Apply the same logic of commit 00cd15a ("build: properly detect clang
version in Ubuntu") in scx_utils as well.
This allows to build scx_utils properly in Ubuntu.
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
The scx_nest scheduler seems to be behaving well. Let's merge it to the
scx repo so that CachyOS can package and use it more easily.
Signed-off-by: David Vernet <void@manifault.com>
Some distro may add their own prefix to the version string of clang, for
example in Ubuntu:
$ clang --version
Ubuntu clang version 17.0.5 (1ubuntu1)
...
That triggers the following meson error during the setup phase:
meson.build:25:44: ERROR: String '' cannot be converted to int
Change the regexp used to evaluate the clang version to avoid this
build failure.
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
We were assigning curr to prev stats, and vice versa, in calc_util().
This was causing the following crash on debug builds:
[void@maniforge scheds]$ sudo RUST_BACKTRACE=1 scx_rusty
00:00:56 [INFO] CPUs: online/possible = 32/32
00:00:56 [INFO] DOM[00] cpumask 0000000000FF00FF (16 cpus)
00:00:56 [INFO] DOM[01] cpumask 00000000FF00FF00 (16 cpus)
00:00:56 [INFO] Rusty Scheduler Attached
thread 'main' panicked at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/core/src/ops/arith.rs:217:1:
attempt to subtract with overflow
stack backtrace:
0: rust_begin_unwind
at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/std/src/panicking.rs:597:5
1: core::panicking::panic_fmt
at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/core/src/panicking.rs:72:14
2: core::panicking::panic
at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/core/src/panicking.rs:127:5
3: <u64 as core::ops::arith::Sub>::sub
at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/core/src/ops/arith.rs:217:1
4: <&u64 as core::ops::arith::Sub<&u64>>::sub
at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/core/src/internal_macros.rs:55:17
5: scx_rusty::calc_util
at ./rust-user/scx_rusty/src/main.rs:216:29
6: scx_rusty::Tuner::step
at ./rust-user/scx_rusty/src/main.rs:444:38
7: scx_rusty::Scheduler::run
at ./rust-user/scx_rusty/src/main.rs:1198:17
8: scx_rusty::main
at ./rust-user/scx_rusty/src/main.rs:1261:5
9: core::ops::function::FnOnce::call_once
at /rustc/475c71da0710fd1d40c046f9cee04b733b5b2b51/library/core/src/ops/function.rs:250:5
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
Flip them to avoid the crash. Rusty now runs fine.
Signed-off-by: David Vernet <void@manifault.com>
There's a fairly comprehensive README in the kernel's tools/sched_ext
directory which describes each of the example schedulers. Let's pull it
into this repository, and split it into the various subdirectories
containing the kernele-examples/ schedulers, and the rust-user/
schedulers.
Signed-off-by: David Vernet <void@manifault.com>
The root README files have some grammatical mistakes, and/or need to be
updated to not be in the context of being sent as a patch set. Update
them.
Signed-off-by: David Vernet <void@manifault.com>
SCX_DSQ_GLOBAL now does not support vtime dispatching. scx_simple uses
it to do vtime scheduling, so let's update it to create and use a
separate DSQ that it can both FIFO and PRIQ dispatch to.
Signed-off-by: David Vernet <void@manifault.com>
tp_cgroup_attach_task() walks p->thread_group to visit all member threads
and set tctx->refresh_layer. However, the upstream kernel has removed
p->thread_group recently in 8e1f385104ac ("kill task_struct->thread_group")
as it was mostly a duplicate of p->signal->thread_head list which goes
through p->thread_node.
Switch to iterate via p->thread_node instead, add a comment explaining why
it's using the cgroup TP instead of scx_ops.cgroup_move(), and make
iteration failure non-fatal as the iteration is racy.
As in scx_layered, bpf_map_delete_elem() can fail due to recursion
protection triggering spuriously which can then lead to task_ctx creation
failure after PIDs wrap. Work around by dropping BPF_NOEXIST.
The scx repo is going to serve as the source of truth for sched_ext
schedulers. Reverse the sync direction and include syncing rust-user
schedulers too.
bpf_cflags doesn't contain '-target bpf' because SkeletonBuilder does so
internally. However, bindgen doesn't and we end up specifying options which
are specific to bpf target without actually specifying the target, this
leads to warnings like the followings:
[scx_layered 0.0.1] warning: unknown warning option '-Wno-compare-distinct-pointer-types''; did you mean '-Wno-compare-distinct-pointer-types'? [-Wunknown-warning-option]
[scx_layered 0.0.1] clang diag: warning: argument unused during compilation: '-mcpu=v3' [-Wunused-command-line-argument]
[scx_layered 0.0.1] clang diag: warning: unknown warning option '-Wno-compare-distinct-pointer-types''; did you mean '-Wno-compare-distinct-pointer-types'? [-Wunknown-warning-option]
Fix it by adding '-target bpf' when invoking bindgen.
