[Linuxppc-users] Fedora 28-1.1 taking 30 seconds to discover/enable PCIe adapter after link disable/enable

[Mike Bieker]

Thanks for the help.  Once I get access to the system again will get the
logs, etc.

-----Original Message-----
From: Benjamin Herrenschmidt [mailto:benh at au1.ibm.com]
Sent: Tuesday, June 05, 2018 9:34 AM
To: Mike Bieker; Michael Neuling; Steve Pittman
Cc: Linuxppc-users; Brian King; linuxppc-users at lists.ozlabs.org
Subject: Re: [Linuxppc-users] Fedora 28-1.1 taking 30 seconds to
discover/enable PCIe adapter after link disable/enable

On Tue, 2018-06-05 at 09:17 -0600, Mike Bieker wrote:
> I've tested retrains on the system also and those worked fine.  For our HW
> validations we do a full set of thousands:
> - Hot Resets (Secondary Bus Resets)
> - Link Retrains
> - Speed Changes
> - Power Management L1
> - Link Disable/Enable
>
> I believe both the Link Disable/Enable and the Hot Reset tests on the PPC
> system failed in the same way where it took 30 to 60 seconds for the link
> to
> come back up after the reset/disable.  We really need to be able to
> perform
> these to validate our HW and Serdes, so if there is some hack that would
> be
> great.

Right so we need to block triggering an error on link down. At the
moment we trigger a fence, which cause the whole EEH machinery to kick
in, reset the whole host bridge etc... It should take 30 to 60s though,
more like 5s ... so there might be something else going on there... but
still too long anyway.

Can you give us an example of one of those "30s" cases so we can give
it a try here and see what else might be going on ? (Do you have kernel
logs ?)

Cheers,
Ben.