Linux Audio

Check our new training course

Loading...
v3.15
 
   1/****************************************************************************
   2 * Driver for Solarflare network controllers and boards
   3 * Copyright 2005-2006 Fen Systems Ltd.
   4 * Copyright 2005-2013 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/* Common definitions for all Efx net driver code */
  12
  13#ifndef EFX_NET_DRIVER_H
  14#define EFX_NET_DRIVER_H
  15
  16#include <linux/netdevice.h>
  17#include <linux/etherdevice.h>
  18#include <linux/ethtool.h>
  19#include <linux/if_vlan.h>
  20#include <linux/timer.h>
  21#include <linux/mdio.h>
  22#include <linux/list.h>
  23#include <linux/pci.h>
  24#include <linux/device.h>
  25#include <linux/highmem.h>
  26#include <linux/workqueue.h>
  27#include <linux/mutex.h>
 
  28#include <linux/vmalloc.h>
  29#include <linux/i2c.h>
  30#include <linux/mtd/mtd.h>
 
  31
  32#include "enum.h"
  33#include "bitfield.h"
  34#include "filter.h"
  35
  36/**************************************************************************
  37 *
  38 * Build definitions
  39 *
  40 **************************************************************************/
  41
  42#define EFX_DRIVER_VERSION	"4.0"
  43
  44#ifdef DEBUG
  45#define EFX_BUG_ON_PARANOID(x) BUG_ON(x)
  46#define EFX_WARN_ON_PARANOID(x) WARN_ON(x)
  47#else
  48#define EFX_BUG_ON_PARANOID(x) do {} while (0)
  49#define EFX_WARN_ON_PARANOID(x) do {} while (0)
  50#endif
  51
  52/**************************************************************************
  53 *
  54 * Efx data structures
  55 *
  56 **************************************************************************/
  57
  58#define EFX_MAX_CHANNELS 32U
  59#define EFX_MAX_RX_QUEUES EFX_MAX_CHANNELS
  60#define EFX_EXTRA_CHANNEL_IOV	0
  61#define EFX_EXTRA_CHANNEL_PTP	1
  62#define EFX_MAX_EXTRA_CHANNELS	2U
  63
  64/* Checksum generation is a per-queue option in hardware, so each
  65 * queue visible to the networking core is backed by two hardware TX
  66 * queues. */
  67#define EFX_MAX_TX_TC		2
  68#define EFX_MAX_CORE_TX_QUEUES	(EFX_MAX_TX_TC * EFX_MAX_CHANNELS)
  69#define EFX_TXQ_TYPE_OFFLOAD	1	/* flag */
  70#define EFX_TXQ_TYPE_HIGHPRI	2	/* flag */
  71#define EFX_TXQ_TYPES		4
  72#define EFX_MAX_TX_QUEUES	(EFX_TXQ_TYPES * EFX_MAX_CHANNELS)
  73
  74/* Maximum possible MTU the driver supports */
  75#define EFX_MAX_MTU (9 * 1024)
  76
 
 
 
  77/* Size of an RX scatter buffer.  Small enough to pack 2 into a 4K page,
  78 * and should be a multiple of the cache line size.
  79 */
  80#define EFX_RX_USR_BUF_SIZE	(2048 - 256)
  81
  82/* If possible, we should ensure cache line alignment at start and end
  83 * of every buffer.  Otherwise, we just need to ensure 4-byte
  84 * alignment of the network header.
  85 */
  86#if NET_IP_ALIGN == 0
  87#define EFX_RX_BUF_ALIGNMENT	L1_CACHE_BYTES
  88#else
  89#define EFX_RX_BUF_ALIGNMENT	4
  90#endif
  91
  92/* Forward declare Precision Time Protocol (PTP) support structure. */
  93struct efx_ptp_data;
  94struct hwtstamp_config;
  95
  96struct efx_self_tests;
  97
  98/**
  99 * struct efx_buffer - A general-purpose DMA buffer
 100 * @addr: host base address of the buffer
 101 * @dma_addr: DMA base address of the buffer
 102 * @len: Buffer length, in bytes
 103 *
 104 * The NIC uses these buffers for its interrupt status registers and
 105 * MAC stats dumps.
 106 */
 107struct efx_buffer {
 108	void *addr;
 109	dma_addr_t dma_addr;
 110	unsigned int len;
 111};
 112
 113/**
 114 * struct efx_special_buffer - DMA buffer entered into buffer table
 115 * @buf: Standard &struct efx_buffer
 116 * @index: Buffer index within controller;s buffer table
 117 * @entries: Number of buffer table entries
 118 *
 119 * The NIC has a buffer table that maps buffers of size %EFX_BUF_SIZE.
 120 * Event and descriptor rings are addressed via one or more buffer
 121 * table entries (and so can be physically non-contiguous, although we
 122 * currently do not take advantage of that).  On Falcon and Siena we
 123 * have to take care of allocating and initialising the entries
 124 * ourselves.  On later hardware this is managed by the firmware and
 125 * @index and @entries are left as 0.
 126 */
 127struct efx_special_buffer {
 128	struct efx_buffer buf;
 129	unsigned int index;
 130	unsigned int entries;
 131};
 132
 133/**
 134 * struct efx_tx_buffer - buffer state for a TX descriptor
 135 * @skb: When @flags & %EFX_TX_BUF_SKB, the associated socket buffer to be
 136 *	freed when descriptor completes
 137 * @heap_buf: When @flags & %EFX_TX_BUF_HEAP, the associated heap buffer to be
 138 *	freed when descriptor completes.
 139 * @option: When @flags & %EFX_TX_BUF_OPTION, a NIC-specific option descriptor.
 140 * @dma_addr: DMA address of the fragment.
 141 * @flags: Flags for allocation and DMA mapping type
 142 * @len: Length of this fragment.
 143 *	This field is zero when the queue slot is empty.
 144 * @unmap_len: Length of this fragment to unmap
 145 * @dma_offset: Offset of @dma_addr from the address of the backing DMA mapping.
 146 * Only valid if @unmap_len != 0.
 147 */
 148struct efx_tx_buffer {
 149	union {
 150		const struct sk_buff *skb;
 151		void *heap_buf;
 152	};
 153	union {
 154		efx_qword_t option;
 155		dma_addr_t dma_addr;
 156	};
 157	unsigned short flags;
 158	unsigned short len;
 159	unsigned short unmap_len;
 160	unsigned short dma_offset;
 161};
 162#define EFX_TX_BUF_CONT		1	/* not last descriptor of packet */
 163#define EFX_TX_BUF_SKB		2	/* buffer is last part of skb */
 164#define EFX_TX_BUF_HEAP		4	/* buffer was allocated with kmalloc() */
 165#define EFX_TX_BUF_MAP_SINGLE	8	/* buffer was mapped with dma_map_single() */
 166#define EFX_TX_BUF_OPTION	0x10	/* empty buffer for option descriptor */
 167
 168/**
 169 * struct efx_tx_queue - An Efx TX queue
 170 *
 171 * This is a ring buffer of TX fragments.
 172 * Since the TX completion path always executes on the same
 173 * CPU and the xmit path can operate on different CPUs,
 174 * performance is increased by ensuring that the completion
 175 * path and the xmit path operate on different cache lines.
 176 * This is particularly important if the xmit path is always
 177 * executing on one CPU which is different from the completion
 178 * path.  There is also a cache line for members which are
 179 * read but not written on the fast path.
 180 *
 181 * @efx: The associated Efx NIC
 182 * @queue: DMA queue number
 
 183 * @channel: The associated channel
 184 * @core_txq: The networking core TX queue structure
 185 * @buffer: The software buffer ring
 186 * @tsoh_page: Array of pages of TSO header buffers
 
 187 * @txd: The hardware descriptor ring
 188 * @ptr_mask: The size of the ring minus 1.
 189 * @piobuf: PIO buffer region for this TX queue (shared with its partner).
 190 *	Size of the region is efx_piobuf_size.
 191 * @piobuf_offset: Buffer offset to be specified in PIO descriptors
 192 * @initialised: Has hardware queue been initialised?
 
 
 
 193 * @read_count: Current read pointer.
 194 *	This is the number of buffers that have been removed from both rings.
 195 * @old_write_count: The value of @write_count when last checked.
 196 *	This is here for performance reasons.  The xmit path will
 197 *	only get the up-to-date value of @write_count if this
 198 *	variable indicates that the queue is empty.  This is to
 199 *	avoid cache-line ping-pong between the xmit path and the
 200 *	completion path.
 201 * @merge_events: Number of TX merged completion events
 
 
 
 
 202 * @insert_count: Current insert pointer
 203 *	This is the number of buffers that have been added to the
 204 *	software ring.
 205 * @write_count: Current write pointer
 206 *	This is the number of buffers that have been added to the
 207 *	hardware ring.
 
 
 
 
 
 
 208 * @old_read_count: The value of read_count when last checked.
 209 *	This is here for performance reasons.  The xmit path will
 210 *	only get the up-to-date value of read_count if this
 211 *	variable indicates that the queue is full.  This is to
 212 *	avoid cache-line ping-pong between the xmit path and the
 213 *	completion path.
 214 * @tso_bursts: Number of times TSO xmit invoked by kernel
 215 * @tso_long_headers: Number of packets with headers too long for standard
 216 *	blocks
 217 * @tso_packets: Number of packets via the TSO xmit path
 
 218 * @pushes: Number of times the TX push feature has been used
 219 * @pio_packets: Number of times the TX PIO feature has been used
 
 
 220 * @empty_read_count: If the completion path has seen the queue as empty
 221 *	and the transmission path has not yet checked this, the value of
 222 *	@read_count bitwise-added to %EFX_EMPTY_COUNT_VALID; otherwise 0.
 223 */
 224struct efx_tx_queue {
 225	/* Members which don't change on the fast path */
 226	struct efx_nic *efx ____cacheline_aligned_in_smp;
 227	unsigned queue;
 
 228	struct efx_channel *channel;
 229	struct netdev_queue *core_txq;
 230	struct efx_tx_buffer *buffer;
 231	struct efx_buffer *tsoh_page;
 232	struct efx_special_buffer txd;
 233	unsigned int ptr_mask;
 234	void __iomem *piobuf;
 235	unsigned int piobuf_offset;
 236	bool initialised;
 
 
 
 
 237
 238	/* Members used mainly on the completion path */
 239	unsigned int read_count ____cacheline_aligned_in_smp;
 240	unsigned int old_write_count;
 241	unsigned int merge_events;
 
 
 
 
 
 242
 243	/* Members used only on the xmit path */
 244	unsigned int insert_count ____cacheline_aligned_in_smp;
 245	unsigned int write_count;
 
 246	unsigned int old_read_count;
 247	unsigned int tso_bursts;
 248	unsigned int tso_long_headers;
 249	unsigned int tso_packets;
 
 250	unsigned int pushes;
 251	unsigned int pio_packets;
 
 
 
 
 252
 253	/* Members shared between paths and sometimes updated */
 254	unsigned int empty_read_count ____cacheline_aligned_in_smp;
 255#define EFX_EMPTY_COUNT_VALID 0x80000000
 256	atomic_t flush_outstanding;
 257};
 258
 
 
 
 259/**
 260 * struct efx_rx_buffer - An Efx RX data buffer
 261 * @dma_addr: DMA base address of the buffer
 262 * @page: The associated page buffer.
 263 *	Will be %NULL if the buffer slot is currently free.
 264 * @page_offset: If pending: offset in @page of DMA base address.
 265 *	If completed: offset in @page of Ethernet header.
 266 * @len: If pending: length for DMA descriptor.
 267 *	If completed: received length, excluding hash prefix.
 268 * @flags: Flags for buffer and packet state.  These are only set on the
 269 *	first buffer of a scattered packet.
 270 */
 271struct efx_rx_buffer {
 272	dma_addr_t dma_addr;
 273	struct page *page;
 274	u16 page_offset;
 275	u16 len;
 276	u16 flags;
 277};
 278#define EFX_RX_BUF_LAST_IN_PAGE	0x0001
 279#define EFX_RX_PKT_CSUMMED	0x0002
 280#define EFX_RX_PKT_DISCARD	0x0004
 281#define EFX_RX_PKT_TCP		0x0040
 282#define EFX_RX_PKT_PREFIX_LEN	0x0080	/* length is in prefix only */
 
 283
 284/**
 285 * struct efx_rx_page_state - Page-based rx buffer state
 286 *
 287 * Inserted at the start of every page allocated for receive buffers.
 288 * Used to facilitate sharing dma mappings between recycled rx buffers
 289 * and those passed up to the kernel.
 290 *
 291 * @dma_addr: The dma address of this page.
 292 */
 293struct efx_rx_page_state {
 294	dma_addr_t dma_addr;
 295
 296	unsigned int __pad[0] ____cacheline_aligned;
 297};
 298
 299/**
 300 * struct efx_rx_queue - An Efx RX queue
 301 * @efx: The associated Efx NIC
 302 * @core_index:  Index of network core RX queue.  Will be >= 0 iff this
 303 *	is associated with a real RX queue.
 304 * @buffer: The software buffer ring
 305 * @rxd: The hardware descriptor ring
 306 * @ptr_mask: The size of the ring minus 1.
 307 * @refill_enabled: Enable refill whenever fill level is low
 308 * @flush_pending: Set when a RX flush is pending. Has the same lifetime as
 309 *	@rxq_flush_pending.
 310 * @added_count: Number of buffers added to the receive queue.
 311 * @notified_count: Number of buffers given to NIC (<= @added_count).
 312 * @removed_count: Number of buffers removed from the receive queue.
 313 * @scatter_n: Used by NIC specific receive code.
 314 * @scatter_len: Used by NIC specific receive code.
 315 * @page_ring: The ring to store DMA mapped pages for reuse.
 316 * @page_add: Counter to calculate the write pointer for the recycle ring.
 317 * @page_remove: Counter to calculate the read pointer for the recycle ring.
 318 * @page_recycle_count: The number of pages that have been recycled.
 319 * @page_recycle_failed: The number of pages that couldn't be recycled because
 320 *      the kernel still held a reference to them.
 321 * @page_recycle_full: The number of pages that were released because the
 322 *      recycle ring was full.
 323 * @page_ptr_mask: The number of pages in the RX recycle ring minus 1.
 324 * @max_fill: RX descriptor maximum fill level (<= ring size)
 325 * @fast_fill_trigger: RX descriptor fill level that will trigger a fast fill
 326 *	(<= @max_fill)
 327 * @min_fill: RX descriptor minimum non-zero fill level.
 328 *	This records the minimum fill level observed when a ring
 329 *	refill was triggered.
 330 * @recycle_count: RX buffer recycle counter.
 331 * @slow_fill: Timer used to defer efx_nic_generate_fill_event().
 332 */
 333struct efx_rx_queue {
 334	struct efx_nic *efx;
 335	int core_index;
 336	struct efx_rx_buffer *buffer;
 337	struct efx_special_buffer rxd;
 338	unsigned int ptr_mask;
 339	bool refill_enabled;
 340	bool flush_pending;
 341
 342	unsigned int added_count;
 343	unsigned int notified_count;
 344	unsigned int removed_count;
 345	unsigned int scatter_n;
 346	unsigned int scatter_len;
 347	struct page **page_ring;
 348	unsigned int page_add;
 349	unsigned int page_remove;
 350	unsigned int page_recycle_count;
 351	unsigned int page_recycle_failed;
 352	unsigned int page_recycle_full;
 353	unsigned int page_ptr_mask;
 354	unsigned int max_fill;
 355	unsigned int fast_fill_trigger;
 356	unsigned int min_fill;
 357	unsigned int min_overfill;
 358	unsigned int recycle_count;
 359	struct timer_list slow_fill;
 360	unsigned int slow_fill_count;
 
 
 361};
 362
 363enum efx_sync_events_state {
 364	SYNC_EVENTS_DISABLED = 0,
 365	SYNC_EVENTS_QUIESCENT,
 366	SYNC_EVENTS_REQUESTED,
 367	SYNC_EVENTS_VALID,
 368};
 369
 370/**
 371 * struct efx_channel - An Efx channel
 372 *
 373 * A channel comprises an event queue, at least one TX queue, at least
 374 * one RX queue, and an associated tasklet for processing the event
 375 * queue.
 376 *
 377 * @efx: Associated Efx NIC
 378 * @channel: Channel instance number
 379 * @type: Channel type definition
 380 * @eventq_init: Event queue initialised flag
 381 * @enabled: Channel enabled indicator
 382 * @irq: IRQ number (MSI and MSI-X only)
 383 * @irq_moderation: IRQ moderation value (in hardware ticks)
 384 * @napi_dev: Net device used with NAPI
 385 * @napi_str: NAPI control structure
 
 
 386 * @eventq: Event queue buffer
 387 * @eventq_mask: Event queue pointer mask
 388 * @eventq_read_ptr: Event queue read pointer
 389 * @event_test_cpu: Last CPU to handle interrupt or test event for this channel
 390 * @irq_count: Number of IRQs since last adaptive moderation decision
 391 * @irq_mod_score: IRQ moderation score
 
 
 
 392 * @n_rx_tobe_disc: Count of RX_TOBE_DISC errors
 393 * @n_rx_ip_hdr_chksum_err: Count of RX IP header checksum errors
 394 * @n_rx_tcp_udp_chksum_err: Count of RX TCP and UDP checksum errors
 395 * @n_rx_mcast_mismatch: Count of unmatched multicast frames
 396 * @n_rx_frm_trunc: Count of RX_FRM_TRUNC errors
 397 * @n_rx_overlength: Count of RX_OVERLENGTH errors
 398 * @n_skbuff_leaks: Count of skbuffs leaked due to RX overrun
 399 * @n_rx_nodesc_trunc: Number of RX packets truncated and then dropped due to
 400 *	lack of descriptors
 401 * @n_rx_merge_events: Number of RX merged completion events
 402 * @n_rx_merge_packets: Number of RX packets completed by merged events
 403 * @rx_pkt_n_frags: Number of fragments in next packet to be delivered by
 404 *	__efx_rx_packet(), or zero if there is none
 405 * @rx_pkt_index: Ring index of first buffer for next packet to be delivered
 406 *	by __efx_rx_packet(), if @rx_pkt_n_frags != 0
 
 407 * @rx_queue: RX queue for this channel
 408 * @tx_queue: TX queues for this channel
 409 * @sync_events_state: Current state of sync events on this channel
 410 * @sync_timestamp_major: Major part of the last ptp sync event
 411 * @sync_timestamp_minor: Minor part of the last ptp sync event
 412 */
 413struct efx_channel {
 414	struct efx_nic *efx;
 415	int channel;
 416	const struct efx_channel_type *type;
 417	bool eventq_init;
 418	bool enabled;
 419	int irq;
 420	unsigned int irq_moderation;
 421	struct net_device *napi_dev;
 422	struct napi_struct napi_str;
 
 
 
 423	struct efx_special_buffer eventq;
 424	unsigned int eventq_mask;
 425	unsigned int eventq_read_ptr;
 426	int event_test_cpu;
 427
 428	unsigned int irq_count;
 429	unsigned int irq_mod_score;
 430#ifdef CONFIG_RFS_ACCEL
 431	unsigned int rfs_filters_added;
 
 
 
 432#endif
 433
 434	unsigned n_rx_tobe_disc;
 435	unsigned n_rx_ip_hdr_chksum_err;
 436	unsigned n_rx_tcp_udp_chksum_err;
 437	unsigned n_rx_mcast_mismatch;
 438	unsigned n_rx_frm_trunc;
 439	unsigned n_rx_overlength;
 440	unsigned n_skbuff_leaks;
 
 
 
 
 
 441	unsigned int n_rx_nodesc_trunc;
 442	unsigned int n_rx_merge_events;
 443	unsigned int n_rx_merge_packets;
 444
 445	unsigned int rx_pkt_n_frags;
 446	unsigned int rx_pkt_index;
 447
 
 
 448	struct efx_rx_queue rx_queue;
 449	struct efx_tx_queue tx_queue[EFX_TXQ_TYPES];
 450
 451	enum efx_sync_events_state sync_events_state;
 452	u32 sync_timestamp_major;
 453	u32 sync_timestamp_minor;
 454};
 455
 456/**
 457 * struct efx_msi_context - Context for each MSI
 458 * @efx: The associated NIC
 459 * @index: Index of the channel/IRQ
 460 * @name: Name of the channel/IRQ
 461 *
 462 * Unlike &struct efx_channel, this is never reallocated and is always
 463 * safe for the IRQ handler to access.
 464 */
 465struct efx_msi_context {
 466	struct efx_nic *efx;
 467	unsigned int index;
 468	char name[IFNAMSIZ + 6];
 469};
 470
 471/**
 472 * struct efx_channel_type - distinguishes traffic and extra channels
 473 * @handle_no_channel: Handle failure to allocate an extra channel
 474 * @pre_probe: Set up extra state prior to initialisation
 475 * @post_remove: Tear down extra state after finalisation, if allocated.
 476 *	May be called on channels that have not been probed.
 477 * @get_name: Generate the channel's name (used for its IRQ handler)
 478 * @copy: Copy the channel state prior to reallocation.  May be %NULL if
 479 *	reallocation is not supported.
 480 * @receive_skb: Handle an skb ready to be passed to netif_receive_skb()
 
 
 481 * @keep_eventq: Flag for whether event queue should be kept initialised
 482 *	while the device is stopped
 
 
 483 */
 484struct efx_channel_type {
 485	void (*handle_no_channel)(struct efx_nic *);
 486	int (*pre_probe)(struct efx_channel *);
 487	void (*post_remove)(struct efx_channel *);
 488	void (*get_name)(struct efx_channel *, char *buf, size_t len);
 489	struct efx_channel *(*copy)(const struct efx_channel *);
 490	bool (*receive_skb)(struct efx_channel *, struct sk_buff *);
 
 491	bool keep_eventq;
 
 492};
 493
 494enum efx_led_mode {
 495	EFX_LED_OFF	= 0,
 496	EFX_LED_ON	= 1,
 497	EFX_LED_DEFAULT	= 2
 498};
 499
 500#define STRING_TABLE_LOOKUP(val, member) \
 501	((val) < member ## _max) ? member ## _names[val] : "(invalid)"
 502
 503extern const char *const efx_loopback_mode_names[];
 504extern const unsigned int efx_loopback_mode_max;
 505#define LOOPBACK_MODE(efx) \
 506	STRING_TABLE_LOOKUP((efx)->loopback_mode, efx_loopback_mode)
 507
 508extern const char *const efx_reset_type_names[];
 509extern const unsigned int efx_reset_type_max;
 510#define RESET_TYPE(type) \
 511	STRING_TABLE_LOOKUP(type, efx_reset_type)
 512
 
 
 513enum efx_int_mode {
 514	/* Be careful if altering to correct macro below */
 515	EFX_INT_MODE_MSIX = 0,
 516	EFX_INT_MODE_MSI = 1,
 517	EFX_INT_MODE_LEGACY = 2,
 518	EFX_INT_MODE_MAX	/* Insert any new items before this */
 519};
 520#define EFX_INT_MODE_USE_MSI(x) (((x)->interrupt_mode) <= EFX_INT_MODE_MSI)
 521
 522enum nic_state {
 523	STATE_UNINIT = 0,	/* device being probed/removed or is frozen */
 524	STATE_READY = 1,	/* hardware ready and netdev registered */
 525	STATE_DISABLED = 2,	/* device disabled due to hardware errors */
 526	STATE_RECOVERY = 3,	/* device recovering from PCI error */
 527};
 528
 529/* Forward declaration */
 530struct efx_nic;
 531
 532/* Pseudo bit-mask flow control field */
 533#define EFX_FC_RX	FLOW_CTRL_RX
 534#define EFX_FC_TX	FLOW_CTRL_TX
 535#define EFX_FC_AUTO	4
 536
 537/**
 538 * struct efx_link_state - Current state of the link
 539 * @up: Link is up
 540 * @fd: Link is full-duplex
 541 * @fc: Actual flow control flags
 542 * @speed: Link speed (Mbps)
 543 */
 544struct efx_link_state {
 545	bool up;
 546	bool fd;
 547	u8 fc;
 548	unsigned int speed;
 549};
 550
 551static inline bool efx_link_state_equal(const struct efx_link_state *left,
 552					const struct efx_link_state *right)
 553{
 554	return left->up == right->up && left->fd == right->fd &&
 555		left->fc == right->fc && left->speed == right->speed;
 556}
 557
 558/**
 559 * struct efx_phy_operations - Efx PHY operations table
 560 * @probe: Probe PHY and initialise efx->mdio.mode_support, efx->mdio.mmds,
 561 *	efx->loopback_modes.
 562 * @init: Initialise PHY
 563 * @fini: Shut down PHY
 564 * @reconfigure: Reconfigure PHY (e.g. for new link parameters)
 565 * @poll: Update @link_state and report whether it changed.
 566 *	Serialised by the mac_lock.
 567 * @get_settings: Get ethtool settings. Serialised by the mac_lock.
 568 * @set_settings: Set ethtool settings. Serialised by the mac_lock.
 
 
 569 * @set_npage_adv: Set abilities advertised in (Extended) Next Page
 570 *	(only needed where AN bit is set in mmds)
 571 * @test_alive: Test that PHY is 'alive' (online)
 572 * @test_name: Get the name of a PHY-specific test/result
 573 * @run_tests: Run tests and record results as appropriate (offline).
 574 *	Flags are the ethtool tests flags.
 575 */
 576struct efx_phy_operations {
 577	int (*probe) (struct efx_nic *efx);
 578	int (*init) (struct efx_nic *efx);
 579	void (*fini) (struct efx_nic *efx);
 580	void (*remove) (struct efx_nic *efx);
 581	int (*reconfigure) (struct efx_nic *efx);
 582	bool (*poll) (struct efx_nic *efx);
 583	void (*get_settings) (struct efx_nic *efx,
 584			      struct ethtool_cmd *ecmd);
 585	int (*set_settings) (struct efx_nic *efx,
 586			     struct ethtool_cmd *ecmd);
 
 
 
 587	void (*set_npage_adv) (struct efx_nic *efx, u32);
 588	int (*test_alive) (struct efx_nic *efx);
 589	const char *(*test_name) (struct efx_nic *efx, unsigned int index);
 590	int (*run_tests) (struct efx_nic *efx, int *results, unsigned flags);
 591	int (*get_module_eeprom) (struct efx_nic *efx,
 592			       struct ethtool_eeprom *ee,
 593			       u8 *data);
 594	int (*get_module_info) (struct efx_nic *efx,
 595				struct ethtool_modinfo *modinfo);
 596};
 597
 598/**
 599 * enum efx_phy_mode - PHY operating mode flags
 600 * @PHY_MODE_NORMAL: on and should pass traffic
 601 * @PHY_MODE_TX_DISABLED: on with TX disabled
 602 * @PHY_MODE_LOW_POWER: set to low power through MDIO
 603 * @PHY_MODE_OFF: switched off through external control
 604 * @PHY_MODE_SPECIAL: on but will not pass traffic
 605 */
 606enum efx_phy_mode {
 607	PHY_MODE_NORMAL		= 0,
 608	PHY_MODE_TX_DISABLED	= 1,
 609	PHY_MODE_LOW_POWER	= 2,
 610	PHY_MODE_OFF		= 4,
 611	PHY_MODE_SPECIAL	= 8,
 612};
 613
 614static inline bool efx_phy_mode_disabled(enum efx_phy_mode mode)
 615{
 616	return !!(mode & ~PHY_MODE_TX_DISABLED);
 617}
 618
 619/**
 620 * struct efx_hw_stat_desc - Description of a hardware statistic
 621 * @name: Name of the statistic as visible through ethtool, or %NULL if
 622 *	it should not be exposed
 623 * @dma_width: Width in bits (0 for non-DMA statistics)
 624 * @offset: Offset within stats (ignored for non-DMA statistics)
 625 */
 626struct efx_hw_stat_desc {
 627	const char *name;
 628	u16 dma_width;
 629	u16 offset;
 630};
 631
 632/* Number of bits used in a multicast filter hash address */
 633#define EFX_MCAST_HASH_BITS 8
 634
 635/* Number of (single-bit) entries in a multicast filter hash */
 636#define EFX_MCAST_HASH_ENTRIES (1 << EFX_MCAST_HASH_BITS)
 637
 638/* An Efx multicast filter hash */
 639union efx_multicast_hash {
 640	u8 byte[EFX_MCAST_HASH_ENTRIES / 8];
 641	efx_oword_t oword[EFX_MCAST_HASH_ENTRIES / sizeof(efx_oword_t) / 8];
 642};
 643
 644struct efx_vf;
 645struct vfdi_status;
 646
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 647/**
 648 * struct efx_nic - an Efx NIC
 649 * @name: Device name (net device name or bus id before net device registered)
 650 * @pci_dev: The PCI device
 651 * @node: List node for maintaning primary/secondary function lists
 652 * @primary: &struct efx_nic instance for the primary function of this
 653 *	controller.  May be the same structure, and may be %NULL if no
 654 *	primary function is bound.  Serialised by rtnl_lock.
 655 * @secondary_list: List of &struct efx_nic instances for the secondary PCI
 656 *	functions of the controller, if this is for the primary function.
 657 *	Serialised by rtnl_lock.
 658 * @type: Controller type attributes
 659 * @legacy_irq: IRQ number
 660 * @workqueue: Workqueue for port reconfigures and the HW monitor.
 661 *	Work items do not hold and must not acquire RTNL.
 662 * @workqueue_name: Name of workqueue
 663 * @reset_work: Scheduled reset workitem
 664 * @membase_phys: Memory BAR value as physical address
 665 * @membase: Memory BAR value
 
 666 * @interrupt_mode: Interrupt mode
 667 * @timer_quantum_ns: Interrupt timer quantum, in nanoseconds
 
 668 * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues
 669 * @irq_rx_moderation: IRQ moderation time for RX event queues
 
 670 * @msg_enable: Log message enable flags
 671 * @state: Device state number (%STATE_*). Serialised by the rtnl_lock.
 672 * @reset_pending: Bitmask for pending resets
 673 * @tx_queue: TX DMA queues
 674 * @rx_queue: RX DMA queues
 675 * @channel: Channels
 676 * @msi_context: Context for each MSI
 677 * @extra_channel_types: Types of extra (non-traffic) channels that
 678 *	should be allocated for this NIC
 679 * @rxq_entries: Size of receive queues requested by user.
 680 * @txq_entries: Size of transmit queues requested by user.
 681 * @txq_stop_thresh: TX queue fill level at or above which we stop it.
 682 * @txq_wake_thresh: TX queue fill level at or below which we wake it.
 683 * @tx_dc_base: Base qword address in SRAM of TX queue descriptor caches
 684 * @rx_dc_base: Base qword address in SRAM of RX queue descriptor caches
 685 * @sram_lim_qw: Qword address limit of SRAM
 686 * @next_buffer_table: First available buffer table id
 687 * @n_channels: Number of channels in use
 688 * @n_rx_channels: Number of channels used for RX (= number of RX queues)
 689 * @n_tx_channels: Number of channels used for TX
 
 690 * @rx_ip_align: RX DMA address offset to have IP header aligned in
 691 *	in accordance with NET_IP_ALIGN
 692 * @rx_dma_len: Current maximum RX DMA length
 693 * @rx_buffer_order: Order (log2) of number of pages for each RX buffer
 694 * @rx_buffer_truesize: Amortised allocation size of an RX buffer,
 695 *	for use in sk_buff::truesize
 696 * @rx_prefix_size: Size of RX prefix before packet data
 697 * @rx_packet_hash_offset: Offset of RX flow hash from start of packet data
 698 *	(valid only if @rx_prefix_size != 0; always negative)
 699 * @rx_packet_len_offset: Offset of RX packet length from start of packet data
 700 *	(valid only for NICs that set %EFX_RX_PKT_PREFIX_LEN; always negative)
 701 * @rx_packet_ts_offset: Offset of timestamp from start of packet data
 702 *	(valid only if channel->sync_timestamps_enabled; always negative)
 703 * @rx_hash_key: Toeplitz hash key for RSS
 704 * @rx_indir_table: Indirection table for RSS
 705 * @rx_scatter: Scatter mode enabled for receives
 
 
 
 706 * @int_error_count: Number of internal errors seen recently
 707 * @int_error_expire: Time at which error count will be expired
 708 * @irq_soft_enabled: Are IRQs soft-enabled? If not, IRQ handler will
 709 *	acknowledge but do nothing else.
 710 * @irq_status: Interrupt status buffer
 711 * @irq_zero_count: Number of legacy IRQs seen with queue flags == 0
 712 * @irq_level: IRQ level/index for IRQs not triggered by an event queue
 713 * @selftest_work: Work item for asynchronous self-test
 714 * @mtd_list: List of MTDs attached to the NIC
 715 * @nic_data: Hardware dependent state
 716 * @mcdi: Management-Controller-to-Driver Interface state
 717 * @mac_lock: MAC access lock. Protects @port_enabled, @phy_mode,
 718 *	efx_monitor() and efx_reconfigure_port()
 719 * @port_enabled: Port enabled indicator.
 720 *	Serialises efx_stop_all(), efx_start_all(), efx_monitor() and
 721 *	efx_mac_work() with kernel interfaces. Safe to read under any
 722 *	one of the rtnl_lock, mac_lock, or netif_tx_lock, but all three must
 723 *	be held to modify it.
 724 * @port_initialized: Port initialized?
 725 * @net_dev: Operating system network device. Consider holding the rtnl lock
 
 
 
 726 * @stats_buffer: DMA buffer for statistics
 727 * @phy_type: PHY type
 728 * @phy_op: PHY interface
 729 * @phy_data: PHY private data (including PHY-specific stats)
 730 * @mdio: PHY MDIO interface
 731 * @mdio_bus: PHY MDIO bus ID (only used by Siena)
 732 * @phy_mode: PHY operating mode. Serialised by @mac_lock.
 733 * @link_advertising: Autonegotiation advertising flags
 
 
 734 * @link_state: Current state of the link
 735 * @n_link_state_changes: Number of times the link has changed state
 736 * @unicast_filter: Flag for Falcon-arch simple unicast filter.
 737 *	Protected by @mac_lock.
 738 * @multicast_hash: Multicast hash table for Falcon-arch.
 739 *	Protected by @mac_lock.
 740 * @wanted_fc: Wanted flow control flags
 741 * @fc_disable: When non-zero flow control is disabled. Typically used to
 742 *	ensure that network back pressure doesn't delay dma queue flushes.
 743 *	Serialised by the rtnl lock.
 744 * @mac_work: Work item for changing MAC promiscuity and multicast hash
 745 * @loopback_mode: Loopback status
 746 * @loopback_modes: Supported loopback mode bitmask
 747 * @loopback_selftest: Offline self-test private state
 748 * @filter_lock: Filter table lock
 749 * @filter_state: Architecture-dependent filter table state
 750 * @rps_flow_id: Flow IDs of filters allocated for accelerated RFS,
 751 *	indexed by filter ID
 752 * @rps_expire_index: Next index to check for expiry in @rps_flow_id
 
 
 
 
 
 
 
 753 * @active_queues: Count of RX and TX queues that haven't been flushed and drained.
 754 * @rxq_flush_pending: Count of number of receive queues that need to be flushed.
 755 *	Decremented when the efx_flush_rx_queue() is called.
 756 * @rxq_flush_outstanding: Count of number of RX flushes started but not yet
 757 *	completed (either success or failure). Not used when MCDI is used to
 758 *	flush receive queues.
 759 * @flush_wq: wait queue used by efx_nic_flush_queues() to wait for flush completions.
 760 * @vf: Array of &struct efx_vf objects.
 761 * @vf_count: Number of VFs intended to be enabled.
 762 * @vf_init_count: Number of VFs that have been fully initialised.
 763 * @vi_scale: log2 number of vnics per VF.
 764 * @vf_buftbl_base: The zeroth buffer table index used to back VF queues.
 765 * @vfdi_status: Common VFDI status page to be dmad to VF address space.
 766 * @local_addr_list: List of local addresses. Protected by %local_lock.
 767 * @local_page_list: List of DMA addressable pages used to broadcast
 768 *	%local_addr_list. Protected by %local_lock.
 769 * @local_lock: Mutex protecting %local_addr_list and %local_page_list.
 770 * @peer_work: Work item to broadcast peer addresses to VMs.
 771 * @ptp_data: PTP state data
 
 772 * @vpd_sn: Serial number read from VPD
 773 * @monitor_work: Hardware monitor workitem
 774 * @biu_lock: BIU (bus interface unit) lock
 775 * @last_irq_cpu: Last CPU to handle a possible test interrupt.  This
 776 *	field is used by efx_test_interrupts() to verify that an
 777 *	interrupt has occurred.
 778 * @stats_lock: Statistics update lock. Must be held when calling
 779 *	efx_nic_type::{update,start,stop}_stats.
 
 780 *
 781 * This is stored in the private area of the &struct net_device.
 782 */
 783struct efx_nic {
 784	/* The following fields should be written very rarely */
 785
 786	char name[IFNAMSIZ];
 787	struct list_head node;
 788	struct efx_nic *primary;
 789	struct list_head secondary_list;
 790	struct pci_dev *pci_dev;
 791	unsigned int port_num;
 792	const struct efx_nic_type *type;
 793	int legacy_irq;
 794	bool eeh_disabled_legacy_irq;
 795	struct workqueue_struct *workqueue;
 796	char workqueue_name[16];
 797	struct work_struct reset_work;
 798	resource_size_t membase_phys;
 799	void __iomem *membase;
 800
 
 
 801	enum efx_int_mode interrupt_mode;
 802	unsigned int timer_quantum_ns;
 
 803	bool irq_rx_adaptive;
 804	unsigned int irq_rx_moderation;
 
 805	u32 msg_enable;
 806
 807	enum nic_state state;
 808	unsigned long reset_pending;
 809
 810	struct efx_channel *channel[EFX_MAX_CHANNELS];
 811	struct efx_msi_context msi_context[EFX_MAX_CHANNELS];
 812	const struct efx_channel_type *
 813	extra_channel_type[EFX_MAX_EXTRA_CHANNELS];
 814
 815	unsigned rxq_entries;
 816	unsigned txq_entries;
 817	unsigned int txq_stop_thresh;
 818	unsigned int txq_wake_thresh;
 819
 820	unsigned tx_dc_base;
 821	unsigned rx_dc_base;
 822	unsigned sram_lim_qw;
 823	unsigned next_buffer_table;
 824
 825	unsigned int max_channels;
 
 826	unsigned n_channels;
 827	unsigned n_rx_channels;
 828	unsigned rss_spread;
 829	unsigned tx_channel_offset;
 830	unsigned n_tx_channels;
 
 831	unsigned int rx_ip_align;
 832	unsigned int rx_dma_len;
 833	unsigned int rx_buffer_order;
 834	unsigned int rx_buffer_truesize;
 835	unsigned int rx_page_buf_step;
 836	unsigned int rx_bufs_per_page;
 837	unsigned int rx_pages_per_batch;
 838	unsigned int rx_prefix_size;
 839	int rx_packet_hash_offset;
 840	int rx_packet_len_offset;
 841	int rx_packet_ts_offset;
 842	u8 rx_hash_key[40];
 843	u32 rx_indir_table[128];
 844	bool rx_scatter;
 
 
 845
 846	unsigned int_error_count;
 847	unsigned long int_error_expire;
 848
 849	bool irq_soft_enabled;
 850	struct efx_buffer irq_status;
 851	unsigned irq_zero_count;
 852	unsigned irq_level;
 853	struct delayed_work selftest_work;
 854
 855#ifdef CONFIG_SFC_MTD
 856	struct list_head mtd_list;
 857#endif
 858
 859	void *nic_data;
 860	struct efx_mcdi_data *mcdi;
 861
 862	struct mutex mac_lock;
 863	struct work_struct mac_work;
 864	bool port_enabled;
 865
 866	bool mc_bist_for_other_fn;
 867	bool port_initialized;
 868	struct net_device *net_dev;
 869
 
 
 
 870	struct efx_buffer stats_buffer;
 871	u64 rx_nodesc_drops_total;
 872	u64 rx_nodesc_drops_while_down;
 873	bool rx_nodesc_drops_prev_state;
 874
 875	unsigned int phy_type;
 876	const struct efx_phy_operations *phy_op;
 877	void *phy_data;
 878	struct mdio_if_info mdio;
 879	unsigned int mdio_bus;
 880	enum efx_phy_mode phy_mode;
 881
 882	u32 link_advertising;
 
 883	struct efx_link_state link_state;
 884	unsigned int n_link_state_changes;
 885
 886	bool unicast_filter;
 887	union efx_multicast_hash multicast_hash;
 888	u8 wanted_fc;
 889	unsigned fc_disable;
 890
 891	atomic_t rx_reset;
 892	enum efx_loopback_mode loopback_mode;
 893	u64 loopback_modes;
 894
 895	void *loopback_selftest;
 896
 897	spinlock_t filter_lock;
 898	void *filter_state;
 899#ifdef CONFIG_RFS_ACCEL
 900	u32 *rps_flow_id;
 
 901	unsigned int rps_expire_index;
 
 
 
 
 
 902#endif
 903
 904	atomic_t active_queues;
 905	atomic_t rxq_flush_pending;
 906	atomic_t rxq_flush_outstanding;
 907	wait_queue_head_t flush_wq;
 908
 909#ifdef CONFIG_SFC_SRIOV
 910	struct efx_channel *vfdi_channel;
 911	struct efx_vf *vf;
 912	unsigned vf_count;
 913	unsigned vf_init_count;
 914	unsigned vi_scale;
 915	unsigned vf_buftbl_base;
 916	struct efx_buffer vfdi_status;
 917	struct list_head local_addr_list;
 918	struct list_head local_page_list;
 919	struct mutex local_lock;
 920	struct work_struct peer_work;
 921#endif
 922
 923	struct efx_ptp_data *ptp_data;
 
 924
 925	char *vpd_sn;
 926
 927	/* The following fields may be written more often */
 928
 929	struct delayed_work monitor_work ____cacheline_aligned_in_smp;
 930	spinlock_t biu_lock;
 931	int last_irq_cpu;
 932	spinlock_t stats_lock;
 
 933};
 934
 935static inline int efx_dev_registered(struct efx_nic *efx)
 936{
 937	return efx->net_dev->reg_state == NETREG_REGISTERED;
 938}
 939
 940static inline unsigned int efx_port_num(struct efx_nic *efx)
 941{
 942	return efx->port_num;
 943}
 944
 945struct efx_mtd_partition {
 946	struct list_head node;
 947	struct mtd_info mtd;
 948	const char *dev_type_name;
 949	const char *type_name;
 950	char name[IFNAMSIZ + 20];
 951};
 952
 
 
 
 
 
 
 
 
 
 953/**
 954 * struct efx_nic_type - Efx device type definition
 
 955 * @mem_map_size: Get memory BAR mapped size
 956 * @probe: Probe the controller
 957 * @remove: Free resources allocated by probe()
 958 * @init: Initialise the controller
 959 * @dimension_resources: Dimension controller resources (buffer table,
 960 *	and VIs once the available interrupt resources are clear)
 961 * @fini: Shut down the controller
 962 * @monitor: Periodic function for polling link state and hardware monitor
 963 * @map_reset_reason: Map ethtool reset reason to a reset method
 964 * @map_reset_flags: Map ethtool reset flags to a reset method, if possible
 965 * @reset: Reset the controller hardware and possibly the PHY.  This will
 966 *	be called while the controller is uninitialised.
 967 * @probe_port: Probe the MAC and PHY
 968 * @remove_port: Free resources allocated by probe_port()
 969 * @handle_global_event: Handle a "global" event (may be %NULL)
 970 * @fini_dmaq: Flush and finalise DMA queues (RX and TX queues)
 971 * @prepare_flush: Prepare the hardware for flushing the DMA queues
 972 *	(for Falcon architecture)
 973 * @finish_flush: Clean up after flushing the DMA queues (for Falcon
 974 *	architecture)
 975 * @prepare_flr: Prepare for an FLR
 976 * @finish_flr: Clean up after an FLR
 977 * @describe_stats: Describe statistics for ethtool
 978 * @update_stats: Update statistics not provided by event handling.
 979 *	Either argument may be %NULL.
 980 * @start_stats: Start the regular fetching of statistics
 981 * @pull_stats: Pull stats from the NIC and wait until they arrive.
 982 * @stop_stats: Stop the regular fetching of statistics
 983 * @set_id_led: Set state of identifying LED or revert to automatic function
 984 * @push_irq_moderation: Apply interrupt moderation value
 985 * @reconfigure_port: Push loopback/power/txdis changes to the MAC and PHY
 986 * @prepare_enable_fc_tx: Prepare MAC to enable pause frame TX (may be %NULL)
 987 * @reconfigure_mac: Push MAC address, MTU, flow control and filter settings
 988 *	to the hardware.  Serialised by the mac_lock.
 989 * @check_mac_fault: Check MAC fault state. True if fault present.
 990 * @get_wol: Get WoL configuration from driver state
 991 * @set_wol: Push WoL configuration to the NIC
 992 * @resume_wol: Synchronise WoL state between driver and MC (e.g. after resume)
 993 * @test_chip: Test registers.  May use efx_farch_test_registers(), and is
 994 *	expected to reset the NIC.
 995 * @test_nvram: Test validity of NVRAM contents
 996 * @mcdi_request: Send an MCDI request with the given header and SDU.
 997 *	The SDU length may be any value from 0 up to the protocol-
 998 *	defined maximum, but its buffer will be padded to a multiple
 999 *	of 4 bytes.
1000 * @mcdi_poll_response: Test whether an MCDI response is available.
1001 * @mcdi_read_response: Read the MCDI response PDU.  The offset will
1002 *	be a multiple of 4.  The length may not be, but the buffer
1003 *	will be padded so it is safe to round up.
1004 * @mcdi_poll_reboot: Test whether the MCDI has rebooted.  If so,
1005 *	return an appropriate error code for aborting any current
1006 *	request; otherwise return 0.
1007 * @irq_enable_master: Enable IRQs on the NIC.  Each event queue must
1008 *	be separately enabled after this.
1009 * @irq_test_generate: Generate a test IRQ
1010 * @irq_disable_non_ev: Disable non-event IRQs on the NIC.  Each event
1011 *	queue must be separately disabled before this.
1012 * @irq_handle_msi: Handle MSI for a channel.  The @dev_id argument is
1013 *	a pointer to the &struct efx_msi_context for the channel.
1014 * @irq_handle_legacy: Handle legacy interrupt.  The @dev_id argument
1015 *	is a pointer to the &struct efx_nic.
1016 * @tx_probe: Allocate resources for TX queue
1017 * @tx_init: Initialise TX queue on the NIC
1018 * @tx_remove: Free resources for TX queue
1019 * @tx_write: Write TX descriptors and doorbell
1020 * @rx_push_rss_config: Write RSS hash key and indirection table to the NIC
 
 
 
 
 
1021 * @rx_probe: Allocate resources for RX queue
1022 * @rx_init: Initialise RX queue on the NIC
1023 * @rx_remove: Free resources for RX queue
1024 * @rx_write: Write RX descriptors and doorbell
1025 * @rx_defer_refill: Generate a refill reminder event
1026 * @ev_probe: Allocate resources for event queue
1027 * @ev_init: Initialise event queue on the NIC
1028 * @ev_fini: Deinitialise event queue on the NIC
1029 * @ev_remove: Free resources for event queue
1030 * @ev_process: Process events for a queue, up to the given NAPI quota
1031 * @ev_read_ack: Acknowledge read events on a queue, rearming its IRQ
1032 * @ev_test_generate: Generate a test event
1033 * @filter_table_probe: Probe filter capabilities and set up filter software state
1034 * @filter_table_restore: Restore filters removed from hardware
1035 * @filter_table_remove: Remove filters from hardware and tear down software state
1036 * @filter_update_rx_scatter: Update filters after change to rx scatter setting
1037 * @filter_insert: add or replace a filter
1038 * @filter_remove_safe: remove a filter by ID, carefully
1039 * @filter_get_safe: retrieve a filter by ID, carefully
1040 * @filter_clear_rx: Remove all RX filters whose priority is less than or
1041 *	equal to the given priority and is not %EFX_FILTER_PRI_AUTO
1042 * @filter_count_rx_used: Get the number of filters in use at a given priority
1043 * @filter_get_rx_id_limit: Get maximum value of a filter id, plus 1
1044 * @filter_get_rx_ids: Get list of RX filters at a given priority
1045 * @filter_rfs_insert: Add or replace a filter for RFS.  This must be
1046 *	atomic.  The hardware change may be asynchronous but should
1047 *	not be delayed for long.  It may fail if this can't be done
1048 *	atomically.
1049 * @filter_rfs_expire_one: Consider expiring a filter inserted for RFS.
1050 *	This must check whether the specified table entry is used by RFS
1051 *	and that rps_may_expire_flow() returns true for it.
1052 * @mtd_probe: Probe and add MTD partitions associated with this net device,
1053 *	 using efx_mtd_add()
1054 * @mtd_rename: Set an MTD partition name using the net device name
1055 * @mtd_read: Read from an MTD partition
1056 * @mtd_erase: Erase part of an MTD partition
1057 * @mtd_write: Write to an MTD partition
1058 * @mtd_sync: Wait for write-back to complete on MTD partition.  This
1059 *	also notifies the driver that a writer has finished using this
1060 *	partition.
1061 * @ptp_write_host_time: Send host time to MC as part of sync protocol
1062 * @ptp_set_ts_sync_events: Enable or disable sync events for inline RX
1063 *	timestamping, possibly only temporarily for the purposes of a reset.
1064 * @ptp_set_ts_config: Set hardware timestamp configuration.  The flags
1065 *	and tx_type will already have been validated but this operation
1066 *	must validate and update rx_filter.
 
 
 
 
 
 
 
 
1067 * @revision: Hardware architecture revision
1068 * @txd_ptr_tbl_base: TX descriptor ring base address
1069 * @rxd_ptr_tbl_base: RX descriptor ring base address
1070 * @buf_tbl_base: Buffer table base address
1071 * @evq_ptr_tbl_base: Event queue pointer table base address
1072 * @evq_rptr_tbl_base: Event queue read-pointer table base address
1073 * @max_dma_mask: Maximum possible DMA mask
1074 * @rx_prefix_size: Size of RX prefix before packet data
1075 * @rx_hash_offset: Offset of RX flow hash within prefix
1076 * @rx_ts_offset: Offset of timestamp within prefix
1077 * @rx_buffer_padding: Size of padding at end of RX packet
1078 * @can_rx_scatter: NIC is able to scatter packets to multiple buffers
1079 * @always_rx_scatter: NIC will always scatter packets to multiple buffers
 
 
 
1080 * @max_interrupt_mode: Highest capability interrupt mode supported
1081 *	from &enum efx_init_mode.
1082 * @timer_period_max: Maximum period of interrupt timer (in ticks)
1083 * @offload_features: net_device feature flags for protocol offload
1084 *	features implemented in hardware
1085 * @mcdi_max_ver: Maximum MCDI version supported
1086 * @hwtstamp_filters: Mask of hardware timestamp filter types supported
1087 */
1088struct efx_nic_type {
 
 
1089	unsigned int (*mem_map_size)(struct efx_nic *efx);
1090	int (*probe)(struct efx_nic *efx);
1091	void (*remove)(struct efx_nic *efx);
1092	int (*init)(struct efx_nic *efx);
1093	int (*dimension_resources)(struct efx_nic *efx);
1094	void (*fini)(struct efx_nic *efx);
1095	void (*monitor)(struct efx_nic *efx);
1096	enum reset_type (*map_reset_reason)(enum reset_type reason);
1097	int (*map_reset_flags)(u32 *flags);
1098	int (*reset)(struct efx_nic *efx, enum reset_type method);
1099	int (*probe_port)(struct efx_nic *efx);
1100	void (*remove_port)(struct efx_nic *efx);
1101	bool (*handle_global_event)(struct efx_channel *channel, efx_qword_t *);
1102	int (*fini_dmaq)(struct efx_nic *efx);
1103	void (*prepare_flush)(struct efx_nic *efx);
1104	void (*finish_flush)(struct efx_nic *efx);
1105	void (*prepare_flr)(struct efx_nic *efx);
1106	void (*finish_flr)(struct efx_nic *efx);
1107	size_t (*describe_stats)(struct efx_nic *efx, u8 *names);
1108	size_t (*update_stats)(struct efx_nic *efx, u64 *full_stats,
1109			       struct rtnl_link_stats64 *core_stats);
1110	void (*start_stats)(struct efx_nic *efx);
1111	void (*pull_stats)(struct efx_nic *efx);
1112	void (*stop_stats)(struct efx_nic *efx);
1113	void (*set_id_led)(struct efx_nic *efx, enum efx_led_mode mode);
1114	void (*push_irq_moderation)(struct efx_channel *channel);
1115	int (*reconfigure_port)(struct efx_nic *efx);
1116	void (*prepare_enable_fc_tx)(struct efx_nic *efx);
1117	int (*reconfigure_mac)(struct efx_nic *efx);
1118	bool (*check_mac_fault)(struct efx_nic *efx);
1119	void (*get_wol)(struct efx_nic *efx, struct ethtool_wolinfo *wol);
1120	int (*set_wol)(struct efx_nic *efx, u32 type);
1121	void (*resume_wol)(struct efx_nic *efx);
1122	int (*test_chip)(struct efx_nic *efx, struct efx_self_tests *tests);
1123	int (*test_nvram)(struct efx_nic *efx);
1124	void (*mcdi_request)(struct efx_nic *efx,
1125			     const efx_dword_t *hdr, size_t hdr_len,
1126			     const efx_dword_t *sdu, size_t sdu_len);
1127	bool (*mcdi_poll_response)(struct efx_nic *efx);
1128	void (*mcdi_read_response)(struct efx_nic *efx, efx_dword_t *pdu,
1129				   size_t pdu_offset, size_t pdu_len);
1130	int (*mcdi_poll_reboot)(struct efx_nic *efx);
 
1131	void (*irq_enable_master)(struct efx_nic *efx);
1132	void (*irq_test_generate)(struct efx_nic *efx);
1133	void (*irq_disable_non_ev)(struct efx_nic *efx);
1134	irqreturn_t (*irq_handle_msi)(int irq, void *dev_id);
1135	irqreturn_t (*irq_handle_legacy)(int irq, void *dev_id);
1136	int (*tx_probe)(struct efx_tx_queue *tx_queue);
1137	void (*tx_init)(struct efx_tx_queue *tx_queue);
1138	void (*tx_remove)(struct efx_tx_queue *tx_queue);
1139	void (*tx_write)(struct efx_tx_queue *tx_queue);
1140	void (*rx_push_rss_config)(struct efx_nic *efx);
 
 
 
 
 
 
 
 
 
 
 
1141	int (*rx_probe)(struct efx_rx_queue *rx_queue);
1142	void (*rx_init)(struct efx_rx_queue *rx_queue);
1143	void (*rx_remove)(struct efx_rx_queue *rx_queue);
1144	void (*rx_write)(struct efx_rx_queue *rx_queue);
1145	void (*rx_defer_refill)(struct efx_rx_queue *rx_queue);
1146	int (*ev_probe)(struct efx_channel *channel);
1147	int (*ev_init)(struct efx_channel *channel);
1148	void (*ev_fini)(struct efx_channel *channel);
1149	void (*ev_remove)(struct efx_channel *channel);
1150	int (*ev_process)(struct efx_channel *channel, int quota);
1151	void (*ev_read_ack)(struct efx_channel *channel);
1152	void (*ev_test_generate)(struct efx_channel *channel);
1153	int (*filter_table_probe)(struct efx_nic *efx);
1154	void (*filter_table_restore)(struct efx_nic *efx);
1155	void (*filter_table_remove)(struct efx_nic *efx);
1156	void (*filter_update_rx_scatter)(struct efx_nic *efx);
1157	s32 (*filter_insert)(struct efx_nic *efx,
1158			     struct efx_filter_spec *spec, bool replace);
1159	int (*filter_remove_safe)(struct efx_nic *efx,
1160				  enum efx_filter_priority priority,
1161				  u32 filter_id);
1162	int (*filter_get_safe)(struct efx_nic *efx,
1163			       enum efx_filter_priority priority,
1164			       u32 filter_id, struct efx_filter_spec *);
1165	int (*filter_clear_rx)(struct efx_nic *efx,
1166			       enum efx_filter_priority priority);
1167	u32 (*filter_count_rx_used)(struct efx_nic *efx,
1168				    enum efx_filter_priority priority);
1169	u32 (*filter_get_rx_id_limit)(struct efx_nic *efx);
1170	s32 (*filter_get_rx_ids)(struct efx_nic *efx,
1171				 enum efx_filter_priority priority,
1172				 u32 *buf, u32 size);
1173#ifdef CONFIG_RFS_ACCEL
1174	s32 (*filter_rfs_insert)(struct efx_nic *efx,
1175				 struct efx_filter_spec *spec);
1176	bool (*filter_rfs_expire_one)(struct efx_nic *efx, u32 flow_id,
1177				      unsigned int index);
1178#endif
1179#ifdef CONFIG_SFC_MTD
1180	int (*mtd_probe)(struct efx_nic *efx);
1181	void (*mtd_rename)(struct efx_mtd_partition *part);
1182	int (*mtd_read)(struct mtd_info *mtd, loff_t start, size_t len,
1183			size_t *retlen, u8 *buffer);
1184	int (*mtd_erase)(struct mtd_info *mtd, loff_t start, size_t len);
1185	int (*mtd_write)(struct mtd_info *mtd, loff_t start, size_t len,
1186			 size_t *retlen, const u8 *buffer);
1187	int (*mtd_sync)(struct mtd_info *mtd);
1188#endif
1189	void (*ptp_write_host_time)(struct efx_nic *efx, u32 host_time);
1190	int (*ptp_set_ts_sync_events)(struct efx_nic *efx, bool en, bool temp);
1191	int (*ptp_set_ts_config)(struct efx_nic *efx,
1192				 struct hwtstamp_config *init);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1193
1194	int revision;
1195	unsigned int txd_ptr_tbl_base;
1196	unsigned int rxd_ptr_tbl_base;
1197	unsigned int buf_tbl_base;
1198	unsigned int evq_ptr_tbl_base;
1199	unsigned int evq_rptr_tbl_base;
1200	u64 max_dma_mask;
1201	unsigned int rx_prefix_size;
1202	unsigned int rx_hash_offset;
1203	unsigned int rx_ts_offset;
1204	unsigned int rx_buffer_padding;
1205	bool can_rx_scatter;
1206	bool always_rx_scatter;
 
 
1207	unsigned int max_interrupt_mode;
1208	unsigned int timer_period_max;
1209	netdev_features_t offload_features;
1210	int mcdi_max_ver;
1211	unsigned int max_rx_ip_filters;
1212	u32 hwtstamp_filters;
 
1213};
1214
1215/**************************************************************************
1216 *
1217 * Prototypes and inline functions
1218 *
1219 *************************************************************************/
1220
1221static inline struct efx_channel *
1222efx_get_channel(struct efx_nic *efx, unsigned index)
1223{
1224	EFX_BUG_ON_PARANOID(index >= efx->n_channels);
1225	return efx->channel[index];
1226}
1227
1228/* Iterate over all used channels */
1229#define efx_for_each_channel(_channel, _efx)				\
1230	for (_channel = (_efx)->channel[0];				\
1231	     _channel;							\
1232	     _channel = (_channel->channel + 1 < (_efx)->n_channels) ?	\
1233		     (_efx)->channel[_channel->channel + 1] : NULL)
1234
1235/* Iterate over all used channels in reverse */
1236#define efx_for_each_channel_rev(_channel, _efx)			\
1237	for (_channel = (_efx)->channel[(_efx)->n_channels - 1];	\
1238	     _channel;							\
1239	     _channel = _channel->channel ?				\
1240		     (_efx)->channel[_channel->channel - 1] : NULL)
1241
1242static inline struct efx_tx_queue *
1243efx_get_tx_queue(struct efx_nic *efx, unsigned index, unsigned type)
1244{
1245	EFX_BUG_ON_PARANOID(index >= efx->n_tx_channels ||
1246			    type >= EFX_TXQ_TYPES);
1247	return &efx->channel[efx->tx_channel_offset + index]->tx_queue[type];
1248}
1249
1250static inline bool efx_channel_has_tx_queues(struct efx_channel *channel)
1251{
1252	return channel->channel - channel->efx->tx_channel_offset <
1253		channel->efx->n_tx_channels;
1254}
1255
1256static inline struct efx_tx_queue *
1257efx_channel_get_tx_queue(struct efx_channel *channel, unsigned type)
1258{
1259	EFX_BUG_ON_PARANOID(!efx_channel_has_tx_queues(channel) ||
1260			    type >= EFX_TXQ_TYPES);
1261	return &channel->tx_queue[type];
1262}
1263
1264static inline bool efx_tx_queue_used(struct efx_tx_queue *tx_queue)
1265{
1266	return !(tx_queue->efx->net_dev->num_tc < 2 &&
1267		 tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI);
1268}
1269
1270/* Iterate over all TX queues belonging to a channel */
1271#define efx_for_each_channel_tx_queue(_tx_queue, _channel)		\
1272	if (!efx_channel_has_tx_queues(_channel))			\
1273		;							\
1274	else								\
1275		for (_tx_queue = (_channel)->tx_queue;			\
1276		     _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES && \
1277			     efx_tx_queue_used(_tx_queue);		\
1278		     _tx_queue++)
1279
1280/* Iterate over all possible TX queues belonging to a channel */
1281#define efx_for_each_possible_channel_tx_queue(_tx_queue, _channel)	\
1282	if (!efx_channel_has_tx_queues(_channel))			\
1283		;							\
1284	else								\
1285		for (_tx_queue = (_channel)->tx_queue;			\
1286		     _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES;	\
1287		     _tx_queue++)
1288
1289static inline bool efx_channel_has_rx_queue(struct efx_channel *channel)
1290{
1291	return channel->rx_queue.core_index >= 0;
1292}
1293
1294static inline struct efx_rx_queue *
1295efx_channel_get_rx_queue(struct efx_channel *channel)
1296{
1297	EFX_BUG_ON_PARANOID(!efx_channel_has_rx_queue(channel));
1298	return &channel->rx_queue;
1299}
1300
1301/* Iterate over all RX queues belonging to a channel */
1302#define efx_for_each_channel_rx_queue(_rx_queue, _channel)		\
1303	if (!efx_channel_has_rx_queue(_channel))			\
1304		;							\
1305	else								\
1306		for (_rx_queue = &(_channel)->rx_queue;			\
1307		     _rx_queue;						\
1308		     _rx_queue = NULL)
1309
1310static inline struct efx_channel *
1311efx_rx_queue_channel(struct efx_rx_queue *rx_queue)
1312{
1313	return container_of(rx_queue, struct efx_channel, rx_queue);
1314}
1315
1316static inline int efx_rx_queue_index(struct efx_rx_queue *rx_queue)
1317{
1318	return efx_rx_queue_channel(rx_queue)->channel;
1319}
1320
1321/* Returns a pointer to the specified receive buffer in the RX
1322 * descriptor queue.
1323 */
1324static inline struct efx_rx_buffer *efx_rx_buffer(struct efx_rx_queue *rx_queue,
1325						  unsigned int index)
1326{
1327	return &rx_queue->buffer[index];
1328}
1329
1330/**
1331 * EFX_MAX_FRAME_LEN - calculate maximum frame length
1332 *
1333 * This calculates the maximum frame length that will be used for a
1334 * given MTU.  The frame length will be equal to the MTU plus a
1335 * constant amount of header space and padding.  This is the quantity
1336 * that the net driver will program into the MAC as the maximum frame
1337 * length.
1338 *
1339 * The 10G MAC requires 8-byte alignment on the frame
1340 * length, so we round up to the nearest 8.
1341 *
1342 * Re-clocking by the XGXS on RX can reduce an IPG to 32 bits (half an
1343 * XGMII cycle).  If the frame length reaches the maximum value in the
1344 * same cycle, the XMAC can miss the IPG altogether.  We work around
1345 * this by adding a further 16 bytes.
1346 */
 
1347#define EFX_MAX_FRAME_LEN(mtu) \
1348	((((mtu) + ETH_HLEN + VLAN_HLEN + 4/* FCS */ + 7) & ~7) + 16)
1349
1350static inline bool efx_xmit_with_hwtstamp(struct sk_buff *skb)
1351{
1352	return skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP;
1353}
1354static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
1355{
1356	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1357}
1358
1359#endif /* EFX_NET_DRIVER_H */
v5.4
   1/* SPDX-License-Identifier: GPL-2.0-only */
   2/****************************************************************************
   3 * Driver for Solarflare network controllers and boards
   4 * Copyright 2005-2006 Fen Systems Ltd.
   5 * Copyright 2005-2013 Solarflare Communications Inc.
 
 
 
 
   6 */
   7
   8/* Common definitions for all Efx net driver code */
   9
  10#ifndef EFX_NET_DRIVER_H
  11#define EFX_NET_DRIVER_H
  12
  13#include <linux/netdevice.h>
  14#include <linux/etherdevice.h>
  15#include <linux/ethtool.h>
  16#include <linux/if_vlan.h>
  17#include <linux/timer.h>
  18#include <linux/mdio.h>
  19#include <linux/list.h>
  20#include <linux/pci.h>
  21#include <linux/device.h>
  22#include <linux/highmem.h>
  23#include <linux/workqueue.h>
  24#include <linux/mutex.h>
  25#include <linux/rwsem.h>
  26#include <linux/vmalloc.h>
  27#include <linux/i2c.h>
  28#include <linux/mtd/mtd.h>
  29#include <net/busy_poll.h>
  30
  31#include "enum.h"
  32#include "bitfield.h"
  33#include "filter.h"
  34
  35/**************************************************************************
  36 *
  37 * Build definitions
  38 *
  39 **************************************************************************/
  40
  41#define EFX_DRIVER_VERSION	"4.1"
  42
  43#ifdef DEBUG
  44#define EFX_WARN_ON_ONCE_PARANOID(x) WARN_ON_ONCE(x)
  45#define EFX_WARN_ON_PARANOID(x) WARN_ON(x)
  46#else
  47#define EFX_WARN_ON_ONCE_PARANOID(x) do {} while (0)
  48#define EFX_WARN_ON_PARANOID(x) do {} while (0)
  49#endif
  50
  51/**************************************************************************
  52 *
  53 * Efx data structures
  54 *
  55 **************************************************************************/
  56
  57#define EFX_MAX_CHANNELS 32U
  58#define EFX_MAX_RX_QUEUES EFX_MAX_CHANNELS
  59#define EFX_EXTRA_CHANNEL_IOV	0
  60#define EFX_EXTRA_CHANNEL_PTP	1
  61#define EFX_MAX_EXTRA_CHANNELS	2U
  62
  63/* Checksum generation is a per-queue option in hardware, so each
  64 * queue visible to the networking core is backed by two hardware TX
  65 * queues. */
  66#define EFX_MAX_TX_TC		2
  67#define EFX_MAX_CORE_TX_QUEUES	(EFX_MAX_TX_TC * EFX_MAX_CHANNELS)
  68#define EFX_TXQ_TYPE_OFFLOAD	1	/* flag */
  69#define EFX_TXQ_TYPE_HIGHPRI	2	/* flag */
  70#define EFX_TXQ_TYPES		4
  71#define EFX_MAX_TX_QUEUES	(EFX_TXQ_TYPES * EFX_MAX_CHANNELS)
  72
  73/* Maximum possible MTU the driver supports */
  74#define EFX_MAX_MTU (9 * 1024)
  75
  76/* Minimum MTU, from RFC791 (IP) */
  77#define EFX_MIN_MTU 68
  78
  79/* Size of an RX scatter buffer.  Small enough to pack 2 into a 4K page,
  80 * and should be a multiple of the cache line size.
  81 */
  82#define EFX_RX_USR_BUF_SIZE	(2048 - 256)
  83
  84/* If possible, we should ensure cache line alignment at start and end
  85 * of every buffer.  Otherwise, we just need to ensure 4-byte
  86 * alignment of the network header.
  87 */
  88#if NET_IP_ALIGN == 0
  89#define EFX_RX_BUF_ALIGNMENT	L1_CACHE_BYTES
  90#else
  91#define EFX_RX_BUF_ALIGNMENT	4
  92#endif
  93
  94/* Forward declare Precision Time Protocol (PTP) support structure. */
  95struct efx_ptp_data;
  96struct hwtstamp_config;
  97
  98struct efx_self_tests;
  99
 100/**
 101 * struct efx_buffer - A general-purpose DMA buffer
 102 * @addr: host base address of the buffer
 103 * @dma_addr: DMA base address of the buffer
 104 * @len: Buffer length, in bytes
 105 *
 106 * The NIC uses these buffers for its interrupt status registers and
 107 * MAC stats dumps.
 108 */
 109struct efx_buffer {
 110	void *addr;
 111	dma_addr_t dma_addr;
 112	unsigned int len;
 113};
 114
 115/**
 116 * struct efx_special_buffer - DMA buffer entered into buffer table
 117 * @buf: Standard &struct efx_buffer
 118 * @index: Buffer index within controller;s buffer table
 119 * @entries: Number of buffer table entries
 120 *
 121 * The NIC has a buffer table that maps buffers of size %EFX_BUF_SIZE.
 122 * Event and descriptor rings are addressed via one or more buffer
 123 * table entries (and so can be physically non-contiguous, although we
 124 * currently do not take advantage of that).  On Falcon and Siena we
 125 * have to take care of allocating and initialising the entries
 126 * ourselves.  On later hardware this is managed by the firmware and
 127 * @index and @entries are left as 0.
 128 */
 129struct efx_special_buffer {
 130	struct efx_buffer buf;
 131	unsigned int index;
 132	unsigned int entries;
 133};
 134
 135/**
 136 * struct efx_tx_buffer - buffer state for a TX descriptor
 137 * @skb: When @flags & %EFX_TX_BUF_SKB, the associated socket buffer to be
 138 *	freed when descriptor completes
 
 
 139 * @option: When @flags & %EFX_TX_BUF_OPTION, a NIC-specific option descriptor.
 140 * @dma_addr: DMA address of the fragment.
 141 * @flags: Flags for allocation and DMA mapping type
 142 * @len: Length of this fragment.
 143 *	This field is zero when the queue slot is empty.
 144 * @unmap_len: Length of this fragment to unmap
 145 * @dma_offset: Offset of @dma_addr from the address of the backing DMA mapping.
 146 * Only valid if @unmap_len != 0.
 147 */
 148struct efx_tx_buffer {
 149	const struct sk_buff *skb;
 
 
 
 150	union {
 151		efx_qword_t option;
 152		dma_addr_t dma_addr;
 153	};
 154	unsigned short flags;
 155	unsigned short len;
 156	unsigned short unmap_len;
 157	unsigned short dma_offset;
 158};
 159#define EFX_TX_BUF_CONT		1	/* not last descriptor of packet */
 160#define EFX_TX_BUF_SKB		2	/* buffer is last part of skb */
 
 161#define EFX_TX_BUF_MAP_SINGLE	8	/* buffer was mapped with dma_map_single() */
 162#define EFX_TX_BUF_OPTION	0x10	/* empty buffer for option descriptor */
 163
 164/**
 165 * struct efx_tx_queue - An Efx TX queue
 166 *
 167 * This is a ring buffer of TX fragments.
 168 * Since the TX completion path always executes on the same
 169 * CPU and the xmit path can operate on different CPUs,
 170 * performance is increased by ensuring that the completion
 171 * path and the xmit path operate on different cache lines.
 172 * This is particularly important if the xmit path is always
 173 * executing on one CPU which is different from the completion
 174 * path.  There is also a cache line for members which are
 175 * read but not written on the fast path.
 176 *
 177 * @efx: The associated Efx NIC
 178 * @queue: DMA queue number
 179 * @tso_version: Version of TSO in use for this queue.
 180 * @channel: The associated channel
 181 * @core_txq: The networking core TX queue structure
 182 * @buffer: The software buffer ring
 183 * @cb_page: Array of pages of copy buffers.  Carved up according to
 184 *	%EFX_TX_CB_ORDER into %EFX_TX_CB_SIZE-sized chunks.
 185 * @txd: The hardware descriptor ring
 186 * @ptr_mask: The size of the ring minus 1.
 187 * @piobuf: PIO buffer region for this TX queue (shared with its partner).
 188 *	Size of the region is efx_piobuf_size.
 189 * @piobuf_offset: Buffer offset to be specified in PIO descriptors
 190 * @initialised: Has hardware queue been initialised?
 191 * @timestamping: Is timestamping enabled for this channel?
 192 * @handle_tso: TSO xmit preparation handler.  Sets up the TSO metadata and
 193 *	may also map tx data, depending on the nature of the TSO implementation.
 194 * @read_count: Current read pointer.
 195 *	This is the number of buffers that have been removed from both rings.
 196 * @old_write_count: The value of @write_count when last checked.
 197 *	This is here for performance reasons.  The xmit path will
 198 *	only get the up-to-date value of @write_count if this
 199 *	variable indicates that the queue is empty.  This is to
 200 *	avoid cache-line ping-pong between the xmit path and the
 201 *	completion path.
 202 * @merge_events: Number of TX merged completion events
 203 * @completed_desc_ptr: Most recent completed pointer - only used with
 204 *      timestamping.
 205 * @completed_timestamp_major: Top part of the most recent tx timestamp.
 206 * @completed_timestamp_minor: Low part of the most recent tx timestamp.
 207 * @insert_count: Current insert pointer
 208 *	This is the number of buffers that have been added to the
 209 *	software ring.
 210 * @write_count: Current write pointer
 211 *	This is the number of buffers that have been added to the
 212 *	hardware ring.
 213 * @packet_write_count: Completable write pointer
 214 *	This is the write pointer of the last packet written.
 215 *	Normally this will equal @write_count, but as option descriptors
 216 *	don't produce completion events, they won't update this.
 217 *	Filled in iff @efx->type->option_descriptors; only used for PIO.
 218 *	Thus, this is written and used on EF10, and neither on farch.
 219 * @old_read_count: The value of read_count when last checked.
 220 *	This is here for performance reasons.  The xmit path will
 221 *	only get the up-to-date value of read_count if this
 222 *	variable indicates that the queue is full.  This is to
 223 *	avoid cache-line ping-pong between the xmit path and the
 224 *	completion path.
 225 * @tso_bursts: Number of times TSO xmit invoked by kernel
 226 * @tso_long_headers: Number of packets with headers too long for standard
 227 *	blocks
 228 * @tso_packets: Number of packets via the TSO xmit path
 229 * @tso_fallbacks: Number of times TSO fallback used
 230 * @pushes: Number of times the TX push feature has been used
 231 * @pio_packets: Number of times the TX PIO feature has been used
 232 * @xmit_more_available: Are any packets waiting to be pushed to the NIC
 233 * @cb_packets: Number of times the TX copybreak feature has been used
 234 * @empty_read_count: If the completion path has seen the queue as empty
 235 *	and the transmission path has not yet checked this, the value of
 236 *	@read_count bitwise-added to %EFX_EMPTY_COUNT_VALID; otherwise 0.
 237 */
 238struct efx_tx_queue {
 239	/* Members which don't change on the fast path */
 240	struct efx_nic *efx ____cacheline_aligned_in_smp;
 241	unsigned queue;
 242	unsigned int tso_version;
 243	struct efx_channel *channel;
 244	struct netdev_queue *core_txq;
 245	struct efx_tx_buffer *buffer;
 246	struct efx_buffer *cb_page;
 247	struct efx_special_buffer txd;
 248	unsigned int ptr_mask;
 249	void __iomem *piobuf;
 250	unsigned int piobuf_offset;
 251	bool initialised;
 252	bool timestamping;
 253
 254	/* Function pointers used in the fast path. */
 255	int (*handle_tso)(struct efx_tx_queue*, struct sk_buff*, bool *);
 256
 257	/* Members used mainly on the completion path */
 258	unsigned int read_count ____cacheline_aligned_in_smp;
 259	unsigned int old_write_count;
 260	unsigned int merge_events;
 261	unsigned int bytes_compl;
 262	unsigned int pkts_compl;
 263	unsigned int completed_desc_ptr;
 264	u32 completed_timestamp_major;
 265	u32 completed_timestamp_minor;
 266
 267	/* Members used only on the xmit path */
 268	unsigned int insert_count ____cacheline_aligned_in_smp;
 269	unsigned int write_count;
 270	unsigned int packet_write_count;
 271	unsigned int old_read_count;
 272	unsigned int tso_bursts;
 273	unsigned int tso_long_headers;
 274	unsigned int tso_packets;
 275	unsigned int tso_fallbacks;
 276	unsigned int pushes;
 277	unsigned int pio_packets;
 278	bool xmit_more_available;
 279	unsigned int cb_packets;
 280	/* Statistics to supplement MAC stats */
 281	unsigned long tx_packets;
 282
 283	/* Members shared between paths and sometimes updated */
 284	unsigned int empty_read_count ____cacheline_aligned_in_smp;
 285#define EFX_EMPTY_COUNT_VALID 0x80000000
 286	atomic_t flush_outstanding;
 287};
 288
 289#define EFX_TX_CB_ORDER	7
 290#define EFX_TX_CB_SIZE	(1 << EFX_TX_CB_ORDER) - NET_IP_ALIGN
 291
 292/**
 293 * struct efx_rx_buffer - An Efx RX data buffer
 294 * @dma_addr: DMA base address of the buffer
 295 * @page: The associated page buffer.
 296 *	Will be %NULL if the buffer slot is currently free.
 297 * @page_offset: If pending: offset in @page of DMA base address.
 298 *	If completed: offset in @page of Ethernet header.
 299 * @len: If pending: length for DMA descriptor.
 300 *	If completed: received length, excluding hash prefix.
 301 * @flags: Flags for buffer and packet state.  These are only set on the
 302 *	first buffer of a scattered packet.
 303 */
 304struct efx_rx_buffer {
 305	dma_addr_t dma_addr;
 306	struct page *page;
 307	u16 page_offset;
 308	u16 len;
 309	u16 flags;
 310};
 311#define EFX_RX_BUF_LAST_IN_PAGE	0x0001
 312#define EFX_RX_PKT_CSUMMED	0x0002
 313#define EFX_RX_PKT_DISCARD	0x0004
 314#define EFX_RX_PKT_TCP		0x0040
 315#define EFX_RX_PKT_PREFIX_LEN	0x0080	/* length is in prefix only */
 316#define EFX_RX_PKT_CSUM_LEVEL	0x0200
 317
 318/**
 319 * struct efx_rx_page_state - Page-based rx buffer state
 320 *
 321 * Inserted at the start of every page allocated for receive buffers.
 322 * Used to facilitate sharing dma mappings between recycled rx buffers
 323 * and those passed up to the kernel.
 324 *
 325 * @dma_addr: The dma address of this page.
 326 */
 327struct efx_rx_page_state {
 328	dma_addr_t dma_addr;
 329
 330	unsigned int __pad[0] ____cacheline_aligned;
 331};
 332
 333/**
 334 * struct efx_rx_queue - An Efx RX queue
 335 * @efx: The associated Efx NIC
 336 * @core_index:  Index of network core RX queue.  Will be >= 0 iff this
 337 *	is associated with a real RX queue.
 338 * @buffer: The software buffer ring
 339 * @rxd: The hardware descriptor ring
 340 * @ptr_mask: The size of the ring minus 1.
 341 * @refill_enabled: Enable refill whenever fill level is low
 342 * @flush_pending: Set when a RX flush is pending. Has the same lifetime as
 343 *	@rxq_flush_pending.
 344 * @added_count: Number of buffers added to the receive queue.
 345 * @notified_count: Number of buffers given to NIC (<= @added_count).
 346 * @removed_count: Number of buffers removed from the receive queue.
 347 * @scatter_n: Used by NIC specific receive code.
 348 * @scatter_len: Used by NIC specific receive code.
 349 * @page_ring: The ring to store DMA mapped pages for reuse.
 350 * @page_add: Counter to calculate the write pointer for the recycle ring.
 351 * @page_remove: Counter to calculate the read pointer for the recycle ring.
 352 * @page_recycle_count: The number of pages that have been recycled.
 353 * @page_recycle_failed: The number of pages that couldn't be recycled because
 354 *      the kernel still held a reference to them.
 355 * @page_recycle_full: The number of pages that were released because the
 356 *      recycle ring was full.
 357 * @page_ptr_mask: The number of pages in the RX recycle ring minus 1.
 358 * @max_fill: RX descriptor maximum fill level (<= ring size)
 359 * @fast_fill_trigger: RX descriptor fill level that will trigger a fast fill
 360 *	(<= @max_fill)
 361 * @min_fill: RX descriptor minimum non-zero fill level.
 362 *	This records the minimum fill level observed when a ring
 363 *	refill was triggered.
 364 * @recycle_count: RX buffer recycle counter.
 365 * @slow_fill: Timer used to defer efx_nic_generate_fill_event().
 366 */
 367struct efx_rx_queue {
 368	struct efx_nic *efx;
 369	int core_index;
 370	struct efx_rx_buffer *buffer;
 371	struct efx_special_buffer rxd;
 372	unsigned int ptr_mask;
 373	bool refill_enabled;
 374	bool flush_pending;
 375
 376	unsigned int added_count;
 377	unsigned int notified_count;
 378	unsigned int removed_count;
 379	unsigned int scatter_n;
 380	unsigned int scatter_len;
 381	struct page **page_ring;
 382	unsigned int page_add;
 383	unsigned int page_remove;
 384	unsigned int page_recycle_count;
 385	unsigned int page_recycle_failed;
 386	unsigned int page_recycle_full;
 387	unsigned int page_ptr_mask;
 388	unsigned int max_fill;
 389	unsigned int fast_fill_trigger;
 390	unsigned int min_fill;
 391	unsigned int min_overfill;
 392	unsigned int recycle_count;
 393	struct timer_list slow_fill;
 394	unsigned int slow_fill_count;
 395	/* Statistics to supplement MAC stats */
 396	unsigned long rx_packets;
 397};
 398
 399enum efx_sync_events_state {
 400	SYNC_EVENTS_DISABLED = 0,
 401	SYNC_EVENTS_QUIESCENT,
 402	SYNC_EVENTS_REQUESTED,
 403	SYNC_EVENTS_VALID,
 404};
 405
 406/**
 407 * struct efx_channel - An Efx channel
 408 *
 409 * A channel comprises an event queue, at least one TX queue, at least
 410 * one RX queue, and an associated tasklet for processing the event
 411 * queue.
 412 *
 413 * @efx: Associated Efx NIC
 414 * @channel: Channel instance number
 415 * @type: Channel type definition
 416 * @eventq_init: Event queue initialised flag
 417 * @enabled: Channel enabled indicator
 418 * @irq: IRQ number (MSI and MSI-X only)
 419 * @irq_moderation_us: IRQ moderation value (in microseconds)
 420 * @napi_dev: Net device used with NAPI
 421 * @napi_str: NAPI control structure
 422 * @state: state for NAPI vs busy polling
 423 * @state_lock: lock protecting @state
 424 * @eventq: Event queue buffer
 425 * @eventq_mask: Event queue pointer mask
 426 * @eventq_read_ptr: Event queue read pointer
 427 * @event_test_cpu: Last CPU to handle interrupt or test event for this channel
 428 * @irq_count: Number of IRQs since last adaptive moderation decision
 429 * @irq_mod_score: IRQ moderation score
 430 * @filter_work: Work item for efx_filter_rfs_expire()
 431 * @rps_flow_id: Flow IDs of filters allocated for accelerated RFS,
 432 *      indexed by filter ID
 433 * @n_rx_tobe_disc: Count of RX_TOBE_DISC errors
 434 * @n_rx_ip_hdr_chksum_err: Count of RX IP header checksum errors
 435 * @n_rx_tcp_udp_chksum_err: Count of RX TCP and UDP checksum errors
 436 * @n_rx_mcast_mismatch: Count of unmatched multicast frames
 437 * @n_rx_frm_trunc: Count of RX_FRM_TRUNC errors
 438 * @n_rx_overlength: Count of RX_OVERLENGTH errors
 439 * @n_skbuff_leaks: Count of skbuffs leaked due to RX overrun
 440 * @n_rx_nodesc_trunc: Number of RX packets truncated and then dropped due to
 441 *	lack of descriptors
 442 * @n_rx_merge_events: Number of RX merged completion events
 443 * @n_rx_merge_packets: Number of RX packets completed by merged events
 444 * @rx_pkt_n_frags: Number of fragments in next packet to be delivered by
 445 *	__efx_rx_packet(), or zero if there is none
 446 * @rx_pkt_index: Ring index of first buffer for next packet to be delivered
 447 *	by __efx_rx_packet(), if @rx_pkt_n_frags != 0
 448 * @rx_list: list of SKBs from current RX, awaiting processing
 449 * @rx_queue: RX queue for this channel
 450 * @tx_queue: TX queues for this channel
 451 * @sync_events_state: Current state of sync events on this channel
 452 * @sync_timestamp_major: Major part of the last ptp sync event
 453 * @sync_timestamp_minor: Minor part of the last ptp sync event
 454 */
 455struct efx_channel {
 456	struct efx_nic *efx;
 457	int channel;
 458	const struct efx_channel_type *type;
 459	bool eventq_init;
 460	bool enabled;
 461	int irq;
 462	unsigned int irq_moderation_us;
 463	struct net_device *napi_dev;
 464	struct napi_struct napi_str;
 465#ifdef CONFIG_NET_RX_BUSY_POLL
 466	unsigned long busy_poll_state;
 467#endif
 468	struct efx_special_buffer eventq;
 469	unsigned int eventq_mask;
 470	unsigned int eventq_read_ptr;
 471	int event_test_cpu;
 472
 473	unsigned int irq_count;
 474	unsigned int irq_mod_score;
 475#ifdef CONFIG_RFS_ACCEL
 476	unsigned int rfs_filters_added;
 477	struct work_struct filter_work;
 478#define RPS_FLOW_ID_INVALID 0xFFFFFFFF
 479	u32 *rps_flow_id;
 480#endif
 481
 482	unsigned int n_rx_tobe_disc;
 483	unsigned int n_rx_ip_hdr_chksum_err;
 484	unsigned int n_rx_tcp_udp_chksum_err;
 485	unsigned int n_rx_outer_ip_hdr_chksum_err;
 486	unsigned int n_rx_outer_tcp_udp_chksum_err;
 487	unsigned int n_rx_inner_ip_hdr_chksum_err;
 488	unsigned int n_rx_inner_tcp_udp_chksum_err;
 489	unsigned int n_rx_eth_crc_err;
 490	unsigned int n_rx_mcast_mismatch;
 491	unsigned int n_rx_frm_trunc;
 492	unsigned int n_rx_overlength;
 493	unsigned int n_skbuff_leaks;
 494	unsigned int n_rx_nodesc_trunc;
 495	unsigned int n_rx_merge_events;
 496	unsigned int n_rx_merge_packets;
 497
 498	unsigned int rx_pkt_n_frags;
 499	unsigned int rx_pkt_index;
 500
 501	struct list_head *rx_list;
 502
 503	struct efx_rx_queue rx_queue;
 504	struct efx_tx_queue tx_queue[EFX_TXQ_TYPES];
 505
 506	enum efx_sync_events_state sync_events_state;
 507	u32 sync_timestamp_major;
 508	u32 sync_timestamp_minor;
 509};
 510
 511/**
 512 * struct efx_msi_context - Context for each MSI
 513 * @efx: The associated NIC
 514 * @index: Index of the channel/IRQ
 515 * @name: Name of the channel/IRQ
 516 *
 517 * Unlike &struct efx_channel, this is never reallocated and is always
 518 * safe for the IRQ handler to access.
 519 */
 520struct efx_msi_context {
 521	struct efx_nic *efx;
 522	unsigned int index;
 523	char name[IFNAMSIZ + 6];
 524};
 525
 526/**
 527 * struct efx_channel_type - distinguishes traffic and extra channels
 528 * @handle_no_channel: Handle failure to allocate an extra channel
 529 * @pre_probe: Set up extra state prior to initialisation
 530 * @post_remove: Tear down extra state after finalisation, if allocated.
 531 *	May be called on channels that have not been probed.
 532 * @get_name: Generate the channel's name (used for its IRQ handler)
 533 * @copy: Copy the channel state prior to reallocation.  May be %NULL if
 534 *	reallocation is not supported.
 535 * @receive_skb: Handle an skb ready to be passed to netif_receive_skb()
 536 * @want_txqs: Determine whether this channel should have TX queues
 537 *	created.  If %NULL, TX queues are not created.
 538 * @keep_eventq: Flag for whether event queue should be kept initialised
 539 *	while the device is stopped
 540 * @want_pio: Flag for whether PIO buffers should be linked to this
 541 *	channel's TX queues.
 542 */
 543struct efx_channel_type {
 544	void (*handle_no_channel)(struct efx_nic *);
 545	int (*pre_probe)(struct efx_channel *);
 546	void (*post_remove)(struct efx_channel *);
 547	void (*get_name)(struct efx_channel *, char *buf, size_t len);
 548	struct efx_channel *(*copy)(const struct efx_channel *);
 549	bool (*receive_skb)(struct efx_channel *, struct sk_buff *);
 550	bool (*want_txqs)(struct efx_channel *);
 551	bool keep_eventq;
 552	bool want_pio;
 553};
 554
 555enum efx_led_mode {
 556	EFX_LED_OFF	= 0,
 557	EFX_LED_ON	= 1,
 558	EFX_LED_DEFAULT	= 2
 559};
 560
 561#define STRING_TABLE_LOOKUP(val, member) \
 562	((val) < member ## _max) ? member ## _names[val] : "(invalid)"
 563
 564extern const char *const efx_loopback_mode_names[];
 565extern const unsigned int efx_loopback_mode_max;
 566#define LOOPBACK_MODE(efx) \
 567	STRING_TABLE_LOOKUP((efx)->loopback_mode, efx_loopback_mode)
 568
 569extern const char *const efx_reset_type_names[];
 570extern const unsigned int efx_reset_type_max;
 571#define RESET_TYPE(type) \
 572	STRING_TABLE_LOOKUP(type, efx_reset_type)
 573
 574void efx_get_udp_tunnel_type_name(u16 type, char *buf, size_t buflen);
 575
 576enum efx_int_mode {
 577	/* Be careful if altering to correct macro below */
 578	EFX_INT_MODE_MSIX = 0,
 579	EFX_INT_MODE_MSI = 1,
 580	EFX_INT_MODE_LEGACY = 2,
 581	EFX_INT_MODE_MAX	/* Insert any new items before this */
 582};
 583#define EFX_INT_MODE_USE_MSI(x) (((x)->interrupt_mode) <= EFX_INT_MODE_MSI)
 584
 585enum nic_state {
 586	STATE_UNINIT = 0,	/* device being probed/removed or is frozen */
 587	STATE_READY = 1,	/* hardware ready and netdev registered */
 588	STATE_DISABLED = 2,	/* device disabled due to hardware errors */
 589	STATE_RECOVERY = 3,	/* device recovering from PCI error */
 590};
 591
 592/* Forward declaration */
 593struct efx_nic;
 594
 595/* Pseudo bit-mask flow control field */
 596#define EFX_FC_RX	FLOW_CTRL_RX
 597#define EFX_FC_TX	FLOW_CTRL_TX
 598#define EFX_FC_AUTO	4
 599
 600/**
 601 * struct efx_link_state - Current state of the link
 602 * @up: Link is up
 603 * @fd: Link is full-duplex
 604 * @fc: Actual flow control flags
 605 * @speed: Link speed (Mbps)
 606 */
 607struct efx_link_state {
 608	bool up;
 609	bool fd;
 610	u8 fc;
 611	unsigned int speed;
 612};
 613
 614static inline bool efx_link_state_equal(const struct efx_link_state *left,
 615					const struct efx_link_state *right)
 616{
 617	return left->up == right->up && left->fd == right->fd &&
 618		left->fc == right->fc && left->speed == right->speed;
 619}
 620
 621/**
 622 * struct efx_phy_operations - Efx PHY operations table
 623 * @probe: Probe PHY and initialise efx->mdio.mode_support, efx->mdio.mmds,
 624 *	efx->loopback_modes.
 625 * @init: Initialise PHY
 626 * @fini: Shut down PHY
 627 * @reconfigure: Reconfigure PHY (e.g. for new link parameters)
 628 * @poll: Update @link_state and report whether it changed.
 629 *	Serialised by the mac_lock.
 630 * @get_link_ksettings: Get ethtool settings. Serialised by the mac_lock.
 631 * @set_link_ksettings: Set ethtool settings. Serialised by the mac_lock.
 632 * @get_fecparam: Get Forward Error Correction settings. Serialised by mac_lock.
 633 * @set_fecparam: Set Forward Error Correction settings. Serialised by mac_lock.
 634 * @set_npage_adv: Set abilities advertised in (Extended) Next Page
 635 *	(only needed where AN bit is set in mmds)
 636 * @test_alive: Test that PHY is 'alive' (online)
 637 * @test_name: Get the name of a PHY-specific test/result
 638 * @run_tests: Run tests and record results as appropriate (offline).
 639 *	Flags are the ethtool tests flags.
 640 */
 641struct efx_phy_operations {
 642	int (*probe) (struct efx_nic *efx);
 643	int (*init) (struct efx_nic *efx);
 644	void (*fini) (struct efx_nic *efx);
 645	void (*remove) (struct efx_nic *efx);
 646	int (*reconfigure) (struct efx_nic *efx);
 647	bool (*poll) (struct efx_nic *efx);
 648	void (*get_link_ksettings)(struct efx_nic *efx,
 649				   struct ethtool_link_ksettings *cmd);
 650	int (*set_link_ksettings)(struct efx_nic *efx,
 651				  const struct ethtool_link_ksettings *cmd);
 652	int (*get_fecparam)(struct efx_nic *efx, struct ethtool_fecparam *fec);
 653	int (*set_fecparam)(struct efx_nic *efx,
 654			    const struct ethtool_fecparam *fec);
 655	void (*set_npage_adv) (struct efx_nic *efx, u32);
 656	int (*test_alive) (struct efx_nic *efx);
 657	const char *(*test_name) (struct efx_nic *efx, unsigned int index);
 658	int (*run_tests) (struct efx_nic *efx, int *results, unsigned flags);
 659	int (*get_module_eeprom) (struct efx_nic *efx,
 660			       struct ethtool_eeprom *ee,
 661			       u8 *data);
 662	int (*get_module_info) (struct efx_nic *efx,
 663				struct ethtool_modinfo *modinfo);
 664};
 665
 666/**
 667 * enum efx_phy_mode - PHY operating mode flags
 668 * @PHY_MODE_NORMAL: on and should pass traffic
 669 * @PHY_MODE_TX_DISABLED: on with TX disabled
 670 * @PHY_MODE_LOW_POWER: set to low power through MDIO
 671 * @PHY_MODE_OFF: switched off through external control
 672 * @PHY_MODE_SPECIAL: on but will not pass traffic
 673 */
 674enum efx_phy_mode {
 675	PHY_MODE_NORMAL		= 0,
 676	PHY_MODE_TX_DISABLED	= 1,
 677	PHY_MODE_LOW_POWER	= 2,
 678	PHY_MODE_OFF		= 4,
 679	PHY_MODE_SPECIAL	= 8,
 680};
 681
 682static inline bool efx_phy_mode_disabled(enum efx_phy_mode mode)
 683{
 684	return !!(mode & ~PHY_MODE_TX_DISABLED);
 685}
 686
 687/**
 688 * struct efx_hw_stat_desc - Description of a hardware statistic
 689 * @name: Name of the statistic as visible through ethtool, or %NULL if
 690 *	it should not be exposed
 691 * @dma_width: Width in bits (0 for non-DMA statistics)
 692 * @offset: Offset within stats (ignored for non-DMA statistics)
 693 */
 694struct efx_hw_stat_desc {
 695	const char *name;
 696	u16 dma_width;
 697	u16 offset;
 698};
 699
 700/* Number of bits used in a multicast filter hash address */
 701#define EFX_MCAST_HASH_BITS 8
 702
 703/* Number of (single-bit) entries in a multicast filter hash */
 704#define EFX_MCAST_HASH_ENTRIES (1 << EFX_MCAST_HASH_BITS)
 705
 706/* An Efx multicast filter hash */
 707union efx_multicast_hash {
 708	u8 byte[EFX_MCAST_HASH_ENTRIES / 8];
 709	efx_oword_t oword[EFX_MCAST_HASH_ENTRIES / sizeof(efx_oword_t) / 8];
 710};
 711
 
 712struct vfdi_status;
 713
 714/* The reserved RSS context value */
 715#define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
 716/**
 717 * struct efx_rss_context - A user-defined RSS context for filtering
 718 * @list: node of linked list on which this struct is stored
 719 * @context_id: the RSS_CONTEXT_ID returned by MC firmware, or
 720 *	%EFX_EF10_RSS_CONTEXT_INVALID if this context is not present on the NIC.
 721 *	For Siena, 0 if RSS is active, else %EFX_EF10_RSS_CONTEXT_INVALID.
 722 * @user_id: the rss_context ID exposed to userspace over ethtool.
 723 * @rx_hash_udp_4tuple: UDP 4-tuple hashing enabled
 724 * @rx_hash_key: Toeplitz hash key for this RSS context
 725 * @indir_table: Indirection table for this RSS context
 726 */
 727struct efx_rss_context {
 728	struct list_head list;
 729	u32 context_id;
 730	u32 user_id;
 731	bool rx_hash_udp_4tuple;
 732	u8 rx_hash_key[40];
 733	u32 rx_indir_table[128];
 734};
 735
 736#ifdef CONFIG_RFS_ACCEL
 737/* Order of these is important, since filter_id >= %EFX_ARFS_FILTER_ID_PENDING
 738 * is used to test if filter does or will exist.
 739 */
 740#define EFX_ARFS_FILTER_ID_PENDING	-1
 741#define EFX_ARFS_FILTER_ID_ERROR	-2
 742#define EFX_ARFS_FILTER_ID_REMOVING	-3
 743/**
 744 * struct efx_arfs_rule - record of an ARFS filter and its IDs
 745 * @node: linkage into hash table
 746 * @spec: details of the filter (used as key for hash table).  Use efx->type to
 747 *	determine which member to use.
 748 * @rxq_index: channel to which the filter will steer traffic.
 749 * @arfs_id: filter ID which was returned to ARFS
 750 * @filter_id: index in software filter table.  May be
 751 *	%EFX_ARFS_FILTER_ID_PENDING if filter was not inserted yet,
 752 *	%EFX_ARFS_FILTER_ID_ERROR if filter insertion failed, or
 753 *	%EFX_ARFS_FILTER_ID_REMOVING if expiry is currently removing the filter.
 754 */
 755struct efx_arfs_rule {
 756	struct hlist_node node;
 757	struct efx_filter_spec spec;
 758	u16 rxq_index;
 759	u16 arfs_id;
 760	s32 filter_id;
 761};
 762
 763/* Size chosen so that the table is one page (4kB) */
 764#define EFX_ARFS_HASH_TABLE_SIZE	512
 765
 766/**
 767 * struct efx_async_filter_insertion - Request to asynchronously insert a filter
 768 * @net_dev: Reference to the netdevice
 769 * @spec: The filter to insert
 770 * @work: Workitem for this request
 771 * @rxq_index: Identifies the channel for which this request was made
 772 * @flow_id: Identifies the kernel-side flow for which this request was made
 773 */
 774struct efx_async_filter_insertion {
 775	struct net_device *net_dev;
 776	struct efx_filter_spec spec;
 777	struct work_struct work;
 778	u16 rxq_index;
 779	u32 flow_id;
 780};
 781
 782/* Maximum number of ARFS workitems that may be in flight on an efx_nic */
 783#define EFX_RPS_MAX_IN_FLIGHT	8
 784#endif /* CONFIG_RFS_ACCEL */
 785
 786/**
 787 * struct efx_nic - an Efx NIC
 788 * @name: Device name (net device name or bus id before net device registered)
 789 * @pci_dev: The PCI device
 790 * @node: List node for maintaning primary/secondary function lists
 791 * @primary: &struct efx_nic instance for the primary function of this
 792 *	controller.  May be the same structure, and may be %NULL if no
 793 *	primary function is bound.  Serialised by rtnl_lock.
 794 * @secondary_list: List of &struct efx_nic instances for the secondary PCI
 795 *	functions of the controller, if this is for the primary function.
 796 *	Serialised by rtnl_lock.
 797 * @type: Controller type attributes
 798 * @legacy_irq: IRQ number
 799 * @workqueue: Workqueue for port reconfigures and the HW monitor.
 800 *	Work items do not hold and must not acquire RTNL.
 801 * @workqueue_name: Name of workqueue
 802 * @reset_work: Scheduled reset workitem
 803 * @membase_phys: Memory BAR value as physical address
 804 * @membase: Memory BAR value
 805 * @vi_stride: step between per-VI registers / memory regions
 806 * @interrupt_mode: Interrupt mode
 807 * @timer_quantum_ns: Interrupt timer quantum, in nanoseconds
 808 * @timer_max_ns: Interrupt timer maximum value, in nanoseconds
 809 * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues
 810 * @irq_rx_mod_step_us: Step size for IRQ moderation for RX event queues
 811 * @irq_rx_moderation_us: IRQ moderation time for RX event queues
 812 * @msg_enable: Log message enable flags
 813 * @state: Device state number (%STATE_*). Serialised by the rtnl_lock.
 814 * @reset_pending: Bitmask for pending resets
 815 * @tx_queue: TX DMA queues
 816 * @rx_queue: RX DMA queues
 817 * @channel: Channels
 818 * @msi_context: Context for each MSI
 819 * @extra_channel_types: Types of extra (non-traffic) channels that
 820 *	should be allocated for this NIC
 821 * @rxq_entries: Size of receive queues requested by user.
 822 * @txq_entries: Size of transmit queues requested by user.
 823 * @txq_stop_thresh: TX queue fill level at or above which we stop it.
 824 * @txq_wake_thresh: TX queue fill level at or below which we wake it.
 825 * @tx_dc_base: Base qword address in SRAM of TX queue descriptor caches
 826 * @rx_dc_base: Base qword address in SRAM of RX queue descriptor caches
 827 * @sram_lim_qw: Qword address limit of SRAM
 828 * @next_buffer_table: First available buffer table id
 829 * @n_channels: Number of channels in use
 830 * @n_rx_channels: Number of channels used for RX (= number of RX queues)
 831 * @n_tx_channels: Number of channels used for TX
 832 * @n_extra_tx_channels: Number of extra channels with TX queues
 833 * @rx_ip_align: RX DMA address offset to have IP header aligned in
 834 *	in accordance with NET_IP_ALIGN
 835 * @rx_dma_len: Current maximum RX DMA length
 836 * @rx_buffer_order: Order (log2) of number of pages for each RX buffer
 837 * @rx_buffer_truesize: Amortised allocation size of an RX buffer,
 838 *	for use in sk_buff::truesize
 839 * @rx_prefix_size: Size of RX prefix before packet data
 840 * @rx_packet_hash_offset: Offset of RX flow hash from start of packet data
 841 *	(valid only if @rx_prefix_size != 0; always negative)
 842 * @rx_packet_len_offset: Offset of RX packet length from start of packet data
 843 *	(valid only for NICs that set %EFX_RX_PKT_PREFIX_LEN; always negative)
 844 * @rx_packet_ts_offset: Offset of timestamp from start of packet data
 845 *	(valid only if channel->sync_timestamps_enabled; always negative)
 
 
 846 * @rx_scatter: Scatter mode enabled for receives
 847 * @rss_context: Main RSS context.  Its @list member is the head of the list of
 848 *	RSS contexts created by user requests
 849 * @rss_lock: Protects custom RSS context software state in @rss_context.list
 850 * @int_error_count: Number of internal errors seen recently
 851 * @int_error_expire: Time at which error count will be expired
 852 * @irq_soft_enabled: Are IRQs soft-enabled? If not, IRQ handler will
 853 *	acknowledge but do nothing else.
 854 * @irq_status: Interrupt status buffer
 855 * @irq_zero_count: Number of legacy IRQs seen with queue flags == 0
 856 * @irq_level: IRQ level/index for IRQs not triggered by an event queue
 857 * @selftest_work: Work item for asynchronous self-test
 858 * @mtd_list: List of MTDs attached to the NIC
 859 * @nic_data: Hardware dependent state
 860 * @mcdi: Management-Controller-to-Driver Interface state
 861 * @mac_lock: MAC access lock. Protects @port_enabled, @phy_mode,
 862 *	efx_monitor() and efx_reconfigure_port()
 863 * @port_enabled: Port enabled indicator.
 864 *	Serialises efx_stop_all(), efx_start_all(), efx_monitor() and
 865 *	efx_mac_work() with kernel interfaces. Safe to read under any
 866 *	one of the rtnl_lock, mac_lock, or netif_tx_lock, but all three must
 867 *	be held to modify it.
 868 * @port_initialized: Port initialized?
 869 * @net_dev: Operating system network device. Consider holding the rtnl lock
 870 * @fixed_features: Features which cannot be turned off
 871 * @num_mac_stats: Number of MAC stats reported by firmware (MAC_STATS_NUM_STATS
 872 *	field of %MC_CMD_GET_CAPABILITIES_V4 response, or %MC_CMD_MAC_NSTATS)
 873 * @stats_buffer: DMA buffer for statistics
 874 * @phy_type: PHY type
 875 * @phy_op: PHY interface
 876 * @phy_data: PHY private data (including PHY-specific stats)
 877 * @mdio: PHY MDIO interface
 878 * @mdio_bus: PHY MDIO bus ID (only used by Siena)
 879 * @phy_mode: PHY operating mode. Serialised by @mac_lock.
 880 * @link_advertising: Autonegotiation advertising flags
 881 * @fec_config: Forward Error Correction configuration flags.  For bit positions
 882 *	see &enum ethtool_fec_config_bits.
 883 * @link_state: Current state of the link
 884 * @n_link_state_changes: Number of times the link has changed state
 885 * @unicast_filter: Flag for Falcon-arch simple unicast filter.
 886 *	Protected by @mac_lock.
 887 * @multicast_hash: Multicast hash table for Falcon-arch.
 888 *	Protected by @mac_lock.
 889 * @wanted_fc: Wanted flow control flags
 890 * @fc_disable: When non-zero flow control is disabled. Typically used to
 891 *	ensure that network back pressure doesn't delay dma queue flushes.
 892 *	Serialised by the rtnl lock.
 893 * @mac_work: Work item for changing MAC promiscuity and multicast hash
 894 * @loopback_mode: Loopback status
 895 * @loopback_modes: Supported loopback mode bitmask
 896 * @loopback_selftest: Offline self-test private state
 897 * @filter_sem: Filter table rw_semaphore, protects existence of @filter_state
 898 * @filter_state: Architecture-dependent filter table state
 899 * @rps_mutex: Protects RPS state of all channels
 900 * @rps_expire_channel: Next channel to check for expiry
 901 * @rps_expire_index: Next index to check for expiry in
 902 *	@rps_expire_channel's @rps_flow_id
 903 * @rps_slot_map: bitmap of in-flight entries in @rps_slot
 904 * @rps_slot: array of ARFS insertion requests for efx_filter_rfs_work()
 905 * @rps_hash_lock: Protects ARFS filter mapping state (@rps_hash_table and
 906 *	@rps_next_id).
 907 * @rps_hash_table: Mapping between ARFS filters and their various IDs
 908 * @rps_next_id: next arfs_id for an ARFS filter
 909 * @active_queues: Count of RX and TX queues that haven't been flushed and drained.
 910 * @rxq_flush_pending: Count of number of receive queues that need to be flushed.
 911 *	Decremented when the efx_flush_rx_queue() is called.
 912 * @rxq_flush_outstanding: Count of number of RX flushes started but not yet
 913 *	completed (either success or failure). Not used when MCDI is used to
 914 *	flush receive queues.
 915 * @flush_wq: wait queue used by efx_nic_flush_queues() to wait for flush completions.
 
 916 * @vf_count: Number of VFs intended to be enabled.
 917 * @vf_init_count: Number of VFs that have been fully initialised.
 918 * @vi_scale: log2 number of vnics per VF.
 
 
 
 
 
 
 
 919 * @ptp_data: PTP state data
 920 * @ptp_warned: has this NIC seen and warned about unexpected PTP events?
 921 * @vpd_sn: Serial number read from VPD
 922 * @monitor_work: Hardware monitor workitem
 923 * @biu_lock: BIU (bus interface unit) lock
 924 * @last_irq_cpu: Last CPU to handle a possible test interrupt.  This
 925 *	field is used by efx_test_interrupts() to verify that an
 926 *	interrupt has occurred.
 927 * @stats_lock: Statistics update lock. Must be held when calling
 928 *	efx_nic_type::{update,start,stop}_stats.
 929 * @n_rx_noskb_drops: Count of RX packets dropped due to failure to allocate an skb
 930 *
 931 * This is stored in the private area of the &struct net_device.
 932 */
 933struct efx_nic {
 934	/* The following fields should be written very rarely */
 935
 936	char name[IFNAMSIZ];
 937	struct list_head node;
 938	struct efx_nic *primary;
 939	struct list_head secondary_list;
 940	struct pci_dev *pci_dev;
 941	unsigned int port_num;
 942	const struct efx_nic_type *type;
 943	int legacy_irq;
 944	bool eeh_disabled_legacy_irq;
 945	struct workqueue_struct *workqueue;
 946	char workqueue_name[16];
 947	struct work_struct reset_work;
 948	resource_size_t membase_phys;
 949	void __iomem *membase;
 950
 951	unsigned int vi_stride;
 952
 953	enum efx_int_mode interrupt_mode;
 954	unsigned int timer_quantum_ns;
 955	unsigned int timer_max_ns;
 956	bool irq_rx_adaptive;
 957	unsigned int irq_mod_step_us;
 958	unsigned int irq_rx_moderation_us;
 959	u32 msg_enable;
 960
 961	enum nic_state state;
 962	unsigned long reset_pending;
 963
 964	struct efx_channel *channel[EFX_MAX_CHANNELS];
 965	struct efx_msi_context msi_context[EFX_MAX_CHANNELS];
 966	const struct efx_channel_type *
 967	extra_channel_type[EFX_MAX_EXTRA_CHANNELS];
 968
 969	unsigned rxq_entries;
 970	unsigned txq_entries;
 971	unsigned int txq_stop_thresh;
 972	unsigned int txq_wake_thresh;
 973
 974	unsigned tx_dc_base;
 975	unsigned rx_dc_base;
 976	unsigned sram_lim_qw;
 977	unsigned next_buffer_table;
 978
 979	unsigned int max_channels;
 980	unsigned int max_tx_channels;
 981	unsigned n_channels;
 982	unsigned n_rx_channels;
 983	unsigned rss_spread;
 984	unsigned tx_channel_offset;
 985	unsigned n_tx_channels;
 986	unsigned n_extra_tx_channels;
 987	unsigned int rx_ip_align;
 988	unsigned int rx_dma_len;
 989	unsigned int rx_buffer_order;
 990	unsigned int rx_buffer_truesize;
 991	unsigned int rx_page_buf_step;
 992	unsigned int rx_bufs_per_page;
 993	unsigned int rx_pages_per_batch;
 994	unsigned int rx_prefix_size;
 995	int rx_packet_hash_offset;
 996	int rx_packet_len_offset;
 997	int rx_packet_ts_offset;
 
 
 998	bool rx_scatter;
 999	struct efx_rss_context rss_context;
1000	struct mutex rss_lock;
1001
1002	unsigned int_error_count;
1003	unsigned long int_error_expire;
1004
1005	bool irq_soft_enabled;
1006	struct efx_buffer irq_status;
1007	unsigned irq_zero_count;
1008	unsigned irq_level;
1009	struct delayed_work selftest_work;
1010
1011#ifdef CONFIG_SFC_MTD
1012	struct list_head mtd_list;
1013#endif
1014
1015	void *nic_data;
1016	struct efx_mcdi_data *mcdi;
1017
1018	struct mutex mac_lock;
1019	struct work_struct mac_work;
1020	bool port_enabled;
1021
1022	bool mc_bist_for_other_fn;
1023	bool port_initialized;
1024	struct net_device *net_dev;
1025
1026	netdev_features_t fixed_features;
1027
1028	u16 num_mac_stats;
1029	struct efx_buffer stats_buffer;
1030	u64 rx_nodesc_drops_total;
1031	u64 rx_nodesc_drops_while_down;
1032	bool rx_nodesc_drops_prev_state;
1033
1034	unsigned int phy_type;
1035	const struct efx_phy_operations *phy_op;
1036	void *phy_data;
1037	struct mdio_if_info mdio;
1038	unsigned int mdio_bus;
1039	enum efx_phy_mode phy_mode;
1040
1041	__ETHTOOL_DECLARE_LINK_MODE_MASK(link_advertising);
1042	u32 fec_config;
1043	struct efx_link_state link_state;
1044	unsigned int n_link_state_changes;
1045
1046	bool unicast_filter;
1047	union efx_multicast_hash multicast_hash;
1048	u8 wanted_fc;
1049	unsigned fc_disable;
1050
1051	atomic_t rx_reset;
1052	enum efx_loopback_mode loopback_mode;
1053	u64 loopback_modes;
1054
1055	void *loopback_selftest;
1056
1057	struct rw_semaphore filter_sem;
1058	void *filter_state;
1059#ifdef CONFIG_RFS_ACCEL
1060	struct mutex rps_mutex;
1061	unsigned int rps_expire_channel;
1062	unsigned int rps_expire_index;
1063	unsigned long rps_slot_map;
1064	struct efx_async_filter_insertion rps_slot[EFX_RPS_MAX_IN_FLIGHT];
1065	spinlock_t rps_hash_lock;
1066	struct hlist_head *rps_hash_table;
1067	u32 rps_next_id;
1068#endif
1069
1070	atomic_t active_queues;
1071	atomic_t rxq_flush_pending;
1072	atomic_t rxq_flush_outstanding;
1073	wait_queue_head_t flush_wq;
1074
1075#ifdef CONFIG_SFC_SRIOV
 
 
1076	unsigned vf_count;
1077	unsigned vf_init_count;
1078	unsigned vi_scale;
 
 
 
 
 
 
1079#endif
1080
1081	struct efx_ptp_data *ptp_data;
1082	bool ptp_warned;
1083
1084	char *vpd_sn;
1085
1086	/* The following fields may be written more often */
1087
1088	struct delayed_work monitor_work ____cacheline_aligned_in_smp;
1089	spinlock_t biu_lock;
1090	int last_irq_cpu;
1091	spinlock_t stats_lock;
1092	atomic_t n_rx_noskb_drops;
1093};
1094
1095static inline int efx_dev_registered(struct efx_nic *efx)
1096{
1097	return efx->net_dev->reg_state == NETREG_REGISTERED;
1098}
1099
1100static inline unsigned int efx_port_num(struct efx_nic *efx)
1101{
1102	return efx->port_num;
1103}
1104
1105struct efx_mtd_partition {
1106	struct list_head node;
1107	struct mtd_info mtd;
1108	const char *dev_type_name;
1109	const char *type_name;
1110	char name[IFNAMSIZ + 20];
1111};
1112
1113struct efx_udp_tunnel {
1114	u16 type; /* TUNNEL_ENCAP_UDP_PORT_ENTRY_foo, see mcdi_pcol.h */
1115	__be16 port;
1116	/* Count of repeated adds of the same port.  Used only inside the list,
1117	 * not in request arguments.
1118	 */
1119	u16 count;
1120};
1121
1122/**
1123 * struct efx_nic_type - Efx device type definition
1124 * @mem_bar: Get the memory BAR
1125 * @mem_map_size: Get memory BAR mapped size
1126 * @probe: Probe the controller
1127 * @remove: Free resources allocated by probe()
1128 * @init: Initialise the controller
1129 * @dimension_resources: Dimension controller resources (buffer table,
1130 *	and VIs once the available interrupt resources are clear)
1131 * @fini: Shut down the controller
1132 * @monitor: Periodic function for polling link state and hardware monitor
1133 * @map_reset_reason: Map ethtool reset reason to a reset method
1134 * @map_reset_flags: Map ethtool reset flags to a reset method, if possible
1135 * @reset: Reset the controller hardware and possibly the PHY.  This will
1136 *	be called while the controller is uninitialised.
1137 * @probe_port: Probe the MAC and PHY
1138 * @remove_port: Free resources allocated by probe_port()
1139 * @handle_global_event: Handle a "global" event (may be %NULL)
1140 * @fini_dmaq: Flush and finalise DMA queues (RX and TX queues)
1141 * @prepare_flush: Prepare the hardware for flushing the DMA queues
1142 *	(for Falcon architecture)
1143 * @finish_flush: Clean up after flushing the DMA queues (for Falcon
1144 *	architecture)
1145 * @prepare_flr: Prepare for an FLR
1146 * @finish_flr: Clean up after an FLR
1147 * @describe_stats: Describe statistics for ethtool
1148 * @update_stats: Update statistics not provided by event handling.
1149 *	Either argument may be %NULL.
1150 * @start_stats: Start the regular fetching of statistics
1151 * @pull_stats: Pull stats from the NIC and wait until they arrive.
1152 * @stop_stats: Stop the regular fetching of statistics
1153 * @set_id_led: Set state of identifying LED or revert to automatic function
1154 * @push_irq_moderation: Apply interrupt moderation value
1155 * @reconfigure_port: Push loopback/power/txdis changes to the MAC and PHY
1156 * @prepare_enable_fc_tx: Prepare MAC to enable pause frame TX (may be %NULL)
1157 * @reconfigure_mac: Push MAC address, MTU, flow control and filter settings
1158 *	to the hardware.  Serialised by the mac_lock.
1159 * @check_mac_fault: Check MAC fault state. True if fault present.
1160 * @get_wol: Get WoL configuration from driver state
1161 * @set_wol: Push WoL configuration to the NIC
1162 * @resume_wol: Synchronise WoL state between driver and MC (e.g. after resume)
1163 * @test_chip: Test registers.  May use efx_farch_test_registers(), and is
1164 *	expected to reset the NIC.
1165 * @test_nvram: Test validity of NVRAM contents
1166 * @mcdi_request: Send an MCDI request with the given header and SDU.
1167 *	The SDU length may be any value from 0 up to the protocol-
1168 *	defined maximum, but its buffer will be padded to a multiple
1169 *	of 4 bytes.
1170 * @mcdi_poll_response: Test whether an MCDI response is available.
1171 * @mcdi_read_response: Read the MCDI response PDU.  The offset will
1172 *	be a multiple of 4.  The length may not be, but the buffer
1173 *	will be padded so it is safe to round up.
1174 * @mcdi_poll_reboot: Test whether the MCDI has rebooted.  If so,
1175 *	return an appropriate error code for aborting any current
1176 *	request; otherwise return 0.
1177 * @irq_enable_master: Enable IRQs on the NIC.  Each event queue must
1178 *	be separately enabled after this.
1179 * @irq_test_generate: Generate a test IRQ
1180 * @irq_disable_non_ev: Disable non-event IRQs on the NIC.  Each event
1181 *	queue must be separately disabled before this.
1182 * @irq_handle_msi: Handle MSI for a channel.  The @dev_id argument is
1183 *	a pointer to the &struct efx_msi_context for the channel.
1184 * @irq_handle_legacy: Handle legacy interrupt.  The @dev_id argument
1185 *	is a pointer to the &struct efx_nic.
1186 * @tx_probe: Allocate resources for TX queue
1187 * @tx_init: Initialise TX queue on the NIC
1188 * @tx_remove: Free resources for TX queue
1189 * @tx_write: Write TX descriptors and doorbell
1190 * @rx_push_rss_config: Write RSS hash key and indirection table to the NIC
1191 * @rx_pull_rss_config: Read RSS hash key and indirection table back from the NIC
1192 * @rx_push_rss_context_config: Write RSS hash key and indirection table for
1193 *	user RSS context to the NIC
1194 * @rx_pull_rss_context_config: Read RSS hash key and indirection table for user
1195 *	RSS context back from the NIC
1196 * @rx_probe: Allocate resources for RX queue
1197 * @rx_init: Initialise RX queue on the NIC
1198 * @rx_remove: Free resources for RX queue
1199 * @rx_write: Write RX descriptors and doorbell
1200 * @rx_defer_refill: Generate a refill reminder event
1201 * @ev_probe: Allocate resources for event queue
1202 * @ev_init: Initialise event queue on the NIC
1203 * @ev_fini: Deinitialise event queue on the NIC
1204 * @ev_remove: Free resources for event queue
1205 * @ev_process: Process events for a queue, up to the given NAPI quota
1206 * @ev_read_ack: Acknowledge read events on a queue, rearming its IRQ
1207 * @ev_test_generate: Generate a test event
1208 * @filter_table_probe: Probe filter capabilities and set up filter software state
1209 * @filter_table_restore: Restore filters removed from hardware
1210 * @filter_table_remove: Remove filters from hardware and tear down software state
1211 * @filter_update_rx_scatter: Update filters after change to rx scatter setting
1212 * @filter_insert: add or replace a filter
1213 * @filter_remove_safe: remove a filter by ID, carefully
1214 * @filter_get_safe: retrieve a filter by ID, carefully
1215 * @filter_clear_rx: Remove all RX filters whose priority is less than or
1216 *	equal to the given priority and is not %EFX_FILTER_PRI_AUTO
1217 * @filter_count_rx_used: Get the number of filters in use at a given priority
1218 * @filter_get_rx_id_limit: Get maximum value of a filter id, plus 1
1219 * @filter_get_rx_ids: Get list of RX filters at a given priority
 
 
 
 
1220 * @filter_rfs_expire_one: Consider expiring a filter inserted for RFS.
1221 *	This must check whether the specified table entry is used by RFS
1222 *	and that rps_may_expire_flow() returns true for it.
1223 * @mtd_probe: Probe and add MTD partitions associated with this net device,
1224 *	 using efx_mtd_add()
1225 * @mtd_rename: Set an MTD partition name using the net device name
1226 * @mtd_read: Read from an MTD partition
1227 * @mtd_erase: Erase part of an MTD partition
1228 * @mtd_write: Write to an MTD partition
1229 * @mtd_sync: Wait for write-back to complete on MTD partition.  This
1230 *	also notifies the driver that a writer has finished using this
1231 *	partition.
1232 * @ptp_write_host_time: Send host time to MC as part of sync protocol
1233 * @ptp_set_ts_sync_events: Enable or disable sync events for inline RX
1234 *	timestamping, possibly only temporarily for the purposes of a reset.
1235 * @ptp_set_ts_config: Set hardware timestamp configuration.  The flags
1236 *	and tx_type will already have been validated but this operation
1237 *	must validate and update rx_filter.
1238 * @get_phys_port_id: Get the underlying physical port id.
1239 * @set_mac_address: Set the MAC address of the device
1240 * @tso_versions: Returns mask of firmware-assisted TSO versions supported.
1241 *	If %NULL, then device does not support any TSO version.
1242 * @udp_tnl_push_ports: Push the list of UDP tunnel ports to the NIC if required.
1243 * @udp_tnl_add_port: Add a UDP tunnel port
1244 * @udp_tnl_has_port: Check if a port has been added as UDP tunnel
1245 * @udp_tnl_del_port: Remove a UDP tunnel port
1246 * @revision: Hardware architecture revision
1247 * @txd_ptr_tbl_base: TX descriptor ring base address
1248 * @rxd_ptr_tbl_base: RX descriptor ring base address
1249 * @buf_tbl_base: Buffer table base address
1250 * @evq_ptr_tbl_base: Event queue pointer table base address
1251 * @evq_rptr_tbl_base: Event queue read-pointer table base address
1252 * @max_dma_mask: Maximum possible DMA mask
1253 * @rx_prefix_size: Size of RX prefix before packet data
1254 * @rx_hash_offset: Offset of RX flow hash within prefix
1255 * @rx_ts_offset: Offset of timestamp within prefix
1256 * @rx_buffer_padding: Size of padding at end of RX packet
1257 * @can_rx_scatter: NIC is able to scatter packets to multiple buffers
1258 * @always_rx_scatter: NIC will always scatter packets to multiple buffers
1259 * @option_descriptors: NIC supports TX option descriptors
1260 * @min_interrupt_mode: Lowest capability interrupt mode supported
1261 *	from &enum efx_int_mode.
1262 * @max_interrupt_mode: Highest capability interrupt mode supported
1263 *	from &enum efx_int_mode.
1264 * @timer_period_max: Maximum period of interrupt timer (in ticks)
1265 * @offload_features: net_device feature flags for protocol offload
1266 *	features implemented in hardware
1267 * @mcdi_max_ver: Maximum MCDI version supported
1268 * @hwtstamp_filters: Mask of hardware timestamp filter types supported
1269 */
1270struct efx_nic_type {
1271	bool is_vf;
1272	unsigned int (*mem_bar)(struct efx_nic *efx);
1273	unsigned int (*mem_map_size)(struct efx_nic *efx);
1274	int (*probe)(struct efx_nic *efx);
1275	void (*remove)(struct efx_nic *efx);
1276	int (*init)(struct efx_nic *efx);
1277	int (*dimension_resources)(struct efx_nic *efx);
1278	void (*fini)(struct efx_nic *efx);
1279	void (*monitor)(struct efx_nic *efx);
1280	enum reset_type (*map_reset_reason)(enum reset_type reason);
1281	int (*map_reset_flags)(u32 *flags);
1282	int (*reset)(struct efx_nic *efx, enum reset_type method);
1283	int (*probe_port)(struct efx_nic *efx);
1284	void (*remove_port)(struct efx_nic *efx);
1285	bool (*handle_global_event)(struct efx_channel *channel, efx_qword_t *);
1286	int (*fini_dmaq)(struct efx_nic *efx);
1287	void (*prepare_flush)(struct efx_nic *efx);
1288	void (*finish_flush)(struct efx_nic *efx);
1289	void (*prepare_flr)(struct efx_nic *efx);
1290	void (*finish_flr)(struct efx_nic *efx);
1291	size_t (*describe_stats)(struct efx_nic *efx, u8 *names);
1292	size_t (*update_stats)(struct efx_nic *efx, u64 *full_stats,
1293			       struct rtnl_link_stats64 *core_stats);
1294	void (*start_stats)(struct efx_nic *efx);
1295	void (*pull_stats)(struct efx_nic *efx);
1296	void (*stop_stats)(struct efx_nic *efx);
1297	void (*set_id_led)(struct efx_nic *efx, enum efx_led_mode mode);
1298	void (*push_irq_moderation)(struct efx_channel *channel);
1299	int (*reconfigure_port)(struct efx_nic *efx);
1300	void (*prepare_enable_fc_tx)(struct efx_nic *efx);
1301	int (*reconfigure_mac)(struct efx_nic *efx);
1302	bool (*check_mac_fault)(struct efx_nic *efx);
1303	void (*get_wol)(struct efx_nic *efx, struct ethtool_wolinfo *wol);
1304	int (*set_wol)(struct efx_nic *efx, u32 type);
1305	void (*resume_wol)(struct efx_nic *efx);
1306	int (*test_chip)(struct efx_nic *efx, struct efx_self_tests *tests);
1307	int (*test_nvram)(struct efx_nic *efx);
1308	void (*mcdi_request)(struct efx_nic *efx,
1309			     const efx_dword_t *hdr, size_t hdr_len,
1310			     const efx_dword_t *sdu, size_t sdu_len);
1311	bool (*mcdi_poll_response)(struct efx_nic *efx);
1312	void (*mcdi_read_response)(struct efx_nic *efx, efx_dword_t *pdu,
1313				   size_t pdu_offset, size_t pdu_len);
1314	int (*mcdi_poll_reboot)(struct efx_nic *efx);
1315	void (*mcdi_reboot_detected)(struct efx_nic *efx);
1316	void (*irq_enable_master)(struct efx_nic *efx);
1317	int (*irq_test_generate)(struct efx_nic *efx);
1318	void (*irq_disable_non_ev)(struct efx_nic *efx);
1319	irqreturn_t (*irq_handle_msi)(int irq, void *dev_id);
1320	irqreturn_t (*irq_handle_legacy)(int irq, void *dev_id);
1321	int (*tx_probe)(struct efx_tx_queue *tx_queue);
1322	void (*tx_init)(struct efx_tx_queue *tx_queue);
1323	void (*tx_remove)(struct efx_tx_queue *tx_queue);
1324	void (*tx_write)(struct efx_tx_queue *tx_queue);
1325	unsigned int (*tx_limit_len)(struct efx_tx_queue *tx_queue,
1326				     dma_addr_t dma_addr, unsigned int len);
1327	int (*rx_push_rss_config)(struct efx_nic *efx, bool user,
1328				  const u32 *rx_indir_table, const u8 *key);
1329	int (*rx_pull_rss_config)(struct efx_nic *efx);
1330	int (*rx_push_rss_context_config)(struct efx_nic *efx,
1331					  struct efx_rss_context *ctx,
1332					  const u32 *rx_indir_table,
1333					  const u8 *key);
1334	int (*rx_pull_rss_context_config)(struct efx_nic *efx,
1335					  struct efx_rss_context *ctx);
1336	void (*rx_restore_rss_contexts)(struct efx_nic *efx);
1337	int (*rx_probe)(struct efx_rx_queue *rx_queue);
1338	void (*rx_init)(struct efx_rx_queue *rx_queue);
1339	void (*rx_remove)(struct efx_rx_queue *rx_queue);
1340	void (*rx_write)(struct efx_rx_queue *rx_queue);
1341	void (*rx_defer_refill)(struct efx_rx_queue *rx_queue);
1342	int (*ev_probe)(struct efx_channel *channel);
1343	int (*ev_init)(struct efx_channel *channel);
1344	void (*ev_fini)(struct efx_channel *channel);
1345	void (*ev_remove)(struct efx_channel *channel);
1346	int (*ev_process)(struct efx_channel *channel, int quota);
1347	void (*ev_read_ack)(struct efx_channel *channel);
1348	void (*ev_test_generate)(struct efx_channel *channel);
1349	int (*filter_table_probe)(struct efx_nic *efx);
1350	void (*filter_table_restore)(struct efx_nic *efx);
1351	void (*filter_table_remove)(struct efx_nic *efx);
1352	void (*filter_update_rx_scatter)(struct efx_nic *efx);
1353	s32 (*filter_insert)(struct efx_nic *efx,
1354			     struct efx_filter_spec *spec, bool replace);
1355	int (*filter_remove_safe)(struct efx_nic *efx,
1356				  enum efx_filter_priority priority,
1357				  u32 filter_id);
1358	int (*filter_get_safe)(struct efx_nic *efx,
1359			       enum efx_filter_priority priority,
1360			       u32 filter_id, struct efx_filter_spec *);
1361	int (*filter_clear_rx)(struct efx_nic *efx,
1362			       enum efx_filter_priority priority);
1363	u32 (*filter_count_rx_used)(struct efx_nic *efx,
1364				    enum efx_filter_priority priority);
1365	u32 (*filter_get_rx_id_limit)(struct efx_nic *efx);
1366	s32 (*filter_get_rx_ids)(struct efx_nic *efx,
1367				 enum efx_filter_priority priority,
1368				 u32 *buf, u32 size);
1369#ifdef CONFIG_RFS_ACCEL
 
 
1370	bool (*filter_rfs_expire_one)(struct efx_nic *efx, u32 flow_id,
1371				      unsigned int index);
1372#endif
1373#ifdef CONFIG_SFC_MTD
1374	int (*mtd_probe)(struct efx_nic *efx);
1375	void (*mtd_rename)(struct efx_mtd_partition *part);
1376	int (*mtd_read)(struct mtd_info *mtd, loff_t start, size_t len,
1377			size_t *retlen, u8 *buffer);
1378	int (*mtd_erase)(struct mtd_info *mtd, loff_t start, size_t len);
1379	int (*mtd_write)(struct mtd_info *mtd, loff_t start, size_t len,
1380			 size_t *retlen, const u8 *buffer);
1381	int (*mtd_sync)(struct mtd_info *mtd);
1382#endif
1383	void (*ptp_write_host_time)(struct efx_nic *efx, u32 host_time);
1384	int (*ptp_set_ts_sync_events)(struct efx_nic *efx, bool en, bool temp);
1385	int (*ptp_set_ts_config)(struct efx_nic *efx,
1386				 struct hwtstamp_config *init);
1387	int (*sriov_configure)(struct efx_nic *efx, int num_vfs);
1388	int (*vlan_rx_add_vid)(struct efx_nic *efx, __be16 proto, u16 vid);
1389	int (*vlan_rx_kill_vid)(struct efx_nic *efx, __be16 proto, u16 vid);
1390	int (*get_phys_port_id)(struct efx_nic *efx,
1391				struct netdev_phys_item_id *ppid);
1392	int (*sriov_init)(struct efx_nic *efx);
1393	void (*sriov_fini)(struct efx_nic *efx);
1394	bool (*sriov_wanted)(struct efx_nic *efx);
1395	void (*sriov_reset)(struct efx_nic *efx);
1396	void (*sriov_flr)(struct efx_nic *efx, unsigned vf_i);
1397	int (*sriov_set_vf_mac)(struct efx_nic *efx, int vf_i, u8 *mac);
1398	int (*sriov_set_vf_vlan)(struct efx_nic *efx, int vf_i, u16 vlan,
1399				 u8 qos);
1400	int (*sriov_set_vf_spoofchk)(struct efx_nic *efx, int vf_i,
1401				     bool spoofchk);
1402	int (*sriov_get_vf_config)(struct efx_nic *efx, int vf_i,
1403				   struct ifla_vf_info *ivi);
1404	int (*sriov_set_vf_link_state)(struct efx_nic *efx, int vf_i,
1405				       int link_state);
1406	int (*vswitching_probe)(struct efx_nic *efx);
1407	int (*vswitching_restore)(struct efx_nic *efx);
1408	void (*vswitching_remove)(struct efx_nic *efx);
1409	int (*get_mac_address)(struct efx_nic *efx, unsigned char *perm_addr);
1410	int (*set_mac_address)(struct efx_nic *efx);
1411	u32 (*tso_versions)(struct efx_nic *efx);
1412	int (*udp_tnl_push_ports)(struct efx_nic *efx);
1413	int (*udp_tnl_add_port)(struct efx_nic *efx, struct efx_udp_tunnel tnl);
1414	bool (*udp_tnl_has_port)(struct efx_nic *efx, __be16 port);
1415	int (*udp_tnl_del_port)(struct efx_nic *efx, struct efx_udp_tunnel tnl);
1416
1417	int revision;
1418	unsigned int txd_ptr_tbl_base;
1419	unsigned int rxd_ptr_tbl_base;
1420	unsigned int buf_tbl_base;
1421	unsigned int evq_ptr_tbl_base;
1422	unsigned int evq_rptr_tbl_base;
1423	u64 max_dma_mask;
1424	unsigned int rx_prefix_size;
1425	unsigned int rx_hash_offset;
1426	unsigned int rx_ts_offset;
1427	unsigned int rx_buffer_padding;
1428	bool can_rx_scatter;
1429	bool always_rx_scatter;
1430	bool option_descriptors;
1431	unsigned int min_interrupt_mode;
1432	unsigned int max_interrupt_mode;
1433	unsigned int timer_period_max;
1434	netdev_features_t offload_features;
1435	int mcdi_max_ver;
1436	unsigned int max_rx_ip_filters;
1437	u32 hwtstamp_filters;
1438	unsigned int rx_hash_key_size;
1439};
1440
1441/**************************************************************************
1442 *
1443 * Prototypes and inline functions
1444 *
1445 *************************************************************************/
1446
1447static inline struct efx_channel *
1448efx_get_channel(struct efx_nic *efx, unsigned index)
1449{
1450	EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_channels);
1451	return efx->channel[index];
1452}
1453
1454/* Iterate over all used channels */
1455#define efx_for_each_channel(_channel, _efx)				\
1456	for (_channel = (_efx)->channel[0];				\
1457	     _channel;							\
1458	     _channel = (_channel->channel + 1 < (_efx)->n_channels) ?	\
1459		     (_efx)->channel[_channel->channel + 1] : NULL)
1460
1461/* Iterate over all used channels in reverse */
1462#define efx_for_each_channel_rev(_channel, _efx)			\
1463	for (_channel = (_efx)->channel[(_efx)->n_channels - 1];	\
1464	     _channel;							\
1465	     _channel = _channel->channel ?				\
1466		     (_efx)->channel[_channel->channel - 1] : NULL)
1467
1468static inline struct efx_tx_queue *
1469efx_get_tx_queue(struct efx_nic *efx, unsigned index, unsigned type)
1470{
1471	EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_tx_channels ||
1472				  type >= EFX_TXQ_TYPES);
1473	return &efx->channel[efx->tx_channel_offset + index]->tx_queue[type];
1474}
1475
1476static inline bool efx_channel_has_tx_queues(struct efx_channel *channel)
1477{
1478	return channel->type && channel->type->want_txqs &&
1479				channel->type->want_txqs(channel);
1480}
1481
1482static inline struct efx_tx_queue *
1483efx_channel_get_tx_queue(struct efx_channel *channel, unsigned type)
1484{
1485	EFX_WARN_ON_ONCE_PARANOID(!efx_channel_has_tx_queues(channel) ||
1486				  type >= EFX_TXQ_TYPES);
1487	return &channel->tx_queue[type];
1488}
1489
1490static inline bool efx_tx_queue_used(struct efx_tx_queue *tx_queue)
1491{
1492	return !(tx_queue->efx->net_dev->num_tc < 2 &&
1493		 tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI);
1494}
1495
1496/* Iterate over all TX queues belonging to a channel */
1497#define efx_for_each_channel_tx_queue(_tx_queue, _channel)		\
1498	if (!efx_channel_has_tx_queues(_channel))			\
1499		;							\
1500	else								\
1501		for (_tx_queue = (_channel)->tx_queue;			\
1502		     _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES && \
1503			     efx_tx_queue_used(_tx_queue);		\
1504		     _tx_queue++)
1505
1506/* Iterate over all possible TX queues belonging to a channel */
1507#define efx_for_each_possible_channel_tx_queue(_tx_queue, _channel)	\
1508	if (!efx_channel_has_tx_queues(_channel))			\
1509		;							\
1510	else								\
1511		for (_tx_queue = (_channel)->tx_queue;			\
1512		     _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES;	\
1513		     _tx_queue++)
1514
1515static inline bool efx_channel_has_rx_queue(struct efx_channel *channel)
1516{
1517	return channel->rx_queue.core_index >= 0;
1518}
1519
1520static inline struct efx_rx_queue *
1521efx_channel_get_rx_queue(struct efx_channel *channel)
1522{
1523	EFX_WARN_ON_ONCE_PARANOID(!efx_channel_has_rx_queue(channel));
1524	return &channel->rx_queue;
1525}
1526
1527/* Iterate over all RX queues belonging to a channel */
1528#define efx_for_each_channel_rx_queue(_rx_queue, _channel)		\
1529	if (!efx_channel_has_rx_queue(_channel))			\
1530		;							\
1531	else								\
1532		for (_rx_queue = &(_channel)->rx_queue;			\
1533		     _rx_queue;						\
1534		     _rx_queue = NULL)
1535
1536static inline struct efx_channel *
1537efx_rx_queue_channel(struct efx_rx_queue *rx_queue)
1538{
1539	return container_of(rx_queue, struct efx_channel, rx_queue);
1540}
1541
1542static inline int efx_rx_queue_index(struct efx_rx_queue *rx_queue)
1543{
1544	return efx_rx_queue_channel(rx_queue)->channel;
1545}
1546
1547/* Returns a pointer to the specified receive buffer in the RX
1548 * descriptor queue.
1549 */
1550static inline struct efx_rx_buffer *efx_rx_buffer(struct efx_rx_queue *rx_queue,
1551						  unsigned int index)
1552{
1553	return &rx_queue->buffer[index];
1554}
1555
1556/**
1557 * EFX_MAX_FRAME_LEN - calculate maximum frame length
1558 *
1559 * This calculates the maximum frame length that will be used for a
1560 * given MTU.  The frame length will be equal to the MTU plus a
1561 * constant amount of header space and padding.  This is the quantity
1562 * that the net driver will program into the MAC as the maximum frame
1563 * length.
1564 *
1565 * The 10G MAC requires 8-byte alignment on the frame
1566 * length, so we round up to the nearest 8.
1567 *
1568 * Re-clocking by the XGXS on RX can reduce an IPG to 32 bits (half an
1569 * XGMII cycle).  If the frame length reaches the maximum value in the
1570 * same cycle, the XMAC can miss the IPG altogether.  We work around
1571 * this by adding a further 16 bytes.
1572 */
1573#define EFX_FRAME_PAD	16
1574#define EFX_MAX_FRAME_LEN(mtu) \
1575	(ALIGN(((mtu) + ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN + EFX_FRAME_PAD), 8))
1576
1577static inline bool efx_xmit_with_hwtstamp(struct sk_buff *skb)
1578{
1579	return skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP;
1580}
1581static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
1582{
1583	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1584}
1585
1586/* Get all supported features.
1587 * If a feature is not fixed, it is present in hw_features.
1588 * If a feature is fixed, it does not present in hw_features, but
1589 * always in features.
1590 */
1591static inline netdev_features_t efx_supported_features(const struct efx_nic *efx)
1592{
1593	const struct net_device *net_dev = efx->net_dev;
1594
1595	return net_dev->features | net_dev->hw_features;
1596}
1597
1598/* Get the current TX queue insert index. */
1599static inline unsigned int
1600efx_tx_queue_get_insert_index(const struct efx_tx_queue *tx_queue)
1601{
1602	return tx_queue->insert_count & tx_queue->ptr_mask;
1603}
1604
1605/* Get a TX buffer. */
1606static inline struct efx_tx_buffer *
1607__efx_tx_queue_get_insert_buffer(const struct efx_tx_queue *tx_queue)
1608{
1609	return &tx_queue->buffer[efx_tx_queue_get_insert_index(tx_queue)];
1610}
1611
1612/* Get a TX buffer, checking it's not currently in use. */
1613static inline struct efx_tx_buffer *
1614efx_tx_queue_get_insert_buffer(const struct efx_tx_queue *tx_queue)
1615{
1616	struct efx_tx_buffer *buffer =
1617		__efx_tx_queue_get_insert_buffer(tx_queue);
1618
1619	EFX_WARN_ON_ONCE_PARANOID(buffer->len);
1620	EFX_WARN_ON_ONCE_PARANOID(buffer->flags);
1621	EFX_WARN_ON_ONCE_PARANOID(buffer->unmap_len);
1622
1623	return buffer;
1624}
1625
1626#endif /* EFX_NET_DRIVER_H */