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 "ice_type.h"
8 
9 #define ICE_DFLT_IRQ_WORK	256
10 #define ICE_RXBUF_3072		3072
11 #define ICE_RXBUF_2048		2048
12 #define ICE_RXBUF_1664		1664
13 #define ICE_RXBUF_1536		1536
14 #define ICE_MAX_CHAINED_RX_BUFS	5
15 #define ICE_MAX_BUF_TXD		8
16 #define ICE_MIN_TX_LEN		17
17 #define ICE_MAX_FRAME_LEGACY_RX 8320
18 
19 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
20  * In order to align with the read requests we will align the value to
21  * the nearest 4K which represents our maximum read request size.
22  */
23 #define ICE_MAX_READ_REQ_SIZE	4096
24 #define ICE_MAX_DATA_PER_TXD	(16 * 1024 - 1)
25 #define ICE_MAX_DATA_PER_TXD_ALIGNED \
26 	(~(ICE_MAX_READ_REQ_SIZE - 1) & ICE_MAX_DATA_PER_TXD)
27 
28 #define ICE_MAX_TXQ_PER_TXQG	128
29 
30 /* Attempt to maximize the headroom available for incoming frames. We use a 2K
31  * buffer for MTUs <= 1500 and need 1536/1534 to store the data for the frame.
32  * This leaves us with 512 bytes of room.  From that we need to deduct the
33  * space needed for the shared info and the padding needed to IP align the
34  * frame.
35  *
36  * Note: For cache line sizes 256 or larger this value is going to end
37  *	 up negative.  In these cases we should fall back to the legacy
38  *	 receive path.
39  */
40 #if (PAGE_SIZE < 8192)
41 #define ICE_2K_TOO_SMALL_WITH_PADDING \
42 	((unsigned int)(NET_SKB_PAD + ICE_RXBUF_1536) > \
43 			SKB_WITH_OVERHEAD(ICE_RXBUF_2048))
44 
45 /**
46  * ice_compute_pad - compute the padding
47  * @rx_buf_len: buffer length
48  *
49  * Figure out the size of half page based on given buffer length and
50  * then subtract the skb_shared_info followed by subtraction of the
51  * actual buffer length; this in turn results in the actual space that
52  * is left for padding usage
53  */
54 static inline int ice_compute_pad(int rx_buf_len)
55 {
56 	int half_page_size;
57 
58 	half_page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
59 	return SKB_WITH_OVERHEAD(half_page_size) - rx_buf_len;
60 }
61 
62 /**
63  * ice_skb_pad - determine the padding that we can supply
64  *
65  * Figure out the right Rx buffer size and based on that calculate the
66  * padding
67  */
68 static inline int ice_skb_pad(void)
69 {
70 	int rx_buf_len;
71 
72 	/* If a 2K buffer cannot handle a standard Ethernet frame then
73 	 * optimize padding for a 3K buffer instead of a 1.5K buffer.
74 	 *
75 	 * For a 3K buffer we need to add enough padding to allow for
76 	 * tailroom due to NET_IP_ALIGN possibly shifting us out of
77 	 * cache-line alignment.
78 	 */
79 	if (ICE_2K_TOO_SMALL_WITH_PADDING)
80 		rx_buf_len = ICE_RXBUF_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
81 	else
82 		rx_buf_len = ICE_RXBUF_1536;
83 
84 	/* if needed make room for NET_IP_ALIGN */
85 	rx_buf_len -= NET_IP_ALIGN;
86 
87 	return ice_compute_pad(rx_buf_len);
88 }
89 
90 #define ICE_SKB_PAD ice_skb_pad()
91 #else
92 #define ICE_2K_TOO_SMALL_WITH_PADDING false
93 #define ICE_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
94 #endif
95 
96 /* We are assuming that the cache line is always 64 Bytes here for ice.
97  * In order to make sure that is a correct assumption there is a check in probe
98  * to print a warning if the read from GLPCI_CNF2 tells us that the cache line
99  * size is 128 bytes. We do it this way because we do not want to read the
100  * GLPCI_CNF2 register or a variable containing the value on every pass through
101  * the Tx path.
102  */
103 #define ICE_CACHE_LINE_BYTES		64
104 #define ICE_DESCS_PER_CACHE_LINE	(ICE_CACHE_LINE_BYTES / \
105 					 sizeof(struct ice_tx_desc))
106 #define ICE_DESCS_FOR_CTX_DESC		1
107 #define ICE_DESCS_FOR_SKB_DATA_PTR	1
108 /* Tx descriptors needed, worst case */
109 #define DESC_NEEDED (MAX_SKB_FRAGS + ICE_DESCS_FOR_CTX_DESC + \
110 		     ICE_DESCS_PER_CACHE_LINE + ICE_DESCS_FOR_SKB_DATA_PTR)
111 #define ICE_DESC_UNUSED(R)	\
112 	(u16)((((R)->next_to_clean > (R)->next_to_use) ? 0 : (R)->count) + \
113 	      (R)->next_to_clean - (R)->next_to_use - 1)
114 
115 #define ICE_RING_QUARTER(R) ((R)->count >> 2)
116 
117 #define ICE_TX_FLAGS_TSO	BIT(0)
118 #define ICE_TX_FLAGS_HW_VLAN	BIT(1)
119 #define ICE_TX_FLAGS_SW_VLAN	BIT(2)
120 /* ICE_TX_FLAGS_DUMMY_PKT is used to mark dummy packets that should be
121  * freed instead of returned like skb packets.
122  */
123 #define ICE_TX_FLAGS_DUMMY_PKT	BIT(3)
124 #define ICE_TX_FLAGS_TSYN	BIT(4)
125 #define ICE_TX_FLAGS_IPV4	BIT(5)
126 #define ICE_TX_FLAGS_IPV6	BIT(6)
127 #define ICE_TX_FLAGS_TUNNEL	BIT(7)
128 #define ICE_TX_FLAGS_HW_OUTER_SINGLE_VLAN	BIT(8)
129 #define ICE_TX_FLAGS_VLAN_M	0xffff0000
130 #define ICE_TX_FLAGS_VLAN_PR_M	0xe0000000
131 #define ICE_TX_FLAGS_VLAN_PR_S	29
132 #define ICE_TX_FLAGS_VLAN_S	16
133 
134 #define ICE_XDP_PASS		0
135 #define ICE_XDP_CONSUMED	BIT(0)
136 #define ICE_XDP_TX		BIT(1)
137 #define ICE_XDP_REDIR		BIT(2)
138 #define ICE_XDP_EXIT		BIT(3)
139 
140 #define ICE_RX_DMA_ATTR \
141 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
142 
143 #define ICE_ETH_PKT_HDR_PAD	(ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
144 
145 #define ICE_TXD_LAST_DESC_CMD (ICE_TX_DESC_CMD_EOP | ICE_TX_DESC_CMD_RS)
146 
147 struct ice_tx_buf {
148 	struct ice_tx_desc *next_to_watch;
149 	union {
150 		struct sk_buff *skb;
151 		void *raw_buf; /* used for XDP */
152 	};
153 	unsigned int bytecount;
154 	unsigned short gso_segs;
155 	u32 tx_flags;
156 	DEFINE_DMA_UNMAP_LEN(len);
157 	DEFINE_DMA_UNMAP_ADDR(dma);
158 };
159 
160 struct ice_tx_offload_params {
161 	u64 cd_qw1;
162 	struct ice_tx_ring *tx_ring;
163 	u32 td_cmd;
164 	u32 td_offset;
165 	u32 td_l2tag1;
166 	u32 cd_tunnel_params;
167 	u16 cd_l2tag2;
168 	u8 header_len;
169 };
170 
171 struct ice_rx_buf {
172 	dma_addr_t dma;
173 	struct page *page;
174 	unsigned int page_offset;
175 	u16 pagecnt_bias;
176 };
177 
178 struct ice_q_stats {
179 	u64 pkts;
180 	u64 bytes;
181 };
182 
183 struct ice_txq_stats {
184 	u64 restart_q;
185 	u64 tx_busy;
186 	u64 tx_linearize;
187 	int prev_pkt; /* negative if no pending Tx descriptors */
188 };
189 
190 struct ice_rxq_stats {
191 	u64 non_eop_descs;
192 	u64 alloc_page_failed;
193 	u64 alloc_buf_failed;
194 };
195 
196 struct ice_ring_stats {
197 	struct rcu_head rcu;	/* to avoid race on free */
198 	struct ice_q_stats stats;
199 	struct u64_stats_sync syncp;
200 	union {
201 		struct ice_txq_stats tx_stats;
202 		struct ice_rxq_stats rx_stats;
203 	};
204 };
205 
206 enum ice_ring_state_t {
207 	ICE_TX_XPS_INIT_DONE,
208 	ICE_TX_NBITS,
209 };
210 
211 /* this enum matches hardware bits and is meant to be used by DYN_CTLN
212  * registers and QINT registers or more generally anywhere in the manual
213  * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
214  * register but instead is a special value meaning "don't update" ITR0/1/2.
215  */
216 enum ice_dyn_idx_t {
217 	ICE_IDX_ITR0 = 0,
218 	ICE_IDX_ITR1 = 1,
219 	ICE_IDX_ITR2 = 2,
220 	ICE_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
221 };
222 
223 /* Header split modes defined by DTYPE field of Rx RLAN context */
224 enum ice_rx_dtype {
225 	ICE_RX_DTYPE_NO_SPLIT		= 0,
226 	ICE_RX_DTYPE_HEADER_SPLIT	= 1,
227 	ICE_RX_DTYPE_SPLIT_ALWAYS	= 2,
228 };
229 
230 /* indices into GLINT_ITR registers */
231 #define ICE_RX_ITR	ICE_IDX_ITR0
232 #define ICE_TX_ITR	ICE_IDX_ITR1
233 #define ICE_ITR_8K	124
234 #define ICE_ITR_20K	50
235 #define ICE_ITR_MAX	8160 /* 0x1FE0 */
236 #define ICE_DFLT_TX_ITR	ICE_ITR_20K
237 #define ICE_DFLT_RX_ITR	ICE_ITR_20K
238 enum ice_dynamic_itr {
239 	ITR_STATIC = 0,
240 	ITR_DYNAMIC = 1
241 };
242 
243 #define ITR_IS_DYNAMIC(rc) ((rc)->itr_mode == ITR_DYNAMIC)
244 #define ICE_ITR_GRAN_S		1	/* ITR granularity is always 2us */
245 #define ICE_ITR_GRAN_US		BIT(ICE_ITR_GRAN_S)
246 #define ICE_ITR_MASK		0x1FFE	/* ITR register value alignment mask */
247 #define ITR_REG_ALIGN(setting)	((setting) & ICE_ITR_MASK)
248 
249 #define ICE_DFLT_INTRL	0
250 #define ICE_MAX_INTRL	236
251 
252 #define ICE_IN_WB_ON_ITR_MODE	255
253 /* Sets WB_ON_ITR and assumes INTENA bit is already cleared, which allows
254  * setting the MSK_M bit to tell hardware to ignore the INTENA_M bit. Also,
255  * set the write-back latency to the usecs passed in.
256  */
257 #define ICE_GLINT_DYN_CTL_WB_ON_ITR(usecs, itr_idx)	\
258 	((((usecs) << (GLINT_DYN_CTL_INTERVAL_S - ICE_ITR_GRAN_S)) & \
259 	  GLINT_DYN_CTL_INTERVAL_M) | \
260 	 (((itr_idx) << GLINT_DYN_CTL_ITR_INDX_S) & \
261 	  GLINT_DYN_CTL_ITR_INDX_M) | GLINT_DYN_CTL_INTENA_MSK_M | \
262 	 GLINT_DYN_CTL_WB_ON_ITR_M)
263 
264 /* Legacy or Advanced Mode Queue */
265 #define ICE_TX_ADVANCED	0
266 #define ICE_TX_LEGACY	1
267 
268 /* descriptor ring, associated with a VSI */
269 struct ice_rx_ring {
270 	/* CL1 - 1st cacheline starts here */
271 	struct ice_rx_ring *next;	/* pointer to next ring in q_vector */
272 	void *desc;			/* Descriptor ring memory */
273 	struct device *dev;		/* Used for DMA mapping */
274 	struct net_device *netdev;	/* netdev ring maps to */
275 	struct ice_vsi *vsi;		/* Backreference to associated VSI */
276 	struct ice_q_vector *q_vector;	/* Backreference to associated vector */
277 	u8 __iomem *tail;
278 	union {
279 		struct ice_rx_buf *rx_buf;
280 		struct xdp_buff **xdp_buf;
281 	};
282 	/* CL2 - 2nd cacheline starts here */
283 	struct xdp_rxq_info xdp_rxq;
284 	/* CL3 - 3rd cacheline starts here */
285 	u16 q_index;			/* Queue number of ring */
286 
287 	u16 count;			/* Number of descriptors */
288 	u16 reg_idx;			/* HW register index of the ring */
289 
290 	/* used in interrupt processing */
291 	u16 next_to_use;
292 	u16 next_to_clean;
293 	u16 next_to_alloc;
294 	u16 rx_offset;
295 	u16 rx_buf_len;
296 
297 	/* stats structs */
298 	struct ice_ring_stats *ring_stats;
299 
300 	struct rcu_head rcu;		/* to avoid race on free */
301 	/* CL4 - 3rd cacheline starts here */
302 	struct ice_channel *ch;
303 	struct bpf_prog *xdp_prog;
304 	struct ice_tx_ring *xdp_ring;
305 	struct xsk_buff_pool *xsk_pool;
306 	struct xdp_buff xdp;
307 	struct sk_buff *skb;
308 	dma_addr_t dma;			/* physical address of ring */
309 	u64 cached_phctime;
310 	u8 dcb_tc;			/* Traffic class of ring */
311 	u8 ptp_rx;
312 #define ICE_RX_FLAGS_RING_BUILD_SKB	BIT(1)
313 #define ICE_RX_FLAGS_CRC_STRIP_DIS	BIT(2)
314 	u8 flags;
315 } ____cacheline_internodealigned_in_smp;
316 
317 struct ice_tx_ring {
318 	/* CL1 - 1st cacheline starts here */
319 	struct ice_tx_ring *next;	/* pointer to next ring in q_vector */
320 	void *desc;			/* Descriptor ring memory */
321 	struct device *dev;		/* Used for DMA mapping */
322 	u8 __iomem *tail;
323 	struct ice_tx_buf *tx_buf;
324 	struct ice_q_vector *q_vector;	/* Backreference to associated vector */
325 	struct net_device *netdev;	/* netdev ring maps to */
326 	struct ice_vsi *vsi;		/* Backreference to associated VSI */
327 	/* CL2 - 2nd cacheline starts here */
328 	dma_addr_t dma;			/* physical address of ring */
329 	struct xsk_buff_pool *xsk_pool;
330 	u16 next_to_use;
331 	u16 next_to_clean;
332 	u16 next_rs;
333 	u16 next_dd;
334 	u16 q_handle;			/* Queue handle per TC */
335 	u16 reg_idx;			/* HW register index of the ring */
336 	u16 count;			/* Number of descriptors */
337 	u16 q_index;			/* Queue number of ring */
338 	/* stats structs */
339 	struct ice_ring_stats *ring_stats;
340 	/* CL3 - 3rd cacheline starts here */
341 	struct rcu_head rcu;		/* to avoid race on free */
342 	DECLARE_BITMAP(xps_state, ICE_TX_NBITS);	/* XPS Config State */
343 	struct ice_channel *ch;
344 	struct ice_ptp_tx *tx_tstamps;
345 	spinlock_t tx_lock;
346 	u32 txq_teid;			/* Added Tx queue TEID */
347 	/* CL4 - 4th cacheline starts here */
348 	u16 xdp_tx_active;
349 #define ICE_TX_FLAGS_RING_XDP		BIT(0)
350 #define ICE_TX_FLAGS_RING_VLAN_L2TAG1	BIT(1)
351 #define ICE_TX_FLAGS_RING_VLAN_L2TAG2	BIT(2)
352 	u8 flags;
353 	u8 dcb_tc;			/* Traffic class of ring */
354 	u8 ptp_tx;
355 } ____cacheline_internodealigned_in_smp;
356 
357 static inline bool ice_ring_uses_build_skb(struct ice_rx_ring *ring)
358 {
359 	return !!(ring->flags & ICE_RX_FLAGS_RING_BUILD_SKB);
360 }
361 
362 static inline void ice_set_ring_build_skb_ena(struct ice_rx_ring *ring)
363 {
364 	ring->flags |= ICE_RX_FLAGS_RING_BUILD_SKB;
365 }
366 
367 static inline void ice_clear_ring_build_skb_ena(struct ice_rx_ring *ring)
368 {
369 	ring->flags &= ~ICE_RX_FLAGS_RING_BUILD_SKB;
370 }
371 
372 static inline bool ice_ring_ch_enabled(struct ice_tx_ring *ring)
373 {
374 	return !!ring->ch;
375 }
376 
377 static inline bool ice_ring_is_xdp(struct ice_tx_ring *ring)
378 {
379 	return !!(ring->flags & ICE_TX_FLAGS_RING_XDP);
380 }
381 
382 enum ice_container_type {
383 	ICE_RX_CONTAINER,
384 	ICE_TX_CONTAINER,
385 };
386 
387 struct ice_ring_container {
388 	/* head of linked-list of rings */
389 	union {
390 		struct ice_rx_ring *rx_ring;
391 		struct ice_tx_ring *tx_ring;
392 	};
393 	struct dim dim;		/* data for net_dim algorithm */
394 	u16 itr_idx;		/* index in the interrupt vector */
395 	/* this matches the maximum number of ITR bits, but in usec
396 	 * values, so it is shifted left one bit (bit zero is ignored)
397 	 */
398 	union {
399 		struct {
400 			u16 itr_setting:13;
401 			u16 itr_reserved:2;
402 			u16 itr_mode:1;
403 		};
404 		u16 itr_settings;
405 	};
406 	enum ice_container_type type;
407 };
408 
409 struct ice_coalesce_stored {
410 	u16 itr_tx;
411 	u16 itr_rx;
412 	u8 intrl;
413 	u8 tx_valid;
414 	u8 rx_valid;
415 };
416 
417 /* iterator for handling rings in ring container */
418 #define ice_for_each_rx_ring(pos, head) \
419 	for (pos = (head).rx_ring; pos; pos = pos->next)
420 
421 #define ice_for_each_tx_ring(pos, head) \
422 	for (pos = (head).tx_ring; pos; pos = pos->next)
423 
424 static inline unsigned int ice_rx_pg_order(struct ice_rx_ring *ring)
425 {
426 #if (PAGE_SIZE < 8192)
427 	if (ring->rx_buf_len > (PAGE_SIZE / 2))
428 		return 1;
429 #endif
430 	return 0;
431 }
432 
433 #define ice_rx_pg_size(_ring) (PAGE_SIZE << ice_rx_pg_order(_ring))
434 
435 union ice_32b_rx_flex_desc;
436 
437 bool ice_alloc_rx_bufs(struct ice_rx_ring *rxr, u16 cleaned_count);
438 netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev);
439 u16
440 ice_select_queue(struct net_device *dev, struct sk_buff *skb,
441 		 struct net_device *sb_dev);
442 void ice_clean_tx_ring(struct ice_tx_ring *tx_ring);
443 void ice_clean_rx_ring(struct ice_rx_ring *rx_ring);
444 int ice_setup_tx_ring(struct ice_tx_ring *tx_ring);
445 int ice_setup_rx_ring(struct ice_rx_ring *rx_ring);
446 void ice_free_tx_ring(struct ice_tx_ring *tx_ring);
447 void ice_free_rx_ring(struct ice_rx_ring *rx_ring);
448 int ice_napi_poll(struct napi_struct *napi, int budget);
449 int
450 ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc,
451 		   u8 *raw_packet);
452 int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget);
453 void ice_clean_ctrl_tx_irq(struct ice_tx_ring *tx_ring);
454 #endif /* _ICE_TXRX_H_ */
455