On Fri, Aug 10, 2018 at 02:00:44AM -0300, Philippe Mathieu-Daudé wrote:
> On 08/03/2018 11:47 AM, Stefan Hajnoczi wrote:
> > + parser->rom_start_address = parser->next_address_to_write;
> > + parser->current_rom_index = 0;
> > + break;
> > +
> > + case START_SEG_ADDR_RECORD:
> > + if (line->byte_count != 4 && line->address != 0) {
> > + return -1;
> > + }
> > +
> > + /* x86 16-bit CS:IP segmented addressing */
> > + *(parser->start_addr) = (((line->data[0] << 8) | line->data[1]) <<
> > 4) |
> > + (line->data[2] << 8) | line->data[3];
>
> Can you add a qtest for this case?
> For the HEX loader I understand the specs as this is the same parsing as
> the START_LINEAR_ADDR_RECORD case; so I disagree with data[0] and
> data[1] shifts.x86 real-mode CS:IP addressing means (CS << 4) + IP. It produces 24-bit addresses on 80286 and later. This is not the same as START_LINEAR_ADDR_RECORD. GNU bfd implements it as follows: abfd->start_address += (HEX4 (buf) << 4) + HEX4 (buf + 4); https://sourceware.org/git/?p=binutils.git;a=blob;f=bfd/ihex.c;h=09f756a1c2c11e57c5da15a6ff96491275575b86;hb=HEAD#l415 Thanks for commenting on this, I had written (CS << 4) | IP instead of (CS << 4) + IP. This will be fixed in the next revision. > This is different for the consumer (i386 expects 2 16-bit registers but > HexParser->start_addr is a hwaddr, used as a single (at least) 32-bit > register. I think you're saying that on x86 the guest CS:IP registers should be set. This loader is never called from hw/i386/ so it doesn't matter at this time. GNU bfd uses START_SEG_ADDR_RECORD records on non-x86 architectures where there are no segment registers and that's where we've encountered them.
signature.asc
Description: PGP signature
