Bug 3398

Summary: NForce3-250Gb IDE (PATA) lockup during boot
Product: IO/Storage Reporter: Bryan O'Sullivan (bos)
Component: IDEAssignee: Bartlomiej Zolnierkiewicz (bzolnier)
Status: REJECTED INSUFFICIENT_DATA    
Severity: normal CC: bunk
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.8.1, 2.6.9 Subsystem:
Regression: --- Bisected commit-id:

Description Bryan O'Sullivan 2004-09-14 10:01:33 UTC
Distribution:
        Fedora Core 2
Hardware Environment:
        MSI K8N Neo Platinum mobo (NForce3-250Gb chipset)
        Athlon 64 3200+
        Maxtor 250GB ATA-133 hard disk
Software Environment:
        nothing out of the ordinary
Problem Description:

I built a system around the above-described hardware last week.  It did not make
it all the way through a vanilla Fedora Core 2 (with 2.6.5 kernel) install,
spontaneously switching to runlevel 6 (reboot) partway through the install.

I managed to work around this by putting the hard disk into PIO mode, which
allowed the install to succeed.

However, on subsequently trying to boot the machine, it hung early in userspace
during boot about 90% of the time.  I made sure that I was booting with
apic=off, but it made no difference to stability.

A Google search indicated that this looked similar to an NForce2 IDE
interrupt-related freeze on recent kernels, so I ran "hdparm -X udma3" during
boot (as suggested in the posting I found), and now the system appears to be
rock-solid.

Steps to reproduce:

Let a recent kernel choose the UDMA level itself, and the machine freezes during
boot more than 90% of the time, and never survives heavy disk activity.
Comment 1 Bryan O'Sullivan 2004-09-14 10:02:15 UTC
By the way, I know this looks a bit like bug 3071, but that one is
SATA-specific, while I've only got PATA gear inside my system.
Comment 2 Bryan O'Sullivan 2004-11-17 10:25:04 UTC
This still occurs with 2.6.9.
Comment 3 Bryan O'Sullivan 2004-11-17 10:29:02 UTC
A final note: even forcing the hard disk to PIO mode doesn't necessarily help. 
For example, I/O hung three times during an upgrade from Fedora Core 2 to Fedora
Core 3.  The hang happened much earlier with DMA enabled, but still occurred
with PIO instead.

Each hang required a reboot, even though the kernel was still somewhat alive. 
Any interactions with processes that required disk I/O during each hang caused
whatever process was involved to get wedged.
Comment 4 Adrian Bunk 2006-01-12 14:46:15 UTC
Is this issue still present in kernel 2.6.15?
Comment 5 Bryan O'Sullivan 2006-01-12 14:48:17 UTC
I don't know.  I threw out my PATA drives in frustration.
Comment 6 Bartlomiej Zolnierkiewicz 2006-01-12 15:11:05 UTC
Hehe.  I'm closing the bug for now.