Re: kern/38742: ahcisata/jmide hang

[Gary Duzan <gary%duzan.org@localhost>]

The following reply was made to PR kern/38742; it has been noted by GNATS.

From: Gary Duzan <gary%duzan.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/38742: ahcisata/jmide hang 
Date: Tue, 03 Jun 2008 07:11:10 -0400

    It turns out that some local changes I made early on to try to
 work around the problem were left in place and distorting the
 analysis. Here is what I know now.
 
    The problem started with the changes on March 18 to "Split
 device_t and softc for ATA devices, as well as wd(4)." With the
 3/21 change to "set the atac dev pointer in the softc" in jmide.c
 (1.5), the behavior changed from a hang to a panic, where the trace
 shows:
 
 ahci_intr_port() at netbsd:ahci_intr_port+0x29
 ahci_intr() at netbsd:ahci_intr+0xaf
 jmide_intr() at netbsd:jmide_intr+0x28
 Xintr_ioapic_level5() at netbsd:Xintr_ioapic_level5+0xf3
 --- interrupt ---
 Xspllower() at netbsd:Xspllower+0xe
 configure() at netbsd:configure+0x3b
 main()
 
    ahci_intr_port+0x29 is ahcisata_core.c:435:
 
 435             struct ata_xfer *xfer = chp->ch_queue->active_xfer;
 
 I did some testing a while back to determine whether chp or ch_queue
 was null, but I can't find my notes, so I'll have to redo that.
 
    If I disable "ahcisata at jmide", the system makes it to single
 user without the SATA or CD drive. If I disable "jmide", the system
 makes it to single user without the CD, but with the SATA attached
 to pciide.
 
                                        Gary Duzan