1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Virtual Function 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)
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 */
60 #define I40E_QUEUE_END_OF_LIST 0x7FF
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.
71 I40E_ITR_NONE = 3 /* ITR_NONE must not be used as an index */
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
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))
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))
101 /* Supported Rx Buffer Sizes (a multiple of 128) */
102 #define I40E_RXBUFFER_256 256
103 #define I40E_RXBUFFER_1536 1536 /* 128B aligned standard Ethernet frame */
104 #define I40E_RXBUFFER_2048 2048
105 #define I40E_RXBUFFER_3072 3072 /* Used for large frames w/ padding */
106 #define I40E_MAX_RXBUFFER 9728 /* largest size for single descriptor */
108 /* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
109 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
110 * this adds up to 512 bytes of extra data meaning the smallest allocation
111 * we could have is 1K.
112 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
113 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
115 #define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
116 #define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
117 #define i40e_rx_desc i40e_32byte_rx_desc
119 #define I40E_RX_DMA_ATTR \
120 (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
122 /* Attempt to maximize the headroom available for incoming frames. We
123 * use a 2K buffer for receives and need 1536/1534 to store the data for
124 * the frame. This leaves us with 512 bytes of room. From that we need
125 * to deduct the space needed for the shared info and the padding needed
126 * to IP align the frame.
128 * Note: For cache line sizes 256 or larger this value is going to end
129 * up negative. In these cases we should fall back to the legacy
132 #if (PAGE_SIZE < 8192)
133 #define I40E_2K_TOO_SMALL_WITH_PADDING \
134 ((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
136 static inline int i40e_compute_pad(int rx_buf_len)
138 int page_size, pad_size;
140 page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
141 pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
146 static inline int i40e_skb_pad(void)
150 /* If a 2K buffer cannot handle a standard Ethernet frame then
151 * optimize padding for a 3K buffer instead of a 1.5K buffer.
153 * For a 3K buffer we need to add enough padding to allow for
154 * tailroom due to NET_IP_ALIGN possibly shifting us out of
155 * cache-line alignment.
157 if (I40E_2K_TOO_SMALL_WITH_PADDING)
158 rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
160 rx_buf_len = I40E_RXBUFFER_1536;
162 /* if needed make room for NET_IP_ALIGN */
163 rx_buf_len -= NET_IP_ALIGN;
165 return i40e_compute_pad(rx_buf_len);
168 #define I40E_SKB_PAD i40e_skb_pad()
170 #define I40E_2K_TOO_SMALL_WITH_PADDING false
171 #define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
175 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
176 * @rx_desc: pointer to receive descriptor (in le64 format)
177 * @stat_err_bits: value to mask
179 * This function does some fast chicanery in order to return the
180 * value of the mask which is really only used for boolean tests.
181 * The status_error_len doesn't need to be shifted because it begins
184 static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
185 const u64 stat_err_bits)
187 return !!(rx_desc->wb.qword1.status_error_len &
188 cpu_to_le64(stat_err_bits));
191 /* How many Rx Buffers do we bundle into one write to the hardware ? */
192 #define I40E_RX_BUFFER_WRITE 16 /* Must be power of 2 */
193 #define I40E_RX_INCREMENT(r, i) \
196 if ((i) == (r)->count) \
198 r->next_to_clean = i; \
201 #define I40E_RX_NEXT_DESC(r, i, n) \
204 if ((i) == (r)->count) \
206 (n) = I40E_RX_DESC((r), (i)); \
209 #define I40E_RX_NEXT_DESC_PREFETCH(r, i, n) \
211 I40E_RX_NEXT_DESC((r), (i), (n)); \
215 #define I40E_MAX_BUFFER_TXD 8
216 #define I40E_MIN_TX_LEN 17
218 /* The size limit for a transmit buffer in a descriptor is (16K - 1).
219 * In order to align with the read requests we will align the value to
220 * the nearest 4K which represents our maximum read request size.
222 #define I40E_MAX_READ_REQ_SIZE 4096
223 #define I40E_MAX_DATA_PER_TXD (16 * 1024 - 1)
224 #define I40E_MAX_DATA_PER_TXD_ALIGNED \
225 (I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
228 * i40e_txd_use_count - estimate the number of descriptors needed for Tx
229 * @size: transmit request size in bytes
231 * Due to hardware alignment restrictions (4K alignment), we need to
232 * assume that we can have no more than 12K of data per descriptor, even
233 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
234 * Thus, we need to divide by 12K. But division is slow! Instead,
235 * we decompose the operation into shifts and one relatively cheap
236 * multiply operation.
238 * To divide by 12K, we first divide by 4K, then divide by 3:
239 * To divide by 4K, shift right by 12 bits
240 * To divide by 3, multiply by 85, then divide by 256
241 * (Divide by 256 is done by shifting right by 8 bits)
242 * Finally, we add one to round up. Because 256 isn't an exact multiple of
243 * 3, we'll underestimate near each multiple of 12K. This is actually more
244 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
245 * segment. For our purposes this is accurate out to 1M which is orders of
246 * magnitude greater than our largest possible GSO size.
248 * This would then be implemented as:
249 * return (((size >> 12) * 85) >> 8) + 1;
251 * Since multiplication and division are commutative, we can reorder
253 * return ((size * 85) >> 20) + 1;
255 static inline unsigned int i40e_txd_use_count(unsigned int size)
257 return ((size * 85) >> 20) + 1;
260 /* Tx Descriptors needed, worst case */
261 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
262 #define I40E_MIN_DESC_PENDING 4
264 #define I40E_TX_FLAGS_HW_VLAN BIT(1)
265 #define I40E_TX_FLAGS_SW_VLAN BIT(2)
266 #define I40E_TX_FLAGS_TSO BIT(3)
267 #define I40E_TX_FLAGS_IPV4 BIT(4)
268 #define I40E_TX_FLAGS_IPV6 BIT(5)
269 #define I40E_TX_FLAGS_FCCRC BIT(6)
270 #define I40E_TX_FLAGS_FSO BIT(7)
271 #define I40E_TX_FLAGS_FD_SB BIT(9)
272 #define I40E_TX_FLAGS_VXLAN_TUNNEL BIT(10)
273 #define I40E_TX_FLAGS_VLAN_MASK 0xffff0000
274 #define I40E_TX_FLAGS_VLAN_PRIO_MASK 0xe0000000
275 #define I40E_TX_FLAGS_VLAN_PRIO_SHIFT 29
276 #define I40E_TX_FLAGS_VLAN_SHIFT 16
278 struct i40e_tx_buffer {
279 struct i40e_tx_desc *next_to_watch;
284 unsigned int bytecount;
285 unsigned short gso_segs;
287 DEFINE_DMA_UNMAP_ADDR(dma);
288 DEFINE_DMA_UNMAP_LEN(len);
292 struct i40e_rx_buffer {
295 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
303 struct i40e_queue_stats {
308 struct i40e_tx_queue_stats {
314 u64 tx_lost_interrupt;
317 struct i40e_rx_queue_stats {
319 u64 alloc_page_failed;
320 u64 alloc_buff_failed;
321 u64 page_reuse_count;
325 enum i40e_ring_state_t {
326 __I40E_TX_FDIR_INIT_DONE,
327 __I40E_TX_XPS_INIT_DONE,
330 /* some useful defines for virtchannel interface, which
331 * is the only remaining user of header split
333 #define I40E_RX_DTYPE_NO_SPLIT 0
334 #define I40E_RX_DTYPE_HEADER_SPLIT 1
335 #define I40E_RX_DTYPE_SPLIT_ALWAYS 2
336 #define I40E_RX_SPLIT_L2 0x1
337 #define I40E_RX_SPLIT_IP 0x2
338 #define I40E_RX_SPLIT_TCP_UDP 0x4
339 #define I40E_RX_SPLIT_SCTP 0x8
341 /* struct that defines a descriptor ring, associated with a VSI */
343 struct i40e_ring *next; /* pointer to next ring in q_vector */
344 void *desc; /* Descriptor ring memory */
345 struct device *dev; /* Used for DMA mapping */
346 struct net_device *netdev; /* netdev ring maps to */
348 struct i40e_tx_buffer *tx_bi;
349 struct i40e_rx_buffer *rx_bi;
352 u16 queue_index; /* Queue number of ring */
353 u8 dcb_tc; /* Traffic class of ring */
356 /* high bit set means dynamic, use accessors routines to read/write.
357 * hardware only supports 2us resolution for the ITR registers.
358 * these values always store the USER setting, and must be converted
359 * before programming to a register.
364 u16 count; /* Number of descriptors */
365 u16 reg_idx; /* HW register index of the ring */
368 /* used in interrupt processing */
375 bool ring_active; /* is ring online or not */
376 bool arm_wb; /* do something to arm write back */
380 #define I40E_TXR_FLAGS_WB_ON_ITR BIT(0)
381 #define I40E_RXR_FLAGS_BUILD_SKB_ENABLED BIT(1)
384 struct i40e_queue_stats stats;
385 struct u64_stats_sync syncp;
387 struct i40e_tx_queue_stats tx_stats;
388 struct i40e_rx_queue_stats rx_stats;
391 unsigned int size; /* length of descriptor ring in bytes */
392 dma_addr_t dma; /* physical address of ring */
394 struct i40e_vsi *vsi; /* Backreference to associated VSI */
395 struct i40e_q_vector *q_vector; /* Backreference to associated vector */
397 struct rcu_head rcu; /* to avoid race on free */
399 struct sk_buff *skb; /* When i40evf_clean_rx_ring_irq() must
400 * return before it sees the EOP for
401 * the current packet, we save that skb
402 * here and resume receiving this
403 * packet the next time
404 * i40evf_clean_rx_ring_irq() is called
407 } ____cacheline_internodealigned_in_smp;
409 static inline bool ring_uses_build_skb(struct i40e_ring *ring)
411 return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
414 static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
416 ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
419 static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
421 ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
424 enum i40e_latency_range {
425 I40E_LOWEST_LATENCY = 0,
426 I40E_LOW_LATENCY = 1,
427 I40E_BULK_LATENCY = 2,
430 struct i40e_ring_container {
431 /* array of pointers to rings */
432 struct i40e_ring *ring;
433 unsigned int total_bytes; /* total bytes processed this int */
434 unsigned int total_packets; /* total packets processed this int */
435 unsigned long last_itr_update; /* jiffies of last ITR update */
437 enum i40e_latency_range latency_range;
441 /* iterator for handling rings in ring container */
442 #define i40e_for_each_ring(pos, head) \
443 for (pos = (head).ring; pos != NULL; pos = pos->next)
445 static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
447 #if (PAGE_SIZE < 8192)
448 if (ring->rx_buf_len > (PAGE_SIZE / 2))
454 #define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
456 bool i40evf_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
457 netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
458 void i40evf_clean_tx_ring(struct i40e_ring *tx_ring);
459 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring);
460 int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring);
461 int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring);
462 void i40evf_free_tx_resources(struct i40e_ring *tx_ring);
463 void i40evf_free_rx_resources(struct i40e_ring *rx_ring);
464 int i40evf_napi_poll(struct napi_struct *napi, int budget);
465 void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
466 u32 i40evf_get_tx_pending(struct i40e_ring *ring, bool in_sw);
467 int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
468 bool __i40evf_chk_linearize(struct sk_buff *skb);
471 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
473 * @tx_ring: ring to send buffer on
475 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
476 * there is not enough descriptors available in this ring since we need at least
479 static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
481 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
482 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
483 int count = 0, size = skb_headlen(skb);
486 count += i40e_txd_use_count(size);
491 size = skb_frag_size(frag++);
498 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
499 * @tx_ring: the ring to be checked
500 * @size: the size buffer we want to assure is available
502 * Returns 0 if stop is not needed
504 static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
506 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
508 return __i40evf_maybe_stop_tx(tx_ring, size);
512 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
514 * @count: number of buffers used
516 * Note: Our HW can't scatter-gather more than 8 fragments to build
517 * a packet on the wire and so we need to figure out the cases where we
518 * need to linearize the skb.
520 static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
522 /* Both TSO and single send will work if count is less than 8 */
523 if (likely(count < I40E_MAX_BUFFER_TXD))
527 return __i40evf_chk_linearize(skb);
529 /* we can support up to 8 data buffers for a single send */
530 return count != I40E_MAX_BUFFER_TXD;
533 * @ring: Tx ring to find the netdev equivalent of
535 static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
537 return netdev_get_tx_queue(ring->netdev, ring->queue_index);
539 #endif /* _I40E_TXRX_H_ */