On Fri, Jun 01, 2007 at 12:33:29PM -1000, J. David Beutel wrote: > Excellent! Thanks! I've gleaned the following from your blog. Is this > correct? > > * A week ago you committed a change that will: > ** get current SMART parameters and faults for SATA on x86 via a single > function in a private library using SCSI emulation; > ** decide whether they indicate any problem I need to be aware of in > terms of over temperature, predictive failure, and self-test failure; > ** periodically check for above problems and generate an FMA ereport and > fault.
Yep, that's about the gist of it. > I haven't used FMA yet, but this all sounds like just what I'm looking > for! Questions: > > * What Nevada build will this be in? This will be in build 65 of Nevada. > * No support for PATA? Nope. Only devices that use the SATA framework (Marvell, Silicon Image, and others - I don't remember the full list) use the SCSI emulation required to make this work. > * Do I need any special SATA configuration to get the SCSI emulation? No, but your SATA HBA must be one of those supported by the SATA framework. Otherwise it will operate in PATA legacy mode and the information will not be available. > * Does ZFS mirror boot work on SATA? Currently, there is a bug that prevents this from working. Basically, ZFS requires device IDs or /dev paths to open devices. For some unknown reason, ldi_open_by_devid() doesn't work for SATA devices early in boot, and /dev is obviously not available. I've fixed this in my upcoming FMA wad by also storing the /devices path with the vdev and falling back to that. Your mileage may vary, but at least for marvell SATA devices, ZFS boot doesn't work at the moment. > * Self-test failures are reported, but self-tests cannot be run? They are not explicitly scheduled by software. The drive firmware itself often runs tests independent of any software control. > * Is there a utility to output the raw disk_status_get() results? No. At one point I had an option to fmtopo to dump the output but I ripped it out during codereview because there had been plans to port the method to become a property. Now the fate is somewhat up in the air. In the meantime, a 20-line C program will do the trick (use nvlist_print to print the results). Sorry about that. > * Are there HDD-model specific configurations for SMART parameters? Not currently. Hope that helps, - Eric -- Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
