xref: /linux/drivers/net/ethernet/google/gve/gve.h (revision 07993df560917357610e0625a9a2e7531c3211fc)
1 /* SPDX-License-Identifier: (GPL-2.0 OR MIT)
2  * Google virtual Ethernet (gve) driver
3  *
4  * Copyright (C) 2015-2024 Google LLC
5  */
6 
7 #ifndef _GVE_H_
8 #define _GVE_H_
9 
10 #include <linux/dma-mapping.h>
11 #include <linux/dmapool.h>
12 #include <linux/ethtool_netlink.h>
13 #include <linux/netdevice.h>
14 #include <linux/net_tstamp.h>
15 #include <linux/pci.h>
16 #include <linux/ptp_clock_kernel.h>
17 #include <linux/u64_stats_sync.h>
18 #include <net/page_pool/helpers.h>
19 #include <net/xdp.h>
20 
21 #include "gve_desc.h"
22 #include "gve_desc_dqo.h"
23 
24 #ifndef PCI_VENDOR_ID_GOOGLE
25 #define PCI_VENDOR_ID_GOOGLE	0x1ae0
26 #endif
27 
28 #define PCI_DEV_ID_GVNIC	0x0042
29 
30 #define GVE_REGISTER_BAR	0
31 #define GVE_DOORBELL_BAR	2
32 
33 /* Driver can alloc up to 2 segments for the header and 2 for the payload. */
34 #define GVE_TX_MAX_IOVEC	4
35 /* 1 for management, 1 for rx, 1 for tx */
36 #define GVE_MIN_MSIX 3
37 
38 /* Numbers of gve tx/rx stats in stats report. */
39 #define GVE_TX_STATS_REPORT_NUM	6
40 #define GVE_RX_STATS_REPORT_NUM	2
41 
42 /* Interval to schedule a stats report update, 20000ms. */
43 #define GVE_STATS_REPORT_TIMER_PERIOD	20000
44 
45 /* Numbers of NIC tx/rx stats in stats report. */
46 #define NIC_TX_STATS_REPORT_NUM	0
47 #define NIC_RX_STATS_REPORT_NUM	4
48 
49 #define GVE_ADMINQ_BUFFER_SIZE 4096
50 
51 #define GVE_DATA_SLOT_ADDR_PAGE_MASK (~(PAGE_SIZE - 1))
52 
53 /* PTYPEs are always 10 bits. */
54 #define GVE_NUM_PTYPES	1024
55 
56 /* Default minimum ring size */
57 #define GVE_DEFAULT_MIN_TX_RING_SIZE 256
58 #define GVE_DEFAULT_MIN_RX_RING_SIZE 512
59 
60 #define GVE_DEFAULT_RX_BUFFER_SIZE 2048
61 
62 #define GVE_XDP_RX_BUFFER_SIZE_DQO 4096
63 
64 #define GVE_DEFAULT_RX_BUFFER_OFFSET 2048
65 
66 #define GVE_PAGE_POOL_SIZE_MULTIPLIER 4
67 
68 #define GVE_FLOW_RULES_CACHE_SIZE \
69 	(GVE_ADMINQ_BUFFER_SIZE / sizeof(struct gve_adminq_queried_flow_rule))
70 #define GVE_FLOW_RULE_IDS_CACHE_SIZE \
71 	(GVE_ADMINQ_BUFFER_SIZE / sizeof(((struct gve_adminq_queried_flow_rule *)0)->location))
72 
73 #define GVE_RSS_KEY_SIZE	40
74 #define GVE_RSS_INDIR_SIZE	128
75 
76 #define GVE_XDP_ACTIONS 5
77 
78 #define GVE_GQ_TX_MIN_PKT_DESC_BYTES 182
79 
80 #define GVE_DEFAULT_HEADER_BUFFER_SIZE 128
81 
82 /* Maximum TSO size supported on DQO */
83 #define GVE_DQO_TX_MAX	0x3FFFF
84 
85 #define GVE_TX_BUF_SHIFT_DQO 11
86 
87 /* 2K buffers for DQO-QPL */
88 #define GVE_TX_BUF_SIZE_DQO BIT(GVE_TX_BUF_SHIFT_DQO)
89 #define GVE_TX_BUFS_PER_PAGE_DQO (PAGE_SIZE >> GVE_TX_BUF_SHIFT_DQO)
90 #define GVE_MAX_TX_BUFS_PER_PKT (DIV_ROUND_UP(GVE_DQO_TX_MAX, GVE_TX_BUF_SIZE_DQO))
91 
92 /* If number of free/recyclable buffers are less than this threshold; driver
93  * allocs and uses a non-qpl page on the receive path of DQO QPL to free
94  * up buffers.
95  * Value is set big enough to post at least 3 64K LRO packet via 2K buffer to NIC.
96  */
97 #define GVE_DQO_QPL_ONDEMAND_ALLOC_THRESHOLD 96
98 
99 #define GVE_DQO_RX_HWTSTAMP_VALID 0x1
100 
101 /* Each slot in the desc ring has a 1:1 mapping to a slot in the data ring */
102 struct gve_rx_desc_queue {
103 	struct gve_rx_desc *desc_ring; /* the descriptor ring */
104 	dma_addr_t bus; /* the bus for the desc_ring */
105 	u8 seqno; /* the next expected seqno for this desc*/
106 };
107 
108 /* The page info for a single slot in the RX data queue */
109 struct gve_rx_slot_page_info {
110 	/* netmem is used for DQO RDA mode
111 	 * page is used in all other modes
112 	 */
113 	union {
114 		struct page *page;
115 		netmem_ref netmem;
116 	};
117 	void *page_address;
118 	u32 page_offset; /* offset to write to in page */
119 	unsigned int buf_size;
120 	int pagecnt_bias; /* expected pagecnt if only the driver has a ref */
121 	u16 pad; /* adjustment for rx padding */
122 	u8 can_flip; /* tracks if the networking stack is using the page */
123 };
124 
125 /* A list of pages registered with the device during setup and used by a queue
126  * as buffers
127  */
128 struct gve_queue_page_list {
129 	u32 id; /* unique id */
130 	u32 num_entries;
131 	struct page **pages; /* list of num_entries pages */
132 	dma_addr_t *page_buses; /* the dma addrs of the pages */
133 };
134 
135 /* Each slot in the data ring has a 1:1 mapping to a slot in the desc ring */
136 struct gve_rx_data_queue {
137 	union gve_rx_data_slot *data_ring; /* read by NIC */
138 	dma_addr_t data_bus; /* dma mapping of the slots */
139 	struct gve_rx_slot_page_info *page_info; /* page info of the buffers */
140 	struct gve_queue_page_list *qpl; /* qpl assigned to this queue */
141 	u8 raw_addressing; /* use raw_addressing? */
142 };
143 
144 struct gve_priv;
145 
146 /* RX buffer queue for posting buffers to HW.
147  * Each RX (completion) queue has a corresponding buffer queue.
148  */
149 struct gve_rx_buf_queue_dqo {
150 	struct gve_rx_desc_dqo *desc_ring;
151 	dma_addr_t bus;
152 	u32 head; /* Pointer to start cleaning buffers at. */
153 	u32 tail; /* Last posted buffer index + 1 */
154 	u32 mask; /* Mask for indices to the size of the ring */
155 };
156 
157 /* RX completion queue to receive packets from HW. */
158 struct gve_rx_compl_queue_dqo {
159 	struct gve_rx_compl_desc_dqo *desc_ring;
160 	dma_addr_t bus;
161 
162 	/* Number of slots which did not have a buffer posted yet. We should not
163 	 * post more buffers than the queue size to avoid HW overrunning the
164 	 * queue.
165 	 */
166 	int num_free_slots;
167 
168 	/* HW uses a "generation bit" to notify SW of new descriptors. When a
169 	 * descriptor's generation bit is different from the current generation,
170 	 * that descriptor is ready to be consumed by SW.
171 	 */
172 	u8 cur_gen_bit;
173 
174 	/* Pointer into desc_ring where the next completion descriptor will be
175 	 * received.
176 	 */
177 	u32 head;
178 	u32 mask; /* Mask for indices to the size of the ring */
179 };
180 
181 struct gve_header_buf {
182 	u8 *data;
183 	dma_addr_t addr;
184 };
185 
186 /* Stores state for tracking buffers posted to HW */
187 struct gve_rx_buf_state_dqo {
188 	/* The page posted to HW. */
189 	struct gve_rx_slot_page_info page_info;
190 
191 	/* XSK buffer */
192 	struct xdp_buff *xsk_buff;
193 
194 	/* The DMA address corresponding to `page_info`. */
195 	dma_addr_t addr;
196 
197 	/* Last offset into the page when it only had a single reference, at
198 	 * which point every other offset is free to be reused.
199 	 */
200 	u32 last_single_ref_offset;
201 
202 	/* Linked list index to next element in the list, or -1 if none */
203 	s16 next;
204 };
205 
206 /* Wrapper for XDP Rx metadata */
207 struct gve_xdp_buff {
208 	struct xdp_buff xdp;
209 	struct gve_priv *gve;
210 	const struct gve_rx_compl_desc_dqo *compl_desc;
211 };
212 
213 /* `head` and `tail` are indices into an array, or -1 if empty. */
214 struct gve_index_list {
215 	s16 head;
216 	s16 tail;
217 };
218 
219 /* A single received packet split across multiple buffers may be
220  * reconstructed using the information in this structure.
221  */
222 struct gve_rx_ctx {
223 	/* head and tail of skb chain for the current packet or NULL if none */
224 	struct sk_buff *skb_head;
225 	struct sk_buff *skb_tail;
226 	u32 total_size;
227 	u8 frag_cnt;
228 	bool drop_pkt;
229 };
230 
231 struct gve_rx_cnts {
232 	u32 ok_pkt_bytes;
233 	u16 ok_pkt_cnt;
234 	u16 total_pkt_cnt;
235 	u16 cont_pkt_cnt;
236 	u16 desc_err_pkt_cnt;
237 };
238 
239 /* Contains datapath state used to represent an RX queue. */
240 struct gve_rx_ring {
241 	struct gve_priv *gve;
242 
243 	u16 packet_buffer_size;		/* Size of buffer posted to NIC */
244 	u16 packet_buffer_truesize;	/* Total size of RX buffer */
245 	u16 rx_headroom;
246 
247 	union {
248 		/* GQI fields */
249 		struct {
250 			struct gve_rx_desc_queue desc;
251 			struct gve_rx_data_queue data;
252 
253 			/* threshold for posting new buffs and descs */
254 			u32 db_threshold;
255 
256 			u32 qpl_copy_pool_mask;
257 			u32 qpl_copy_pool_head;
258 			struct gve_rx_slot_page_info *qpl_copy_pool;
259 		};
260 
261 		/* DQO fields. */
262 		struct {
263 			struct gve_rx_buf_queue_dqo bufq;
264 			struct gve_rx_compl_queue_dqo complq;
265 
266 			struct gve_rx_buf_state_dqo *buf_states;
267 			u16 num_buf_states;
268 
269 			/* Linked list of gve_rx_buf_state_dqo. Index into
270 			 * buf_states, or -1 if empty.
271 			 */
272 			s16 free_buf_states;
273 
274 			/* Linked list of gve_rx_buf_state_dqo. Indexes into
275 			 * buf_states, or -1 if empty.
276 			 *
277 			 * This list contains buf_states which are pointing to
278 			 * valid buffers.
279 			 *
280 			 * We use a FIFO here in order to increase the
281 			 * probability that buffers can be reused by increasing
282 			 * the time between usages.
283 			 */
284 			struct gve_index_list recycled_buf_states;
285 
286 			/* Linked list of gve_rx_buf_state_dqo. Indexes into
287 			 * buf_states, or -1 if empty.
288 			 *
289 			 * This list contains buf_states which have buffers
290 			 * which cannot be reused yet.
291 			 */
292 			struct gve_index_list used_buf_states;
293 
294 			/* qpl assigned to this queue */
295 			struct gve_queue_page_list *qpl;
296 
297 			/* index into queue page list */
298 			u32 next_qpl_page_idx;
299 
300 			/* track number of used buffers */
301 			u16 used_buf_states_cnt;
302 
303 			/* Address info of the buffers for header-split */
304 			struct gve_header_buf hdr_bufs;
305 
306 			struct page_pool *page_pool;
307 		} dqo;
308 	};
309 
310 	u64 rbytes; /* free-running bytes received */
311 	u64 rx_hsplit_bytes; /* free-running header bytes received */
312 	u64 rpackets; /* free-running packets received */
313 	u32 cnt; /* free-running total number of completed packets */
314 	u32 fill_cnt; /* free-running total number of descs and buffs posted */
315 	u32 mask; /* masks the cnt and fill_cnt to the size of the ring */
316 	u64 rx_hsplit_pkt; /* free-running packets with headers split */
317 	u64 rx_copybreak_pkt; /* free-running count of copybreak packets */
318 	u64 rx_copied_pkt; /* free-running total number of copied packets */
319 	u64 rx_skb_alloc_fail; /* free-running count of skb alloc fails */
320 	u64 rx_buf_alloc_fail; /* free-running count of buffer alloc fails */
321 	u64 rx_desc_err_dropped_pkt; /* free-running count of packets dropped by descriptor error */
322 	/* free-running count of unsplit packets due to header buffer overflow or hdr_len is 0 */
323 	u64 rx_hsplit_unsplit_pkt;
324 	u64 rx_cont_packet_cnt; /* free-running multi-fragment packets received */
325 	u64 rx_frag_flip_cnt; /* free-running count of rx segments where page_flip was used */
326 	u64 rx_frag_copy_cnt; /* free-running count of rx segments copied */
327 	u64 rx_frag_alloc_cnt; /* free-running count of rx page allocations */
328 	u64 xdp_tx_errors;
329 	u64 xdp_redirect_errors;
330 	u64 xdp_alloc_fails;
331 	u64 xdp_actions[GVE_XDP_ACTIONS];
332 	u32 q_num; /* queue index */
333 	u32 ntfy_id; /* notification block index */
334 	struct gve_queue_resources *q_resources; /* head and tail pointer idx */
335 	dma_addr_t q_resources_bus; /* dma address for the queue resources */
336 	struct u64_stats_sync statss; /* sync stats for 32bit archs */
337 
338 	struct gve_rx_ctx ctx; /* Info for packet currently being processed in this ring. */
339 
340 	/* XDP stuff */
341 	struct xdp_rxq_info xdp_rxq;
342 	struct xsk_buff_pool *xsk_pool;
343 	struct page_frag_cache page_cache; /* Page cache to allocate XDP frames */
344 };
345 
346 /* A TX desc ring entry */
347 union gve_tx_desc {
348 	struct gve_tx_pkt_desc pkt; /* first desc for a packet */
349 	struct gve_tx_mtd_desc mtd; /* optional metadata descriptor */
350 	struct gve_tx_seg_desc seg; /* subsequent descs for a packet */
351 };
352 
353 /* Tracks the memory in the fifo occupied by a segment of a packet */
354 struct gve_tx_iovec {
355 	u32 iov_offset; /* offset into this segment */
356 	u32 iov_len; /* length */
357 	u32 iov_padding; /* padding associated with this segment */
358 };
359 
360 /* Tracks the memory in the fifo occupied by the skb. Mapped 1:1 to a desc
361  * ring entry but only used for a pkt_desc not a seg_desc
362  */
363 struct gve_tx_buffer_state {
364 	union {
365 		struct sk_buff *skb; /* skb for this pkt */
366 		struct xdp_frame *xdp_frame; /* xdp_frame */
367 	};
368 	struct {
369 		u16 size; /* size of xmitted xdp pkt */
370 		u8 is_xsk; /* xsk buff */
371 	} xdp;
372 	union {
373 		struct gve_tx_iovec iov[GVE_TX_MAX_IOVEC]; /* segments of this pkt */
374 		struct {
375 			DEFINE_DMA_UNMAP_ADDR(dma);
376 			DEFINE_DMA_UNMAP_LEN(len);
377 		};
378 	};
379 };
380 
381 /* A TX buffer - each queue has one */
382 struct gve_tx_fifo {
383 	void *base; /* address of base of FIFO */
384 	u32 size; /* total size */
385 	atomic_t available; /* how much space is still available */
386 	u32 head; /* offset to write at */
387 	struct gve_queue_page_list *qpl; /* QPL mapped into this FIFO */
388 };
389 
390 /* TX descriptor for DQO format */
391 union gve_tx_desc_dqo {
392 	struct gve_tx_pkt_desc_dqo pkt;
393 	struct gve_tx_tso_context_desc_dqo tso_ctx;
394 	struct gve_tx_general_context_desc_dqo general_ctx;
395 };
396 
397 enum gve_packet_state {
398 	/* Packet is in free list, available to be allocated.
399 	 * This should always be zero since state is not explicitly initialized.
400 	 */
401 	GVE_PACKET_STATE_UNALLOCATED,
402 	/* Packet is expecting a regular data completion or miss completion */
403 	GVE_PACKET_STATE_PENDING_DATA_COMPL,
404 	/* Packet has received a miss completion and is expecting a
405 	 * re-injection completion.
406 	 */
407 	GVE_PACKET_STATE_PENDING_REINJECT_COMPL,
408 	/* No valid completion received within the specified timeout. */
409 	GVE_PACKET_STATE_TIMED_OUT_COMPL,
410 	/* XSK pending packet has received a packet/reinjection completion, or
411 	 * has timed out. At this point, the pending packet can be counted by
412 	 * xsk_tx_complete and freed.
413 	 */
414 	GVE_PACKET_STATE_XSK_COMPLETE,
415 };
416 
417 enum gve_tx_pending_packet_dqo_type {
418 	GVE_TX_PENDING_PACKET_DQO_SKB,
419 	GVE_TX_PENDING_PACKET_DQO_XDP_FRAME,
420 	GVE_TX_PENDING_PACKET_DQO_XSK,
421 };
422 
423 struct gve_tx_pending_packet_dqo {
424 	union {
425 		struct sk_buff *skb;
426 		struct xdp_frame *xdpf;
427 	};
428 
429 	/* 0th element corresponds to the linear portion of `skb`, should be
430 	 * unmapped with `dma_unmap_single`.
431 	 *
432 	 * All others correspond to `skb`'s frags and should be unmapped with
433 	 * `dma_unmap_page`.
434 	 */
435 	union {
436 		struct {
437 			DEFINE_DMA_UNMAP_ADDR(dma[MAX_SKB_FRAGS + 1]);
438 			DEFINE_DMA_UNMAP_LEN(len[MAX_SKB_FRAGS + 1]);
439 		};
440 		s16 tx_qpl_buf_ids[GVE_MAX_TX_BUFS_PER_PKT];
441 	};
442 
443 	u16 num_bufs;
444 
445 	/* Linked list index to next element in the list, or -1 if none */
446 	s16 next;
447 
448 	/* Linked list index to prev element in the list, or -1 if none.
449 	 * Used for tracking either outstanding miss completions or prematurely
450 	 * freed packets.
451 	 */
452 	s16 prev;
453 
454 	/* Identifies the current state of the packet as defined in
455 	 * `enum gve_packet_state`.
456 	 */
457 	u8 state : 3;
458 
459 	/* gve_tx_pending_packet_dqo_type */
460 	u8 type : 2;
461 
462 	/* If packet is an outstanding miss completion, then the packet is
463 	 * freed if the corresponding re-injection completion is not received
464 	 * before kernel jiffies exceeds timeout_jiffies.
465 	 */
466 	unsigned long timeout_jiffies;
467 };
468 
469 /* Contains datapath state used to represent a TX queue. */
470 struct gve_tx_ring {
471 	/* Cacheline 0 -- Accessed & dirtied during transmit */
472 	union {
473 		/* GQI fields */
474 		struct {
475 			struct gve_tx_fifo tx_fifo;
476 			u32 req; /* driver tracked head pointer */
477 			u32 done; /* driver tracked tail pointer */
478 		};
479 
480 		/* DQO fields. */
481 		struct {
482 			/* Spinlock for XDP tx traffic */
483 			spinlock_t xdp_lock;
484 
485 			/* Linked list of gve_tx_pending_packet_dqo. Index into
486 			 * pending_packets, or -1 if empty.
487 			 *
488 			 * This is a consumer list owned by the TX path. When it
489 			 * runs out, the producer list is stolen from the
490 			 * completion handling path
491 			 * (dqo_compl.free_pending_packets).
492 			 */
493 			s16 free_pending_packets;
494 
495 			/* Cached value of `dqo_compl.hw_tx_head` */
496 			u32 head;
497 			u32 tail; /* Last posted buffer index + 1 */
498 
499 			/* Index of the last descriptor with "report event" bit
500 			 * set.
501 			 */
502 			u32 last_re_idx;
503 
504 			/* free running number of packet buf descriptors posted */
505 			u16 posted_packet_desc_cnt;
506 			/* free running number of packet buf descriptors completed */
507 			u16 completed_packet_desc_cnt;
508 
509 			/* QPL fields */
510 			struct {
511 			       /* Linked list of gve_tx_buf_dqo. Index into
512 				* tx_qpl_buf_next, or -1 if empty.
513 				*
514 				* This is a consumer list owned by the TX path. When it
515 				* runs out, the producer list is stolen from the
516 				* completion handling path
517 				* (dqo_compl.free_tx_qpl_buf_head).
518 				*/
519 				s16 free_tx_qpl_buf_head;
520 
521 			       /* Free running count of the number of QPL tx buffers
522 				* allocated
523 				*/
524 				u32 alloc_tx_qpl_buf_cnt;
525 
526 				/* Cached value of `dqo_compl.free_tx_qpl_buf_cnt` */
527 				u32 free_tx_qpl_buf_cnt;
528 			};
529 
530 			atomic_t xsk_reorder_queue_tail;
531 		} dqo_tx;
532 	};
533 
534 	/* Cacheline 1 -- Accessed & dirtied during gve_clean_tx_done */
535 	union {
536 		/* GQI fields */
537 		struct {
538 			/* Spinlock for when cleanup in progress */
539 			spinlock_t clean_lock;
540 			/* Spinlock for XDP tx traffic */
541 			spinlock_t xdp_lock;
542 		};
543 
544 		/* DQO fields. */
545 		struct {
546 			u32 head; /* Last read on compl_desc */
547 
548 			/* Tracks the current gen bit of compl_q */
549 			u8 cur_gen_bit;
550 
551 			/* Linked list of gve_tx_pending_packet_dqo. Index into
552 			 * pending_packets, or -1 if empty.
553 			 *
554 			 * This is the producer list, owned by the completion
555 			 * handling path. When the consumer list
556 			 * (dqo_tx.free_pending_packets) is runs out, this list
557 			 * will be stolen.
558 			 */
559 			atomic_t free_pending_packets;
560 
561 			/* Last TX ring index fetched by HW */
562 			atomic_t hw_tx_head;
563 
564 			u16 xsk_reorder_queue_head;
565 			u16 xsk_reorder_queue_tail;
566 
567 			/* List to track pending packets which received a miss
568 			 * completion but not a corresponding reinjection.
569 			 */
570 			struct gve_index_list miss_completions;
571 
572 			/* List to track pending packets that were completed
573 			 * before receiving a valid completion because they
574 			 * reached a specified timeout.
575 			 */
576 			struct gve_index_list timed_out_completions;
577 
578 			/* QPL fields */
579 			struct {
580 				/* Linked list of gve_tx_buf_dqo. Index into
581 				 * tx_qpl_buf_next, or -1 if empty.
582 				 *
583 				 * This is the producer list, owned by the completion
584 				 * handling path. When the consumer list
585 				 * (dqo_tx.free_tx_qpl_buf_head) is runs out, this list
586 				 * will be stolen.
587 				 */
588 				atomic_t free_tx_qpl_buf_head;
589 
590 				/* Free running count of the number of tx buffers
591 				 * freed
592 				 */
593 				atomic_t free_tx_qpl_buf_cnt;
594 			};
595 		} dqo_compl;
596 	} ____cacheline_aligned;
597 	u64 pkt_done; /* free-running - total packets completed */
598 	u64 bytes_done; /* free-running - total bytes completed */
599 	u64 dropped_pkt; /* free-running - total packets dropped */
600 	u64 dma_mapping_error; /* count of dma mapping errors */
601 
602 	/* Cacheline 2 -- Read-mostly fields */
603 	union {
604 		/* GQI fields */
605 		struct {
606 			union gve_tx_desc *desc;
607 
608 			/* Maps 1:1 to a desc */
609 			struct gve_tx_buffer_state *info;
610 		};
611 
612 		/* DQO fields. */
613 		struct {
614 			union gve_tx_desc_dqo *tx_ring;
615 			struct gve_tx_compl_desc *compl_ring;
616 
617 			struct gve_tx_pending_packet_dqo *pending_packets;
618 			s16 num_pending_packets;
619 
620 			u16 *xsk_reorder_queue;
621 
622 			u32 complq_mask; /* complq size is complq_mask + 1 */
623 
624 			/* QPL fields */
625 			struct {
626 				/* qpl assigned to this queue */
627 				struct gve_queue_page_list *qpl;
628 
629 				/* Each QPL page is divided into TX bounce buffers
630 				 * of size GVE_TX_BUF_SIZE_DQO. tx_qpl_buf_next is
631 				 * an array to manage linked lists of TX buffers.
632 				 * An entry j at index i implies that j'th buffer
633 				 * is next on the list after i
634 				 */
635 				s16 *tx_qpl_buf_next;
636 				u32 num_tx_qpl_bufs;
637 			};
638 		} dqo;
639 	} ____cacheline_aligned;
640 	struct netdev_queue *netdev_txq;
641 	struct gve_queue_resources *q_resources; /* head and tail pointer idx */
642 	struct device *dev;
643 	u32 mask; /* masks req and done down to queue size */
644 	u8 raw_addressing; /* use raw_addressing? */
645 
646 	/* Slow-path fields */
647 	u32 q_num ____cacheline_aligned; /* queue idx */
648 	u32 stop_queue; /* count of queue stops */
649 	u32 wake_queue; /* count of queue wakes */
650 	u32 queue_timeout; /* count of queue timeouts */
651 	u32 ntfy_id; /* notification block index */
652 	u32 last_kick_msec; /* Last time the queue was kicked */
653 	dma_addr_t bus; /* dma address of the descr ring */
654 	dma_addr_t q_resources_bus; /* dma address of the queue resources */
655 	dma_addr_t complq_bus_dqo; /* dma address of the dqo.compl_ring */
656 	struct u64_stats_sync statss; /* sync stats for 32bit archs */
657 	struct xsk_buff_pool *xsk_pool;
658 	u64 xdp_xsk_sent;
659 	u64 xdp_xmit;
660 	u64 xdp_xmit_errors;
661 } ____cacheline_aligned;
662 
663 /* Wraps the info for one irq including the napi struct and the queues
664  * associated with that irq.
665  */
666 struct gve_notify_block {
667 	__be32 *irq_db_index; /* pointer to idx into Bar2 */
668 	char name[IFNAMSIZ + 16]; /* name registered with the kernel */
669 	struct napi_struct napi; /* kernel napi struct for this block */
670 	struct gve_priv *priv;
671 	struct gve_tx_ring *tx; /* tx rings on this block */
672 	struct gve_rx_ring *rx; /* rx rings on this block */
673 	u32 irq;
674 };
675 
676 /* Tracks allowed and current rx queue settings */
677 struct gve_rx_queue_config {
678 	u16 max_queues;
679 	u16 num_queues;
680 	u16 packet_buffer_size;
681 };
682 
683 /* Tracks allowed and current tx queue settings */
684 struct gve_tx_queue_config {
685 	u16 max_queues;
686 	u16 num_queues; /* number of TX queues, excluding XDP queues */
687 	u16 num_xdp_queues;
688 };
689 
690 /* Tracks the available and used qpl IDs */
691 struct gve_qpl_config {
692 	u32 qpl_map_size; /* map memory size */
693 	unsigned long *qpl_id_map; /* bitmap of used qpl ids */
694 };
695 
696 struct gve_irq_db {
697 	__be32 index;
698 } ____cacheline_aligned;
699 
700 struct gve_ptype {
701 	u8 l3_type;  /* `gve_l3_type` in gve_adminq.h */
702 	u8 l4_type;  /* `gve_l4_type` in gve_adminq.h */
703 };
704 
705 struct gve_ptype_lut {
706 	struct gve_ptype ptypes[GVE_NUM_PTYPES];
707 };
708 
709 /* Parameters for allocating resources for tx queues */
710 struct gve_tx_alloc_rings_cfg {
711 	struct gve_tx_queue_config *qcfg;
712 	u16 pages_per_qpl;
713 
714 	u16 num_xdp_rings;
715 
716 	u16 ring_size;
717 	bool raw_addressing;
718 
719 	/* Allocated resources are returned here */
720 	struct gve_tx_ring *tx;
721 };
722 
723 /* Parameters for allocating resources for rx queues */
724 struct gve_rx_alloc_rings_cfg {
725 	/* tx config is also needed to determine QPL ids */
726 	struct gve_rx_queue_config *qcfg_rx;
727 	struct gve_tx_queue_config *qcfg_tx;
728 	u16 pages_per_qpl;
729 
730 	u16 ring_size;
731 	u16 packet_buffer_size;
732 	bool raw_addressing;
733 	bool enable_header_split;
734 	bool reset_rss;
735 	bool xdp;
736 
737 	/* Allocated resources are returned here */
738 	struct gve_rx_ring *rx;
739 };
740 
741 /* GVE_QUEUE_FORMAT_UNSPECIFIED must be zero since 0 is the default value
742  * when the entire configure_device_resources command is zeroed out and the
743  * queue_format is not specified.
744  */
745 enum gve_queue_format {
746 	GVE_QUEUE_FORMAT_UNSPECIFIED	= 0x0,
747 	GVE_GQI_RDA_FORMAT		= 0x1,
748 	GVE_GQI_QPL_FORMAT		= 0x2,
749 	GVE_DQO_RDA_FORMAT		= 0x3,
750 	GVE_DQO_QPL_FORMAT		= 0x4,
751 };
752 
753 struct gve_flow_spec {
754 	__be32 src_ip[4];
755 	__be32 dst_ip[4];
756 	union {
757 		struct {
758 			__be16 src_port;
759 			__be16 dst_port;
760 		};
761 		__be32 spi;
762 	};
763 	union {
764 		u8 tos;
765 		u8 tclass;
766 	};
767 };
768 
769 struct gve_flow_rule {
770 	u32 location;
771 	u16 flow_type;
772 	u16 action;
773 	struct gve_flow_spec key;
774 	struct gve_flow_spec mask;
775 };
776 
777 struct gve_flow_rules_cache {
778 	bool rules_cache_synced; /* False if the driver's rules_cache is outdated */
779 	struct gve_adminq_queried_flow_rule *rules_cache;
780 	__be32 *rule_ids_cache;
781 	/* The total number of queried rules that stored in the caches */
782 	u32 rules_cache_num;
783 	u32 rule_ids_cache_num;
784 };
785 
786 struct gve_rss_config {
787 	u8 *hash_key;
788 	u32 *hash_lut;
789 };
790 
791 struct gve_ptp {
792 	struct ptp_clock_info info;
793 	struct ptp_clock *clock;
794 	struct gve_priv *priv;
795 };
796 
797 struct gve_priv {
798 	struct net_device *dev;
799 	struct gve_tx_ring *tx; /* array of tx_cfg.num_queues */
800 	struct gve_rx_ring *rx; /* array of rx_cfg.num_queues */
801 	struct gve_notify_block *ntfy_blocks; /* array of num_ntfy_blks */
802 	struct gve_irq_db *irq_db_indices; /* array of num_ntfy_blks */
803 	dma_addr_t irq_db_indices_bus;
804 	struct msix_entry *msix_vectors; /* array of num_ntfy_blks + 1 */
805 	char mgmt_msix_name[IFNAMSIZ + 16];
806 	u32 mgmt_msix_idx;
807 	__be32 *counter_array; /* array of num_event_counters */
808 	dma_addr_t counter_array_bus;
809 
810 	u16 num_event_counters;
811 	u16 tx_desc_cnt; /* num desc per ring */
812 	u16 rx_desc_cnt; /* num desc per ring */
813 	u16 max_tx_desc_cnt;
814 	u16 max_rx_desc_cnt;
815 	u16 min_tx_desc_cnt;
816 	u16 min_rx_desc_cnt;
817 	bool modify_ring_size_enabled;
818 	bool default_min_ring_size;
819 	u16 tx_pages_per_qpl;
820 	u16 rx_pages_per_qpl;
821 	u64 max_registered_pages;
822 	u64 num_registered_pages; /* num pages registered with NIC */
823 	struct bpf_prog *xdp_prog; /* XDP BPF program */
824 	u32 rx_copybreak; /* copy packets smaller than this */
825 	u16 default_num_queues; /* default num queues to set up */
826 
827 	struct gve_tx_queue_config tx_cfg;
828 	struct gve_rx_queue_config rx_cfg;
829 	unsigned long *xsk_pools; /* bitmap of RX queues with XSK pools */
830 	u32 num_ntfy_blks; /* split between TX and RX so must be even */
831 	int numa_node;
832 
833 	struct gve_registers __iomem *reg_bar0; /* see gve_register.h */
834 	__be32 __iomem *db_bar2; /* "array" of doorbells */
835 	u32 msg_enable;	/* level for netif* netdev print macros	*/
836 	struct pci_dev *pdev;
837 
838 	/* metrics */
839 	u32 tx_timeo_cnt;
840 
841 	/* Admin queue - see gve_adminq.h*/
842 	union gve_adminq_command *adminq;
843 	dma_addr_t adminq_bus_addr;
844 	struct dma_pool *adminq_pool;
845 	struct mutex adminq_lock; /* Protects adminq command execution */
846 	u32 adminq_mask; /* masks prod_cnt to adminq size */
847 	u32 adminq_prod_cnt; /* free-running count of AQ cmds executed */
848 	u32 adminq_cmd_fail; /* free-running count of AQ cmds failed */
849 	u32 adminq_timeouts; /* free-running count of AQ cmds timeouts */
850 	/* free-running count of per AQ cmd executed */
851 	u32 adminq_describe_device_cnt;
852 	u32 adminq_cfg_device_resources_cnt;
853 	u32 adminq_register_page_list_cnt;
854 	u32 adminq_unregister_page_list_cnt;
855 	u32 adminq_create_tx_queue_cnt;
856 	u32 adminq_create_rx_queue_cnt;
857 	u32 adminq_destroy_tx_queue_cnt;
858 	u32 adminq_destroy_rx_queue_cnt;
859 	u32 adminq_dcfg_device_resources_cnt;
860 	u32 adminq_set_driver_parameter_cnt;
861 	u32 adminq_report_stats_cnt;
862 	u32 adminq_report_link_speed_cnt;
863 	u32 adminq_report_nic_timestamp_cnt;
864 	u32 adminq_get_ptype_map_cnt;
865 	u32 adminq_verify_driver_compatibility_cnt;
866 	u32 adminq_query_flow_rules_cnt;
867 	u32 adminq_cfg_flow_rule_cnt;
868 	u32 adminq_cfg_rss_cnt;
869 	u32 adminq_query_rss_cnt;
870 
871 	/* Global stats */
872 	u32 interface_up_cnt; /* count of times interface turned up since last reset */
873 	u32 interface_down_cnt; /* count of times interface turned down since last reset */
874 	u32 reset_cnt; /* count of reset */
875 	u32 page_alloc_fail; /* count of page alloc fails */
876 	u32 dma_mapping_error; /* count of dma mapping errors */
877 	u32 stats_report_trigger_cnt; /* count of device-requested stats-reports since last reset */
878 	u32 suspend_cnt; /* count of times suspended */
879 	u32 resume_cnt; /* count of times resumed */
880 	struct workqueue_struct *gve_wq;
881 	struct work_struct service_task;
882 	struct work_struct stats_report_task;
883 	unsigned long service_task_flags;
884 	unsigned long state_flags;
885 
886 	struct gve_stats_report *stats_report;
887 	u64 stats_report_len;
888 	dma_addr_t stats_report_bus; /* dma address for the stats report */
889 	unsigned long ethtool_flags;
890 
891 	unsigned long stats_report_timer_period;
892 	struct timer_list stats_report_timer;
893 
894 	/* Gvnic device link speed from hypervisor. */
895 	u64 link_speed;
896 	bool up_before_suspend; /* True if dev was up before suspend */
897 
898 	struct gve_ptype_lut *ptype_lut_dqo;
899 
900 	/* Must be a power of two. */
901 	u16 max_rx_buffer_size; /* device limit */
902 
903 	enum gve_queue_format queue_format;
904 
905 	/* Interrupt coalescing settings */
906 	u32 tx_coalesce_usecs;
907 	u32 rx_coalesce_usecs;
908 
909 	u16 header_buf_size; /* device configured, header-split supported if non-zero */
910 	bool header_split_enabled; /* True if the header split is enabled by the user */
911 
912 	u32 max_flow_rules;
913 	u32 num_flow_rules;
914 
915 	struct gve_flow_rules_cache flow_rules_cache;
916 
917 	u16 rss_key_size;
918 	u16 rss_lut_size;
919 	bool cache_rss_config;
920 	struct gve_rss_config rss_config;
921 
922 	/* True if the device supports reading the nic clock */
923 	bool nic_timestamp_supported;
924 	struct gve_ptp *ptp;
925 	struct kernel_hwtstamp_config ts_config;
926 	struct gve_nic_ts_report *nic_ts_report;
927 	dma_addr_t nic_ts_report_bus;
928 	u64 last_sync_nic_counter; /* Clock counter from last NIC TS report */
929 };
930 
931 enum gve_service_task_flags_bit {
932 	GVE_PRIV_FLAGS_DO_RESET			= 1,
933 	GVE_PRIV_FLAGS_RESET_IN_PROGRESS	= 2,
934 	GVE_PRIV_FLAGS_PROBE_IN_PROGRESS	= 3,
935 	GVE_PRIV_FLAGS_DO_REPORT_STATS = 4,
936 };
937 
938 enum gve_state_flags_bit {
939 	GVE_PRIV_FLAGS_ADMIN_QUEUE_OK		= 1,
940 	GVE_PRIV_FLAGS_DEVICE_RESOURCES_OK	= 2,
941 	GVE_PRIV_FLAGS_DEVICE_RINGS_OK		= 3,
942 	GVE_PRIV_FLAGS_NAPI_ENABLED		= 4,
943 };
944 
945 enum gve_ethtool_flags_bit {
946 	GVE_PRIV_FLAGS_REPORT_STATS		= 0,
947 };
948 
949 static inline bool gve_get_do_reset(struct gve_priv *priv)
950 {
951 	return test_bit(GVE_PRIV_FLAGS_DO_RESET, &priv->service_task_flags);
952 }
953 
954 static inline void gve_set_do_reset(struct gve_priv *priv)
955 {
956 	set_bit(GVE_PRIV_FLAGS_DO_RESET, &priv->service_task_flags);
957 }
958 
959 static inline void gve_clear_do_reset(struct gve_priv *priv)
960 {
961 	clear_bit(GVE_PRIV_FLAGS_DO_RESET, &priv->service_task_flags);
962 }
963 
964 static inline bool gve_get_reset_in_progress(struct gve_priv *priv)
965 {
966 	return test_bit(GVE_PRIV_FLAGS_RESET_IN_PROGRESS,
967 			&priv->service_task_flags);
968 }
969 
970 static inline void gve_set_reset_in_progress(struct gve_priv *priv)
971 {
972 	set_bit(GVE_PRIV_FLAGS_RESET_IN_PROGRESS, &priv->service_task_flags);
973 }
974 
975 static inline void gve_clear_reset_in_progress(struct gve_priv *priv)
976 {
977 	clear_bit(GVE_PRIV_FLAGS_RESET_IN_PROGRESS, &priv->service_task_flags);
978 }
979 
980 static inline bool gve_get_probe_in_progress(struct gve_priv *priv)
981 {
982 	return test_bit(GVE_PRIV_FLAGS_PROBE_IN_PROGRESS,
983 			&priv->service_task_flags);
984 }
985 
986 static inline void gve_set_probe_in_progress(struct gve_priv *priv)
987 {
988 	set_bit(GVE_PRIV_FLAGS_PROBE_IN_PROGRESS, &priv->service_task_flags);
989 }
990 
991 static inline void gve_clear_probe_in_progress(struct gve_priv *priv)
992 {
993 	clear_bit(GVE_PRIV_FLAGS_PROBE_IN_PROGRESS, &priv->service_task_flags);
994 }
995 
996 static inline bool gve_get_do_report_stats(struct gve_priv *priv)
997 {
998 	return test_bit(GVE_PRIV_FLAGS_DO_REPORT_STATS,
999 			&priv->service_task_flags);
1000 }
1001 
1002 static inline void gve_set_do_report_stats(struct gve_priv *priv)
1003 {
1004 	set_bit(GVE_PRIV_FLAGS_DO_REPORT_STATS, &priv->service_task_flags);
1005 }
1006 
1007 static inline void gve_clear_do_report_stats(struct gve_priv *priv)
1008 {
1009 	clear_bit(GVE_PRIV_FLAGS_DO_REPORT_STATS, &priv->service_task_flags);
1010 }
1011 
1012 static inline bool gve_get_admin_queue_ok(struct gve_priv *priv)
1013 {
1014 	return test_bit(GVE_PRIV_FLAGS_ADMIN_QUEUE_OK, &priv->state_flags);
1015 }
1016 
1017 static inline void gve_set_admin_queue_ok(struct gve_priv *priv)
1018 {
1019 	set_bit(GVE_PRIV_FLAGS_ADMIN_QUEUE_OK, &priv->state_flags);
1020 }
1021 
1022 static inline void gve_clear_admin_queue_ok(struct gve_priv *priv)
1023 {
1024 	clear_bit(GVE_PRIV_FLAGS_ADMIN_QUEUE_OK, &priv->state_flags);
1025 }
1026 
1027 static inline bool gve_get_device_resources_ok(struct gve_priv *priv)
1028 {
1029 	return test_bit(GVE_PRIV_FLAGS_DEVICE_RESOURCES_OK, &priv->state_flags);
1030 }
1031 
1032 static inline void gve_set_device_resources_ok(struct gve_priv *priv)
1033 {
1034 	set_bit(GVE_PRIV_FLAGS_DEVICE_RESOURCES_OK, &priv->state_flags);
1035 }
1036 
1037 static inline void gve_clear_device_resources_ok(struct gve_priv *priv)
1038 {
1039 	clear_bit(GVE_PRIV_FLAGS_DEVICE_RESOURCES_OK, &priv->state_flags);
1040 }
1041 
1042 static inline bool gve_get_device_rings_ok(struct gve_priv *priv)
1043 {
1044 	return test_bit(GVE_PRIV_FLAGS_DEVICE_RINGS_OK, &priv->state_flags);
1045 }
1046 
1047 static inline void gve_set_device_rings_ok(struct gve_priv *priv)
1048 {
1049 	set_bit(GVE_PRIV_FLAGS_DEVICE_RINGS_OK, &priv->state_flags);
1050 }
1051 
1052 static inline void gve_clear_device_rings_ok(struct gve_priv *priv)
1053 {
1054 	clear_bit(GVE_PRIV_FLAGS_DEVICE_RINGS_OK, &priv->state_flags);
1055 }
1056 
1057 static inline bool gve_get_napi_enabled(struct gve_priv *priv)
1058 {
1059 	return test_bit(GVE_PRIV_FLAGS_NAPI_ENABLED, &priv->state_flags);
1060 }
1061 
1062 static inline void gve_set_napi_enabled(struct gve_priv *priv)
1063 {
1064 	set_bit(GVE_PRIV_FLAGS_NAPI_ENABLED, &priv->state_flags);
1065 }
1066 
1067 static inline void gve_clear_napi_enabled(struct gve_priv *priv)
1068 {
1069 	clear_bit(GVE_PRIV_FLAGS_NAPI_ENABLED, &priv->state_flags);
1070 }
1071 
1072 static inline bool gve_get_report_stats(struct gve_priv *priv)
1073 {
1074 	return test_bit(GVE_PRIV_FLAGS_REPORT_STATS, &priv->ethtool_flags);
1075 }
1076 
1077 static inline void gve_clear_report_stats(struct gve_priv *priv)
1078 {
1079 	clear_bit(GVE_PRIV_FLAGS_REPORT_STATS, &priv->ethtool_flags);
1080 }
1081 
1082 /* Returns the address of the ntfy_blocks irq doorbell
1083  */
1084 static inline __be32 __iomem *gve_irq_doorbell(struct gve_priv *priv,
1085 					       struct gve_notify_block *block)
1086 {
1087 	return &priv->db_bar2[be32_to_cpu(*block->irq_db_index)];
1088 }
1089 
1090 /* Returns the index into ntfy_blocks of the given tx ring's block
1091  */
1092 static inline u32 gve_tx_idx_to_ntfy(struct gve_priv *priv, u32 queue_idx)
1093 {
1094 	return queue_idx;
1095 }
1096 
1097 /* Returns the index into ntfy_blocks of the given rx ring's block
1098  */
1099 static inline u32 gve_rx_idx_to_ntfy(struct gve_priv *priv, u32 queue_idx)
1100 {
1101 	return (priv->num_ntfy_blks / 2) + queue_idx;
1102 }
1103 
1104 static inline bool gve_is_qpl(struct gve_priv *priv)
1105 {
1106 	return priv->queue_format == GVE_GQI_QPL_FORMAT ||
1107 		priv->queue_format == GVE_DQO_QPL_FORMAT;
1108 }
1109 
1110 /* Returns the number of tx queue page lists */
1111 static inline u32 gve_num_tx_qpls(const struct gve_tx_queue_config *tx_cfg,
1112 				  bool is_qpl)
1113 {
1114 	if (!is_qpl)
1115 		return 0;
1116 	return tx_cfg->num_queues + tx_cfg->num_xdp_queues;
1117 }
1118 
1119 /* Returns the number of rx queue page lists */
1120 static inline u32 gve_num_rx_qpls(const struct gve_rx_queue_config *rx_cfg,
1121 				  bool is_qpl)
1122 {
1123 	if (!is_qpl)
1124 		return 0;
1125 	return rx_cfg->num_queues;
1126 }
1127 
1128 static inline u32 gve_tx_qpl_id(struct gve_priv *priv, int tx_qid)
1129 {
1130 	return tx_qid;
1131 }
1132 
1133 static inline u32 gve_rx_qpl_id(struct gve_priv *priv, int rx_qid)
1134 {
1135 	return priv->tx_cfg.max_queues + rx_qid;
1136 }
1137 
1138 static inline u32 gve_get_rx_qpl_id(const struct gve_tx_queue_config *tx_cfg,
1139 				    int rx_qid)
1140 {
1141 	return tx_cfg->max_queues + rx_qid;
1142 }
1143 
1144 static inline u32 gve_tx_start_qpl_id(struct gve_priv *priv)
1145 {
1146 	return gve_tx_qpl_id(priv, 0);
1147 }
1148 
1149 static inline u32 gve_rx_start_qpl_id(const struct gve_tx_queue_config *tx_cfg)
1150 {
1151 	return gve_get_rx_qpl_id(tx_cfg, 0);
1152 }
1153 
1154 /* Returns the correct dma direction for tx and rx qpls */
1155 static inline enum dma_data_direction gve_qpl_dma_dir(struct gve_priv *priv,
1156 						      int id)
1157 {
1158 	if (id < gve_rx_start_qpl_id(&priv->tx_cfg))
1159 		return DMA_TO_DEVICE;
1160 	else
1161 		return DMA_FROM_DEVICE;
1162 }
1163 
1164 static inline bool gve_is_gqi(struct gve_priv *priv)
1165 {
1166 	return priv->queue_format == GVE_GQI_RDA_FORMAT ||
1167 		priv->queue_format == GVE_GQI_QPL_FORMAT;
1168 }
1169 
1170 static inline bool gve_is_dqo(struct gve_priv *priv)
1171 {
1172 	return priv->queue_format == GVE_DQO_RDA_FORMAT ||
1173 	       priv->queue_format == GVE_DQO_QPL_FORMAT;
1174 }
1175 
1176 static inline u32 gve_num_tx_queues(struct gve_priv *priv)
1177 {
1178 	return priv->tx_cfg.num_queues + priv->tx_cfg.num_xdp_queues;
1179 }
1180 
1181 static inline u32 gve_xdp_tx_queue_id(struct gve_priv *priv, u32 queue_id)
1182 {
1183 	return priv->tx_cfg.num_queues + queue_id;
1184 }
1185 
1186 static inline u32 gve_xdp_tx_start_queue_id(struct gve_priv *priv)
1187 {
1188 	return gve_xdp_tx_queue_id(priv, 0);
1189 }
1190 
1191 static inline bool gve_supports_xdp_xmit(struct gve_priv *priv)
1192 {
1193 	switch (priv->queue_format) {
1194 	case GVE_GQI_QPL_FORMAT:
1195 	case GVE_DQO_RDA_FORMAT:
1196 		return true;
1197 	default:
1198 		return false;
1199 	}
1200 }
1201 
1202 static inline bool gve_is_clock_enabled(struct gve_priv *priv)
1203 {
1204 	return priv->nic_ts_report;
1205 }
1206 
1207 /* gqi napi handler defined in gve_main.c */
1208 int gve_napi_poll(struct napi_struct *napi, int budget);
1209 
1210 /* buffers */
1211 int gve_alloc_page(struct gve_priv *priv, struct device *dev,
1212 		   struct page **page, dma_addr_t *dma,
1213 		   enum dma_data_direction, gfp_t gfp_flags);
1214 void gve_free_page(struct device *dev, struct page *page, dma_addr_t dma,
1215 		   enum dma_data_direction);
1216 /* qpls */
1217 struct gve_queue_page_list *gve_alloc_queue_page_list(struct gve_priv *priv,
1218 						      u32 id, int pages);
1219 void gve_free_queue_page_list(struct gve_priv *priv,
1220 			      struct gve_queue_page_list *qpl,
1221 			      u32 id);
1222 /* tx handling */
1223 netdev_tx_t gve_tx(struct sk_buff *skb, struct net_device *dev);
1224 int gve_xdp_xmit_gqi(struct net_device *dev, int n, struct xdp_frame **frames,
1225 		     u32 flags);
1226 int gve_xdp_xmit_dqo(struct net_device *dev, int n, struct xdp_frame **frames,
1227 		     u32 flags);
1228 int gve_xdp_xmit_one(struct gve_priv *priv, struct gve_tx_ring *tx,
1229 		     void *data, int len, void *frame_p);
1230 void gve_xdp_tx_flush(struct gve_priv *priv, u32 xdp_qid);
1231 int gve_xdp_xmit_one_dqo(struct gve_priv *priv, struct gve_tx_ring *tx,
1232 			 struct xdp_frame *xdpf);
1233 bool gve_tx_poll(struct gve_notify_block *block, int budget);
1234 bool gve_xdp_poll(struct gve_notify_block *block, int budget);
1235 int gve_xsk_tx_poll(struct gve_notify_block *block, int budget);
1236 int gve_tx_alloc_rings_gqi(struct gve_priv *priv,
1237 			   struct gve_tx_alloc_rings_cfg *cfg);
1238 void gve_tx_free_rings_gqi(struct gve_priv *priv,
1239 			   struct gve_tx_alloc_rings_cfg *cfg);
1240 void gve_tx_start_ring_gqi(struct gve_priv *priv, int idx);
1241 void gve_tx_stop_ring_gqi(struct gve_priv *priv, int idx);
1242 u32 gve_tx_load_event_counter(struct gve_priv *priv,
1243 			      struct gve_tx_ring *tx);
1244 bool gve_tx_clean_pending(struct gve_priv *priv, struct gve_tx_ring *tx);
1245 /* rx handling */
1246 void gve_rx_write_doorbell(struct gve_priv *priv, struct gve_rx_ring *rx);
1247 int gve_rx_poll(struct gve_notify_block *block, int budget);
1248 bool gve_rx_work_pending(struct gve_rx_ring *rx);
1249 int gve_rx_alloc_ring_gqi(struct gve_priv *priv,
1250 			  struct gve_rx_alloc_rings_cfg *cfg,
1251 			  struct gve_rx_ring *rx,
1252 			  int idx);
1253 void gve_rx_free_ring_gqi(struct gve_priv *priv, struct gve_rx_ring *rx,
1254 			  struct gve_rx_alloc_rings_cfg *cfg);
1255 int gve_rx_alloc_rings_gqi(struct gve_priv *priv,
1256 			   struct gve_rx_alloc_rings_cfg *cfg);
1257 void gve_rx_free_rings_gqi(struct gve_priv *priv,
1258 			   struct gve_rx_alloc_rings_cfg *cfg);
1259 void gve_rx_start_ring_gqi(struct gve_priv *priv, int idx);
1260 void gve_rx_stop_ring_gqi(struct gve_priv *priv, int idx);
1261 bool gve_header_split_supported(const struct gve_priv *priv);
1262 int gve_set_rx_buf_len_config(struct gve_priv *priv, u32 rx_buf_len,
1263 			      struct netlink_ext_ack *extack,
1264 			      struct gve_rx_alloc_rings_cfg *rx_alloc_cfg);
1265 int gve_set_hsplit_config(struct gve_priv *priv, u8 tcp_data_split,
1266 			  struct gve_rx_alloc_rings_cfg *rx_alloc_cfg);
1267 /* rx buffer handling */
1268 int gve_buf_ref_cnt(struct gve_rx_buf_state_dqo *bs);
1269 void gve_free_page_dqo(struct gve_priv *priv, struct gve_rx_buf_state_dqo *bs,
1270 		       bool free_page);
1271 struct gve_rx_buf_state_dqo *gve_alloc_buf_state(struct gve_rx_ring *rx);
1272 bool gve_buf_state_is_allocated(struct gve_rx_ring *rx,
1273 				struct gve_rx_buf_state_dqo *buf_state);
1274 void gve_free_buf_state(struct gve_rx_ring *rx,
1275 			struct gve_rx_buf_state_dqo *buf_state);
1276 struct gve_rx_buf_state_dqo *gve_dequeue_buf_state(struct gve_rx_ring *rx,
1277 						   struct gve_index_list *list);
1278 void gve_enqueue_buf_state(struct gve_rx_ring *rx, struct gve_index_list *list,
1279 			   struct gve_rx_buf_state_dqo *buf_state);
1280 struct gve_rx_buf_state_dqo *gve_get_recycled_buf_state(struct gve_rx_ring *rx);
1281 void gve_try_recycle_buf(struct gve_priv *priv, struct gve_rx_ring *rx,
1282 			 struct gve_rx_buf_state_dqo *buf_state);
1283 void gve_free_to_page_pool(struct gve_rx_ring *rx,
1284 			   struct gve_rx_buf_state_dqo *buf_state,
1285 			   bool allow_direct);
1286 int gve_alloc_qpl_page_dqo(struct gve_rx_ring *rx,
1287 			   struct gve_rx_buf_state_dqo *buf_state);
1288 void gve_free_qpl_page_dqo(struct gve_rx_buf_state_dqo *buf_state);
1289 void gve_reuse_buffer(struct gve_rx_ring *rx,
1290 		      struct gve_rx_buf_state_dqo *buf_state);
1291 void gve_free_buffer(struct gve_rx_ring *rx,
1292 		     struct gve_rx_buf_state_dqo *buf_state);
1293 int gve_alloc_buffer(struct gve_rx_ring *rx, struct gve_rx_desc_dqo *desc);
1294 struct page_pool *gve_rx_create_page_pool(struct gve_priv *priv,
1295 					  struct gve_rx_ring *rx,
1296 					  bool xdp);
1297 
1298 /* Reset */
1299 void gve_schedule_reset(struct gve_priv *priv);
1300 int gve_reset(struct gve_priv *priv, bool attempt_teardown);
1301 void gve_get_curr_alloc_cfgs(struct gve_priv *priv,
1302 			     struct gve_tx_alloc_rings_cfg *tx_alloc_cfg,
1303 			     struct gve_rx_alloc_rings_cfg *rx_alloc_cfg);
1304 void gve_update_num_qpl_pages(struct gve_priv *priv,
1305 			      struct gve_rx_alloc_rings_cfg *rx_alloc_cfg,
1306 			      struct gve_tx_alloc_rings_cfg *tx_alloc_cfg);
1307 int gve_adjust_config(struct gve_priv *priv,
1308 		      struct gve_tx_alloc_rings_cfg *tx_alloc_cfg,
1309 		      struct gve_rx_alloc_rings_cfg *rx_alloc_cfg);
1310 int gve_adjust_queues(struct gve_priv *priv,
1311 		      struct gve_rx_queue_config new_rx_config,
1312 		      struct gve_tx_queue_config new_tx_config,
1313 		      bool reset_rss);
1314 /* flow steering rule */
1315 int gve_get_flow_rule_entry(struct gve_priv *priv, struct ethtool_rxnfc *cmd);
1316 int gve_get_flow_rule_ids(struct gve_priv *priv, struct ethtool_rxnfc *cmd, u32 *rule_locs);
1317 int gve_add_flow_rule(struct gve_priv *priv, struct ethtool_rxnfc *cmd);
1318 int gve_del_flow_rule(struct gve_priv *priv, struct ethtool_rxnfc *cmd);
1319 int gve_flow_rules_reset(struct gve_priv *priv);
1320 /* RSS config */
1321 int gve_init_rss_config(struct gve_priv *priv, u16 num_queues);
1322 /* PTP and timestamping */
1323 #if IS_ENABLED(CONFIG_PTP_1588_CLOCK)
1324 int gve_clock_nic_ts_read(struct gve_priv *priv);
1325 int gve_init_clock(struct gve_priv *priv);
1326 void gve_teardown_clock(struct gve_priv *priv);
1327 #else /* CONFIG_PTP_1588_CLOCK */
1328 static inline int gve_clock_nic_ts_read(struct gve_priv *priv)
1329 {
1330 	return -EOPNOTSUPP;
1331 }
1332 
1333 static inline int gve_init_clock(struct gve_priv *priv)
1334 {
1335 	return 0;
1336 }
1337 
1338 static inline void gve_teardown_clock(struct gve_priv *priv) { }
1339 #endif /* CONFIG_PTP_1588_CLOCK */
1340 /* report stats handling */
1341 void gve_handle_report_stats(struct gve_priv *priv);
1342 /* exported by ethtool.c */
1343 extern const struct ethtool_ops gve_ethtool_ops;
1344 /* needed by ethtool */
1345 extern char gve_driver_name[];
1346 extern const char gve_version_str[];
1347 #endif /* _GVE_H_ */
1348