Kernel 3.7 doesn't boot using EFI Stub

cfr · 2013-04-06 16:14:21

mcloaked wrote:

I have been using rEFInd without any problems - but initially I did not understand how to install the various parts - the better option would be to have really clear and explicit guidance on getting rEFInd installed in the beginners guide - including the potential problems with efibootmgr and using the uefi shell as an alternative to write NVRAM entries. If I get some time I might try to write some wiki advice for rEFInd. I think a clean up of the guidance would be better than simply advising another alternative bootloader?

The trouble is that that still won't help people bitten by the bug discussed in this thread because people are unable to boot even though everything is installed and configured correctly. That's not to say that clarifying the instructions would not be a service to the community - I definitely think that would be a great thing to do because there certainly are people who fail because they've tried to follow the instructions but missed some detail. But that still leaves the people for whom it just won't work no matter what.

dhave · 2013-04-08 16:01:53

Arch kernel 3.8.6 is booting fine using rEFInd.

mcloaked · 2013-04-08 16:13:32

dhave wrote:

Arch kernel 3.8.6 is booting fine using rEFInd.

And for me also - on two machines both using rEFInd.

srs5694 · 2013-04-08 16:39:53

I just got an e-mail saying that bug #34358 has been closed because it's been fixed, but there's no explanation on the bug page about what the root cause of the problem was.

mcloaked · 2013-04-08 16:59:44

srs5694 wrote:

I just got an e-mail saying that bug #34358 has been closed because it's been fixed, but there's no explanation on the bug page about what the root cause of the problem was.

The bug was closed by Tobias Powalowski (tpowa) - so presumably he knows why he closed it and what was fixed?

chetwisniewski · 2013-04-08 18:02:02

mcloaked wrote:

dhave wrote:
Arch kernel 3.8.6 is booting fine using rEFInd.
And for me also - on two machines both using rEFInd.

All good here with 3.8.6 from testing.

cfr · 2013-04-09 02:10:13

3.8.6 will not boot for me - even with the alternative rEFInd binary compiled with the different EFI .

srs5694 · 2013-04-09 16:05:09

cfr wrote:

3.8.6 will not boot for me - even with the alternative rEFInd binary compiled with the different EFI .

Then I recommend you re-open this bug report. I e-mailed Tobias Powalowski, asking what the ultimate fix was, and got this reply back:

Tobias Powalowski wrote:

I closed this because users requested it.
The new kernels are now compiled with gcc 4.8, 3.8.6 and LTS 3.0.72
I cannot help you much on this issue, I don't have any UEFI system.
I already changed to tiano-core building because gnu-efi-libs seems to
trigger bugs in gummiboot too.
grub2 is the only bootloader that seems not to trigger issues with uefi
at the moment.

Clearly, this issue needs more attention from Arch developers/packagers.

I'd like to reiterate a suggestion I made some time ago in this thread: There are at least two Arch packagers who release various Arch kernel packages. Those with the problem should look for a pattern in whose kernels are causing problems. For that, you need to study who packaged each kernel and correlate that with whether it caused problems for you. So: Everybody who's experienced this problem should check their kernel packages to see who packaged them (as revealed by the "Packager:" line in "pacman -Si" output) and compare that with whether it causes a problem. Somebody should then collect all this data and build a 2D database (user/motherboard vs. packager), with good/bad marks for each. With any luck a pattern will emerge.

Also, I note that Tobias implied that there's been a recent change in the compiler used to compile the kernels, to GCC 4.8. I don't have more data, but it's conceivable that this change is at least part of the cause of the problem. If somebody here is familiar with kernel compilation, some tests using two or more GCC versions is worth doing.

cfr · 2013-04-09 19:35:21

srs5694 wrote:

cfr wrote:
3.8.6 will not boot for me - even with the alternative rEFInd binary compiled with the different EFI .
Then I recommend you re-open this bug report.

Done. Though the fact that the last comment on the bug was me saying it wasn't fixed for me does not make me very hopeful that my request will be granted .

I'd like to reiterate a suggestion I made some time ago in this thread: There are at least two Arch packagers who release various Arch kernel packages. Those with the problem should look for a pattern in whose kernels are causing problems. For that, you need to study who packaged each kernel and correlate that with whether it caused problems for you. So: Everybody who's experienced this problem should check their kernel packages to see who packaged them (as revealed by the "Packager:" line in "pacman -Si" output) and compare that with whether it causes a problem. Somebody should then collect all this data and build a 2D database (user/motherboard vs. packager), with good/bad marks for each. With any luck a pattern will emerge.
Also, I note that Tobias implied that there's been a recent change in the compiler used to compile the kernels, to GCC 4.8. I don't have more data, but it's conceivable that this change is at least part of the cause of the problem. If somebody here is familiar with kernel compilation, some tests using two or more GCC versions is worth doing.

Does this mean that particular kernel versions are compiled by different developers or that a particular version is compiled by one but another by the other?

Note that pacman -Qi would be a better command to get this info as that will make sure that the information matches the currently installed package and not an update.

Maybe a wiki page would be a good way to collate this information? I will try to set a page up if people think that's a good way to do it.

I have never compiled a kernel and wouldn't trust any results based on my experiments in that direction.

srs5694 · 2013-04-09 20:20:27

FWIW, I've now run into this problem on my 32-bit Mac Mini and at least begun tracing its cause: It seems that the exit_boot() call in the Linux kernel's arch/x86/boot/compressed/eboot.c file never returns. Replacing this file with the equivalent file from a 3.6 kernel gives me a bootable kernel. I haven't yet checked the revision history of this file, but unless there have been a lot of back-and-forth changes in the 3.7.x and 3.8.x series, it seems unlikely that this is the whole and complete cause; I still suspect that it's interacting with something about the build environment.

I've e-mailed Matt Fleming, the EFI stub loader's author, alerting him to this problem. With any luck that will result in some action.

Last edited by srs5694 (2013-04-09 20:21:03)

cfr · 2013-04-09 21:12:29

FWIW, the problem does not correlate with the developer building the kernel in my case. Tobias Powalowski built all of the kernels I've had issues with but he also built a lot of those which have been fine, including the last kernel that worked for me:

linux-3.7.10-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # OK last working kernel
linux-3.7.4-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # OK
linux-3.7.5-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # OK
linux-3.7.6-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # OK
linux-3.7.7-1-x86_64.pkg.tar.xz-PKGINFO:packager = Evangelos Foutras <evangelos@foutrelis.com> # OK
linux-3.7.8-1-x86_64.pkg.tar.xz-PKGINFO:packager = Evangelos Foutras <evangelos@foutrelis.com> # OK
linux-3.7.9-1-x86_64.pkg.tar.xz-PKGINFO:packager = Evangelos Foutras <evangelos@foutrelis.com> # OK
linux-3.7.9-2-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # OK
linux-3.8.3-2-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # first bad kernel
linux-3.8.4-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # bad
linux-3.8.5-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # bad
linux-3.8.6-1-x86_64.pkg.tar.xz-PKGINFO:packager = Tobias Powalowski <tpowa@archlinux.org> # bad

Unlike some people, I haven't seen the good-bad-good-bad pattern: everything 3.7.* was fine; nothing 3.8.* works.

Firmware is Phoenix SecureCore Tiano. I have only tested 3.8.6-1 with firmware version 1.15. The rest were tested with version 1.16. (Note: this IS the correct way around despite the apparent improbability.)

EDIT: 3.8.6-1 now tested with firmware version 1.18 with same result. Stops in the middle of printing the kernel command parameters. Boots fine with GRUB. (Shouldn't have updated - now I get an error on every boot although the upgrade went fine. I think this is Lenovo's fault, though. All works fine after a few beeps, an error message and a pause. Pah.)

Last edited by cfr (2013-04-10 23:53:29)

dhave · 2013-04-09 21:59:01

srs5694 wrote:

I'd like to reiterate a suggestion I made some time ago in this thread: There are at least two Arch packagers who release various Arch kernel packages. Those with the problem should look for a pattern in whose kernels are causing problems. For that, you need to study who packaged each kernel and correlate that with whether it caused problems for you. So: Everybody who's experienced this problem should check their kernel packages to see who packaged them (as revealed by the "Packager:" line in "pacman -Si" output) and compare that with whether it causes a problem. Somebody should then collect all this data and build a 2D database (user/motherboard vs. packager), with good/bad marks for each. With any luck a pattern will emerge.

O.K., here's some data. "Yes" means booted successfully in UEFI mode with rEFInd. "No" means did not boot successfully in UEFI with rEFInd. The name at the end of each line is the packager as listed in the output of "pacman -Qi linux".

3.8.6-1 yes Tobias Powaloski

3.8.5-1 yes Tobias Powaloski

3.8.4-1 yes Tobias Powaloski

3.8.3-2 no Tobias Powaloski

3.7.10-1 no Tobias Powaloski

3.7.9-2 yes Tobias Powaloski

3.7.9-1 no Evangelos Foutras

3.7.8-1 no Evangelos Foutras

3.7.7-1 yes Evangelos Foutras

3.7.6-1 no Tobias Powaloski

3.7.5-1 yes Tobias Powaloski

3.7.4-1 yes Tobias Powaloski

3.7.3-1 yes Tobias Powaloski

Machine is a Lenovo Thinkpad T420 running UEFI version 1.44. All tests were done in one session. No other changes were made in the test system's configuration.

Last edited by dhave (2013-04-09 21:59:46)

cfr · 2013-04-09 22:38:37

How do you know what version of UEFI you have? Is that different from the firmware version?

dhave · 2013-04-09 23:04:40

cfr wrote:

How do you know what version of UEFI you have? Is that different from the firmware version?

That's the firmware version I'm citing, AFAIK. The version is shown on my BIOS setup screen, but it's also in the release notes of the most recent BIOS-UEFI update from Lenovo.

Last edited by dhave (2013-04-09 23:07:15)

FoxBuru · 2013-04-10 07:45:59

Hi all:

I repost this message here, because i think it could help to solve the problem.

I have an Inspiron 14z (5423). I'm trying since i've installed Arch to put it to boot with UEFI, but since the first time i configured it with a UEFI bootloader, i've had the same problem. Tried with gummiboot, GRUB, rEFInd, EFILINUX and almost all UEFI capable bootloaders.

Actually all of those loaders start on the PC, but in the moment the kernel starts charging, the full PC just hangs up and stays there forever. I have to apply a "Quick n' Dirty" solution to get it somehow working: I copy a Grub efi bootloader from the Ubuntu 12.10 live CD and put it on a USB key. Once load, i load the kernel first using the linux instructions, and then some linuxefi one, as below indicated:

set root=(hd2,gpt2)
linux /vmlinuz-linux root=/dev/sdb3 ro  quiet
initrd /initramfs-linux.img
boot

[Here appears some error of UEFI, that cannot terminate UEFI mode, or something like that. I'll take a screencap later]

linuxefi /vmlinuz-linux root=/dev/sdb3 ro  quiet
initrdefi /initramfs-linux.img
boot

And after that, Archlinux starts booting normally. If i don't do that steps in the exact way, i cannot start Arch in any way. In fact, i've tried to put the same "grubx64.efi" on the EFI partition of Arch installation, but I don't get the same results than booting from USB.

Some ideas that could be causing this?

Greetings.

P.S: With this method, i've managed to boot kernel since 3.5 and upwards.

Last edited by FoxBuru (2013-04-10 07:47:00)

srs5694 · 2013-04-10 14:24:55

FoxBuru, GRUB doesn't rely on the EFI stub loader to work, and this thread is explicitly about problems with the EFI stub loader in recent kernels. Although using GRUB is a valid alternative to using the EFI stub loader, and one that may be necessary for some users with the problem under discussion in this thread, problems with GRUB (such as you're reporting) are best discussed elsewhere, IMHO. This thread is long and complex enough as it is without adding a completely unrelated problem to the mix under discussion. Therefore, I recommend you start a new thread about your GRUB problem.

Incidentally, the bug report on this problem has been re-opened.

jordi · 2013-04-10 20:02:41

MacBook Air 5.1

3.7.9        - yes
3.7.10     - untested
3.8.3       - yes
3.8.4       - no
3.8.5       - no
3.8.6       - yes

Last edited by jordi (2013-04-10 20:03:20)

srs5694 · 2013-04-11 16:53:38

I've received a reply from Matt Fleming, the author of the EFI stub loader code. He was unaware of the problem, and he says he wants samples of the kernel binaries that do and that do not work. I went rummaging through an Arch Linux repository site, but I could only find the latest (3.8.6-1) kernel. If somebody could post links to working and non-working binaries, I can forward the data; or you could e-mail him directly (matt at console-pimps.org).

Scimmia · 2013-04-11 17:49:46

Check here: https://wiki.archlinux.org/index.php/Do … ckages#ARM

srs5694 · 2013-04-11 20:40:41

Scimmia wrote:

Check here: https://wiki.archlinux.org/index.php/Do … ckages#ARM

Thanks for the pointer. I've forwarded links to 3.8.4, 3.8.5, and 3.8.6 from that site to Matt Fleming.

dhave · 2013-04-11 20:44:43

srs5694 wrote:

Scimmia wrote:
Check here: https://wiki.archlinux.org/index.php/Do … ckages#ARM
Thanks for the pointer. I've forwarded links to 3.8.4, 3.8.5, and 3.8.6 from that site to Matt Fleming.

Kernels 3.8.4, 3.8.5, and 3.8.6 all work fine for me.

The troublesome ones for me are 3.8.3-2, 3.7.10, 3.7.9-1 (3.7.9-2 was O.K.), 3.7.8 and 3.7.6.

cfr · 2013-04-11 20:46:31

If the ARM isn't suitable, I have all of the kernels which haven't worked for me and a lot that have in pacman's cache. Let me know if you need them where might be a good place to post them. I don't have a website or anything like that so I'm not sure where to put them.

cfr · 2013-04-11 20:48:29

srs5694 wrote:

Scimmia wrote:
Check here: https://wiki.archlinux.org/index.php/Do … ckages#ARM
Thanks for the pointer. I've forwarded links to 3.8.4, 3.8.5, and 3.8.6 from that site to Matt Fleming.

3.8.3 also failed for me. All 3.7.* worked fine.

I'm not quite sure what we are providing in terms of "do" and "don't" work since it seems as if different kernels work for different people. So kernel 1 may work for me and not you but kernel 2 work for you and not me... It isn't a consistent pattern.

dhave · 2013-04-11 20:49:23

With different people getting different results, it really looks like this problem is manufacturer specific. Either the UEFI "standard" is a moving target or the specs aren't really being followed uniformly.

Last edited by dhave (2013-04-11 20:50:16)

dhave · 2013-04-15 15:13:46

Arch kernel 3.8.7 is booting fine via rEFInd on my uefi-only machine.

Arch Linux

#176 2013-04-06 16:14:21

Re: Kernel 3.7 doesn't boot using EFI Stub

#177 2013-04-08 16:01:53

Re: Kernel 3.7 doesn't boot using EFI Stub

#178 2013-04-08 16:13:32

Re: Kernel 3.7 doesn't boot using EFI Stub

#179 2013-04-08 16:39:53

Re: Kernel 3.7 doesn't boot using EFI Stub

#180 2013-04-08 16:59:44

Re: Kernel 3.7 doesn't boot using EFI Stub

#181 2013-04-08 18:02:02

Re: Kernel 3.7 doesn't boot using EFI Stub

#182 2013-04-09 02:10:13

Re: Kernel 3.7 doesn't boot using EFI Stub

#183 2013-04-09 16:05:09

Re: Kernel 3.7 doesn't boot using EFI Stub

#184 2013-04-09 19:35:21

Re: Kernel 3.7 doesn't boot using EFI Stub

#185 2013-04-09 20:20:27

Re: Kernel 3.7 doesn't boot using EFI Stub

#186 2013-04-09 21:12:29

Re: Kernel 3.7 doesn't boot using EFI Stub

#187 2013-04-09 21:59:01

Re: Kernel 3.7 doesn't boot using EFI Stub

#188 2013-04-09 22:38:37

Re: Kernel 3.7 doesn't boot using EFI Stub

#189 2013-04-09 23:04:40

Re: Kernel 3.7 doesn't boot using EFI Stub

#190 2013-04-10 07:45:59

Re: Kernel 3.7 doesn't boot using EFI Stub

#191 2013-04-10 14:24:55

Re: Kernel 3.7 doesn't boot using EFI Stub

#192 2013-04-10 20:02:41

Re: Kernel 3.7 doesn't boot using EFI Stub

#193 2013-04-11 16:53:38

Re: Kernel 3.7 doesn't boot using EFI Stub

#194 2013-04-11 17:49:46

Re: Kernel 3.7 doesn't boot using EFI Stub

#195 2013-04-11 20:40:41

Re: Kernel 3.7 doesn't boot using EFI Stub

#196 2013-04-11 20:44:43

Re: Kernel 3.7 doesn't boot using EFI Stub

#197 2013-04-11 20:46:31

Re: Kernel 3.7 doesn't boot using EFI Stub

#198 2013-04-11 20:48:29

Re: Kernel 3.7 doesn't boot using EFI Stub

#199 2013-04-11 20:49:23

Re: Kernel 3.7 doesn't boot using EFI Stub

#200 2013-04-15 15:13:46

Re: Kernel 3.7 doesn't boot using EFI Stub

Board footer