Skip to contentSkip to conversation

A "new driver model" and EXPORT_SYMBOL_GPL question

DISCUSSION Michal Jaegermann 5 participants 9 messages Jul 3, 2005 – Jul 6, 2005

lore
Share & tools
lore
#0 lore raw

It dawned on me only now that a "new driver model" introduced in patches from GKH export symbols like that:

EXPORT_SYMBOL_GPL(class_create);
EXPORT_SYMBOL_GPL(class_destroy);

and so on. The problem is that corresponding old symbols, which are still present in 2.6.12, were exported

EXPORT_SYMBOL(class_simple_create);
EXPORT_SYMBOL(class_simple_destroy);
....

This creates a problem. There exist out-of-tree drivers which are using those symbols and, even if sources are available, are not licensed GPL for whatever reasons. No, I am not the author of any of those so I cannot do very much about re-licensing. As an effect a conversion to a "new driver model", even if simple, does not work. In particular I bumped into that with Myrinet card drivers.

Was a decision to use EXPORT_SYMBOL_GPL deliberate and if yes then what considerations dictated it, other then the patch author wrote it that way, and what drivers in question are supposed to use when this change will show up in the mainline? It looks that 2.6.13 will do this.

Thanks,
Michal

#1 lore raw
Quoted 9 lines · Michal Jaegermann

It dawned on me only now that a "new driver model" introduced in patches from GKH export symbols like that:

EXPORT_SYMBOL_GPL(class_create);
EXPORT_SYMBOL_GPL(class_destroy);

and so on. The problem is that corresponding old symbols, which are still present in 2.6.12, were exported

Quoted 6 lines

EXPORT_SYMBOL(class_simple_create);
EXPORT_SYMBOL(class_simple_destroy);
....

This creates a problem. There exist out-of-tree drivers which are using those symbols

Where? What drivers?

Quoted 5 lines

and, even if sources are available, are not
licensed GPL for whatever reasons. No, I am not the author of any of those so I cannot do very much about re-licensing. As an effect a conversion to a "new driver model", even if simple, does not work. In particular I bumped into that with Myrinet card drivers.

Then take it up with them. Users of those symbols have had many months advance notice that this was going to happen.

Quoted 5 lines

Was a decision to use EXPORT_SYMBOL_GPL deliberate and if yes then what considerations dictated it, other then the patch author wrote it that way, and what drivers in question are supposed to use when this change will show up in the mainline? It looks that 2.6.13 will do this.

Please see the archives for the answers to these questions.

thanks,

greg k-h

#2 lore raw
Quoted 9 lines · Greg KH
Quoted 5 lines · Michal Jaegermann

Was a decision to use EXPORT_SYMBOL_GPL deliberate and if yes then what considerations dictated it, other then the patch author wrote it that way, and what drivers in question are supposed to use when this change will show up in the mainline? It looks that 2.6.13 will do this.

Please see the archives for the answers to these questions.

I actually tried that before posting. Maybe I attempted to look for wrong things but, beyond conversion examples, I found some postings with a general theme "there is no point to make life easy for binary-only modules" and not much else. I am afraid that this leaves me not much wiser.

Also, at least when dealing with 2.6.13-r1, neither Documentation/feature-removal-schedule.txt, nor files in Documentation/driver-model/ directory, mention anything about a switch to EXPORT_SYMBOL_GPL for relevant symbols. The only thing I can find about the later in "feature-removal..." is a note about RCU API.

Michal

#3 lore raw
In reply to Greg KH
Quoted 1 line · Greg KH

On Sun, Jul 03, 2005 at 05:12:02PM -0600, Michal Jaegermann wrote:

[snip]

Quoted 10 lines

Then take it up with them. Users of those symbols have had many months advance notice that this was going to happen.

Quoted 5 lines

Was a decision to use EXPORT_SYMBOL_GPL deliberate and if yes then what considerations dictated it, other then the patch author wrote it that way, and what drivers in question are supposed to use when this change will show up in the mainline? It looks that 2.6.13 will do this.

Please see the archives for the answers to these questions.

The archives say:

Quoted 6 lines · Greg KH

I have been recently advised that I should not change these symbols, and so I will not.

Sorry for the noise and wasted bandwidth, will not happen again.

greg k-h

Sourced from here:
http://hulllug.principalhosting.net/archive/index.php/t-52440.html

That was the way it was as of 2.6.10-mm1 and it stayed that way through 2.6.12. When did that decision change? If it was there in the archives, I missed it in the search.

If this was a Greg-only decision, perhaps a patch reversing the change addressed to Linus would get a solid yes/no decision from the top.

From what I gather in the archives, the last time this happened it was just a leak from Greg's tree and not an official policy change. It isn't in the feature removal schedule, even though other _GPL changes are listed there.

Signature · 1 line
Zan Lynx <[email protected]>
#4 lore raw
Quoted 2 lines · Zan Lynx

Sourced from here:
http://hulllug.principalhosting.net/archive/index.php/t-52440.html

No, that is not the same topic or thread.

Quoted 3 lines

That was the way it was as of 2.6.10-mm1 and it stayed that way through 2.6.12. When did that decision change? If it was there in the archives, I missed it in the search.

The code was totally rewritten from what was in 2.6.10-mm1, it was not just a simple "license change". All in-kernel users were converted, and the known closed-source users of this code were also contacted and they have already changed their code (nvidia being one of these users.)

If you know of any closed source code, using those functions, please put them in contact with me.

Quoted 2 lines

If this was a Greg-only decision, perhaps a patch reversing the change addressed to Linus would get a solid yes/no decision from the top.

What problem is this change causing you? Do you have code that calls these old, now gone, functions?

thanks,

greg k-h

#5 lore raw
Quoted 5 lines · Greg KH
Quoted 2 lines · Zan Lynx

Sourced from here:
http://hulllug.principalhosting.net/archive/index.php/t-52440.html

No, that is not the same topic or thread.

Formally you are correct but from my POV this sounds casuistic and fit for a patent lawyer. You were "recently advised not to change these symbols" and you stated that you will not. So instead you did an end run and you removed an old interface and introduced a replacement; but this time with EXPORT_SYMBOL_GPL - which has the same effect as what you told you will not do.

Quoted 2 lines

If you know of any closed source code, using those functions, please put them in contact with me.

Well, I gave an example in my original question. Yes, I asked them to contact you. If they will do that I have no idea.

Michal

#6 lore raw
Quoted 9 lines · Michal Jaegermann
Quoted 5 lines · Greg KH
Quoted 2 lines · Zan Lynx

Sourced from here:
http://hulllug.principalhosting.net/archive/index.php/t-52440.html

No, that is not the same topic or thread.

Formally you are correct but from my POV this sounds casuistic and fit for a patent lawyer.

You must have have dealt with many patent lawyers. I have, and they operate in a league of their own on how to intrepret the english language in patent claims :)

Quoted 5 lines

You were "recently advised not to change
these symbols" and you stated that you will not. So instead you did an end run and you removed an old interface and introduced a replacement; but this time with EXPORT_SYMBOL_GPL - which has the same effect as what you told you will not do.

I was "advised" to not do so, until the parties that were complaining about the change, had time to convert their code. They then did so, and so, a number of MONTHS later, I did a rewrite of the code, which caused me to create new functions and delete old, non-used functions. Those new functions were marked with the EXPORT_SYMBOL_GPL mark, as ALL of the driver core functions are. This change had lived in the -mm tree for a number of months now, with no objections.

Quoted 5 lines
Quoted 2 lines

If you know of any closed source code, using those functions, please put them in contact with me.

Well, I gave an example in my original question. Yes, I asked them to contact you. If they will do that I have no idea.

Thanks for doing so.

greg k-h

#7 lore raw
In reply to Michal Jaegermann
Quoted 26 lines · Michal Jaegermann

It dawned on me only now that a "new driver model" introduced in patches from GKH export symbols like that:

EXPORT_SYMBOL_GPL(class_create);
EXPORT_SYMBOL_GPL(class_destroy);

and so on. The problem is that corresponding old symbols, which are still present in 2.6.12, were exported

EXPORT_SYMBOL(class_simple_create);
EXPORT_SYMBOL(class_simple_destroy);
....

This creates a problem. There exist out-of-tree drivers which are using those symbols and, even if sources are available, are not licensed GPL for whatever reasons. No, I am not the author of any of those so I cannot do very much about re-licensing. As an effect a conversion to a "new driver model", even if simple, does not work. In particular I bumped into that with Myrinet card drivers.

Was a decision to use EXPORT_SYMBOL_GPL deliberate and if yes then what considerations dictated it, other then the patch author wrote it that way, and what drivers in question are supposed to use when this change will show up in the mainline? It looks that 2.6.13 will do this.

Quoted 3 lines

Thanks,
Michal
-

This problem will continue. Eventually there will be no general exported symbols. The apparent idea is to prevent the use of the kernel in proprietary systems.

Not to worry. The tools provided with a typical Linux distribution are capable of resolving those symbols. You can make a script that greps System.map for the correct offsets of those symbols. You can use those offsets in a linker script.

Cheers,
Dick Johnson
Penguin : Linux version 2.6.12 on an i686 machine (5537.79 BogoMips).
Notice : All mail here is now cached for review by Dictator Bush.
98.36% of all statistics are fiction.

#8 lore raw
Quoted 3 lines · Richard B. Johnson

This problem will continue. Eventually there will be no general exported symbols. The apparent idea is to prevent the use of the kernel in proprietary systems.

... with proprietary kernel extensions. There's a difference.

Quoted 4 lines

Not to worry. The tools provided with a typical Linux distribution are capable of resolving those symbols. You can make a script that greps System.map for the correct offsets of those symbols. You can use those offsets in a linker script.

If it wasn't you, I'd be assuming you'd be joking when suggesting to subvert the clear wishes of and licensing granted by the copyright holders and authors.

Sincerely,
Lars Marowsky-Brée <[email protected]>

Signature · 4 lines
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"