[PATCH 1/1] lib: utils: identify supported GPIO reset methods

Anup Patel anup at brainfault.org
Thu Sep 30 06:13:27 PDT 2021


On Thu, Sep 30, 2021 at 3:59 PM Heinrich Schuchardt
<heinrich.schuchardt at canonical.com> wrote:
>
> On 9/29/21 14:38, Anup Patel wrote:
> > On Wed, Sep 29, 2021 at 12:31 PM Heinrich Schuchardt
> > <heinrich.schuchardt at canonical.com> wrote:
> >>
> >>
> >>
> >> On 9/29/21 6:16 AM, Anup Patel wrote:
> >>> +Nikita
> >>>
> >>> On Tue, Sep 28, 2021 at 5:13 PM Heinrich Schuchardt
> >>> <heinrich.schuchardt at canonical.com> wrote:
> >>>>
> >>>> The GPIO reset driver supports reset and poweroff. But not all boards
> >>>> support both. gpio_system_reset_check() must detect this situation.
> >>>>
> >>>> Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt at canonical.com>
> >>>
> >>> Looks good to me.
> >>>
> >>> Reviewed-by: Anup Patel <anup.patel at wdc.com>
> >>>
> >>> I had mentioned on the PMIC reset series that we need to improve
> >>> the sbi_system.h device registration such that reset drivers can
> >>> register a reset device for a range of reset types. This will allow
> >>> separate reset drivers (e.g. PMIC+GPIO) for SiFive Unmatched.
> >>> Also, reset_check() callback will not be required anymore.
> >>
> >> Hello Anup,
> >>
> >> thanks for reviewing.
> >>
> >> Currently we keep a list of drivers in a static array. As we are
> >> acquiring more and drivers I would like to move to linker generated
> >> lists as Linux and U-Boot use the for enumerating drivers.
> >>
> >> So a driver could look like:
> >>
> >> int drv1_check(unsigned int type, unsigned int reason) {...}
> >>
> >> void drv1_reset(unsigned int type, unsigned int reason) {...}
> >>
> >> SRST_DRIVER(drv1) = {
> >>           .name = "driver_1",
> >>           .check = drv1_check,
> >>           .reset = drv1_reset,
> >> };
> >>
> >> The linker will create a list of all SRST_DRIVER() instances over which
> >> we can iterate in our code.
> >>
> >> Does this make sense to you?
> >>
> >> Example code is available in
> >> https://github.com/xypron/linker_generated_lists
> >
> > I certainly like the Linker list approach. In fact, I have
> > implemented something similar in-past for Xvisor built-in
> > modules.
> >
> > We can go ahead with Linker list approach with
> > following considerations:
> > 1) The generic SBI library (i.e lib/sbi and include/sbi/
> > directories) should not use linker list because we have
> > couple of projects (e.g. Microchip HSS and EDK2)
> > which directly link with the OpenSBI generic library
> > instead of using OpenSBI firmwares.
>
> The platform specification requires:
>
> "the UEFI ResetSystem() service must be implemented via the SBI System
> Reset Extension."
>
> Is this compatible with linking part of SBI as a library but excluding
> reset drivers?

For Microchip HSS, I had added a compile-time option to
use external OpenSBI firmware so they have a way to
stay compliant.

For EDK2, we have two problems:
1) It's boot-flow assumes to start from M-mode so we
can't use it inside Guest/VM (VS-mode)
2) the UEFI ResetSystem() compliance with platform
spec (which you mentioned)

Changing EDK2 boot-flow to use external OpenSBI
firmware is more involved. Someone need to get this
done soon.

>
> > 2) This should only impact FDT based drivers in
> > SBI utils (i.e. lib/utils and include/sbi_utils directories).
> >
> > Basically, all FDT based driver frameworks under
> > lib/utils can happily move to the Linker list approach
> > so that we don't have to maintain various xyz_drivers[]
> > arrays across different FDT based driver frameworks.
> >
> > Agree ??
>
> While trying to implement the linker list for OpenSBI I ran into some
> problem with the linker:
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=28398

That's unfortunate.

>
> So I would suggest we start by converting the interface using an array.

How about we write a script which:
1) Takes list of C files as input
2) It will scan for a macro in each C file and output
an array into a generated C file
3) The generated C file is compiled and linked with
OpenSBI
4) Each driver framework will have it's own generated
C file

OR

Do you have some other ideas?

Regards,
Anup

>
> Best regards
>
> Heinrich
>
> >
> > Regards,
> > Anup
> >
>



More information about the opensbi mailing list