
 749}
 750
 751/* Reset queue and flush accounting after FLR
 752 *
 753 * One possible cause of FLR recovery is that DMA may be failing (eg. if bus
 754 * mastering was disabled), in which case we don't receive (RXQ) flush
 755 * completion events.  This means that efx->rxq_flush_outstanding remained at 4
 756 * after the FLR; also, efx->active_queues was non-zero (as no flush completion
 757 * events were received, and we didn't go through efx_check_tx_flush_complete())
 758 * If we don't fix this up, on the next call to efx_realloc_channels() we won't
 759 * flush any RX queues because efx->rxq_flush_outstanding is at the limit of 4
 760 * for batched flush requests; and the efx->active_queues gets messed up because
 761 * we keep incrementing for the newly initialised queues, but it never went to
 762 * zero previously.  Then we get a timeout every time we try to restart the
 763 * queues, as it doesn't go back to zero when we should be flushing the queues.
 764 */
 765void efx_farch_finish_flr(struct efx_nic *efx)
 766{
 767        atomic_set(&efx->rxq_flush_pending, 0);
 768        atomic_set(&efx->rxq_flush_outstanding, 0);
 769        atomic_set(&efx->active_queues, 0);
 770}
 771
 772
 773/**************************************************************************
 774 *
 775 * Event queue processing
 776 * Event queues are processed by per-channel tasklets.
 777 *
 778 **************************************************************************/
 779
 780/* Update a channel's event queue's read pointer (RPTR) register
 781 *
 782 * This writes the EVQ_RPTR_REG register for the specified channel's
 783 * event queue.
 784 */
 785void efx_farch_ev_read_ack(struct efx_channel *channel)
 786{
 787        efx_dword_t reg;
 788        struct efx_nic *efx = channel->efx;
 789
 790        EFX_POPULATE_DWORD_1(reg, FRF_AZ_EVQ_RPTR,
 791                             channel->eventq_read_ptr & channel->eventq_mask);
 792
 793        /* For Falcon A1, EVQ_RPTR_KER is documented as having a step size
 794         * of 4 bytes, but it is really 16 bytes just like later revisions.
 795         */
 796        efx_writed(efx, &reg,
 797                   efx->type->evq_rptr_tbl_base +
 798                   FR_BZ_EVQ_RPTR_STEP * channel->channel);
 799}
 800
 801/* Use HW to insert a SW defined event */
 802void efx_farch_generate_event(struct efx_nic *efx, unsigned int evq,
 803                              efx_qword_t *event)
 804{
 805        efx_oword_t drv_ev_reg;
 806
 807        BUILD_BUG_ON(FRF_AZ_DRV_EV_DATA_LBN != 0 ||
 808                     FRF_AZ_DRV_EV_DATA_WIDTH != 64);
 809        drv_ev_reg.u32[0] = event->u32[0];
 810        drv_ev_reg.u32[1] = event->u32[1];
 811        drv_ev_reg.u32[2] = 0;
 812        drv_ev_reg.u32[3] = 0;
 813        EFX_SET_OWORD_FIELD(drv_ev_reg, FRF_AZ_DRV_EV_QID, evq);
 814        efx_writeo(efx, &drv_ev_reg, FR_AZ_DRV_EV);
 815}
 816
 817static void efx_farch_magic_event(struct efx_channel *channel, u32 magic)
 818{
 819        efx_qword_t event;
 820
 821        EFX_POPULATE_QWORD_2(event, FSF_AZ_EV_CODE,
 822                             FSE_AZ_EV_CODE_DRV_GEN_EV,
 823                             FSF_AZ_DRV_GEN_EV_MAGIC, magic);
 824        efx_farch_generate_event(channel->efx, channel->channel, &event);
 825}
 826
 827/* Handle a transmit completion event
 828 *
 829 * The NIC batches TX completion events; the message we receive is of
 830 * the form "complete all TX events up to this index".
 831 */
 832static int
 833efx_farch_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
 834{
 835        unsigned int tx_ev_desc_ptr;
 836        unsigned int tx_ev_q_label;
 837        struct efx_tx_queue *tx_queue;
 838        struct efx_nic *efx = channel->efx;
 839        int tx_packets = 0;
 840
 841        if (unlikely(ACCESS_ONCE(efx->reset_pending)))
 842                return 0;
 843
 844        if (likely(EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_COMP))) {
 845                /* Transmit completion */
 846                tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_DESC_PTR);
 847                tx_ev_q_label = EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_Q_LABEL);
 848                tx_queue = efx_channel_get_tx_queue(
 849                        channel, tx_ev_q_label % EFX_TXQ_TYPES);
 850                tx_packets = ((tx_ev_desc_ptr - tx_queue->read_count) &
 851                              tx_queue->ptr_mask);
 852                efx_xmit_done(tx_queue, tx_ev_desc_ptr);
 853        } else if (EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_WQ_FF_FULL)) {
 854                /* Rewrite the FIFO write pointer */
 855                tx_ev_q_label = EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_Q_LABEL);
 856                tx_queue = efx_channel_get_tx_queue(
 857                        channel, tx_ev_q_label % EFX_TXQ_TYPES);
 858
 859                netif_tx_lock(efx->net_dev);
 860                efx_farch_notify_tx_desc(tx_queue);
 861                netif_tx_unlock(efx->net_dev);
 862        } else if (EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_PKT_ERR)) {
 863                efx_schedule_reset(efx, RESET_TYPE_DMA_ERROR);
 864        } else {
 865                netif_err(efx, tx_err, efx->net_dev,
 866                          "channel %d unexpected TX event "
 867                          EFX_QWORD_FMT"\n", channel->channel,
 868                          EFX_QWORD_VAL(*event));
 869        }
 870
 871        return tx_packets;
 872}
 873
 874/* Detect errors included in the rx_evt_pkt_ok bit. */
 875static u16 efx_farch_handle_rx_not_ok(struct efx_rx_queue *rx_queue,
 876                                      const efx_qword_t *event)
 877{
 878        struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
 879        struct efx_nic *efx = rx_queue->efx;
 880        bool rx_ev_buf_owner_id_err, rx_ev_ip_hdr_chksum_err;
 881        bool rx_ev_tcp_udp_chksum_err, rx_ev_eth_crc_err;
 882        bool rx_ev_frm_trunc, rx_ev_drib_nib, rx_ev_tobe_disc;
 883        bool rx_ev_other_err, rx_ev_pause_frm;
 884        bool rx_ev_hdr_type, rx_ev_mcast_pkt;
 885        unsigned rx_ev_pkt_type;
 886
 887        rx_ev_hdr_type = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_HDR_TYPE);
 888        rx_ev_mcast_pkt = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_MCAST_PKT);
 889        rx_ev_tobe_disc = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_TOBE_DISC);
 890        rx_ev_pkt_type = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_PKT_TYPE);
 891        rx_ev_buf_owner_id_err = EFX_QWORD_FIELD(*event,
 892                                                 FSF_AZ_RX_EV_BUF_OWNER_ID_ERR);
 893        rx_ev_ip_hdr_chksum_err = EFX_QWORD_FIELD(*event,
 894                                                  FSF_AZ_RX_EV_IP_HDR_CHKSUM_ERR);
 895        rx_ev_tcp_udp_chksum_err = EFX_QWORD_FIELD(*event,
 896                                                   FSF_AZ_RX_EV_TCP_UDP_CHKSUM_ERR);
 897        rx_ev_eth_crc_err = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_ETH_CRC_ERR);
 898        rx_ev_frm_trunc = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_FRM_TRUNC);
 899        rx_ev_drib_nib = ((efx_nic_rev(efx) >= EFX_REV_FALCON_B0) ?
 900                          0 : EFX_QWORD_FIELD(*event, FSF_AA_RX_EV_DRIB_NIB));
 901        rx_ev_pause_frm = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_PAUSE_FRM_ERR);
 902
 903        /* Every error apart from tobe_disc and pause_frm */
 904        rx_ev_other_err = (rx_ev_drib_nib | rx_ev_tcp_udp_chksum_err |
 905                           rx_ev_buf_owner_id_err | rx_ev_eth_crc_err |
 906                           rx_ev_frm_trunc | rx_ev_ip_hdr_chksum_err);
 907
 908        /* Count errors that are not in MAC stats.  Ignore expected
 909         * checksum errors during self-test. */
 910        if (rx_ev_frm_trunc)
 911                ++channel->n_rx_frm_trunc;
 912        else if (rx_ev_tobe_disc)
 913                ++channel->n_rx_tobe_disc;
 914        else if (!efx->loopback_selftest) {
 915                if (rx_ev_ip_hdr_chksum_err)
 916                        ++channel->n_rx_ip_hdr_chksum_err;
 917                else if (rx_ev_tcp_udp_chksum_err)
 918                        ++channel->n_rx_tcp_udp_chksum_err;
 919        }
 920
 921        /* TOBE_DISC is expected on unicast mismatches; don't print out an
 922         * error message.  FRM_TRUNC indicates RXDP dropped the packet due
 923         * to a FIFO overflow.
 924         */
 925#ifdef DEBUG
 926        if (rx_ev_other_err && net_ratelimit()) {
 927                netif_dbg(efx, rx_err, efx->net_dev,
 928                          " RX queue %d unexpected RX event "
 929                          EFX_QWORD_FMT "%s%s%s%s%s%s%s%s\n",
 930                          efx_rx_queue_index(rx_queue), EFX_QWORD_VAL(*event),
 931                          rx_ev_buf_owner_id_err ? " [OWNER_ID_ERR]" : "",
 932                          rx_ev_ip_hdr_chksum_err ?
 933                          " [IP_HDR_CHKSUM_ERR]" : "",
 934                          rx_ev_tcp_udp_chksum_err ?
 935                          " [TCP_UDP_CHKSUM_ERR]" : "",
 936                          rx_ev_eth_crc_err ? " [ETH_CRC_ERR]" : "",
 937                          rx_ev_frm_trunc ? " [FRM_TRUNC]" : "",
 938                          rx_ev_drib_nib ? " [DRIB_NIB]" : "",
 939                          rx_ev_tobe_disc ? " [TOBE_DISC]" : "",
 940                          rx_ev_pause_frm ? " [PAUSE]" : "");
 941        }
 942#endif
 943
 944        /* The frame must be discarded if any of these are true. */
 945        return (rx_ev_eth_crc_err | rx_ev_frm_trunc | rx_ev_drib_nib |
 946                rx_ev_tobe_disc | rx_ev_pause_frm) ?
 947                EFX_RX_PKT_DISCARD : 0;
 948}
 949
 950/* Handle receive events that are not in-order. Return true if this
 951 * can be handled as a partial packet discard, false if it's more
 952 * serious.
 953 */
 954static bool
 955efx_farch_handle_rx_bad_index(struct efx_rx_queue *rx_queue, unsigned index)
 956{
 957        struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
 958        struct efx_nic *efx = rx_queue->efx;
 959        unsigned expected, dropped;
 960
 961        if (rx_queue->scatter_n &&
 962            index == ((rx_queue->removed_count + rx_queue->scatter_n - 1) &
 963                      rx_queue->ptr_mask)) {
 964                ++channel->n_rx_nodesc_trunc;
 965                return true;
 966        }
 967
 968        expected = rx_queue->removed_count & rx_queue->ptr_mask;
 969        dropped = (index - expected) & rx_queue->ptr_mask;
 970        netif_info(efx, rx_err, efx->net_dev,
 971                   "dropped %d events (index=%d expected=%d)\n",
 972                   dropped, index, expected);
 973
 974        efx_schedule_reset(efx, EFX_WORKAROUND_5676(efx) ?
 975                           RESET_TYPE_RX_RECOVERY : RESET_TYPE_DISABLE);
 976        return false;
 977}
 978
 979/* Handle a packet received event
 980 *
 981 * The NIC gives a "discard" flag if it's a unicast packet with the
 982 * wrong destination address
 983 * Also "is multicast" and "matches multicast filter" flags can be used to
 984 * discard non-matching multicast packets.
 985 */
 986static void
 987efx_farch_handle_rx_event(struct efx_channel *channel, const efx_qword_t *event)
 988{
 989        unsigned int rx_ev_desc_ptr, rx_ev_byte_cnt;
 990        unsigned int rx_ev_hdr_type, rx_ev_mcast_pkt;
 991        unsigned expected_ptr;
 992        bool rx_ev_pkt_ok, rx_ev_sop, rx_ev_cont;
 993        u16 flags;
 994        struct efx_rx_queue *rx_queue;
 995        struct efx_nic *efx = channel->efx;
 996
 997        if (unlikely(ACCESS_ONCE(efx->reset_pending)))
 998                return;
 999
1000        rx_ev_cont = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_JUMBO_CONT);
1001        rx_ev_sop = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_SOP);
1002        WARN_ON(EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_Q_LABEL) !=
1003                channel->channel);
1004
1005        rx_queue = efx_channel_get_rx_queue(channel);
1006
1007        rx_ev_desc_ptr = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_DESC_PTR);
1008        expected_ptr = ((rx_queue->removed_count + rx_queue->scatter_n) &
1009                        rx_queue->ptr_mask);
1010
1011        /* Check for partial drops and other errors */
1012        if (unlikely(rx_ev_desc_ptr != expected_ptr) ||
1013            unlikely(rx_ev_sop != (rx_queue->scatter_n == 0))) {
1014                if (rx_ev_desc_ptr != expected_ptr &&
1015                    !efx_farch_handle_rx_bad_index(rx_queue, rx_ev_desc_ptr))
1016                        return;
1017
1018                /* Discard all pending fragments */
1019                if (rx_queue->scatter_n) {
1020                        efx_rx_packet(
1021                                rx_queue,
1022                                rx_queue->removed_count & rx_queue->ptr_mask,
1023                                rx_queue->scatter_n, 0, EFX_RX_PKT_DISCARD);
1024                        rx_queue->removed_count += rx_queue->scatter_n;
1025                        rx_queue->scatter_n = 0;
1026                }
1027
1028                /* Return if there is no new fragment */
1029                if (rx_ev_desc_ptr != expected_ptr)
1030                        return;
1031
1032                /* Discard new fragment if not SOP */
1033                if (!rx_ev_sop) {
1034                        efx_rx_packet(
1035                                rx_queue,
1036                                rx_queue->removed_count & rx_queue->ptr_mask,
1037                                1, 0, EFX_RX_PKT_DISCARD);
1038                        ++rx_queue->removed_count;
1039                        return;
1040                }
1041        }
1042
1043        ++rx_queue->scatter_n;
1044        if (rx_ev_cont)
1045                return;
1046
1047        rx_ev_byte_cnt = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_BYTE_CNT);
1048        rx_ev_pkt_ok = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_PKT_OK);
1049        rx_ev_hdr_type = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_HDR_TYPE);
1050
1051        if (likely(rx_ev_pkt_ok)) {
1052                /* If packet is marked as OK then we can rely on the
1053                 * hardware checksum and classification.
1054                 */
1055                flags = 0;
1056                switch (rx_ev_hdr_type) {
1057                case FSE_CZ_RX_EV_HDR_TYPE_IPV4V6_TCP:
1058                        flags |= EFX_RX_PKT_TCP;
1059                        /* fall through */
1060                case FSE_CZ_RX_EV_HDR_TYPE_IPV4V6_UDP:
1061                        flags |= EFX_RX_PKT_CSUMMED;
1062                        /* fall through */
1063                case FSE_CZ_RX_EV_HDR_TYPE_IPV4V6_OTHER:
1064                case FSE_AZ_RX_EV_HDR_TYPE_OTHER:
1065                        break;
1066                }
1067        } else {
1068                flags = efx_farch_handle_rx_not_ok(rx_queue, event);
1069        }
1070
1071        /* Detect multicast packets that didn't match the filter */
1072        rx_ev_mcast_pkt = EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_MCAST_PKT);
1073        if (rx_ev_mcast_pkt) {
1074                unsigned int rx_ev_mcast_hash_match =
1075                        EFX_QWORD_FIELD(*event, FSF_AZ_RX_EV_MCAST_HASH_MATCH);
1076
1077                if (unlikely(!rx_ev_mcast_hash_match)) {
1078                        ++channel->n_rx_mcast_mismatch;
1079                        flags |= EFX_RX_PKT_DISCARD;
1080                }
1081        }
1082
1083        channel->irq_mod_score += 2;
1084
1085        /* Handle received packet */
1086        efx_rx_packet(rx_queue,
1087                      rx_queue->removed_count & rx_queue->ptr_mask,
1088                      rx_queue->scatter_n, rx_ev_byte_cnt, flags);
1089        rx_queue->removed_count += rx_queue->scatter_n;
1090        rx_queue->scatter_n = 0;
1091}
1092
1093/* If this flush done event corresponds to a &struct efx_tx_queue, then
1094 * send an %EFX_CHANNEL_MAGIC_TX_DRAIN event to drain the event queue
1095 * of all transmit completions.
1096 */
1097static void
1098efx_farch_handle_tx_flush_done(struct efx_nic *efx, efx_qword_t *event)
1099{
1100        struct efx_tx_queue *tx_queue;
1101        int qid;
1102
1103        qid = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_SUBDATA);
1104        if (qid < EFX_TXQ_TYPES * efx->n_tx_channels) {
1105                tx_queue = efx_get_tx_queue(efx, qid / EFX_TXQ_TYPES,
1106                                            qid % EFX_TXQ_TYPES);
1107                if (atomic_cmpxchg(&tx_queue->flush_outstanding, 1, 0)) {
1108                        efx_farch_magic_event(tx_queue->channel,
1109                                              EFX_CHANNEL_MAGIC_TX_DRAIN(tx_queue));
1110                }
1111        }
1112}
1113
1114/* If this flush done event corresponds to a &struct efx_rx_queue: If the flush
1115 * was successful then send an %EFX_CHANNEL_MAGIC_RX_DRAIN, otherwise add
1116 * the RX queue back to the mask of RX queues in need of flushing.
1117 */
1118static void
1119efx_farch_handle_rx_flush_done(struct efx_nic *efx, efx_qword_t *event)
1120{
1121        struct efx_channel *channel;
1122        struct efx_rx_queue *rx_queue;
1123        int qid;
1124        bool failed;
1125
1126        qid = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_RX_DESCQ_ID);
1127        failed = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_RX_FLUSH_FAIL);
1128        if (qid >= efx->n_channels)
1129                return;
1130        channel = efx_get_channel(efx, qid);
1131        if (!efx_channel_has_rx_queue(channel))
1132                return;
1133        rx_queue = efx_channel_get_rx_queue(channel);
1134
1135        if (failed) {
1136                netif_info(efx, hw, efx->net_dev,
1137                           "RXQ %d flush retry\n", qid);
1138                rx_queue->flush_pending = true;
1139                atomic_inc(&efx->rxq_flush_pending);
1140        } else {
1141                efx_farch_magic_event(efx_rx_queue_channel(rx_queue),
1142                                      EFX_CHANNEL_MAGIC_RX_DRAIN(rx_queue));
1143        }
1144        atomic_dec(&efx->rxq_flush_outstanding);
1145        if (efx_farch_flush_wake(efx))
1146                wake_up(&efx->flush_wq);
1147}
1148
1149static void
1150efx_farch_handle_drain_event(struct efx_channel *channel)
1151{
1152        struct efx_nic *efx = channel->efx;
1153
1154        WARN_ON(atomic_read(&efx->active_queues) == 0);
1155        atomic_dec(&efx->active_queues);
1156        if (efx_farch_flush_wake(efx))
1157                wake_up(&efx->flush_wq);
1158}
1159
1160static void efx_farch_handle_generated_event(struct efx_channel *channel,
1161                                             efx_qword_t *event)
1162{
1163        struct efx_nic *efx = channel->efx;
1164        struct efx_rx_queue *rx_queue =
1165                efx_channel_has_rx_queue(channel) ?
1166                efx_channel_get_rx_queue(channel) : NULL;
1167        unsigned magic, code;
1168
1169        magic = EFX_QWORD_FIELD(*event, FSF_AZ_DRV_GEN_EV_MAGIC);
1170        code = _EFX_CHANNEL_MAGIC_CODE(magic);
1171
1172        if (magic == EFX_CHANNEL_MAGIC_TEST(channel)) {
1173                channel->event_test_cpu = raw_smp_processor_id();
1174        } else if (rx_queue && magic == EFX_CHANNEL_MAGIC_FILL(rx_queue)) {
1175                /* The queue must be empty, so we won't receive any rx
1176                 * events, so efx_process_channel() won't refill the
1177                 * queue. Refill it here */
1178                efx_fast_push_rx_descriptors(rx_queue, true);
1179        } else if (rx_queue && magic == EFX_CHANNEL_MAGIC_RX_DRAIN(rx_queue)) {
1180                efx_farch_handle_drain_event(channel);
1181        } else if (code == _EFX_CHANNEL_MAGIC_TX_DRAIN) {
1182                efx_farch_handle_drain_event(channel);
1183        } else {
1184                netif_dbg(efx, hw, efx->net_dev, "channel %d received "
1185                          "generated event "EFX_QWORD_FMT"\n",
1186                          channel->channel, EFX_QWORD_VAL(*event));
1187        }
1188}
1189
1190static void
1191efx_farch_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
1192{
1193        struct efx_nic *efx = channel->efx;
1194        unsigned int ev_sub_code;
1195        unsigned int ev_sub_data;
1196
1197        ev_sub_code = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_SUBCODE);
1198        ev_sub_data = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_SUBDATA);
1199
1200        switch (ev_sub_code) {
1201        case FSE_AZ_TX_DESCQ_FLS_DONE_EV:
1202                netif_vdbg(efx, hw, efx->net_dev, "channel %d TXQ %d flushed\n",
1203                           channel->channel, ev_sub_data);
1204                efx_farch_handle_tx_flush_done(efx, event);
1205#ifdef CONFIG_SFC_SRIOV
1206                efx_siena_sriov_tx_flush_done(efx, event);
1207#endif
1208                break;
1209        case FSE_AZ_RX_DESCQ_FLS_DONE_EV:
1210                netif_vdbg(efx, hw, efx->net_dev, "channel %d RXQ %d flushed\n",
1211                           channel->channel, ev_sub_data);
1212                efx_farch_handle_rx_flush_done(efx, event);
1213#ifdef CONFIG_SFC_SRIOV
1214                efx_siena_sriov_rx_flush_done(efx, event);
1215#endif
1216                break;
1217        case FSE_AZ_EVQ_INIT_DONE_EV:
1218                netif_dbg(efx, hw, efx->net_dev,
1219                          "channel %d EVQ %d initialised\n",
1220                          channel->channel, ev_sub_data);
1221                break;
1222        case FSE_AZ_SRM_UPD_DONE_EV:
1223                netif_vdbg(efx, hw, efx->net_dev,
1224                           "channel %d SRAM update done\n", channel->channel);
1225                break;
1226        case FSE_AZ_WAKE_UP_EV:
1227                netif_vdbg(efx, hw, efx->net_dev,
1228                           "channel %d RXQ %d wakeup event\n",
1229                           channel->channel, ev_sub_data);
1230                break;
1231        case FSE_AZ_TIMER_EV:
1232                netif_vdbg(efx, hw, efx->net_dev,
1233                           "channel %d RX queue %d timer expired\n",
1234                           channel->channel, ev_sub_data);
1235                break;
1236        case FSE_AA_RX_RECOVER_EV:
1237                netif_err(efx, rx_err, efx->net_dev,
1238                          "channel %d seen DRIVER RX_RESET event. "
1239                        "Resetting.\n", channel->channel);
1240                atomic_inc(&efx->rx_reset);
1241                efx_schedule_reset(efx,
1242                                   EFX_WORKAROUND_6555(efx) ?
1243                                   RESET_TYPE_RX_RECOVERY :
1244                                   RESET_TYPE_DISABLE);
1245                break;
1246        case FSE_BZ_RX_DSC_ERROR_EV:
1247                if (ev_sub_data < EFX_VI_BASE) {
1248                        netif_err(efx, rx_err, efx->net_dev,
1249                                  "RX DMA Q %d reports descriptor fetch error."
1250                                  " RX Q %d is disabled.\n", ev_sub_data,
1251                                  ev_sub_data);
1252                        efx_schedule_reset(efx, RESET_TYPE_DMA_ERROR);
1253                }
1254#ifdef CONFIG_SFC_SRIOV
1255                else
1256                        efx_siena_sriov_desc_fetch_err(efx, ev_sub_data);
1257#endif
1258                break;
1259        case FSE_BZ_TX_DSC_ERROR_EV:
1260                if (ev_sub_data < EFX_VI_BASE) {
1261                        netif_err(efx, tx_err, efx->net_dev,
1262                                  "TX DMA Q %d reports descriptor fetch error."
1263                                  " TX Q %d is disabled.\n", ev_sub_data,
1264                                  ev_sub_data);
1265                        efx_schedule_reset(efx, RESET_TYPE_DMA_ERROR);
1266                }
1267#ifdef CONFIG_SFC_SRIOV
1268                else
1269                        efx_siena_sriov_desc_fetch_err(efx, ev_sub_data);
1270#endif
1271                break;
1272        default:
1273                netif_vdbg(efx, hw, efx->net_dev,
1274                           "channel %d unknown driver event code %d "
1275                           "data %04x\n", channel->channel, ev_sub_code,
1276                           ev_sub_data);
1277                break;
1278        }
1279}
1280
1281int efx_farch_ev_process(struct efx_channel *channel, int budget)
1282{
1283        struct efx_nic *efx = channel->efx;
1284        unsigned int read_ptr;
1285        efx_qword_t event, *p_event;
1286        int ev_code;
1287        int tx_packets = 0;
1288        int spent = 0;
1289
1290        if (budget <= 0)
1291                return spent;
1292
1293        read_ptr = channel->eventq_read_ptr;
1294
1295        for (;;) {
1296                p_event = efx_event(channel, read_ptr);
1297                event = *p_event;
1298
1299                if (!efx_event_present(&event))
1300                        /* End of events */
1301                        break;
1302
1303                netif_vdbg(channel->efx, intr, channel->efx->net_dev,
1304                           "channel %d event is "EFX_QWORD_FMT"\n",
1305                           channel->channel, EFX_QWORD_VAL(event));
1306
1307                /* Clear this event by marking it all ones */
1308                EFX_SET_QWORD(*p_event);
1309
1310                ++read_ptr;
1311
1312                ev_code = EFX_QWORD_FIELD(event, FSF_AZ_EV_CODE);
1313
1314                switch (ev_code) {
1315                case FSE_AZ_EV_CODE_RX_EV:
1316                        efx_farch_handle_rx_event(channel, &event);
1317                        if (++spent == budget)
1318                                goto out;
1319                        break;
1320                case FSE_AZ_EV_CODE_TX_EV:
1321                        tx_packets += efx_farch_handle_tx_event(channel,
1322                                                                &event);
1323                        if (tx_packets > efx->txq_entries) {
1324                                spent = budget;
1325                                goto out;
1326                        }
1327                        break;
1328                case FSE_AZ_EV_CODE_DRV_GEN_EV:
1329                        efx_farch_handle_generated_event(channel, &event);
1330                        break;
1331                case FSE_AZ_EV_CODE_DRIVER_EV:
1332                        efx_farch_handle_driver_event(channel, &event);
1333                        break;
1334#ifdef CONFIG_SFC_SRIOV
1335                case FSE_CZ_EV_CODE_USER_EV:
1336                        efx_siena_sriov_event(channel, &event);
1337                        break;
1338#endif
1339                case FSE_CZ_EV_CODE_MCDI_EV:
1340                        efx_mcdi_process_event(channel, &event);
1341                        break;
1342                case FSE_AZ_EV_CODE_GLOBAL_EV:
1343                        if (efx->type->handle_global_event &&
1344                            efx->type->handle_global_event(channel, &event))
1345                                break;
1346                        /* else fall through */
1347                default:
1348                        netif_err(channel->efx, hw, channel->efx->net_dev,
1349                                  "channel %d unknown event type %d (data "
1350                                  EFX_QWORD_FMT ")\n", channel->channel,
1351                                  ev_code, EFX_QWORD_VAL(event));
1352                }
1353        }
1354
1355out:
1356        channel->eventq_read_ptr = read_ptr;
1357        return spent;
1358}
1359
1360/* Allocate buffer table entries for event queue */
1361int efx_farch_ev_probe(struct efx_channel *channel)
1362{
1363        struct efx_nic *efx = channel->efx;
1364        unsigned entries;
1365
1366        entries = channel->eventq_mask + 1;
1367        return efx_alloc_special_buffer(efx, &channel->eventq,
1368                                        entries * sizeof(efx_qword_t));
1369}
1370
1371int efx_farch_ev_init(struct efx_channel *channel)
1372{
1373        efx_oword_t reg;
1374        struct efx_nic *efx = channel->efx;
1375
1376        netif_dbg(efx, hw, efx->net_dev,
1377                  "channel %d event queue in special buffers %d-%d\n",
1378                  channel->channel, channel->eventq.index,
1379                  channel->eventq.index + channel->eventq.entries - 1);
1380
1381        if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
1382                EFX_POPULATE_OWORD_3(reg,
1383                                     FRF_CZ_TIMER_Q_EN, 1,
1384                                     FRF_CZ_HOST_NOTIFY_MODE, 0,
1385                                     FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);
1386                efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL, channel->channel);
1387        }
1388
1389        /* Pin event queue buffer */
1390        efx_init_special_buffer(efx, &channel->eventq);
1391
1392        /* Fill event queue with all ones (i.e. empty events) */
1393        memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
1394
1395        /* Push event queue to card */
1396        EFX_POPULATE_OWORD_3(reg,
1397                             FRF_AZ_EVQ_EN, 1,
1398                             FRF_AZ_EVQ_SIZE, __ffs(channel->eventq.entries),
1399                             FRF_AZ_EVQ_BUF_BASE_ID, channel->eventq.index);
1400        efx_writeo_table(efx, &reg, efx->type->evq_ptr_tbl_base,
1401                         channel->channel);
1402
1403        return 0;
1404}
1405
1406void efx_farch_ev_fini(struct efx_channel *channel)
1407{
1408        efx_oword_t reg;
1409        struct efx_nic *efx = channel->efx;
1410
1411        /* Remove event queue from card */
1412        EFX_ZERO_OWORD(reg);
1413        efx_writeo_table(efx, &reg, efx->type->evq_ptr_tbl_base,
1414                         channel->channel);
1415        if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0)
1416                efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL, channel->channel);
1417
1418        /* Unpin event queue */
1419        efx_fini_special_buffer(efx, &channel->eventq);
1420}
1421
1422/* Free buffers backing event queue */
1423void efx_farch_ev_remove(struct efx_channel *channel)
1424{
1425        efx_free_special_buffer(channel->efx, &channel->eventq);
1426}
1427
1428
1429void efx_farch_ev_test_generate(struct efx_channel *channel)
1430{
1431        efx_farch_magic_event(channel, EFX_CHANNEL_MAGIC_TEST(channel));
1432}
1433
1434void efx_farch_rx_defer_refill(struct efx_rx_queue *rx_queue)
1435{
1436        efx_farch_magic_event(efx_rx_queue_channel(rx_queue),
1437                              EFX_CHANNEL_MAGIC_FILL(rx_queue));
1438}
1439
1440/**************************************************************************
1441 *
1442 * Hardware interrupts
1443 * The hardware interrupt handler does very little work; all the event
1444 * queue processing is carried out by per-channel tasklets.
1445 *
1446 **************************************************************************/
1447
1448/* Enable/disable/generate interrupts */
1449static inline void efx_farch_interrupts(struct efx_nic *efx,
1450                                      bool enabled, bool force)
1451{
1452        efx_oword_t int_en_reg_ker;
1453
1454        EFX_POPULATE_OWORD_3(int_en_reg_ker,
1455                             FRF_AZ_KER_INT_LEVE_SEL, efx->irq_level,
1456                             FRF_AZ_KER_INT_KER, force,
1457                             FRF_AZ_DRV_INT_EN_KER, enabled);
1458        efx_writeo(efx, &int_en_reg_ker, FR_AZ_INT_EN_KER);
1459}
1460
1461void efx_farch_irq_enable_master(struct efx_nic *efx)
1462{
1463        EFX_ZERO_OWORD(*((efx_oword_t *) efx->irq_status.addr));
1464        wmb(); /* Ensure interrupt vector is clear before interrupts enabled */
1465
1466        efx_farch_interrupts(efx, true, false);
1467}
1468
1469void efx_farch_irq_disable_master(struct efx_nic *efx)
1470{
1471        /* Disable interrupts */
1472        efx_farch_interrupts(efx, false, false);
1473}
1474
1475/* Generate a test interrupt
1476 * Interrupt must already have been enabled, otherwise nasty things
1477 * may happen.
1478 */
1479void efx_farch_irq_test_generate(struct efx_nic *efx)
1480{
1481        efx_farch_interrupts(efx, true, true);
1482}
1483
1484/* Process a fatal interrupt
1485 * Disable bus mastering ASAP and schedule a reset
1486 */
1487irqreturn_t efx_farch_fatal_interrupt(struct efx_nic *efx)
1488{
1489        struct falcon_nic_data *nic_data = efx->nic_data;
1490        efx_oword_t *int_ker = efx->irq_status.addr;
1491        efx_oword_t fatal_intr;
1492        int error, mem_perr;
1493
1494        efx_reado(efx, &fatal_intr, FR_AZ_FATAL_INTR_KER);
1495        error = EFX_OWORD_FIELD(fatal_intr, FRF_AZ_FATAL_INTR);
1496
1497        netif_err(efx, hw, efx->net_dev, "SYSTEM ERROR "EFX_OWORD_FMT" status "
1498                  EFX_OWORD_FMT ": %s\n", EFX_OWORD_VAL(*int_ker),
1499                  EFX_OWORD_VAL(fatal_intr),
1500                  error ? "disabling bus mastering" : "no recognised error");
1501
1502        /* If this is a memory parity error dump which blocks are offending */
1503        mem_perr = (EFX_OWORD_FIELD(fatal_intr, FRF_AZ_MEM_PERR_INT_KER) ||
1504                    EFX_OWORD_FIELD(fatal_intr, FRF_AZ_SRM_PERR_INT_KER));
1505        if (mem_perr) {
1506                efx_oword_t reg;
1507                efx_reado(efx, &reg, FR_AZ_MEM_STAT);
1508                netif_err(efx, hw, efx->net_dev,
1509                          "SYSTEM ERROR: memory parity error "EFX_OWORD_FMT"\n",
1510                          EFX_OWORD_VAL(reg));
1511        }
1512
1513        /* Disable both devices */
1514        pci_clear_master(efx->pci_dev);
1515        if (efx_nic_is_dual_func(efx))
1516                pci_clear_master(nic_data->pci_dev2);
1517        efx_farch_irq_disable_master(efx);
1518
1519        /* Count errors and reset or disable the NIC accordingly */
1520        if (efx->int_error_count == 0 ||
1521            time_after(jiffies, efx->int_error_expire)) {
1522                efx->int_error_count = 0;
1523                efx->int_error_expire =
1524                        jiffies + EFX_INT_ERROR_EXPIRE * HZ;
1525        }
1526        if (++efx->int_error_count < EFX_MAX_INT_ERRORS) {
1527                netif_err(efx, hw, efx->net_dev,
1528                          "SYSTEM ERROR - reset scheduled\n");
1529                efx_schedule_reset(efx, RESET_TYPE_INT_ERROR);
1530        } else {
1531                netif_err(efx, hw, efx->net_dev,
1532                          "SYSTEM ERROR - max number of errors seen."
1533                          "NIC will be disabled\n");
1534                efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1535        }
1536
1537        return IRQ_HANDLED;
1538}
1539
1540/* Handle a legacy interrupt
1541 * Acknowledges the interrupt and schedule event queue processing.
1542 */
1543irqreturn_t efx_farch_legacy_interrupt(int irq, void *dev_id)
1544{
1545        struct efx_nic *efx = dev_id;
1546        bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1547        efx_oword_t *int_ker = efx->irq_status.addr;
1548        irqreturn_t result = IRQ_NONE;
1549        struct efx_channel *channel;
1550        efx_dword_t reg;
1551        u32 queues;
1552        int syserr;
1553
1554        /* Read the ISR which also ACKs the interrupts */
1555        efx_readd(efx, &reg, FR_BZ_INT_ISR0);
1556        queues = EFX_EXTRACT_DWORD(reg, 0, 31);
1557
1558        /* Legacy interrupts are disabled too late by the EEH kernel
1559         * code. Disable them earlier.
1560         * If an EEH error occurred, the read will have returned all ones.
1561         */
1562        if (EFX_DWORD_IS_ALL_ONES(reg) && efx_try_recovery(efx) &&
1563            !efx->eeh_disabled_legacy_irq) {
1564                disable_irq_nosync(efx->legacy_irq);
1565                efx->eeh_disabled_legacy_irq = true;
1566        }
1567
1568        /* Handle non-event-queue sources */
1569        if (queues & (1U << efx->irq_level) && soft_enabled) {
1570                syserr = EFX_OWORD_FIELD(*int_ker, FSF_AZ_NET_IVEC_FATAL_INT);
1571                if (unlikely(syserr))
1572                        return efx_farch_fatal_interrupt(efx);
1573                efx->last_irq_cpu = raw_smp_processor_id();
1574        }
1575
1576        if (queues != 0) {
1577                efx->irq_zero_count = 0;
1578
1579                /* Schedule processing of any interrupting queues */
1580                if (likely(soft_enabled)) {
1581                        efx_for_each_channel(channel, efx) {
1582                                if (queues & 1)
1583                                        efx_schedule_channel_irq(channel);
1584                                queues >>= 1;
1585                        }
1586                }
1587                result = IRQ_HANDLED;
1588
1589        } else {
1590                efx_qword_t *event;
1591
1592                /* Legacy ISR read can return zero once (SF bug 15783) */
1593
1594                /* We can't return IRQ_HANDLED more than once on seeing ISR=0
1595                 * because this might be a shared interrupt. */
1596                if (efx->irq_zero_count++ == 0)
1597                        result = IRQ_HANDLED;
1598
1599                /* Ensure we schedule or rearm all event queues */
1600                if (likely(soft_enabled)) {
1601                        efx_for_each_channel(channel, efx) {
1602                                event = efx_event(channel,
1603                                                  channel->eventq_read_ptr);
1604                                if (efx_event_present(event))
1605                                        efx_schedule_channel_irq(channel);
1606                                else
1607                                        efx_farch_ev_read_ack(channel);
1608                        }
1609                }
1610        }
1611
1612        if (result == IRQ_HANDLED)
1613                netif_vdbg(efx, intr, efx->net_dev,
1614                           "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1615                           irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1616
1617        return result;
1618}
1619
1620/* Handle an MSI interrupt
1621 *
1622 * Handle an MSI hardware interrupt.  This routine schedules event
1623 * queue processing.  No interrupt acknowledgement cycle is necessary.
1624 * Also, we never need to check that the interrupt is for us, since
1625 * MSI interrupts cannot be shared.
1626 */
1627irqreturn_t efx_farch_msi_interrupt(int irq, void *dev_id)
1628{
1629        struct efx_msi_context *context = dev_id;
1630        struct efx_nic *efx = context->efx;
1631        efx_oword_t *int_ker = efx->irq_status.addr;
1632        int syserr;
1633
1634        netif_vdbg(efx, intr, efx->net_dev,
1635                   "IRQ %d on CPU %d status " EFX_OWORD_FMT "\n",
1636                   irq, raw_smp_processor_id(), EFX_OWORD_VAL(*int_ker));
1637
1638        if (!likely(ACCESS_ONCE(efx->irq_soft_enabled)))
1639                return IRQ_HANDLED;
1640
1641        /* Handle non-event-queue sources */
1642        if (context->index == efx->irq_level) {
1643                syserr = EFX_OWORD_FIELD(*int_ker, FSF_AZ_NET_IVEC_FATAL_INT);
1644                if (unlikely(syserr))
1645                        return efx_farch_fatal_interrupt(efx);
1646                efx->last_irq_cpu = raw_smp_processor_id();
1647        }
1648
1649        /* Schedule processing of the channel */
1650        efx_schedule_channel_irq(efx->channel[context->index]);
1651
1652        return IRQ_HANDLED;
1653}
1654
1655/* Setup RSS indirection table.
1656 * This maps from the hash value of the packet to RXQ
1657 */
1658void efx_farch_rx_push_indir_table(struct efx_nic *efx)
1659{
1660        size_t i = 0;
1661        efx_dword_t dword;
1662
1663        BUG_ON(efx_nic_rev(efx) < EFX_REV_FALCON_B0);
1664
1665        BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1666                     FR_BZ_RX_INDIRECTION_TBL_ROWS);
1667
1668        for (i = 0; i < FR_BZ_RX_INDIRECTION_TBL_ROWS; i++) {
1669                EFX_POPULATE_DWORD_1(dword, FRF_BZ_IT_QUEUE,
1670                                     efx->rx_indir_table[i]);
1671                efx_writed(efx, &dword,
1672                           FR_BZ_RX_INDIRECTION_TBL +
1673                           FR_BZ_RX_INDIRECTION_TBL_STEP * i);
1674        }
1675}
1676
1677/* Looks at available SRAM resources and works out how many queues we
1678 * can support, and where things like descriptor caches should live.
1679 *
1680 * SRAM is split up as follows:
1681 * 0                          buftbl entries for channels
1682 * efx->vf_buftbl_base        buftbl entries for SR-IOV
1683 * efx->rx_dc_base            RX descriptor caches
1684 * efx->tx_dc_base            TX descriptor caches
1685 */
1686void efx_farch_dimension_resources(struct efx_nic *efx, unsigned sram_lim_qw)
1687{
1688        unsigned vi_count, buftbl_min;
1689
1690#ifdef CONFIG_SFC_SRIOV
1691        struct siena_nic_data *nic_data = efx->nic_data;
1692#endif
1693
1694        /* Account for the buffer table entries backing the datapath channels
1695         * and the descriptor caches for those channels.
1696         */
1697        buftbl_min = ((efx->n_rx_channels * EFX_MAX_DMAQ_SIZE +
1698                       efx->n_tx_channels * EFX_TXQ_TYPES * EFX_MAX_DMAQ_SIZE +
1699                       efx->n_channels * EFX_MAX_EVQ_SIZE)
1700                      * sizeof(efx_qword_t) / EFX_BUF_SIZE);
1701        vi_count = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
1702
1703#ifdef CONFIG_SFC_SRIOV
1704        if (efx->type->sriov_wanted) {
1705                if (efx->type->sriov_wanted(efx)) {
1706                        unsigned vi_dc_entries, buftbl_free;
1707                        unsigned entries_per_vf, vf_limit;
1708
1709                        nic_data->vf_buftbl_base = buftbl_min;
1710
1711                        vi_dc_entries = RX_DC_ENTRIES + TX_DC_ENTRIES;
1712                        vi_count = max(vi_count, EFX_VI_BASE);
1713                        buftbl_free = (sram_lim_qw - buftbl_min -
1714                                       vi_count * vi_dc_entries);
1715
1716                        entries_per_vf = ((vi_dc_entries +
1717                                           EFX_VF_BUFTBL_PER_VI) *
1718                                          efx_vf_size(efx));
1719                        vf_limit = min(buftbl_free / entries_per_vf,
1720                                       (1024U - EFX_VI_BASE) >> efx->vi_scale);
1721
1722                        if (efx->vf_count > vf_limit) {
1723                                netif_err(efx, probe, efx->net_dev,
1724                                          "Reducing VF count from from %d to %d\n",
1725                                          efx->vf_count, vf_limit);
1726                                efx->vf_count = vf_limit;
1727                        }
1728                        vi_count += efx->vf_count * efx_vf_size(efx);
1729                }
1730        }
1731#endif
1732
1733        efx->tx_dc_base = sram_lim_qw - vi_count * TX_DC_ENTRIES;
1734        efx->rx_dc_base = efx->tx_dc_base - vi_count * RX_DC_ENTRIES;
1735}
1736
1737u32 efx_farch_fpga_ver(struct efx_nic *efx)
1738{
1739        efx_oword_t altera_build;
1740        efx_reado(efx, &altera_build, FR_AZ_ALTERA_BUILD);
1741        return EFX_OWORD_FIELD(altera_build, FRF_AZ_ALTERA_BUILD_VER);
1742}
1743
1744void efx_farch_init_common(struct efx_nic *efx)
1745{
1746        efx_oword_t temp;
1747
1748        /* Set positions of descriptor caches in SRAM. */
1749        EFX_POPULATE_OWORD_1(temp, FRF_AZ_SRM_TX_DC_BASE_ADR, efx->tx_dc_base);
1750        efx_writeo(efx, &temp, FR_AZ_SRM_TX_DC_CFG);
1751        EFX_POPULATE_OWORD_1(temp, FRF_AZ_SRM_RX_DC_BASE_ADR, efx->rx_dc_base);
1752        efx_writeo(efx, &temp, FR_AZ_SRM_RX_DC_CFG);
1753
1754        /* Set TX descriptor cache size. */
1755        BUILD_BUG_ON(TX_DC_ENTRIES != (8 << TX_DC_ENTRIES_ORDER));
1756        EFX_POPULATE_OWORD_1(temp, FRF_AZ_TX_DC_SIZE, TX_DC_ENTRIES_ORDER);
1757        efx_writeo(efx, &temp, FR_AZ_TX_DC_CFG);
1758
1759        /* Set RX descriptor cache size.  Set low watermark to size-8, as
1760         * this allows most efficient prefetching.
1761         */
1762        BUILD_BUG_ON(RX_DC_ENTRIES != (8 << RX_DC_ENTRIES_ORDER));
1763        EFX_POPULATE_OWORD_1(temp, FRF_AZ_RX_DC_SIZE, RX_DC_ENTRIES_ORDER);
1764        efx_writeo(efx, &temp, FR_AZ_RX_DC_CFG);
1765        EFX_POPULATE_OWORD_1(temp, FRF_AZ_RX_DC_PF_LWM, RX_DC_ENTRIES - 8);
1766        efx_writeo(efx, &temp, FR_AZ_RX_DC_PF_WM);
1767
1768        /* Program INT_KER address */
1769        EFX_POPULATE_OWORD_2(temp,
1770                             FRF_AZ_NORM_INT_VEC_DIS_KER,
1771                             EFX_INT_MODE_USE_MSI(efx),
1772                             FRF_AZ_INT_ADR_KER, efx->irq_status.dma_addr);
1773        efx_writeo(efx, &temp, FR_AZ_INT_ADR_KER);
1774
1775        if (EFX_WORKAROUND_17213(efx) && !EFX_INT_MODE_USE_MSI(efx))
1776                /* Use an interrupt level unused by event queues */
1777                efx->irq_level = 0x1f;
1778        else
1779                /* Use a valid MSI-X vector */
1780                efx->irq_level = 0;
1781
1782        /* Enable all the genuinely fatal interrupts.  (They are still
1783         * masked by the overall interrupt mask, controlled by
1784         * falcon_interrupts()).
1785         *
1786         * Note: All other fatal interrupts are enabled
1787         */
1788        EFX_POPULATE_OWORD_3(temp,
1789                             FRF_AZ_ILL_ADR_INT_KER_EN, 1,
1790                             FRF_AZ_RBUF_OWN_INT_KER_EN, 1,
1791                             FRF_AZ_TBUF_OWN_INT_KER_EN, 1);
1792        if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0)
1793                EFX_SET_OWORD_FIELD(temp, FRF_CZ_SRAM_PERR_INT_P_KER_EN, 1);
1794        EFX_INVERT_OWORD(temp);
1795        efx_writeo(efx, &temp, FR_AZ_FATAL_INTR_KER);
1796
1797        /* Disable the ugly timer-based TX DMA backoff and allow TX DMA to be
1798         * controlled by the RX FIFO fill level. Set arbitration to one pkt/Q.
1799         */
1800        efx_reado(efx, &temp, FR_AZ_TX_RESERVED);
1801        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_RX_SPACER, 0xfe);
1802        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_RX_SPACER_EN, 1);
1803        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_ONE_PKT_PER_Q, 1);
1804        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_PUSH_EN, 1);
1805        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_DIS_NON_IP_EV, 1);
1806        /* Enable SW_EV to inherit in char driver - assume harmless here */
1807        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_SOFT_EVT_EN, 1);
1808        /* Prefetch threshold 2 => fetch when descriptor cache half empty */
1809        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_PREF_THRESHOLD, 2);
1810        /* Disable hardware watchdog which can misfire */
1811        EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_PREF_WD_TMR, 0x3fffff);
1812        /* Squash TX of packets of 16 bytes or less */
1813        if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0)
1814                EFX_SET_OWORD_FIELD(temp, FRF_BZ_TX_FLUSH_MIN_LEN_EN, 1);
1815        efx_writeo(efx, &temp, FR_AZ_TX_RESERVED);
1816
1817        if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
1818                EFX_POPULATE_OWORD_4(temp,
1819                                     /* Default values */
1820                                     FRF_BZ_TX_PACE_SB_NOT_AF, 0x15,
1821                                     FRF_BZ_TX_PACE_SB_AF, 0xb,
1822                                     FRF_BZ_TX_PACE_FB_BASE, 0,
1823                                     /* Allow large pace values in the
1824                                      * fast bin. */
1825                                     FRF_BZ_TX_PACE_BIN_TH,
1826                                     FFE_BZ_TX_PACE_RESERVED);
1827                efx_writeo(efx, &temp, FR_BZ_TX_PACE);
1828        }
1829}
1830
1831/**************************************************************************
1832 *
1833 * Filter tables
1834 *
1835 **************************************************************************
1836 */
1837
1838/* "Fudge factors" - difference between programmed value and actual depth.
1839 * Due to pipelined implementation we need to program H/W with a value that
1840 * is larger than the hop limit we want.
1841 */
1842#define EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD 3
1843#define EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL 1
1844
1845/* Hard maximum search limit.  Hardware will time-out beyond 200-something.
1846 * We also need to avoid infinite loops in efx_farch_filter_search() when the
1847 * table is full.
1848 */
1849#define EFX_FARCH_FILTER_CTL_SRCH_MAX 200
1850
1851/* Don't try very hard to find space for performance hints, as this is
1852 * counter-productive. */
1853#define EFX_FARCH_FILTER_CTL_SRCH_HINT_MAX 5
1854
1855enum efx_farch_filter_type {
1856        EFX_FARCH_FILTER_TCP_FULL = 0,
1857        EFX_FARCH_FILTER_TCP_WILD,
1858        EFX_FARCH_FILTER_UDP_FULL,
1859        EFX_FARCH_FILTER_UDP_WILD,
1860        EFX_FARCH_FILTER_MAC_FULL = 4,
1861        EFX_FARCH_FILTER_MAC_WILD,
1862        EFX_FARCH_FILTER_UC_DEF = 8,
1863        EFX_FARCH_FILTER_MC_DEF,
1864        EFX_FARCH_FILTER_TYPE_COUNT,            /* number of specific types */
1865};
1866
1867enum efx_farch_filter_table_id {
1868        EFX_FARCH_FILTER_TABLE_RX_IP = 0,
1869        EFX_FARCH_FILTER_TABLE_RX_MAC,
1870        EFX_FARCH_FILTER_TABLE_RX_DEF,
1871        EFX_FARCH_FILTER_TABLE_TX_MAC,
1872        EFX_FARCH_FILTER_TABLE_COUNT,
1873};
1874
1875enum efx_farch_filter_index {
1876        EFX_FARCH_FILTER_INDEX_UC_DEF,
1877        EFX_FARCH_FILTER_INDEX_MC_DEF,
1878        EFX_FARCH_FILTER_SIZE_RX_DEF,
1879};
1880
1881struct efx_farch_filter_spec {
1882        u8      type:4;
1883        u8      priority:4;
1884        u8      flags;
1885        u16     dmaq_id;
1886        u32     data[3];
1887};
1888
1889struct efx_farch_filter_table {
1890        enum efx_farch_filter_table_id id;
1891        u32             offset;         /* address of table relative to BAR */
1892        unsigned        size;           /* number of entries */
1893        unsigned        step;           /* step between entries */
1894        unsigned        used;           /* number currently used */
1895        unsigned long   *used_bitmap;
1896        struct efx_farch_filter_spec *spec;
1897        unsigned        search_limit[EFX_FARCH_FILTER_TYPE_COUNT];
1898};
1899
1900struct efx_farch_filter_state {
1901        struct efx_farch_filter_table table[EFX_FARCH_FILTER_TABLE_COUNT];
1902};
1903
1904static void
1905efx_farch_filter_table_clear_entry(struct efx_nic *efx,
1906                                   struct efx_farch_filter_table *table,
1907                                   unsigned int filter_idx);
1908
1909/* The filter hash function is LFSR polynomial x^16 + x^3 + 1 of a 32-bit
1910 * key derived from the n-tuple.  The initial LFSR state is 0xffff. */
1911static u16 efx_farch_filter_hash(u32 key)
1912{
1913        u16 tmp;
1914
1915        /* First 16 rounds */
1916        tmp = 0x1fff ^ key >> 16;
1917        tmp = tmp ^ tmp >> 3 ^ tmp >> 6;
1918        tmp = tmp ^ tmp >> 9;
1919        /* Last 16 rounds */
1920        tmp = tmp ^ tmp << 13 ^ key;
1921        tmp = tmp ^ tmp >> 3 ^ tmp >> 6;
1922        return tmp ^ tmp >> 9;
1923}
1924
1925/* To allow for hash collisions, filter search continues at these
1926 * increments from the first possible entry selected by the hash. */
1927static u16 efx_farch_filter_increment(u32 key)
1928{
1929        return key * 2 - 1;
1930}
1931
1932static enum efx_farch_filter_table_id
1933efx_farch_filter_spec_table_id(const struct efx_farch_filter_spec *spec)
1934{
1935        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
1936                     (EFX_FARCH_FILTER_TCP_FULL >> 2));
1937        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
1938                     (EFX_FARCH_FILTER_TCP_WILD >> 2));
1939        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
1940                     (EFX_FARCH_FILTER_UDP_FULL >> 2));
1941        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
1942                     (EFX_FARCH_FILTER_UDP_WILD >> 2));
1943        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_MAC !=
1944                     (EFX_FARCH_FILTER_MAC_FULL >> 2));
1945        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_MAC !=
1946                     (EFX_FARCH_FILTER_MAC_WILD >> 2));
1947        BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_TX_MAC !=
1948                     EFX_FARCH_FILTER_TABLE_RX_MAC + 2);
1949        return (spec->type >> 2) + ((spec->flags & EFX_FILTER_FLAG_TX) ? 2 : 0);
1950}
1951
1952static void efx_farch_filter_push_rx_config(struct efx_nic *efx)
1953{
1954        struct efx_farch_filter_state *state = efx->filter_state;
1955        struct efx_farch_filter_table *table;
1956        efx_oword_t filter_ctl;
1957
1958        efx_reado(efx, &filter_ctl, FR_BZ_RX_FILTER_CTL);
1959
1960        table = &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
1961        EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_TCP_FULL_SRCH_LIMIT,
1962                            table->search_limit[EFX_FARCH_FILTER_TCP_FULL] +
1963                            EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
1964        EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_TCP_WILD_SRCH_LIMIT,
1965                            table->search_limit[EFX_FARCH_FILTER_TCP_WILD] +
1966                            EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
1967        EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_UDP_FULL_SRCH_LIMIT,
1968                            table->search_limit[EFX_FARCH_FILTER_UDP_FULL] +
1969                            EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
1970        EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_UDP_WILD_SRCH_LIMIT,
1971                            table->search_limit[EFX_FARCH_FILTER_UDP_WILD] +
1972                            EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
1973
1974        table = &state->table[EFX_FARCH_FILTER_TABLE_RX_MAC];
1975        if (table->size) {
1976                EFX_SET_OWORD_FIELD(
1977                        filter_ctl, FRF_CZ_ETHERNET_FULL_SEARCH_LIMIT,
1978                        table->search_limit[EFX_FARCH_FILTER_MAC_FULL] +
1979                        EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
1980                EFX_SET_OWORD_FIELD(
1981                        filter_ctl, FRF_CZ_ETHERNET_WILDCARD_SEARCH_LIMIT,
1982                        table->search_limit[EFX_FARCH_FILTER_MAC_WILD] +
1983                        EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
1984        }
1985
1986        table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
1987        if (table->size) {
1988                EFX_SET_OWORD_FIELD(
1989                        filter_ctl, FRF_CZ_UNICAST_NOMATCH_Q_ID,
1990                        table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].dmaq_id);
1991                EFX_SET_OWORD_FIELD(
1992                        filter_ctl, FRF_CZ_UNICAST_NOMATCH_RSS_ENABLED,
1993                        !!(table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].flags &
1994                           EFX_FILTER_FLAG_RX_RSS));
1995                EFX_SET_OWORD_FIELD(
1996                        filter_ctl, FRF_CZ_MULTICAST_NOMATCH_Q_ID,
1997                        table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].dmaq_id);
1998                EFX_SET_OWORD_FIELD(
1999                        filter_ctl, FRF_CZ_MULTICAST_NOMATCH_RSS_ENABLED,
2000                        !!(table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].flags &
2001                           EFX_FILTER_FLAG_RX_RSS));
2002
2003                /* There is a single bit to enable RX scatter for all
2004                 * unmatched packets.  Only set it if scatter is
2005                 * enabled in both filter specs.
2006                 */
2007                EFX_SET_OWORD_FIELD(
2008                        filter_ctl, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q,
2009                        !!(table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].flags &
2010                           table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].flags &
2011                           EFX_FILTER_FLAG_RX_SCATTER));
2012        } else if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
2013                /* We don't expose 'default' filters because unmatched
2014                 * packets always go to the queue number found in the
2015                 * RSS table.  But we still need to set the RX scatter
2016                 * bit here.
2017                 */
2018                EFX_SET_OWORD_FIELD(
2019                        filter_ctl, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q,
2020                        efx->rx_scatter);
2021        }
2022
2023        efx_writeo(efx, &filter_ctl, FR_BZ_RX_FILTER_CTL);
2024}
2025
2026static void efx_farch_filter_push_tx_limits(struct efx_nic *efx)
2027{
2028        struct efx_farch_filter_state *state = efx->filter_state;
2029        struct efx_farch_filter_table *table;
2030        efx_oword_t tx_cfg;
2031
2032        efx_reado(efx, &tx_cfg, FR_AZ_TX_CFG);
2033
2034        table = &state->table[EFX_FARCH_FILTER_TABLE_TX_MAC];
2035        if (table->size) {
2036                EFX_SET_OWORD_FIELD(
2037                        tx_cfg, FRF_CZ_TX_ETH_FILTER_FULL_SEARCH_RANGE,
2038                        table->search_limit[EFX_FARCH_FILTER_MAC_FULL] +
2039                        EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
2040                EFX_SET_OWORD_FIELD(
2041                        tx_cfg, FRF_CZ_TX_ETH_FILTER_WILD_SEARCH_RANGE,
2042                        table->search_limit[EFX_FARCH_FILTER_MAC_WILD] +
2043                        EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
2044        }
2045
2046        efx_writeo(efx, &tx_cfg, FR_AZ_TX_CFG);
2047}
2048
2049static int
2050efx_farch_filter_from_gen_spec(struct efx_farch_filter_spec *spec,
2051                               const struct efx_filter_spec *gen_spec)
2052{
2053        bool is_full = false;
2054
2055        if ((gen_spec->flags & EFX_FILTER_FLAG_RX_RSS) &&
2056            gen_spec->rss_context != EFX_FILTER_RSS_CONTEXT_DEFAULT)
2057                return -EINVAL;
2058
2059        spec->priority = gen_spec->priority;
2060        spec->flags = gen_spec->flags;
2061        spec->dmaq_id = gen_spec->dmaq_id;
2062
2063        switch (gen_spec->match_flags) {
2064        case (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO |
2065              EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT |
2066              EFX_FILTER_MATCH_REM_HOST | EFX_FILTER_MATCH_REM_PORT):
2067                is_full = true;
2068                /* fall through */
2069        case (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO |
2070              EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT): {
2071                __be32 rhost, host1, host2;
2072                __be16 rport, port1, port2;
2073
2074                EFX_BUG_ON_PARANOID(!(gen_spec->flags & EFX_FILTER_FLAG_RX));
2075
2076                if (gen_spec->ether_type != htons(ETH_P_IP))
2077                        return -EPROTONOSUPPORT;
2078                if (gen_spec->loc_port == 0 ||
2079                    (is_full && gen_spec->rem_port == 0))
2080                        return -EADDRNOTAVAIL;
2081                switch (gen_spec->ip_proto) {
2082                case IPPROTO_TCP:
2083                        spec->type = (is_full ? EFX_FARCH_FILTER_TCP_FULL :
2084                                      EFX_FARCH_FILTER_TCP_WILD);
2085                        break;
2086                case IPPROTO_UDP:
2087                        spec->type = (is_full ? EFX_FARCH_FILTER_UDP_FULL :
2088                                      EFX_FARCH_FILTER_UDP_WILD);
2089                        break;
2090                default:
2091                        return -EPROTONOSUPPORT;
2092                }
2093
2094                /* Filter is constructed in terms of source and destination,
2095                 * with the odd wrinkle that the ports are swapped in a UDP
2096                 * wildcard filter.  We need to convert from local and remote
2097                 * (= zero for wildcard) addresses.
2098                 */
2099                rhost = is_full ? gen_spec->rem_host[0] : 0;
2100                rport = is_full ? gen_spec->rem_port : 0;
2101                host1 = rhost;
2102                host2 = gen_spec->loc_host[0];
2103                if (!is_full && gen_spec->ip_proto == IPPROTO_UDP) {
2104                        port1 = gen_spec->loc_port;
2105                        port2 = rport;
2106                } else {
2107                        port1 = rport;
2108                        port2 = gen_spec->loc_port;
2109                }
2110                spec->data[0] = ntohl(host1) << 16 | ntohs(port1);
2111                spec->data[1] = ntohs(port2) << 16 | ntohl(host1) >> 16;
2112                spec->data[2] = ntohl(host2);
2113
2114                break;
2115        }
2116
2117        case EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_OUTER_VID:
2118                is_full = true;
2119                /* fall through */
2120        case EFX_FILTER_MATCH_LOC_MAC:
2121                spec->type = (is_full ? EFX_FARCH_FILTER_MAC_FULL :
2122                              EFX_FARCH_FILTER_MAC_WILD);
2123                spec->data[0] = is_full ? ntohs(gen_spec->outer_vid) : 0;
2124                spec->data[1] = (gen_spec->loc_mac[2] << 24 |
2125                                 gen_spec->loc_mac[3] << 16 |
2126                                 gen_spec->loc_mac[4] << 8 |
2127                                 gen_spec->loc_mac[5]);
2128                spec->data[2] = (gen_spec->loc_mac[0] << 8 |
2129                                 gen_spec->loc_mac[1]);
2130                break;
2131
2132        case EFX_FILTER_MATCH_LOC_MAC_IG:
2133                spec->type = (is_multicast_ether_addr(gen_spec->loc_mac) ?
2134                              EFX_FARCH_FILTER_MC_DEF :
2135                              EFX_FARCH_FILTER_UC_DEF);
2136                memset(spec->data, 0, sizeof(spec->data)); /* ensure equality */
2137                break;
2138
2139        default:
2140                return -EPROTONOSUPPORT;
2141        }
2142
2143        return 0;
2144}
2145
2146static void
2147efx_farch_filter_to_gen_spec(struct efx_filter_spec *gen_spec,
2148                             const struct efx_farch_filter_spec *spec)
2149{
2150        bool is_full = false;
2151
2152        /* *gen_spec should be completely initialised, to be consistent
2153         * with efx_filter_init_{rx,tx}() and in case we want to copy
2154         * it back to userland.
2155         */
2156        memset(gen_spec, 0, sizeof(*gen_spec));
2157
2158        gen_spec->priority = spec->priority;
2159        gen_spec->flags = spec->flags;
2160        gen_spec->dmaq_id = spec->dmaq_id;
2161
2162        switch (spec->type) {
2163        case EFX_FARCH_FILTER_TCP_FULL:
2164        case EFX_FARCH_FILTER_UDP_FULL:
2165                is_full = true;
2166                /* fall through */
2167        case EFX_FARCH_FILTER_TCP_WILD:
2168        case EFX_FARCH_FILTER_UDP_WILD: {
2169                __be32 host1, host2;
2170                __be16 port1, port2;
2171
2172                gen_spec->match_flags =
2173                        EFX_FILTER_MATCH_ETHER_TYPE |
2174                        EFX_FILTER_MATCH_IP_PROTO |
2175                        EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT;
2176                if (is_full)
2177                        gen_spec->match_flags |= (EFX_FILTER_MATCH_REM_HOST |
2178                                                  EFX_FILTER_MATCH_REM_PORT);
2179                gen_spec->ether_type = htons(ETH_P_IP);
2180                gen_spec->ip_proto =
2181                        (spec->type == EFX_FARCH_FILTER_TCP_FULL ||
2182                         spec->type == EFX_FARCH_FILTER_TCP_WILD) ?
2183                        IPPROTO_TCP : IPPROTO_UDP;
2184
2185                host1 = htonl(spec->data[0] >> 16 | spec->data[1] << 16);
2186                port1 = htons(spec->data[0]);
2187                host2 = htonl(spec->data[2]);
2188                port2 = htons(spec->data[1] >> 16);
2189                if (spec->flags & EFX_FILTER_FLAG_TX) {
2190                        gen_spec->loc_host[0] = host1;
2191                        gen_spec->rem_host[0] = host2;
2192                } else {
2193                        gen_spec->loc_host[0] = host2;
2194                        gen_spec->rem_host[0] = host1;
2195                }
2196                if (!!(gen_spec->flags & EFX_FILTER_FLAG_TX) ^
2197                    (!is_full && gen_spec->ip_proto == IPPROTO_UDP)) {
2198                        gen_spec->loc_port = port1;
2199                        gen_spec->rem_port = port2;
2200                } else {
2201                        gen_spec->loc_port = port2;
2202                        gen_spec->rem_port = port1;
2203                }
2204
2205                break;
2206        }
2207
2208        case EFX_FARCH_FILTER_MAC_FULL:
2209                is_full = true;
2210                /* fall through */
2211        case EFX_FARCH_FILTER_MAC_WILD:
2212                gen_spec->match_flags = EFX_FILTER_MATCH_LOC_MAC;
2213                if (is_full)
2214                        gen_spec->match_flags |= EFX_FILTER_MATCH_OUTER_VID;
2215                gen_spec->loc_mac[0] = spec->data[2] >> 8;
2216                gen_spec->loc_mac[1] = spec->data[2];
2217                gen_spec->loc_mac[2] = spec->data[1] >> 24;
2218                gen_spec->loc_mac[3] = spec->data[1] >> 16;
2219                gen_spec->loc_mac[4] = spec->data[1] >> 8;
2220                gen_spec->loc_mac[5] = spec->data[1];
2221                gen_spec->outer_vid = htons(spec->data[0]);
2222                break;
2223
2224        case EFX_FARCH_FILTER_UC_DEF:
2225        case EFX_FARCH_FILTER_MC_DEF:
2226                gen_spec->match_flags = EFX_FILTER_MATCH_LOC_MAC_IG;
2227                gen_spec->loc_mac[0] = spec->type == EFX_FARCH_FILTER_MC_DEF;
2228                break;
2229
2230        default:
2231                WARN_ON(1);
2232                break;
2233        }
2234}
2235
2236static void
2237efx_farch_filter_init_rx_auto(struct efx_nic *efx,
2238                              struct efx_farch_filter_spec *spec)
2239{
2240        /* If there's only one channel then disable RSS for non VF
2241         * traffic, thereby allowing VFs to use RSS when the PF can't.
2242         */
2243        spec->priority = EFX_FILTER_PRI_AUTO;
2244        spec->flags = (EFX_FILTER_FLAG_RX |
2245                       (efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0) |
2246                       (efx->rx_scatter ? EFX_FILTER_FLAG_RX_SCATTER : 0));
2247        spec->dmaq_id = 0;
2248}
2249
2250/* Build a filter entry and return its n-tuple key. */
2251static u32 efx_farch_filter_build(efx_oword_t *filter,
2252                                  struct efx_farch_filter_spec *spec)
2253{
2254        u32 data3;
2255
2256        switch (efx_farch_filter_spec_table_id(spec)) {
2257        case EFX_FARCH_FILTER_TABLE_RX_IP: {
2258                bool is_udp = (spec->type == EFX_FARCH_FILTER_UDP_FULL ||
2259                               spec->type == EFX_FARCH_FILTER_UDP_WILD);
2260                EFX_POPULATE_OWORD_7(
2261                        *filter,
2262                        FRF_BZ_RSS_EN,
2263                        !!(spec->flags & EFX_FILTER_FLAG_RX_RSS),
2264                        FRF_BZ_SCATTER_EN,
2265                        !!(spec->flags & EFX_FILTER_FLAG_RX_SCATTER),
2266                        FRF_BZ_TCP_UDP, is_udp,
2267                        FRF_BZ_RXQ_ID, spec->dmaq_id,
2268                        EFX_DWORD_2, spec->data[2],
2269                        EFX_DWORD_1, spec->data[1],
2270                        EFX_DWORD_0, spec->data[0]);
2271                data3 = is_udp;
2272                break;
2273        }
2274
2275        case EFX_FARCH_FILTER_TABLE_RX_MAC: {
2276                bool is_wild = spec->type == EFX_FARCH_FILTER_MAC_WILD;
2277                EFX_POPULATE_OWORD_7(
2278                        *filter,
2279                        FRF_CZ_RMFT_RSS_EN,
2280                        !!(spec->flags & EFX_FILTER_FLAG_RX_RSS),
2281                        FRF_CZ_RMFT_SCATTER_EN,
2282                        !!(spec->flags & EFX_FILTER_FLAG_RX_SCATTER),
2283                        FRF_CZ_RMFT_RXQ_ID, spec->dmaq_id,
2284                        FRF_CZ_RMFT_WILDCARD_MATCH, is_wild,
2285                        FRF_CZ_RMFT_DEST_MAC_HI, spec->data[2],
2286                        FRF_CZ_RMFT_DEST_MAC_LO, spec->data[1],
2287                        FRF_CZ_RMFT_VLAN_ID, spec->data[0]);
2288                data3 = is_wild;
2289                break;
2290        }
2291
2292        case EFX_FARCH_FILTER_TABLE_TX_MAC: {
2293                bool is_wild = spec->type == EFX_FARCH_FILTER_MAC_WILD;
2294                EFX_POPULATE_OWORD_5(*filter,
2295                                     FRF_CZ_TMFT_TXQ_ID, spec->dmaq_id,
2296                                     FRF_CZ_TMFT_WILDCARD_MATCH, is_wild,
2297                                     FRF_CZ_TMFT_SRC_MAC_HI, spec->data[2],
2298                                     FRF_CZ_TMFT_SRC_MAC_LO, spec->data[1],
2299                                     FRF_CZ_TMFT_VLAN_ID, spec->data[0]);
2300                data3 = is_wild | spec->dmaq_id << 1;
2301                break;
2302        }
2303
2304        default:
2305                BUG();
2306        }
2307
2308        return spec->data[0] ^ spec->data[1] ^ spec->data[2] ^ data3;
2309}
2310
2311static bool efx_farch_filter_equal(const struct efx_farch_filter_spec *left,
2312                                   const struct efx_farch_filter_spec *right)
2313{
2314        if (left->type != right->type ||
2315            memcmp(left->data, right->data, sizeof(left->data)))
2316                return false;
2317
2318        if (left->flags & EFX_FILTER_FLAG_TX &&
2319            left->dmaq_id != right->dmaq_id)
2320                return false;
2321
2322        return true;
2323}
2324
2325/*
2326 * Construct/deconstruct external filter IDs.  At least the RX filter
2327 * IDs must be ordered by matching priority, for RX NFC semantics.
2328 *
2329 * Deconstruction needs to be robust against invalid IDs so that
2330 * efx_filter_remove_id_safe() and efx_filter_get_filter_safe() can
2331 * accept user-provided IDs.
2332 */
2333
2334#define EFX_FARCH_FILTER_MATCH_PRI_COUNT        5
2335
2336static const u8 efx_farch_filter_type_match_pri[EFX_FARCH_FILTER_TYPE_COUNT] = {
2337        [EFX_FARCH_FILTER_TCP_FULL]     = 0,
2338        [EFX_FARCH_FILTER_UDP_FULL]     = 0,
2339        [EFX_FARCH_FILTER_TCP_WILD]     = 1,
2340        [EFX_FARCH_FILTER_UDP_WILD]     = 1,
2341        [EFX_FARCH_FILTER_MAC_FULL]     = 2,
2342        [EFX_FARCH_FILTER_MAC_WILD]     = 3,
2343        [EFX_FARCH_FILTER_UC_DEF]       = 4,
2344        [EFX_FARCH_FILTER_MC_DEF]       = 4,
2345};
2346
2347static const enum efx_farch_filter_table_id efx_farch_filter_range_table[] = {
2348        EFX_FARCH_FILTER_TABLE_RX_IP,   /* RX match pri 0 */
2349        EFX_FARCH_FILTER_TABLE_RX_IP,
2350        EFX_FARCH_FILTER_TABLE_RX_MAC,
2351        EFX_FARCH_FILTER_TABLE_RX_MAC,
2352        EFX_FARCH_FILTER_TABLE_RX_DEF,  /* RX match pri 4 */
2353        EFX_FARCH_FILTER_TABLE_TX_MAC,  /* TX match pri 0 */
2354        EFX_FARCH_FILTER_TABLE_TX_MAC,  /* TX match pri 1 */
2355};
2356
2357#define EFX_FARCH_FILTER_INDEX_WIDTH 13
2358#define EFX_FARCH_FILTER_INDEX_MASK ((1 << EFX_FARCH_FILTER_INDEX_WIDTH) - 1)
2359
2360static inline u32
2361efx_farch_filter_make_id(const struct efx_farch_filter_spec *spec,
2362                         unsigned int index)
2363{
2364        unsigned int range;
2365
2366        range = efx_farch_filter_type_match_pri[spec->type];
2367        if (!(spec->flags & EFX_FILTER_FLAG_RX))
2368                range += EFX_FARCH_FILTER_MATCH_PRI_COUNT;
2369
2370        return range << EFX_FARCH_FILTER_INDEX_WIDTH | index;
2371}
2372
2373static inline enum efx_farch_filter_table_id
2374efx_farch_filter_id_table_id(u32 id)
2375{
2376        unsigned int range = id >> EFX_FARCH_FILTER_INDEX_WIDTH;
2377
2378        if (range < ARRAY_SIZE(efx_farch_filter_range_table))
2379                return efx_farch_filter_range_table[range];
2380        else
2381                return EFX_FARCH_FILTER_TABLE_COUNT; /* invalid */
2382}
2383
2384static inline unsigned int efx_farch_filter_id_index(u32 id)
2385{
2386        return id & EFX_FARCH_FILTER_INDEX_MASK;
2387}
2388
2389u32 efx_farch_filter_get_rx_id_limit(struct efx_nic *efx)
2390{
2391        struct efx_farch_filter_state *state = efx->filter_state;
2392        unsigned int range = EFX_FARCH_FILTER_MATCH_PRI_COUNT - 1;
2393        enum efx_farch_filter_table_id table_id;
2394
2395        do {
2396                table_id = efx_farch_filter_range_table[range];
2397                if (state->table[table_id].size != 0)
2398                        return range << EFX_FARCH_FILTER_INDEX_WIDTH |
2399                                state->table[table_id].size;
2400        } while (range--);
2401
2402        return 0;
2403}
2404
2405s32 efx_farch_filter_insert(struct efx_nic *efx,
2406                            struct efx_filter_spec *gen_spec,
2407                            bool replace_equal)
2408{
2409        struct efx_farch_filter_state *state = efx->filter_state;
2410        struct efx_farch_filter_table *table;
2411        struct efx_farch_filter_spec spec;
2412        efx_oword_t filter;
2413        int rep_index, ins_index;
2414        unsigned int depth = 0;
2415        int rc;
2416
2417        rc = efx_farch_filter_from_gen_spec(&spec, gen_spec);
2418        if (rc)
2419                return rc;
2420
2421        table = &state->table[efx_farch_filter_spec_table_id(&spec)];
2422        if (table->size == 0)
2423                return -EINVAL;
2424
2425        netif_vdbg(efx, hw, efx->net_dev,
2426                   "%s: type %d search_limit=%d", __func__, spec.type,
2427                   table->search_limit[spec.type]);
2428
2429        if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
2430                /* One filter spec per type */
2431                BUILD_BUG_ON(EFX_FARCH_FILTER_INDEX_UC_DEF != 0);
2432                BUILD_BUG_ON(EFX_FARCH_FILTER_INDEX_MC_DEF !=
2433                             EFX_FARCH_FILTER_MC_DEF - EFX_FARCH_FILTER_UC_DEF);
2434                rep_index = spec.type - EFX_FARCH_FILTER_UC_DEF;
2435                ins_index = rep_index;
2436
2437                spin_lock_bh(&efx->filter_lock);
2438        } else {
2439                /* Search concurrently for
2440                 * (1) a filter to be replaced (rep_index): any filter
2441                 *     with the same match values, up to the current
2442                 *     search depth for this type, and
2443                 * (2) the insertion point (ins_index): (1) or any
2444                 *     free slot before it or up to the maximum search
2445                 *     depth for this priority
2446                 * We fail if we cannot find (2).
2447                 *
2448                 * We can stop once either
2449                 * (a) we find (1), in which case we have definitely
2450                 *     found (2) as well; or
2451                 * (b) we have searched exhaustively for (1), and have
2452                 *     either found (2) or searched exhaustively for it
2453                 */
2454                u32 key = efx_farch_filter_build(&filter, &spec);
2455                unsigned int hash = efx_farch_filter_hash(key);
2456                unsigned int incr = efx_farch_filter_increment(key);
2457                unsigned int max_rep_depth = table->search_limit[spec.type];
2458                unsigned int max_ins_depth =
2459                        spec.priority <= EFX_FILTER_PRI_HINT ?
2460                        EFX_FARCH_FILTER_CTL_SRCH_HINT_MAX :
2461                        EFX_FARCH_FILTER_CTL_SRCH_MAX;
2462                unsigned int i = hash & (table->size - 1);
2463
2464                ins_index = -1;
2465                depth = 1;
2466
2467                spin_lock_bh(&efx->filter_lock);
2468
2469                for (;;) {
2470                        if (!test_bit(i, table->used_bitmap)) {
2471                                if (ins_index < 0)
2472                                        ins_index = i;
2473                        } else if (efx_farch_filter_equal(&spec,
2474                                                          &table->spec[i])) {
2475                                /* Case (a) */
2476                                if (ins_index < 0)
2477                                        ins_index = i;
2478                                rep_index = i;
2479                                break;
2480                        }
2481
2482                        if (depth >= max_rep_depth &&
2483                            (ins_index >= 0 || depth >= max_ins_depth)) {
2484                                /* Case (b) */
2485                                if (ins_index < 0) {
2486                                        rc = -EBUSY;
2487                                        goto out;
2488                                }
2489                                rep_index = -1;
2490                                break;
2491                        }
2492
2493                        i = (i + incr) & (table->size - 1);
2494                        ++depth;
2495                }
2496        }
2497
2498        /* If we found a filter to be replaced, check whether we
2499         * should do so
2500         */
2501        if (rep_index >= 0) {
2502                struct efx_farch_filter_spec *saved_spec =
2503                        &table->spec[rep_index];
2504
2505                if (spec.priority == saved_spec->priority && !replace_equal) {
2506                        rc = -EEXIST;
2507                        goto out;
2508                }
2509                if (spec.priority < saved_spec->priority) {
2510                        rc = -EPERM;
2511                        goto out;
2512                }
2513                if (saved_spec->priority == EFX_FILTER_PRI_AUTO ||
2514                    saved_spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO)
2515                        spec.flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
2516        }
2517
2518        /* Insert the filter */
2519        if (ins_index != rep_index) {
2520                __set_bit(ins_index, table->used_bitmap);
2521                ++table->used;
2522        }
2523        table->spec[ins_index] = spec;
2524
2525        if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
2526                efx_farch_filter_push_rx_config(efx);
2527        } else {
2528                if (table->search_limit[spec.type] < depth) {
2529                        table->search_limit[spec.type] = depth;
2530                        if (spec.flags & EFX_FILTER_FLAG_TX)
2531                                efx_farch_filter_push_tx_limits(efx);
2532                        else
2533                                efx_farch_filter_push_rx_config(efx);
2534                }
2535
2536                efx_writeo(efx, &filter,
2537                           table->offset + table->step * ins_index);
2538
2539                /* If we were able to replace a filter by inserting
2540                 * at a lower depth, clear the replaced filter
2541                 */
2542                if (ins_index != rep_index && rep_index >= 0)
2543                        efx_farch_filter_table_clear_entry(efx, table,
2544                                                           rep_index);
2545        }
2546
2547        netif_vdbg(efx, hw, efx->net_dev,
2548                   "%s: filter type %d index %d rxq %u set",
2549                   __func__, spec.type, ins_index, spec.dmaq_id);
2550        rc = efx_farch_filter_make_id(&spec, ins_index);
2551
2552out:
2553        spin_unlock_bh(&efx->filter_lock);
2554        return rc;
2555}
2556
2557static void
2558efx_farch_filter_table_clear_entry(struct efx_nic *efx,
2559                                   struct efx_farch_filter_table *table,
2560                                   unsigned int filter_idx)
2561{
2562        static efx_oword_t filter;
2563
2564        EFX_WARN_ON_PARANOID(!test_bit(filter_idx, table->used_bitmap));
2565        BUG_ON(table->offset == 0); /* can't clear MAC default filters */
2566
2567        __clear_bit(filter_idx, table->used_bitmap);
2568        --table->used;
2569        memset(&table->spec[filter_idx], 0, sizeof(table->spec[0]));
2570
2571        efx_writeo(efx, &filter, table->offset + table->step * filter_idx);
2572
2573        /* If this filter required a greater search depth than
2574         * any other, the search limit for its type can now be
2575         * decreased.  However, it is hard to determine that
2576         * unless the table has become completely empty - in
2577         * which case, all its search limits can be set to 0.
2578         */
2579        if (unlikely(table->used == 0)) {
2580                memset(table->search_limit, 0, sizeof(table->search_limit));
2581                if (table->id == EFX_FARCH_FILTER_TABLE_TX_MAC)
2582                        efx_farch_filter_push_tx_limits(efx);
2583                else
2584                        efx_farch_filter_push_rx_config(efx);
2585        }
2586}
2587
2588static int efx_farch_filter_remove(struct efx_nic *efx,
2589                                   struct efx_farch_filter_table *table,
2590                                   unsigned int filter_idx,
2591                                   enum efx_filter_priority priority)
2592{
2593        struct efx_farch_filter_spec *spec = &table->spec[filter_idx];
2594
2595        if (!test_bit(filter_idx, table->used_bitmap) ||
2596            spec->priority != priority)
2597                return -ENOENT;
2598
2599        if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
2600                efx_farch_filter_init_rx_auto(efx, spec);
2601                efx_farch_filter_push_rx_config(efx);
2602        } else {
2603                efx_farch_filter_table_clear_entry(efx, table, filter_idx);
2604        }
2605
2606        return 0;
2607}
2608
2609int efx_farch_filter_remove_safe(struct efx_nic *efx,
2610                                 enum efx_filter_priority priority,
2611                                 u32 filter_id)
2612{
2613        struct efx_farch_filter_state *state = efx->filter_state;
2614        enum efx_farch_filter_table_id table_id;
2615        struct efx_farch_filter_table *table;
2616        unsigned int filter_idx;
2617        struct efx_farch_filter_spec *spec;
2618        int rc;
2619
2620        table_id = efx_farch_filter_id_table_id(filter_id);
2621        if ((unsigned int)table_id >= EFX_FARCH_FILTER_TABLE_COUNT)
2622                return -ENOENT;
2623        table = &state->table[table_id];
2624
2625        filter_idx = efx_farch_filter_id_index(filter_id);
2626        if (filter_idx >= table->size)
2627                return -ENOENT;
2628        spec = &table->spec[filter_idx];
2629
2630        spin_lock_bh(&efx->filter_lock);
2631        rc = efx_farch_filter_remove(efx, table, filter_idx, priority);
2632        spin_unlock_bh(&efx->filter_lock);
2633
2634        return rc;
2635}
2636
2637int efx_farch_filter_get_safe(struct efx_nic *efx,
2638                              enum efx_filter_priority priority,
2639                              u32 filter_id, struct efx_filter_spec *spec_buf)
2640{
2641        struct efx_farch_filter_state *state = efx->filter_state;
2642        enum efx_farch_filter_table_id table_id;
2643        struct efx_farch_filter_table *table;
2644        struct efx_farch_filter_spec *spec;
2645        unsigned int filter_idx;
2646        int rc;
2647
2648        table_id = efx_farch_filter_id_table_id(filter_id);
2649        if ((unsigned int)table_id >= EFX_FARCH_FILTER_TABLE_COUNT)
2650                return -ENOENT;
2651        table = &state->table[table_id];
2652
2653        filter_idx = efx_farch_filter_id_index(filter_id);
2654        if (filter_idx >= table->size)
2655                return -ENOENT;
2656        spec = &table->spec[filter_idx];
2657
2658        spin_lock_bh(&efx->filter_lock);
2659
2660        if (test_bit(filter_idx, table->used_bitmap) &&
2661            spec->priority == priority) {
2662                efx_farch_filter_to_gen_spec(spec_buf, spec);
2663                rc = 0;
2664        } else {
2665                rc = -ENOENT;
2666        }
2667
2668        spin_unlock_bh(&efx->filter_lock);
2669
2670        return rc;
2671}
2672
2673static void
2674efx_farch_filter_table_clear(struct efx_nic *efx,
2675                             enum efx_farch_filter_table_id table_id,
2676                             enum efx_filter_priority priority)
2677{
2678        struct efx_farch_filter_state *state = efx->filter_state;
2679        struct efx_farch_filter_table *table = &state->table[table_id];
2680        unsigned int filter_idx;
2681
2682        spin_lock_bh(&efx->filter_lock);
2683        for (filter_idx = 0; filter_idx < table->size; ++filter_idx) {
2684                if (table->spec[filter_idx].priority != EFX_FILTER_PRI_AUTO)
2685                        efx_farch_filter_remove(efx, table,
2686                                                filter_idx, priority);
2687        }
2688        spin_unlock_bh(&efx->filter_lock);
2689}
2690
2691int efx_farch_filter_clear_rx(struct efx_nic *efx,
2692                               enum efx_filter_priority priority)
2693{
2694        efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_IP,
2695                                     priority);
2696        efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_MAC,
2697                                     priority);
2698        efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_DEF,
2699                                     priority);
2700        return 0;
2701}
2702
2703u32 efx_farch_filter_count_rx_used(struct efx_nic *efx,
2704                                   enum efx_filter_priority priority)
2705{
2706        struct efx_farch_filter_state *state = efx->filter_state;
2707        enum efx_farch_filter_table_id table_id;
2708        struct efx_farch_filter_table *table;
2709        unsigned int filter_idx;
2710        u32 count = 0;
2711
2712        spin_lock_bh(&efx->filter_lock);
2713
2714        for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
2715             table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
2716             table_id++) {
2717                table = &state->table[table_id];
2718                for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
2719                        if (test_bit(filter_idx, table->used_bitmap) &&
2720                            table->spec[filter_idx].priority == priority)
2721                                ++count;
2722                }
2723        }
2724
2725        spin_unlock_bh(&efx->filter_lock);
2726
2727        return count;
2728}
2729
2730s32 efx_farch_filter_get_rx_ids(struct efx_nic *efx,
2731                                enum efx_filter_priority priority,
2732                                u32 *buf, u32 size)
2733{
2734        struct efx_farch_filter_state *state = efx->filter_state;
2735        enum efx_farch_filter_table_id table_id;
2736        struct efx_farch_filter_table *table;
2737        unsigned int filter_idx;
2738        s32 count = 0;
2739
2740        spin_lock_bh(&efx->filter_lock);
2741
2742        for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
2743             table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
2744             table_id++) {
2745                table = &state->table[table_id];
2746                for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
2747                        if (test_bit(filter_idx, table->used_bitmap) &&
2748                            table->spec[filter_idx].priority == priority) {
2749                                if (count == size) {
2750                                        count = -EMSGSIZE;
2751                                        goto out;
2752                                }
2753                                buf[count++] = efx_farch_filter_make_id(
2754                                        &table->spec[filter_idx], filter_idx);
2755                        }
2756                }
2757        }
2758out:
2759        spin_unlock_bh(&efx->filter_lock);
2760
2761        return count;
2762}
2763
2764/* Restore filter stater after reset */
2765void efx_farch_filter_table_restore(struct efx_nic *efx)
2766{
2767        struct efx_farch_filter_state *state = efx->filter_state;
2768        enum efx_farch_filter_table_id table_id;
2769        struct efx_farch_filter_table *table;
2770        efx_oword_t filter;
2771        unsigned int filter_idx;
2772
2773        spin_lock_bh(&efx->filter_lock);
2774
2775        for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
2776                table = &state->table[table_id];
2777
2778                /* Check whether this is a regular register table */
2779                if (table->step == 0)
2780                        continue;
2781
2782                for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
2783                        if (!test_bit(filter_idx, table->used_bitmap))
2784                                continue;
2785                        efx_farch_filter_build(&filter, &table->spec[filter_idx]);
2786                        efx_writeo(efx, &filter,
2787                                   table->offset + table->step * filter_idx);
2788                }
2789        }
2790
2791        efx_farch_filter_push_rx_config(efx);
2792        efx_farch_filter_push_tx_limits(efx);
2793
2794        spin_unlock_bh(&efx->filter_lock);
2795}
2796
2797void efx_farch_filter_table_remove(struct efx_nic *efx)
2798{
2799        struct efx_farch_filter_state *state = efx->filter_state;
2800        enum efx_farch_filter_table_id table_id;
2801
2802        for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
2803                kfree(state->table[table_id].used_bitmap);
2804                vfree(state->table[table_id].spec);
2805        }
2806        kfree(state);
2807}
2808
2809int efx_farch_filter_table_probe(struct efx_nic *efx)
2810{
2811        struct efx_farch_filter_state *state;
2812        struct efx_farch_filter_table *table;
2813        unsigned table_id;
2814
2815        state = kzalloc(sizeof(struct efx_farch_filter_state), GFP_KERNEL);
2816        if (!state)
2817                return -ENOMEM;
2818        efx->filter_state = state;
2819
2820        if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
2821                table = &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
2822                table->id = EFX_FARCH_FILTER_TABLE_RX_IP;
2823                table->offset = FR_BZ_RX_FILTER_TBL0;
2824                table->size = FR_BZ_RX_FILTER_TBL0_ROWS;
2825                table->step = FR_BZ_RX_FILTER_TBL0_STEP;
2826        }
2827
2828        if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
2829                table = &state->table[EFX_FARCH_FILTER_TABLE_RX_MAC];
2830                table->id = EFX_FARCH_FILTER_TABLE_RX_MAC;
2831                table->offset = FR_CZ_RX_MAC_FILTER_TBL0;
2832                table->size = FR_CZ_RX_MAC_FILTER_TBL0_ROWS;
2833                table->step = FR_CZ_RX_MAC_FILTER_TBL0_STEP;
2834
2835                table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
2836                table->id = EFX_FARCH_FILTER_TABLE_RX_DEF;
2837                table->size = EFX_FARCH_FILTER_SIZE_RX_DEF;
2838
2839                table = &state->table[EFX_FARCH_FILTER_TABLE_TX_MAC];
2840                table->id = EFX_FARCH_FILTER_TABLE_TX_MAC;
2841                table->offset = FR_CZ_TX_MAC_FILTER_TBL0;
2842                table->size = FR_CZ_TX_MAC_FILTER_TBL0_ROWS;
2843                table->step = FR_CZ_TX_MAC_FILTER_TBL0_STEP;
2844        }
2845
2846        for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
2847                table = &state->table[table_id];
2848                if (table->size == 0)
2849                        continue;
2850                table->used_bitmap = kcalloc(BITS_TO_LONGS(table->size),
2851                                             sizeof(unsigned long),
2852                                             GFP_KERNEL);
2853                if (!table->used_bitmap)
2854                        goto fail;
2855                table->spec = vzalloc(table->size * sizeof(*table->spec));
2856                if (!table->spec)
2857                        goto fail;
2858        }
2859
2860        table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
2861        if (table->size) {
2862                /* RX default filters must always exist */
2863                struct efx_farch_filter_spec *spec;
2864                unsigned i;
2865
2866                for (i = 0; i < EFX_FARCH_FILTER_SIZE_RX_DEF; i++) {
2867                        spec = &table->spec[i];
2868                        spec->type = EFX_FARCH_FILTER_UC_DEF + i;
2869                        efx_farch_filter_init_rx_auto(efx, spec);
2870                        __set_bit(i, table->used_bitmap);
2871                }
2872        }
2873
2874        efx_farch_filter_push_rx_config(efx);
2875
2876        return 0;
2877
2878fail:
2879        efx_farch_filter_table_remove(efx);
2880        return -ENOMEM;
2881}
2882
2883/* Update scatter enable flags for filters pointing to our own RX queues */
2884void efx_farch_filter_update_rx_scatter(struct efx_nic *efx)
2885{
2886        struct efx_farch_filter_state *state = efx->filter_state;
2887        enum efx_farch_filter_table_id table_id;
2888        struct efx_farch_filter_table *table;
2889        efx_oword_t filter;
2890        unsigned int filter_idx;
2891
2892        spin_lock_bh(&efx->filter_lock);
2893
2894        for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
2895             table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
2896             table_id++) {
2897                table = &state->table[table_id];
2898
2899                for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
2900                        if (!test_bit(filter_idx, table->used_bitmap) ||
2901                            table->spec[filter_idx].dmaq_id >=
2902                            efx->n_rx_channels)
2903                                continue;
2904
2905                        if (efx->rx_scatter)
2906                                table->spec[filter_idx].flags |=
2907                                        EFX_FILTER_FLAG_RX_SCATTER;
2908                        else
2909                                table->spec[filter_idx].flags &=
2910                                        ~EFX_FILTER_FLAG_RX_SCATTER;
2911
2912                        if (table_id == EFX_FARCH_FILTER_TABLE_RX_DEF)
2913                                /* Pushed by efx_farch_filter_push_rx_config() */
2914                                continue;
2915
2916                        efx_farch_filter_build(&filter, &table->spec[filter_idx]);
2917                        efx_writeo(efx, &filter,
2918                                   table->offset + table->step * filter_idx);
2919                }
2920        }
2921
2922        efx_farch_filter_push_rx_config(efx);
2923
2924        spin_unlock_bh(&efx->filter_lock);
2925}
2926
2927#ifdef CONFIG_RFS_ACCEL
2928
2929s32 efx_farch_filter_rfs_insert(struct efx_nic *efx,
2930                                struct efx_filter_spec *gen_spec)
2931{
2932        return efx_farch_filter_insert(efx, gen_spec, true);
2933}
2934
2935bool efx_farch_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
2936                                     unsigned int index)
2937{
2938        struct efx_farch_filter_state *state = efx->filter_state;
2939        struct efx_farch_filter_table *table =
2940                &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
2941
2942        if (test_bit(index, table->used_bitmap) &&
2943            table->spec[index].priority == EFX_FILTER_PRI_HINT &&
2944            rps_may_expire_flow(efx->net_dev, table->spec[index].dmaq_id,
2945                                flow_id, index)) {
2946                efx_farch_filter_table_clear_entry(efx, table, index);
2947                return true;
2948        }
2949
2950        return false;
2951}
2952
2953#endif /* CONFIG_RFS_ACCEL */
2954
2955void efx_farch_filter_sync_rx_mode(struct efx_nic *efx)
2956{
2957        struct net_device *net_dev = efx->net_dev;
2958        struct netdev_hw_addr *ha;
2959        union efx_multicast_hash *mc_hash = &efx->multicast_hash;
2960        u32 crc;
2961        int bit;
2962
2963        if (!efx_dev_registered(efx))
2964                return;
2965
2966        netif_addr_lock_bh(net_dev);
2967
2968        efx->unicast_filter = !(net_dev->flags & IFF_PROMISC);
2969
2970        /* Build multicast hash table */
2971        if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
2972                memset(mc_hash, 0xff, sizeof(*mc_hash));
2973        } else {
2974                memset(mc_hash, 0x00, sizeof(*mc_hash));
2975                netdev_for_each_mc_addr(ha, net_dev) {
2976                        crc = ether_crc_le(ETH_ALEN, ha->addr);
2977                        bit = crc & (EFX_MCAST_HASH_ENTRIES - 1);
2978                        __set_bit_le(bit, mc_hash);
2979                }
2980
2981                /* Broadcast packets go through the multicast hash filter.
2982                 * ether_crc_le() of the broadcast address is 0xbe2612ff
2983                 * so we always add bit 0xff to the mask.
2984                 */
2985                __set_bit_le(0xff, mc_hash);
2986        }
2987
2988        netif_addr_unlock_bh(net_dev);
2989}
2990