Firefox 83 vs. Chrome 87 On Intel Tiger Lake + AMD Renoir Under Linux

[pmorph]

Firefox has been pretty solid here as well for the last couple of years (before that there indeed was a crashy period).
I also believe Mozilla is on the right track with Rust. Just betting on skillful programmers to avoid security issues won't cut it, there needs to be progress on all fronts.

[birdie]

Oh wow! Just wow!

You are so wrong it's literally stupid.
nVidia does in fact have the right to write any code they want and to license it in any way they want. But, they definitely do -NOT- have the write to -distribute- GPL infringing code. It is -NOT- the end users responsibility it -IS- nVidia's!

Here is just one of many of Linus' explanations...
https://groups.google.com/g/fa.linux..._P2xaGEJ?pli=1

EDIT: In that same conversation, here is Linus explaining what a derived work is. Any module that uses internal interfaces are clearly derived works, which -IS- a GPL violation. Which would be fine as long as they didn't distribute that modules to end users, but nVidia -DOES- distribute their module to end users...
https://groups.google.com/g/fa.linux...m/QiaTaMqKmcMJ

I don't need no Torvald's explanations. He's not a God, he himself has admitted on several occasions that Linux has been a total failure on the desktop where you are f-king using it right now which could indicate that maybe something is wrong about the Linux kernel license or development model with the propensity of Linux kernel developers to break APIs every other moon which doesn't really help anyone: NVIDIA/VMWare/VirtualBox/etc. users are all f-ked - and the latter two, OMG, have 100% open source Linux kernel modules.

One can say using any Linux kernel APIs (since GPL symbols are assigned seemingly at random) is derived work which means binary only Linux kernel modules may not exist (which is not very far from the truth). VMWare has already been sued for that - don't remember how and if it all ended. Great! Go sue NVIDIA into oblivion and make people using NVIDIA GPUs lose any support and ability to run Linux. I thought Linux was about inclusion (first POSIX compatibility, then projects like Wine) but its fans make it abundantly clear: play by own rules, no matter how insane they are, or f off from our puny obscure OS who most people have never heard of. And let Linux on the desktop remain a f-king niche OS - that's what you're craving for. Over and out.

[duby229]

I don't need no Torvald's explanations. He's not a God, he himself has admitted on several occasions that Linux has been a total failure on the desktop where you are f-king using it right now which could indicate that maybe something is wrong about the Linux kernel license or development model with the propensity of Linux kernel developers to break APIs every other moon which doesn't really help anyone: NVIDIA/VMWare/VirtualBox/etc. users are all f-ked - and the latter two, OMG, have 100% open source Linux kernel modules.

One can say using any Linux kernel APIs (since GPL symbols are assigned seemingly at random) is derived work which means binary only Linux kernel modules may not exist (which is not very far from the truth). VMWare has already been sued for that - don't remember how and if it all ended. Great! Go sue NVIDIA into oblivion and make people using NVIDIA GPUs lose any support and ability to run Linux. I thought Linux was about inclusion (first POSIX compatibility, then projects like Wine) but its fans make it abundantly clear: play by own rules, no matter how insane they are, or f off from our puny obscure OS who most people have never heard of. And let Linux on the desktop remain a f-king niche OS - that's what you're craving for. Over and out.

Symbols only get marked as GPL-only after somebody gets caught using them in a binary module to hide kernel modifications. It's -not- random. nVidia and VmWare/VBox use shims to interface their proprietary binaries, only the shims are open source, the binaries are not. If they need to make modifications then they need to share them upstream and not hide them inside of their binaries.

Linux is modular for a good reason, anybody can distribute a binary kernel module as long as it can stand on its own. Any modifications they need to make so that their binary can stand on its own needs to be GPL compatible so that it can be shared upstream. It isn't random, it's totally logical. nVidia can fix this situation, all they have to do is contribute what they need. They won't, but that's their fault.

[birdie]

gets caught using them in a binary module to hide kernel modifications.

You keep insisting on that without providing citations. Some random Torvald's interviews, a discussion of GPL on LWN, etc. - nothing concrete. All Linux NVIDIA drivers are publicly available. Go fetch any and show proofs that NVIDIA is modifying the Linux kernel. I dare you. I don't need random URLs or quotes from random people no matter how high they are in the ranking system of kernel development. I deal with facts - not emotions or attitudes.

[jrch2k8]

This is your system somehow.

This Ubuntu 20.04 system with a Ryzen 3900X and Vega 56 can run Firefox basically forever. I may have had a crash a few weeks ago. I don't remember. It's certainly not common. Gnome Wayland. I also run Fedora 33 on a Dell XPS15 laptop. Also Gnome Wayland. No Firefox crashes.

One thing I've been finding is people running Ryzen systems that set their RAM to high speeds and it *seems* to test well. But in real life it does not always work.

Some of these things that break the system are high load on all SMT threads. And PCIe DMA transfers. A GPU even at PCIe version 3 can suck down a ton of RAM bandwidth. It's even worse with PCIe 4.0. With a RAM overclock and even the slightest amount of sag in the SoC power the Ryzen starts to provide corrupted RAM data.

I used to have IOMMU faults from the Vega 56. When I switched from 3,600 MHz RAM using ASUS DOCP settings (that's how they convert XMP profiles to AMD) to 2,666 MHz ECC using standard JEDEC, all of my RAM and GPU problems disappeared. And yet the 3,600 MHz RAM passed Memtest perfectly for 24 hours. Probably because it was not simultaneously using 24 compute threads and doing NVMe and GPU DMA.

I have not done a lot of RAM overclocking on Intel systems so I don't know how they operate. But overclocking is overclocking. It is untested and unstable. Weird problems is why I stopped running my Intel 5960X at 4.1 GHz and now just run it at stock 3.3. The weird and strange issues were not worth it.

Yeap, all true and like you i run everything stock and my RAM are factory 2666.

But as i said someone from mozilla is mailing me and it seems my crashes are even nastier cuz the crash are not even registering.

The weird thing is whatever it is, only affects Firefox and Firefox only but well eventually we should find out what the mystery is, for now Brave does the work perfectly and finally migrated most of my data to it, so my workflow is back to normal so later on i can go harder at debugging FF.

[jrch2k8]

You keep insisting on that without providing citations. Some random Torvald's interviews, a discussion of GPL on LWN, etc. - nothing concrete. All Linux NVIDIA drivers are publicly available. Go fetch any and show proofs that NVIDIA is modifying the Linux kernel. I dare you. I don't need random URLs or quotes from random people no matter how high they are in the ranking system of kernel development. I deal with facts - not emotions or attitudes.

Well, since nVidia BLOB is closed you cannot prove wrongdoings directly without a legal settlement to audit the code so a 100% ligation valid proof is almost impossible at this point but since the kernel is open you can detect violation attempts, so if nVidia is not trying to shim GPL code this commit https://github.com/torvalds/linux/co...d5c2cbba2bb991 should not affect it but it seems it breaks OpenCL support on the BLOB at least

[duby229]

You keep insisting on that without providing citations. Some random Torvald's interviews, a discussion of GPL on LWN, etc. - nothing concrete. All Linux NVIDIA drivers are publicly available. Go fetch any and show proofs that NVIDIA is modifying the Linux kernel. I dare you. I don't need random URLs or quotes from random people no matter how high they are in the ranking system of kernel development. I deal with facts - not emotions or attitudes.

I already did many multiple times in this thread. The Linus conversation I linked to was him explaining what derived works mean. Every single time that nVidia uses internal interfaces it becomes a derived work. Which is actually fine, the GPL does actually allow that as long as they don't distribute it, but they do distribute it. A derived work that is distributed as a proprietary binary is a GPL violation. It's the entire reason why symbols they use in their distributed binaries keep getting marked as GPL-only. I even linked to a web search that showes dozens of examples where nVidia used interfaces that required the kernel devs to mark symbols as GPL-only. You say it's random, but it's -not-, it's because they got caught red handed. I've posted the proofs multiple times already.

I dare you to actually -read- what he said. You say you deal in facts, except they've already been presented to you and you have completely ignored the facts many multiple times in this thread alone.

[birdie]

I already did many multiple times in this thread. The Linus conversation I linked to was him explaining what derived works mean. Every single time that nVidia uses internal interfaces it becomes a derived work. Which is actually fine, the GPL does actually allow that as long as they don't distribute it, but they do distribute it. A derived work that is distributed as a proprietary binary is a GPL violation. It's the entire reason why symbols they use in their distributed binaries keep getting marked as GPL-only. I even linked to a web search that showes dozens of examples where nVidia used interfaces that required the kernel devs to mark symbols as GPL-only. I've posted the proofs multiple times already.

I dare you to actually -read- what he said. You say you deal in facts, except they've already been presented to you and you have completely ignored the facts many multiple times in this thread alone.

You f--king said, "NVIDIA drivers are modifying the kernel" and then proceeded to provide a random Torvalds interview to prove the statement. The f--k is wrong with you? Do you know how to argue? I will now utter some complete and utter nonsense and provide some random tangentially related interview as to prove my statement. LMAO.

I no longer want to waste my time dealing with a person full of bullcrap. You can download NVIDIA drivers in under a few minutes - go inspect, disassemble and debug them as much as you want and f--king prove: "NVIDIA drivers are modifying the kernel". I don't need no f--king interviews and random posts on the internet. Oh, God.

Well, since nVidia BLOB is closed you cannot prove wrongdoings directly without a legal settlement to audit the code so a 100% ligation valid proof is almost impossible at this point but since the kernel is open you can detect violation attempts, so if nVidia is not trying to shim GPL code this commit https://github.com/torvalds/linux/co...d5c2cbba2bb991 should not affect it but it seems it breaks OpenCL support on the BLOB at least

That's hilarious. I'm out of this insanity. Two Open Source fans are 100% sure about something with zero evidence proving it.

[jrch2k8]

That's hilarious. I'm out of this insanity. Two Open Source fans are 100% sure about something with zero evidence proving it.

Well, that patch breaks nVidia drivers when trying to shim gpl only symbols, so is at least evidence of a possible wrong doing on nVidia side. Now for an actual proof you need access to nVidia source code to prove materially that shim is doing something in violation of the license, hence the kernel have 2 choices:

1.) Actually use that evidence to bring nvidia to trial and request an code audit (i can guess that won't be pretty and will be expensive AF)

2.) keep hiding GPL_only symbols until nVidia have no other way that find an alternative course

[duby229]

You f--king said, "NVIDIA drivers are modifying the kernel" and then proceeded to provide a random Torvalds interview to prove the statement. The f--k is wrong with you? Do you know how to argue? I will now utter some complete and utter nonsense and provide some random tangentially related interview as to prove my statement. LMAO.

I no longer want to waste my time dealing with a person full of bullcrap. You can download NVIDIA drivers in under a few minutes - go inspect, disassemble and debug them as much as you want and f--king prove: "NVIDIA drivers are modifying the kernel". I don't need no f--king interviews and random posts on the internet. Oh, God.



That's hilarious. I'm out of this insanity. Two Open Source fans are 100% sure about something with zero evidence proving it.

Every single time time nVidia gets caught using internal interfaces in their proprietary binary, that -IS- proof of them modifying the kernel and hiding their modifications. They've been caught red handed dozens of times.

Just search for "nvidia gpl-only". I already linked to a duckduckgo search with that exact search term and it came up with dozens of examples. Each and every single one of those examples is -absolute- proof.

It's your own damn fault that you refuse to read the links I provided to you. It's your own damn fault that you refuse to look for your own fucking self.

Nvm, I'm out, you can go ahead and stuff your head in sand all you want.