Re: [Qemu-devel] [PATCH v3] Add optionrom compatible with fw_cfg DMA version

["Marc Marí" <markmb@redhat.com>]

On Wed, 27 Jan 2016 12:17:27 -0500
"Kevin O'Connor" <kevin@koconnor.net> wrote:

> On Wed, Jan 27, 2016 at 05:57:28PM +0100, Marc Marí wrote:
> > On Wed, 27 Jan 2016 16:43:29 +0000
> > Stefan Hajnoczi <stefanha@redhat.com> wrote:
> >   
> > > On Tue, Jan 26, 2016 at 12:26:12PM +0100, Gerd Hoffmann wrote:  
> > > > On Di, 2016-01-26 at 12:20 +0100, Marc Marí wrote:    
> > > > > On Tue, 26 Jan 2016 11:11:54 +0000
> > > > > Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > > > >     
> > > > > > On Mon, Jan 25, 2016 at 02:17:48PM +0100, Marc Marí
> > > > > > wrote:    
> > > > > > > +linuxboot_dma.img: linuxboot_dma.o
> > > > > > > +	$(call quiet-command,$(LD) $(LDFLAGS_NOPIE) -m
> > > > > > > elf_i386 -Ttext 0 -e _start -s -o $@ $<,"  Building
> > > > > > > $(TARGET_DIR)$@") + %.img: %.o
> > > > > > >  	$(call quiet-command,$(LD) $(LDFLAGS_NOPIE)
> > > > > > > -Ttext 0 -e _start -s -o $@ $<,"  Building
> > > > > > > $(TARGET_DIR)$@")      
> > > > > > 
> > > > > > Why is -m elf_i386 necessary for linuxboot_dma.img but not
> > > > > > for the other *.img files?    
> > > > > 
> > > > > I cannot give a precise explanation. But if I don't force an
> > > > > output type, I get this error:
> > > > > 
> > > > > Building optionrom/linuxboot_dma.img
> > > > > ld: i386 architecture of input file `linuxboot_dma.o' is
> > > > > incompatible with i386:x86-64 output    
> > > > 
> > > > Any chance the linker needs -m32 too?    
> > > 
> > > I wonder why this isn't a problem for the existing firmware
> > > code.  Are we really building x86_64 ELF files for our firmware?  
> > 
> > Yes they are x86_64:
> > 
> > pc-bios/optionrom/linuxboot.img: ELF 64-bit LSB executable, x86-64,
> > version 1 (SYSV), statically linked, stripped
> > 
> > But as they are written directly in assembly, it does not give any
> > problem. Whereas when mixing C and ASM, it does give problems.  
> 
> Would it hurt anything to add "-m elf_i386" to the generic assembler
> rule and then use that for both targets?  (Just to keep the makefile a
> little simpler.)

It doesn't seem to hurt, and it is simpler.

Marc