Re: 2.6.21-rc4-mm1

[Eric Rannaud <eric.rannaud@gmail.com>]

On Tue, Mar 27, 2007 at 07:17:49PM +0200, Cornelia Huck wrote:
> On Tue, 27 Mar 2007 11:25:57 +0200,
> "Kay Sievers" <kay.sievers@vrfy.org> wrote:
> > Checking the uevent return value, will not prevent any malfunction,
> > usually this kind of "error handling" just prevents bringing up a
> > whole subsystem, or booting-up a box, because the needed device does
> > not exist at all.
> 
> OK, if we consider uevents to be non-vital to a functioning device.

The reason for that original patch was that it is actually possible for the
uevent functions to return -ENOMEM, the uevent buffer being statically
allocated to BUFFER_SIZE (2048). It used to be 1024 but that was not
always enough and it was doubled a while ago [1]. Using add_uevent_var()
makes this less of a problem as such an overflow should be catched
cleanly [2].

> OTOH, I think using something like uevent_suppress (maybe via
> dev_uevent_filter?) is a saner way to suppress a uevent than to return
> an error code in the uevent function.

That makes sense, I guess. I will try that.

Thanks.


[1] http://marc.info/?t=113797361200002&r=1&w=2
[2] uevent-use-add_uevent_var-instead-of-open-coding-it.patch in rc4-mm1