> I'd like to draft a proposal to EOF/eliminate the handling of persistent > kstats. IMO, a kstat should be cleaned up once kstat_destroy() is > called, unconditionally. (The definition of KSTAT_FLAG_PERSISTENT > should be left alone, with admonishments against its use.) This does > mean that update_drv and driver detaches will result in loss of > historical information.
The notion of all such data being "historical" seems a bit odd. In particular, especially on DEBUG systems, the administrator may not be aware of when attach and detach operations are occurring. So the result will be that kstats for occasionally-needed kernel modules will be constantly getting cleared. That doesn't seem good to me. However, I could see an argument for having update_drv firing off an ioctl to torch persistent kstats so that the updated driver can reliably reattach. -- meem _______________________________________________ opensolaris-code mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
