Hi Ryan/Leif,
I was initially happy enough with the solution #4 and wanted to push your change to SVN. And yes, I can only be agree on the fact the EFI stub is the way we want to go to boot Linux kernel on a UEFI system.
But I am not sure how the EFI stub will solve your problem.
Even with the EFI stub, you will still need to change you Continuous Integration Infrastructure to start the Linux kernel with different parameters for Android/Ubuntu/OpenEmbedded.
So, it looks the only solution is #1.
I do not mind to push your patch, but that will not help you in the future neither.
Thanks,
Olivier
From: Ryan Harkin [mailto:ryan.harkin@linaro.org] Sent: 19 November 2013 11:23 To: Leif Lindholm Cc: boot-architecture@lists.linaro.org; edk2-devel@lists.sourceforge.net; Olivier Martin; Steven Kinney Subject: Re: ARM/BDS: skip initrd if not found
On 19 November 2013 10:34, Leif Lindholm leif.lindholm@linaro.org wrote:
Hi Ryan,
On Tue, Nov 19, 2013 at 08:30:25AM +0000, Ryan Harkin wrote:
Hi Olivier/Steve/Leif/whoever is interested,
I have a problem I'm trying to solve and I don't think there is a proper solution using the current ARM BDS.
Basically, some of Linaro's releases are failing to boot "out of the box" due to incorrect default BDS config. The default assumes that
the
image has an initrd.
Android (the main focus of our releases) and Ubuntu images use an initrd. OpenEmbedded images do not.
I'd like a single UEFI binary that can work on all three without reconfiguration.
The obvious solutions are:
- there is no default config that always works and the user should
configure
the board themselves each time they want to boot a different image type. Our LAVA automated test environment and people like myself who boot
test
many images daily don't favour this option.
- change OpenEmbedded to use an initrd It's not my image to change and the owners don't want to do that
because
it's also wrong.
Have different UEFI binaries for each image type This isn't ideal because I (and LAVA) would be forever reflashing UEFI.
Make BDS continue if it can't load the initrd This isn't ideal because if there is no initrd, it could be for a bad
reason. By continuing, we aren't giving the user the change to immediately correct the config. However, the likelihood of the initrd being completely missing, whilst
a
valid kernel and FDT is provided seems slim. If it is missing, it's most likely on purpose.
Of the options above, I prefer #4 and have provided a patch below for discussion. I suspect that if it's not going to cause other problems, it
could
be like my other BDS hacks, fixes and improvements and only live in the
Linaro
tree, which would be fine with me too.
Opinions on a way forward and/or this patch?
Medium-term, I would say that the correct thing to do will be to simply use the UEFI stub loader version of the kernel image. But you may not want to take on the tediousness of having to sync this not-yet-upstream patchset across the various kernel branches things will build from for each new release until this code does go upstream?
I agree, moving to a proper boot solution is the end goal. But a lot of that end-goal is out of my control/domain, so I'm hacking what I have to make it more usable until the cool stuff hits mainline.
I would say that the solution your patch introduces is wrong, but it is less wrong than the current situation - so I have no fundamental objection.
I agree, it's not ideal, but as you say, it's less wrong.
If we wanted to keep the built-in Linux loader, I would say that the correct fix would be to add a "has initrd" property, or a NULL string chack for the path. But we don't, so we shouldn't spend time trying to to improve a way too old stop-gap solution.
The initrd can be configured out in the default config, so UEFI does not attempt to even load it. However, in that case, a single UEFI binary's default config will fail to load an Android or Ubuntu image, because they need an initrd.
Really, I'm working around the fact that we cannot provide a config to BDS; the config either has to be initialised by the UEFI binary or hand edited by the user. If we had that feature, each image (Android, Ubuntu, OE) could provide a config that it knew would work.
/ Leif
p.s. The built-in Linux loader delenda est!
Yes please :-)