Re: [PATCH 16/18] llseek: automatically add .llseek fop

[Arnd Bergmann <arnd@arndb.de>]

On Wednesday 07 July 2010 23:55:25 Mike Frysinger wrote:
> > New drivers should normally not use
> > noop_llseek but instead use no_llseek and
> > call nonseekable_open at open time.
> > Existing drivers can be converted to do
> > the same when the maintainer knows for
> > certain that no user code relies on calling
> > seek on the device file.
> 
> can we skip the transition for devices that we already know ?

Yes, we could do that for the cases that call nonseekable_open
either by putting it into file_operations or by calling it
from their own open() function in the same file. These
are the two cases where the semantic patch currently adds
no_llseek.

The reason I added the explicit no_llseek was so I could
tell the difference between those file operations that
use nonseekable_open and those that spatch could for
some reason not identify as belonging into any of the
categories.

> > --- a/arch/blackfin/kernel/kgdb_test.c
> > +++ b/arch/blackfin/kernel/kgdb_test.c
> > static const struct file_operations kgdb_test_proc_fops = {
> > +       .llseek = noop_llseek,/* read and write both use no f_pos */
> > --- a/arch/blackfin/mach-bf561/coreb.c
> > +++ b/arch/blackfin/mach-bf561/coreb.c
> >  static const struct file_operations coreb_fops = {
> > +       .llseek = noop_llseek,/* no read or write fn */
> 
> neither of these drivers are seekable, so attempts to do so should be
> an error ...

Ok, then you should add ".open = nonseekable_open," to the file
operations and put that change into a tree that gets pulled into
linux-next. The semantic patch will do the right thing then,
whatever we decide.

	Arnd