Xen devel - May 2006 - Questions about VIRT_BASE and ELF_PADDR_OFFSET in __xen_guest

Keir,

I was looking at your changeset:
http://xenbits.xensource.com/xen-unstable.hg?cs=dbe396fd6fd8

I am finding that by making VIRT_BASE=0x0 is causing the builder to seg
fault. The same thing happens when VIRT_BASE and ELF_PADDR_OFFSET is set
to 0. I don''t understand why dsi->elf_paddr_offset is being set to
VIRT_BASE and then overridden with ELF_PADDR_OFFSET if it is set. Does
ELF_PADDR_OFFSET have any real use? Does it correlate with anything in
the ELF program header? I am a little confused here :-), so could you
please explain.

Thanks,
Aravindh Puthiyaparambil
Xen Development Team 
Unisys, Tredyffrin PA

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 18 May 2006, at 19:15, Puthiyaparambil, Aravindh wrote:
> I was looking at your changeset:
> http://xenbits.xensource.com/xen-unstable.hg?cs=dbe396fd6fd8
>
> I am finding that by making VIRT_BASE=0x0 is causing the builder to seg
> fault. The same thing happens when VIRT_BASE and ELF_PADDR_OFFSET is 
> set
> to 0. I don''t understand why dsi->elf_paddr_offset is being set
to
> VIRT_BASE and then overridden with ELF_PADDR_OFFSET if it is set. Does
> ELF_PADDR_OFFSET have any real use? Does it correlate with anything in
> the ELF program header? I am a little confused here :-), so could you
> please explain.
VIRT_BASE is the linear address you want your memory allocation mapped 
at.

ELF_PADDR_OFFSET is a value subtracted from addresses in the Elf 
headers to produce a pseudo-physical address. We usually set it to 
VIRT_BASE since we write virtual (linear) addresses into those Elf 
header fields. The default of setting it to VIRT_BASE is simply to 
maintain backward compatibility with old guest images which do not 
explicitly define it.

There is no reason really why VIRT_BASE=0 should not work. If it 
crashes there is presumably some underlying bug which disallowing 
VIRT_BASE=0 does not fix.

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> There is no reason really why VIRT_BASE=0 should not work. If it
> crashes there is presumably some underlying bug which disallowing
> VIRT_BASE=0 does not fix.
You are right. The experiment I was trying with Mini-OS was flawed. I
had forgotten to fix up the minios_x86_64.lds file to reflect 0x0 (bang
head on table). Once I did that things worked fine. 

Do we care about the situation where there is a mismatch in ELF header
and __xen_guest section? When this happens the var "pa" is calculated
incorrectly causing "parray" to go out of bounds.

pa = (phdr->p_paddr + done) - dsi->elf_paddr_offset; 
va = xc_map_foreign_range(xch, dom, PAGE_SIZE, PROT_WRITE, 
                          parray[pa>>PAGE_SHIFT]);

[line 227-228 xc_load_elf.c loadelfimage()]

(In my flawed test, p_addr was 0xffffffff80000000 and elf_paddr_offset
was 0 due to obvious reasons)

I know this is rarely possible unless someone does something stupid like
I did :-) which is why I am wondering if we should test for this case. 

[ASIDE]
Due to this I think I should fix x86_xx.S in Mini-OS so that it picks up
&_text from minios_x86_xx.lds.

Cheers,
Aravindh

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> > There is no reason really why VIRT_BASE=0 should not work. If it
> > crashes there is presumably some underlying bug which disallowing
> > VIRT_BASE=0 does not fix.
> 
> You are right. The experiment I was trying with Mini-OS was flawed. I
> had forgotten to fix up the minios_x86_64.lds file to reflect 0x0 (bang
> head on table). Once I did that things worked fine. 
> 
> Do we care about the situation where there is a mismatch in ELF header
> and __xen_guest section? When this happens the var "pa" is
calculated
> incorrectly causing "parray" to go out of bounds.
Well, that''s the bug. We should perform bounds checks on indexes into
parray. I would very much like to see a patch to fix this!

 -- Keir
> pa = (phdr->p_paddr + done) - dsi->elf_paddr_offset; 
> va = xc_map_foreign_range(xch, dom, PAGE_SIZE, PROT_WRITE, 
>                           parray[pa>>PAGE_SHIFT]);
> 
> [line 227-228 xc_load_elf.c loadelfimage()]
> 
> (In my flawed test, p_addr was 0xffffffff80000000 and elf_paddr_offset
> was 0 due to obvious reasons)
> 
> I know this is rarely possible unless someone does something stupid like
> I did :-) which is why I am wondering if we should test for this case. 
> 
> [ASIDE]
> Due to this I think I should fix x86_xx.S in Mini-OS so that it picks up
> &_text from minios_x86_xx.lds.
> 
> Cheers,
> Aravindh
> 
 -=- MIME -=- 
> There is no reason really why VIRT_BASE=3D0 should not work. If it
> crashes there is presumably some underlying bug which disallowing
> VIRT_BASE=3D0 does not fix.
You are right. The experiment I was trying with Mini-OS was flawed. I
had forgotten to fix up the minios_x86_64.lds file to reflect 0x0 (bang
head on table). Once I did that things worked fine.=20

Do we care about the situation where there is a mismatch in ELF header
and __xen_guest section? When this happens the var "pa" is calculated
incorrectly causing "parray" to go out of bounds.

pa =3D (phdr->p_paddr + done) - dsi->elf_paddr_offset;=20
va =3D xc_map_foreign_range(xch, dom, PAGE_SIZE, PROT_WRITE,=20
                          parray[pa>>PAGE_SHIFT]);

[line 227-228 xc_load_elf.c loadelfimage()]

(In my flawed test, p_addr was 0xffffffff80000000 and elf_paddr_offset
was 0 due to obvious reasons)

I know this is rarely possible unless someone does something stupid like
I did :-) which is why I am wondering if we should test for this case.=20

[ASIDE]
Due to this I think I should fix x86_xx.S in Mini-OS so that it picks up
&_text from minios_x86_xx.lds.

Cheers,
Aravindh



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel