The Open-Source NVIDIA Driver Will End Out 2019 In Poor Shape Still For Newer GPUs

[starshipeleven]

I don't understand why anyone hasn't reversed the firmware out of the official driver yet. Of course it should require some effort from a qualified developer, but then it doesn't matter if it's signed or unsigned. It's all software and software can be reversed.

Extracting the firmware from the driver blob is still hard to do (you cannot just use a script that cuts and pastes like it was done in the past), you cannot redistribute the firmware because of legal permissions, and you cannot automate a firmware extraction process in the user system as it requires a custom hacked kernel to bypass the secure firmware load.

And you cannot "reverse-engineer" the firmware because it is signed. The hardware will only accept and load firmware signed from NVIDIA. You could make a perfectly working firmware but the GPU will not accept it. Breaking signature is breaking a secure crypto validation system, not software reverse-engineer

I suspect that nobody just cares.

Mostly correct. Businness customers use the proprietary driver, and that means there is near-zero pressure to reverse engineer it

[vitalif]

Extracting the firmware from the driver blob is still hard to do (you cannot just use a script that cuts and pastes like it was done in the past), you cannot redistribute the firmware because of legal permissions, and you cannot automate a firmware extraction process in the user system as it requires a custom hacked kernel to bypass the secure firmware load.

This was never an obstacle with drivers like broadcom etc... In fact there were scripts that successfully extracted the firmware from broadcom windows driver. The driver is able to decrypt and extract firmware itself, so an external script should also be able to. I mean it's there somewhere, someone just needs to find it

[starshipeleven]

This was never an obstacle with drivers like broadcom etc... In fact there were scripts that successfully extracted the firmware from broadcom windows driver. The driver is able to decrypt and extract firmware itself, so an external script should also be able to. I mean it's there somewhere, someone just needs to find it

This was done also with some older NVIDIA drivers/cards, but it is not possible to do anymore, the easiest method is to intercept the firmware load when it is done by the kernel, but it is not really easy. This is what noveau developers said some time ago.

In other words, NVIDIA has complicated the process on purpose, pretty much none else gives 2 shits about protecting the firmware like that, they just bundle it with the driver and that's it. Easy to extract.

[Qaridarium]

Minor correction in green.

Nvidia could just build a big-firmware then in the end for the end-user it would be the same like AMD solution with tiny-firmware instead of big one.

in the end only fully open-source without closed-firmware is the true solution for anyone who are sick of this closed-source game

[starshipeleven]

Nvidia could just build a big-firmware

No they can't, stop this bs.

Firmware runs on the device, the driver on the CPU. You can't move the driver in the device as it is doing jobs that are far more efficiently done by the CPU.

[Qaridarium]

No they can't, stop this bs.
Firmware runs on the device, the driver on the CPU. You can't move the driver in the device as it is doing jobs that are far more efficiently done by the CPU.

other drivers do this for years... like "wlan" device drivers or lan chip drivers.
just send in the raw data into the device and it will make the rest.

bigger firmware vs smaller firmware is not bullshit it is the reality.

AMD did the Tiny closed source firmware model and Nvidia could try the big one. in the end the closed source would stay closed source

and for the end-users there is nearly zero difference in installing bigger or smaller firmware. debian for example will not inklude it .

[starshipeleven]

other drivers do this for years... like "wlan" device drivers or lan chip drivers.

GPU is not wlan or lan chip.

GPU hardware processor is not efficient for all tasks it has to do.

GPU drivers run a bunch of tasks on the host CPU.

If you want a self-sufficient GPU firmware you must integrate x86 CPU in GPU.

Does not matter how many times you repeat stupid ideas. It remains stupid for GPU to move all code to a firmware. You just show everyone you don't know what you are talking about.

[Qaridarium]

GPU is not wlan or lan chip.
GPU hardware processor is not efficient for all tasks it has to do.
GPU drivers run a bunch of tasks on the host CPU.
If you want a self-sufficient GPU firmware you must integrate x86 CPU in GPU.
Does not matter how many times you repeat stupid ideas. It remains stupid for GPU to move all code to a firmware. You just show everyone you don't know what you are talking about.

well but you already gives Nvidia the answer in future (not for past products) why not put a x86 cpu in the GPU and then make a big closed-source firmware?
in the end the customer would have same problem like with AMD closed source firmware that for example Debian does not ship the firmware.
but also for customers it would make zero different if they go with small-closed-source-firmware or the big-closed-source-firmware.

in the end the secrets of the nvidia company would stay secret and they can claim nearly the same level of open-source-ness than AMD do.

and your point: "If you want a self-sufficient GPU firmware you must integrate x86 CPU in GPU." who cares ? for future products they can do it just get some x86 lizence and put a 86core on the gpu and make big-firmware with fake open-source drivers.