Re: [PATCH v2] generic-loader: check that binary file target location exists

[Damien Hedde <damien.hedde@greensocs.com>]

On 11/2/21 15:15, Peter Maydell wrote:
> On Tue, 2 Nov 2021 at 14:04, Damien Hedde <damien.hedde@greensocs.com> wrote:
>>
>>
>>
>> On 11/1/21 11:53, Peter Maydell wrote:
>>> Won't this break the case of loading a file that spans two
>>> consecutive-but-different memory regions ?
>>
>> yes. I did not thought about that.
>>
>>>                                             I think if we want
>>> to catch "we tried to load something to an address not backed
>>> by something" we should do that by making load_image_targetphys_as()
>>> correctly handle errors from the memory accesses it makes and
>>> propagate an error-return to the caller.
>>>
>>
>> The problem is that load_image_targetphys_as() just uses rom_load_...().
>> And these macros/functions only put some "rom to load" objects in a list
>> without doing any real accesses.
> 
> Oh yes, I'd forgotten that this was all using the rom blob
> mechanism underneath.
> 
> I think in that case that whatever we do we should aim
> to be consistent between the different load-image functions;
> we shouldn't add a check to load_image_targetphys_as() which
> we don't do in eg load_elf_as(). That probably means we should
> only be looking to warn or error at the point where we actually
> load the ROM blobs into memory.
> 
> -- PMM
> 

I'll look into that.

Thanks,
Damien