On Sun, 19 Apr 2020 at 14:55, Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
>
> Hi,
>
> On 4/19/20 1:59 PM, Ard Biesheuvel wrote:
> > On Sun, 19 Apr 2020 at 13:41, Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
> >>
> >> Hi Ard,
> >>
> >> I got a bug report by email that "grub-install" was throwing errors
> >> on 32 bit UEFI (64 bit kernel, mixed mode) systems starting with 5.6 .
> >> After that I noticed that the Fedora installer also threw an error about
> >> it being unable to setup the bootloader (which I could luckily skip), this
> >> was also a 32 bit UEFI mixed-mode system.
> >>
> >> So I've been running some tests with efibootmgr from the shell. Adding
> >> new entries works, but removing an entry involves unlinking the EFI var
> >> for the old which fails, here is the relevant part from a strace run:
> >>
> >> unlink("/sys/firmware/efi/efivars/Boot0000-8be4df61-93ca-11d2-aa0d-00e098032b8c") = -1 EINVAL (Invalid argument)
> >>
> >> With kernel 5.5 the same unlink succeeds, I guess this is also the
> >> cause for the grub-install and Fedora 32 installer errors. At least
> >> the Fedora 32 install error was seen on a system which already had
> >> a "Fedora" efibootmgr entry, so likely the installer tried to remove
> >> the old entry first and that caused the failure.
> >>
> >> I've also tried removing efibootmgr entires on a 64 bit UEFI system running
> >> 5.6.4 and that works fine. On 32 bit UEFI mixed-mode systems the problem
> >> persists with 5.7-rc1.
> >>
> >> Ard, any idea where to start with pinpointing the cause of this
> >> regression ?
> >>
> >
> > Can you start by trying this patch please?
> >
> > https://git.kernel.org/pub/scm/linux/kernel/git/efi/efi.git/commit/?h=urgent&id=9461aa3b44ac21668100067939d24a6ffa810eae
>
> Cool, I can confirm that this fixes deleting boot entries with
> efibootmgr.
>
Thanks for confirming, and apologies for the breakage.
> Regards,
>
> Hans
>
>
> p.s.
>
> In Torvald's tree this seems to have gotten a different commit-id:
> a4b81ccfd4caba017d2b84720b6de4edd16911a0
>
Yes, as you know, Ingo prefers to re-apply EFI patches piecemeal
rather than merging my branches directly, so the branches in the EFI
tree are not stable, unfortunately.
