1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Driver
4 * Copyright(c) 2013 - 2016 Intel Corporation.
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.
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
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/>.
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
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
25 ******************************************************************************/
30 /* Interrupt Throttling and Rate Limiting Goodies */
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
53 #define INTRL_ENA BIT(6)
54 #define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
56 * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
57 * @intrl: interrupt rate limit to convert
59 * This function converts a decimal interrupt rate limit to the appropriate
60 * register format expected by the firmware when setting interrupt rate limit.
62 static inline u16 i40e_intrl_usec_to_reg(int intrl)
65 return ((intrl >> 2) | INTRL_ENA);
69 #define I40E_INTRL_8K 125 /* 8000 ints/sec */
70 #define I40E_INTRL_62K 16 /* 62500 ints/sec */
71 #define I40E_INTRL_83K 12 /* 83333 ints/sec */
73 #define I40E_QUEUE_END_OF_LIST 0x7FF
75 /* this enum matches hardware bits and is meant to be used by DYN_CTLN
76 * registers and QINT registers or more generally anywhere in the manual
77 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
78 * register but instead is a special value meaning "don't update" ITR0/1/2.
84 I40E_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
87 /* these are indexes into ITRN registers */
88 #define I40E_RX_ITR I40E_IDX_ITR0
89 #define I40E_TX_ITR I40E_IDX_ITR1
90 #define I40E_PE_ITR I40E_IDX_ITR2
92 /* Supported RSS offloads */
93 #define I40E_DEFAULT_RSS_HENA ( \
94 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
95 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
96 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
97 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
98 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
99 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
100 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
101 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
102 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
103 BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
104 BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
106 #define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
107 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
108 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
109 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
110 BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
111 BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
112 BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
114 #define i40e_pf_get_default_rss_hena(pf) \
115 (((pf)->hw_features & I40E_HW_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
116 I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
118 /* Supported Rx Buffer Sizes (a multiple of 128) */
119 #define I40E_RXBUFFER_256 256
120 #define I40E_RXBUFFER_1536 1536 /* 128B aligned standard Ethernet frame */
121 #define I40E_RXBUFFER_2048 2048
122 #define I40E_RXBUFFER_3072 3072 /* Used for large frames w/ padding */
123 #define I40E_MAX_RXBUFFER 9728 /* largest size for single descriptor */
125 /* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
126 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
127 * this adds up to 512 bytes of extra data meaning the smallest allocation
128 * we could have is 1K.
129 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
130 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
132 #define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
133 #define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
134 #define i40e_rx_desc i40e_32byte_rx_desc
136 #define I40E_RX_DMA_ATTR \
137 (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
139 /* Attempt to maximize the headroom available for incoming frames. We
140 * use a 2K buffer for receives and need 1536/1534 to store the data for
141 * the frame. This leaves us with 512 bytes of room. From that we need
142 * to deduct the space needed for the shared info and the padding needed
143 * to IP align the frame.
145 * Note: For cache line sizes 256 or larger this value is going to end
146 * up negative. In these cases we should fall back to the legacy
149 #if (PAGE_SIZE < 8192)
150 #define I40E_2K_TOO_SMALL_WITH_PADDING \
151 ((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
153 static inline int i40e_compute_pad(int rx_buf_len)
155 int page_size, pad_size;
157 page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
158 pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
163 static inline int i40e_skb_pad(void)
167 /* If a 2K buffer cannot handle a standard Ethernet frame then
168 * optimize padding for a 3K buffer instead of a 1.5K buffer.
170 * For a 3K buffer we need to add enough padding to allow for
171 * tailroom due to NET_IP_ALIGN possibly shifting us out of
172 * cache-line alignment.
174 if (I40E_2K_TOO_SMALL_WITH_PADDING)
175 rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
177 rx_buf_len = I40E_RXBUFFER_1536;
179 /* if needed make room for NET_IP_ALIGN */
180 rx_buf_len -= NET_IP_ALIGN;
182 return i40e_compute_pad(rx_buf_len);
185 #define I40E_SKB_PAD i40e_skb_pad()
187 #define I40E_2K_TOO_SMALL_WITH_PADDING false
188 #define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
192 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
193 * @rx_desc: pointer to receive descriptor (in le64 format)
194 * @stat_err_bits: value to mask
196 * This function does some fast chicanery in order to return the
197 * value of the mask which is really only used for boolean tests.
198 * The status_error_len doesn't need to be shifted because it begins
201 static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
202 const u64 stat_err_bits)
204 return !!(rx_desc->wb.qword1.status_error_len &
205 cpu_to_le64(stat_err_bits));
208 /* How many Rx Buffers do we bundle into one write to the hardware ? */
209 #define I40E_RX_BUFFER_WRITE 16 /* Must be power of 2 */
210 #define I40E_RX_INCREMENT(r, i) \
213 if ((i) == (r)->count) \
215 r->next_to_clean = i; \
218 #define I40E_RX_NEXT_DESC(r, i, n) \
221 if ((i) == (r)->count) \
223 (n) = I40E_RX_DESC((r), (i)); \
226 #define I40E_RX_NEXT_DESC_PREFETCH(r, i, n) \
228 I40E_RX_NEXT_DESC((r), (i), (n)); \
232 #define I40E_MAX_BUFFER_TXD 8
233 #define I40E_MIN_TX_LEN 17
235 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
236 * In order to align with the read requests we will align the value to
237 * the nearest 4K which represents our maximum read request size.
239 #define I40E_MAX_READ_REQ_SIZE 4096
240 #define I40E_MAX_DATA_PER_TXD (16 * 1024 - 1)
241 #define I40E_MAX_DATA_PER_TXD_ALIGNED \
242 (I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
245 * i40e_txd_use_count - estimate the number of descriptors needed for Tx
246 * @size: transmit request size in bytes
248 * Due to hardware alignment restrictions (4K alignment), we need to
249 * assume that we can have no more than 12K of data per descriptor, even
250 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
251 * Thus, we need to divide by 12K. But division is slow! Instead,
252 * we decompose the operation into shifts and one relatively cheap
253 * multiply operation.
255 * To divide by 12K, we first divide by 4K, then divide by 3:
256 * To divide by 4K, shift right by 12 bits
257 * To divide by 3, multiply by 85, then divide by 256
258 * (Divide by 256 is done by shifting right by 8 bits)
259 * Finally, we add one to round up. Because 256 isn't an exact multiple of
260 * 3, we'll underestimate near each multiple of 12K. This is actually more
261 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
262 * segment. For our purposes this is accurate out to 1M which is orders of
263 * magnitude greater than our largest possible GSO size.
265 * This would then be implemented as:
266 * return (((size >> 12) * 85) >> 8) + 1;
268 * Since multiplication and division are commutative, we can reorder
270 * return ((size * 85) >> 20) + 1;
272 static inline unsigned int i40e_txd_use_count(unsigned int size)
274 return ((size * 85) >> 20) + 1;
277 /* Tx Descriptors needed, worst case */
278 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
279 #define I40E_MIN_DESC_PENDING 4
281 #define I40E_TX_FLAGS_HW_VLAN BIT(1)
282 #define I40E_TX_FLAGS_SW_VLAN BIT(2)
283 #define I40E_TX_FLAGS_TSO BIT(3)
284 #define I40E_TX_FLAGS_IPV4 BIT(4)
285 #define I40E_TX_FLAGS_IPV6 BIT(5)
286 #define I40E_TX_FLAGS_FCCRC BIT(6)
287 #define I40E_TX_FLAGS_FSO BIT(7)
288 #define I40E_TX_FLAGS_TSYN BIT(8)
289 #define I40E_TX_FLAGS_FD_SB BIT(9)
290 #define I40E_TX_FLAGS_UDP_TUNNEL BIT(10)
291 #define I40E_TX_FLAGS_VLAN_MASK 0xffff0000
292 #define I40E_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
293 #define I40E_TX_FLAGS_VLAN_PRIO_SHIFT 29
294 #define I40E_TX_FLAGS_VLAN_SHIFT 16
296 struct i40e_tx_buffer {
297 struct i40e_tx_desc *next_to_watch;
302 unsigned int bytecount;
303 unsigned short gso_segs;
305 DEFINE_DMA_UNMAP_ADDR(dma);
306 DEFINE_DMA_UNMAP_LEN(len);
310 struct i40e_rx_buffer {
313 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
321 struct i40e_queue_stats {
326 struct i40e_tx_queue_stats {
334 struct i40e_rx_queue_stats {
336 u64 alloc_page_failed;
337 u64 alloc_buff_failed;
338 u64 page_reuse_count;
342 enum i40e_ring_state_t {
343 __I40E_TX_FDIR_INIT_DONE,
344 __I40E_TX_XPS_INIT_DONE,
347 /* some useful defines for virtchannel interface, which
348 * is the only remaining user of header split
350 #define I40E_RX_DTYPE_NO_SPLIT 0
351 #define I40E_RX_DTYPE_HEADER_SPLIT 1
352 #define I40E_RX_DTYPE_SPLIT_ALWAYS 2
353 #define I40E_RX_SPLIT_L2 0x1
354 #define I40E_RX_SPLIT_IP 0x2
355 #define I40E_RX_SPLIT_TCP_UDP 0x4
356 #define I40E_RX_SPLIT_SCTP 0x8
358 /* struct that defines a descriptor ring, associated with a VSI */
360 struct i40e_ring *next; /* pointer to next ring in q_vector */
361 void *desc; /* Descriptor ring memory */
362 struct device *dev; /* Used for DMA mapping */
363 struct net_device *netdev; /* netdev ring maps to */
364 struct bpf_prog *xdp_prog;
366 struct i40e_tx_buffer *tx_bi;
367 struct i40e_rx_buffer *rx_bi;
370 u16 queue_index; /* Queue number of ring */
371 u8 dcb_tc; /* Traffic class of ring */
374 /* high bit set means dynamic, use accessor routines to read/write.
375 * hardware only supports 2us resolution for the ITR registers.
376 * these values always store the USER setting, and must be converted
377 * before programming to a register.
382 u16 count; /* Number of descriptors */
383 u16 reg_idx; /* HW register index of the ring */
386 /* used in interrupt processing */
393 bool ring_active; /* is ring online or not */
394 bool arm_wb; /* do something to arm write back */
398 #define I40E_TXR_FLAGS_WB_ON_ITR BIT(0)
399 #define I40E_RXR_FLAGS_BUILD_SKB_ENABLED BIT(1)
400 #define I40E_TXR_FLAGS_XDP BIT(2)
403 struct i40e_queue_stats stats;
404 struct u64_stats_sync syncp;
406 struct i40e_tx_queue_stats tx_stats;
407 struct i40e_rx_queue_stats rx_stats;
410 unsigned int size; /* length of descriptor ring in bytes */
411 dma_addr_t dma; /* physical address of ring */
413 struct i40e_vsi *vsi; /* Backreference to associated VSI */
414 struct i40e_q_vector *q_vector; /* Backreference to associated vector */
416 struct rcu_head rcu; /* to avoid race on free */
418 struct sk_buff *skb; /* When i40e_clean_rx_ring_irq() must
419 * return before it sees the EOP for
420 * the current packet, we save that skb
421 * here and resume receiving this
422 * packet the next time
423 * i40e_clean_rx_ring_irq() is called
426 } ____cacheline_internodealigned_in_smp;
428 static inline bool ring_uses_build_skb(struct i40e_ring *ring)
430 return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
433 static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
435 ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
438 static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
440 ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
443 static inline bool ring_is_xdp(struct i40e_ring *ring)
445 return !!(ring->flags & I40E_TXR_FLAGS_XDP);
448 static inline void set_ring_xdp(struct i40e_ring *ring)
450 ring->flags |= I40E_TXR_FLAGS_XDP;
453 enum i40e_latency_range {
454 I40E_LOWEST_LATENCY = 0,
455 I40E_LOW_LATENCY = 1,
456 I40E_BULK_LATENCY = 2,
459 struct i40e_ring_container {
460 /* array of pointers to rings */
461 struct i40e_ring *ring;
462 unsigned int total_bytes; /* total bytes processed this int */
463 unsigned int total_packets; /* total packets processed this int */
464 unsigned long last_itr_update; /* jiffies of last ITR update */
466 enum i40e_latency_range latency_range;
470 /* iterator for handling rings in ring container */
471 #define i40e_for_each_ring(pos, head) \
472 for (pos = (head).ring; pos != NULL; pos = pos->next)
474 static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
476 #if (PAGE_SIZE < 8192)
477 if (ring->rx_buf_len > (PAGE_SIZE / 2))
483 #define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
485 bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
486 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
487 void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
488 void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
489 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
490 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
491 void i40e_free_tx_resources(struct i40e_ring *tx_ring);
492 void i40e_free_rx_resources(struct i40e_ring *rx_ring);
493 int i40e_napi_poll(struct napi_struct *napi, int budget);
494 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
495 u32 i40e_get_tx_pending(struct i40e_ring *ring);
496 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
497 bool __i40e_chk_linearize(struct sk_buff *skb);
500 * i40e_get_head - Retrieve head from head writeback
501 * @tx_ring: tx ring to fetch head of
503 * Returns value of Tx ring head based on value stored
504 * in head write-back location
506 static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
508 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
510 return le32_to_cpu(*(volatile __le32 *)head);
514 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
516 * @tx_ring: ring to send buffer on
518 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
519 * there is not enough descriptors available in this ring since we need at least
522 static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
524 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
525 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
526 int count = 0, size = skb_headlen(skb);
529 count += i40e_txd_use_count(size);
534 size = skb_frag_size(frag++);
541 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
542 * @tx_ring: the ring to be checked
543 * @size: the size buffer we want to assure is available
545 * Returns 0 if stop is not needed
547 static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
549 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
551 return __i40e_maybe_stop_tx(tx_ring, size);
555 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
557 * @count: number of buffers used
559 * Note: Our HW can't scatter-gather more than 8 fragments to build
560 * a packet on the wire and so we need to figure out the cases where we
561 * need to linearize the skb.
563 static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
565 /* Both TSO and single send will work if count is less than 8 */
566 if (likely(count < I40E_MAX_BUFFER_TXD))
570 return __i40e_chk_linearize(skb);
572 /* we can support up to 8 data buffers for a single send */
573 return count != I40E_MAX_BUFFER_TXD;
577 * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
578 * @ring: Tx ring to find the netdev equivalent of
580 static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
582 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
584 #endif /* _I40E_TXRX_H_ */