Re: kexec, x86/purgatory: Cleanup the unholy mess
From: Vivek Goyal
Date: Fri Mar 10 2017 - 09:24:31 EST
On Fri, Mar 10, 2017 at 02:57:38PM +0100, Thomas Gleixner wrote:
> On Fri, 10 Mar 2017, Mike Galbraith wrote:
> > On Fri, 2017-03-10 at 13:17 +0100, Thomas Gleixner wrote:
> > > The purgatory code defines global variables which are referenced via a
> > > symbol lookup in the kexec code (core and arch).
> > >
> > > A recent commit addressing sparse warning made these static and thereby
> > > broke kexec file.
> > >
> > > Why did this happen? Simply because the whole machinery is undocumented and
> > > lacks any form of forward declarations. The variable names are unspecific
> > > and lack a prefix, so adding forward declarations creates shadow variables
> > > in the core code. Aside of that the code relies on magic constants and
> > > duplicate struct definitions with no way to ensure that these things stay
> > > in sync.
> > >
> > > Unbreak kexec and cleanup the mess by:
> > >
> > > - Adding proper forward declarations and document the usage
> > > - Use the proper common defines instead of magic constants
> > > - Add a purgatory_ prefix to have a proper name space
> > > - Use ARRAY_SIZE() instead of a homebrewn reimplementation
> > >
> > > Fixes: 72042a8c7b01 ("x86/purgatory: Make functions and variables static")
> >
> > Well, almost fixes.
> >
> > [ 15.118820] kexec: symbol 'purgatory_sha_regions' in common section
> > [ 15.119187] kexec-bzImage64: Loading purgatory failed
>
> Bah. /me goes to investigate.
I think we probably will have to initialize these global variables in
purgatory itself and that puts them in .data section and relocation
works.
That's how the code was intially. I initialized value of
purgatory_sha256_digest in the code and then did "readelf -a purgatory.o"
and symbol section index changed from COM to 3.
13: 0000000000000000 32 OBJECT GLOBAL DEFAULT 3
purgatory_sha256_digest
[ 3] .data PROGBITS 0000000000000000 00000120
0000000000000020 0000000000000000 WA 0 0 32
Thanks
Vivek