[syslinux] Problem with pxelinux 5.0 and memtest

[H. Peter Anvin]

On 04/03/2013 10:05 AM, H. Peter Anvin wrote:
> On 04/03/2013 09:54 AM, Matt Fleming wrote:
>> On 03/04/13 15:11, H. Peter Anvin wrote:
>>> I don't think we can... in an error condition we need it.
>>
>> In that case, how about building logic into the shuffler code so that it
>> understands it can reclaim the PXE stack region, even those it's
>> currently marked as reserved?
>>
> 
> Well, we'd mark that in the map we feed the shuffler.  That part is easy
> enough.
> 
> The problem is that if the unload is successful, then we're ok, but if
> the unload is not successful -- and that does happen on real systems --
> we are stuck.
> 
> Overall it feels like we don't have any good options here.  We have
> similar issues with stubs that need to be in high DOS memory too (used
> by the chainloader), so it isn't entirely unique to this case.  It's a
> fundamental problem with PXE.
> 
> What we *might* be able to do is to see if we have a memory map which is
> impossible and in that case, switch to "optimistic mode".
> 

Come to think of it... the sanest possibility is probably to introduce
into the shuffler a new memory type "destination only".  That is, this
memory can ONLY be used to move something into its final resting place.

We should still avoid using it for any dynamic allocations, but for
stuff that is stuck at a fixed address like true legacy kernels it is
better than nothing.

Should also see if we can get the memtest people to slap a modern
bzImage header on...

	-hpa