Re: kern/56737: WDCTL_RST errors in 9.99.92 and 9.99.93

[David Brownlee <abs%absd.org@localhost>]

The following reply was made to PR kern/56737; it has been noted by GNATS.

From: David Brownlee <abs%absd.org@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/56737: WDCTL_RST errors in 9.99.92 and 9.99.93
Date: Thu, 31 Aug 2023 19:32:23 +0100

 I am in the process of moving an 8 disk ZFS setup from an older Dell
 T320 with an mpii controller to a newer machine with onboard Intel 600
 SATA and pcie AM1061 SATA cards.
 
 On a standard netbsd-10 kernel some of the disks (usually, but not
 exclusively 8TB Seagate Barracuda Compute) sometimes fail to probe on
 boot with:
 
 autoconfiguration error: ahcisata0 port 5: setting WDCTL_RST failed for drive 0
 wd1 at atabus1 drive 0
 autoconfiguration error: ahcisata0 port 5: setting WDCTL_RST failed for drive 0
 wd1: autoconfiguration error: IDENTIFY failed
 wd1: fixing 0 sector size
 wd1: secperunit and ncylinders are zero
 wd1: secperunit and (sectors or tracks) are zero
 autoconfiguration error: wd1: unable to open device, error = 19
 wd1(ahcisata0:5:0): using PIO mode 0
 
 This happens on both the onboard Intel and AM1061 SATA cards
 
 Adding AHCISATA_EXTRA_DELAY appears to allows the disks to reliably
 probe on every boot.
 
 I think in the short term we should default AHCISATA_EXTRA_DELAY on
 for netbsd-10, and possibly drop AHCI_QUIRK_EXTRA_DELAY as it seems to
 be more drive than controller related.
 
 It may make sense to have an opt-in faster timeout enabled on
 virtualised interfaces, or potentially if a timeout is too short have
 extra logic to reset the drive and retry the whole process with very
 generous timeouts.
 
 David