Patchwork [v2] i2c: core: make it possible to match a pure device tree driver

login
register
mail settings
Submitter Linus Walleij
Date May 13, 2013, 8:18 p.m.
Message ID <1368476301-10495-1-git-send-email-linus.walleij@linaro.org>
Download mbox | patch
Permalink /patch/243523/
State Accepted
Headers show

Comments

Linus Walleij - May 13, 2013, 8:18 p.m.
This tries to address an issue found when writing an MFD driver
for the Nomadik STw481x PMICs: as the platform is using device
tree exclusively I want to specify the driver matching like
this:

static const struct of_device_id stw481x_match[] = {
	{ .compatible = "st,stw4810", },
	{ .compatible = "st,stw4811", },
	{},
};

static struct i2c_driver stw481x_driver = {
	.driver = {
		.name	= "stw481x",
		.of_match_table = stw481x_match,
	},
	.probe		= stw481x_probe,
	.remove		= stw481x_remove,
};

However that turns out not to be possible: the I2C probe code
is written so that the probe() call is always passed a match
from i2c_match_id() using non-devicetree matches.

This is probably why most devices using device tree for I2C
clients currently will pass no .of_match_table *at all* but
instead just use .id_table from struct i2c_driver to match
the device. As you realize that means that the whole idea with
compatible strings is discarded, and that is why we find strange
device tree I2C device compatible strings like "product" instead
of "vendor,product" as you could expect.

Let's figure out how to fix this before the mess spreads. This
patch will allow probeing devices with only an of_match_table
as per above, and will pass NULL as the second argument to the
probe() function. If the driver wants to deduce secondary info
from the struct of_device_id .data field, it has to call
of_match_device() on its own match table in the probe function
device tree probe path.

If drivers define both an .of_match_table *AND* a i2c_driver
.id_table, the .of_match_table will take precedence, just
as is done in the i2c_device_match() function in i2c-core.c.

I2C devices probed from device tree should subsequently be
fixed to handle the case where of_match_table() is
used (I think none of them do that today), and platforms should
fix their device trees to use compatible strings for I2C devices
instead of setting the name to Linux device driver names as is
done in multiple cases today.

Cc: Rob Herring <rob.herring@calxeda.com>
Cc: Grant Likely <grant.likely@linaro.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
---
ChangeLog v1->v2:
- Use of_match_device() to determine if there is a DT match in
  the probe code. If there is a match we pass NULL for the
  id_table match parameter.
---
 drivers/i2c/i2c-core.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
Linus Walleij - May 22, 2013, 7:56 a.m.
On Mon, May 13, 2013 at 10:18 PM, Linus Walleij
<linus.walleij@linaro.org> wrote:

> This tries to address an issue found when writing an MFD driver
> for the Nomadik STw481x PMICs: as the platform is using device
> tree exclusively I want to specify the driver matching like
> this:
(...)
> ---
> ChangeLog v1->v2:
> - Use of_match_device() to determine if there is a DT match in
>   the probe code. If there is a match we pass NULL for the
>   id_table match parameter.

Ping on this.

v2 should be doing what Grant suggested...

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Linus Walleij - May 30, 2013, 8 a.m.
On Wed, May 22, 2013 at 9:56 AM, Linus Walleij <linus.walleij@linaro.org> wrote:
> On Mon, May 13, 2013 at 10:18 PM, Linus Walleij
> <linus.walleij@linaro.org> wrote:
>
>> This tries to address an issue found when writing an MFD driver
>> for the Nomadik STw481x PMICs: as the platform is using device
>> tree exclusively I want to specify the driver matching like
>> this:
> (...)
>> ---
>> ChangeLog v1->v2:
>> - Use of_match_device() to determine if there is a DT match in
>>   the probe code. If there is a match we pass NULL for the
>>   id_table match parameter.
>
> Ping on this.
>
> v2 should be doing what Grant suggested...

Ping, nocheinmal.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang - June 7, 2013, 9:32 p.m.
...
> I2C devices probed from device tree should subsequently be
> fixed to handle the case where of_match_table() is
> used (I think none of them do that today), and platforms should
> fix their device trees to use compatible strings for I2C devices
> instead of setting the name to Linux device driver names as is
> done in multiple cases today.

I guess your solution is the least intrusive one. Still, it could happen
that of_match_table is scanned three times (by driver core, i2c layer,
and i2c driver) which is IMO an indication to look for a more elegant
solution tp find out what really matched?
Linus Walleij - June 10, 2013, 12:18 p.m.
On Fri, Jun 7, 2013 at 11:32 PM, Wolfram Sang <wsa@the-dreams.de> wrote:
> ...
>> I2C devices probed from device tree should subsequently be
>> fixed to handle the case where of_match_table() is
>> used (I think none of them do that today), and platforms should
>> fix their device trees to use compatible strings for I2C devices
>> instead of setting the name to Linux device driver names as is
>> done in multiple cases today.
>
> I guess your solution is the least intrusive one. Still, it could happen
> that of_match_table is scanned three times (by driver core, i2c layer,
> and i2c driver) which is IMO an indication to look for a more elegant
> solution tp find out what really matched?

I think that is a generic problem with the device tree
being completely stateless, and rather a comment on the
of_match_device() intrinsics being inelegant than on this
patch?

Do you see it as a blocker for the patch?

What happens before this patch is that instead of looping
over the of_match_table, the id_table is always iterated
to the end also in the DT case, yielding NULL as the last
argument to driver->probe() anyway.

Maybe the OF people have some comments on this...
I cannot really see how it could be any different given
the way the FDT works :-/

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang - June 10, 2013, 1:40 p.m.
On Mon, Jun 10, 2013 at 02:18:17PM +0200, Linus Walleij wrote:
> On Fri, Jun 7, 2013 at 11:32 PM, Wolfram Sang <wsa@the-dreams.de> wrote:
> > ...
> >> I2C devices probed from device tree should subsequently be
> >> fixed to handle the case where of_match_table() is
> >> used (I think none of them do that today), and platforms should
> >> fix their device trees to use compatible strings for I2C devices
> >> instead of setting the name to Linux device driver names as is
> >> done in multiple cases today.
> >
> > I guess your solution is the least intrusive one. Still, it could happen
> > that of_match_table is scanned three times (by driver core, i2c layer,
> > and i2c driver) which is IMO an indication to look for a more elegant
> > solution tp find out what really matched?
> 
> I think that is a generic problem with the device tree
> being completely stateless, and rather a comment on the
> of_match_device() intrinsics being inelegant than on this
> patch?

Yes.

> Do you see it as a blocker for the patch?

No blocker. Yet, I was hoping for somebody perhaps having a good idea.
Platform devices have 'id_entry', for example. Sadly, I don't have the
resources to pick up a similar idea.
Grant Likely - June 12, 2013, 1:20 p.m.
On Fri, 7 Jun 2013 23:32:42 +0200, Wolfram Sang <wsa@the-dreams.de> wrote:
> ...
> > I2C devices probed from device tree should subsequently be
> > fixed to handle the case where of_match_table() is
> > used (I think none of them do that today), and platforms should
> > fix their device trees to use compatible strings for I2C devices
> > instead of setting the name to Linux device driver names as is
> > done in multiple cases today.
> 
> I guess your solution is the least intrusive one. Still, it could happen
> that of_match_table is scanned three times (by driver core, i2c layer,
> and i2c driver) which is IMO an indication to look for a more elegant
> solution tp find out what really matched?

It's what we do on platform_devices. It really isn't an expensive
operation so I haven't pushed anyone to go optimize it.

g.
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Grant Likely - June 12, 2013, 1:22 p.m.
On Mon, 10 Jun 2013 15:40:14 +0200, Wolfram Sang <wsa@the-dreams.de> wrote:
> On Mon, Jun 10, 2013 at 02:18:17PM +0200, Linus Walleij wrote:
> > On Fri, Jun 7, 2013 at 11:32 PM, Wolfram Sang <wsa@the-dreams.de> wrote:
> > > ...
> > >> I2C devices probed from device tree should subsequently be
> > >> fixed to handle the case where of_match_table() is
> > >> used (I think none of them do that today), and platforms should
> > >> fix their device trees to use compatible strings for I2C devices
> > >> instead of setting the name to Linux device driver names as is
> > >> done in multiple cases today.
> > >
> > > I guess your solution is the least intrusive one. Still, it could happen
> > > that of_match_table is scanned three times (by driver core, i2c layer,
> > > and i2c driver) which is IMO an indication to look for a more elegant
> > > solution tp find out what really matched?
> > 
> > I think that is a generic problem with the device tree
> > being completely stateless, and rather a comment on the
> > of_match_device() intrinsics being inelegant than on this
> > patch?
> 
> Yes.
> 
> > Do you see it as a blocker for the patch?
> 
> No blocker. Yet, I was hoping for somebody perhaps having a good idea.
> Platform devices have 'id_entry', for example. Sadly, I don't have the
> resources to pick up a similar idea.
> 

I did at one time have a patch that kept the pointer around in the
struct device if an of_match_table succeeded, but it caused subtle race
conditions that weren't easy to solve. I reverted back to just calling
of_match_table() several times because it was easy.

g.

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang - June 12, 2013, 5:09 p.m.
On Mon, May 13, 2013 at 10:18:21PM +0200, Linus Walleij wrote:
> This tries to address an issue found when writing an MFD driver
> for the Nomadik STw481x PMICs: as the platform is using device
> tree exclusively I want to specify the driver matching like
> this:
> 
> static const struct of_device_id stw481x_match[] = {
> 	{ .compatible = "st,stw4810", },
> 	{ .compatible = "st,stw4811", },
> 	{},
> };
> 
> static struct i2c_driver stw481x_driver = {
> 	.driver = {
> 		.name	= "stw481x",
> 		.of_match_table = stw481x_match,
> 	},
> 	.probe		= stw481x_probe,
> 	.remove		= stw481x_remove,
> };
> 
> However that turns out not to be possible: the I2C probe code
> is written so that the probe() call is always passed a match
> from i2c_match_id() using non-devicetree matches.
> 
> This is probably why most devices using device tree for I2C
> clients currently will pass no .of_match_table *at all* but
> instead just use .id_table from struct i2c_driver to match
> the device. As you realize that means that the whole idea with
> compatible strings is discarded, and that is why we find strange
> device tree I2C device compatible strings like "product" instead
> of "vendor,product" as you could expect.
> 
> Let's figure out how to fix this before the mess spreads. This
> patch will allow probeing devices with only an of_match_table
> as per above, and will pass NULL as the second argument to the
> probe() function. If the driver wants to deduce secondary info
> from the struct of_device_id .data field, it has to call
> of_match_device() on its own match table in the probe function
> device tree probe path.
> 
> If drivers define both an .of_match_table *AND* a i2c_driver
> .id_table, the .of_match_table will take precedence, just
> as is done in the i2c_device_match() function in i2c-core.c.
> 
> I2C devices probed from device tree should subsequently be
> fixed to handle the case where of_match_table() is
> used (I think none of them do that today), and platforms should
> fix their device trees to use compatible strings for I2C devices
> instead of setting the name to Linux device driver names as is
> done in multiple cases today.
> 
> Cc: Rob Herring <rob.herring@calxeda.com>
> Cc: Grant Likely <grant.likely@linaro.org>
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>

Applied to for-next, thanks!

Thanks also to Grant for the insight.
Linus Walleij - June 13, 2013, 9:02 a.m.
On Wed, Jun 12, 2013 at 3:20 PM, Grant Likely <grant.likely@linaro.org> wrote:
> On Fri, 7 Jun 2013 23:32:42 +0200, Wolfram Sang <wsa@the-dreams.de> wrote:

>> I guess your solution is the least intrusive one. Still, it could happen
>> that of_match_table is scanned three times (by driver core, i2c layer,
>> and i2c driver) which is IMO an indication to look for a more elegant
>> solution tp find out what really matched?
>
> It's what we do on platform_devices. It really isn't an expensive
> operation so I haven't pushed anyone to go optimize it.

I tried to think of something and it ended up with ideas like
decorating the device tree representation and it was just ...
ouch.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Stephen Warren - June 17, 2013, 3:48 p.m.
On 05/22/2013 01:56 AM, Linus Walleij wrote:
> On Mon, May 13, 2013 at 10:18 PM, Linus Walleij
> <linus.walleij@linaro.org> wrote:
> 
>> This tries to address an issue found when writing an MFD driver
>> for the Nomadik STw481x PMICs: as the platform is using device
>> tree exclusively I want to specify the driver matching like
>> this:
> (...)
>> ---
>> ChangeLog v1->v2:
>> - Use of_match_device() to determine if there is a DT match in
>>   the probe code. If there is a match we pass NULL for the
>>   id_table match parameter.
> 
> Ping on this.
> 
> v2 should be doing what Grant suggested...

This has just shown up in next-20130617, and breaks at least the
TPS65910 and TPS62360 drivers, since they assume that the id parameter
passed to probe is non-NULL. However, now the parameter is NULL since
these drivers have both an ID table and an OF match table.

I'd like to suggest this patch be reverted an re-introduced immediately
after the merge window. That should give enough time for everyone to get
a heads-up on fixing any drivers with a similar problem, rather than
trying to cram all that in immediately before the merge window. I'd also
suggest that this patch should be accompanied by fixes for any similarly
broken drivers, based on code inspection.

Do people agree? If not, please let us know ASAP so we can post patches
to fix this.

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Linus Walleij - June 17, 2013, 4:33 p.m.
On Mon, Jun 17, 2013 at 5:48 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:

> This has just shown up in next-20130617, and breaks at least the
> TPS65910 and TPS62360 drivers, since they assume that the id parameter
> passed to probe is non-NULL. However, now the parameter is NULL since
> these drivers have both an ID table and an OF match table.

So you mean they come in through the DT boot path and assume
that parameter is non-null even though they should not make use of
it?

> I'd like to suggest this patch be reverted an re-introduced immediately
> after the merge window. That should give enough time for everyone to get
> a heads-up on fixing any drivers with a similar problem, rather than
> trying to cram all that in immediately before the merge window.

OK that works for me, I'm not in any hurry.

Wolfram get to decide how to handle this...

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Stephen Warren - June 17, 2013, 4:49 p.m.
On 06/17/2013 10:33 AM, Linus Walleij wrote:
> On Mon, Jun 17, 2013 at 5:48 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
> 
>> This has just shown up in next-20130617, and breaks at least the
>> TPS65910 and TPS62360 drivers, since they assume that the id parameter
>> passed to probe is non-NULL. However, now the parameter is NULL since
>> these drivers have both an ID table and an OF match table.
> 
> So you mean they come in through the DT boot path and assume
> that parameter is non-null even though they should not make use of
> it?

Yes. Since this wasn't true/enforced before, it probably wasn't clear
that rule existed.
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Grant Likely - June 17, 2013, 10:15 p.m.
On Mon, Jun 17, 2013 at 5:33 PM, Linus Walleij <linus.walleij@linaro.org> wrote:
> On Mon, Jun 17, 2013 at 5:48 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
>
>> This has just shown up in next-20130617, and breaks at least the
>> TPS65910 and TPS62360 drivers, since they assume that the id parameter
>> passed to probe is non-NULL. However, now the parameter is NULL since
>> these drivers have both an ID table and an OF match table.
>
> So you mean they come in through the DT boot path and assume
> that parameter is non-null even though they should not make use of
> it?
>
>> I'd like to suggest this patch be reverted an re-introduced immediately
>> after the merge window. That should give enough time for everyone to get
>> a heads-up on fixing any drivers with a similar problem, rather than
>> trying to cram all that in immediately before the merge window.
>
> OK that works for me, I'm not in any hurry.

Deferring by a merge window isn't going to make it any less painful.
Do your best to find all the users that need to be changed. Use a
coccinelle search perhaps, but I think it should be merged anyway.

g.
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Stephen Warren - June 17, 2013, 11:25 p.m.
On 06/17/2013 04:15 PM, Grant Likely wrote:
> On Mon, Jun 17, 2013 at 5:33 PM, Linus Walleij <linus.walleij@linaro.org> wrote:
>> On Mon, Jun 17, 2013 at 5:48 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
>>
>>> This has just shown up in next-20130617, and breaks at least the
>>> TPS65910 and TPS62360 drivers, since they assume that the id parameter
>>> passed to probe is non-NULL. However, now the parameter is NULL since
>>> these drivers have both an ID table and an OF match table.
>>
>> So you mean they come in through the DT boot path and assume
>> that parameter is non-null even though they should not make use of
>> it?
>>
>>> I'd like to suggest this patch be reverted an re-introduced immediately
>>> after the merge window. That should give enough time for everyone to get
>>> a heads-up on fixing any drivers with a similar problem, rather than
>>> trying to cram all that in immediately before the merge window.
>>
>> OK that works for me, I'm not in any hurry.
> 
> Deferring by a merge window isn't going to make it any less painful.

It'll give a lot more time for people to be exposed to the change and
hence find/fix it linux-next rather than seeing it for the first time in
Linus's tree.
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang - June 18, 2013, 7:33 a.m.
On Mon, Jun 17, 2013 at 11:15:30PM +0100, Grant Likely wrote:
> On Mon, Jun 17, 2013 at 5:33 PM, Linus Walleij <linus.walleij@linaro.org> wrote:
> > On Mon, Jun 17, 2013 at 5:48 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
> >
> >> This has just shown up in next-20130617, and breaks at least the
> >> TPS65910 and TPS62360 drivers, since they assume that the id parameter
> >> passed to probe is non-NULL. However, now the parameter is NULL since
> >> these drivers have both an ID table and an OF match table.
> >
> > So you mean they come in through the DT boot path and assume
> > that parameter is non-null even though they should not make use of
> > it?
> >
> >> I'd like to suggest this patch be reverted an re-introduced immediately
> >> after the merge window. That should give enough time for everyone to get
> >> a heads-up on fixing any drivers with a similar problem, rather than
> >> trying to cram all that in immediately before the merge window.
> >
> > OK that works for me, I'm not in any hurry.
> 
> Deferring by a merge window isn't going to make it any less painful.
> Do your best to find all the users that need to be changed. Use a
> coccinelle search perhaps, but I think it should be merged anyway.

I'll try a bit of my coccinelle-foo today and then decide.
Linus Walleij - June 18, 2013, 7:44 a.m.
On Tue, Jun 18, 2013 at 9:33 AM, Wolfram Sang <wsa@the-dreams.de> wrote:
> On Mon, Jun 17, 2013 at 11:15:30PM +0100, Grant Likely wrote:
>> On Mon, Jun 17, 2013 at 5:33 PM, Linus Walleij <linus.walleij@linaro.org> wrote:
>> > OK that works for me, I'm not in any hurry.
>>
>> Deferring by a merge window isn't going to make it any less painful.
>> Do your best to find all the users that need to be changed. Use a
>> coccinelle search perhaps, but I think it should be merged anyway.
>
> I'll try a bit of my coccinelle-foo today and then decide.

Thanks Wolfram, much appreciated.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfram Sang - June 18, 2013, 3:55 p.m.
On Tue, Jun 18, 2013 at 09:44:17AM +0200, Linus Walleij wrote:
> On Tue, Jun 18, 2013 at 9:33 AM, Wolfram Sang <wsa@the-dreams.de> wrote:
> > On Mon, Jun 17, 2013 at 11:15:30PM +0100, Grant Likely wrote:
> >> On Mon, Jun 17, 2013 at 5:33 PM, Linus Walleij <linus.walleij@linaro.org> wrote:
> >> > OK that works for me, I'm not in any hurry.
> >>
> >> Deferring by a merge window isn't going to make it any less painful.
> >> Do your best to find all the users that need to be changed. Use a
> >> coccinelle search perhaps, but I think it should be merged anyway.
> >
> > I'll try a bit of my coccinelle-foo today and then decide.
> 
> Thanks Wolfram, much appreciated.

I am going to revert that commit. I was thinking back and forth, even
playing with the idea to remove the id as a parameter to probe for i2c
drivers and let them request the id from the i2c core when needed. But
now I found more side-effects. E.g. run-time based instantiation for i2c
devices is depending on an id-table. So, for now I keep insisting that
an id-table must exist. Looks like DT-only drivers need more thinking
and this is too late for 3.11.

Regards,

   Wolfram

Patch

diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
index 6b63cc7..17cd22a 100644
--- a/drivers/i2c/i2c-core.c
+++ b/drivers/i2c/i2c-core.c
@@ -240,7 +240,7 @@  static int i2c_device_probe(struct device *dev)
 		return 0;
 
 	driver = to_i2c_driver(dev->driver);
-	if (!driver->probe || !driver->id_table)
+	if (!driver->probe || (!driver->id_table && !dev->driver->of_match_table))
 		return -ENODEV;
 	client->driver = driver;
 	if (!device_can_wakeup(&client->dev))
@@ -248,7 +248,12 @@  static int i2c_device_probe(struct device *dev)
 					client->flags & I2C_CLIENT_WAKE);
 	dev_dbg(dev, "probe\n");
 
-	status = driver->probe(client, i2c_match_id(driver->id_table, client));
+	if (of_match_device(dev->driver->of_match_table, dev))
+		/* Device tree matching */
+		status = driver->probe(client, NULL);
+	else
+		/* Fall back to matching the id_table */
+		status = driver->probe(client, i2c_match_id(driver->id_table, client));
 	if (status) {
 		client->driver = NULL;
 		i2c_set_clientdata(client, NULL);