xref: /dpdk/drivers/net/ena/ena_ethdev.h (revision 97b914f4e715565d53d38ac6e04815b9be5e58a9)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright (c) 2015-2020 Amazon.com, Inc. or its affiliates.
3  * All rights reserved.
4  */
5 
6 #ifndef _ENA_ETHDEV_H_
7 #define _ENA_ETHDEV_H_
8 
9 #include <rte_atomic.h>
10 #include <rte_ether.h>
11 #include <ethdev_driver.h>
12 #include <ethdev_pci.h>
13 #include <rte_cycles.h>
14 #include <rte_pci.h>
15 #include <rte_bus_pci.h>
16 #include <rte_timer.h>
17 #include <rte_dev.h>
18 #include <rte_net.h>
19 
20 #include "ena_com.h"
21 
22 #define ENA_REGS_BAR	0
23 #define ENA_MEM_BAR	2
24 
25 #define ENA_MAX_NUM_QUEUES	128
26 #define ENA_MIN_FRAME_LEN	64
27 #define ENA_NAME_MAX_LEN	20
28 #define ENA_PKT_MAX_BUFS	17
29 #define ENA_RX_BUF_MIN_SIZE	1400
30 #define ENA_DEFAULT_RING_SIZE	1024
31 
32 #define ENA_RX_RSS_TABLE_LOG_SIZE	7
33 #define ENA_RX_RSS_TABLE_SIZE		(1 << ENA_RX_RSS_TABLE_LOG_SIZE)
34 
35 #define ENA_MIN_MTU		128
36 
37 #define ENA_MMIO_DISABLE_REG_READ	BIT(0)
38 
39 #define ENA_WD_TIMEOUT_SEC	3
40 #define ENA_DEVICE_KALIVE_TIMEOUT (ENA_WD_TIMEOUT_SEC * rte_get_timer_hz())
41 
42 #define ENA_TX_TIMEOUT			(5 * rte_get_timer_hz())
43 #define ENA_MAX_TX_TIMEOUT_SECONDS	60
44 #define ENA_MONITORED_TX_QUEUES		3
45 #define ENA_DEFAULT_MISSING_COMP	256U
46 
47 /* While processing submitted and completed descriptors (rx and tx path
48  * respectively) in a loop it is desired to:
49  *  - perform batch submissions while populating submission queue
50  *  - avoid blocking transmission of other packets during cleanup phase
51  * Hence the utilization ratio of 1/8 of a queue size or max value if the size
52  * of the ring is very big - like 8k Rx rings.
53  */
54 #define ENA_REFILL_THRESH_DIVIDER      8
55 #define ENA_REFILL_THRESH_PACKET       256
56 
57 #define ENA_IDX_NEXT_MASKED(idx, mask) (((idx) + 1) & (mask))
58 #define ENA_IDX_ADD_MASKED(idx, n, mask) (((idx) + (n)) & (mask))
59 
60 #define ENA_RX_RSS_TABLE_LOG_SIZE	7
61 #define ENA_RX_RSS_TABLE_SIZE		(1 << ENA_RX_RSS_TABLE_LOG_SIZE)
62 
63 #define ENA_HASH_KEY_SIZE		40
64 
65 #define ENA_ALL_RSS_HF (RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_NONFRAG_IPV4_UDP | \
66 			RTE_ETH_RSS_NONFRAG_IPV6_TCP | RTE_ETH_RSS_NONFRAG_IPV6_UDP)
67 
68 #define ENA_IO_TXQ_IDX(q)		(2 * (q))
69 #define ENA_IO_RXQ_IDX(q)		(2 * (q) + 1)
70 /* Reversed version of ENA_IO_RXQ_IDX */
71 #define ENA_IO_RXQ_IDX_REV(q)		(((q) - 1) / 2)
72 
73 extern struct ena_shared_data *ena_shared_data;
74 
75 struct ena_adapter;
76 
77 enum ena_ring_type {
78 	ENA_RING_TYPE_RX = 1,
79 	ENA_RING_TYPE_TX = 2,
80 };
81 
82 struct ena_tx_buffer {
83 	struct rte_mbuf *mbuf;
84 	unsigned int tx_descs;
85 	unsigned int num_of_bufs;
86 	uint64_t timestamp;
87 	bool print_once;
88 	struct ena_com_buf bufs[ENA_PKT_MAX_BUFS];
89 };
90 
91 /* Rx buffer holds only pointer to the mbuf - may be expanded in the future */
92 struct ena_rx_buffer {
93 	struct rte_mbuf *mbuf;
94 	struct ena_com_buf ena_buf;
95 };
96 
97 struct ena_calc_queue_size_ctx {
98 	struct ena_com_dev_get_features_ctx *get_feat_ctx;
99 	struct ena_com_dev *ena_dev;
100 	u32 max_rx_queue_size;
101 	u32 max_tx_queue_size;
102 	u16 max_tx_sgl_size;
103 	u16 max_rx_sgl_size;
104 };
105 
106 struct ena_stats_tx {
107 	u64 cnt;
108 	u64 bytes;
109 	u64 prepare_ctx_err;
110 	u64 tx_poll;
111 	u64 doorbells;
112 	u64 bad_req_id;
113 	u64 available_desc;
114 	u64 missed_tx;
115 };
116 
117 struct ena_stats_rx {
118 	u64 cnt;
119 	u64 bytes;
120 	u64 refill_partial;
121 	u64 l3_csum_bad;
122 	u64 l4_csum_bad;
123 	u64 l4_csum_good;
124 	u64 mbuf_alloc_fail;
125 	u64 bad_desc_num;
126 	u64 bad_req_id;
127 };
128 
129 struct ena_ring {
130 	u16 next_to_use;
131 	u16 next_to_clean;
132 	uint64_t last_cleanup_ticks;
133 
134 	enum ena_ring_type type;
135 	enum ena_admin_placement_policy_type tx_mem_queue_type;
136 
137 	/* Indicate there are Tx packets pushed to the device and wait for db */
138 	bool pkts_without_db;
139 
140 	/* Holds the empty requests for TX/RX OOO completions */
141 	union {
142 		uint16_t *empty_tx_reqs;
143 		uint16_t *empty_rx_reqs;
144 	};
145 
146 	union {
147 		struct ena_tx_buffer *tx_buffer_info; /* contex of tx packet */
148 		struct ena_rx_buffer *rx_buffer_info; /* contex of rx packet */
149 	};
150 	struct rte_mbuf **rx_refill_buffer;
151 	unsigned int ring_size; /* number of tx/rx_buffer_info's entries */
152 	unsigned int size_mask;
153 
154 	struct ena_com_io_cq *ena_com_io_cq;
155 	struct ena_com_io_sq *ena_com_io_sq;
156 
157 	union {
158 		uint16_t tx_free_thresh;
159 		uint16_t rx_free_thresh;
160 	};
161 
162 	struct ena_com_rx_buf_info ena_bufs[ENA_PKT_MAX_BUFS]
163 						__rte_cache_aligned;
164 
165 	struct rte_mempool *mb_pool;
166 	unsigned int port_id;
167 	unsigned int id;
168 	/* Max length PMD can push to device for LLQ */
169 	uint8_t tx_max_header_size;
170 	int configured;
171 
172 	uint8_t *push_buf_intermediate_buf;
173 
174 	struct ena_adapter *adapter;
175 	uint64_t offloads;
176 	u16 sgl_size;
177 
178 	bool disable_meta_caching;
179 
180 	union {
181 		struct ena_stats_rx rx_stats;
182 		struct ena_stats_tx tx_stats;
183 	};
184 
185 	unsigned int numa_socket_id;
186 
187 	uint32_t missing_tx_completion_threshold;
188 } __rte_cache_aligned;
189 
190 enum ena_adapter_state {
191 	ENA_ADAPTER_STATE_FREE    = 0,
192 	ENA_ADAPTER_STATE_INIT    = 1,
193 	ENA_ADAPTER_STATE_RUNNING = 2,
194 	ENA_ADAPTER_STATE_STOPPED = 3,
195 	ENA_ADAPTER_STATE_CONFIG  = 4,
196 	ENA_ADAPTER_STATE_CLOSED  = 5,
197 };
198 
199 struct ena_driver_stats {
200 	rte_atomic64_t ierrors;
201 	rte_atomic64_t oerrors;
202 	rte_atomic64_t rx_nombuf;
203 	u64 rx_drops;
204 };
205 
206 struct ena_stats_dev {
207 	u64 wd_expired;
208 	u64 dev_start;
209 	u64 dev_stop;
210 	/*
211 	 * Tx drops cannot be reported as the driver statistic, because DPDK
212 	 * rte_eth_stats structure isn't providing appropriate field for that.
213 	 * As a workaround it is being published as an extended statistic.
214 	 */
215 	u64 tx_drops;
216 };
217 
218 struct ena_stats_eni {
219 	/*
220 	 * The number of packets shaped due to inbound aggregate BW
221 	 * allowance being exceeded
222 	 */
223 	uint64_t bw_in_allowance_exceeded;
224 	/*
225 	 * The number of packets shaped due to outbound aggregate BW
226 	 * allowance being exceeded
227 	 */
228 	uint64_t bw_out_allowance_exceeded;
229 	/* The number of packets shaped due to PPS allowance being exceeded */
230 	uint64_t pps_allowance_exceeded;
231 	/*
232 	 * The number of packets shaped due to connection tracking
233 	 * allowance being exceeded and leading to failure in establishment
234 	 * of new connections
235 	 */
236 	uint64_t conntrack_allowance_exceeded;
237 	/*
238 	 * The number of packets shaped due to linklocal packet rate
239 	 * allowance being exceeded
240 	 */
241 	uint64_t linklocal_allowance_exceeded;
242 };
243 
244 struct ena_offloads {
245 	uint32_t tx_offloads;
246 	uint32_t rx_offloads;
247 };
248 
249 /* board specific private data structure */
250 struct ena_adapter {
251 	/* OS defined structs */
252 	struct rte_eth_dev_data *edev_data;
253 
254 	struct ena_com_dev ena_dev __rte_cache_aligned;
255 
256 	/* TX */
257 	struct ena_ring tx_ring[ENA_MAX_NUM_QUEUES] __rte_cache_aligned;
258 	u32 max_tx_ring_size;
259 	u16 max_tx_sgl_size;
260 
261 	/* RX */
262 	struct ena_ring rx_ring[ENA_MAX_NUM_QUEUES] __rte_cache_aligned;
263 	u32 max_rx_ring_size;
264 	u16 max_rx_sgl_size;
265 
266 	u32 max_num_io_queues;
267 	u16 max_mtu;
268 	struct ena_offloads offloads;
269 
270 	/* The admin queue isn't protected by the lock and is used to
271 	 * retrieve statistics from the device. As there is no guarantee that
272 	 * application won't try to get statistics from multiple threads, it is
273 	 * safer to lock the queue to avoid admin queue failure.
274 	 */
275 	rte_spinlock_t admin_lock;
276 
277 	int id_number;
278 	char name[ENA_NAME_MAX_LEN];
279 	u8 mac_addr[RTE_ETHER_ADDR_LEN];
280 
281 	void *regs;
282 	void *dev_mem_base;
283 
284 	struct ena_driver_stats *drv_stats;
285 	enum ena_adapter_state state;
286 
287 	bool link_status;
288 
289 	enum ena_regs_reset_reason_types reset_reason;
290 
291 	struct rte_timer timer_wd;
292 	uint64_t timestamp_wd;
293 	uint64_t keep_alive_timeout;
294 
295 	struct ena_stats_dev dev_stats;
296 	struct ena_stats_eni eni_stats;
297 	struct ena_admin_basic_stats basic_stats;
298 
299 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE];
300 
301 	uint32_t all_aenq_groups;
302 	uint32_t active_aenq_groups;
303 
304 	bool trigger_reset;
305 
306 	bool use_large_llq_hdr;
307 
308 	uint32_t last_tx_comp_qid;
309 	uint64_t missing_tx_completion_to;
310 	uint64_t missing_tx_completion_budget;
311 	uint64_t tx_cleanup_stall_delay;
312 
313 	uint64_t memzone_cnt;
314 };
315 
316 int ena_mp_indirect_table_set(struct ena_adapter *adapter);
317 int ena_mp_indirect_table_get(struct ena_adapter *adapter,
318 			      uint32_t *indirect_table);
319 int ena_rss_reta_update(struct rte_eth_dev *dev,
320 			struct rte_eth_rss_reta_entry64 *reta_conf,
321 			uint16_t reta_size);
322 int ena_rss_reta_query(struct rte_eth_dev *dev,
323 		       struct rte_eth_rss_reta_entry64 *reta_conf,
324 		       uint16_t reta_size);
325 int ena_rss_hash_update(struct rte_eth_dev *dev,
326 			struct rte_eth_rss_conf *rss_conf);
327 int ena_rss_hash_conf_get(struct rte_eth_dev *dev,
328 			  struct rte_eth_rss_conf *rss_conf);
329 int ena_rss_configure(struct ena_adapter *adapter);
330 
331 #endif /* _ENA_ETHDEV_H_ */
332