Re: [PATCH 00/28] Initial implementation of bootmethod/bootflow

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

Hi Simon,

On Thu, Aug 19, 2021 at 08:25:33AM -0600, Simon Glass wrote:
> Hi Tom,
> 
> On Thu, 19 Aug 2021 at 07:59, Tom Rini <trini@konsulko.com> wrote:
> >
> > On Wed, Aug 18, 2021 at 09:45:33PM -0600, Simon Glass wrote:
> >
> > > Bootmethod and bootflow provide a built-in way for U-Boot to automatically boot
> > > an Operating System without custom scripting and other customisation:
> > >
> > >   - bootmethod - a method to scan a device to find bootflows (owned by U-Boot)
> > >   - bootflow - a description of how to boot (owned by the distro)
> > >
> > > This series provides an initial implementation of these, enable to scan
> > > for bootflows from MMC and Ethernet. The only bootflow supported is
> > > distro boot, i.e. an extlinux.conf file included on a filesystem or
> > > tftp server. It works similiarly to the existing script-based approach,
> > > but is native to U-Boot.
> > >
> > > With this we can boot on a Raspberry Pi 3 with just one command:
> > >
> > >    bootflow scan -lb
> > >
> > > which means to scan, listing (-l) each bootflow and trying to boot each
> > > one (-b). The final patch shows this.
> > >
> > > It is intended that this approach be expanded to support mechanisms other
> > > than distro boot, including EFI-related ones. With a standard way to
> > > identify boot devices, these features become easier. It also should
> > > support U-Boot scripts, for backwards compatibility only.
> > >
> > > The first patch of this series moves boot-related code out of common/ and
> > > into a new boot/ directory. This helps to collect these related files
> > > in one place, as common/ is quite large.
> > >
> > > Like sysboot, this feature makes use of the existing PXE implementation.
> > > Much of this series consists of cleaning up that code and refactoring it
> > > into something closer to a module that can be called, teasing apart its
> > > reliance on the command-line interpreter to access filesystems and the
> > > like. Also it now uses function arguments and its own context struct
> > > internally rather than environment variables, which is very hard to
> > > follow. No core functional change is included in the included PXE patches.
> > >
> > > For documentation, see the 'doc' patch.
> > >
> > > There is quite a long list of future work included in the documentation.
> > > One question is the choice of naming. Since this is a bootloader, should
> > > we just call this a 'method' and a 'flow' ? The 'boot' prefix is already
> > > shared by other commands like bootm, booti, etc.

Regarding the naming, I'm still a bit confused with bootmethod vs.
bootflow. Personally, I prefer a more intuitive name against bootmethod,
say, bootmedia or bootdevice (as the original distro_bootcmd uses).

> > > The design is described here:
> > >
> > > https://drive.google.com/file/d/1ggW0KJpUOR__vBkj3l61L2dav4ZkNC12/view?usp=sharing
> > >
> > > The series is available at u-boot-dm/bmea-working
> >
> > My question / concern is this.  Would the next step here be to
> > implement the generic UEFI boot path?  Today, I can write Fedora 34 for
> > AArch64 to a USB stick, boot U-Boot off of uSD card and the installer
> > automatically boots.  I'm sure I could do the same with the BSDs.
> > Reading the documentation left me with the impression that every OSV
> > would be expected to write something, so that their installer / OS boot.
> > But there's already standards for that, which they do, and we should be
> > implementing (and do, via the current distro_boot) or making easier to
> > enable.  Thanks!

I had the same concern.

> Here you are talking about scanning for a bootflow. It is not actually
> OS-specific. If it were, there would be no point to this :-)
> 
> If you look in the distro scripts you will see 'scan_dev_for_efi' (and
> also scan_dev_for_scrips). At least the first needs to be implemented
> a bit like the distro boot is at present. So far I have only
> implemented scan_dev_for_extlinux (plus pxe) as it is enough to show
> the concept.
> 
> Adding EFI it's likely to be about the same amount of code as distro.c
> at present, perhaps a little less since we don't have the network
> case. It is used by Fedora 34, I believe, so is easy enough for me to
> do.  But I wanted to get something out as the concept is visible in
> this series.

If I correctly understand this framework, we will have to
write several functions:
(Here I will ignore "UEFI boot manager" sequence.)

1)boot/distro_efi.c
   distro_boot_efi_setup()  ; almost the same as distro_boot_setup()
   distro_boot_efi()        ; search for /EFI/BOOT/BOOTAA64.EFI (and dtb)

2)boot/bootflow.c
   bootmethod_find_efi_in_blk()
      call distro_boot_efi_setup()
   bootflow_boot()
      CASE BOOTFLOWT_DISTRO_EFI: ; new bootflow type
          call distro_boot_efi()

3)drivers/mmc/mmc_bootmethod.c
   mmc_efi_get_bootflow()
      call bootmethod_find_efi_in_blk()
   U_BOOT_DRIVER(mmc_efi_bootmethod) =
      .name           = "mmc_efi_bootmethod",
      .id             = UCLASS_BOOTMETHOD,

4)drivers/mm/Makefile
   obj-$(CONFIG_$(SPL_TPL_)BOOTMETHOD) += mmc_bootmethod.o mmc_efi_bootmethod.o

Do you think that the above is correct?
If so, does (3) (+ (1)/(2)) inevitably increase the code size
as we need functions for all devices?

-Takahiro Akashi


> Regards,
> Simon