Skip to content

docs(hardware): firmware and os tables as of 2026-10-21 - #115

Open
mgheorghe wants to merge 25 commits into
opiproject:mainfrom
mgheorghe:fw-refresh-2026-10
Open

mgheorghe wants to merge 25 commits into
opiproject:mainfrom
mgheorghe:fw-refresh-2026-10

Conversation

@mgheorghe

Copy link
Copy Markdown
Contributor

Refresh the hardware READMEs to what the lab runs after the 2026-10 update round.

Depends on #114: this branch is stacked on it. Until #114 merges, review only the last 5 commits.

Changes

Area Change
hardware/*/README.md "Firmware and OS" tables as of 2026-10-21
OS mgmt, dh1-dh5, tgen1 on Ubuntu 26.04.1 / 7.0.0-38; dh9 Fedora 43 bootc 7.0.7
mgmt Kea DHCPv4 3.0.3 replaces isc-dhcp-server; NAT, BIOS F48, BMC 13.06.29
dh1 known issue and workaround for megaraid_sas on 6.17+ (CVE-2026-90008)
dh1, dh2 1G forced on the copper SFP ports (Broadcom firmware 38.x)
dh3 CN106 SDK 12.24.03 and U-Boot 2023.01 rows; console how-to refreshed
dh4 the card is the Intel IPU E2100 (MacKay Peak), not Dayton Peak
dh9 Mount Evans (DYP) row: installed, needs its auxiliary power cable
A100G EOS 4.33.10M (the ToR stays on 4.32.0F for now)
TS, ips.md terminal server at 172.22.0.30 from Kea, firmware 2.32.6
server-setup.md Ubuntu 26.04 install media

Copy /etc/dhcp/dhcpd.conf from 172.22.0.1 verbatim (2026-10-19) so later
changes are reviewable diffs against what is actually running.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
The tor entry used the VLAN 102 CPU MAC (:65) and 172.22.1.250, which is
already reserved for arista100g. Point it at the ToR's Management1 MAC
28:99:3a:9c:39:64 and its real address 172.22.0.5.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
The empty 172.22.0.0/16 block was a leftover from the ISC sample file and
duplicated the real subnet that carries the 172.22.222.0/24 pool.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
BF3 in DH5 (ToR Et36) had no reservation and was leasing 172.22.222.2/.3
from the unknown pool, so 172.22.3.5 and 172.22.4.5 looked down. Pin the
Arm OOB (:f4) to 172.22.3.5 and the on-card BMC (:f5) to 172.22.4.5.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
TBD02 (Senao OUI d4:e5:c9:00:ec:30, ToR Et34) is the OpenBMC on the DH9
DPU, next to dh9dpu (:31). Name it dh9dpubmc and give it the DPU-BMC
address matching its host number.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
TBD03 (a0:36:bc:cb:cf:f8) is DH9's AMI BMC answering through the shared
host NIC on ToR Et28 (dedicated port is dh9bmc on Et29). Move it out of
the DPU-BMC range into the BMC range as dh9bmcshared.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
TBD04 (Avocent OUI 00:e0:86:37:59:9c, ToR Et3 = TS port on the cabling
diagram) is the ACS8000. Give it its planned 172.22.0.30 address and drop
the old commented ts entry whose MAC was never seen on the ToR.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
TBD01 (02:04:9f:70:8e:be, ToR Et2) is the CN106 management MCU (Zephyr,
'net iface' on DH3 /dev/ttyUSB1), not the DPU OS. Replace the placeholder
MAC in dh3cn106mcu with it and drop TBD01. The CN106 Linux has no OOB
link and is reached on DH3 /dev/ttyUSB0.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
The DH4 IPU E2100 IMC now presents 10:2e:00:00:fe:fe and was leasing
172.22.222.4, which broke the ansible jump host (root@172.22.4.4) for mev.
Point the reservation at the new MAC and record the MAC history.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
eno12409np1 (d4:04:e6:2a:b0:21, ToR Et30) was leasing 172.22.222.1 on the
same /16 as eno12399np0, giving two default routes and ARP flux. Add a
netplan override (sorts after 50-cloud-init.yaml so cloud-init cannot undo
it). Applied on DH4 2026-10-19; netplan apply alone kept the old lease,
'networkctl reconfigure eno12409np1' removed it.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Correct the dh4 IMC comment from 00715dc: DH4 carries an Intel IPU
Adapter E2100-CCQDA2HL (MacKay Peak), and the Mount Evans (DYP) card was
moved to DH9, so the old MACs predate a card swap. Rename dh4dypimc to
dh4e2100imc and label the DH9 DPU entries.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
The repo copy predated the EOS 4.32.0F upgrade and was missing VLANs
100/102/200, the Et24/Et32 trunks to the BIG-IP pair and the Et54/1
VLAN 102 link to the A100G.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Et1 (he.net uplink), Et6 (mgmt public NIC), Et7 (mgmt BMC) and Et23/Et31
(BIG-IP external) move to VLAN 10. Lab VLAN 1 no longer shares an L2
domain with the internet. The BIG-IP trunks Et24/Et32 are pinned to
VLANs 1,100,102,200 so VLAN 10 never reaches them.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
INTERFACESv4 was empty, so dhcpd listened on every interface including
the public one. Restrict it to enp1s0f0np0 (172.22.0.1/16).

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Ingress ACL PUBLIC-IN on the he.net uplink: from the internet the mgmt
BMC (216.218.200.249) only accepts HTTPS, replies to its own sessions,
NTP/DNS answers and ICMP. SSH and IPMI (623) from the internet are
dropped. Lab and VPN users still reach it through mgmt on VLAN 10,
which does not cross Et1.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
mgmt sshd on the public IP saw ~17k failed root logins a day. Ban for
1h after 5 failures in 10 min; lab, WireGuard, BIG-IP management and
the public /27 are never banned.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
…0-19 sweep

Fix the F5 internal addresses (.227 F5A, .228 F5B, .229 floating), add
the dhcpd host names, the CN106 MCU, the DH4 E2100 IMC, the DH9 BMC
(dedicated and shared) and the Mount Evans card now in DH9. Keep the
mev/mevbmc inventory names, which telegraf and Grafana depend on, and
add the DH9 DPU commented out until its credentials are known.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
…0-19 sweep

- move DH9 to the bottom group (U37-40, above DH5)
- relabel DH4 DPU as Intel IPU E2100 (MacKay Peak), IMC 172.22.4.4
- add DH9 Mount Evans 172.22.3.9 / BMC 172.22.4.9, fix DH5 title and host IP
- fix F5 model typo, add BIG-IP-01/02 names; drop the bogus mgmt IPMI 172.22.2.0
- rewire TS consoles: 1 PDU, 2 A100G, 4 F5B, 5 ToR, 7 F5A; label TS 172.22.4.104
- hardware/TS: ACS8032 model, firmware, SSH access and the port table
- ips.md: TS answers on 172.22.4.104

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
mgmt moved from isc-dhcp-server to kea-dhcp4 on 2026-10-20 (3.0.3 since the Ubuntu 26.04 upgrade).
Add the live kea-dhcp4.conf, drop dhcpd.conf and /etc/default/isc-dhcp-server, and update ips.md,
server-setup.md, the sztp guide (option 143 via flex_option) and the mgmt README (Kea, NAT,
fail2ban, firmware and OS versions).

The terminal server now takes 172.22.0.30 from Kea and runs firmware 2.32.6.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Broadcom firmware 38.x / 238.1.x will not link the FS 1000BASE-T SFP with
autonegotiation on. Add the systemd .link files deployed on both hosts.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Add or refresh the Firmware and OS table in the A100G, dh1-dh5, dh9 and tgen1 READMEs from a
read-only snapshot of the lab (Redfish, SSH, EOS), after the move to Ubuntu 26.04.1 and the
2026-10 firmware round. Name the dh4 card correctly (Intel IPU E2100, MacKay Peak), add the CN106
SDK and U-Boot rows on dh3, note the dh9 Mount Evans power cable, and update the KVM and nPDU
firmware versions.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>


On kernel 6.17 and later megaraid_sas overruns its PRP chain frame on requests over about 2 MiB,
and dh1's NVMe virtual disk behind the PERC H755N wedges. dh1 stays on 7.0 with
max_sectors_kb capped at 1024 through a udev rule and an initramfs hook.

Signed-off-by: Mircea Dan Gheorghe <mircea-dan.gheorghe@keysight.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant