xref: /openbmc/linux/drivers/net/ethernet/intel/i40e/i40e_txrx.h (revision 4f139972b489f8bc2c821aa25ac65018d92af3f7)
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 /**
56  * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
57  * @intrl: interrupt rate limit to convert
58  *
59  * This function converts a decimal interrupt rate limit to the appropriate
60  * register format expected by the firmware when setting interrupt rate limit.
61  */
62 static inline u16 i40e_intrl_usec_to_reg(int intrl)
63 {
64 	if (intrl >> 2)
65 		return ((intrl >> 2) | INTRL_ENA);
66 	else
67 		return 0;
68 }
69 #define I40E_INTRL_8K              125     /* 8000 ints/sec */
70 #define I40E_INTRL_62K             16      /* 62500 ints/sec */
71 #define I40E_INTRL_83K             12      /* 83333 ints/sec */
72 
73 #define I40E_QUEUE_END_OF_LIST 0x7FF
74 
75 /* this enum matches hardware bits and is meant to be used by DYN_CTLN
76  * registers and QINT registers or more generally anywhere in the manual
77  * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
78  * register but instead is a special value meaning "don't update" ITR0/1/2.
79  */
80 enum i40e_dyn_idx_t {
81 	I40E_IDX_ITR0 = 0,
82 	I40E_IDX_ITR1 = 1,
83 	I40E_IDX_ITR2 = 2,
84 	I40E_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
85 };
86 
87 /* these are indexes into ITRN registers */
88 #define I40E_RX_ITR    I40E_IDX_ITR0
89 #define I40E_TX_ITR    I40E_IDX_ITR1
90 #define I40E_PE_ITR    I40E_IDX_ITR2
91 
92 /* Supported RSS offloads */
93 #define I40E_DEFAULT_RSS_HENA ( \
94 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
95 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
96 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
97 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
98 	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
99 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
100 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
101 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
102 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
103 	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
104 	BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
105 
106 #define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
107 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
108 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
109 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
110 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
111 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
112 	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
113 
114 #define i40e_pf_get_default_rss_hena(pf) \
115 	(((pf)->flags & I40E_FLAG_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
116 	  I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
117 
118 /* Supported Rx Buffer Sizes (a multiple of 128) */
119 #define I40E_RXBUFFER_256   256
120 #define I40E_RXBUFFER_1536  1536  /* 128B aligned standard Ethernet frame */
121 #define I40E_RXBUFFER_2048  2048
122 #define I40E_MAX_RXBUFFER   9728  /* largest size for single descriptor */
123 
124 /* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
125  * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
126  * this adds up to 512 bytes of extra data meaning the smallest allocation
127  * we could have is 1K.
128  * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
129  * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
130  */
131 #define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
132 #define i40e_rx_desc i40e_32byte_rx_desc
133 
134 #define I40E_RX_DMA_ATTR \
135 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
136 
137 /**
138  * i40e_test_staterr - tests bits in Rx descriptor status and error fields
139  * @rx_desc: pointer to receive descriptor (in le64 format)
140  * @stat_err_bits: value to mask
141  *
142  * This function does some fast chicanery in order to return the
143  * value of the mask which is really only used for boolean tests.
144  * The status_error_len doesn't need to be shifted because it begins
145  * at offset zero.
146  */
147 static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
148 				     const u64 stat_err_bits)
149 {
150 	return !!(rx_desc->wb.qword1.status_error_len &
151 		  cpu_to_le64(stat_err_bits));
152 }
153 
154 /* How many Rx Buffers do we bundle into one write to the hardware ? */
155 #define I40E_RX_BUFFER_WRITE	16	/* Must be power of 2 */
156 #define I40E_RX_INCREMENT(r, i) \
157 	do {					\
158 		(i)++;				\
159 		if ((i) == (r)->count)		\
160 			i = 0;			\
161 		r->next_to_clean = i;		\
162 	} while (0)
163 
164 #define I40E_RX_NEXT_DESC(r, i, n)		\
165 	do {					\
166 		(i)++;				\
167 		if ((i) == (r)->count)		\
168 			i = 0;			\
169 		(n) = I40E_RX_DESC((r), (i));	\
170 	} while (0)
171 
172 #define I40E_RX_NEXT_DESC_PREFETCH(r, i, n)		\
173 	do {						\
174 		I40E_RX_NEXT_DESC((r), (i), (n));	\
175 		prefetch((n));				\
176 	} while (0)
177 
178 #define I40E_MAX_BUFFER_TXD	8
179 #define I40E_MIN_TX_LEN		17
180 
181 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
182  * In order to align with the read requests we will align the value to
183  * the nearest 4K which represents our maximum read request size.
184  */
185 #define I40E_MAX_READ_REQ_SIZE		4096
186 #define I40E_MAX_DATA_PER_TXD		(16 * 1024 - 1)
187 #define I40E_MAX_DATA_PER_TXD_ALIGNED \
188 	(I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
189 
190 /**
191  * i40e_txd_use_count  - estimate the number of descriptors needed for Tx
192  * @size: transmit request size in bytes
193  *
194  * Due to hardware alignment restrictions (4K alignment), we need to
195  * assume that we can have no more than 12K of data per descriptor, even
196  * though each descriptor can take up to 16K - 1 bytes of aligned memory.
197  * Thus, we need to divide by 12K. But division is slow! Instead,
198  * we decompose the operation into shifts and one relatively cheap
199  * multiply operation.
200  *
201  * To divide by 12K, we first divide by 4K, then divide by 3:
202  *     To divide by 4K, shift right by 12 bits
203  *     To divide by 3, multiply by 85, then divide by 256
204  *     (Divide by 256 is done by shifting right by 8 bits)
205  * Finally, we add one to round up. Because 256 isn't an exact multiple of
206  * 3, we'll underestimate near each multiple of 12K. This is actually more
207  * accurate as we have 4K - 1 of wiggle room that we can fit into the last
208  * segment.  For our purposes this is accurate out to 1M which is orders of
209  * magnitude greater than our largest possible GSO size.
210  *
211  * This would then be implemented as:
212  *     return (((size >> 12) * 85) >> 8) + 1;
213  *
214  * Since multiplication and division are commutative, we can reorder
215  * operations into:
216  *     return ((size * 85) >> 20) + 1;
217  */
218 static inline unsigned int i40e_txd_use_count(unsigned int size)
219 {
220 	return ((size * 85) >> 20) + 1;
221 }
222 
223 /* Tx Descriptors needed, worst case */
224 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
225 #define I40E_MIN_DESC_PENDING	4
226 
227 #define I40E_TX_FLAGS_HW_VLAN		BIT(1)
228 #define I40E_TX_FLAGS_SW_VLAN		BIT(2)
229 #define I40E_TX_FLAGS_TSO		BIT(3)
230 #define I40E_TX_FLAGS_IPV4		BIT(4)
231 #define I40E_TX_FLAGS_IPV6		BIT(5)
232 #define I40E_TX_FLAGS_FCCRC		BIT(6)
233 #define I40E_TX_FLAGS_FSO		BIT(7)
234 #define I40E_TX_FLAGS_TSYN		BIT(8)
235 #define I40E_TX_FLAGS_FD_SB		BIT(9)
236 #define I40E_TX_FLAGS_UDP_TUNNEL	BIT(10)
237 #define I40E_TX_FLAGS_VLAN_MASK		0xffff0000
238 #define I40E_TX_FLAGS_VLAN_PRIO_MASK	0xe0000000
239 #define I40E_TX_FLAGS_VLAN_PRIO_SHIFT	29
240 #define I40E_TX_FLAGS_VLAN_SHIFT	16
241 
242 struct i40e_tx_buffer {
243 	struct i40e_tx_desc *next_to_watch;
244 	union {
245 		struct sk_buff *skb;
246 		void *raw_buf;
247 	};
248 	unsigned int bytecount;
249 	unsigned short gso_segs;
250 
251 	DEFINE_DMA_UNMAP_ADDR(dma);
252 	DEFINE_DMA_UNMAP_LEN(len);
253 	u32 tx_flags;
254 };
255 
256 struct i40e_rx_buffer {
257 	dma_addr_t dma;
258 	struct page *page;
259 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
260 	__u32 page_offset;
261 #else
262 	__u16 page_offset;
263 #endif
264 	__u16 pagecnt_bias;
265 };
266 
267 struct i40e_queue_stats {
268 	u64 packets;
269 	u64 bytes;
270 };
271 
272 struct i40e_tx_queue_stats {
273 	u64 restart_queue;
274 	u64 tx_busy;
275 	u64 tx_done_old;
276 	u64 tx_linearize;
277 	u64 tx_force_wb;
278 	u64 tx_lost_interrupt;
279 };
280 
281 struct i40e_rx_queue_stats {
282 	u64 non_eop_descs;
283 	u64 alloc_page_failed;
284 	u64 alloc_buff_failed;
285 	u64 page_reuse_count;
286 	u64 realloc_count;
287 };
288 
289 enum i40e_ring_state_t {
290 	__I40E_TX_FDIR_INIT_DONE,
291 	__I40E_TX_XPS_INIT_DONE,
292 };
293 
294 /* some useful defines for virtchannel interface, which
295  * is the only remaining user of header split
296  */
297 #define I40E_RX_DTYPE_NO_SPLIT      0
298 #define I40E_RX_DTYPE_HEADER_SPLIT  1
299 #define I40E_RX_DTYPE_SPLIT_ALWAYS  2
300 #define I40E_RX_SPLIT_L2      0x1
301 #define I40E_RX_SPLIT_IP      0x2
302 #define I40E_RX_SPLIT_TCP_UDP 0x4
303 #define I40E_RX_SPLIT_SCTP    0x8
304 
305 /* struct that defines a descriptor ring, associated with a VSI */
306 struct i40e_ring {
307 	struct i40e_ring *next;		/* pointer to next ring in q_vector */
308 	void *desc;			/* Descriptor ring memory */
309 	struct device *dev;		/* Used for DMA mapping */
310 	struct net_device *netdev;	/* netdev ring maps to */
311 	union {
312 		struct i40e_tx_buffer *tx_bi;
313 		struct i40e_rx_buffer *rx_bi;
314 	};
315 	unsigned long state;
316 	u16 queue_index;		/* Queue number of ring */
317 	u8 dcb_tc;			/* Traffic class of ring */
318 	u8 __iomem *tail;
319 
320 	/* high bit set means dynamic, use accessor routines to read/write.
321 	 * hardware only supports 2us resolution for the ITR registers.
322 	 * these values always store the USER setting, and must be converted
323 	 * before programming to a register.
324 	 */
325 	u16 rx_itr_setting;
326 	u16 tx_itr_setting;
327 
328 	u16 count;			/* Number of descriptors */
329 	u16 reg_idx;			/* HW register index of the ring */
330 	u16 rx_buf_len;
331 
332 	/* used in interrupt processing */
333 	u16 next_to_use;
334 	u16 next_to_clean;
335 
336 	u8 atr_sample_rate;
337 	u8 atr_count;
338 
339 	bool ring_active;		/* is ring online or not */
340 	bool arm_wb;		/* do something to arm write back */
341 	u8 packet_stride;
342 
343 	u16 flags;
344 #define I40E_TXR_FLAGS_WB_ON_ITR	BIT(0)
345 
346 	/* stats structs */
347 	struct i40e_queue_stats	stats;
348 	struct u64_stats_sync syncp;
349 	union {
350 		struct i40e_tx_queue_stats tx_stats;
351 		struct i40e_rx_queue_stats rx_stats;
352 	};
353 
354 	unsigned int size;		/* length of descriptor ring in bytes */
355 	dma_addr_t dma;			/* physical address of ring */
356 
357 	struct i40e_vsi *vsi;		/* Backreference to associated VSI */
358 	struct i40e_q_vector *q_vector;	/* Backreference to associated vector */
359 
360 	struct rcu_head rcu;		/* to avoid race on free */
361 	u16 next_to_alloc;
362 	struct sk_buff *skb;		/* When i40e_clean_rx_ring_irq() must
363 					 * return before it sees the EOP for
364 					 * the current packet, we save that skb
365 					 * here and resume receiving this
366 					 * packet the next time
367 					 * i40e_clean_rx_ring_irq() is called
368 					 * for this ring.
369 					 */
370 } ____cacheline_internodealigned_in_smp;
371 
372 enum i40e_latency_range {
373 	I40E_LOWEST_LATENCY = 0,
374 	I40E_LOW_LATENCY = 1,
375 	I40E_BULK_LATENCY = 2,
376 	I40E_ULTRA_LATENCY = 3,
377 };
378 
379 struct i40e_ring_container {
380 	/* array of pointers to rings */
381 	struct i40e_ring *ring;
382 	unsigned int total_bytes;	/* total bytes processed this int */
383 	unsigned int total_packets;	/* total packets processed this int */
384 	u16 count;
385 	enum i40e_latency_range latency_range;
386 	u16 itr;
387 };
388 
389 /* iterator for handling rings in ring container */
390 #define i40e_for_each_ring(pos, head) \
391 	for (pos = (head).ring; pos != NULL; pos = pos->next)
392 
393 bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
394 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
395 void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
396 void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
397 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
398 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
399 void i40e_free_tx_resources(struct i40e_ring *tx_ring);
400 void i40e_free_rx_resources(struct i40e_ring *rx_ring);
401 int i40e_napi_poll(struct napi_struct *napi, int budget);
402 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
403 u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
404 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
405 bool __i40e_chk_linearize(struct sk_buff *skb);
406 
407 /**
408  * i40e_get_head - Retrieve head from head writeback
409  * @tx_ring:  tx ring to fetch head of
410  *
411  * Returns value of Tx ring head based on value stored
412  * in head write-back location
413  **/
414 static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
415 {
416 	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
417 
418 	return le32_to_cpu(*(volatile __le32 *)head);
419 }
420 
421 /**
422  * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
423  * @skb:     send buffer
424  * @tx_ring: ring to send buffer on
425  *
426  * Returns number of data descriptors needed for this skb. Returns 0 to indicate
427  * there is not enough descriptors available in this ring since we need at least
428  * one descriptor.
429  **/
430 static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
431 {
432 	const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
433 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
434 	int count = 0, size = skb_headlen(skb);
435 
436 	for (;;) {
437 		count += i40e_txd_use_count(size);
438 
439 		if (!nr_frags--)
440 			break;
441 
442 		size = skb_frag_size(frag++);
443 	}
444 
445 	return count;
446 }
447 
448 /**
449  * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
450  * @tx_ring: the ring to be checked
451  * @size:    the size buffer we want to assure is available
452  *
453  * Returns 0 if stop is not needed
454  **/
455 static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
456 {
457 	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
458 		return 0;
459 	return __i40e_maybe_stop_tx(tx_ring, size);
460 }
461 
462 /**
463  * i40e_chk_linearize - Check if there are more than 8 fragments per packet
464  * @skb:      send buffer
465  * @count:    number of buffers used
466  *
467  * Note: Our HW can't scatter-gather more than 8 fragments to build
468  * a packet on the wire and so we need to figure out the cases where we
469  * need to linearize the skb.
470  **/
471 static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
472 {
473 	/* Both TSO and single send will work if count is less than 8 */
474 	if (likely(count < I40E_MAX_BUFFER_TXD))
475 		return false;
476 
477 	if (skb_is_gso(skb))
478 		return __i40e_chk_linearize(skb);
479 
480 	/* we can support up to 8 data buffers for a single send */
481 	return count != I40E_MAX_BUFFER_TXD;
482 }
483 
484 /**
485  * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
486  * @ring: Tx ring to find the netdev equivalent of
487  **/
488 static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
489 {
490 	return netdev_get_tx_queue(ring->netdev, ring->queue_index);
491 }
492 #endif /* _I40E_TXRX_H_ */
493