s64ilp32: ebpf_jit: Fix the incorrect jump addresses and index offset in tail_call

[cp0613]

This patch solves the 10 tail_call testing issues in test_bpf. At this point, all tests of test_bpf in BPF_JIT mode have passed. The comparison between s64lp64 and s32ilp32 is as follows:

• s64lp64

test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

• s64ilp32

test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

• s32ilp32

test_bpf: Summary: 1027 PASSED, 0 FAILED, [832/1015 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

Actually, s64ilp32 and s64lp64 perform consistently, both in terms of the number that can be executed by JIT and execution time, while s32ilp32 has a relatively small number that can be executed by JIT, and the execution time is also longer under the same JIT execution situation.

[guoren83]

Great Job! But some comments:

This patch solves the 10 tail_call testing issues in test_bpf. At this point, all tests of test_bpf in BPF_JIT mode have passed. The comparison between s64lp64 and s32ilp32 is as follows:

Here is the comparison between s64ilp32, s64lp64 and s32ilp32:

• s64lp64

test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

• s64ilp32

test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

• s32ilp32

test_bpf: Summary: 1027 PASSED, 0 FAILED, [832/1015 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

Actually, s64ilp32 and s64lp64 perform consistently, both in terms of the number that can be executed by JIT and execution time,
while s32ilp32 has a relatively small number that can be executed by JIT,

832/1015 means 80+%, which is not a small number, right? How we describe it adequately.

and the execution time is also longer under the same JIT execution situation.

What is the longer? We should give an example to show.

[cp0613]

This patch solves the 10 tail_call testing issues in test_bpf. At this point, all tests of test_bpf in BPF_JIT mode have passed. Here is the comparison between s64ilp32, s64lp64 and s32ilp32:

• s64lp64

...
test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
test_bpf: #0 Tail call leaf jited:1 188 PASS
test_bpf: #1 Tail call 2 jited:1 180 PASS
test_bpf: #2 Tail call 3 jited:1 203 PASS
test_bpf: #3 Tail call 4 jited:1 225 PASS
test_bpf: #4 Tail call load/store leaf jited:1 145 PASS
test_bpf: #5 Tail call load/store jited:1 195 PASS
test_bpf: #6 Tail call error path, max count reached jited:1 997 PASS
test_bpf: #7 Tail call count preserved across function calls jited:1 155563 PASS
test_bpf: #8 Tail call error path, NULL target jited:1 164 PASS
test_bpf: #9 Tail call error path, index out of range jited:1 136 PASS
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
...
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

• s64ilp32

...
test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
test_bpf: #0 Tail call leaf jited:1 160 PASS
test_bpf: #1 Tail call 2 jited:1 221 PASS
test_bpf: #2 Tail call 3 jited:1 251 PASS
test_bpf: #3 Tail call 4 jited:1 275 PASS
test_bpf: #4 Tail call load/store leaf jited:1 198 PASS
test_bpf: #5 Tail call load/store jited:1 262 PASS
test_bpf: #6 Tail call error path, max count reached jited:1 1390 PASS
test_bpf: #7 Tail call count preserved across function calls jited:1 204492 PASS
test_bpf: #8 Tail call error path, NULL target jited:1 199 PASS
test_bpf: #9 Tail call error path, index out of range jited:1 168 PASS
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
...
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

• s32ilp32

...
test_bpf: Summary: 1027 PASSED, 0 FAILED, [832/1015 JIT'ed]
test_bpf: #0 Tail call leaf jited:1 266 PASS
test_bpf: #1 Tail call 2 jited:1 409 PASS
test_bpf: #2 Tail call 3 jited:1 481 PASS
test_bpf: #3 Tail call 4 jited:1 537 PASS
test_bpf: #4 Tail call load/store leaf jited:1 325 PASS
test_bpf: #5 Tail call load/store jited:1 427 PASS
test_bpf: #6 Tail call error path, max count reached jited:1 3050 PASS
test_bpf: #7 Tail call count preserved across function calls jited:1 255522 PASS
test_bpf: #8 Tail call error path, NULL target jited:1 315 PASS
test_bpf: #9 Tail call error path, index out of range jited:1 280 PASS
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
...
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

Actually, s64ilp32 and s64lp64 perform consistently, both in terms of the number that can be executed by JIT and execution time. while, only 80% of cases in s32ilp32 can be executed by JIT, and the execution time is also longer under the same JIT execution situation.