Skip PCIe link-speed warnings for disabled devices

Disabled PCIe devices (sysfs enable==0) carry no data traffic; their
link state has no operational impact. Switchtec PCIe switch management
endpoints on NVIDIA HGX H100 baseboards (and similar fabric controllers)
train at reduced speed intentionally and were producing spurious warnings.

Check is vendor-agnostic: reads enable attribute via existing helper,
no vendor/device ID hardcoding.

Documented in bible-local/decisions/2026-06-12-pcie-disabled-device-link-warning.md.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-06-12 03:42:19 +03:00
parent e169a7722c
commit 2320925433
3 changed files with 52 additions and 0 deletions

View File

@@ -278,6 +278,11 @@ func readPCIStringAttribute(bdf, attribute string) (string, bool) {
// below the device maximum. Regular PCIe slots get Warning; NVLink bridge cards
// get Critical because they are fixed internal connectors that must always train
// to max speed — any downgrade signals a hardware fault.
//
// Disabled devices (sysfs enable==0) are skipped: they carry no data traffic and
// their link state has no operational impact. This covers management endpoints
// (e.g. PCIe switch fabric controllers on HGX baseboards) that the kernel never
// activates but that lspci still reports with link stats.
func applyPCIeLinkSpeedWarning(dev *schema.HardwarePCIeDevice) {
if dev.LinkSpeed == nil || dev.MaxLinkSpeed == nil {
return
@@ -285,6 +290,11 @@ func applyPCIeLinkSpeedWarning(dev *schema.HardwarePCIeDevice) {
if pcieLinkSpeedRank(*dev.LinkSpeed) >= pcieLinkSpeedRank(*dev.MaxLinkSpeed) {
return
}
if dev.BDF != nil {
if enabled, ok := readPCIIntAttribute(*dev.BDF, "enable"); ok && enabled == 0 {
return
}
}
desc := fmt.Sprintf("PCIe link speed degraded: running at %s, capable of %s", *dev.LinkSpeed, *dev.MaxLinkSpeed)
dev.ErrorDescription = &desc

View File

@@ -0,0 +1,41 @@
# Decision: Skip PCIe link-speed warnings for disabled devices
**Date:** 2026-06-12
**Status:** active
## Context
On HGX H100 SXM5 baseboards, the Microchip Switchtec PM41028 PSX PCIe switch
(vendor 11F8, device 4128, NVIDIA subsystem 10DE:1643) appears in `lspci` as a
"Memory controller". Its upstream link trains at Gen3 x2 while the device is
capable of Gen4 x16. The device is permanently in a disabled state: memory access
and bus-mastering are both off (Mem-, BusMaster-); `/sys/bus/pci/devices/<bdf>/enable`
reads `0`.
This chip is the PCIe fabric management endpoint for the NVSwitch interconnect — it
carries only management traffic at low bandwidth and is intentionally not activated
by any Linux driver. The bee audit was reporting a `statusWarning` with message
"PCIe link speed degraded" for this device, which is misleading because the device
is not in the data path.
## Decision
`applyPCIeLinkSpeedWarning` reads `/sys/bus/pci/devices/<bdf>/enable` via the
existing `readPCIIntAttribute` helper. If the value is `0` the function returns
early without setting any warning status.
The check is vendor-agnostic: it applies to any PCIe device that Linux has not
activated, regardless of make or model. This is consistent with the
`no-hardcoded-vendors` contract — no vendor ID, device ID, or name string is
used as a condition.
## Consequences
- PCIe fabric management endpoints, IPMI virtual devices, and other permanently
disabled PCIe functions no longer produce spurious link-degradation warnings.
- Real link degradation on active devices (GPUs, NICs, NVMe, NVLink bridges)
continues to be detected and reported as before.
- NVLink bridge cards retain their existing `statusCritical` path (they are always
enabled, so the early return is never taken for them).
- The Switchtec device on HGX H100 boards shows `statusOK` with no
`error_description` in the audit JSON.

View File

@@ -7,3 +7,4 @@ One file per decision, named `YYYY-MM-DD-short-topic.md`.
| 2026-03-05 | Use NVIDIA proprietary driver | active |
| 2026-04-01 | Treat memtest as explicit ISO content | active |
| 2026-04-29 | Treat embedded submodules as read-only | active |
| 2026-06-12 | Skip PCIe link-speed warnings for disabled devices | active |