The main problem is not the huge amount of errors in the syslog. The main problem is the contention caused by the locate record ccws that end up in error. I have seen this happen anytime there are counts with the wrong cylinder anywhere on the disk and the first cylinder of the disk is in recognizable format. The only fast way I have found that prevents Linux from issuing locate records all over the device when the device is brought online is to write data to the first track of the first cylinder in order to make it unrecognizable as a format known to Linux. Then Linux will treat the disk as unformatted and the contention problem is no longer there. One thing I did not explain is why I'm insisting on doing the format from Linux: I have user interface that reports progress of a long format to the user and I have infrastructure for this reporting, which is in Linux. There isn't time to port this to CMS. I can format from CMS, but only as long as the format is quick. If I CMS format whole disks, then I can't report progress (within my existing infrastructure).
Tomas -----Original Message----- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Mark Post Sent: Wednesday, March 13, 2013 9:05 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: DASD format from Linux only >>> On 3/13/2013 at 12:20 PM, Stefan Haberland <s...@linux.vnet.ibm.com> wrote: > Hi Tomas, > > I have a possible solution for you from within Linux. You can set the > device online with raw_track_access enabled. > > $ echo 1 > /sys/bus/ccw/devices/0.0.XXXX/raw_track_access > $ chccwdev -e XXXX > > Please ignore the few Buffer I/O errors in syslog. I can't say this made a huge amount of difference. For a 1,000 cylinder TDISK, I got 572 lines of output from bringing a mis-formatted disk online. Using the raw_track_access, I got 444 lines of output. I don't know if the number of messages is proportional to the number of cylinders or not, but I suspect it is. So, either method is going to be generating a lot of output regardless. I think the best method is to CPFMTXA at _least_ cylinder 0 before giving it to a guest. It really should be the entire volume, or use the DIRMAINT function to erase things for you. Mark Post ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
