Talos II - second CPU support and test automation

Another post about our adventures with porting coreboot for Talos II. This phase focused on enabling second CPU and its internal devices. We also expanded our test suite.

Binaries and their signatures can be found on Talos II release page. In installation instructions you can find steps needed both for permanent flashing as well as for temporary use of different firmware image.

Second CPU initialization

First of the problems we had to overcome was accessing SCOM (Serial COMmunication) registers on the second CPU. For first CPU we were using XSCOM (eXtended SCOM) which was conveniently set up for us by SBE (Self Boot Engine, small PowerPC-lite core inside POWER9 SoC), now we have to do something similar for the other processor. Before that happens, all SCOM accesses are done through FSI (Flexible Service Interface).

This added another layer in accessing buses that are already accessed through SCOM. An example of such bus is I2C, which is used (among other uses) for accessing MVPD (Module VPD). MVPD holds important data about CPU, like its operating frequencies and voltages or list of enabled cores. It must be read before XSCOM is enabled, meaning it has to happen using slow FSI communication. How slow are we talking about? Reading 64 KB of MVPD takes 4 seconds when done through XSCOM (primary CPU), but 9 seconds over FSI (secondary CPU). It is used both in romstage and ramstage, so we quickly decided it is worth to cache MVPD between stages. Next step will be saving it to flash and read it through I2C only if necessary, i.e. after flashing whole PNOR or when processor was swapped and saved image doesn’t match hardware. Hostboot does it this way, so should we. This could save about 13 seconds of boot time.

Next item on the list was enabling and training X bus links. An X bus is the socket-to-socket SMP interconnect between two POWER9 processors. Before this step each CPU worked independently of the other one, in so-called island mode. In that mode it would be impossible to directly access address spaces (memory and MMIO) behind remote CPU. Enabling SMP makes XSCOM access to secondary processor’s devices possible.

With SCOM, I2C and X Bus out of the way, most of initialization consisted of repeating existing steps for the second CPU. There were some exceptions where simple repetition was not enough, most notably:

  • TOD (time of day) - in case of discrepancy between two clocks one of them takes precedence in setting the other. Complex calculations are done during the initialization to choose “more important” clock source - the one with higher delay, as it is more inert.
  • memory controller - even though memory initialization and training is repeated for secondary processor, only main CPU sets a common memory space address translation maps.
  • PCIe controller - links are split differently between ports connected to each of processor sockets.

In addition to various initialization steps, we had to update device tree passed to Skiboot to include information about new CPU and devices it adds. We also switched away from FIT payload and started using ELF. Main motivation for this change was fact that FIT decompression code does not work as expected in coreboot. With payload compressed with LZMA, flash memory footprint of whole coreboot and Skiboot has shrunk to just below 700 KB. For comparison, Hostboot takes about 30 MB of space.

New tests

In order to achieve greater reliability of the firmware and to avoid bugs test base on the Talos II platform has been improved.

Currently, before each firmware release, the following test suites will be performed:

All test cases documentation is available under this link. Full test matrix for Talos II is available under this link

Summary

We will soon start long and tedious process of upstreaming those changes, after some additional cleanup. This may put most of further development on hold to keep rebasing effort as low as possible. Nevertheless, we will in parallel start researching what it would take to add TPM support.

If you think we can help in improving the security of your firmware or you are looking for someone who can boost your product by leveraging advanced features of used hardware platform, feel free to book a call with us or drop us email to contact<at>3mdeb<dot>com. If you are interested in similar content feel free to sign up to our newsletter


Krystian Hebel
Firmware Engineer interested mostly in things deliberately omitted from documentation. Thinks C is high-level language, can write something in Verilog but doesn't know how to read it yet.