Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/* Copyright(c) 2013 - 2018 Intel Corporation. */
3
4#ifndef _I40E_TXRX_H_
5#define _I40E_TXRX_H_
6
7#include <net/xdp.h>
8#include "i40e_type.h"
9
10/* Interrupt Throttling and Rate Limiting Goodies */
11#define I40E_DEFAULT_IRQ_WORK 256
12
13/* The datasheet for the X710 and XL710 indicate that the maximum value for
14 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
15 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
16 * the register value which is divided by 2 lets use the actual values and
17 * avoid an excessive amount of translation.
18 */
19#define I40E_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
20#define I40E_ITR_MASK 0x1FFE /* mask for ITR register value */
21#define I40E_MIN_ITR 2 /* reg uses 2 usec resolution */
22#define I40E_ITR_20K 50
23#define I40E_ITR_8K 122
24#define I40E_MAX_ITR 8160 /* maximum value as per datasheet */
25#define ITR_TO_REG(setting) ((setting) & ~I40E_ITR_DYNAMIC)
26#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~I40E_ITR_MASK)
27#define ITR_IS_DYNAMIC(setting) (!!((setting) & I40E_ITR_DYNAMIC))
28
29#define I40E_ITR_RX_DEF (I40E_ITR_20K | I40E_ITR_DYNAMIC)
30#define I40E_ITR_TX_DEF (I40E_ITR_20K | I40E_ITR_DYNAMIC)
31
32/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
33 * the value of the rate limit is non-zero
34 */
35#define INTRL_ENA BIT(6)
36#define I40E_MAX_INTRL 0x3B /* reg uses 4 usec resolution */
37#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
38
39/**
40 * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
41 * @intrl: interrupt rate limit to convert
42 *
43 * This function converts a decimal interrupt rate limit to the appropriate
44 * register format expected by the firmware when setting interrupt rate limit.
45 */
46static inline u16 i40e_intrl_usec_to_reg(int intrl)
47{
48 if (intrl >> 2)
49 return ((intrl >> 2) | INTRL_ENA);
50 else
51 return 0;
52}
53
54#define I40E_QUEUE_END_OF_LIST 0x7FF
55
56/* this enum matches hardware bits and is meant to be used by DYN_CTLN
57 * registers and QINT registers or more generally anywhere in the manual
58 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
59 * register but instead is a special value meaning "don't update" ITR0/1/2.
60 */
61enum i40e_dyn_idx {
62 I40E_IDX_ITR0 = 0,
63 I40E_IDX_ITR1 = 1,
64 I40E_IDX_ITR2 = 2,
65 I40E_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
66};
67
68/* these are indexes into ITRN registers */
69#define I40E_RX_ITR I40E_IDX_ITR0
70#define I40E_TX_ITR I40E_IDX_ITR1
71#define I40E_SW_ITR I40E_IDX_ITR2
72
73/* Supported RSS offloads */
74#define I40E_DEFAULT_RSS_HENA ( \
75 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
76 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
77 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
78 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
79 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
80 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
81 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
82 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
83 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
84 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
85 BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
86
87#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
88 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
89 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
90 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
91 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
92 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
93 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
94
95#define i40e_pf_get_default_rss_hena(pf) \
96 (test_bit(I40E_HW_CAP_MULTI_TCP_UDP_RSS_PCTYPE, (pf)->hw.caps) ? \
97 I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
98
99/* Supported Rx Buffer Sizes (a multiple of 128) */
100#define I40E_RXBUFFER_256 256
101#define I40E_RXBUFFER_1536 1536 /* 128B aligned standard Ethernet frame */
102#define I40E_RXBUFFER_2048 2048
103#define I40E_RXBUFFER_3072 3072 /* Used for large frames w/ padding */
104#define I40E_MAX_RXBUFFER 9728 /* largest size for single descriptor */
105
106/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
107 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
108 * this adds up to 512 bytes of extra data meaning the smallest allocation
109 * we could have is 1K.
110 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
111 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
112 */
113#define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
114#define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
115#define i40e_rx_desc i40e_16byte_rx_desc
116
117#define I40E_RX_DMA_ATTR \
118 (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
119
120/* Attempt to maximize the headroom available for incoming frames. We
121 * use a 2K buffer for receives and need 1536/1534 to store the data for
122 * the frame. This leaves us with 512 bytes of room. From that we need
123 * to deduct the space needed for the shared info and the padding needed
124 * to IP align the frame.
125 *
126 * Note: For cache line sizes 256 or larger this value is going to end
127 * up negative. In these cases we should fall back to the legacy
128 * receive path.
129 */
130#if (PAGE_SIZE < 8192)
131#define I40E_2K_TOO_SMALL_WITH_PADDING \
132((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
133
134static inline int i40e_compute_pad(int rx_buf_len)
135{
136 int page_size, pad_size;
137
138 page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
139 pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
140
141 return pad_size;
142}
143
144static inline int i40e_skb_pad(void)
145{
146 int rx_buf_len;
147
148 /* If a 2K buffer cannot handle a standard Ethernet frame then
149 * optimize padding for a 3K buffer instead of a 1.5K buffer.
150 *
151 * For a 3K buffer we need to add enough padding to allow for
152 * tailroom due to NET_IP_ALIGN possibly shifting us out of
153 * cache-line alignment.
154 */
155 if (I40E_2K_TOO_SMALL_WITH_PADDING)
156 rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
157 else
158 rx_buf_len = I40E_RXBUFFER_1536;
159
160 /* if needed make room for NET_IP_ALIGN */
161 rx_buf_len -= NET_IP_ALIGN;
162
163 return i40e_compute_pad(rx_buf_len);
164}
165
166#define I40E_SKB_PAD i40e_skb_pad()
167#else
168#define I40E_2K_TOO_SMALL_WITH_PADDING false
169#define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
170#endif
171
172/**
173 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
174 * @rx_desc: pointer to receive descriptor (in le64 format)
175 * @stat_err_bits: value to mask
176 *
177 * This function does some fast chicanery in order to return the
178 * value of the mask which is really only used for boolean tests.
179 * The status_error_len doesn't need to be shifted because it begins
180 * at offset zero.
181 */
182static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
183 const u64 stat_err_bits)
184{
185 return !!(rx_desc->wb.qword1.status_error_len &
186 cpu_to_le64(stat_err_bits));
187}
188
189/* How many Rx Buffers do we bundle into one write to the hardware ? */
190#define I40E_RX_BUFFER_WRITE 32 /* Must be power of 2 */
191
192#define I40E_RX_NEXT_DESC(r, i, n) \
193 do { \
194 (i)++; \
195 if ((i) == (r)->count) \
196 i = 0; \
197 (n) = I40E_RX_DESC((r), (i)); \
198 } while (0)
199
200
201#define I40E_MAX_BUFFER_TXD 8
202#define I40E_MIN_TX_LEN 17
203
204/* The size limit for a transmit buffer in a descriptor is (16K - 1).
205 * In order to align with the read requests we will align the value to
206 * the nearest 4K which represents our maximum read request size.
207 */
208#define I40E_MAX_READ_REQ_SIZE 4096
209#define I40E_MAX_DATA_PER_TXD (16 * 1024 - 1)
210#define I40E_MAX_DATA_PER_TXD_ALIGNED \
211 (I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
212
213/**
214 * i40e_txd_use_count - estimate the number of descriptors needed for Tx
215 * @size: transmit request size in bytes
216 *
217 * Due to hardware alignment restrictions (4K alignment), we need to
218 * assume that we can have no more than 12K of data per descriptor, even
219 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
220 * Thus, we need to divide by 12K. But division is slow! Instead,
221 * we decompose the operation into shifts and one relatively cheap
222 * multiply operation.
223 *
224 * To divide by 12K, we first divide by 4K, then divide by 3:
225 * To divide by 4K, shift right by 12 bits
226 * To divide by 3, multiply by 85, then divide by 256
227 * (Divide by 256 is done by shifting right by 8 bits)
228 * Finally, we add one to round up. Because 256 isn't an exact multiple of
229 * 3, we'll underestimate near each multiple of 12K. This is actually more
230 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
231 * segment. For our purposes this is accurate out to 1M which is orders of
232 * magnitude greater than our largest possible GSO size.
233 *
234 * This would then be implemented as:
235 * return (((size >> 12) * 85) >> 8) + 1;
236 *
237 * Since multiplication and division are commutative, we can reorder
238 * operations into:
239 * return ((size * 85) >> 20) + 1;
240 */
241static inline unsigned int i40e_txd_use_count(unsigned int size)
242{
243 return ((size * 85) >> 20) + 1;
244}
245
246/* Tx Descriptors needed, worst case */
247#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
248
249#define I40E_TX_FLAGS_HW_VLAN BIT(1)
250#define I40E_TX_FLAGS_SW_VLAN BIT(2)
251#define I40E_TX_FLAGS_TSO BIT(3)
252#define I40E_TX_FLAGS_IPV4 BIT(4)
253#define I40E_TX_FLAGS_IPV6 BIT(5)
254#define I40E_TX_FLAGS_TSYN BIT(8)
255#define I40E_TX_FLAGS_FD_SB BIT(9)
256#define I40E_TX_FLAGS_UDP_TUNNEL BIT(10)
257#define I40E_TX_FLAGS_VLAN_MASK 0xffff0000
258#define I40E_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
259#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT 29
260#define I40E_TX_FLAGS_VLAN_SHIFT 16
261
262struct i40e_tx_buffer {
263 struct i40e_tx_desc *next_to_watch;
264 union {
265 struct xdp_frame *xdpf;
266 struct sk_buff *skb;
267 void *raw_buf;
268 };
269 unsigned int bytecount;
270 unsigned short gso_segs;
271
272 DEFINE_DMA_UNMAP_ADDR(dma);
273 DEFINE_DMA_UNMAP_LEN(len);
274 u32 tx_flags;
275};
276
277struct i40e_rx_buffer {
278 dma_addr_t dma;
279 struct page *page;
280 __u32 page_offset;
281 __u16 pagecnt_bias;
282 __u32 page_count;
283};
284
285struct i40e_queue_stats {
286 u64 packets;
287 u64 bytes;
288};
289
290struct i40e_tx_queue_stats {
291 u64 restart_queue;
292 u64 tx_busy;
293 u64 tx_done_old;
294 u64 tx_linearize;
295 u64 tx_force_wb;
296 u64 tx_stopped;
297 int prev_pkt_ctr;
298};
299
300struct i40e_rx_queue_stats {
301 u64 non_eop_descs;
302 u64 alloc_page_failed;
303 u64 alloc_buff_failed;
304 u64 page_reuse_count;
305 u64 page_alloc_count;
306 u64 page_waive_count;
307 u64 page_busy_count;
308};
309
310enum i40e_ring_state {
311 __I40E_TX_FDIR_INIT_DONE,
312 __I40E_TX_XPS_INIT_DONE,
313 __I40E_RING_STATE_NBITS /* must be last */
314};
315
316/* some useful defines for virtchannel interface, which
317 * is the only remaining user of header split
318 */
319#define I40E_RX_DTYPE_HEADER_SPLIT 1
320#define I40E_RX_SPLIT_L2 0x1
321#define I40E_RX_SPLIT_IP 0x2
322#define I40E_RX_SPLIT_TCP_UDP 0x4
323#define I40E_RX_SPLIT_SCTP 0x8
324
325/* struct that defines a descriptor ring, associated with a VSI */
326struct i40e_ring {
327 struct i40e_ring *next; /* pointer to next ring in q_vector */
328 void *desc; /* Descriptor ring memory */
329 struct device *dev; /* Used for DMA mapping */
330 struct net_device *netdev; /* netdev ring maps to */
331 struct bpf_prog *xdp_prog;
332 union {
333 struct i40e_tx_buffer *tx_bi;
334 struct i40e_rx_buffer *rx_bi;
335 struct xdp_buff **rx_bi_zc;
336 };
337 DECLARE_BITMAP(state, __I40E_RING_STATE_NBITS);
338 u16 queue_index; /* Queue number of ring */
339 u8 dcb_tc; /* Traffic class of ring */
340 u8 __iomem *tail;
341
342 /* Storing xdp_buff on ring helps in saving the state of partially built
343 * packet when i40e_clean_rx_ring_irq() must return before it sees EOP
344 * and to resume packet building for this ring in the next call to
345 * i40e_clean_rx_ring_irq().
346 */
347 struct xdp_buff xdp;
348
349 /* Next descriptor to be processed; next_to_clean is updated only on
350 * processing EOP descriptor
351 */
352 u16 next_to_process;
353 /* high bit set means dynamic, use accessor routines to read/write.
354 * hardware only supports 2us resolution for the ITR registers.
355 * these values always store the USER setting, and must be converted
356 * before programming to a register.
357 */
358 u16 itr_setting;
359
360 u16 count; /* Number of descriptors */
361 u16 reg_idx; /* HW register index of the ring */
362 u16 rx_buf_len;
363
364 /* used in interrupt processing */
365 u16 next_to_use;
366 u16 next_to_clean;
367 u16 xdp_tx_active;
368
369 u8 atr_sample_rate;
370 u8 atr_count;
371
372 bool ring_active; /* is ring online or not */
373 bool arm_wb; /* do something to arm write back */
374 u8 packet_stride;
375
376 u16 flags;
377#define I40E_TXR_FLAGS_WB_ON_ITR BIT(0)
378#define I40E_RXR_FLAGS_BUILD_SKB_ENABLED BIT(1)
379#define I40E_TXR_FLAGS_XDP BIT(2)
380
381 /* stats structs */
382 struct i40e_queue_stats stats;
383 struct u64_stats_sync syncp;
384 union {
385 struct i40e_tx_queue_stats tx_stats;
386 struct i40e_rx_queue_stats rx_stats;
387 };
388
389 unsigned int size; /* length of descriptor ring in bytes */
390 dma_addr_t dma; /* physical address of ring */
391
392 struct i40e_vsi *vsi; /* Backreference to associated VSI */
393 struct i40e_q_vector *q_vector; /* Backreference to associated vector */
394
395 struct rcu_head rcu; /* to avoid race on free */
396 u16 next_to_alloc;
397
398 struct i40e_channel *ch;
399 u16 rx_offset;
400 struct xdp_rxq_info xdp_rxq;
401 struct xsk_buff_pool *xsk_pool;
402} ____cacheline_internodealigned_in_smp;
403
404static inline bool ring_uses_build_skb(struct i40e_ring *ring)
405{
406 return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
407}
408
409static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
410{
411 ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
412}
413
414static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
415{
416 ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
417}
418
419static inline bool ring_is_xdp(struct i40e_ring *ring)
420{
421 return !!(ring->flags & I40E_TXR_FLAGS_XDP);
422}
423
424static inline void set_ring_xdp(struct i40e_ring *ring)
425{
426 ring->flags |= I40E_TXR_FLAGS_XDP;
427}
428
429#define I40E_ITR_ADAPTIVE_MIN_INC 0x0002
430#define I40E_ITR_ADAPTIVE_MIN_USECS 0x0002
431#define I40E_ITR_ADAPTIVE_MAX_USECS 0x007e
432#define I40E_ITR_ADAPTIVE_LATENCY 0x8000
433#define I40E_ITR_ADAPTIVE_BULK 0x0000
434
435struct i40e_ring_container {
436 struct i40e_ring *ring; /* pointer to linked list of ring(s) */
437 unsigned long next_update; /* jiffies value of next update */
438 unsigned int total_bytes; /* total bytes processed this int */
439 unsigned int total_packets; /* total packets processed this int */
440 u16 count;
441 u16 target_itr; /* target ITR setting for ring(s) */
442 u16 current_itr; /* current ITR setting for ring(s) */
443};
444
445/* iterator for handling rings in ring container */
446#define i40e_for_each_ring(pos, head) \
447 for (pos = (head).ring; pos != NULL; pos = pos->next)
448
449static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
450{
451#if (PAGE_SIZE < 8192)
452 if (ring->rx_buf_len > (PAGE_SIZE / 2))
453 return 1;
454#endif
455 return 0;
456}
457
458#define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
459
460bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
461netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
462u16 i40e_lan_select_queue(struct net_device *netdev, struct sk_buff *skb,
463 struct net_device *sb_dev);
464void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
465void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
466int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
467int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
468void i40e_free_tx_resources(struct i40e_ring *tx_ring);
469void i40e_free_rx_resources(struct i40e_ring *rx_ring);
470int i40e_napi_poll(struct napi_struct *napi, int budget);
471void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
472u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
473void i40e_detect_recover_hung(struct i40e_pf *pf);
474int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
475bool __i40e_chk_linearize(struct sk_buff *skb);
476int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
477 u32 flags);
478bool i40e_is_non_eop(struct i40e_ring *rx_ring,
479 union i40e_rx_desc *rx_desc);
480
481/**
482 * i40e_get_head - Retrieve head from head writeback
483 * @tx_ring: tx ring to fetch head of
484 *
485 * Returns value of Tx ring head based on value stored
486 * in head write-back location
487 **/
488static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
489{
490 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
491
492 return le32_to_cpu(*(volatile __le32 *)head);
493}
494
495/**
496 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
497 * @skb: send buffer
498 *
499 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
500 * there is not enough descriptors available in this ring since we need at least
501 * one descriptor.
502 **/
503static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
504{
505 const skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
506 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
507 int count = 0, size = skb_headlen(skb);
508
509 for (;;) {
510 count += i40e_txd_use_count(size);
511
512 if (!nr_frags--)
513 break;
514
515 size = skb_frag_size(frag++);
516 }
517
518 return count;
519}
520
521/**
522 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
523 * @tx_ring: the ring to be checked
524 * @size: the size buffer we want to assure is available
525 *
526 * Returns 0 if stop is not needed
527 **/
528static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
529{
530 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
531 return 0;
532 return __i40e_maybe_stop_tx(tx_ring, size);
533}
534
535/**
536 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
537 * @skb: send buffer
538 * @count: number of buffers used
539 *
540 * Note: Our HW can't scatter-gather more than 8 fragments to build
541 * a packet on the wire and so we need to figure out the cases where we
542 * need to linearize the skb.
543 **/
544static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
545{
546 /* Both TSO and single send will work if count is less than 8 */
547 if (likely(count < I40E_MAX_BUFFER_TXD))
548 return false;
549
550 if (skb_is_gso(skb))
551 return __i40e_chk_linearize(skb);
552
553 /* we can support up to 8 data buffers for a single send */
554 return count != I40E_MAX_BUFFER_TXD;
555}
556
557/**
558 * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
559 * @ring: Tx ring to find the netdev equivalent of
560 **/
561static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
562{
563 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
564}
565#endif /* _I40E_TXRX_H_ */
1/*******************************************************************************
2 *
3 * Intel Ethernet Controller XL710 Family Linux Driver
4 * Copyright(c) 2013 - 2016 Intel Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program. If not, see <http://www.gnu.org/licenses/>.
17 *
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
20 *
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 *
25 ******************************************************************************/
26
27#ifndef _I40E_TXRX_H_
28#define _I40E_TXRX_H_
29
30/* Interrupt Throttling and Rate Limiting Goodies */
31
32#define I40E_MAX_ITR 0x0FF0 /* reg uses 2 usec resolution */
33#define I40E_MIN_ITR 0x0001 /* reg uses 2 usec resolution */
34#define I40E_ITR_100K 0x0005
35#define I40E_ITR_50K 0x000A
36#define I40E_ITR_20K 0x0019
37#define I40E_ITR_18K 0x001B
38#define I40E_ITR_8K 0x003E
39#define I40E_ITR_4K 0x007A
40#define I40E_MAX_INTRL 0x3B /* reg uses 4 usec resolution */
41#define I40E_ITR_RX_DEF I40E_ITR_20K
42#define I40E_ITR_TX_DEF I40E_ITR_20K
43#define I40E_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
44#define I40E_MIN_INT_RATE 250 /* ~= 1000000 / (I40E_MAX_ITR * 2) */
45#define I40E_MAX_INT_RATE 500000 /* == 1000000 / (I40E_MIN_ITR * 2) */
46#define I40E_DEFAULT_IRQ_WORK 256
47#define ITR_TO_REG(setting) ((setting & ~I40E_ITR_DYNAMIC) >> 1)
48#define ITR_IS_DYNAMIC(setting) (!!(setting & I40E_ITR_DYNAMIC))
49#define ITR_REG_TO_USEC(itr_reg) (itr_reg << 1)
50/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
51 * the value of the rate limit is non-zero
52 */
53#define INTRL_ENA BIT(6)
54#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
55#define INTRL_USEC_TO_REG(set) ((set) ? ((set) >> 2) | INTRL_ENA : 0)
56#define I40E_INTRL_8K 125 /* 8000 ints/sec */
57#define I40E_INTRL_62K 16 /* 62500 ints/sec */
58#define I40E_INTRL_83K 12 /* 83333 ints/sec */
59
60#define I40E_QUEUE_END_OF_LIST 0x7FF
61
62/* this enum matches hardware bits and is meant to be used by DYN_CTLN
63 * registers and QINT registers or more generally anywhere in the manual
64 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
65 * register but instead is a special value meaning "don't update" ITR0/1/2.
66 */
67enum i40e_dyn_idx_t {
68 I40E_IDX_ITR0 = 0,
69 I40E_IDX_ITR1 = 1,
70 I40E_IDX_ITR2 = 2,
71 I40E_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
72};
73
74/* these are indexes into ITRN registers */
75#define I40E_RX_ITR I40E_IDX_ITR0
76#define I40E_TX_ITR I40E_IDX_ITR1
77#define I40E_PE_ITR I40E_IDX_ITR2
78
79/* Supported RSS offloads */
80#define I40E_DEFAULT_RSS_HENA ( \
81 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
82 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
83 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
84 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
85 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
86 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
87 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
88 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
89 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
90 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
91 BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
92
93#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
94 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
95 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
96 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
97 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
98 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
99 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
100
101#define i40e_pf_get_default_rss_hena(pf) \
102 (((pf)->flags & I40E_FLAG_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
103 I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
104
105/* Supported Rx Buffer Sizes (a multiple of 128) */
106#define I40E_RXBUFFER_256 256
107#define I40E_RXBUFFER_2048 2048
108#define I40E_RXBUFFER_3072 3072 /* For FCoE MTU of 2158 */
109#define I40E_RXBUFFER_4096 4096
110#define I40E_RXBUFFER_8192 8192
111#define I40E_MAX_RXBUFFER 9728 /* largest size for single descriptor */
112
113/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
114 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
115 * this adds up to 512 bytes of extra data meaning the smallest allocation
116 * we could have is 1K.
117 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
118 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
119 */
120#define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
121#define i40e_rx_desc i40e_32byte_rx_desc
122
123/**
124 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
125 * @rx_desc: pointer to receive descriptor (in le64 format)
126 * @stat_err_bits: value to mask
127 *
128 * This function does some fast chicanery in order to return the
129 * value of the mask which is really only used for boolean tests.
130 * The status_error_len doesn't need to be shifted because it begins
131 * at offset zero.
132 */
133static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
134 const u64 stat_err_bits)
135{
136 return !!(rx_desc->wb.qword1.status_error_len &
137 cpu_to_le64(stat_err_bits));
138}
139
140/* How many Rx Buffers do we bundle into one write to the hardware ? */
141#define I40E_RX_BUFFER_WRITE 16 /* Must be power of 2 */
142#define I40E_RX_INCREMENT(r, i) \
143 do { \
144 (i)++; \
145 if ((i) == (r)->count) \
146 i = 0; \
147 r->next_to_clean = i; \
148 } while (0)
149
150#define I40E_RX_NEXT_DESC(r, i, n) \
151 do { \
152 (i)++; \
153 if ((i) == (r)->count) \
154 i = 0; \
155 (n) = I40E_RX_DESC((r), (i)); \
156 } while (0)
157
158#define I40E_RX_NEXT_DESC_PREFETCH(r, i, n) \
159 do { \
160 I40E_RX_NEXT_DESC((r), (i), (n)); \
161 prefetch((n)); \
162 } while (0)
163
164#define I40E_MAX_BUFFER_TXD 8
165#define I40E_MIN_TX_LEN 17
166
167/* The size limit for a transmit buffer in a descriptor is (16K - 1).
168 * In order to align with the read requests we will align the value to
169 * the nearest 4K which represents our maximum read request size.
170 */
171#define I40E_MAX_READ_REQ_SIZE 4096
172#define I40E_MAX_DATA_PER_TXD (16 * 1024 - 1)
173#define I40E_MAX_DATA_PER_TXD_ALIGNED \
174 (I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
175
176/**
177 * i40e_txd_use_count - estimate the number of descriptors needed for Tx
178 * @size: transmit request size in bytes
179 *
180 * Due to hardware alignment restrictions (4K alignment), we need to
181 * assume that we can have no more than 12K of data per descriptor, even
182 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
183 * Thus, we need to divide by 12K. But division is slow! Instead,
184 * we decompose the operation into shifts and one relatively cheap
185 * multiply operation.
186 *
187 * To divide by 12K, we first divide by 4K, then divide by 3:
188 * To divide by 4K, shift right by 12 bits
189 * To divide by 3, multiply by 85, then divide by 256
190 * (Divide by 256 is done by shifting right by 8 bits)
191 * Finally, we add one to round up. Because 256 isn't an exact multiple of
192 * 3, we'll underestimate near each multiple of 12K. This is actually more
193 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
194 * segment. For our purposes this is accurate out to 1M which is orders of
195 * magnitude greater than our largest possible GSO size.
196 *
197 * This would then be implemented as:
198 * return (((size >> 12) * 85) >> 8) + 1;
199 *
200 * Since multiplication and division are commutative, we can reorder
201 * operations into:
202 * return ((size * 85) >> 20) + 1;
203 */
204static inline unsigned int i40e_txd_use_count(unsigned int size)
205{
206 return ((size * 85) >> 20) + 1;
207}
208
209/* Tx Descriptors needed, worst case */
210#define DESC_NEEDED (MAX_SKB_FRAGS + 4)
211#define I40E_MIN_DESC_PENDING 4
212
213#define I40E_TX_FLAGS_HW_VLAN BIT(1)
214#define I40E_TX_FLAGS_SW_VLAN BIT(2)
215#define I40E_TX_FLAGS_TSO BIT(3)
216#define I40E_TX_FLAGS_IPV4 BIT(4)
217#define I40E_TX_FLAGS_IPV6 BIT(5)
218#define I40E_TX_FLAGS_FCCRC BIT(6)
219#define I40E_TX_FLAGS_FSO BIT(7)
220#define I40E_TX_FLAGS_TSYN BIT(8)
221#define I40E_TX_FLAGS_FD_SB BIT(9)
222#define I40E_TX_FLAGS_UDP_TUNNEL BIT(10)
223#define I40E_TX_FLAGS_VLAN_MASK 0xffff0000
224#define I40E_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
225#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT 29
226#define I40E_TX_FLAGS_VLAN_SHIFT 16
227
228struct i40e_tx_buffer {
229 struct i40e_tx_desc *next_to_watch;
230 union {
231 struct sk_buff *skb;
232 void *raw_buf;
233 };
234 unsigned int bytecount;
235 unsigned short gso_segs;
236
237 DEFINE_DMA_UNMAP_ADDR(dma);
238 DEFINE_DMA_UNMAP_LEN(len);
239 u32 tx_flags;
240};
241
242struct i40e_rx_buffer {
243 struct sk_buff *skb;
244 dma_addr_t dma;
245 struct page *page;
246 unsigned int page_offset;
247};
248
249struct i40e_queue_stats {
250 u64 packets;
251 u64 bytes;
252};
253
254struct i40e_tx_queue_stats {
255 u64 restart_queue;
256 u64 tx_busy;
257 u64 tx_done_old;
258 u64 tx_linearize;
259 u64 tx_force_wb;
260 u64 tx_lost_interrupt;
261};
262
263struct i40e_rx_queue_stats {
264 u64 non_eop_descs;
265 u64 alloc_page_failed;
266 u64 alloc_buff_failed;
267 u64 page_reuse_count;
268 u64 realloc_count;
269};
270
271enum i40e_ring_state_t {
272 __I40E_TX_FDIR_INIT_DONE,
273 __I40E_TX_XPS_INIT_DONE,
274};
275
276/* some useful defines for virtchannel interface, which
277 * is the only remaining user of header split
278 */
279#define I40E_RX_DTYPE_NO_SPLIT 0
280#define I40E_RX_DTYPE_HEADER_SPLIT 1
281#define I40E_RX_DTYPE_SPLIT_ALWAYS 2
282#define I40E_RX_SPLIT_L2 0x1
283#define I40E_RX_SPLIT_IP 0x2
284#define I40E_RX_SPLIT_TCP_UDP 0x4
285#define I40E_RX_SPLIT_SCTP 0x8
286
287/* struct that defines a descriptor ring, associated with a VSI */
288struct i40e_ring {
289 struct i40e_ring *next; /* pointer to next ring in q_vector */
290 void *desc; /* Descriptor ring memory */
291 struct device *dev; /* Used for DMA mapping */
292 struct net_device *netdev; /* netdev ring maps to */
293 union {
294 struct i40e_tx_buffer *tx_bi;
295 struct i40e_rx_buffer *rx_bi;
296 };
297 unsigned long state;
298 u16 queue_index; /* Queue number of ring */
299 u8 dcb_tc; /* Traffic class of ring */
300 u8 __iomem *tail;
301
302 /* high bit set means dynamic, use accessor routines to read/write.
303 * hardware only supports 2us resolution for the ITR registers.
304 * these values always store the USER setting, and must be converted
305 * before programming to a register.
306 */
307 u16 rx_itr_setting;
308 u16 tx_itr_setting;
309
310 u16 count; /* Number of descriptors */
311 u16 reg_idx; /* HW register index of the ring */
312 u16 rx_buf_len;
313
314 /* used in interrupt processing */
315 u16 next_to_use;
316 u16 next_to_clean;
317
318 u8 atr_sample_rate;
319 u8 atr_count;
320
321 bool ring_active; /* is ring online or not */
322 bool arm_wb; /* do something to arm write back */
323 u8 packet_stride;
324
325 u16 flags;
326#define I40E_TXR_FLAGS_WB_ON_ITR BIT(0)
327
328 /* stats structs */
329 struct i40e_queue_stats stats;
330 struct u64_stats_sync syncp;
331 union {
332 struct i40e_tx_queue_stats tx_stats;
333 struct i40e_rx_queue_stats rx_stats;
334 };
335
336 unsigned int size; /* length of descriptor ring in bytes */
337 dma_addr_t dma; /* physical address of ring */
338
339 struct i40e_vsi *vsi; /* Backreference to associated VSI */
340 struct i40e_q_vector *q_vector; /* Backreference to associated vector */
341
342 struct rcu_head rcu; /* to avoid race on free */
343 u16 next_to_alloc;
344} ____cacheline_internodealigned_in_smp;
345
346enum i40e_latency_range {
347 I40E_LOWEST_LATENCY = 0,
348 I40E_LOW_LATENCY = 1,
349 I40E_BULK_LATENCY = 2,
350 I40E_ULTRA_LATENCY = 3,
351};
352
353struct i40e_ring_container {
354 /* array of pointers to rings */
355 struct i40e_ring *ring;
356 unsigned int total_bytes; /* total bytes processed this int */
357 unsigned int total_packets; /* total packets processed this int */
358 u16 count;
359 enum i40e_latency_range latency_range;
360 u16 itr;
361};
362
363/* iterator for handling rings in ring container */
364#define i40e_for_each_ring(pos, head) \
365 for (pos = (head).ring; pos != NULL; pos = pos->next)
366
367bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
368netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
369void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
370void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
371int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
372int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
373void i40e_free_tx_resources(struct i40e_ring *tx_ring);
374void i40e_free_rx_resources(struct i40e_ring *rx_ring);
375int i40e_napi_poll(struct napi_struct *napi, int budget);
376#ifdef I40E_FCOE
377void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
378 struct i40e_tx_buffer *first, u32 tx_flags,
379 const u8 hdr_len, u32 td_cmd, u32 td_offset);
380int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
381 struct i40e_ring *tx_ring, u32 *flags);
382#endif
383void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
384u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
385int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
386bool __i40e_chk_linearize(struct sk_buff *skb);
387
388/**
389 * i40e_get_head - Retrieve head from head writeback
390 * @tx_ring: tx ring to fetch head of
391 *
392 * Returns value of Tx ring head based on value stored
393 * in head write-back location
394 **/
395static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
396{
397 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
398
399 return le32_to_cpu(*(volatile __le32 *)head);
400}
401
402/**
403 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
404 * @skb: send buffer
405 * @tx_ring: ring to send buffer on
406 *
407 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
408 * there is not enough descriptors available in this ring since we need at least
409 * one descriptor.
410 **/
411static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
412{
413 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
414 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
415 int count = 0, size = skb_headlen(skb);
416
417 for (;;) {
418 count += i40e_txd_use_count(size);
419
420 if (!nr_frags--)
421 break;
422
423 size = skb_frag_size(frag++);
424 }
425
426 return count;
427}
428
429/**
430 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
431 * @tx_ring: the ring to be checked
432 * @size: the size buffer we want to assure is available
433 *
434 * Returns 0 if stop is not needed
435 **/
436static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
437{
438 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
439 return 0;
440 return __i40e_maybe_stop_tx(tx_ring, size);
441}
442
443/**
444 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
445 * @skb: send buffer
446 * @count: number of buffers used
447 *
448 * Note: Our HW can't scatter-gather more than 8 fragments to build
449 * a packet on the wire and so we need to figure out the cases where we
450 * need to linearize the skb.
451 **/
452static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
453{
454 /* Both TSO and single send will work if count is less than 8 */
455 if (likely(count < I40E_MAX_BUFFER_TXD))
456 return false;
457
458 if (skb_is_gso(skb))
459 return __i40e_chk_linearize(skb);
460
461 /* we can support up to 8 data buffers for a single send */
462 return count != I40E_MAX_BUFFER_TXD;
463}
464
465/**
466 * i40e_rx_is_fcoe - returns true if the Rx packet type is FCoE
467 * @ptype: the packet type field from Rx descriptor write-back
468 **/
469static inline bool i40e_rx_is_fcoe(u16 ptype)
470{
471 return (ptype >= I40E_RX_PTYPE_L2_FCOE_PAY3) &&
472 (ptype <= I40E_RX_PTYPE_L2_FCOE_VFT_FCOTHER);
473}
474
475/**
476 * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
477 * @ring: Tx ring to find the netdev equivalent of
478 **/
479static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
480{
481 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
482}
483#endif /* _I40E_TXRX_H_ */