Hi,
On 2024/4/26 03:10, Andy Shevchenko wrote:
> On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
>> On 2024/4/25 22:26, Andy Shevchenko wrote:
>>> It seems driver missed the point of proper use of device property APIs.
>>> Correct this by updating headers and calls respectively.
>> You are using the 'seems' here exactly saying that you are not 100% sure.
>>
>> Please allow me to tell you the truth: This patch again has ZERO effect.
>> It fix nothing. And this patch is has the risks to be wrong.
> Huh?! Really, stop commenting the stuff you do not understand.
I'm actually a professional display drivers developer at the downstream
in the past, despite my contribution to upstream is less. But I believe
that all panel driver developers know what I'm talking about. So please
have take a look at my replies.
>> Simple because the "ili9341_probe() ---> ili9341_dbi_prob()" code path
>> is DT dependent.
>>
>> First of all, the devm_of_find_backlight() is called in ili9341_dbi_probe()
>> under *non-DT* environment, devm_of_find_backlight() is just a just a
>> no-op and will return NULL. NULL is not an error code, so ili9341_dbi_probe()
>> won't rage quit. But the several side effect is that the backlight will
>> NOT works at all.
> Is it a problem?
Yes, it is.
The core problem is that the driver you are modifying has *implicit* *dependency* on DT.
The implicit dependency is due to the calling of devm_of_find_backlight(). This function
is a no-op under non-DT systems. Therefore, before the devm_of_find_backlight() and
the device_get_match_data() function can truly DT independent.
Removing the "OF" dependency just let the tigers run out from the jail.
It is not really meant to targeting at you, but I thinks, all of drm_panel drivers
that has the devm_of_find_backlight() invoked will suffer such concerns.
In short, the reason is that the *implicit* *dependency* populates and
the undefined behavior gets triggered.
I'm sure you know that device_get_match_data() is same with of_device_get_match_data()
for DT based systems. For non DT based systems, device_get_match_data() is just *undefined*
Note that ACPI is not in the scope of the discussion here, as all of the drm bridges and
panels driver under drivers/gpu/drm/ hasn't the ACPI support yet. Therefore, at present,
it safe to say that device_get_match_data() is *undefined* under no-DT environment.
Removing the "OF" dependency hints to us that it allows the driver to be probed as a
pure SPI device under non DT systems. When device_get_match_data() is called, it returns
NULL to us now. As a result, the drm driver being modified will tears down.
See bellow code snippet extracted frompanel-ilitek-ili9341.c:
```
ili->conf = of_device_get_match_data(dev);
if (!ili->conf) {
dev_err(dev, "missing device configuration\n");
return -ENODEV;
}
```
>> It is actually considered as fatal bug for *panels* if the backlight of
>> it is not light up, at least the brightness of *won't* be able to adjust.
>> What's worse, if there is no sane platform setup code at the firmware
>> or boot loader stage to set a proper initial state. The screen is complete
>> dark. Even though the itself panel is refreshing framebuffers, it can not
>> be seen by human's eye. Simple because of no backlight.
> Can you imagine that I may have different hardware that considered
> this is non-fatal error?
>
Yes, I can imagine.
I believe you have the hardware which make you patch correct to run
in 99.9% of all cases. But as long as there one bug happened, you patch
are going to be blamed.
Because its your patch that open the door, both from the perceptive of
practice and from the perceptive of the concept (static analysis).
>> Second, the ili9341_dbi_probe() requires additional device properties to
>> be able to works very well on the rotation screen case. See the calling
>> of "device_property_read_u32(dev, "rotation", &rotation)" in
>> ili9341_dbi_probe() function.
> Yes, exactly, and how does it object the purpose of this patch?
Because under *non-DT* environment, your commit message do not give a
valid description, how does the additional device property can be acquired
is not demonstrated.
And it is exactly your patch open the non-DT code path (way or possibility).
It isn't has such risks before your patch is applied. In other words,
previously, the driver has the 'OF' dependency as the guard, all of the
potential risk(or problem) are suppressed. It is a extremely safe policy,
and it is also a extremely perfect defend.
And suddenly, you patch release the dangerous tiger from the cage.
So I think you can imagine...
>> Combine with those two factors, it is actually can conclude that the
>> panel-ilitek-ili9394 driver has the *implicit* dependency on 'OF'.
>> Removing the 'OF' dependency from its Kconfig just trigger the
>> leakage of such risks.
> What?!
>
Posting a patch is actually doing the defensive works, such a saying
may not sound fair for you, but this is just the hash cruel reality.
Sorry for saying that. :(
>> My software node related patches can help to reduce part of the potential
>> risks, but it still need some extra work. And it is not landed yet.
> Your patch has nothing to do with this series.
With my patch applied, this is way to meet the gap under non-DT systems.
Users of this driver could managed to attach(complete) absent properties
to the SPI device with software node properties. Register the swnode
properties group into the system prior the panel driver is probed. There
may need some quirk. But at the least there has a way to go. When there
has a way to go, things become self-consistent. Viewed from both the
practice of viewpoint and the concept of viewpoint.
And the dangerous tiger will steer its way to the direction of "ACPI
support is missing". But both of will be safe then.
--
Best regards,
Sui
On 2024/4/26 04:43, Sui Jingfeng wrote:
> But both of will be safe then.
But both of us will be safe then.
--
Best regards,
Sui
Hi,
On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
> On 2024/4/26 03:10, Andy Shevchenko wrote:
> > On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
> > > On 2024/4/25 22:26, Andy Shevchenko wrote:
> > > > It seems driver missed the point of proper use of device property APIs.
> > > > Correct this by updating headers and calls respectively.
> > > You are using the 'seems' here exactly saying that you are not 100% sure.
> > >
> > > Please allow me to tell you the truth: This patch again has ZERO effect.
> > > It fix nothing. And this patch is has the risks to be wrong.
> > Huh?! Really, stop commenting the stuff you do not understand.
>
> I'm actually a professional display drivers developer at the downstream
> in the past, despite my contribution to upstream is less. But I believe
> that all panel driver developers know what I'm talking about. So please
> have take a look at my replies.
Most of the interactions you had in this series has been uncalled for.
You might be against a patch, but there's no need to go to such length.
As far as I'm concerned, this patch is fine to me in itself, and I don't
see anything that would prevent us from merging it.
Maxime
Hi,
On 2024/4/26 14:23, Maxime Ripard wrote:
> Hi,
>
> On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
>> On 2024/4/26 03:10, Andy Shevchenko wrote:
>>> On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
>>>> On 2024/4/25 22:26, Andy Shevchenko wrote:
>>>>> It seems driver missed the point of proper use of device property APIs.
>>>>> Correct this by updating headers and calls respectively.
>>>> You are using the 'seems' here exactly saying that you are not 100% sure.
>>>>
>>>> Please allow me to tell you the truth: This patch again has ZERO effect.
>>>> It fix nothing. And this patch is has the risks to be wrong.
>>> Huh?! Really, stop commenting the stuff you do not understand.
>> I'm actually a professional display drivers developer at the downstream
>> in the past, despite my contribution to upstream is less. But I believe
>> that all panel driver developers know what I'm talking about. So please
>> have take a look at my replies.
> Most of the interactions you had in this series has been uncalled for.
> You might be against a patch, but there's no need to go to such length.
>
> As far as I'm concerned, this patch is fine to me in itself, and I don't
> see anything that would prevent us from merging it.
No one is preventing you, as long as don't misunderstanding what other
people's technical replies intentionally. I'm just a usual and normal
contributor, I hope the world will better than yesterday. Saying such
thing to me may not proper, I guess you may want to talk to peoples
who has the push rights, just make sure it isn't a insult to the
professionalism of drm bridge community itself though.
We still grateful for you help and admire you numerous contribution,
thanks.
--
Best regards,
Sui
On Sat, Apr 27, 2024 at 01:57:46PM +0800, Sui Jingfeng wrote:
> Hi,
>
>
> On 2024/4/26 14:23, Maxime Ripard wrote:
> > Hi,
> >
> > On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
> > > On 2024/4/26 03:10, Andy Shevchenko wrote:
> > > > On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
> > > > > On 2024/4/25 22:26, Andy Shevchenko wrote:
> > > > > > It seems driver missed the point of proper use of device property APIs.
> > > > > > Correct this by updating headers and calls respectively.
> > > > > You are using the 'seems' here exactly saying that you are not 100% sure.
> > > > >
> > > > > Please allow me to tell you the truth: This patch again has ZERO effect.
> > > > > It fix nothing. And this patch is has the risks to be wrong.
> > > > Huh?! Really, stop commenting the stuff you do not understand.
> > > I'm actually a professional display drivers developer at the downstream
> > > in the past, despite my contribution to upstream is less. But I believe
> > > that all panel driver developers know what I'm talking about. So please
> > > have take a look at my replies.
> > Most of the interactions you had in this series has been uncalled for.
> > You might be against a patch, but there's no need to go to such length.
> >
> > As far as I'm concerned, this patch is fine to me in itself, and I don't
> > see anything that would prevent us from merging it.
>
> No one is preventing you, as long as don't misunderstanding what other
> people's technical replies intentionally. I'm just a usual and normal
> contributor, I hope the world will better than yesterday.
You should seriously consider your tone when replying then.
> Saying such thing to me may not proper, I guess you may want to talk
> to peoples who has the push rights
I think you misunderstood me. My point was that your several rants were
uncalled for and aren't the kind of things we're doing here.
I know very well how to get a patch merged, thanks.
> just make sure it isn't a insult to the professionalism of drm bridge
> community itself though.
I'm not sure why you're bringing the bridge community or its
professionalism. It's a panel, not a bridge, and I never doubted the
professionalism of anyone.
Maxime
Hi,
On 2024/4/29 19:55, Maxime Ripard wrote:
> On Sat, Apr 27, 2024 at 01:57:46PM +0800, Sui Jingfeng wrote:
>> Hi,
>>
>>
>> On 2024/4/26 14:23, Maxime Ripard wrote:
>>> Hi,
>>>
>>> On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
>>>> On 2024/4/26 03:10, Andy Shevchenko wrote:
>>>>> On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
>>>>>> On 2024/4/25 22:26, Andy Shevchenko wrote:
>>>>>>> It seems driver missed the point of proper use of device property APIs.
>>>>>>> Correct this by updating headers and calls respectively.
>>>>>> You are using the 'seems' here exactly saying that you are not 100% sure.
>>>>>>
>>>>>> Please allow me to tell you the truth: This patch again has ZERO effect.
>>>>>> It fix nothing. And this patch is has the risks to be wrong.
>>>>> Huh?! Really, stop commenting the stuff you do not understand.
>>>> I'm actually a professional display drivers developer at the downstream
>>>> in the past, despite my contribution to upstream is less. But I believe
>>>> that all panel driver developers know what I'm talking about. So please
>>>> have take a look at my replies.
>>> Most of the interactions you had in this series has been uncalled for.
>>> You might be against a patch, but there's no need to go to such length.
>>>
>>> As far as I'm concerned, this patch is fine to me in itself, and I don't
>>> see anything that would prevent us from merging it.
>> No one is preventing you, as long as don't misunderstanding what other
>> people's technical replies intentionally. I'm just a usual and normal
>> contributor, I hope the world will better than yesterday.
> You should seriously consider your tone when replying then.
>
>> Saying such thing to me may not proper, I guess you may want to talk
>> to peoples who has the push rights
> I think you misunderstood me. My point was that your several rants were
> uncalled for and aren't the kind of things we're doing here.
>
> I know very well how to get a patch merged, thanks.
>
>> just make sure it isn't a insult to the professionalism of drm bridge
>> community itself though.
> I'm not sure why you're bringing the bridge community or its
> professionalism. It's a panel, not a bridge, and I never doubted the
> professionalism of anyone.
I means that the code itself could be adopted, as newer and younger
programmer (like Andy) need to be encouraged to contribute. I express
no obvious objections, just hints him that something else probably
should also be taken into consideration as well.
On the other hand, we probably should allow other people participate
in discussion so that it is sufficient discussed and ensure that it
won't be reverted by someone in the future for some reasons. Backing
to out case happens here, we may need to move things forward. Therefore,
it definitely deserve to have a try. It is not a big deal even though
it gets reverted someday.
In the end, I don't mind if you think there is nothing that could
prevent you from merge it, but I still suggest you have a glance at
peoples siting at the Cc list. I'm busy now and I have a lot of other
tasks to do, and may not be able to reply you emails on time. So it up
to you and other maintainers to decide.
Thank you.
> Maxime
--
Best regards,
Sui
On Tue, Apr 30, 2024 at 12:54:39AM +0800, Sui Jingfeng wrote:
> On 2024/4/29 19:55, Maxime Ripard wrote:
> > On Sat, Apr 27, 2024 at 01:57:46PM +0800, Sui Jingfeng wrote:
> > > On 2024/4/26 14:23, Maxime Ripard wrote:
> > > > On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
> > > > > On 2024/4/26 03:10, Andy Shevchenko wrote:
> > > > > > On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
> > > > > > > On 2024/4/25 22:26, Andy Shevchenko wrote:
> > > > > > > > It seems driver missed the point of proper use of device property APIs.
> > > > > > > > Correct this by updating headers and calls respectively.
> > > > > > > You are using the 'seems' here exactly saying that you are not 100% sure.
> > > > > > >
> > > > > > > Please allow me to tell you the truth: This patch again has ZERO effect.
> > > > > > > It fix nothing. And this patch is has the risks to be wrong.
> > > > > > Huh?! Really, stop commenting the stuff you do not understand.
> > > > > I'm actually a professional display drivers developer at the downstream
> > > > > in the past, despite my contribution to upstream is less. But I believe
> > > > > that all panel driver developers know what I'm talking about. So please
> > > > > have take a look at my replies.
> > > > Most of the interactions you had in this series has been uncalled for.
> > > > You might be against a patch, but there's no need to go to such length.
> > > >
> > > > As far as I'm concerned, this patch is fine to me in itself, and I don't
> > > > see anything that would prevent us from merging it.
> > > No one is preventing you, as long as don't misunderstanding what other
> > > people's technical replies intentionally. I'm just a usual and normal
> > > contributor, I hope the world will better than yesterday.
> > You should seriously consider your tone when replying then.
> >
> > > Saying such thing to me may not proper, I guess you may want to talk
> > > to peoples who has the push rights
> > I think you misunderstood me. My point was that your several rants were
> > uncalled for and aren't the kind of things we're doing here.
> >
> > I know very well how to get a patch merged, thanks.
> >
> > > just make sure it isn't a insult to the professionalism of drm bridge
> > > community itself though.
> > I'm not sure why you're bringing the bridge community or its
> > professionalism. It's a panel, not a bridge, and I never doubted the
> > professionalism of anyone.
>
>
> I means that the code itself could be adopted, as newer and younger
> programmer (like Andy) need to be encouraged to contribute.
Andy has thousands of commits in Linux. He's *very* far from being a new
contributor.
> I express no obvious objections, just hints him that something else
> probably should also be taken into consideration as well.
That might be what you wanted to express, but you definitely didn't
express it that way.
> On the other hand, we probably should allow other people participate
> in discussion so that it is sufficient discussed and ensure that it
> won't be reverted by someone in the future for some reasons. Backing
> to out case happens here, we may need to move things forward. Therefore,
> it definitely deserve to have a try. It is not a big deal even though
> it gets reverted someday.
>
> In the end, I don't mind if you think there is nothing that could
> prevent you from merge it, but I still suggest you have a glance at
> peoples siting at the Cc list. I'm busy now and I have a lot of other
> tasks to do, and may not be able to reply you emails on time. So it up
> to you and other maintainers to decide.
> Thank you.
So far, you're the only one who reviewed those patches. I'm not sure
what you're talking about here.
Maxime
On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
> On 2024/4/26 03:10, Andy Shevchenko wrote:
> > On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
> > > On 2024/4/25 22:26, Andy Shevchenko wrote:
> > > > It seems driver missed the point of proper use of device property APIs.
> > > > Correct this by updating headers and calls respectively.
> > > You are using the 'seems' here exactly saying that you are not 100% sure.
To add here, "seems" is used to show that I have no knowledge on what was
the idea behind this implementation by the original author. Plus my knowledge
in the firmware node / device property APIs and use cases in Linux kernel.
> > > Please allow me to tell you the truth: This patch again has ZERO effect.
> > > It fix nothing. And this patch is has the risks to be wrong.
> > Huh?! Really, stop commenting the stuff you do not understand.
>
> I'm actually a professional display drivers developer at the downstream
> in the past, despite my contribution to upstream is less. But I believe
> that all panel driver developers know what I'm talking about. So please
> have take a look at my replies.
Okay.
> > > Simple because the "ili9341_probe() ---> ili9341_dbi_prob()" code path
> > > is DT dependent.
> > >
> > > First of all, the devm_of_find_backlight() is called in ili9341_dbi_probe()
> > > under *non-DT* environment, devm_of_find_backlight() is just a just a
> > > no-op and will return NULL. NULL is not an error code, so ili9341_dbi_probe()
> > > won't rage quit. But the several side effect is that the backlight will
> > > NOT works at all.
> > Is it a problem?
>
> Yes, it is.
>
> The core problem is that the driver you are modifying has *implicit* *dependency* on DT.
> The implicit dependency is due to the calling of devm_of_find_backlight(). This function
> is a no-op under non-DT systems.
Okay.
> Therefore, before the devm_of_find_backlight() and
> the device_get_match_data() function can truly DT independent.
True for the first part, not true for the second.
> Removing the "OF" dependency just let the tigers run out from the jail.
>
> It is not really meant to targeting at you, but I thinks, all of drm_panel drivers
> that has the devm_of_find_backlight() invoked will suffer such concerns.
> In short, the reason is that the *implicit* *dependency* populates and
> the undefined behavior gets triggered.
Still no problem statement. My hardware works nicely on non-DT environment.
(And since it's Arduino-based one, I assume it will work on DT environments
the very same way.)
> I'm sure you know that device_get_match_data() is same with of_device_get_match_data()
> for DT based systems. For non DT based systems, device_get_match_data() is just *undefined*
> Note that ACPI is not in the scope of the discussion here, as all of the drm bridges and
> panels driver under drivers/gpu/drm/ hasn't the ACPI support yet.
This patch shows exactly how to bring back the ACPI support to one of them
(as it's done for tinyDRM cases).
> Therefore, at present,
> it safe to say that device_get_match_data() is *undefined* under no-DT environment.
This is not true.
> Removing the "OF" dependency hints to us that it allows the driver to be probed as a
> pure SPI device under non DT systems. When device_get_match_data() is called, it returns
> NULL to us now. As a result, the drm driver being modified will tears down.
>
> See bellow code snippet extracted frompanel-ilitek-ili9341.c:
>
>
> ```
> ili->conf = of_device_get_match_data(dev);
> if (!ili->conf) {
> dev_err(dev, "missing device configuration\n");
> return -ENODEV;
> }
> ```
>
> > > It is actually considered as fatal bug for *panels* if the backlight of
> > > it is not light up, at least the brightness of *won't* be able to adjust.
> > > What's worse, if there is no sane platform setup code at the firmware
> > > or boot loader stage to set a proper initial state. The screen is complete
> > > dark. Even though the itself panel is refreshing framebuffers, it can not
> > > be seen by human's eye. Simple because of no backlight.
> > Can you imagine that I may have different hardware that considered
> > this is non-fatal error?
> >
> Yes, I can imagine.
>
> I believe you have the hardware which make you patch correct to run
> in 99.9% of all cases. But as long as there one bug happened, you patch
> are going to be blamed.
>
> Because its your patch that open the door, both from the perceptive of
> practice and from the perceptive of the concept (static analysis).
>
> > > Second, the ili9341_dbi_probe() requires additional device properties to
> > > be able to works very well on the rotation screen case. See the calling
> > > of "device_property_read_u32(dev, "rotation", &rotation)" in
> > > ili9341_dbi_probe() function.
> > Yes, exactly, and how does it object the purpose of this patch?
>
> Because under *non-DT* environment, your commit message do not give a
> valid description, how does the additional device property can be acquired
> is not demonstrated.
>
> And it is exactly your patch open the non-DT code path (way or possibility).
> It isn't has such risks before your patch is applied. In other words,
> previously, the driver has the 'OF' dependency as the guard, all of the
> potential risk(or problem) are suppressed. It is a extremely safe policy,
> and it is also a extremely perfect defend.
>
> And suddenly, you patch release the dangerous tiger from the cage.
> So I think you can imagine...
No, I can't, sorry. I don't see how dangerous will be the use of DRM panel
in a wrong configuration. The same can very well happen on improperly working
hardware (backlight part) or simply when somebody didn't correctly set a DT
or manually use it when it should not be. But again I see *no* problem
statement, only some worries.
And on top of that I made tinyDRM drivers to be accessible on ACPI platforms
and so far I have none complains about the tigers that left the cage.
> > > Combine with those two factors, it is actually can conclude that the
> > > panel-ilitek-ili9394 driver has the *implicit* dependency on 'OF'.
> > > Removing the 'OF' dependency from its Kconfig just trigger the
> > > leakage of such risks.
> > What?!
> >
> Posting a patch is actually doing the defensive works, such a saying
> may not sound fair for you, but this is just the hash cruel reality.
> Sorry for saying that. :(
So, the summary of your message is that:
- there's no understanding how ACPI (or any other non-DT fwnode based
environment) can utilise the driver
- there's a worry about some problems which can't be stated clearly
- there's a neglecting of the previous successful cases specific for DRM
(tinyDRM drivers)
As a result of the false input, the non-constructive conclusion was given.
And note, I converted dozens if not hundredth of drivers that used to be
OF-only and haven't heart any negative feedback before this case. Maybe
we (reviewers of my patches and maintainers who applied them and end users)
miss a BIG DEAL here? Please, elaborate how dropping OF dependency can be
dangerous as a free walking tiger.
> > > My software node related patches can help to reduce part of the potential
> > > risks, but it still need some extra work. And it is not landed yet.
> > Your patch has nothing to do with this series.
I am not going to repeat the above.
> With my patch applied, this is way to meet the gap under non-DT systems.
> Users of this driver could managed to attach(complete) absent properties
> to the SPI device with software node properties. Register the swnode
> properties group into the system prior the panel driver is probed. There
> may need some quirk. But at the least there has a way to go. When there
> has a way to go, things become self-consistent. Viewed from both the
> practice of viewpoint and the concept of viewpoint.
>
> And the dangerous tiger will steer its way to the direction of "ACPI
> support is missing". But both of will be safe then.
--
With Best Regards,
Andy Shevchenko
Hi,
On 2024/4/30 22:32, Andy Shevchenko wrote:
> On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
>> On 2024/4/26 03:10, Andy Shevchenko wrote:
>>> On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
>>>> On 2024/4/25 22:26, Andy Shevchenko wrote:
>>>>> It seems driver missed the point of proper use of device property APIs.
>>>>> Correct this by updating headers and calls respectively.
>>>> You are using the 'seems' here exactly saying that you are not 100% sure.
> To add here, "seems" is used to show that I have no knowledge on what was
> the idea behind this implementation by the original author. Plus my knowledge
> in the firmware node / device property APIs and use cases in Linux kernel.
>
>>>> Please allow me to tell you the truth: This patch again has ZERO effect.
>>>> It fix nothing. And this patch is has the risks to be wrong.
>>> Huh?! Really, stop commenting the stuff you do not understand.
>> I'm actually a professional display drivers developer at the downstream
>> in the past, despite my contribution to upstream is less. But I believe
>> that all panel driver developers know what I'm talking about. So please
>> have take a look at my replies.
> Okay.
>
>>>> Simple because the "ili9341_probe() ---> ili9341_dbi_prob()" code path
>>>> is DT dependent.
>>>>
>>>> First of all, the devm_of_find_backlight() is called in ili9341_dbi_probe()
>>>> under *non-DT* environment, devm_of_find_backlight() is just a just a
>>>> no-op and will return NULL. NULL is not an error code, so ili9341_dbi_probe()
>>>> won't rage quit. But the several side effect is that the backlight will
>>>> NOT works at all.
>>> Is it a problem?
>> Yes, it is.
>>
>> The core problem is that the driver you are modifying has *implicit* *dependency* on DT.
>> The implicit dependency is due to the calling of devm_of_find_backlight(). This function
>> is a no-op under non-DT systems.
> Okay.
>
>> Therefore, before the devm_of_find_backlight() and
>> the device_get_match_data() function can truly DT independent.
> True for the first part, not true for the second.
>
>> Removing the "OF" dependency just let the tigers run out from the jail.
>>
>> It is not really meant to targeting at you, but I thinks, all of drm_panel drivers
>> that has the devm_of_find_backlight() invoked will suffer such concerns.
>> In short, the reason is that the *implicit* *dependency* populates and
>> the undefined behavior gets triggered.
> Still no problem statement. My hardware works nicely on non-DT environment.
> (And since it's Arduino-based one, I assume it will work on DT environments
> the very same way.)
>
>> I'm sure you know that device_get_match_data() is same with of_device_get_match_data()
>> for DT based systems. For non DT based systems, device_get_match_data() is just *undefined*
>> Note that ACPI is not in the scope of the discussion here, as all of the drm bridges and
>> panels driver under drivers/gpu/drm/ hasn't the ACPI support yet.
> This patch shows exactly how to bring back the ACPI support to one of them
> (as it's done for tinyDRM cases).
>
>> Therefore, at present,
>> it safe to say that device_get_match_data() is *undefined* under no-DT environment.
> This is not true.
>
>> Removing the "OF" dependency hints to us that it allows the driver to be probed as a
>> pure SPI device under non DT systems. When device_get_match_data() is called, it returns
>> NULL to us now. As a result, the drm driver being modified will tears down.
>>
>> See bellow code snippet extracted frompanel-ilitek-ili9341.c:
>>
>>
>> ```
>> ili->conf = of_device_get_match_data(dev);
>> if (!ili->conf) {
>> dev_err(dev, "missing device configuration\n");
>> return -ENODEV;
>> }
>> ```
>>
>>>> It is actually considered as fatal bug for *panels* if the backlight of
>>>> it is not light up, at least the brightness of *won't* be able to adjust.
>>>> What's worse, if there is no sane platform setup code at the firmware
>>>> or boot loader stage to set a proper initial state. The screen is complete
>>>> dark. Even though the itself panel is refreshing framebuffers, it can not
>>>> be seen by human's eye. Simple because of no backlight.
>>> Can you imagine that I may have different hardware that considered
>>> this is non-fatal error?
>>>
>> Yes, I can imagine.
>>
>> I believe you have the hardware which make you patch correct to run
>> in 99.9% of all cases. But as long as there one bug happened, you patch
>> are going to be blamed.
>>
>> Because its your patch that open the door, both from the perceptive of
>> practice and from the perceptive of the concept (static analysis).
>>
>>>> Second, the ili9341_dbi_probe() requires additional device properties to
>>>> be able to works very well on the rotation screen case. See the calling
>>>> of "device_property_read_u32(dev, "rotation", &rotation)" in
>>>> ili9341_dbi_probe() function.
>>> Yes, exactly, and how does it object the purpose of this patch?
>> Because under *non-DT* environment, your commit message do not give a
>> valid description, how does the additional device property can be acquired
>> is not demonstrated.
>>
>> And it is exactly your patch open the non-DT code path (way or possibility).
>> It isn't has such risks before your patch is applied. In other words,
>> previously, the driver has the 'OF' dependency as the guard, all of the
>> potential risk(or problem) are suppressed. It is a extremely safe policy,
>> and it is also a extremely perfect defend.
>>
>> And suddenly, you patch release the dangerous tiger from the cage.
>> So I think you can imagine...
> No, I can't, sorry. I don't see how dangerous will be the use of DRM panel
> in a wrong configuration. The same can very well happen on improperly working
> hardware (backlight part) or simply when somebody didn't correctly set a DT
> or manually use it when it should not be. But again I see *no* problem
> statement, only some worries.
>
> And on top of that I made tinyDRM drivers to be accessible on ACPI platforms
> and so far I have none complains about the tigers that left the cage.
>
>>>> Combine with those two factors, it is actually can conclude that the
>>>> panel-ilitek-ili9394 driver has the *implicit* dependency on 'OF'.
>>>> Removing the 'OF' dependency from its Kconfig just trigger the
>>>> leakage of such risks.
>>> What?!
>>>
>> Posting a patch is actually doing the defensive works, such a saying
>> may not sound fair for you, but this is just the hash cruel reality.
>> Sorry for saying that. :(
> So, the summary of your message is that:
> - there's no understanding how ACPI (or any other non-DT fwnode based
> environment) can utilise the driver
> - there's a worry about some problems which can't be stated clearly
> - there's a neglecting of the previous successful cases specific for DRM
> (tinyDRM drivers)
>
> As a result of the false input, the non-constructive conclusion was given.
>
> And note, I converted dozens if not hundredth of drivers that used to be
> OF-only and haven't heart any negative feedback before this case. Maybe
> we (reviewers of my patches and maintainers who applied them and end users)
> miss a BIG DEAL here? Please, elaborate how dropping OF dependency can be
> dangerous as a free walking tiger.
>
>>>> My software node related patches can help to reduce part of the potential
>>>> risks, but it still need some extra work. And it is not landed yet.
>>> Your patch has nothing to do with this series.
> I am not going to repeat the above.
>
>> With my patch applied, this is way to meet the gap under non-DT systems.
>> Users of this driver could managed to attach(complete) absent properties
>> to the SPI device with software node properties. Register the swnode
>> properties group into the system prior the panel driver is probed. There
>> may need some quirk. But at the least there has a way to go. When there
>> has a way to go, things become self-consistent. Viewed from both the
>> practice of viewpoint and the concept of viewpoint.
>>
>> And the dangerous tiger will steer its way to the direction of "ACPI
>> support is missing". But both of will be safe then.
I have no obvious opinions then, code inside this patch seems no obvious problem
for majority applications. Sorry about the noise and thanks for reply.
--
Best regards,
Sui
On 30/04/2024 11:34, Maxime Ripard wrote:
> On Tue, Apr 30, 2024 at 12:54:39AM +0800, Sui Jingfeng wrote:
>> On 2024/4/29 19:55, Maxime Ripard wrote:
>>> On Sat, Apr 27, 2024 at 01:57:46PM +0800, Sui Jingfeng wrote:
>>>> On 2024/4/26 14:23, Maxime Ripard wrote:
>>>>> On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
>>>>>> On 2024/4/26 03:10, Andy Shevchenko wrote:
>>>>>>> On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
>>>>>>>> On 2024/4/25 22:26, Andy Shevchenko wrote:
>>>>>>>>> It seems driver missed the point of proper use of device property APIs.
>>>>>>>>> Correct this by updating headers and calls respectively.
>>>>>>>> You are using the 'seems' here exactly saying that you are not 100% sure.
>>>>>>>>
>>>>>>>> Please allow me to tell you the truth: This patch again has ZERO effect.
>>>>>>>> It fix nothing. And this patch is has the risks to be wrong.
>>>>>>> Huh?! Really, stop commenting the stuff you do not understand.
>>>>>> I'm actually a professional display drivers developer at the downstream
>>>>>> in the past, despite my contribution to upstream is less. But I believe
>>>>>> that all panel driver developers know what I'm talking about. So please
>>>>>> have take a look at my replies.
>>>>> Most of the interactions you had in this series has been uncalled for.
>>>>> You might be against a patch, but there's no need to go to such length.
>>>>>
>>>>> As far as I'm concerned, this patch is fine to me in itself, and I don't
>>>>> see anything that would prevent us from merging it.
>>>> No one is preventing you, as long as don't misunderstanding what other
>>>> people's technical replies intentionally. I'm just a usual and normal
>>>> contributor, I hope the world will better than yesterday.
>>> You should seriously consider your tone when replying then.
>>>
>>>> Saying such thing to me may not proper, I guess you may want to talk
>>>> to peoples who has the push rights
>>> I think you misunderstood me. My point was that your several rants were
>>> uncalled for and aren't the kind of things we're doing here.
>>>
>>> I know very well how to get a patch merged, thanks.
>>>
>>>> just make sure it isn't a insult to the professionalism of drm bridge
>>>> community itself though.
>>> I'm not sure why you're bringing the bridge community or its
>>> professionalism. It's a panel, not a bridge, and I never doubted the
>>> professionalism of anyone.
>>
>>
>> I means that the code itself could be adopted, as newer and younger
>> programmer (like Andy) need to be encouraged to contribute.
>
> Andy has thousands of commits in Linux. He's *very* far from being a new
> contributor.
>
>> I express no obvious objections, just hints him that something else
>> probably should also be taken into consideration as well.
>
> That might be what you wanted to express, but you definitely didn't
> express it that way.
>
>> On the other hand, we probably should allow other people participate
>> in discussion so that it is sufficient discussed and ensure that it
>> won't be reverted by someone in the future for some reasons. Backing
>> to out case happens here, we may need to move things forward. Therefore,
>> it definitely deserve to have a try. It is not a big deal even though
>> it gets reverted someday.
>>
>> In the end, I don't mind if you think there is nothing that could
>> prevent you from merge it, but I still suggest you have a glance at
>> peoples siting at the Cc list. I'm busy now and I have a lot of other
>> tasks to do, and may not be able to reply you emails on time. So it up
>> to you and other maintainers to decide.
>> Thank you.
>
> So far, you're the only one who reviewed those patches. I'm not sure
> what you're talking about here.
Well I (as drm-panel maintainer) did review them positively because the patches looked
perfectly correct in regards of the commit message and the patchset motivation and
because I trust Andy being a long time contributor with a lot of expertise.
Anyway since the rant is finished I'll land the patches.
Neil
>
> Maxime
On Thu, May 02, 2024 at 09:34:17AM +0200, Neil Armstrong wrote:
> On 30/04/2024 11:34, Maxime Ripard wrote:
..
> Anyway since the rant is finished I'll land the patches.
Thank you all for the review and discussion!
--
With Best Regards,
Andy Shevchenko
Hi,
On 2024/5/2 15:34, Neil Armstrong wrote:
> On 30/04/2024 11:34, Maxime Ripard wrote:
>> On Tue, Apr 30, 2024 at 12:54:39AM +0800, Sui Jingfeng wrote:
>>> On 2024/4/29 19:55, Maxime Ripard wrote:
>>>> On Sat, Apr 27, 2024 at 01:57:46PM +0800, Sui Jingfeng wrote:
>>>>> On 2024/4/26 14:23, Maxime Ripard wrote:
>>>>>> On Fri, Apr 26, 2024 at 04:43:18AM +0800, Sui Jingfeng wrote:
>>>>>>> On 2024/4/26 03:10, Andy Shevchenko wrote:
>>>>>>>> On Fri, Apr 26, 2024 at 02:08:16AM +0800, Sui Jingfeng wrote:
>>>>>>>>> On 2024/4/25 22:26, Andy Shevchenko wrote:
>>>>>>>>>> It seems driver missed the point of proper use of device
>>>>>>>>>> property APIs.
>>>>>>>>>> Correct this by updating headers and calls respectively.
>>>>>>>>> You are using the 'seems' here exactly saying that you are not
>>>>>>>>> 100% sure.
>>>>>>>>>
>>>>>>>>> Please allow me to tell you the truth: This patch again has
>>>>>>>>> ZERO effect.
>>>>>>>>> It fix nothing. And this patch is has the risks to be wrong.
>>>>>>>> Huh?! Really, stop commenting the stuff you do not understand.
>>>>>>> I'm actually a professional display drivers developer at the
>>>>>>> downstream
>>>>>>> in the past, despite my contribution to upstream is less. But I
>>>>>>> believe
>>>>>>> that all panel driver developers know what I'm talking about. So
>>>>>>> please
>>>>>>> have take a look at my replies.
>>>>>> Most of the interactions you had in this series has been uncalled
>>>>>> for.
>>>>>> You might be against a patch, but there's no need to go to such
>>>>>> length.
>>>>>>
>>>>>> As far as I'm concerned, this patch is fine to me in itself, and
>>>>>> I don't
>>>>>> see anything that would prevent us from merging it.
>>>>> No one is preventing you, as long as don't misunderstanding what
>>>>> other
>>>>> people's technical replies intentionally. I'm just a usual and normal
>>>>> contributor, I hope the world will better than yesterday.
>>>> You should seriously consider your tone when replying then.
>>>>
>>>>> Saying such thing to me may not proper, I guess you may want to talk
>>>>> to peoples who has the push rights
>>>> I think you misunderstood me. My point was that your several rants
>>>> were
>>>> uncalled for and aren't the kind of things we're doing here.
>>>>
>>>> I know very well how to get a patch merged, thanks.
>>>>
>>>>> just make sure it isn't a insult to the professionalism of drm bridge
>>>>> community itself though.
>>>> I'm not sure why you're bringing the bridge community or its
>>>> professionalism. It's a panel, not a bridge, and I never doubted the
>>>> professionalism of anyone.
>>>
>>>
>>> I means that the code itself could be adopted, as newer and younger
>>> programmer (like Andy) need to be encouraged to contribute.
>>
>> Andy has thousands of commits in Linux. He's *very* far from being a new
>> contributor.
>>
>>> I express no obvious objections, just hints him that something else
>>> probably should also be taken into consideration as well.
>>
>> That might be what you wanted to express, but you definitely didn't
>> express it that way.
>>
>>> On the other hand, we probably should allow other people participate
>>> in discussion so that it is sufficient discussed and ensure that it
>>> won't be reverted by someone in the future for some reasons. Backing
>>> to out case happens here, we may need to move things forward.
>>> Therefore,
>>> it definitely deserve to have a try. It is not a big deal even though
>>> it gets reverted someday.
>>>
>>> In the end, I don't mind if you think there is nothing that could
>>> prevent you from merge it, but I still suggest you have a glance at
>>> peoples siting at the Cc list. I'm busy now and I have a lot of other
>>> tasks to do, and may not be able to reply you emails on time. So it up
>>> to you and other maintainers to decide.
>>> Thank you.
>>
>> So far, you're the only one who reviewed those patches. I'm not sure
>> what you're talking about here.
>
> Well I (as drm-panel maintainer) did review them positively
[...]
> because the patches looked perfectly correct in regards of the commit
> message
The point is the 'fixes' tag.
Then, can I ask what's the issue it fixes? I'm asking because I see the
submitting-patches.html [1] documentation told us that a fixes tag indicates
that the patch fixes an issue in a previous commit.
Previously, the driver only meant to be used on the DT systems, so what's issue?
[1] https://www.kernel.org/doc/html/latest/process/submitting-patches.html#reviewer-s-statement-of-oversight
I copy & paste the paragraph from link [1] for easier to read. See below:
"A Fixes: tag indicates that the patch fixes an issue in a previous commit. It
is used to make it easy to determine where a bug originated, which can help
review a bug fix. This tag also assists the stable kernel team in determining
which stable kernel versions should receive your fix. This is the preferred
method for indicating a bug fixed by the patch."
> and the patchset motivation and
OK, the motivation is good, I agree and I admit.
> because I trust Andy being a long time contributor with a lot of
> expertise.
>
Does this means that you are going to merge patches from the experts without have a glance and
reject or ignore novice's patches unconditionally?
I'm asking because I see there still have a lot of other panel drivers use of_device_get_match_data()
function to get a match, and most of them has the 'OF' guard. However, in theory, panel should be
able to use on any CPU architecture if necessary. Does the remains has the similar issue? or Do we
need to fixed them together?
$ find . -name "*.c" -type f | xargs grep "of_device_get_match_data"
/panel-ilitek-ili9882t.c: desc = of_device_get_match_data(&dsi->dev);
/panel-innolux-p079zca.c: desc = of_device_get_match_data(&dsi->dev);
/panel-simple.c: desc = of_device_get_match_data(&pdev->dev);
/panel-simple.c: desc = of_device_get_match_data(&dsi->dev);
/panel-novatek-nt39016.c: panel->panel_info =
of_device_get_match_data(dev);
/panel-novatek-nt35950.c: nt->desc = of_device_get_match_data(dev);
/panel-boe-himax8279d.c: desc = of_device_get_match_data(&dsi->dev);
/panel-sitronix-st7703.c: ctx->desc = of_device_get_match_data(dev);
/panel-sony-td4353-jdi.c: ctx->type =
(uintptr_t)of_device_get_match_data(dev);
/panel-samsung-sofef00.c: ctx->mode = of_device_get_match_data(dev);
/panel-synaptics-r63353.c: panel->pdata = (struct r63353_desc
*)of_device_get_match_data(dev);
/panel-abt-y030xx067a.c: priv->panel_info =
of_device_get_match_data(dev);
/panel-ilitek-ili9881c.c: ctx->desc =
of_device_get_match_data(&dsi->dev);
/panel-newvision-nv3052c.c: priv->panel_info =
of_device_get_match_data(dev);
/panel-mantix-mlaf057we51.c: ctx->default_mode =
of_device_get_match_data(dev);
/panel-himax-hx8394.c: ctx->desc = of_device_get_match_data(dev);
/panel-ilitek-ili9805.c: ctx->desc =
of_device_get_match_data(&dsi->dev);
/panel-boe-tv101wum-nl6.c: desc = of_device_get_match_data(&dsi->dev);
/panel-samsung-s6d7aa0.c: ctx->desc = of_device_get_match_data(dev);
/panel-novatek-nt36523.c: pinfo->desc = of_device_get_match_data(dev);
/panel-novatek-nt35510.c: nt->conf = of_device_get_match_data(dev);
/panel-newvision-nv3051d.c: ctx->panel_info =
of_device_get_match_data(dev);
/panel-khadas-ts050.c: const void *data =
of_device_get_match_data(&dsi->dev);
/panel-leadtek-ltk500hd1829.c: ctx->panel_desc =
of_device_get_match_data(dev);
/panel-truly-nt35597.c: ctx->config = of_device_get_match_data(dev);
/panel-innolux-ej030na.c: priv->panel_info =
of_device_get_match_data(dev);
/panel-magnachip-d53e6ea8966.c: db->panel_info =
of_device_get_match_data(dev);
/panel-novatek-nt36672e.c: ctx->desc = of_device_get_match_data(dev);
/panel-sitronix-st7701.c: desc = of_device_get_match_data(&dsi->dev);
/panel-dsi-cm.c: ddata->panel_data = of_device_get_match_data(dev);
/panel-novatek-nt36672a.c: desc = of_device_get_match_data(&dsi->dev);
/panel-novatek-nt35560.c: nt->conf = of_device_get_match_data(dev);
/panel-ilitek-ili9341.c: ili->conf = of_device_get_match_data(dev);
/panel-jadard-jd9365da-h3.c: desc = of_device_get_match_data(dev);
/panel-leadtek-ltk050h3146w.c: ctx->panel_desc =
of_device_get_match_data(dev);
/panel-ilitek-ili9322.c: ili->conf = of_device_get_match_data(dev);
/panel-samsung-s6e3ha2.c: ctx->desc = of_device_get_match_data(dev);
> Anyway since the rant is finished I'll land the patches.
>
It's just *comments* or *remarks*, there really no need to use the 'rant'
to insult and/or devalue other peoples, as it make no sense.
> Neil
>
>>
>> Maxime
>
--
Best regards,
Sui