On Mon, Jan 14, 2013 at 08:16:44AM -0800, Greg Kroah-Hartman wrote:
> On Mon, Jan 14, 2013 at 04:49:59PM +0100, Thierry Reding wrote:
> > On Wed, Jan 09, 2013 at 10:23:52AM +0100, Thierry Reding wrote:
> > > On Wed, Jan 09, 2013 at 01:19:39AM -0800, Dmitry Torokhov wrote:
> > > > On Wed, Jan 09, 2013 at 08:07:45AM +0100, Thierry Reding wrote:
> > > > > On Sun, Jan 06, 2013 at 11:57:48AM -0800, Dmitry Torokhov wrote:
> > > > > > On Sun, Jan 06, 2013 at 08:27:39PM +0100, Thierry Reding wrote:
> > > > > > > On Sat, Jan 05, 2013 at 12:06:58AM -0800, Dmitry Torokhov wrote:
> > > > > > > > On Sat, Jan 05, 2013 at 01:15:08PM +0530, Laxman Dewangan wrote:
> > > > > > > [...]
> > > > > > > > > @@ -735,25 +738,16 @@ static int tegra_kbc_probe(struct platform_device *pdev)
> > > > > > > > > spin_lock_init(&kbc->lock);
> > > > > > > > > setup_timer(&kbc->timer, tegra_kbc_keypress_timer, (unsigned long)kbc);
> > > > > > > > >
> > > > > > > > > - res = request_mem_region(res->start, resource_size(res), pdev->name);
> > > > > > > > > - if (!res) {
> > > > > > > > > - dev_err(&pdev->dev, "failed to request I/O memory\n");
> > > > > > > > > - err = -EBUSY;
> > > > > > > > > - goto err_free_mem;
> > > > > > > > > - }
> > > > > > > > > -
> > > > > > > > > - kbc->mmio = ioremap(res->start, resource_size(res));
> > > > > > > > > + kbc->mmio = devm_request_and_ioremap(&pdev->dev, res);
> > > > > > > > > if (!kbc->mmio) {
> > > > > > > > > - dev_err(&pdev->dev, "failed to remap I/O memory\n");
> > > > > > > > > - err = -ENXIO;
> > > > > > > > > - goto err_free_mem_region;
> > > > > > > > > + dev_err(&pdev->dev, "Cannot request memregion/iomap address\n");
> > > > > > > > > + return -EADDRNOTAVAIL;
> > > > > > > >
> > > > > > > > Erm, no, -EBUSY please.
> > > > > > >
> > > > > > > EADDRNOTAVAIL is the canonical error for devm_request_and_ioremap()
> > > > > > > failure. The kerneldoc comment in lib/devres.c even gives a short
> > > > > > > example that uses this error code.
> > > > > >
> > > > > > I am sorry, but I do not consider a function that was added a little
> > > > > > over a year ago as a canon. If you look at the uses of EADDRNOTAVAIL it
> > > > > > is used predominantly in networking code to indicate that attempted
> > > > > > _network_ address is not available.
> > > > >
> > > > > EBUSY might be misleading, though. devm_request_and_ioremap() can fail
> > > > > in both the request_mem_region() and ioremap() calls. Furthermore it'd
> > > > > be good to settle on a consistent error-code instead of doing it
> > > > > differently depending on subsystem and/or driver. Currently the various
> > > > > error codes used are:
> > > > >
> > > > > EBUSY, EADDRNOTAVAIL, ENXIO, ENOMEM, ENODEV, ENOENT, EINVAL,
> > > > > EIO, EFAULT, EADDRINUSE
> > > > >
> > > > > Also if we can settle on one error code we should follow up with a patch
> > > > > to make it consistent across the tree and also update that kerneldoc
> > > > > comment. I volunteer to do that if nobody else steps up. I'm also Cc'ing
> > > > > Wolfram (the original author), maybe he has some thoughts on this.
> > > > >
> > > >
> > > > If you going to change all drivers make devm_request_and_ioremap()
> > > > return ERR_PTR()-encoded errors and then we can differentiate what
> > > > part of it failed.
> > >
> > > Yeah, that thought also crossed my mind. I'll give other people some
> > > time to comment before hurling myself into preparing patches.
> >
> > I've prepared a patch that changes devm_request_and_ioremap() to return
> > ERR_PTR()-encoded errors and adjusts all callers. As you can imagine it
> > is a bit on the huge side. scripts/get_maintainer.pl lists 156 people
> > and mailing lists. I've gone through the list, and as far as I can tell
> > everyone on that list is actually affected by the patch, so there's not
> > much potential for tuning it down.
> >
> > There is also the issue of whose tree this should go into. Unfortunately
> > the patch can't be broken down into smaller chunks because it changes
> > how the devm_request_and_ioremap() function's return value is handled in
> > an incompatible way, so it won't be possible to gradually phase this in.
> > Furthermore I can imagine that until the end of the release cycle new
> > code may be added on which the same transformations need to be done. I
> > have a semantic patch to do the bulk of the work, but quite a bit of
> > coordination will be required.
> >
> > I'm adding Arnd and Greg on Cc, maybe they can advise on how best to
> > handle this kind of patch.
>
> You should provide a "wrapper" function that does the correct return
> value, convert drivers over to it, and then, when everyone is changed,
> drop the old call. To change 156 different drivers all at once, in a
> way that is not detectable by the compiler breaking the build, is not a
> good thing to do at all.
>
> By doing it in this manner, it will take longer, but you can push the
> patches through the different maintainer's trees. If they are slow,
> I'll be glad to take the remaining patches in my driver-core tree to do
> the final bits.
It certainly sounds like a less complicated way to do it. But it also
involves adding a function with a made up name and drop a function with
a perfectly good name instead. I wouldn't even know what name to choose
for the new API.
Thierry
Attachment:
pgpxcEd1YqE4r.pgp
Description: PGP signature
