xref: /linux/drivers/net/ethernet/intel/idpf/idpf_txrx.h (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /* Copyright (C) 2023 Intel Corporation */
3 
4 #ifndef _IDPF_TXRX_H_
5 #define _IDPF_TXRX_H_
6 
7 #include <net/page_pool/helpers.h>
8 #include <net/tcp.h>
9 #include <net/netdev_queues.h>
10 
11 #define IDPF_LARGE_MAX_Q			256
12 #define IDPF_MAX_Q				16
13 #define IDPF_MIN_Q				2
14 /* Mailbox Queue */
15 #define IDPF_MAX_MBXQ				1
16 
17 #define IDPF_MIN_TXQ_DESC			64
18 #define IDPF_MIN_RXQ_DESC			64
19 #define IDPF_MIN_TXQ_COMPLQ_DESC		256
20 #define IDPF_MAX_QIDS				256
21 
22 /* Number of descriptors in a queue should be a multiple of 32. RX queue
23  * descriptors alone should be a multiple of IDPF_REQ_RXQ_DESC_MULTIPLE
24  * to achieve BufQ descriptors aligned to 32
25  */
26 #define IDPF_REQ_DESC_MULTIPLE			32
27 #define IDPF_REQ_RXQ_DESC_MULTIPLE (IDPF_MAX_BUFQS_PER_RXQ_GRP * 32)
28 #define IDPF_MIN_TX_DESC_NEEDED (MAX_SKB_FRAGS + 6)
29 #define IDPF_TX_WAKE_THRESH ((u16)IDPF_MIN_TX_DESC_NEEDED * 2)
30 
31 #define IDPF_MAX_DESCS				8160
32 #define IDPF_MAX_TXQ_DESC ALIGN_DOWN(IDPF_MAX_DESCS, IDPF_REQ_DESC_MULTIPLE)
33 #define IDPF_MAX_RXQ_DESC ALIGN_DOWN(IDPF_MAX_DESCS, IDPF_REQ_RXQ_DESC_MULTIPLE)
34 #define MIN_SUPPORT_TXDID (\
35 	VIRTCHNL2_TXDID_FLEX_FLOW_SCHED |\
36 	VIRTCHNL2_TXDID_FLEX_TSO_CTX)
37 
38 #define IDPF_DFLT_SINGLEQ_TX_Q_GROUPS		1
39 #define IDPF_DFLT_SINGLEQ_RX_Q_GROUPS		1
40 #define IDPF_DFLT_SINGLEQ_TXQ_PER_GROUP		4
41 #define IDPF_DFLT_SINGLEQ_RXQ_PER_GROUP		4
42 
43 #define IDPF_COMPLQ_PER_GROUP			1
44 #define IDPF_SINGLE_BUFQ_PER_RXQ_GRP		1
45 #define IDPF_MAX_BUFQS_PER_RXQ_GRP		2
46 #define IDPF_BUFQ2_ENA				1
47 #define IDPF_NUMQ_PER_CHUNK			1
48 
49 #define IDPF_DFLT_SPLITQ_TXQ_PER_GROUP		1
50 #define IDPF_DFLT_SPLITQ_RXQ_PER_GROUP		1
51 
52 /* Default vector sharing */
53 #define IDPF_MBX_Q_VEC		1
54 #define IDPF_MIN_Q_VEC		1
55 
56 #define IDPF_DFLT_TX_Q_DESC_COUNT		512
57 #define IDPF_DFLT_TX_COMPLQ_DESC_COUNT		512
58 #define IDPF_DFLT_RX_Q_DESC_COUNT		512
59 
60 /* IMPORTANT: We absolutely _cannot_ have more buffers in the system than a
61  * given RX completion queue has descriptors. This includes _ALL_ buffer
62  * queues. E.g.: If you have two buffer queues of 512 descriptors and buffers,
63  * you have a total of 1024 buffers so your RX queue _must_ have at least that
64  * many descriptors. This macro divides a given number of RX descriptors by
65  * number of buffer queues to calculate how many descriptors each buffer queue
66  * can have without overrunning the RX queue.
67  *
68  * If you give hardware more buffers than completion descriptors what will
69  * happen is that if hardware gets a chance to post more than ring wrap of
70  * descriptors before SW gets an interrupt and overwrites SW head, the gen bit
71  * in the descriptor will be wrong. Any overwritten descriptors' buffers will
72  * be gone forever and SW has no reasonable way to tell that this has happened.
73  * From SW perspective, when we finally get an interrupt, it looks like we're
74  * still waiting for descriptor to be done, stalling forever.
75  */
76 #define IDPF_RX_BUFQ_DESC_COUNT(RXD, NUM_BUFQ)	((RXD) / (NUM_BUFQ))
77 
78 #define IDPF_RX_BUFQ_WORKING_SET(rxq)		((rxq)->desc_count - 1)
79 
80 #define IDPF_RX_BUMP_NTC(rxq, ntc)				\
81 do {								\
82 	if (unlikely(++(ntc) == (rxq)->desc_count)) {		\
83 		ntc = 0;					\
84 		change_bit(__IDPF_Q_GEN_CHK, (rxq)->flags);	\
85 	}							\
86 } while (0)
87 
88 #define IDPF_SINGLEQ_BUMP_RING_IDX(q, idx)			\
89 do {								\
90 	if (unlikely(++(idx) == (q)->desc_count))		\
91 		idx = 0;					\
92 } while (0)
93 
94 #define IDPF_RX_HDR_SIZE			256
95 #define IDPF_RX_BUF_2048			2048
96 #define IDPF_RX_BUF_4096			4096
97 #define IDPF_RX_BUF_STRIDE			32
98 #define IDPF_RX_BUF_POST_STRIDE			16
99 #define IDPF_LOW_WATERMARK			64
100 /* Size of header buffer specifically for header split */
101 #define IDPF_HDR_BUF_SIZE			256
102 #define IDPF_PACKET_HDR_PAD	\
103 	(ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN * 2)
104 #define IDPF_TX_TSO_MIN_MSS			88
105 
106 /* Minimum number of descriptors between 2 descriptors with the RE bit set;
107  * only relevant in flow scheduling mode
108  */
109 #define IDPF_TX_SPLITQ_RE_MIN_GAP	64
110 
111 #define IDPF_RX_BI_BUFID_S		0
112 #define IDPF_RX_BI_BUFID_M		GENMASK(14, 0)
113 #define IDPF_RX_BI_GEN_S		15
114 #define IDPF_RX_BI_GEN_M		BIT(IDPF_RX_BI_GEN_S)
115 #define IDPF_RXD_EOF_SPLITQ		VIRTCHNL2_RX_FLEX_DESC_ADV_STATUS0_EOF_M
116 #define IDPF_RXD_EOF_SINGLEQ		VIRTCHNL2_RX_BASE_DESC_STATUS_EOF_M
117 
118 #define IDPF_SINGLEQ_RX_BUF_DESC(rxq, i)	\
119 	(&(((struct virtchnl2_singleq_rx_buf_desc *)((rxq)->desc_ring))[i]))
120 #define IDPF_SPLITQ_RX_BUF_DESC(rxq, i)	\
121 	(&(((struct virtchnl2_splitq_rx_buf_desc *)((rxq)->desc_ring))[i]))
122 #define IDPF_SPLITQ_RX_BI_DESC(rxq, i) ((((rxq)->ring))[i])
123 
124 #define IDPF_BASE_TX_DESC(txq, i)	\
125 	(&(((struct idpf_base_tx_desc *)((txq)->desc_ring))[i]))
126 #define IDPF_BASE_TX_CTX_DESC(txq, i) \
127 	(&(((struct idpf_base_tx_ctx_desc *)((txq)->desc_ring))[i]))
128 #define IDPF_SPLITQ_TX_COMPLQ_DESC(txcq, i)	\
129 	(&(((struct idpf_splitq_tx_compl_desc *)((txcq)->desc_ring))[i]))
130 
131 #define IDPF_FLEX_TX_DESC(txq, i) \
132 	(&(((union idpf_tx_flex_desc *)((txq)->desc_ring))[i]))
133 #define IDPF_FLEX_TX_CTX_DESC(txq, i)	\
134 	(&(((struct idpf_flex_tx_ctx_desc *)((txq)->desc_ring))[i]))
135 
136 #define IDPF_DESC_UNUSED(txq)     \
137 	((((txq)->next_to_clean > (txq)->next_to_use) ? 0 : (txq)->desc_count) + \
138 	(txq)->next_to_clean - (txq)->next_to_use - 1)
139 
140 #define IDPF_TX_BUF_RSV_UNUSED(txq)	((txq)->buf_stack.top)
141 #define IDPF_TX_BUF_RSV_LOW(txq)	(IDPF_TX_BUF_RSV_UNUSED(txq) < \
142 					 (txq)->desc_count >> 2)
143 
144 #define IDPF_TX_COMPLQ_OVERFLOW_THRESH(txcq)	((txcq)->desc_count >> 1)
145 /* Determine the absolute number of completions pending, i.e. the number of
146  * completions that are expected to arrive on the TX completion queue.
147  */
148 #define IDPF_TX_COMPLQ_PENDING(txq)	\
149 	(((txq)->num_completions_pending >= (txq)->complq->num_completions ? \
150 	0 : U64_MAX) + \
151 	(txq)->num_completions_pending - (txq)->complq->num_completions)
152 
153 #define IDPF_TX_SPLITQ_COMPL_TAG_WIDTH	16
154 #define IDPF_SPLITQ_TX_INVAL_COMPL_TAG	-1
155 /* Adjust the generation for the completion tag and wrap if necessary */
156 #define IDPF_TX_ADJ_COMPL_TAG_GEN(txq) \
157 	((++(txq)->compl_tag_cur_gen) >= (txq)->compl_tag_gen_max ? \
158 	0 : (txq)->compl_tag_cur_gen)
159 
160 #define IDPF_TXD_LAST_DESC_CMD (IDPF_TX_DESC_CMD_EOP | IDPF_TX_DESC_CMD_RS)
161 
162 #define IDPF_TX_FLAGS_TSO		BIT(0)
163 #define IDPF_TX_FLAGS_IPV4		BIT(1)
164 #define IDPF_TX_FLAGS_IPV6		BIT(2)
165 #define IDPF_TX_FLAGS_TUNNEL		BIT(3)
166 
167 union idpf_tx_flex_desc {
168 	struct idpf_flex_tx_desc q; /* queue based scheduling */
169 	struct idpf_flex_tx_sched_desc flow; /* flow based scheduling */
170 };
171 
172 /**
173  * struct idpf_tx_buf
174  * @next_to_watch: Next descriptor to clean
175  * @skb: Pointer to the skb
176  * @dma: DMA address
177  * @len: DMA length
178  * @bytecount: Number of bytes
179  * @gso_segs: Number of GSO segments
180  * @compl_tag: Splitq only, unique identifier for a buffer. Used to compare
181  *	       with completion tag returned in buffer completion event.
182  *	       Because the completion tag is expected to be the same in all
183  *	       data descriptors for a given packet, and a single packet can
184  *	       span multiple buffers, we need this field to track all
185  *	       buffers associated with this completion tag independently of
186  *	       the buf_id. The tag consists of a N bit buf_id and M upper
187  *	       order "generation bits". See compl_tag_bufid_m and
188  *	       compl_tag_gen_s in struct idpf_queue. We'll use a value of -1
189  *	       to indicate the tag is not valid.
190  * @ctx_entry: Singleq only. Used to indicate the corresponding entry
191  *	       in the descriptor ring was used for a context descriptor and
192  *	       this buffer entry should be skipped.
193  */
194 struct idpf_tx_buf {
195 	void *next_to_watch;
196 	struct sk_buff *skb;
197 	DEFINE_DMA_UNMAP_ADDR(dma);
198 	DEFINE_DMA_UNMAP_LEN(len);
199 	unsigned int bytecount;
200 	unsigned short gso_segs;
201 
202 	union {
203 		int compl_tag;
204 
205 		bool ctx_entry;
206 	};
207 };
208 
209 struct idpf_tx_stash {
210 	struct hlist_node hlist;
211 	struct idpf_tx_buf buf;
212 };
213 
214 /**
215  * struct idpf_buf_lifo - LIFO for managing OOO completions
216  * @top: Used to know how many buffers are left
217  * @size: Total size of LIFO
218  * @bufs: Backing array
219  */
220 struct idpf_buf_lifo {
221 	u16 top;
222 	u16 size;
223 	struct idpf_tx_stash **bufs;
224 };
225 
226 /**
227  * struct idpf_tx_offload_params - Offload parameters for a given packet
228  * @tx_flags: Feature flags enabled for this packet
229  * @hdr_offsets: Offset parameter for single queue model
230  * @cd_tunneling: Type of tunneling enabled for single queue model
231  * @tso_len: Total length of payload to segment
232  * @mss: Segment size
233  * @tso_segs: Number of segments to be sent
234  * @tso_hdr_len: Length of headers to be duplicated
235  * @td_cmd: Command field to be inserted into descriptor
236  */
237 struct idpf_tx_offload_params {
238 	u32 tx_flags;
239 
240 	u32 hdr_offsets;
241 	u32 cd_tunneling;
242 
243 	u32 tso_len;
244 	u16 mss;
245 	u16 tso_segs;
246 	u16 tso_hdr_len;
247 
248 	u16 td_cmd;
249 };
250 
251 /**
252  * struct idpf_tx_splitq_params
253  * @dtype: General descriptor info
254  * @eop_cmd: Type of EOP
255  * @compl_tag: Associated tag for completion
256  * @td_tag: Descriptor tunneling tag
257  * @offload: Offload parameters
258  */
259 struct idpf_tx_splitq_params {
260 	enum idpf_tx_desc_dtype_value dtype;
261 	u16 eop_cmd;
262 	union {
263 		u16 compl_tag;
264 		u16 td_tag;
265 	};
266 
267 	struct idpf_tx_offload_params offload;
268 };
269 
270 enum idpf_tx_ctx_desc_eipt_offload {
271 	IDPF_TX_CTX_EXT_IP_NONE         = 0x0,
272 	IDPF_TX_CTX_EXT_IP_IPV6         = 0x1,
273 	IDPF_TX_CTX_EXT_IP_IPV4_NO_CSUM = 0x2,
274 	IDPF_TX_CTX_EXT_IP_IPV4         = 0x3
275 };
276 
277 /* Checksum offload bits decoded from the receive descriptor. */
278 struct idpf_rx_csum_decoded {
279 	u32 l3l4p : 1;
280 	u32 ipe : 1;
281 	u32 eipe : 1;
282 	u32 eudpe : 1;
283 	u32 ipv6exadd : 1;
284 	u32 l4e : 1;
285 	u32 pprs : 1;
286 	u32 nat : 1;
287 	u32 raw_csum_inv : 1;
288 	u32 raw_csum : 16;
289 };
290 
291 struct idpf_rx_extracted {
292 	unsigned int size;
293 	u16 rx_ptype;
294 };
295 
296 #define IDPF_TX_COMPLQ_CLEAN_BUDGET	256
297 #define IDPF_TX_MIN_PKT_LEN		17
298 #define IDPF_TX_DESCS_FOR_SKB_DATA_PTR	1
299 #define IDPF_TX_DESCS_PER_CACHE_LINE	(L1_CACHE_BYTES / \
300 					 sizeof(struct idpf_flex_tx_desc))
301 #define IDPF_TX_DESCS_FOR_CTX		1
302 /* TX descriptors needed, worst case */
303 #define IDPF_TX_DESC_NEEDED (MAX_SKB_FRAGS + IDPF_TX_DESCS_FOR_CTX + \
304 			     IDPF_TX_DESCS_PER_CACHE_LINE + \
305 			     IDPF_TX_DESCS_FOR_SKB_DATA_PTR)
306 
307 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
308  * In order to align with the read requests we will align the value to
309  * the nearest 4K which represents our maximum read request size.
310  */
311 #define IDPF_TX_MAX_READ_REQ_SIZE	SZ_4K
312 #define IDPF_TX_MAX_DESC_DATA		(SZ_16K - 1)
313 #define IDPF_TX_MAX_DESC_DATA_ALIGNED \
314 	ALIGN_DOWN(IDPF_TX_MAX_DESC_DATA, IDPF_TX_MAX_READ_REQ_SIZE)
315 
316 #define IDPF_RX_DMA_ATTR \
317 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
318 #define IDPF_RX_DESC(rxq, i)	\
319 	(&(((union virtchnl2_rx_desc *)((rxq)->desc_ring))[i]))
320 
321 struct idpf_rx_buf {
322 	struct page *page;
323 	unsigned int page_offset;
324 	u16 truesize;
325 };
326 
327 #define IDPF_RX_MAX_PTYPE_PROTO_IDS    32
328 #define IDPF_RX_MAX_PTYPE_SZ	(sizeof(struct virtchnl2_ptype) + \
329 				 (sizeof(u16) * IDPF_RX_MAX_PTYPE_PROTO_IDS))
330 #define IDPF_RX_PTYPE_HDR_SZ	sizeof(struct virtchnl2_get_ptype_info)
331 #define IDPF_RX_MAX_PTYPES_PER_BUF	\
332 	DIV_ROUND_DOWN_ULL((IDPF_CTLQ_MAX_BUF_LEN - IDPF_RX_PTYPE_HDR_SZ), \
333 			   IDPF_RX_MAX_PTYPE_SZ)
334 
335 #define IDPF_GET_PTYPE_SIZE(p) struct_size((p), proto_id, (p)->proto_id_count)
336 
337 #define IDPF_TUN_IP_GRE (\
338 	IDPF_PTYPE_TUNNEL_IP |\
339 	IDPF_PTYPE_TUNNEL_IP_GRENAT)
340 
341 #define IDPF_TUN_IP_GRE_MAC (\
342 	IDPF_TUN_IP_GRE |\
343 	IDPF_PTYPE_TUNNEL_IP_GRENAT_MAC)
344 
345 #define IDPF_RX_MAX_PTYPE	1024
346 #define IDPF_RX_MAX_BASE_PTYPE	256
347 #define IDPF_INVALID_PTYPE_ID	0xFFFF
348 
349 /* Packet type non-ip values */
350 enum idpf_rx_ptype_l2 {
351 	IDPF_RX_PTYPE_L2_RESERVED	= 0,
352 	IDPF_RX_PTYPE_L2_MAC_PAY2	= 1,
353 	IDPF_RX_PTYPE_L2_TIMESYNC_PAY2	= 2,
354 	IDPF_RX_PTYPE_L2_FIP_PAY2	= 3,
355 	IDPF_RX_PTYPE_L2_OUI_PAY2	= 4,
356 	IDPF_RX_PTYPE_L2_MACCNTRL_PAY2	= 5,
357 	IDPF_RX_PTYPE_L2_LLDP_PAY2	= 6,
358 	IDPF_RX_PTYPE_L2_ECP_PAY2	= 7,
359 	IDPF_RX_PTYPE_L2_EVB_PAY2	= 8,
360 	IDPF_RX_PTYPE_L2_QCN_PAY2	= 9,
361 	IDPF_RX_PTYPE_L2_EAPOL_PAY2	= 10,
362 	IDPF_RX_PTYPE_L2_ARP		= 11,
363 };
364 
365 enum idpf_rx_ptype_outer_ip {
366 	IDPF_RX_PTYPE_OUTER_L2	= 0,
367 	IDPF_RX_PTYPE_OUTER_IP	= 1,
368 };
369 
370 #define IDPF_RX_PTYPE_TO_IPV(ptype, ipv)			\
371 	(((ptype)->outer_ip == IDPF_RX_PTYPE_OUTER_IP) &&	\
372 	 ((ptype)->outer_ip_ver == (ipv)))
373 
374 enum idpf_rx_ptype_outer_ip_ver {
375 	IDPF_RX_PTYPE_OUTER_NONE	= 0,
376 	IDPF_RX_PTYPE_OUTER_IPV4	= 1,
377 	IDPF_RX_PTYPE_OUTER_IPV6	= 2,
378 };
379 
380 enum idpf_rx_ptype_outer_fragmented {
381 	IDPF_RX_PTYPE_NOT_FRAG	= 0,
382 	IDPF_RX_PTYPE_FRAG	= 1,
383 };
384 
385 enum idpf_rx_ptype_tunnel_type {
386 	IDPF_RX_PTYPE_TUNNEL_NONE		= 0,
387 	IDPF_RX_PTYPE_TUNNEL_IP_IP		= 1,
388 	IDPF_RX_PTYPE_TUNNEL_IP_GRENAT		= 2,
389 	IDPF_RX_PTYPE_TUNNEL_IP_GRENAT_MAC	= 3,
390 	IDPF_RX_PTYPE_TUNNEL_IP_GRENAT_MAC_VLAN	= 4,
391 };
392 
393 enum idpf_rx_ptype_tunnel_end_prot {
394 	IDPF_RX_PTYPE_TUNNEL_END_NONE	= 0,
395 	IDPF_RX_PTYPE_TUNNEL_END_IPV4	= 1,
396 	IDPF_RX_PTYPE_TUNNEL_END_IPV6	= 2,
397 };
398 
399 enum idpf_rx_ptype_inner_prot {
400 	IDPF_RX_PTYPE_INNER_PROT_NONE		= 0,
401 	IDPF_RX_PTYPE_INNER_PROT_UDP		= 1,
402 	IDPF_RX_PTYPE_INNER_PROT_TCP		= 2,
403 	IDPF_RX_PTYPE_INNER_PROT_SCTP		= 3,
404 	IDPF_RX_PTYPE_INNER_PROT_ICMP		= 4,
405 	IDPF_RX_PTYPE_INNER_PROT_TIMESYNC	= 5,
406 };
407 
408 enum idpf_rx_ptype_payload_layer {
409 	IDPF_RX_PTYPE_PAYLOAD_LAYER_NONE	= 0,
410 	IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY2	= 1,
411 	IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY3	= 2,
412 	IDPF_RX_PTYPE_PAYLOAD_LAYER_PAY4	= 3,
413 };
414 
415 enum idpf_tunnel_state {
416 	IDPF_PTYPE_TUNNEL_IP                    = BIT(0),
417 	IDPF_PTYPE_TUNNEL_IP_GRENAT             = BIT(1),
418 	IDPF_PTYPE_TUNNEL_IP_GRENAT_MAC         = BIT(2),
419 };
420 
421 struct idpf_ptype_state {
422 	bool outer_ip;
423 	bool outer_frag;
424 	u8 tunnel_state;
425 };
426 
427 struct idpf_rx_ptype_decoded {
428 	u32 ptype:10;
429 	u32 known:1;
430 	u32 outer_ip:1;
431 	u32 outer_ip_ver:2;
432 	u32 outer_frag:1;
433 	u32 tunnel_type:3;
434 	u32 tunnel_end_prot:2;
435 	u32 tunnel_end_frag:1;
436 	u32 inner_prot:4;
437 	u32 payload_layer:3;
438 };
439 
440 /**
441  * enum idpf_queue_flags_t
442  * @__IDPF_Q_GEN_CHK: Queues operating in splitq mode use a generation bit to
443  *		      identify new descriptor writebacks on the ring. HW sets
444  *		      the gen bit to 1 on the first writeback of any given
445  *		      descriptor. After the ring wraps, HW sets the gen bit of
446  *		      those descriptors to 0, and continues flipping
447  *		      0->1 or 1->0 on each ring wrap. SW maintains its own
448  *		      gen bit to know what value will indicate writebacks on
449  *		      the next pass around the ring. E.g. it is initialized
450  *		      to 1 and knows that reading a gen bit of 1 in any
451  *		      descriptor on the initial pass of the ring indicates a
452  *		      writeback. It also flips on every ring wrap.
453  * @__IDPF_RFLQ_GEN_CHK: Refill queues are SW only, so Q_GEN acts as the HW bit
454  *			 and RFLGQ_GEN is the SW bit.
455  * @__IDPF_Q_FLOW_SCH_EN: Enable flow scheduling
456  * @__IDPF_Q_SW_MARKER: Used to indicate TX queue marker completions
457  * @__IDPF_Q_POLL_MODE: Enable poll mode
458  * @__IDPF_Q_FLAGS_NBITS: Must be last
459  */
460 enum idpf_queue_flags_t {
461 	__IDPF_Q_GEN_CHK,
462 	__IDPF_RFLQ_GEN_CHK,
463 	__IDPF_Q_FLOW_SCH_EN,
464 	__IDPF_Q_SW_MARKER,
465 	__IDPF_Q_POLL_MODE,
466 
467 	__IDPF_Q_FLAGS_NBITS,
468 };
469 
470 /**
471  * struct idpf_vec_regs
472  * @dyn_ctl_reg: Dynamic control interrupt register offset
473  * @itrn_reg: Interrupt Throttling Rate register offset
474  * @itrn_index_spacing: Register spacing between ITR registers of the same
475  *			vector
476  */
477 struct idpf_vec_regs {
478 	u32 dyn_ctl_reg;
479 	u32 itrn_reg;
480 	u32 itrn_index_spacing;
481 };
482 
483 /**
484  * struct idpf_intr_reg
485  * @dyn_ctl: Dynamic control interrupt register
486  * @dyn_ctl_intena_m: Mask for dyn_ctl interrupt enable
487  * @dyn_ctl_itridx_s: Register bit offset for ITR index
488  * @dyn_ctl_itridx_m: Mask for ITR index
489  * @dyn_ctl_intrvl_s: Register bit offset for ITR interval
490  * @rx_itr: RX ITR register
491  * @tx_itr: TX ITR register
492  * @icr_ena: Interrupt cause register offset
493  * @icr_ena_ctlq_m: Mask for ICR
494  */
495 struct idpf_intr_reg {
496 	void __iomem *dyn_ctl;
497 	u32 dyn_ctl_intena_m;
498 	u32 dyn_ctl_itridx_s;
499 	u32 dyn_ctl_itridx_m;
500 	u32 dyn_ctl_intrvl_s;
501 	void __iomem *rx_itr;
502 	void __iomem *tx_itr;
503 	void __iomem *icr_ena;
504 	u32 icr_ena_ctlq_m;
505 };
506 
507 /**
508  * struct idpf_q_vector
509  * @vport: Vport back pointer
510  * @affinity_mask: CPU affinity mask
511  * @napi: napi handler
512  * @v_idx: Vector index
513  * @intr_reg: See struct idpf_intr_reg
514  * @num_txq: Number of TX queues
515  * @tx: Array of TX queues to service
516  * @tx_dim: Data for TX net_dim algorithm
517  * @tx_itr_value: TX interrupt throttling rate
518  * @tx_intr_mode: Dynamic ITR or not
519  * @tx_itr_idx: TX ITR index
520  * @num_rxq: Number of RX queues
521  * @rx: Array of RX queues to service
522  * @rx_dim: Data for RX net_dim algorithm
523  * @rx_itr_value: RX interrupt throttling rate
524  * @rx_intr_mode: Dynamic ITR or not
525  * @rx_itr_idx: RX ITR index
526  * @num_bufq: Number of buffer queues
527  * @bufq: Array of buffer queues to service
528  * @total_events: Number of interrupts processed
529  * @name: Queue vector name
530  */
531 struct idpf_q_vector {
532 	struct idpf_vport *vport;
533 	cpumask_t affinity_mask;
534 	struct napi_struct napi;
535 	u16 v_idx;
536 	struct idpf_intr_reg intr_reg;
537 
538 	u16 num_txq;
539 	struct idpf_queue **tx;
540 	struct dim tx_dim;
541 	u16 tx_itr_value;
542 	bool tx_intr_mode;
543 	u32 tx_itr_idx;
544 
545 	u16 num_rxq;
546 	struct idpf_queue **rx;
547 	struct dim rx_dim;
548 	u16 rx_itr_value;
549 	bool rx_intr_mode;
550 	u32 rx_itr_idx;
551 
552 	u16 num_bufq;
553 	struct idpf_queue **bufq;
554 
555 	u16 total_events;
556 	char *name;
557 };
558 
559 struct idpf_rx_queue_stats {
560 	u64_stats_t packets;
561 	u64_stats_t bytes;
562 	u64_stats_t rsc_pkts;
563 	u64_stats_t hw_csum_err;
564 	u64_stats_t hsplit_pkts;
565 	u64_stats_t hsplit_buf_ovf;
566 	u64_stats_t bad_descs;
567 };
568 
569 struct idpf_tx_queue_stats {
570 	u64_stats_t packets;
571 	u64_stats_t bytes;
572 	u64_stats_t lso_pkts;
573 	u64_stats_t linearize;
574 	u64_stats_t q_busy;
575 	u64_stats_t skb_drops;
576 	u64_stats_t dma_map_errs;
577 };
578 
579 struct idpf_cleaned_stats {
580 	u32 packets;
581 	u32 bytes;
582 };
583 
584 union idpf_queue_stats {
585 	struct idpf_rx_queue_stats rx;
586 	struct idpf_tx_queue_stats tx;
587 };
588 
589 #define IDPF_ITR_DYNAMIC	1
590 #define IDPF_ITR_MAX		0x1FE0
591 #define IDPF_ITR_20K		0x0032
592 #define IDPF_ITR_GRAN_S		1	/* Assume ITR granularity is 2us */
593 #define IDPF_ITR_MASK		0x1FFE  /* ITR register value alignment mask */
594 #define ITR_REG_ALIGN(setting)	((setting) & IDPF_ITR_MASK)
595 #define IDPF_ITR_IS_DYNAMIC(itr_mode) (itr_mode)
596 #define IDPF_ITR_TX_DEF		IDPF_ITR_20K
597 #define IDPF_ITR_RX_DEF		IDPF_ITR_20K
598 /* Index used for 'No ITR' update in DYN_CTL register */
599 #define IDPF_NO_ITR_UPDATE_IDX	3
600 #define IDPF_ITR_IDX_SPACING(spacing, dflt)	(spacing ? spacing : dflt)
601 #define IDPF_DIM_DEFAULT_PROFILE_IX		1
602 
603 /**
604  * struct idpf_queue
605  * @dev: Device back pointer for DMA mapping
606  * @vport: Back pointer to associated vport
607  * @txq_grp: See struct idpf_txq_group
608  * @rxq_grp: See struct idpf_rxq_group
609  * @idx: For buffer queue, it is used as group id, either 0 or 1. On clean,
610  *	 buffer queue uses this index to determine which group of refill queues
611  *	 to clean.
612  *	 For TX queue, it is used as index to map between TX queue group and
613  *	 hot path TX pointers stored in vport. Used in both singleq/splitq.
614  *	 For RX queue, it is used to index to total RX queue across groups and
615  *	 used for skb reporting.
616  * @tail: Tail offset. Used for both queue models single and split. In splitq
617  *	  model relevant only for TX queue and RX queue.
618  * @tx_buf: See struct idpf_tx_buf
619  * @rx_buf: Struct with RX buffer related members
620  * @rx_buf.buf: See struct idpf_rx_buf
621  * @rx_buf.hdr_buf_pa: DMA handle
622  * @rx_buf.hdr_buf_va: Virtual address
623  * @pp: Page pool pointer
624  * @skb: Pointer to the skb
625  * @q_type: Queue type (TX, RX, TX completion, RX buffer)
626  * @q_id: Queue id
627  * @desc_count: Number of descriptors
628  * @next_to_use: Next descriptor to use. Relevant in both split & single txq
629  *		 and bufq.
630  * @next_to_clean: Next descriptor to clean. In split queue model, only
631  *		   relevant to TX completion queue and RX queue.
632  * @next_to_alloc: RX buffer to allocate at. Used only for RX. In splitq model
633  *		   only relevant to RX queue.
634  * @flags: See enum idpf_queue_flags_t
635  * @q_stats: See union idpf_queue_stats
636  * @stats_sync: See struct u64_stats_sync
637  * @cleaned_bytes: Splitq only, TXQ only: When a TX completion is received on
638  *		   the TX completion queue, it can be for any TXQ associated
639  *		   with that completion queue. This means we can clean up to
640  *		   N TXQs during a single call to clean the completion queue.
641  *		   cleaned_bytes|pkts tracks the clean stats per TXQ during
642  *		   that single call to clean the completion queue. By doing so,
643  *		   we can update BQL with aggregate cleaned stats for each TXQ
644  *		   only once at the end of the cleaning routine.
645  * @cleaned_pkts: Number of packets cleaned for the above said case
646  * @rx_hsplit_en: RX headsplit enable
647  * @rx_hbuf_size: Header buffer size
648  * @rx_buf_size: Buffer size
649  * @rx_max_pkt_size: RX max packet size
650  * @rx_buf_stride: RX buffer stride
651  * @rx_buffer_low_watermark: RX buffer low watermark
652  * @rxdids: Supported RX descriptor ids
653  * @q_vector: Backreference to associated vector
654  * @size: Length of descriptor ring in bytes
655  * @dma: Physical address of ring
656  * @desc_ring: Descriptor ring memory
657  * @tx_max_bufs: Max buffers that can be transmitted with scatter-gather
658  * @tx_min_pkt_len: Min supported packet length
659  * @num_completions: Only relevant for TX completion queue. It tracks the
660  *		     number of completions received to compare against the
661  *		     number of completions pending, as accumulated by the
662  *		     TX queues.
663  * @buf_stack: Stack of empty buffers to store buffer info for out of order
664  *	       buffer completions. See struct idpf_buf_lifo.
665  * @compl_tag_bufid_m: Completion tag buffer id mask
666  * @compl_tag_gen_s: Completion tag generation bit
667  *	The format of the completion tag will change based on the TXQ
668  *	descriptor ring size so that we can maintain roughly the same level
669  *	of "uniqueness" across all descriptor sizes. For example, if the
670  *	TXQ descriptor ring size is 64 (the minimum size supported), the
671  *	completion tag will be formatted as below:
672  *	15                 6 5         0
673  *	--------------------------------
674  *	|    GEN=0-1023     |IDX = 0-63|
675  *	--------------------------------
676  *
677  *	This gives us 64*1024 = 65536 possible unique values. Similarly, if
678  *	the TXQ descriptor ring size is 8160 (the maximum size supported),
679  *	the completion tag will be formatted as below:
680  *	15 13 12                       0
681  *	--------------------------------
682  *	|GEN |       IDX = 0-8159      |
683  *	--------------------------------
684  *
685  *	This gives us 8*8160 = 65280 possible unique values.
686  * @compl_tag_cur_gen: Used to keep track of current completion tag generation
687  * @compl_tag_gen_max: To determine when compl_tag_cur_gen should be reset
688  * @sched_buf_hash: Hash table to stores buffers
689  */
690 struct idpf_queue {
691 	struct device *dev;
692 	struct idpf_vport *vport;
693 	union {
694 		struct idpf_txq_group *txq_grp;
695 		struct idpf_rxq_group *rxq_grp;
696 	};
697 	u16 idx;
698 	void __iomem *tail;
699 	union {
700 		struct idpf_tx_buf *tx_buf;
701 		struct {
702 			struct idpf_rx_buf *buf;
703 			dma_addr_t hdr_buf_pa;
704 			void *hdr_buf_va;
705 		} rx_buf;
706 	};
707 	struct page_pool *pp;
708 	struct sk_buff *skb;
709 	u16 q_type;
710 	u32 q_id;
711 	u16 desc_count;
712 
713 	u16 next_to_use;
714 	u16 next_to_clean;
715 	u16 next_to_alloc;
716 	DECLARE_BITMAP(flags, __IDPF_Q_FLAGS_NBITS);
717 
718 	union idpf_queue_stats q_stats;
719 	struct u64_stats_sync stats_sync;
720 
721 	u32 cleaned_bytes;
722 	u16 cleaned_pkts;
723 
724 	bool rx_hsplit_en;
725 	u16 rx_hbuf_size;
726 	u16 rx_buf_size;
727 	u16 rx_max_pkt_size;
728 	u16 rx_buf_stride;
729 	u8 rx_buffer_low_watermark;
730 	u64 rxdids;
731 	struct idpf_q_vector *q_vector;
732 	unsigned int size;
733 	dma_addr_t dma;
734 	void *desc_ring;
735 
736 	u16 tx_max_bufs;
737 	u8 tx_min_pkt_len;
738 
739 	u32 num_completions;
740 
741 	struct idpf_buf_lifo buf_stack;
742 
743 	u16 compl_tag_bufid_m;
744 	u16 compl_tag_gen_s;
745 
746 	u16 compl_tag_cur_gen;
747 	u16 compl_tag_gen_max;
748 
749 	DECLARE_HASHTABLE(sched_buf_hash, 12);
750 } ____cacheline_internodealigned_in_smp;
751 
752 /**
753  * struct idpf_sw_queue
754  * @next_to_clean: Next descriptor to clean
755  * @next_to_alloc: Buffer to allocate at
756  * @flags: See enum idpf_queue_flags_t
757  * @ring: Pointer to the ring
758  * @desc_count: Descriptor count
759  * @dev: Device back pointer for DMA mapping
760  *
761  * Software queues are used in splitq mode to manage buffers between rxq
762  * producer and the bufq consumer.  These are required in order to maintain a
763  * lockless buffer management system and are strictly software only constructs.
764  */
765 struct idpf_sw_queue {
766 	u16 next_to_clean;
767 	u16 next_to_alloc;
768 	DECLARE_BITMAP(flags, __IDPF_Q_FLAGS_NBITS);
769 	u16 *ring;
770 	u16 desc_count;
771 	struct device *dev;
772 } ____cacheline_internodealigned_in_smp;
773 
774 /**
775  * struct idpf_rxq_set
776  * @rxq: RX queue
777  * @refillq0: Pointer to refill queue 0
778  * @refillq1: Pointer to refill queue 1
779  *
780  * Splitq only.  idpf_rxq_set associates an rxq with at an array of refillqs.
781  * Each rxq needs a refillq to return used buffers back to the respective bufq.
782  * Bufqs then clean these refillqs for buffers to give to hardware.
783  */
784 struct idpf_rxq_set {
785 	struct idpf_queue rxq;
786 	struct idpf_sw_queue *refillq0;
787 	struct idpf_sw_queue *refillq1;
788 };
789 
790 /**
791  * struct idpf_bufq_set
792  * @bufq: Buffer queue
793  * @num_refillqs: Number of refill queues. This is always equal to num_rxq_sets
794  *		  in idpf_rxq_group.
795  * @refillqs: Pointer to refill queues array.
796  *
797  * Splitq only. idpf_bufq_set associates a bufq to an array of refillqs.
798  * In this bufq_set, there will be one refillq for each rxq in this rxq_group.
799  * Used buffers received by rxqs will be put on refillqs which bufqs will
800  * clean to return new buffers back to hardware.
801  *
802  * Buffers needed by some number of rxqs associated in this rxq_group are
803  * managed by at most two bufqs (depending on performance configuration).
804  */
805 struct idpf_bufq_set {
806 	struct idpf_queue bufq;
807 	int num_refillqs;
808 	struct idpf_sw_queue *refillqs;
809 };
810 
811 /**
812  * struct idpf_rxq_group
813  * @vport: Vport back pointer
814  * @singleq: Struct with single queue related members
815  * @singleq.num_rxq: Number of RX queues associated
816  * @singleq.rxqs: Array of RX queue pointers
817  * @splitq: Struct with split queue related members
818  * @splitq.num_rxq_sets: Number of RX queue sets
819  * @splitq.rxq_sets: Array of RX queue sets
820  * @splitq.bufq_sets: Buffer queue set pointer
821  *
822  * In singleq mode, an rxq_group is simply an array of rxqs.  In splitq, a
823  * rxq_group contains all the rxqs, bufqs and refillqs needed to
824  * manage buffers in splitq mode.
825  */
826 struct idpf_rxq_group {
827 	struct idpf_vport *vport;
828 
829 	union {
830 		struct {
831 			u16 num_rxq;
832 			struct idpf_queue *rxqs[IDPF_LARGE_MAX_Q];
833 		} singleq;
834 		struct {
835 			u16 num_rxq_sets;
836 			struct idpf_rxq_set *rxq_sets[IDPF_LARGE_MAX_Q];
837 			struct idpf_bufq_set *bufq_sets;
838 		} splitq;
839 	};
840 };
841 
842 /**
843  * struct idpf_txq_group
844  * @vport: Vport back pointer
845  * @num_txq: Number of TX queues associated
846  * @txqs: Array of TX queue pointers
847  * @complq: Associated completion queue pointer, split queue only
848  * @num_completions_pending: Total number of completions pending for the
849  *			     completion queue, acculumated for all TX queues
850  *			     associated with that completion queue.
851  *
852  * Between singleq and splitq, a txq_group is largely the same except for the
853  * complq. In splitq a single complq is responsible for handling completions
854  * for some number of txqs associated in this txq_group.
855  */
856 struct idpf_txq_group {
857 	struct idpf_vport *vport;
858 
859 	u16 num_txq;
860 	struct idpf_queue *txqs[IDPF_LARGE_MAX_Q];
861 
862 	struct idpf_queue *complq;
863 
864 	u32 num_completions_pending;
865 };
866 
867 /**
868  * idpf_size_to_txd_count - Get number of descriptors needed for large Tx frag
869  * @size: transmit request size in bytes
870  *
871  * In the case where a large frag (>= 16K) needs to be split across multiple
872  * descriptors, we need to assume that we can have no more than 12K of data
873  * per descriptor due to hardware alignment restrictions (4K alignment).
874  */
875 static inline u32 idpf_size_to_txd_count(unsigned int size)
876 {
877 	return DIV_ROUND_UP(size, IDPF_TX_MAX_DESC_DATA_ALIGNED);
878 }
879 
880 /**
881  * idpf_tx_singleq_build_ctob - populate command tag offset and size
882  * @td_cmd: Command to be filled in desc
883  * @td_offset: Offset to be filled in desc
884  * @size: Size of the buffer
885  * @td_tag: td tag to be filled
886  *
887  * Returns the 64 bit value populated with the input parameters
888  */
889 static inline __le64 idpf_tx_singleq_build_ctob(u64 td_cmd, u64 td_offset,
890 						unsigned int size, u64 td_tag)
891 {
892 	return cpu_to_le64(IDPF_TX_DESC_DTYPE_DATA |
893 			   (td_cmd << IDPF_TXD_QW1_CMD_S) |
894 			   (td_offset << IDPF_TXD_QW1_OFFSET_S) |
895 			   ((u64)size << IDPF_TXD_QW1_TX_BUF_SZ_S) |
896 			   (td_tag << IDPF_TXD_QW1_L2TAG1_S));
897 }
898 
899 void idpf_tx_splitq_build_ctb(union idpf_tx_flex_desc *desc,
900 			      struct idpf_tx_splitq_params *params,
901 			      u16 td_cmd, u16 size);
902 void idpf_tx_splitq_build_flow_desc(union idpf_tx_flex_desc *desc,
903 				    struct idpf_tx_splitq_params *params,
904 				    u16 td_cmd, u16 size);
905 /**
906  * idpf_tx_splitq_build_desc - determine which type of data descriptor to build
907  * @desc: descriptor to populate
908  * @params: pointer to tx params struct
909  * @td_cmd: command to be filled in desc
910  * @size: size of buffer
911  */
912 static inline void idpf_tx_splitq_build_desc(union idpf_tx_flex_desc *desc,
913 					     struct idpf_tx_splitq_params *params,
914 					     u16 td_cmd, u16 size)
915 {
916 	if (params->dtype == IDPF_TX_DESC_DTYPE_FLEX_L2TAG1_L2TAG2)
917 		idpf_tx_splitq_build_ctb(desc, params, td_cmd, size);
918 	else
919 		idpf_tx_splitq_build_flow_desc(desc, params, td_cmd, size);
920 }
921 
922 /**
923  * idpf_alloc_page - Allocate a new RX buffer from the page pool
924  * @pool: page_pool to allocate from
925  * @buf: metadata struct to populate with page info
926  * @buf_size: 2K or 4K
927  *
928  * Returns &dma_addr_t to be passed to HW for Rx, %DMA_MAPPING_ERROR otherwise.
929  */
930 static inline dma_addr_t idpf_alloc_page(struct page_pool *pool,
931 					 struct idpf_rx_buf *buf,
932 					 unsigned int buf_size)
933 {
934 	if (buf_size == IDPF_RX_BUF_2048)
935 		buf->page = page_pool_dev_alloc_frag(pool, &buf->page_offset,
936 						     buf_size);
937 	else
938 		buf->page = page_pool_dev_alloc_pages(pool);
939 
940 	if (!buf->page)
941 		return DMA_MAPPING_ERROR;
942 
943 	buf->truesize = buf_size;
944 
945 	return page_pool_get_dma_addr(buf->page) + buf->page_offset +
946 	       pool->p.offset;
947 }
948 
949 /**
950  * idpf_rx_put_page - Return RX buffer page to pool
951  * @rx_buf: RX buffer metadata struct
952  */
953 static inline void idpf_rx_put_page(struct idpf_rx_buf *rx_buf)
954 {
955 	page_pool_put_page(rx_buf->page->pp, rx_buf->page,
956 			   rx_buf->truesize, true);
957 	rx_buf->page = NULL;
958 }
959 
960 /**
961  * idpf_rx_sync_for_cpu - Synchronize DMA buffer
962  * @rx_buf: RX buffer metadata struct
963  * @len: frame length from descriptor
964  */
965 static inline void idpf_rx_sync_for_cpu(struct idpf_rx_buf *rx_buf, u32 len)
966 {
967 	struct page *page = rx_buf->page;
968 	struct page_pool *pp = page->pp;
969 
970 	dma_sync_single_range_for_cpu(pp->p.dev,
971 				      page_pool_get_dma_addr(page),
972 				      rx_buf->page_offset + pp->p.offset, len,
973 				      page_pool_get_dma_dir(pp));
974 }
975 
976 int idpf_vport_singleq_napi_poll(struct napi_struct *napi, int budget);
977 void idpf_vport_init_num_qs(struct idpf_vport *vport,
978 			    struct virtchnl2_create_vport *vport_msg);
979 void idpf_vport_calc_num_q_desc(struct idpf_vport *vport);
980 int idpf_vport_calc_total_qs(struct idpf_adapter *adapter, u16 vport_index,
981 			     struct virtchnl2_create_vport *vport_msg,
982 			     struct idpf_vport_max_q *max_q);
983 void idpf_vport_calc_num_q_groups(struct idpf_vport *vport);
984 int idpf_vport_queues_alloc(struct idpf_vport *vport);
985 void idpf_vport_queues_rel(struct idpf_vport *vport);
986 void idpf_vport_intr_rel(struct idpf_vport *vport);
987 int idpf_vport_intr_alloc(struct idpf_vport *vport);
988 void idpf_vport_intr_update_itr_ena_irq(struct idpf_q_vector *q_vector);
989 void idpf_vport_intr_deinit(struct idpf_vport *vport);
990 int idpf_vport_intr_init(struct idpf_vport *vport);
991 enum pkt_hash_types idpf_ptype_to_htype(const struct idpf_rx_ptype_decoded *decoded);
992 int idpf_config_rss(struct idpf_vport *vport);
993 int idpf_init_rss(struct idpf_vport *vport);
994 void idpf_deinit_rss(struct idpf_vport *vport);
995 int idpf_rx_bufs_init_all(struct idpf_vport *vport);
996 void idpf_rx_add_frag(struct idpf_rx_buf *rx_buf, struct sk_buff *skb,
997 		      unsigned int size);
998 struct sk_buff *idpf_rx_construct_skb(struct idpf_queue *rxq,
999 				      struct idpf_rx_buf *rx_buf,
1000 				      unsigned int size);
1001 bool idpf_init_rx_buf_hw_alloc(struct idpf_queue *rxq, struct idpf_rx_buf *buf);
1002 void idpf_rx_buf_hw_update(struct idpf_queue *rxq, u32 val);
1003 void idpf_tx_buf_hw_update(struct idpf_queue *tx_q, u32 val,
1004 			   bool xmit_more);
1005 unsigned int idpf_size_to_txd_count(unsigned int size);
1006 netdev_tx_t idpf_tx_drop_skb(struct idpf_queue *tx_q, struct sk_buff *skb);
1007 void idpf_tx_dma_map_error(struct idpf_queue *txq, struct sk_buff *skb,
1008 			   struct idpf_tx_buf *first, u16 ring_idx);
1009 unsigned int idpf_tx_desc_count_required(struct idpf_queue *txq,
1010 					 struct sk_buff *skb);
1011 bool idpf_chk_linearize(struct sk_buff *skb, unsigned int max_bufs,
1012 			unsigned int count);
1013 int idpf_tx_maybe_stop_common(struct idpf_queue *tx_q, unsigned int size);
1014 void idpf_tx_timeout(struct net_device *netdev, unsigned int txqueue);
1015 netdev_tx_t idpf_tx_splitq_start(struct sk_buff *skb,
1016 				 struct net_device *netdev);
1017 netdev_tx_t idpf_tx_singleq_start(struct sk_buff *skb,
1018 				  struct net_device *netdev);
1019 bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_queue *rxq,
1020 				      u16 cleaned_count);
1021 int idpf_tso(struct sk_buff *skb, struct idpf_tx_offload_params *off);
1022 
1023 #endif /* !_IDPF_TXRX_H_ */
1024