[PATCH 3/5] sysinfo: Require that sysinfo_detect be called before other methods

[Simon Glass]

Hi Sean,

On Fri, 5 Mar 2021 at 08:12, Sean Anderson <sean.anderson at seco.com> wrote:
>
>
>
> On 3/4/21 11:08 PM, Simon Glass wrote:
>  > Hi Sean,
>  >
>  > On Mon, 1 Mar 2021 at 15:46, Sean Anderson <sean.anderson at seco.com> wrote:
>  >>
>  >> This updates sysinfo documentation to document that detect() must be
>  >> called first. This allows drivers to cache information in detect() and
>  >> perform (cheaper) retrieval in the other accessors. This also modifies
>  >> the only instance where this sequencing was not followed.
>  >>
>  >> Signed-off-by: Sean Anderson <sean.anderson at seco.com>
>  >> ---
>  >>
>  >>   common/spl/spl_fit.c |  4 ++++
>  >>   include/sysinfo.h    | 29 ++++++++++++++++++++---------
>  >>   2 files changed, 24 insertions(+), 9 deletions(-)
>  >
>  > Reviewed-by: Simon Glass <sjg at chromium.org>
>  >
>  > Is it possible to enforce this in the uclass?
>
> So have the uclass call detect() if it hasn't been called? Should there
> be a way to re-detect?

I mean have a check function (that all off the uclass methods call)
that checks it has been called and returns an error immediately if
not.

I don't think automatically detecting is a great idea - as you say it
makes it hard to re-detect. But just a bool flag in the device's
uclass-private data should be enough.

>
>  > Also I think -EPERM or -EACCES might be better than -ENOENT.
>
> Hm, I thought I saw gazerbeam using ENOENT for this purpose, but on
> further review it looks like it does not. I think EPERM is good for
> this.

OK.

Regards,
Simon