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