[RFC PATCH v13 0/4] blk-iocost: BPF struct_ops cost model

[Tao Cui]

From: Tao Cui <cuitao@xxxxxxxxxx>

This series lets a BPF program take over the cost model of one
device.  The BPF model provides calc_cost() for charged IOs plus two
transfer coefficients for the completion-time sizing, io.cost.model
selects between the attached model and the builtin one, and enabling
and disabling the controller stays with io.cost.qos.

The builtin model is a linear formula derived from a handful of device
parameters, and the right formula is device specific.  A BPF model can
be adapted to the device and updated without rebuilding the kernel.

The series is based on next-20261105.

Tao Cui (4):
  blk-iocost: add BPF struct_ops cost model support
  blk-iocost: add iocost_ioc_tick tracepoint for per-period device
    summary
  selftests/bpf: add iocost cost model test
  docs: cgroup-v2: document the iocost BPF cost model attachment

 Documentation/admin-guide/cgroup-v2.rst       |  49 +-
 MAINTAINERS                                   |   1 +
 block/Kconfig                                 |   4 +
 block/blk-iocost.c                            | 284 +++++-
 include/trace/events/iocost.h                 |  43 +
 tools/testing/selftests/bpf/config            |   2 +
 .../selftests/bpf/prog_tests/iocost_model.c   | 812 ++++++++++++++++++
 .../selftests/bpf/progs/iocost_model.c        | 180 ++++
 tools/testing/selftests/bpf/progs/iocost_ms.c | 197 +++++
 9 files changed, 1560 insertions(+), 12 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/iocost_model.c
 create mode 100644 tools/testing/selftests/bpf/progs/iocost_model.c
 create mode 100644 tools/testing/selftests/bpf/progs/iocost_ms.c

Changes since v12:

- the iocost_ioc_tick emission now sits inside the
  trace_iocost_ioc_tick_enabled() check, as tracepoints.rst requires,
  and the usage_pct trace field is widened to u64
- detaching a model resets the autop stepping timestamps, so the
  cooldown restarts fresh when the builtin model comes back
- iocost_bpf_supported() probes kernel BTF for the struct_ops wrapper
  type instead of /proc/config.gz, so an unavailable config no longer
  skips the tests, and the probe result is cached across cases
- the selftest sets up its private mount namespace and configfs mount
  once instead of stacking them across cases
- the multi-stream case feeds two interleaved streams and asserts that
  calc_cost() actually priced them
- the selftest cgroup names are created inside the per-pid work
  directory (leading '/'), nullb_create() rolls back the instance on
  its own error paths, and the streams test cleans up in the right
  order
- docs: document the attach rejection codes (-EOPNOTSUPP, -ENODEV,
  -EBUSY) and that root-cgroup IOs still use the attached model's
  transfer coefficients for completion-time sizing

Tested on next-20261105 (qemu x86_64): the selftests pass, and the 2x
example model doubles the charged cost over the builtin on a
rotational device (2.00x measured), returning to 1.00x after
"model=linear".

Link: https://lore.kernel.org/r/20261103132613.1952975-1-cui.tao@xxxxxxxxx # v12