AMD Open-Source GPU Kernel Driver Above 5 Million Lines, Entire Linux Kernel At 34.8 Million

[phoronix]

Phoronix: AMD Open-Source GPU Kernel Driver Above 5 Million Lines, Entire Linux Kernel At 34.8 Million

With the in-development Linux 6.6 kernel adding support for more upcoming Radeon graphics processors, that means more auto-generated header files for the new IP blocks... I was curious to see the overall size now of the AMDGPU kernel driver along with its associated code like the AMDKFD compute driver. It's now above 5 million lines for the kernel driver portion...

https://www.phoronix.com/news/AMD-5-Million-Lines

[HD7950]

It would be great to be able to compile amdgpu by architecture, for example CONFIG_AMDGPU_VEGA=m and remove the support for the rest of generations, it would save a lot of time for those who compile the kernel.

[avis]

It would be great to be able to compile amdgpu by architecture, for example CONFIG_AMDGPU_VEGA=m and remove the support for the rest of generations, it would save a lot of time for those who compile the kernel.

It would even be greater if drivers were decoupled from the kernel and had their own release pace/schedule. This will make the kernel a ton leaner and easier to manage.

Nope, I'm dreaming, not gonna happen.

[user1]

It would even be greater if drivers were decoupled from the kernel and had their own release pace/schedule. This will make the kernel a ton leaner and easier to manage.

Nope, I'm dreaming, not gonna happen.

While it will definitely be a huge benefit for proper early driver support for new hardware (not having to wait for distros shipping the required kernel / firmware for new hardware), I would also not want Linux to have a driver situation similar to Windows, where it's a huge f**king mess. You get ancient drivers here and there because there is no proper straightforward way to update them, other than trying to hunt for them in the web. But like you said, without a stable kernel abi that's all just dreaming.

[andyprough]

Size of tarballs (xz compression):
Current Linux 6.5 from kernel.org: 132mb
Current Linux-libre 6.5 from linux-libre.fsfla.org: 124mb
Most recent (8-30-2023) Linux 4.19 from kernel.org: 102mb
Most recent (8-16-2023) Linux-libre 4.19 from linux-libre.fsfla.org: 98mb

Makes for kind of an interesting comparison, although it's probably fairly meaningless. The oldest one I have access to in the xz tarball format is a 4.4 Linux-libre kernel from February 2022, which weighed in at 81mb.

[mlau]

It would even be greater if drivers were decoupled from the kernel and had their own release pace/schedule. This will make the kernel a ton leaner and easier to manage.

Here you go: https://support.microsoft.com/ru-ru/windows

[avis]

Here you go: https://support.microsoft.com/ru-ru/windows

So funny, huh? Specially when you need to use new hardware and that's only possibly when running the latest kernel which also has criticial regressions which make it impossible to use it. So freaking funny. Here, enjoy this:

https://bugzilla.kernel.org/show_bug.cgi?id=217848

https://bugzilla.kernel.org/show_bug.cgi?id=217846

And this is such a common occurrence, yeah, you're right Windows users have got it right as their hardware just works. They are free to use whatever driver is available during the 10 or so years that a particular Windows release is supported. The luxury the Linux users have never heard about.

While it will definitely be a huge benefit for proper early driver support for new hardware (not having to wait for distros shipping the required kernel / firmware for new hardware), I would also not want Linux to have a driver situation similar to Windows, where it's a huge f**king mess. You get ancient drivers here and there because there is no proper straightforward way to update them, other than trying to hunt for them in the web. But like you said, without a stable kernel abi that's all just dreaming.

It's a mess for those who don't use Windows. The problem you're describing I last had maybe during Windows 8.1 life time.

Windows 10/11 nowadays install pretty much all the drivers automatically, including firmware updates. I have yet to come across a PC or laptop for which Windows 10/11 has failed to install drivers from WU.

It's funny to see ages old Windows myths being perpetuated again and again by people who don't use Windows.

[Myownfriend]

While it will definitely be a huge benefit for proper early driver support for new hardware (not having to wait for distros shipping the required kernel / firmware for new hardware), I would also not want Linux to have a driver situation similar to Windows, where it's a huge f**king mess. You get ancient drivers here and there because there is no proper straightforward way to update them, other than trying to hunt for them in the web. But like you said, without a stable kernel abi that's all just dreaming.

Wouldn't part of that be solved by having a Linux driver repository?

[chromer]

I wonder where is going AMD with this approach, by this way in year 2030 we will have about 15 millions line of code of auto generated headers in kernel

[oleid]

Windows 10/11 nowadays install pretty much all the drivers automatically, including firmware updates. I have yet to come across a PC or laptop for which Windows 10/11 has failed to install drivers from WU.

It's funny to see ages old Windows myths being perpetuated again and again by people who don't use Windows.

Except if there is no driver for current windows versions.