scx_flatcg: Keep cgroup rb nodes stashed #33
Conversation
|
Note: I have validated this against an older |
|
LGTM, please merge once tested. Thanks. |
This PR has been sitting as the series it depends on didn't make it into 6.7. It's in |
The flatcg scheduler uses a rb_node type - struct cgv_node - to keep track of vtime. On cgroup init, a cgv_node is created and stashed in a hashmap - cgv_node_stash - for later use. In cgroup_enqueued and try_pick_next_cgroup, the node is inserted into the rbtree, which required removing it from the stash before this patch's changes. This patch makes cgv_node refcounted, which allows keeping it in the stash for the entirety of the cgroup's lifetime. Unnecessary bpf_kptr_xchg's and other boilerplate can be removed as a result. Note that in addition to bpf_refcount patches, which have been upstream for quite some time, this change depends on a more recent series [0]. [0]: https://lore.kernel.org/bpf/20231107085639.3016113-1-davemarchevsky@fb.com/ Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
davemarchevsky
force-pushed
the
fewer_kptr_xchg
branch
from
919edeb
to
3b7f33e
Compare
|
6.8 has been cut, rebased + addressed nit. Think this is good-to-go |
|
Would this work with the currently packaged kernels on arch and ubuntu? |
|
@htejun Looks like @arighi 's ubuntu kernel has backported the necessary series - it contains Since 6.8 has only been out for a day, shall we wait a bit for CachyOS folks to update before poking / trying to submit a PR? |
|
Yeah, let's wait a bit more so that the released kernels have the support. @jordalgo, how about fedora? What's the version situation there? |
|
@htejun Still working on both the c-schedulers package and I plan on starting on the actual kernel changes this week or next. |
6.8 will land on 13.03.2024 from our side. Everything seems fine so far. |
|
Looks like the packaged kernels should be ready by the time scx is packaged the next time. Merging now. Thanks. |
The flatcg scheduler uses a rb_node type - struct cgv_node - to keep track of vtime. On cgroup init, a cgv_node is created and stashed in a hashmap - cgv_node_stash - for later use. In cgroup_enqueued and try_pick_next_cgroup, the node is inserted into the rbtree, which required removing it from the stash before this patch's changes.
This patch makes cgv_node refcounted, which allows keeping it in the stash for the entirety of the cgroup's lifetime. Unnecessary bpf_kptr_xchg's and other boilerplate can be removed as a result.
Note that in addition to bpf_refcount patches, which have been upstream for quite some time, this change depends on a more recent series 0.