> One error message is ignorable. Hundreds are a problem that should be 
> fixed.

I just had to learn this the hard way... I started playing with 
raw_track_access, noticed the large number of I/O errors reported, but paid it 
no mind and made it work. I wrote a CKD track that was not recognized as a 
known format and I fixed my problem. Then I started to do more large scale 
testing and found out that while I can write the track to a minidisk on one 
real device, others would not let me write the track. I have noticed that the 
vendor (as reported in /sys) of the good device is IBM and the one that I 
cannot write to is HTC. And again I have a large wall of I/O errors and sense 
data...
I'm on the verge of giving up on trying to drive all this from Linux. We 
already have a working directory manager user exit that puts a label on every 
new disk to let Linux know that the disk is not formatted in any known format. 
That takes care of the problem.

The problem is that when Linux puts a disk online, it does analysis of the disk 
to determine if it is in a known format. If it finds a known format it starts 
reading all over the disk. If there are errors, like bad cylinders in the count 
field, all kinds of bad things can happen. There is no way to bring the disk 
online without doing the analysis, which means that Linux does not have a way 
to safely format its own disks.
If there was a way to tell the kernel to bring the device online without the 
analysis, the problem would be solved. There is a state called "unformatted" 
where the disk goes if kernel can't identify a known label. I would like to 
have a way to explicitly put a disk online in the unformatted state. For 
example, writing 1 to /sys/bus/ccw/devices/<vaddr>/online would bring the disk 
online with format detection and writing 2 would put it online without 
detection, ending up in "unformatted" state. If the user knew that they intend 
to format a disk with potentially unsafe data on it they could just bring it 
online unformatted. (By unsafe I mean data that may be partially in a good 
format and partially in bad).

Tomas

----------------------------------------------------------------------
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/

Reply via email to