Message-ID: <CABFh=a7EZQanSX+=of5A41Kwy83YDmu-AoGJ5hCns8YNkJ2FxA@mail.gmail.com>
Date: Thu, 23 Oct 2025 10:03:58 -0400
From: Emil Tsalapatis <emil@...alapatis.com>
To: Andrea Righi <arighi@...dia.com>
Cc: Tejun Heo <tj@...nel.org>, David Vernet <void@...ifault.com>, 
	Changwoo Min <changwoo@...lia.com>, sched-ext@...ts.linux.dev, 
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 sched_ext/for-6.19] sched_ext/tools: Add compat layer
 for kfuncs using struct *_args

On Thu, Oct 23, 2025 at 4:01 AM Andrea Righi <arighi@...dia.com> wrote:
>
> Commit c0d630ba347c7 ("sched_ext: Wrap kfunc args in struct to prepare
> for aux__prog") introduced new kfuncs that take their scalar arguments
> packed into dedicated *_args structs.
>
> However, these new kfuncs need to coexist with the previous scalar
> variants, which causes build failures in sched_ext schedulers when
> building against older kernels.
>
> To address this, mirror the struct definitions adding a __COMPAT prefix,
> ensuring schedulers can still build also on old kernels that don't have
> the BTF definitions for these structs.
>
> Moreover, add a __COMPAT_bpf_ksym_exists() helper, which extends
> bpf_ksym_exists() to handle static inline kfunc wrappers and fall back
> to the *___compat() variants, where available.
>
> Fixes: c0d630ba347c7 ("sched_ext: Wrap kfunc args in struct to prepare for aux__prog")
> Signed-off-by: Andrea Righi <arighi@...dia.com>
> ---

Tested the patch on a 6.17 and a 6.19 system, binaries run on both
when compiled on either.

Reviewed-by: Emil Tsalapatis <emil@...alapatis.com>

> Changes v2 -> v3:
>  - Drop the ___v2() suffix from the new kfuncs (Tejun)
>  - Add __COMPAT_bpf_ksym_exists() to handle kfunc detection with static
>    inline wrappers
>  - Link to v2: https://lore.kernel.org/all/20251022153610.20111-1-arighi@nvidia.com/
>
> ChangeLog v1 -> v2:
>  - Introduce __COMPAT_* struct to fix build error with v6.19 (thanks Emil)
>
>  tools/sched_ext/include/scx/common.bpf.h |  3 --
>  tools/sched_ext/include/scx/compat.bpf.h | 36 +++++++++++++++++++++++-
>  2 files changed, 35 insertions(+), 4 deletions(-)
>
> diff --git a/tools/sched_ext/include/scx/common.bpf.h b/tools/sched_ext/include/scx/common.bpf.h
> index e65b1eb668ea5..64e5411d04c04 100644
> --- a/tools/sched_ext/include/scx/common.bpf.h
> +++ b/tools/sched_ext/include/scx/common.bpf.h
> @@ -60,9 +60,6 @@ static inline void ___vmlinux_h_sanity_check___(void)
>
>  s32 scx_bpf_create_dsq(u64 dsq_id, s32 node) __ksym;
>  s32 scx_bpf_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags, bool *is_idle) __ksym;
> -s32 __scx_bpf_select_cpu_and(struct task_struct *p, const struct cpumask *cpus_allowed,
> -                            struct scx_bpf_select_cpu_and_args *args) __ksym __weak;
> -bool __scx_bpf_dsq_insert_vtime(struct task_struct *p, struct scx_bpf_dsq_insert_vtime_args *args) __ksym __weak;
>  u32 scx_bpf_dispatch_nr_slots(void) __ksym;
>  void scx_bpf_dispatch_cancel(void) __ksym;
>  bool scx_bpf_dsq_move_to_local(u64 dsq_id) __ksym __weak;
> diff --git a/tools/sched_ext/include/scx/compat.bpf.h b/tools/sched_ext/include/scx/compat.bpf.h
> index a023b71991a6a..d534cacb94066 100644
> --- a/tools/sched_ext/include/scx/compat.bpf.h
> +++ b/tools/sched_ext/include/scx/compat.bpf.h
> @@ -15,6 +15,13 @@
>         __ret;                                                                  \
>  })
>
> +/*
> + * Extended version of bpf_ksym_exists() that handles static inline kfunc
> + * wrappers, falling back to the *___compat versions.
> + */
> +#define __COMPAT_bpf_ksym_exists(sym) \
> +       (__builtin_constant_p(!!sym) ? bpf_ksym_exists(sym##___compat) : !!sym)
> +
>  /*
>   * v6.15: 950ad93df2fc ("bpf: add kfunc for populating cpumask bits")
>   *
> @@ -161,6 +168,27 @@ static inline struct task_struct *__COMPAT_scx_bpf_cpu_curr(int cpu)
>         return rq ? rq->curr : NULL;
>  }
>
> +/*
> + * v6.19: Mirror the following _args structs, to prevent build errors in
> + * kernels that don't have these structs defined yet.
> + *
> + * The kernel will carry these __COMPAT_* structs until v6.23 (see below).
> + */
> +#define scx_bpf_select_cpu_and_args __COMPAT_scx_bpf_select_cpu_and_args
> +struct __COMPAT_scx_bpf_select_cpu_and_args {
> +       s32 prev_cpu;
> +       u64 wake_flags;
> +       u64 flags;
> +};
> +
> +#define scx_bpf_dsq_insert_vtime_args __COMPAT_scx_bpf_dsq_insert_vtime_args
> +struct __COMPAT_scx_bpf_dsq_insert_vtime_args {
> +       u64 dsq_id;
> +       u64 slice;
> +       u64 vtime;
> +       u64 enq_flags;
> +};
> +
>  /*
>   * v6.19: To work around BPF maximum parameter limit, the following kfuncs are
>   * replaced with variants that pack scalar arguments in a struct. Wrappers are
> @@ -170,9 +198,15 @@ static inline struct task_struct *__COMPAT_scx_bpf_cpu_curr(int cpu)
>   * compatibility. After v6.23 release, remove the compat handling and move the
>   * wrappers to common.bpf.h.
>   */
> +s32 __scx_bpf_select_cpu_and(struct task_struct *p, const struct cpumask *cpus_allowed,
> +                            struct scx_bpf_select_cpu_and_args *args) __ksym __weak;
>  s32 scx_bpf_select_cpu_and___compat(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
>                                     const struct cpumask *cpus_allowed, u64 flags) __ksym __weak;
> -void scx_bpf_dsq_insert_vtime___compat(struct task_struct *p, u64 dsq_id, u64 slice, u64 vtime, u64 enq_flags) __ksym __weak;
> +
> +bool __scx_bpf_dsq_insert_vtime(struct task_struct *p,
> +                               struct scx_bpf_dsq_insert_vtime_args *args) __ksym __weak;
> +void scx_bpf_dsq_insert_vtime___compat(struct task_struct *p,
> +                               u64 dsq_id, u64 slice, u64 vtime, u64 enq_flags) __ksym __weak;
>
>  /**
>   * scx_bpf_select_cpu_and - Pick an idle CPU usable by task @p
> --
> 2.51.1
>

Powered by blists - more mailing lists