Description of Problem: Using PowerEdge 500 and Seagate TR5 IDE tape drive (10/20GB capacity with firmware 8A51), dumping/restoring data hangs the system. Version-Release number of selected component (if applicable): Fairfax RC1 How Reproducible: Always Steps to Reproduce: 1. Boot PowerEdge 500 that has Seagate TR5 IDE tape drive with Fairfax RC1 2. "tar -cvf /dev/ht0 /usr/share/doc" 3. The system hangs Actual Results: As soon as the tape is accessed, the system hangs. Expected Results: Dumping and restoring data using an IDE tape should complete without any problem. Additional Information: DMA for the tape drive is enabled by default and the issue occures with DMA enabled. Once DMA is disabled and ide-scsi module used ("rmmod ide- tape" and "modprobe ide-scsi"), all works fine. This issue was also seen with RH7.1. Other issues are also seen with IDE tape, see: https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=38404 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=36628
This defect is considered MUST-FIX for Fairfax.
Did the new kernel work ?
DMA is still enabled with kernel-2.4.7-2.7.i686.rpm. I downloaded this RPM from your site, installed it, remake my initrd image, rebooted the system with this new image, but DMA was still enabled for the tape drive and the system hanged when trying to access it. If you need to add it to the ide-dma.c black list, `cat /proc/ide/hdd/model` gives "Seagate STT20000A".
can you attach the /proc/ide/hdd/settings file please ? (I've added it to blacklist as well)
Created attachment 29395 [details] Output of `cat /proc/ide/hdd/settings` where hdd is the ide tape
Quite sure fixed for RC2, please test as soon as you get it (or test the -2.18 kernel Arjan posted to roswell-list about earlier, that will also have the fix)
Issue still exists with Fairfax RC2. DMA is enabled for the ide-tape device by default.
It is failing with kernel 2.4.7-6 (Fairfax RC2).
Using 2.4.7-7 kernel (from Rawhide): DMA was disabled for this device during install (2.4.7-7BOOT kernel). But DMA is enabled when booting system with this kernel, in which case the system hangs as soon as the tape is accessed (with DMA enabled).
Does this also persist in the final release?
I have now reproduced this on a seagate TR5 on a PE500 with 2.4.7-10. /proc/ide/hdc/settings (it's a master in my setup) shows using_dma as "0" Using "ide=nodma" works around the problem without bothering with ide-scsi at all. Well, not quite reproduced. I can do some I/O to the tape; it took about a minute for the system to hang. Once it hangs, it hangs with interrupts disabled. Since this hasn't been reproduced on anything but a PE500 as far as I know, I suspect OSB5 driver.
Typo, I means "CSB5" driver, not "OSB5" driver. I have also tested this with 2.4.9-0.8.2, which has some IDE cleanups, but the results were exactly the same; ide=nodma is required to make the device work. Have you reproduced this on any other machine besides the PE500?
The hang occurs only on PE500.
That's your only hardware that uses the CSB5 driver, right? More evidence that it is a bug in that particular driver.
Yes, at this time, that is the only system that uses CSB5.
Trimming the cc: list.
Still happens in Hampton beta 3
initscripts have been modified to disable DMA on tapes and cdrom drives by default.
Unfortunately, at least in our testing here, disabling DMA only for the tape/cdrom was not sufficient. This is either a CSB5 driver bug or a simple hardware bug.
Closing as hardware bug.