[PATCH] disk: part_efi: update partition table entries after write

[AKASHI Takahiro <takahiro.akashi@linaro.org>]

On Tue, Jan 26, 2021 at 04:08:13PM +0100, Heinrich Schuchardt wrote:
> On 26.01.21 14:56, Gary Bisson wrote:
> > Fixes fastboot issues when switching from mbr to gpt partition tables.
> >
> > Signed-off-by: Gary Bisson <gary.bisson@boundarydevices.com>
> > ---
> > Hi,
> >
> > I hesitated calling this a RFC as I'm not sure everyone will like the
> > idea.
> >
> > Basically the issue encountered was the following:
> > - Device with its storage flashed with a MBR partition table
> > - Run fastboot to flash a new OS (and new partition table)
> > - After flashing a GPT partition table, U-Boot couldn't find any of the
> >   partition names from the GPT
> >   -> reason is that the partition table entries aren't updated
> >   -> so U-Boot still believes it's a MBR table inside and so its parsing
> >   fails
> >
> > This commit adds an update of the table entries inside the
> > write_mbr_and_gpt_partitions() function. At first I thought maybe this
> > should be added in fastboot file (fb_mmc.c) but couldn't think of a good
> > reason why this shouldn't happened in part_efi directly.
> >
> > Let me know if you have any questions.
> >
> > Regards,
> > Gary
> > ---
> >  disk/part_efi.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/disk/part_efi.c b/disk/part_efi.c
> > index 2f922662e6e..7a24e33d189 100644
> > --- a/disk/part_efi.c
> > +++ b/disk/part_efi.c
> > @@ -867,6 +867,9 @@ int write_mbr_and_gpt_partitions(struct blk_desc *dev_desc, void *buf)
> >  		return 1;
> >  	}
> >
> > +	/* Update the partition table entries*/
> > +	part_init(dev_desc);
> > +
> >  	return 0;
> >  }
> >  #endif
> >
> 
> This change is for GPT only. Don't we need the same for MBR partition
> tables (write_mbr_partitions())?
> 
> @Simon:
> 
> Essentially this reflashing is similar to changing media. What I am
> missing in part_init() is setting a field with a counter to indicate the
> media change. The EFI_BLOCK_IO_PROTOCOL has a u32 field media_id. Would
> struct blk_desc be the right place to add such a field.

As far as "efi_loader" is concerned, none of partitions on the media,
either in case of repartitioning or changing media, will no longer be
recognized as a "efi disk" after efi subsystem is initialized.
All we can and have to do is to just reboot the system.

I think we should take one step further to fixing the issue along with
this patch. I have posted a couple of proposals in the past, for example,
see [1].

[1] https://lists.denx.de/pipermail/u-boot/2019-January/356499.html

-Takahiro Akashi

> Best regards
> 
> Heinrich
>