Samuel Thibault, le ven. 26 mai 2023 13:37:11 +0200, a ecrit:
> Luca, le ven. 26 mai 2023 13:18:08 +0200, a ecrit:
> > Il 26/05/23 12:56, Samuel Thibault ha scritto:
> > > Luca, le ven. 26 mai 2023 12:55:32 +0200, a ecrit:
> > > > read_exec() it seems that there is something wrong with the elf
> > > > parsing, as
> > > > the size parameters look like kernel addresses:
> > > >
> > > > (gdb) p/x file_size
> > > > $1 = 0xffffffff8106f4d3
> > >
> > > Perhaps a spurious signedness bit extension?
> >
> > Could be, the elf header seems ok when calling read_exec:
> >
> > (gdb) frame 1
> > #1 0xffffffff8107a712 in exec_load (read=0xffffffff8106f45c <boot_read>,
> > read_exec=0xffffffff8106f4e5 <read_exec>, handle=0xffffffffbd8faea0,
> > out_info=0xffffffffbd8f0f70)
> > at ../kern/elf-load.c:87
> > 87 result = (*read_exec)(handle,
> > (gdb) p/x *ph
> > $9 = {
> > p_type = 0x1,
> > p_flags = 0x5,
> > p_offset = 0x0,
> > p_vaddr = 0x400000,
> > p_paddr = 0x400000,
> > p_filesz = 0x1cfd58,
> > p_memsz = 0x1cfd58,
> > p_align = 0x1000
> > }
> >
> > but I can't explain why I don't see the same issue with my build. I 've now
> > configured gnumach just with
> >
> > ../configure --enable-silent-rules MIG=/path_to_my/mig --disable-user32
> > --enable-kdb
> >
> > and I'm building on Debian/Linux stable.
>
> Indeed, when I use that configuration, I don't get the issue.
>
> Heh, when I tell my students never to hide anything because you never
> know what might be relevant...
>
> CFLAGS="" MIG=/usr/bin/x86_64-gnu-mig ../configure --disable-user32
> --enable-kdb
>
> does get the issue,
I.e. gnumach without optimizations gets things wrong. We do want to fix
that :)
Samuel
