Re: [PATCH v2 03/11] dump: Split write of section headers and data and add a prepare step

[Janosch Frank]

On 7/13/22 17:31, Marc-André Lureau wrote:
> Hi
>
> On Wed, Jul 13, 2022 at 5:07 PM Janosch Frank <frankja@linux.ibm.com> wrote:
> > By splitting the writing of the section headers and (future) section
> > data we prepare for the addition of a string table section and
> > architecture sections.
> >
> > Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
> > ---

[...]

> > @@ -557,12 +600,22 @@ static void dump_begin(DumpState *s, Error **errp)
> >       /* Write elf header to buffer */
> >       prepare_elf_header(s);
> >
> > +    prepare_elf_sections(s, errp);
> > +    if (*errp) {
> > +        return;
> > +    }
> > +
> >       /* Start to write stuff into files*/
> >       write_elf_header(s, errp);
> >       if (*errp) {
> >           return;
> >       }
> >
> > +    write_elf_section_headers(s, errp);
>
> Why do you reorder the sections? Could you explain in the commit
> message why? Is this is format compliant? and update the comment
> above? thanks


Having the section data at the end of the file is unfortunately a s390 
PV requirement since we can only grab the encrypted page tweaks and 
counts *after* all of the memory has been encrypted.

The sections are the most obvious way to add such data to the file since 
they are basically unused right now and we're able to write a string 
table at the very end after everyone registered their strings.

All of this is ELF compliant AFAIK, that's why elf specifies offsets of 
the headers and the data. From what I see only the main elf header needs 
to start at offset 0.