FD.io VPP  v18.07-34-g55fbdb9
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  _ ("net_qede", QEDE)
89 
90 typedef enum
91 {
93 #define _(s,f) VNET_DPDK_PMD_##f,
95 #undef _
96  VNET_DPDK_PMD_UNKNOWN, /* must be last */
97 } dpdk_pmd_t;
98 
99 typedef enum
100 {
120 
121 typedef uint16_t dpdk_portid_t;
122 
123 typedef struct
124 {
125  /* Required for vec_validate_aligned */
126  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
127 
128  struct rte_ring *swq;
129 
139  u32 hqos_tc_table[64];
141 
142 typedef struct
143 {
144  /* Required for vec_validate_aligned */
145  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
146  struct rte_ring **swq;
147  struct rte_mbuf **pkts_enq;
148  struct rte_mbuf **pkts_deq;
149  struct rte_sched_port *hqos;
156 
157 #define foreach_dpdk_device_flags \
158  _( 0, ADMIN_UP, "admin-up") \
159  _( 1, PROMISC, "promisc") \
160  _( 2, PMD, "pmd") \
161  _( 3, PMD_INIT_FAIL, "pmd-init-fail") \
162  _( 4, MAYBE_MULTISEG, "maybe-multiseg") \
163  _( 5, HAVE_SUBIF, "subif") \
164  _( 6, HQOS, "hqos") \
165  _( 7, BOND_SLAVE, "bond-slave") \
166  _( 8, BOND_SLAVE_UP, "bond-slave-up") \
167  _( 9, TX_OFFLOAD, "tx-offload") \
168  _(10, INTEL_PHDR_CKSUM, "intel-phdr-cksum") \
169  _(11, RX_FLOW_OFFLOAD, "rx-flow-offload")
170 
171 enum
172 {
173 #define _(a, b, c) DPDK_DEVICE_FLAG_##b = (1 << a),
175 #undef _
176 };
177 
178 typedef struct
179 {
182  struct rte_flow *handle;
184 
185 typedef struct
186 {
191 
192 typedef struct
193 {
194  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
195  volatile u32 **lockp;
196 
197  /* Instance ID to access internal device array. */
199 
200  /* DPDK device port number */
202 
205 
206  /* next node index if we decide to steal the rx graph arc */
208 
209  dpdk_pmd_t pmd:8;
211 
213 
215  CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
216 
218 
219  /* number of sub-interfaces */
221 
222  /* PMD related */
228  struct rte_eth_conf port_conf;
229  struct rte_eth_txconf tx_conf;
230 
231  /* flow related */
235  u32 *parked_lookup_indexes; /* vector */
237  struct rte_flow_error last_flow_error;
238 
239  /* HQoS related */
242 
243  /* af_packet or BondEthernet instance number */
246 
247  /* Bonded interface port# of a slave -
248  only valid if DPDK_DEVICE_FLAG_BOND_SLAVE bit is set */
250 
251  struct rte_eth_link link;
253 
254  struct rte_eth_stats stats;
255  struct rte_eth_stats last_stats;
256  struct rte_eth_stats last_cleared_stats;
257  struct rte_eth_xstat *xstats;
258  struct rte_eth_xstat *last_cleared_xstats;
261 
262  /* mac address */
264 
265  /* error string */
267 } dpdk_device_t;
268 
269 #define DPDK_STATS_POLL_INTERVAL (10.0)
270 #define DPDK_MIN_STATS_POLL_INTERVAL (0.001) /* 1msec */
271 
272 #define DPDK_LINK_POLL_INTERVAL (3.0)
273 #define DPDK_MIN_LINK_POLL_INTERVAL (0.001) /* 1msec */
274 
275 typedef struct
276 {
280 
281 #ifndef DPDK_HQOS_DBG_BYPASS
282 #define DPDK_HQOS_DBG_BYPASS 0
283 #endif
284 
285 #ifndef HQOS_FLUSH_COUNT_THRESHOLD
286 #define HQOS_FLUSH_COUNT_THRESHOLD 100000
287 #endif
288 
290 {
293 
297 
305 
306  struct rte_sched_port_params port;
307  struct rte_sched_subport_params *subport;
308  struct rte_sched_pipe_params *pipe;
309  uint32_t *pipe_map;
311 
312 int dpdk_hqos_validate_mask (u64 mask, u32 n);
314  hqos, u32 pipe_profile_id);
319  struct rte_mbuf **pkts, u32 n_pkts);
320 
321 #define foreach_dpdk_device_config_item \
322  _ (num_rx_queues) \
323  _ (num_tx_queues) \
324  _ (num_rx_desc) \
325  _ (num_tx_desc) \
326  _ (rss_fn)
327 
328 typedef struct
329 {
330  vlib_pci_addr_t pci_addr;
333 #define DPDK_DEVICE_VLAN_STRIP_DEFAULT 0
334 #define DPDK_DEVICE_VLAN_STRIP_OFF 1
335 #define DPDK_DEVICE_VLAN_STRIP_ON 2
336 
337 #define _(x) uword x;
339 #undef _
344 
345 typedef struct
346 {
347 
348  /* Config stuff */
355 
356  /* Required config parameters */
362 
363  /*
364  * format interface names ala xxxEthernet%d/%d/%d instead of
365  * xxxEthernet%x/%x/%x.
366  */
368 
369  /* per-device config */
373 
375 
377 
378 #define DPDK_RX_BURST_SZ VLIB_FRAME_SIZE
379 
380 typedef struct
381 {
382  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
383  struct rte_mbuf *mbufs[DPDK_RX_BURST_SZ];
390 
391 typedef struct
392 {
398 } dpdk_pcap_t;
399 
400 typedef struct
401 {
402 
403  /* Devices */
407 
408  /* per-thread recycle lists */
410 
411  /* buffer flags template, configurable to enable/disable tcp / udp cksum */
413 
414  /* pcap tracing */
416 
422 
423  /*
424  * flag indicating that a posted admin up/down
425  * (via post_sw_interface_set_flags) is in progress
426  */
428 
429  /* which cpus are running I/O TX */
432 
433  /* control interval of dpdk link state and stat polling */
436 
437  /* convenience */
441 
442  /* mempool */
443  struct rte_mempool **pktmbuf_pools;
444 
445  /* API message ID base */
447 
448  /* logging */
450 } dpdk_main_t;
451 
452 extern dpdk_main_t dpdk_main;
453 
454 typedef struct
455 {
459  struct rte_mbuf mb;
460  /* Copy of VLIB buffer; packet data stored in pre_data. */
462  u8 data[256]; /* First 256 data bytes, used for hexdump */
464 
465 typedef struct
466 {
470  struct rte_mbuf mb;
471  vlib_buffer_t buffer; /* Copy of VLIB buffer; pkt data stored in pre_data. */
472  u8 data[256]; /* First 256 data bytes, used for hexdump */
474 
475 void dpdk_device_setup (dpdk_device_t * xd);
476 void dpdk_device_start (dpdk_device_t * xd);
477 void dpdk_device_stop (dpdk_device_t * xd);
478 
480  enum rte_eth_event_type type,
481  void *param, void *ret_param);
482 
483 #define foreach_dpdk_error \
484  _(NONE, "no error") \
485  _(RX_PACKET_ERROR, "Rx packet errors") \
486  _(RX_BAD_FCS, "Rx bad fcs") \
487  _(IP_CHECKSUM_ERROR, "Rx ip checksum errors") \
488  _(RX_ALLOC_FAIL, "rx buf alloc from free list failed") \
489  _(RX_ALLOC_NO_PHYSMEM, "rx buf alloc failed no physmem") \
490  _(RX_ALLOC_DROP_PKTS, "rx packets dropped due to alloc error")
491 
492 typedef enum
493 {
494 #define _(f,s) DPDK_ERROR_##f,
496 #undef _
498 } dpdk_error_t;
499 
500 #define dpdk_log_err(...) \
501  vlib_log(VLIB_LOG_LEVEL_ERR, dpdk_main.log_default, __VA_ARGS__)
502 #define dpdk_log_warn(...) \
503  vlib_log(VLIB_LOG_LEVEL_WARNING, dpdk_main.log_default, __VA_ARGS__)
504 #define dpdk_log_notice(...) \
505  vlib_log(VLIB_LOG_LEVEL_NOTICE, dpdk_main.log_default, __VA_ARGS__)
506 #define dpdk_log_info(...) \
507  vlib_log(VLIB_LOG_LEVEL_INFO, dpdk_main.log_default, __VA_ARGS__)
508 
509 void dpdk_update_link_state (dpdk_device_t * xd, f64 now);
510 
521 
525 
526 uword
529 
531  u32 elt_size, u32 num_elts,
532  u32 pool_priv_size, u16 cache_size, u8 numa,
533  struct rte_mempool **_mp,
535 
536 clib_error_t *dpdk_buffer_pool_create (vlib_main_t * vm, unsigned num_mbufs,
537  unsigned socket_id);
538 
539 #if CLI_DEBUG
540 int dpdk_buffer_validate_trajectory_all (u32 * uninitialized);
541 void dpdk_buffer_poison_trajectory_all (void);
542 #endif
543 
544 #endif /* __included_dpdk_h__ */
545 
546 /*
547  * fd.io coding-style-patch-verification: ON
548  *
549  * Local Variables:
550  * eval: (c-set-style "gnu")
551  * End:
552  */
#define foreach_dpdk_device_flags
Definition: dpdk.h:157
u8 * default_mac_address
Definition: dpdk.h:263
format_function_t format_dpdk_tx_trace
Definition: dpdk.h:514
f64 time_last_link_update
Definition: dpdk.h:252
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:232
format_function_t format_dpdk_flow
Definition: dpdk.h:518
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:350
u8 interface_name_format_decimal
Definition: dpdk.h:367
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:204
#define foreach_dpdk_error
Definition: dpdk.h:483
dpdk_error_t
Definition: dpdk.h:492
PCAP utility definitions.
u16 flags
Definition: dpdk.h:212
dpdk_device_and_queue_t ** devices_by_hqos_cpu
Definition: dpdk.h:405
clib_error_t * errors
Definition: dpdk.h:266
struct rte_sched_port_params port
Definition: dpdk.h:306
u32 per_interface_next_index
Definition: dpdk.h:207
u8 enable_tcp_udp_checksum
Definition: dpdk.h:353
struct rte_mbuf ** pkts_enq
Definition: dpdk.h:147
struct rte_ring ** swq
Definition: dpdk.h:146
struct rte_sched_port * hqos
Definition: dpdk.h:149
u32 supported_flow_actions
Definition: dpdk.h:232
u8 *( format_function_t)(u8 *s, va_list *args)
Definition: format.h:48
u16 bond_instance_num
Definition: dpdk.h:245
struct rte_flow * handle
Definition: dpdk.h:182
struct rte_eth_xstat * last_cleared_xstats
Definition: dpdk.h:258
format_function_t format_dpdk_rte_mbuf
Definition: dpdk.h:516
dpdk_device_config_hqos_t hqos
Definition: dpdk.h:342
u16 num_subifs
Definition: dpdk.h:220
struct _vnet_device_class vnet_device_class_t
u8 * pcap_filename
Definition: dpdk.h:419
unsigned char u8
Definition: types.h:56
double f64
Definition: types.h:142
struct rte_sched_pipe_params * pipe
Definition: dpdk.h:308
foreach_dpdk_device_config_item clib_bitmap_t * workers
Definition: dpdk.h:340
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:201
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:148
dpdk_device_config_t default_devconf
Definition: dpdk.h:370
u32 buffer_index
Definition: dpdk.h:456
f64 stat_poll_interval
Definition: dpdk.h:435
u32 pcap_pkts_to_capture
Definition: dpdk.h:421
u16 rx_q_used
Definition: dpdk.h:224
unsigned int u32
Definition: types.h:88
i16 buffer_advance
Definition: dpdk.h:189
format_function_t format_dpdk_rx_rte_mbuf
Definition: dpdk.h:517
clib_error_t * unformat_rss_fn(unformat_input_t *input, uword *rss_fn)
Definition: format.c:835
u32 parked_loop_count
Definition: dpdk.h:236
dpdk_device_hqos_per_worker_thread_t * hqos_wt
Definition: dpdk.h:240
#define DPDK_RX_BURST_SZ
Definition: dpdk.h:378
f64 time_last_stats_update
Definition: dpdk.h:259
u32 pcap_sw_if_index
Definition: dpdk.h:420
vlib_pci_addr_t pci_addr
Definition: dpdk.h:330
dpdk_portid_t bond_port
Definition: dpdk.h:249
u32 pcap_sw_if_index
Definition: dpdk.h:396
u8 ** eal_init_args
Definition: dpdk.h:349
dpdk_per_thread_data_t * per_thread_data
Definition: dpdk.h:406
void dpdk_device_stop(dpdk_device_t *xd)
Definition: common.c:190
pcap_main_t pcap_main
Definition: dpdk.h:418
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:388
dpdk_port_type_t port_type
Definition: dpdk.h:260
signed char i8
Definition: types.h:45
u16 tx_q_used
Definition: dpdk.h:223
u16 nb_rx_desc
Definition: dpdk.h:225
uint16_t dpdk_portid_t
Definition: dpdk.h:121
u32 hw_if_index
Definition: dpdk.h:203
u32 flags
Definition: vhost_user.h:110
u32 ** recycle
Definition: dpdk.h:409
u16 af_packet_instance_num
Definition: dpdk.h:244
Definition: dpdk.h:178
dpdk_device_t * devices
Definition: dpdk.h:404
vlib_main_t * vm
Definition: buffer.c:294
dpdk_flow_lookup_entry_t * flow_lookup_entries
Definition: dpdk.h:234
u8 nchannels_set_manually
Definition: dpdk.h:358
u32 * parked_lookup_indexes
Definition: dpdk.h:235
u16 * cpu_socket_id_by_queue
Definition: dpdk.h:226
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:195
dpdk_device_config_t * dev_confs
Definition: dpdk.h:371
u32 flow_index
Definition: dpdk.h:180
format_function_t format_dpdk_device_errors
Definition: dpdk.h:513
void dpdk_device_setup(dpdk_device_t *xd)
Definition: common.c:40
u8 coremask_set_manually
Definition: dpdk.h:357
dpdk_pmd_t
Definition: dpdk.h:90
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:512
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:257
u8 * interface_name_suffix
Definition: dpdk.h:217
u16 device_index
Definition: dpdk.h:468
format_function_t format_dpdk_device_name
Definition: dpdk.h:511
void dpdk_device_config_hqos_default(dpdk_device_config_hqos_t *hqos)
Definition: hqos.c:205
int hqos_cpu_count
Definition: dpdk.h:431
vlib_log_class_t log_default
Definition: dpdk.h:449
struct rte_mempool ** pktmbuf_pools
Definition: dpdk.h:443
dpdk_device_hqos_per_hqos_thread_t * hqos_ht
Definition: dpdk.h:241
vlib_buffer_t buffer
Definition: dpdk.h:471
void dpdk_update_link_state(dpdk_device_t *xd, f64 now)
Definition: init.c:1421
dpdk_portid_t device_index
Definition: dpdk.h:198
u32 buffer_index
Definition: dpdk.h:467
format_function_t format_dpdk_rx_trace
Definition: dpdk.h:515
int pcap_enable
Definition: dpdk.h:417
u8 * pcap_filename
Definition: dpdk.h:395
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:461
dpdk_port_type_t
Definition: dpdk.h:99
struct _vlib_node_registration vlib_node_registration_t
Definition: dpdk.h:185
void dpdk_device_start(dpdk_device_t *xd)
Definition: common.c:143
pcap_main_t pcap_main
Definition: dpdk.h:394
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:427
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:354
u64 uword
Definition: types.h:112
uint32_t * pipe_map
Definition: dpdk.h:309
u16 device_index
Definition: dpdk.h:457
u32 buffer_flags_template
Definition: dpdk.h:412
u16 next_index
Definition: dpdk.h:188
u32 pcap_pkts_to_capture
Definition: dpdk.h:397
vnet_flow_dev_ops_function_t dpdk_flow_ops_fn
Definition: dpdk.h:520
int hqos_cpu_first_index
Definition: dpdk.h:430
struct rte_sched_subport_params * subport
Definition: dpdk.h:307
#define foreach_dpdk_device_config_item
Definition: dpdk.h:321
uword clib_bitmap_t
Definition: bitmap.h:50
dpdk_flow_entry_t * flow_entries
Definition: dpdk.h:233
i8 cpu_socket
Definition: dpdk.h:210
u16 msg_id_base
Definition: dpdk.h:446
u8 * uio_driver_name
Definition: dpdk.h:351
unformat_function_t unformat_dpdk_log_level
Definition: dpdk.h:519
u8 queue_index
Definition: dpdk.h:458
vnet_main_t * vnet_main
Definition: dpdk.h:439
u16 nb_tx_desc
Definition: dpdk.h:214
clib_error_t * unformat_hqos(unformat_input_t *input, dpdk_device_config_hqos_t *hqos)
Definition: format.c:872
u8 vlib_physmem_region_index_t
Definition: physmem.h:43
uword * device_config_index_by_pci_addr
Definition: dpdk.h:372
u8 * buffer_pool_for_queue
Definition: dpdk.h:227
u32 flow_id
Definition: dpdk.h:187
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:434
u32 mark
Definition: dpdk.h:181
dpdk_config_main_t * conf
Definition: dpdk.h:440
int pcap_enable
Definition: dpdk.h:393
signed short i16
Definition: types.h:46
u16 queue_index
Definition: dpdk.h:469
vlib_main_t * vlib_main
Definition: dpdk.h:438