Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/* Copyright(c) 2013 - 2018 Intel Corporation. */
3
4#ifndef _IAVF_TXRX_H_
5#define _IAVF_TXRX_H_
6
7/* Interrupt Throttling and Rate Limiting Goodies */
8#define IAVF_DEFAULT_IRQ_WORK 256
9
10/* The datasheet for the X710 and XL710 indicate that the maximum value for
11 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
12 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
13 * the register value which is divided by 2 lets use the actual values and
14 * avoid an excessive amount of translation.
15 */
16#define IAVF_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
17#define IAVF_ITR_MASK 0x1FFE /* mask for ITR register value */
18#define IAVF_ITR_100K 10 /* all values below must be even */
19#define IAVF_ITR_50K 20
20#define IAVF_ITR_20K 50
21#define IAVF_ITR_18K 60
22#define IAVF_ITR_8K 122
23#define IAVF_MAX_ITR 8160 /* maximum value as per datasheet */
24#define ITR_TO_REG(setting) ((setting) & ~IAVF_ITR_DYNAMIC)
25#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~IAVF_ITR_MASK)
26#define ITR_IS_DYNAMIC(setting) (!!((setting) & IAVF_ITR_DYNAMIC))
27
28#define IAVF_ITR_RX_DEF (IAVF_ITR_20K | IAVF_ITR_DYNAMIC)
29#define IAVF_ITR_TX_DEF (IAVF_ITR_20K | IAVF_ITR_DYNAMIC)
30
31/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
32 * the value of the rate limit is non-zero
33 */
34#define INTRL_ENA BIT(6)
35#define IAVF_MAX_INTRL 0x3B /* reg uses 4 usec resolution */
36#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
37#define INTRL_USEC_TO_REG(set) ((set) ? ((set) >> 2) | INTRL_ENA : 0)
38#define IAVF_INTRL_8K 125 /* 8000 ints/sec */
39#define IAVF_INTRL_62K 16 /* 62500 ints/sec */
40#define IAVF_INTRL_83K 12 /* 83333 ints/sec */
41
42#define IAVF_QUEUE_END_OF_LIST 0x7FF
43
44/* this enum matches hardware bits and is meant to be used by DYN_CTLN
45 * registers and QINT registers or more generally anywhere in the manual
46 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
47 * register but instead is a special value meaning "don't update" ITR0/1/2.
48 */
49enum iavf_dyn_idx_t {
50 IAVF_IDX_ITR0 = 0,
51 IAVF_IDX_ITR1 = 1,
52 IAVF_IDX_ITR2 = 2,
53 IAVF_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
54};
55
56/* these are indexes into ITRN registers */
57#define IAVF_RX_ITR IAVF_IDX_ITR0
58#define IAVF_TX_ITR IAVF_IDX_ITR1
59#define IAVF_PE_ITR IAVF_IDX_ITR2
60
61/* Supported RSS offloads */
62#define IAVF_DEFAULT_RSS_HENA ( \
63 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_UDP) | \
64 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
65 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_TCP) | \
66 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
67 BIT_ULL(IAVF_FILTER_PCTYPE_FRAG_IPV4) | \
68 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_UDP) | \
69 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_TCP) | \
70 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
71 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
72 BIT_ULL(IAVF_FILTER_PCTYPE_FRAG_IPV6) | \
73 BIT_ULL(IAVF_FILTER_PCTYPE_L2_PAYLOAD))
74
75#define IAVF_DEFAULT_RSS_HENA_EXPANDED (IAVF_DEFAULT_RSS_HENA | \
76 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
77 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
78 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
79 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
80 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
81 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
82
83#define iavf_rx_desc iavf_32byte_rx_desc
84
85/**
86 * iavf_test_staterr - tests bits in Rx descriptor status and error fields
87 * @rx_desc: pointer to receive descriptor (in le64 format)
88 * @stat_err_bits: value to mask
89 *
90 * This function does some fast chicanery in order to return the
91 * value of the mask which is really only used for boolean tests.
92 * The status_error_len doesn't need to be shifted because it begins
93 * at offset zero.
94 */
95static inline bool iavf_test_staterr(union iavf_rx_desc *rx_desc,
96 const u64 stat_err_bits)
97{
98 return !!(rx_desc->wb.qword1.status_error_len &
99 cpu_to_le64(stat_err_bits));
100}
101
102/* How many Rx Buffers do we bundle into one write to the hardware ? */
103#define IAVF_RX_INCREMENT(r, i) \
104 do { \
105 (i)++; \
106 if ((i) == (r)->count) \
107 i = 0; \
108 r->next_to_clean = i; \
109 } while (0)
110
111#define IAVF_RX_NEXT_DESC(r, i, n) \
112 do { \
113 (i)++; \
114 if ((i) == (r)->count) \
115 i = 0; \
116 (n) = IAVF_RX_DESC((r), (i)); \
117 } while (0)
118
119#define IAVF_RX_NEXT_DESC_PREFETCH(r, i, n) \
120 do { \
121 IAVF_RX_NEXT_DESC((r), (i), (n)); \
122 prefetch((n)); \
123 } while (0)
124
125#define IAVF_MAX_BUFFER_TXD 8
126#define IAVF_MIN_TX_LEN 17
127
128/* The size limit for a transmit buffer in a descriptor is (16K - 1).
129 * In order to align with the read requests we will align the value to
130 * the nearest 4K which represents our maximum read request size.
131 */
132#define IAVF_MAX_READ_REQ_SIZE 4096
133#define IAVF_MAX_DATA_PER_TXD (16 * 1024 - 1)
134#define IAVF_MAX_DATA_PER_TXD_ALIGNED \
135 (IAVF_MAX_DATA_PER_TXD & ~(IAVF_MAX_READ_REQ_SIZE - 1))
136
137/**
138 * iavf_txd_use_count - estimate the number of descriptors needed for Tx
139 * @size: transmit request size in bytes
140 *
141 * Due to hardware alignment restrictions (4K alignment), we need to
142 * assume that we can have no more than 12K of data per descriptor, even
143 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
144 * Thus, we need to divide by 12K. But division is slow! Instead,
145 * we decompose the operation into shifts and one relatively cheap
146 * multiply operation.
147 *
148 * To divide by 12K, we first divide by 4K, then divide by 3:
149 * To divide by 4K, shift right by 12 bits
150 * To divide by 3, multiply by 85, then divide by 256
151 * (Divide by 256 is done by shifting right by 8 bits)
152 * Finally, we add one to round up. Because 256 isn't an exact multiple of
153 * 3, we'll underestimate near each multiple of 12K. This is actually more
154 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
155 * segment. For our purposes this is accurate out to 1M which is orders of
156 * magnitude greater than our largest possible GSO size.
157 *
158 * This would then be implemented as:
159 * return (((size >> 12) * 85) >> 8) + 1;
160 *
161 * Since multiplication and division are commutative, we can reorder
162 * operations into:
163 * return ((size * 85) >> 20) + 1;
164 */
165static inline unsigned int iavf_txd_use_count(unsigned int size)
166{
167 return ((size * 85) >> 20) + 1;
168}
169
170/* Tx Descriptors needed, worst case */
171#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
172#define IAVF_MIN_DESC_PENDING 4
173
174#define IAVF_TX_FLAGS_HW_VLAN BIT(1)
175#define IAVF_TX_FLAGS_SW_VLAN BIT(2)
176#define IAVF_TX_FLAGS_TSO BIT(3)
177#define IAVF_TX_FLAGS_IPV4 BIT(4)
178#define IAVF_TX_FLAGS_IPV6 BIT(5)
179#define IAVF_TX_FLAGS_FCCRC BIT(6)
180#define IAVF_TX_FLAGS_FSO BIT(7)
181#define IAVF_TX_FLAGS_FD_SB BIT(9)
182#define IAVF_TX_FLAGS_VXLAN_TUNNEL BIT(10)
183#define IAVF_TX_FLAGS_HW_OUTER_SINGLE_VLAN BIT(11)
184#define IAVF_TX_FLAGS_VLAN_MASK 0xffff0000
185#define IAVF_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
186#define IAVF_TX_FLAGS_VLAN_PRIO_SHIFT 29
187#define IAVF_TX_FLAGS_VLAN_SHIFT 16
188
189struct iavf_tx_buffer {
190 struct iavf_tx_desc *next_to_watch;
191 union {
192 struct sk_buff *skb;
193 void *raw_buf;
194 };
195 unsigned int bytecount;
196 unsigned short gso_segs;
197
198 DEFINE_DMA_UNMAP_ADDR(dma);
199 DEFINE_DMA_UNMAP_LEN(len);
200 u32 tx_flags;
201};
202
203struct iavf_queue_stats {
204 u64 packets;
205 u64 bytes;
206};
207
208struct iavf_tx_queue_stats {
209 u64 restart_queue;
210 u64 tx_busy;
211 u64 tx_done_old;
212 u64 tx_linearize;
213 u64 tx_force_wb;
214 u64 tx_lost_interrupt;
215};
216
217struct iavf_rx_queue_stats {
218 u64 non_eop_descs;
219 u64 alloc_page_failed;
220 u64 alloc_buff_failed;
221};
222
223/* some useful defines for virtchannel interface, which
224 * is the only remaining user of header split
225 */
226#define IAVF_RX_DTYPE_NO_SPLIT 0
227#define IAVF_RX_DTYPE_HEADER_SPLIT 1
228#define IAVF_RX_DTYPE_SPLIT_ALWAYS 2
229#define IAVF_RX_SPLIT_L2 0x1
230#define IAVF_RX_SPLIT_IP 0x2
231#define IAVF_RX_SPLIT_TCP_UDP 0x4
232#define IAVF_RX_SPLIT_SCTP 0x8
233
234/* struct that defines a descriptor ring, associated with a VSI */
235struct iavf_ring {
236 struct iavf_ring *next; /* pointer to next ring in q_vector */
237 void *desc; /* Descriptor ring memory */
238 union {
239 struct page_pool *pp; /* Used on Rx for buffer management */
240 struct device *dev; /* Used on Tx for DMA mapping */
241 };
242 struct net_device *netdev; /* netdev ring maps to */
243 union {
244 struct libeth_fqe *rx_fqes;
245 struct iavf_tx_buffer *tx_bi;
246 };
247 u8 __iomem *tail;
248 u32 truesize;
249
250 u16 queue_index; /* Queue number of ring */
251
252 /* high bit set means dynamic, use accessors routines to read/write.
253 * hardware only supports 2us resolution for the ITR registers.
254 * these values always store the USER setting, and must be converted
255 * before programming to a register.
256 */
257 u16 itr_setting;
258
259 u16 count; /* Number of descriptors */
260
261 /* used in interrupt processing */
262 u16 next_to_use;
263 u16 next_to_clean;
264
265 u16 flags;
266#define IAVF_TXR_FLAGS_WB_ON_ITR BIT(0)
267#define IAVF_TXR_FLAGS_ARM_WB BIT(1)
268/* BIT(2) is free */
269#define IAVF_TXRX_FLAGS_VLAN_TAG_LOC_L2TAG1 BIT(3)
270#define IAVF_TXR_FLAGS_VLAN_TAG_LOC_L2TAG2 BIT(4)
271#define IAVF_RXR_FLAGS_VLAN_TAG_LOC_L2TAG2_2 BIT(5)
272
273 /* stats structs */
274 struct iavf_queue_stats stats;
275 struct u64_stats_sync syncp;
276 union {
277 struct iavf_tx_queue_stats tx_stats;
278 struct iavf_rx_queue_stats rx_stats;
279 };
280
281 int prev_pkt_ctr; /* For Tx stall detection */
282 unsigned int size; /* length of descriptor ring in bytes */
283 dma_addr_t dma; /* physical address of ring */
284
285 struct iavf_vsi *vsi; /* Backreference to associated VSI */
286 struct iavf_q_vector *q_vector; /* Backreference to associated vector */
287
288 struct rcu_head rcu; /* to avoid race on free */
289 struct sk_buff *skb; /* When iavf_clean_rx_ring_irq() must
290 * return before it sees the EOP for
291 * the current packet, we save that skb
292 * here and resume receiving this
293 * packet the next time
294 * iavf_clean_rx_ring_irq() is called
295 * for this ring.
296 */
297
298 u32 rx_buf_len;
299 struct net_shaper q_shaper;
300 bool q_shaper_update;
301} ____cacheline_internodealigned_in_smp;
302
303#define IAVF_ITR_ADAPTIVE_MIN_INC 0x0002
304#define IAVF_ITR_ADAPTIVE_MIN_USECS 0x0002
305#define IAVF_ITR_ADAPTIVE_MAX_USECS 0x007e
306#define IAVF_ITR_ADAPTIVE_LATENCY 0x8000
307#define IAVF_ITR_ADAPTIVE_BULK 0x0000
308#define ITR_IS_BULK(x) (!((x) & IAVF_ITR_ADAPTIVE_LATENCY))
309
310struct iavf_ring_container {
311 struct iavf_ring *ring; /* pointer to linked list of ring(s) */
312 unsigned long next_update; /* jiffies value of next update */
313 unsigned int total_bytes; /* total bytes processed this int */
314 unsigned int total_packets; /* total packets processed this int */
315 u16 count;
316 u16 target_itr; /* target ITR setting for ring(s) */
317 u16 current_itr; /* current ITR setting for ring(s) */
318};
319
320/* iterator for handling rings in ring container */
321#define iavf_for_each_ring(pos, head) \
322 for (pos = (head).ring; pos != NULL; pos = pos->next)
323
324bool iavf_alloc_rx_buffers(struct iavf_ring *rxr, u16 cleaned_count);
325netdev_tx_t iavf_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
326int iavf_setup_tx_descriptors(struct iavf_ring *tx_ring);
327int iavf_setup_rx_descriptors(struct iavf_ring *rx_ring);
328void iavf_free_tx_resources(struct iavf_ring *tx_ring);
329void iavf_free_rx_resources(struct iavf_ring *rx_ring);
330int iavf_napi_poll(struct napi_struct *napi, int budget);
331void iavf_detect_recover_hung(struct iavf_vsi *vsi);
332int __iavf_maybe_stop_tx(struct iavf_ring *tx_ring, int size);
333bool __iavf_chk_linearize(struct sk_buff *skb);
334
335/**
336 * iavf_xmit_descriptor_count - calculate number of Tx descriptors needed
337 * @skb: send buffer
338 *
339 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
340 * there is not enough descriptors available in this ring since we need at least
341 * one descriptor.
342 **/
343static inline int iavf_xmit_descriptor_count(struct sk_buff *skb)
344{
345 const skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
346 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
347 int count = 0, size = skb_headlen(skb);
348
349 for (;;) {
350 count += iavf_txd_use_count(size);
351
352 if (!nr_frags--)
353 break;
354
355 size = skb_frag_size(frag++);
356 }
357
358 return count;
359}
360
361/**
362 * iavf_maybe_stop_tx - 1st level check for Tx stop conditions
363 * @tx_ring: the ring to be checked
364 * @size: the size buffer we want to assure is available
365 *
366 * Returns 0 if stop is not needed
367 **/
368static inline int iavf_maybe_stop_tx(struct iavf_ring *tx_ring, int size)
369{
370 if (likely(IAVF_DESC_UNUSED(tx_ring) >= size))
371 return 0;
372 return __iavf_maybe_stop_tx(tx_ring, size);
373}
374
375/**
376 * iavf_chk_linearize - Check if there are more than 8 fragments per packet
377 * @skb: send buffer
378 * @count: number of buffers used
379 *
380 * Note: Our HW can't scatter-gather more than 8 fragments to build
381 * a packet on the wire and so we need to figure out the cases where we
382 * need to linearize the skb.
383 **/
384static inline bool iavf_chk_linearize(struct sk_buff *skb, int count)
385{
386 /* Both TSO and single send will work if count is less than 8 */
387 if (likely(count < IAVF_MAX_BUFFER_TXD))
388 return false;
389
390 if (skb_is_gso(skb))
391 return __iavf_chk_linearize(skb);
392
393 /* we can support up to 8 data buffers for a single send */
394 return count != IAVF_MAX_BUFFER_TXD;
395}
396/**
397 * txring_txq - helper to convert from a ring to a queue
398 * @ring: Tx ring to find the netdev equivalent of
399 **/
400static inline struct netdev_queue *txring_txq(const struct iavf_ring *ring)
401{
402 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
403}
404#endif /* _IAVF_TXRX_H_ */
1/* SPDX-License-Identifier: GPL-2.0 */
2/* Copyright(c) 2013 - 2018 Intel Corporation. */
3
4#ifndef _IAVF_TXRX_H_
5#define _IAVF_TXRX_H_
6
7/* Interrupt Throttling and Rate Limiting Goodies */
8#define IAVF_DEFAULT_IRQ_WORK 256
9
10/* The datasheet for the X710 and XL710 indicate that the maximum value for
11 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
12 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
13 * the register value which is divided by 2 lets use the actual values and
14 * avoid an excessive amount of translation.
15 */
16#define IAVF_ITR_DYNAMIC 0x8000 /* use top bit as a flag */
17#define IAVF_ITR_MASK 0x1FFE /* mask for ITR register value */
18#define IAVF_MIN_ITR 2 /* reg uses 2 usec resolution */
19#define IAVF_ITR_100K 10 /* all values below must be even */
20#define IAVF_ITR_50K 20
21#define IAVF_ITR_20K 50
22#define IAVF_ITR_18K 60
23#define IAVF_ITR_8K 122
24#define IAVF_MAX_ITR 8160 /* maximum value as per datasheet */
25#define ITR_TO_REG(setting) ((setting) & ~IAVF_ITR_DYNAMIC)
26#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~IAVF_ITR_MASK)
27#define ITR_IS_DYNAMIC(setting) (!!((setting) & IAVF_ITR_DYNAMIC))
28
29#define IAVF_ITR_RX_DEF (IAVF_ITR_20K | IAVF_ITR_DYNAMIC)
30#define IAVF_ITR_TX_DEF (IAVF_ITR_20K | IAVF_ITR_DYNAMIC)
31
32/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
33 * the value of the rate limit is non-zero
34 */
35#define INTRL_ENA BIT(6)
36#define IAVF_MAX_INTRL 0x3B /* reg uses 4 usec resolution */
37#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
38#define INTRL_USEC_TO_REG(set) ((set) ? ((set) >> 2) | INTRL_ENA : 0)
39#define IAVF_INTRL_8K 125 /* 8000 ints/sec */
40#define IAVF_INTRL_62K 16 /* 62500 ints/sec */
41#define IAVF_INTRL_83K 12 /* 83333 ints/sec */
42
43#define IAVF_QUEUE_END_OF_LIST 0x7FF
44
45/* this enum matches hardware bits and is meant to be used by DYN_CTLN
46 * registers and QINT registers or more generally anywhere in the manual
47 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
48 * register but instead is a special value meaning "don't update" ITR0/1/2.
49 */
50enum iavf_dyn_idx_t {
51 IAVF_IDX_ITR0 = 0,
52 IAVF_IDX_ITR1 = 1,
53 IAVF_IDX_ITR2 = 2,
54 IAVF_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
55};
56
57/* these are indexes into ITRN registers */
58#define IAVF_RX_ITR IAVF_IDX_ITR0
59#define IAVF_TX_ITR IAVF_IDX_ITR1
60#define IAVF_PE_ITR IAVF_IDX_ITR2
61
62/* Supported RSS offloads */
63#define IAVF_DEFAULT_RSS_HENA ( \
64 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_UDP) | \
65 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
66 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_TCP) | \
67 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
68 BIT_ULL(IAVF_FILTER_PCTYPE_FRAG_IPV4) | \
69 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_UDP) | \
70 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_TCP) | \
71 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
72 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
73 BIT_ULL(IAVF_FILTER_PCTYPE_FRAG_IPV6) | \
74 BIT_ULL(IAVF_FILTER_PCTYPE_L2_PAYLOAD))
75
76#define IAVF_DEFAULT_RSS_HENA_EXPANDED (IAVF_DEFAULT_RSS_HENA | \
77 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
78 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
79 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
80 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
81 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
82 BIT_ULL(IAVF_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
83
84/* Supported Rx Buffer Sizes (a multiple of 128) */
85#define IAVF_RXBUFFER_256 256
86#define IAVF_RXBUFFER_1536 1536 /* 128B aligned standard Ethernet frame */
87#define IAVF_RXBUFFER_2048 2048
88#define IAVF_RXBUFFER_3072 3072 /* Used for large frames w/ padding */
89#define IAVF_MAX_RXBUFFER 9728 /* largest size for single descriptor */
90
91/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
92 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
93 * this adds up to 512 bytes of extra data meaning the smallest allocation
94 * we could have is 1K.
95 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
96 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
97 */
98#define IAVF_RX_HDR_SIZE IAVF_RXBUFFER_256
99#define IAVF_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
100#define iavf_rx_desc iavf_32byte_rx_desc
101
102#define IAVF_RX_DMA_ATTR \
103 (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
104
105/* Attempt to maximize the headroom available for incoming frames. We
106 * use a 2K buffer for receives and need 1536/1534 to store the data for
107 * the frame. This leaves us with 512 bytes of room. From that we need
108 * to deduct the space needed for the shared info and the padding needed
109 * to IP align the frame.
110 *
111 * Note: For cache line sizes 256 or larger this value is going to end
112 * up negative. In these cases we should fall back to the legacy
113 * receive path.
114 */
115#if (PAGE_SIZE < 8192)
116#define IAVF_2K_TOO_SMALL_WITH_PADDING \
117((NET_SKB_PAD + IAVF_RXBUFFER_1536) > SKB_WITH_OVERHEAD(IAVF_RXBUFFER_2048))
118
119static inline int iavf_compute_pad(int rx_buf_len)
120{
121 int page_size, pad_size;
122
123 page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
124 pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
125
126 return pad_size;
127}
128
129static inline int iavf_skb_pad(void)
130{
131 int rx_buf_len;
132
133 /* If a 2K buffer cannot handle a standard Ethernet frame then
134 * optimize padding for a 3K buffer instead of a 1.5K buffer.
135 *
136 * For a 3K buffer we need to add enough padding to allow for
137 * tailroom due to NET_IP_ALIGN possibly shifting us out of
138 * cache-line alignment.
139 */
140 if (IAVF_2K_TOO_SMALL_WITH_PADDING)
141 rx_buf_len = IAVF_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
142 else
143 rx_buf_len = IAVF_RXBUFFER_1536;
144
145 /* if needed make room for NET_IP_ALIGN */
146 rx_buf_len -= NET_IP_ALIGN;
147
148 return iavf_compute_pad(rx_buf_len);
149}
150
151#define IAVF_SKB_PAD iavf_skb_pad()
152#else
153#define IAVF_2K_TOO_SMALL_WITH_PADDING false
154#define IAVF_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
155#endif
156
157/**
158 * iavf_test_staterr - tests bits in Rx descriptor status and error fields
159 * @rx_desc: pointer to receive descriptor (in le64 format)
160 * @stat_err_bits: value to mask
161 *
162 * This function does some fast chicanery in order to return the
163 * value of the mask which is really only used for boolean tests.
164 * The status_error_len doesn't need to be shifted because it begins
165 * at offset zero.
166 */
167static inline bool iavf_test_staterr(union iavf_rx_desc *rx_desc,
168 const u64 stat_err_bits)
169{
170 return !!(rx_desc->wb.qword1.status_error_len &
171 cpu_to_le64(stat_err_bits));
172}
173
174/* How many Rx Buffers do we bundle into one write to the hardware ? */
175#define IAVF_RX_INCREMENT(r, i) \
176 do { \
177 (i)++; \
178 if ((i) == (r)->count) \
179 i = 0; \
180 r->next_to_clean = i; \
181 } while (0)
182
183#define IAVF_RX_NEXT_DESC(r, i, n) \
184 do { \
185 (i)++; \
186 if ((i) == (r)->count) \
187 i = 0; \
188 (n) = IAVF_RX_DESC((r), (i)); \
189 } while (0)
190
191#define IAVF_RX_NEXT_DESC_PREFETCH(r, i, n) \
192 do { \
193 IAVF_RX_NEXT_DESC((r), (i), (n)); \
194 prefetch((n)); \
195 } while (0)
196
197#define IAVF_MAX_BUFFER_TXD 8
198#define IAVF_MIN_TX_LEN 17
199
200/* The size limit for a transmit buffer in a descriptor is (16K - 1).
201 * In order to align with the read requests we will align the value to
202 * the nearest 4K which represents our maximum read request size.
203 */
204#define IAVF_MAX_READ_REQ_SIZE 4096
205#define IAVF_MAX_DATA_PER_TXD (16 * 1024 - 1)
206#define IAVF_MAX_DATA_PER_TXD_ALIGNED \
207 (IAVF_MAX_DATA_PER_TXD & ~(IAVF_MAX_READ_REQ_SIZE - 1))
208
209/**
210 * iavf_txd_use_count - estimate the number of descriptors needed for Tx
211 * @size: transmit request size in bytes
212 *
213 * Due to hardware alignment restrictions (4K alignment), we need to
214 * assume that we can have no more than 12K of data per descriptor, even
215 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
216 * Thus, we need to divide by 12K. But division is slow! Instead,
217 * we decompose the operation into shifts and one relatively cheap
218 * multiply operation.
219 *
220 * To divide by 12K, we first divide by 4K, then divide by 3:
221 * To divide by 4K, shift right by 12 bits
222 * To divide by 3, multiply by 85, then divide by 256
223 * (Divide by 256 is done by shifting right by 8 bits)
224 * Finally, we add one to round up. Because 256 isn't an exact multiple of
225 * 3, we'll underestimate near each multiple of 12K. This is actually more
226 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
227 * segment. For our purposes this is accurate out to 1M which is orders of
228 * magnitude greater than our largest possible GSO size.
229 *
230 * This would then be implemented as:
231 * return (((size >> 12) * 85) >> 8) + 1;
232 *
233 * Since multiplication and division are commutative, we can reorder
234 * operations into:
235 * return ((size * 85) >> 20) + 1;
236 */
237static inline unsigned int iavf_txd_use_count(unsigned int size)
238{
239 return ((size * 85) >> 20) + 1;
240}
241
242/* Tx Descriptors needed, worst case */
243#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
244#define IAVF_MIN_DESC_PENDING 4
245
246#define IAVF_TX_FLAGS_HW_VLAN BIT(1)
247#define IAVF_TX_FLAGS_SW_VLAN BIT(2)
248#define IAVF_TX_FLAGS_TSO BIT(3)
249#define IAVF_TX_FLAGS_IPV4 BIT(4)
250#define IAVF_TX_FLAGS_IPV6 BIT(5)
251#define IAVF_TX_FLAGS_FCCRC BIT(6)
252#define IAVF_TX_FLAGS_FSO BIT(7)
253#define IAVF_TX_FLAGS_FD_SB BIT(9)
254#define IAVF_TX_FLAGS_VXLAN_TUNNEL BIT(10)
255#define IAVF_TX_FLAGS_VLAN_MASK 0xffff0000
256#define IAVF_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
257#define IAVF_TX_FLAGS_VLAN_PRIO_SHIFT 29
258#define IAVF_TX_FLAGS_VLAN_SHIFT 16
259
260struct iavf_tx_buffer {
261 struct iavf_tx_desc *next_to_watch;
262 union {
263 struct sk_buff *skb;
264 void *raw_buf;
265 };
266 unsigned int bytecount;
267 unsigned short gso_segs;
268
269 DEFINE_DMA_UNMAP_ADDR(dma);
270 DEFINE_DMA_UNMAP_LEN(len);
271 u32 tx_flags;
272};
273
274struct iavf_rx_buffer {
275 dma_addr_t dma;
276 struct page *page;
277#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
278 __u32 page_offset;
279#else
280 __u16 page_offset;
281#endif
282 __u16 pagecnt_bias;
283};
284
285struct iavf_queue_stats {
286 u64 packets;
287 u64 bytes;
288};
289
290struct iavf_tx_queue_stats {
291 u64 restart_queue;
292 u64 tx_busy;
293 u64 tx_done_old;
294 u64 tx_linearize;
295 u64 tx_force_wb;
296 int prev_pkt_ctr;
297 u64 tx_lost_interrupt;
298};
299
300struct iavf_rx_queue_stats {
301 u64 non_eop_descs;
302 u64 alloc_page_failed;
303 u64 alloc_buff_failed;
304 u64 page_reuse_count;
305 u64 realloc_count;
306};
307
308enum iavf_ring_state_t {
309 __IAVF_TX_FDIR_INIT_DONE,
310 __IAVF_TX_XPS_INIT_DONE,
311 __IAVF_RING_STATE_NBITS /* must be last */
312};
313
314/* some useful defines for virtchannel interface, which
315 * is the only remaining user of header split
316 */
317#define IAVF_RX_DTYPE_NO_SPLIT 0
318#define IAVF_RX_DTYPE_HEADER_SPLIT 1
319#define IAVF_RX_DTYPE_SPLIT_ALWAYS 2
320#define IAVF_RX_SPLIT_L2 0x1
321#define IAVF_RX_SPLIT_IP 0x2
322#define IAVF_RX_SPLIT_TCP_UDP 0x4
323#define IAVF_RX_SPLIT_SCTP 0x8
324
325/* struct that defines a descriptor ring, associated with a VSI */
326struct iavf_ring {
327 struct iavf_ring *next; /* pointer to next ring in q_vector */
328 void *desc; /* Descriptor ring memory */
329 struct device *dev; /* Used for DMA mapping */
330 struct net_device *netdev; /* netdev ring maps to */
331 union {
332 struct iavf_tx_buffer *tx_bi;
333 struct iavf_rx_buffer *rx_bi;
334 };
335 DECLARE_BITMAP(state, __IAVF_RING_STATE_NBITS);
336 u16 queue_index; /* Queue number of ring */
337 u8 dcb_tc; /* Traffic class of ring */
338 u8 __iomem *tail;
339
340 /* high bit set means dynamic, use accessors routines to read/write.
341 * hardware only supports 2us resolution for the ITR registers.
342 * these values always store the USER setting, and must be converted
343 * before programming to a register.
344 */
345 u16 itr_setting;
346
347 u16 count; /* Number of descriptors */
348 u16 reg_idx; /* HW register index of the ring */
349 u16 rx_buf_len;
350
351 /* used in interrupt processing */
352 u16 next_to_use;
353 u16 next_to_clean;
354
355 u8 atr_sample_rate;
356 u8 atr_count;
357
358 bool ring_active; /* is ring online or not */
359 bool arm_wb; /* do something to arm write back */
360 u8 packet_stride;
361
362 u16 flags;
363#define IAVF_TXR_FLAGS_WB_ON_ITR BIT(0)
364#define IAVF_RXR_FLAGS_BUILD_SKB_ENABLED BIT(1)
365
366 /* stats structs */
367 struct iavf_queue_stats stats;
368 struct u64_stats_sync syncp;
369 union {
370 struct iavf_tx_queue_stats tx_stats;
371 struct iavf_rx_queue_stats rx_stats;
372 };
373
374 unsigned int size; /* length of descriptor ring in bytes */
375 dma_addr_t dma; /* physical address of ring */
376
377 struct iavf_vsi *vsi; /* Backreference to associated VSI */
378 struct iavf_q_vector *q_vector; /* Backreference to associated vector */
379
380 struct rcu_head rcu; /* to avoid race on free */
381 u16 next_to_alloc;
382 struct sk_buff *skb; /* When iavf_clean_rx_ring_irq() must
383 * return before it sees the EOP for
384 * the current packet, we save that skb
385 * here and resume receiving this
386 * packet the next time
387 * iavf_clean_rx_ring_irq() is called
388 * for this ring.
389 */
390} ____cacheline_internodealigned_in_smp;
391
392static inline bool ring_uses_build_skb(struct iavf_ring *ring)
393{
394 return !!(ring->flags & IAVF_RXR_FLAGS_BUILD_SKB_ENABLED);
395}
396
397static inline void set_ring_build_skb_enabled(struct iavf_ring *ring)
398{
399 ring->flags |= IAVF_RXR_FLAGS_BUILD_SKB_ENABLED;
400}
401
402static inline void clear_ring_build_skb_enabled(struct iavf_ring *ring)
403{
404 ring->flags &= ~IAVF_RXR_FLAGS_BUILD_SKB_ENABLED;
405}
406
407#define IAVF_ITR_ADAPTIVE_MIN_INC 0x0002
408#define IAVF_ITR_ADAPTIVE_MIN_USECS 0x0002
409#define IAVF_ITR_ADAPTIVE_MAX_USECS 0x007e
410#define IAVF_ITR_ADAPTIVE_LATENCY 0x8000
411#define IAVF_ITR_ADAPTIVE_BULK 0x0000
412#define ITR_IS_BULK(x) (!((x) & IAVF_ITR_ADAPTIVE_LATENCY))
413
414struct iavf_ring_container {
415 struct iavf_ring *ring; /* pointer to linked list of ring(s) */
416 unsigned long next_update; /* jiffies value of next update */
417 unsigned int total_bytes; /* total bytes processed this int */
418 unsigned int total_packets; /* total packets processed this int */
419 u16 count;
420 u16 target_itr; /* target ITR setting for ring(s) */
421 u16 current_itr; /* current ITR setting for ring(s) */
422};
423
424/* iterator for handling rings in ring container */
425#define iavf_for_each_ring(pos, head) \
426 for (pos = (head).ring; pos != NULL; pos = pos->next)
427
428static inline unsigned int iavf_rx_pg_order(struct iavf_ring *ring)
429{
430#if (PAGE_SIZE < 8192)
431 if (ring->rx_buf_len > (PAGE_SIZE / 2))
432 return 1;
433#endif
434 return 0;
435}
436
437#define iavf_rx_pg_size(_ring) (PAGE_SIZE << iavf_rx_pg_order(_ring))
438
439bool iavf_alloc_rx_buffers(struct iavf_ring *rxr, u16 cleaned_count);
440netdev_tx_t iavf_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
441void iavf_clean_tx_ring(struct iavf_ring *tx_ring);
442void iavf_clean_rx_ring(struct iavf_ring *rx_ring);
443int iavf_setup_tx_descriptors(struct iavf_ring *tx_ring);
444int iavf_setup_rx_descriptors(struct iavf_ring *rx_ring);
445void iavf_free_tx_resources(struct iavf_ring *tx_ring);
446void iavf_free_rx_resources(struct iavf_ring *rx_ring);
447int iavf_napi_poll(struct napi_struct *napi, int budget);
448void iavf_force_wb(struct iavf_vsi *vsi, struct iavf_q_vector *q_vector);
449u32 iavf_get_tx_pending(struct iavf_ring *ring, bool in_sw);
450void iavf_detect_recover_hung(struct iavf_vsi *vsi);
451int __iavf_maybe_stop_tx(struct iavf_ring *tx_ring, int size);
452bool __iavf_chk_linearize(struct sk_buff *skb);
453
454/**
455 * iavf_xmit_descriptor_count - calculate number of Tx descriptors needed
456 * @skb: send buffer
457 * @tx_ring: ring to send buffer on
458 *
459 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
460 * there is not enough descriptors available in this ring since we need at least
461 * one descriptor.
462 **/
463static inline int iavf_xmit_descriptor_count(struct sk_buff *skb)
464{
465 const skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
466 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
467 int count = 0, size = skb_headlen(skb);
468
469 for (;;) {
470 count += iavf_txd_use_count(size);
471
472 if (!nr_frags--)
473 break;
474
475 size = skb_frag_size(frag++);
476 }
477
478 return count;
479}
480
481/**
482 * iavf_maybe_stop_tx - 1st level check for Tx stop conditions
483 * @tx_ring: the ring to be checked
484 * @size: the size buffer we want to assure is available
485 *
486 * Returns 0 if stop is not needed
487 **/
488static inline int iavf_maybe_stop_tx(struct iavf_ring *tx_ring, int size)
489{
490 if (likely(IAVF_DESC_UNUSED(tx_ring) >= size))
491 return 0;
492 return __iavf_maybe_stop_tx(tx_ring, size);
493}
494
495/**
496 * iavf_chk_linearize - Check if there are more than 8 fragments per packet
497 * @skb: send buffer
498 * @count: number of buffers used
499 *
500 * Note: Our HW can't scatter-gather more than 8 fragments to build
501 * a packet on the wire and so we need to figure out the cases where we
502 * need to linearize the skb.
503 **/
504static inline bool iavf_chk_linearize(struct sk_buff *skb, int count)
505{
506 /* Both TSO and single send will work if count is less than 8 */
507 if (likely(count < IAVF_MAX_BUFFER_TXD))
508 return false;
509
510 if (skb_is_gso(skb))
511 return __iavf_chk_linearize(skb);
512
513 /* we can support up to 8 data buffers for a single send */
514 return count != IAVF_MAX_BUFFER_TXD;
515}
516/**
517 * @ring: Tx ring to find the netdev equivalent of
518 **/
519static inline struct netdev_queue *txring_txq(const struct iavf_ring *ring)
520{
521 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
522}
523#endif /* _IAVF_TXRX_H_ */