On Wed, Jul 29, 2015 at 12:19:57PM +0200, Michal Suchanek wrote:
Please use subject lines matching the style for the subsytsem so people
can spot that the patch is in some way relevant.
> The controller-data subnode has no compatible. This can lead to other
> drivers getting confused by it. Add a compatible to make devicetreee
> unambiguous.
I can't tell from this commit message what the issue you're trying to
fix is, sorry. Nodes without compatible strings are entirely normal and
don't need compatible strings. It sounds like a bug in whatever other
driver is becoming confused.
> + if (!of_get_property(data_np, "compatible", NULL) ||
> + strcmp(of_get_property(data_np, "compatible", NULL),
> + "samsung,s3c-controller-data"))
> + dev_err(&spi->dev, "child node 'controller-data' does not have correct compatible\n");
This will break all existing users which is not acceptable for
mainline, we need to preserve compatibility with existing device trees.
On 29 July 2015 at 16:00, Mark Brown <[email protected]> wrote:
> On Wed, Jul 29, 2015 at 12:19:57PM +0200, Michal Suchanek wrote:
>
> Please use subject lines matching the style for the subsytsem so people
> can spot that the patch is in some way relevant.
>
>> The controller-data subnode has no compatible. This can lead to other
>> drivers getting confused by it. Add a compatible to make devicetreee
>> unambiguous.
>
> I can't tell from this commit message what the issue you're trying to
> fix is, sorry. Nodes without compatible strings are entirely normal and
> don't need compatible strings. It sounds like a bug in whatever other
> driver is becoming confused.
The driver that gets confused is ofpart.
The two-line patch to allow it to just ignore controller-data has been
rejected on the basis that s3c64xx should use a compatible string
because ofpart monopolizes all nodes without compatible which are
children of a mtd device. Devicetrees containing such nodes that are
not partitions are presumably invalid and should be rejected when
ofpart is compiled into the kernel.
>
>> + if (!of_get_property(data_np, "compatible", NULL) ||
>> + strcmp(of_get_property(data_np, "compatible", NULL),
>> + "samsung,s3c-controller-data"))
>> + dev_err(&spi->dev, "child node 'controller-data' does not have correct compatible\n");
>
> This will break all existing users which is not acceptable for
> mainline, we need to preserve compatibility with existing device trees.
It will not break anything. It will just spam dmesg.
Thanks
Michal
On Wed, Jul 29, 2015 at 06:19:24PM +0200, Michal Suchanek wrote:
> On 29 July 2015 at 16:00, Mark Brown <[email protected]> wrote:
> > On Wed, Jul 29, 2015 at 12:19:57PM +0200, Michal Suchanek wrote:
> >
> > Please use subject lines matching the style for the subsytsem so people
> > can spot that the patch is in some way relevant.
> >
> >> The controller-data subnode has no compatible. This can lead to other
> >> drivers getting confused by it. Add a compatible to make devicetreee
> >> unambiguous.
> >
> > I can't tell from this commit message what the issue you're trying to
> > fix is, sorry. Nodes without compatible strings are entirely normal and
> > don't need compatible strings. It sounds like a bug in whatever other
> > driver is becoming confused.
>
> The driver that gets confused is ofpart.
>
> The two-line patch to allow it to just ignore controller-data has been
> rejected on the basis that s3c64xx should use a compatible string
It wasn't outright rejected, but it was questioned.
> because ofpart monopolizes all nodes without compatible which are
> children of a mtd device. Devicetrees containing such nodes that are
> not partitions are presumably invalid and should be rejected when
> ofpart is compiled into the kernel.
That characterization is currently correct. I'm not a fan of what ofpart
does in the first place; I'd really like it if its binding was written
in a way that made it more precise, but we have to live with what we
have.
I also didn't like that you were designing your system such that we were
now keying off the fact that your node has no 'reg' property. This
seemed awfully fragile. (Admittedly, so is ofpart.c.)
All in all, if you have better suggestions, I'm all ears. (At first
glance, I'm liking your patch 1 to help disambiguate both MTD partitions
and the 'controller-data' node going forward.)
Brian
On Wed, Jul 29, 2015 at 06:19:24PM +0200, Michal Suchanek wrote:
> On 29 July 2015 at 16:00, Mark Brown <[email protected]> wrote:
> > I can't tell from this commit message what the issue you're trying to
> > fix is, sorry. Nodes without compatible strings are entirely normal and
> > don't need compatible strings. It sounds like a bug in whatever other
> > driver is becoming confused.
> The driver that gets confused is ofpart.
> The two-line patch to allow it to just ignore controller-data has been
> rejected on the basis that s3c64xx should use a compatible string
> because ofpart monopolizes all nodes without compatible which are
> children of a mtd device. Devicetrees containing such nodes that are
> not partitions are presumably invalid and should be rejected when
> ofpart is compiled into the kernel.
That seems like an extremely limited binding, the normal thing here
would be to create a specifically named node to contain the collection
of subnodes like many PMICs do for their regulators. As a fix I'd
suggest just silently ignoring nodes it can't understand, or printing a
warning if that's a serious issue.
> >> + if (!of_get_property(data_np, "compatible", NULL) ||
> >> + strcmp(of_get_property(data_np, "compatible", NULL),
> >> + "samsung,s3c-controller-data"))
> >> + dev_err(&spi->dev, "child node 'controller-data' does not have correct compatible\n");
> > This will break all existing users which is not acceptable for
> > mainline, we need to preserve compatibility with existing device trees.
> It will not break anything. It will just spam dmesg.
I'm confused - if all this change does is to spam dmesg then what's the
point?
On 29 July 2015 at 19:16, Mark Brown <[email protected]> wrote:
> On Wed, Jul 29, 2015 at 06:19:24PM +0200, Michal Suchanek wrote:
>> On 29 July 2015 at 16:00, Mark Brown <[email protected]> wrote:
>
>> > I can't tell from this commit message what the issue you're trying to
>> > fix is, sorry. Nodes without compatible strings are entirely normal and
>> > don't need compatible strings. It sounds like a bug in whatever other
>> > driver is becoming confused.
>
>> The driver that gets confused is ofpart.
>
>> The two-line patch to allow it to just ignore controller-data has been
>> rejected on the basis that s3c64xx should use a compatible string
>> because ofpart monopolizes all nodes without compatible which are
>> children of a mtd device. Devicetrees containing such nodes that are
>> not partitions are presumably invalid and should be rejected when
>> ofpart is compiled into the kernel.
>
> That seems like an extremely limited binding, the normal thing here
> would be to create a specifically named node to contain the collection
> of subnodes like many PMICs do for their regulators. As a fix I'd
> suggest just silently ignoring nodes it can't understand, or printing a
> warning if that's a serious issue.
>
>> >> + if (!of_get_property(data_np, "compatible", NULL) ||
>> >> + strcmp(of_get_property(data_np, "compatible", NULL),
>> >> + "samsung,s3c-controller-data"))
>> >> + dev_err(&spi->dev, "child node 'controller-data' does not have correct compatible\n");
>
>> > This will break all existing users which is not acceptable for
>> > mainline, we need to preserve compatibility with existing device trees.
>
>> It will not break anything. It will just spam dmesg.
>
> I'm confused - if all this change does is to spam dmesg then what's the
> point?
Presumably when your SPI NOR flash fails to probe this message will be
just above and you will look into the binding doc and add the
compatible.
Thanks
Michal
On Wed, Jul 29, 2015 at 08:21:34PM +0200, Michal Suchanek wrote:
> On 29 July 2015 at 19:16, Mark Brown <[email protected]> wrote:
> >> It will not break anything. It will just spam dmesg.
> > I'm confused - if all this change does is to spam dmesg then what's the
> > point?
> Presumably when your SPI NOR flash fails to probe this message will be
> just above and you will look into the binding doc and add the
> compatible.
If you're looking to add a warning message when the flash device fails
to probe then add that in the flash driver when it detects an error,
this will cause needless noise for everyone else using this controller
purely to work around the broken binding.
And like I say compatible really seems like it isn't an appropriate
property here.
On 29 July 2015 at 20:40, Mark Brown <[email protected]> wrote:
> On Wed, Jul 29, 2015 at 08:21:34PM +0200, Michal Suchanek wrote:
>> On 29 July 2015 at 19:16, Mark Brown <[email protected]> wrote:
>
>> >> It will not break anything. It will just spam dmesg.
>
>> > I'm confused - if all this change does is to spam dmesg then what's the
>> > point?
>
>> Presumably when your SPI NOR flash fails to probe this message will be
>> just above and you will look into the binding doc and add the
>> compatible.
>
> If you're looking to add a warning message when the flash device fails
> to probe then add that in the flash driver when it detects an error,
> this will cause needless noise for everyone else using this controller
> purely to work around the broken binding.
Technically nobody needs to see the warning with in-tree boards since
the dts can be amended with the compatible.
There is no error in flash device driver.
There is only error parsing partition scheme. In my opinion this
should never cause an error. With disk drives failure to parse
partition table results in unpartitioned disk. As there are number of
partitioning schemes failure to parse one still does not prevent other
in succeding.
>
> And like I say compatible really seems like it isn't an appropriate
> property here.
So to sum up the discussion adding compatible to s3c64xx
controller-data is not desirable and making ofpart more robust is
desirable.
I think the suggestion to use a subnode for ofpart gives the most
robust solution.
Even adding compatibles to the partition subnodes ofpart still
monopolizes the address and cells property of the mtd node which does
not pass the 'if another driver did the same would it work together?'
test.
So my suggestion is to
1) search of ofpart subnode in mtd node. If present read address and
reg from it and search partitions as subnodes of the ofpart node. In
this case unknown nodes can cause error.
2) failing that issue a warning and try to parse ofpart partitions
from the mtd node itself. In this case unknown nodes cannot cause
error for compatibility with other drivers including the already
exisitn s3c64xx controller-data node.
The parser code can be the same for both cases and only operate on
different node with a flag to reject unknown subnodes or not.
Thanks
Michal
On Thu, Jul 30, 2015 at 10:24:37AM +0200, Michal Suchanek wrote:
> So my suggestion is to
> 1) search of ofpart subnode in mtd node. If present read address and
> reg from it and search partitions as subnodes of the ofpart node. In
> this case unknown nodes can cause error.
> 2) failing that issue a warning and try to parse ofpart partitions
> from the mtd node itself. In this case unknown nodes cannot cause
> error for compatibility with other drivers including the already
> exisitn s3c64xx controller-data node.
> The parser code can be the same for both cases and only operate on
> different node with a flag to reject unknown subnodes or not.
That seems reasonable to me.