Saveoops: Making Kexec purgatory position-independent?

[Eric W. Biederman]

"H. Peter Anvin" <hpa at zytor.com> writes:

> On 02/26/2011 08:20 AM, Ahmed S. Darwish wrote:
>> Hi,
>> 
>> I'm continuing work on 'Saveoops', saving both early and normal Linux
>> oops log to disk upon panic [*], using Kexec and bootloaders this time.
>> 
>> Purgatory, the transitional mini-kernel used by kexec, is a relocatable
>> ELF file. Userspace kexec-tools finds the final load address of such
>> code (by parsing /proc/iomem, etc) and then applies the relocations
>> itself before passing the now-ready executable image to kernel.
>> 
>> Since capturing early oopses is the major goal, doing such relocation
>> in userspace won't fit my purposes. Two options remain:
>>    - relocate purgatory entries in the kernel early boot path
>>    - or compile purgatory as position-independent, thus simplifying
>>      the kernel load logic
>> 
>> The former will add extra logic in a sensitive path (early boot), while
>> the latter will require changes inside the purgatory code itself,
>> especially i386 assembly files.
>> 
>> Any preferable option from our kexec and x86 maintainers?
>> 
>> thanks!
>> 
>> [*] http://news.gmane.org/find-root.php?message_id=<20110125134748.GA10051@laptop>
>>     http://news.gmane.org/find-root.php?message_id=<20110126124954.GC24527@laptop>
>> 
>
> I can't see any sane reason to *not* make kexec purgatory
> position-independent.  It is the obvious thing to do.

This isn't a case of the code not being position independent.  This is
case of where the relocations are applied.

I can see a couple of handling this with different tradeoffs.

1) We teach bootloaders how to load two kernels at once.  This
   completely avoids the purgatory, as it is replaced by code in the
   bootloader that already exists to load the primary kernel and setup
   it's arguments.

2) We add minimal relocation processing to purgatory, allowing us to do
   the setup for the second kernel extremely early and allow it to be
   compiled into the first kernel.

3) We come up with a scheme where we don't share code and the first
   kernel copies the firmware information to place where the second
   kernel can get at it, and uses it's own home grown stub and not
   purgatory.

I think this whole thing can be prototyped easily with a getting /sbin/kexec
to load to a fixed address and then baking that section into the primary
kernel.  I'm not convinced that directly using /sbin/kexec is the right
way forward to handle the general case.  This is something where the
devil is in the details.

Eric