[Qemu-devel] PPC: Which instruction to fetch at Power-On?

[Qemu-devel] PPC: Which instruction to fetch at Power-On?

[Fabien Chouteau]

While working on a patch to remove env->hreset_excp_prefix, I'm looking
at which instruction should be the fetched first at power-on. I'm lost
in all the PPC version and configuration, can anyone (Alex :) can help
me with this mess?

BTW, is there somewhere some kind of table that gives for all PPC cores
which specification(s) they implements?

Thanks,

-- 
Fabien Chouteau

Re: [Qemu-devel] PPC: Which instruction to fetch at Power-On?

[Alexander Graf]

On 28.03.2013, at 12:42, Fabien Chouteau wrote:

> 
> While working on a patch to remove env->hreset_excp_prefix, I'm looking
> at which instruction should be the fetched first at power-on. I'm lost
> in all the PPC version and configuration, can anyone (Alex :) can help
> me with this mess?

Phew - I'd say just keep the targets you don't know exactly as they are :). That way you at least don't break anything that wasn't broken before.

For the targets where you do know better (like 7x0), change its behavior to be correct according to the respective spec.

> BTW, is there somewhere some kind of table that gives for all PPC cores
> which specification(s) they implements?

All I have are a number of core specs, but I really only have a fraction of what QEMU implements. Most of that code heavily predates me. I don't think there's any document that contains information for multiple different cores.


Alex

Re: [Qemu-devel] PPC: Which instruction to fetch at Power-On?

[Fabien Chouteau]

On 03/28/2013 12:46 PM, Alexander Graf wrote:
> 
> On 28.03.2013, at 12:42, Fabien Chouteau wrote:
> 
>>
>> While working on a patch to remove env->hreset_excp_prefix, I'm looking
>> at which instruction should be the fetched first at power-on. I'm lost
>> in all the PPC version and configuration, can anyone (Alex :) can help
>> me with this mess?
> 
> Phew - I'd say just keep the targets you don't know exactly as they are :). That way you at least don't break anything that wasn't broken before.
> 
> For the targets where you do know better (like 7x0), change its behavior to be correct according to the respective spec.
> 

Here's what I found in the UM of almost all the cores supported by QEMU:

Every CPU should start at the hreset address (i.e. excp_prefix + 0x100).

However, some cores don't have a hreset exception (4xx and Book E (e200,
e500)), those cores start at 0xFFFF_FFFC.

-- 
Fabien Chouteau

Re: [Qemu-devel] PPC: Which instruction to fetch at Power-On?

[Alexander Graf]

On 28.03.2013, at 16:10, Fabien Chouteau wrote:

> On 03/28/2013 12:46 PM, Alexander Graf wrote:
>> 
>> On 28.03.2013, at 12:42, Fabien Chouteau wrote:
>> 
>>> 
>>> While working on a patch to remove env->hreset_excp_prefix, I'm looking
>>> at which instruction should be the fetched first at power-on. I'm lost
>>> in all the PPC version and configuration, can anyone (Alex :) can help
>>> me with this mess?
>> 
>> Phew - I'd say just keep the targets you don't know exactly as they are :). That way you at least don't break anything that wasn't broken before.
>> 
>> For the targets where you do know better (like 7x0), change its behavior to be correct according to the respective spec.
>> 
> 
> Here's what I found in the UM of almost all the cores supported by QEMU:
> 
> Every CPU should start at the hreset address (i.e. excp_prefix + 0x100).
> 
> However, some cores don't have a hreset exception (4xx and Book E (e200,
> e500)), those cores start at 0xFFFF_FFFC.

At least the G5 and I think also more recent POWER CPUs have either a small command stream or a full blown service processor that initialize initial CPU state like the PC. So there, the CPU starts at a board defined address.

I don't think we need to worry too much about that for now though.


Alex

Re: [Qemu-devel] PPC: Which instruction to fetch at Power-On?

[Fabien Chouteau]

On 03/28/2013 05:04 PM, Alexander Graf wrote:
> 
> On 28.03.2013, at 16:10, Fabien Chouteau wrote:
> 
>> On 03/28/2013 12:46 PM, Alexander Graf wrote:
>>>
>>> On 28.03.2013, at 12:42, Fabien Chouteau wrote:
>>>
>>>>
>>>> While working on a patch to remove env->hreset_excp_prefix, I'm looking
>>>> at which instruction should be the fetched first at power-on. I'm lost
>>>> in all the PPC version and configuration, can anyone (Alex :) can help
>>>> me with this mess?
>>>
>>> Phew - I'd say just keep the targets you don't know exactly as they are :). That way you at least don't break anything that wasn't broken before.
>>>
>>> For the targets where you do know better (like 7x0), change its behavior to be correct according to the respective spec.
>>>
>>
>> Here's what I found in the UM of almost all the cores supported by QEMU:
>>
>> Every CPU should start at the hreset address (i.e. excp_prefix + 0x100).
>>
>> However, some cores don't have a hreset exception (4xx and Book E (e200,
>> e500)), those cores start at 0xFFFF_FFFC.
> 
> At least the G5 and I think also more recent POWER CPUs have either a small command stream or a full blown service processor that initialize initial CPU state like the PC. So there, the CPU starts at a board defined address.
>

You're right, if the start address doesn't follow core's specification,
it's up to the board.


> I don't think we need to worry too much about that for now though.
>

The problem is that my patch may break support of some platforms, we
will have to fix the boards later on.

-- 
Fabien Chouteau