Re: ATA write-dma interrupt was seen but timeout fired LBA=53346288

On Mon, Aug 16, 2004 at 04:19:11PM -0700, Jake Khuon wrote:
> ### On Mon, 16 Aug 2004 08:10:00 +0200, Marc van Kempen <marc_at_bowtie.nl>
> ### casually decided to expound upon freebsd-current_at_freebsd.org the
> ### following thoughts about "ATA write-dma interrupt was seen but timeout
> ### fired LBA=53346288":
> 
> MvK> After I resume I get these warnings:
> MvK> 
> MvK> ATA write-dma interrupt was seen but timeout fired LBA=53346288
> MvK> ATA write-dma interrupt was seen but timeout fired LBA=53346288
> MvK> ATA write-dma interrupt was seen but taskqueue stalled LBA=53346288
> MvK> 
> MvK> This basically results in the harddisk being unaccessible and the systems 
> MvK> hangs eventually. Rebooting does not work because the harddisk can't be 
> MvK> synced, etc.
> MvK> 
> MvK> Does anyone know what is going on?
> 
> I had this same problem on my IBM T30 although I'm using APM.  I couldn't
> get ACPI suspend to play nice with XFree86 and GLX.
> 
try:

hw.syscons.sc_no_suspend_vtswitch=1
hw.acpi.reset_video=0

works for me, glx enabled and very fast resuming/suspending. In X and Console.

> It was very frustrating to not have reliable suspend.  However, I think I
> might have solved it.  Do you have a CDROM (or similar) device attached?
> 
> I added "atacontrol detach 1" to /etc/rc.suspend to detach the CDROM drive
> before suspending.  Likewise, I added a complementary "atacontrol attach 1"
> in /etc/rc.resume.  Since doing that, I have not seen the interrupt
> problems.  I do notice now that shortly before the re-attachment kicks in
> after coming back from suspend, I will see:
> 
> Aug 16 15:45:02 Mocha kernel: stray irq15
> 

I see exactly the same thing (from some quick testing), manually attaching
detaching the cdrom drive (actually dvd/cdrom-writer) makes suspend/resume
work reliably!

> I'm not sure if this is normal or is an indication of some weird IRQ storm. 
> At anyrate, detaching and reattaching during the suspend/resume cycle seems
> to damp it and stablised the whole ATA subsystem.
> 

The only problem left is that with acpi the file /etc/rc.{suspend,resume}
don't seem to get called.

Cheers,
Marc.