Re: ELF bugfixes

[Robert Millan]

On Fri, Mar 13, 2009 at 09:41:42PM +0100, phcoder wrote:
> Robert Millan wrote:
>> On Wed, Mar 11, 2009 at 10:21:41PM +0100, phcoder wrote:
>>> Robert Millan wrote:
>>>> On Mon, Mar 02, 2009 at 01:35:06AM +0100, phcoder wrote:
>>>>> + * include/grub/elf.h: added missing attributes
>>>> This should be a bit more descriptive.
>>>>
>>>>>    for (i = 0; i < ehdr->e_phnum; i++)
>>>>>      if (phdr(i)->p_type == PT_LOAD && phdr(i)->p_filesz != 0)
>>>>>        {
>>>>> - if (phdr(i)->p_paddr < phdr(lowest_segment)->p_paddr)
>>>>> + if (lowest_segment == -1 +          || phdr(i)->p_paddr <  
>>>>> phdr(lowest_segment)->p_paddr)
>>>>>     lowest_segment = i;
>>>>> - if (phdr(i)->p_paddr > phdr(highest_segment)->p_paddr)
>>>>> + if (highest_segment == -1
>>>>> +     || phdr(i)->p_paddr > phdr(highest_segment)->p_paddr)
>>>>>     highest_segment = i;
>>>>>        }
>>>> Why?
>>> Because if first segment doesn't have the PT_LOAD attribute set then 
>>> it  should be considered in this comparison
>>
>> But you didn't remove the PT_LOAD check.  And in the routine below that
>> does the actual segment load, we still check for PT_LOAD.  Those should be
>> consistent, right?
>>
>
> No I expressed myself badly. Original code assumed that first segment  
> has PT_LOAD always set (lowest_segment is 0 initally). I removed this  
> assumption

Why do we care about non-PT_LOAD segments?

>>>>> -  grub_multiboot_payload_entry_offset = ehdr->e_entry - 
>>>>> phdr(lowest_segment)->p_vaddr;
>>>>> +  grub_multiboot_payload_entry_offset = ehdr->e_entry - 
>>>>> phdr(lowest_segment)->p_paddr;
>>>> Are you sure about this?  IIRC e_entry is in the virtual address space.  I
>>>> think we had some trouble with this (with NetBSD?), which lead to the 
>>>> current
>>>> use of p_vaddr in this line.
>>>>
>>> Actually now thinking I see that the problem is more deep. The 
>>> section  which is loaded at the lowest address isn't necessarily the 
>>> section  which contains entry point. I'll fix this part cleanly and 
>>> will resubmit  the patch
>>
>> No, but AFAICT the entry point is defined relative to that address, 
>> regardless
>> of which segment contains it.
>>
> Actually our segment table is also our table for transforming between  
> virtual and physical address. I don't see why entry point would be  
> defined against virtual address of lowest physical segement

I think entry point is supposed to be defined in virtual address space.  As
to why do we check for physical addresses earlier, I'm not entirely sure.  I
think the idea was that we store the entry point as an offset, so that it
can be applied to physical addresses, despite the fact that we obtained it
by comparing e_entry with a virtual address.

ISTR this being an issue for NetBSD.  We should be certain what we do before
changing it.  In particular, the following commit seem relevant:

2008-02-05  Bean  <address@hidden>

        * loader/i386/pc/multiboot.c (grub_multiboot_load_elf32): Get physical
        address of entry.

I'd also recommend testing your changes with NetBSD's kernel.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."