Loading...
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#define ICE_DFLT_IRQ_WORK 256
8#define ICE_RXBUF_2048 2048
9#define ICE_MAX_CHAINED_RX_BUFS 5
10#define ICE_MAX_BUF_TXD 8
11#define ICE_MIN_TX_LEN 17
12
13/* The size limit for a transmit buffer in a descriptor is (16K - 1).
14 * In order to align with the read requests we will align the value to
15 * the nearest 4K which represents our maximum read request size.
16 */
17#define ICE_MAX_READ_REQ_SIZE 4096
18#define ICE_MAX_DATA_PER_TXD (16 * 1024 - 1)
19#define ICE_MAX_DATA_PER_TXD_ALIGNED \
20 (~(ICE_MAX_READ_REQ_SIZE - 1) & ICE_MAX_DATA_PER_TXD)
21
22#define ICE_RX_BUF_WRITE 16 /* Must be power of 2 */
23#define ICE_MAX_TXQ_PER_TXQG 128
24
25/* Tx Descriptors needed, worst case */
26#define DESC_NEEDED (MAX_SKB_FRAGS + 4)
27#define ICE_DESC_UNUSED(R) \
28 ((((R)->next_to_clean > (R)->next_to_use) ? 0 : (R)->count) + \
29 (R)->next_to_clean - (R)->next_to_use - 1)
30
31#define ICE_TX_FLAGS_TSO BIT(0)
32#define ICE_TX_FLAGS_HW_VLAN BIT(1)
33#define ICE_TX_FLAGS_SW_VLAN BIT(2)
34#define ICE_TX_FLAGS_VLAN_M 0xffff0000
35#define ICE_TX_FLAGS_VLAN_S 16
36
37struct ice_tx_buf {
38 struct ice_tx_desc *next_to_watch;
39 struct sk_buff *skb;
40 unsigned int bytecount;
41 unsigned short gso_segs;
42 u32 tx_flags;
43 DEFINE_DMA_UNMAP_ADDR(dma);
44 DEFINE_DMA_UNMAP_LEN(len);
45};
46
47struct ice_tx_offload_params {
48 u8 header_len;
49 u32 td_cmd;
50 u32 td_offset;
51 u32 td_l2tag1;
52 u16 cd_l2tag2;
53 u32 cd_tunnel_params;
54 u64 cd_qw1;
55 struct ice_ring *tx_ring;
56};
57
58struct ice_rx_buf {
59 struct sk_buff *skb;
60 dma_addr_t dma;
61 struct page *page;
62 unsigned int page_offset;
63};
64
65struct ice_q_stats {
66 u64 pkts;
67 u64 bytes;
68};
69
70struct ice_txq_stats {
71 u64 restart_q;
72 u64 tx_busy;
73 u64 tx_linearize;
74};
75
76struct ice_rxq_stats {
77 u64 non_eop_descs;
78 u64 alloc_page_failed;
79 u64 alloc_buf_failed;
80 u64 page_reuse_count;
81};
82
83/* this enum matches hardware bits and is meant to be used by DYN_CTLN
84 * registers and QINT registers or more generally anywhere in the manual
85 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
86 * register but instead is a special value meaning "don't update" ITR0/1/2.
87 */
88enum ice_dyn_idx_t {
89 ICE_IDX_ITR0 = 0,
90 ICE_IDX_ITR1 = 1,
91 ICE_IDX_ITR2 = 2,
92 ICE_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
93};
94
95/* Header split modes defined by DTYPE field of Rx RLAN context */
96enum ice_rx_dtype {
97 ICE_RX_DTYPE_NO_SPLIT = 0,
98 ICE_RX_DTYPE_HEADER_SPLIT = 1,
99 ICE_RX_DTYPE_SPLIT_ALWAYS = 2,
100};
101
102/* indices into GLINT_ITR registers */
103#define ICE_RX_ITR ICE_IDX_ITR0
104#define ICE_TX_ITR ICE_IDX_ITR1
105#define ICE_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
106#define ICE_ITR_8K 0x003E
107
108/* apply ITR HW granularity translation to program the HW registers */
109#define ITR_TO_REG(val, itr_gran) (((val) & ~ICE_ITR_DYNAMIC) >> (itr_gran))
110
111/* Legacy or Advanced Mode Queue */
112#define ICE_TX_ADVANCED 0
113#define ICE_TX_LEGACY 1
114
115/* descriptor ring, associated with a VSI */
116struct ice_ring {
117 struct ice_ring *next; /* pointer to next ring in q_vector */
118 void *desc; /* Descriptor ring memory */
119 struct device *dev; /* Used for DMA mapping */
120 struct net_device *netdev; /* netdev ring maps to */
121 struct ice_vsi *vsi; /* Backreference to associated VSI */
122 struct ice_q_vector *q_vector; /* Backreference to associated vector */
123 u8 __iomem *tail;
124 union {
125 struct ice_tx_buf *tx_buf;
126 struct ice_rx_buf *rx_buf;
127 };
128 u16 q_index; /* Queue number of ring */
129 u32 txq_teid; /* Added Tx queue TEID */
130
131 /* high bit set means dynamic, use accessor routines to read/write.
132 * hardware supports 2us/1us resolution for the ITR registers.
133 * these values always store the USER setting, and must be converted
134 * before programming to a register.
135 */
136 u16 rx_itr_setting;
137 u16 tx_itr_setting;
138
139 u16 count; /* Number of descriptors */
140 u16 reg_idx; /* HW register index of the ring */
141
142 /* used in interrupt processing */
143 u16 next_to_use;
144 u16 next_to_clean;
145
146 bool ring_active; /* is ring online or not */
147
148 /* stats structs */
149 struct ice_q_stats stats;
150 struct u64_stats_sync syncp;
151 union {
152 struct ice_txq_stats tx_stats;
153 struct ice_rxq_stats rx_stats;
154 };
155
156 unsigned int size; /* length of descriptor ring in bytes */
157 dma_addr_t dma; /* physical address of ring */
158 struct rcu_head rcu; /* to avoid race on free */
159 u16 next_to_alloc;
160} ____cacheline_internodealigned_in_smp;
161
162enum ice_latency_range {
163 ICE_LOWEST_LATENCY = 0,
164 ICE_LOW_LATENCY = 1,
165 ICE_BULK_LATENCY = 2,
166 ICE_ULTRA_LATENCY = 3,
167};
168
169struct ice_ring_container {
170 /* array of pointers to rings */
171 struct ice_ring *ring;
172 unsigned int total_bytes; /* total bytes processed this int */
173 unsigned int total_pkts; /* total packets processed this int */
174 enum ice_latency_range latency_range;
175 u16 itr;
176};
177
178/* iterator for handling rings in ring container */
179#define ice_for_each_ring(pos, head) \
180 for (pos = (head).ring; pos; pos = pos->next)
181
182bool ice_alloc_rx_bufs(struct ice_ring *rxr, u16 cleaned_count);
183netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev);
184void ice_clean_tx_ring(struct ice_ring *tx_ring);
185void ice_clean_rx_ring(struct ice_ring *rx_ring);
186int ice_setup_tx_ring(struct ice_ring *tx_ring);
187int ice_setup_rx_ring(struct ice_ring *rx_ring);
188void ice_free_tx_ring(struct ice_ring *tx_ring);
189void ice_free_rx_ring(struct ice_ring *rx_ring);
190int ice_napi_poll(struct napi_struct *napi, int budget);
191
192#endif /* _ICE_TXRX_H_ */
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 */
53static 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 */
67static 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_TSYN BIT(4)
122#define ICE_TX_FLAGS_IPV4 BIT(5)
123#define ICE_TX_FLAGS_IPV6 BIT(6)
124#define ICE_TX_FLAGS_TUNNEL BIT(7)
125#define ICE_TX_FLAGS_VLAN_M 0xffff0000
126#define ICE_TX_FLAGS_VLAN_PR_M 0xe0000000
127#define ICE_TX_FLAGS_VLAN_PR_S 29
128#define ICE_TX_FLAGS_VLAN_S 16
129
130#define ICE_XDP_PASS 0
131#define ICE_XDP_CONSUMED BIT(0)
132#define ICE_XDP_TX BIT(1)
133#define ICE_XDP_REDIR BIT(2)
134
135#define ICE_RX_DMA_ATTR \
136 (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
137
138#define ICE_ETH_PKT_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
139
140#define ICE_TXD_LAST_DESC_CMD (ICE_TX_DESC_CMD_EOP | ICE_TX_DESC_CMD_RS)
141
142struct ice_tx_buf {
143 struct ice_tx_desc *next_to_watch;
144 union {
145 struct sk_buff *skb;
146 void *raw_buf; /* used for XDP */
147 };
148 unsigned int bytecount;
149 unsigned short gso_segs;
150 u32 tx_flags;
151 DEFINE_DMA_UNMAP_LEN(len);
152 DEFINE_DMA_UNMAP_ADDR(dma);
153};
154
155struct ice_tx_offload_params {
156 u64 cd_qw1;
157 struct ice_ring *tx_ring;
158 u32 td_cmd;
159 u32 td_offset;
160 u32 td_l2tag1;
161 u32 cd_tunnel_params;
162 u16 cd_l2tag2;
163 u8 header_len;
164};
165
166struct ice_rx_buf {
167 union {
168 struct {
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
180struct ice_q_stats {
181 u64 pkts;
182 u64 bytes;
183};
184
185struct 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
192struct ice_rxq_stats {
193 u64 non_eop_descs;
194 u64 alloc_page_failed;
195 u64 alloc_buf_failed;
196};
197
198enum ice_ring_state_t {
199 ICE_TX_XPS_INIT_DONE,
200 ICE_TX_NBITS,
201};
202
203/* this enum matches hardware bits and is meant to be used by DYN_CTLN
204 * registers and QINT registers or more generally anywhere in the manual
205 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
206 * register but instead is a special value meaning "don't update" ITR0/1/2.
207 */
208enum ice_dyn_idx_t {
209 ICE_IDX_ITR0 = 0,
210 ICE_IDX_ITR1 = 1,
211 ICE_IDX_ITR2 = 2,
212 ICE_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
213};
214
215/* Header split modes defined by DTYPE field of Rx RLAN context */
216enum ice_rx_dtype {
217 ICE_RX_DTYPE_NO_SPLIT = 0,
218 ICE_RX_DTYPE_HEADER_SPLIT = 1,
219 ICE_RX_DTYPE_SPLIT_ALWAYS = 2,
220};
221
222/* indices into GLINT_ITR registers */
223#define ICE_RX_ITR ICE_IDX_ITR0
224#define ICE_TX_ITR ICE_IDX_ITR1
225#define ICE_ITR_8K 124
226#define ICE_ITR_20K 50
227#define ICE_ITR_MAX 8160 /* 0x1FE0 */
228#define ICE_DFLT_TX_ITR ICE_ITR_20K
229#define ICE_DFLT_RX_ITR ICE_ITR_20K
230enum ice_dynamic_itr {
231 ITR_STATIC = 0,
232 ITR_DYNAMIC = 1
233};
234
235#define ITR_IS_DYNAMIC(rc) ((rc)->itr_mode == ITR_DYNAMIC)
236#define ICE_ITR_GRAN_S 1 /* ITR granularity is always 2us */
237#define ICE_ITR_GRAN_US BIT(ICE_ITR_GRAN_S)
238#define ICE_ITR_MASK 0x1FFE /* ITR register value alignment mask */
239#define ITR_REG_ALIGN(setting) ((setting) & ICE_ITR_MASK)
240
241#define ICE_DFLT_INTRL 0
242#define ICE_MAX_INTRL 236
243
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 */
261struct 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 DECLARE_BITMAP(xps_state, ICE_TX_NBITS); /* XPS Config State */
298 struct bpf_prog *xdp_prog;
299 struct xsk_buff_pool *xsk_pool;
300 u16 rx_offset;
301 /* CL3 - 3rd cacheline starts here */
302 struct xdp_rxq_info xdp_rxq;
303 struct sk_buff *skb;
304 /* CLX - the below items are only accessed infrequently and should be
305 * in their own cache line if possible
306 */
307#define ICE_TX_FLAGS_RING_XDP BIT(0)
308#define ICE_RX_FLAGS_RING_BUILD_SKB BIT(1)
309 u8 flags;
310 dma_addr_t dma; /* physical address of ring */
311 unsigned int size; /* length of descriptor ring in bytes */
312 u32 txq_teid; /* Added Tx queue TEID */
313 u16 rx_buf_len;
314 u8 dcb_tc; /* Traffic class of ring */
315 struct ice_ptp_tx *tx_tstamps;
316 u64 cached_phctime;
317 u8 ptp_rx:1;
318 u8 ptp_tx:1;
319} ____cacheline_internodealigned_in_smp;
320
321static inline bool ice_ring_uses_build_skb(struct ice_ring *ring)
322{
323 return !!(ring->flags & ICE_RX_FLAGS_RING_BUILD_SKB);
324}
325
326static inline void ice_set_ring_build_skb_ena(struct ice_ring *ring)
327{
328 ring->flags |= ICE_RX_FLAGS_RING_BUILD_SKB;
329}
330
331static inline void ice_clear_ring_build_skb_ena(struct ice_ring *ring)
332{
333 ring->flags &= ~ICE_RX_FLAGS_RING_BUILD_SKB;
334}
335
336static inline bool ice_ring_is_xdp(struct ice_ring *ring)
337{
338 return !!(ring->flags & ICE_TX_FLAGS_RING_XDP);
339}
340
341struct ice_ring_container {
342 /* head of linked-list of rings */
343 struct ice_ring *ring;
344 struct dim dim; /* data for net_dim algorithm */
345 u16 itr_idx; /* index in the interrupt vector */
346 /* this matches the maximum number of ITR bits, but in usec
347 * values, so it is shifted left one bit (bit zero is ignored)
348 */
349 u16 itr_setting:13;
350 u16 itr_reserved:2;
351 u16 itr_mode:1;
352};
353
354struct ice_coalesce_stored {
355 u16 itr_tx;
356 u16 itr_rx;
357 u8 intrl;
358 u8 tx_valid;
359 u8 rx_valid;
360};
361
362/* iterator for handling rings in ring container */
363#define ice_for_each_ring(pos, head) \
364 for (pos = (head).ring; pos; pos = pos->next)
365
366static inline unsigned int ice_rx_pg_order(struct ice_ring *ring)
367{
368#if (PAGE_SIZE < 8192)
369 if (ring->rx_buf_len > (PAGE_SIZE / 2))
370 return 1;
371#endif
372 return 0;
373}
374
375#define ice_rx_pg_size(_ring) (PAGE_SIZE << ice_rx_pg_order(_ring))
376
377union ice_32b_rx_flex_desc;
378
379bool ice_alloc_rx_bufs(struct ice_ring *rxr, u16 cleaned_count);
380netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev);
381void ice_clean_tx_ring(struct ice_ring *tx_ring);
382void ice_clean_rx_ring(struct ice_ring *rx_ring);
383int ice_setup_tx_ring(struct ice_ring *tx_ring);
384int ice_setup_rx_ring(struct ice_ring *rx_ring);
385void ice_free_tx_ring(struct ice_ring *tx_ring);
386void ice_free_rx_ring(struct ice_ring *rx_ring);
387int ice_napi_poll(struct napi_struct *napi, int budget);
388int
389ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc,
390 u8 *raw_packet);
391int ice_clean_rx_irq(struct ice_ring *rx_ring, int budget);
392void ice_clean_ctrl_tx_irq(struct ice_ring *tx_ring);
393#endif /* _ICE_TXRX_H_ */