1 /* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */
2 /* Copyright (C) 2015-2018 Netronome Systems, Inc. */
3
4 /*
5 * nfp_net.h
6 * Declarations for Netronome network device driver.
7 * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
8 * Jason McMullan <jason.mcmullan@netronome.com>
9 * Rolf Neugebauer <rolf.neugebauer@netronome.com>
10 */
11
12 #ifndef _NFP_NET_H_
13 #define _NFP_NET_H_
14
15 #include <linux/atomic.h>
16 #include <linux/interrupt.h>
17 #include <linux/list.h>
18 #include <linux/netdevice.h>
19 #include <linux/pci.h>
20 #include <linux/io-64-nonatomic-hi-lo.h>
21 #include <linux/semaphore.h>
22 #include <linux/workqueue.h>
23 #include <net/xdp.h>
24
25 #include "nfp_net_ctrl.h"
26
27 #define nn_pr(nn, lvl, fmt, args...) \
28 ({ \
29 struct nfp_net *__nn = (nn); \
30 \
31 if (__nn->dp.netdev) \
32 netdev_printk(lvl, __nn->dp.netdev, fmt, ## args); \
33 else \
34 dev_printk(lvl, __nn->dp.dev, "ctrl: " fmt, ## args); \
35 })
36
37 #define nn_err(nn, fmt, args...) nn_pr(nn, KERN_ERR, fmt, ## args)
38 #define nn_warn(nn, fmt, args...) nn_pr(nn, KERN_WARNING, fmt, ## args)
39 #define nn_info(nn, fmt, args...) nn_pr(nn, KERN_INFO, fmt, ## args)
40 #define nn_dbg(nn, fmt, args...) nn_pr(nn, KERN_DEBUG, fmt, ## args)
41
42 #define nn_dp_warn(dp, fmt, args...) \
43 ({ \
44 struct nfp_net_dp *__dp = (dp); \
45 \
46 if (unlikely(net_ratelimit())) { \
47 if (__dp->netdev) \
48 netdev_warn(__dp->netdev, fmt, ## args); \
49 else \
50 dev_warn(__dp->dev, fmt, ## args); \
51 } \
52 })
53
54 /* Max time to wait for NFP to respond on updates (in seconds) */
55 #define NFP_NET_POLL_TIMEOUT 5
56
57 /* Interval for reading offloaded filter stats */
58 #define NFP_NET_STAT_POLL_IVL msecs_to_jiffies(100)
59
60 /* Bar allocation */
61 #define NFP_NET_CTRL_BAR 0
62 #define NFP_NET_Q0_BAR 2
63 #define NFP_NET_Q1_BAR 4 /* OBSOLETE */
64
65 /* Max bits in DMA address */
66 #define NFP_NET_MAX_DMA_BITS 40
67
68 /* Default size for MTU and freelist buffer sizes */
69 #define NFP_NET_DEFAULT_MTU 1500U
70
71 /* Maximum number of bytes prepended to a packet */
72 #define NFP_NET_MAX_PREPEND 64
73
74 /* Interrupt definitions */
75 #define NFP_NET_NON_Q_VECTORS 2
76 #define NFP_NET_IRQ_LSC_IDX 0
77 #define NFP_NET_IRQ_EXN_IDX 1
78 #define NFP_NET_MIN_VNIC_IRQS (NFP_NET_NON_Q_VECTORS + 1)
79
80 /* Queue/Ring definitions */
81 #define NFP_NET_MAX_TX_RINGS 64 /* Max. # of Tx rings per device */
82 #define NFP_NET_MAX_RX_RINGS 64 /* Max. # of Rx rings per device */
83 #define NFP_NET_MAX_R_VECS (NFP_NET_MAX_TX_RINGS > NFP_NET_MAX_RX_RINGS ? \
84 NFP_NET_MAX_TX_RINGS : NFP_NET_MAX_RX_RINGS)
85 #define NFP_NET_MAX_IRQS (NFP_NET_NON_Q_VECTORS + NFP_NET_MAX_R_VECS)
86
87 #define NFP_NET_MIN_TX_DESCS 256 /* Min. # of Tx descs per ring */
88 #define NFP_NET_MIN_RX_DESCS 256 /* Min. # of Rx descs per ring */
89 #define NFP_NET_MAX_TX_DESCS (256 * 1024) /* Max. # of Tx descs per ring */
90 #define NFP_NET_MAX_RX_DESCS (256 * 1024) /* Max. # of Rx descs per ring */
91
92 #define NFP_NET_TX_DESCS_DEFAULT 4096 /* Default # of Tx descs per ring */
93 #define NFP_NET_RX_DESCS_DEFAULT 4096 /* Default # of Rx descs per ring */
94
95 #define NFP_NET_FL_BATCH 16 /* Add freelist in this Batch size */
96 #define NFP_NET_XDP_MAX_COMPLETE 2048 /* XDP bufs to reclaim in NAPI poll */
97
98 /* Offload definitions */
99 #define NFP_NET_N_VXLAN_PORTS (NFP_NET_CFG_VXLAN_SZ / sizeof(__be16))
100
101 #define NFP_NET_RX_BUF_HEADROOM (NET_SKB_PAD + NET_IP_ALIGN)
102 #define NFP_NET_RX_BUF_NON_DATA (NFP_NET_RX_BUF_HEADROOM + \
103 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
104
105 /* Forward declarations */
106 struct nfp_cpp;
107 struct nfp_eth_table_port;
108 struct nfp_net;
109 struct nfp_net_r_vector;
110 struct nfp_port;
111
112 /* Convenience macro for wrapping descriptor index on ring size */
113 #define D_IDX(ring, idx) ((idx) & ((ring)->cnt - 1))
114
115 /* Convenience macro for writing dma address into RX/TX descriptors */
116 #define nfp_desc_set_dma_addr(desc, dma_addr) \
117 do { \
118 __typeof(desc) __d = (desc); \
119 dma_addr_t __addr = (dma_addr); \
120 \
121 __d->dma_addr_lo = cpu_to_le32(lower_32_bits(__addr)); \
122 __d->dma_addr_hi = upper_32_bits(__addr) & 0xff; \
123 } while (0)
124
125 /* TX descriptor format */
126
127 #define PCIE_DESC_TX_EOP BIT(7)
128 #define PCIE_DESC_TX_OFFSET_MASK GENMASK(6, 0)
129 #define PCIE_DESC_TX_MSS_MASK GENMASK(13, 0)
130
131 /* Flags in the host TX descriptor */
132 #define PCIE_DESC_TX_CSUM BIT(7)
133 #define PCIE_DESC_TX_IP4_CSUM BIT(6)
134 #define PCIE_DESC_TX_TCP_CSUM BIT(5)
135 #define PCIE_DESC_TX_UDP_CSUM BIT(4)
136 #define PCIE_DESC_TX_VLAN BIT(3)
137 #define PCIE_DESC_TX_LSO BIT(2)
138 #define PCIE_DESC_TX_ENCAP BIT(1)
139 #define PCIE_DESC_TX_O_IP4_CSUM BIT(0)
140
141 struct nfp_net_tx_desc {
142 union {
143 struct {
144 u8 dma_addr_hi; /* High bits of host buf address */
145 __le16 dma_len; /* Length to DMA for this desc */
146 u8 offset_eop; /* Offset in buf where pkt starts +
147 * highest bit is eop flag.
148 */
149 __le32 dma_addr_lo; /* Low 32bit of host buf addr */
150
151 __le16 mss; /* MSS to be used for LSO */
152 u8 lso_hdrlen; /* LSO, TCP payload offset */
153 u8 flags; /* TX Flags, see @PCIE_DESC_TX_* */
154 union {
155 struct {
156 u8 l3_offset; /* L3 header offset */
157 u8 l4_offset; /* L4 header offset */
158 };
159 __le16 vlan; /* VLAN tag to add if indicated */
160 };
161 __le16 data_len; /* Length of frame + meta data */
162 } __packed;
163 __le32 vals[4];
164 __le64 vals8[2];
165 };
166 };
167
168 /**
169 * struct nfp_net_tx_buf - software TX buffer descriptor
170 * @skb: normal ring, sk_buff associated with this buffer
171 * @frag: XDP ring, page frag associated with this buffer
172 * @dma_addr: DMA mapping address of the buffer
173 * @fidx: Fragment index (-1 for the head and [0..nr_frags-1] for frags)
174 * @pkt_cnt: Number of packets to be produced out of the skb associated
175 * with this buffer (valid only on the head's buffer).
176 * Will be 1 for all non-TSO packets.
177 * @real_len: Number of bytes which to be produced out of the skb (valid only
178 * on the head's buffer). Equal to skb->len for non-TSO packets.
179 */
180 struct nfp_net_tx_buf {
181 union {
182 struct sk_buff *skb;
183 void *frag;
184 };
185 dma_addr_t dma_addr;
186 short int fidx;
187 u16 pkt_cnt;
188 u32 real_len;
189 };
190
191 /**
192 * struct nfp_net_tx_ring - TX ring structure
193 * @r_vec: Back pointer to ring vector structure
194 * @idx: Ring index from Linux's perspective
195 * @qcidx: Queue Controller Peripheral (QCP) queue index for the TX queue
196 * @qcp_q: Pointer to base of the QCP TX queue
197 * @cnt: Size of the queue in number of descriptors
198 * @wr_p: TX ring write pointer (free running)
199 * @rd_p: TX ring read pointer (free running)
200 * @qcp_rd_p: Local copy of QCP TX queue read pointer
201 * @wr_ptr_add: Accumulated number of buffers to add to QCP write pointer
202 * (used for .xmit_more delayed kick)
203 * @txbufs: Array of transmitted TX buffers, to free on transmit
204 * @txds: Virtual address of TX ring in host memory
205 * @dma: DMA address of the TX ring
206 * @size: Size, in bytes, of the TX ring (needed to free)
207 * @is_xdp: Is this a XDP TX ring?
208 */
209 struct nfp_net_tx_ring {
210 struct nfp_net_r_vector *r_vec;
211
212 u32 idx;
213 int qcidx;
214 u8 __iomem *qcp_q;
215
216 u32 cnt;
217 u32 wr_p;
218 u32 rd_p;
219 u32 qcp_rd_p;
220
221 u32 wr_ptr_add;
222
223 struct nfp_net_tx_buf *txbufs;
224 struct nfp_net_tx_desc *txds;
225
226 dma_addr_t dma;
227 size_t size;
228 bool is_xdp;
229 } ____cacheline_aligned;
230
231 /* RX and freelist descriptor format */
232
233 #define PCIE_DESC_RX_DD BIT(7)
234 #define PCIE_DESC_RX_META_LEN_MASK GENMASK(6, 0)
235
236 /* Flags in the RX descriptor */
237 #define PCIE_DESC_RX_RSS cpu_to_le16(BIT(15))
238 #define PCIE_DESC_RX_I_IP4_CSUM cpu_to_le16(BIT(14))
239 #define PCIE_DESC_RX_I_IP4_CSUM_OK cpu_to_le16(BIT(13))
240 #define PCIE_DESC_RX_I_TCP_CSUM cpu_to_le16(BIT(12))
241 #define PCIE_DESC_RX_I_TCP_CSUM_OK cpu_to_le16(BIT(11))
242 #define PCIE_DESC_RX_I_UDP_CSUM cpu_to_le16(BIT(10))
243 #define PCIE_DESC_RX_I_UDP_CSUM_OK cpu_to_le16(BIT(9))
244 #define PCIE_DESC_RX_DECRYPTED cpu_to_le16(BIT(8))
245 #define PCIE_DESC_RX_EOP cpu_to_le16(BIT(7))
246 #define PCIE_DESC_RX_IP4_CSUM cpu_to_le16(BIT(6))
247 #define PCIE_DESC_RX_IP4_CSUM_OK cpu_to_le16(BIT(5))
248 #define PCIE_DESC_RX_TCP_CSUM cpu_to_le16(BIT(4))
249 #define PCIE_DESC_RX_TCP_CSUM_OK cpu_to_le16(BIT(3))
250 #define PCIE_DESC_RX_UDP_CSUM cpu_to_le16(BIT(2))
251 #define PCIE_DESC_RX_UDP_CSUM_OK cpu_to_le16(BIT(1))
252 #define PCIE_DESC_RX_VLAN cpu_to_le16(BIT(0))
253
254 #define PCIE_DESC_RX_CSUM_ALL (PCIE_DESC_RX_IP4_CSUM | \
255 PCIE_DESC_RX_TCP_CSUM | \
256 PCIE_DESC_RX_UDP_CSUM | \
257 PCIE_DESC_RX_I_IP4_CSUM | \
258 PCIE_DESC_RX_I_TCP_CSUM | \
259 PCIE_DESC_RX_I_UDP_CSUM)
260 #define PCIE_DESC_RX_CSUM_OK_SHIFT 1
261 #define __PCIE_DESC_RX_CSUM_ALL le16_to_cpu(PCIE_DESC_RX_CSUM_ALL)
262 #define __PCIE_DESC_RX_CSUM_ALL_OK (__PCIE_DESC_RX_CSUM_ALL >> \
263 PCIE_DESC_RX_CSUM_OK_SHIFT)
264
265 struct nfp_net_rx_desc {
266 union {
267 struct {
268 u8 dma_addr_hi; /* High bits of the buf address */
269 __le16 reserved; /* Must be zero */
270 u8 meta_len_dd; /* Must be zero */
271
272 __le32 dma_addr_lo; /* Low bits of the buffer address */
273 } __packed fld;
274
275 struct {
276 __le16 data_len; /* Length of the frame + meta data */
277 u8 reserved;
278 u8 meta_len_dd; /* Length of meta data prepended +
279 * descriptor done flag.
280 */
281
282 __le16 flags; /* RX flags. See @PCIE_DESC_RX_* */
283 __le16 vlan; /* VLAN if stripped */
284 } __packed rxd;
285
286 __le32 vals[2];
287 };
288 };
289
290 #define NFP_NET_META_FIELD_MASK GENMASK(NFP_NET_META_FIELD_SIZE - 1, 0)
291
292 struct nfp_meta_parsed {
293 u8 hash_type;
294 u8 csum_type;
295 u32 hash;
296 u32 mark;
297 u32 portid;
298 __wsum csum;
299 };
300
301 struct nfp_net_rx_hash {
302 __be32 hash_type;
303 __be32 hash;
304 };
305
306 /**
307 * struct nfp_net_rx_buf - software RX buffer descriptor
308 * @frag: page fragment buffer
309 * @dma_addr: DMA mapping address of the buffer
310 */
311 struct nfp_net_rx_buf {
312 void *frag;
313 dma_addr_t dma_addr;
314 };
315
316 /**
317 * struct nfp_net_rx_ring - RX ring structure
318 * @r_vec: Back pointer to ring vector structure
319 * @cnt: Size of the queue in number of descriptors
320 * @wr_p: FL/RX ring write pointer (free running)
321 * @rd_p: FL/RX ring read pointer (free running)
322 * @idx: Ring index from Linux's perspective
323 * @fl_qcidx: Queue Controller Peripheral (QCP) queue index for the freelist
324 * @qcp_fl: Pointer to base of the QCP freelist queue
325 * @rxbufs: Array of transmitted FL/RX buffers
326 * @rxds: Virtual address of FL/RX ring in host memory
327 * @xdp_rxq: RX-ring info avail for XDP
328 * @dma: DMA address of the FL/RX ring
329 * @size: Size, in bytes, of the FL/RX ring (needed to free)
330 */
331 struct nfp_net_rx_ring {
332 struct nfp_net_r_vector *r_vec;
333
334 u32 cnt;
335 u32 wr_p;
336 u32 rd_p;
337
338 u32 idx;
339
340 int fl_qcidx;
341 u8 __iomem *qcp_fl;
342
343 struct nfp_net_rx_buf *rxbufs;
344 struct nfp_net_rx_desc *rxds;
345
346 struct xdp_rxq_info xdp_rxq;
347
348 dma_addr_t dma;
349 size_t size;
350 } ____cacheline_aligned;
351
352 /**
353 * struct nfp_net_r_vector - Per ring interrupt vector configuration
354 * @nfp_net: Backpointer to nfp_net structure
355 * @napi: NAPI structure for this ring vec
356 * @tasklet: ctrl vNIC, tasklet for servicing the r_vec
357 * @queue: ctrl vNIC, send queue
358 * @lock: ctrl vNIC, r_vec lock protects @queue
359 * @tx_ring: Pointer to TX ring
360 * @rx_ring: Pointer to RX ring
361 * @xdp_ring: Pointer to an extra TX ring for XDP
362 * @irq_entry: MSI-X table entry (use for talking to the device)
363 * @rx_sync: Seqlock for atomic updates of RX stats
364 * @rx_pkts: Number of received packets
365 * @rx_bytes: Number of received bytes
366 * @rx_drops: Number of packets dropped on RX due to lack of resources
367 * @hw_csum_rx_ok: Counter of packets where the HW checksum was OK
368 * @hw_csum_rx_inner_ok: Counter of packets where the inner HW checksum was OK
369 * @hw_csum_rx_complete: Counter of packets with CHECKSUM_COMPLETE reported
370 * @hw_csum_rx_error: Counter of packets with bad checksums
371 * @hw_tls_rx: Number of packets with TLS decrypted by hardware
372 * @tx_sync: Seqlock for atomic updates of TX stats
373 * @tx_pkts: Number of Transmitted packets
374 * @tx_bytes: Number of Transmitted bytes
375 * @hw_csum_tx: Counter of packets with TX checksum offload requested
376 * @hw_csum_tx_inner: Counter of inner TX checksum offload requests
377 * @tx_gather: Counter of packets with Gather DMA
378 * @tx_lso: Counter of LSO packets sent
379 * @hw_tls_tx: Counter of TLS packets sent with crypto offloaded to HW
380 * @tls_tx_fallback: Counter of TLS packets sent which had to be encrypted
381 * by the fallback path because packets came out of order
382 * @tls_tx_no_fallback: Counter of TLS packets not sent because the fallback
383 * path could not encrypt them
384 * @tx_errors: How many TX errors were encountered
385 * @tx_busy: How often was TX busy (no space)?
386 * @rx_replace_buf_alloc_fail: Counter of RX buffer allocation failures
387 * @irq_vector: Interrupt vector number (use for talking to the OS)
388 * @handler: Interrupt handler for this ring vector
389 * @name: Name of the interrupt vector
390 * @affinity_mask: SMP affinity mask for this vector
391 *
392 * This structure ties RX and TX rings to interrupt vectors and a NAPI
393 * context. This currently only supports one RX and TX ring per
394 * interrupt vector but might be extended in the future to allow
395 * association of multiple rings per vector.
396 */
397 struct nfp_net_r_vector {
398 struct nfp_net *nfp_net;
399 union {
400 struct napi_struct napi;
401 struct {
402 struct tasklet_struct tasklet;
403 struct sk_buff_head queue;
404 spinlock_t lock;
405 };
406 };
407
408 struct nfp_net_tx_ring *tx_ring;
409 struct nfp_net_rx_ring *rx_ring;
410
411 u16 irq_entry;
412
413 struct u64_stats_sync rx_sync;
414 u64 rx_pkts;
415 u64 rx_bytes;
416 u64 rx_drops;
417 u64 hw_csum_rx_ok;
418 u64 hw_csum_rx_inner_ok;
419 u64 hw_csum_rx_complete;
420 u64 hw_tls_rx;
421
422 u64 hw_csum_rx_error;
423 u64 rx_replace_buf_alloc_fail;
424
425 struct nfp_net_tx_ring *xdp_ring;
426
427 struct u64_stats_sync tx_sync;
428 u64 tx_pkts;
429 u64 tx_bytes;
430
431 u64 ____cacheline_aligned_in_smp hw_csum_tx;
432 u64 hw_csum_tx_inner;
433 u64 tx_gather;
434 u64 tx_lso;
435 u64 hw_tls_tx;
436
437 u64 tls_tx_fallback;
438 u64 tls_tx_no_fallback;
439 u64 tx_errors;
440 u64 tx_busy;
441
442 /* Cold data follows */
443
444 u32 irq_vector;
445 irq_handler_t handler;
446 char name[IFNAMSIZ + 8];
447 cpumask_t affinity_mask;
448 } ____cacheline_aligned;
449
450 /* Firmware version as it is written in the 32bit value in the BAR */
451 struct nfp_net_fw_version {
452 u8 minor;
453 u8 major;
454 u8 class;
455 u8 resv;
456 } __packed;
457
458 static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver,
459 u8 resv, u8 class, u8 major, u8 minor)
460 {
461 return fw_ver->resv == resv &&
462 fw_ver->class == class &&
463 fw_ver->major == major &&
464 fw_ver->minor == minor;
465 }
466
467 struct nfp_stat_pair {
468 u64 pkts;
469 u64 bytes;
470 };
471
472 /**
473 * struct nfp_net_dp - NFP network device datapath data structure
474 * @dev: Backpointer to struct device
475 * @netdev: Backpointer to net_device structure
476 * @is_vf: Is the driver attached to a VF?
477 * @chained_metadata_format: Firemware will use new metadata format
478 * @ktls_tx: Is kTLS TX enabled?
479 * @rx_dma_dir: Mapping direction for RX buffers
480 * @rx_dma_off: Offset at which DMA packets (for XDP headroom)
481 * @rx_offset: Offset in the RX buffers where packet data starts
482 * @ctrl: Local copy of the control register/word.
483 * @fl_bufsz: Currently configured size of the freelist buffers
484 * @xdp_prog: Installed XDP program
485 * @tx_rings: Array of pre-allocated TX ring structures
486 * @rx_rings: Array of pre-allocated RX ring structures
487 * @ctrl_bar: Pointer to mapped control BAR
488 *
489 * @txd_cnt: Size of the TX ring in number of descriptors
490 * @rxd_cnt: Size of the RX ring in number of descriptors
491 * @num_r_vecs: Number of used ring vectors
492 * @num_tx_rings: Currently configured number of TX rings
493 * @num_stack_tx_rings: Number of TX rings used by the stack (not XDP)
494 * @num_rx_rings: Currently configured number of RX rings
495 * @mtu: Device MTU
496 */
497 struct nfp_net_dp {
498 struct device *dev;
499 struct net_device *netdev;
500
501 u8 is_vf:1;
502 u8 chained_metadata_format:1;
503 u8 ktls_tx:1;
504
505 u8 rx_dma_dir;
506 u8 rx_offset;
507
508 u32 rx_dma_off;
509
510 u32 ctrl;
511 u32 fl_bufsz;
512
513 struct bpf_prog *xdp_prog;
514
515 struct nfp_net_tx_ring *tx_rings;
516 struct nfp_net_rx_ring *rx_rings;
517
518 u8 __iomem *ctrl_bar;
519
520 /* Cold data follows */
521
522 unsigned int txd_cnt;
523 unsigned int rxd_cnt;
524
525 unsigned int num_r_vecs;
526
527 unsigned int num_tx_rings;
528 unsigned int num_stack_tx_rings;
529 unsigned int num_rx_rings;
530
531 unsigned int mtu;
532 };
533
534 /**
535 * struct nfp_net - NFP network device structure
536 * @dp: Datapath structure
537 * @id: vNIC id within the PF (0 for VFs)
538 * @fw_ver: Firmware version
539 * @cap: Capabilities advertised by the Firmware
540 * @max_mtu: Maximum support MTU advertised by the Firmware
541 * @rss_hfunc: RSS selected hash function
542 * @rss_cfg: RSS configuration
543 * @rss_key: RSS secret key
544 * @rss_itbl: RSS indirection table
545 * @xdp: Information about the driver XDP program
546 * @xdp_hw: Information about the HW XDP program
547 * @max_r_vecs: Number of allocated interrupt vectors for RX/TX
548 * @max_tx_rings: Maximum number of TX rings supported by the Firmware
549 * @max_rx_rings: Maximum number of RX rings supported by the Firmware
550 * @stride_rx: Queue controller RX queue spacing
551 * @stride_tx: Queue controller TX queue spacing
552 * @r_vecs: Pre-allocated array of ring vectors
553 * @irq_entries: Pre-allocated array of MSI-X entries
554 * @lsc_handler: Handler for Link State Change interrupt
555 * @lsc_name: Name for Link State Change interrupt
556 * @exn_handler: Handler for Exception interrupt
557 * @exn_name: Name for Exception interrupt
558 * @shared_handler: Handler for shared interrupts
559 * @shared_name: Name for shared interrupt
560 * @me_freq_mhz: ME clock_freq (MHz)
561 * @reconfig_lock: Protects @reconfig_posted, @reconfig_timer_active,
562 * @reconfig_sync_present and HW reconfiguration request
563 * regs/machinery from async requests (sync must take
564 * @bar_lock)
565 * @reconfig_posted: Pending reconfig bits coming from async sources
566 * @reconfig_timer_active: Timer for reading reconfiguration results is pending
567 * @reconfig_sync_present: Some thread is performing synchronous reconfig
568 * @reconfig_timer: Timer for async reading of reconfig results
569 * @reconfig_in_progress_update: Update FW is processing now (debug only)
570 * @bar_lock: vNIC config BAR access lock, protects: update,
571 * mailbox area, crypto TLV
572 * @link_up: Is the link up?
573 * @link_status_lock: Protects @link_* and ensures atomicity with BAR reading
574 * @rx_coalesce_usecs: RX interrupt moderation usecs delay parameter
575 * @rx_coalesce_max_frames: RX interrupt moderation frame count parameter
576 * @tx_coalesce_usecs: TX interrupt moderation usecs delay parameter
577 * @tx_coalesce_max_frames: TX interrupt moderation frame count parameter
578 * @vxlan_ports: VXLAN ports for RX inner csum offload communicated to HW
579 * @vxlan_usecnt: IPv4/IPv6 VXLAN port use counts
580 * @qcp_cfg: Pointer to QCP queue used for configuration notification
581 * @tx_bar: Pointer to mapped TX queues
582 * @rx_bar: Pointer to mapped FL/RX queues
583 * @tlv_caps: Parsed TLV capabilities
584 * @ktls_tx_conn_cnt: Number of offloaded kTLS TX connections
585 * @ktls_rx_conn_cnt: Number of offloaded kTLS RX connections
586 * @ktls_conn_id_gen: Trivial generator for kTLS connection ids (for TX)
587 * @ktls_no_space: Counter of firmware rejecting kTLS connection due to
588 * lack of space
589 * @mbox_cmsg: Common Control Message via vNIC mailbox state
590 * @mbox_cmsg.queue: CCM mbox queue of pending messages
591 * @mbox_cmsg.wq: CCM mbox wait queue of waiting processes
592 * @mbox_cmsg.workq: CCM mbox work queue for @wait_work and @runq_work
593 * @mbox_cmsg.wait_work: CCM mbox posted msg reconfig wait work
594 * @mbox_cmsg.runq_work: CCM mbox posted msg queue runner work
595 * @mbox_cmsg.tag: CCM mbox message tag allocator
596 * @debugfs_dir: Device directory in debugfs
597 * @vnic_list: Entry on device vNIC list
598 * @pdev: Backpointer to PCI device
599 * @app: APP handle if available
600 * @vnic_no_name: For non-port PF vNIC make ndo_get_phys_port_name return
601 * -EOPNOTSUPP to keep backwards compatibility (set by app)
602 * @port: Pointer to nfp_port structure if vNIC is a port
603 * @app_priv: APP private data for this vNIC
604 */
605 struct nfp_net {
606 struct nfp_net_dp dp;
607
608 struct nfp_net_fw_version fw_ver;
609
610 u32 id;
611
612 u32 cap;
613 u32 max_mtu;
614
615 u8 rss_hfunc;
616 u32 rss_cfg;
617 u8 rss_key[NFP_NET_CFG_RSS_KEY_SZ];
618 u8 rss_itbl[NFP_NET_CFG_RSS_ITBL_SZ];
619
620 struct xdp_attachment_info xdp;
621 struct xdp_attachment_info xdp_hw;
622
623 unsigned int max_tx_rings;
624 unsigned int max_rx_rings;
625
626 int stride_tx;
627 int stride_rx;
628
629 unsigned int max_r_vecs;
630 struct nfp_net_r_vector r_vecs[NFP_NET_MAX_R_VECS];
631 struct msix_entry irq_entries[NFP_NET_MAX_IRQS];
632
633 irq_handler_t lsc_handler;
634 char lsc_name[IFNAMSIZ + 8];
635
636 irq_handler_t exn_handler;
637 char exn_name[IFNAMSIZ + 8];
638
639 irq_handler_t shared_handler;
640 char shared_name[IFNAMSIZ + 8];
641
642 u32 me_freq_mhz;
643
644 bool link_up;
645 spinlock_t link_status_lock;
646
647 spinlock_t reconfig_lock;
648 u32 reconfig_posted;
649 bool reconfig_timer_active;
650 bool reconfig_sync_present;
651 struct timer_list reconfig_timer;
652 u32 reconfig_in_progress_update;
653
654 struct semaphore bar_lock;
655
656 u32 rx_coalesce_usecs;
657 u32 rx_coalesce_max_frames;
658 u32 tx_coalesce_usecs;
659 u32 tx_coalesce_max_frames;
660
661 __be16 vxlan_ports[NFP_NET_N_VXLAN_PORTS];
662 u8 vxlan_usecnt[NFP_NET_N_VXLAN_PORTS];
663
664 u8 __iomem *qcp_cfg;
665
666 u8 __iomem *tx_bar;
667 u8 __iomem *rx_bar;
668
669 struct nfp_net_tlv_caps tlv_caps;
670
671 unsigned int ktls_tx_conn_cnt;
672 unsigned int ktls_rx_conn_cnt;
673
674 atomic64_t ktls_conn_id_gen;
675
676 atomic_t ktls_no_space;
677
678 struct {
679 struct sk_buff_head queue;
680 wait_queue_head_t wq;
681 struct workqueue_struct *workq;
682 struct work_struct wait_work;
683 struct work_struct runq_work;
684 u16 tag;
685 } mbox_cmsg;
686
687 struct dentry *debugfs_dir;
688
689 struct list_head vnic_list;
690
691 struct pci_dev *pdev;
692 struct nfp_app *app;
693
694 bool vnic_no_name;
695
696 struct nfp_port *port;
697
698 void *app_priv;
699 };
700
701 /* Functions to read/write from/to a BAR
702 * Performs any endian conversion necessary.
703 */
704 static inline u16 nn_readb(struct nfp_net *nn, int off)
705 {
706 return readb(nn->dp.ctrl_bar + off);
707 }
708
709 static inline void nn_writeb(struct nfp_net *nn, int off, u8 val)
710 {
711 writeb(val, nn->dp.ctrl_bar + off);
712 }
713
714 static inline u16 nn_readw(struct nfp_net *nn, int off)
715 {
716 return readw(nn->dp.ctrl_bar + off);
717 }
718
719 static inline void nn_writew(struct nfp_net *nn, int off, u16 val)
720 {
721 writew(val, nn->dp.ctrl_bar + off);
722 }
723
724 static inline u32 nn_readl(struct nfp_net *nn, int off)
725 {
726 return readl(nn->dp.ctrl_bar + off);
727 }
728
729 static inline void nn_writel(struct nfp_net *nn, int off, u32 val)
730 {
731 writel(val, nn->dp.ctrl_bar + off);
732 }
733
734 static inline u64 nn_readq(struct nfp_net *nn, int off)
735 {
736 return readq(nn->dp.ctrl_bar + off);
737 }
738
739 static inline void nn_writeq(struct nfp_net *nn, int off, u64 val)
740 {
741 writeq(val, nn->dp.ctrl_bar + off);
742 }
743
744 /* Flush posted PCI writes by reading something without side effects */
745 static inline void nn_pci_flush(struct nfp_net *nn)
746 {
747 nn_readl(nn, NFP_NET_CFG_VERSION);
748 }
749
750 /* Queue Controller Peripheral access functions and definitions.
751 *
752 * Some of the BARs of the NFP are mapped to portions of the Queue
753 * Controller Peripheral (QCP) address space on the NFP. A QCP queue
754 * has a read and a write pointer (as well as a size and flags,
755 * indicating overflow etc). The QCP offers a number of different
756 * operation on queue pointers, but here we only offer function to
757 * either add to a pointer or to read the pointer value.
758 */
759 #define NFP_QCP_QUEUE_ADDR_SZ 0x800
760 #define NFP_QCP_QUEUE_AREA_SZ 0x80000
761 #define NFP_QCP_QUEUE_OFF(_x) ((_x) * NFP_QCP_QUEUE_ADDR_SZ)
762 #define NFP_QCP_QUEUE_ADD_RPTR 0x0000
763 #define NFP_QCP_QUEUE_ADD_WPTR 0x0004
764 #define NFP_QCP_QUEUE_STS_LO 0x0008
765 #define NFP_QCP_QUEUE_STS_LO_READPTR_mask 0x3ffff
766 #define NFP_QCP_QUEUE_STS_HI 0x000c
767 #define NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask 0x3ffff
768
769 /* The offset of a QCP queues in the PCIe Target */
770 #define NFP_PCIE_QUEUE(_q) (0x80000 + (NFP_QCP_QUEUE_ADDR_SZ * ((_q) & 0xff)))
771
772 /* nfp_qcp_ptr - Read or Write Pointer of a queue */
773 enum nfp_qcp_ptr {
774 NFP_QCP_READ_PTR = 0,
775 NFP_QCP_WRITE_PTR
776 };
777
778 /* There appear to be an *undocumented* upper limit on the value which
779 * one can add to a queue and that value is either 0x3f or 0x7f. We
780 * go with 0x3f as a conservative measure.
781 */
782 #define NFP_QCP_MAX_ADD 0x3f
783
784 static inline void _nfp_qcp_ptr_add(u8 __iomem *q,
785 enum nfp_qcp_ptr ptr, u32 val)
786 {
787 u32 off;
788
789 if (ptr == NFP_QCP_READ_PTR)
790 off = NFP_QCP_QUEUE_ADD_RPTR;
791 else
792 off = NFP_QCP_QUEUE_ADD_WPTR;
793
794 while (val > NFP_QCP_MAX_ADD) {
795 writel(NFP_QCP_MAX_ADD, q + off);
796 val -= NFP_QCP_MAX_ADD;
797 }
798
799 writel(val, q + off);
800 }
801
802 /**
803 * nfp_qcp_rd_ptr_add() - Add the value to the read pointer of a queue
804 *
805 * @q: Base address for queue structure
806 * @val: Value to add to the queue pointer
807 *
808 * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed.
809 */
810 static inline void nfp_qcp_rd_ptr_add(u8 __iomem *q, u32 val)
811 {
812 _nfp_qcp_ptr_add(q, NFP_QCP_READ_PTR, val);
813 }
814
815 /**
816 * nfp_qcp_wr_ptr_add() - Add the value to the write pointer of a queue
817 *
818 * @q: Base address for queue structure
819 * @val: Value to add to the queue pointer
820 *
821 * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed.
822 */
823 static inline void nfp_qcp_wr_ptr_add(u8 __iomem *q, u32 val)
824 {
825 _nfp_qcp_ptr_add(q, NFP_QCP_WRITE_PTR, val);
826 }
827
828 static inline u32 _nfp_qcp_read(u8 __iomem *q, enum nfp_qcp_ptr ptr)
829 {
830 u32 off;
831 u32 val;
832
833 if (ptr == NFP_QCP_READ_PTR)
834 off = NFP_QCP_QUEUE_STS_LO;
835 else
836 off = NFP_QCP_QUEUE_STS_HI;
837
838 val = readl(q + off);
839
840 if (ptr == NFP_QCP_READ_PTR)
841 return val & NFP_QCP_QUEUE_STS_LO_READPTR_mask;
842 else
843 return val & NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask;
844 }
845
846 /**
847 * nfp_qcp_rd_ptr_read() - Read the current read pointer value for a queue
848 * @q: Base address for queue structure
849 *
850 * Return: Value read.
851 */
852 static inline u32 nfp_qcp_rd_ptr_read(u8 __iomem *q)
853 {
854 return _nfp_qcp_read(q, NFP_QCP_READ_PTR);
855 }
856
857 /**
858 * nfp_qcp_wr_ptr_read() - Read the current write pointer value for a queue
859 * @q: Base address for queue structure
860 *
861 * Return: Value read.
862 */
863 static inline u32 nfp_qcp_wr_ptr_read(u8 __iomem *q)
864 {
865 return _nfp_qcp_read(q, NFP_QCP_WRITE_PTR);
866 }
867
868 static inline bool nfp_net_is_data_vnic(struct nfp_net *nn)
869 {
870 WARN_ON_ONCE(!nn->dp.netdev && nn->port);
871 return !!nn->dp.netdev;
872 }
873
874 static inline bool nfp_net_running(struct nfp_net *nn)
875 {
876 return nn->dp.ctrl & NFP_NET_CFG_CTRL_ENABLE;
877 }
878
879 static inline const char *nfp_net_name(struct nfp_net *nn)
880 {
881 return nn->dp.netdev ? nn->dp.netdev->name : "ctrl";
882 }
883
884 static inline void nfp_ctrl_lock(struct nfp_net *nn)
885 __acquires(&nn->r_vecs[0].lock)
886 {
887 spin_lock_bh(&nn->r_vecs[0].lock);
888 }
889
890 static inline void nfp_ctrl_unlock(struct nfp_net *nn)
891 __releases(&nn->r_vecs[0].lock)
892 {
893 spin_unlock_bh(&nn->r_vecs[0].lock);
894 }
895
896 static inline void nn_ctrl_bar_lock(struct nfp_net *nn)
897 {
898 down(&nn->bar_lock);
899 }
900
901 static inline bool nn_ctrl_bar_trylock(struct nfp_net *nn)
902 {
903 return !down_trylock(&nn->bar_lock);
904 }
905
906 static inline void nn_ctrl_bar_unlock(struct nfp_net *nn)
907 {
908 up(&nn->bar_lock);
909 }
910
911 /* Globals */
912 extern const char nfp_driver_version[];
913
914 extern const struct net_device_ops nfp_net_netdev_ops;
915
916 static inline bool nfp_netdev_is_nfp_net(struct net_device *netdev)
917 {
918 return netdev->netdev_ops == &nfp_net_netdev_ops;
919 }
920
921 /* Prototypes */
922 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
923 void __iomem *ctrl_bar);
924
925 struct nfp_net *
926 nfp_net_alloc(struct pci_dev *pdev, void __iomem *ctrl_bar, bool needs_netdev,
927 unsigned int max_tx_rings, unsigned int max_rx_rings);
928 void nfp_net_free(struct nfp_net *nn);
929
930 int nfp_net_init(struct nfp_net *nn);
931 void nfp_net_clean(struct nfp_net *nn);
932
933 int nfp_ctrl_open(struct nfp_net *nn);
934 void nfp_ctrl_close(struct nfp_net *nn);
935
936 void nfp_net_set_ethtool_ops(struct net_device *netdev);
937 void nfp_net_info(struct nfp_net *nn);
938 int __nfp_net_reconfig(struct nfp_net *nn, u32 update);
939 int nfp_net_reconfig(struct nfp_net *nn, u32 update);
940 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn);
941 void nfp_net_rss_write_itbl(struct nfp_net *nn);
942 void nfp_net_rss_write_key(struct nfp_net *nn);
943 void nfp_net_coalesce_write_cfg(struct nfp_net *nn);
944 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size);
945 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd);
946 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd);
947 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 update);
948 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn);
949
950 unsigned int
951 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
952 unsigned int min_irqs, unsigned int want_irqs);
953 void nfp_net_irqs_disable(struct pci_dev *pdev);
954 void
955 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
956 unsigned int n);
957
958 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn);
959 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *new,
960 struct netlink_ext_ack *extack);
961
962 #ifdef CONFIG_NFP_DEBUG
963 void nfp_net_debugfs_create(void);
964 void nfp_net_debugfs_destroy(void);
965 struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev);
966 void nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir);
967 void nfp_net_debugfs_dir_clean(struct dentry **dir);
968 #else
969 static inline void nfp_net_debugfs_create(void)
970 {
971 }
972
973 static inline void nfp_net_debugfs_destroy(void)
974 {
975 }
976
977 static inline struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev)
978 {
979 return NULL;
980 }
981
982 static inline void
983 nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir)
984 {
985 }
986
987 static inline void nfp_net_debugfs_dir_clean(struct dentry **dir)
988 {
989 }
990 #endif /* CONFIG_NFP_DEBUG */
991
992 #endif /* _NFP_NET_H_ */