Re: Unable to boot UEFI from 2.04

[David Huffman]

I have created a new thread in the bug section of the mailing list that 
provides some more detail and updated findings. I have also provided a link to 
a screen grab video that shows the entire output before grub appears to crash.

https://lists.gnu.org/archive/html/bug-grub/2020-12/msg00002.html

It is more correct to say I am unable to LOAD the kernel rather than execute 
it. For older kernels I don’t get beyond the load stage.

 - David



> On Dec 4, 2020, at 12:54 AM, Andrei Borzenkov <arvidjaar@gmail.com> wrote:
> 
> 03.12.2020 21:51, David Huffman пишет:
>> I’m not using the configuration file (for now) to remove variables. I am 
>> doing everything from the command shell and I am not trying to boot the 
>> system. Just trying to execute the kernel. Babysteps.
>> 
>> insmod efi_gop
>> linux /boot/vmlinux
>> boot
>> 
> 
> And what happens? Do you get some error message, system freezes, or
> what? What happens when running with debug enabled?
> 
>> I just compiled 2.02 and it works fine, so there is something about the 
>> updates in 2.04 that seemed to have broken.
> 
> If you have reproducible test case when 2.02 works and 2.04 does not you
> should report it to grub development list or bug tracker.
> 
>> 
>> - David
>> 
>>> On Dec 3, 2020, at 9:48 AM, Hanson Char <hanson.char@gmail.com> wrote:
>>> 
>>> I suppose you have done something like:
>>> 
>>>     sudo grub-mkconfig -o /boot/grub/grub.cfg
>>> 
>>> ?  Can you see the grub menu entries during UEFI boot?
>>> 
>>>> On Dec 3, 2020, at 9:13 AM, David Huffman <dhuffmansd@gmail.com> wrote:
>>>> 
>>>> I’ve changed the partition ID, but made no difference. I found that a 
>>>> different kernel level was able to execute.
>>>> 
>>>> I am able to execute kernel 4.19.0-12-amd64 (debian 10.5), but am unable 
>>>> to execute kernel 2.6.32-358.el6.x86_64 (RHEL 6.0). Is there a minimum 
>>>> Linux kernel level supported with grub 2.04? 
>>>> 
>>>> Thanks,
>>>> David
>>>> 
>>>>> On Dec 2, 2020, at 2:38 PM, Hanson Char <hanson.char@gmail.com> wrote:
>>>>> 
>>>>> Seems you are missing an EFI System partition which is necessary for UEFI 
>>>>> boot (EF00).
>>>>> 
>>>>> FWIW, I’ve had success creating an EFI system partition using gdisk, 
>>>>> building+installing grub 2.04 from source, and UEFI boot on both Debian 
>>>>> and Centos.
>>>>> 
>>>>> Regards,
>>>>> Hanson
>>>>> 
>>>>>> On Dec 2, 2020, at 12:55 PM, David Huffman <dhuffmansd@gmail.com> wrote:
>>>>>> 
>>>>>> I have built 2.04 from source (no errors). I have a script to create a 
>>>>>> BIOS/UEFI bootable hard drive. The grub-install command I am running 
>>>>>> succeeds without errors, but the kernel does not seem to execute when 
>>>>>> booting from UEFI (BIOS is fine).
>>>>>> 
>>>>>> Adding debug=all to the configuration file shows the execution stops at:
>>>>>> (...last three lines)
>>>>>> 
>>>>>> diskefiefidisk.c:595: reading 0x40 sectors at sector 0x48dc0 from hd1
>>>>>> diskefiefidisk.c:595: reading 0x40 sectors at sector 0x48e00 from hd1
>>>>>> diskefiefidisk.c:595: reading 0x40 sectors at sector 0x48e40 from hd1
>>>>>> 
> 
> Last three lines after *what*? Without knowing what you did it is
> impossible to interpret debug output.
> 
> Anyway, here grub reads some file from disk. When you start Linux kernel
> with boot command, it is already loaded in memory and no disk read
> should happen. You should see debug output from boot command.
> 
>>>>>> If I use the grub /usr/lib/grub/x86_64-efi files (*.mod, kernel.img, 
>>>>>> etc) poached from Debian 10.5, the system executes the kernel properly. 
>>>>>> If I just swap out the x86_64-efi directory with the files I compiled, 
>>>>>> the kernel does not execute. I am using the grub-install program I 
>>>>>> compiled from source in both cases.The only difference are the files in 
>>>>>> lib/grub/x86_64-efi/.
>>>>>> 
>>>>>> I have found references that linuxefi.mod was removed from grub and is a 
>>>>>> “distro patch”. This module appears to be missing from by source build 
>>>>>> but removing it from the debian grub files didn’t seem to make a 
>>>>>> difference.
>>>>>> 
>>>>>> Here are the commands used to build:
>>>>>> 
>>>>>> configure --with-platform=efi --target=x86_64 --disable-device-mapper 
>>>>>> —prefix=$GRUBDIR
>>>>>> make
>>>>>> make install
>>>>>> 
>>>>>> Inside $GRUBDIR I have all of the files I would expect from the build.
>>>>>> 
>>>>>> The disk has three partitions with an msdos partition table:
>>>>>> 
>>>>>> # sfdisk -l /dev/sdb
>>>>>> 
>>>>>> Disk /dev/sdb: 1305 cylinders, 255 heads, 63 sectors/track
>>>>>> Units = cylinders of 8225280 bytes, blocks of 1024 bytes, counting from 0
>>>>>> 
>>>>>> Device Boot Start     End   #cyls    #blocks   Id  System
>>>>>> /dev/sdb1   *      0+     12-     13-    102400   83  Linux      
>>>>>> /dev/sdb2         12+     25-     13-    102400   83  Linux
>>>>>> /dev/sdb3         25+   1305-   1280-  10279936   83  Linux
>>>>>> /dev/sdb4          0       -       0          0    0  Empty
>>>>>> 
>>>>>> /dev/sdb2 on /mnt type ext2 (rw)
>>>>>> /dev/sdb1 on /mnt/boot/EFI type vfat (rw)
>>>>>> 
>>>>>> Here is the grub-install commands used:
>>>>>> 
>>>>>> FOR BIOS:
>>>>>> grub-install —force --boot-directory=$TMPMNT/boot --target=i386-pc 
>>>>>> —directory=$GRUBDIR/i386-pc  /dev/sdb
>>>>>> 
>>>>>> FOR UEFI       
>>>>>> grub-install --removable --efi-directory=$TMPMNT/boot/EFI 
>>>>>> --boot-directory=$TMPMNT/boot --target=x86_64-efi 
>>>>>> --directory=$GRUBDIR/x86_64-efi /dev/sdb
>>>>>> 
>>>>>> At this point I am not sure what else to look at to find out what is 
>>>>>> different between the modules and kernel.img file I compile and what is 
>>>>>> supplied with debian. Any assistance in tracking down the problem would 
>>>>>> be appreciated.
>>>>>> 
>>>>>> 
>>>>>> - David