> So: create a fake task and preload it into our fake PACA. Load the paca
> just into r13 (local_paca) before we call into dt_cpu_ftrs_init. This fake
> task persists just for the first part of the setup process before we set
> up the real PACAs.

mpe has asked for this to be fixed in a different way, so I'll respin
with that change.

Daniel

>
> Translations get switched on once we leave early_setup, so I think we'd
> already catch any other cases where the PACA or task aren't set up.
>
> Fixes: fb0b0a73b223 ("powerpc: Enable kcov")
> Cc: Andrew Donnellan <[email protected]>
> Signed-off-by: Daniel Axtens <[email protected]>
>
> ---
>
> I haven't made the setup conditional on kcov being compiled in, but I
> guess I could if we think it's worth it?
> ---
>  arch/powerpc/kernel/setup_64.c | 13 ++++++++++++-
>  1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/arch/powerpc/kernel/setup_64.c b/arch/powerpc/kernel/setup_64.c
> index e05e6dd67ae6..26f1b8539f8e 100644
> --- a/arch/powerpc/kernel/setup_64.c
> +++ b/arch/powerpc/kernel/setup_64.c
> @@ -281,7 +281,18 @@ void __init record_spr_defaults(void)
>  
>  void __init early_setup(unsigned long dt_ptr)
>  {
> -     static __initdata struct paca_struct boot_paca;
> +     /*
> +      * We need to get something valid into local_paca/r13 asap if we
> +      * are using kcov. dt_cpu_ftrs_init will call coverage-enabled code
> +      * in the generic dt library, and that will try to call in_task().
> +      * We need a minimal paca that at least provides a valid __current.
> +      * We can't use the usual initialise/setup/fixup path as that relies
> +      * on a CPU feature.
> +      */
> +     static __initdata struct task_struct task = {};
> +     static __initdata struct paca_struct boot_paca = { .__current = &task };
> +
> +     local_paca = &boot_paca;
>  
>       /* -------- printk is _NOT_ safe to use here ! ------- */
>  
> -- 
> 2.20.1

Reply via email to