On Tue, 3 Aug 2010, haad wrote:
Well, still no hurry, but it would be nice if some additional eyes were
pointed this way. I've got the recursive-module-load test case added
to the existing tests/modules/ stuff (and it works)!
BTW, since this is changing the kernel ABI, I guess it will need a
version bump. Do we have any other bumps coming soon that we can
share?
I talked with Andrew today and he said that basicaly we can use
mutex_owned for this particular case. That should make your code much
simpler :) sorry for any additional work :)
Well, we've come full circle! :)
It seems to me that, even if it is OK to use mutex_owned() for this
particular case, perhaps we should not. Most everyone seemed to agree
that your condvar solution was technically correct, and if we did use
the mutex_owned() solution it would only serve as a temptation for
someone else in the future!
We now have two solutions, both of which actually work (tested on my
home machine). One solution (mutex_owned) might be slightly simpler and
slightly more efficient than the other (condvar), but this section of
code is certainly not performance-critical. Therefore, is there any
reason for preferring the mutex_owned solution?
I'm happy to go either way based on concensus here, but I'm leaning
towards the condvar solution. It doesn't require any exemptions or
long explanations of why it is OK to use mutex_owned in ways that
would appear to violate our own documentation. And it also addresses
eeh's (and mrg's) concerns of keeping the lock across potentially
long-duration operations.
-
| Paul Goyette | PGP Key fingerprint: | E-mail addresses: |
| Customer Service | FA29 0E3B 35AF E8AE 6651 | paul at whooppee.com|
| Network Engineer | 0786 F758 55DE 53BA 7731 | pgoyette at juniper.net |
| Kernel Developer | | pgoyette at netbsd.org |
-
