dpdk

Name: dpdk
Author: mohitmishra786

141stars19forks

How to install dpdk?

mohitmishra786/low-level-dev-skills/dpdk

$npx -y skills add mohitmishra786/low-level-dev-skills --skill dpdk

Installs into the current project.

Summary

DPDK skill for userspace packet I/O. Use when initializing EAL, configuring PMD drivers, using mbuf pools and rte_ring, setting up huge pages, RSS, or testpmd validation. Activates on queries about DPDK, EAL, rte_eth_rx_burst, hugepages, PMD, or testpmd.

Files · 1

View on GitHub

SKILL.md

1	# DPDK
2
3	## Purpose
4
5	Guide agents through DPDK (Data Plane Development Kit): EAL initialization, poll-mode driver (PMD) concepts, `rte_eth_rx_burst`/`tx_burst`, mbuf mempools, `rte_ring` queues, huge page setup, RSS configuration, testpmd validation, QEMU virtio testing, and pipeline vs run-to-completion models.
6
7	## When to Use
8
9	- Building a userspace packet forwarder bypassing the kernel network stack
10	- Achieving line-rate on 10/25/100 GbE NICs
11	- Prototyping NFV/vSwitch data plane components
12	- Testing NIC configuration with testpmd before custom code
13	- Comparing DPDK throughput with kernel networking or AF_XDP
14	- Running DPDK in VMs with virtio for development
15
16	## Workflow
17
18	### 1. Huge pages setup
19
20	```bash
21	# 2MB hugepages (common)
22	echo 1024 \| sudo tee /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages
23
24	# 1GB hugepages (better TLB perf on large memory)
25	echo 4 \| sudo tee /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages
26
27	# Mount hugetlbfs
28	sudo mkdir -p /mnt/huge
29	sudo mount -t hugetlbfs nodev /mnt/huge
30
31	# Verify
32	grep Huge /proc/meminfo
33	```
34
35	DPDK EAL maps hugepages at startup — insufficient pages cause init failure.
36
37	### 2. EAL initialization
38
39	```c
40	#include <rte_eal.h>
41	#include <rte_ethdev.h>
42
43	int main(int argc, char **argv) {
44	int ret = rte_eal_init(argc, argv);
45	if (ret < 0)
46	rte_exit(EXIT_FAILURE, "EAL init failed\n");
47	// argc/argv adjusted — remaining args for app
48	return run_dataplane(argc - ret, argv + ret);
49	}
50	```
51
52	```bash
53	# Typical EAL args
54	./dpdk_app -l 0-3 -n 4 --huge-dir=/mnt/huge -- -p 0x3
55
56	# Flags:
57	# -l 0-3 — cores for DPDK (lcore mask)
58	# -n 4 — memory channels
59	# --proc-type=primary
60	# --file-prefix=myapp — multi-instance
61	```
62
63	### 3. Port configuration and PMD
64
65	```c
66	#include <rte_ethdev.h>
67
68	#define RX_RING_SIZE 1024
69	#define TX_RING_SIZE 1024
70	#define NUM_MBUFS 8191
71	#define MBUF_CACHE 250
72	#define BURST_SIZE 32
73
74	static const struct rte_eth_conf port_conf = {
75	.rxmode = { .max_lro_pkt_size = RTE_ETHER_MAX_LEN },
76	};
77
78	struct rte_mempool *mbuf_pool;
79
80	int port_init(uint16_t port) {
81	mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", NUM_MBUFS,
82	MBUF_CACHE, 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
83
84	struct rte_eth_rxconf rxq_conf = dev_info.default_rxconf;
85	ret = rte_eth_rx_queue_setup(port, 0, RX_RING_SIZE,
86	rte_eth_dev_socket_id(port), &rxq_conf, mbuf_pool);
87	// ... tx queue setup ...
88	ret = rte_eth_dev_start(port);
89	rte_eth_promiscuous_enable(port);
90	return 0;
91	}
92	```
93
94	```bash
95	# List available PMDs
96	dpdk-devbind.py --status
97
98	# Bind NIC to vfio-pci (required for physical NICs)
99	sudo modprobe vfio-pci
100	sudo dpdk-devbind.py --bind=vfio-pci 0000:03:00.0
101	```
102
103	### 4. RX/TX burst loop
104
105	```c
106	static inline void lcore_main(void) {
107	const uint16_t port = 0;
108	struct rte_mbuf *bufs[BURST_SIZE];
109
110	while (1) {
111	uint16_t nb_rx = rte_eth_rx_burst(port, 0, bufs, BURST_SIZE);
112	if (nb_rx == 0)
113	continue;
114
115	uint16_t nb_tx = rte_eth_tx_burst(port, 0, bufs, nb_rx);
116	// Free unsent mbufs
117	if (nb_tx < nb_rx) {
118	for (uint16_t i = nb_tx; i < nb_rx; i++)
119	rte_pktmbuf_free(bufs[i]);
120	}
121	}
122	}
123	```
124
125	Poll-mode: no interrupts — cores spin for packets. Assign one core per queue for scaling.
126
127	### 5. rte_ring for inter-core queues
128
129	```c
130	#include <rte_ring.h>
131
132	struct rte_ring *ring = rte_ring_create("RX_RING", 1024,
133	rte_socket_id(), RING_F_SP_ENQ \| RING_F_SC_DEQ);
134
135	// Producer (RX core)
136	rte_ring_enqueue_bulk(ring, (void **)bufs, nb_rx, NULL);
137
138	// Consumer (worker core)
139	uint16_t nb_deq = rte_ring_dequeue_bulk(ring, (void **)bufs, BURST_SIZE, NULL);
140	```
141
142	### 6. RSS configuration
143
144	```c
145	static const struct rte_eth_rss_conf rss_conf = {
146	.rss_key = NULL,
147	.rss_hf = RTE_ETH_RSS_IP \| RTE_ETH_RSS_TCP \| RTE_ETH_RSS_UDP,
148	};
149
150	struct rte_eth_conf port_conf = {
151	.rxmode = { .mq_mode = RTE_ETH_MQ_RX_RSS },
152	.rx_adv_conf = { .rss_conf = rss_conf },
153	};
154	```
155
156	Distributes flows across RX queues — map queues to cores for parallelism.
157
158	### 7. testpmd validation
159
160	```bash
161	# Build testpmd
162	cd dpdk/build
163	ninja -C build dpdk-testpmd
164
165	# Forward mode test
166	sudo ./build/app/dpdk-testpmd -l 0-3 -n 4 -- -i --forward-mode=io
167
168	testpmd> show port stats all
169	testpmd> start tx_first
170	testpmd> show port stats all
171	```
172
173	Forward modes: `io`, `mac`, `macswap`, `flowgen` for throughput testing.
174
175	### 8. QEMU virtio testing
176
177	```bash
178	qemu-system-x86_64 \
179	-cpu host -m 4096 -smp 4 \
180	-netdev user,id=net0 \
181	-device virtio-net-pci,netdev=net0,mq=on,vectors=10 \
182	-object memory-backend-file,id=mem,size=2G,mem-path=/dev/hugepages,share=on \
183	-numa node,memdev=mem \
184	...
185	```
186
187	Inside VM: bind virtio PCI to `vfio-pci` or

People who liked this also viewed

6 picks

Type

dpdk

141stars19forks

View on GitHub

How to install dpdk?

mohitmishra786/low-level-dev-skills/dpdk

$npx -y skills add mohitmishra786/low-level-dev-skills --skill dpdk

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# DPDK
2
3	## Purpose
4
5	Guide agents through DPDK (Data Plane Development Kit): EAL initialization, poll-mode driver (PMD) concepts, `rte_eth_rx_burst`/`tx_burst`, mbuf mempools, `rte_ring` queues, huge page setup, RSS configuration, testpmd validation, QEMU virtio testing, and pipeline vs run-to-completion models.
6
7	## When to Use
8
9	- Building a userspace packet forwarder bypassing the kernel network stack
10	- Achieving line-rate on 10/25/100 GbE NICs
11	- Prototyping NFV/vSwitch data plane components
12	- Testing NIC configuration with testpmd before custom code
13	- Comparing DPDK throughput with kernel networking or AF_XDP
14	- Running DPDK in VMs with virtio for development
15
16	## Workflow
17
18	### 1. Huge pages setup
19
20	```bash
21	# 2MB hugepages (common)
22	echo 1024 \| sudo tee /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages
23
24	# 1GB hugepages (better TLB perf on large memory)
25	echo 4 \| sudo tee /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages
26
27	# Mount hugetlbfs
28	sudo mkdir -p /mnt/huge
29	sudo mount -t hugetlbfs nodev /mnt/huge
30
31	# Verify
32	grep Huge /proc/meminfo
33	```
34
35	DPDK EAL maps hugepages at startup — insufficient pages cause init failure.
36
37	### 2. EAL initialization
38
39	```c
40	#include <rte_eal.h>
41	#include <rte_ethdev.h>
42
43	int main(int argc, char **argv) {
44	int ret = rte_eal_init(argc, argv);
45	if (ret < 0)
46	rte_exit(EXIT_FAILURE, "EAL init failed\n");
47	// argc/argv adjusted — remaining args for app
48	return run_dataplane(argc - ret, argv + ret);
49	}
50	```
51
52	```bash
53	# Typical EAL args
54	./dpdk_app -l 0-3 -n 4 --huge-dir=/mnt/huge -- -p 0x3
55
56	# Flags:
57	# -l 0-3 — cores for DPDK (lcore mask)
58	# -n 4 — memory channels
59	# --proc-type=primary
60	# --file-prefix=myapp — multi-instance
61	```
62
63	### 3. Port configuration and PMD
64
65	```c
66	#include <rte_ethdev.h>
67
68	#define RX_RING_SIZE 1024
69	#define TX_RING_SIZE 1024
70	#define NUM_MBUFS 8191
71	#define MBUF_CACHE 250
72	#define BURST_SIZE 32
73
74	static const struct rte_eth_conf port_conf = {
75	.rxmode = { .max_lro_pkt_size = RTE_ETHER_MAX_LEN },
76	};
77
78	struct rte_mempool *mbuf_pool;
79
80	int port_init(uint16_t port) {
81	mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", NUM_MBUFS,
82	MBUF_CACHE, 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
83
84	struct rte_eth_rxconf rxq_conf = dev_info.default_rxconf;
85	ret = rte_eth_rx_queue_setup(port, 0, RX_RING_SIZE,
86	rte_eth_dev_socket_id(port), &rxq_conf, mbuf_pool);
87	// ... tx queue setup ...
88	ret = rte_eth_dev_start(port);
89	rte_eth_promiscuous_enable(port);
90	return 0;
91	}
92	```
93
94	```bash
95	# List available PMDs
96	dpdk-devbind.py --status
97
98	# Bind NIC to vfio-pci (required for physical NICs)
99	sudo modprobe vfio-pci
100	sudo dpdk-devbind.py --bind=vfio-pci 0000:03:00.0
101	```
102
103	### 4. RX/TX burst loop
104
105	```c
106	static inline void lcore_main(void) {
107	const uint16_t port = 0;
108	struct rte_mbuf *bufs[BURST_SIZE];
109
110	while (1) {
111	uint16_t nb_rx = rte_eth_rx_burst(port, 0, bufs, BURST_SIZE);
112	if (nb_rx == 0)
113	continue;
114
115	uint16_t nb_tx = rte_eth_tx_burst(port, 0, bufs, nb_rx);
116	// Free unsent mbufs
117	if (nb_tx < nb_rx) {
118	for (uint16_t i = nb_tx; i < nb_rx; i++)
119	rte_pktmbuf_free(bufs[i]);
120	}
121	}
122	}
123	```
124
125	Poll-mode: no interrupts — cores spin for packets. Assign one core per queue for scaling.
126
127	### 5. rte_ring for inter-core queues
128
129	```c
130	#include <rte_ring.h>
131
132	struct rte_ring *ring = rte_ring_create("RX_RING", 1024,
133	rte_socket_id(), RING_F_SP_ENQ \| RING_F_SC_DEQ);
134
135	// Producer (RX core)
136	rte_ring_enqueue_bulk(ring, (void **)bufs, nb_rx, NULL);
137
138	// Consumer (worker core)
139	uint16_t nb_deq = rte_ring_dequeue_bulk(ring, (void **)bufs, BURST_SIZE, NULL);
140	```
141
142	### 6. RSS configuration
143
144	```c
145	static const struct rte_eth_rss_conf rss_conf = {
146	.rss_key = NULL,
147	.rss_hf = RTE_ETH_RSS_IP \| RTE_ETH_RSS_TCP \| RTE_ETH_RSS_UDP,
148	};
149
150	struct rte_eth_conf port_conf = {
151	.rxmode = { .mq_mode = RTE_ETH_MQ_RX_RSS },
152	.rx_adv_conf = { .rss_conf = rss_conf },
153	};
154	```
155
156	Distributes flows across RX queues — map queues to cores for parallelism.
157
158	### 7. testpmd validation
159
160	```bash
161	# Build testpmd
162	cd dpdk/build
163	ninja -C build dpdk-testpmd
164
165	# Forward mode test
166	sudo ./build/app/dpdk-testpmd -l 0-3 -n 4 -- -i --forward-mode=io
167
168	testpmd> show port stats all
169	testpmd> start tx_first
170	testpmd> show port stats all
171	```
172
173	Forward modes: `io`, `mac`, `macswap`, `flowgen` for throughput testing.
174
175	### 8. QEMU virtio testing
176
177	```bash
178	qemu-system-x86_64 \
179	-cpu host -m 4096 -smp 4 \
180	-netdev user,id=net0 \
181	-device virtio-net-pci,netdev=net0,mq=on,vectors=10 \
182	-object memory-backend-file,id=mem,size=2G,mem-path=/dev/hugepages,share=on \
183	-numa node,memdev=mem \
184	...
185	```
186
187	Inside VM: bind virtio PCI to `vfio-pci` or

People who liked this also viewed

6 picks

Type