[PATCH bpf-next v1 0/2] veristat: memory accounting for bpf programs

[Eduard Zingerman <eddyz87@gmail.com>]

When working on the verifier, it is sometimes interesting to know how a
particular change affects memory consumption. This patch-set modifies
veristat to provide such information. As a collateral, kernel needs an
update to make allocations reachable from BPF program load accountable
in memcg statistics.

Here is a sample output:

    File       Program               Peak states  Peak memory (KiB)
    ---------  --------------------  -----------  -----------------
    bpf.bpf.o  lavd_select_cpu              1311              26256
    bpf.bpf.o  lavd_enqueue                 1140              22720
    bpf.bpf.o  layered_enqueue               777              11504
    ...

Technically, this is implemented by creating and entering a new cgroup
before verifying each program. The difference in memory.peak values
before and after bpf_object__load() is reported as the metric.

This incurs some overhead in veristat runtime. For example:
- increase from 82s to 102s on test_progs BPF binaries;
- increase from 42s to 47s on sched_ext BPF binaries.

Measurements are not completely stable and might change from run to
run by +-256Kb or something close. For sched_ext binaries I observe a
rate of 3 changes per run from a total of 188 programs. Mostly affects
very small programs.

I tried a different scheme, where new cgroup was allocated only once,
and then a combination of "echo 32G > memory.reclaim" and
"echo reset > memory.peak" (via fd) was executed before each program
load. For reasons unclear this approach did not produce stable
measurements.

Eduard Zingerman (2):
  bpf: include verifier memory allocations in memcg statistics
  veristat: memory accounting for bpf programs

 kernel/bpf/btf.c                       |  15 +-
 kernel/bpf/verifier.c                  |  49 ++---
 tools/testing/selftests/bpf/veristat.c | 249 ++++++++++++++++++++++++-
 3 files changed, 275 insertions(+), 38 deletions(-)

-- 
2.48.1