re: CVS commit: [pgoyette-compat] src/sys/compat/netbsd32

[Paul Goyette <paul%whooppee.com@localhost>]

On Thu, 24 Jan 2019, matthew green wrote:

> > Module Name:	src
> > Committed By:	pgoyette
> > Date:		Thu Jan 24 04:24:52 UTC 2019
> >
> > Modified Files:
> > 	src/sys/compat/netbsd32 [pgoyette-compat]: netbsd32_sysctl.c
> >
> > Log Message:
> > Use the hook to get the value of machine32
>
> hm.  why does the module need to use the hook?  doesn't it know the
> value internally, since it has to publish it via the hook?
>
> i don't understand the value of using hooks inside the publisher.
> they are useful for external consumers, i thought.

In this case the symbol is defined as __weak and might not exist.  So
we use a hook function to return a value if the hook has been set, or
a default value if the hook has not been set.  The hook's struct
itself is always present/allocated.

I suppose I could have avoided the hook by simply having another
machine32_ptr variable, rather than a hook function.  But I wasn't
totally clear on whether there were any race conditions during a
module unload which could result in retrieving a pointer to a value
which just got removed from memory.

Of course, if we had a reasonable way of dealing with weak symbols,
this mess would not be needed.



+------------------+--------------------------+----------------------------+
| Paul Goyette     | PGP Key fingerprint:     | E-mail addresses:          |
| (Retired)        | FA29 0E3B 35AF E8AE 6651 | paul at whooppee dot com   |
| Kernel Developer | 0786 F758 55DE 53BA 7731 | pgoyette at netbsd dot org |
+------------------+--------------------------+----------------------------+