[Uboot-stm32] [PATCH 0/7] arm: cache: cp15: don't map reserved region with no-map property

[Jerome Forissier]

On 10/29/20 5:06 PM, Etienne Carriere wrote:
> On Thu, 29 Oct 2020 at 12:26, Ard Biesheuvel <ardb at kernel.org> wrote:
>> The point I made before was that secure and non-secure are two
>> disjoint address spaces. The fact that TZ firewalls exist where you
>> can move things from one side to the other does not imply that things
>> works like this in the general case.
>>
>> E.g., you could have
>>
>> secure DRAM at S 0x0
>> non-secure DRAM at NS 0x0
>>
>> where the ranges are backed by *different* memory. Since the DT
>> description does not include the S/NS distinction, only the address
>> range, the only thing we can assume when looking at memory@ and
>> /reserved-memory is that everything it describes is NS.
> 
> From Arm Trustzone stand point, both secure and non-secure worlds
> share the very same physical address space. I your example, physical
> address 0x0 would refer to the same DRAM cell. Whether this cell is secure
> or non-secure is a configuration set in the DRAM firmwall.

No, like Ard said it is a possibility but it doesn't have to be the
case. See the Armv8-A ARM (DDI 0487F.c) section D5.1.3 VMSA address
types and address spaces, "Physical address (PA)".

If we need to differentiate between non-secure and secure PA I suppose
we could use the status and secure-status properties in the memory
nodes, consistent with the usual usage described in [1].

As Etienne says, it seems that a majority of systems actually have a
single PA space with access control added on top, and by default the
secure state can access non-secure memory. That goes well with memory
nodes without a status nor a secure-status property, yet other
configurations can easily be supported.

[1]
https://www.kernel.org/doc/Documentation/devicetree/bindings/arm/secure.txt

-- 
Jerome