Modify Direct-Memory-Ptr to False in gs-memory for Tracking Memory Access in LiqBox Example

[alpha-beta-user]

I have a situation where the gs_memory provides a direct-mem-ptr to the CPU, enabling fast memory access.

Now, I need to modify direct-mem-ptr to false to track every memory access. However, when doing this, the CPU halts when attempting to access 0x0.
if (p_dmi) txn.set_dmi_allowed(false);

I am currently using the liqbox example, specifically the simpleWriteTest case.

Looking forward to your assistance. Thank you!

[markfoodyburton]

Are you using KVM, HVF or TCG?
Do you REALLY want all instruction bus accesses to go over the TLM fabric? That will make the thing unbearable slow?

I'd guess this may not be the first instruction executed, but that you end up branching to 0 because of some sort of fault. Note that ld/st exclusives may not be correctly implemented outside of DMI memory, you may need to (re-)plumb that through QEMU

[alpha-beta-user]

I am using TCG.

I am considering using the get_mem_ptr method to start the firmware. Once the operating system is up and running, the interaction between the CPU and DDR can be conducted in TLM form. This approach allows transaction tracing for further analysis.

[alpha-beta-user]

but I think it should not get wrong. I just chang p_dmi to false.

it will just cause slow but not wrong I think.

[markfoodyburton]

True - please let us know if you find the issue !