Re: [PATCH v6 2/5] s390/vmcore: Use ELF header in new memory feature

[Vivek Goyal <vgoyal@redhat.com>]

On Wed, Jul 03, 2013 at 04:39:44PM +0200, Michael Holzheu wrote:
> On Wed, 3 Jul 2013 10:15:29 -0400
> Vivek Goyal <vgoyal@redhat.com> wrote:
> 
> > On Wed, Jul 03, 2013 at 09:59:13AM +0200, Michael Holzheu wrote:
> > > On Tue, 2 Jul 2013 12:23:23 -0400
> > > Vivek Goyal <vgoyal@redhat.com> wrote:
> > > 
> > > > On Mon, Jul 01, 2013 at 09:32:36PM +0200, Michael Holzheu wrote:
> > > > 
> > > > [..]
> > > > > +ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
> > > > > +{
> > > > > +	void *src = (void *)(unsigned long)*ppos;
> > > > > +
> > > > > +	src = elfcorehdr_newmem ? src : src - OLDMEM_BASE;
> > > > 
> > > > Seriously, we need to get rid of all this OLDMEM_BASE logic in s390
> > > > specific code. For regular kdump, it is no different than x86. Only
> > > > special handling required for zfcpdump for HSA region.
> > > > 
> > > > Why do we need above. Is it to cover the case where elfcorehdr have
> > > > been prepared by user space? Are elf headers initially stored in
> > > > reserved region and then swapped. Why do we need to swap these or
> > > > why kexec-tools could not take care of swapping it.
> > > 
> > > I know it is confusing. The "src - OLDMEM_BASE" term is currently
> > > needed because of the swap issue that we have discussed already. We
> > > load the ELF header into reserved memory
> > > [OLDMEM_BASE, OLDMEM_BASE + OLDMEM_SIZE] that is swapped with
> > > [0, OLDMEM_SIZE]. So the ELF header address has to be adjusted.
> > 
> > Can't kexec-tools could easily do this swapping and modify elfcorehdr=
> > command line accordingly so that second kernel does not have to do
> > swapping for ELF headers.
> > 
> > And for PT_LOAD segment swapping, we could use ELF header swapping trick
> > (again in kexec-tools).
> > 
> > After above two changes I think all the OLD_MEMBASE magic will go away
> > from s390 code and only HSA region special handling will remain.
> > 
> > This brings it inline with x86 code and it becomes easier to understand
> > the s390 code. Otherwise there so may special corner cases that it is
> > easy to get lost.
> 
> Right, I agree that it is possible to do the swap in the kexec tool.
> Then we would load in the kexec tool the ELF header to address
> "OLDMEM_BASE + addr" (or "crashkernel base + addr") and would specify
> the kernel parameter as "elfcorehdr=addr".

Exactly. purgatory does swap and kexec-tools prepares the purgatory and 
it knows that swap will take place so it can pass elfcorehdr=addr,
instead of elfcorehdr=OLDMEM_BASE.

Hardcoding this knowledge in second kernel is not a very good idea.

Thanks
Vivek