non-free firmware in kernel modules, aggregation and unclear copyright notice.
A message that is not in the archive (sent off-list or not yet received): lLj-vC.A.92G.w4pUCB@murphy
Quoted 7 lines · Josselin Mouette
You are mixing apples and oranges. The fact that the GFDL sucks has nothing to do with the firmware issue. With the current situation of firmwares in the kernel, it is illegal to redistribute binary images of the kernel. Full stop. End of story. Bye bye. Redhat and SuSE may still be willing to distribute such binary images, but it isn't our problem.
Yes, GFDL has nothing to do with the main issue. No, it is not necessarily illegal to redistribute binary images of the kernel as they are today (see below). The first problem is that they (the complete w/firmware kernel binary images) are not DFSG-free, anyway. The second problem is that some firmware blobs don't have explicitly stated in the kernel tree which exactly are their licensing terms for redistribution -- those are, in principle, undistributable.
Quoted 5 lines
Putting the firmwares outside the kernel makes them distributable. Some distributions will want to include them, some others not. But the important point is that it makes that redistribution legal.
If putting the firmwares outside the kernel makes *them* distributable, then the binary kernel image is already distributable -- just not DFSG-free. The important fact WRT Debian, IMHO, is that putting the firmwares outside the kernel makes the kernel binary image DFSG-free.
HTH,
Massa
Quoted 8 lines · Humberto Massa
Quoted 6 lines · Josselin Mouette
You are mixing apples and oranges. The fact that the GFDL sucks has nothing to do with the firmware issue. With the current situation of firmwares in the kernel, it is illegal to redistribute binary images of the kernel. Full stop. End of story. Bye bye. Redhat and SuSE may still be willing to distribute such binary images, but it isn't our problem.
Wrong! It is perfectly legal in the United States, and I'm pretty sure in your country, to distribute or redistribute copyrighted works. Otherwise there wouldn't be any bookstores or newspaper stands.
There is nothing about firmware that is any different than any other component of a product. If the product was legally obtained and it requires firmware to run, then there are no special considerations about how one inserts the firmware into the product.
If you are a GPL-religious-zealot who believes that you are supposed to get the technical design (i.e. the software schematics) of the hardware device for free so you can copy it, then you are going to have to learn something about intellectual property.
The firmware, in most cases, are the bits generated by a design program that creates the function of the device. It's what the manufacturer paid 5-10 engineers over a period of a year or so to produce. The rest of the design is just some chips you can get off-the-shelf. Even if the manufacturer said; "Here you are.... You can have the design....". You don't have the "compilers" and other stuff necessary to turn this design into the firmware unless you planned to steal the design.
So, you either accept the firmware component, thanking the manufacturer for it, or you go cry foul someplace else. This whole firmware thing is a non-issue, blown way out of proportion by people who don't have a clue.
Sometimes a manufacturer doesn't have a separate bag-of-bits to supply competing operating systems. Instead, only one "driver" for one OS was produced by the manufacturer. Extracting those bits, from offset-N to offset-M in that driver likely constitutes fair use as long as the product wasn't stolen and the driver was distributed with the product, or was publicly available.
Quoted 14 lines
Quoted 1 line
Yes, GFDL has nothing to do with the main issue. No, it is not necessarily illegal to redistribute binary images of the kernel as they are today (see below). The first problem is that they (the complete w/firmware kernel binary images) are not DFSG-free, anyway. The second problem is that some firmware blobs don't have explicitly stated in the kernel tree which exactly are their licensing terms for redistribution -- those are, in principle, undistributable.
Quoted 5 lines
Putting the firmwares outside the kernel makes them distributable. Some distributions will want to include them, some others not. But the important point is that it makes that redistribution legal.
Quoted 7 lines
If putting the firmwares outside the kernel makes *them* distributable, then the binary kernel image is already distributable -- just not DFSG-free. The important fact WRT Debian, IMHO, is that putting the firmwares outside the kernel makes the kernel binary image DFSG-free.
HTH,
Massa
Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 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.
Quoted 11 lines · Richard B. Johnson
Quoted 6 lines
Quoted 6 lines
You are mixing apples and oranges. The fact that the GFDL sucks has nothing to do with the firmware issue. With the current situation of firmwares in the kernel, it is illegal to redistribute binary images of the kernel. Full stop. End of story. Bye bye. Redhat and SuSE may still be willing to distribute such binary images, but it isn't our problem.
Wrong! It is perfectly legal in the United States, and I'm pretty sure in your country, to distribute or redistribute copyrighted works. Otherwise there wouldn't be any bookstores or newspaper stands.
It is not legal to distribute the mix of a GPL software (the Linux kernel) and a proprietary file (the firmware). I wasn't aware of the "mere aggregation" interpretation, and I'm probably a bit late to say I disagree with it - mainly because you'd have a hard time convincing a court this is the case.
Quoted 5 lines
There is nothing about firmware that is any different than any other component of a product. If the product was legally obtained and it requires firmware to run, then there are no special considerations about how one inserts the firmware into the product.
Indeed, but that's not what I'm talking about.
Quoted 4 lines
If you are a GPL-religious-zealot who believes that you are supposed to get the technical design (i.e. the software schematics) of the hardware device for free so you can copy it, then you are going to have to learn something about intellectual property.
Maybe you should try to understand what people are saying before teaching them anything.
Quoted 9 lines
The firmware, in most cases, are the bits generated by a design program that creates the function of the device. It's what the manufacturer paid 5-10 engineers over a period of a year or so to produce. The rest of the design is just some chips you can get off-the-shelf. Even if the manufacturer said; "Here you are.... You can have the design....". You don't have the "compilers" and other stuff necessary to turn this design into the firmware unless you planned to steal the design.
Quoted 4 lines
So, you either accept the firmware component, thanking the manufacturer for it, or you go cry foul someplace else. This whole firmware thing is a non-issue, blown way out of proportion by people who don't have a clue.
You are completely missing the point. I don't care whether the firmwares should be free, or whether they could be free. The fact is they are not free, and Debian doesn't distribute non-free software in the "main" archive. The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
Signature · 4 lines
.''`. Josselin Mouette /\./\ : :' : [email protected] `. `' [email protected] `- Debian GNU/Linux -- The power of freedom
Quoted 49 lines · Josselin Mouette
Le mardi 05 avril 2005 �� 11:50 -0400, Richard B. Johnson a ��crit :
Quoted 11 lines
Quoted 6 lines
Quoted 6 lines
You are mixing apples and oranges. The fact that the GFDL sucks has nothing to do with the firmware issue. With the current situation of firmwares in the kernel, it is illegal to redistribute binary images of the kernel. Full stop. End of story. Bye bye. Redhat and SuSE may still be willing to distribute such binary images, but it isn't our problem.
Wrong! It is perfectly legal in the United States, and I'm pretty sure in your country, to distribute or redistribute copyrighted works. Otherwise there wouldn't be any bookstores or newspaper stands.
It is not legal to distribute the mix of a GPL software (the Linux kernel) and a proprietary file (the firmware). I wasn't aware of the "mere aggregation" interpretation, and I'm probably a bit late to say I disagree with it - mainly because you'd have a hard time convincing a court this is the case.
Quoted 5 lines
There is nothing about firmware that is any different than any other component of a product. If the product was legally obtained and it requires firmware to run, then there are no special considerations about how one inserts the firmware into the product.
Indeed, but that's not what I'm talking about.
Quoted 4 lines
If you are a GPL-religious-zealot who believes that you are supposed to get the technical design (i.e. the software schematics) of the hardware device for free so you can copy it, then you are going to have to learn something about intellectual property.
Maybe you should try to understand what people are saying before teaching them anything.
Quoted 9 lines
The firmware, in most cases, are the bits generated by a design program that creates the function of the device. It's what the manufacturer paid 5-10 engineers over a period of a year or so to produce. The rest of the design is just some chips you can get off-the-shelf. Even if the manufacturer said; "Here you are.... You can have the design....". You don't have the "compilers" and other stuff necessary to turn this design into the firmware unless you planned to steal the design.
Quoted 4 lines
So, you either accept the firmware component, thanking the manufacturer for it, or you go cry foul someplace else. This whole firmware thing is a non-issue, blown way out of proportion by people who don't have a clue.
Quoted 7 lines
You are completely missing the point. I don't care whether the firmwares should be free, or whether they could be free. The fact is they are not free, and Debian doesn't distribute non-free software in the "main" archive. The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
As previously explained, if I buy a screen-card I get a driver that will allow it to run under Windows. If I extract the stuff from that driver that allows me to run it under Linux, that constitutes fair use. Otherwise there are criminal issues like restraint-of-trade and similar problems for the manufacturer. That firmware is free for use on/in the device you purchased.
Quoted 6 lines
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
Throwing in a bit of truth to a pile of bullshit still leaves the bullshit. It isn't relevant to the issue whether or not upgrading firmware as a separate function from loading a module is "good" or "bad".
Quoted 6 lines
Signature · 4 lines
.''`. Josselin Mouette /\./\ : :' : [email protected] `. `' [email protected] `- Debian GNU/Linux -- The power of freedom
Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 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.
Quoted 7 lines · Richard B. Johnson
Quoted 6 lines
You are completely missing the point. I don't care whether the firmwares should be free, or whether they could be free. The fact is they are not free, and Debian doesn't distribute non-free software in the "main" archive. The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
Quoted 6 lines
As previously explained, if I buy a screen-card I get a driver that will allow it to run under Windows. If I extract the stuff from that driver that allows me to run it under Linux, that constitutes fair use. Otherwise there are criminal issues like restraint-of-trade and similar problems for the manufacturer. That firmware is free for use on/in the device you purchased.
You are mixing free beer and free speech. Of course I'm free to use it in the device I purchased, but it is nevertheless unsuitable for the Debian main archive, where there is only free software.
Quoted 11 lines
Quoted 6 lines
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
Throwing in a bit of truth to a pile of bullshit still leaves the bullshit. It isn't relevant to the issue whether or not upgrading firmware as a separate function from loading a module is "good" or "bad".
Of course it is. If the proposed solution is technically better and pleases the so-called zealots, why are you discussing it at all?
Signature · 4 lines
.''`. Josselin Mouette /\./\ : :' : [email protected] `. `' [email protected] `- Debian GNU/Linux -- The power of freedom
Quoted 3 lines · Josselin Mouette
The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
This is only true if the result is considered a "derivative work" of the gpl'd code.
The GPL states "In addition, mere aggregation of another work not based on the Program with the Program (or with a work based on the Program) on a volume of a storage or distribution medium does not bring the other work under the scope of this License."
Since the main cpu does not actually run the binary firmware, the fact that it lives in main memory with the code that the cpu *does* run is irrelevent. In this case, the Debian stance is that the kernel proper and the binary firmware are "merely aggregated" in a volume of storage ( ie. system memory).
Chris
Quoted 14 lines · Chris Friesen
Quoted 3 lines · Josselin Mouette
The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
This is only true if the result is considered a "derivative work" of the gpl'd code.
The GPL states "In addition, mere aggregation of another work not based on the Program with the Program (or with a work based on the Program) on a volume of a storage or distribution medium does not bring the other work under the scope of this License."
Quoted 5 lines
Since the main cpu does not actually run the binary firmware, the fact that it lives in main memory with the code that the cpu *does* run is irrelevent. In this case, the Debian stance is that the kernel proper and the binary firmware are "merely aggregated" in a volume of storage ( ie. system memory).
It merely depends on the definition of "aggregation". I'd say that two works that are only aggregated can be easily distinguished and separated. This is not the case for a binary kernel module, from which you cannot easily extract the firmware and code parts.
Signature · 4 lines
.''`. Josselin Mouette /\./\ : :' : [email protected] `. `' [email protected] `- Debian GNU/Linux -- The power of freedom
Quoted 21 lines · Josselin Mouette
Quoted 14 lines · Chris Friesen
Quoted 3 lines · Josselin Mouette
The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
This is only true if the result is considered a "derivative work" of the gpl'd code.
The GPL states "In addition, mere aggregation of another work not based on the Program with the Program (or with a work based on the Program) on a volume of a storage or distribution medium does not bring the other work under the scope of this License."
Quoted 5 lines
Since the main cpu does not actually run the binary firmware, the fact that it lives in main memory with the code that the cpu *does* run is irrelevent. In this case, the Debian stance is that the kernel proper and the binary firmware are "merely aggregated" in a volume of storage ( ie. system memory).
Quoted 4 lines
It merely depends on the definition of "aggregation". I'd say that two works that are only aggregated can be easily distinguished and separated. This is not the case for a binary kernel module, from which you cannot easily extract the firmware and code parts.
Josselin, please read the thread i linked to in debian-legal, and as nobody really gave reason to oppose it, i believe we have consensus that those firmware blobs constitute mere agregation, provided they are clearly identified and properly licenced, which they are not always.
Let's take this to debian-legal only if you want to further discuss it.
Friendly,
Sven Luther
Quoted 9 lines · Sven Luther
Quoted 4 lines
It merely depends on the definition of "aggregation". I'd say that two works that are only aggregated can be easily distinguished and separated. This is not the case for a binary kernel module, from which you cannot easily extract the firmware and code parts.
Josselin, please read the thread i linked to in debian-legal, and as nobody really gave reason to oppose it, i believe we have consensus that those firmware blobs constitute mere agregation, provided they are clearly identified and properly licenced, which they are not always.
The fact that nobody cared to answer you shouldn't be considered as any kind of approval for your sayings.
Signature · 4 lines
.''`. Josselin Mouette /\./\ : :' : [email protected] `. `' [email protected] `- Debian GNU/Linux -- The power of freedom
Quoted 6 lines · Josselin Mouette
Quoted 5 lines · Sven Luther
Quoted 4 lines
It merely depends on the definition of "aggregation". I'd say that two works that are only aggregated can be easily distinguished and separated. This is not the case for a binary kernel module, from which you cannot easily extract the firmware and code parts.
Quoted 7 lines
Quoted 4 lines
Josselin, please read the thread i linked to in debian-legal, and as nobody really gave reason to oppose it, i believe we have consensus that those firmware blobs constitute mere agregation, provided they are clearly identified and properly licenced, which they are not always.
The fact that nobody cared to answer you shouldn't be considered as any kind of approval for your sayings.
There were a couple of replies, but if you are going to argue this, please read the analysis i made, and reply to it. Read in particular :
http://lists.debian.org/debian-legal/2005/03/msg00288.html
Which contains a more formal analysis from Humberto Massa.
So, given that this thread together with the GPLed firmware flasher thread got a respectable amount of replies, i believe we can claim consensus, and this is something the debian-kernel team has been acting upon, and i believe even aknowledged by the release managers and ftp-masters.
If you have strong evidence that this is not the case, it would really have been nice to comment on it before the kernel team (not only me which you may dislike for past dealings or whatever) waste effort on something which is wrong in the first place, and i commend you to participate in the above thread asap, voicing your concerns (or remain silent forever thereafter :).
Friendly,
Sven Luther
Quoted 14 lines · Chris Friesen
Quoted 3 lines · Josselin Mouette
The fact is also that mixing them with a GPLed software gives an result you can't redistribute - although it seems many people disagree with that assertion now.
This is only true if the result is considered a "derivative work" of the gpl'd code.
The GPL states "In addition, mere aggregation of another work not based on the Program with the Program (or with a work based on the Program) on a volume of a storage or distribution medium does not bring the other work under the scope of this License."
Quoted 5 lines
Since the main cpu does not actually run the binary firmware, the fact that it lives in main memory with the code that the cpu *does* run is irrelevent. In this case, the Debian stance is that the kernel proper and the binary firmware are "merely aggregated" in a volume of storage ( ie. system memory).
The problem is that you can only argue it is mere agregation, if the copyright notice doesn't de-facto put said firmware blobs under the GPL, thus making them undistributable by the selfsame definition of the GPL.
Friendly,
Sven Luther
Quoted 6 lines · Josselin Mouette
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
This is highly amusing, coming from someone who does not maintain a driver with a firmware.
The current firmware infrastructure is too primitive. Compiling the firmware into the driver is much easier on the driver maintainers and users, presently.
Repeating myself,
* Most firmwares are a -collection- of images and data. The firmware infrastructure should load an -archive- of firmwares and associated data values.
* The firmware distribution infrastructure is basically non-existent. There is no standard way to make sure that a firmware separated from the driver gets to all users.
* The firmware bundling infrastructure is basically non-existent. (Arjan talked about this) There needs to be a a way to ensure that the needed firmwares are automatically added to initramfs/initrd.
* There is no chicken-and-egg problem as Arjan mentions. Once the above technical problems are resolved, its trivial to apply a firmware loading patch. I believe in hard transitions, not shipping tg3 with firmware -and- a firmware loading patch.
* Firmwares such as tg3 should be shipped with the kernel tarball.
In short, there are plenty of technical problems to resolve before this is even a reasonable request. Currently, a user upgrading to a tg3 sans firmware will simply get tg3 sans firmware.
Jeff
Quoted 10 lines
* The firmware distribution infrastructure is basically non-existent. There is no standard way to make sure that a firmware separated from the driver gets to all users.
* The firmware bundling infrastructure is basically non-existent. (Arjan talked about this) There needs to be a a way to ensure that the needed firmwares are automatically added to initramfs/initrd.
* There is no chicken-and-egg problem as Arjan mentions.
actually there is; you just perfectly described it. Until we have drivers that can use such firmware (and need it in initrds and the like) infrastructure for that is unlikely to come into existence, and until there is such infrastructure, driver authors like you are unlikely to want to transition to loading firmware. Now there is also your other point about the request_firmware() interface being too primitive, and that sure sounds valid. So to move forward two things need to happen
1) the infrastructure needs expanding to be useful for more drivers
2) the chicken and egg stalemate needs breaking. One way to do that is to have dual-mode drivers for a while (eg drivers that request_firmware () and if that fails, use the built-in firmware) so that the other outside-the-kernel infrastructure can be developed and deployed.
Quoted 19 lines · Arjan van de Ven
Quoted 10 lines
* The firmware distribution infrastructure is basically non-existent. There is no standard way to make sure that a firmware separated from the driver gets to all users.
* The firmware bundling infrastructure is basically non-existent. (Arjan talked about this) There needs to be a a way to ensure that the needed firmwares are automatically added to initramfs/initrd.
* There is no chicken-and-egg problem as Arjan mentions.
actually there is; you just perfectly described it. Until we have drivers that can use such firmware (and need it in initrds and the like) infrastructure for that is unlikely to come into existence, and until there is such infrastructure, driver authors like you are unlikely to want to transition to loading firmware. Now there is also your other point about the
request_firmware()interface being too primitive, and that sure sounds valid. So to move forward two things need to happen
Quoted 6 lines
1) the infrastructure needs expanding to be useful for more drivers
2) the chicken and egg stalemate needs breaking. One way to do that is to have dual-mode drivers for a while (eg drivers that
request_firmware() and if that fails, use the built-in firmware) so that the other outside-the-kernel infrastructure can be developed and deployed.
The tg3 firmware should be delivered with the kernel; and any infrastructure that does not continue to work seamlessly with firmware-separate-from-tg3 is a non-starter. That's why I say there's no chicken-and-egg: presumption of such implies half a solution.
Dual mode drivers are only useful for the 1-2 developers working on firmware loading.
Someone needs to make the effort to create and test a solution, rather than half measures which -do- imply a c-and-e problem.
Jeff
Quoted 11 lines
Quoted 11 lines
actually there is; you just perfectly described it. Until we have drivers that can use such firmware (and need it in initrds and the like) infrastructure for that is unlikely to come into existence, and until there is such infrastructure, driver authors like you are unlikely to want to transition to loading firmware. Now there is also your other point about the
request_firmware()interface being too primitive, and that sure sounds valid. So to move forward two things need to happen1) the infrastructure needs expanding to be useful for more drivers
Quoted 8 lines
Quoted 4 lines
2) the chicken and egg stalemate needs breaking. One way to do that is to have dual-mode drivers for a while (eg drivers that
request_firmware() and if that fails, use the built-in firmware) so that the other outside-the-kernel infrastructure can be developed and deployed.The tg3 firmware should be delivered with the kernel; and any infrastructure that does not continue to work seamlessly with firmware-separate-from-tg3 is a non-starter.
I'm not arguing that
Quoted 2 lines
That's why I say there's
no chicken-and-egg: presumption of such implies half a solution.
I think you haven't read what I wrote. The part that makes up the chicken-and-egg problem is the part where there is no infrastructure to USE such firmware in initrds, to distribute it with the kernel image etc.
Quoted 2 lines
Dual mode drivers are only useful for the 1-2 developers working on firmware loading.
and for the distro people who need to get their distro to do this sort of thing very smoothly. It also, to put it bluntly, shuts the "fanatics" up because they have their solution immediately, while others that may need more time to get their distro ready, have such more time. I don't have the illusion that any distro will do the infrastructure work if there are no drivers that are going to use it. Nor do I have the illusion that driver writers will make a hard switch until such infrastructure is out there en masse.
Note that I don't even mention anything about not shipping the firmware in the kernel tarbal. It might be an option in the end, depending on how the infrastructure developers; but that is not part of the stalemate that I see. You seem to be focused on only that thing, but to be honest, for me that is the least interesting/controversial part.
Quoted 10 lines · Jeff Garzik
Quoted 6 lines · Josselin Mouette
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
Quoted 7 lines
This is highly amusing, coming from someone who does not maintain a driver with a firmware.
The current firmware infrastructure is too primitive. Compiling the firmware into the driver is much easier on the driver maintainers and users, presently.
Repeating myself,
* Most firmwares are a -collection- of images and data. The firmware infrastructure should load an -archive- of firmwares and associated data values.
The firmware interface should only be concerned with getting the blob of data into kernel space. Once it is in kernel space the driver can parse out whatever archive format it is in. Take for example the ihex code that was posted recently. Similar code could be written to parse out a tarball, cpio archive, etc.
--
Brian Gerst
Quoted 15 lines · Brian Gerst
Quoted 13 lines · Jeff Garzik
Quoted 6 lines · Josselin Mouette
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
This is highly amusing, coming from someone who does not maintain a driver with a firmware.
Quoted 8 lines
Quoted 5 lines
The current firmware infrastructure is too primitive. Compiling the firmware into the driver is much easier on the driver maintainers and users, presently.
Repeating myself,
* Most firmwares are a -collection- of images and data. The firmware infrastructure should load an -archive- of firmwares and associated data values.
The firmware interface should only be concerned with getting the blob of data into kernel space. Once it is in kernel space the driver can parse out whatever archive format it is in. Take for example the ihex code that was posted recently. Similar code could be written to parse out a tarball, cpio archive, etc.
The archive format for firmware data collections must be standardized, and generic code for loading such collections needs to be written, not duplicated into each driver.
Obviously the driver-specific data inside the archive is, as the phrase implies, driver-specific.
Jeff
Quoted 22 lines · Jeff Garzik
Quoted 20 lines · Brian Gerst
Quoted 18 lines · Jeff Garzik
Quoted 6 lines · Josselin Mouette
Finally, you shouldn't forget that, technically speaking, using hotplug for uploading the firmware is much more flexible and elegant than including it in the kernel. Upgrading the firmware and the module should be two independent operations. People who are advocating the current situation are refusing technical improvements just because they are brought by people they find convenient to call "zealots".
This is highly amusing, coming from someone who does not maintain a driver with a firmware.
The current firmware infrastructure is too primitive. Compiling the firmware into the driver is much easier on the driver maintainers and users, presently.
Repeating myself,
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
* Most firmwares are a -collection- of images and data. The firmware infrastructure should load an -archive- of firmwares and associated data values.
The firmware interface should only be concerned with getting the blob of data into kernel space. Once it is in kernel space the driver can parse out whatever archive format it is in. Take for example the ihex code that was posted recently. Similar code could be written to parse out a tarball, cpio archive, etc.
The archive format for firmware data collections must be standardized, and generic code for loading such collections needs to be written, not duplicated into each driver.
I never said code shouldn't be standardized or shared. I just said that it is a seperate function from the process of transferring the data from userspace to kernel space. That said, initramfs already uses cpio, so that would be a good choice to use.
--
Brian Gerst
Quoted 2 lines · Jeff Garzik
* Firmwares such as tg3 should be shipped with the kernel tarball.
As in /usr/src/linux/firmware/tg3.tar? Would be a simple patch to add that one.
Jörn
Signature · 3 lines
The cost of changing business rules is much more expensive for software than for a secretaty. -- unknown
Quoted 3 lines · Jeff Garzik
* Most firmwares are a -collection- of images and data. The firmware infrastructure should load an -archive- of firmwares and associated data values.
Why don't you use multiple firmware loading calls with different names? Maybe adding a "group" name, or simply adding a '/' in the name. That way the userspace half will have the choice between using directories, tar, zip or whatever. The speedtouch driver already requests multiple files, having them together in one file is a userspace problem which the kernel can help but shouldn't try to solve.
Quoted 4 lines
* The firmware distribution infrastructure is basically non-existent. There is no standard way to make sure that a firmware separated from the driver gets to all users.
That's the price how having non-gpl compatible firmware though.
Quoted 3 lines
* The firmware bundling infrastructure is basically non-existent. (Arjan talked about this) There needs to be a a way to ensure that the needed firmwares are automatically added to initramfs/initrd.
Yes. See following too.
Quoted 4 lines
* There is no chicken-and-egg problem as Arjan mentions. Once the above technical problems are resolved, its trivial to apply a firmware loading patch. I believe in hard transitions, not shipping tg3 with firmware -and- a firmware loading patch.
An infrastructure to add a number of files to the kernel image (_not_ the initrd/initramfs) which can be found through internal kernel calls, firmware loading and probably an export in /proc would be nice. Unifying DSDT override, config.gz and early firmware could be nice.
Quoted 1 line
* Firmwares such as tg3 should be shipped with the kernel tarball.
Does it change between kernel versions? How often?
OG.
Quoted 11 lines · Richard B. Johnson
Quoted 8 lines · Humberto Massa
Quoted 6 lines · Josselin Mouette
You are mixing apples and oranges. The fact that the GFDL sucks has nothing to do with the firmware issue. With the current situation of firmwares in the kernel, it is illegal to redistribute binary images of the kernel. Full stop. End of story. Bye bye. Redhat and SuSE may still be willing to distribute such binary images, but it isn't our problem.
Quoted 4 lines
Wrong! It is perfectly legal in the United States, and I'm pretty sure in your country, to distribute or redistribute copyrighted works. Otherwise there wouldn't be any bookstores or newspaper stands.
Mmm, so you are claiming it is perfectly right to make copies of the windows installation CD, or for that matter to duplicate music CDs ?
I would be rather interested in knowing how you came to that conclusion :)
Friendly,
Sven Luther