Re: Re: [PATCH] Xen Guest Kexec

[Gerd Hoffmann <kraxel@suse.de>]

Mark Williamson wrote:
>> As close as possible to normal i386 kexec ...
>>
>> Dis- and reconnecting should be ok by now I guess.  I expect the paging
>> setup being the most tricky part:  First because the pseudophysical
>> memory (probably not a major issue though).  Second because unlike i386
>> kexec we'll have to run with paging enabled all the time ...
> 
> I doubt having paging enabled would be too painful.  i386 kexec disables 
> paging right at the end of the process so that the new kernel will have a 
> sensible start-of-day.  We'd just need start-of-day to contain bootstrap 
> pagetables, same as for normal Linux.  Ideally you'd need to find a slot in 
> the bootstrap tables for the trampoline code to live, if you take that 
> approach.

x86-64 has paging enabled while trampoline is running too, so I can get
some ideas there ;)

> You've got a load of other things to worry about in this approach, like 
> un-type-pinning all pages you own, etc.

Yep, that is a problem.  What pages are pinned (other than pgd)?  I've
seen plenty of pages with PG_pinned set, but can't figure easily what
pages that are ...

Also switching page tables seems to be not so easy.  Is it possible to
switch atomically to a new, completely independant page table tree?
i.e. old tree is valid (of cource), new tree is too, but the pages of
the old tree are not mapped read-only in the new tree (and visa versa).

> The generic kexec code doesn't understand phys vs machine memory, IIRC, so you 
> may need to worry about it mis-allocating your trampoline page (this is an 
> issue because you need to identity map the trampoline page later on in the 
> process).

Not a big issue if paging is enabled anyway, we can use a identity map
then (virtual == physical, not virtual == machine), so kexec doesn't
even notice outside the arch-specific code.

>> Right now linux kexec depends on the new kernel having a different
>> physical (and virtual) start address, so taking the very same approach
>> likely doesn't work.
> 
> I'm not convinced: the reboot kernel doesn't need to be any different from the 
> standard kernel *unless* you're running kdump (when the kernel will need to 
> live in a different place so that it doesn't stomp on the main kernel - not a 
> limitation of kexec).  Or am I misunderstanding what you meant?

I think it's needed in both cases, at least I had problems making normal
kexec work (without xen) when both kernels had the same physical
address.  Might have been a simple bug though.

cheers,

  Gerd

-- 
Gerd 'just married' Hoffmann <kraxel@suse.de>
I'm the hacker formerly known as Gerd Knorr.
http://www.suse.de/~kraxel/just-married.jpeg