Re: [PATCH bpf-next v8 7/9] bpf: task work scheduling kfuncs

[Heiko Carstens <hca@linux.ibm.com>]

On Fri, Jul 24, 2026 at 02:06:58PM +0100, Mykyta Yatsenko wrote:
> On 7/24/26 12:22 PM, Heiko Carstens wrote:
> > On Tue, Sep 23, 2025 at 12:24:02PM +0100, Mykyta Yatsenko wrote:
> >> From: Mykyta Yatsenko <yatsenko@meta.com>
> >>
> >> Implementation of the new bpf_task_work_schedule kfuncs, that let a BPF
> >> program schedule task_work callbacks for a target task:
> >>  * bpf_task_work_schedule_signal() - schedules with TWA_SIGNAL
> >>  * bpf_task_work_schedule_resume() - schedules with TWA_RESUME
> > 
> > ...
> > 
> >> Signed-off-by: Mykyta Yatsenko <yatsenko@meta.com>
> >> Reviewed-by: Andrii Nakryiko <andrii@kernel.org>
> >> Reviewed-by: Eduard Zingerman <eddyz87@gmail.com>
> >> Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
> >> ---
> >>  kernel/bpf/helpers.c | 292 ++++++++++++++++++++++++++++++++++++++++++-
> >>  1 file changed, 290 insertions(+), 2 deletions(-)
> > 
> > ...
> > 
> >> +static void bpf_task_work_callback(struct callback_head *cb)
> >> +{
> > ...
> >> +	migrate_disable();
> >> +	ctx->callback_fn(ctx->map, key, ctx->map_val);
> >> +	migrate_enable();
> > 
> > The above indirect call results in a CFI failure - while trying to make kCFI
> > work on s390, and running the bpf selftests I stumbled across this, which can
> > also be reproduced on x86:
> > 
> 
> Thanks for reporting, I'll take a look.

FWIW, there is another thing to consider: architectures which select
ARCH_SUPPORTS_CFI only optionally need to support kCFI for BPF.

In order to avoid CFI failures because of missing CFI hashes of BPF programs,
functions which call such programs indirectly need to have the overridable
__bpfcall attribute. So something like this is required as well:

-static void bpf_task_work_callback(struct callback_head *cb)
+static __bpfcall void bpf_task_work_callback(struct callback_head *cb)