On Mon, Feb 19, 2024 at 06:35:03PM +0000, Caleb Connolly wrote:
> The current ordering always results in func pointing to the next
> function in the init_sequence. e.g. if fdtdec_setup() fails, ret will
> be set to the error code, then func will be updated to point to
> initf_malloc(), only then is ret checked and the loop broken. The end
> result of this is that the "initcall failed at ..." error will point you
> to initf_malloc(), when the error actually occured in fdtdec_setup()!
>
> This can be quite confusing and result in a lot of time wasted debugging
> code that has nothing to do with the failure (ask me how I know :P).
Heh. Subtle.
>
> Adjust the for loop to check ret immediately after the call and break
> early so that func will correctly reference the failed function.
>
> Signed-off-by: Caleb Connolly <caleb.conno...@linaro.org>
> ---
> lib/initcall.c | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/lib/initcall.c b/lib/initcall.c
> index ce317af213ab..128242e5ff55 100644
> --- a/lib/initcall.c
> +++ b/lib/initcall.c
> @@ -52,11 +52,11 @@ int initcall_run_list(const init_fnc_t init_sequence[])
> ulong reloc_ofs = calc_reloc_ofs();
> const init_fnc_t *ptr;
> enum event_t type;
> init_fnc_t func;
> - int ret = 0;
> + int ret;
We need to keep this initialization. The bug would be when we're
building for CONFIG_ARC and this is called from board_init_f_r(). The
array is empty except for the NULL sentinal at the end. We don't
enter the loop in that case.
regards,
dan carpenter
>
> - for (ptr = init_sequence; func = *ptr, !ret && func; ptr++) {
> + for (ptr = init_sequence; func = *ptr, func; ptr++) {
> type = initcall_is_event(func);
>
> if (type) {
> if (!CONFIG_IS_ENABLED(EVENT))
> @@ -70,8 +70,10 @@ int initcall_run_list(const init_fnc_t init_sequence[])
> debug("initcall: %p\n", (char *)func - reloc_ofs);
> }
>
> ret = type ? event_notify_null(type) : func();
> + if (ret)
> + break;
> }
>
> if (ret) {
> if (CONFIG_IS_ENABLED(EVENT)) {
> --
> 2.43.1
>
