Loading...
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/mtd/mtd.h>
28#include <net/busy_poll.h>
29#include <net/xdp.h>
30#include <net/netevent.h>
31
32#include "enum.h"
33#include "bitfield.h"
34#include "filter.h"
35
36/**************************************************************************
37 *
38 * Build definitions
39 *
40 **************************************************************************/
41
42#ifdef DEBUG
43#define EFX_WARN_ON_ONCE_PARANOID(x) WARN_ON_ONCE(x)
44#define EFX_WARN_ON_PARANOID(x) WARN_ON(x)
45#else
46#define EFX_WARN_ON_ONCE_PARANOID(x) do {} while (0)
47#define EFX_WARN_ON_PARANOID(x) do {} while (0)
48#endif
49
50/**************************************************************************
51 *
52 * Efx data structures
53 *
54 **************************************************************************/
55
56#define EFX_MAX_CHANNELS 32U
57#define EFX_MAX_RX_QUEUES EFX_MAX_CHANNELS
58#define EFX_EXTRA_CHANNEL_IOV 0
59#define EFX_EXTRA_CHANNEL_PTP 1
60#define EFX_EXTRA_CHANNEL_TC 2
61#define EFX_MAX_EXTRA_CHANNELS 3U
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_OUTER_CSUM 1 /* Outer checksum offload */
69#define EFX_TXQ_TYPE_INNER_CSUM 2 /* Inner checksum offload */
70#define EFX_TXQ_TYPES 4
71#define EFX_MAX_TXQ_PER_CHANNEL 4
72#define EFX_MAX_TX_QUEUES (EFX_MAX_TXQ_PER_CHANNEL * EFX_MAX_CHANNELS)
73
74/* Maximum possible MTU the driver supports */
75#define EFX_MAX_MTU (9 * 1024)
76
77/* Minimum MTU, from RFC791 (IP) */
78#define EFX_MIN_MTU 68
79
80/* Maximum total header length for TSOv2 */
81#define EFX_TSO2_MAX_HDRLEN 208
82
83/* Size of an RX scatter buffer. Small enough to pack 2 into a 4K page,
84 * and should be a multiple of the cache line size.
85 */
86#define EFX_RX_USR_BUF_SIZE (2048 - 256)
87
88/* If possible, we should ensure cache line alignment at start and end
89 * of every buffer. Otherwise, we just need to ensure 4-byte
90 * alignment of the network header.
91 */
92#if NET_IP_ALIGN == 0
93#define EFX_RX_BUF_ALIGNMENT L1_CACHE_BYTES
94#else
95#define EFX_RX_BUF_ALIGNMENT 4
96#endif
97
98/* Non-standard XDP_PACKET_HEADROOM and tailroom to satisfy XDP_REDIRECT and
99 * still fit two standard MTU size packets into a single 4K page.
100 */
101#define EFX_XDP_HEADROOM 128
102#define EFX_XDP_TAILROOM SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
103
104/* Forward declare Precision Time Protocol (PTP) support structure. */
105struct efx_ptp_data;
106struct hwtstamp_config;
107
108struct efx_self_tests;
109
110/**
111 * struct efx_buffer - A general-purpose DMA buffer
112 * @addr: host base address of the buffer
113 * @dma_addr: DMA base address of the buffer
114 * @len: Buffer length, in bytes
115 *
116 * The NIC uses these buffers for its interrupt status registers and
117 * MAC stats dumps.
118 */
119struct efx_buffer {
120 void *addr;
121 dma_addr_t dma_addr;
122 unsigned int len;
123};
124
125/**
126 * struct efx_tx_buffer - buffer state for a TX descriptor
127 * @skb: When @flags & %EFX_TX_BUF_SKB, the associated socket buffer to be
128 * freed when descriptor completes
129 * @xdpf: When @flags & %EFX_TX_BUF_XDP, the XDP frame information; its @data
130 * member is the associated buffer to drop a page reference on.
131 * @option: When @flags & %EFX_TX_BUF_OPTION, an EF10-specific option
132 * descriptor.
133 * @dma_addr: DMA address of the fragment.
134 * @flags: Flags for allocation and DMA mapping type
135 * @len: Length of this fragment.
136 * This field is zero when the queue slot is empty.
137 * @unmap_len: Length of this fragment to unmap
138 * @dma_offset: Offset of @dma_addr from the address of the backing DMA mapping.
139 * Only valid if @unmap_len != 0.
140 */
141struct efx_tx_buffer {
142 union {
143 const struct sk_buff *skb;
144 struct xdp_frame *xdpf;
145 };
146 union {
147 efx_qword_t option; /* EF10 */
148 dma_addr_t dma_addr;
149 };
150 unsigned short flags;
151 unsigned short len;
152 unsigned short unmap_len;
153 unsigned short dma_offset;
154};
155#define EFX_TX_BUF_CONT 1 /* not last descriptor of packet */
156#define EFX_TX_BUF_SKB 2 /* buffer is last part of skb */
157#define EFX_TX_BUF_MAP_SINGLE 8 /* buffer was mapped with dma_map_single() */
158#define EFX_TX_BUF_OPTION 0x10 /* empty buffer for option descriptor */
159#define EFX_TX_BUF_XDP 0x20 /* buffer was sent with XDP */
160#define EFX_TX_BUF_TSO_V3 0x40 /* empty buffer for a TSO_V3 descriptor */
161#define EFX_TX_BUF_EFV 0x100 /* buffer was sent from representor */
162
163/**
164 * struct efx_tx_queue - An Efx TX queue
165 *
166 * This is a ring buffer of TX fragments.
167 * Since the TX completion path always executes on the same
168 * CPU and the xmit path can operate on different CPUs,
169 * performance is increased by ensuring that the completion
170 * path and the xmit path operate on different cache lines.
171 * This is particularly important if the xmit path is always
172 * executing on one CPU which is different from the completion
173 * path. There is also a cache line for members which are
174 * read but not written on the fast path.
175 *
176 * @efx: The associated Efx NIC
177 * @queue: DMA queue number
178 * @label: Label for TX completion events.
179 * Is our index within @channel->tx_queue array.
180 * @type: configuration type of this TX queue. A bitmask of %EFX_TXQ_TYPE_* flags.
181 * @tso_version: Version of TSO in use for this queue.
182 * @tso_encap: Is encapsulated TSO supported? Supported in TSOv2 on 8000 series.
183 * @channel: The associated channel
184 * @core_txq: The networking core TX queue structure
185 * @buffer: The software buffer ring
186 * @cb_page: Array of pages of copy buffers. Carved up according to
187 * %EFX_TX_CB_ORDER into %EFX_TX_CB_SIZE-sized chunks.
188 * @txd: The hardware descriptor ring
189 * @ptr_mask: The size of the ring minus 1.
190 * @piobuf: PIO buffer region for this TX queue (shared with its partner).
191 * Size of the region is efx_piobuf_size.
192 * @piobuf_offset: Buffer offset to be specified in PIO descriptors
193 * @initialised: Has hardware queue been initialised?
194 * @timestamping: Is timestamping enabled for this channel?
195 * @xdp_tx: Is this an XDP tx queue?
196 * @read_count: Current read pointer.
197 * This is the number of buffers that have been removed from both rings.
198 * @old_write_count: The value of @write_count when last checked.
199 * This is here for performance reasons. The xmit path will
200 * only get the up-to-date value of @write_count if this
201 * variable indicates that the queue is empty. This is to
202 * avoid cache-line ping-pong between the xmit path and the
203 * completion path.
204 * @merge_events: Number of TX merged completion events
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 only written and used on EF10.
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_pending: Are any packets waiting to be pushed to the NIC
233 * @cb_packets: Number of times the TX copybreak feature has been used
234 * @notify_count: Count of notified descriptors to the NIC
235 * @empty_read_count: If the completion path has seen the queue as empty
236 * and the transmission path has not yet checked this, the value of
237 * @read_count bitwise-added to %EFX_EMPTY_COUNT_VALID; otherwise 0.
238 */
239struct efx_tx_queue {
240 /* Members which don't change on the fast path */
241 struct efx_nic *efx ____cacheline_aligned_in_smp;
242 unsigned int queue;
243 unsigned int label;
244 unsigned int type;
245 unsigned int tso_version;
246 bool tso_encap;
247 struct efx_channel *channel;
248 struct netdev_queue *core_txq;
249 struct efx_tx_buffer *buffer;
250 struct efx_buffer *cb_page;
251 struct efx_buffer txd;
252 unsigned int ptr_mask;
253 void __iomem *piobuf;
254 unsigned int piobuf_offset;
255 bool initialised;
256 bool timestamping;
257 bool xdp_tx;
258
259 /* Members used mainly on the completion path */
260 unsigned int read_count ____cacheline_aligned_in_smp;
261 unsigned int old_write_count;
262 unsigned int merge_events;
263 unsigned int bytes_compl;
264 unsigned int pkts_compl;
265 u32 completed_timestamp_major;
266 u32 completed_timestamp_minor;
267
268 /* Members used only on the xmit path */
269 unsigned int insert_count ____cacheline_aligned_in_smp;
270 unsigned int write_count;
271 unsigned int packet_write_count;
272 unsigned int old_read_count;
273 unsigned int tso_bursts;
274 unsigned int tso_long_headers;
275 unsigned int tso_packets;
276 unsigned int tso_fallbacks;
277 unsigned int pushes;
278 unsigned int pio_packets;
279 bool xmit_pending;
280 unsigned int cb_packets;
281 unsigned int notify_count;
282 /* Statistics to supplement MAC stats */
283 unsigned long tx_packets;
284
285 /* Members shared between paths and sometimes updated */
286 unsigned int empty_read_count ____cacheline_aligned_in_smp;
287#define EFX_EMPTY_COUNT_VALID 0x80000000
288 atomic_t flush_outstanding;
289};
290
291#define EFX_TX_CB_ORDER 7
292#define EFX_TX_CB_SIZE (1 << EFX_TX_CB_ORDER) - NET_IP_ALIGN
293
294/**
295 * struct efx_rx_buffer - An Efx RX data buffer
296 * @dma_addr: DMA base address of the buffer
297 * @page: The associated page buffer.
298 * Will be %NULL if the buffer slot is currently free.
299 * @page_offset: If pending: offset in @page of DMA base address.
300 * If completed: offset in @page of Ethernet header.
301 * @len: If pending: length for DMA descriptor.
302 * If completed: received length, excluding hash prefix.
303 * @flags: Flags for buffer and packet state. These are only set on the
304 * first buffer of a scattered packet.
305 */
306struct efx_rx_buffer {
307 dma_addr_t dma_addr;
308 struct page *page;
309 u16 page_offset;
310 u16 len;
311 u16 flags;
312};
313#define EFX_RX_BUF_LAST_IN_PAGE 0x0001
314#define EFX_RX_PKT_CSUMMED 0x0002
315#define EFX_RX_PKT_DISCARD 0x0004
316#define EFX_RX_PKT_TCP 0x0040
317#define EFX_RX_PKT_PREFIX_LEN 0x0080 /* length is in prefix only */
318#define EFX_RX_PKT_CSUM_LEVEL 0x0200
319
320/**
321 * struct efx_rx_page_state - Page-based rx buffer state
322 *
323 * Inserted at the start of every page allocated for receive buffers.
324 * Used to facilitate sharing dma mappings between recycled rx buffers
325 * and those passed up to the kernel.
326 *
327 * @dma_addr: The dma address of this page.
328 */
329struct efx_rx_page_state {
330 dma_addr_t dma_addr;
331
332 unsigned int __pad[] ____cacheline_aligned;
333};
334
335/**
336 * struct efx_rx_queue - An Efx RX queue
337 * @efx: The associated Efx NIC
338 * @core_index: Index of network core RX queue. Will be >= 0 iff this
339 * is associated with a real RX queue.
340 * @buffer: The software buffer ring
341 * @rxd: The hardware descriptor ring
342 * @ptr_mask: The size of the ring minus 1.
343 * @refill_enabled: Enable refill whenever fill level is low
344 * @flush_pending: Set when a RX flush is pending. Has the same lifetime as
345 * @rxq_flush_pending.
346 * @grant_credits: Posted RX descriptors need to be granted to the MAE with
347 * %MC_CMD_MAE_COUNTERS_STREAM_GIVE_CREDITS. For %EFX_EXTRA_CHANNEL_TC,
348 * and only supported on EF100.
349 * @added_count: Number of buffers added to the receive queue.
350 * @notified_count: Number of buffers given to NIC (<= @added_count).
351 * @granted_count: Number of buffers granted to the MAE (<= @notified_count).
352 * @removed_count: Number of buffers removed from the receive queue.
353 * @scatter_n: Used by NIC specific receive code.
354 * @scatter_len: Used by NIC specific receive code.
355 * @page_ring: The ring to store DMA mapped pages for reuse.
356 * @page_add: Counter to calculate the write pointer for the recycle ring.
357 * @page_remove: Counter to calculate the read pointer for the recycle ring.
358 * @page_recycle_count: The number of pages that have been recycled.
359 * @page_recycle_failed: The number of pages that couldn't be recycled because
360 * the kernel still held a reference to them.
361 * @page_recycle_full: The number of pages that were released because the
362 * recycle ring was full.
363 * @page_ptr_mask: The number of pages in the RX recycle ring minus 1.
364 * @max_fill: RX descriptor maximum fill level (<= ring size)
365 * @fast_fill_trigger: RX descriptor fill level that will trigger a fast fill
366 * (<= @max_fill)
367 * @min_fill: RX descriptor minimum non-zero fill level.
368 * This records the minimum fill level observed when a ring
369 * refill was triggered.
370 * @recycle_count: RX buffer recycle counter.
371 * @slow_fill: Timer used to defer efx_nic_generate_fill_event().
372 * @grant_work: workitem used to grant credits to the MAE if @grant_credits
373 * @xdp_rxq_info: XDP specific RX queue information.
374 * @xdp_rxq_info_valid: Is xdp_rxq_info valid data?.
375 */
376struct efx_rx_queue {
377 struct efx_nic *efx;
378 int core_index;
379 struct efx_rx_buffer *buffer;
380 struct efx_buffer rxd;
381 unsigned int ptr_mask;
382 bool refill_enabled;
383 bool flush_pending;
384 bool grant_credits;
385
386 unsigned int added_count;
387 unsigned int notified_count;
388 unsigned int granted_count;
389 unsigned int removed_count;
390 unsigned int scatter_n;
391 unsigned int scatter_len;
392 struct page **page_ring;
393 unsigned int page_add;
394 unsigned int page_remove;
395 unsigned int page_recycle_count;
396 unsigned int page_recycle_failed;
397 unsigned int page_recycle_full;
398 unsigned int page_ptr_mask;
399 unsigned int max_fill;
400 unsigned int fast_fill_trigger;
401 unsigned int min_fill;
402 unsigned int min_overfill;
403 unsigned int recycle_count;
404 struct timer_list slow_fill;
405 unsigned int slow_fill_count;
406 struct work_struct grant_work;
407 /* Statistics to supplement MAC stats */
408 unsigned long rx_packets;
409 struct xdp_rxq_info xdp_rxq_info;
410 bool xdp_rxq_info_valid;
411};
412
413enum efx_sync_events_state {
414 SYNC_EVENTS_DISABLED = 0,
415 SYNC_EVENTS_QUIESCENT,
416 SYNC_EVENTS_REQUESTED,
417 SYNC_EVENTS_VALID,
418};
419
420/**
421 * struct efx_channel - An Efx channel
422 *
423 * A channel comprises an event queue, at least one TX queue, at least
424 * one RX queue, and an associated tasklet for processing the event
425 * queue.
426 *
427 * @efx: Associated Efx NIC
428 * @channel: Channel instance number
429 * @type: Channel type definition
430 * @eventq_init: Event queue initialised flag
431 * @enabled: Channel enabled indicator
432 * @irq: IRQ number (MSI and MSI-X only)
433 * @irq_moderation_us: IRQ moderation value (in microseconds)
434 * @napi_dev: Net device used with NAPI
435 * @napi_str: NAPI control structure
436 * @state: state for NAPI vs busy polling
437 * @state_lock: lock protecting @state
438 * @eventq: Event queue buffer
439 * @eventq_mask: Event queue pointer mask
440 * @eventq_read_ptr: Event queue read pointer
441 * @event_test_cpu: Last CPU to handle interrupt or test event for this channel
442 * @irq_count: Number of IRQs since last adaptive moderation decision
443 * @irq_mod_score: IRQ moderation score
444 * @rfs_filter_count: number of accelerated RFS filters currently in place;
445 * equals the count of @rps_flow_id slots filled
446 * @rfs_last_expiry: value of jiffies last time some accelerated RFS filters
447 * were checked for expiry
448 * @rfs_expire_index: next accelerated RFS filter ID to check for expiry
449 * @n_rfs_succeeded: number of successful accelerated RFS filter insertions
450 * @n_rfs_failed: number of failed accelerated RFS filter insertions
451 * @filter_work: Work item for efx_filter_rfs_expire()
452 * @rps_flow_id: Flow IDs of filters allocated for accelerated RFS,
453 * indexed by filter ID
454 * @n_rx_tobe_disc: Count of RX_TOBE_DISC errors
455 * @n_rx_ip_hdr_chksum_err: Count of RX IP header checksum errors
456 * @n_rx_tcp_udp_chksum_err: Count of RX TCP and UDP checksum errors
457 * @n_rx_mcast_mismatch: Count of unmatched multicast frames
458 * @n_rx_frm_trunc: Count of RX_FRM_TRUNC errors
459 * @n_rx_overlength: Count of RX_OVERLENGTH errors
460 * @n_skbuff_leaks: Count of skbuffs leaked due to RX overrun
461 * @n_rx_nodesc_trunc: Number of RX packets truncated and then dropped due to
462 * lack of descriptors
463 * @n_rx_merge_events: Number of RX merged completion events
464 * @n_rx_merge_packets: Number of RX packets completed by merged events
465 * @n_rx_xdp_drops: Count of RX packets intentionally dropped due to XDP
466 * @n_rx_xdp_bad_drops: Count of RX packets dropped due to XDP errors
467 * @n_rx_xdp_tx: Count of RX packets retransmitted due to XDP
468 * @n_rx_xdp_redirect: Count of RX packets redirected to a different NIC by XDP
469 * @n_rx_mport_bad: Count of RX packets dropped because their ingress mport was
470 * not recognised
471 * @rx_pkt_n_frags: Number of fragments in next packet to be delivered by
472 * __efx_rx_packet(), or zero if there is none
473 * @rx_pkt_index: Ring index of first buffer for next packet to be delivered
474 * by __efx_rx_packet(), if @rx_pkt_n_frags != 0
475 * @rx_list: list of SKBs from current RX, awaiting processing
476 * @rx_queue: RX queue for this channel
477 * @tx_queue: TX queues for this channel
478 * @tx_queue_by_type: pointers into @tx_queue, or %NULL, indexed by txq type
479 * @sync_events_state: Current state of sync events on this channel
480 * @sync_timestamp_major: Major part of the last ptp sync event
481 * @sync_timestamp_minor: Minor part of the last ptp sync event
482 */
483struct efx_channel {
484 struct efx_nic *efx;
485 int channel;
486 const struct efx_channel_type *type;
487 bool eventq_init;
488 bool enabled;
489 int irq;
490 unsigned int irq_moderation_us;
491 struct net_device *napi_dev;
492 struct napi_struct napi_str;
493#ifdef CONFIG_NET_RX_BUSY_POLL
494 unsigned long busy_poll_state;
495#endif
496 struct efx_buffer eventq;
497 unsigned int eventq_mask;
498 unsigned int eventq_read_ptr;
499 int event_test_cpu;
500
501 unsigned int irq_count;
502 unsigned int irq_mod_score;
503#ifdef CONFIG_RFS_ACCEL
504 unsigned int rfs_filter_count;
505 unsigned int rfs_last_expiry;
506 unsigned int rfs_expire_index;
507 unsigned int n_rfs_succeeded;
508 unsigned int n_rfs_failed;
509 struct delayed_work filter_work;
510#define RPS_FLOW_ID_INVALID 0xFFFFFFFF
511 u32 *rps_flow_id;
512#endif
513
514 unsigned int n_rx_tobe_disc;
515 unsigned int n_rx_ip_hdr_chksum_err;
516 unsigned int n_rx_tcp_udp_chksum_err;
517 unsigned int n_rx_outer_ip_hdr_chksum_err;
518 unsigned int n_rx_outer_tcp_udp_chksum_err;
519 unsigned int n_rx_inner_ip_hdr_chksum_err;
520 unsigned int n_rx_inner_tcp_udp_chksum_err;
521 unsigned int n_rx_eth_crc_err;
522 unsigned int n_rx_mcast_mismatch;
523 unsigned int n_rx_frm_trunc;
524 unsigned int n_rx_overlength;
525 unsigned int n_skbuff_leaks;
526 unsigned int n_rx_nodesc_trunc;
527 unsigned int n_rx_merge_events;
528 unsigned int n_rx_merge_packets;
529 unsigned int n_rx_xdp_drops;
530 unsigned int n_rx_xdp_bad_drops;
531 unsigned int n_rx_xdp_tx;
532 unsigned int n_rx_xdp_redirect;
533 unsigned int n_rx_mport_bad;
534
535 unsigned int rx_pkt_n_frags;
536 unsigned int rx_pkt_index;
537
538 struct list_head *rx_list;
539
540 struct efx_rx_queue rx_queue;
541 struct efx_tx_queue tx_queue[EFX_MAX_TXQ_PER_CHANNEL];
542 struct efx_tx_queue *tx_queue_by_type[EFX_TXQ_TYPES];
543
544 enum efx_sync_events_state sync_events_state;
545 u32 sync_timestamp_major;
546 u32 sync_timestamp_minor;
547};
548
549/**
550 * struct efx_msi_context - Context for each MSI
551 * @efx: The associated NIC
552 * @index: Index of the channel/IRQ
553 * @name: Name of the channel/IRQ
554 *
555 * Unlike &struct efx_channel, this is never reallocated and is always
556 * safe for the IRQ handler to access.
557 */
558struct efx_msi_context {
559 struct efx_nic *efx;
560 unsigned int index;
561 char name[IFNAMSIZ + 6];
562};
563
564/**
565 * struct efx_channel_type - distinguishes traffic and extra channels
566 * @handle_no_channel: Handle failure to allocate an extra channel
567 * @pre_probe: Set up extra state prior to initialisation
568 * @start: called early in efx_start_channels()
569 * @stop: called early in efx_stop_channels()
570 * @post_remove: Tear down extra state after finalisation, if allocated.
571 * May be called on channels that have not been probed.
572 * @get_name: Generate the channel's name (used for its IRQ handler)
573 * @copy: Copy the channel state prior to reallocation. May be %NULL if
574 * reallocation is not supported.
575 * @receive_skb: Handle an skb ready to be passed to netif_receive_skb()
576 * @receive_raw: Handle an RX buffer ready to be passed to __efx_rx_packet()
577 * @want_txqs: Determine whether this channel should have TX queues
578 * created. If %NULL, TX queues are not created.
579 * @keep_eventq: Flag for whether event queue should be kept initialised
580 * while the device is stopped
581 * @want_pio: Flag for whether PIO buffers should be linked to this
582 * channel's TX queues.
583 */
584struct efx_channel_type {
585 void (*handle_no_channel)(struct efx_nic *);
586 int (*pre_probe)(struct efx_channel *);
587 int (*start)(struct efx_channel *);
588 void (*stop)(struct efx_channel *);
589 void (*post_remove)(struct efx_channel *);
590 void (*get_name)(struct efx_channel *, char *buf, size_t len);
591 struct efx_channel *(*copy)(const struct efx_channel *);
592 bool (*receive_skb)(struct efx_channel *, struct sk_buff *);
593 bool (*receive_raw)(struct efx_rx_queue *, u32);
594 bool (*want_txqs)(struct efx_channel *);
595 bool keep_eventq;
596 bool want_pio;
597};
598
599enum efx_led_mode {
600 EFX_LED_OFF = 0,
601 EFX_LED_ON = 1,
602 EFX_LED_DEFAULT = 2
603};
604
605#define STRING_TABLE_LOOKUP(val, member) \
606 ((val) < member ## _max) ? member ## _names[val] : "(invalid)"
607
608extern const char *const efx_loopback_mode_names[];
609extern const unsigned int efx_loopback_mode_max;
610#define LOOPBACK_MODE(efx) \
611 STRING_TABLE_LOOKUP((efx)->loopback_mode, efx_loopback_mode)
612
613enum efx_int_mode {
614 /* Be careful if altering to correct macro below */
615 EFX_INT_MODE_MSIX = 0,
616 EFX_INT_MODE_MSI = 1,
617 EFX_INT_MODE_LEGACY = 2,
618 EFX_INT_MODE_MAX /* Insert any new items before this */
619};
620#define EFX_INT_MODE_USE_MSI(x) (((x)->interrupt_mode) <= EFX_INT_MODE_MSI)
621
622enum nic_state {
623 STATE_UNINIT = 0, /* device being probed/removed */
624 STATE_PROBED, /* hardware probed */
625 STATE_NET_DOWN, /* netdev registered */
626 STATE_NET_UP, /* ready for traffic */
627 STATE_DISABLED, /* device disabled due to hardware errors */
628
629 STATE_RECOVERY = 0x100,/* recovering from PCI error */
630 STATE_FROZEN = 0x200, /* frozen by power management */
631};
632
633static inline bool efx_net_active(enum nic_state state)
634{
635 return state == STATE_NET_DOWN || state == STATE_NET_UP;
636}
637
638static inline bool efx_frozen(enum nic_state state)
639{
640 return state & STATE_FROZEN;
641}
642
643static inline bool efx_recovering(enum nic_state state)
644{
645 return state & STATE_RECOVERY;
646}
647
648static inline enum nic_state efx_freeze(enum nic_state state)
649{
650 WARN_ON(!efx_net_active(state));
651 return state | STATE_FROZEN;
652}
653
654static inline enum nic_state efx_thaw(enum nic_state state)
655{
656 WARN_ON(!efx_frozen(state));
657 return state & ~STATE_FROZEN;
658}
659
660static inline enum nic_state efx_recover(enum nic_state state)
661{
662 WARN_ON(!efx_net_active(state));
663 return state | STATE_RECOVERY;
664}
665
666static inline enum nic_state efx_recovered(enum nic_state state)
667{
668 WARN_ON(!efx_recovering(state));
669 return state & ~STATE_RECOVERY;
670}
671
672/* Forward declaration */
673struct efx_nic;
674
675/* Pseudo bit-mask flow control field */
676#define EFX_FC_RX FLOW_CTRL_RX
677#define EFX_FC_TX FLOW_CTRL_TX
678#define EFX_FC_AUTO 4
679
680/**
681 * struct efx_link_state - Current state of the link
682 * @up: Link is up
683 * @fd: Link is full-duplex
684 * @fc: Actual flow control flags
685 * @speed: Link speed (Mbps)
686 */
687struct efx_link_state {
688 bool up;
689 bool fd;
690 u8 fc;
691 unsigned int speed;
692};
693
694static inline bool efx_link_state_equal(const struct efx_link_state *left,
695 const struct efx_link_state *right)
696{
697 return left->up == right->up && left->fd == right->fd &&
698 left->fc == right->fc && left->speed == right->speed;
699}
700
701/**
702 * enum efx_phy_mode - PHY operating mode flags
703 * @PHY_MODE_NORMAL: on and should pass traffic
704 * @PHY_MODE_TX_DISABLED: on with TX disabled
705 * @PHY_MODE_LOW_POWER: set to low power through MDIO
706 * @PHY_MODE_OFF: switched off through external control
707 * @PHY_MODE_SPECIAL: on but will not pass traffic
708 */
709enum efx_phy_mode {
710 PHY_MODE_NORMAL = 0,
711 PHY_MODE_TX_DISABLED = 1,
712 PHY_MODE_LOW_POWER = 2,
713 PHY_MODE_OFF = 4,
714 PHY_MODE_SPECIAL = 8,
715};
716
717static inline bool efx_phy_mode_disabled(enum efx_phy_mode mode)
718{
719 return !!(mode & ~PHY_MODE_TX_DISABLED);
720}
721
722/**
723 * struct efx_hw_stat_desc - Description of a hardware statistic
724 * @name: Name of the statistic as visible through ethtool, or %NULL if
725 * it should not be exposed
726 * @dma_width: Width in bits (0 for non-DMA statistics)
727 * @offset: Offset within stats (ignored for non-DMA statistics)
728 */
729struct efx_hw_stat_desc {
730 const char *name;
731 u16 dma_width;
732 u16 offset;
733};
734
735struct vfdi_status;
736
737/* The reserved RSS context value */
738#define EFX_MCDI_RSS_CONTEXT_INVALID 0xffffffff
739/**
740 * struct efx_rss_context - A user-defined RSS context for filtering
741 * @list: node of linked list on which this struct is stored
742 * @context_id: the RSS_CONTEXT_ID returned by MC firmware, or
743 * %EFX_MCDI_RSS_CONTEXT_INVALID if this context is not present on the NIC.
744 * For Siena, 0 if RSS is active, else %EFX_MCDI_RSS_CONTEXT_INVALID.
745 * @user_id: the rss_context ID exposed to userspace over ethtool.
746 * @rx_hash_udp_4tuple: UDP 4-tuple hashing enabled
747 * @rx_hash_key: Toeplitz hash key for this RSS context
748 * @indir_table: Indirection table for this RSS context
749 */
750struct efx_rss_context {
751 struct list_head list;
752 u32 context_id;
753 u32 user_id;
754 bool rx_hash_udp_4tuple;
755 u8 rx_hash_key[40];
756 u32 rx_indir_table[128];
757};
758
759#ifdef CONFIG_RFS_ACCEL
760/* Order of these is important, since filter_id >= %EFX_ARFS_FILTER_ID_PENDING
761 * is used to test if filter does or will exist.
762 */
763#define EFX_ARFS_FILTER_ID_PENDING -1
764#define EFX_ARFS_FILTER_ID_ERROR -2
765#define EFX_ARFS_FILTER_ID_REMOVING -3
766/**
767 * struct efx_arfs_rule - record of an ARFS filter and its IDs
768 * @node: linkage into hash table
769 * @spec: details of the filter (used as key for hash table). Use efx->type to
770 * determine which member to use.
771 * @rxq_index: channel to which the filter will steer traffic.
772 * @arfs_id: filter ID which was returned to ARFS
773 * @filter_id: index in software filter table. May be
774 * %EFX_ARFS_FILTER_ID_PENDING if filter was not inserted yet,
775 * %EFX_ARFS_FILTER_ID_ERROR if filter insertion failed, or
776 * %EFX_ARFS_FILTER_ID_REMOVING if expiry is currently removing the filter.
777 */
778struct efx_arfs_rule {
779 struct hlist_node node;
780 struct efx_filter_spec spec;
781 u16 rxq_index;
782 u16 arfs_id;
783 s32 filter_id;
784};
785
786/* Size chosen so that the table is one page (4kB) */
787#define EFX_ARFS_HASH_TABLE_SIZE 512
788
789/**
790 * struct efx_async_filter_insertion - Request to asynchronously insert a filter
791 * @net_dev: Reference to the netdevice
792 * @spec: The filter to insert
793 * @work: Workitem for this request
794 * @rxq_index: Identifies the channel for which this request was made
795 * @flow_id: Identifies the kernel-side flow for which this request was made
796 */
797struct efx_async_filter_insertion {
798 struct net_device *net_dev;
799 struct efx_filter_spec spec;
800 struct work_struct work;
801 u16 rxq_index;
802 u32 flow_id;
803};
804
805/* Maximum number of ARFS workitems that may be in flight on an efx_nic */
806#define EFX_RPS_MAX_IN_FLIGHT 8
807#endif /* CONFIG_RFS_ACCEL */
808
809enum efx_xdp_tx_queues_mode {
810 EFX_XDP_TX_QUEUES_DEDICATED, /* one queue per core, locking not needed */
811 EFX_XDP_TX_QUEUES_SHARED, /* each queue used by more than 1 core */
812 EFX_XDP_TX_QUEUES_BORROWED /* queues borrowed from net stack */
813};
814
815struct efx_mae;
816
817/**
818 * struct efx_nic - an Efx NIC
819 * @name: Device name (net device name or bus id before net device registered)
820 * @pci_dev: The PCI device
821 * @node: List node for maintaning primary/secondary function lists
822 * @primary: &struct efx_nic instance for the primary function of this
823 * controller. May be the same structure, and may be %NULL if no
824 * primary function is bound. Serialised by rtnl_lock.
825 * @secondary_list: List of &struct efx_nic instances for the secondary PCI
826 * functions of the controller, if this is for the primary function.
827 * Serialised by rtnl_lock.
828 * @type: Controller type attributes
829 * @legacy_irq: IRQ number
830 * @workqueue: Workqueue for port reconfigures and the HW monitor.
831 * Work items do not hold and must not acquire RTNL.
832 * @workqueue_name: Name of workqueue
833 * @reset_work: Scheduled reset workitem
834 * @membase_phys: Memory BAR value as physical address
835 * @membase: Memory BAR value
836 * @vi_stride: step between per-VI registers / memory regions
837 * @interrupt_mode: Interrupt mode
838 * @timer_quantum_ns: Interrupt timer quantum, in nanoseconds
839 * @timer_max_ns: Interrupt timer maximum value, in nanoseconds
840 * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues
841 * @irqs_hooked: Channel interrupts are hooked
842 * @irq_rx_mod_step_us: Step size for IRQ moderation for RX event queues
843 * @irq_rx_moderation_us: IRQ moderation time for RX event queues
844 * @msg_enable: Log message enable flags
845 * @state: Device state number (%STATE_*). Serialised by the rtnl_lock.
846 * @reset_pending: Bitmask for pending resets
847 * @tx_queue: TX DMA queues
848 * @rx_queue: RX DMA queues
849 * @channel: Channels
850 * @msi_context: Context for each MSI
851 * @extra_channel_types: Types of extra (non-traffic) channels that
852 * should be allocated for this NIC
853 * @mae: Details of the Match Action Engine
854 * @xdp_tx_queue_count: Number of entries in %xdp_tx_queues.
855 * @xdp_tx_queues: Array of pointers to tx queues used for XDP transmit.
856 * @xdp_txq_queues_mode: XDP TX queues sharing strategy.
857 * @rxq_entries: Size of receive queues requested by user.
858 * @txq_entries: Size of transmit queues requested by user.
859 * @txq_stop_thresh: TX queue fill level at or above which we stop it.
860 * @txq_wake_thresh: TX queue fill level at or below which we wake it.
861 * @tx_dc_base: Base qword address in SRAM of TX queue descriptor caches
862 * @rx_dc_base: Base qword address in SRAM of RX queue descriptor caches
863 * @sram_lim_qw: Qword address limit of SRAM
864 * @n_channels: Number of channels in use
865 * @n_rx_channels: Number of channels used for RX (= number of RX queues)
866 * @n_tx_channels: Number of channels used for TX
867 * @n_extra_tx_channels: Number of extra channels with TX queues
868 * @tx_queues_per_channel: number of TX queues probed on each channel
869 * @n_xdp_channels: Number of channels used for XDP TX
870 * @xdp_channel_offset: Offset of zeroth channel used for XPD TX.
871 * @xdp_tx_per_channel: Max number of TX queues on an XDP TX channel.
872 * @rx_ip_align: RX DMA address offset to have IP header aligned in
873 * in accordance with NET_IP_ALIGN
874 * @rx_dma_len: Current maximum RX DMA length
875 * @rx_buffer_order: Order (log2) of number of pages for each RX buffer
876 * @rx_buffer_truesize: Amortised allocation size of an RX buffer,
877 * for use in sk_buff::truesize
878 * @rx_prefix_size: Size of RX prefix before packet data
879 * @rx_packet_hash_offset: Offset of RX flow hash from start of packet data
880 * (valid only if @rx_prefix_size != 0; always negative)
881 * @rx_packet_len_offset: Offset of RX packet length from start of packet data
882 * (valid only for NICs that set %EFX_RX_PKT_PREFIX_LEN; always negative)
883 * @rx_packet_ts_offset: Offset of timestamp from start of packet data
884 * (valid only if channel->sync_timestamps_enabled; always negative)
885 * @rx_scatter: Scatter mode enabled for receives
886 * @rss_context: Main RSS context. Its @list member is the head of the list of
887 * RSS contexts created by user requests
888 * @rss_lock: Protects custom RSS context software state in @rss_context.list
889 * @vport_id: The function's vport ID, only relevant for PFs
890 * @int_error_count: Number of internal errors seen recently
891 * @int_error_expire: Time at which error count will be expired
892 * @must_realloc_vis: Flag: VIs have yet to be reallocated after MC reboot
893 * @irq_soft_enabled: Are IRQs soft-enabled? If not, IRQ handler will
894 * acknowledge but do nothing else.
895 * @irq_status: Interrupt status buffer
896 * @irq_zero_count: Number of legacy IRQs seen with queue flags == 0
897 * @irq_level: IRQ level/index for IRQs not triggered by an event queue
898 * @selftest_work: Work item for asynchronous self-test
899 * @mtd_list: List of MTDs attached to the NIC
900 * @nic_data: Hardware dependent state
901 * @mcdi: Management-Controller-to-Driver Interface state
902 * @mac_lock: MAC access lock. Protects @port_enabled, @phy_mode,
903 * efx_monitor() and efx_reconfigure_port()
904 * @port_enabled: Port enabled indicator.
905 * Serialises efx_stop_all(), efx_start_all(), efx_monitor() and
906 * efx_mac_work() with kernel interfaces. Safe to read under any
907 * one of the rtnl_lock, mac_lock, or netif_tx_lock, but all three must
908 * be held to modify it.
909 * @port_initialized: Port initialized?
910 * @net_dev: Operating system network device. Consider holding the rtnl lock
911 * @fixed_features: Features which cannot be turned off
912 * @num_mac_stats: Number of MAC stats reported by firmware (MAC_STATS_NUM_STATS
913 * field of %MC_CMD_GET_CAPABILITIES_V4 response, or %MC_CMD_MAC_NSTATS)
914 * @stats_buffer: DMA buffer for statistics
915 * @phy_type: PHY type
916 * @phy_data: PHY private data (including PHY-specific stats)
917 * @mdio: PHY MDIO interface
918 * @mdio_bus: PHY MDIO bus ID (only used by Siena)
919 * @phy_mode: PHY operating mode. Serialised by @mac_lock.
920 * @link_advertising: Autonegotiation advertising flags
921 * @fec_config: Forward Error Correction configuration flags. For bit positions
922 * see &enum ethtool_fec_config_bits.
923 * @link_state: Current state of the link
924 * @n_link_state_changes: Number of times the link has changed state
925 * @wanted_fc: Wanted flow control flags
926 * @fc_disable: When non-zero flow control is disabled. Typically used to
927 * ensure that network back pressure doesn't delay dma queue flushes.
928 * Serialised by the rtnl lock.
929 * @mac_work: Work item for changing MAC promiscuity and multicast hash
930 * @loopback_mode: Loopback status
931 * @loopback_modes: Supported loopback mode bitmask
932 * @loopback_selftest: Offline self-test private state
933 * @xdp_prog: Current XDP programme for this interface
934 * @filter_sem: Filter table rw_semaphore, protects existence of @filter_state
935 * @filter_state: Architecture-dependent filter table state
936 * @rps_mutex: Protects RPS state of all channels
937 * @rps_slot_map: bitmap of in-flight entries in @rps_slot
938 * @rps_slot: array of ARFS insertion requests for efx_filter_rfs_work()
939 * @rps_hash_lock: Protects ARFS filter mapping state (@rps_hash_table and
940 * @rps_next_id).
941 * @rps_hash_table: Mapping between ARFS filters and their various IDs
942 * @rps_next_id: next arfs_id for an ARFS filter
943 * @active_queues: Count of RX and TX queues that haven't been flushed and drained.
944 * @rxq_flush_pending: Count of number of receive queues that need to be flushed.
945 * Decremented when the efx_flush_rx_queue() is called.
946 * @rxq_flush_outstanding: Count of number of RX flushes started but not yet
947 * completed (either success or failure). Not used when MCDI is used to
948 * flush receive queues.
949 * @flush_wq: wait queue used by efx_nic_flush_queues() to wait for flush completions.
950 * @vf_count: Number of VFs intended to be enabled.
951 * @vf_init_count: Number of VFs that have been fully initialised.
952 * @vi_scale: log2 number of vnics per VF.
953 * @vf_reps_lock: Protects vf_reps list
954 * @vf_reps: local VF reps
955 * @ptp_data: PTP state data
956 * @ptp_warned: has this NIC seen and warned about unexpected PTP events?
957 * @vpd_sn: Serial number read from VPD
958 * @xdp_rxq_info_failed: Have any of the rx queues failed to initialise their
959 * xdp_rxq_info structures?
960 * @netdev_notifier: Netdevice notifier.
961 * @netevent_notifier: Netevent notifier (for neighbour updates).
962 * @tc: state for TC offload (EF100).
963 * @devlink: reference to devlink structure owned by this device
964 * @dl_port: devlink port associated with the PF
965 * @mem_bar: The BAR that is mapped into membase.
966 * @reg_base: Offset from the start of the bar to the function control window.
967 * @monitor_work: Hardware monitor workitem
968 * @biu_lock: BIU (bus interface unit) lock
969 * @last_irq_cpu: Last CPU to handle a possible test interrupt. This
970 * field is used by efx_test_interrupts() to verify that an
971 * interrupt has occurred.
972 * @stats_lock: Statistics update lock. Must be held when calling
973 * efx_nic_type::{update,start,stop}_stats.
974 * @n_rx_noskb_drops: Count of RX packets dropped due to failure to allocate an skb
975 *
976 * This is stored in the private area of the &struct net_device.
977 */
978struct efx_nic {
979 /* The following fields should be written very rarely */
980
981 char name[IFNAMSIZ];
982 struct list_head node;
983 struct efx_nic *primary;
984 struct list_head secondary_list;
985 struct pci_dev *pci_dev;
986 unsigned int port_num;
987 const struct efx_nic_type *type;
988 int legacy_irq;
989 bool eeh_disabled_legacy_irq;
990 struct workqueue_struct *workqueue;
991 char workqueue_name[16];
992 struct work_struct reset_work;
993 resource_size_t membase_phys;
994 void __iomem *membase;
995
996 unsigned int vi_stride;
997
998 enum efx_int_mode interrupt_mode;
999 unsigned int timer_quantum_ns;
1000 unsigned int timer_max_ns;
1001 bool irq_rx_adaptive;
1002 bool irqs_hooked;
1003 unsigned int irq_mod_step_us;
1004 unsigned int irq_rx_moderation_us;
1005 u32 msg_enable;
1006
1007 enum nic_state state;
1008 unsigned long reset_pending;
1009
1010 struct efx_channel *channel[EFX_MAX_CHANNELS];
1011 struct efx_msi_context msi_context[EFX_MAX_CHANNELS];
1012 const struct efx_channel_type *
1013 extra_channel_type[EFX_MAX_EXTRA_CHANNELS];
1014 struct efx_mae *mae;
1015
1016 unsigned int xdp_tx_queue_count;
1017 struct efx_tx_queue **xdp_tx_queues;
1018 enum efx_xdp_tx_queues_mode xdp_txq_queues_mode;
1019
1020 unsigned rxq_entries;
1021 unsigned txq_entries;
1022 unsigned int txq_stop_thresh;
1023 unsigned int txq_wake_thresh;
1024
1025 unsigned tx_dc_base;
1026 unsigned rx_dc_base;
1027 unsigned sram_lim_qw;
1028
1029 unsigned int max_channels;
1030 unsigned int max_vis;
1031 unsigned int max_tx_channels;
1032 unsigned n_channels;
1033 unsigned n_rx_channels;
1034 unsigned rss_spread;
1035 unsigned tx_channel_offset;
1036 unsigned n_tx_channels;
1037 unsigned n_extra_tx_channels;
1038 unsigned int tx_queues_per_channel;
1039 unsigned int n_xdp_channels;
1040 unsigned int xdp_channel_offset;
1041 unsigned int xdp_tx_per_channel;
1042 unsigned int rx_ip_align;
1043 unsigned int rx_dma_len;
1044 unsigned int rx_buffer_order;
1045 unsigned int rx_buffer_truesize;
1046 unsigned int rx_page_buf_step;
1047 unsigned int rx_bufs_per_page;
1048 unsigned int rx_pages_per_batch;
1049 unsigned int rx_prefix_size;
1050 int rx_packet_hash_offset;
1051 int rx_packet_len_offset;
1052 int rx_packet_ts_offset;
1053 bool rx_scatter;
1054 struct efx_rss_context rss_context;
1055 struct mutex rss_lock;
1056 u32 vport_id;
1057
1058 unsigned int_error_count;
1059 unsigned long int_error_expire;
1060
1061 bool must_realloc_vis;
1062 bool irq_soft_enabled;
1063 struct efx_buffer irq_status;
1064 unsigned irq_zero_count;
1065 unsigned irq_level;
1066 struct delayed_work selftest_work;
1067
1068#ifdef CONFIG_SFC_MTD
1069 struct list_head mtd_list;
1070#endif
1071
1072 void *nic_data;
1073 struct efx_mcdi_data *mcdi;
1074
1075 struct mutex mac_lock;
1076 struct work_struct mac_work;
1077 bool port_enabled;
1078
1079 bool mc_bist_for_other_fn;
1080 bool port_initialized;
1081 struct net_device *net_dev;
1082
1083 netdev_features_t fixed_features;
1084
1085 u16 num_mac_stats;
1086 struct efx_buffer stats_buffer;
1087 u64 rx_nodesc_drops_total;
1088 u64 rx_nodesc_drops_while_down;
1089 bool rx_nodesc_drops_prev_state;
1090
1091 unsigned int phy_type;
1092 void *phy_data;
1093 struct mdio_if_info mdio;
1094 unsigned int mdio_bus;
1095 enum efx_phy_mode phy_mode;
1096
1097 __ETHTOOL_DECLARE_LINK_MODE_MASK(link_advertising);
1098 u32 fec_config;
1099 struct efx_link_state link_state;
1100 unsigned int n_link_state_changes;
1101
1102 u8 wanted_fc;
1103 unsigned fc_disable;
1104
1105 atomic_t rx_reset;
1106 enum efx_loopback_mode loopback_mode;
1107 u64 loopback_modes;
1108
1109 void *loopback_selftest;
1110 /* We access loopback_selftest immediately before running XDP,
1111 * so we want them next to each other.
1112 */
1113 struct bpf_prog __rcu *xdp_prog;
1114
1115 struct rw_semaphore filter_sem;
1116 void *filter_state;
1117#ifdef CONFIG_RFS_ACCEL
1118 struct mutex rps_mutex;
1119 unsigned long rps_slot_map;
1120 struct efx_async_filter_insertion rps_slot[EFX_RPS_MAX_IN_FLIGHT];
1121 spinlock_t rps_hash_lock;
1122 struct hlist_head *rps_hash_table;
1123 u32 rps_next_id;
1124#endif
1125
1126 atomic_t active_queues;
1127 atomic_t rxq_flush_pending;
1128 atomic_t rxq_flush_outstanding;
1129 wait_queue_head_t flush_wq;
1130
1131#ifdef CONFIG_SFC_SRIOV
1132 unsigned vf_count;
1133 unsigned vf_init_count;
1134 unsigned vi_scale;
1135#endif
1136 spinlock_t vf_reps_lock;
1137 struct list_head vf_reps;
1138
1139 struct efx_ptp_data *ptp_data;
1140 bool ptp_warned;
1141
1142 char *vpd_sn;
1143 bool xdp_rxq_info_failed;
1144
1145 struct notifier_block netdev_notifier;
1146 struct notifier_block netevent_notifier;
1147 struct efx_tc_state *tc;
1148
1149 struct devlink *devlink;
1150 struct devlink_port *dl_port;
1151 unsigned int mem_bar;
1152 u32 reg_base;
1153
1154 /* The following fields may be written more often */
1155
1156 struct delayed_work monitor_work ____cacheline_aligned_in_smp;
1157 spinlock_t biu_lock;
1158 int last_irq_cpu;
1159 spinlock_t stats_lock;
1160 atomic_t n_rx_noskb_drops;
1161};
1162
1163/**
1164 * struct efx_probe_data - State after hardware probe
1165 * @pci_dev: The PCI device
1166 * @efx: Efx NIC details
1167 */
1168struct efx_probe_data {
1169 struct pci_dev *pci_dev;
1170 struct efx_nic efx;
1171};
1172
1173static inline struct efx_nic *efx_netdev_priv(struct net_device *dev)
1174{
1175 struct efx_probe_data **probe_ptr = netdev_priv(dev);
1176 struct efx_probe_data *probe_data = *probe_ptr;
1177
1178 return &probe_data->efx;
1179}
1180
1181static inline int efx_dev_registered(struct efx_nic *efx)
1182{
1183 return efx->net_dev->reg_state == NETREG_REGISTERED;
1184}
1185
1186static inline unsigned int efx_port_num(struct efx_nic *efx)
1187{
1188 return efx->port_num;
1189}
1190
1191struct efx_mtd_partition {
1192 struct list_head node;
1193 struct mtd_info mtd;
1194 const char *dev_type_name;
1195 const char *type_name;
1196 char name[IFNAMSIZ + 20];
1197};
1198
1199struct efx_udp_tunnel {
1200#define TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID 0xffff
1201 u16 type; /* TUNNEL_ENCAP_UDP_PORT_ENTRY_foo, see mcdi_pcol.h */
1202 __be16 port;
1203};
1204
1205/**
1206 * struct efx_nic_type - Efx device type definition
1207 * @mem_bar: Get the memory BAR
1208 * @mem_map_size: Get memory BAR mapped size
1209 * @probe: Probe the controller
1210 * @remove: Free resources allocated by probe()
1211 * @init: Initialise the controller
1212 * @dimension_resources: Dimension controller resources (buffer table,
1213 * and VIs once the available interrupt resources are clear)
1214 * @fini: Shut down the controller
1215 * @monitor: Periodic function for polling link state and hardware monitor
1216 * @map_reset_reason: Map ethtool reset reason to a reset method
1217 * @map_reset_flags: Map ethtool reset flags to a reset method, if possible
1218 * @reset: Reset the controller hardware and possibly the PHY. This will
1219 * be called while the controller is uninitialised.
1220 * @probe_port: Probe the MAC and PHY
1221 * @remove_port: Free resources allocated by probe_port()
1222 * @handle_global_event: Handle a "global" event (may be %NULL)
1223 * @fini_dmaq: Flush and finalise DMA queues (RX and TX queues)
1224 * @prepare_flr: Prepare for an FLR
1225 * @finish_flr: Clean up after an FLR
1226 * @describe_stats: Describe statistics for ethtool
1227 * @update_stats: Update statistics not provided by event handling.
1228 * Either argument may be %NULL.
1229 * @update_stats_atomic: Update statistics while in atomic context, if that
1230 * is more limiting than @update_stats. Otherwise, leave %NULL and
1231 * driver core will call @update_stats.
1232 * @start_stats: Start the regular fetching of statistics
1233 * @pull_stats: Pull stats from the NIC and wait until they arrive.
1234 * @stop_stats: Stop the regular fetching of statistics
1235 * @push_irq_moderation: Apply interrupt moderation value
1236 * @reconfigure_port: Push loopback/power/txdis changes to the MAC and PHY
1237 * @prepare_enable_fc_tx: Prepare MAC to enable pause frame TX (may be %NULL)
1238 * @reconfigure_mac: Push MAC address, MTU, flow control and filter settings
1239 * to the hardware. Serialised by the mac_lock.
1240 * @check_mac_fault: Check MAC fault state. True if fault present.
1241 * @get_wol: Get WoL configuration from driver state
1242 * @set_wol: Push WoL configuration to the NIC
1243 * @resume_wol: Synchronise WoL state between driver and MC (e.g. after resume)
1244 * @get_fec_stats: Get standard FEC statistics.
1245 * @test_chip: Test registers. This is expected to reset the NIC.
1246 * @test_nvram: Test validity of NVRAM contents
1247 * @mcdi_request: Send an MCDI request with the given header and SDU.
1248 * The SDU length may be any value from 0 up to the protocol-
1249 * defined maximum, but its buffer will be padded to a multiple
1250 * of 4 bytes.
1251 * @mcdi_poll_response: Test whether an MCDI response is available.
1252 * @mcdi_read_response: Read the MCDI response PDU. The offset will
1253 * be a multiple of 4. The length may not be, but the buffer
1254 * will be padded so it is safe to round up.
1255 * @mcdi_poll_reboot: Test whether the MCDI has rebooted. If so,
1256 * return an appropriate error code for aborting any current
1257 * request; otherwise return 0.
1258 * @irq_enable_master: Enable IRQs on the NIC. Each event queue must
1259 * be separately enabled after this.
1260 * @irq_test_generate: Generate a test IRQ
1261 * @irq_disable_non_ev: Disable non-event IRQs on the NIC. Each event
1262 * queue must be separately disabled before this.
1263 * @irq_handle_msi: Handle MSI for a channel. The @dev_id argument is
1264 * a pointer to the &struct efx_msi_context for the channel.
1265 * @irq_handle_legacy: Handle legacy interrupt. The @dev_id argument
1266 * is a pointer to the &struct efx_nic.
1267 * @tx_probe: Allocate resources for TX queue (and select TXQ type)
1268 * @tx_init: Initialise TX queue on the NIC
1269 * @tx_remove: Free resources for TX queue
1270 * @tx_write: Write TX descriptors and doorbell
1271 * @tx_enqueue: Add an SKB to TX queue
1272 * @rx_push_rss_config: Write RSS hash key and indirection table to the NIC
1273 * @rx_pull_rss_config: Read RSS hash key and indirection table back from the NIC
1274 * @rx_push_rss_context_config: Write RSS hash key and indirection table for
1275 * user RSS context to the NIC
1276 * @rx_pull_rss_context_config: Read RSS hash key and indirection table for user
1277 * RSS context back from the NIC
1278 * @rx_probe: Allocate resources for RX queue
1279 * @rx_init: Initialise RX queue on the NIC
1280 * @rx_remove: Free resources for RX queue
1281 * @rx_write: Write RX descriptors and doorbell
1282 * @rx_defer_refill: Generate a refill reminder event
1283 * @rx_packet: Receive the queued RX buffer on a channel
1284 * @rx_buf_hash_valid: Determine whether the RX prefix contains a valid hash
1285 * @ev_probe: Allocate resources for event queue
1286 * @ev_init: Initialise event queue on the NIC
1287 * @ev_fini: Deinitialise event queue on the NIC
1288 * @ev_remove: Free resources for event queue
1289 * @ev_process: Process events for a queue, up to the given NAPI quota
1290 * @ev_read_ack: Acknowledge read events on a queue, rearming its IRQ
1291 * @ev_test_generate: Generate a test event
1292 * @filter_table_probe: Probe filter capabilities and set up filter software state
1293 * @filter_table_restore: Restore filters removed from hardware
1294 * @filter_table_remove: Remove filters from hardware and tear down software state
1295 * @filter_update_rx_scatter: Update filters after change to rx scatter setting
1296 * @filter_insert: add or replace a filter
1297 * @filter_remove_safe: remove a filter by ID, carefully
1298 * @filter_get_safe: retrieve a filter by ID, carefully
1299 * @filter_clear_rx: Remove all RX filters whose priority is less than or
1300 * equal to the given priority and is not %EFX_FILTER_PRI_AUTO
1301 * @filter_count_rx_used: Get the number of filters in use at a given priority
1302 * @filter_get_rx_id_limit: Get maximum value of a filter id, plus 1
1303 * @filter_get_rx_ids: Get list of RX filters at a given priority
1304 * @filter_rfs_expire_one: Consider expiring a filter inserted for RFS.
1305 * This must check whether the specified table entry is used by RFS
1306 * and that rps_may_expire_flow() returns true for it.
1307 * @mtd_probe: Probe and add MTD partitions associated with this net device,
1308 * using efx_mtd_add()
1309 * @mtd_rename: Set an MTD partition name using the net device name
1310 * @mtd_read: Read from an MTD partition
1311 * @mtd_erase: Erase part of an MTD partition
1312 * @mtd_write: Write to an MTD partition
1313 * @mtd_sync: Wait for write-back to complete on MTD partition. This
1314 * also notifies the driver that a writer has finished using this
1315 * partition.
1316 * @ptp_write_host_time: Send host time to MC as part of sync protocol
1317 * @ptp_set_ts_sync_events: Enable or disable sync events for inline RX
1318 * timestamping, possibly only temporarily for the purposes of a reset.
1319 * @ptp_set_ts_config: Set hardware timestamp configuration. The flags
1320 * and tx_type will already have been validated but this operation
1321 * must validate and update rx_filter.
1322 * @get_phys_port_id: Get the underlying physical port id.
1323 * @set_mac_address: Set the MAC address of the device
1324 * @tso_versions: Returns mask of firmware-assisted TSO versions supported.
1325 * If %NULL, then device does not support any TSO version.
1326 * @udp_tnl_push_ports: Push the list of UDP tunnel ports to the NIC if required.
1327 * @udp_tnl_has_port: Check if a port has been added as UDP tunnel
1328 * @print_additional_fwver: Dump NIC-specific additional FW version info
1329 * @sensor_event: Handle a sensor event from MCDI
1330 * @rx_recycle_ring_size: Size of the RX recycle ring
1331 * @revision: Hardware architecture revision
1332 * @txd_ptr_tbl_base: TX descriptor ring base address
1333 * @rxd_ptr_tbl_base: RX descriptor ring base address
1334 * @buf_tbl_base: Buffer table base address
1335 * @evq_ptr_tbl_base: Event queue pointer table base address
1336 * @evq_rptr_tbl_base: Event queue read-pointer table base address
1337 * @max_dma_mask: Maximum possible DMA mask
1338 * @rx_prefix_size: Size of RX prefix before packet data
1339 * @rx_hash_offset: Offset of RX flow hash within prefix
1340 * @rx_ts_offset: Offset of timestamp within prefix
1341 * @rx_buffer_padding: Size of padding at end of RX packet
1342 * @can_rx_scatter: NIC is able to scatter packets to multiple buffers
1343 * @always_rx_scatter: NIC will always scatter packets to multiple buffers
1344 * @option_descriptors: NIC supports TX option descriptors
1345 * @min_interrupt_mode: Lowest capability interrupt mode supported
1346 * from &enum efx_int_mode.
1347 * @timer_period_max: Maximum period of interrupt timer (in ticks)
1348 * @offload_features: net_device feature flags for protocol offload
1349 * features implemented in hardware
1350 * @mcdi_max_ver: Maximum MCDI version supported
1351 * @hwtstamp_filters: Mask of hardware timestamp filter types supported
1352 */
1353struct efx_nic_type {
1354 bool is_vf;
1355 unsigned int (*mem_bar)(struct efx_nic *efx);
1356 unsigned int (*mem_map_size)(struct efx_nic *efx);
1357 int (*probe)(struct efx_nic *efx);
1358 void (*remove)(struct efx_nic *efx);
1359 int (*init)(struct efx_nic *efx);
1360 int (*dimension_resources)(struct efx_nic *efx);
1361 void (*fini)(struct efx_nic *efx);
1362 void (*monitor)(struct efx_nic *efx);
1363 enum reset_type (*map_reset_reason)(enum reset_type reason);
1364 int (*map_reset_flags)(u32 *flags);
1365 int (*reset)(struct efx_nic *efx, enum reset_type method);
1366 int (*probe_port)(struct efx_nic *efx);
1367 void (*remove_port)(struct efx_nic *efx);
1368 bool (*handle_global_event)(struct efx_channel *channel, efx_qword_t *);
1369 int (*fini_dmaq)(struct efx_nic *efx);
1370 void (*prepare_flr)(struct efx_nic *efx);
1371 void (*finish_flr)(struct efx_nic *efx);
1372 size_t (*describe_stats)(struct efx_nic *efx, u8 *names);
1373 size_t (*update_stats)(struct efx_nic *efx, u64 *full_stats,
1374 struct rtnl_link_stats64 *core_stats);
1375 size_t (*update_stats_atomic)(struct efx_nic *efx, u64 *full_stats,
1376 struct rtnl_link_stats64 *core_stats);
1377 void (*start_stats)(struct efx_nic *efx);
1378 void (*pull_stats)(struct efx_nic *efx);
1379 void (*stop_stats)(struct efx_nic *efx);
1380 void (*push_irq_moderation)(struct efx_channel *channel);
1381 int (*reconfigure_port)(struct efx_nic *efx);
1382 void (*prepare_enable_fc_tx)(struct efx_nic *efx);
1383 int (*reconfigure_mac)(struct efx_nic *efx, bool mtu_only);
1384 bool (*check_mac_fault)(struct efx_nic *efx);
1385 void (*get_wol)(struct efx_nic *efx, struct ethtool_wolinfo *wol);
1386 int (*set_wol)(struct efx_nic *efx, u32 type);
1387 void (*resume_wol)(struct efx_nic *efx);
1388 void (*get_fec_stats)(struct efx_nic *efx,
1389 struct ethtool_fec_stats *fec_stats);
1390 unsigned int (*check_caps)(const struct efx_nic *efx,
1391 u8 flag,
1392 u32 offset);
1393 int (*test_chip)(struct efx_nic *efx, struct efx_self_tests *tests);
1394 int (*test_nvram)(struct efx_nic *efx);
1395 void (*mcdi_request)(struct efx_nic *efx,
1396 const efx_dword_t *hdr, size_t hdr_len,
1397 const efx_dword_t *sdu, size_t sdu_len);
1398 bool (*mcdi_poll_response)(struct efx_nic *efx);
1399 void (*mcdi_read_response)(struct efx_nic *efx, efx_dword_t *pdu,
1400 size_t pdu_offset, size_t pdu_len);
1401 int (*mcdi_poll_reboot)(struct efx_nic *efx);
1402 void (*mcdi_reboot_detected)(struct efx_nic *efx);
1403 void (*irq_enable_master)(struct efx_nic *efx);
1404 int (*irq_test_generate)(struct efx_nic *efx);
1405 void (*irq_disable_non_ev)(struct efx_nic *efx);
1406 irqreturn_t (*irq_handle_msi)(int irq, void *dev_id);
1407 irqreturn_t (*irq_handle_legacy)(int irq, void *dev_id);
1408 int (*tx_probe)(struct efx_tx_queue *tx_queue);
1409 void (*tx_init)(struct efx_tx_queue *tx_queue);
1410 void (*tx_remove)(struct efx_tx_queue *tx_queue);
1411 void (*tx_write)(struct efx_tx_queue *tx_queue);
1412 netdev_tx_t (*tx_enqueue)(struct efx_tx_queue *tx_queue, struct sk_buff *skb);
1413 unsigned int (*tx_limit_len)(struct efx_tx_queue *tx_queue,
1414 dma_addr_t dma_addr, unsigned int len);
1415 int (*rx_push_rss_config)(struct efx_nic *efx, bool user,
1416 const u32 *rx_indir_table, const u8 *key);
1417 int (*rx_pull_rss_config)(struct efx_nic *efx);
1418 int (*rx_push_rss_context_config)(struct efx_nic *efx,
1419 struct efx_rss_context *ctx,
1420 const u32 *rx_indir_table,
1421 const u8 *key);
1422 int (*rx_pull_rss_context_config)(struct efx_nic *efx,
1423 struct efx_rss_context *ctx);
1424 void (*rx_restore_rss_contexts)(struct efx_nic *efx);
1425 int (*rx_probe)(struct efx_rx_queue *rx_queue);
1426 void (*rx_init)(struct efx_rx_queue *rx_queue);
1427 void (*rx_remove)(struct efx_rx_queue *rx_queue);
1428 void (*rx_write)(struct efx_rx_queue *rx_queue);
1429 void (*rx_defer_refill)(struct efx_rx_queue *rx_queue);
1430 void (*rx_packet)(struct efx_channel *channel);
1431 bool (*rx_buf_hash_valid)(const u8 *prefix);
1432 int (*ev_probe)(struct efx_channel *channel);
1433 int (*ev_init)(struct efx_channel *channel);
1434 void (*ev_fini)(struct efx_channel *channel);
1435 void (*ev_remove)(struct efx_channel *channel);
1436 int (*ev_process)(struct efx_channel *channel, int quota);
1437 void (*ev_read_ack)(struct efx_channel *channel);
1438 void (*ev_test_generate)(struct efx_channel *channel);
1439 int (*filter_table_probe)(struct efx_nic *efx);
1440 void (*filter_table_restore)(struct efx_nic *efx);
1441 void (*filter_table_remove)(struct efx_nic *efx);
1442 void (*filter_update_rx_scatter)(struct efx_nic *efx);
1443 s32 (*filter_insert)(struct efx_nic *efx,
1444 struct efx_filter_spec *spec, bool replace);
1445 int (*filter_remove_safe)(struct efx_nic *efx,
1446 enum efx_filter_priority priority,
1447 u32 filter_id);
1448 int (*filter_get_safe)(struct efx_nic *efx,
1449 enum efx_filter_priority priority,
1450 u32 filter_id, struct efx_filter_spec *);
1451 int (*filter_clear_rx)(struct efx_nic *efx,
1452 enum efx_filter_priority priority);
1453 u32 (*filter_count_rx_used)(struct efx_nic *efx,
1454 enum efx_filter_priority priority);
1455 u32 (*filter_get_rx_id_limit)(struct efx_nic *efx);
1456 s32 (*filter_get_rx_ids)(struct efx_nic *efx,
1457 enum efx_filter_priority priority,
1458 u32 *buf, u32 size);
1459#ifdef CONFIG_RFS_ACCEL
1460 bool (*filter_rfs_expire_one)(struct efx_nic *efx, u32 flow_id,
1461 unsigned int index);
1462#endif
1463#ifdef CONFIG_SFC_MTD
1464 int (*mtd_probe)(struct efx_nic *efx);
1465 void (*mtd_rename)(struct efx_mtd_partition *part);
1466 int (*mtd_read)(struct mtd_info *mtd, loff_t start, size_t len,
1467 size_t *retlen, u8 *buffer);
1468 int (*mtd_erase)(struct mtd_info *mtd, loff_t start, size_t len);
1469 int (*mtd_write)(struct mtd_info *mtd, loff_t start, size_t len,
1470 size_t *retlen, const u8 *buffer);
1471 int (*mtd_sync)(struct mtd_info *mtd);
1472#endif
1473 void (*ptp_write_host_time)(struct efx_nic *efx, u32 host_time);
1474 int (*ptp_set_ts_sync_events)(struct efx_nic *efx, bool en, bool temp);
1475 int (*ptp_set_ts_config)(struct efx_nic *efx,
1476 struct kernel_hwtstamp_config *init);
1477 int (*sriov_configure)(struct efx_nic *efx, int num_vfs);
1478 int (*vlan_rx_add_vid)(struct efx_nic *efx, __be16 proto, u16 vid);
1479 int (*vlan_rx_kill_vid)(struct efx_nic *efx, __be16 proto, u16 vid);
1480 int (*get_phys_port_id)(struct efx_nic *efx,
1481 struct netdev_phys_item_id *ppid);
1482 int (*sriov_init)(struct efx_nic *efx);
1483 void (*sriov_fini)(struct efx_nic *efx);
1484 bool (*sriov_wanted)(struct efx_nic *efx);
1485 int (*sriov_set_vf_mac)(struct efx_nic *efx, int vf_i, const u8 *mac);
1486 int (*sriov_set_vf_vlan)(struct efx_nic *efx, int vf_i, u16 vlan,
1487 u8 qos);
1488 int (*sriov_set_vf_spoofchk)(struct efx_nic *efx, int vf_i,
1489 bool spoofchk);
1490 int (*sriov_get_vf_config)(struct efx_nic *efx, int vf_i,
1491 struct ifla_vf_info *ivi);
1492 int (*sriov_set_vf_link_state)(struct efx_nic *efx, int vf_i,
1493 int link_state);
1494 int (*vswitching_probe)(struct efx_nic *efx);
1495 int (*vswitching_restore)(struct efx_nic *efx);
1496 void (*vswitching_remove)(struct efx_nic *efx);
1497 int (*get_mac_address)(struct efx_nic *efx, unsigned char *perm_addr);
1498 int (*set_mac_address)(struct efx_nic *efx);
1499 u32 (*tso_versions)(struct efx_nic *efx);
1500 int (*udp_tnl_push_ports)(struct efx_nic *efx);
1501 bool (*udp_tnl_has_port)(struct efx_nic *efx, __be16 port);
1502 size_t (*print_additional_fwver)(struct efx_nic *efx, char *buf,
1503 size_t len);
1504 void (*sensor_event)(struct efx_nic *efx, efx_qword_t *ev);
1505 unsigned int (*rx_recycle_ring_size)(const struct efx_nic *efx);
1506
1507 int revision;
1508 unsigned int txd_ptr_tbl_base;
1509 unsigned int rxd_ptr_tbl_base;
1510 unsigned int buf_tbl_base;
1511 unsigned int evq_ptr_tbl_base;
1512 unsigned int evq_rptr_tbl_base;
1513 u64 max_dma_mask;
1514 unsigned int rx_prefix_size;
1515 unsigned int rx_hash_offset;
1516 unsigned int rx_ts_offset;
1517 unsigned int rx_buffer_padding;
1518 bool can_rx_scatter;
1519 bool always_rx_scatter;
1520 bool option_descriptors;
1521 unsigned int min_interrupt_mode;
1522 unsigned int timer_period_max;
1523 netdev_features_t offload_features;
1524 int mcdi_max_ver;
1525 unsigned int max_rx_ip_filters;
1526 u32 hwtstamp_filters;
1527 unsigned int rx_hash_key_size;
1528};
1529
1530/**************************************************************************
1531 *
1532 * Prototypes and inline functions
1533 *
1534 *************************************************************************/
1535
1536static inline struct efx_channel *
1537efx_get_channel(struct efx_nic *efx, unsigned index)
1538{
1539 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_channels);
1540 return efx->channel[index];
1541}
1542
1543/* Iterate over all used channels */
1544#define efx_for_each_channel(_channel, _efx) \
1545 for (_channel = (_efx)->channel[0]; \
1546 _channel; \
1547 _channel = (_channel->channel + 1 < (_efx)->n_channels) ? \
1548 (_efx)->channel[_channel->channel + 1] : NULL)
1549
1550/* Iterate over all used channels in reverse */
1551#define efx_for_each_channel_rev(_channel, _efx) \
1552 for (_channel = (_efx)->channel[(_efx)->n_channels - 1]; \
1553 _channel; \
1554 _channel = _channel->channel ? \
1555 (_efx)->channel[_channel->channel - 1] : NULL)
1556
1557static inline struct efx_channel *
1558efx_get_tx_channel(struct efx_nic *efx, unsigned int index)
1559{
1560 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_tx_channels);
1561 return efx->channel[efx->tx_channel_offset + index];
1562}
1563
1564static inline struct efx_channel *
1565efx_get_xdp_channel(struct efx_nic *efx, unsigned int index)
1566{
1567 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_xdp_channels);
1568 return efx->channel[efx->xdp_channel_offset + index];
1569}
1570
1571static inline bool efx_channel_is_xdp_tx(struct efx_channel *channel)
1572{
1573 return channel->channel - channel->efx->xdp_channel_offset <
1574 channel->efx->n_xdp_channels;
1575}
1576
1577static inline bool efx_channel_has_tx_queues(struct efx_channel *channel)
1578{
1579 return channel && channel->channel >= channel->efx->tx_channel_offset;
1580}
1581
1582static inline unsigned int efx_channel_num_tx_queues(struct efx_channel *channel)
1583{
1584 if (efx_channel_is_xdp_tx(channel))
1585 return channel->efx->xdp_tx_per_channel;
1586 return channel->efx->tx_queues_per_channel;
1587}
1588
1589static inline struct efx_tx_queue *
1590efx_channel_get_tx_queue(struct efx_channel *channel, unsigned int type)
1591{
1592 EFX_WARN_ON_ONCE_PARANOID(type >= EFX_TXQ_TYPES);
1593 return channel->tx_queue_by_type[type];
1594}
1595
1596static inline struct efx_tx_queue *
1597efx_get_tx_queue(struct efx_nic *efx, unsigned int index, unsigned int type)
1598{
1599 struct efx_channel *channel = efx_get_tx_channel(efx, index);
1600
1601 return efx_channel_get_tx_queue(channel, type);
1602}
1603
1604/* Iterate over all TX queues belonging to a channel */
1605#define efx_for_each_channel_tx_queue(_tx_queue, _channel) \
1606 if (!efx_channel_has_tx_queues(_channel)) \
1607 ; \
1608 else \
1609 for (_tx_queue = (_channel)->tx_queue; \
1610 _tx_queue < (_channel)->tx_queue + \
1611 efx_channel_num_tx_queues(_channel); \
1612 _tx_queue++)
1613
1614static inline bool efx_channel_has_rx_queue(struct efx_channel *channel)
1615{
1616 return channel->rx_queue.core_index >= 0;
1617}
1618
1619static inline struct efx_rx_queue *
1620efx_channel_get_rx_queue(struct efx_channel *channel)
1621{
1622 EFX_WARN_ON_ONCE_PARANOID(!efx_channel_has_rx_queue(channel));
1623 return &channel->rx_queue;
1624}
1625
1626/* Iterate over all RX queues belonging to a channel */
1627#define efx_for_each_channel_rx_queue(_rx_queue, _channel) \
1628 if (!efx_channel_has_rx_queue(_channel)) \
1629 ; \
1630 else \
1631 for (_rx_queue = &(_channel)->rx_queue; \
1632 _rx_queue; \
1633 _rx_queue = NULL)
1634
1635static inline struct efx_channel *
1636efx_rx_queue_channel(struct efx_rx_queue *rx_queue)
1637{
1638 return container_of(rx_queue, struct efx_channel, rx_queue);
1639}
1640
1641static inline int efx_rx_queue_index(struct efx_rx_queue *rx_queue)
1642{
1643 return efx_rx_queue_channel(rx_queue)->channel;
1644}
1645
1646/* Returns a pointer to the specified receive buffer in the RX
1647 * descriptor queue.
1648 */
1649static inline struct efx_rx_buffer *efx_rx_buffer(struct efx_rx_queue *rx_queue,
1650 unsigned int index)
1651{
1652 return &rx_queue->buffer[index];
1653}
1654
1655static inline struct efx_rx_buffer *
1656efx_rx_buf_next(struct efx_rx_queue *rx_queue, struct efx_rx_buffer *rx_buf)
1657{
1658 if (unlikely(rx_buf == efx_rx_buffer(rx_queue, rx_queue->ptr_mask)))
1659 return efx_rx_buffer(rx_queue, 0);
1660 else
1661 return rx_buf + 1;
1662}
1663
1664/**
1665 * EFX_MAX_FRAME_LEN - calculate maximum frame length
1666 *
1667 * This calculates the maximum frame length that will be used for a
1668 * given MTU. The frame length will be equal to the MTU plus a
1669 * constant amount of header space and padding. This is the quantity
1670 * that the net driver will program into the MAC as the maximum frame
1671 * length.
1672 *
1673 * The 10G MAC requires 8-byte alignment on the frame
1674 * length, so we round up to the nearest 8.
1675 *
1676 * Re-clocking by the XGXS on RX can reduce an IPG to 32 bits (half an
1677 * XGMII cycle). If the frame length reaches the maximum value in the
1678 * same cycle, the XMAC can miss the IPG altogether. We work around
1679 * this by adding a further 16 bytes.
1680 */
1681#define EFX_FRAME_PAD 16
1682#define EFX_MAX_FRAME_LEN(mtu) \
1683 (ALIGN(((mtu) + ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN + EFX_FRAME_PAD), 8))
1684
1685static inline bool efx_xmit_with_hwtstamp(struct sk_buff *skb)
1686{
1687 return skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP;
1688}
1689static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
1690{
1691 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1692}
1693
1694/* Get the max fill level of the TX queues on this channel */
1695static inline unsigned int
1696efx_channel_tx_fill_level(struct efx_channel *channel)
1697{
1698 struct efx_tx_queue *tx_queue;
1699 unsigned int fill_level = 0;
1700
1701 efx_for_each_channel_tx_queue(tx_queue, channel)
1702 fill_level = max(fill_level,
1703 tx_queue->insert_count - tx_queue->read_count);
1704
1705 return fill_level;
1706}
1707
1708/* Conservative approximation of efx_channel_tx_fill_level using cached value */
1709static inline unsigned int
1710efx_channel_tx_old_fill_level(struct efx_channel *channel)
1711{
1712 struct efx_tx_queue *tx_queue;
1713 unsigned int fill_level = 0;
1714
1715 efx_for_each_channel_tx_queue(tx_queue, channel)
1716 fill_level = max(fill_level,
1717 tx_queue->insert_count - tx_queue->old_read_count);
1718
1719 return fill_level;
1720}
1721
1722/* Get all supported features.
1723 * If a feature is not fixed, it is present in hw_features.
1724 * If a feature is fixed, it does not present in hw_features, but
1725 * always in features.
1726 */
1727static inline netdev_features_t efx_supported_features(const struct efx_nic *efx)
1728{
1729 const struct net_device *net_dev = efx->net_dev;
1730
1731 return net_dev->features | net_dev->hw_features;
1732}
1733
1734/* Get the current TX queue insert index. */
1735static inline unsigned int
1736efx_tx_queue_get_insert_index(const struct efx_tx_queue *tx_queue)
1737{
1738 return tx_queue->insert_count & tx_queue->ptr_mask;
1739}
1740
1741/* Get a TX buffer. */
1742static inline struct efx_tx_buffer *
1743__efx_tx_queue_get_insert_buffer(const struct efx_tx_queue *tx_queue)
1744{
1745 return &tx_queue->buffer[efx_tx_queue_get_insert_index(tx_queue)];
1746}
1747
1748/* Get a TX buffer, checking it's not currently in use. */
1749static inline struct efx_tx_buffer *
1750efx_tx_queue_get_insert_buffer(const struct efx_tx_queue *tx_queue)
1751{
1752 struct efx_tx_buffer *buffer =
1753 __efx_tx_queue_get_insert_buffer(tx_queue);
1754
1755 EFX_WARN_ON_ONCE_PARANOID(buffer->len);
1756 EFX_WARN_ON_ONCE_PARANOID(buffer->flags);
1757 EFX_WARN_ON_ONCE_PARANOID(buffer->unmap_len);
1758
1759 return buffer;
1760}
1761
1762#endif /* EFX_NET_DRIVER_H */
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/mtd/mtd.h>
28#include <net/busy_poll.h>
29#include <net/xdp.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/* Non-standard XDP_PACKET_HEADROOM and tailroom to satisfy XDP_REDIRECT and
95 * still fit two standard MTU size packets into a single 4K page.
96 */
97#define EFX_XDP_HEADROOM 128
98#define EFX_XDP_TAILROOM SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
99
100/* Forward declare Precision Time Protocol (PTP) support structure. */
101struct efx_ptp_data;
102struct hwtstamp_config;
103
104struct efx_self_tests;
105
106/**
107 * struct efx_buffer - A general-purpose DMA buffer
108 * @addr: host base address of the buffer
109 * @dma_addr: DMA base address of the buffer
110 * @len: Buffer length, in bytes
111 *
112 * The NIC uses these buffers for its interrupt status registers and
113 * MAC stats dumps.
114 */
115struct efx_buffer {
116 void *addr;
117 dma_addr_t dma_addr;
118 unsigned int len;
119};
120
121/**
122 * struct efx_special_buffer - DMA buffer entered into buffer table
123 * @buf: Standard &struct efx_buffer
124 * @index: Buffer index within controller;s buffer table
125 * @entries: Number of buffer table entries
126 *
127 * The NIC has a buffer table that maps buffers of size %EFX_BUF_SIZE.
128 * Event and descriptor rings are addressed via one or more buffer
129 * table entries (and so can be physically non-contiguous, although we
130 * currently do not take advantage of that). On Falcon and Siena we
131 * have to take care of allocating and initialising the entries
132 * ourselves. On later hardware this is managed by the firmware and
133 * @index and @entries are left as 0.
134 */
135struct efx_special_buffer {
136 struct efx_buffer buf;
137 unsigned int index;
138 unsigned int entries;
139};
140
141/**
142 * struct efx_tx_buffer - buffer state for a TX descriptor
143 * @skb: When @flags & %EFX_TX_BUF_SKB, the associated socket buffer to be
144 * freed when descriptor completes
145 * @xdpf: When @flags & %EFX_TX_BUF_XDP, the XDP frame information; its @data
146 * member is the associated buffer to drop a page reference on.
147 * @option: When @flags & %EFX_TX_BUF_OPTION, an EF10-specific option
148 * descriptor.
149 * @dma_addr: DMA address of the fragment.
150 * @flags: Flags for allocation and DMA mapping type
151 * @len: Length of this fragment.
152 * This field is zero when the queue slot is empty.
153 * @unmap_len: Length of this fragment to unmap
154 * @dma_offset: Offset of @dma_addr from the address of the backing DMA mapping.
155 * Only valid if @unmap_len != 0.
156 */
157struct efx_tx_buffer {
158 union {
159 const struct sk_buff *skb;
160 struct xdp_frame *xdpf;
161 };
162 union {
163 efx_qword_t option; /* EF10 */
164 dma_addr_t dma_addr;
165 };
166 unsigned short flags;
167 unsigned short len;
168 unsigned short unmap_len;
169 unsigned short dma_offset;
170};
171#define EFX_TX_BUF_CONT 1 /* not last descriptor of packet */
172#define EFX_TX_BUF_SKB 2 /* buffer is last part of skb */
173#define EFX_TX_BUF_MAP_SINGLE 8 /* buffer was mapped with dma_map_single() */
174#define EFX_TX_BUF_OPTION 0x10 /* empty buffer for option descriptor */
175#define EFX_TX_BUF_XDP 0x20 /* buffer was sent with XDP */
176#define EFX_TX_BUF_TSO_V3 0x40 /* empty buffer for a TSO_V3 descriptor */
177
178/**
179 * struct efx_tx_queue - An Efx TX queue
180 *
181 * This is a ring buffer of TX fragments.
182 * Since the TX completion path always executes on the same
183 * CPU and the xmit path can operate on different CPUs,
184 * performance is increased by ensuring that the completion
185 * path and the xmit path operate on different cache lines.
186 * This is particularly important if the xmit path is always
187 * executing on one CPU which is different from the completion
188 * path. There is also a cache line for members which are
189 * read but not written on the fast path.
190 *
191 * @efx: The associated Efx NIC
192 * @queue: DMA queue number
193 * @label: Label for TX completion events.
194 * Is our index within @channel->tx_queue array.
195 * @tso_version: Version of TSO in use for this queue.
196 * @channel: The associated channel
197 * @core_txq: The networking core TX queue structure
198 * @buffer: The software buffer ring
199 * @cb_page: Array of pages of copy buffers. Carved up according to
200 * %EFX_TX_CB_ORDER into %EFX_TX_CB_SIZE-sized chunks.
201 * @txd: The hardware descriptor ring
202 * @ptr_mask: The size of the ring minus 1.
203 * @piobuf: PIO buffer region for this TX queue (shared with its partner).
204 * Size of the region is efx_piobuf_size.
205 * @piobuf_offset: Buffer offset to be specified in PIO descriptors
206 * @initialised: Has hardware queue been initialised?
207 * @timestamping: Is timestamping enabled for this channel?
208 * @xdp_tx: Is this an XDP tx queue?
209 * @handle_tso: TSO xmit preparation handler. Sets up the TSO metadata and
210 * may also map tx data, depending on the nature of the TSO implementation.
211 * @read_count: Current read pointer.
212 * This is the number of buffers that have been removed from both rings.
213 * @old_write_count: The value of @write_count when last checked.
214 * This is here for performance reasons. The xmit path will
215 * only get the up-to-date value of @write_count if this
216 * variable indicates that the queue is empty. This is to
217 * avoid cache-line ping-pong between the xmit path and the
218 * completion path.
219 * @merge_events: Number of TX merged completion events
220 * @completed_timestamp_major: Top part of the most recent tx timestamp.
221 * @completed_timestamp_minor: Low part of the most recent tx timestamp.
222 * @insert_count: Current insert pointer
223 * This is the number of buffers that have been added to the
224 * software ring.
225 * @write_count: Current write pointer
226 * This is the number of buffers that have been added to the
227 * hardware ring.
228 * @packet_write_count: Completable write pointer
229 * This is the write pointer of the last packet written.
230 * Normally this will equal @write_count, but as option descriptors
231 * don't produce completion events, they won't update this.
232 * Filled in iff @efx->type->option_descriptors; only used for PIO.
233 * Thus, this is written and used on EF10, and neither on farch.
234 * @old_read_count: The value of read_count when last checked.
235 * This is here for performance reasons. The xmit path will
236 * only get the up-to-date value of read_count if this
237 * variable indicates that the queue is full. This is to
238 * avoid cache-line ping-pong between the xmit path and the
239 * completion path.
240 * @tso_bursts: Number of times TSO xmit invoked by kernel
241 * @tso_long_headers: Number of packets with headers too long for standard
242 * blocks
243 * @tso_packets: Number of packets via the TSO xmit path
244 * @tso_fallbacks: Number of times TSO fallback used
245 * @pushes: Number of times the TX push feature has been used
246 * @pio_packets: Number of times the TX PIO feature has been used
247 * @xmit_more_available: Are any packets waiting to be pushed to the NIC
248 * @cb_packets: Number of times the TX copybreak feature has been used
249 * @notify_count: Count of notified descriptors to the NIC
250 * @empty_read_count: If the completion path has seen the queue as empty
251 * and the transmission path has not yet checked this, the value of
252 * @read_count bitwise-added to %EFX_EMPTY_COUNT_VALID; otherwise 0.
253 */
254struct efx_tx_queue {
255 /* Members which don't change on the fast path */
256 struct efx_nic *efx ____cacheline_aligned_in_smp;
257 unsigned int queue;
258 unsigned int label;
259 unsigned int tso_version;
260 struct efx_channel *channel;
261 struct netdev_queue *core_txq;
262 struct efx_tx_buffer *buffer;
263 struct efx_buffer *cb_page;
264 struct efx_special_buffer txd;
265 unsigned int ptr_mask;
266 void __iomem *piobuf;
267 unsigned int piobuf_offset;
268 bool initialised;
269 bool timestamping;
270 bool xdp_tx;
271
272 /* Function pointers used in the fast path. */
273 int (*handle_tso)(struct efx_tx_queue*, struct sk_buff*, bool *);
274
275 /* Members used mainly on the completion path */
276 unsigned int read_count ____cacheline_aligned_in_smp;
277 unsigned int old_write_count;
278 unsigned int merge_events;
279 unsigned int bytes_compl;
280 unsigned int pkts_compl;
281 u32 completed_timestamp_major;
282 u32 completed_timestamp_minor;
283
284 /* Members used only on the xmit path */
285 unsigned int insert_count ____cacheline_aligned_in_smp;
286 unsigned int write_count;
287 unsigned int packet_write_count;
288 unsigned int old_read_count;
289 unsigned int tso_bursts;
290 unsigned int tso_long_headers;
291 unsigned int tso_packets;
292 unsigned int tso_fallbacks;
293 unsigned int pushes;
294 unsigned int pio_packets;
295 bool xmit_more_available;
296 unsigned int cb_packets;
297 unsigned int notify_count;
298 /* Statistics to supplement MAC stats */
299 unsigned long tx_packets;
300
301 /* Members shared between paths and sometimes updated */
302 unsigned int empty_read_count ____cacheline_aligned_in_smp;
303#define EFX_EMPTY_COUNT_VALID 0x80000000
304 atomic_t flush_outstanding;
305};
306
307#define EFX_TX_CB_ORDER 7
308#define EFX_TX_CB_SIZE (1 << EFX_TX_CB_ORDER) - NET_IP_ALIGN
309
310/**
311 * struct efx_rx_buffer - An Efx RX data buffer
312 * @dma_addr: DMA base address of the buffer
313 * @page: The associated page buffer.
314 * Will be %NULL if the buffer slot is currently free.
315 * @page_offset: If pending: offset in @page of DMA base address.
316 * If completed: offset in @page of Ethernet header.
317 * @len: If pending: length for DMA descriptor.
318 * If completed: received length, excluding hash prefix.
319 * @flags: Flags for buffer and packet state. These are only set on the
320 * first buffer of a scattered packet.
321 */
322struct efx_rx_buffer {
323 dma_addr_t dma_addr;
324 struct page *page;
325 u16 page_offset;
326 u16 len;
327 u16 flags;
328};
329#define EFX_RX_BUF_LAST_IN_PAGE 0x0001
330#define EFX_RX_PKT_CSUMMED 0x0002
331#define EFX_RX_PKT_DISCARD 0x0004
332#define EFX_RX_PKT_TCP 0x0040
333#define EFX_RX_PKT_PREFIX_LEN 0x0080 /* length is in prefix only */
334#define EFX_RX_PKT_CSUM_LEVEL 0x0200
335
336/**
337 * struct efx_rx_page_state - Page-based rx buffer state
338 *
339 * Inserted at the start of every page allocated for receive buffers.
340 * Used to facilitate sharing dma mappings between recycled rx buffers
341 * and those passed up to the kernel.
342 *
343 * @dma_addr: The dma address of this page.
344 */
345struct efx_rx_page_state {
346 dma_addr_t dma_addr;
347
348 unsigned int __pad[] ____cacheline_aligned;
349};
350
351/**
352 * struct efx_rx_queue - An Efx RX queue
353 * @efx: The associated Efx NIC
354 * @core_index: Index of network core RX queue. Will be >= 0 iff this
355 * is associated with a real RX queue.
356 * @buffer: The software buffer ring
357 * @rxd: The hardware descriptor ring
358 * @ptr_mask: The size of the ring minus 1.
359 * @refill_enabled: Enable refill whenever fill level is low
360 * @flush_pending: Set when a RX flush is pending. Has the same lifetime as
361 * @rxq_flush_pending.
362 * @added_count: Number of buffers added to the receive queue.
363 * @notified_count: Number of buffers given to NIC (<= @added_count).
364 * @removed_count: Number of buffers removed from the receive queue.
365 * @scatter_n: Used by NIC specific receive code.
366 * @scatter_len: Used by NIC specific receive code.
367 * @page_ring: The ring to store DMA mapped pages for reuse.
368 * @page_add: Counter to calculate the write pointer for the recycle ring.
369 * @page_remove: Counter to calculate the read pointer for the recycle ring.
370 * @page_recycle_count: The number of pages that have been recycled.
371 * @page_recycle_failed: The number of pages that couldn't be recycled because
372 * the kernel still held a reference to them.
373 * @page_recycle_full: The number of pages that were released because the
374 * recycle ring was full.
375 * @page_ptr_mask: The number of pages in the RX recycle ring minus 1.
376 * @max_fill: RX descriptor maximum fill level (<= ring size)
377 * @fast_fill_trigger: RX descriptor fill level that will trigger a fast fill
378 * (<= @max_fill)
379 * @min_fill: RX descriptor minimum non-zero fill level.
380 * This records the minimum fill level observed when a ring
381 * refill was triggered.
382 * @recycle_count: RX buffer recycle counter.
383 * @slow_fill: Timer used to defer efx_nic_generate_fill_event().
384 * @xdp_rxq_info: XDP specific RX queue information.
385 * @xdp_rxq_info_valid: Is xdp_rxq_info valid data?.
386 */
387struct efx_rx_queue {
388 struct efx_nic *efx;
389 int core_index;
390 struct efx_rx_buffer *buffer;
391 struct efx_special_buffer rxd;
392 unsigned int ptr_mask;
393 bool refill_enabled;
394 bool flush_pending;
395
396 unsigned int added_count;
397 unsigned int notified_count;
398 unsigned int removed_count;
399 unsigned int scatter_n;
400 unsigned int scatter_len;
401 struct page **page_ring;
402 unsigned int page_add;
403 unsigned int page_remove;
404 unsigned int page_recycle_count;
405 unsigned int page_recycle_failed;
406 unsigned int page_recycle_full;
407 unsigned int page_ptr_mask;
408 unsigned int max_fill;
409 unsigned int fast_fill_trigger;
410 unsigned int min_fill;
411 unsigned int min_overfill;
412 unsigned int recycle_count;
413 struct timer_list slow_fill;
414 unsigned int slow_fill_count;
415 /* Statistics to supplement MAC stats */
416 unsigned long rx_packets;
417 struct xdp_rxq_info xdp_rxq_info;
418 bool xdp_rxq_info_valid;
419};
420
421enum efx_sync_events_state {
422 SYNC_EVENTS_DISABLED = 0,
423 SYNC_EVENTS_QUIESCENT,
424 SYNC_EVENTS_REQUESTED,
425 SYNC_EVENTS_VALID,
426};
427
428/**
429 * struct efx_channel - An Efx channel
430 *
431 * A channel comprises an event queue, at least one TX queue, at least
432 * one RX queue, and an associated tasklet for processing the event
433 * queue.
434 *
435 * @efx: Associated Efx NIC
436 * @channel: Channel instance number
437 * @type: Channel type definition
438 * @eventq_init: Event queue initialised flag
439 * @enabled: Channel enabled indicator
440 * @irq: IRQ number (MSI and MSI-X only)
441 * @irq_moderation_us: IRQ moderation value (in microseconds)
442 * @napi_dev: Net device used with NAPI
443 * @napi_str: NAPI control structure
444 * @state: state for NAPI vs busy polling
445 * @state_lock: lock protecting @state
446 * @eventq: Event queue buffer
447 * @eventq_mask: Event queue pointer mask
448 * @eventq_read_ptr: Event queue read pointer
449 * @event_test_cpu: Last CPU to handle interrupt or test event for this channel
450 * @irq_count: Number of IRQs since last adaptive moderation decision
451 * @irq_mod_score: IRQ moderation score
452 * @rfs_filter_count: number of accelerated RFS filters currently in place;
453 * equals the count of @rps_flow_id slots filled
454 * @rfs_last_expiry: value of jiffies last time some accelerated RFS filters
455 * were checked for expiry
456 * @rfs_expire_index: next accelerated RFS filter ID to check for expiry
457 * @n_rfs_succeeded: number of successful accelerated RFS filter insertions
458 * @n_rfs_failed; number of failed accelerated RFS filter insertions
459 * @filter_work: Work item for efx_filter_rfs_expire()
460 * @rps_flow_id: Flow IDs of filters allocated for accelerated RFS,
461 * indexed by filter ID
462 * @n_rx_tobe_disc: Count of RX_TOBE_DISC errors
463 * @n_rx_ip_hdr_chksum_err: Count of RX IP header checksum errors
464 * @n_rx_tcp_udp_chksum_err: Count of RX TCP and UDP checksum errors
465 * @n_rx_mcast_mismatch: Count of unmatched multicast frames
466 * @n_rx_frm_trunc: Count of RX_FRM_TRUNC errors
467 * @n_rx_overlength: Count of RX_OVERLENGTH errors
468 * @n_skbuff_leaks: Count of skbuffs leaked due to RX overrun
469 * @n_rx_nodesc_trunc: Number of RX packets truncated and then dropped due to
470 * lack of descriptors
471 * @n_rx_merge_events: Number of RX merged completion events
472 * @n_rx_merge_packets: Number of RX packets completed by merged events
473 * @n_rx_xdp_drops: Count of RX packets intentionally dropped due to XDP
474 * @n_rx_xdp_bad_drops: Count of RX packets dropped due to XDP errors
475 * @n_rx_xdp_tx: Count of RX packets retransmitted due to XDP
476 * @n_rx_xdp_redirect: Count of RX packets redirected to a different NIC by XDP
477 * @rx_pkt_n_frags: Number of fragments in next packet to be delivered by
478 * __efx_rx_packet(), or zero if there is none
479 * @rx_pkt_index: Ring index of first buffer for next packet to be delivered
480 * by __efx_rx_packet(), if @rx_pkt_n_frags != 0
481 * @rx_list: list of SKBs from current RX, awaiting processing
482 * @rx_queue: RX queue for this channel
483 * @tx_queue: TX queues for this channel
484 * @sync_events_state: Current state of sync events on this channel
485 * @sync_timestamp_major: Major part of the last ptp sync event
486 * @sync_timestamp_minor: Minor part of the last ptp sync event
487 */
488struct efx_channel {
489 struct efx_nic *efx;
490 int channel;
491 const struct efx_channel_type *type;
492 bool eventq_init;
493 bool enabled;
494 int irq;
495 unsigned int irq_moderation_us;
496 struct net_device *napi_dev;
497 struct napi_struct napi_str;
498#ifdef CONFIG_NET_RX_BUSY_POLL
499 unsigned long busy_poll_state;
500#endif
501 struct efx_special_buffer eventq;
502 unsigned int eventq_mask;
503 unsigned int eventq_read_ptr;
504 int event_test_cpu;
505
506 unsigned int irq_count;
507 unsigned int irq_mod_score;
508#ifdef CONFIG_RFS_ACCEL
509 unsigned int rfs_filter_count;
510 unsigned int rfs_last_expiry;
511 unsigned int rfs_expire_index;
512 unsigned int n_rfs_succeeded;
513 unsigned int n_rfs_failed;
514 struct delayed_work filter_work;
515#define RPS_FLOW_ID_INVALID 0xFFFFFFFF
516 u32 *rps_flow_id;
517#endif
518
519 unsigned int n_rx_tobe_disc;
520 unsigned int n_rx_ip_hdr_chksum_err;
521 unsigned int n_rx_tcp_udp_chksum_err;
522 unsigned int n_rx_outer_ip_hdr_chksum_err;
523 unsigned int n_rx_outer_tcp_udp_chksum_err;
524 unsigned int n_rx_inner_ip_hdr_chksum_err;
525 unsigned int n_rx_inner_tcp_udp_chksum_err;
526 unsigned int n_rx_eth_crc_err;
527 unsigned int n_rx_mcast_mismatch;
528 unsigned int n_rx_frm_trunc;
529 unsigned int n_rx_overlength;
530 unsigned int n_skbuff_leaks;
531 unsigned int n_rx_nodesc_trunc;
532 unsigned int n_rx_merge_events;
533 unsigned int n_rx_merge_packets;
534 unsigned int n_rx_xdp_drops;
535 unsigned int n_rx_xdp_bad_drops;
536 unsigned int n_rx_xdp_tx;
537 unsigned int n_rx_xdp_redirect;
538
539 unsigned int rx_pkt_n_frags;
540 unsigned int rx_pkt_index;
541
542 struct list_head *rx_list;
543
544 struct efx_rx_queue rx_queue;
545 struct efx_tx_queue tx_queue[EFX_TXQ_TYPES];
546
547 enum efx_sync_events_state sync_events_state;
548 u32 sync_timestamp_major;
549 u32 sync_timestamp_minor;
550};
551
552/**
553 * struct efx_msi_context - Context for each MSI
554 * @efx: The associated NIC
555 * @index: Index of the channel/IRQ
556 * @name: Name of the channel/IRQ
557 *
558 * Unlike &struct efx_channel, this is never reallocated and is always
559 * safe for the IRQ handler to access.
560 */
561struct efx_msi_context {
562 struct efx_nic *efx;
563 unsigned int index;
564 char name[IFNAMSIZ + 6];
565};
566
567/**
568 * struct efx_channel_type - distinguishes traffic and extra channels
569 * @handle_no_channel: Handle failure to allocate an extra channel
570 * @pre_probe: Set up extra state prior to initialisation
571 * @post_remove: Tear down extra state after finalisation, if allocated.
572 * May be called on channels that have not been probed.
573 * @get_name: Generate the channel's name (used for its IRQ handler)
574 * @copy: Copy the channel state prior to reallocation. May be %NULL if
575 * reallocation is not supported.
576 * @receive_skb: Handle an skb ready to be passed to netif_receive_skb()
577 * @want_txqs: Determine whether this channel should have TX queues
578 * created. If %NULL, TX queues are not created.
579 * @keep_eventq: Flag for whether event queue should be kept initialised
580 * while the device is stopped
581 * @want_pio: Flag for whether PIO buffers should be linked to this
582 * channel's TX queues.
583 */
584struct efx_channel_type {
585 void (*handle_no_channel)(struct efx_nic *);
586 int (*pre_probe)(struct efx_channel *);
587 void (*post_remove)(struct efx_channel *);
588 void (*get_name)(struct efx_channel *, char *buf, size_t len);
589 struct efx_channel *(*copy)(const struct efx_channel *);
590 bool (*receive_skb)(struct efx_channel *, struct sk_buff *);
591 bool (*want_txqs)(struct efx_channel *);
592 bool keep_eventq;
593 bool want_pio;
594};
595
596enum efx_led_mode {
597 EFX_LED_OFF = 0,
598 EFX_LED_ON = 1,
599 EFX_LED_DEFAULT = 2
600};
601
602#define STRING_TABLE_LOOKUP(val, member) \
603 ((val) < member ## _max) ? member ## _names[val] : "(invalid)"
604
605extern const char *const efx_loopback_mode_names[];
606extern const unsigned int efx_loopback_mode_max;
607#define LOOPBACK_MODE(efx) \
608 STRING_TABLE_LOOKUP((efx)->loopback_mode, efx_loopback_mode)
609
610extern const char *const efx_reset_type_names[];
611extern const unsigned int efx_reset_type_max;
612#define RESET_TYPE(type) \
613 STRING_TABLE_LOOKUP(type, efx_reset_type)
614
615enum efx_int_mode {
616 /* Be careful if altering to correct macro below */
617 EFX_INT_MODE_MSIX = 0,
618 EFX_INT_MODE_MSI = 1,
619 EFX_INT_MODE_LEGACY = 2,
620 EFX_INT_MODE_MAX /* Insert any new items before this */
621};
622#define EFX_INT_MODE_USE_MSI(x) (((x)->interrupt_mode) <= EFX_INT_MODE_MSI)
623
624enum nic_state {
625 STATE_UNINIT = 0, /* device being probed/removed or is frozen */
626 STATE_READY = 1, /* hardware ready and netdev registered */
627 STATE_DISABLED = 2, /* device disabled due to hardware errors */
628 STATE_RECOVERY = 3, /* device recovering from PCI error */
629};
630
631/* Forward declaration */
632struct efx_nic;
633
634/* Pseudo bit-mask flow control field */
635#define EFX_FC_RX FLOW_CTRL_RX
636#define EFX_FC_TX FLOW_CTRL_TX
637#define EFX_FC_AUTO 4
638
639/**
640 * struct efx_link_state - Current state of the link
641 * @up: Link is up
642 * @fd: Link is full-duplex
643 * @fc: Actual flow control flags
644 * @speed: Link speed (Mbps)
645 */
646struct efx_link_state {
647 bool up;
648 bool fd;
649 u8 fc;
650 unsigned int speed;
651};
652
653static inline bool efx_link_state_equal(const struct efx_link_state *left,
654 const struct efx_link_state *right)
655{
656 return left->up == right->up && left->fd == right->fd &&
657 left->fc == right->fc && left->speed == right->speed;
658}
659
660/**
661 * struct efx_phy_operations - Efx PHY operations table
662 * @probe: Probe PHY and initialise efx->mdio.mode_support, efx->mdio.mmds,
663 * efx->loopback_modes.
664 * @init: Initialise PHY
665 * @fini: Shut down PHY
666 * @reconfigure: Reconfigure PHY (e.g. for new link parameters)
667 * @poll: Update @link_state and report whether it changed.
668 * Serialised by the mac_lock.
669 * @get_link_ksettings: Get ethtool settings. Serialised by the mac_lock.
670 * @set_link_ksettings: Set ethtool settings. Serialised by the mac_lock.
671 * @get_fecparam: Get Forward Error Correction settings. Serialised by mac_lock.
672 * @set_fecparam: Set Forward Error Correction settings. Serialised by mac_lock.
673 * @set_npage_adv: Set abilities advertised in (Extended) Next Page
674 * (only needed where AN bit is set in mmds)
675 * @test_alive: Test that PHY is 'alive' (online)
676 * @test_name: Get the name of a PHY-specific test/result
677 * @run_tests: Run tests and record results as appropriate (offline).
678 * Flags are the ethtool tests flags.
679 */
680struct efx_phy_operations {
681 int (*probe) (struct efx_nic *efx);
682 int (*init) (struct efx_nic *efx);
683 void (*fini) (struct efx_nic *efx);
684 void (*remove) (struct efx_nic *efx);
685 int (*reconfigure) (struct efx_nic *efx);
686 bool (*poll) (struct efx_nic *efx);
687 void (*get_link_ksettings)(struct efx_nic *efx,
688 struct ethtool_link_ksettings *cmd);
689 int (*set_link_ksettings)(struct efx_nic *efx,
690 const struct ethtool_link_ksettings *cmd);
691 int (*get_fecparam)(struct efx_nic *efx, struct ethtool_fecparam *fec);
692 int (*set_fecparam)(struct efx_nic *efx,
693 const struct ethtool_fecparam *fec);
694 void (*set_npage_adv) (struct efx_nic *efx, u32);
695 int (*test_alive) (struct efx_nic *efx);
696 const char *(*test_name) (struct efx_nic *efx, unsigned int index);
697 int (*run_tests) (struct efx_nic *efx, int *results, unsigned flags);
698 int (*get_module_eeprom) (struct efx_nic *efx,
699 struct ethtool_eeprom *ee,
700 u8 *data);
701 int (*get_module_info) (struct efx_nic *efx,
702 struct ethtool_modinfo *modinfo);
703};
704
705/**
706 * enum efx_phy_mode - PHY operating mode flags
707 * @PHY_MODE_NORMAL: on and should pass traffic
708 * @PHY_MODE_TX_DISABLED: on with TX disabled
709 * @PHY_MODE_LOW_POWER: set to low power through MDIO
710 * @PHY_MODE_OFF: switched off through external control
711 * @PHY_MODE_SPECIAL: on but will not pass traffic
712 */
713enum efx_phy_mode {
714 PHY_MODE_NORMAL = 0,
715 PHY_MODE_TX_DISABLED = 1,
716 PHY_MODE_LOW_POWER = 2,
717 PHY_MODE_OFF = 4,
718 PHY_MODE_SPECIAL = 8,
719};
720
721static inline bool efx_phy_mode_disabled(enum efx_phy_mode mode)
722{
723 return !!(mode & ~PHY_MODE_TX_DISABLED);
724}
725
726/**
727 * struct efx_hw_stat_desc - Description of a hardware statistic
728 * @name: Name of the statistic as visible through ethtool, or %NULL if
729 * it should not be exposed
730 * @dma_width: Width in bits (0 for non-DMA statistics)
731 * @offset: Offset within stats (ignored for non-DMA statistics)
732 */
733struct efx_hw_stat_desc {
734 const char *name;
735 u16 dma_width;
736 u16 offset;
737};
738
739/* Number of bits used in a multicast filter hash address */
740#define EFX_MCAST_HASH_BITS 8
741
742/* Number of (single-bit) entries in a multicast filter hash */
743#define EFX_MCAST_HASH_ENTRIES (1 << EFX_MCAST_HASH_BITS)
744
745/* An Efx multicast filter hash */
746union efx_multicast_hash {
747 u8 byte[EFX_MCAST_HASH_ENTRIES / 8];
748 efx_oword_t oword[EFX_MCAST_HASH_ENTRIES / sizeof(efx_oword_t) / 8];
749};
750
751struct vfdi_status;
752
753/* The reserved RSS context value */
754#define EFX_MCDI_RSS_CONTEXT_INVALID 0xffffffff
755/**
756 * struct efx_rss_context - A user-defined RSS context for filtering
757 * @list: node of linked list on which this struct is stored
758 * @context_id: the RSS_CONTEXT_ID returned by MC firmware, or
759 * %EFX_MCDI_RSS_CONTEXT_INVALID if this context is not present on the NIC.
760 * For Siena, 0 if RSS is active, else %EFX_MCDI_RSS_CONTEXT_INVALID.
761 * @user_id: the rss_context ID exposed to userspace over ethtool.
762 * @rx_hash_udp_4tuple: UDP 4-tuple hashing enabled
763 * @rx_hash_key: Toeplitz hash key for this RSS context
764 * @indir_table: Indirection table for this RSS context
765 */
766struct efx_rss_context {
767 struct list_head list;
768 u32 context_id;
769 u32 user_id;
770 bool rx_hash_udp_4tuple;
771 u8 rx_hash_key[40];
772 u32 rx_indir_table[128];
773};
774
775#ifdef CONFIG_RFS_ACCEL
776/* Order of these is important, since filter_id >= %EFX_ARFS_FILTER_ID_PENDING
777 * is used to test if filter does or will exist.
778 */
779#define EFX_ARFS_FILTER_ID_PENDING -1
780#define EFX_ARFS_FILTER_ID_ERROR -2
781#define EFX_ARFS_FILTER_ID_REMOVING -3
782/**
783 * struct efx_arfs_rule - record of an ARFS filter and its IDs
784 * @node: linkage into hash table
785 * @spec: details of the filter (used as key for hash table). Use efx->type to
786 * determine which member to use.
787 * @rxq_index: channel to which the filter will steer traffic.
788 * @arfs_id: filter ID which was returned to ARFS
789 * @filter_id: index in software filter table. May be
790 * %EFX_ARFS_FILTER_ID_PENDING if filter was not inserted yet,
791 * %EFX_ARFS_FILTER_ID_ERROR if filter insertion failed, or
792 * %EFX_ARFS_FILTER_ID_REMOVING if expiry is currently removing the filter.
793 */
794struct efx_arfs_rule {
795 struct hlist_node node;
796 struct efx_filter_spec spec;
797 u16 rxq_index;
798 u16 arfs_id;
799 s32 filter_id;
800};
801
802/* Size chosen so that the table is one page (4kB) */
803#define EFX_ARFS_HASH_TABLE_SIZE 512
804
805/**
806 * struct efx_async_filter_insertion - Request to asynchronously insert a filter
807 * @net_dev: Reference to the netdevice
808 * @spec: The filter to insert
809 * @work: Workitem for this request
810 * @rxq_index: Identifies the channel for which this request was made
811 * @flow_id: Identifies the kernel-side flow for which this request was made
812 */
813struct efx_async_filter_insertion {
814 struct net_device *net_dev;
815 struct efx_filter_spec spec;
816 struct work_struct work;
817 u16 rxq_index;
818 u32 flow_id;
819};
820
821/* Maximum number of ARFS workitems that may be in flight on an efx_nic */
822#define EFX_RPS_MAX_IN_FLIGHT 8
823#endif /* CONFIG_RFS_ACCEL */
824
825/**
826 * struct efx_nic - an Efx NIC
827 * @name: Device name (net device name or bus id before net device registered)
828 * @pci_dev: The PCI device
829 * @node: List node for maintaning primary/secondary function lists
830 * @primary: &struct efx_nic instance for the primary function of this
831 * controller. May be the same structure, and may be %NULL if no
832 * primary function is bound. Serialised by rtnl_lock.
833 * @secondary_list: List of &struct efx_nic instances for the secondary PCI
834 * functions of the controller, if this is for the primary function.
835 * Serialised by rtnl_lock.
836 * @type: Controller type attributes
837 * @legacy_irq: IRQ number
838 * @workqueue: Workqueue for port reconfigures and the HW monitor.
839 * Work items do not hold and must not acquire RTNL.
840 * @workqueue_name: Name of workqueue
841 * @reset_work: Scheduled reset workitem
842 * @membase_phys: Memory BAR value as physical address
843 * @membase: Memory BAR value
844 * @vi_stride: step between per-VI registers / memory regions
845 * @interrupt_mode: Interrupt mode
846 * @timer_quantum_ns: Interrupt timer quantum, in nanoseconds
847 * @timer_max_ns: Interrupt timer maximum value, in nanoseconds
848 * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues
849 * @irqs_hooked: Channel interrupts are hooked
850 * @irq_rx_mod_step_us: Step size for IRQ moderation for RX event queues
851 * @irq_rx_moderation_us: IRQ moderation time for RX event queues
852 * @msg_enable: Log message enable flags
853 * @state: Device state number (%STATE_*). Serialised by the rtnl_lock.
854 * @reset_pending: Bitmask for pending resets
855 * @tx_queue: TX DMA queues
856 * @rx_queue: RX DMA queues
857 * @channel: Channels
858 * @msi_context: Context for each MSI
859 * @extra_channel_types: Types of extra (non-traffic) channels that
860 * should be allocated for this NIC
861 * @xdp_tx_queue_count: Number of entries in %xdp_tx_queues.
862 * @xdp_tx_queues: Array of pointers to tx queues used for XDP transmit.
863 * @rxq_entries: Size of receive queues requested by user.
864 * @txq_entries: Size of transmit queues requested by user.
865 * @txq_stop_thresh: TX queue fill level at or above which we stop it.
866 * @txq_wake_thresh: TX queue fill level at or below which we wake it.
867 * @tx_dc_base: Base qword address in SRAM of TX queue descriptor caches
868 * @rx_dc_base: Base qword address in SRAM of RX queue descriptor caches
869 * @sram_lim_qw: Qword address limit of SRAM
870 * @next_buffer_table: First available buffer table id
871 * @n_channels: Number of channels in use
872 * @n_rx_channels: Number of channels used for RX (= number of RX queues)
873 * @n_tx_channels: Number of channels used for TX
874 * @n_extra_tx_channels: Number of extra channels with TX queues
875 * @tx_queues_per_channel: number of TX queues probed on each channel
876 * @n_xdp_channels: Number of channels used for XDP TX
877 * @xdp_channel_offset: Offset of zeroth channel used for XPD TX.
878 * @xdp_tx_per_channel: Max number of TX queues on an XDP TX channel.
879 * @rx_ip_align: RX DMA address offset to have IP header aligned in
880 * in accordance with NET_IP_ALIGN
881 * @rx_dma_len: Current maximum RX DMA length
882 * @rx_buffer_order: Order (log2) of number of pages for each RX buffer
883 * @rx_buffer_truesize: Amortised allocation size of an RX buffer,
884 * for use in sk_buff::truesize
885 * @rx_prefix_size: Size of RX prefix before packet data
886 * @rx_packet_hash_offset: Offset of RX flow hash from start of packet data
887 * (valid only if @rx_prefix_size != 0; always negative)
888 * @rx_packet_len_offset: Offset of RX packet length from start of packet data
889 * (valid only for NICs that set %EFX_RX_PKT_PREFIX_LEN; always negative)
890 * @rx_packet_ts_offset: Offset of timestamp from start of packet data
891 * (valid only if channel->sync_timestamps_enabled; always negative)
892 * @rx_scatter: Scatter mode enabled for receives
893 * @rss_context: Main RSS context. Its @list member is the head of the list of
894 * RSS contexts created by user requests
895 * @rss_lock: Protects custom RSS context software state in @rss_context.list
896 * @vport_id: The function's vport ID, only relevant for PFs
897 * @int_error_count: Number of internal errors seen recently
898 * @int_error_expire: Time at which error count will be expired
899 * @must_realloc_vis: Flag: VIs have yet to be reallocated after MC reboot
900 * @irq_soft_enabled: Are IRQs soft-enabled? If not, IRQ handler will
901 * acknowledge but do nothing else.
902 * @irq_status: Interrupt status buffer
903 * @irq_zero_count: Number of legacy IRQs seen with queue flags == 0
904 * @irq_level: IRQ level/index for IRQs not triggered by an event queue
905 * @selftest_work: Work item for asynchronous self-test
906 * @mtd_list: List of MTDs attached to the NIC
907 * @nic_data: Hardware dependent state
908 * @mcdi: Management-Controller-to-Driver Interface state
909 * @mac_lock: MAC access lock. Protects @port_enabled, @phy_mode,
910 * efx_monitor() and efx_reconfigure_port()
911 * @port_enabled: Port enabled indicator.
912 * Serialises efx_stop_all(), efx_start_all(), efx_monitor() and
913 * efx_mac_work() with kernel interfaces. Safe to read under any
914 * one of the rtnl_lock, mac_lock, or netif_tx_lock, but all three must
915 * be held to modify it.
916 * @port_initialized: Port initialized?
917 * @net_dev: Operating system network device. Consider holding the rtnl lock
918 * @fixed_features: Features which cannot be turned off
919 * @num_mac_stats: Number of MAC stats reported by firmware (MAC_STATS_NUM_STATS
920 * field of %MC_CMD_GET_CAPABILITIES_V4 response, or %MC_CMD_MAC_NSTATS)
921 * @stats_buffer: DMA buffer for statistics
922 * @phy_type: PHY type
923 * @phy_op: PHY interface
924 * @phy_data: PHY private data (including PHY-specific stats)
925 * @mdio: PHY MDIO interface
926 * @mdio_bus: PHY MDIO bus ID (only used by Siena)
927 * @phy_mode: PHY operating mode. Serialised by @mac_lock.
928 * @link_advertising: Autonegotiation advertising flags
929 * @fec_config: Forward Error Correction configuration flags. For bit positions
930 * see &enum ethtool_fec_config_bits.
931 * @link_state: Current state of the link
932 * @n_link_state_changes: Number of times the link has changed state
933 * @unicast_filter: Flag for Falcon-arch simple unicast filter.
934 * Protected by @mac_lock.
935 * @multicast_hash: Multicast hash table for Falcon-arch.
936 * Protected by @mac_lock.
937 * @wanted_fc: Wanted flow control flags
938 * @fc_disable: When non-zero flow control is disabled. Typically used to
939 * ensure that network back pressure doesn't delay dma queue flushes.
940 * Serialised by the rtnl lock.
941 * @mac_work: Work item for changing MAC promiscuity and multicast hash
942 * @loopback_mode: Loopback status
943 * @loopback_modes: Supported loopback mode bitmask
944 * @loopback_selftest: Offline self-test private state
945 * @xdp_prog: Current XDP programme for this interface
946 * @filter_sem: Filter table rw_semaphore, protects existence of @filter_state
947 * @filter_state: Architecture-dependent filter table state
948 * @rps_mutex: Protects RPS state of all channels
949 * @rps_slot_map: bitmap of in-flight entries in @rps_slot
950 * @rps_slot: array of ARFS insertion requests for efx_filter_rfs_work()
951 * @rps_hash_lock: Protects ARFS filter mapping state (@rps_hash_table and
952 * @rps_next_id).
953 * @rps_hash_table: Mapping between ARFS filters and their various IDs
954 * @rps_next_id: next arfs_id for an ARFS filter
955 * @active_queues: Count of RX and TX queues that haven't been flushed and drained.
956 * @rxq_flush_pending: Count of number of receive queues that need to be flushed.
957 * Decremented when the efx_flush_rx_queue() is called.
958 * @rxq_flush_outstanding: Count of number of RX flushes started but not yet
959 * completed (either success or failure). Not used when MCDI is used to
960 * flush receive queues.
961 * @flush_wq: wait queue used by efx_nic_flush_queues() to wait for flush completions.
962 * @vf_count: Number of VFs intended to be enabled.
963 * @vf_init_count: Number of VFs that have been fully initialised.
964 * @vi_scale: log2 number of vnics per VF.
965 * @ptp_data: PTP state data
966 * @ptp_warned: has this NIC seen and warned about unexpected PTP events?
967 * @vpd_sn: Serial number read from VPD
968 * @xdp_rxq_info_failed: Have any of the rx queues failed to initialise their
969 * xdp_rxq_info structures?
970 * @netdev_notifier: Netdevice notifier.
971 * @mem_bar: The BAR that is mapped into membase.
972 * @reg_base: Offset from the start of the bar to the function control window.
973 * @monitor_work: Hardware monitor workitem
974 * @biu_lock: BIU (bus interface unit) lock
975 * @last_irq_cpu: Last CPU to handle a possible test interrupt. This
976 * field is used by efx_test_interrupts() to verify that an
977 * interrupt has occurred.
978 * @stats_lock: Statistics update lock. Must be held when calling
979 * efx_nic_type::{update,start,stop}_stats.
980 * @n_rx_noskb_drops: Count of RX packets dropped due to failure to allocate an skb
981 *
982 * This is stored in the private area of the &struct net_device.
983 */
984struct efx_nic {
985 /* The following fields should be written very rarely */
986
987 char name[IFNAMSIZ];
988 struct list_head node;
989 struct efx_nic *primary;
990 struct list_head secondary_list;
991 struct pci_dev *pci_dev;
992 unsigned int port_num;
993 const struct efx_nic_type *type;
994 int legacy_irq;
995 bool eeh_disabled_legacy_irq;
996 struct workqueue_struct *workqueue;
997 char workqueue_name[16];
998 struct work_struct reset_work;
999 resource_size_t membase_phys;
1000 void __iomem *membase;
1001
1002 unsigned int vi_stride;
1003
1004 enum efx_int_mode interrupt_mode;
1005 unsigned int timer_quantum_ns;
1006 unsigned int timer_max_ns;
1007 bool irq_rx_adaptive;
1008 bool irqs_hooked;
1009 unsigned int irq_mod_step_us;
1010 unsigned int irq_rx_moderation_us;
1011 u32 msg_enable;
1012
1013 enum nic_state state;
1014 unsigned long reset_pending;
1015
1016 struct efx_channel *channel[EFX_MAX_CHANNELS];
1017 struct efx_msi_context msi_context[EFX_MAX_CHANNELS];
1018 const struct efx_channel_type *
1019 extra_channel_type[EFX_MAX_EXTRA_CHANNELS];
1020
1021 unsigned int xdp_tx_queue_count;
1022 struct efx_tx_queue **xdp_tx_queues;
1023
1024 unsigned rxq_entries;
1025 unsigned txq_entries;
1026 unsigned int txq_stop_thresh;
1027 unsigned int txq_wake_thresh;
1028
1029 unsigned tx_dc_base;
1030 unsigned rx_dc_base;
1031 unsigned sram_lim_qw;
1032 unsigned next_buffer_table;
1033
1034 unsigned int max_channels;
1035 unsigned int max_vis;
1036 unsigned int max_tx_channels;
1037 unsigned n_channels;
1038 unsigned n_rx_channels;
1039 unsigned rss_spread;
1040 unsigned tx_channel_offset;
1041 unsigned n_tx_channels;
1042 unsigned n_extra_tx_channels;
1043 unsigned int tx_queues_per_channel;
1044 unsigned int n_xdp_channels;
1045 unsigned int xdp_channel_offset;
1046 unsigned int xdp_tx_per_channel;
1047 unsigned int rx_ip_align;
1048 unsigned int rx_dma_len;
1049 unsigned int rx_buffer_order;
1050 unsigned int rx_buffer_truesize;
1051 unsigned int rx_page_buf_step;
1052 unsigned int rx_bufs_per_page;
1053 unsigned int rx_pages_per_batch;
1054 unsigned int rx_prefix_size;
1055 int rx_packet_hash_offset;
1056 int rx_packet_len_offset;
1057 int rx_packet_ts_offset;
1058 bool rx_scatter;
1059 struct efx_rss_context rss_context;
1060 struct mutex rss_lock;
1061 u32 vport_id;
1062
1063 unsigned int_error_count;
1064 unsigned long int_error_expire;
1065
1066 bool must_realloc_vis;
1067 bool irq_soft_enabled;
1068 struct efx_buffer irq_status;
1069 unsigned irq_zero_count;
1070 unsigned irq_level;
1071 struct delayed_work selftest_work;
1072
1073#ifdef CONFIG_SFC_MTD
1074 struct list_head mtd_list;
1075#endif
1076
1077 void *nic_data;
1078 struct efx_mcdi_data *mcdi;
1079
1080 struct mutex mac_lock;
1081 struct work_struct mac_work;
1082 bool port_enabled;
1083
1084 bool mc_bist_for_other_fn;
1085 bool port_initialized;
1086 struct net_device *net_dev;
1087
1088 netdev_features_t fixed_features;
1089
1090 u16 num_mac_stats;
1091 struct efx_buffer stats_buffer;
1092 u64 rx_nodesc_drops_total;
1093 u64 rx_nodesc_drops_while_down;
1094 bool rx_nodesc_drops_prev_state;
1095
1096 unsigned int phy_type;
1097 const struct efx_phy_operations *phy_op;
1098 void *phy_data;
1099 struct mdio_if_info mdio;
1100 unsigned int mdio_bus;
1101 enum efx_phy_mode phy_mode;
1102
1103 __ETHTOOL_DECLARE_LINK_MODE_MASK(link_advertising);
1104 u32 fec_config;
1105 struct efx_link_state link_state;
1106 unsigned int n_link_state_changes;
1107
1108 bool unicast_filter;
1109 union efx_multicast_hash multicast_hash;
1110 u8 wanted_fc;
1111 unsigned fc_disable;
1112
1113 atomic_t rx_reset;
1114 enum efx_loopback_mode loopback_mode;
1115 u64 loopback_modes;
1116
1117 void *loopback_selftest;
1118 /* We access loopback_selftest immediately before running XDP,
1119 * so we want them next to each other.
1120 */
1121 struct bpf_prog __rcu *xdp_prog;
1122
1123 struct rw_semaphore filter_sem;
1124 void *filter_state;
1125#ifdef CONFIG_RFS_ACCEL
1126 struct mutex rps_mutex;
1127 unsigned long rps_slot_map;
1128 struct efx_async_filter_insertion rps_slot[EFX_RPS_MAX_IN_FLIGHT];
1129 spinlock_t rps_hash_lock;
1130 struct hlist_head *rps_hash_table;
1131 u32 rps_next_id;
1132#endif
1133
1134 atomic_t active_queues;
1135 atomic_t rxq_flush_pending;
1136 atomic_t rxq_flush_outstanding;
1137 wait_queue_head_t flush_wq;
1138
1139#ifdef CONFIG_SFC_SRIOV
1140 unsigned vf_count;
1141 unsigned vf_init_count;
1142 unsigned vi_scale;
1143#endif
1144
1145 struct efx_ptp_data *ptp_data;
1146 bool ptp_warned;
1147
1148 char *vpd_sn;
1149 bool xdp_rxq_info_failed;
1150
1151 struct notifier_block netdev_notifier;
1152
1153 unsigned int mem_bar;
1154 u32 reg_base;
1155
1156 /* The following fields may be written more often */
1157
1158 struct delayed_work monitor_work ____cacheline_aligned_in_smp;
1159 spinlock_t biu_lock;
1160 int last_irq_cpu;
1161 spinlock_t stats_lock;
1162 atomic_t n_rx_noskb_drops;
1163};
1164
1165static inline int efx_dev_registered(struct efx_nic *efx)
1166{
1167 return efx->net_dev->reg_state == NETREG_REGISTERED;
1168}
1169
1170static inline unsigned int efx_port_num(struct efx_nic *efx)
1171{
1172 return efx->port_num;
1173}
1174
1175struct efx_mtd_partition {
1176 struct list_head node;
1177 struct mtd_info mtd;
1178 const char *dev_type_name;
1179 const char *type_name;
1180 char name[IFNAMSIZ + 20];
1181};
1182
1183struct efx_udp_tunnel {
1184#define TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID 0xffff
1185 u16 type; /* TUNNEL_ENCAP_UDP_PORT_ENTRY_foo, see mcdi_pcol.h */
1186 __be16 port;
1187};
1188
1189/**
1190 * struct efx_nic_type - Efx device type definition
1191 * @mem_bar: Get the memory BAR
1192 * @mem_map_size: Get memory BAR mapped size
1193 * @probe: Probe the controller
1194 * @remove: Free resources allocated by probe()
1195 * @init: Initialise the controller
1196 * @dimension_resources: Dimension controller resources (buffer table,
1197 * and VIs once the available interrupt resources are clear)
1198 * @fini: Shut down the controller
1199 * @monitor: Periodic function for polling link state and hardware monitor
1200 * @map_reset_reason: Map ethtool reset reason to a reset method
1201 * @map_reset_flags: Map ethtool reset flags to a reset method, if possible
1202 * @reset: Reset the controller hardware and possibly the PHY. This will
1203 * be called while the controller is uninitialised.
1204 * @probe_port: Probe the MAC and PHY
1205 * @remove_port: Free resources allocated by probe_port()
1206 * @handle_global_event: Handle a "global" event (may be %NULL)
1207 * @fini_dmaq: Flush and finalise DMA queues (RX and TX queues)
1208 * @prepare_flush: Prepare the hardware for flushing the DMA queues
1209 * (for Falcon architecture)
1210 * @finish_flush: Clean up after flushing the DMA queues (for Falcon
1211 * architecture)
1212 * @prepare_flr: Prepare for an FLR
1213 * @finish_flr: Clean up after an FLR
1214 * @describe_stats: Describe statistics for ethtool
1215 * @update_stats: Update statistics not provided by event handling.
1216 * Either argument may be %NULL.
1217 * @start_stats: Start the regular fetching of statistics
1218 * @pull_stats: Pull stats from the NIC and wait until they arrive.
1219 * @stop_stats: Stop the regular fetching of statistics
1220 * @set_id_led: Set state of identifying LED or revert to automatic function
1221 * @push_irq_moderation: Apply interrupt moderation value
1222 * @reconfigure_port: Push loopback/power/txdis changes to the MAC and PHY
1223 * @prepare_enable_fc_tx: Prepare MAC to enable pause frame TX (may be %NULL)
1224 * @reconfigure_mac: Push MAC address, MTU, flow control and filter settings
1225 * to the hardware. Serialised by the mac_lock.
1226 * @check_mac_fault: Check MAC fault state. True if fault present.
1227 * @get_wol: Get WoL configuration from driver state
1228 * @set_wol: Push WoL configuration to the NIC
1229 * @resume_wol: Synchronise WoL state between driver and MC (e.g. after resume)
1230 * @test_chip: Test registers. May use efx_farch_test_registers(), and is
1231 * expected to reset the NIC.
1232 * @test_nvram: Test validity of NVRAM contents
1233 * @mcdi_request: Send an MCDI request with the given header and SDU.
1234 * The SDU length may be any value from 0 up to the protocol-
1235 * defined maximum, but its buffer will be padded to a multiple
1236 * of 4 bytes.
1237 * @mcdi_poll_response: Test whether an MCDI response is available.
1238 * @mcdi_read_response: Read the MCDI response PDU. The offset will
1239 * be a multiple of 4. The length may not be, but the buffer
1240 * will be padded so it is safe to round up.
1241 * @mcdi_poll_reboot: Test whether the MCDI has rebooted. If so,
1242 * return an appropriate error code for aborting any current
1243 * request; otherwise return 0.
1244 * @irq_enable_master: Enable IRQs on the NIC. Each event queue must
1245 * be separately enabled after this.
1246 * @irq_test_generate: Generate a test IRQ
1247 * @irq_disable_non_ev: Disable non-event IRQs on the NIC. Each event
1248 * queue must be separately disabled before this.
1249 * @irq_handle_msi: Handle MSI for a channel. The @dev_id argument is
1250 * a pointer to the &struct efx_msi_context for the channel.
1251 * @irq_handle_legacy: Handle legacy interrupt. The @dev_id argument
1252 * is a pointer to the &struct efx_nic.
1253 * @tx_probe: Allocate resources for TX queue
1254 * @tx_init: Initialise TX queue on the NIC
1255 * @tx_remove: Free resources for TX queue
1256 * @tx_write: Write TX descriptors and doorbell
1257 * @tx_enqueue: Add an SKB to TX queue
1258 * @rx_push_rss_config: Write RSS hash key and indirection table to the NIC
1259 * @rx_pull_rss_config: Read RSS hash key and indirection table back from the NIC
1260 * @rx_push_rss_context_config: Write RSS hash key and indirection table for
1261 * user RSS context to the NIC
1262 * @rx_pull_rss_context_config: Read RSS hash key and indirection table for user
1263 * RSS context back from the NIC
1264 * @rx_probe: Allocate resources for RX queue
1265 * @rx_init: Initialise RX queue on the NIC
1266 * @rx_remove: Free resources for RX queue
1267 * @rx_write: Write RX descriptors and doorbell
1268 * @rx_defer_refill: Generate a refill reminder event
1269 * @rx_packet: Receive the queued RX buffer on a channel
1270 * @rx_buf_hash_valid: Determine whether the RX prefix contains a valid hash
1271 * @ev_probe: Allocate resources for event queue
1272 * @ev_init: Initialise event queue on the NIC
1273 * @ev_fini: Deinitialise event queue on the NIC
1274 * @ev_remove: Free resources for event queue
1275 * @ev_process: Process events for a queue, up to the given NAPI quota
1276 * @ev_read_ack: Acknowledge read events on a queue, rearming its IRQ
1277 * @ev_test_generate: Generate a test event
1278 * @filter_table_probe: Probe filter capabilities and set up filter software state
1279 * @filter_table_restore: Restore filters removed from hardware
1280 * @filter_table_remove: Remove filters from hardware and tear down software state
1281 * @filter_update_rx_scatter: Update filters after change to rx scatter setting
1282 * @filter_insert: add or replace a filter
1283 * @filter_remove_safe: remove a filter by ID, carefully
1284 * @filter_get_safe: retrieve a filter by ID, carefully
1285 * @filter_clear_rx: Remove all RX filters whose priority is less than or
1286 * equal to the given priority and is not %EFX_FILTER_PRI_AUTO
1287 * @filter_count_rx_used: Get the number of filters in use at a given priority
1288 * @filter_get_rx_id_limit: Get maximum value of a filter id, plus 1
1289 * @filter_get_rx_ids: Get list of RX filters at a given priority
1290 * @filter_rfs_expire_one: Consider expiring a filter inserted for RFS.
1291 * This must check whether the specified table entry is used by RFS
1292 * and that rps_may_expire_flow() returns true for it.
1293 * @mtd_probe: Probe and add MTD partitions associated with this net device,
1294 * using efx_mtd_add()
1295 * @mtd_rename: Set an MTD partition name using the net device name
1296 * @mtd_read: Read from an MTD partition
1297 * @mtd_erase: Erase part of an MTD partition
1298 * @mtd_write: Write to an MTD partition
1299 * @mtd_sync: Wait for write-back to complete on MTD partition. This
1300 * also notifies the driver that a writer has finished using this
1301 * partition.
1302 * @ptp_write_host_time: Send host time to MC as part of sync protocol
1303 * @ptp_set_ts_sync_events: Enable or disable sync events for inline RX
1304 * timestamping, possibly only temporarily for the purposes of a reset.
1305 * @ptp_set_ts_config: Set hardware timestamp configuration. The flags
1306 * and tx_type will already have been validated but this operation
1307 * must validate and update rx_filter.
1308 * @get_phys_port_id: Get the underlying physical port id.
1309 * @set_mac_address: Set the MAC address of the device
1310 * @tso_versions: Returns mask of firmware-assisted TSO versions supported.
1311 * If %NULL, then device does not support any TSO version.
1312 * @udp_tnl_push_ports: Push the list of UDP tunnel ports to the NIC if required.
1313 * @udp_tnl_has_port: Check if a port has been added as UDP tunnel
1314 * @print_additional_fwver: Dump NIC-specific additional FW version info
1315 * @sensor_event: Handle a sensor event from MCDI
1316 * @revision: Hardware architecture revision
1317 * @txd_ptr_tbl_base: TX descriptor ring base address
1318 * @rxd_ptr_tbl_base: RX descriptor ring base address
1319 * @buf_tbl_base: Buffer table base address
1320 * @evq_ptr_tbl_base: Event queue pointer table base address
1321 * @evq_rptr_tbl_base: Event queue read-pointer table base address
1322 * @max_dma_mask: Maximum possible DMA mask
1323 * @rx_prefix_size: Size of RX prefix before packet data
1324 * @rx_hash_offset: Offset of RX flow hash within prefix
1325 * @rx_ts_offset: Offset of timestamp within prefix
1326 * @rx_buffer_padding: Size of padding at end of RX packet
1327 * @can_rx_scatter: NIC is able to scatter packets to multiple buffers
1328 * @always_rx_scatter: NIC will always scatter packets to multiple buffers
1329 * @option_descriptors: NIC supports TX option descriptors
1330 * @min_interrupt_mode: Lowest capability interrupt mode supported
1331 * from &enum efx_int_mode.
1332 * @timer_period_max: Maximum period of interrupt timer (in ticks)
1333 * @offload_features: net_device feature flags for protocol offload
1334 * features implemented in hardware
1335 * @mcdi_max_ver: Maximum MCDI version supported
1336 * @hwtstamp_filters: Mask of hardware timestamp filter types supported
1337 */
1338struct efx_nic_type {
1339 bool is_vf;
1340 unsigned int (*mem_bar)(struct efx_nic *efx);
1341 unsigned int (*mem_map_size)(struct efx_nic *efx);
1342 int (*probe)(struct efx_nic *efx);
1343 void (*remove)(struct efx_nic *efx);
1344 int (*init)(struct efx_nic *efx);
1345 int (*dimension_resources)(struct efx_nic *efx);
1346 void (*fini)(struct efx_nic *efx);
1347 void (*monitor)(struct efx_nic *efx);
1348 enum reset_type (*map_reset_reason)(enum reset_type reason);
1349 int (*map_reset_flags)(u32 *flags);
1350 int (*reset)(struct efx_nic *efx, enum reset_type method);
1351 int (*probe_port)(struct efx_nic *efx);
1352 void (*remove_port)(struct efx_nic *efx);
1353 bool (*handle_global_event)(struct efx_channel *channel, efx_qword_t *);
1354 int (*fini_dmaq)(struct efx_nic *efx);
1355 void (*prepare_flush)(struct efx_nic *efx);
1356 void (*finish_flush)(struct efx_nic *efx);
1357 void (*prepare_flr)(struct efx_nic *efx);
1358 void (*finish_flr)(struct efx_nic *efx);
1359 size_t (*describe_stats)(struct efx_nic *efx, u8 *names);
1360 size_t (*update_stats)(struct efx_nic *efx, u64 *full_stats,
1361 struct rtnl_link_stats64 *core_stats);
1362 void (*start_stats)(struct efx_nic *efx);
1363 void (*pull_stats)(struct efx_nic *efx);
1364 void (*stop_stats)(struct efx_nic *efx);
1365 void (*set_id_led)(struct efx_nic *efx, enum efx_led_mode mode);
1366 void (*push_irq_moderation)(struct efx_channel *channel);
1367 int (*reconfigure_port)(struct efx_nic *efx);
1368 void (*prepare_enable_fc_tx)(struct efx_nic *efx);
1369 int (*reconfigure_mac)(struct efx_nic *efx, bool mtu_only);
1370 bool (*check_mac_fault)(struct efx_nic *efx);
1371 void (*get_wol)(struct efx_nic *efx, struct ethtool_wolinfo *wol);
1372 int (*set_wol)(struct efx_nic *efx, u32 type);
1373 void (*resume_wol)(struct efx_nic *efx);
1374 unsigned int (*check_caps)(const struct efx_nic *efx,
1375 u8 flag,
1376 u32 offset);
1377 int (*test_chip)(struct efx_nic *efx, struct efx_self_tests *tests);
1378 int (*test_nvram)(struct efx_nic *efx);
1379 void (*mcdi_request)(struct efx_nic *efx,
1380 const efx_dword_t *hdr, size_t hdr_len,
1381 const efx_dword_t *sdu, size_t sdu_len);
1382 bool (*mcdi_poll_response)(struct efx_nic *efx);
1383 void (*mcdi_read_response)(struct efx_nic *efx, efx_dword_t *pdu,
1384 size_t pdu_offset, size_t pdu_len);
1385 int (*mcdi_poll_reboot)(struct efx_nic *efx);
1386 void (*mcdi_reboot_detected)(struct efx_nic *efx);
1387 void (*irq_enable_master)(struct efx_nic *efx);
1388 int (*irq_test_generate)(struct efx_nic *efx);
1389 void (*irq_disable_non_ev)(struct efx_nic *efx);
1390 irqreturn_t (*irq_handle_msi)(int irq, void *dev_id);
1391 irqreturn_t (*irq_handle_legacy)(int irq, void *dev_id);
1392 int (*tx_probe)(struct efx_tx_queue *tx_queue);
1393 void (*tx_init)(struct efx_tx_queue *tx_queue);
1394 void (*tx_remove)(struct efx_tx_queue *tx_queue);
1395 void (*tx_write)(struct efx_tx_queue *tx_queue);
1396 netdev_tx_t (*tx_enqueue)(struct efx_tx_queue *tx_queue, struct sk_buff *skb);
1397 unsigned int (*tx_limit_len)(struct efx_tx_queue *tx_queue,
1398 dma_addr_t dma_addr, unsigned int len);
1399 int (*rx_push_rss_config)(struct efx_nic *efx, bool user,
1400 const u32 *rx_indir_table, const u8 *key);
1401 int (*rx_pull_rss_config)(struct efx_nic *efx);
1402 int (*rx_push_rss_context_config)(struct efx_nic *efx,
1403 struct efx_rss_context *ctx,
1404 const u32 *rx_indir_table,
1405 const u8 *key);
1406 int (*rx_pull_rss_context_config)(struct efx_nic *efx,
1407 struct efx_rss_context *ctx);
1408 void (*rx_restore_rss_contexts)(struct efx_nic *efx);
1409 int (*rx_probe)(struct efx_rx_queue *rx_queue);
1410 void (*rx_init)(struct efx_rx_queue *rx_queue);
1411 void (*rx_remove)(struct efx_rx_queue *rx_queue);
1412 void (*rx_write)(struct efx_rx_queue *rx_queue);
1413 void (*rx_defer_refill)(struct efx_rx_queue *rx_queue);
1414 void (*rx_packet)(struct efx_channel *channel);
1415 bool (*rx_buf_hash_valid)(const u8 *prefix);
1416 int (*ev_probe)(struct efx_channel *channel);
1417 int (*ev_init)(struct efx_channel *channel);
1418 void (*ev_fini)(struct efx_channel *channel);
1419 void (*ev_remove)(struct efx_channel *channel);
1420 int (*ev_process)(struct efx_channel *channel, int quota);
1421 void (*ev_read_ack)(struct efx_channel *channel);
1422 void (*ev_test_generate)(struct efx_channel *channel);
1423 int (*filter_table_probe)(struct efx_nic *efx);
1424 void (*filter_table_restore)(struct efx_nic *efx);
1425 void (*filter_table_remove)(struct efx_nic *efx);
1426 void (*filter_update_rx_scatter)(struct efx_nic *efx);
1427 s32 (*filter_insert)(struct efx_nic *efx,
1428 struct efx_filter_spec *spec, bool replace);
1429 int (*filter_remove_safe)(struct efx_nic *efx,
1430 enum efx_filter_priority priority,
1431 u32 filter_id);
1432 int (*filter_get_safe)(struct efx_nic *efx,
1433 enum efx_filter_priority priority,
1434 u32 filter_id, struct efx_filter_spec *);
1435 int (*filter_clear_rx)(struct efx_nic *efx,
1436 enum efx_filter_priority priority);
1437 u32 (*filter_count_rx_used)(struct efx_nic *efx,
1438 enum efx_filter_priority priority);
1439 u32 (*filter_get_rx_id_limit)(struct efx_nic *efx);
1440 s32 (*filter_get_rx_ids)(struct efx_nic *efx,
1441 enum efx_filter_priority priority,
1442 u32 *buf, u32 size);
1443#ifdef CONFIG_RFS_ACCEL
1444 bool (*filter_rfs_expire_one)(struct efx_nic *efx, u32 flow_id,
1445 unsigned int index);
1446#endif
1447#ifdef CONFIG_SFC_MTD
1448 int (*mtd_probe)(struct efx_nic *efx);
1449 void (*mtd_rename)(struct efx_mtd_partition *part);
1450 int (*mtd_read)(struct mtd_info *mtd, loff_t start, size_t len,
1451 size_t *retlen, u8 *buffer);
1452 int (*mtd_erase)(struct mtd_info *mtd, loff_t start, size_t len);
1453 int (*mtd_write)(struct mtd_info *mtd, loff_t start, size_t len,
1454 size_t *retlen, const u8 *buffer);
1455 int (*mtd_sync)(struct mtd_info *mtd);
1456#endif
1457 void (*ptp_write_host_time)(struct efx_nic *efx, u32 host_time);
1458 int (*ptp_set_ts_sync_events)(struct efx_nic *efx, bool en, bool temp);
1459 int (*ptp_set_ts_config)(struct efx_nic *efx,
1460 struct hwtstamp_config *init);
1461 int (*sriov_configure)(struct efx_nic *efx, int num_vfs);
1462 int (*vlan_rx_add_vid)(struct efx_nic *efx, __be16 proto, u16 vid);
1463 int (*vlan_rx_kill_vid)(struct efx_nic *efx, __be16 proto, u16 vid);
1464 int (*get_phys_port_id)(struct efx_nic *efx,
1465 struct netdev_phys_item_id *ppid);
1466 int (*sriov_init)(struct efx_nic *efx);
1467 void (*sriov_fini)(struct efx_nic *efx);
1468 bool (*sriov_wanted)(struct efx_nic *efx);
1469 void (*sriov_reset)(struct efx_nic *efx);
1470 void (*sriov_flr)(struct efx_nic *efx, unsigned vf_i);
1471 int (*sriov_set_vf_mac)(struct efx_nic *efx, int vf_i, u8 *mac);
1472 int (*sriov_set_vf_vlan)(struct efx_nic *efx, int vf_i, u16 vlan,
1473 u8 qos);
1474 int (*sriov_set_vf_spoofchk)(struct efx_nic *efx, int vf_i,
1475 bool spoofchk);
1476 int (*sriov_get_vf_config)(struct efx_nic *efx, int vf_i,
1477 struct ifla_vf_info *ivi);
1478 int (*sriov_set_vf_link_state)(struct efx_nic *efx, int vf_i,
1479 int link_state);
1480 int (*vswitching_probe)(struct efx_nic *efx);
1481 int (*vswitching_restore)(struct efx_nic *efx);
1482 void (*vswitching_remove)(struct efx_nic *efx);
1483 int (*get_mac_address)(struct efx_nic *efx, unsigned char *perm_addr);
1484 int (*set_mac_address)(struct efx_nic *efx);
1485 u32 (*tso_versions)(struct efx_nic *efx);
1486 int (*udp_tnl_push_ports)(struct efx_nic *efx);
1487 bool (*udp_tnl_has_port)(struct efx_nic *efx, __be16 port);
1488 size_t (*print_additional_fwver)(struct efx_nic *efx, char *buf,
1489 size_t len);
1490 void (*sensor_event)(struct efx_nic *efx, efx_qword_t *ev);
1491
1492 int revision;
1493 unsigned int txd_ptr_tbl_base;
1494 unsigned int rxd_ptr_tbl_base;
1495 unsigned int buf_tbl_base;
1496 unsigned int evq_ptr_tbl_base;
1497 unsigned int evq_rptr_tbl_base;
1498 u64 max_dma_mask;
1499 unsigned int rx_prefix_size;
1500 unsigned int rx_hash_offset;
1501 unsigned int rx_ts_offset;
1502 unsigned int rx_buffer_padding;
1503 bool can_rx_scatter;
1504 bool always_rx_scatter;
1505 bool option_descriptors;
1506 unsigned int min_interrupt_mode;
1507 unsigned int timer_period_max;
1508 netdev_features_t offload_features;
1509 int mcdi_max_ver;
1510 unsigned int max_rx_ip_filters;
1511 u32 hwtstamp_filters;
1512 unsigned int rx_hash_key_size;
1513};
1514
1515/**************************************************************************
1516 *
1517 * Prototypes and inline functions
1518 *
1519 *************************************************************************/
1520
1521static inline struct efx_channel *
1522efx_get_channel(struct efx_nic *efx, unsigned index)
1523{
1524 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_channels);
1525 return efx->channel[index];
1526}
1527
1528/* Iterate over all used channels */
1529#define efx_for_each_channel(_channel, _efx) \
1530 for (_channel = (_efx)->channel[0]; \
1531 _channel; \
1532 _channel = (_channel->channel + 1 < (_efx)->n_channels) ? \
1533 (_efx)->channel[_channel->channel + 1] : NULL)
1534
1535/* Iterate over all used channels in reverse */
1536#define efx_for_each_channel_rev(_channel, _efx) \
1537 for (_channel = (_efx)->channel[(_efx)->n_channels - 1]; \
1538 _channel; \
1539 _channel = _channel->channel ? \
1540 (_efx)->channel[_channel->channel - 1] : NULL)
1541
1542static inline struct efx_channel *
1543efx_get_tx_channel(struct efx_nic *efx, unsigned int index)
1544{
1545 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_tx_channels);
1546 return efx->channel[efx->tx_channel_offset + index];
1547}
1548
1549static inline struct efx_tx_queue *
1550efx_get_tx_queue(struct efx_nic *efx, unsigned index, unsigned type)
1551{
1552 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_tx_channels ||
1553 type >= efx->tx_queues_per_channel);
1554 return &efx->channel[efx->tx_channel_offset + index]->tx_queue[type];
1555}
1556
1557static inline struct efx_channel *
1558efx_get_xdp_channel(struct efx_nic *efx, unsigned int index)
1559{
1560 EFX_WARN_ON_ONCE_PARANOID(index >= efx->n_xdp_channels);
1561 return efx->channel[efx->xdp_channel_offset + index];
1562}
1563
1564static inline bool efx_channel_is_xdp_tx(struct efx_channel *channel)
1565{
1566 return channel->channel - channel->efx->xdp_channel_offset <
1567 channel->efx->n_xdp_channels;
1568}
1569
1570static inline bool efx_channel_has_tx_queues(struct efx_channel *channel)
1571{
1572 return true;
1573}
1574
1575static inline unsigned int efx_channel_num_tx_queues(struct efx_channel *channel)
1576{
1577 if (efx_channel_is_xdp_tx(channel))
1578 return channel->efx->xdp_tx_per_channel;
1579 return channel->efx->tx_queues_per_channel;
1580}
1581
1582static inline struct efx_tx_queue *
1583efx_channel_get_tx_queue(struct efx_channel *channel, unsigned type)
1584{
1585 EFX_WARN_ON_ONCE_PARANOID(type >= efx_channel_num_tx_queues(channel));
1586 return &channel->tx_queue[type];
1587}
1588
1589/* Iterate over all TX queues belonging to a channel */
1590#define efx_for_each_channel_tx_queue(_tx_queue, _channel) \
1591 if (!efx_channel_has_tx_queues(_channel)) \
1592 ; \
1593 else \
1594 for (_tx_queue = (_channel)->tx_queue; \
1595 _tx_queue < (_channel)->tx_queue + \
1596 efx_channel_num_tx_queues(_channel); \
1597 _tx_queue++)
1598
1599static inline bool efx_channel_has_rx_queue(struct efx_channel *channel)
1600{
1601 return channel->rx_queue.core_index >= 0;
1602}
1603
1604static inline struct efx_rx_queue *
1605efx_channel_get_rx_queue(struct efx_channel *channel)
1606{
1607 EFX_WARN_ON_ONCE_PARANOID(!efx_channel_has_rx_queue(channel));
1608 return &channel->rx_queue;
1609}
1610
1611/* Iterate over all RX queues belonging to a channel */
1612#define efx_for_each_channel_rx_queue(_rx_queue, _channel) \
1613 if (!efx_channel_has_rx_queue(_channel)) \
1614 ; \
1615 else \
1616 for (_rx_queue = &(_channel)->rx_queue; \
1617 _rx_queue; \
1618 _rx_queue = NULL)
1619
1620static inline struct efx_channel *
1621efx_rx_queue_channel(struct efx_rx_queue *rx_queue)
1622{
1623 return container_of(rx_queue, struct efx_channel, rx_queue);
1624}
1625
1626static inline int efx_rx_queue_index(struct efx_rx_queue *rx_queue)
1627{
1628 return efx_rx_queue_channel(rx_queue)->channel;
1629}
1630
1631/* Returns a pointer to the specified receive buffer in the RX
1632 * descriptor queue.
1633 */
1634static inline struct efx_rx_buffer *efx_rx_buffer(struct efx_rx_queue *rx_queue,
1635 unsigned int index)
1636{
1637 return &rx_queue->buffer[index];
1638}
1639
1640static inline struct efx_rx_buffer *
1641efx_rx_buf_next(struct efx_rx_queue *rx_queue, struct efx_rx_buffer *rx_buf)
1642{
1643 if (unlikely(rx_buf == efx_rx_buffer(rx_queue, rx_queue->ptr_mask)))
1644 return efx_rx_buffer(rx_queue, 0);
1645 else
1646 return rx_buf + 1;
1647}
1648
1649/**
1650 * EFX_MAX_FRAME_LEN - calculate maximum frame length
1651 *
1652 * This calculates the maximum frame length that will be used for a
1653 * given MTU. The frame length will be equal to the MTU plus a
1654 * constant amount of header space and padding. This is the quantity
1655 * that the net driver will program into the MAC as the maximum frame
1656 * length.
1657 *
1658 * The 10G MAC requires 8-byte alignment on the frame
1659 * length, so we round up to the nearest 8.
1660 *
1661 * Re-clocking by the XGXS on RX can reduce an IPG to 32 bits (half an
1662 * XGMII cycle). If the frame length reaches the maximum value in the
1663 * same cycle, the XMAC can miss the IPG altogether. We work around
1664 * this by adding a further 16 bytes.
1665 */
1666#define EFX_FRAME_PAD 16
1667#define EFX_MAX_FRAME_LEN(mtu) \
1668 (ALIGN(((mtu) + ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN + EFX_FRAME_PAD), 8))
1669
1670static inline bool efx_xmit_with_hwtstamp(struct sk_buff *skb)
1671{
1672 return skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP;
1673}
1674static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
1675{
1676 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1677}
1678
1679/* Get the max fill level of the TX queues on this channel */
1680static inline unsigned int
1681efx_channel_tx_fill_level(struct efx_channel *channel)
1682{
1683 struct efx_tx_queue *tx_queue;
1684 unsigned int fill_level = 0;
1685
1686 /* This function is currently only used by EF100, which maybe
1687 * could do something simpler and just compute the fill level
1688 * of the single TXQ that's really in use.
1689 */
1690 efx_for_each_channel_tx_queue(tx_queue, channel)
1691 fill_level = max(fill_level,
1692 tx_queue->insert_count - tx_queue->read_count);
1693
1694 return fill_level;
1695}
1696
1697/* Get all supported features.
1698 * If a feature is not fixed, it is present in hw_features.
1699 * If a feature is fixed, it does not present in hw_features, but
1700 * always in features.
1701 */
1702static inline netdev_features_t efx_supported_features(const struct efx_nic *efx)
1703{
1704 const struct net_device *net_dev = efx->net_dev;
1705
1706 return net_dev->features | net_dev->hw_features;
1707}
1708
1709/* Get the current TX queue insert index. */
1710static inline unsigned int
1711efx_tx_queue_get_insert_index(const struct efx_tx_queue *tx_queue)
1712{
1713 return tx_queue->insert_count & tx_queue->ptr_mask;
1714}
1715
1716/* Get a TX buffer. */
1717static inline struct efx_tx_buffer *
1718__efx_tx_queue_get_insert_buffer(const struct efx_tx_queue *tx_queue)
1719{
1720 return &tx_queue->buffer[efx_tx_queue_get_insert_index(tx_queue)];
1721}
1722
1723/* Get a TX buffer, checking it's not currently in use. */
1724static inline struct efx_tx_buffer *
1725efx_tx_queue_get_insert_buffer(const struct efx_tx_queue *tx_queue)
1726{
1727 struct efx_tx_buffer *buffer =
1728 __efx_tx_queue_get_insert_buffer(tx_queue);
1729
1730 EFX_WARN_ON_ONCE_PARANOID(buffer->len);
1731 EFX_WARN_ON_ONCE_PARANOID(buffer->flags);
1732 EFX_WARN_ON_ONCE_PARANOID(buffer->unmap_len);
1733
1734 return buffer;
1735}
1736
1737#endif /* EFX_NET_DRIVER_H */