BISECTED f3866909e350 ("distro_bootcmd: call EFI bootmgr even without having /EFI/boot")

AKASHI Takahiro takahiro.akashi at linaro.org
Sat Jun 12 04:05:51 CEST 2021 

Hi Matwey,

On Fri, Jun 11, 2021 at 06:08:28PM +0300, Matwey V. Kornilov wrote:
> чт, 10 июн. 2021 г. в 23:05, Heinrich Schuchardt <xypron.glpk at gmx.de>:
> >
> > On 6/7/21 7:51 PM, Matwey V. Kornilov wrote:
> > > вс, 6 июн. 2021 г. в 19:47, Heinrich Schuchardt <xypron.glpk at gmx.de>:
> > >>
> > >> On 6/6/21 6:21 PM, Heinrich Schuchardt wrote:
> > >>> On 6/6/21 5:42 PM, Matwey V. Kornilov wrote:
> > >>>> вс, 6 июн. 2021 г. в 18:20, Heinrich Schuchardt <xypron.glpk at gmx.de>:
> > >>>>>
> > >>>>> On 6/6/21 4:37 PM, Matwey V. Kornilov wrote:
> > >>>>>> Hi,
> > >>>>>>
> > >>>>>> I've found that
> > >>>>>>
> > >>>>>> f3866909e350 ("distro_bootcmd: call EFI bootmgr even without having
> > >>>>>> /EFI/boot")
> > >>>>>>
> > >>>>>> breaks running EFI application from USB device on BeagleBone Black
> > >>>>>> (am335x) device.
> > >>>>>>
> > >>>>>> With this patch I see the following:
> > >>>>>>
> > >>>>>> Booting /efi\boot\bootarm.efi
> > >>>>>> Welcome to GRUB!
> > >>>>>>
> > >>>>>> data abort
> > >>>>>> pc : [<9ce0b6d0>]          lr : [<9ffab7c7>]
> > >>>>>> reloc pc : [<7d69d6d0>]    lr : [<8083d7c7>]
> > >>>>>> sp : 9df44e28  ip : 9ffdfe90     fp : 00000003
> > >>>>>> r10: 9ffe3300  r9 : 00000000     r8 : 9df6fe88
> > >>>>>> r7 : 00000000  r6 : 9ce5da08     r5 : 9ce571f8  r4 : 9ce2c040
> > >>>>>> r3 : 00000000  r2 : 00000001     r1 : 9ce56598  r0 : 00000000
> > >>>>>> Flags: NzCv  IRQs off  FIQs on  Mode SVC_32
> > >>>>>> Code: e3500000 0a000015 e590000c eb00f96e (e5d03000)
> > >>>>>    > UEFI image [0x9ce46000:0x9cf28fff] '/efi\boot\bootarm.efi'
> > >>>>>    > Resetting CPU ...
> > >>>>>
> > >>>>> Hello Matwey,
> > >>>>>
> > >>>>> thank you for reporting the issue.
> > >>>>>
> > >>>>> $ echo 'Code: e3500000 0a000015 e590000c eb00f96e (e5d03000)' |
> > >>>>> CROSS_COMPILE=arm-linux-gnueabihf- ARCH=arm scripts/decodecode
> > >>>>>
> > >>>>> Code: e3500000 0a000015 e590000c eb00f96e (e5d03000)
> > >>>>> All code
> > >>>>> ========
> > >>>>>       0:   e3500000        cmp     r0, #0
> > >>>>>       4:   0a000015        beq     0x60
> > >>>>>       8:   e590000c        ldr     r0, [r0, #12]
> > >>>>>       c:   eb00f96e        bl      0x3e5cc
> > >>>>>      10:*  e5d03000        ldrb    r3, [r0]                <-- trapping
> > >>>>> instruction
> > >>>>>
> > >>>>> Code starting with the faulting instruction
> > >>>>> ===========================================
> > >>>>>       0:   e5d03000        ldrb    r3, [r0]
> > >>>>>
> > >>>>> Looking at the disassembly above we see that reading memory location
> > >>>>> NULL fails.
> > >>>>>
> > >>>>> We need to find out where the exception occurs. The code position is
> > >>>>> neither in bootarm.efi nor in U-Boot (9ce0b6d0 is lower than the load
> > >>>>> position of bootarm.efi, so it is below the relocated U-Boot code).
> > >>>>>
> > >>>>> Please, add the following line at the start of grub.cfg to get more
> > >>>>> output from GRUB:
> > >>>>>
> > >>>>>           debug=all
> > >>>>
> > >>>> This doesn't provide any additional output from GRUB :(
> > >>>>
> > >>>>>
> > >>>>> When building U-Boot, please, add
> > >>>>>
> > >>>>>           #define DEBUG 1
> > >>>>>
> > >>>>> in lib/efi_loader/efi_disk.c and lib/efi_loader_file.c a line before
> > >>>>> #include <common.h>.
> > >>>>
> > >>>>
> > >>>> This doesn't provide much output as well:
> > >>>>
> > >>>> Scanning disk mmc at 48060000.blk...
> > >>>> EFI: Call: efi_install_multiple_protocol_interfaces( &handle,
> > >>>> &efi_guid_device_path, diskobj->dp, &efi_block_io_guid, &diskobj->ops,
> > >>>> NULL)
> > >>>> EFI: 0 returned by efi_install_multiple_protocol_interfaces( &handle,
> > >>>> &efi_guid_device_path, diskobj->dp, &efi_block_io_guid, &diskobj->ops,
> > >>>> NULL)
> > >>>> ** Unrecognized filesystem type **
> > >>>> Scanning disk mmc at 481d8000.blk...
> > >>>> EFI: Call: efi_install_multiple_protocol_interfaces( &handle,
> > >>>> &efi_guid_device_path, diskobj->dp, &efi_block_io_guid, &diskobj->ops,
> > >>>> NULL)
> > >>>> EFI: 0 returned by efi_install_multiple_protocol_interfaces( &handle,
> > >>>> &efi_guid_device_path, diskobj->dp, &efi_block_io_guid, &diskobj->ops,
> > >>>> NULL)
> > >>>> EFI: Call: efi_install_multiple_protocol_interfaces( &handle,
> > >>>> &efi_guid_device_path, diskobj->dp, &efi_block_io_guid, &diskobj->ops,
> > >>>> NULL)
> > >>>> EFI: 0 returned by efi_install_multiple_protocol_interfaces( &handle,
> > >>>> &efi_guid_device_path, diskobj->dp, &efi_block_io_guid, &diskobj->ops,
> > >>>> NULL)
> > >>>> Found 3 disks
> > >>>
> > >>> This implies that GRUB is crashing before even accessing the file system
> > >>> (including grub.cfg).
> > >>>
> > >>> On an OrangePi PC I deleted /boot.scr and moved grubarm.efi to
> > >>> /EFI/boot/bootarm.efi. It boots without problem.
> > >>>
> > >>> What version of GRUB are you using?
> > >>> How were you booting before updating U-Boot?
> > >>> What version of U-Boot are you using where the error occurs?
> > >>> Why do you have grub in /EFI/boot/bootarm.efi and not in a distro
> > >>> specific path, e.g. /EFI/debian/grubarm.efi? /EFI/boot is typically only
> > >>> used by installers.
> > >>>
> > >>> If the boot manager is started by distroboot it may not have an
> > >>> appropriate device path. It tries to load the file given by environment
> > >>> variable $fdtfile from the boot device.
> > >>>
> > >>>   From the U-Boot console could you, please, try:
> > >>>
> > >>> 1)
> > >>> load usb 0:1 $kernel_addr_r EFI/boot/bootarm.efi
> > >>> bootefi bootmgr
> > >>>
> > >>>
> > >>> 2)
> > >>> load usb 0:1 $kernel_addr_r EFI/boot/bootarm.efi
> > >>> load usb 0:2 $fdt_addr_r dtb
> > >>> bootefi bootmgr $fdt_addr_r
> > >>>
> > >>> where you need to replace dtb by the correct device tree file and adjust
> > >>> the partition numbers.
> > >>>
> > >>> Best regards
> > >>>
> > >>> Heinrich
> > >>
> > >> To catch the earlier EFI API calls you can add
> > >>
> > >> #define DEBUG 1
> > >>
> > >> to lib/efi_loader/efi_boottime.c
> > >
> > >
> > > Welcome to GRUB!
> > >
> > >      EFI: Entry efi_locate_handle_ext(2,
> > > 9042a9de-23dc-4a38-96fb-7aded080516a, 00000000, 9df40dfc, 9ce2d660)
> >
> > EFI_GRAPHICS_OUTPUT_PROTOCOL_GUID
> >
> > This call could be from grub-core/video/efi_gop.c, check_protocol().
> >
> > >      EFI: Exit: efi_locate_handle_ext: 14
> >
> > EFI_NOT_FOUND
> >
> > >      EFI: Entry efi_open_protocol(9df5f298,
> > > 5b1b31a1-9562-11d2-8e3f-00a0c969723b, 9df40e14, 9df5f298, 00000000,
> > > 0x2)
> >
> > EFI_LOADED_IMAGE_PROTOCOL_GUID
> >
> > This call could be from grub-core/kern/efi/efi.c,
> > grub_efi_get_loaded_image().
> >
> > >      EFI: Exit: efi_open_protocol: 0
> > >      EFI: Entry efi_open_protocol(00000000,
> >
> > The parameter @handle must not be NULL.
> >
> > > 09576e91-6d3f-11d2-8e39-00a0c969723b, 9df40e14, 9df5f298, 00000000,
> >
> > EFI_DEVICE_PATH_PROTOCOL_GUID
> >
> > This could be called from grub-core/kern/efi/efi.c,
> > grub_efi_get_device_path() which is invoked from
> > grub_machine_get_bootlocation().
> >
> > > 0x2)
> >
> > EFI_INVALID_PARAMETER is returned because the handle is NULL.
> >
> > I could partially reproduce the problem by setting
> >
> >     info->device_handle = NULL;
> >
> > at the end of efi_setup_loaded_image():
> >
> > Welcome to GRUB!
> >
> >      EFI: Entry efi_open_protocol(79fdea40,
> > 5b1b31a1-9562-11d2-8e3f-00a0c969723b, 79f570e4, 79fdea40, 00000000, 0x2)
> >      EFI: Exit: efi_open_protocol: 0
> >      EFI: Entry efi_open_protocol(00000000,
> > 09576e91-6d3f-11d2-8e39-00a0c969723b, 79f570b4, 79fdea40, 00000000, 0x2)
> >      EFI: Exit: efi_open_protocol: 2
> > error: disk `,msdos2' not found.
> > grub rescue> >
> >
> > This leaves me with two questions:
> >
> > Why does GRUB not handle
> >
> >      *device = grub_efidisk_get_device_name (NULL);
> >
> > gracefully? Maybe it is because it tries to print via the graphical
> > output protocol which does not exist?
> >
> > Why is image->device_handle NULL?
> >
> > Next step is to verify that image->device_handle is really NULL.
> >
> > Please apply the following change to efi/efi_loader/efi_boottime.c
> >
> > @@ -2060,6 +2069,7 @@ efi_status_t EFIAPI efi_load_image(bool boot_policy,
> >                  free(info);
> >          }
> >   error:
> > +       printf("*** %p\n", info->device_handle);
> >          return EFI_EXIT(ret);
> >   }
> 
> 
> Booting /efi\boot\bootarm.efi
> EFI: Entry efi_load_image(0, 9df56d90,
> /VenHw(e61d73b9-a384-4acc-aeab-82e828f3628b)/UsbClass(0x13fe,0x4123,0x0,0x0,0x0)/HD(1,GPT,c3d6ff69-0ec0-4be1-91a3-e5fbeed5c4b9,0x2000,0x8000)/efi\boot\bootarm.efi,
> 82000000, 929792, 9df40ea0)
> *** 00000000

The commit message of f3866909e350 ("distro_bootcmd: call EFI bootmgr
even without having /EFI/boot") says:
    Thus, call the bootmgr before we try to boot the EFI binary inside
    the removable media path.

Given the above, your log messages suggest:
    - bootmgr didn't boot grub
    - So grub in USB:/efi/boot/bootarm.efi was booted by bootefi
      after enumerating/probing USB devices
    - Then, grub tried to open a protocol using the handle of
      the (boot?) device (i.e. USB partition) to access some file
      and got corrupted.
    - the handle was NULL because efi_dp_find_obj() in
      efi_setup_loaded_image() failed and returned NULL
      when booting grub

The last problem happened because:
- When bootmgr was first called, uefi subsystem was initialized
  and all the U-Boot disks at this point were enumerated as UEFI disks.
  Unfortunately, any USB devices had not yet enumerated due to the commit
  above.
- Then, bootefi was called after enumerating USB devices in
  distro_bootcmd, but the result was never reflected to UEFI
  subsystem due to the limitation of the current UEFI implementation.
- So grub image was successfully loaded from USB, yet UEFI never recognized
  the device.
- Please note that efi_set_bootdev() doesn't care whether the device is
  already recognized by UEFI or not.

If my guess is right, this is another example of the defect in
UEFI disk implementation.
FYI, see:
https://lists.denx.de/pipermail/u-boot/2021-June/451828.html

-Takahiro Akashi


> EFI: Exit: efi_load_image: 0
> EFI: Entry efi_start_image(9df5f298, 9df40e78, 9df40e7c)
>   EFI: Call: efi_open_protocol(image_handle, &efi_guid_loaded_image,
> &info, NULL, NULL, EFI_OPEN_PROTOCOL_GET_PROTOCOL)
>     EFI: Entry efi_open_protocol(9df5f298,
> 5b1b31a1-9562-11d2-8e3f-00a0c969723b, 9df40e60, 00000000, 00000000,
> 0x2)
>     EFI: Exit: efi_open_protocol: 0
> 
> 
> >
> > Best regards
> >
> > Heinrich
> >
> > >      EFI: Exit: efi_open_protocol: 2
> > > data abort
> > > pc : [<9ce076d0>]          lr : [<9ffa85b3>]
> > > reloc pc : [<7d69d6d0>]    lr : [<8083e5b3>]
> > > sp : 9df40e28  ip : 00000000     fp : 00000003
> > > r10: 9ffe2df8  r9 : 00000000     r8 : 9df5f298

> > > r7 : 00000000  r6 : 9ce59a08     r5 : 9ce531f8  r4 : 9ce28040
> > > r3 : 00000000  r2 : 9ffeb328     r1 : 00000000  r0 : 00000000
> > > Flags: NzCv  IRQs off  FIQs on  Mode SVC_32
> > > Code: e3500000 0a000015 e590000c eb00f96e (e5d03000)
> > > UEFI image [0x9ce42000:0x9cf24fff] '/efi\boot\bootarm.efi'
> > > Resetting CPU ...
> > >
> > >
> > >>
> > >> Best regards
> > >>
> > >> Heinrich
> > >>
> > >>
> > >
> > >
> >
> >
> 
> 
> -- 
> With best regards,
> Matwey V. Kornilov

More information about the U-Boot mailing list