Skip to contentSkip to conversation

patch driver-core-warn-about-duplicate-driver-names-on-the-same-bus.patch added to gregkh-2.6 tree

DISCUSSION Stas Sergeev 3 participants 14 messages Apr 29, 2008 – May 2, 2008 1 message not in the archive

lore
Share & tools
lore
#1 lore raw
In reply to a message not in the archive

Hello.

Quoted 3 lines · [email protected] wrote:

This is a note to let you know that I've just added the patch titled
Subject: driver core: warn about duplicate driver names on the same bus
to my gregkh-2.6 tree. Its filename is

Thanks.

Takashi, do you think this fixes the pcspkr/snd-pcsp conflict sufficienly well? I think the Kconfig protection is no longer needed. For the better convinience the modprobe rule can be created that will remove one driver before inserting another. But this is optional.
What do you think? Can we now remove the Kconfig protection?

Quoted 59 lines

driver-core-warn-about-duplicate-driver-names-on-the-same-bus.patch

This tree can be found at
http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/patches/

Quoted 1 line

From [email protected] Sat Apr 26 08:52:51 2008

From: Stas Sergeev <[email protected]>
Date: Sat, 26 Apr 2008 19:52:35 +0400
Subject: driver core: warn about duplicate driver names on the same bus
To: Andrew Morton <[email protected]>
Cc: Greg KH <[email protected]>, Takashi Iwai <[email protected]>
Message-ID: <[email protected]>

Currently an attempt to register multiple
drivers with the same name causes the
stack trace with some cryptic error message.
The attached patch adds the necessary check
and the clear error message.

---
 drivers/base/driver.c |   10 ++++++++++
 1 file changed, 10 insertions(+)
drivers/base/driver.c +10 -0
@@ -217,12 +217,22 @@ static void driver_remove_groups(struct 
 int driver_register(struct device_driver *drv)
 {
 	int ret;
+	struct device_driver *other;
 
 	if ((drv->bus->probe && drv->probe) ||
 	    (drv->bus->remove && drv->remove) ||
 	    (drv->bus->shutdown && drv->shutdown))
 		printk(KERN_WARNING "Driver '%s' needs updating - please use "
 			"bus_type methods\n", drv->name);
+
+	other = driver_find(drv->name, drv->bus);
+	if (other) {
+		put_driver(other);
+		printk(KERN_ERR "Error: Driver '%s' is already registered, "
+			"aborting...\n", drv->name);
+		return -EEXIST;
+	}
+
 	ret = bus_add_driver(drv);
 	if (ret)
 		return ret;

Patches currently in gregkh-2.6 which might be from [email protected] are

driver-core/driver-core-warn-about-duplicate-driver-names-on-the-same-bus.patch

#2 lore raw
Quoted 8 lines · Stas Sergeev

Hello.

Quoted 3 lines · [email protected] wrote:

This is a note to let you know that I've just added the patch titled
Subject: driver core: warn about duplicate driver names on the same bus
to my gregkh-2.6 tree. Its filename is

Thanks.

Quoted 8 lines

Takashi, do you think this fixes the pcspkr/snd-pcsp conflict sufficienly well? I think the Kconfig protection is no longer needed. For the better convinience the modprobe rule can be created that will remove one driver before inserting another. But this is optional.

No, don't rely on the driver core warning and catching code errors like this, that's not a good idea at all. I like the warning and your patch, but don't assume it's always going to be there.

thanks,

greg k-h

#3 lore raw

At Mon, 28 Apr 2008 21:58:49 -0700,

Quoted 19 lines · Greg KH
Quoted 8 lines · Stas Sergeev

Hello.

Quoted 3 lines · [email protected] wrote:

This is a note to let you know that I've just added the patch titled
Subject: driver core: warn about duplicate driver names on the same bus
to my gregkh-2.6 tree. Its filename is

Thanks.

Quoted 8 lines

Takashi, do you think this fixes the pcspkr/snd-pcsp conflict sufficienly well? I think the Kconfig protection is no longer needed. For the better convinience the modprobe rule can be created that will remove one driver before inserting another. But this is optional.

Quoted 3 lines

No, don't rely on the driver core warning and catching code errors like this, that's not a good idea at all. I like the warning and your patch, but don't assume it's always going to be there.

But multiple drivers for the same device are allowed on other buses like PCI. I think this is no reason to prohibit the multiple platform drivers for the same platform device.

Though, I think the snd-pcsp driver could be better built on the top of input pcspkr driver, or coexist with it. Then we'll have no more conflict about platform name space.

When you compare input pcspkr.c and sound pcsp_input.c, you can find that most of codes are identical. What we'd need is a hook on pcspkr.c that adds a dynamic check whether snd-pcsp (or any ohter) is running.

thanks,

Takashi

#4 lore raw
Quoted 21 lines · Takashi Iwai

At Mon, 28 Apr 2008 21:58:49 -0700,

Quoted 19 lines · Greg KH
Quoted 16 lines · Stas Sergeev

Hello.

Quoted 3 lines · [email protected] wrote:

This is a note to let you know that I've just added the patch titled
Subject: driver core: warn about duplicate driver names on the same bus
to my gregkh-2.6 tree. Its filename is

Thanks.

Takashi, do you think this fixes the pcspkr/snd-pcsp conflict sufficienly well? I think the Kconfig protection is no longer needed. For the better convinience the modprobe rule can be created that will remove one driver before inserting another. But this is optional.

Quoted 7 lines
Quoted 3 lines

No, don't rely on the driver core warning and catching code errors like this, that's not a good idea at all. I like the warning and your patch, but don't assume it's always going to be there.

But multiple drivers for the same device are allowed on other buses like PCI. I think this is no reason to prohibit the multiple platform drivers for the same platform device.

No, the driver core doesn't allow that at all right now (I'm working on fixing that though...)

What this patch did is just not allow you to have the same "name" for the driver on the same bus, which is reasonable as without the check, sysfs complains about trying to create a duplicate name in the same directory.

Different names is fine, and I'll continue to work on the ability to allow multiple drivers for the same device at the same time...

thanks,

greg k-h

#5 lore raw

At Tue, 29 Apr 2008 08:14:36 -0700,

Quoted 34 lines · Greg KH
Quoted 21 lines · Takashi Iwai

At Mon, 28 Apr 2008 21:58:49 -0700,

Quoted 19 lines · Greg KH
Quoted 16 lines · Stas Sergeev

Hello.

Quoted 3 lines · [email protected] wrote:

This is a note to let you know that I've just added the patch titled
Subject: driver core: warn about duplicate driver names on the same bus
to my gregkh-2.6 tree. Its filename is

Thanks.

Takashi, do you think this fixes the pcspkr/snd-pcsp conflict sufficienly well? I think the Kconfig protection is no longer needed. For the better convinience the modprobe rule can be created that will remove one driver before inserting another. But this is optional.

Quoted 7 lines
Quoted 3 lines

No, don't rely on the driver core warning and catching code errors like this, that's not a good idea at all. I like the warning and your patch, but don't assume it's always going to be there.

But multiple drivers for the same device are allowed on other buses like PCI. I think this is no reason to prohibit the multiple platform drivers for the same platform device.

No, the driver core doesn't allow that at all right now (I'm working on fixing that though...)

Quoted 7 lines

What this patch did is just not allow you to have the same "name" for the driver on the same bus, which is reasonable as without the check, sysfs complains about trying to create a duplicate name in the same directory.

Different names is fine, and I'll continue to work on the ability to allow multiple drivers for the same device at the same time...

Well, I wasn't clear enough -- currently we prohibit the *build* of snd-pcsp driver when input pcspkr driver is built, regardless whether module or built-in. The reason is that sysfs spews an error with a stack trace when the platform name conflicts, and this appears to be a serious error. Disabling the build is the simplest solution in such a case (when considering built-in driver).

Now, with this patch, the platform name conflict appears non-critical, and the driver should handle properly for the returned error code. Thus, there shouldn't be a big obstacle to *build* both drivers.

Takashi

#6 lore raw
Quoted 51 lines · Takashi Iwai

At Tue, 29 Apr 2008 08:14:36 -0700,

Quoted 34 lines · Greg KH
Quoted 21 lines · Takashi Iwai

At Mon, 28 Apr 2008 21:58:49 -0700,

Quoted 19 lines · Greg KH
Quoted 16 lines · Stas Sergeev

Hello.

Quoted 3 lines · [email protected] wrote:

This is a note to let you know that I've just added the patch titled
Subject: driver core: warn about duplicate driver names on the same bus
to my gregkh-2.6 tree. Its filename is

Thanks.

Takashi, do you think this fixes the pcspkr/snd-pcsp conflict sufficienly well? I think the Kconfig protection is no longer needed. For the better convinience the modprobe rule can be created that will remove one driver before inserting another. But this is optional.

Quoted 7 lines
Quoted 3 lines

No, don't rely on the driver core warning and catching code errors like this, that's not a good idea at all. I like the warning and your patch, but don't assume it's always going to be there.

But multiple drivers for the same device are allowed on other buses like PCI. I think this is no reason to prohibit the multiple platform drivers for the same platform device.

No, the driver core doesn't allow that at all right now (I'm working on fixing that though...)

Quoted 7 lines

What this patch did is just not allow you to have the same "name" for the driver on the same bus, which is reasonable as without the check, sysfs complains about trying to create a duplicate name in the same directory.

Different names is fine, and I'll continue to work on the ability to allow multiple drivers for the same device at the same time...

Well, I wasn't clear enough -- currently we prohibit the *build* of snd-pcsp driver when input pcspkr driver is built, regardless whether module or built-in. The reason is that sysfs spews an error with a stack trace when the platform name conflicts, and this appears to be a serious error. Disabling the build is the simplest solution in such a case (when considering built-in driver).

Quoted 3 lines

Now, with this patch, the platform name conflict appears non-critical, and the driver should handle properly for the returned error code. Thus, there shouldn't be a big obstacle to *build* both drivers.

Ah, ok, that makes more sense :)

thanks,

greg k-h

#7 lore raw
In reply to Takashi Iwai

Hello.

Quoted 3 lines · Takashi Iwai

Though, I think the snd-pcsp driver could be better built on the top of input pcspkr driver, or coexist with it. Then we'll have no more conflict about platform name space.

I was trying this in the past. This never worked out very well. I disliked the dependancies. Either snd-pcsp was loading pcspkr, or there had to be the global variable to prevent the concurrent access, and that hurts modularity.

Quoted 2 lines

When you compare input pcspkr.c and sound pcsp_input.c, you can find that most of codes are identical.

Yep, its a copy/paste mainly.
I wanted a complete replacement. Back then, I've been told that multiple drivers controlling the same device is never a good idea. But I won't be surprised if that have changed since. :)

Quoted 3 lines

What we'd need is a hook on
pcspkr.c that adds a dynamic check whether snd-pcsp (or any ohter) is running.

How?
And also, with snd-pcsp you have a mixer control to disable the beeps, which I find sometimes even more usefull than the pcm sound itself. :)

#8 lore raw

At Tue, 29 Apr 2008 23:28:42 +0400,

Quoted 9 lines · Stas Sergeev

Hello.

Quoted 3 lines · Takashi Iwai

Though, I think the snd-pcsp driver could be better built on the top of input pcspkr driver, or coexist with it. Then we'll have no more conflict about platform name space.

I was trying this in the past. This never worked out very well.

Why?

Quoted 5 lines

I disliked the dependancies. Either snd-pcsp was loading pcspkr, or there had to be the global variable to prevent the concurrent access, and that hurts modularity.

But you anyway enable the input pcspkr feature in your snd-pcsp code. So, basically you depend on (or build on) it.

Quoted 9 lines
Quoted 2 lines

When you compare input pcspkr.c and sound pcsp_input.c, you can find that most of codes are identical.

Yep, its a copy/paste mainly.
I wanted a complete replacement. Back then, I've been told that multiple drivers controlling the same device is never a good idea. But I won't be surprised if that have changed since. :)

Quoted 4 lines
Quoted 3 lines

What we'd need is a hook on
pcspkr.c that adds a dynamic check whether snd-pcsp (or any ohter) is running.

How?

What you need is a way to check whether input pcspkr can be usable or not. You can add a function pointer, for example.

Quoted 4 lines

And also, with snd-pcsp you have a mixer control to disable the beeps, which I find sometimes even more usefull than the pcm sound itself. :)

Yes, that seems useful.

Takashi

#9 lore raw

Hello.

Quoted 3 lines · Takashi Iwai
Quoted 2 lines

I was trying this in the past. This never worked out very well.

Why?

Mainly because I was not able to
come up with the good hooks for the pcspkr driver, and those I tried, were not applied.
There was a lengthy thread about that. Now I can't find its beginning and its end, but some is here:
http://www.ussg.iu.edu/hypermail/linux/kernel/0603.2/3096.html
I also think you were CCed, but maybe not.

Quoted 7 lines
Quoted 5 lines

I disliked the dependancies. Either snd-pcsp was loading pcspkr, or there had to be the global variable to prevent the concurrent access, and that hurts modularity.

But you anyway enable the input pcspkr feature in your snd-pcsp code. So, basically you depend on (or build on) it.

If they are separate, then "rmmod pcspkr" should disable the beeps. I don't want to fuzzy that logic up to something like
- Check if snd-pcsp is loaded
- Use alsamixer to disable beeps, if it is.
- Use rmmod pcspkr if it is not.
I think there should be always a single way for the user to disable the beeps. Now he can choose it by chosing the driver.

Quoted 6 lines
Quoted 4 lines
Quoted 3 lines

What we'd need is a hook on
pcspkr.c that adds a dynamic check whether snd-pcsp (or any ohter) is running.

How?

What you need is a way to check whether input pcspkr can be usable or not. You can add a function pointer, for example.

Could you please clarify?
- Should snd-pcsp then forcibly select pcspkr.c to compile?
- What exactly function pointer, and where to add?

Quoted 5 lines
Quoted 4 lines

And also, with snd-pcsp you have a mixer control to disable the beeps, which I find sometimes even more usefull than the pcm sound itself. :)

Yes, that seems useful.

Yes, but problematic when they are separate. I was trying to add an input event to shut up pcspkr.c, but that was rejected. Everything else will introduce the dependancy. The dependancy will block rmmod, obfuscating the logic of disabling the beeps.
Just for the record, what problems do you see with the current solution, where only one driver drives the device? That looks rather logical to me. And I also can remember the complains about pcspkr driver being in an input drivers section. Some people had problems finding it there and were asking to move it to sound menu.

#10 lore raw

At Wed, 30 Apr 2008 21:45:59 +0400,

Quoted 23 lines · Stas Sergeev

Hello.

Quoted 3 lines · Takashi Iwai
Quoted 2 lines

I was trying this in the past. This never worked out very well.

Why?

Mainly because I was not able to
come up with the good hooks for the pcspkr driver, and those I tried,
were not applied.
There was a lengthy thread about that. Now I can't find its beginning and its end, but some is here:
http://www.ussg.iu.edu/hypermail/linux/kernel/0603.2/3096.html
I also think you were CCed, but maybe not.

Quoted 5 lines
Quoted 5 lines

I disliked the dependancies. Either snd-pcsp was loading pcspkr, or there had to be the global variable to prevent the concurrent access, and that hurts modularity.

Quoted 4 lines
Quoted 2 lines

But you anyway enable the input pcspkr feature in your snd-pcsp code. So, basically you depend on (or build on) it.

If they are separate, then "rmmod pcspkr" should disable the beeps.

But, they are *not* seperate right now. snd-pcsp contains pcspkr functionality.

Quoted 9 lines

I don't want
to fuzzy that logic up to something like
- Check if snd-pcsp is loaded
- Use alsamixer to disable beeps, if it is.
- Use rmmod pcspkr if it is not.
I think there should be always a single way for the user to disable the beeps. Now he can choose it by chosing the driver.

And this won't work in most cases. People don't want to replace the existing pcspkr driver with snd-pcsp. They don't want to load the sound subsystem on their systems just because of beep.

Quoted 11 lines
Quoted 6 lines
Quoted 4 lines
Quoted 3 lines

What we'd need is a hook on
pcspkr.c that adds a dynamic check whether snd-pcsp (or any ohter) is running.

How?

What you need is a way to check whether input pcspkr can be usable or not. You can add a function pointer, for example.

Could you please clarify?
- Should snd-pcsp then forcibly select pcspkr.c to compile?
- What exactly function pointer, and where to add?

If you compare pcspkr.c and pcsp_input.c, it's found that the only essential difference is the additional check at the head of the event handler:

	if (atomic_read(&pcsp_chip.timer_active) || !pcsp_chip.pcspkr)
		return 0;

If this can be added dynamically to input pcspkr.c, no big point to have duped codes.

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines

And also, with snd-pcsp you have a mixer control to disable the beeps, which I find sometimes even more usefull than the pcm sound itself. :)

Yes, that seems useful.

Quoted 14 lines

Yes, but problematic when they are separate. I was trying to add an input event to shut up pcspkr.c, but that was rejected. Everything else will introduce the dependancy. The dependancy will block rmmod, obfuscating the logic of disabling the beeps.
Just for the record, what problems do you see with the current solution, where only one driver drives the device? That looks rather logical to me. And I also can remember the complains about pcspkr driver being in an input drivers section. Some people had problems finding it there and were asking to move it to sound menu.

Distros usually make input-pcspkr as built-in, not as module. So, snd-pcsp is practically unusable on standard kernels of major distros as is, unfortunately...

Takashi

#11 lore raw

Hello.

Quoted 3 lines · Takashi Iwai

And this won't work in most cases. People don't want to replace the existing pcspkr driver with snd-pcsp. They don't want to load the sound subsystem on their systems just because of beep.

Why should they? They may just stick with pcspkr driver if they want.

Quoted 7 lines

If you compare pcspkr.c and pcsp_input.c, it's found that the only essential difference is the additional check at the head of the event handler:

	if (atomic_read(&pcsp_chip.timer_active) || !pcsp_chip.pcspkr)
		return 0;

If this can be added dynamically to input pcspkr.c, no big point to have duped codes.

Another point is PM callbacks. Somehow snd-pcsp will have to register them with pcspkr.

Quoted 3 lines

Distros usually make input-pcspkr as built-in, not as module. So, snd-pcsp is practically unusable on standard kernels of major distros as is, unfortunately...

Oh, that's really bad, I didn't know they do. For what reason? And how then people disable the beeps?
Btw, could you please name a few? At least Fedora has it as a module.

OK, I'll see about using pcspkr.c. But it looks like the needed hooks won't be too small (call to disable the beeps and a call to register the PM callbacks) and unlikely to be accepted upstream...

#12 lore raw

At Fri, 02 May 2008 19:38:03 +0400,

Quoted 9 lines · Stas Sergeev

Hello.

Quoted 3 lines · Takashi Iwai

And this won't work in most cases. People don't want to replace the existing pcspkr driver with snd-pcsp. They don't want to load the sound subsystem on their systems just because of beep.

Why should they? They may just stick with pcspkr driver if they want.

... if it's not built-in. And you'd need extra to manage selectively loading the driver for the very same platform name.

Quoted 10 lines
Quoted 7 lines

If you compare pcspkr.c and pcsp_input.c, it's found that the only essential difference is the additional check at the head of the event handler:

	if (atomic_read(&pcsp_chip.timer_active) || !pcsp_chip.pcspkr)
		return 0;

If this can be added dynamically to input pcspkr.c, no big point to have duped codes.

Another point is PM callbacks. Somehow snd-pcsp will have to register them with pcspkr.

Hmm, isn't platform PM hook enough?
pcspkr.c just sends an event at suspend or shutdown to the event handler. So, if the event was shut up at the beginning in the handler, nothing would be needed.

Quoted 6 lines
Quoted 3 lines

Distros usually make input-pcspkr as built-in, not as module. So, snd-pcsp is practically unusable on standard kernels of major distros as is, unfortunately...

Oh, that's really bad, I didn't know they do. For what reason? And how then people disable the beeps?

Why real men need to disable beep? :)

Quoted 2 lines

Btw, could you please name a few? At least Fedora has it as a module.

Well, then I remember wrongly or an old information...

Takashi

#13 lore raw

Hello.

Quoted 7 lines · Takashi Iwai
Quoted 3 lines

Another point is PM callbacks. Somehow
snd-pcsp will have to register them with
pcspkr.

Hmm, isn't platform PM hook enough?
pcspkr.c just sends an event at suspend or shutdown to the event handler. So, if the event was shut up at the beginning in the handler, nothing would be needed.

I don't think I understand what you mean... Currently, pcspkr.c just stops the timer counter to shut up the beep. snd-pcsp needs more. snd_pcm_suspend_all() for instance. Since it doesn't register itself as a platform driver any more, he have to register itself into pcspkr.c. Or?
I mean, sending a "shut up" event to the pcspkr's handler is not what snd-pcsp needs to do.

#14 lore raw

At Fri, 02 May 2008 20:22:23 +0400,

Quoted 11 lines · Stas Sergeev

Hello.

Quoted 7 lines · Takashi Iwai
Quoted 3 lines

Another point is PM callbacks. Somehow
snd-pcsp will have to register them with
pcspkr.

Hmm, isn't platform PM hook enough?
pcspkr.c just sends an event at suspend or shutdown to the event handler. So, if the event was shut up at the beginning in the handler, nothing would be needed.

Quoted 8 lines

I don't think I understand what you mean... Currently, pcspkr.c just stops the timer counter to shut up the beep. snd-pcsp needs more. snd_pcm_suspend_all() for instance. Since it doesn't register itself as a platform driver any more, he have to register itself into pcspkr.c. Or?

If snd-pcsp can hook up input pcspkr dynamically, you can (should) register snd-pcsp as a different name, as an individual platform driver. Then the whole PM stuff would still belong to snd-pcsp.

Takashi