1 /*
2  * Copyright (C) 2015 Cavium, Inc.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License
6  * as published by the Free Software Foundation.
7  */
8 
9 #ifndef NICVF_QUEUES_H
10 #define NICVF_QUEUES_H
11 
12 #include <linux/netdevice.h>
13 #include <linux/iommu.h>
14 #include <net/xdp.h>
15 #include "q_struct.h"
16 
17 #define MAX_QUEUE_SET			128
18 #define MAX_RCV_QUEUES_PER_QS		8
19 #define MAX_RCV_BUF_DESC_RINGS_PER_QS	2
20 #define MAX_SND_QUEUES_PER_QS		8
21 #define MAX_CMP_QUEUES_PER_QS		8
22 
23 /* VF's queue interrupt ranges */
24 #define	NICVF_INTR_ID_CQ		0
25 #define	NICVF_INTR_ID_SQ		8
26 #define	NICVF_INTR_ID_RBDR		16
27 #define	NICVF_INTR_ID_MISC		18
28 #define	NICVF_INTR_ID_QS_ERR		19
29 
30 #define	for_each_cq_irq(irq)	\
31 	for (irq = NICVF_INTR_ID_CQ; irq < NICVF_INTR_ID_SQ; irq++)
32 #define	for_each_sq_irq(irq)	\
33 	for (irq = NICVF_INTR_ID_SQ; irq < NICVF_INTR_ID_RBDR; irq++)
34 #define	for_each_rbdr_irq(irq)	\
35 	for (irq = NICVF_INTR_ID_RBDR; irq < NICVF_INTR_ID_MISC; irq++)
36 
37 #define RBDR_SIZE0		0ULL /* 8K entries */
38 #define RBDR_SIZE1		1ULL /* 16K entries */
39 #define RBDR_SIZE2		2ULL /* 32K entries */
40 #define RBDR_SIZE3		3ULL /* 64K entries */
41 #define RBDR_SIZE4		4ULL /* 126K entries */
42 #define RBDR_SIZE5		5ULL /* 256K entries */
43 #define RBDR_SIZE6		6ULL /* 512K entries */
44 
45 #define SND_QUEUE_SIZE0		0ULL /* 1K entries */
46 #define SND_QUEUE_SIZE1		1ULL /* 2K entries */
47 #define SND_QUEUE_SIZE2		2ULL /* 4K entries */
48 #define SND_QUEUE_SIZE3		3ULL /* 8K entries */
49 #define SND_QUEUE_SIZE4		4ULL /* 16K entries */
50 #define SND_QUEUE_SIZE5		5ULL /* 32K entries */
51 #define SND_QUEUE_SIZE6		6ULL /* 64K entries */
52 
53 #define CMP_QUEUE_SIZE0		0ULL /* 1K entries */
54 #define CMP_QUEUE_SIZE1		1ULL /* 2K entries */
55 #define CMP_QUEUE_SIZE2		2ULL /* 4K entries */
56 #define CMP_QUEUE_SIZE3		3ULL /* 8K entries */
57 #define CMP_QUEUE_SIZE4		4ULL /* 16K entries */
58 #define CMP_QUEUE_SIZE5		5ULL /* 32K entries */
59 #define CMP_QUEUE_SIZE6		6ULL /* 64K entries */
60 
61 /* Default queue count per QS, its lengths and threshold values */
62 #define DEFAULT_RBDR_CNT	1
63 
64 #define SND_QSIZE		SND_QUEUE_SIZE0
65 #define SND_QUEUE_LEN		(1ULL << (SND_QSIZE + 10))
66 #define MIN_SND_QUEUE_LEN	(1ULL << (SND_QUEUE_SIZE0 + 10))
67 #define MAX_SND_QUEUE_LEN	(1ULL << (SND_QUEUE_SIZE6 + 10))
68 #define SND_QUEUE_THRESH	2ULL
69 #define MIN_SQ_DESC_PER_PKT_XMIT	2
70 /* Since timestamp not enabled, otherwise 2 */
71 #define MAX_CQE_PER_PKT_XMIT		1
72 
73 /* Keep CQ and SQ sizes same, if timestamping
74  * is enabled this equation will change.
75  */
76 #define CMP_QSIZE		CMP_QUEUE_SIZE0
77 #define CMP_QUEUE_LEN		(1ULL << (CMP_QSIZE + 10))
78 #define MIN_CMP_QUEUE_LEN	(1ULL << (CMP_QUEUE_SIZE0 + 10))
79 #define MAX_CMP_QUEUE_LEN	(1ULL << (CMP_QUEUE_SIZE6 + 10))
80 #define CMP_QUEUE_CQE_THRESH	(NAPI_POLL_WEIGHT / 2)
81 #define CMP_QUEUE_TIMER_THRESH	80 /* ~2usec */
82 
83 /* No of CQEs that might anyway gets used by HW due to pipelining
84  * effects irrespective of PASS/DROP/LEVELS being configured
85  */
86 #define CMP_QUEUE_PIPELINE_RSVD 544
87 
88 #define RBDR_SIZE		RBDR_SIZE0
89 #define RCV_BUF_COUNT		(1ULL << (RBDR_SIZE + 13))
90 #define MAX_RCV_BUF_COUNT	(1ULL << (RBDR_SIZE6 + 13))
91 #define RBDR_THRESH		(RCV_BUF_COUNT / 2)
92 #define DMA_BUFFER_LEN		1536 /* In multiples of 128bytes */
93 #define RCV_FRAG_LEN	 (SKB_DATA_ALIGN(DMA_BUFFER_LEN + NET_SKB_PAD) + \
94 			 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
95 
96 #define MAX_CQES_FOR_TX		((SND_QUEUE_LEN / MIN_SQ_DESC_PER_PKT_XMIT) * \
97 				 MAX_CQE_PER_PKT_XMIT)
98 
99 /* RED and Backpressure levels of CQ for pkt reception
100  * For CQ, level is a measure of emptiness i.e 0x0 means full
101  * eg: For CQ of size 4K, and for pass/drop levels of 160/144
102  * HW accepts pkt if unused CQE >= 2560
103  * RED accepts pkt if unused CQE < 2304 & >= 2560
104  * DROPs pkts if unused CQE < 2304
105  */
106 #define RQ_PASS_CQ_LVL         192ULL
107 #define RQ_DROP_CQ_LVL         184ULL
108 
109 /* RED and Backpressure levels of RBDR for pkt reception
110  * For RBDR, level is a measure of fullness i.e 0x0 means empty
111  * eg: For RBDR of size 8K, and for pass/drop levels of 4/0
112  * HW accepts pkt if unused RBs >= 256
113  * RED accepts pkt if unused RBs < 256 & >= 0
114  * DROPs pkts if unused RBs < 0
115  */
116 #define RQ_PASS_RBDR_LVL	8ULL
117 #define RQ_DROP_RBDR_LVL	0ULL
118 
119 /* Descriptor size in bytes */
120 #define SND_QUEUE_DESC_SIZE	16
121 #define CMP_QUEUE_DESC_SIZE	512
122 
123 /* Buffer / descriptor alignments */
124 #define NICVF_RCV_BUF_ALIGN		7
125 #define NICVF_RCV_BUF_ALIGN_BYTES	(1ULL << NICVF_RCV_BUF_ALIGN)
126 #define NICVF_CQ_BASE_ALIGN_BYTES	512  /* 9 bits */
127 #define NICVF_SQ_BASE_ALIGN_BYTES	128  /* 7 bits */
128 
129 #define NICVF_ALIGNED_ADDR(ADDR, ALIGN_BYTES)	ALIGN(ADDR, ALIGN_BYTES)
130 
131 /* Queue enable/disable */
132 #define NICVF_SQ_EN		BIT_ULL(19)
133 
134 /* Queue reset */
135 #define NICVF_CQ_RESET		BIT_ULL(41)
136 #define NICVF_SQ_RESET		BIT_ULL(17)
137 #define NICVF_RBDR_RESET	BIT_ULL(43)
138 
139 enum CQ_RX_ERRLVL_E {
140 	CQ_ERRLVL_MAC,
141 	CQ_ERRLVL_L2,
142 	CQ_ERRLVL_L3,
143 	CQ_ERRLVL_L4,
144 };
145 
146 enum CQ_RX_ERROP_E {
147 	CQ_RX_ERROP_RE_NONE = 0x0,
148 	CQ_RX_ERROP_RE_PARTIAL = 0x1,
149 	CQ_RX_ERROP_RE_JABBER = 0x2,
150 	CQ_RX_ERROP_RE_FCS = 0x7,
151 	CQ_RX_ERROP_RE_TERMINATE = 0x9,
152 	CQ_RX_ERROP_RE_RX_CTL = 0xb,
153 	CQ_RX_ERROP_PREL2_ERR = 0x1f,
154 	CQ_RX_ERROP_L2_FRAGMENT = 0x20,
155 	CQ_RX_ERROP_L2_OVERRUN = 0x21,
156 	CQ_RX_ERROP_L2_PFCS = 0x22,
157 	CQ_RX_ERROP_L2_PUNY = 0x23,
158 	CQ_RX_ERROP_L2_MAL = 0x24,
159 	CQ_RX_ERROP_L2_OVERSIZE = 0x25,
160 	CQ_RX_ERROP_L2_UNDERSIZE = 0x26,
161 	CQ_RX_ERROP_L2_LENMISM = 0x27,
162 	CQ_RX_ERROP_L2_PCLP = 0x28,
163 	CQ_RX_ERROP_IP_NOT = 0x41,
164 	CQ_RX_ERROP_IP_CSUM_ERR = 0x42,
165 	CQ_RX_ERROP_IP_MAL = 0x43,
166 	CQ_RX_ERROP_IP_MALD = 0x44,
167 	CQ_RX_ERROP_IP_HOP = 0x45,
168 	CQ_RX_ERROP_L3_ICRC = 0x46,
169 	CQ_RX_ERROP_L3_PCLP = 0x47,
170 	CQ_RX_ERROP_L4_MAL = 0x61,
171 	CQ_RX_ERROP_L4_CHK = 0x62,
172 	CQ_RX_ERROP_UDP_LEN = 0x63,
173 	CQ_RX_ERROP_L4_PORT = 0x64,
174 	CQ_RX_ERROP_TCP_FLAG = 0x65,
175 	CQ_RX_ERROP_TCP_OFFSET = 0x66,
176 	CQ_RX_ERROP_L4_PCLP = 0x67,
177 	CQ_RX_ERROP_RBDR_TRUNC = 0x70,
178 };
179 
180 enum CQ_TX_ERROP_E {
181 	CQ_TX_ERROP_GOOD = 0x0,
182 	CQ_TX_ERROP_DESC_FAULT = 0x10,
183 	CQ_TX_ERROP_HDR_CONS_ERR = 0x11,
184 	CQ_TX_ERROP_SUBDC_ERR = 0x12,
185 	CQ_TX_ERROP_MAX_SIZE_VIOL = 0x13,
186 	CQ_TX_ERROP_IMM_SIZE_OFLOW = 0x80,
187 	CQ_TX_ERROP_DATA_SEQUENCE_ERR = 0x81,
188 	CQ_TX_ERROP_MEM_SEQUENCE_ERR = 0x82,
189 	CQ_TX_ERROP_LOCK_VIOL = 0x83,
190 	CQ_TX_ERROP_DATA_FAULT = 0x84,
191 	CQ_TX_ERROP_TSTMP_CONFLICT = 0x85,
192 	CQ_TX_ERROP_TSTMP_TIMEOUT = 0x86,
193 	CQ_TX_ERROP_MEM_FAULT = 0x87,
194 	CQ_TX_ERROP_CK_OVERLAP = 0x88,
195 	CQ_TX_ERROP_CK_OFLOW = 0x89,
196 	CQ_TX_ERROP_ENUM_LAST = 0x8a,
197 };
198 
199 enum RQ_SQ_STATS {
200 	RQ_SQ_STATS_OCTS,
201 	RQ_SQ_STATS_PKTS,
202 };
203 
204 struct rx_tx_queue_stats {
205 	u64	bytes;
206 	u64	pkts;
207 } ____cacheline_aligned_in_smp;
208 
209 struct q_desc_mem {
210 	dma_addr_t	dma;
211 	u64		size;
212 	u32		q_len;
213 	dma_addr_t	phys_base;
214 	void		*base;
215 	void		*unalign_base;
216 };
217 
218 struct pgcache {
219 	struct page	*page;
220 	int		ref_count;
221 	u64		dma_addr;
222 };
223 
224 struct rbdr {
225 	bool		enable;
226 	u32		dma_size;
227 	u32		frag_len;
228 	u32		thresh;		/* Threshold level for interrupt */
229 	void		*desc;
230 	u32		head;
231 	u32		tail;
232 	struct q_desc_mem   dmem;
233 	bool		is_xdp;
234 
235 	/* For page recycling */
236 	int		pgidx;
237 	int		pgcnt;
238 	int		pgalloc;
239 	struct pgcache	*pgcache;
240 } ____cacheline_aligned_in_smp;
241 
242 struct rcv_queue {
243 	bool		enable;
244 	struct	rbdr	*rbdr_start;
245 	struct	rbdr	*rbdr_cont;
246 	bool		en_tcp_reassembly;
247 	u8		cq_qs;  /* CQ's QS to which this RQ is assigned */
248 	u8		cq_idx; /* CQ index (0 to 7) in the QS */
249 	u8		cont_rbdr_qs;      /* Continue buffer ptrs - QS num */
250 	u8		cont_qs_rbdr_idx;  /* RBDR idx in the cont QS */
251 	u8		start_rbdr_qs;     /* First buffer ptrs - QS num */
252 	u8		start_qs_rbdr_idx; /* RBDR idx in the above QS */
253 	u8		caching;
254 	struct		rx_tx_queue_stats stats;
255 	struct xdp_rxq_info xdp_rxq;
256 } ____cacheline_aligned_in_smp;
257 
258 struct cmp_queue {
259 	bool		enable;
260 	u16		thresh;
261 	spinlock_t	lock;  /* lock to serialize processing CQEs */
262 	void		*desc;
263 	struct q_desc_mem   dmem;
264 	int		irq;
265 } ____cacheline_aligned_in_smp;
266 
267 struct snd_queue {
268 	bool		enable;
269 	u8		cq_qs;  /* CQ's QS to which this SQ is pointing */
270 	u8		cq_idx; /* CQ index (0 to 7) in the above QS */
271 	u16		thresh;
272 	atomic_t	free_cnt;
273 	u32		head;
274 	u32		tail;
275 	u64		*skbuff;
276 	void		*desc;
277 	u64		*xdp_page;
278 	u16		xdp_desc_cnt;
279 	u16		xdp_free_cnt;
280 	bool		is_xdp;
281 
282 	/* For TSO segment's header */
283 	char		*tso_hdrs;
284 	dma_addr_t	tso_hdrs_phys;
285 
286 	cpumask_t	affinity_mask;
287 	struct q_desc_mem   dmem;
288 	struct rx_tx_queue_stats stats;
289 } ____cacheline_aligned_in_smp;
290 
291 struct queue_set {
292 	bool		enable;
293 	bool		be_en;
294 	u8		vnic_id;
295 	u8		rq_cnt;
296 	u8		cq_cnt;
297 	u64		cq_len;
298 	u8		sq_cnt;
299 	u64		sq_len;
300 	u8		rbdr_cnt;
301 	u64		rbdr_len;
302 	struct	rcv_queue	rq[MAX_RCV_QUEUES_PER_QS];
303 	struct	cmp_queue	cq[MAX_CMP_QUEUES_PER_QS];
304 	struct	snd_queue	sq[MAX_SND_QUEUES_PER_QS];
305 	struct	rbdr		rbdr[MAX_RCV_BUF_DESC_RINGS_PER_QS];
306 } ____cacheline_aligned_in_smp;
307 
308 #define GET_RBDR_DESC(RING, idx)\
309 		(&(((struct rbdr_entry_t *)((RING)->desc))[idx]))
310 #define GET_SQ_DESC(RING, idx)\
311 		(&(((struct sq_hdr_subdesc *)((RING)->desc))[idx]))
312 #define GET_CQ_DESC(RING, idx)\
313 		(&(((union cq_desc_t *)((RING)->desc))[idx]))
314 
315 /* CQ status bits */
316 #define	CQ_WR_FULL	BIT(26)
317 #define	CQ_WR_DISABLE	BIT(25)
318 #define	CQ_WR_FAULT	BIT(24)
319 #define	CQ_CQE_COUNT	(0xFFFF << 0)
320 
321 #define	CQ_ERR_MASK	(CQ_WR_FULL | CQ_WR_DISABLE | CQ_WR_FAULT)
322 
323 static inline u64 nicvf_iova_to_phys(struct nicvf *nic, dma_addr_t dma_addr)
324 {
325 	/* Translation is installed only when IOMMU is present */
326 	if (nic->iommu_domain)
327 		return iommu_iova_to_phys(nic->iommu_domain, dma_addr);
328 	return dma_addr;
329 }
330 
331 void nicvf_unmap_sndq_buffers(struct nicvf *nic, struct snd_queue *sq,
332 			      int hdr_sqe, u8 subdesc_cnt);
333 void nicvf_config_vlan_stripping(struct nicvf *nic,
334 				 netdev_features_t features);
335 int nicvf_set_qset_resources(struct nicvf *nic);
336 int nicvf_config_data_transfer(struct nicvf *nic, bool enable);
337 void nicvf_qset_config(struct nicvf *nic, bool enable);
338 void nicvf_cmp_queue_config(struct nicvf *nic, struct queue_set *qs,
339 			    int qidx, bool enable);
340 
341 void nicvf_sq_enable(struct nicvf *nic, struct snd_queue *sq, int qidx);
342 void nicvf_sq_disable(struct nicvf *nic, int qidx);
343 void nicvf_put_sq_desc(struct snd_queue *sq, int desc_cnt);
344 void nicvf_sq_free_used_descs(struct net_device *netdev,
345 			      struct snd_queue *sq, int qidx);
346 int nicvf_sq_append_skb(struct nicvf *nic, struct snd_queue *sq,
347 			struct sk_buff *skb, u8 sq_num);
348 int nicvf_xdp_sq_append_pkt(struct nicvf *nic, struct snd_queue *sq,
349 			    u64 bufaddr, u64 dma_addr, u16 len);
350 void nicvf_xdp_sq_doorbell(struct nicvf *nic, struct snd_queue *sq, int sq_num);
351 
352 struct sk_buff *nicvf_get_rcv_skb(struct nicvf *nic,
353 				  struct cqe_rx_t *cqe_rx, bool xdp);
354 void nicvf_rbdr_task(unsigned long data);
355 void nicvf_rbdr_work(struct work_struct *work);
356 
357 void nicvf_enable_intr(struct nicvf *nic, int int_type, int q_idx);
358 void nicvf_disable_intr(struct nicvf *nic, int int_type, int q_idx);
359 void nicvf_clear_intr(struct nicvf *nic, int int_type, int q_idx);
360 int nicvf_is_intr_enabled(struct nicvf *nic, int int_type, int q_idx);
361 
362 /* Register access APIs */
363 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val);
364 u64  nicvf_reg_read(struct nicvf *nic, u64 offset);
365 void nicvf_qset_reg_write(struct nicvf *nic, u64 offset, u64 val);
366 u64 nicvf_qset_reg_read(struct nicvf *nic, u64 offset);
367 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset,
368 			   u64 qidx, u64 val);
369 u64  nicvf_queue_reg_read(struct nicvf *nic,
370 			  u64 offset, u64 qidx);
371 
372 /* Stats */
373 void nicvf_update_rq_stats(struct nicvf *nic, int rq_idx);
374 void nicvf_update_sq_stats(struct nicvf *nic, int sq_idx);
375 int nicvf_check_cqe_rx_errs(struct nicvf *nic, struct cqe_rx_t *cqe_rx);
376 int nicvf_check_cqe_tx_errs(struct nicvf *nic, struct cqe_send_t *cqe_tx);
377 #endif /* NICVF_QUEUES_H */
378