Open rsnikhil opened 4 years ago
@rsnikhil : have got Carnyx back in sync ('fake' CPU meant to stress-test the memory system on scripted tests and random tests).
@gameboo : looking into resurrecting Cambridge's AXE tool
This task (rigorous testing of the memory system) was originally expected to start after getting FreeBSD to boot with virtio (based on ad hoc testing), but has already begun in order to debug the FreeBSD boot.
However, we've started hammering on the memory system using Carnyx, described in more detail in https://github.com/DARPA-SSITH-Demonstrators/BESSPIN-CloudGFE/issues/118#issuecomment-693672892
This has already uncovered 4+ bugs that have now been diagnosed and fixed.
This activity continues.
@jrtc27 we are booting FreeBSD on Cissr and the GFE to try shed some light on the failing FreeBSD boot on AWSteria/Verilator. The GFE repository contains a FreeBSD image, but is it old compared to the image you provide to Nikhil for AWSteria testing? If you, could you provide a FreeBSD image for GFE that is a close as possible to the one Nikhil is debugging on AWSteria? Thanks. CC: @rsnikhil @quark17 @joestoy
The one Nikhil has been trying is our GFE configuration, which works (well, should work, I guess) on both the GFE and AWSteria. I don’t know what the state of the image is that Galois have built, but it will be based on a different commit and not the exact same kernel config.
Thank you @jrtc27. I imagine that the device tree may need to change between AWSteria and GFE. If so, is there an easy way to change device trees? (Sorry for this very basic question).
It’s passed to BBL and FreeBSD by the boot ROM embedded in a BRAM at synthesis time, rather than being hard-coded in FeeeBSD itself like sometimes happens, so it should just work.
Got it. Thank you.
In addition to the usual tests (ISA tests, booting OS, ...), set up @rsnikhil 's 'RISCV_Mem_Test' system for stress-testing and adversarial-testing the memory system.