Linux Audio

Check our new training course

Loading...
v5.9
  1/* SPDX-License-Identifier: GPL-2.0 */
  2/* Copyright(c) 2013 - 2018 Intel Corporation. */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  3
  4#ifndef _I40E_TXRX_H_
  5#define _I40E_TXRX_H_
  6
  7#include <net/xdp.h>
  8
  9/* Interrupt Throttling and Rate Limiting Goodies */
 10#define I40E_DEFAULT_IRQ_WORK      256
 11
 12/* The datasheet for the X710 and XL710 indicate that the maximum value for
 13 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
 14 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
 15 * the register value which is divided by 2 lets use the actual values and
 16 * avoid an excessive amount of translation.
 17 */
 18#define I40E_ITR_DYNAMIC	0x8000	/* use top bit as a flag */
 19#define I40E_ITR_MASK		0x1FFE	/* mask for ITR register value */
 20#define I40E_MIN_ITR		     2	/* reg uses 2 usec resolution */
 21#define I40E_ITR_20K		    50
 22#define I40E_ITR_8K		   122
 23#define I40E_MAX_ITR		  8160	/* maximum value as per datasheet */
 24#define ITR_TO_REG(setting) ((setting) & ~I40E_ITR_DYNAMIC)
 25#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~I40E_ITR_MASK)
 26#define ITR_IS_DYNAMIC(setting) (!!((setting) & I40E_ITR_DYNAMIC))
 27
 28#define I40E_ITR_RX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
 29#define I40E_ITR_TX_DEF		(I40E_ITR_20K | I40E_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 I40E_MAX_INTRL             0x3B    /* reg uses 4 usec resolution */
 36#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
 37
 38/**
 39 * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
 40 * @intrl: interrupt rate limit to convert
 41 *
 42 * This function converts a decimal interrupt rate limit to the appropriate
 43 * register format expected by the firmware when setting interrupt rate limit.
 44 */
 45static inline u16 i40e_intrl_usec_to_reg(int intrl)
 46{
 47	if (intrl >> 2)
 48		return ((intrl >> 2) | INTRL_ENA);
 49	else
 50		return 0;
 51}
 52
 53#define I40E_QUEUE_END_OF_LIST 0x7FF
 54
 55/* this enum matches hardware bits and is meant to be used by DYN_CTLN
 56 * registers and QINT registers or more generally anywhere in the manual
 57 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
 58 * register but instead is a special value meaning "don't update" ITR0/1/2.
 59 */
 60enum i40e_dyn_idx_t {
 61	I40E_IDX_ITR0 = 0,
 62	I40E_IDX_ITR1 = 1,
 63	I40E_IDX_ITR2 = 2,
 64	I40E_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
 65};
 66
 67/* these are indexes into ITRN registers */
 68#define I40E_RX_ITR    I40E_IDX_ITR0
 69#define I40E_TX_ITR    I40E_IDX_ITR1
 
 70
 71/* Supported RSS offloads */
 72#define I40E_DEFAULT_RSS_HENA ( \
 73	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
 74	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
 75	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
 76	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
 77	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
 78	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
 79	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
 80	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
 81	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
 82	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
 83	BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
 84
 85#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
 86	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
 87	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
 88	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
 89	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
 90	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
 91	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
 92
 93#define i40e_pf_get_default_rss_hena(pf) \
 94	(((pf)->hw_features & I40E_HW_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
 95	  I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
 96
 97/* Supported Rx Buffer Sizes (a multiple of 128) */
 98#define I40E_RXBUFFER_256   256
 99#define I40E_RXBUFFER_1536  1536  /* 128B aligned standard Ethernet frame */
100#define I40E_RXBUFFER_2048  2048
101#define I40E_RXBUFFER_3072  3072  /* Used for large frames w/ padding */
 
 
102#define I40E_MAX_RXBUFFER   9728  /* largest size for single descriptor */
103
104/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
105 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
106 * this adds up to 512 bytes of extra data meaning the smallest allocation
107 * we could have is 1K.
108 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
109 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
110 */
111#define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
112#define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
113#define i40e_rx_desc i40e_32byte_rx_desc
114
115#define I40E_RX_DMA_ATTR \
116	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
117
118/* Attempt to maximize the headroom available for incoming frames.  We
119 * use a 2K buffer for receives and need 1536/1534 to store the data for
120 * the frame.  This leaves us with 512 bytes of room.  From that we need
121 * to deduct the space needed for the shared info and the padding needed
122 * to IP align the frame.
123 *
124 * Note: For cache line sizes 256 or larger this value is going to end
125 *	 up negative.  In these cases we should fall back to the legacy
126 *	 receive path.
127 */
128#if (PAGE_SIZE < 8192)
129#define I40E_2K_TOO_SMALL_WITH_PADDING \
130((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
131
132static inline int i40e_compute_pad(int rx_buf_len)
133{
134	int page_size, pad_size;
135
136	page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
137	pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
138
139	return pad_size;
140}
141
142static inline int i40e_skb_pad(void)
143{
144	int rx_buf_len;
145
146	/* If a 2K buffer cannot handle a standard Ethernet frame then
147	 * optimize padding for a 3K buffer instead of a 1.5K buffer.
148	 *
149	 * For a 3K buffer we need to add enough padding to allow for
150	 * tailroom due to NET_IP_ALIGN possibly shifting us out of
151	 * cache-line alignment.
152	 */
153	if (I40E_2K_TOO_SMALL_WITH_PADDING)
154		rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
155	else
156		rx_buf_len = I40E_RXBUFFER_1536;
157
158	/* if needed make room for NET_IP_ALIGN */
159	rx_buf_len -= NET_IP_ALIGN;
160
161	return i40e_compute_pad(rx_buf_len);
162}
163
164#define I40E_SKB_PAD i40e_skb_pad()
165#else
166#define I40E_2K_TOO_SMALL_WITH_PADDING false
167#define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
168#endif
169
170/**
171 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
172 * @rx_desc: pointer to receive descriptor (in le64 format)
173 * @stat_err_bits: value to mask
174 *
175 * This function does some fast chicanery in order to return the
176 * value of the mask which is really only used for boolean tests.
177 * The status_error_len doesn't need to be shifted because it begins
178 * at offset zero.
179 */
180static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
181				     const u64 stat_err_bits)
182{
183	return !!(rx_desc->wb.qword1.status_error_len &
184		  cpu_to_le64(stat_err_bits));
185}
186
187/* How many Rx Buffers do we bundle into one write to the hardware ? */
188#define I40E_RX_BUFFER_WRITE	32	/* Must be power of 2 */
 
 
 
 
 
 
 
189
190#define I40E_RX_NEXT_DESC(r, i, n)		\
191	do {					\
192		(i)++;				\
193		if ((i) == (r)->count)		\
194			i = 0;			\
195		(n) = I40E_RX_DESC((r), (i));	\
196	} while (0)
197
 
 
 
 
 
 
 
198
199#define I40E_MAX_BUFFER_TXD	8
200#define I40E_MIN_TX_LEN		17
201
202/* The size limit for a transmit buffer in a descriptor is (16K - 1).
203 * In order to align with the read requests we will align the value to
204 * the nearest 4K which represents our maximum read request size.
205 */
206#define I40E_MAX_READ_REQ_SIZE		4096
207#define I40E_MAX_DATA_PER_TXD		(16 * 1024 - 1)
208#define I40E_MAX_DATA_PER_TXD_ALIGNED \
209	(I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
210
211/**
212 * i40e_txd_use_count  - estimate the number of descriptors needed for Tx
213 * @size: transmit request size in bytes
214 *
215 * Due to hardware alignment restrictions (4K alignment), we need to
216 * assume that we can have no more than 12K of data per descriptor, even
217 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
218 * Thus, we need to divide by 12K. But division is slow! Instead,
219 * we decompose the operation into shifts and one relatively cheap
220 * multiply operation.
221 *
222 * To divide by 12K, we first divide by 4K, then divide by 3:
223 *     To divide by 4K, shift right by 12 bits
224 *     To divide by 3, multiply by 85, then divide by 256
225 *     (Divide by 256 is done by shifting right by 8 bits)
226 * Finally, we add one to round up. Because 256 isn't an exact multiple of
227 * 3, we'll underestimate near each multiple of 12K. This is actually more
228 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
229 * segment.  For our purposes this is accurate out to 1M which is orders of
230 * magnitude greater than our largest possible GSO size.
231 *
232 * This would then be implemented as:
233 *     return (((size >> 12) * 85) >> 8) + 1;
234 *
235 * Since multiplication and division are commutative, we can reorder
236 * operations into:
237 *     return ((size * 85) >> 20) + 1;
238 */
239static inline unsigned int i40e_txd_use_count(unsigned int size)
240{
241	return ((size * 85) >> 20) + 1;
242}
243
244/* Tx Descriptors needed, worst case */
245#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
 
 
246
247#define I40E_TX_FLAGS_HW_VLAN		BIT(1)
248#define I40E_TX_FLAGS_SW_VLAN		BIT(2)
249#define I40E_TX_FLAGS_TSO		BIT(3)
250#define I40E_TX_FLAGS_IPV4		BIT(4)
251#define I40E_TX_FLAGS_IPV6		BIT(5)
 
 
252#define I40E_TX_FLAGS_TSYN		BIT(8)
253#define I40E_TX_FLAGS_FD_SB		BIT(9)
254#define I40E_TX_FLAGS_UDP_TUNNEL	BIT(10)
255#define I40E_TX_FLAGS_VLAN_MASK		0xffff0000
256#define I40E_TX_FLAGS_VLAN_PRIO_MASK	0xe0000000
257#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT	29
258#define I40E_TX_FLAGS_VLAN_SHIFT	16
259
260struct i40e_tx_buffer {
261	struct i40e_tx_desc *next_to_watch;
262	union {
263		struct xdp_frame *xdpf;
264		struct sk_buff *skb;
265		void *raw_buf;
266	};
267	unsigned int bytecount;
268	unsigned short gso_segs;
269
270	DEFINE_DMA_UNMAP_ADDR(dma);
271	DEFINE_DMA_UNMAP_LEN(len);
272	u32 tx_flags;
273};
274
275struct i40e_rx_buffer {
 
 
276	dma_addr_t dma;
277	struct page *page;
278	__u32 page_offset;
279	__u16 pagecnt_bias;
280};
281
282struct i40e_queue_stats {
283	u64 packets;
284	u64 bytes;
285};
286
287struct i40e_tx_queue_stats {
288	u64 restart_queue;
289	u64 tx_busy;
290	u64 tx_done_old;
291	u64 tx_linearize;
292	u64 tx_force_wb;
293	int prev_pkt_ctr;
294};
295
296struct i40e_rx_queue_stats {
297	u64 non_eop_descs;
298	u64 alloc_page_failed;
299	u64 alloc_buff_failed;
300	u64 page_reuse_count;
301	u64 realloc_count;
302};
303
304enum i40e_ring_state_t {
305	__I40E_TX_FDIR_INIT_DONE,
306	__I40E_TX_XPS_INIT_DONE,
307	__I40E_RING_STATE_NBITS /* must be last */
 
308};
309
310/* some useful defines for virtchannel interface, which
311 * is the only remaining user of header split
312 */
313#define I40E_RX_DTYPE_HEADER_SPLIT  1
314#define I40E_RX_SPLIT_L2      0x1
315#define I40E_RX_SPLIT_IP      0x2
316#define I40E_RX_SPLIT_TCP_UDP 0x4
317#define I40E_RX_SPLIT_SCTP    0x8
 
 
 
 
318
319/* struct that defines a descriptor ring, associated with a VSI */
320struct i40e_ring {
321	struct i40e_ring *next;		/* pointer to next ring in q_vector */
322	void *desc;			/* Descriptor ring memory */
323	struct device *dev;		/* Used for DMA mapping */
324	struct net_device *netdev;	/* netdev ring maps to */
325	struct bpf_prog *xdp_prog;
326	union {
327		struct i40e_tx_buffer *tx_bi;
328		struct i40e_rx_buffer *rx_bi;
329		struct xdp_buff **rx_bi_zc;
330	};
331	DECLARE_BITMAP(state, __I40E_RING_STATE_NBITS);
332	u16 queue_index;		/* Queue number of ring */
333	u8 dcb_tc;			/* Traffic class of ring */
334	u8 __iomem *tail;
335
336	/* high bit set means dynamic, use accessor routines to read/write.
337	 * hardware only supports 2us resolution for the ITR registers.
338	 * these values always store the USER setting, and must be converted
339	 * before programming to a register.
340	 */
341	u16 itr_setting;
 
342
343	u16 count;			/* Number of descriptors */
344	u16 reg_idx;			/* HW register index of the ring */
 
345	u16 rx_buf_len;
 
 
 
 
 
 
 
 
346
347	/* used in interrupt processing */
348	u16 next_to_use;
349	u16 next_to_clean;
350	u16 xdp_tx_active;
351
352	u8 atr_sample_rate;
353	u8 atr_count;
354
 
 
355	bool ring_active;		/* is ring online or not */
356	bool arm_wb;		/* do something to arm write back */
357	u8 packet_stride;
358
359	u16 flags;
360#define I40E_TXR_FLAGS_WB_ON_ITR		BIT(0)
361#define I40E_RXR_FLAGS_BUILD_SKB_ENABLED	BIT(1)
362#define I40E_TXR_FLAGS_XDP			BIT(2)
363
364	/* stats structs */
365	struct i40e_queue_stats	stats;
366	struct u64_stats_sync syncp;
367	union {
368		struct i40e_tx_queue_stats tx_stats;
369		struct i40e_rx_queue_stats rx_stats;
370	};
371
372	unsigned int size;		/* length of descriptor ring in bytes */
373	dma_addr_t dma;			/* physical address of ring */
374
375	struct i40e_vsi *vsi;		/* Backreference to associated VSI */
376	struct i40e_q_vector *q_vector;	/* Backreference to associated vector */
377
378	struct rcu_head rcu;		/* to avoid race on free */
379	u16 next_to_alloc;
380	struct sk_buff *skb;		/* When i40e_clean_rx_ring_irq() must
381					 * return before it sees the EOP for
382					 * the current packet, we save that skb
383					 * here and resume receiving this
384					 * packet the next time
385					 * i40e_clean_rx_ring_irq() is called
386					 * for this ring.
387					 */
388
389	struct i40e_channel *ch;
390	struct xdp_rxq_info xdp_rxq;
391	struct xdp_umem *xsk_umem;
392} ____cacheline_internodealigned_in_smp;
393
394static inline bool ring_uses_build_skb(struct i40e_ring *ring)
395{
396	return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
397}
398
399static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
400{
401	ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
402}
403
404static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
405{
406	ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
407}
408
409static inline bool ring_is_xdp(struct i40e_ring *ring)
410{
411	return !!(ring->flags & I40E_TXR_FLAGS_XDP);
412}
413
414static inline void set_ring_xdp(struct i40e_ring *ring)
415{
416	ring->flags |= I40E_TXR_FLAGS_XDP;
417}
418
419#define I40E_ITR_ADAPTIVE_MIN_INC	0x0002
420#define I40E_ITR_ADAPTIVE_MIN_USECS	0x0002
421#define I40E_ITR_ADAPTIVE_MAX_USECS	0x007e
422#define I40E_ITR_ADAPTIVE_LATENCY	0x8000
423#define I40E_ITR_ADAPTIVE_BULK		0x0000
424
425struct i40e_ring_container {
426	struct i40e_ring *ring;		/* pointer to linked list of ring(s) */
427	unsigned long next_update;	/* jiffies value of next update */
428	unsigned int total_bytes;	/* total bytes processed this int */
429	unsigned int total_packets;	/* total packets processed this int */
430	u16 count;
431	u16 target_itr;			/* target ITR setting for ring(s) */
432	u16 current_itr;		/* current ITR setting for ring(s) */
433};
434
435/* iterator for handling rings in ring container */
436#define i40e_for_each_ring(pos, head) \
437	for (pos = (head).ring; pos != NULL; pos = pos->next)
438
439static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
440{
441#if (PAGE_SIZE < 8192)
442	if (ring->rx_buf_len > (PAGE_SIZE / 2))
443		return 1;
444#endif
445	return 0;
446}
447
448#define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
449
450bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
451netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
452void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
453void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
454int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
455int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
456void i40e_free_tx_resources(struct i40e_ring *tx_ring);
457void i40e_free_rx_resources(struct i40e_ring *rx_ring);
458int i40e_napi_poll(struct napi_struct *napi, int budget);
 
 
 
 
 
 
 
459void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
460u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
461void i40e_detect_recover_hung(struct i40e_vsi *vsi);
462int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
463bool __i40e_chk_linearize(struct sk_buff *skb);
464int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
465		  u32 flags);
466int i40e_alloc_rx_bi(struct i40e_ring *rx_ring);
467
468/**
469 * i40e_get_head - Retrieve head from head writeback
470 * @tx_ring:  tx ring to fetch head of
471 *
472 * Returns value of Tx ring head based on value stored
473 * in head write-back location
474 **/
475static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
476{
477	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
478
479	return le32_to_cpu(*(volatile __le32 *)head);
480}
481
482/**
483 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
484 * @skb:     send buffer
485 * @tx_ring: ring to send buffer on
486 *
487 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
488 * there is not enough descriptors available in this ring since we need at least
489 * one descriptor.
490 **/
491static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
492{
493	const skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
494	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
495	int count = 0, size = skb_headlen(skb);
496
497	for (;;) {
498		count += i40e_txd_use_count(size);
499
500		if (!nr_frags--)
501			break;
502
503		size = skb_frag_size(frag++);
504	}
505
506	return count;
507}
508
509/**
510 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
511 * @tx_ring: the ring to be checked
512 * @size:    the size buffer we want to assure is available
513 *
514 * Returns 0 if stop is not needed
515 **/
516static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
517{
518	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
519		return 0;
520	return __i40e_maybe_stop_tx(tx_ring, size);
521}
522
523/**
524 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
525 * @skb:      send buffer
526 * @count:    number of buffers used
527 *
528 * Note: Our HW can't scatter-gather more than 8 fragments to build
529 * a packet on the wire and so we need to figure out the cases where we
530 * need to linearize the skb.
531 **/
532static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
533{
534	/* Both TSO and single send will work if count is less than 8 */
535	if (likely(count < I40E_MAX_BUFFER_TXD))
536		return false;
537
538	if (skb_is_gso(skb))
539		return __i40e_chk_linearize(skb);
540
541	/* we can support up to 8 data buffers for a single send */
542	return count != I40E_MAX_BUFFER_TXD;
543}
544
545/**
546 * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
547 * @ring: Tx ring to find the netdev equivalent of
548 **/
549static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
550{
551	return netdev_get_tx_queue(ring->netdev, ring->queue_index);
552}
553#endif /* _I40E_TXRX_H_ */
v4.6
  1/*******************************************************************************
  2 *
  3 * Intel Ethernet Controller XL710 Family Linux 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_TSYN		BIT(8)
164#define I40E_TX_FLAGS_FD_SB		BIT(9)
165#define I40E_TX_FLAGS_UDP_TUNNEL	BIT(10)
166#define I40E_TX_FLAGS_VLAN_MASK		0xffff0000
167#define I40E_TX_FLAGS_VLAN_PRIO_MASK	0xe0000000
168#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT	29
169#define I40E_TX_FLAGS_VLAN_SHIFT	16
170
171struct i40e_tx_buffer {
172	struct i40e_tx_desc *next_to_watch;
173	union {
 
174		struct sk_buff *skb;
175		void *raw_buf;
176	};
177	unsigned int bytecount;
178	unsigned short gso_segs;
179
180	DEFINE_DMA_UNMAP_ADDR(dma);
181	DEFINE_DMA_UNMAP_LEN(len);
182	u32 tx_flags;
183};
184
185struct i40e_rx_buffer {
186	struct sk_buff *skb;
187	void *hdr_buf;
188	dma_addr_t dma;
189	struct page *page;
190	dma_addr_t page_dma;
191	unsigned int page_offset;
192};
193
194struct i40e_queue_stats {
195	u64 packets;
196	u64 bytes;
197};
198
199struct i40e_tx_queue_stats {
200	u64 restart_queue;
201	u64 tx_busy;
202	u64 tx_done_old;
203	u64 tx_linearize;
204	u64 tx_force_wb;
205	u64 tx_lost_interrupt;
206};
207
208struct i40e_rx_queue_stats {
209	u64 non_eop_descs;
210	u64 alloc_page_failed;
211	u64 alloc_buff_failed;
212	u64 page_reuse_count;
213	u64 realloc_count;
214};
215
216enum i40e_ring_state_t {
217	__I40E_TX_FDIR_INIT_DONE,
218	__I40E_TX_XPS_INIT_DONE,
219	__I40E_RX_PS_ENABLED,
220	__I40E_RX_16BYTE_DESC_ENABLED,
221};
222
223#define ring_is_ps_enabled(ring) \
224	test_bit(__I40E_RX_PS_ENABLED, &(ring)->state)
225#define set_ring_ps_enabled(ring) \
226	set_bit(__I40E_RX_PS_ENABLED, &(ring)->state)
227#define clear_ring_ps_enabled(ring) \
228	clear_bit(__I40E_RX_PS_ENABLED, &(ring)->state)
229#define ring_is_16byte_desc_enabled(ring) \
230	test_bit(__I40E_RX_16BYTE_DESC_ENABLED, &(ring)->state)
231#define set_ring_16byte_desc_enabled(ring) \
232	set_bit(__I40E_RX_16BYTE_DESC_ENABLED, &(ring)->state)
233#define clear_ring_16byte_desc_enabled(ring) \
234	clear_bit(__I40E_RX_16BYTE_DESC_ENABLED, &(ring)->state)
235
236/* struct that defines a descriptor ring, associated with a VSI */
237struct i40e_ring {
238	struct i40e_ring *next;		/* pointer to next ring in q_vector */
239	void *desc;			/* Descriptor ring memory */
240	struct device *dev;		/* Used for DMA mapping */
241	struct net_device *netdev;	/* netdev ring maps to */
 
242	union {
243		struct i40e_tx_buffer *tx_bi;
244		struct i40e_rx_buffer *rx_bi;
 
245	};
246	unsigned long state;
247	u16 queue_index;		/* Queue number of ring */
248	u8 dcb_tc;			/* Traffic class of ring */
249	u8 __iomem *tail;
250
251	/* high bit set means dynamic, use accessor routines to read/write.
252	 * hardware only supports 2us resolution for the ITR registers.
253	 * these values always store the USER setting, and must be converted
254	 * before programming to a register.
255	 */
256	u16 rx_itr_setting;
257	u16 tx_itr_setting;
258
259	u16 count;			/* Number of descriptors */
260	u16 reg_idx;			/* HW register index of the ring */
261	u16 rx_hdr_len;
262	u16 rx_buf_len;
263	u8  dtype;
264#define I40E_RX_DTYPE_NO_SPLIT      0
265#define I40E_RX_DTYPE_HEADER_SPLIT  1
266#define I40E_RX_DTYPE_SPLIT_ALWAYS  2
267#define I40E_RX_SPLIT_L2      0x1
268#define I40E_RX_SPLIT_IP      0x2
269#define I40E_RX_SPLIT_TCP_UDP 0x4
270#define I40E_RX_SPLIT_SCTP    0x8
271
272	/* used in interrupt processing */
273	u16 next_to_use;
274	u16 next_to_clean;
 
275
276	u8 atr_sample_rate;
277	u8 atr_count;
278
279	unsigned long last_rx_timestamp;
280
281	bool ring_active;		/* is ring online or not */
282	bool arm_wb;		/* do something to arm write back */
283	u8 packet_stride;
284
285	u16 flags;
286#define I40E_TXR_FLAGS_WB_ON_ITR	BIT(0)
287#define I40E_TXR_FLAGS_LAST_XMIT_MORE_SET BIT(2)
 
288
289	/* stats structs */
290	struct i40e_queue_stats	stats;
291	struct u64_stats_sync syncp;
292	union {
293		struct i40e_tx_queue_stats tx_stats;
294		struct i40e_rx_queue_stats rx_stats;
295	};
296
297	unsigned int size;		/* length of descriptor ring in bytes */
298	dma_addr_t dma;			/* physical address of ring */
299
300	struct i40e_vsi *vsi;		/* Backreference to associated VSI */
301	struct i40e_q_vector *q_vector;	/* Backreference to associated vector */
302
303	struct rcu_head rcu;		/* to avoid race on free */
 
 
 
 
 
 
 
 
 
 
 
 
 
304} ____cacheline_internodealigned_in_smp;
305
306enum i40e_latency_range {
307	I40E_LOWEST_LATENCY = 0,
308	I40E_LOW_LATENCY = 1,
309	I40E_BULK_LATENCY = 2,
310	I40E_ULTRA_LATENCY = 3,
311};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
312
313struct i40e_ring_container {
314	/* array of pointers to rings */
315	struct i40e_ring *ring;
316	unsigned int total_bytes;	/* total bytes processed this int */
317	unsigned int total_packets;	/* total packets processed this int */
318	u16 count;
319	enum i40e_latency_range latency_range;
320	u16 itr;
321};
322
323/* iterator for handling rings in ring container */
324#define i40e_for_each_ring(pos, head) \
325	for (pos = (head).ring; pos != NULL; pos = pos->next)
326
327bool i40e_alloc_rx_buffers_ps(struct i40e_ring *rxr, u16 cleaned_count);
328bool i40e_alloc_rx_buffers_1buf(struct i40e_ring *rxr, u16 cleaned_count);
329void i40e_alloc_rx_headers(struct i40e_ring *rxr);
 
 
 
 
 
 
 
 
 
330netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
331void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
332void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
333int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
334int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
335void i40e_free_tx_resources(struct i40e_ring *tx_ring);
336void i40e_free_rx_resources(struct i40e_ring *rx_ring);
337int i40e_napi_poll(struct napi_struct *napi, int budget);
338#ifdef I40E_FCOE
339void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
340		 struct i40e_tx_buffer *first, u32 tx_flags,
341		 const u8 hdr_len, u32 td_cmd, u32 td_offset);
342int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
343			       struct i40e_ring *tx_ring, u32 *flags);
344#endif
345void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
346u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
 
347int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
348bool __i40e_chk_linearize(struct sk_buff *skb);
 
 
 
349
350/**
351 * i40e_get_head - Retrieve head from head writeback
352 * @tx_ring:  tx ring to fetch head of
353 *
354 * Returns value of Tx ring head based on value stored
355 * in head write-back location
356 **/
357static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
358{
359	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
360
361	return le32_to_cpu(*(volatile __le32 *)head);
362}
363
364/**
365 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
366 * @skb:     send buffer
367 * @tx_ring: ring to send buffer on
368 *
369 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
370 * there is not enough descriptors available in this ring since we need at least
371 * one descriptor.
372 **/
373static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
374{
375	const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
376	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
377	int count = 0, size = skb_headlen(skb);
378
379	for (;;) {
380		count += TXD_USE_COUNT(size);
381
382		if (!nr_frags--)
383			break;
384
385		size = skb_frag_size(frag++);
386	}
387
388	return count;
389}
390
391/**
392 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
393 * @tx_ring: the ring to be checked
394 * @size:    the size buffer we want to assure is available
395 *
396 * Returns 0 if stop is not needed
397 **/
398static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
399{
400	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
401		return 0;
402	return __i40e_maybe_stop_tx(tx_ring, size);
403}
404
405/**
406 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
407 * @skb:      send buffer
408 * @count:    number of buffers used
409 *
410 * Note: Our HW can't scatter-gather more than 8 fragments to build
411 * a packet on the wire and so we need to figure out the cases where we
412 * need to linearize the skb.
413 **/
414static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
415{
416	/* Both TSO and single send will work if count is less than 8 */
417	if (likely(count < I40E_MAX_BUFFER_TXD))
418		return false;
419
420	if (skb_is_gso(skb))
421		return __i40e_chk_linearize(skb);
422
423	/* we can support up to 8 data buffers for a single send */
424	return count != I40E_MAX_BUFFER_TXD;
 
 
 
 
 
 
 
 
 
425}
426#endif /* _I40E_TXRX_H_ */