On Fri, Jul 4, 2025 at 1:53 PM jianghaoran <jianghaoran@xxxxxxxxxx> wrote: > > > > > > 在 2025-07-03星期四的 20:31 +0800，Hengqi Chen写道： > > On Tue, Jul 1, 2025 at 3:41 PM Haoran Jiang < > > jianghaoran@xxxxxxxxxx > > > wrote: > > > 1,Fix the jmp_offset calculation error in the > > > emit_bpf_tail_call function. > > > 2,Fix the issue that MAX_TAIL_CALL_CNT limit bypass in hybrid > > > tailcall and BPF-to-BPF call > > > > > > After applying this patch, testing results are as follows: > > > > > > ./test_progs --allow=tailcalls/tailcall_bpf2bpf_hierarchy_1 > > > 413/18 tailcalls/tailcall_bpf2bpf_hierarchy_1:OK > > > 413 tailcalls:OK > > > Summary: 1/1 PASSED, 0 SKIPPED, 0 FAILED > > > > > > ./test_progs --allow=tailcalls/tailcall_bpf2bpf_hierarchy_2 > > > 413/23 tailcalls/tailcall_bpf2bpf_hierarchy_2:OK > > > 413 tailcalls:OK > > > Summary: 1/1 PASSED, 0 SKIPPED, 0 FAILED > > > > > > ./test_progs --allow=tailcalls/tailcall_bpf2bpf_hierarchy_3 > > > 413/24 tailcalls/tailcall_bpf2bpf_hierarchy_3:OK > > > 413 tailcalls:OK > > > Summary: 1/1 PASSED, 0 SKIPPED, 0 FAILED > > > > > > > Thanks for the fixes. Will review this series soon. > > BTW, do you test other tailcall test cases ? > > > > Cheers, > > --- > > Hengqi > > > > tailcall_1/tailcall_2/tailcall_3/tailcall_4/tailcall_5/tailcall_6/t > ailcall_bpf2bpf_1/tailcall_bpf2bpf_2/tailcall_bpf2bpf_3/tailcall_bp > f2bpf_4/tailcall_bpf2bpf_5/tailcall_bpf2bpf_6 > /tailcall_bpf2bpf_hierarchy_1/tailcall_bpf2bpf_hierarchy_2/tailcall > _bpf2bpf_hierarchy_3/tailcall_failure > These test cases passed > > tailcall_bpf2bpf_fentry/tailcall_bpf2bpf_fexit/tailcall_bpf2bpf_fen > try_fexit/tailcall_bpf2bpf_fentry_entry/tailcall_bpf2bpf_hierarchy_ > fentry/tailcall_bpf2bpf_hierarchy_fexit > /tailcall_bpf2bpf_hierarchy_fentry_fexit/tailcall_bpf2bpf_hierarchy > _fentry_entry/tailcall_freplace/tailcall_bpf2bpf_freplace > These test cases depend on the trampoline capability, which is > currently under review in the Linux kernel. > Please post the full test result of the tailcall test cases for future references. We can safely ignore those ENOTSUPP cases. Thanks. > These two patches are relatively independent. Could we prioritize > reviewing the fixes above first? > Trampoline-dependent changes will be implemented after > trampoline is merged. > > thanks > > > > > > Haoran Jiang (2): > > > LoongArch: BPF: Optimize the calculation method of jmp_offset in the > > > emit_bpf_tail_call function > > > LoongArch: BPF: Fix tailcall hierarchy > > > > > > arch/loongarch/net/bpf_jit.c | 140 ++++++++++++++++++++--------------- > > > 1 file changed, 80 insertions(+), 60 deletions(-) > > > > > > -- > > > 2.43.0 > > > >
