Hi, I've read about some ATA DMA timeout problems recently here. I just want to tell you, I've got same problems recently. I had to install the kernel.old back to get rid of severe problems. Affected kernel version: FreeBSD 5.4-STABLE #0: Tue May 31 00:19:34 CEST 2005 Last known working version: FreeBSD 5.4-STABLE #0: Mon May 9 00:50:46 CEST 2005 Problems with yesterdays kernel: ad0: TIMEOUT - READ_DMA retrying (2 retries left) LBA=200649278 ad0: TIMEOUT - WRITE_DMA retrying (2 retries left) LBA=225457534 The system was slowed down. I noticed it while starting some desktop applications. Then I ran cvsup to update my ports tree. This led to a slowdown, which looked like the system was frozen. It was still moving (a bit), but I lost my patience and pressed the reset switch. Martin
Hello Martin, M> Affected kernel version: M> FreeBSD 5.4-STABLE #0: Tue May 31 00:19:34 CEST 2005 M> Last known working version: M> FreeBSD 5.4-STABLE #0: Mon May 9 00:50:46 CEST 2005 This looks to be the same problem alright. Can you confirm that the problem goes away when you revert to the earlier kernel? Also, as a matter of interest what size HD is your ad0? Regards, Tony. -- Tony Byrne
Hi, thank you, whoever fixed it and whatever it was, but -STABLE is stable again for me. The ATA DMA timeouts have disappeared with yesterdays kernel. Looking at the CVS tree, I notice that there weren't many changes to the ATA code before the problems appeared and also when it has been fixed. More info: I didn't update world to build this latest kernel and there were no changes in my kernel configuration. There is still one and only difference. I updated the nvidia-driver this time. I wonder if an "old" nvidia-driver can lead to ATA DMA errors. I don't think so, but I just want to tell you to make my report complete here. (Btw, nvidia-driver triggers a "panic: spinlock rm.mutex_mtx not in order list, when using "option WITNESS" without "option WITNESS_SKIPSPIN".) Thanks, Martin