Hardware Qualification for riscv64

Page to explain how riscv64 meets Teams/DSA's hardware requirements for buildds and porter boxes, as explained on parent page HardwareQualification containing links to the actual requirements.

Requirements for individual hosts: Must have

Out of band management [...]

[... with dedicated network port, preferably a BMC, or failing that, serial console and networked power bars]

The boards (SiFive's HiFive Unmatched) do not have integrated BMC, however:

• They have serial console and JTAG through USB. The boards installed at OSUOSL are connected to one another in a loop, but it should be possible to use a console server instead.
• The boards at OSUOSLs are controlled by networked power bars.

• The secondary location for buildd machines is manda

Rackmount

They are mounted on racks on standard 1U cases.

No human intervention to power on

The boards boot without manual intervention, controlled by a switched PDU.

They need the help of an ad-hoc device (*), simulating the press of a button when the power comes up, but it's been working flawlessly for all installed machines for months/years.

(*) By default the HiFive Unmatched doesn't start automatically when receiving power, it needs a button pressed to start. The ad-hoc device consists basically of two wires, two connectors and a 47µF capacitor capacitor between the motherboard and the front panel button of the chassis. It sends a pulse when power is applied, and then the board starts. It is therefore fully transparent when installed. The boards installed at OSUOSL are used that way with switched PDUs without issue.

The boards can be rebooted through the PDU or through the JTAG interface.

Warranty or post-warranty hardware support, preferably provided by the sponsor

The existing hardware (SiFive's HiFive Unmatched) is relatively inexpensive, they will not provide warranty in the usual sense but they might agree to replace the machines. Also, since it's being produced in small batches, this exact hardware might be difficult to replace with the same units as time goes on. But at the moment the boards are still available.

However we will have ~10 boards, which is not great in terms of having to manage more machines, but in terms of failures it is a bit of an advantage because if one (or a few) fail the capacity as a whole will degrade progressively -- it's not like if there's only a few big servers and so you can loose a big % of capacity in a single case of hardware failure.

The other good part is that new and relatively inexpensive hardware, and probably more powerful, is going to be available over the next months. So replacements would also be hardware upgrades, and they might actually happen even if the hardware doesn't fail.

Under the 'ownership' of Debian (a long-term loan can also work)

Yes.

Requirements for individual hosts: Would like to have

Production quality rather than pre-production hardware [...]

[... This can be waived in cases such as when there is no production hardware available at all.]

The hardware available so far has been produced in small batches and can be purchased by end users, but the small batches mean that they might not have the same level of QA of typical hardware sold in the millions. However the hardware has been very very stable.

There are alternatives on the market right now, but they don't have full upstream support in the Linux kernel, so they are not usable at the moment, but getting there.

And as hinted above, there are also plans from several vendors, including the ones from the current machines, to release new hardware in the coming months. Some of these platforms are marketed for PC/workstation (rather than SBCs) so they should have similar features in terms of hosting them, and be more powerful than the current machines.

Support for multiple drives (so we can do RAID)

The boards uses SSD for storage, which should be quite reliable. They have a single M2 2280 slot for an SSD.

If RAID is really required we can try to install a second (or more) drives through the PCIe slot, but it might mean changing the 1U case for a bigger one.

Requirements for ports: Must have/be

At least two machines to act as buildds

At the moment we have 5 HiFive Unmatched boards installed at OSUOSL acting as buildds.

At least one machine to act as a porter box

We have another HiFive Unmatched board installed at OSUOSL acting as porterbox:

• https://db.debian.org/machines.cgi?host=debian-riscv64-porterbox-01

Hosted in two different locations

At the moment most of the machines are at OSUOSL in Oregon, USA.

The manda is secondary location for riscv64 buildd machines

Supported in Debian stable [...]

[... Waived for ports approved by ftp-masters and the release team for inclusion in a future stable release.]

It is not supported in Debian stable as being managed by DSA is a pre-requisite to supported in Debian stable, but RMs do not have concerns about it being included in future stable releases.

The machine's architecture has an actively-maintained stable kernel in the archive

Yes. All the firmwares are also available in the archive.

Packages critical for DSA operations available: puppet, samhain, syslog-ng, ferm/pf, etc

Yes to the mentioned above.

The debian.org, debian.org-recommended, debian.org-recommended-bullseye and debian.org-recommended-linux packages are installed on the porterbox.

Requirements for ports: Would like to have

Hosted in two different legal jurisdictions

See "Hosted in two different locations".