Re: [Xen-devel] [PATCH v2] xen/arm: build as zImage

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Tue, 2012-11-27 at 13:58 +0000, Tim Deegan wrote:
> At 12:11 +0000 on 27 Nov (1354018273), Stefano Stabellini wrote:
> > On Fri, 23 Nov 2012, Ian Campbell wrote:
> > > On Fri, 2012-11-23 at 18:04 +0000, Stefano Stabellini wrote:
> > > > The zImage format is extremely simple: it only consists of a magic
> > > > number and 2 addresses in a specific position (see
> > > > http://www.simtec.co.uk/products/SWLINUX/files/booting_article.html#d0e309).
> > > > 
> > > > Some bootloaders expect a zImage; considering that it doesn't cost us
> > > > much to build Xen compatible with the format, make it so.
> > > > 
> > > > Changes in v2:
> > > > - code style;
> > > > - pass 0 a start address.
> > > > 
> > > > Signed-off-by: Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>
> > > > 
> > > > diff --git a/xen/arch/arm/head.S b/xen/arch/arm/head.S
> > > > index 25c4cfe..347ddcf 100644
> > > > --- a/xen/arch/arm/head.S
> > > > +++ b/xen/arch/arm/head.S
> > > > @@ -22,6 +22,8 @@
> > > >  #include <asm/processor-ca15.h>
> > > >  #include <asm/asm_defns.h>
> > > >  
> > > > +#define ZIMAGE_MAGIC_NUMBER 0x016f2818
> > > > +
> > > >  #define PT_PT  0xe7f /* nG=1, AF=1, SH=10, AP=01, NS=1, ATTR=111, T=1, 
> > > > P=1 */
> > > >  #define PT_MEM 0xe7d /* nG=1, AF=1, SH=10, AP=01, NS=1, ATTR=111, T=0, 
> > > > P=1 */
> > > >  #define PT_DEV 0xe71 /* nG=1, AF=1, SH=10, AP=01, NS=1, ATTR=100, T=0, 
> > > > P=1 */
> > > > @@ -52,6 +54,18 @@
> > > >          * or the initial pagetable code below will need adjustment. */
> > > >         .global start
> > > >  start:
> > > > +
> > > > +       /* zImage magic header, see:
> > > > +        * 
> > > > http://www.simtec.co.uk/products/SWLINUX/files/booting_article.html#d0e309
> > > > +        */
> > > > +       .skip 0x20
> > > 
> > > This was better as the explicit repeated nops, as it stands it's not at
> > > all clear that you can start executing at start. It might even rely on
> > > the specific assemblers behaviour?
> > >
> > > Also this wasn't what Tim suggested, which was branch then skip, not
> > > skip then branch. I think the nops are better for consistency with Linux
> > > though.
> > 
> > You are right. Thinking twice about it, I would rather keep the code as
> > it was originally, for consistency with Linux.
> 
> Does consistency with linux buy us anything (beyond untidy code)?
> I very much prefer the other way.

I suspect the reason for this run of nops is some buggy bootloader which
jumped to byte 4*N instead of the start, for whatever reason. They may
well be historical but I don't see any reason to risk it.

Given that we are mimicing Linux's protocol I think remaining blindly
compatible is worth it.

>   And if we do need to fill with nops,
> why 7 repeated and one standalone?

In the original Linux it is actually:
                mov     r0, r0
                .endr
   ARM(         mov     r0, r0          )
   ARM(         b       1f              )
 THUMB(         adr     r12, BSYM(1f)   )
 THUMB(         bx      r12             )

i.e. once you strip out the thumb stuff (because we don't do thumb mode
hypervisor) you end up with what Stefano had. So indeed we could just do
8 in our case.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel