[PATCH 0/4] introduce EFI_RAM_DISK_PROTOCOL

[Masahisa Kojima]

Hi Akashi-san,

On Fri, 7 Jul 2023 at 16:16, AKASHI Takahiro <takahiro.akashi at linaro.org> wrote:
>
> On Fri, Jul 07, 2023 at 08:29:12AM +0200, Heinrich Schuchardt wrote:
> > On 7/7/23 06:00, Masahisa Kojima wrote:
> > > This series introduces the EFI_RAM_DISK_PROTOCOL implementation.
> > > The major purpose of this series is a preparation for EFI HTTP(S) boot.
> > >
> > > Now U-Boot can download the distro installer ISO image
> > > via wget or tftpboot commands, but U-Boot can not mount
> > > the downloaded ISO image.
> > > By calling EFI_RAM_DISK_PROTOCOL->register API, user can
> > > mount the ISO image and boot the distro installer.
> >
> > If I understand you correctly, with your design RAM disks will only
> > exist in the EFI sub-system.
>
> Probably, not. As Kojima-san's "dm tree" shows, there is a U-Boot
> block device (and associated partition devices) thanks to your
> efi_driver framework.

Read/Write the RAM disk is expected to be called from the EFI context, so
native U-Boot can not access the RAM disk.
 # Honestly speaking, I'm not sure how U-Boot prohibits the access to
the EFI RAM disk
    because the udevices are created for the RAM disk.

Thanks,
Masahisa Kojima

>
> > We strive to synchronize what is happening on the driver model level and
> > on the UEFI level. I would prefer a design where we have a UCLASS_BLK
> > driver ram disks and the EFI_RAM_DISK_PROTOCOL is a one method of
> > managing ram disk devices.
>
> That said, I agree to starting with U-Boot block device implementation,
> UEFI_DISK comes automatically. It benefits both U-Boot proper and
> UEFI subsystem equally as well.
> (That is also what I meant to say in my first response.)
>
> > A big issue we have is RAM management. malloc() can only assign limited
> > amount of memory which is probably too small for the RAM disk you are
> > looking at. We manage page sized memory in the EFI sub-system but this
> > is not integrated with the LMB memory checks.
>
> Not sure, is it enough simply to add some restrictions on the start size
> and the size when a memory region is specified for a raw disk?
>
> -Takahiro Akashi
>
> > The necessary sequence of development is probably:
> >
> > * Rework memory management
> > * Implement ramdisks as UCLASS_BLK driver
> > * Implement the EFI_RAM_DISK_PROTOCOL based on the UCLASS_BLK driver.
> >
> > Best regards
> >
> > Heinrich
> >
> > >
> > > Note that the installation process may not proceed
> > > after the distro installer calls ExitBootServices()
> > > since there is no hand-off process for the block device of
> > > memory mapped ISO image.
> > >
> > > The EFI_RAM_DISK_PROTOCOL was tested with the
> > > debian network installer[1] on qemu_arm64 platform.
> > > The example procedure is as follows.
> > >   => tftpboot 41000000 mini.iso
> > >   => efidebug disk load 41000000 $filesize
> > >
> > > After these commands, ISO image is mounted like:
> > >
> > >   => efidebug dh
> > >
> > >      000000007eec5910 (efiblk#0)
> > >        /RamDisk(0x41000000,4974afff,3d5abd30-4175-87ce-6d64-d2ade523c4bb,0x0)
> > >        Block IO
> > >
> > >      000000007eec6140 (efiblk#0:1)
> > >        /RamDisk(0x41000000,4974afff,3d5abd30-4175-87ce-6d64-d2ade523c4bb,0x0)/HD(1,0x01,0,0x0,0x33800)
> > >        Block IO
> > >
> > >      000000007eec62b0 (efiblk#0:2)
> > >        /RamDisk(0x41000000,4974afff,3d5abd30-4175-87ce-6d64-d2ade523c4bb,0x0)/HD(2,0x01,0,0x33800,0x10000)
> > >        Block IO
> > >        System Partition
> > >        Simple File System
> > >
> > >    => dm tree
> > >
> > >      blk           0  [ + ]   efi_blk               `-- efiblk#0
> > >      partition     0  [ + ]   blk_partition             |-- efiblk#0:1
> > >      partition     1  [ + ]   blk_partition             `-- efiblk#0:2
> > >
> > > Debian can be successfully installed with this RAM disk on QEMU.
> > >
> > > [TODO]
> > > - udevices created in ./lib/efi_driver/efi_block_device.c::efi_bl_bind()
> > >    are not removed when the efi_handle is removed.
> > >    So after unload the RAM disk, udevices still exist.
> > >    I plan to add a udevice removal process in ./lib/efi_driver/efi_uclass.c::efi_uc_stop().
> > >    In addition, I also plan to add unbind() callback in struct efi_driver_ops.
> > >
> > >
> > > [1] https://deb.debian.org/debian/dists/bookworm/main/installer-arm64/current/images/netboot/mini.iso
> > >
> > > Masahisa Kojima (4):
> > >    efi_loader: add RAM disk device path
> > >    efi_loader: add EFI_RAM_DISK_PROTOCOL implementation
> > >    cmd: efidebug: add RAM disk mount command
> > >    efi_selftest: add EFI_RAM_DISK_PROTOCOL selftest
> > >
> > >   cmd/efidebug.c                           | 117 ++++++
> > >   include/efi_api.h                        |  32 ++
> > >   include/efi_loader.h                     |   4 +
> > >   lib/efi_driver/efi_uclass.c              |   7 +-
> > >   lib/efi_loader/Kconfig                   |   6 +
> > >   lib/efi_loader/Makefile                  |   1 +
> > >   lib/efi_loader/efi_device_path_to_text.c |  14 +
> > >   lib/efi_loader/efi_ram_disk.c            | 334 +++++++++++++++
> > >   lib/efi_loader/efi_setup.c               |   6 +
> > >   lib/efi_selftest/Makefile                |   1 +
> > >   lib/efi_selftest/efi_selftest_ram_disk.c | 511 +++++++++++++++++++++++
> > >   lib/uuid.c                               |   4 +
> > >   12 files changed, 1035 insertions(+), 2 deletions(-)
> > >   create mode 100644 lib/efi_loader/efi_ram_disk.c
> > >   create mode 100644 lib/efi_selftest/efi_selftest_ram_disk.c
> > >
> > >
> > > base-commit: e2e2aea5733f0d23cd9593bbefe5c803c552dcb9
> >