selftests/seccomp: Try to fit runtime of benchmark into timeout

The seccomp benchmark runs five scenarios, one calibration run with no
seccomp filters enabled then four further runs each adding a filter. The
calibration run times itself for 15s and then each additional run executes
for the same number of times.

Currently the seccomp tests, including the benchmark, run with an extended
120s timeout but this is not sufficient to robustly run the tests on a lot
of platforms. Sample timings from some recent runs:

Platform Run 1 Run 2 Run 3 Run 4
--------- ----- ----- ----- -----
PowerEdge R200 16.6s 16.6s 31.6s 37.4s
BBB (arm) 20.4s 20.4s 54.5s
Synquacer (arm64) 20.7s 23.7s 40.3s

The x86 runs from the PowerEdge are quite marginal and routinely fail, for
the successful run reported here the timed portions of the run are at
117.2s leaving less than 3s of margin which is frequently breached. The
added overhead of adding filters on the other platforms is such that there
is no prospect of their runs fitting into the 120s timeout, especially
on 32 bit arm where there is no BPF JIT.

While we could lower the time we calibrate for I'm also already seeing the
currently completing runs reporting issues with the per filter overheads
not matching expectations:

Let's instead raise the timeout to 180s which is only a 50% increase on the
current timeout which is itself not *too* large given that there's only two
tests in this suite.

Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>

selftests/seccomp: Compare bitmap vs filter overhead

As part of the seccomp benchmarking, include the expectations with
regard to the timing behavior of the constant action bitmaps, and report
inconsistencies better.

Example output with constant action bitmaps on x86:

$ sudo ./seccomp_benchmark 100000000
Current BPF sysctl settings:
net.core.bpf_jit_enable = 1
net.core.bpf_jit_harden = 0
Benchmarking 200000000 syscalls...
129.359381409 - 0.008724424 = 129350656985 (129.4s)
getpid native: 646 ns
264.385890006 - 129.360453229 = 135025436777 (135.0s)
getpid RET_ALLOW 1 filter (bitmap): 675 ns
399.400511893 - 264.387045901 = 135013465992 (135.0s)
getpid RET_ALLOW 2 filters (bitmap): 675 ns
545.872866260 - 399.401718327 = 146471147933 (146.5s)
getpid RET_ALLOW 3 filters (full): 732 ns
696.337101319 - 545.874097681 = 150463003638 (150.5s)
getpid RET_ALLOW 4 filters (full): 752 ns
Estimated total seccomp overhead for 1 bitmapped filter: 29 ns
Estimated total seccomp overhead for 2 bitmapped filters: 29 ns
Estimated total seccomp overhead for 3 full filters: 86 ns
Estimated total seccomp overhead for 4 full filters: 106 ns
Estimated seccomp entry overhead: 29 ns
Estimated seccomp per-filter overhead (last 2 diff): 20 ns
Estimated seccomp per-filter overhead (filters / 4): 19 ns
Expectations:
native ≤ 1 bitmap (646 ≤ 675): ✔️
native ≤ 1 filter (646 ≤ 732): ✔️
per-filter (last 2 diff) ≈ per-filter (filters / 4) (20 ≈ 19): ✔️
1 bitmapped ≈ 2 bitmapped (29 ≈ 29): ✔️
entry ≈ 1 bitmapped (29 ≈ 29): ✔️
entry ≈ 2 bitmapped (29 ≈ 29): ✔️
native + entry + (per filter * 4) ≈ 4 filters total (755 ≈ 752): ✔️

[YiFei: Changed commit message to show stats for this patch series]
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/1b61df3db85c5f7f1b9202722c45e7b39df73ef2.1602431034.git.yifeifz2@illinois.edu

selftests/seccomp: use 90s as timeout

As seccomp_benchmark tries to calibrate how many samples will take more
than 5 seconds to execute, it may end up picking up a number of samples
that take 10 (but up to 12) seconds. As the calibration will take double
that time, it takes around 20 seconds. Then, it executes the whole thing
again, and then once more, with some added overhead. So, the thing might
take more than 40 seconds, which is too close to the 45s timeout.

That is very dependent on the system where it's executed, so may not be
observed always, but it has been observed on x86 VMs. Using a 90s timeout
seems safe enough.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Link: https://lore.kernel.org/r/20200601123202.1183526-1-cascardo@canonical.com
Signed-off-by: Kees Cook <keescook@chromium.org>