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.
