|
|
Subscribe / Log in / New account

Making life (even) harder for proprietary modules

Making life (even) harder for proprietary modules

Posted Aug 9, 2023 21:02 UTC (Wed) by intelfx (subscriber, #130118)
In reply to: Making life (even) harder for proprietary modules by Cyberax
Parent article: Making life (even) harder for proprietary modules

> No. The firmware has always contained a ton of functionality (power management, command processing, etc.), and the new open source driver works fine with "old" firmware.

I don't believe that's true. The new open-source driver only works with the GSP-enabled generations, starting with Turing.

> Some of "secret sauce" remains in the closed userspace part, just like with AMD. But I don't believe that they actually moved functionality that is important for the kernel driver into the firmware.

Let's discuss an example.

For instance, one of the long-standing issues with NVIDIA's GPUs was power management, specifically reclocking. On those generations where reclocking was successfully reverse-engineered and implemented in nouveau, the driver was able to change the clocks at will (and overclocking was likely possible).

Now, I did not look at the "new" open-source kernel driver code, but I bet that the reclocking decisions are made entirely by the "new" proprietary firmware, while the kernel driver merely indicates the desired power state (or maybe even less than that, maybe it's only possible to indicate the upper limit). I also bet that there is zero support for overclocking, such that it remains a Windows-exclusive feature.

Does this "new" open-source code make it possible for nouveau to implement "some" support for reclocking? Yes. Will it be even remotely as useful as the hypothetical (reverse-)engineered interface to the low-level firmware? Not at all.


to post comments

Making life (even) harder for proprietary modules

Posted Aug 9, 2023 21:35 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link] (1 responses)

> I don't believe that's true. The new open-source driver only works with the GSP-enabled generations, starting with Turing.

I mean that the OpenSource driver doesn't need a special firmware for it.

> For instance, one of the long-standing issues with NVIDIA's GPUs was power management, specifically reclocking. On those generations where reclocking was successfully reverse-engineered and implemented in nouveau, the driver was able to change the clocks at will (and overclocking was likely possible).

Noveau still used firmware for reclocking, even on older devices. See: https://www.phoronix.com/forums/forum/linux-graphics-x-or...

Making life (even) harder for proprietary modules

Posted Aug 10, 2023 2:11 UTC (Thu) by intelfx (subscriber, #130118) [Link]

> I mean that the OpenSource driver doesn't need a special firmware for it.

Then this does not address GP's concerns at all.

> Noveau still used firmware for reclocking, even on older devices

Nope, they wrote their own PMU firmware for supported devices.

Making life (even) harder for proprietary modules

Posted Aug 9, 2023 21:39 UTC (Wed) by pizza (subscriber, #46) [Link] (1 responses)

> Does this "new" open-source code make it possible for nouveau to implement "some" support for reclocking? Yes. Will it be even remotely as useful as the hypothetical (reverse-)engineered interface to the low-level firmware? Not at all.

This "new" codebase is infinitely more useful than the current status quo, which prevents _any_ reclocking because nVidia locked things down cryptographically. Furthermore, attempting to tinker with those cryptographic locks exposes oneself to massive civil (and possibly criminal) liabilities in most jurisdictions.

nVidia has gone from actively impeding development of functionally useful F/OSS GPU drivers for their hardware to actively helping. This represents a massive improvement; why do you feel the need to dump on it for not being perfect?

Making life (even) harder for proprietary modules

Posted Aug 9, 2023 21:59 UTC (Wed) by intelfx (subscriber, #130118) [Link]

> Furthermore, attempting to tinker with those cryptographic locks exposes oneself to massive civil (and possibly criminal) liabilities in most jurisdictions

Could you please point out how exactly "attempting to tinker with cryptographic locks" on the hardware I legally own exposes me to liabilities "in most jurisdictions"? I fear I might need to turn myself in and report most of my friends and relatives :-)

If this is about DMCA, then, thankfully, USA != most jurisdictions. And even then, jailbreaking is exempt.


Copyright © 2025, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds