xref: /openbmc/linux/drivers/net/ethernet/intel/i40e/i40e_txrx.h (revision 93707cbabcc8baf2b2b5f4a99c1f08ee83eb7abd)
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 #include <net/xdp.h>
31 
32 /* Interrupt Throttling and Rate Limiting Goodies */
33 #define I40E_DEFAULT_IRQ_WORK      256
34 
35 /* The datasheet for the X710 and XL710 indicate that the maximum value for
36  * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
37  * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
38  * the register value which is divided by 2 lets use the actual values and
39  * avoid an excessive amount of translation.
40  */
41 #define I40E_ITR_DYNAMIC	0x8000	/* use top bit as a flag */
42 #define I40E_ITR_MASK		0x1FFE	/* mask for ITR register value */
43 #define I40E_MIN_ITR		     2	/* reg uses 2 usec resolution */
44 #define I40E_ITR_100K		    10	/* all values below must be even */
45 #define I40E_ITR_50K		    20
46 #define I40E_ITR_20K		    50
47 #define I40E_ITR_18K		    60
48 #define I40E_ITR_8K		   122
49 #define I40E_MAX_ITR		  8160	/* maximum value as per datasheet */
50 #define ITR_TO_REG(setting) ((setting) & ~I40E_ITR_DYNAMIC)
51 #define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~I40E_ITR_MASK)
52 #define ITR_IS_DYNAMIC(setting) (!!((setting) & I40E_ITR_DYNAMIC))
53 
54 #define I40E_ITR_RX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
55 #define I40E_ITR_TX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
56 
57 /* 0x40 is the enable bit for interrupt rate limiting, and must be set if
58  * the value of the rate limit is non-zero
59  */
60 #define INTRL_ENA                  BIT(6)
61 #define I40E_MAX_INTRL             0x3B    /* reg uses 4 usec resolution */
62 #define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
63 
64 /**
65  * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
66  * @intrl: interrupt rate limit to convert
67  *
68  * This function converts a decimal interrupt rate limit to the appropriate
69  * register format expected by the firmware when setting interrupt rate limit.
70  */
71 static inline u16 i40e_intrl_usec_to_reg(int intrl)
72 {
73 	if (intrl >> 2)
74 		return ((intrl >> 2) | INTRL_ENA);
75 	else
76 		return 0;
77 }
78 #define I40E_INTRL_8K              125     /* 8000 ints/sec */
79 #define I40E_INTRL_62K             16      /* 62500 ints/sec */
80 #define I40E_INTRL_83K             12      /* 83333 ints/sec */
81 
82 #define I40E_QUEUE_END_OF_LIST 0x7FF
83 
84 /* this enum matches hardware bits and is meant to be used by DYN_CTLN
85  * registers and QINT registers or more generally anywhere in the manual
86  * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
87  * register but instead is a special value meaning "don't update" ITR0/1/2.
88  */
89 enum i40e_dyn_idx_t {
90 	I40E_IDX_ITR0 = 0,
91 	I40E_IDX_ITR1 = 1,
92 	I40E_IDX_ITR2 = 2,
93 	I40E_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
94 };
95 
96 /* these are indexes into ITRN registers */
97 #define I40E_RX_ITR    I40E_IDX_ITR0
98 #define I40E_TX_ITR    I40E_IDX_ITR1
99 #define I40E_PE_ITR    I40E_IDX_ITR2
100 
101 /* Supported RSS offloads */
102 #define I40E_DEFAULT_RSS_HENA ( \
103 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
104 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
105 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
106 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
107 	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
108 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
109 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
110 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
111 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
112 	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
113 	BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
114 
115 #define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
116 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
117 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
118 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
119 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
120 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
121 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
122 
123 #define i40e_pf_get_default_rss_hena(pf) \
124 	(((pf)->hw_features & I40E_HW_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
125 	  I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
126 
127 /* Supported Rx Buffer Sizes (a multiple of 128) */
128 #define I40E_RXBUFFER_256   256
129 #define I40E_RXBUFFER_1536  1536  /* 128B aligned standard Ethernet frame */
130 #define I40E_RXBUFFER_2048  2048
131 #define I40E_RXBUFFER_3072  3072  /* Used for large frames w/ padding */
132 #define I40E_MAX_RXBUFFER   9728  /* largest size for single descriptor */
133 
134 /* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
135  * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
136  * this adds up to 512 bytes of extra data meaning the smallest allocation
137  * we could have is 1K.
138  * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
139  * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
140  */
141 #define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
142 #define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
143 #define i40e_rx_desc i40e_32byte_rx_desc
144 
145 #define I40E_RX_DMA_ATTR \
146 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
147 
148 /* Attempt to maximize the headroom available for incoming frames.  We
149  * use a 2K buffer for receives and need 1536/1534 to store the data for
150  * the frame.  This leaves us with 512 bytes of room.  From that we need
151  * to deduct the space needed for the shared info and the padding needed
152  * to IP align the frame.
153  *
154  * Note: For cache line sizes 256 or larger this value is going to end
155  *	 up negative.  In these cases we should fall back to the legacy
156  *	 receive path.
157  */
158 #if (PAGE_SIZE < 8192)
159 #define I40E_2K_TOO_SMALL_WITH_PADDING \
160 ((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
161 
162 static inline int i40e_compute_pad(int rx_buf_len)
163 {
164 	int page_size, pad_size;
165 
166 	page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
167 	pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
168 
169 	return pad_size;
170 }
171 
172 static inline int i40e_skb_pad(void)
173 {
174 	int rx_buf_len;
175 
176 	/* If a 2K buffer cannot handle a standard Ethernet frame then
177 	 * optimize padding for a 3K buffer instead of a 1.5K buffer.
178 	 *
179 	 * For a 3K buffer we need to add enough padding to allow for
180 	 * tailroom due to NET_IP_ALIGN possibly shifting us out of
181 	 * cache-line alignment.
182 	 */
183 	if (I40E_2K_TOO_SMALL_WITH_PADDING)
184 		rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
185 	else
186 		rx_buf_len = I40E_RXBUFFER_1536;
187 
188 	/* if needed make room for NET_IP_ALIGN */
189 	rx_buf_len -= NET_IP_ALIGN;
190 
191 	return i40e_compute_pad(rx_buf_len);
192 }
193 
194 #define I40E_SKB_PAD i40e_skb_pad()
195 #else
196 #define I40E_2K_TOO_SMALL_WITH_PADDING false
197 #define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
198 #endif
199 
200 /**
201  * i40e_test_staterr - tests bits in Rx descriptor status and error fields
202  * @rx_desc: pointer to receive descriptor (in le64 format)
203  * @stat_err_bits: value to mask
204  *
205  * This function does some fast chicanery in order to return the
206  * value of the mask which is really only used for boolean tests.
207  * The status_error_len doesn't need to be shifted because it begins
208  * at offset zero.
209  */
210 static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
211 				     const u64 stat_err_bits)
212 {
213 	return !!(rx_desc->wb.qword1.status_error_len &
214 		  cpu_to_le64(stat_err_bits));
215 }
216 
217 /* How many Rx Buffers do we bundle into one write to the hardware ? */
218 #define I40E_RX_BUFFER_WRITE	32	/* Must be power of 2 */
219 #define I40E_RX_INCREMENT(r, i) \
220 	do {					\
221 		(i)++;				\
222 		if ((i) == (r)->count)		\
223 			i = 0;			\
224 		r->next_to_clean = i;		\
225 	} while (0)
226 
227 #define I40E_RX_NEXT_DESC(r, i, n)		\
228 	do {					\
229 		(i)++;				\
230 		if ((i) == (r)->count)		\
231 			i = 0;			\
232 		(n) = I40E_RX_DESC((r), (i));	\
233 	} while (0)
234 
235 #define I40E_RX_NEXT_DESC_PREFETCH(r, i, n)		\
236 	do {						\
237 		I40E_RX_NEXT_DESC((r), (i), (n));	\
238 		prefetch((n));				\
239 	} while (0)
240 
241 #define I40E_MAX_BUFFER_TXD	8
242 #define I40E_MIN_TX_LEN		17
243 
244 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
245  * In order to align with the read requests we will align the value to
246  * the nearest 4K which represents our maximum read request size.
247  */
248 #define I40E_MAX_READ_REQ_SIZE		4096
249 #define I40E_MAX_DATA_PER_TXD		(16 * 1024 - 1)
250 #define I40E_MAX_DATA_PER_TXD_ALIGNED \
251 	(I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
252 
253 /**
254  * i40e_txd_use_count  - estimate the number of descriptors needed for Tx
255  * @size: transmit request size in bytes
256  *
257  * Due to hardware alignment restrictions (4K alignment), we need to
258  * assume that we can have no more than 12K of data per descriptor, even
259  * though each descriptor can take up to 16K - 1 bytes of aligned memory.
260  * Thus, we need to divide by 12K. But division is slow! Instead,
261  * we decompose the operation into shifts and one relatively cheap
262  * multiply operation.
263  *
264  * To divide by 12K, we first divide by 4K, then divide by 3:
265  *     To divide by 4K, shift right by 12 bits
266  *     To divide by 3, multiply by 85, then divide by 256
267  *     (Divide by 256 is done by shifting right by 8 bits)
268  * Finally, we add one to round up. Because 256 isn't an exact multiple of
269  * 3, we'll underestimate near each multiple of 12K. This is actually more
270  * accurate as we have 4K - 1 of wiggle room that we can fit into the last
271  * segment.  For our purposes this is accurate out to 1M which is orders of
272  * magnitude greater than our largest possible GSO size.
273  *
274  * This would then be implemented as:
275  *     return (((size >> 12) * 85) >> 8) + 1;
276  *
277  * Since multiplication and division are commutative, we can reorder
278  * operations into:
279  *     return ((size * 85) >> 20) + 1;
280  */
281 static inline unsigned int i40e_txd_use_count(unsigned int size)
282 {
283 	return ((size * 85) >> 20) + 1;
284 }
285 
286 /* Tx Descriptors needed, worst case */
287 #define DESC_NEEDED (MAX_SKB_FRAGS + 6)
288 #define I40E_MIN_DESC_PENDING	4
289 
290 #define I40E_TX_FLAGS_HW_VLAN		BIT(1)
291 #define I40E_TX_FLAGS_SW_VLAN		BIT(2)
292 #define I40E_TX_FLAGS_TSO		BIT(3)
293 #define I40E_TX_FLAGS_IPV4		BIT(4)
294 #define I40E_TX_FLAGS_IPV6		BIT(5)
295 #define I40E_TX_FLAGS_FCCRC		BIT(6)
296 #define I40E_TX_FLAGS_FSO		BIT(7)
297 #define I40E_TX_FLAGS_TSYN		BIT(8)
298 #define I40E_TX_FLAGS_FD_SB		BIT(9)
299 #define I40E_TX_FLAGS_UDP_TUNNEL	BIT(10)
300 #define I40E_TX_FLAGS_VLAN_MASK		0xffff0000
301 #define I40E_TX_FLAGS_VLAN_PRIO_MASK	0xe0000000
302 #define I40E_TX_FLAGS_VLAN_PRIO_SHIFT	29
303 #define I40E_TX_FLAGS_VLAN_SHIFT	16
304 
305 struct i40e_tx_buffer {
306 	struct i40e_tx_desc *next_to_watch;
307 	union {
308 		struct sk_buff *skb;
309 		void *raw_buf;
310 	};
311 	unsigned int bytecount;
312 	unsigned short gso_segs;
313 
314 	DEFINE_DMA_UNMAP_ADDR(dma);
315 	DEFINE_DMA_UNMAP_LEN(len);
316 	u32 tx_flags;
317 };
318 
319 struct i40e_rx_buffer {
320 	dma_addr_t dma;
321 	struct page *page;
322 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
323 	__u32 page_offset;
324 #else
325 	__u16 page_offset;
326 #endif
327 	__u16 pagecnt_bias;
328 };
329 
330 struct i40e_queue_stats {
331 	u64 packets;
332 	u64 bytes;
333 };
334 
335 struct i40e_tx_queue_stats {
336 	u64 restart_queue;
337 	u64 tx_busy;
338 	u64 tx_done_old;
339 	u64 tx_linearize;
340 	u64 tx_force_wb;
341 	int prev_pkt_ctr;
342 };
343 
344 struct i40e_rx_queue_stats {
345 	u64 non_eop_descs;
346 	u64 alloc_page_failed;
347 	u64 alloc_buff_failed;
348 	u64 page_reuse_count;
349 	u64 realloc_count;
350 };
351 
352 enum i40e_ring_state_t {
353 	__I40E_TX_FDIR_INIT_DONE,
354 	__I40E_TX_XPS_INIT_DONE,
355 	__I40E_RING_STATE_NBITS /* must be last */
356 };
357 
358 /* some useful defines for virtchannel interface, which
359  * is the only remaining user of header split
360  */
361 #define I40E_RX_DTYPE_NO_SPLIT      0
362 #define I40E_RX_DTYPE_HEADER_SPLIT  1
363 #define I40E_RX_DTYPE_SPLIT_ALWAYS  2
364 #define I40E_RX_SPLIT_L2      0x1
365 #define I40E_RX_SPLIT_IP      0x2
366 #define I40E_RX_SPLIT_TCP_UDP 0x4
367 #define I40E_RX_SPLIT_SCTP    0x8
368 
369 /* struct that defines a descriptor ring, associated with a VSI */
370 struct i40e_ring {
371 	struct i40e_ring *next;		/* pointer to next ring in q_vector */
372 	void *desc;			/* Descriptor ring memory */
373 	struct device *dev;		/* Used for DMA mapping */
374 	struct net_device *netdev;	/* netdev ring maps to */
375 	struct bpf_prog *xdp_prog;
376 	union {
377 		struct i40e_tx_buffer *tx_bi;
378 		struct i40e_rx_buffer *rx_bi;
379 	};
380 	DECLARE_BITMAP(state, __I40E_RING_STATE_NBITS);
381 	u16 queue_index;		/* Queue number of ring */
382 	u8 dcb_tc;			/* Traffic class of ring */
383 	u8 __iomem *tail;
384 
385 	/* high bit set means dynamic, use accessor routines to read/write.
386 	 * hardware only supports 2us resolution for the ITR registers.
387 	 * these values always store the USER setting, and must be converted
388 	 * before programming to a register.
389 	 */
390 	u16 itr_setting;
391 
392 	u16 count;			/* Number of descriptors */
393 	u16 reg_idx;			/* HW register index of the ring */
394 	u16 rx_buf_len;
395 
396 	/* used in interrupt processing */
397 	u16 next_to_use;
398 	u16 next_to_clean;
399 
400 	u8 atr_sample_rate;
401 	u8 atr_count;
402 
403 	bool ring_active;		/* is ring online or not */
404 	bool arm_wb;		/* do something to arm write back */
405 	u8 packet_stride;
406 
407 	u16 flags;
408 #define I40E_TXR_FLAGS_WB_ON_ITR		BIT(0)
409 #define I40E_RXR_FLAGS_BUILD_SKB_ENABLED	BIT(1)
410 #define I40E_TXR_FLAGS_XDP			BIT(2)
411 
412 	/* stats structs */
413 	struct i40e_queue_stats	stats;
414 	struct u64_stats_sync syncp;
415 	union {
416 		struct i40e_tx_queue_stats tx_stats;
417 		struct i40e_rx_queue_stats rx_stats;
418 	};
419 
420 	unsigned int size;		/* length of descriptor ring in bytes */
421 	dma_addr_t dma;			/* physical address of ring */
422 
423 	struct i40e_vsi *vsi;		/* Backreference to associated VSI */
424 	struct i40e_q_vector *q_vector;	/* Backreference to associated vector */
425 
426 	struct rcu_head rcu;		/* to avoid race on free */
427 	u16 next_to_alloc;
428 	struct sk_buff *skb;		/* When i40e_clean_rx_ring_irq() must
429 					 * return before it sees the EOP for
430 					 * the current packet, we save that skb
431 					 * here and resume receiving this
432 					 * packet the next time
433 					 * i40e_clean_rx_ring_irq() is called
434 					 * for this ring.
435 					 */
436 
437 	struct i40e_channel *ch;
438 	struct xdp_rxq_info xdp_rxq;
439 } ____cacheline_internodealigned_in_smp;
440 
441 static inline bool ring_uses_build_skb(struct i40e_ring *ring)
442 {
443 	return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
444 }
445 
446 static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
447 {
448 	ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
449 }
450 
451 static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
452 {
453 	ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
454 }
455 
456 static inline bool ring_is_xdp(struct i40e_ring *ring)
457 {
458 	return !!(ring->flags & I40E_TXR_FLAGS_XDP);
459 }
460 
461 static inline void set_ring_xdp(struct i40e_ring *ring)
462 {
463 	ring->flags |= I40E_TXR_FLAGS_XDP;
464 }
465 
466 #define I40E_ITR_ADAPTIVE_MIN_INC	0x0002
467 #define I40E_ITR_ADAPTIVE_MIN_USECS	0x0002
468 #define I40E_ITR_ADAPTIVE_MAX_USECS	0x007e
469 #define I40E_ITR_ADAPTIVE_LATENCY	0x8000
470 #define I40E_ITR_ADAPTIVE_BULK		0x0000
471 #define ITR_IS_BULK(x) (!((x) & I40E_ITR_ADAPTIVE_LATENCY))
472 
473 struct i40e_ring_container {
474 	struct i40e_ring *ring;		/* pointer to linked list of ring(s) */
475 	unsigned long next_update;	/* jiffies value of next update */
476 	unsigned int total_bytes;	/* total bytes processed this int */
477 	unsigned int total_packets;	/* total packets processed this int */
478 	u16 count;
479 	u16 target_itr;			/* target ITR setting for ring(s) */
480 	u16 current_itr;		/* current ITR setting for ring(s) */
481 };
482 
483 /* iterator for handling rings in ring container */
484 #define i40e_for_each_ring(pos, head) \
485 	for (pos = (head).ring; pos != NULL; pos = pos->next)
486 
487 static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
488 {
489 #if (PAGE_SIZE < 8192)
490 	if (ring->rx_buf_len > (PAGE_SIZE / 2))
491 		return 1;
492 #endif
493 	return 0;
494 }
495 
496 #define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
497 
498 bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
499 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
500 void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
501 void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
502 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
503 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
504 void i40e_free_tx_resources(struct i40e_ring *tx_ring);
505 void i40e_free_rx_resources(struct i40e_ring *rx_ring);
506 int i40e_napi_poll(struct napi_struct *napi, int budget);
507 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
508 u32 i40e_get_tx_pending(struct i40e_ring *ring);
509 void i40e_detect_recover_hung(struct i40e_vsi *vsi);
510 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
511 bool __i40e_chk_linearize(struct sk_buff *skb);
512 
513 /**
514  * i40e_get_head - Retrieve head from head writeback
515  * @tx_ring:  tx ring to fetch head of
516  *
517  * Returns value of Tx ring head based on value stored
518  * in head write-back location
519  **/
520 static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
521 {
522 	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
523 
524 	return le32_to_cpu(*(volatile __le32 *)head);
525 }
526 
527 /**
528  * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
529  * @skb:     send buffer
530  * @tx_ring: ring to send buffer on
531  *
532  * Returns number of data descriptors needed for this skb. Returns 0 to indicate
533  * there is not enough descriptors available in this ring since we need at least
534  * one descriptor.
535  **/
536 static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
537 {
538 	const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
539 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
540 	int count = 0, size = skb_headlen(skb);
541 
542 	for (;;) {
543 		count += i40e_txd_use_count(size);
544 
545 		if (!nr_frags--)
546 			break;
547 
548 		size = skb_frag_size(frag++);
549 	}
550 
551 	return count;
552 }
553 
554 /**
555  * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
556  * @tx_ring: the ring to be checked
557  * @size:    the size buffer we want to assure is available
558  *
559  * Returns 0 if stop is not needed
560  **/
561 static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
562 {
563 	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
564 		return 0;
565 	return __i40e_maybe_stop_tx(tx_ring, size);
566 }
567 
568 /**
569  * i40e_chk_linearize - Check if there are more than 8 fragments per packet
570  * @skb:      send buffer
571  * @count:    number of buffers used
572  *
573  * Note: Our HW can't scatter-gather more than 8 fragments to build
574  * a packet on the wire and so we need to figure out the cases where we
575  * need to linearize the skb.
576  **/
577 static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
578 {
579 	/* Both TSO and single send will work if count is less than 8 */
580 	if (likely(count < I40E_MAX_BUFFER_TXD))
581 		return false;
582 
583 	if (skb_is_gso(skb))
584 		return __i40e_chk_linearize(skb);
585 
586 	/* we can support up to 8 data buffers for a single send */
587 	return count != I40E_MAX_BUFFER_TXD;
588 }
589 
590 /**
591  * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
592  * @ring: Tx ring to find the netdev equivalent of
593  **/
594 static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
595 {
596 	return netdev_get_tx_queue(ring->netdev, ring->queue_index);
597 }
598 #endif /* _I40E_TXRX_H_ */
599