FD.io VPP  v18.07-rc0-415-g6c78436
Vector Packet Processing
dpdk.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2015 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 #ifndef __included_dpdk_h__
16 #define __included_dpdk_h__
17 
18 /* $$$$ We should rename always_inline -> clib_always_inline */
19 #undef always_inline
20 
21 #define ALLOW_EXPERIMENTAL_API
22 
23 #include <rte_config.h>
24 
25 #include <rte_common.h>
26 #include <rte_dev.h>
27 #include <rte_memory.h>
28 #include <rte_eal.h>
29 #include <rte_per_lcore.h>
30 #include <rte_cycles.h>
31 #include <rte_lcore.h>
32 #include <rte_per_lcore.h>
33 #include <rte_interrupts.h>
34 #include <rte_pci.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
37 #include <rte_ring.h>
38 #include <rte_mempool.h>
39 #include <rte_mbuf.h>
40 #include <rte_version.h>
41 #include <rte_eth_bond.h>
42 #include <rte_sched.h>
43 #include <rte_net.h>
44 #include <rte_bus_pci.h>
45 #include <rte_flow.h>
46 
47 #include <vnet/unix/pcap.h>
48 #include <vnet/devices/devices.h>
49 
50 #if CLIB_DEBUG > 0
51 #define always_inline static inline
52 #else
53 #define always_inline static inline __attribute__ ((__always_inline__))
54 #endif
55 
56 #include <vlib/pci/pci.h>
57 #include <vnet/flow/flow.h>
58 
59 #define NB_MBUF (16<<10)
60 
63 
64 #define foreach_dpdk_pmd \
65  _ ("net_thunderx", THUNDERX) \
66  _ ("net_e1000_em", E1000EM) \
67  _ ("net_e1000_igb", IGB) \
68  _ ("net_e1000_igb_vf", IGBVF) \
69  _ ("net_ixgbe", IXGBE) \
70  _ ("net_ixgbe_vf", IXGBEVF) \
71  _ ("net_i40e", I40E) \
72  _ ("net_i40e_vf", I40EVF) \
73  _ ("net_virtio", VIRTIO) \
74  _ ("net_enic", ENIC) \
75  _ ("net_vmxnet3", VMXNET3) \
76  _ ("AF_PACKET PMD", AF_PACKET) \
77  _ ("net_bonding", BOND) \
78  _ ("net_fm10k", FM10K) \
79  _ ("net_cxgbe", CXGBE) \
80  _ ("net_mlx4", MLX4) \
81  _ ("net_mlx5", MLX5) \
82  _ ("net_dpaa2", DPAA2) \
83  _ ("net_virtio_user", VIRTIO_USER) \
84  _ ("net_vhost", VHOST_ETHER) \
85  _ ("net_ena", ENA) \
86  _ ("net_failsafe", FAILSAFE) \
87  _ ("net_liovf", LIOVF_ETHER)
88 
89 typedef enum
90 {
92 #define _(s,f) VNET_DPDK_PMD_##f,
94 #undef _
95  VNET_DPDK_PMD_UNKNOWN, /* must be last */
96 } dpdk_pmd_t;
97 
98 typedef enum
99 {
119 
120 typedef uint16_t dpdk_portid_t;
121 
122 typedef struct
123 {
124  /* Required for vec_validate_aligned */
125  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
126 
127  struct rte_ring *swq;
128 
138  u32 hqos_tc_table[64];
140 
141 typedef struct
142 {
143  /* Required for vec_validate_aligned */
144  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
145  struct rte_ring **swq;
146  struct rte_mbuf **pkts_enq;
147  struct rte_mbuf **pkts_deq;
148  struct rte_sched_port *hqos;
155 
156 #define foreach_dpdk_device_flags \
157  _( 0, ADMIN_UP, "admin-up") \
158  _( 1, PROMISC, "promisc") \
159  _( 2, PMD, "pmd") \
160  _( 3, PMD_INIT_FAIL, "pmd-init-fail") \
161  _( 4, MAYBE_MULTISEG, "maybe-multiseg") \
162  _( 5, HAVE_SUBIF, "subif") \
163  _( 6, HQOS, "hqos") \
164  _( 7, BOND_SLAVE, "bond-slave") \
165  _( 8, BOND_SLAVE_UP, "bond-slave-up") \
166  _( 9, TX_OFFLOAD, "tx-offload") \
167  _(10, INTEL_PHDR_CKSUM, "intel-phdr-cksum") \
168  _(11, RX_FLOW_OFFLOAD, "rx-flow-offload")
169 
170 enum
171 {
172 #define _(a, b, c) DPDK_DEVICE_FLAG_##b = (1 << a),
174 #undef _
175 };
176 
177 typedef struct
178 {
181  struct rte_flow *handle;
183 
184 typedef struct
185 {
190 
191 typedef struct
192 {
193  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
194  volatile u32 **lockp;
195 
196  /* Instance ID to access internal device array. */
198 
199  /* DPDK device port number */
201 
204 
205  /* next node index if we decide to steal the rx graph arc */
207 
208  dpdk_pmd_t pmd:8;
210 
212 
214  CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
215 
217 
218  /* number of sub-interfaces */
220 
221  /* PMD related */
227  struct rte_eth_conf port_conf;
228  struct rte_eth_txconf tx_conf;
229 
230  /* flow related */
234  u32 *parked_lookup_indexes; /* vector */
236  struct rte_flow_error last_flow_error;
237 
238  /* HQoS related */
241 
242  /* af_packet or BondEthernet instance number */
245 
246  /* Bonded interface port# of a slave -
247  only valid if DPDK_DEVICE_FLAG_BOND_SLAVE bit is set */
249 
250  struct rte_eth_link link;
252 
253  struct rte_eth_stats stats;
254  struct rte_eth_stats last_stats;
255  struct rte_eth_stats last_cleared_stats;
256  struct rte_eth_xstat *xstats;
257  struct rte_eth_xstat *last_cleared_xstats;
260 
261  /* mac address */
263 
264  /* error string */
266 } dpdk_device_t;
267 
268 #define DPDK_STATS_POLL_INTERVAL (10.0)
269 #define DPDK_MIN_STATS_POLL_INTERVAL (0.001) /* 1msec */
270 
271 #define DPDK_LINK_POLL_INTERVAL (3.0)
272 #define DPDK_MIN_LINK_POLL_INTERVAL (0.001) /* 1msec */
273 
274 typedef struct
275 {
279 
280 #ifndef DPDK_HQOS_DBG_BYPASS
281 #define DPDK_HQOS_DBG_BYPASS 0
282 #endif
283 
284 #ifndef HQOS_FLUSH_COUNT_THRESHOLD
285 #define HQOS_FLUSH_COUNT_THRESHOLD 100000
286 #endif
287 
289 {
292 
296 
304 
305  struct rte_sched_port_params port;
306  struct rte_sched_subport_params *subport;
307  struct rte_sched_pipe_params *pipe;
308  uint32_t *pipe_map;
310 
311 int dpdk_hqos_validate_mask (u64 mask, u32 n);
313  hqos, u32 pipe_profile_id);
318  struct rte_mbuf **pkts, u32 n_pkts);
319 
320 #define foreach_dpdk_device_config_item \
321  _ (num_rx_queues) \
322  _ (num_tx_queues) \
323  _ (num_rx_desc) \
324  _ (num_tx_desc) \
325  _ (rss_fn)
326 
327 typedef struct
328 {
329  vlib_pci_addr_t pci_addr;
332 #define DPDK_DEVICE_VLAN_STRIP_DEFAULT 0
333 #define DPDK_DEVICE_VLAN_STRIP_OFF 1
334 #define DPDK_DEVICE_VLAN_STRIP_ON 2
335 
336 #define _(x) uword x;
338 #undef _
343 
344 typedef struct
345 {
346 
347  /* Config stuff */
354 
355  /* Required config parameters */
361 
362  /*
363  * format interface names ala xxxEthernet%d/%d/%d instead of
364  * xxxEthernet%x/%x/%x.
365  */
367 
368  /* per-device config */
372 
374 
376 
377 #define DPDK_RX_BURST_SZ VLIB_FRAME_SIZE
378 
379 typedef struct
380 {
381  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
382  struct rte_mbuf *mbufs[DPDK_RX_BURST_SZ];
389 
390 typedef struct
391 {
397 } dpdk_pcap_t;
398 
399 typedef struct
400 {
401 
402  /* Devices */
406 
407  /* per-thread recycle lists */
409 
410  /* buffer flags template, configurable to enable/disable tcp / udp cksum */
412 
413  /* pcap tracing */
415 
421 
422  /*
423  * flag indicating that a posted admin up/down
424  * (via post_sw_interface_set_flags) is in progress
425  */
427 
428  /* which cpus are running I/O TX */
431 
432  /* control interval of dpdk link state and stat polling */
435 
436  /* convenience */
440 
441  /* mempool */
442  struct rte_mempool **pktmbuf_pools;
443 
444  /* API message ID base */
446 
447  /* logging */
449 } dpdk_main_t;
450 
451 extern dpdk_main_t dpdk_main;
452 
453 typedef struct
454 {
458  struct rte_mbuf mb;
459  /* Copy of VLIB buffer; packet data stored in pre_data. */
461  u8 data[256]; /* First 256 data bytes, used for hexdump */
463 
464 typedef struct
465 {
469  struct rte_mbuf mb;
470  vlib_buffer_t buffer; /* Copy of VLIB buffer; pkt data stored in pre_data. */
471  u8 data[256]; /* First 256 data bytes, used for hexdump */
473 
474 void dpdk_device_setup (dpdk_device_t * xd);
475 void dpdk_device_start (dpdk_device_t * xd);
476 void dpdk_device_stop (dpdk_device_t * xd);
477 
479  enum rte_eth_event_type type,
480  void *param, void *ret_param);
481 
482 #define foreach_dpdk_error \
483  _(NONE, "no error") \
484  _(RX_PACKET_ERROR, "Rx packet errors") \
485  _(RX_BAD_FCS, "Rx bad fcs") \
486  _(IP_CHECKSUM_ERROR, "Rx ip checksum errors") \
487  _(RX_ALLOC_FAIL, "rx buf alloc from free list failed") \
488  _(RX_ALLOC_NO_PHYSMEM, "rx buf alloc failed no physmem") \
489  _(RX_ALLOC_DROP_PKTS, "rx packets dropped due to alloc error")
490 
491 typedef enum
492 {
493 #define _(f,s) DPDK_ERROR_##f,
495 #undef _
497 } dpdk_error_t;
498 
499 #define dpdk_log_err(...) \
500  vlib_log(VLIB_LOG_LEVEL_ERR, dpdk_main.log_default, __VA_ARGS__)
501 #define dpdk_log_warn(...) \
502  vlib_log(VLIB_LOG_LEVEL_WARNING, dpdk_main.log_default, __VA_ARGS__)
503 #define dpdk_log_notice(...) \
504  vlib_log(VLIB_LOG_LEVEL_NOTICE, dpdk_main.log_default, __VA_ARGS__)
505 #define dpdk_log_info(...) \
506  vlib_log(VLIB_LOG_LEVEL_INFO, dpdk_main.log_default, __VA_ARGS__)
507 
508 void dpdk_update_link_state (dpdk_device_t * xd, f64 now);
509 
520 
524 
525 uword
528 
530  u32 elt_size, u32 num_elts,
531  u32 pool_priv_size, u16 cache_size, u8 numa,
532  struct rte_mempool **_mp,
534 
535 clib_error_t *dpdk_buffer_pool_create (vlib_main_t * vm, unsigned num_mbufs,
536  unsigned socket_id);
537 
538 #if CLI_DEBUG
539 int dpdk_buffer_validate_trajectory_all (u32 * uninitialized);
540 void dpdk_buffer_poison_trajectory_all (void);
541 #endif
542 
543 #endif /* __included_dpdk_h__ */
544 
545 /*
546  * fd.io coding-style-patch-verification: ON
547  *
548  * Local Variables:
549  * eval: (c-set-style "gnu")
550  * End:
551  */
#define foreach_dpdk_device_flags
Definition: dpdk.h:156
u8 * default_mac_address
Definition: dpdk.h:262
format_function_t format_dpdk_tx_trace
Definition: dpdk.h:513
f64 time_last_link_update
Definition: dpdk.h:251
struct dpdk_device_config_hqos_t dpdk_device_config_hqos_t
uword( unformat_function_t)(unformat_input_t *input, va_list *args)
Definition: format.h:231
format_function_t format_dpdk_flow
Definition: dpdk.h:517
u32 vlib_log_class_t
Definition: log.h:21
#define CLIB_CACHE_LINE_ALIGN_MARK(mark)
Definition: cache.h:63
u8 * eal_init_args_str
Definition: dpdk.h:349
u8 interface_name_format_decimal
Definition: dpdk.h:366
dpdk_main_t dpdk_main
Definition: init.c:42
int dpdk_hqos_validate_mask(u64 mask, u32 n)
Definition: hqos.c:168
vnet_device_class_t dpdk_device_class
unsigned long u64
Definition: types.h:89
u32 sw_if_index
Definition: dpdk.h:203
#define foreach_dpdk_error
Definition: dpdk.h:482
dpdk_error_t
Definition: dpdk.h:491
PCAP utility definitions.
u16 flags
Definition: dpdk.h:211
dpdk_device_and_queue_t ** devices_by_hqos_cpu
Definition: dpdk.h:404
clib_error_t * errors
Definition: dpdk.h:265
struct rte_sched_port_params port
Definition: dpdk.h:305
u32 per_interface_next_index
Definition: dpdk.h:206
u8 enable_tcp_udp_checksum
Definition: dpdk.h:352
struct rte_mbuf ** pkts_enq
Definition: dpdk.h:146
struct rte_ring ** swq
Definition: dpdk.h:145
struct rte_sched_port * hqos
Definition: dpdk.h:148
u32 supported_flow_actions
Definition: dpdk.h:231
u8 *( format_function_t)(u8 *s, va_list *args)
Definition: format.h:48
u16 bond_instance_num
Definition: dpdk.h:244
struct rte_flow * handle
Definition: dpdk.h:181
struct rte_eth_xstat * last_cleared_xstats
Definition: dpdk.h:257
format_function_t format_dpdk_rte_mbuf
Definition: dpdk.h:515
dpdk_device_config_hqos_t hqos
Definition: dpdk.h:341
u16 num_subifs
Definition: dpdk.h:219
struct _vnet_device_class vnet_device_class_t
u8 * pcap_filename
Definition: dpdk.h:418
unsigned char u8
Definition: types.h:56
double f64
Definition: types.h:142
struct rte_sched_pipe_params * pipe
Definition: dpdk.h:307
foreach_dpdk_device_config_item clib_bitmap_t * workers
Definition: dpdk.h:339
PCAP main state data structure.
Definition: pcap.h:124
dpdk_config_main_t dpdk_config_main
Definition: init.c:43
dpdk_portid_t port_id
Definition: dpdk.h:200
vlib_node_registration_t dpdk_input_node
(constructor) VLIB_REGISTER_NODE (dpdk_input_node)
Definition: node.c:649
struct rte_mbuf ** pkts_deq
Definition: dpdk.h:147
dpdk_device_config_t default_devconf
Definition: dpdk.h:369
u32 buffer_index
Definition: dpdk.h:455
f64 stat_poll_interval
Definition: dpdk.h:434
u32 pcap_pkts_to_capture
Definition: dpdk.h:420
u16 rx_q_used
Definition: dpdk.h:223
unsigned int u32
Definition: types.h:88
i16 buffer_advance
Definition: dpdk.h:188
format_function_t format_dpdk_rx_rte_mbuf
Definition: dpdk.h:516
clib_error_t * unformat_rss_fn(unformat_input_t *input, uword *rss_fn)
Definition: format.c:811
u32 parked_loop_count
Definition: dpdk.h:235
dpdk_device_hqos_per_worker_thread_t * hqos_wt
Definition: dpdk.h:239
#define DPDK_RX_BURST_SZ
Definition: dpdk.h:377
f64 time_last_stats_update
Definition: dpdk.h:258
u32 pcap_sw_if_index
Definition: dpdk.h:419
vlib_pci_addr_t pci_addr
Definition: dpdk.h:329
dpdk_portid_t bond_port
Definition: dpdk.h:248
u32 pcap_sw_if_index
Definition: dpdk.h:395
u8 ** eal_init_args
Definition: dpdk.h:348
dpdk_per_thread_data_t * per_thread_data
Definition: dpdk.h:405
void dpdk_device_stop(dpdk_device_t *xd)
Definition: common.c:190
pcap_main_t pcap_main
Definition: dpdk.h:417
struct _unformat_input_t unformat_input_t
unsigned short u16
Definition: types.h:57
void dpdk_hqos_metadata_set(dpdk_device_hqos_per_worker_thread_t *hqos, struct rte_mbuf **pkts, u32 n_pkts)
Definition: hqos.c:640
#define foreach_dpdk_pmd
Definition: dpdk.h:64
vlib_buffer_t buffer_template
Definition: dpdk.h:387
dpdk_port_type_t port_type
Definition: dpdk.h:259
signed char i8
Definition: types.h:45
u16 tx_q_used
Definition: dpdk.h:222
u16 nb_rx_desc
Definition: dpdk.h:224
uint16_t dpdk_portid_t
Definition: dpdk.h:120
u32 hw_if_index
Definition: dpdk.h:202
u32 ** recycle
Definition: dpdk.h:408
u16 af_packet_instance_num
Definition: dpdk.h:243
Definition: dpdk.h:177
dpdk_device_t * devices
Definition: dpdk.h:403
vlib_main_t * vm
Definition: buffer.c:294
dpdk_flow_lookup_entry_t * flow_lookup_entries
Definition: dpdk.h:233
u8 nchannels_set_manually
Definition: dpdk.h:357
u32 * parked_lookup_indexes
Definition: dpdk.h:234
u16 * cpu_socket_id_by_queue
Definition: dpdk.h:225
void dpdk_device_config_hqos_pipe_profile_default(dpdk_device_config_hqos_t *hqos, u32 pipe_profile_id)
Definition: hqos.c:197
volatile u32 ** lockp
Definition: dpdk.h:194
dpdk_device_config_t * dev_confs
Definition: dpdk.h:370
u32 flow_index
Definition: dpdk.h:179
format_function_t format_dpdk_device_errors
Definition: dpdk.h:512
void dpdk_device_setup(dpdk_device_t *xd)
Definition: common.c:40
u8 coremask_set_manually
Definition: dpdk.h:356
dpdk_pmd_t
Definition: dpdk.h:89
uword admin_up_down_process(vlib_main_t *vm, vlib_node_runtime_t *rt, vlib_frame_t *f)
Definition: device.c:700
format_function_t format_dpdk_device
Definition: dpdk.h:511
int dpdk_port_state_callback(dpdk_portid_t port_id, enum rte_eth_event_type type, void *param, void *ret_param)
Definition: common.c:319
struct rte_eth_xstat * xstats
Definition: dpdk.h:256
u8 * interface_name_suffix
Definition: dpdk.h:216
u16 device_index
Definition: dpdk.h:467
format_function_t format_dpdk_device_name
Definition: dpdk.h:510
void dpdk_device_config_hqos_default(dpdk_device_config_hqos_t *hqos)
Definition: hqos.c:205
int hqos_cpu_count
Definition: dpdk.h:430
vlib_log_class_t log_default
Definition: dpdk.h:448
struct rte_mempool ** pktmbuf_pools
Definition: dpdk.h:442
dpdk_device_hqos_per_hqos_thread_t * hqos_ht
Definition: dpdk.h:240
vlib_buffer_t buffer
Definition: dpdk.h:470
void dpdk_update_link_state(dpdk_device_t *xd, f64 now)
Definition: init.c:1424
dpdk_portid_t device_index
Definition: dpdk.h:197
u32 buffer_index
Definition: dpdk.h:466
format_function_t format_dpdk_rx_trace
Definition: dpdk.h:514
int pcap_enable
Definition: dpdk.h:416
u8 * pcap_filename
Definition: dpdk.h:394
clib_error_t * dpdk_port_setup_hqos(dpdk_device_t *xd, dpdk_device_config_hqos_t *hqos)
Definition: hqos.c:247
vlib_buffer_t buffer
Definition: dpdk.h:460
dpdk_port_type_t
Definition: dpdk.h:98
struct _vlib_node_registration vlib_node_registration_t
Definition: dpdk.h:184
void dpdk_device_start(dpdk_device_t *xd)
Definition: common.c:143
pcap_main_t pcap_main
Definition: dpdk.h:393
clib_error_t * dpdk_buffer_pool_create(vlib_main_t *vm, unsigned num_mbufs, unsigned socket_id)
Definition: buffer.c:542
u8 admin_up_down_in_progress
Definition: dpdk.h:426
clib_error_t * dpdk_pool_create(vlib_main_t *vm, u8 *pool_name, u32 elt_size, u32 num_elts, u32 pool_priv_size, u16 cache_size, u8 numa, struct rte_mempool **_mp, vlib_physmem_region_index_t *pri)
Definition: buffer.c:454
u8 no_tx_checksum_offload
Definition: dpdk.h:353
u64 uword
Definition: types.h:112
uint32_t * pipe_map
Definition: dpdk.h:308
u16 device_index
Definition: dpdk.h:456
u32 buffer_flags_template
Definition: dpdk.h:411
u16 next_index
Definition: dpdk.h:187
u32 pcap_pkts_to_capture
Definition: dpdk.h:396
vnet_flow_dev_ops_function_t dpdk_flow_ops_fn
Definition: dpdk.h:519
int hqos_cpu_first_index
Definition: dpdk.h:429
struct rte_sched_subport_params * subport
Definition: dpdk.h:306
#define foreach_dpdk_device_config_item
Definition: dpdk.h:320
uword clib_bitmap_t
Definition: bitmap.h:50
dpdk_flow_entry_t * flow_entries
Definition: dpdk.h:232
i8 cpu_socket
Definition: dpdk.h:209
u16 msg_id_base
Definition: dpdk.h:445
u8 * uio_driver_name
Definition: dpdk.h:350
u32 flags
Definition: vhost-user.h:77
unformat_function_t unformat_dpdk_log_level
Definition: dpdk.h:518
u8 queue_index
Definition: dpdk.h:457
vnet_main_t * vnet_main
Definition: dpdk.h:438
u16 nb_tx_desc
Definition: dpdk.h:213
clib_error_t * unformat_hqos(unformat_input_t *input, dpdk_device_config_hqos_t *hqos)
Definition: format.c:848
u8 vlib_physmem_region_index_t
Definition: physmem.h:43
uword * device_config_index_by_pci_addr
Definition: dpdk.h:371
u8 * buffer_pool_for_queue
Definition: dpdk.h:226
u32 flow_id
Definition: dpdk.h:186
int( vnet_flow_dev_ops_function_t)(struct vnet_main_t *vnm, vnet_flow_dev_op_t op, u32 hw_if_index, u32 index, uword *private_data)
Definition: interface.h:93
f64 link_state_poll_interval
Definition: dpdk.h:433
u32 mark
Definition: dpdk.h:180
dpdk_config_main_t * conf
Definition: dpdk.h:439
int pcap_enable
Definition: dpdk.h:392
signed short i16
Definition: types.h:46
u16 queue_index
Definition: dpdk.h:468
vlib_main_t * vlib_main
Definition: dpdk.h:437