+
+u32 bpf_direct_xsk(const struct bpf_prog *prog, struct xdp_buff *xdp)
+{
+ struct xdp_sock *xsk;
+
+ xsk = xdp_get_xsk_from_qid(xdp->rxq->dev, xdp->rxq->queue_index);
+ if (xsk) {
+ struct bpf_redirect_info *ri =
+ this_cpu_ptr(&bpf_redirect_info);
+
+ ri->xsk = xsk;
+ return XDP_REDIRECT;
+ }
+
+ return XDP_PASS;
+}
+EXPORT_SYMBOL(bpf_direct_xsk);
So you're saying there is a:
"""
xdpsock rxdrop 1 core (both app and queue's irq pinned to the same core)
default : taskset -c 1 ./xdpsock -i enp66s0f0 -r -q 1
direct-xsk :taskset -c 1 ./xdpsock -i enp66s0f0 -r -q 1 6.1x improvement in drop rate """
6.1x gain running above C code vs exactly equivalent BPF code?
How is that possible?
It seems to be due to the overhead of __bpf_prog_run on older processors
(Ivybridge). The overhead is smaller on newer processors, but even on
skylake i see around 1.5x improvement.
perf report with default xdpsock
================================
Samples: 2K of event 'cycles:ppp', Event count (approx.): 8437658090
Overhead Command Shared Object Symbol
34.57% xdpsock xdpsock [.] main
17.19% ksoftirqd/1 [kernel.vmlinux] [k] ___bpf_prog_run
13.12% xdpsock [kernel.vmlinux] [k] ___bpf_prog_run
4.09% ksoftirqd/1 [kernel.vmlinux] [k] __x86_indirect_thunk_rax
3.08% xdpsock [kernel.vmlinux] [k] nmi
2.76% ksoftirqd/1 [kernel.vmlinux] [k] xsk_map_lookup_elem
2.33% xdpsock [kernel.vmlinux] [k] __x86_indirect_thunk_rax
2.33% ksoftirqd/1 [i40e] [k] i40e_clean_rx_irq_zc
2.16% xdpsock [kernel.vmlinux] [k] bpf_map_lookup_elem
1.82% ksoftirqd/1 [kernel.vmlinux] [k] xdp_do_redirect
1.41% ksoftirqd/1 [kernel.vmlinux] [k] xsk_rcv
1.39% ksoftirqd/1 [kernel.vmlinux] [k] update_curr
1.09% ksoftirqd/1 [kernel.vmlinux] [k] bpf_xdp_redirect_map
1.09% xdpsock [i40e] [k] i40e_clean_rx_irq_zc
1.08% ksoftirqd/1 [kernel.vmlinux] [k] __xsk_map_redirect
1.07% swapper [kernel.vmlinux] [k] xsk_umem_peek_addr
1.05% ksoftirqd/1 [kernel.vmlinux] [k] xsk_umem_peek_addr
0.89% swapper [kernel.vmlinux] [k] __xsk_map_redirect
0.87% ksoftirqd/1 [kernel.vmlinux] [k] __bpf_prog_run32
0.87% swapper [kernel.vmlinux] [k] intel_idle
0.67% xdpsock [kernel.vmlinux] [k] bpf_xdp_redirect_map
0.57% xdpsock [kernel.vmlinux] [k] xdp_do_redirect
perf report with direct xdpsock
===============================
Samples: 2K of event 'cycles:ppp', Event count (approx.): 17996091975
Overhead Command Shared Object Symbol
18.44% xdpsock [i40e] [k] i40e_clean_rx_irq_zc
15.14% ksoftirqd/1 [i40e] [k] i40e_clean_rx_irq_zc
6.87% xdpsock [kernel.vmlinux] [k] xsk_umem_peek_addr
5.03% ksoftirqd/1 [kernel.vmlinux] [k] xdp_do_redirect
4.21% xdpsock xdpsock [.] main
4.13% ksoftirqd/1 [i40e] [k]
i40e_clean_programming_status
3.71% xdpsock [kernel.vmlinux] [k] xsk_rcv
3.44% ksoftirqd/1 [kernel.vmlinux] [k] nmi
3.41% xdpsock [kernel.vmlinux] [k] nmi
3.20% ksoftirqd/1 [kernel.vmlinux] [k] xsk_rcv
2.45% xdpsock [kernel.vmlinux] [k] xdp_get_xsk_from_qid
2.35% ksoftirqd/1 [kernel.vmlinux] [k] xsk_umem_peek_addr
2.33% ksoftirqd/1 [kernel.vmlinux] [k] net_rx_action
2.16% ksoftirqd/1 [kernel.vmlinux] [k] xsk_umem_consume_tx
2.10% swapper [kernel.vmlinux] [k] __softirqentry_text_start
2.06% xdpsock [kernel.vmlinux] [k] native_irq_return_iret
1.43% xdpsock [kernel.vmlinux] [k] check_preempt_wakeup
1.42% xdpsock [kernel.vmlinux] [k] xsk_umem_consume_tx
1.22% xdpsock [kernel.vmlinux] [k] xdp_do_redirect
1.21% xdpsock [kernel.vmlinux] [k]
dma_direct_sync_single_for_device
1.16% ksoftirqd/1 [kernel.vmlinux] [k] irqtime_account_irq
1.09% xdpsock [kernel.vmlinux] [k] sock_def_readable
0.99% swapper [kernel.vmlinux] [k] intel_idle
0.88% xdpsock [i40e] [k]
i40e_clean_programming_status
0.74% ksoftirqd/1 [kernel.vmlinux] [k] xsk_umem_discard_addr
0.71% ksoftirqd/1 [kernel.vmlinux] [k] __switch_to
0.50% ksoftirqd/1 [kernel.vmlinux] [k]
dma_direct_sync_single_for_device
