xref: /linux/drivers/net/ethernet/intel/ice/ice_txrx.h (revision 1fd1dc41724319406b0aff221a352a400b0ddfc5)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright (c) 2018, Intel Corporation. */
3 
4 #ifndef _ICE_TXRX_H_
5 #define _ICE_TXRX_H_
6 
7 #include <net/libeth/types.h>
8 
9 #include "ice_type.h"
10 
11 #define ICE_DFLT_IRQ_WORK	256
12 #define ICE_RXBUF_3072		3072
13 #define ICE_RXBUF_2048		2048
14 #define ICE_RXBUF_1664		1664
15 #define ICE_RXBUF_1536		1536
16 #define ICE_MAX_CHAINED_RX_BUFS	5
17 #define ICE_MAX_BUF_TXD		8
18 #define ICE_MIN_TX_LEN		17
19 #define ICE_MAX_FRAME_LEGACY_RX 8320
20 
21 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
22  * In order to align with the read requests we will align the value to
23  * the nearest 4K which represents our maximum read request size.
24  */
25 #define ICE_MAX_READ_REQ_SIZE	4096
26 #define ICE_MAX_DATA_PER_TXD	(16 * 1024 - 1)
27 #define ICE_MAX_DATA_PER_TXD_ALIGNED \
28 	(~(ICE_MAX_READ_REQ_SIZE - 1) & ICE_MAX_DATA_PER_TXD)
29 
30 #define ICE_MAX_TXQ_PER_TXQG	128
31 
32 /* We are assuming that the cache line is always 64 Bytes here for ice.
33  * In order to make sure that is a correct assumption there is a check in probe
34  * to print a warning if the read from GLPCI_CNF2 tells us that the cache line
35  * size is 128 bytes. We do it this way because we do not want to read the
36  * GLPCI_CNF2 register or a variable containing the value on every pass through
37  * the Tx path.
38  */
39 #define ICE_CACHE_LINE_BYTES		64
40 #define ICE_DESCS_PER_CACHE_LINE	(ICE_CACHE_LINE_BYTES / \
41 					 sizeof(struct ice_tx_desc))
42 #define ICE_DESCS_FOR_CTX_DESC		1
43 #define ICE_DESCS_FOR_SKB_DATA_PTR	1
44 /* Tx descriptors needed, worst case */
45 #define DESC_NEEDED (MAX_SKB_FRAGS + ICE_DESCS_FOR_CTX_DESC + \
46 		     ICE_DESCS_PER_CACHE_LINE + ICE_DESCS_FOR_SKB_DATA_PTR)
47 #define ICE_DESC_UNUSED(R)	\
48 	(u16)((((R)->next_to_clean > (R)->next_to_use) ? 0 : (R)->count) + \
49 	      (R)->next_to_clean - (R)->next_to_use - 1)
50 
51 #define ICE_RING_QUARTER(R) ((R)->count >> 2)
52 
53 #define ICE_TX_FLAGS_TSO	BIT(0)
54 #define ICE_TX_FLAGS_HW_VLAN	BIT(1)
55 #define ICE_TX_FLAGS_SW_VLAN	BIT(2)
56 /* Free, was ICE_TX_FLAGS_DUMMY_PKT */
57 #define ICE_TX_FLAGS_TSYN	BIT(4)
58 #define ICE_TX_FLAGS_IPV4	BIT(5)
59 #define ICE_TX_FLAGS_IPV6	BIT(6)
60 #define ICE_TX_FLAGS_TUNNEL	BIT(7)
61 #define ICE_TX_FLAGS_HW_OUTER_SINGLE_VLAN	BIT(8)
62 
63 #define ICE_XDP_PASS		0
64 #define ICE_XDP_CONSUMED	BIT(0)
65 #define ICE_XDP_TX		BIT(1)
66 #define ICE_XDP_REDIR		BIT(2)
67 #define ICE_XDP_EXIT		BIT(3)
68 #define ICE_SKB_CONSUMED	ICE_XDP_CONSUMED
69 
70 #define ICE_RX_DMA_ATTR \
71 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
72 
73 #define ICE_ETH_PKT_HDR_PAD	(ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
74 
75 #define ICE_TXD_LAST_DESC_CMD (ICE_TX_DESC_CMD_EOP | ICE_TX_DESC_CMD_RS)
76 
77 /**
78  * enum ice_tx_buf_type - type of &ice_tx_buf to act on Tx completion
79  * @ICE_TX_BUF_EMPTY: unused OR XSk frame, no action required
80  * @ICE_TX_BUF_DUMMY: dummy Flow Director packet, unmap and kfree()
81  * @ICE_TX_BUF_FRAG: mapped skb OR &xdp_buff frag, only unmap DMA
82  * @ICE_TX_BUF_SKB: &sk_buff, unmap and consume_skb(), update stats
83  * @ICE_TX_BUF_XDP_TX: &xdp_buff, unmap and page_frag_free(), stats
84  * @ICE_TX_BUF_XDP_XMIT: &xdp_frame, unmap and xdp_return_frame(), stats
85  * @ICE_TX_BUF_XSK_TX: &xdp_buff on XSk queue, xsk_buff_free(), stats
86  */
87 enum ice_tx_buf_type {
88 	ICE_TX_BUF_EMPTY	= 0U,
89 	ICE_TX_BUF_DUMMY,
90 	ICE_TX_BUF_FRAG,
91 	ICE_TX_BUF_SKB,
92 	ICE_TX_BUF_XDP_TX,
93 	ICE_TX_BUF_XDP_XMIT,
94 	ICE_TX_BUF_XSK_TX,
95 };
96 
97 struct ice_tx_buf {
98 	union {
99 		struct ice_tx_desc *next_to_watch;
100 		u32 rs_idx;
101 	};
102 	union {
103 		void *raw_buf;		/* used for XDP_TX and FDir rules */
104 		struct sk_buff *skb;	/* used for .ndo_start_xmit() */
105 		struct xdp_frame *xdpf;	/* used for .ndo_xdp_xmit() */
106 		struct xdp_buff *xdp;	/* used for XDP_TX ZC */
107 	};
108 	unsigned int bytecount;
109 	union {
110 		unsigned int gso_segs;
111 		unsigned int nr_frags;	/* used for mbuf XDP */
112 	};
113 	u32 tx_flags:12;
114 	u32 type:4;			/* &ice_tx_buf_type */
115 	u32 vid:16;
116 	DEFINE_DMA_UNMAP_LEN(len);
117 	DEFINE_DMA_UNMAP_ADDR(dma);
118 };
119 
120 struct ice_tx_offload_params {
121 	u64 cd_qw1;
122 	struct ice_tx_ring *tx_ring;
123 	u32 td_cmd;
124 	u32 td_offset;
125 	u32 td_l2tag1;
126 	u32 cd_tunnel_params;
127 	u16 cd_l2tag2;
128 	u16 cd_gcs_params;
129 	u8 header_len;
130 };
131 
132 struct ice_ring_stats {
133 	struct rcu_head rcu;	/* to avoid race on free */
134 	struct u64_stats_sync syncp;
135 	struct_group(stats,
136 		u64_stats_t pkts;
137 		u64_stats_t bytes;
138 		union {
139 			struct_group(tx,
140 				u64_stats_t tx_restart_q;
141 				u64_stats_t tx_busy;
142 				u64_stats_t tx_linearize;
143 				/* negative if no pending Tx descriptors */
144 				int prev_pkt;
145 			);
146 			struct_group(rx,
147 				u64_stats_t rx_non_eop_descs;
148 				u64_stats_t rx_page_failed;
149 				u64_stats_t rx_buf_failed;
150 			);
151 		};
152 	);
153 };
154 
155 /**
156  * ice_stats_read - Read a single ring stat value
157  * @stats: pointer to ring_stats structure for a queue
158  * @member: the ice_ring_stats member to read
159  *
160  * Shorthand for reading a single 64-bit stat value from struct
161  * ice_ring_stats.
162  *
163  * Return: the value of the requested stat.
164  */
165 #define ice_stats_read(stats, member) ({				\
166 	struct ice_ring_stats *__stats = (stats);			\
167 	unsigned int start;						\
168 	u64 val;							\
169 	do {								\
170 		start = u64_stats_fetch_begin(&__stats->syncp);		\
171 		val = u64_stats_read(&__stats->member);			\
172 	} while (u64_stats_fetch_retry(&__stats->syncp, start));	\
173 	val;								\
174 })
175 
176 /**
177  * ice_stats_inc - Increment a single ring stat value
178  * @stats: pointer to the ring_stats structure for a queue
179  * @member: the ice_ring_stats member to increment
180  *
181  * Shorthand for incrementing a single 64-bit stat value in struct
182  * ice_ring_stats.
183  */
184 #define ice_stats_inc(stats, member) do {				\
185 	struct ice_ring_stats *__stats = (stats);			\
186 	u64_stats_update_begin(&__stats->syncp);			\
187 	u64_stats_inc(&__stats->member);				\
188 	u64_stats_update_end(&__stats->syncp);				\
189 } while (0)
190 
191 enum ice_ring_state_t {
192 	ICE_TX_XPS_INIT_DONE,
193 	ICE_TX_NBITS,
194 };
195 
196 /* this enum matches hardware bits and is meant to be used by DYN_CTLN
197  * registers and QINT registers or more generally anywhere in the manual
198  * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
199  * register but instead is a special value meaning "don't update" ITR0/1/2.
200  */
201 enum ice_dyn_idx_t {
202 	ICE_IDX_ITR0 = 0,
203 	ICE_IDX_ITR1 = 1,
204 	ICE_IDX_ITR2 = 2,
205 	ICE_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
206 };
207 
208 /* Header split modes defined by DTYPE field of Rx RLAN context */
209 enum ice_rx_dtype {
210 	ICE_RX_DTYPE_NO_SPLIT		= 0,
211 	ICE_RX_DTYPE_HEADER_SPLIT	= 1,
212 	ICE_RX_DTYPE_SPLIT_ALWAYS	= 2,
213 };
214 
215 struct ice_pkt_ctx {
216 	u64 cached_phctime;
217 	__be16 vlan_proto;
218 };
219 
220 /* indices into GLINT_ITR registers */
221 #define ICE_RX_ITR	ICE_IDX_ITR0
222 #define ICE_TX_ITR	ICE_IDX_ITR1
223 #define ICE_ITR_8K	124
224 #define ICE_ITR_20K	50
225 #define ICE_ITR_MAX	8160 /* 0x1FE0 */
226 #define ICE_DFLT_TX_ITR	ICE_ITR_20K
227 #define ICE_DFLT_RX_ITR	ICE_ITR_20K
228 enum ice_dynamic_itr {
229 	ITR_STATIC = 0,
230 	ITR_DYNAMIC = 1
231 };
232 
233 #define ITR_IS_DYNAMIC(rc) ((rc)->itr_mode == ITR_DYNAMIC)
234 #define ICE_ITR_GRAN_S		1	/* ITR granularity is always 2us */
235 #define ICE_ITR_GRAN_US		BIT(ICE_ITR_GRAN_S)
236 #define ICE_ITR_MASK		0x1FFE	/* ITR register value alignment mask */
237 #define ITR_REG_ALIGN(setting)	((setting) & ICE_ITR_MASK)
238 
239 #define ICE_DFLT_INTRL	0
240 #define ICE_MAX_INTRL	236
241 
242 #define ICE_IN_WB_ON_ITR_MODE	255
243 /* Sets WB_ON_ITR and assumes INTENA bit is already cleared, which allows
244  * setting the MSK_M bit to tell hardware to ignore the INTENA_M bit. Also,
245  * set the write-back latency to the usecs passed in.
246  */
247 #define ICE_GLINT_DYN_CTL_WB_ON_ITR(usecs, itr_idx)	\
248 	((((usecs) << (GLINT_DYN_CTL_INTERVAL_S - ICE_ITR_GRAN_S)) & \
249 	  GLINT_DYN_CTL_INTERVAL_M) | \
250 	 (((itr_idx) << GLINT_DYN_CTL_ITR_INDX_S) & \
251 	  GLINT_DYN_CTL_ITR_INDX_M) | GLINT_DYN_CTL_INTENA_MSK_M | \
252 	 GLINT_DYN_CTL_WB_ON_ITR_M)
253 
254 /* Legacy or Advanced Mode Queue */
255 #define ICE_TX_ADVANCED	0
256 #define ICE_TX_LEGACY	1
257 
258 /* descriptor ring, associated with a VSI */
259 struct ice_tstamp_ring {
260 	struct ice_tx_ring *tx_ring;	/* Backreference to associated Tx ring */
261 	dma_addr_t dma;			/* physical address of ring */
262 	struct rcu_head rcu;            /* to avoid race on free */
263 	u8 __iomem *tail;
264 	void *desc;
265 	u16 next_to_use;
266 	u16 count;
267 } ____cacheline_internodealigned_in_smp;
268 
269 struct ice_rx_ring {
270 	__cacheline_group_begin_aligned(read_mostly);
271 	void *desc;			/* Descriptor ring memory */
272 	struct page_pool *pp;
273 	struct net_device *netdev;	/* netdev ring maps to */
274 	struct ice_q_vector *q_vector;	/* Backreference to associated vector */
275 	u8 __iomem *tail;
276 
277 	union {
278 		struct libeth_fqe *rx_fqes;
279 		struct xdp_buff **xdp_buf;
280 	};
281 
282 	u16 count;			/* Number of descriptors */
283 	u8 ptp_rx;
284 
285 	u8 flags;
286 #define ICE_RX_FLAGS_CRC_STRIP_DIS	BIT(2)
287 #define ICE_RX_FLAGS_MULTIDEV		BIT(3)
288 #define ICE_RX_FLAGS_RING_GCS		BIT(4)
289 
290 	u32 truesize;
291 
292 	struct page_pool *hdr_pp;
293 	struct libeth_fqe *hdr_fqes;
294 
295 	struct bpf_prog *xdp_prog;
296 	struct ice_tx_ring *xdp_ring;
297 	struct xsk_buff_pool *xsk_pool;
298 
299 	/* stats structs */
300 	struct ice_ring_stats *ring_stats;
301 	struct ice_rx_ring *next;	/* pointer to next ring in q_vector */
302 
303 	u32 hdr_truesize;
304 
305 	struct xdp_rxq_info xdp_rxq;
306 	__cacheline_group_end_aligned(read_mostly);
307 
308 	__cacheline_group_begin_aligned(read_write);
309 	union {
310 		struct libeth_xdp_buff_stash xdp;
311 		struct libeth_xdp_buff *xsk;
312 	};
313 	union {
314 		struct ice_pkt_ctx pkt_ctx;
315 		struct {
316 			u64 cached_phctime;
317 			__be16 vlan_proto;
318 		};
319 	};
320 
321 	/* used in interrupt processing */
322 	u16 next_to_use;
323 	u16 next_to_clean;
324 	__cacheline_group_end_aligned(read_write);
325 
326 	__cacheline_group_begin_aligned(cold);
327 	struct rcu_head rcu;		/* to avoid race on free */
328 	struct ice_vsi *vsi;		/* Backreference to associated VSI */
329 	struct ice_channel *ch;
330 
331 	dma_addr_t dma;			/* physical address of ring */
332 	u16 q_index;			/* Queue number of ring */
333 	u16 reg_idx;			/* HW register index of the ring */
334 	u8 dcb_tc;			/* Traffic class of ring */
335 
336 	u16 rx_hdr_len;
337 	u16 rx_buf_len;
338 	__cacheline_group_end_aligned(cold);
339 } ____cacheline_internodealigned_in_smp;
340 
341 struct ice_tx_ring {
342 	__cacheline_group_begin_aligned(read_mostly);
343 	void *desc;			/* Descriptor ring memory */
344 	struct device *dev;		/* Used for DMA mapping */
345 	u8 __iomem *tail;
346 	struct ice_tx_buf *tx_buf;
347 
348 	struct ice_q_vector *q_vector;	/* Backreference to associated vector */
349 	struct net_device *netdev;	/* netdev ring maps to */
350 	struct ice_vsi *vsi;		/* Backreference to associated VSI */
351 
352 	u16 count;			/* Number of descriptors */
353 	u16 q_index;			/* Queue number of ring */
354 
355 	u8 flags;
356 #define ICE_TX_FLAGS_RING_XDP		BIT(0)
357 #define ICE_TX_FLAGS_RING_VLAN_L2TAG1	BIT(1)
358 #define ICE_TX_FLAGS_RING_VLAN_L2TAG2	BIT(2)
359 #define ICE_TX_FLAGS_TXTIME		BIT(3)
360 
361 	struct xsk_buff_pool *xsk_pool;
362 
363 	/* stats structs */
364 	struct ice_ring_stats *ring_stats;
365 	struct ice_tx_ring *next;	/* pointer to next ring in q_vector */
366 
367 	struct ice_tstamp_ring *tstamp_ring;
368 	struct ice_ptp_tx *tx_tstamps;
369 	__cacheline_group_end_aligned(read_mostly);
370 
371 	__cacheline_group_begin_aligned(read_write);
372 	u16 next_to_use;
373 	u16 next_to_clean;
374 
375 	u16 xdp_tx_active;
376 	spinlock_t tx_lock;
377 	__cacheline_group_end_aligned(read_write);
378 
379 	__cacheline_group_begin_aligned(cold);
380 	struct rcu_head rcu;		/* to avoid race on free */
381 	DECLARE_BITMAP(xps_state, ICE_TX_NBITS);	/* XPS Config State */
382 	struct ice_channel *ch;
383 
384 	dma_addr_t dma;			/* physical address of ring */
385 	u16 q_handle;			/* Queue handle per TC */
386 	u16 reg_idx;			/* HW register index of the ring */
387 	u8 dcb_tc;			/* Traffic class of ring */
388 
389 	u16 quanta_prof_id;
390 	u32 txq_teid;			/* Added Tx queue TEID */
391 	__cacheline_group_end_aligned(cold);
392 } ____cacheline_internodealigned_in_smp;
393 
394 static inline bool ice_ring_ch_enabled(struct ice_tx_ring *ring)
395 {
396 	return !!ring->ch;
397 }
398 
399 static inline bool ice_ring_is_xdp(struct ice_tx_ring *ring)
400 {
401 	return !!(ring->flags & ICE_TX_FLAGS_RING_XDP);
402 }
403 
404 enum ice_container_type {
405 	ICE_RX_CONTAINER,
406 	ICE_TX_CONTAINER,
407 };
408 
409 struct ice_ring_container {
410 	/* head of linked-list of rings */
411 	union {
412 		struct ice_rx_ring *rx_ring;
413 		struct ice_tx_ring *tx_ring;
414 	};
415 	struct dim dim;		/* data for net_dim algorithm */
416 	u16 itr_idx;		/* index in the interrupt vector */
417 	/* this matches the maximum number of ITR bits, but in usec
418 	 * values, so it is shifted left one bit (bit zero is ignored)
419 	 */
420 	union {
421 		struct {
422 			u16 itr_setting:13;
423 			u16 itr_reserved:2;
424 			u16 itr_mode:1;
425 		};
426 		u16 itr_settings;
427 	};
428 	enum ice_container_type type;
429 };
430 
431 struct ice_coalesce_stored {
432 	u16 itr_tx;
433 	u16 itr_rx;
434 	u8 intrl;
435 	u8 tx_valid;
436 	u8 rx_valid;
437 };
438 
439 /* iterator for handling rings in ring container */
440 #define ice_for_each_rx_ring(pos, head) \
441 	for (pos = (head).rx_ring; pos; pos = pos->next)
442 
443 #define ice_for_each_tx_ring(pos, head) \
444 	for (pos = (head).tx_ring; pos; pos = pos->next)
445 
446 static inline unsigned int ice_rx_pg_order(struct ice_rx_ring *ring)
447 {
448 	return 0;
449 }
450 
451 union ice_32b_rx_flex_desc;
452 
453 void ice_init_ctrl_rx_descs(struct ice_rx_ring *rx_ring, u32 num_descs);
454 void ice_rxq_pp_destroy(struct ice_rx_ring *rq);
455 bool ice_alloc_rx_bufs(struct ice_rx_ring *rxr, unsigned int cleaned_count);
456 netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev);
457 u16
458 ice_select_queue(struct net_device *dev, struct sk_buff *skb,
459 		 struct net_device *sb_dev);
460 void ice_clean_tx_ring(struct ice_tx_ring *tx_ring);
461 void ice_clean_rx_ring(struct ice_rx_ring *rx_ring);
462 int ice_setup_tx_ring(struct ice_tx_ring *tx_ring);
463 int ice_setup_rx_ring(struct ice_rx_ring *rx_ring);
464 int ice_alloc_setup_tstamp_ring(struct ice_tx_ring *tx_ring);
465 void ice_free_tx_ring(struct ice_tx_ring *tx_ring);
466 void ice_free_rx_ring(struct ice_rx_ring *rx_ring);
467 int ice_napi_poll(struct napi_struct *napi, int budget);
468 int
469 ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc,
470 		   u8 *raw_packet);
471 void ice_clean_ctrl_tx_irq(struct ice_tx_ring *tx_ring);
472 void ice_clean_ctrl_rx_irq(struct ice_rx_ring *rx_ring);
473 void ice_free_tx_tstamp_ring(struct ice_tx_ring *tx_ring);
474 void ice_free_tstamp_ring(struct ice_tx_ring *tx_ring);
475 #endif /* _ICE_TXRX_H_ */
476