x86-64/kexec: crashkernel= without @xM suffix

x86-64/kexec: crashkernel= without @xM suffix

[Jan Beulich]

Was this ever tested by anyone? Has it ever worked with any Linux
as secondary kernel? We're getting reports (and analysis confirms
this) that the fact that Xen places the hole at the highest suitable
address below 4G prevents kexec from working.

A possible fix is non-obvious, since there doesn't appear to be a
hard boundary below which the hole must reside (especially for a
32-bit secondary kernel; for a 64-bit one it seems like any place
below 2G would be acceptable).

Simply using Linux' strategy and allocating at the lowest possible
address doesn't seem too nice a solution either, as we try to
conserve on the use of low memory as much as possible almost
everywhere else.

Thanks, Jan

Re: x86-64/kexec: crashkernel= without @xM suffix

[Simon Horman]

On Fri, Apr 30, 2010 at 07:42:06AM +0100, Jan Beulich wrote:
> Was this ever tested by anyone? Has it ever worked with any Linux
> as secondary kernel? We're getting reports (and analysis confirms
> this) that the fact that Xen places the hole at the highest suitable
> address below 4G prevents kexec from working.
> 
> A possible fix is non-obvious, since there doesn't appear to be a
> hard boundary below which the hole must reside (especially for a
> 32-bit secondary kernel; for a 64-bit one it seems like any place
> below 2G would be acceptable).
> 
> Simply using Linux' strategy and allocating at the lowest possible
> address doesn't seem too nice a solution either, as we try to
> conserve on the use of low memory as much as possible almost
> everywhere else.

Hi Jan,

I don't recall specifically testing this combination.
I think that your suggestion of simply locating the hole as
low as possible is a good one though I fear there is
a gotcha in there somewhere.