Loading...
1/*******************************************************************************
2 *
3 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
4 * Copyright(c) 2013 - 2016 Intel Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program. If not, see <http://www.gnu.org/licenses/>.
17 *
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
20 *
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 *
25 ******************************************************************************/
26
27#ifndef _I40E_TXRX_H_
28#define _I40E_TXRX_H_
29
30/* Interrupt Throttling and Rate Limiting Goodies */
31
32#define I40E_MAX_ITR 0x0FF0 /* reg uses 2 usec resolution */
33#define I40E_MIN_ITR 0x0001 /* reg uses 2 usec resolution */
34#define I40E_ITR_100K 0x0005
35#define I40E_ITR_50K 0x000A
36#define I40E_ITR_20K 0x0019
37#define I40E_ITR_18K 0x001B
38#define I40E_ITR_8K 0x003E
39#define I40E_ITR_4K 0x007A
40#define I40E_MAX_INTRL 0x3B /* reg uses 4 usec resolution */
41#define I40E_ITR_RX_DEF I40E_ITR_20K
42#define I40E_ITR_TX_DEF I40E_ITR_20K
43#define I40E_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
44#define I40E_MIN_INT_RATE 250 /* ~= 1000000 / (I40E_MAX_ITR * 2) */
45#define I40E_MAX_INT_RATE 500000 /* == 1000000 / (I40E_MIN_ITR * 2) */
46#define I40E_DEFAULT_IRQ_WORK 256
47#define ITR_TO_REG(setting) ((setting & ~I40E_ITR_DYNAMIC) >> 1)
48#define ITR_IS_DYNAMIC(setting) (!!(setting & I40E_ITR_DYNAMIC))
49#define ITR_REG_TO_USEC(itr_reg) (itr_reg << 1)
50/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
51 * the value of the rate limit is non-zero
52 */
53#define INTRL_ENA BIT(6)
54#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
55#define INTRL_USEC_TO_REG(set) ((set) ? ((set) >> 2) | INTRL_ENA : 0)
56#define I40E_INTRL_8K 125 /* 8000 ints/sec */
57#define I40E_INTRL_62K 16 /* 62500 ints/sec */
58#define I40E_INTRL_83K 12 /* 83333 ints/sec */
59
60#define I40E_QUEUE_END_OF_LIST 0x7FF
61
62/* this enum matches hardware bits and is meant to be used by DYN_CTLN
63 * registers and QINT registers or more generally anywhere in the manual
64 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
65 * register but instead is a special value meaning "don't update" ITR0/1/2.
66 */
67enum i40e_dyn_idx_t {
68 I40E_IDX_ITR0 = 0,
69 I40E_IDX_ITR1 = 1,
70 I40E_IDX_ITR2 = 2,
71 I40E_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
72};
73
74/* these are indexes into ITRN registers */
75#define I40E_RX_ITR I40E_IDX_ITR0
76#define I40E_TX_ITR I40E_IDX_ITR1
77#define I40E_PE_ITR I40E_IDX_ITR2
78
79/* Supported RSS offloads */
80#define I40E_DEFAULT_RSS_HENA ( \
81 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
82 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
83 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
84 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
85 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
86 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
87 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
88 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
89 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
90 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
91 BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
92
93#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
94 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
95 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
96 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
97 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
98 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
99 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
100
101#define i40e_pf_get_default_rss_hena(pf) \
102 (((pf)->flags & I40E_FLAG_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
103 I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
104
105/* Supported Rx Buffer Sizes */
106#define I40E_RXBUFFER_512 512 /* Used for packet split */
107#define I40E_RXBUFFER_2048 2048
108#define I40E_RXBUFFER_3072 3072 /* For FCoE MTU of 2158 */
109#define I40E_RXBUFFER_4096 4096
110#define I40E_RXBUFFER_8192 8192
111#define I40E_MAX_RXBUFFER 9728 /* largest size for single descriptor */
112
113/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
114 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
115 * this adds up to 512 bytes of extra data meaning the smallest allocation
116 * we could have is 1K.
117 * i.e. RXBUFFER_512 --> size-1024 slab
118 */
119#define I40E_RX_HDR_SIZE I40E_RXBUFFER_512
120
121/* How many Rx Buffers do we bundle into one write to the hardware ? */
122#define I40E_RX_BUFFER_WRITE 16 /* Must be power of 2 */
123#define I40E_RX_INCREMENT(r, i) \
124 do { \
125 (i)++; \
126 if ((i) == (r)->count) \
127 i = 0; \
128 r->next_to_clean = i; \
129 } while (0)
130
131#define I40E_RX_NEXT_DESC(r, i, n) \
132 do { \
133 (i)++; \
134 if ((i) == (r)->count) \
135 i = 0; \
136 (n) = I40E_RX_DESC((r), (i)); \
137 } while (0)
138
139#define I40E_RX_NEXT_DESC_PREFETCH(r, i, n) \
140 do { \
141 I40E_RX_NEXT_DESC((r), (i), (n)); \
142 prefetch((n)); \
143 } while (0)
144
145#define i40e_rx_desc i40e_32byte_rx_desc
146
147#define I40E_MAX_BUFFER_TXD 8
148#define I40E_MIN_TX_LEN 17
149#define I40E_MAX_DATA_PER_TXD 8192
150
151/* Tx Descriptors needed, worst case */
152#define TXD_USE_COUNT(S) DIV_ROUND_UP((S), I40E_MAX_DATA_PER_TXD)
153#define DESC_NEEDED (MAX_SKB_FRAGS + 4)
154#define I40E_MIN_DESC_PENDING 4
155
156#define I40E_TX_FLAGS_HW_VLAN BIT(1)
157#define I40E_TX_FLAGS_SW_VLAN BIT(2)
158#define I40E_TX_FLAGS_TSO BIT(3)
159#define I40E_TX_FLAGS_IPV4 BIT(4)
160#define I40E_TX_FLAGS_IPV6 BIT(5)
161#define I40E_TX_FLAGS_FCCRC BIT(6)
162#define I40E_TX_FLAGS_FSO BIT(7)
163#define I40E_TX_FLAGS_FD_SB BIT(9)
164#define I40E_TX_FLAGS_VXLAN_TUNNEL BIT(10)
165#define I40E_TX_FLAGS_VLAN_MASK 0xffff0000
166#define I40E_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
167#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT 29
168#define I40E_TX_FLAGS_VLAN_SHIFT 16
169
170struct i40e_tx_buffer {
171 struct i40e_tx_desc *next_to_watch;
172 union {
173 struct sk_buff *skb;
174 void *raw_buf;
175 };
176 unsigned int bytecount;
177 unsigned short gso_segs;
178
179 DEFINE_DMA_UNMAP_ADDR(dma);
180 DEFINE_DMA_UNMAP_LEN(len);
181 u32 tx_flags;
182};
183
184struct i40e_rx_buffer {
185 struct sk_buff *skb;
186 void *hdr_buf;
187 dma_addr_t dma;
188 struct page *page;
189 dma_addr_t page_dma;
190 unsigned int page_offset;
191};
192
193struct i40e_queue_stats {
194 u64 packets;
195 u64 bytes;
196};
197
198struct i40e_tx_queue_stats {
199 u64 restart_queue;
200 u64 tx_busy;
201 u64 tx_done_old;
202 u64 tx_linearize;
203 u64 tx_force_wb;
204 u64 tx_lost_interrupt;
205};
206
207struct i40e_rx_queue_stats {
208 u64 non_eop_descs;
209 u64 alloc_page_failed;
210 u64 alloc_buff_failed;
211 u64 page_reuse_count;
212 u64 realloc_count;
213};
214
215enum i40e_ring_state_t {
216 __I40E_TX_FDIR_INIT_DONE,
217 __I40E_TX_XPS_INIT_DONE,
218 __I40E_RX_PS_ENABLED,
219 __I40E_RX_16BYTE_DESC_ENABLED,
220};
221
222#define ring_is_ps_enabled(ring) \
223 test_bit(__I40E_RX_PS_ENABLED, &(ring)->state)
224#define set_ring_ps_enabled(ring) \
225 set_bit(__I40E_RX_PS_ENABLED, &(ring)->state)
226#define clear_ring_ps_enabled(ring) \
227 clear_bit(__I40E_RX_PS_ENABLED, &(ring)->state)
228#define ring_is_16byte_desc_enabled(ring) \
229 test_bit(__I40E_RX_16BYTE_DESC_ENABLED, &(ring)->state)
230#define set_ring_16byte_desc_enabled(ring) \
231 set_bit(__I40E_RX_16BYTE_DESC_ENABLED, &(ring)->state)
232#define clear_ring_16byte_desc_enabled(ring) \
233 clear_bit(__I40E_RX_16BYTE_DESC_ENABLED, &(ring)->state)
234
235/* struct that defines a descriptor ring, associated with a VSI */
236struct i40e_ring {
237 struct i40e_ring *next; /* pointer to next ring in q_vector */
238 void *desc; /* Descriptor ring memory */
239 struct device *dev; /* Used for DMA mapping */
240 struct net_device *netdev; /* netdev ring maps to */
241 union {
242 struct i40e_tx_buffer *tx_bi;
243 struct i40e_rx_buffer *rx_bi;
244 };
245 unsigned long state;
246 u16 queue_index; /* Queue number of ring */
247 u8 dcb_tc; /* Traffic class of ring */
248 u8 __iomem *tail;
249
250 u16 count; /* Number of descriptors */
251 u16 reg_idx; /* HW register index of the ring */
252 u16 rx_hdr_len;
253 u16 rx_buf_len;
254 u8 dtype;
255#define I40E_RX_DTYPE_NO_SPLIT 0
256#define I40E_RX_DTYPE_HEADER_SPLIT 1
257#define I40E_RX_DTYPE_SPLIT_ALWAYS 2
258#define I40E_RX_SPLIT_L2 0x1
259#define I40E_RX_SPLIT_IP 0x2
260#define I40E_RX_SPLIT_TCP_UDP 0x4
261#define I40E_RX_SPLIT_SCTP 0x8
262
263 /* used in interrupt processing */
264 u16 next_to_use;
265 u16 next_to_clean;
266
267 u8 atr_sample_rate;
268 u8 atr_count;
269
270 bool ring_active; /* is ring online or not */
271 bool arm_wb; /* do something to arm write back */
272 u8 packet_stride;
273#define I40E_TXR_FLAGS_LAST_XMIT_MORE_SET BIT(2)
274
275 u16 flags;
276#define I40E_TXR_FLAGS_WB_ON_ITR BIT(0)
277
278 /* stats structs */
279 struct i40e_queue_stats stats;
280 struct u64_stats_sync syncp;
281 union {
282 struct i40e_tx_queue_stats tx_stats;
283 struct i40e_rx_queue_stats rx_stats;
284 };
285
286 unsigned int size; /* length of descriptor ring in bytes */
287 dma_addr_t dma; /* physical address of ring */
288
289 struct i40e_vsi *vsi; /* Backreference to associated VSI */
290 struct i40e_q_vector *q_vector; /* Backreference to associated vector */
291
292 struct rcu_head rcu; /* to avoid race on free */
293} ____cacheline_internodealigned_in_smp;
294
295enum i40e_latency_range {
296 I40E_LOWEST_LATENCY = 0,
297 I40E_LOW_LATENCY = 1,
298 I40E_BULK_LATENCY = 2,
299 I40E_ULTRA_LATENCY = 3,
300};
301
302struct i40e_ring_container {
303 /* array of pointers to rings */
304 struct i40e_ring *ring;
305 unsigned int total_bytes; /* total bytes processed this int */
306 unsigned int total_packets; /* total packets processed this int */
307 u16 count;
308 enum i40e_latency_range latency_range;
309 u16 itr;
310};
311
312/* iterator for handling rings in ring container */
313#define i40e_for_each_ring(pos, head) \
314 for (pos = (head).ring; pos != NULL; pos = pos->next)
315
316bool i40evf_alloc_rx_buffers_ps(struct i40e_ring *rxr, u16 cleaned_count);
317bool i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rxr, u16 cleaned_count);
318void i40evf_alloc_rx_headers(struct i40e_ring *rxr);
319netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
320void i40evf_clean_tx_ring(struct i40e_ring *tx_ring);
321void i40evf_clean_rx_ring(struct i40e_ring *rx_ring);
322int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring);
323int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring);
324void i40evf_free_tx_resources(struct i40e_ring *tx_ring);
325void i40evf_free_rx_resources(struct i40e_ring *rx_ring);
326int i40evf_napi_poll(struct napi_struct *napi, int budget);
327void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
328u32 i40evf_get_tx_pending(struct i40e_ring *ring, bool in_sw);
329int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
330bool __i40evf_chk_linearize(struct sk_buff *skb);
331
332/**
333 * i40e_get_head - Retrieve head from head writeback
334 * @tx_ring: Tx ring to fetch head of
335 *
336 * Returns value of Tx ring head based on value stored
337 * in head write-back location
338 **/
339static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
340{
341 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
342
343 return le32_to_cpu(*(volatile __le32 *)head);
344}
345
346/**
347 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
348 * @skb: send buffer
349 * @tx_ring: ring to send buffer on
350 *
351 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
352 * there is not enough descriptors available in this ring since we need at least
353 * one descriptor.
354 **/
355static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
356{
357 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
358 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
359 int count = 0, size = skb_headlen(skb);
360
361 for (;;) {
362 count += TXD_USE_COUNT(size);
363
364 if (!nr_frags--)
365 break;
366
367 size = skb_frag_size(frag++);
368 }
369
370 return count;
371}
372
373/**
374 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
375 * @tx_ring: the ring to be checked
376 * @size: the size buffer we want to assure is available
377 *
378 * Returns 0 if stop is not needed
379 **/
380static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
381{
382 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
383 return 0;
384 return __i40evf_maybe_stop_tx(tx_ring, size);
385}
386
387/**
388 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
389 * @skb: send buffer
390 * @count: number of buffers used
391 *
392 * Note: Our HW can't scatter-gather more than 8 fragments to build
393 * a packet on the wire and so we need to figure out the cases where we
394 * need to linearize the skb.
395 **/
396static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
397{
398 /* Both TSO and single send will work if count is less than 8 */
399 if (likely(count < I40E_MAX_BUFFER_TXD))
400 return false;
401
402 if (skb_is_gso(skb))
403 return __i40evf_chk_linearize(skb);
404
405 /* we can support up to 8 data buffers for a single send */
406 return count != I40E_MAX_BUFFER_TXD;
407}
408#endif /* _I40E_TXRX_H_ */
1/* SPDX-License-Identifier: GPL-2.0 */
2/*******************************************************************************
3 *
4 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
5 * Copyright(c) 2013 - 2016 Intel Corporation.
6 *
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms and conditions of the GNU General Public License,
9 * version 2, as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with this program. If not, see <http://www.gnu.org/licenses/>.
18 *
19 * The full GNU General Public License is included in this distribution in
20 * the file called "COPYING".
21 *
22 * Contact Information:
23 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 *
26 ******************************************************************************/
27
28#ifndef _I40E_TXRX_H_
29#define _I40E_TXRX_H_
30
31/* Interrupt Throttling and Rate Limiting Goodies */
32#define I40E_DEFAULT_IRQ_WORK 256
33
34/* The datasheet for the X710 and XL710 indicate that the maximum value for
35 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
36 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
37 * the register value which is divided by 2 lets use the actual values and
38 * avoid an excessive amount of translation.
39 */
40#define I40E_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
41#define I40E_ITR_MASK 0x1FFE /* mask for ITR register value */
42#define I40E_MIN_ITR 2 /* reg uses 2 usec resolution */
43#define I40E_ITR_100K 10 /* all values below must be even */
44#define I40E_ITR_50K 20
45#define I40E_ITR_20K 50
46#define I40E_ITR_18K 60
47#define I40E_ITR_8K 122
48#define I40E_MAX_ITR 8160 /* maximum value as per datasheet */
49#define ITR_TO_REG(setting) ((setting) & ~I40E_ITR_DYNAMIC)
50#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~I40E_ITR_MASK)
51#define ITR_IS_DYNAMIC(setting) (!!((setting) & I40E_ITR_DYNAMIC))
52
53#define I40E_ITR_RX_DEF (I40E_ITR_20K | I40E_ITR_DYNAMIC)
54#define I40E_ITR_TX_DEF (I40E_ITR_20K | I40E_ITR_DYNAMIC)
55
56/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
57 * the value of the rate limit is non-zero
58 */
59#define INTRL_ENA BIT(6)
60#define I40E_MAX_INTRL 0x3B /* reg uses 4 usec resolution */
61#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
62#define INTRL_USEC_TO_REG(set) ((set) ? ((set) >> 2) | INTRL_ENA : 0)
63#define I40E_INTRL_8K 125 /* 8000 ints/sec */
64#define I40E_INTRL_62K 16 /* 62500 ints/sec */
65#define I40E_INTRL_83K 12 /* 83333 ints/sec */
66
67#define I40E_QUEUE_END_OF_LIST 0x7FF
68
69/* this enum matches hardware bits and is meant to be used by DYN_CTLN
70 * registers and QINT registers or more generally anywhere in the manual
71 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
72 * register but instead is a special value meaning "don't update" ITR0/1/2.
73 */
74enum i40e_dyn_idx_t {
75 I40E_IDX_ITR0 = 0,
76 I40E_IDX_ITR1 = 1,
77 I40E_IDX_ITR2 = 2,
78 I40E_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
79};
80
81/* these are indexes into ITRN registers */
82#define I40E_RX_ITR I40E_IDX_ITR0
83#define I40E_TX_ITR I40E_IDX_ITR1
84#define I40E_PE_ITR I40E_IDX_ITR2
85
86/* Supported RSS offloads */
87#define I40E_DEFAULT_RSS_HENA ( \
88 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
89 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
90 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
91 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
92 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
93 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
94 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
95 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
96 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
97 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
98 BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
99
100#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
101 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
102 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
103 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
104 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
105 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
106 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
107
108/* Supported Rx Buffer Sizes (a multiple of 128) */
109#define I40E_RXBUFFER_256 256
110#define I40E_RXBUFFER_1536 1536 /* 128B aligned standard Ethernet frame */
111#define I40E_RXBUFFER_2048 2048
112#define I40E_RXBUFFER_3072 3072 /* Used for large frames w/ padding */
113#define I40E_MAX_RXBUFFER 9728 /* largest size for single descriptor */
114
115/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
116 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
117 * this adds up to 512 bytes of extra data meaning the smallest allocation
118 * we could have is 1K.
119 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
120 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
121 */
122#define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
123#define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
124#define i40e_rx_desc i40e_32byte_rx_desc
125
126#define I40E_RX_DMA_ATTR \
127 (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
128
129/* Attempt to maximize the headroom available for incoming frames. We
130 * use a 2K buffer for receives and need 1536/1534 to store the data for
131 * the frame. This leaves us with 512 bytes of room. From that we need
132 * to deduct the space needed for the shared info and the padding needed
133 * to IP align the frame.
134 *
135 * Note: For cache line sizes 256 or larger this value is going to end
136 * up negative. In these cases we should fall back to the legacy
137 * receive path.
138 */
139#if (PAGE_SIZE < 8192)
140#define I40E_2K_TOO_SMALL_WITH_PADDING \
141((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
142
143static inline int i40e_compute_pad(int rx_buf_len)
144{
145 int page_size, pad_size;
146
147 page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
148 pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
149
150 return pad_size;
151}
152
153static inline int i40e_skb_pad(void)
154{
155 int rx_buf_len;
156
157 /* If a 2K buffer cannot handle a standard Ethernet frame then
158 * optimize padding for a 3K buffer instead of a 1.5K buffer.
159 *
160 * For a 3K buffer we need to add enough padding to allow for
161 * tailroom due to NET_IP_ALIGN possibly shifting us out of
162 * cache-line alignment.
163 */
164 if (I40E_2K_TOO_SMALL_WITH_PADDING)
165 rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
166 else
167 rx_buf_len = I40E_RXBUFFER_1536;
168
169 /* if needed make room for NET_IP_ALIGN */
170 rx_buf_len -= NET_IP_ALIGN;
171
172 return i40e_compute_pad(rx_buf_len);
173}
174
175#define I40E_SKB_PAD i40e_skb_pad()
176#else
177#define I40E_2K_TOO_SMALL_WITH_PADDING false
178#define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
179#endif
180
181/**
182 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
183 * @rx_desc: pointer to receive descriptor (in le64 format)
184 * @stat_err_bits: value to mask
185 *
186 * This function does some fast chicanery in order to return the
187 * value of the mask which is really only used for boolean tests.
188 * The status_error_len doesn't need to be shifted because it begins
189 * at offset zero.
190 */
191static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
192 const u64 stat_err_bits)
193{
194 return !!(rx_desc->wb.qword1.status_error_len &
195 cpu_to_le64(stat_err_bits));
196}
197
198/* How many Rx Buffers do we bundle into one write to the hardware ? */
199#define I40E_RX_BUFFER_WRITE 32 /* Must be power of 2 */
200#define I40E_RX_INCREMENT(r, i) \
201 do { \
202 (i)++; \
203 if ((i) == (r)->count) \
204 i = 0; \
205 r->next_to_clean = i; \
206 } while (0)
207
208#define I40E_RX_NEXT_DESC(r, i, n) \
209 do { \
210 (i)++; \
211 if ((i) == (r)->count) \
212 i = 0; \
213 (n) = I40E_RX_DESC((r), (i)); \
214 } while (0)
215
216#define I40E_RX_NEXT_DESC_PREFETCH(r, i, n) \
217 do { \
218 I40E_RX_NEXT_DESC((r), (i), (n)); \
219 prefetch((n)); \
220 } while (0)
221
222#define I40E_MAX_BUFFER_TXD 8
223#define I40E_MIN_TX_LEN 17
224
225/* The size limit for a transmit buffer in a descriptor is (16K - 1).
226 * In order to align with the read requests we will align the value to
227 * the nearest 4K which represents our maximum read request size.
228 */
229#define I40E_MAX_READ_REQ_SIZE 4096
230#define I40E_MAX_DATA_PER_TXD (16 * 1024 - 1)
231#define I40E_MAX_DATA_PER_TXD_ALIGNED \
232 (I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
233
234/**
235 * i40e_txd_use_count - estimate the number of descriptors needed for Tx
236 * @size: transmit request size in bytes
237 *
238 * Due to hardware alignment restrictions (4K alignment), we need to
239 * assume that we can have no more than 12K of data per descriptor, even
240 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
241 * Thus, we need to divide by 12K. But division is slow! Instead,
242 * we decompose the operation into shifts and one relatively cheap
243 * multiply operation.
244 *
245 * To divide by 12K, we first divide by 4K, then divide by 3:
246 * To divide by 4K, shift right by 12 bits
247 * To divide by 3, multiply by 85, then divide by 256
248 * (Divide by 256 is done by shifting right by 8 bits)
249 * Finally, we add one to round up. Because 256 isn't an exact multiple of
250 * 3, we'll underestimate near each multiple of 12K. This is actually more
251 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
252 * segment. For our purposes this is accurate out to 1M which is orders of
253 * magnitude greater than our largest possible GSO size.
254 *
255 * This would then be implemented as:
256 * return (((size >> 12) * 85) >> 8) + 1;
257 *
258 * Since multiplication and division are commutative, we can reorder
259 * operations into:
260 * return ((size * 85) >> 20) + 1;
261 */
262static inline unsigned int i40e_txd_use_count(unsigned int size)
263{
264 return ((size * 85) >> 20) + 1;
265}
266
267/* Tx Descriptors needed, worst case */
268#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
269#define I40E_MIN_DESC_PENDING 4
270
271#define I40E_TX_FLAGS_HW_VLAN BIT(1)
272#define I40E_TX_FLAGS_SW_VLAN BIT(2)
273#define I40E_TX_FLAGS_TSO BIT(3)
274#define I40E_TX_FLAGS_IPV4 BIT(4)
275#define I40E_TX_FLAGS_IPV6 BIT(5)
276#define I40E_TX_FLAGS_FCCRC BIT(6)
277#define I40E_TX_FLAGS_FSO BIT(7)
278#define I40E_TX_FLAGS_FD_SB BIT(9)
279#define I40E_TX_FLAGS_VXLAN_TUNNEL BIT(10)
280#define I40E_TX_FLAGS_VLAN_MASK 0xffff0000
281#define I40E_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
282#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT 29
283#define I40E_TX_FLAGS_VLAN_SHIFT 16
284
285struct i40e_tx_buffer {
286 struct i40e_tx_desc *next_to_watch;
287 union {
288 struct sk_buff *skb;
289 void *raw_buf;
290 };
291 unsigned int bytecount;
292 unsigned short gso_segs;
293
294 DEFINE_DMA_UNMAP_ADDR(dma);
295 DEFINE_DMA_UNMAP_LEN(len);
296 u32 tx_flags;
297};
298
299struct i40e_rx_buffer {
300 dma_addr_t dma;
301 struct page *page;
302#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
303 __u32 page_offset;
304#else
305 __u16 page_offset;
306#endif
307 __u16 pagecnt_bias;
308};
309
310struct i40e_queue_stats {
311 u64 packets;
312 u64 bytes;
313};
314
315struct i40e_tx_queue_stats {
316 u64 restart_queue;
317 u64 tx_busy;
318 u64 tx_done_old;
319 u64 tx_linearize;
320 u64 tx_force_wb;
321 int prev_pkt_ctr;
322 u64 tx_lost_interrupt;
323};
324
325struct i40e_rx_queue_stats {
326 u64 non_eop_descs;
327 u64 alloc_page_failed;
328 u64 alloc_buff_failed;
329 u64 page_reuse_count;
330 u64 realloc_count;
331};
332
333enum i40e_ring_state_t {
334 __I40E_TX_FDIR_INIT_DONE,
335 __I40E_TX_XPS_INIT_DONE,
336 __I40E_RING_STATE_NBITS /* must be last */
337};
338
339/* some useful defines for virtchannel interface, which
340 * is the only remaining user of header split
341 */
342#define I40E_RX_DTYPE_NO_SPLIT 0
343#define I40E_RX_DTYPE_HEADER_SPLIT 1
344#define I40E_RX_DTYPE_SPLIT_ALWAYS 2
345#define I40E_RX_SPLIT_L2 0x1
346#define I40E_RX_SPLIT_IP 0x2
347#define I40E_RX_SPLIT_TCP_UDP 0x4
348#define I40E_RX_SPLIT_SCTP 0x8
349
350/* struct that defines a descriptor ring, associated with a VSI */
351struct i40e_ring {
352 struct i40e_ring *next; /* pointer to next ring in q_vector */
353 void *desc; /* Descriptor ring memory */
354 struct device *dev; /* Used for DMA mapping */
355 struct net_device *netdev; /* netdev ring maps to */
356 union {
357 struct i40e_tx_buffer *tx_bi;
358 struct i40e_rx_buffer *rx_bi;
359 };
360 DECLARE_BITMAP(state, __I40E_RING_STATE_NBITS);
361 u16 queue_index; /* Queue number of ring */
362 u8 dcb_tc; /* Traffic class of ring */
363 u8 __iomem *tail;
364
365 /* high bit set means dynamic, use accessors routines to read/write.
366 * hardware only supports 2us resolution for the ITR registers.
367 * these values always store the USER setting, and must be converted
368 * before programming to a register.
369 */
370 u16 itr_setting;
371
372 u16 count; /* Number of descriptors */
373 u16 reg_idx; /* HW register index of the ring */
374 u16 rx_buf_len;
375
376 /* used in interrupt processing */
377 u16 next_to_use;
378 u16 next_to_clean;
379
380 u8 atr_sample_rate;
381 u8 atr_count;
382
383 bool ring_active; /* is ring online or not */
384 bool arm_wb; /* do something to arm write back */
385 u8 packet_stride;
386
387 u16 flags;
388#define I40E_TXR_FLAGS_WB_ON_ITR BIT(0)
389#define I40E_RXR_FLAGS_BUILD_SKB_ENABLED BIT(1)
390
391 /* stats structs */
392 struct i40e_queue_stats stats;
393 struct u64_stats_sync syncp;
394 union {
395 struct i40e_tx_queue_stats tx_stats;
396 struct i40e_rx_queue_stats rx_stats;
397 };
398
399 unsigned int size; /* length of descriptor ring in bytes */
400 dma_addr_t dma; /* physical address of ring */
401
402 struct i40e_vsi *vsi; /* Backreference to associated VSI */
403 struct i40e_q_vector *q_vector; /* Backreference to associated vector */
404
405 struct rcu_head rcu; /* to avoid race on free */
406 u16 next_to_alloc;
407 struct sk_buff *skb; /* When i40evf_clean_rx_ring_irq() must
408 * return before it sees the EOP for
409 * the current packet, we save that skb
410 * here and resume receiving this
411 * packet the next time
412 * i40evf_clean_rx_ring_irq() is called
413 * for this ring.
414 */
415} ____cacheline_internodealigned_in_smp;
416
417static inline bool ring_uses_build_skb(struct i40e_ring *ring)
418{
419 return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
420}
421
422static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
423{
424 ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
425}
426
427static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
428{
429 ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
430}
431
432#define I40E_ITR_ADAPTIVE_MIN_INC 0x0002
433#define I40E_ITR_ADAPTIVE_MIN_USECS 0x0002
434#define I40E_ITR_ADAPTIVE_MAX_USECS 0x007e
435#define I40E_ITR_ADAPTIVE_LATENCY 0x8000
436#define I40E_ITR_ADAPTIVE_BULK 0x0000
437#define ITR_IS_BULK(x) (!((x) & I40E_ITR_ADAPTIVE_LATENCY))
438
439struct i40e_ring_container {
440 struct i40e_ring *ring; /* pointer to linked list of ring(s) */
441 unsigned long next_update; /* jiffies value of next update */
442 unsigned int total_bytes; /* total bytes processed this int */
443 unsigned int total_packets; /* total packets processed this int */
444 u16 count;
445 u16 target_itr; /* target ITR setting for ring(s) */
446 u16 current_itr; /* current ITR setting for ring(s) */
447};
448
449/* iterator for handling rings in ring container */
450#define i40e_for_each_ring(pos, head) \
451 for (pos = (head).ring; pos != NULL; pos = pos->next)
452
453static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
454{
455#if (PAGE_SIZE < 8192)
456 if (ring->rx_buf_len > (PAGE_SIZE / 2))
457 return 1;
458#endif
459 return 0;
460}
461
462#define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
463
464bool i40evf_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
465netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
466void i40evf_clean_tx_ring(struct i40e_ring *tx_ring);
467void i40evf_clean_rx_ring(struct i40e_ring *rx_ring);
468int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring);
469int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring);
470void i40evf_free_tx_resources(struct i40e_ring *tx_ring);
471void i40evf_free_rx_resources(struct i40e_ring *rx_ring);
472int i40evf_napi_poll(struct napi_struct *napi, int budget);
473void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
474u32 i40evf_get_tx_pending(struct i40e_ring *ring, bool in_sw);
475void i40evf_detect_recover_hung(struct i40e_vsi *vsi);
476int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
477bool __i40evf_chk_linearize(struct sk_buff *skb);
478
479/**
480 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
481 * @skb: send buffer
482 * @tx_ring: ring to send buffer on
483 *
484 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
485 * there is not enough descriptors available in this ring since we need at least
486 * one descriptor.
487 **/
488static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
489{
490 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
491 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
492 int count = 0, size = skb_headlen(skb);
493
494 for (;;) {
495 count += i40e_txd_use_count(size);
496
497 if (!nr_frags--)
498 break;
499
500 size = skb_frag_size(frag++);
501 }
502
503 return count;
504}
505
506/**
507 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
508 * @tx_ring: the ring to be checked
509 * @size: the size buffer we want to assure is available
510 *
511 * Returns 0 if stop is not needed
512 **/
513static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
514{
515 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
516 return 0;
517 return __i40evf_maybe_stop_tx(tx_ring, size);
518}
519
520/**
521 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
522 * @skb: send buffer
523 * @count: number of buffers used
524 *
525 * Note: Our HW can't scatter-gather more than 8 fragments to build
526 * a packet on the wire and so we need to figure out the cases where we
527 * need to linearize the skb.
528 **/
529static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
530{
531 /* Both TSO and single send will work if count is less than 8 */
532 if (likely(count < I40E_MAX_BUFFER_TXD))
533 return false;
534
535 if (skb_is_gso(skb))
536 return __i40evf_chk_linearize(skb);
537
538 /* we can support up to 8 data buffers for a single send */
539 return count != I40E_MAX_BUFFER_TXD;
540}
541/**
542 * @ring: Tx ring to find the netdev equivalent of
543 **/
544static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
545{
546 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
547}
548#endif /* _I40E_TXRX_H_ */