bug#40590:

bug#40590:

[Keyhenge]

Hi, I'm a new Guix user (well, trying to be) and am trying to 
install
Guix onto my desktop machine. Unfortunately, I'm getting four 
errors in
the installer which are preventing me from installing it. Also, 
apologies
if I format this wrong, this is my first time mailing in a bug 
report.

Info about my machine:
CPU: AMD Ryzen 2700X
GPU: AMD 5700 XT
SSD: Samsung SSD 860 2TB
Memory: 32GB

The first error comes when trying to partition the SSD via the 
graphical
installer, choosing a guided installation. This will happen 
choosing with
or without encryption, and with or without a separate /home 
partition.

In ice-9/boot-9.scm:
  829:9 19 (catch srfi-34 #procedure 2ea60c0 at
./gnu/installer/steps.scm:144:7()> #procedure 2e251e0 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 18 (catch srfi-34 #procedure 2eb8440 at
./gnu/installer/steps.scm:144:7()> #procedure 2f27e60 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 17 (catch srfi-34 #procedure 2eb8240 at
./gnu/installer/steps.scm:144:7()> #procedure 2f27e10 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 16 (catch srfi-34 #procedure 2eb8040 at
./gnu/installer/steps.scm:144:7()> #procedure 2f27dc0 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 15 (catch srfi-34 #procedure 286e200 at
./gnu/installer/steps.scm:144:7()> #procedure 17468c0 at
./gnu/installer/steps.scm:144:7 (key c)> _)
In ./gnu/installer/steps.scm
  182:21 14 (_)
In ./gnu/installer/newt/partition.scm:
  755:33 13 (run-partitioning-page)
In ./gnu/installer/parted.scm:
  1010:14 12 (auto-partition! #<<disk> bytestructure: 
  #<bytestructure
0x185ba50>>#:scheme _)
  863:17 11 (loop _ 3905980593 1048575)
  771:25 10 (mkpart #<<disk> bytestructure: #<bytestructure 
  0x185ba50>> _
#:previous-partition _)
In parted/structs.scm
  552:19 9 (pointer->partition _)
  132:3 8 (pointer->bytestructure #<pointer 0x0> 
  #<bytestructure-descriptor
0x31071c0>)
In unknown file:
  7 (pointer->bytevector #<pointer 0x0> 88 #<undefined> 
  #<undefined>)
In ice-9/boot-9.scm:
  751:25 6 (dispatch-exception 5 null-pointer-error 
  ("pointer->bytevector"
"null pointer dereference" () ()))
In ice-9/eval.scm:
  619:8 5 (_ #(#(#<directory (guile-user) 1253140> #<<installer> 
  name: newt
init: #<procedure init ()> exit: #<procedure exit ()> exit-error:
#<procedure exit-error (file key args)> final-...>) ...))
  619:8 4 (_ #(#(#(#<directory (guile-user) 1253140> #<<installer> 
  name:
newt
init: #<procedure init ()> exit: #<procedure exit ()> exit-error:
#<procedure exit-error (file key args)> f...>) ...) #))
In ice-9/ports.scm:
  462:17 3 (call-with-output-file _ _ #:binary _ #:encoding _)
In ice-9/eval.scm:
  619:8 2 (_ #(#(#<directory (guile-user) 1253140> 
  null-pointer-error
("pointer->bytevector" "null pointer dereference" () ())) 
#<output:
/tmp/last-installer-error 12>))
  159:9 1 (_ #(#(#<directory (guile-user) 1253140> 
  null-pointer-error
("pointer->bytevector" "null pointer dereference" () ())) 
#<output:
/tmp/last-installer-error 12>))
In unknown file:
  0 (make-stack #t)
ice-9/eval.scm:159:9: In procedure pointer->bytevector: null 
pointer
dereference

The second error I get unfortunately doesn't have an error 
message, and
comes when trying to do a manual partition via the graphical 
installer.
I'll be able to make a boot partition (around 512 MB) just fine, 
but when
trying to make a partition on the rest of the disk (or even most 
of it),
the installer will immediately reset and take me back to picking 
the
language.

I then tried the CLI installation method which fared quite a bit 
better.
For one, I got past the partitioning stage, mounted the partitions 
to the
appropriate places, and made a configuration. It even gets through 
most of
the installation progress, but unfortunately fails on what seems 
to be the
very last step, initializing the operating system under /mnt, 
specifically
populating /mnt. bootloader-installer fails with the following 
error:

Initializing operating system under '/mnt'...
copying to '/mnt'...
populating '/mnt'...
error: '/gnu/store/[hash]-grub-efi-2.02/sbin/grub-install 
--boot-directory
/mnt/boot --bootloader-id=Guix --efi-directory /mnt/boot/efi' 
exited with
status 1; output follows:
  Installing for x86_64-efi platform.
  Could not prepare Boot variable: No such file or directory
  /gnu/store/[hash]-grub-efi-2.02/sbin/grub-install: error: 
  efibootmgr
failed to register the boot entry: Input/output error.
guix system: error: failed to install bootloader
/gnu/store/[hash]-bootloader-installer

If you head back to the graphical install after this, you'll be 
treated to
another error when trying to pick your locale/keymap.

In ice-9/eval.scm:
  619:8 19 (_ #(#(#<directory (guile-user) 133e140> #<<installer> 
  name: newt
init: #<procedure init ()> exit: #<procedure exit ()> exit-error:
#<procedure exit-error (file key args)> final-...>) #))
In ice-9/boot-9.scm:
  829:9 18 (catch #t #<procedure 294a2e0 at ice-9/eval.scm:330:13 
  ()>
#<procedure 294a140 at ice-9/boot-9.scm:1048:2 _> _)
In ice-9/eval.scm:
  619:8 17 (_ #(#(#(#<directory (guile-user) 133e140> 
  #<<installer> name:
newt
init: #<procedure init ()> exit: #<procedure exit ()> exit-error:
#<procedure exit-error (file key args)> fi...> ...)) ...))
  626:19 16 (_ #(#(#(#<directory (guile-user) 133e140> 
  #<<installer> name:
newtinit: #<procedure init ()> exit: #<procedure exit ()> 
exit-error:
#<procedure exit-error (file key args)> fi...> ...)) ...))
In ./gnu/installer/steps.scm:
  189:6 15 (run-installer-steps #:steps _ #:rewind-strategy _ 
  #:menu-proc _)
In ice-9/boot-9.scm:
  829:9 14 (catch srfi-34 #<procedure 2e79b80 at
./gnu/installer/steps.scm:144:7()> #<procedure 2df4320 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 13 (catch srfi-34 #<procedure 2efaf80 at
./gnu/installer/steps.scm:144:7()> #<procedure 2df4050 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 12 (catch srfi-34 #<procedure 2efad40 at
./gnu/installer/steps.scm:144:7()> #<procedure 2df4000 at
./gnu/installer/steps.scm:144:7 (key c)> _)
  829:9 11 (catch srfi-34 #<procedure 2efab40 at
./gnu/installer/steps.scm:144:7()> #<procedure 2ff1fa0 at
./gnu/installer/steps.scm:144:7 (key c)> _)
In ./gnu/installer/keymap.scm:
  163:7 8 (_ _)
In unknown file:
  7 (scm-error misc-error #f "~A" ("Unable to locate keymap update 
  file")
#f)
In ice-9/boot-9.scm:
  751:25 6 (dispatch-exception 4 misc-error (#f "~A" ("Unable to 
  locate
keymap update file") #f))
In ice-9/eval.scm:
  619:8 5 (_ #(#(#<directory (guile-user) 133e140> #<<installer> 
  name: newt
init: #<procedure init ()> exit: #<procedure exit ()> exit-error:
#<procedure exit-error (file key args)> final-...>) ...))
  619:8 4 (_ #(#(#(#<directory (guile-user) 133e140> #<<installer> 
  name:
newt
init: #<procedure init ()> exit: #<procedure exit ()> exit-error:
#<procedure exit-error (file key args)> f...>) ...) #))
In ice-9/ports.scm:
  462:17 3 (call-with-output-file _ _ #:binary _ #:encoding _)
In ice-9/eval.scm:
  619:8 2 (_ #(#(#<directory (guile-user) 133e140> misc-error (#f 
  "~A"
("Unable to locate keymap update file") #f)) #<output:
/tmp/last-installer-error 17>))
  159:9 1 (_ #(#(#<directory (guile-user) 133e140> misc-error (#f 
  "~A"
("Unable to locate keymap update file") #f)) #<output:
/tmp/last-installer-error 17>))
In unknown file:
  0 (make-stack #t)
ice-9/eval.scm:159:9: Unable to locate keymap update file

Some side notes:
- I've redownloaded, verified through gpg, and dd'd the iso onto 
  my USB a
couple of times to ensure it wasn't just a bad download
- While I have NixOS on another SSD, these errors persist even 
  when I
physically remove it, so I don't think there's any /dev/sdX 
ordering issues
involved.
- Doing the partitions via CLI, then going back through the manual 
  install
process results in normal behavior up until the grub-install error
- Restarting the graphical installer after the grub-install error, 
  then
going to the CLI will let you see that all of the partitions have 
been
unmounted, except for the root partition which will instead be 
mounted to
/tmp and have a guix-install folder.
- When viewing what's under /mnt/boot/efi, you'll get the path
/mnt/boot/efi/EFI/Guix/grubx64.efi. This seems like a weird path 
to me with
efi/EFI, but I don't really have experience with tinkering with 
grub.
Changing the path to be /mnt/boot/efi/Guix/grubx64.efi doesn't 
change the
bootloader error.
- I've transcribed the above errors by hand, but since they seem 
  to be
exporting the output to /tmp, there's got to be a better way. Is 
it
possible to email this list from inside the installation image?

bug#40590:

[Mathieu Othacehe]

Hello,

> The first error comes when trying to partition the SSD via the graphical
> installer, choosing a guided installation. This will happen choosing with
> or without encryption, and with or without a separate /home partition.

Sorry for all the errors you encountered, but many thanks for the
feedback!  Let's try to see what went wrong.

Did you use 1.0.1 image or a more recent one?

>  619:8 2 (_ #(#(#<directory (guile-user) 1253140> null-pointer-error
> ("pointer->bytevector" "null pointer dereference" () ())) #<output:
> /tmp/last-installer-error 12>))
>  159:9 1 (_ #(#(#<directory (guile-user) 1253140> null-pointer-error
> ("pointer->bytevector" "null pointer dereference" () ())) #<output:
> /tmp/last-installer-error 12>))
> In unknown file:
>  0 (make-stack #t)
> ice-9/eval.scm:159:9: In procedure pointer->bytevector: null pointer
> dereference

Ok, so I suspect that this one is caused by the size of your hard
drive (2TiB). It's an unsolved issue[1] that fell into the cracks.

> error: '/gnu/store/[hash]-grub-efi-2.02/sbin/grub-install --boot-directory
> /mnt/boot --bootloader-id=Guix --efi-directory /mnt/boot/efi' exited with
> status 1; output follows:
>  Installing for x86_64-efi platform.
>  Could not prepare Boot variable: No such file or directory
>  /gnu/store/[hash]-grub-efi-2.02/sbin/grub-install: error: efibootmgr
> failed to register the boot entry: Input/output error.
> guix system: error: failed to install bootloader
> /gnu/store/[hash]-bootloader-installer

Did you mount your EFI partition to "/mnt/boot/efi" before starting the
install?

> If you head back to the graphical install after this, you'll be treated to
> another error when trying to pick your locale/keymap.

Well, the graphical installer does not support very well to be
restarted, especially in 1.0.1 release.

> - I've transcribed the above errors by hand, but since they seem to be
> exporting the output to /tmp, there's got to be a better way. Is it
> possible to email this list from inside the installation image?

That's heroic! If you have another machine around you may use SSH to get
the error file (located in /tmp/last-installer-error). Otherwise you can
take some pictures of your screen and send them as attachment.

Now, if you could try again with:

* The 1.1.0-rc2 release available here[2].

* This custom release[3]: 1.1.0-rc2 + the patch suggested by Danny for
partitions > 1TiB.

that would be really appreciated :)

Thanks,

Mathieu

[1]: https://issues.guix.gnu.org/issue/36783
[2]: https://web.fdn.fr/~lcourtes/software/guix/1.1.0rc2/
[3]: Via ipfs: QmbbkQZc7fbsDrutNwVuvpFep3dBjZx6muni3WgXjRneq7

bug#40590:

[Keyhenge]

First off, thanks for your help so far!

> Did you use 1.0.1 image or a more recent one?
I was using the Guix System 1.0.1 x86_64 image here:
https://guix.gnu.org/download/

> Did you mount your EFI partition to "/mnt/boot/efi" before 
> starting the
> install?
Yep. Not mounting it yields a different error (cannot find 
canonical
path of /mnt/boot/efi)
I'll also confirm I'm setting the esp flag on the partition.

> Now, if you could try again with:
>
> * The 1.1.0-rc2 release available here[2].
- The graphical installer displays normal behavior up to when 
  making the
partitions, where it resets the installation without displaying an
error
- Installation through TTY has the same behavior as before, 
  efibootmgr
failed to register the boot entry: Input/output error

> * This custom release[3]: 1.1.0-rc2 + the patch suggested by 
> Danny for
> partitions > 1TiB.
- Graphical installation has an oddity, my Samsung drive has 
  disappeared
from the disk selection in both guided and manual partitioning
- Strangely, my WD Blue 2TB drive (also an SSD) with the NixOS
installation shows up just fine
- While it is possible to access the Samsung drive from the TTY, 
  and
even wipe it/make new partitions, trying to populate the 
partitions with
filesystems will result in them saying they're in use or already
mounted. I suspect this is also why it's not showing up in the 
graphical
installation.

bug#40590:

[Keyhenge]

Quick update on the previous email:

> * This custom release[3]: 1.1.0-rc2 + the patch suggested by 
> Danny for
> partitions > 1TiB.
> [3]: Via ipfs: QmbbkQZc7fbsDrutNwVuvpFep3dBjZx6muni3WgXjRneq7
It occured to me that the Samsung drive might have been corrupted 
at
some point, so I dd'd /dev/zero into the first few gigabytes of 
the
drive to completely wipe the partition table. Sure enough, it 
showed up
on the partitioning stage of the graphical installer and I was 
able get
all the way through to the end... except for the same efibootmgr 
error.

bug#40590:

[Mathieu Othacehe]

Hey,

>> * This custom release[3]: 1.1.0-rc2 + the patch suggested by Danny for
>> partitions > 1TiB.
>> [3]: Via ipfs: QmbbkQZc7fbsDrutNwVuvpFep3dBjZx6muni3WgXjRneq7
> It occured to me that the Samsung drive might have been corrupted at
> some point, so I dd'd /dev/zero into the first few gigabytes of the
> drive to completely wipe the partition table. Sure enough, it showed up
> on the partitioning stage of the graphical installer and I was able get
> all the way through to the end... except for the same efibootmgr error.

Ok! So it seems that the additional patch is improving the situation. I
also fear that 1.1.0-rc2 is having the same error as 1.0.1 but is
swallowing it. I will investigate it further.

Regarding the efibootmgr error, I'm not really familiar with EFI. Can
you check that efivarfs is mounted as suggested here[1]?

Thanks,

Mathieu

[1]: https://unix.stackexchange.com/questions/379774/grub-installation-failed

bug#40590:

[Keyhenge]

Progress! But not quite there yet.

>> It occured to me that the Samsung drive might have been 
>> corrupted at
>> some point, so I dd'd /dev/zero into the first few gigabytes of 
>> the
>> drive to completely wipe the partition table. Sure enough, it 
>> showed up
>> on the partitioning stage of the graphical installer and I was 
>> able get
>> all the way through to the end... except for the same 
>> efibootmgr error.
>
> Ok! So it seems that the additional patch is improving the 
> situation. I
> also fear that 1.1.0-rc2 is having the same error as 1.0.1 but 
> is
> swallowing it. I will investigate it further.
>
> Regarding the efibootmgr error, I'm not really familiar with 
> EFI. Can
> you check that efivarfs is mounted as suggested here[1]?
>
> [1]:
> https://unix.stackexchange.com/questions/379774/grub-installation-failed

Followed the advice on the SE answer, specifically:
$ mount -t efivarfs efivarfs /sys/firmware/efi/efivars
$ rm /sys/firmware/efi/efivars/dump-*
And it successfully installed!.. sort of.

Guix now boots to this screen[1] and then hard freezes.

Just in case its relevant, I did the installation twice (both 
through
TTY in order to run the mount/rm commands):
1. 2 partitions, boot and encrypted root w/ swapfile, using 
desktop.scm
as a base
2. 3 partitions, boot, root, and a swap partition, using
lightweight-desktop.scm as a base
Both had the same results.

[1]:
https://imgur.com/lGWmzfL

bug#40590:

[Mathieu Othacehe]

Hello,

> Just in case its relevant, I did the installation twice (both through
> TTY in order to run the mount/rm commands):
> 1. 2 partitions, boot and encrypted root w/ swapfile, using desktop.scm
> as a base
> 2. 3 partitions, boot, root, and a swap partition, using
> lightweight-desktop.scm as a base
> Both had the same results.

Ok, thanks for your perseverance! I see that someone in experimenting
system freeze with a hardware close to yours[1].

Could you try adding the following kernel parameters (via Grub by
pressing 'e', then appending the option at the end of the cmdline):

* nomodeset

and if it doesn't work:

* amd_iommu=pt ivrs_ioapic[32]=00:14.0 iommu=soft

Thanks,

Mathieu

[1]: https://github.com/NixOS/nixpkgs/issues/54378

bug#40590:

[Keyhenge]

Sorry for the late reply on this.

> Could you try adding the following kernel parameters (via Grub 
> by
> pressing 'e', then appending the option at the end of the 
> cmdline):
>
> * nomodeset
>
> and if it doesn't work:
>
> * amd_iommu=pt ivrs_ioapic[32]=00:14.0 iommu=soft

The amd option didn't work, but nomodeset allowed me to get to a 
shell,
at which point I could start experimenting with various settings 
to see
if I could get a graphical environment. I finally did by switching 
over
to the proprietary kernel (which points to a graphical driver 
issue with
AMD Navi cards in the libre kernel), but that's obviously less 
than ideal.

Still, it means I can finally start using the system in 
earnest. Thanks
for all your help!

bug#40590:

[pelzflorian (Florian Pelz)]

On Sat, Apr 18, 2020 at 01:47:11PM -0400, Keyhenge wrote:
> The amd option didn't work, but nomodeset allowed me to get to a shell,
> at which point I could start experimenting with various settings to see
> if I could get a graphical environment. I finally did by switching over
> to the proprietary kernel (which points to a graphical driver issue with
> AMD Navi cards in the libre kernel), but that's obviously less than ideal.
> 

If you have no problems sacrificing all the acceleration by your AMD
GPU, you can try using libre uvesafb for Xorg graphics.  What I wrote
at <https://lists.gnu.org/archive/html/bug-guix/2020-04/msg00320.html>
was not necessary there but maybe is in your case.  That said, this
situation is far from ideal, but I don’t think the AMD firmware will
be freed (if there is no other way).

Regards,
Florian

bug#40590:

[Mathieu Othacehe]

Hello,

Thanks for your answer!

> The amd option didn't work, but nomodeset allowed me to get to a shell,
> at which point I could start experimenting with various settings to see
> if I could get a graphical environment. I finally did by switching over
> to the proprietary kernel (which points to a graphical driver issue with
> AMD Navi cards in the libre kernel), but that's obviously less than ideal.
>
> Still, it means I can finally start using the system in earnest. Thanks
> for all your help!

It is deserved after all the efforts :p. I sent patches for the issues
you had (partitioning issue[1] and efivarfs[2] support). Now for the
graphical issue, if you find a workaround, maybe using uvesafb as
suggested by Florian, please let us know.

Thanks,

Mathieu

[1]: https://issues.guix.gnu.org/issue/36783
[2]: https://lists.gnu.org/archive/html/guix-patches/2020-04/msg00707.html.