[PATCH 1/2] EFI: Support CAPSULE_FLAGS_INITIATE_RESET for capsule update

[Masami Hiramatsu]

2022年1月26日(水) 16:20 AKASHI Takahiro <takahiro.akashi at linaro.org>:
>
> On Wed, Jan 26, 2022 at 11:29:10AM +0900, Masami Hiramatsu wrote:
> > BTW, the original code comes from EDK2 implementation.
>
> What do you mean by "original code"?

I meant the version1, this patch. EDK2 resets the machine in
CapsuleUpdate() EFI API, but as you said the flag itself has different
meaning.

>
> > It seems that this INITIATE_RESET flag meaning in EDK2 is a bit different
> > from what we thought here.
>
> Yeah,
>
> > The flag is only used for resetting system via
> > UpdateCapsule() EFI call. The capsule update process itself will be done
> > at the boot time and warm reset soon after that.
>
> EDK2's UpdateCapsule() seems to
> - immediately process *all* the capsules with !PERSIST_ACROSS_RESET
> - Set "CapsuleUpdateData" variable with a physical address of capsule data
> - If some of capsules have INITIATE_RESET, kick in warm restart
> After restart,
> - process the rest of capsules with PERSIST_ACROSS_RESET
>   using "CapsuleUpdateData" (?)
>
> > (See HandleCapsules()@ArmPkg/Library/PlatformBootManagerLib/PlatformBm.c)
> >
> > Is it OK to change the INITIATE_RESET flag means? or should we forcibly
> > reset the system if we succeeded to update capsule on boot time?
> > (note that capsule on disk must be done at boot time in U-Boot)
>
> Obviously, we are trying to use the flag in a different way;
> initiating a reset *after* processing a capsule.
>
> While I think that the text in the specification is still ambiguous,
> we should not give the flag a different meaning.

Agreed.

>
> In 8.5.5 "Delivery of Capsules via file on Mass Storage device",
>     In all cases that a capsule is identified for processing the system is
>     restarted after capsule processing is completed.
>
> So a reset in case of capsule-on-disk seems mandatory.
> (Personally, I don't like the system to enforce a reset.)

Yeah, so it seems my first attempt (Kconfig to forcibly reset after
applying capsules)
was correct way...

>
> The discussion above also indicates that the current efi_launch_capsules()
> must be reworked so not to use efi_update_capsule() which is to honor
> the flags in a capsule header.

Indeed. If the system will forcibly reset, ESRT also no need to be updated.
Then we can use efi_capsule_update_firmware() directly from
efi_launch_capsules().

Thank you,

> -Takahiro Akashi
>
> > Thank you,
> >
> > 2022年1月26日(水) 7:31 Masami Hiramatsu <masami.hiramatsu at linaro.org>:
> > >
> > > Hi Takahiro,
> > >
> > > 2022年1月25日(火) 21:49 AKASHI Takahiro <takahiro.akashi at linaro.org>:
> > > >
> > > > Hi Masami,
> > > >
> > > > Thank you for this enhancement.
> > > >
> > > > On Tue, Jan 25, 2022 at 08:31:29PM +0900, Masami Hiramatsu wrote:
> > > > > Support CAPSULE_FLAGS_INITIATE_RESET for rebooting uboot soon after
> > > > > updating firmware. Note that the machine will reboot soon after
> > > > > applying the capsule file which has CAPSULE_FLAGS_INITIATE_RESET
> > > > > flag. If there are multiple capsules and one has this flag, the
> > > > > machine may reboot while scanning the capsule files.
> > > > > You can control when the machine reboot by renaming the capsule
> > > > > file because the capsule files will be applied alphabetically.
> > > > >
> > > > > Signed-off-by: Masami Hiramatsu <masami.hiramatsu at linaro.org>
> > > > > ---
> > > > >  lib/efi_loader/efi_capsule.c |   13 +++++++++++++
> > > > >  1 file changed, 13 insertions(+)
> > > > >
> > > > > diff --git a/lib/efi_loader/efi_capsule.c b/lib/efi_loader/efi_capsule.c
> > > > > index 4463ae00fd..24a2a026a9 100644
> > > > > --- a/lib/efi_loader/efi_capsule.c
> > > > > +++ b/lib/efi_loader/efi_capsule.c
> > > > > @@ -407,12 +407,20 @@ static efi_status_t efi_capsule_update_firmware(
> > > > >       struct efi_firmware_management_protocol *fmp;
> > > > >       u16 *abort_reason;
> > > > >       efi_status_t ret = EFI_SUCCESS;
> > > > > +     bool reset = false;
> > > > >
> > > > >       /* sanity check */
> > > > >       if (capsule_data->header_size < sizeof(*capsule) ||
> > > > >           capsule_data->header_size >= capsule_data->capsule_image_size)
> > > > >               return EFI_INVALID_PARAMETER;
> > > > >
> > > > > +     if (capsule_data->flags & CAPSULE_FLAGS_INITIATE_RESET) {
> > > > > +             /* INITIATE_RESET flag requires PERSIST_ACROSS_RESET flag */
> > > > > +             if (!(capsule_data->flags & CAPSULE_FLAGS_PERSIST_ACROSS_RESET))
> > > > > +                     return EFI_INVALID_PARAMETER;
> > > > > +             reset = true;
> > > > > +     }
> > > > > +
> > > > >       capsule = (void *)capsule_data + capsule_data->header_size;
> > > > >       capsule_size = capsule_data->capsule_image_size
> > > > >                       - capsule_data->header_size;
> > > > > @@ -498,6 +506,11 @@ static efi_status_t efi_capsule_update_firmware(
> > > > >  out:
> > > > >       efi_free_pool(handles);
> > > > >
> > > > > +     if (ret == EFI_SUCCESS && reset) {
> > > > > +             log_debug("This capsule has CAPSULE_FLAGS_INITIATE_RESET. Reboot machine.\n");
> > > > > +             do_reset(NULL, 0, 0, NULL);
> > > >
> > > > I don't think this is the right place of calling do_reset().
> > > > Whenever you have processed any capsule file, you have to
> > > >  - delete the capsule file,
> > > >  - create/update "CapsuleXXXX" and "CapsuleLast" variables, and
> > > >  - modify ESRT table
> > > > before initiating a reset.
> > >
> > > Oops, indeed. Let me update the patch.
> > > Thank you!
> > >
> > > >
> > > > -Takahiro Akashi
> > > >
> > > > > +     }
> > > > > +
> > > > >       return ret;
> > > > >  }
> > > > >  #else
> > > > >
> > >
> > >
> > >
> > > --
> > > Masami Hiramatsu
> >
> >
> >
> > --
> > Masami Hiramatsu



-- 
Masami Hiramatsu