Project

General

Profile

Bug #3865

AHCI: 88SE9230 watchdog timeouts, "hung" panic

Added by Alex Wilson about 7 years ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
driver - device drivers
Start date:
2013-07-03
Due date:
% Done:

0%

Estimated time:
Difficulty:
Medium
Tags:
needs-triage
Gerrit CR:

Description

I've been running a box with a Marvell 88SE9230 on a kernel including the recent patches for the 88SE9128. The two chips look very similar on the surface, so I figured it was worth a try.

The card seems to work fine at startup -- it detects disks just fine, I can use cfgadm to hotplug new disks on the ports and it all goes swimmingly.

However, when I put the zpool under some load (long seq writes at 30+MB/sec over SMB is what I've been doing) I start getting messages like this in /var/adm/messages:

Jul  3 23:49:20 metrix ahci: [ID 517647 kern.warning] WARNING: ahci1: watchdog port 3 satapkt 0xffffff02e001e4b8 timed out
Jul  3 23:49:20 metrix ahci: [ID 517647 kern.warning] WARNING: ahci1: watchdog port 3 satapkt 0xffffff02e156f3f0 timed out
Jul  3 23:49:20 metrix ahci: [ID 517647 kern.warning] WARNING: ahci1: watchdog port 3 satapkt 0xffffff02eb763a28 timed out
Jul  3 23:49:20 metrix ahci: [ID 517647 kern.warning] WARNING: ahci1: watchdog port 3 satapkt 0xffffff02eb88d238 timed out
Jul  3 23:49:20 metrix ahci: [ID 517647 kern.warning] WARNING: ahci1: watchdog port 3 satapkt 0xffffff02eb88d858 timed out

These usually come in short bursts like this every 10-20 minutes, during which no I/O to the zpool gets through. Sometimes the bursts are a bit longer -- long enough for zfs to decide to offline the disk. Note that it only seems to happen on one port for me so far. If, however, I take that disk and cable and put it on my Intel AHCI, it works perfectly with no such issues under the same load. If I take the disk+cable and put it on a different port of the same 88SE9230, the watchdog timeouts do occur.

I have had two cases so far, also, where zfs has decided that the pool is "hung" and panicked the whole machine. This seems relatively rare, though, compared to the regular watchdog timeout messages.


Related issues

Related to illumos gate - Feature #8947: Support the Marvell 88SE9230 PCIe to SATA controllerNew2018-01-03

Actions

History

#1

Updated by Marcel Telka over 2 years ago

  • Related to Feature #8947: Support the Marvell 88SE9230 PCIe to SATA controller added

Also available in: Atom PDF