Pass struct scx_enable_cmd to scx_enable() rather than unpacking @ops at every call site and re-packing into a fresh cmd inside. bpf_scx_reg() now builds the cmd on its stack and hands it in; scx_enable() just wires up the kthread work and waits. Relocate struct scx_enable_cmd above scx_alloc_and_add_sched() so upcoming patches that also want the cmd can see it. No behavior change.
Unified · Split
| @@ -6529,6 +6529,19 @@ static struct scx_sched_pnode *alloc_pnode(struct scx_sched *sch, int node) · 6528 unmodified lines | |||
| 6529 | return pnode; | 6529 | return pnode; |
| 6530 | } | 6530 | } |
| 6531 | 6531 | ||
| 6532 | /* | ||
| 6533 | * scx_enable() is offloaded to a dedicated system-wide RT kthread to avoid | ||
| 6534 | * starvation. During the READY -> ENABLED task switching loop, the calling | ||
| 6535 | * thread's sched_class gets switched from fair to ext. As fair has higher | ||
| 6536 | * priority than ext, the calling thread can be indefinitely starved under | ||
| 6537 | * fair-class saturation, leading to a system hang. | ||
| 6538 | */ | ||
| 6539 | struct scx_enable_cmd { | ||
| 6540 | struct kthread_work work; | ||
| 6541 | struct sched_ext_ops *ops; | ||
| 6542 | int ret; | ||
| 6543 | }; | ||
| 6544 | |||
| 6532 | /* | 6545 | /* |
| 6533 | * Allocate and initialize a new scx_sched. @cgrp's reference is always | 6546 | * Allocate and initialize a new scx_sched. @cgrp's reference is always |
| 6534 | * consumed whether the function succeeds or fails. | 6547 | * consumed whether the function succeeds or fails. |
| @@ -6771,19 +6784,6 @@ static int validate_ops(struct scx_sched *sch, const struct sched_ext_ops *ops) · 236 unmodified lines | |||
| 6771 | return 0; | 6784 | return 0; |
| 6772 | } | 6785 | } |
| 6773 | 6786 | ||
| 6774 | /* | ||
| 6775 | * scx_enable() is offloaded to a dedicated system-wide RT kthread to avoid | ||
| 6776 | * starvation. During the READY -> ENABLED task switching loop, the calling | ||
| 6777 | * thread's sched_class gets switched from fair to ext. As fair has higher | ||
| 6778 | * priority than ext, the calling thread can be indefinitely starved under | ||
| 6779 | * fair-class saturation, leading to a system hang. | ||
| 6780 | */ | ||
| 6781 | struct scx_enable_cmd { | ||
| 6782 | struct kthread_work work; | ||
| 6783 | struct sched_ext_ops *ops; | ||
| 6784 | int ret; | ||
| 6785 | }; | ||
| 6786 | |||
| 6787 | static void scx_root_enable_workfn(struct kthread_work *work) | 6787 | static void scx_root_enable_workfn(struct kthread_work *work) |
| 6788 | { | 6788 | { |
| 6789 | struct scx_enable_cmd *cmd = container_of(work, struct scx_enable_cmd, work); | 6789 | struct scx_enable_cmd *cmd = container_of(work, struct scx_enable_cmd, work); |
| @@ -7368,11 +7368,10 @@ static s32 __init scx_cgroup_lifetime_notifier_init(void) · 578 unmodified lines | |||
| 7368 | core_initcall(scx_cgroup_lifetime_notifier_init); | 7368 | core_initcall(scx_cgroup_lifetime_notifier_init); |
| 7369 | #endif /* CONFIG_EXT_SUB_SCHED */ | 7369 | #endif /* CONFIG_EXT_SUB_SCHED */ |
| 7370 | 7370 | ||
| 7371 | static s32 scx_enable(struct sched_ext_ops *ops, struct bpf_link *link) | 7371 | static s32 scx_enable(struct scx_enable_cmd *cmd, struct bpf_link *link) |
| 7372 | { | 7372 | { |
| 7373 | static struct kthread_worker *helper; | 7373 | static struct kthread_worker *helper; |
| 7374 | static DEFINE_MUTEX(helper_mutex); | 7374 | static DEFINE_MUTEX(helper_mutex); |
| 7375 | struct scx_enable_cmd cmd; | ||
| 7376 | 7375 | ||
| 7377 | if (!cpumask_equal(housekeeping_cpumask(HK_TYPE_DOMAIN), | 7376 | if (!cpumask_equal(housekeeping_cpumask(HK_TYPE_DOMAIN), |
| 7378 | cpu_possible_mask)) { | 7377 | cpu_possible_mask)) { |
| @@ -7396,16 +7395,15 @@ static s32 scx_enable(struct sched_ext_ops *ops, struct bpf_link *link) · 17 unmodified lines | |||
| 7396 | } | 7395 | } |
| 7397 | 7396 | ||
| 7398 | #ifdef CONFIG_EXT_SUB_SCHED | 7397 | #ifdef CONFIG_EXT_SUB_SCHED |
| 7399 | if (ops->sub_cgroup_id > 1) | 7398 | if (cmd->ops->sub_cgroup_id > 1) |
| 7400 | kthread_init_work(&cmd.work, scx_sub_enable_workfn); | 7399 | kthread_init_work(&cmd->work, scx_sub_enable_workfn); |
| 7401 | else | 7400 | else |
| 7402 | #endif /* CONFIG_EXT_SUB_SCHED */ | 7401 | #endif /* CONFIG_EXT_SUB_SCHED */ |
| 7403 | kthread_init_work(&cmd.work, scx_root_enable_workfn); | 7402 | kthread_init_work(&cmd->work, scx_root_enable_workfn); |
| 7404 | cmd.ops = ops; | ||
| 7405 | 7403 | ||
| 7406 | kthread_queue_work(READ_ONCE(helper), &cmd.work); | 7404 | kthread_queue_work(READ_ONCE(helper), &cmd->work); |
| 7407 | kthread_flush_work(&cmd.work); | 7405 | kthread_flush_work(&cmd->work); |
| 7408 | return cmd.ret; | 7406 | return cmd->ret; |
| 7409 | } | 7407 | } |
| 7410 | 7408 | ||
| 7411 | 7409 | ||
| @@ -7577,7 +7575,9 @@ static int bpf_scx_check_member(const struct btf_type *t, · 165 unmodified lines | |||
| 7577 | 7575 | ||
| 7578 | static int bpf_scx_reg(void *kdata, struct bpf_link *link) | 7576 | static int bpf_scx_reg(void *kdata, struct bpf_link *link) |
| 7579 | { | 7577 | { |
| 7580 | return scx_enable(kdata, link); | 7578 | struct scx_enable_cmd cmd = { .ops = kdata }; |
| 7579 | |||
| 7580 | return scx_enable(&cmd, link); | ||
| 7581 | } | 7581 | } |
| 7582 | 7582 | ||
| 7583 | static void bpf_scx_unreg(void *kdata, struct bpf_link *link) | 7583 | static void bpf_scx_unreg(void *kdata, struct bpf_link *link) |