Re: [Qemu-devel] [PATCH 5/6] [wip] tseg, part1, not (yet) tested

[Laszlo Ersek <lersek@redhat.com>]

On 04/22/15 10:09, Gerd Hoffmann wrote:
>   Hi,
> 
>> tseg is just normal ram (yes, located at the end of memory), but (once
>> tseg is enabled) only cpus in smm mode are allowed to access it.
>> Likewise busmaster dma access is rejected, so non-smm code can't use the
>> sata controller to access this indirectly.
> 
> Update:  Seems tseg can be anywhere, there is a "tseg memory base"
> register @ 0xac in pci config space.
> 
> Placing it at the end of memory is just what the bios is supposed to do
> by default.  And it makes sense to place it there.
> 
> <quote>
> This register contains the base address of TSEG DRAM memory. BIOS
> determines the base of TSEG memory by subtracting the TSEG size (PCI
> Device 0, offset 9Eh, bits 2:1) from graphics GTT stolen base (PCI
> Device 0, offset A8h, bits 31:20).
> 
> Once D_LCK has been set, these bits becomes read only.
> </quote>
> 
> "GTT stolen base" equals "top of below-4g memory" for us because we
> emulate the chipset variant without graphics in qemu.

Thanks, that sounds good. So, as far as I understand, no changes to what
we've discussed thus far.

But, I have another question -- am I allowed to use "top of below-4g
memory" directly, as discussed earlier, or should I use the above PCI
registers? The tseg size will actually come from me (because I'll select
it), but the top I could take from "top of below-4g memory" (preferably,
see earlier), or reading the 0xA8 register.

Unless, of course 0xA8 won't be implemented at all, *because* "we
emulate the chipset variant without graphics in qemu." In other words,
if the 0xA8 register is dependent on the integrated graphics, then I
don't have a question. :)

Thanks!
Laszlo