On 2/26/20 2:44 PM, Paolo Pisati wrote:
This particular selftest fails on arm64 (x86-64 is fine): $ sudo ./tools/testing/selftests/bpf/test_verifier -v 13 #13/p valid read map access into a read-only array 2 , verifier log: 0: (7a) *(u64 *)(r10 -8) = 0 1: (bf) r2 = r10 2: (07) r2 += -8 3: (18) r1 = 0xffff00becd0d8c00 5: (85) call bpf_map_lookup_elem#1 6: (15) if r0 == 0x0 goto pc+6 R0_w=map_value(id=0,off=0,ks=4,vs=48,imm=0) R10=fp0 fp-8_w=mmmmmmmm 7: (bf) r1 = r0 8: (b7) r2 = 4 9: (b7) r3 = 0 10: (b7) r4 = 0 11: (b7) r5 = 0 12: (85) call bpf_csum_diff#28 R0_w=map_value(id=0,off=0,ks=4,vs=48,imm=0) R1_w=map_value(id=0,off=0,ks=4,vs=48,imm=0) R2_w=inv4 R3_w=inv0 R4_w=inv0 R5_w=inv0 R10=fp0 fp-8_w=mmmmmmmm last_idx 12 first_idx 0 regs=4 stack=0 before 11: (b7) r5 = 0 regs=4 stack=0 before 10: (b7) r4 = 0 regs=4 stack=0 before 9: (b7) r3 = 0 regs=4 stack=0 before 8: (b7) r2 = 4 last_idx 12 first_idx 0 regs=10 stack=0 before 11: (b7) r5 = 0 regs=10 stack=0 before 10: (b7) r4 = 0 13: (95) exit from 6 to 13: safe processed 14 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 1 FAIL retval 65507 != -29 (run 1/1) 0: (7a) *(u64 *)(r10 -8) = 0 1: (bf) r2 = r10 2: (07) r2 += -8 3: (18) r1 = 0xffff00becd0d8c00 5: (85) call bpf_map_lookup_elem#1 6: (15) if r0 == 0x0 goto pc+6 R0_w=map_value(id=0,off=0,ks=4,vs=48,imm=0) R10=fp0 fp-8_w=mmmmmmmm 7: (bf) r1 = r0 8: (b7) r2 = 4 9: (b7) r3 = 0 10: (b7) r4 = 0 11: (b7) r5 = 0 12: (85) call bpf_csum_diff#28 R0_w=map_value(id=0,off=0,ks=4,vs=48,imm=0) R1_w=map_value(id=0,off=0,ks=4,vs=48,imm=0) R2_w=inv4 R3_w=inv0 R4_w=inv0 R5_w=inv0 R10=fp0 fp-8_w=mmmmmmmm last_idx 12 first_idx 0 regs=4 stack=0 before 11: (b7) r5 = 0 regs=4 stack=0 before 10: (b7) r4 = 0 regs=4 stack=0 before 9: (b7) r3 = 0 regs=4 stack=0 before 8: (b7) r2 = 4 last_idx 12 first_idx 0 regs=10 stack=0 before 11: (b7) r5 = 0 regs=10 stack=0 before 10: (b7) r4 = 0 13: (95) exit from 6 to 13: safe processed 14 insns (limit 1000000) max_states_per_insn 0 total_states 1 peak_states 1 mark_read 1 Summary: 0 PASSED, 0 SKIPPED, 1 FAILED Above output without line wrapping: https://paste.ubuntu.com/p/qhCK8nJjKw/ Kernel version 5.4.21, config: https://paste.ubuntu.com/p/G3yxvvjRMS/
Yep, the csum_diff is broken for non-x86_64. Fix is wip on my side, will keep you posted. Thanks, Daniel
