Re: [PATCH v2] hw/i386: place setup_data at fixed place in memory

[Laszlo Ersek]

On 08/05/22 00:56, Jason A. Donenfeld wrote:
> Hey Laszlo,
> 
> On Thu, Aug 04, 2022 at 03:56:54PM +0200, Laszlo Ersek wrote:
>> - do we want setup_data chaining work generally?
>>
>> - or do we want only the random seed injection to stop crashing OVMF guests?
> 
> Preferably the first - generally. Which brings us to your point:
>  
>>> Given we only need 48 bytes or so, isn't there a more subtle place we
>>> could just throw this in ram that doesn't need such complex
>>> coordination?
>>
>> These tricks add up and go wrong after a while. The pedantic
>> reservations in the firmware have proved necessary.
>>
>> IIUC, with v2, the setup_data_base address would (most frequently) be 96
>> KB. edk2 does have uses for very low memory. If OVMF's PlatformPei does
>> not reserve away the area, UefiCpuPkg or other drivers might allocate an
>> overlapping chunk, even if only temporarily. That might not break the
>> firmware, but it could overwrite the random seed. 
> 
> Yea, so we don't want an address that something else might overwrite. So
> my question is: isn't there some 48 bytes or so available in some low
> address (or maybe a high one?) that is traditionally reserved for some
> hardware function, and so software doesn't use it, but it turns out QEMU
> doesn't use it for anything either, so we can get away placing it at
> that address? It seems like there *ought* to be something like that. I
> just don't (yet) know what it is...

I don't know of any such "hidden pocket", unfortunately. On the other
hand, low-level edk2 drivers (usually dealing with x86 intricacies, such
as MTRRs, CPU bringup, ...) have repeatedly surprised me with their
handling of low memory.

Laszlo