Who hasn't been frustrated by some device stuck in a hardware
read or write? Maybe it's a tape drive? It's effectively dead, maybe
because the person who wrote the driver is an idiot or maybe because
the person who designed the hardware is more of an idiot, or more likely
something has just gone tragically wrong and you are stuck. Literally
stuck, because the process doing the reading or writing dove down into
the driver and it's never coming back up for air. Never. Go ahead,
send it a "kill -9" signal. The process will never see that, because
it's way down at the bottom of the ocean waiting for something that
apparently is never going to happen.
On Linux, you'd see the process state is "D for "Defunct"". OS X uses "U" for
"uninterruptible". I don't know about you, but I think "U" makes more sense. After all, we have to suppose that it is at least possible that whatever trouble
is going down, it might resolve itself. If a very large counter is ticking down,
maybe if we wait long enough the driver will finish or give up? Yeah, right: we're
going to power-cycle the box and hope the problem doesn't repeat.
What an awful thing to have to do. Reboot a perfectly good system because a tape drive is stuck?
Wouldn't it be nice if you could kill the stupid thing? Or tell the driver to give up with an ioctl? Well,
don't get too excited, but somebody agrees and is working on just that.. Apparently it was Linus Torvald who
actually suggested this back in 2002, and now it's actually there - well,
somewhat anyway. The problem is that programmers need to use
this and of course unless someone has kindly rewritten a driver for
that purpose, it doesn't. This TASK_KILLABLE: New process state in Linux article from IBM describes its use in NFS.
Update: This was still in the include files as of 3.10, but isn't to be found in my Ubuntu 12.04 include files. I still find mention of it in this fuse, TASK_KILLABLE and daemon deadlocks thread from 2013, though.
The arguments AGAINST TASK_KILLABLE are that it complicates things. I think that among driver developers there's a certain amount of misplaced arrogance too: "MY code can't get stuck sleeping!".
But of course it can, because unexpected hardware flakiness can confuse
any code - well, maybe not "any", but it sure isn't unusual to see
something get stuck like this. Yes, of course these things can be solved by better code: don't assume, don't wait forever for anything no matter how sure you
are that it can't possibly block, cover your bases, yadda, yadda. Trouble
is, all that goes against tight and fast, doesn't it?
So, the author of this thought about it and:
A while back, Matthew Wilcox realized that many of these concerns about application bugs do not really apply if the application is about to be killed anyway. It does not matter if the developer thought about the possibility of an interrupted system call if said system call is doomed to never return to user space.
Makes sense to me, but I'm no kernel programmer.
Then again: if the problem is internal hardware that has gotten itself all confused with registers contradicting one another and the hardware design lacks any way to clear to a default state without removing power, you may have to shut
everything down anyway. Still.. isn't it better to be able to at least try
Got something to add? Send me email.
Increase ad revenue 50-250% with Ezoic
More Articles by Anthony Lawrence
Find me on Google+
© 2009-11-07 Anthony Lawrence