1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright (c) 2018, Intel Corporation. */
3 
4 #ifndef _ICE_TXRX_H_
5 #define _ICE_TXRX_H_
6 
7 #include "ice_type.h"
8 
9 #define ICE_DFLT_IRQ_WORK	256
10 #define ICE_RXBUF_3072		3072
11 #define ICE_RXBUF_2048		2048
12 #define ICE_RXBUF_1536		1536
13 #define ICE_MAX_CHAINED_RX_BUFS	5
14 #define ICE_MAX_BUF_TXD		8
15 #define ICE_MIN_TX_LEN		17
16 
17 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
18  * In order to align with the read requests we will align the value to
19  * the nearest 4K which represents our maximum read request size.
20  */
21 #define ICE_MAX_READ_REQ_SIZE	4096
22 #define ICE_MAX_DATA_PER_TXD	(16 * 1024 - 1)
23 #define ICE_MAX_DATA_PER_TXD_ALIGNED \
24 	(~(ICE_MAX_READ_REQ_SIZE - 1) & ICE_MAX_DATA_PER_TXD)
25 
26 #define ICE_RX_BUF_WRITE	16	/* Must be power of 2 */
27 #define ICE_MAX_TXQ_PER_TXQG	128
28 
29 /* Attempt to maximize the headroom available for incoming frames. We use a 2K
30  * buffer for MTUs <= 1500 and need 1536/1534 to store the data for the frame.
31  * This leaves us with 512 bytes of room.  From that we need to deduct the
32  * space needed for the shared info and the padding needed to IP align the
33  * frame.
34  *
35  * Note: For cache line sizes 256 or larger this value is going to end
36  *	 up negative.  In these cases we should fall back to the legacy
37  *	 receive path.
38  */
39 #if (PAGE_SIZE < 8192)
40 #define ICE_2K_TOO_SMALL_WITH_PADDING \
41 	((unsigned int)(NET_SKB_PAD + ICE_RXBUF_1536) > \
42 			SKB_WITH_OVERHEAD(ICE_RXBUF_2048))
43 
44 /**
45  * ice_compute_pad - compute the padding
46  * rx_buf_len: buffer length
47  *
48  * Figure out the size of half page based on given buffer length and
49  * then subtract the skb_shared_info followed by subtraction of the
50  * actual buffer length; this in turn results in the actual space that
51  * is left for padding usage
52  */
53 static inline int ice_compute_pad(int rx_buf_len)
54 {
55 	int half_page_size;
56 
57 	half_page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
58 	return SKB_WITH_OVERHEAD(half_page_size) - rx_buf_len;
59 }
60 
61 /**
62  * ice_skb_pad - determine the padding that we can supply
63  *
64  * Figure out the right Rx buffer size and based on that calculate the
65  * padding
66  */
67 static inline int ice_skb_pad(void)
68 {
69 	int rx_buf_len;
70 
71 	/* If a 2K buffer cannot handle a standard Ethernet frame then
72 	 * optimize padding for a 3K buffer instead of a 1.5K buffer.
73 	 *
74 	 * For a 3K buffer we need to add enough padding to allow for
75 	 * tailroom due to NET_IP_ALIGN possibly shifting us out of
76 	 * cache-line alignment.
77 	 */
78 	if (ICE_2K_TOO_SMALL_WITH_PADDING)
79 		rx_buf_len = ICE_RXBUF_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
80 	else
81 		rx_buf_len = ICE_RXBUF_1536;
82 
83 	/* if needed make room for NET_IP_ALIGN */
84 	rx_buf_len -= NET_IP_ALIGN;
85 
86 	return ice_compute_pad(rx_buf_len);
87 }
88 
89 #define ICE_SKB_PAD ice_skb_pad()
90 #else
91 #define ICE_2K_TOO_SMALL_WITH_PADDING false
92 #define ICE_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
93 #endif
94 
95 /* We are assuming that the cache line is always 64 Bytes here for ice.
96  * In order to make sure that is a correct assumption there is a check in probe
97  * to print a warning if the read from GLPCI_CNF2 tells us that the cache line
98  * size is 128 bytes. We do it this way because we do not want to read the
99  * GLPCI_CNF2 register or a variable containing the value on every pass through
100  * the Tx path.
101  */
102 #define ICE_CACHE_LINE_BYTES		64
103 #define ICE_DESCS_PER_CACHE_LINE	(ICE_CACHE_LINE_BYTES / \
104 					 sizeof(struct ice_tx_desc))
105 #define ICE_DESCS_FOR_CTX_DESC		1
106 #define ICE_DESCS_FOR_SKB_DATA_PTR	1
107 /* Tx descriptors needed, worst case */
108 #define DESC_NEEDED (MAX_SKB_FRAGS + ICE_DESCS_FOR_CTX_DESC + \
109 		     ICE_DESCS_PER_CACHE_LINE + ICE_DESCS_FOR_SKB_DATA_PTR)
110 #define ICE_DESC_UNUSED(R)	\
111 	(u16)((((R)->next_to_clean > (R)->next_to_use) ? 0 : (R)->count) + \
112 	      (R)->next_to_clean - (R)->next_to_use - 1)
113 
114 #define ICE_TX_FLAGS_TSO	BIT(0)
115 #define ICE_TX_FLAGS_HW_VLAN	BIT(1)
116 #define ICE_TX_FLAGS_SW_VLAN	BIT(2)
117 /* ICE_TX_FLAGS_DUMMY_PKT is used to mark dummy packets that should be
118  * freed instead of returned like skb packets.
119  */
120 #define ICE_TX_FLAGS_DUMMY_PKT	BIT(3)
121 #define ICE_TX_FLAGS_IPV4	BIT(5)
122 #define ICE_TX_FLAGS_IPV6	BIT(6)
123 #define ICE_TX_FLAGS_TUNNEL	BIT(7)
124 #define ICE_TX_FLAGS_VLAN_M	0xffff0000
125 #define ICE_TX_FLAGS_VLAN_PR_M	0xe0000000
126 #define ICE_TX_FLAGS_VLAN_PR_S	29
127 #define ICE_TX_FLAGS_VLAN_S	16
128 
129 #define ICE_XDP_PASS		0
130 #define ICE_XDP_CONSUMED	BIT(0)
131 #define ICE_XDP_TX		BIT(1)
132 #define ICE_XDP_REDIR		BIT(2)
133 
134 #define ICE_RX_DMA_ATTR \
135 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
136 
137 #define ICE_ETH_PKT_HDR_PAD	(ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
138 
139 #define ICE_TXD_LAST_DESC_CMD (ICE_TX_DESC_CMD_EOP | ICE_TX_DESC_CMD_RS)
140 
141 struct ice_tx_buf {
142 	struct ice_tx_desc *next_to_watch;
143 	union {
144 		struct sk_buff *skb;
145 		void *raw_buf; /* used for XDP */
146 	};
147 	unsigned int bytecount;
148 	unsigned short gso_segs;
149 	u32 tx_flags;
150 	DEFINE_DMA_UNMAP_LEN(len);
151 	DEFINE_DMA_UNMAP_ADDR(dma);
152 };
153 
154 struct ice_tx_offload_params {
155 	u64 cd_qw1;
156 	struct ice_ring *tx_ring;
157 	u32 td_cmd;
158 	u32 td_offset;
159 	u32 td_l2tag1;
160 	u32 cd_tunnel_params;
161 	u16 cd_l2tag2;
162 	u8 header_len;
163 };
164 
165 struct ice_rx_buf {
166 	union {
167 		struct {
168 			struct sk_buff *skb;
169 			dma_addr_t dma;
170 			struct page *page;
171 			unsigned int page_offset;
172 			u16 pagecnt_bias;
173 		};
174 		struct {
175 			struct xdp_buff *xdp;
176 		};
177 	};
178 };
179 
180 struct ice_q_stats {
181 	u64 pkts;
182 	u64 bytes;
183 };
184 
185 struct ice_txq_stats {
186 	u64 restart_q;
187 	u64 tx_busy;
188 	u64 tx_linearize;
189 	int prev_pkt; /* negative if no pending Tx descriptors */
190 };
191 
192 struct ice_rxq_stats {
193 	u64 non_eop_descs;
194 	u64 alloc_page_failed;
195 	u64 alloc_buf_failed;
196 	u64 gro_dropped; /* GRO returned dropped */
197 };
198 
199 /* this enum matches hardware bits and is meant to be used by DYN_CTLN
200  * registers and QINT registers or more generally anywhere in the manual
201  * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
202  * register but instead is a special value meaning "don't update" ITR0/1/2.
203  */
204 enum ice_dyn_idx_t {
205 	ICE_IDX_ITR0 = 0,
206 	ICE_IDX_ITR1 = 1,
207 	ICE_IDX_ITR2 = 2,
208 	ICE_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
209 };
210 
211 /* Header split modes defined by DTYPE field of Rx RLAN context */
212 enum ice_rx_dtype {
213 	ICE_RX_DTYPE_NO_SPLIT		= 0,
214 	ICE_RX_DTYPE_HEADER_SPLIT	= 1,
215 	ICE_RX_DTYPE_SPLIT_ALWAYS	= 2,
216 };
217 
218 /* indices into GLINT_ITR registers */
219 #define ICE_RX_ITR	ICE_IDX_ITR0
220 #define ICE_TX_ITR	ICE_IDX_ITR1
221 #define ICE_ITR_8K	124
222 #define ICE_ITR_20K	50
223 #define ICE_ITR_MAX	8160
224 #define ICE_DFLT_TX_ITR	(ICE_ITR_20K | ICE_ITR_DYNAMIC)
225 #define ICE_DFLT_RX_ITR	(ICE_ITR_20K | ICE_ITR_DYNAMIC)
226 #define ICE_ITR_DYNAMIC	0x8000  /* used as flag for itr_setting */
227 #define ITR_IS_DYNAMIC(setting) (!!((setting) & ICE_ITR_DYNAMIC))
228 #define ITR_TO_REG(setting)	((setting) & ~ICE_ITR_DYNAMIC)
229 #define ICE_ITR_GRAN_S		1	/* ITR granularity is always 2us */
230 #define ICE_ITR_GRAN_US		BIT(ICE_ITR_GRAN_S)
231 #define ICE_ITR_MASK		0x1FFE	/* ITR register value alignment mask */
232 #define ITR_REG_ALIGN(setting)	((setting) & ICE_ITR_MASK)
233 
234 #define ICE_ITR_ADAPTIVE_MIN_INC	0x0002
235 #define ICE_ITR_ADAPTIVE_MIN_USECS	0x0002
236 #define ICE_ITR_ADAPTIVE_MAX_USECS	0x00FA
237 #define ICE_ITR_ADAPTIVE_LATENCY	0x8000
238 #define ICE_ITR_ADAPTIVE_BULK		0x0000
239 
240 #define ICE_DFLT_INTRL	0
241 #define ICE_MAX_INTRL	236
242 
243 #define ICE_WB_ON_ITR_USECS	2
244 #define ICE_IN_WB_ON_ITR_MODE	255
245 /* Sets WB_ON_ITR and assumes INTENA bit is already cleared, which allows
246  * setting the MSK_M bit to tell hardware to ignore the INTENA_M bit. Also,
247  * set the write-back latency to the usecs passed in.
248  */
249 #define ICE_GLINT_DYN_CTL_WB_ON_ITR(usecs, itr_idx)	\
250 	((((usecs) << (GLINT_DYN_CTL_INTERVAL_S - ICE_ITR_GRAN_S)) & \
251 	  GLINT_DYN_CTL_INTERVAL_M) | \
252 	 (((itr_idx) << GLINT_DYN_CTL_ITR_INDX_S) & \
253 	  GLINT_DYN_CTL_ITR_INDX_M) | GLINT_DYN_CTL_INTENA_MSK_M | \
254 	 GLINT_DYN_CTL_WB_ON_ITR_M)
255 
256 /* Legacy or Advanced Mode Queue */
257 #define ICE_TX_ADVANCED	0
258 #define ICE_TX_LEGACY	1
259 
260 /* descriptor ring, associated with a VSI */
261 struct ice_ring {
262 	/* CL1 - 1st cacheline starts here */
263 	struct ice_ring *next;		/* pointer to next ring in q_vector */
264 	void *desc;			/* Descriptor ring memory */
265 	struct device *dev;		/* Used for DMA mapping */
266 	struct net_device *netdev;	/* netdev ring maps to */
267 	struct ice_vsi *vsi;		/* Backreference to associated VSI */
268 	struct ice_q_vector *q_vector;	/* Backreference to associated vector */
269 	u8 __iomem *tail;
270 	union {
271 		struct ice_tx_buf *tx_buf;
272 		struct ice_rx_buf *rx_buf;
273 	};
274 	/* CL2 - 2nd cacheline starts here */
275 	u16 q_index;			/* Queue number of ring */
276 	u16 q_handle;			/* Queue handle per TC */
277 
278 	u8 ring_active:1;		/* is ring online or not */
279 
280 	u16 count;			/* Number of descriptors */
281 	u16 reg_idx;			/* HW register index of the ring */
282 
283 	/* used in interrupt processing */
284 	u16 next_to_use;
285 	u16 next_to_clean;
286 	u16 next_to_alloc;
287 
288 	/* stats structs */
289 	struct ice_q_stats	stats;
290 	struct u64_stats_sync syncp;
291 	union {
292 		struct ice_txq_stats tx_stats;
293 		struct ice_rxq_stats rx_stats;
294 	};
295 
296 	struct rcu_head rcu;		/* to avoid race on free */
297 	struct bpf_prog *xdp_prog;
298 	struct xdp_umem *xsk_umem;
299 	/* CL3 - 3rd cacheline starts here */
300 	struct xdp_rxq_info xdp_rxq;
301 	/* CLX - the below items are only accessed infrequently and should be
302 	 * in their own cache line if possible
303 	 */
304 #define ICE_TX_FLAGS_RING_XDP		BIT(0)
305 #define ICE_RX_FLAGS_RING_BUILD_SKB	BIT(1)
306 	u8 flags;
307 	dma_addr_t dma;			/* physical address of ring */
308 	unsigned int size;		/* length of descriptor ring in bytes */
309 	u32 txq_teid;			/* Added Tx queue TEID */
310 	u16 rx_buf_len;
311 	u8 dcb_tc;			/* Traffic class of ring */
312 } ____cacheline_internodealigned_in_smp;
313 
314 static inline bool ice_ring_uses_build_skb(struct ice_ring *ring)
315 {
316 	return !!(ring->flags & ICE_RX_FLAGS_RING_BUILD_SKB);
317 }
318 
319 static inline void ice_set_ring_build_skb_ena(struct ice_ring *ring)
320 {
321 	ring->flags |= ICE_RX_FLAGS_RING_BUILD_SKB;
322 }
323 
324 static inline void ice_clear_ring_build_skb_ena(struct ice_ring *ring)
325 {
326 	ring->flags &= ~ICE_RX_FLAGS_RING_BUILD_SKB;
327 }
328 
329 static inline bool ice_ring_is_xdp(struct ice_ring *ring)
330 {
331 	return !!(ring->flags & ICE_TX_FLAGS_RING_XDP);
332 }
333 
334 struct ice_ring_container {
335 	/* head of linked-list of rings */
336 	struct ice_ring *ring;
337 	unsigned long next_update;	/* jiffies value of next queue update */
338 	unsigned int total_bytes;	/* total bytes processed this int */
339 	unsigned int total_pkts;	/* total packets processed this int */
340 	u16 itr_idx;		/* index in the interrupt vector */
341 	u16 target_itr;		/* value in usecs divided by the hw->itr_gran */
342 	u16 current_itr;	/* value in usecs divided by the hw->itr_gran */
343 	/* high bit set means dynamic ITR, rest is used to store user
344 	 * readable ITR value in usecs and must be converted before programming
345 	 * to a register.
346 	 */
347 	u16 itr_setting;
348 };
349 
350 struct ice_coalesce_stored {
351 	u16 itr_tx;
352 	u16 itr_rx;
353 	u8 intrl;
354 };
355 
356 /* iterator for handling rings in ring container */
357 #define ice_for_each_ring(pos, head) \
358 	for (pos = (head).ring; pos; pos = pos->next)
359 
360 static inline unsigned int ice_rx_pg_order(struct ice_ring *ring)
361 {
362 #if (PAGE_SIZE < 8192)
363 	if (ring->rx_buf_len > (PAGE_SIZE / 2))
364 		return 1;
365 #endif
366 	return 0;
367 }
368 
369 #define ice_rx_pg_size(_ring) (PAGE_SIZE << ice_rx_pg_order(_ring))
370 
371 union ice_32b_rx_flex_desc;
372 
373 bool ice_alloc_rx_bufs(struct ice_ring *rxr, u16 cleaned_count);
374 netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev);
375 void ice_clean_tx_ring(struct ice_ring *tx_ring);
376 void ice_clean_rx_ring(struct ice_ring *rx_ring);
377 int ice_setup_tx_ring(struct ice_ring *tx_ring);
378 int ice_setup_rx_ring(struct ice_ring *rx_ring);
379 void ice_free_tx_ring(struct ice_ring *tx_ring);
380 void ice_free_rx_ring(struct ice_ring *rx_ring);
381 int ice_napi_poll(struct napi_struct *napi, int budget);
382 int
383 ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc,
384 		   u8 *raw_packet);
385 int ice_clean_rx_irq(struct ice_ring *rx_ring, int budget);
386 void ice_clean_ctrl_tx_irq(struct ice_ring *tx_ring);
387 #endif /* _ICE_TXRX_H_ */
388