xref: /linux/drivers/net/ethernet/sfc/net_driver.h (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /****************************************************************************
2  * Driver for Solarflare Solarstorm network controllers and boards
3  * Copyright 2005-2006 Fen Systems Ltd.
4  * Copyright 2005-2011 Solarflare Communications Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation, incorporated herein by reference.
9  */
10 
11 /* Common definitions for all Efx net driver code */
12 
13 #ifndef EFX_NET_DRIVER_H
14 #define EFX_NET_DRIVER_H
15 
16 #include <linux/netdevice.h>
17 #include <linux/etherdevice.h>
18 #include <linux/ethtool.h>
19 #include <linux/if_vlan.h>
20 #include <linux/timer.h>
21 #include <linux/mdio.h>
22 #include <linux/list.h>
23 #include <linux/pci.h>
24 #include <linux/device.h>
25 #include <linux/highmem.h>
26 #include <linux/workqueue.h>
27 #include <linux/mutex.h>
28 #include <linux/vmalloc.h>
29 #include <linux/i2c.h>
30 
31 #include "enum.h"
32 #include "bitfield.h"
33 
34 /**************************************************************************
35  *
36  * Build definitions
37  *
38  **************************************************************************/
39 
40 #define EFX_DRIVER_VERSION	"3.2"
41 
42 #ifdef DEBUG
43 #define EFX_BUG_ON_PARANOID(x) BUG_ON(x)
44 #define EFX_WARN_ON_PARANOID(x) WARN_ON(x)
45 #else
46 #define EFX_BUG_ON_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_MAX_EXTRA_CHANNELS	2U
61 
62 /* Checksum generation is a per-queue option in hardware, so each
63  * queue visible to the networking core is backed by two hardware TX
64  * queues. */
65 #define EFX_MAX_TX_TC		2
66 #define EFX_MAX_CORE_TX_QUEUES	(EFX_MAX_TX_TC * EFX_MAX_CHANNELS)
67 #define EFX_TXQ_TYPE_OFFLOAD	1	/* flag */
68 #define EFX_TXQ_TYPE_HIGHPRI	2	/* flag */
69 #define EFX_TXQ_TYPES		4
70 #define EFX_MAX_TX_QUEUES	(EFX_TXQ_TYPES * EFX_MAX_CHANNELS)
71 
72 /* Forward declare Precision Time Protocol (PTP) support structure. */
73 struct efx_ptp_data;
74 
75 struct efx_self_tests;
76 
77 /**
78  * struct efx_special_buffer - An Efx special buffer
79  * @addr: CPU base address of the buffer
80  * @dma_addr: DMA base address of the buffer
81  * @len: Buffer length, in bytes
82  * @index: Buffer index within controller;s buffer table
83  * @entries: Number of buffer table entries
84  *
85  * Special buffers are used for the event queues and the TX and RX
86  * descriptor queues for each channel.  They are *not* used for the
87  * actual transmit and receive buffers.
88  */
89 struct efx_special_buffer {
90 	void *addr;
91 	dma_addr_t dma_addr;
92 	unsigned int len;
93 	unsigned int index;
94 	unsigned int entries;
95 };
96 
97 /**
98  * struct efx_tx_buffer - buffer state for a TX descriptor
99  * @skb: When @flags & %EFX_TX_BUF_SKB, the associated socket buffer to be
100  *	freed when descriptor completes
101  * @heap_buf: When @flags & %EFX_TX_BUF_HEAP, the associated heap buffer to be
102  *	freed when descriptor completes.
103  * @dma_addr: DMA address of the fragment.
104  * @flags: Flags for allocation and DMA mapping type
105  * @len: Length of this fragment.
106  *	This field is zero when the queue slot is empty.
107  * @unmap_len: Length of this fragment to unmap
108  */
109 struct efx_tx_buffer {
110 	union {
111 		const struct sk_buff *skb;
112 		void *heap_buf;
113 	};
114 	dma_addr_t dma_addr;
115 	unsigned short flags;
116 	unsigned short len;
117 	unsigned short unmap_len;
118 };
119 #define EFX_TX_BUF_CONT		1	/* not last descriptor of packet */
120 #define EFX_TX_BUF_SKB		2	/* buffer is last part of skb */
121 #define EFX_TX_BUF_HEAP		4	/* buffer was allocated with kmalloc() */
122 #define EFX_TX_BUF_MAP_SINGLE	8	/* buffer was mapped with dma_map_single() */
123 
124 /**
125  * struct efx_tx_queue - An Efx TX queue
126  *
127  * This is a ring buffer of TX fragments.
128  * Since the TX completion path always executes on the same
129  * CPU and the xmit path can operate on different CPUs,
130  * performance is increased by ensuring that the completion
131  * path and the xmit path operate on different cache lines.
132  * This is particularly important if the xmit path is always
133  * executing on one CPU which is different from the completion
134  * path.  There is also a cache line for members which are
135  * read but not written on the fast path.
136  *
137  * @efx: The associated Efx NIC
138  * @queue: DMA queue number
139  * @channel: The associated channel
140  * @core_txq: The networking core TX queue structure
141  * @buffer: The software buffer ring
142  * @tsoh_page: Array of pages of TSO header buffers
143  * @txd: The hardware descriptor ring
144  * @ptr_mask: The size of the ring minus 1.
145  * @initialised: Has hardware queue been initialised?
146  * @read_count: Current read pointer.
147  *	This is the number of buffers that have been removed from both rings.
148  * @old_write_count: The value of @write_count when last checked.
149  *	This is here for performance reasons.  The xmit path will
150  *	only get the up-to-date value of @write_count if this
151  *	variable indicates that the queue is empty.  This is to
152  *	avoid cache-line ping-pong between the xmit path and the
153  *	completion path.
154  * @insert_count: Current insert pointer
155  *	This is the number of buffers that have been added to the
156  *	software ring.
157  * @write_count: Current write pointer
158  *	This is the number of buffers that have been added to the
159  *	hardware ring.
160  * @old_read_count: The value of read_count when last checked.
161  *	This is here for performance reasons.  The xmit path will
162  *	only get the up-to-date value of read_count if this
163  *	variable indicates that the queue is full.  This is to
164  *	avoid cache-line ping-pong between the xmit path and the
165  *	completion path.
166  * @tso_bursts: Number of times TSO xmit invoked by kernel
167  * @tso_long_headers: Number of packets with headers too long for standard
168  *	blocks
169  * @tso_packets: Number of packets via the TSO xmit path
170  * @pushes: Number of times the TX push feature has been used
171  * @empty_read_count: If the completion path has seen the queue as empty
172  *	and the transmission path has not yet checked this, the value of
173  *	@read_count bitwise-added to %EFX_EMPTY_COUNT_VALID; otherwise 0.
174  */
175 struct efx_tx_queue {
176 	/* Members which don't change on the fast path */
177 	struct efx_nic *efx ____cacheline_aligned_in_smp;
178 	unsigned queue;
179 	struct efx_channel *channel;
180 	struct netdev_queue *core_txq;
181 	struct efx_tx_buffer *buffer;
182 	struct efx_buffer *tsoh_page;
183 	struct efx_special_buffer txd;
184 	unsigned int ptr_mask;
185 	bool initialised;
186 
187 	/* Members used mainly on the completion path */
188 	unsigned int read_count ____cacheline_aligned_in_smp;
189 	unsigned int old_write_count;
190 
191 	/* Members used only on the xmit path */
192 	unsigned int insert_count ____cacheline_aligned_in_smp;
193 	unsigned int write_count;
194 	unsigned int old_read_count;
195 	unsigned int tso_bursts;
196 	unsigned int tso_long_headers;
197 	unsigned int tso_packets;
198 	unsigned int pushes;
199 
200 	/* Members shared between paths and sometimes updated */
201 	unsigned int empty_read_count ____cacheline_aligned_in_smp;
202 #define EFX_EMPTY_COUNT_VALID 0x80000000
203 	atomic_t flush_outstanding;
204 };
205 
206 /**
207  * struct efx_rx_buffer - An Efx RX data buffer
208  * @dma_addr: DMA base address of the buffer
209  * @skb: The associated socket buffer. Valid iff !(@flags & %EFX_RX_BUF_PAGE).
210  *	Will be %NULL if the buffer slot is currently free.
211  * @page: The associated page buffer. Valif iff @flags & %EFX_RX_BUF_PAGE.
212  *	Will be %NULL if the buffer slot is currently free.
213  * @len: Buffer length, in bytes.
214  * @flags: Flags for buffer and packet state.
215  */
216 struct efx_rx_buffer {
217 	dma_addr_t dma_addr;
218 	union {
219 		struct sk_buff *skb;
220 		struct page *page;
221 	} u;
222 	unsigned int len;
223 	u16 flags;
224 };
225 #define EFX_RX_BUF_PAGE		0x0001
226 #define EFX_RX_PKT_CSUMMED	0x0002
227 #define EFX_RX_PKT_DISCARD	0x0004
228 
229 /**
230  * struct efx_rx_page_state - Page-based rx buffer state
231  *
232  * Inserted at the start of every page allocated for receive buffers.
233  * Used to facilitate sharing dma mappings between recycled rx buffers
234  * and those passed up to the kernel.
235  *
236  * @refcnt: Number of struct efx_rx_buffer's referencing this page.
237  *	When refcnt falls to zero, the page is unmapped for dma
238  * @dma_addr: The dma address of this page.
239  */
240 struct efx_rx_page_state {
241 	unsigned refcnt;
242 	dma_addr_t dma_addr;
243 
244 	unsigned int __pad[0] ____cacheline_aligned;
245 };
246 
247 /**
248  * struct efx_rx_queue - An Efx RX queue
249  * @efx: The associated Efx NIC
250  * @core_index:  Index of network core RX queue.  Will be >= 0 iff this
251  *	is associated with a real RX queue.
252  * @buffer: The software buffer ring
253  * @rxd: The hardware descriptor ring
254  * @ptr_mask: The size of the ring minus 1.
255  * @enabled: Receive queue enabled indicator.
256  * @flush_pending: Set when a RX flush is pending. Has the same lifetime as
257  *	@rxq_flush_pending.
258  * @added_count: Number of buffers added to the receive queue.
259  * @notified_count: Number of buffers given to NIC (<= @added_count).
260  * @removed_count: Number of buffers removed from the receive queue.
261  * @max_fill: RX descriptor maximum fill level (<= ring size)
262  * @fast_fill_trigger: RX descriptor fill level that will trigger a fast fill
263  *	(<= @max_fill)
264  * @min_fill: RX descriptor minimum non-zero fill level.
265  *	This records the minimum fill level observed when a ring
266  *	refill was triggered.
267  * @alloc_page_count: RX allocation strategy counter.
268  * @alloc_skb_count: RX allocation strategy counter.
269  * @slow_fill: Timer used to defer efx_nic_generate_fill_event().
270  */
271 struct efx_rx_queue {
272 	struct efx_nic *efx;
273 	int core_index;
274 	struct efx_rx_buffer *buffer;
275 	struct efx_special_buffer rxd;
276 	unsigned int ptr_mask;
277 	bool enabled;
278 	bool flush_pending;
279 
280 	int added_count;
281 	int notified_count;
282 	int removed_count;
283 	unsigned int max_fill;
284 	unsigned int fast_fill_trigger;
285 	unsigned int min_fill;
286 	unsigned int min_overfill;
287 	unsigned int alloc_page_count;
288 	unsigned int alloc_skb_count;
289 	struct timer_list slow_fill;
290 	unsigned int slow_fill_count;
291 };
292 
293 /**
294  * struct efx_buffer - An Efx general-purpose buffer
295  * @addr: host base address of the buffer
296  * @dma_addr: DMA base address of the buffer
297  * @len: Buffer length, in bytes
298  *
299  * The NIC uses these buffers for its interrupt status registers and
300  * MAC stats dumps.
301  */
302 struct efx_buffer {
303 	void *addr;
304 	dma_addr_t dma_addr;
305 	unsigned int len;
306 };
307 
308 
309 enum efx_rx_alloc_method {
310 	RX_ALLOC_METHOD_AUTO = 0,
311 	RX_ALLOC_METHOD_SKB = 1,
312 	RX_ALLOC_METHOD_PAGE = 2,
313 };
314 
315 /**
316  * struct efx_channel - An Efx channel
317  *
318  * A channel comprises an event queue, at least one TX queue, at least
319  * one RX queue, and an associated tasklet for processing the event
320  * queue.
321  *
322  * @efx: Associated Efx NIC
323  * @channel: Channel instance number
324  * @type: Channel type definition
325  * @enabled: Channel enabled indicator
326  * @irq: IRQ number (MSI and MSI-X only)
327  * @irq_moderation: IRQ moderation value (in hardware ticks)
328  * @napi_dev: Net device used with NAPI
329  * @napi_str: NAPI control structure
330  * @work_pending: Is work pending via NAPI?
331  * @eventq: Event queue buffer
332  * @eventq_mask: Event queue pointer mask
333  * @eventq_read_ptr: Event queue read pointer
334  * @event_test_cpu: Last CPU to handle interrupt or test event for this channel
335  * @irq_count: Number of IRQs since last adaptive moderation decision
336  * @irq_mod_score: IRQ moderation score
337  * @rx_alloc_level: Watermark based heuristic counter for pushing descriptors
338  *	and diagnostic counters
339  * @rx_alloc_push_pages: RX allocation method currently in use for pushing
340  *	descriptors
341  * @n_rx_tobe_disc: Count of RX_TOBE_DISC errors
342  * @n_rx_ip_hdr_chksum_err: Count of RX IP header checksum errors
343  * @n_rx_tcp_udp_chksum_err: Count of RX TCP and UDP checksum errors
344  * @n_rx_mcast_mismatch: Count of unmatched multicast frames
345  * @n_rx_frm_trunc: Count of RX_FRM_TRUNC errors
346  * @n_rx_overlength: Count of RX_OVERLENGTH errors
347  * @n_skbuff_leaks: Count of skbuffs leaked due to RX overrun
348  * @rx_queue: RX queue for this channel
349  * @tx_queue: TX queues for this channel
350  */
351 struct efx_channel {
352 	struct efx_nic *efx;
353 	int channel;
354 	const struct efx_channel_type *type;
355 	bool enabled;
356 	int irq;
357 	unsigned int irq_moderation;
358 	struct net_device *napi_dev;
359 	struct napi_struct napi_str;
360 	bool work_pending;
361 	struct efx_special_buffer eventq;
362 	unsigned int eventq_mask;
363 	unsigned int eventq_read_ptr;
364 	int event_test_cpu;
365 
366 	unsigned int irq_count;
367 	unsigned int irq_mod_score;
368 #ifdef CONFIG_RFS_ACCEL
369 	unsigned int rfs_filters_added;
370 #endif
371 
372 	int rx_alloc_level;
373 	int rx_alloc_push_pages;
374 
375 	unsigned n_rx_tobe_disc;
376 	unsigned n_rx_ip_hdr_chksum_err;
377 	unsigned n_rx_tcp_udp_chksum_err;
378 	unsigned n_rx_mcast_mismatch;
379 	unsigned n_rx_frm_trunc;
380 	unsigned n_rx_overlength;
381 	unsigned n_skbuff_leaks;
382 
383 	/* Used to pipeline received packets in order to optimise memory
384 	 * access with prefetches.
385 	 */
386 	struct efx_rx_buffer *rx_pkt;
387 
388 	struct efx_rx_queue rx_queue;
389 	struct efx_tx_queue tx_queue[EFX_TXQ_TYPES];
390 };
391 
392 /**
393  * struct efx_channel_type - distinguishes traffic and extra channels
394  * @handle_no_channel: Handle failure to allocate an extra channel
395  * @pre_probe: Set up extra state prior to initialisation
396  * @post_remove: Tear down extra state after finalisation, if allocated.
397  *	May be called on channels that have not been probed.
398  * @get_name: Generate the channel's name (used for its IRQ handler)
399  * @copy: Copy the channel state prior to reallocation.  May be %NULL if
400  *	reallocation is not supported.
401  * @receive_skb: Handle an skb ready to be passed to netif_receive_skb()
402  * @keep_eventq: Flag for whether event queue should be kept initialised
403  *	while the device is stopped
404  */
405 struct efx_channel_type {
406 	void (*handle_no_channel)(struct efx_nic *);
407 	int (*pre_probe)(struct efx_channel *);
408 	void (*post_remove)(struct efx_channel *);
409 	void (*get_name)(struct efx_channel *, char *buf, size_t len);
410 	struct efx_channel *(*copy)(const struct efx_channel *);
411 	void (*receive_skb)(struct efx_channel *, struct sk_buff *);
412 	bool keep_eventq;
413 };
414 
415 enum efx_led_mode {
416 	EFX_LED_OFF	= 0,
417 	EFX_LED_ON	= 1,
418 	EFX_LED_DEFAULT	= 2
419 };
420 
421 #define STRING_TABLE_LOOKUP(val, member) \
422 	((val) < member ## _max) ? member ## _names[val] : "(invalid)"
423 
424 extern const char *const efx_loopback_mode_names[];
425 extern const unsigned int efx_loopback_mode_max;
426 #define LOOPBACK_MODE(efx) \
427 	STRING_TABLE_LOOKUP((efx)->loopback_mode, efx_loopback_mode)
428 
429 extern const char *const efx_reset_type_names[];
430 extern const unsigned int efx_reset_type_max;
431 #define RESET_TYPE(type) \
432 	STRING_TABLE_LOOKUP(type, efx_reset_type)
433 
434 enum efx_int_mode {
435 	/* Be careful if altering to correct macro below */
436 	EFX_INT_MODE_MSIX = 0,
437 	EFX_INT_MODE_MSI = 1,
438 	EFX_INT_MODE_LEGACY = 2,
439 	EFX_INT_MODE_MAX	/* Insert any new items before this */
440 };
441 #define EFX_INT_MODE_USE_MSI(x) (((x)->interrupt_mode) <= EFX_INT_MODE_MSI)
442 
443 enum nic_state {
444 	STATE_UNINIT = 0,	/* device being probed/removed or is frozen */
445 	STATE_READY = 1,	/* hardware ready and netdev registered */
446 	STATE_DISABLED = 2,	/* device disabled due to hardware errors */
447 };
448 
449 /*
450  * Alignment of page-allocated RX buffers
451  *
452  * Controls the number of bytes inserted at the start of an RX buffer.
453  * This is the equivalent of NET_IP_ALIGN [which controls the alignment
454  * of the skb->head for hardware DMA].
455  */
456 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
457 #define EFX_PAGE_IP_ALIGN 0
458 #else
459 #define EFX_PAGE_IP_ALIGN NET_IP_ALIGN
460 #endif
461 
462 /*
463  * Alignment of the skb->head which wraps a page-allocated RX buffer
464  *
465  * The skb allocated to wrap an rx_buffer can have this alignment. Since
466  * the data is memcpy'd from the rx_buf, it does not need to be equal to
467  * EFX_PAGE_IP_ALIGN.
468  */
469 #define EFX_PAGE_SKB_ALIGN 2
470 
471 /* Forward declaration */
472 struct efx_nic;
473 
474 /* Pseudo bit-mask flow control field */
475 #define EFX_FC_RX	FLOW_CTRL_RX
476 #define EFX_FC_TX	FLOW_CTRL_TX
477 #define EFX_FC_AUTO	4
478 
479 /**
480  * struct efx_link_state - Current state of the link
481  * @up: Link is up
482  * @fd: Link is full-duplex
483  * @fc: Actual flow control flags
484  * @speed: Link speed (Mbps)
485  */
486 struct efx_link_state {
487 	bool up;
488 	bool fd;
489 	u8 fc;
490 	unsigned int speed;
491 };
492 
493 static inline bool efx_link_state_equal(const struct efx_link_state *left,
494 					const struct efx_link_state *right)
495 {
496 	return left->up == right->up && left->fd == right->fd &&
497 		left->fc == right->fc && left->speed == right->speed;
498 }
499 
500 /**
501  * struct efx_phy_operations - Efx PHY operations table
502  * @probe: Probe PHY and initialise efx->mdio.mode_support, efx->mdio.mmds,
503  *	efx->loopback_modes.
504  * @init: Initialise PHY
505  * @fini: Shut down PHY
506  * @reconfigure: Reconfigure PHY (e.g. for new link parameters)
507  * @poll: Update @link_state and report whether it changed.
508  *	Serialised by the mac_lock.
509  * @get_settings: Get ethtool settings. Serialised by the mac_lock.
510  * @set_settings: Set ethtool settings. Serialised by the mac_lock.
511  * @set_npage_adv: Set abilities advertised in (Extended) Next Page
512  *	(only needed where AN bit is set in mmds)
513  * @test_alive: Test that PHY is 'alive' (online)
514  * @test_name: Get the name of a PHY-specific test/result
515  * @run_tests: Run tests and record results as appropriate (offline).
516  *	Flags are the ethtool tests flags.
517  */
518 struct efx_phy_operations {
519 	int (*probe) (struct efx_nic *efx);
520 	int (*init) (struct efx_nic *efx);
521 	void (*fini) (struct efx_nic *efx);
522 	void (*remove) (struct efx_nic *efx);
523 	int (*reconfigure) (struct efx_nic *efx);
524 	bool (*poll) (struct efx_nic *efx);
525 	void (*get_settings) (struct efx_nic *efx,
526 			      struct ethtool_cmd *ecmd);
527 	int (*set_settings) (struct efx_nic *efx,
528 			     struct ethtool_cmd *ecmd);
529 	void (*set_npage_adv) (struct efx_nic *efx, u32);
530 	int (*test_alive) (struct efx_nic *efx);
531 	const char *(*test_name) (struct efx_nic *efx, unsigned int index);
532 	int (*run_tests) (struct efx_nic *efx, int *results, unsigned flags);
533 	int (*get_module_eeprom) (struct efx_nic *efx,
534 			       struct ethtool_eeprom *ee,
535 			       u8 *data);
536 	int (*get_module_info) (struct efx_nic *efx,
537 				struct ethtool_modinfo *modinfo);
538 };
539 
540 /**
541  * enum efx_phy_mode - PHY operating mode flags
542  * @PHY_MODE_NORMAL: on and should pass traffic
543  * @PHY_MODE_TX_DISABLED: on with TX disabled
544  * @PHY_MODE_LOW_POWER: set to low power through MDIO
545  * @PHY_MODE_OFF: switched off through external control
546  * @PHY_MODE_SPECIAL: on but will not pass traffic
547  */
548 enum efx_phy_mode {
549 	PHY_MODE_NORMAL		= 0,
550 	PHY_MODE_TX_DISABLED	= 1,
551 	PHY_MODE_LOW_POWER	= 2,
552 	PHY_MODE_OFF		= 4,
553 	PHY_MODE_SPECIAL	= 8,
554 };
555 
556 static inline bool efx_phy_mode_disabled(enum efx_phy_mode mode)
557 {
558 	return !!(mode & ~PHY_MODE_TX_DISABLED);
559 }
560 
561 /*
562  * Efx extended statistics
563  *
564  * Not all statistics are provided by all supported MACs.  The purpose
565  * is this structure is to contain the raw statistics provided by each
566  * MAC.
567  */
568 struct efx_mac_stats {
569 	u64 tx_bytes;
570 	u64 tx_good_bytes;
571 	u64 tx_bad_bytes;
572 	u64 tx_packets;
573 	u64 tx_bad;
574 	u64 tx_pause;
575 	u64 tx_control;
576 	u64 tx_unicast;
577 	u64 tx_multicast;
578 	u64 tx_broadcast;
579 	u64 tx_lt64;
580 	u64 tx_64;
581 	u64 tx_65_to_127;
582 	u64 tx_128_to_255;
583 	u64 tx_256_to_511;
584 	u64 tx_512_to_1023;
585 	u64 tx_1024_to_15xx;
586 	u64 tx_15xx_to_jumbo;
587 	u64 tx_gtjumbo;
588 	u64 tx_collision;
589 	u64 tx_single_collision;
590 	u64 tx_multiple_collision;
591 	u64 tx_excessive_collision;
592 	u64 tx_deferred;
593 	u64 tx_late_collision;
594 	u64 tx_excessive_deferred;
595 	u64 tx_non_tcpudp;
596 	u64 tx_mac_src_error;
597 	u64 tx_ip_src_error;
598 	u64 rx_bytes;
599 	u64 rx_good_bytes;
600 	u64 rx_bad_bytes;
601 	u64 rx_packets;
602 	u64 rx_good;
603 	u64 rx_bad;
604 	u64 rx_pause;
605 	u64 rx_control;
606 	u64 rx_unicast;
607 	u64 rx_multicast;
608 	u64 rx_broadcast;
609 	u64 rx_lt64;
610 	u64 rx_64;
611 	u64 rx_65_to_127;
612 	u64 rx_128_to_255;
613 	u64 rx_256_to_511;
614 	u64 rx_512_to_1023;
615 	u64 rx_1024_to_15xx;
616 	u64 rx_15xx_to_jumbo;
617 	u64 rx_gtjumbo;
618 	u64 rx_bad_lt64;
619 	u64 rx_bad_64_to_15xx;
620 	u64 rx_bad_15xx_to_jumbo;
621 	u64 rx_bad_gtjumbo;
622 	u64 rx_overflow;
623 	u64 rx_missed;
624 	u64 rx_false_carrier;
625 	u64 rx_symbol_error;
626 	u64 rx_align_error;
627 	u64 rx_length_error;
628 	u64 rx_internal_error;
629 	u64 rx_good_lt64;
630 };
631 
632 /* Number of bits used in a multicast filter hash address */
633 #define EFX_MCAST_HASH_BITS 8
634 
635 /* Number of (single-bit) entries in a multicast filter hash */
636 #define EFX_MCAST_HASH_ENTRIES (1 << EFX_MCAST_HASH_BITS)
637 
638 /* An Efx multicast filter hash */
639 union efx_multicast_hash {
640 	u8 byte[EFX_MCAST_HASH_ENTRIES / 8];
641 	efx_oword_t oword[EFX_MCAST_HASH_ENTRIES / sizeof(efx_oword_t) / 8];
642 };
643 
644 struct efx_filter_state;
645 struct efx_vf;
646 struct vfdi_status;
647 
648 /**
649  * struct efx_nic - an Efx NIC
650  * @name: Device name (net device name or bus id before net device registered)
651  * @pci_dev: The PCI device
652  * @type: Controller type attributes
653  * @legacy_irq: IRQ number
654  * @legacy_irq_enabled: Are IRQs enabled on NIC (INT_EN_KER register)?
655  * @workqueue: Workqueue for port reconfigures and the HW monitor.
656  *	Work items do not hold and must not acquire RTNL.
657  * @workqueue_name: Name of workqueue
658  * @reset_work: Scheduled reset workitem
659  * @membase_phys: Memory BAR value as physical address
660  * @membase: Memory BAR value
661  * @interrupt_mode: Interrupt mode
662  * @timer_quantum_ns: Interrupt timer quantum, in nanoseconds
663  * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues
664  * @irq_rx_moderation: IRQ moderation time for RX event queues
665  * @msg_enable: Log message enable flags
666  * @state: Device state number (%STATE_*). Serialised by the rtnl_lock.
667  * @reset_pending: Bitmask for pending resets
668  * @tx_queue: TX DMA queues
669  * @rx_queue: RX DMA queues
670  * @channel: Channels
671  * @channel_name: Names for channels and their IRQs
672  * @extra_channel_types: Types of extra (non-traffic) channels that
673  *	should be allocated for this NIC
674  * @rxq_entries: Size of receive queues requested by user.
675  * @txq_entries: Size of transmit queues requested by user.
676  * @txq_stop_thresh: TX queue fill level at or above which we stop it.
677  * @txq_wake_thresh: TX queue fill level at or below which we wake it.
678  * @tx_dc_base: Base qword address in SRAM of TX queue descriptor caches
679  * @rx_dc_base: Base qword address in SRAM of RX queue descriptor caches
680  * @sram_lim_qw: Qword address limit of SRAM
681  * @next_buffer_table: First available buffer table id
682  * @n_channels: Number of channels in use
683  * @n_rx_channels: Number of channels used for RX (= number of RX queues)
684  * @n_tx_channels: Number of channels used for TX
685  * @rx_buffer_len: RX buffer length
686  * @rx_buffer_order: Order (log2) of number of pages for each RX buffer
687  * @rx_hash_key: Toeplitz hash key for RSS
688  * @rx_indir_table: Indirection table for RSS
689  * @int_error_count: Number of internal errors seen recently
690  * @int_error_expire: Time at which error count will be expired
691  * @irq_status: Interrupt status buffer
692  * @irq_zero_count: Number of legacy IRQs seen with queue flags == 0
693  * @irq_level: IRQ level/index for IRQs not triggered by an event queue
694  * @selftest_work: Work item for asynchronous self-test
695  * @mtd_list: List of MTDs attached to the NIC
696  * @nic_data: Hardware dependent state
697  * @mac_lock: MAC access lock. Protects @port_enabled, @phy_mode,
698  *	efx_monitor() and efx_reconfigure_port()
699  * @port_enabled: Port enabled indicator.
700  *	Serialises efx_stop_all(), efx_start_all(), efx_monitor() and
701  *	efx_mac_work() with kernel interfaces. Safe to read under any
702  *	one of the rtnl_lock, mac_lock, or netif_tx_lock, but all three must
703  *	be held to modify it.
704  * @port_initialized: Port initialized?
705  * @net_dev: Operating system network device. Consider holding the rtnl lock
706  * @stats_buffer: DMA buffer for statistics
707  * @phy_type: PHY type
708  * @phy_op: PHY interface
709  * @phy_data: PHY private data (including PHY-specific stats)
710  * @mdio: PHY MDIO interface
711  * @mdio_bus: PHY MDIO bus ID (only used by Siena)
712  * @phy_mode: PHY operating mode. Serialised by @mac_lock.
713  * @link_advertising: Autonegotiation advertising flags
714  * @link_state: Current state of the link
715  * @n_link_state_changes: Number of times the link has changed state
716  * @promiscuous: Promiscuous flag. Protected by netif_tx_lock.
717  * @multicast_hash: Multicast hash table
718  * @wanted_fc: Wanted flow control flags
719  * @fc_disable: When non-zero flow control is disabled. Typically used to
720  *	ensure that network back pressure doesn't delay dma queue flushes.
721  *	Serialised by the rtnl lock.
722  * @mac_work: Work item for changing MAC promiscuity and multicast hash
723  * @loopback_mode: Loopback status
724  * @loopback_modes: Supported loopback mode bitmask
725  * @loopback_selftest: Offline self-test private state
726  * @drain_pending: Count of RX and TX queues that haven't been flushed and drained.
727  * @rxq_flush_pending: Count of number of receive queues that need to be flushed.
728  *	Decremented when the efx_flush_rx_queue() is called.
729  * @rxq_flush_outstanding: Count of number of RX flushes started but not yet
730  *	completed (either success or failure). Not used when MCDI is used to
731  *	flush receive queues.
732  * @flush_wq: wait queue used by efx_nic_flush_queues() to wait for flush completions.
733  * @vf: Array of &struct efx_vf objects.
734  * @vf_count: Number of VFs intended to be enabled.
735  * @vf_init_count: Number of VFs that have been fully initialised.
736  * @vi_scale: log2 number of vnics per VF.
737  * @vf_buftbl_base: The zeroth buffer table index used to back VF queues.
738  * @vfdi_status: Common VFDI status page to be dmad to VF address space.
739  * @local_addr_list: List of local addresses. Protected by %local_lock.
740  * @local_page_list: List of DMA addressable pages used to broadcast
741  *	%local_addr_list. Protected by %local_lock.
742  * @local_lock: Mutex protecting %local_addr_list and %local_page_list.
743  * @peer_work: Work item to broadcast peer addresses to VMs.
744  * @ptp_data: PTP state data
745  * @monitor_work: Hardware monitor workitem
746  * @biu_lock: BIU (bus interface unit) lock
747  * @last_irq_cpu: Last CPU to handle a possible test interrupt.  This
748  *	field is used by efx_test_interrupts() to verify that an
749  *	interrupt has occurred.
750  * @n_rx_nodesc_drop_cnt: RX no descriptor drop count
751  * @mac_stats: MAC statistics. These include all statistics the MACs
752  *	can provide.  Generic code converts these into a standard
753  *	&struct net_device_stats.
754  * @stats_lock: Statistics update lock. Serialises statistics fetches
755  *	and access to @mac_stats.
756  *
757  * This is stored in the private area of the &struct net_device.
758  */
759 struct efx_nic {
760 	/* The following fields should be written very rarely */
761 
762 	char name[IFNAMSIZ];
763 	struct pci_dev *pci_dev;
764 	const struct efx_nic_type *type;
765 	int legacy_irq;
766 	bool legacy_irq_enabled;
767 	struct workqueue_struct *workqueue;
768 	char workqueue_name[16];
769 	struct work_struct reset_work;
770 	resource_size_t membase_phys;
771 	void __iomem *membase;
772 
773 	enum efx_int_mode interrupt_mode;
774 	unsigned int timer_quantum_ns;
775 	bool irq_rx_adaptive;
776 	unsigned int irq_rx_moderation;
777 	u32 msg_enable;
778 
779 	enum nic_state state;
780 	unsigned long reset_pending;
781 
782 	struct efx_channel *channel[EFX_MAX_CHANNELS];
783 	char channel_name[EFX_MAX_CHANNELS][IFNAMSIZ + 6];
784 	const struct efx_channel_type *
785 	extra_channel_type[EFX_MAX_EXTRA_CHANNELS];
786 
787 	unsigned rxq_entries;
788 	unsigned txq_entries;
789 	unsigned int txq_stop_thresh;
790 	unsigned int txq_wake_thresh;
791 
792 	unsigned tx_dc_base;
793 	unsigned rx_dc_base;
794 	unsigned sram_lim_qw;
795 	unsigned next_buffer_table;
796 	unsigned n_channels;
797 	unsigned n_rx_channels;
798 	unsigned rss_spread;
799 	unsigned tx_channel_offset;
800 	unsigned n_tx_channels;
801 	unsigned int rx_buffer_len;
802 	unsigned int rx_buffer_order;
803 	u8 rx_hash_key[40];
804 	u32 rx_indir_table[128];
805 
806 	unsigned int_error_count;
807 	unsigned long int_error_expire;
808 
809 	struct efx_buffer irq_status;
810 	unsigned irq_zero_count;
811 	unsigned irq_level;
812 	struct delayed_work selftest_work;
813 
814 #ifdef CONFIG_SFC_MTD
815 	struct list_head mtd_list;
816 #endif
817 
818 	void *nic_data;
819 
820 	struct mutex mac_lock;
821 	struct work_struct mac_work;
822 	bool port_enabled;
823 
824 	bool port_initialized;
825 	struct net_device *net_dev;
826 
827 	struct efx_buffer stats_buffer;
828 
829 	unsigned int phy_type;
830 	const struct efx_phy_operations *phy_op;
831 	void *phy_data;
832 	struct mdio_if_info mdio;
833 	unsigned int mdio_bus;
834 	enum efx_phy_mode phy_mode;
835 
836 	u32 link_advertising;
837 	struct efx_link_state link_state;
838 	unsigned int n_link_state_changes;
839 
840 	bool promiscuous;
841 	union efx_multicast_hash multicast_hash;
842 	u8 wanted_fc;
843 	unsigned fc_disable;
844 
845 	atomic_t rx_reset;
846 	enum efx_loopback_mode loopback_mode;
847 	u64 loopback_modes;
848 
849 	void *loopback_selftest;
850 
851 	struct efx_filter_state *filter_state;
852 
853 	atomic_t drain_pending;
854 	atomic_t rxq_flush_pending;
855 	atomic_t rxq_flush_outstanding;
856 	wait_queue_head_t flush_wq;
857 
858 #ifdef CONFIG_SFC_SRIOV
859 	struct efx_channel *vfdi_channel;
860 	struct efx_vf *vf;
861 	unsigned vf_count;
862 	unsigned vf_init_count;
863 	unsigned vi_scale;
864 	unsigned vf_buftbl_base;
865 	struct efx_buffer vfdi_status;
866 	struct list_head local_addr_list;
867 	struct list_head local_page_list;
868 	struct mutex local_lock;
869 	struct work_struct peer_work;
870 #endif
871 
872 	struct efx_ptp_data *ptp_data;
873 
874 	/* The following fields may be written more often */
875 
876 	struct delayed_work monitor_work ____cacheline_aligned_in_smp;
877 	spinlock_t biu_lock;
878 	int last_irq_cpu;
879 	unsigned n_rx_nodesc_drop_cnt;
880 	struct efx_mac_stats mac_stats;
881 	spinlock_t stats_lock;
882 };
883 
884 static inline int efx_dev_registered(struct efx_nic *efx)
885 {
886 	return efx->net_dev->reg_state == NETREG_REGISTERED;
887 }
888 
889 static inline unsigned int efx_port_num(struct efx_nic *efx)
890 {
891 	return efx->net_dev->dev_id;
892 }
893 
894 /**
895  * struct efx_nic_type - Efx device type definition
896  * @probe: Probe the controller
897  * @remove: Free resources allocated by probe()
898  * @init: Initialise the controller
899  * @dimension_resources: Dimension controller resources (buffer table,
900  *	and VIs once the available interrupt resources are clear)
901  * @fini: Shut down the controller
902  * @monitor: Periodic function for polling link state and hardware monitor
903  * @map_reset_reason: Map ethtool reset reason to a reset method
904  * @map_reset_flags: Map ethtool reset flags to a reset method, if possible
905  * @reset: Reset the controller hardware and possibly the PHY.  This will
906  *	be called while the controller is uninitialised.
907  * @probe_port: Probe the MAC and PHY
908  * @remove_port: Free resources allocated by probe_port()
909  * @handle_global_event: Handle a "global" event (may be %NULL)
910  * @prepare_flush: Prepare the hardware for flushing the DMA queues
911  * @finish_flush: Clean up after flushing the DMA queues
912  * @update_stats: Update statistics not provided by event handling
913  * @start_stats: Start the regular fetching of statistics
914  * @stop_stats: Stop the regular fetching of statistics
915  * @set_id_led: Set state of identifying LED or revert to automatic function
916  * @push_irq_moderation: Apply interrupt moderation value
917  * @reconfigure_port: Push loopback/power/txdis changes to the MAC and PHY
918  * @reconfigure_mac: Push MAC address, MTU, flow control and filter settings
919  *	to the hardware.  Serialised by the mac_lock.
920  * @check_mac_fault: Check MAC fault state. True if fault present.
921  * @get_wol: Get WoL configuration from driver state
922  * @set_wol: Push WoL configuration to the NIC
923  * @resume_wol: Synchronise WoL state between driver and MC (e.g. after resume)
924  * @test_chip: Test registers.  Should use efx_nic_test_registers(), and is
925  *	expected to reset the NIC.
926  * @test_nvram: Test validity of NVRAM contents
927  * @revision: Hardware architecture revision
928  * @mem_map_size: Memory BAR mapped size
929  * @txd_ptr_tbl_base: TX descriptor ring base address
930  * @rxd_ptr_tbl_base: RX descriptor ring base address
931  * @buf_tbl_base: Buffer table base address
932  * @evq_ptr_tbl_base: Event queue pointer table base address
933  * @evq_rptr_tbl_base: Event queue read-pointer table base address
934  * @max_dma_mask: Maximum possible DMA mask
935  * @rx_buffer_hash_size: Size of hash at start of RX buffer
936  * @rx_buffer_padding: Size of padding at end of RX buffer
937  * @max_interrupt_mode: Highest capability interrupt mode supported
938  *	from &enum efx_init_mode.
939  * @phys_addr_channels: Number of channels with physically addressed
940  *	descriptors
941  * @timer_period_max: Maximum period of interrupt timer (in ticks)
942  * @offload_features: net_device feature flags for protocol offload
943  *	features implemented in hardware
944  */
945 struct efx_nic_type {
946 	int (*probe)(struct efx_nic *efx);
947 	void (*remove)(struct efx_nic *efx);
948 	int (*init)(struct efx_nic *efx);
949 	void (*dimension_resources)(struct efx_nic *efx);
950 	void (*fini)(struct efx_nic *efx);
951 	void (*monitor)(struct efx_nic *efx);
952 	enum reset_type (*map_reset_reason)(enum reset_type reason);
953 	int (*map_reset_flags)(u32 *flags);
954 	int (*reset)(struct efx_nic *efx, enum reset_type method);
955 	int (*probe_port)(struct efx_nic *efx);
956 	void (*remove_port)(struct efx_nic *efx);
957 	bool (*handle_global_event)(struct efx_channel *channel, efx_qword_t *);
958 	void (*prepare_flush)(struct efx_nic *efx);
959 	void (*finish_flush)(struct efx_nic *efx);
960 	void (*update_stats)(struct efx_nic *efx);
961 	void (*start_stats)(struct efx_nic *efx);
962 	void (*stop_stats)(struct efx_nic *efx);
963 	void (*set_id_led)(struct efx_nic *efx, enum efx_led_mode mode);
964 	void (*push_irq_moderation)(struct efx_channel *channel);
965 	int (*reconfigure_port)(struct efx_nic *efx);
966 	int (*reconfigure_mac)(struct efx_nic *efx);
967 	bool (*check_mac_fault)(struct efx_nic *efx);
968 	void (*get_wol)(struct efx_nic *efx, struct ethtool_wolinfo *wol);
969 	int (*set_wol)(struct efx_nic *efx, u32 type);
970 	void (*resume_wol)(struct efx_nic *efx);
971 	int (*test_chip)(struct efx_nic *efx, struct efx_self_tests *tests);
972 	int (*test_nvram)(struct efx_nic *efx);
973 
974 	int revision;
975 	unsigned int mem_map_size;
976 	unsigned int txd_ptr_tbl_base;
977 	unsigned int rxd_ptr_tbl_base;
978 	unsigned int buf_tbl_base;
979 	unsigned int evq_ptr_tbl_base;
980 	unsigned int evq_rptr_tbl_base;
981 	u64 max_dma_mask;
982 	unsigned int rx_buffer_hash_size;
983 	unsigned int rx_buffer_padding;
984 	unsigned int max_interrupt_mode;
985 	unsigned int phys_addr_channels;
986 	unsigned int timer_period_max;
987 	netdev_features_t offload_features;
988 };
989 
990 /**************************************************************************
991  *
992  * Prototypes and inline functions
993  *
994  *************************************************************************/
995 
996 static inline struct efx_channel *
997 efx_get_channel(struct efx_nic *efx, unsigned index)
998 {
999 	EFX_BUG_ON_PARANOID(index >= efx->n_channels);
1000 	return efx->channel[index];
1001 }
1002 
1003 /* Iterate over all used channels */
1004 #define efx_for_each_channel(_channel, _efx)				\
1005 	for (_channel = (_efx)->channel[0];				\
1006 	     _channel;							\
1007 	     _channel = (_channel->channel + 1 < (_efx)->n_channels) ?	\
1008 		     (_efx)->channel[_channel->channel + 1] : NULL)
1009 
1010 /* Iterate over all used channels in reverse */
1011 #define efx_for_each_channel_rev(_channel, _efx)			\
1012 	for (_channel = (_efx)->channel[(_efx)->n_channels - 1];	\
1013 	     _channel;							\
1014 	     _channel = _channel->channel ?				\
1015 		     (_efx)->channel[_channel->channel - 1] : NULL)
1016 
1017 static inline struct efx_tx_queue *
1018 efx_get_tx_queue(struct efx_nic *efx, unsigned index, unsigned type)
1019 {
1020 	EFX_BUG_ON_PARANOID(index >= efx->n_tx_channels ||
1021 			    type >= EFX_TXQ_TYPES);
1022 	return &efx->channel[efx->tx_channel_offset + index]->tx_queue[type];
1023 }
1024 
1025 static inline bool efx_channel_has_tx_queues(struct efx_channel *channel)
1026 {
1027 	return channel->channel - channel->efx->tx_channel_offset <
1028 		channel->efx->n_tx_channels;
1029 }
1030 
1031 static inline struct efx_tx_queue *
1032 efx_channel_get_tx_queue(struct efx_channel *channel, unsigned type)
1033 {
1034 	EFX_BUG_ON_PARANOID(!efx_channel_has_tx_queues(channel) ||
1035 			    type >= EFX_TXQ_TYPES);
1036 	return &channel->tx_queue[type];
1037 }
1038 
1039 static inline bool efx_tx_queue_used(struct efx_tx_queue *tx_queue)
1040 {
1041 	return !(tx_queue->efx->net_dev->num_tc < 2 &&
1042 		 tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI);
1043 }
1044 
1045 /* Iterate over all TX queues belonging to a channel */
1046 #define efx_for_each_channel_tx_queue(_tx_queue, _channel)		\
1047 	if (!efx_channel_has_tx_queues(_channel))			\
1048 		;							\
1049 	else								\
1050 		for (_tx_queue = (_channel)->tx_queue;			\
1051 		     _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES && \
1052 			     efx_tx_queue_used(_tx_queue);		\
1053 		     _tx_queue++)
1054 
1055 /* Iterate over all possible TX queues belonging to a channel */
1056 #define efx_for_each_possible_channel_tx_queue(_tx_queue, _channel)	\
1057 	if (!efx_channel_has_tx_queues(_channel))			\
1058 		;							\
1059 	else								\
1060 		for (_tx_queue = (_channel)->tx_queue;			\
1061 		     _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES;	\
1062 		     _tx_queue++)
1063 
1064 static inline bool efx_channel_has_rx_queue(struct efx_channel *channel)
1065 {
1066 	return channel->rx_queue.core_index >= 0;
1067 }
1068 
1069 static inline struct efx_rx_queue *
1070 efx_channel_get_rx_queue(struct efx_channel *channel)
1071 {
1072 	EFX_BUG_ON_PARANOID(!efx_channel_has_rx_queue(channel));
1073 	return &channel->rx_queue;
1074 }
1075 
1076 /* Iterate over all RX queues belonging to a channel */
1077 #define efx_for_each_channel_rx_queue(_rx_queue, _channel)		\
1078 	if (!efx_channel_has_rx_queue(_channel))			\
1079 		;							\
1080 	else								\
1081 		for (_rx_queue = &(_channel)->rx_queue;			\
1082 		     _rx_queue;						\
1083 		     _rx_queue = NULL)
1084 
1085 static inline struct efx_channel *
1086 efx_rx_queue_channel(struct efx_rx_queue *rx_queue)
1087 {
1088 	return container_of(rx_queue, struct efx_channel, rx_queue);
1089 }
1090 
1091 static inline int efx_rx_queue_index(struct efx_rx_queue *rx_queue)
1092 {
1093 	return efx_rx_queue_channel(rx_queue)->channel;
1094 }
1095 
1096 /* Returns a pointer to the specified receive buffer in the RX
1097  * descriptor queue.
1098  */
1099 static inline struct efx_rx_buffer *efx_rx_buffer(struct efx_rx_queue *rx_queue,
1100 						  unsigned int index)
1101 {
1102 	return &rx_queue->buffer[index];
1103 }
1104 
1105 
1106 /**
1107  * EFX_MAX_FRAME_LEN - calculate maximum frame length
1108  *
1109  * This calculates the maximum frame length that will be used for a
1110  * given MTU.  The frame length will be equal to the MTU plus a
1111  * constant amount of header space and padding.  This is the quantity
1112  * that the net driver will program into the MAC as the maximum frame
1113  * length.
1114  *
1115  * The 10G MAC requires 8-byte alignment on the frame
1116  * length, so we round up to the nearest 8.
1117  *
1118  * Re-clocking by the XGXS on RX can reduce an IPG to 32 bits (half an
1119  * XGMII cycle).  If the frame length reaches the maximum value in the
1120  * same cycle, the XMAC can miss the IPG altogether.  We work around
1121  * this by adding a further 16 bytes.
1122  */
1123 #define EFX_MAX_FRAME_LEN(mtu) \
1124 	((((mtu) + ETH_HLEN + VLAN_HLEN + 4/* FCS */ + 7) & ~7) + 16)
1125 
1126 static inline bool efx_xmit_with_hwtstamp(struct sk_buff *skb)
1127 {
1128 	return skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP;
1129 }
1130 static inline void efx_xmit_hwtstamp_pending(struct sk_buff *skb)
1131 {
1132 	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1133 }
1134 
1135 #endif /* EFX_NET_DRIVER_H */
1136