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/dim.h>
21 #include <linux/io-64-nonatomic-hi-lo.h>
22 #include <linux/semaphore.h>
23 #include <linux/workqueue.h>
24 #include <net/xdp.h>
25 
26 #include "nfp_net_ctrl.h"
27 
28 #define nn_pr(nn, lvl, fmt, args...)					\
29 	({								\
30 		struct nfp_net *__nn = (nn);				\
31 									\
32 		if (__nn->dp.netdev)					\
33 			netdev_printk(lvl, __nn->dp.netdev, fmt, ## args); \
34 		else							\
35 			dev_printk(lvl, __nn->dp.dev, "ctrl: " fmt, ## args); \
36 	})
37 
38 #define nn_err(nn, fmt, args...)	nn_pr(nn, KERN_ERR, fmt, ## args)
39 #define nn_warn(nn, fmt, args...)	nn_pr(nn, KERN_WARNING, fmt, ## args)
40 #define nn_info(nn, fmt, args...)	nn_pr(nn, KERN_INFO, fmt, ## args)
41 #define nn_dbg(nn, fmt, args...)	nn_pr(nn, KERN_DEBUG, fmt, ## args)
42 
43 #define nn_dp_warn(dp, fmt, args...)					\
44 	({								\
45 		struct nfp_net_dp *__dp = (dp);				\
46 									\
47 		if (unlikely(net_ratelimit())) {			\
48 			if (__dp->netdev)				\
49 				netdev_warn(__dp->netdev, fmt, ## args); \
50 			else						\
51 				dev_warn(__dp->dev, fmt, ## args);	\
52 		}							\
53 	})
54 
55 /* Max time to wait for NFP to respond on updates (in seconds) */
56 #define NFP_NET_POLL_TIMEOUT	5
57 
58 /* Interval for reading offloaded filter stats */
59 #define NFP_NET_STAT_POLL_IVL	msecs_to_jiffies(100)
60 
61 /* Bar allocation */
62 #define NFP_NET_CTRL_BAR	0
63 #define NFP_NET_Q0_BAR		2
64 #define NFP_NET_Q1_BAR		4	/* OBSOLETE */
65 
66 /* Max bits in DMA address */
67 #define NFP_NET_MAX_DMA_BITS	40
68 
69 /* Default size for MTU and freelist buffer sizes */
70 #define NFP_NET_DEFAULT_MTU		1500U
71 
72 /* Maximum number of bytes prepended to a packet */
73 #define NFP_NET_MAX_PREPEND		64
74 
75 /* Interrupt definitions */
76 #define NFP_NET_NON_Q_VECTORS		2
77 #define NFP_NET_IRQ_LSC_IDX		0
78 #define NFP_NET_IRQ_EXN_IDX		1
79 #define NFP_NET_MIN_VNIC_IRQS		(NFP_NET_NON_Q_VECTORS + 1)
80 
81 /* Queue/Ring definitions */
82 #define NFP_NET_MAX_TX_RINGS	64	/* Max. # of Tx rings per device */
83 #define NFP_NET_MAX_RX_RINGS	64	/* Max. # of Rx rings per device */
84 #define NFP_NET_MAX_R_VECS	(NFP_NET_MAX_TX_RINGS > NFP_NET_MAX_RX_RINGS ? \
85 				 NFP_NET_MAX_TX_RINGS : NFP_NET_MAX_RX_RINGS)
86 #define NFP_NET_MAX_IRQS	(NFP_NET_NON_Q_VECTORS + NFP_NET_MAX_R_VECS)
87 
88 #define NFP_NET_MIN_TX_DESCS	256	/* Min. # of Tx descs per ring */
89 #define NFP_NET_MIN_RX_DESCS	256	/* Min. # of Rx descs per ring */
90 #define NFP_NET_MAX_TX_DESCS	(256 * 1024) /* Max. # of Tx descs per ring */
91 #define NFP_NET_MAX_RX_DESCS	(256 * 1024) /* Max. # of Rx descs per ring */
92 
93 #define NFP_NET_TX_DESCS_DEFAULT 4096	/* Default # of Tx descs per ring */
94 #define NFP_NET_RX_DESCS_DEFAULT 4096	/* Default # of Rx descs per ring */
95 
96 #define NFP_NET_FL_BATCH	16	/* Add freelist in this Batch size */
97 #define NFP_NET_XDP_MAX_COMPLETE 2048	/* XDP bufs to reclaim in NAPI poll */
98 
99 /* Offload definitions */
100 #define NFP_NET_N_VXLAN_PORTS	(NFP_NET_CFG_VXLAN_SZ / sizeof(__be16))
101 
102 #define NFP_NET_RX_BUF_HEADROOM	(NET_SKB_PAD + NET_IP_ALIGN)
103 #define NFP_NET_RX_BUF_NON_DATA	(NFP_NET_RX_BUF_HEADROOM +		\
104 				 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
105 
106 /* Forward declarations */
107 struct nfp_cpp;
108 struct nfp_eth_table_port;
109 struct nfp_net;
110 struct nfp_net_r_vector;
111 struct nfp_port;
112 struct xsk_buff_pool;
113 
114 /* Convenience macro for wrapping descriptor index on ring size */
115 #define D_IDX(ring, idx)	((idx) & ((ring)->cnt - 1))
116 
117 /* Convenience macro for writing dma address into RX/TX descriptors */
118 #define nfp_desc_set_dma_addr(desc, dma_addr)				\
119 	do {								\
120 		__typeof(desc) __d = (desc);				\
121 		dma_addr_t __addr = (dma_addr);				\
122 									\
123 		__d->dma_addr_lo = cpu_to_le32(lower_32_bits(__addr));	\
124 		__d->dma_addr_hi = upper_32_bits(__addr) & 0xff;	\
125 	} while (0)
126 
127 /* TX descriptor format */
128 
129 #define PCIE_DESC_TX_EOP		BIT(7)
130 #define PCIE_DESC_TX_OFFSET_MASK	GENMASK(6, 0)
131 #define PCIE_DESC_TX_MSS_MASK		GENMASK(13, 0)
132 
133 /* Flags in the host TX descriptor */
134 #define PCIE_DESC_TX_CSUM		BIT(7)
135 #define PCIE_DESC_TX_IP4_CSUM		BIT(6)
136 #define PCIE_DESC_TX_TCP_CSUM		BIT(5)
137 #define PCIE_DESC_TX_UDP_CSUM		BIT(4)
138 #define PCIE_DESC_TX_VLAN		BIT(3)
139 #define PCIE_DESC_TX_LSO		BIT(2)
140 #define PCIE_DESC_TX_ENCAP		BIT(1)
141 #define PCIE_DESC_TX_O_IP4_CSUM	BIT(0)
142 
143 struct nfp_net_tx_desc {
144 	union {
145 		struct {
146 			u8 dma_addr_hi; /* High bits of host buf address */
147 			__le16 dma_len;	/* Length to DMA for this desc */
148 			u8 offset_eop;	/* Offset in buf where pkt starts +
149 					 * highest bit is eop flag.
150 					 */
151 			__le32 dma_addr_lo; /* Low 32bit of host buf addr */
152 
153 			__le16 mss;	/* MSS to be used for LSO */
154 			u8 lso_hdrlen;	/* LSO, TCP payload offset */
155 			u8 flags;	/* TX Flags, see @PCIE_DESC_TX_* */
156 			union {
157 				struct {
158 					u8 l3_offset; /* L3 header offset */
159 					u8 l4_offset; /* L4 header offset */
160 				};
161 				__le16 vlan; /* VLAN tag to add if indicated */
162 			};
163 			__le16 data_len; /* Length of frame + meta data */
164 		} __packed;
165 		__le32 vals[4];
166 		__le64 vals8[2];
167 	};
168 };
169 
170 /**
171  * struct nfp_net_tx_buf - software TX buffer descriptor
172  * @skb:	normal ring, sk_buff associated with this buffer
173  * @frag:	XDP ring, page frag associated with this buffer
174  * @dma_addr:	DMA mapping address of the buffer
175  * @fidx:	Fragment index (-1 for the head and [0..nr_frags-1] for frags)
176  * @pkt_cnt:	Number of packets to be produced out of the skb associated
177  *		with this buffer (valid only on the head's buffer).
178  *		Will be 1 for all non-TSO packets.
179  * @real_len:	Number of bytes which to be produced out of the skb (valid only
180  *		on the head's buffer). Equal to skb->len for non-TSO packets.
181  */
182 struct nfp_net_tx_buf {
183 	union {
184 		struct sk_buff *skb;
185 		void *frag;
186 	};
187 	dma_addr_t dma_addr;
188 	short int fidx;
189 	u16 pkt_cnt;
190 	u32 real_len;
191 };
192 
193 /**
194  * struct nfp_net_tx_ring - TX ring structure
195  * @r_vec:      Back pointer to ring vector structure
196  * @idx:        Ring index from Linux's perspective
197  * @qcidx:      Queue Controller Peripheral (QCP) queue index for the TX queue
198  * @qcp_q:      Pointer to base of the QCP TX queue
199  * @cnt:        Size of the queue in number of descriptors
200  * @wr_p:       TX ring write pointer (free running)
201  * @rd_p:       TX ring read pointer (free running)
202  * @qcp_rd_p:   Local copy of QCP TX queue read pointer
203  * @wr_ptr_add:	Accumulated number of buffers to add to QCP write pointer
204  *		(used for .xmit_more delayed kick)
205  * @txbufs:     Array of transmitted TX buffers, to free on transmit
206  * @txds:       Virtual address of TX ring in host memory
207  * @dma:        DMA address of the TX ring
208  * @size:       Size, in bytes, of the TX ring (needed to free)
209  * @is_xdp:	Is this a XDP TX ring?
210  */
211 struct nfp_net_tx_ring {
212 	struct nfp_net_r_vector *r_vec;
213 
214 	u32 idx;
215 	int qcidx;
216 	u8 __iomem *qcp_q;
217 
218 	u32 cnt;
219 	u32 wr_p;
220 	u32 rd_p;
221 	u32 qcp_rd_p;
222 
223 	u32 wr_ptr_add;
224 
225 	struct nfp_net_tx_buf *txbufs;
226 	struct nfp_net_tx_desc *txds;
227 
228 	dma_addr_t dma;
229 	size_t size;
230 	bool is_xdp;
231 } ____cacheline_aligned;
232 
233 /* RX and freelist descriptor format */
234 
235 #define PCIE_DESC_RX_DD			BIT(7)
236 #define PCIE_DESC_RX_META_LEN_MASK	GENMASK(6, 0)
237 
238 /* Flags in the RX descriptor */
239 #define PCIE_DESC_RX_RSS		cpu_to_le16(BIT(15))
240 #define PCIE_DESC_RX_I_IP4_CSUM		cpu_to_le16(BIT(14))
241 #define PCIE_DESC_RX_I_IP4_CSUM_OK	cpu_to_le16(BIT(13))
242 #define PCIE_DESC_RX_I_TCP_CSUM		cpu_to_le16(BIT(12))
243 #define PCIE_DESC_RX_I_TCP_CSUM_OK	cpu_to_le16(BIT(11))
244 #define PCIE_DESC_RX_I_UDP_CSUM		cpu_to_le16(BIT(10))
245 #define PCIE_DESC_RX_I_UDP_CSUM_OK	cpu_to_le16(BIT(9))
246 #define PCIE_DESC_RX_DECRYPTED		cpu_to_le16(BIT(8))
247 #define PCIE_DESC_RX_EOP		cpu_to_le16(BIT(7))
248 #define PCIE_DESC_RX_IP4_CSUM		cpu_to_le16(BIT(6))
249 #define PCIE_DESC_RX_IP4_CSUM_OK	cpu_to_le16(BIT(5))
250 #define PCIE_DESC_RX_TCP_CSUM		cpu_to_le16(BIT(4))
251 #define PCIE_DESC_RX_TCP_CSUM_OK	cpu_to_le16(BIT(3))
252 #define PCIE_DESC_RX_UDP_CSUM		cpu_to_le16(BIT(2))
253 #define PCIE_DESC_RX_UDP_CSUM_OK	cpu_to_le16(BIT(1))
254 #define PCIE_DESC_RX_VLAN		cpu_to_le16(BIT(0))
255 
256 #define PCIE_DESC_RX_CSUM_ALL		(PCIE_DESC_RX_IP4_CSUM |	\
257 					 PCIE_DESC_RX_TCP_CSUM |	\
258 					 PCIE_DESC_RX_UDP_CSUM |	\
259 					 PCIE_DESC_RX_I_IP4_CSUM |	\
260 					 PCIE_DESC_RX_I_TCP_CSUM |	\
261 					 PCIE_DESC_RX_I_UDP_CSUM)
262 #define PCIE_DESC_RX_CSUM_OK_SHIFT	1
263 #define __PCIE_DESC_RX_CSUM_ALL		le16_to_cpu(PCIE_DESC_RX_CSUM_ALL)
264 #define __PCIE_DESC_RX_CSUM_ALL_OK	(__PCIE_DESC_RX_CSUM_ALL >>	\
265 					 PCIE_DESC_RX_CSUM_OK_SHIFT)
266 
267 struct nfp_net_rx_desc {
268 	union {
269 		struct {
270 			u8 dma_addr_hi;	/* High bits of the buf address */
271 			__le16 reserved; /* Must be zero */
272 			u8 meta_len_dd; /* Must be zero */
273 
274 			__le32 dma_addr_lo; /* Low bits of the buffer address */
275 		} __packed fld;
276 
277 		struct {
278 			__le16 data_len; /* Length of the frame + meta data */
279 			u8 reserved;
280 			u8 meta_len_dd;	/* Length of meta data prepended +
281 					 * descriptor done flag.
282 					 */
283 
284 			__le16 flags;	/* RX flags. See @PCIE_DESC_RX_* */
285 			__le16 vlan;	/* VLAN if stripped */
286 		} __packed rxd;
287 
288 		__le32 vals[2];
289 	};
290 };
291 
292 #define NFP_NET_META_FIELD_MASK GENMASK(NFP_NET_META_FIELD_SIZE - 1, 0)
293 
294 struct nfp_meta_parsed {
295 	u8 hash_type;
296 	u8 csum_type;
297 	u32 hash;
298 	u32 mark;
299 	u32 portid;
300 	__wsum csum;
301 };
302 
303 struct nfp_net_rx_hash {
304 	__be32 hash_type;
305 	__be32 hash;
306 };
307 
308 /**
309  * struct nfp_net_rx_buf - software RX buffer descriptor
310  * @frag:	page fragment buffer
311  * @dma_addr:	DMA mapping address of the buffer
312  */
313 struct nfp_net_rx_buf {
314 	void *frag;
315 	dma_addr_t dma_addr;
316 };
317 
318 /**
319  * struct nfp_net_rx_ring - RX ring structure
320  * @r_vec:      Back pointer to ring vector structure
321  * @cnt:        Size of the queue in number of descriptors
322  * @wr_p:       FL/RX ring write pointer (free running)
323  * @rd_p:       FL/RX ring read pointer (free running)
324  * @idx:        Ring index from Linux's perspective
325  * @fl_qcidx:   Queue Controller Peripheral (QCP) queue index for the freelist
326  * @qcp_fl:     Pointer to base of the QCP freelist queue
327  * @rxbufs:     Array of transmitted FL/RX buffers
328  * @rxds:       Virtual address of FL/RX ring in host memory
329  * @xdp_rxq:    RX-ring info avail for XDP
330  * @dma:        DMA address of the FL/RX ring
331  * @size:       Size, in bytes, of the FL/RX ring (needed to free)
332  */
333 struct nfp_net_rx_ring {
334 	struct nfp_net_r_vector *r_vec;
335 
336 	u32 cnt;
337 	u32 wr_p;
338 	u32 rd_p;
339 
340 	u32 idx;
341 
342 	int fl_qcidx;
343 	u8 __iomem *qcp_fl;
344 
345 	struct nfp_net_rx_buf *rxbufs;
346 	struct nfp_net_rx_desc *rxds;
347 
348 	struct xdp_rxq_info xdp_rxq;
349 
350 	dma_addr_t dma;
351 	size_t size;
352 } ____cacheline_aligned;
353 
354 /**
355  * struct nfp_net_r_vector - Per ring interrupt vector configuration
356  * @nfp_net:        Backpointer to nfp_net structure
357  * @napi:           NAPI structure for this ring vec
358  * @tasklet:        ctrl vNIC, tasklet for servicing the r_vec
359  * @queue:          ctrl vNIC, send queue
360  * @lock:           ctrl vNIC, r_vec lock protects @queue
361  * @tx_ring:        Pointer to TX ring
362  * @rx_ring:        Pointer to RX ring
363  * @xdp_ring:	    Pointer to an extra TX ring for XDP
364  * @irq_entry:      MSI-X table entry (use for talking to the device)
365  * @event_ctr:	    Number of interrupt
366  * @rx_dim:	    Dynamic interrupt moderation structure for RX
367  * @tx_dim:	    Dynamic interrupt moderation structure for TX
368  * @rx_sync:	    Seqlock for atomic updates of RX stats
369  * @rx_pkts:        Number of received packets
370  * @rx_bytes:	    Number of received bytes
371  * @rx_drops:	    Number of packets dropped on RX due to lack of resources
372  * @hw_csum_rx_ok:  Counter of packets where the HW checksum was OK
373  * @hw_csum_rx_inner_ok: Counter of packets where the inner HW checksum was OK
374  * @hw_csum_rx_complete: Counter of packets with CHECKSUM_COMPLETE reported
375  * @hw_csum_rx_error:	 Counter of packets with bad checksums
376  * @hw_tls_rx:	    Number of packets with TLS decrypted by hardware
377  * @tx_sync:	    Seqlock for atomic updates of TX stats
378  * @tx_pkts:	    Number of Transmitted packets
379  * @tx_bytes:	    Number of Transmitted bytes
380  * @hw_csum_tx:	    Counter of packets with TX checksum offload requested
381  * @hw_csum_tx_inner:	 Counter of inner TX checksum offload requests
382  * @tx_gather:	    Counter of packets with Gather DMA
383  * @tx_lso:	    Counter of LSO packets sent
384  * @hw_tls_tx:	    Counter of TLS packets sent with crypto offloaded to HW
385  * @tls_tx_fallback:	Counter of TLS packets sent which had to be encrypted
386  *			by the fallback path because packets came out of order
387  * @tls_tx_no_fallback:	Counter of TLS packets not sent because the fallback
388  *			path could not encrypt them
389  * @tx_errors:	    How many TX errors were encountered
390  * @tx_busy:        How often was TX busy (no space)?
391  * @rx_replace_buf_alloc_fail:	Counter of RX buffer allocation failures
392  * @irq_vector:     Interrupt vector number (use for talking to the OS)
393  * @handler:        Interrupt handler for this ring vector
394  * @name:           Name of the interrupt vector
395  * @affinity_mask:  SMP affinity mask for this vector
396  *
397  * This structure ties RX and TX rings to interrupt vectors and a NAPI
398  * context. This currently only supports one RX and TX ring per
399  * interrupt vector but might be extended in the future to allow
400  * association of multiple rings per vector.
401  */
402 struct nfp_net_r_vector {
403 	struct nfp_net *nfp_net;
404 	union {
405 		struct napi_struct napi;
406 		struct {
407 			struct tasklet_struct tasklet;
408 			struct sk_buff_head queue;
409 			spinlock_t lock;
410 		};
411 	};
412 
413 	struct nfp_net_tx_ring *tx_ring;
414 	struct nfp_net_rx_ring *rx_ring;
415 
416 	u16 irq_entry;
417 
418 	u16 event_ctr;
419 	struct dim rx_dim;
420 	struct dim tx_dim;
421 
422 	struct u64_stats_sync rx_sync;
423 	u64 rx_pkts;
424 	u64 rx_bytes;
425 	u64 rx_drops;
426 	u64 hw_csum_rx_ok;
427 	u64 hw_csum_rx_inner_ok;
428 	u64 hw_csum_rx_complete;
429 	u64 hw_tls_rx;
430 
431 	u64 hw_csum_rx_error;
432 	u64 rx_replace_buf_alloc_fail;
433 
434 	struct nfp_net_tx_ring *xdp_ring;
435 
436 	struct u64_stats_sync tx_sync;
437 	u64 tx_pkts;
438 	u64 tx_bytes;
439 
440 	u64 ____cacheline_aligned_in_smp hw_csum_tx;
441 	u64 hw_csum_tx_inner;
442 	u64 tx_gather;
443 	u64 tx_lso;
444 	u64 hw_tls_tx;
445 
446 	u64 tls_tx_fallback;
447 	u64 tls_tx_no_fallback;
448 	u64 tx_errors;
449 	u64 tx_busy;
450 
451 	/* Cold data follows */
452 
453 	u32 irq_vector;
454 	irq_handler_t handler;
455 	char name[IFNAMSIZ + 8];
456 	cpumask_t affinity_mask;
457 } ____cacheline_aligned;
458 
459 /* Firmware version as it is written in the 32bit value in the BAR */
460 struct nfp_net_fw_version {
461 	u8 minor;
462 	u8 major;
463 	u8 class;
464 	u8 resv;
465 } __packed;
466 
467 static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver,
468 				     u8 resv, u8 class, u8 major, u8 minor)
469 {
470 	return fw_ver->resv == resv &&
471 	       fw_ver->class == class &&
472 	       fw_ver->major == major &&
473 	       fw_ver->minor == minor;
474 }
475 
476 struct nfp_stat_pair {
477 	u64 pkts;
478 	u64 bytes;
479 };
480 
481 /**
482  * struct nfp_net_dp - NFP network device datapath data structure
483  * @dev:		Backpointer to struct device
484  * @netdev:		Backpointer to net_device structure
485  * @is_vf:		Is the driver attached to a VF?
486  * @chained_metadata_format:  Firemware will use new metadata format
487  * @ktls_tx:		Is kTLS TX enabled?
488  * @rx_dma_dir:		Mapping direction for RX buffers
489  * @rx_dma_off:		Offset at which DMA packets (for XDP headroom)
490  * @rx_offset:		Offset in the RX buffers where packet data starts
491  * @ctrl:		Local copy of the control register/word.
492  * @fl_bufsz:		Currently configured size of the freelist buffers
493  * @xdp_prog:		Installed XDP program
494  * @tx_rings:		Array of pre-allocated TX ring structures
495  * @rx_rings:		Array of pre-allocated RX ring structures
496  * @ctrl_bar:		Pointer to mapped control BAR
497  *
498  * @txd_cnt:		Size of the TX ring in number of descriptors
499  * @rxd_cnt:		Size of the RX ring in number of descriptors
500  * @num_r_vecs:		Number of used ring vectors
501  * @num_tx_rings:	Currently configured number of TX rings
502  * @num_stack_tx_rings:	Number of TX rings used by the stack (not XDP)
503  * @num_rx_rings:	Currently configured number of RX rings
504  * @mtu:		Device MTU
505  * @xsk_pools:		AF_XDP UMEM table (@num_r_vecs in size)
506  */
507 struct nfp_net_dp {
508 	struct device *dev;
509 	struct net_device *netdev;
510 
511 	u8 is_vf:1;
512 	u8 chained_metadata_format:1;
513 	u8 ktls_tx:1;
514 
515 	u8 rx_dma_dir;
516 	u8 rx_offset;
517 
518 	u32 rx_dma_off;
519 
520 	u32 ctrl;
521 	u32 fl_bufsz;
522 
523 	struct bpf_prog *xdp_prog;
524 
525 	struct nfp_net_tx_ring *tx_rings;
526 	struct nfp_net_rx_ring *rx_rings;
527 
528 	u8 __iomem *ctrl_bar;
529 
530 	/* Cold data follows */
531 
532 	unsigned int txd_cnt;
533 	unsigned int rxd_cnt;
534 
535 	unsigned int num_r_vecs;
536 
537 	unsigned int num_tx_rings;
538 	unsigned int num_stack_tx_rings;
539 	unsigned int num_rx_rings;
540 
541 	unsigned int mtu;
542 
543 	struct xsk_buff_pool **xsk_pools;
544 };
545 
546 /**
547  * struct nfp_net - NFP network device structure
548  * @dp:			Datapath structure
549  * @id:			vNIC id within the PF (0 for VFs)
550  * @fw_ver:		Firmware version
551  * @cap:                Capabilities advertised by the Firmware
552  * @max_mtu:            Maximum support MTU advertised by the Firmware
553  * @rss_hfunc:		RSS selected hash function
554  * @rss_cfg:            RSS configuration
555  * @rss_key:            RSS secret key
556  * @rss_itbl:           RSS indirection table
557  * @xdp:		Information about the driver XDP program
558  * @xdp_hw:		Information about the HW XDP program
559  * @max_r_vecs:		Number of allocated interrupt vectors for RX/TX
560  * @max_tx_rings:       Maximum number of TX rings supported by the Firmware
561  * @max_rx_rings:       Maximum number of RX rings supported by the Firmware
562  * @stride_rx:		Queue controller RX queue spacing
563  * @stride_tx:		Queue controller TX queue spacing
564  * @r_vecs:             Pre-allocated array of ring vectors
565  * @irq_entries:        Pre-allocated array of MSI-X entries
566  * @lsc_handler:        Handler for Link State Change interrupt
567  * @lsc_name:           Name for Link State Change interrupt
568  * @exn_handler:        Handler for Exception interrupt
569  * @exn_name:           Name for Exception interrupt
570  * @shared_handler:     Handler for shared interrupts
571  * @shared_name:        Name for shared interrupt
572  * @reconfig_lock:	Protects @reconfig_posted, @reconfig_timer_active,
573  *			@reconfig_sync_present and HW reconfiguration request
574  *			regs/machinery from async requests (sync must take
575  *			@bar_lock)
576  * @reconfig_posted:	Pending reconfig bits coming from async sources
577  * @reconfig_timer_active:  Timer for reading reconfiguration results is pending
578  * @reconfig_sync_present:  Some thread is performing synchronous reconfig
579  * @reconfig_timer:	Timer for async reading of reconfig results
580  * @reconfig_in_progress_update:	Update FW is processing now (debug only)
581  * @bar_lock:		vNIC config BAR access lock, protects: update,
582  *			mailbox area, crypto TLV
583  * @link_up:            Is the link up?
584  * @link_status_lock:	Protects @link_* and ensures atomicity with BAR reading
585  * @rx_coalesce_adapt_on:   Is RX interrupt moderation adaptive?
586  * @tx_coalesce_adapt_on:   Is TX interrupt moderation adaptive?
587  * @rx_coalesce_usecs:      RX interrupt moderation usecs delay parameter
588  * @rx_coalesce_max_frames: RX interrupt moderation frame count parameter
589  * @tx_coalesce_usecs:      TX interrupt moderation usecs delay parameter
590  * @tx_coalesce_max_frames: TX interrupt moderation frame count parameter
591  * @qcp_cfg:            Pointer to QCP queue used for configuration notification
592  * @tx_bar:             Pointer to mapped TX queues
593  * @rx_bar:             Pointer to mapped FL/RX queues
594  * @tlv_caps:		Parsed TLV capabilities
595  * @ktls_tx_conn_cnt:	Number of offloaded kTLS TX connections
596  * @ktls_rx_conn_cnt:	Number of offloaded kTLS RX connections
597  * @ktls_conn_id_gen:	Trivial generator for kTLS connection ids (for TX)
598  * @ktls_no_space:	Counter of firmware rejecting kTLS connection due to
599  *			lack of space
600  * @ktls_rx_resync_req:	Counter of TLS RX resync requested
601  * @ktls_rx_resync_ign:	Counter of TLS RX resync requests ignored
602  * @ktls_rx_resync_sent:    Counter of TLS RX resync completed
603  * @mbox_cmsg:		Common Control Message via vNIC mailbox state
604  * @mbox_cmsg.queue:	CCM mbox queue of pending messages
605  * @mbox_cmsg.wq:	CCM mbox wait queue of waiting processes
606  * @mbox_cmsg.workq:	CCM mbox work queue for @wait_work and @runq_work
607  * @mbox_cmsg.wait_work:    CCM mbox posted msg reconfig wait work
608  * @mbox_cmsg.runq_work:    CCM mbox posted msg queue runner work
609  * @mbox_cmsg.tag:	CCM mbox message tag allocator
610  * @debugfs_dir:	Device directory in debugfs
611  * @vnic_list:		Entry on device vNIC list
612  * @pdev:		Backpointer to PCI device
613  * @app:		APP handle if available
614  * @vnic_no_name:	For non-port PF vNIC make ndo_get_phys_port_name return
615  *			-EOPNOTSUPP to keep backwards compatibility (set by app)
616  * @port:		Pointer to nfp_port structure if vNIC is a port
617  * @app_priv:		APP private data for this vNIC
618  */
619 struct nfp_net {
620 	struct nfp_net_dp dp;
621 
622 	struct nfp_net_fw_version fw_ver;
623 
624 	u32 id;
625 
626 	u32 cap;
627 	u32 max_mtu;
628 
629 	u8 rss_hfunc;
630 	u32 rss_cfg;
631 	u8 rss_key[NFP_NET_CFG_RSS_KEY_SZ];
632 	u8 rss_itbl[NFP_NET_CFG_RSS_ITBL_SZ];
633 
634 	struct xdp_attachment_info xdp;
635 	struct xdp_attachment_info xdp_hw;
636 
637 	unsigned int max_tx_rings;
638 	unsigned int max_rx_rings;
639 
640 	int stride_tx;
641 	int stride_rx;
642 
643 	unsigned int max_r_vecs;
644 	struct nfp_net_r_vector r_vecs[NFP_NET_MAX_R_VECS];
645 	struct msix_entry irq_entries[NFP_NET_MAX_IRQS];
646 
647 	irq_handler_t lsc_handler;
648 	char lsc_name[IFNAMSIZ + 8];
649 
650 	irq_handler_t exn_handler;
651 	char exn_name[IFNAMSIZ + 8];
652 
653 	irq_handler_t shared_handler;
654 	char shared_name[IFNAMSIZ + 8];
655 
656 	bool link_up;
657 	spinlock_t link_status_lock;
658 
659 	spinlock_t reconfig_lock;
660 	u32 reconfig_posted;
661 	bool reconfig_timer_active;
662 	bool reconfig_sync_present;
663 	struct timer_list reconfig_timer;
664 	u32 reconfig_in_progress_update;
665 
666 	struct semaphore bar_lock;
667 
668 	bool rx_coalesce_adapt_on;
669 	bool tx_coalesce_adapt_on;
670 	u32 rx_coalesce_usecs;
671 	u32 rx_coalesce_max_frames;
672 	u32 tx_coalesce_usecs;
673 	u32 tx_coalesce_max_frames;
674 
675 	u8 __iomem *qcp_cfg;
676 
677 	u8 __iomem *tx_bar;
678 	u8 __iomem *rx_bar;
679 
680 	struct nfp_net_tlv_caps tlv_caps;
681 
682 	unsigned int ktls_tx_conn_cnt;
683 	unsigned int ktls_rx_conn_cnt;
684 
685 	atomic64_t ktls_conn_id_gen;
686 
687 	atomic_t ktls_no_space;
688 	atomic_t ktls_rx_resync_req;
689 	atomic_t ktls_rx_resync_ign;
690 	atomic_t ktls_rx_resync_sent;
691 
692 	struct {
693 		struct sk_buff_head queue;
694 		wait_queue_head_t wq;
695 		struct workqueue_struct *workq;
696 		struct work_struct wait_work;
697 		struct work_struct runq_work;
698 		u16 tag;
699 	} mbox_cmsg;
700 
701 	struct dentry *debugfs_dir;
702 
703 	struct list_head vnic_list;
704 
705 	struct pci_dev *pdev;
706 	struct nfp_app *app;
707 
708 	bool vnic_no_name;
709 
710 	struct nfp_port *port;
711 
712 	void *app_priv;
713 };
714 
715 /* Functions to read/write from/to a BAR
716  * Performs any endian conversion necessary.
717  */
718 static inline u16 nn_readb(struct nfp_net *nn, int off)
719 {
720 	return readb(nn->dp.ctrl_bar + off);
721 }
722 
723 static inline void nn_writeb(struct nfp_net *nn, int off, u8 val)
724 {
725 	writeb(val, nn->dp.ctrl_bar + off);
726 }
727 
728 static inline u16 nn_readw(struct nfp_net *nn, int off)
729 {
730 	return readw(nn->dp.ctrl_bar + off);
731 }
732 
733 static inline void nn_writew(struct nfp_net *nn, int off, u16 val)
734 {
735 	writew(val, nn->dp.ctrl_bar + off);
736 }
737 
738 static inline u32 nn_readl(struct nfp_net *nn, int off)
739 {
740 	return readl(nn->dp.ctrl_bar + off);
741 }
742 
743 static inline void nn_writel(struct nfp_net *nn, int off, u32 val)
744 {
745 	writel(val, nn->dp.ctrl_bar + off);
746 }
747 
748 static inline u64 nn_readq(struct nfp_net *nn, int off)
749 {
750 	return readq(nn->dp.ctrl_bar + off);
751 }
752 
753 static inline void nn_writeq(struct nfp_net *nn, int off, u64 val)
754 {
755 	writeq(val, nn->dp.ctrl_bar + off);
756 }
757 
758 /* Flush posted PCI writes by reading something without side effects */
759 static inline void nn_pci_flush(struct nfp_net *nn)
760 {
761 	nn_readl(nn, NFP_NET_CFG_VERSION);
762 }
763 
764 /* Queue Controller Peripheral access functions and definitions.
765  *
766  * Some of the BARs of the NFP are mapped to portions of the Queue
767  * Controller Peripheral (QCP) address space on the NFP.  A QCP queue
768  * has a read and a write pointer (as well as a size and flags,
769  * indicating overflow etc).  The QCP offers a number of different
770  * operation on queue pointers, but here we only offer function to
771  * either add to a pointer or to read the pointer value.
772  */
773 #define NFP_QCP_QUEUE_ADDR_SZ			0x800
774 #define NFP_QCP_QUEUE_AREA_SZ			0x80000
775 #define NFP_QCP_QUEUE_OFF(_x)			((_x) * NFP_QCP_QUEUE_ADDR_SZ)
776 #define NFP_QCP_QUEUE_ADD_RPTR			0x0000
777 #define NFP_QCP_QUEUE_ADD_WPTR			0x0004
778 #define NFP_QCP_QUEUE_STS_LO			0x0008
779 #define NFP_QCP_QUEUE_STS_LO_READPTR_mask	0x3ffff
780 #define NFP_QCP_QUEUE_STS_HI			0x000c
781 #define NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask	0x3ffff
782 
783 /* The offset of a QCP queues in the PCIe Target */
784 #define NFP_PCIE_QUEUE(_q) (0x80000 + (NFP_QCP_QUEUE_ADDR_SZ * ((_q) & 0xff)))
785 
786 /* nfp_qcp_ptr - Read or Write Pointer of a queue */
787 enum nfp_qcp_ptr {
788 	NFP_QCP_READ_PTR = 0,
789 	NFP_QCP_WRITE_PTR
790 };
791 
792 /* There appear to be an *undocumented* upper limit on the value which
793  * one can add to a queue and that value is either 0x3f or 0x7f.  We
794  * go with 0x3f as a conservative measure.
795  */
796 #define NFP_QCP_MAX_ADD				0x3f
797 
798 static inline void _nfp_qcp_ptr_add(u8 __iomem *q,
799 				    enum nfp_qcp_ptr ptr, u32 val)
800 {
801 	u32 off;
802 
803 	if (ptr == NFP_QCP_READ_PTR)
804 		off = NFP_QCP_QUEUE_ADD_RPTR;
805 	else
806 		off = NFP_QCP_QUEUE_ADD_WPTR;
807 
808 	while (val > NFP_QCP_MAX_ADD) {
809 		writel(NFP_QCP_MAX_ADD, q + off);
810 		val -= NFP_QCP_MAX_ADD;
811 	}
812 
813 	writel(val, q + off);
814 }
815 
816 /**
817  * nfp_qcp_rd_ptr_add() - Add the value to the read pointer of a queue
818  *
819  * @q:   Base address for queue structure
820  * @val: Value to add to the queue pointer
821  *
822  * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed.
823  */
824 static inline void nfp_qcp_rd_ptr_add(u8 __iomem *q, u32 val)
825 {
826 	_nfp_qcp_ptr_add(q, NFP_QCP_READ_PTR, val);
827 }
828 
829 /**
830  * nfp_qcp_wr_ptr_add() - Add the value to the write pointer of a queue
831  *
832  * @q:   Base address for queue structure
833  * @val: Value to add to the queue pointer
834  *
835  * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed.
836  */
837 static inline void nfp_qcp_wr_ptr_add(u8 __iomem *q, u32 val)
838 {
839 	_nfp_qcp_ptr_add(q, NFP_QCP_WRITE_PTR, val);
840 }
841 
842 static inline u32 _nfp_qcp_read(u8 __iomem *q, enum nfp_qcp_ptr ptr)
843 {
844 	u32 off;
845 	u32 val;
846 
847 	if (ptr == NFP_QCP_READ_PTR)
848 		off = NFP_QCP_QUEUE_STS_LO;
849 	else
850 		off = NFP_QCP_QUEUE_STS_HI;
851 
852 	val = readl(q + off);
853 
854 	if (ptr == NFP_QCP_READ_PTR)
855 		return val & NFP_QCP_QUEUE_STS_LO_READPTR_mask;
856 	else
857 		return val & NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask;
858 }
859 
860 /**
861  * nfp_qcp_rd_ptr_read() - Read the current read pointer value for a queue
862  * @q:  Base address for queue structure
863  *
864  * Return: Value read.
865  */
866 static inline u32 nfp_qcp_rd_ptr_read(u8 __iomem *q)
867 {
868 	return _nfp_qcp_read(q, NFP_QCP_READ_PTR);
869 }
870 
871 /**
872  * nfp_qcp_wr_ptr_read() - Read the current write pointer value for a queue
873  * @q:  Base address for queue structure
874  *
875  * Return: Value read.
876  */
877 static inline u32 nfp_qcp_wr_ptr_read(u8 __iomem *q)
878 {
879 	return _nfp_qcp_read(q, NFP_QCP_WRITE_PTR);
880 }
881 
882 static inline bool nfp_net_is_data_vnic(struct nfp_net *nn)
883 {
884 	WARN_ON_ONCE(!nn->dp.netdev && nn->port);
885 	return !!nn->dp.netdev;
886 }
887 
888 static inline bool nfp_net_running(struct nfp_net *nn)
889 {
890 	return nn->dp.ctrl & NFP_NET_CFG_CTRL_ENABLE;
891 }
892 
893 static inline const char *nfp_net_name(struct nfp_net *nn)
894 {
895 	return nn->dp.netdev ? nn->dp.netdev->name : "ctrl";
896 }
897 
898 static inline void nfp_ctrl_lock(struct nfp_net *nn)
899 	__acquires(&nn->r_vecs[0].lock)
900 {
901 	spin_lock_bh(&nn->r_vecs[0].lock);
902 }
903 
904 static inline void nfp_ctrl_unlock(struct nfp_net *nn)
905 	__releases(&nn->r_vecs[0].lock)
906 {
907 	spin_unlock_bh(&nn->r_vecs[0].lock);
908 }
909 
910 static inline void nn_ctrl_bar_lock(struct nfp_net *nn)
911 {
912 	down(&nn->bar_lock);
913 }
914 
915 static inline bool nn_ctrl_bar_trylock(struct nfp_net *nn)
916 {
917 	return !down_trylock(&nn->bar_lock);
918 }
919 
920 static inline void nn_ctrl_bar_unlock(struct nfp_net *nn)
921 {
922 	up(&nn->bar_lock);
923 }
924 
925 /* Globals */
926 extern const char nfp_driver_version[];
927 
928 extern const struct net_device_ops nfp_net_netdev_ops;
929 
930 static inline bool nfp_netdev_is_nfp_net(struct net_device *netdev)
931 {
932 	return netdev->netdev_ops == &nfp_net_netdev_ops;
933 }
934 
935 static inline int nfp_net_coalesce_para_check(u32 usecs, u32 pkts)
936 {
937 	if ((usecs >= ((1 << 16) - 1)) || (pkts >= ((1 << 16) - 1)))
938 		return -EINVAL;
939 
940 	return 0;
941 }
942 
943 /* Prototypes */
944 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
945 			    void __iomem *ctrl_bar);
946 
947 struct nfp_net *
948 nfp_net_alloc(struct pci_dev *pdev, void __iomem *ctrl_bar, bool needs_netdev,
949 	      unsigned int max_tx_rings, unsigned int max_rx_rings);
950 void nfp_net_free(struct nfp_net *nn);
951 
952 int nfp_net_init(struct nfp_net *nn);
953 void nfp_net_clean(struct nfp_net *nn);
954 
955 int nfp_ctrl_open(struct nfp_net *nn);
956 void nfp_ctrl_close(struct nfp_net *nn);
957 
958 void nfp_net_set_ethtool_ops(struct net_device *netdev);
959 void nfp_net_info(struct nfp_net *nn);
960 int __nfp_net_reconfig(struct nfp_net *nn, u32 update);
961 int nfp_net_reconfig(struct nfp_net *nn, u32 update);
962 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn);
963 void nfp_net_rss_write_itbl(struct nfp_net *nn);
964 void nfp_net_rss_write_key(struct nfp_net *nn);
965 void nfp_net_coalesce_write_cfg(struct nfp_net *nn);
966 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size);
967 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd);
968 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd);
969 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 update);
970 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn);
971 
972 void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr);
973 unsigned int
974 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
975 		   unsigned int min_irqs, unsigned int want_irqs);
976 void nfp_net_irqs_disable(struct pci_dev *pdev);
977 void
978 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
979 		    unsigned int n);
980 
981 void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring);
982 void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget);
983 
984 bool
985 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
986 		   void *data, void *pkt, unsigned int pkt_len, int meta_len);
987 
988 void nfp_net_rx_csum(const struct nfp_net_dp *dp,
989 		     struct nfp_net_r_vector *r_vec,
990 		     const struct nfp_net_rx_desc *rxd,
991 		     const struct nfp_meta_parsed *meta,
992 		     struct sk_buff *skb);
993 
994 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn);
995 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *new,
996 			  struct netlink_ext_ack *extack);
997 
998 #ifdef CONFIG_NFP_DEBUG
999 void nfp_net_debugfs_create(void);
1000 void nfp_net_debugfs_destroy(void);
1001 struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev);
1002 void nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir);
1003 void nfp_net_debugfs_dir_clean(struct dentry **dir);
1004 #else
1005 static inline void nfp_net_debugfs_create(void)
1006 {
1007 }
1008 
1009 static inline void nfp_net_debugfs_destroy(void)
1010 {
1011 }
1012 
1013 static inline struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev)
1014 {
1015 	return NULL;
1016 }
1017 
1018 static inline void
1019 nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir)
1020 {
1021 }
1022 
1023 static inline void nfp_net_debugfs_dir_clean(struct dentry **dir)
1024 {
1025 }
1026 #endif /* CONFIG_NFP_DEBUG */
1027 
1028 #endif /* _NFP_NET_H_ */
1029