Nvidia drivers leak their frame buffers

>>52349837
copyright issues

https://code.google.com/p/chromium/issues/detail?id=477328

>Graphics Card: Dedicated AMD or NVIDIA cards only - this doesn't happen on intel graphics

It affects AMD also, so STFU faggot OP

>>52349892
Which are their own, right?

>using chrome

>>52349837
>Seriously though, why not make the drivers open source? They make more than enough money by designing the actual GPUs, and the drivers you can download for free anyway.
I'm guessing to keep windows still relevant in gaming at least.

>>52349837
AMD shill at it's best
see
>>52349910

File: amd.png (120KB, 625x600px) Image search: [Google]

120KB, 625x600px

>>52349910
Cry baby.

>>52349837
>wordpress
ahaha
oh god
ahahahahahahaha

>>52349993
Second cry baby. See >>52349994

>>52350009
>>52350037
embarrassing, isn't it? cretin

>>52349994
>>52350037
Only crybaby shill is you

Ah yes these are the excellent drivers I've heard so much about.


>Seriously though, why not make the drivers open source? They make more than enough money by designing the actual GPUs, and the drivers you can download for free anyway.
Probably for the same reasons AMD does not. AMD would like to release some of the source code but doing so is not entirely in their control, their drivers have third party IP that has been licensed for internal usage only. Releasing the code that runs on your cards could also reveal how your card works and allow competitors to borrow pieces of your design.

AMD is deprecating the proprietary kernel drivers and putting all focus on the open source kernel drivers though. They're going to make catalyst run on top of the open source drivers

>>52349910
>WontFix

LOL

>>52349837
>giving click to a blog
MODS

>>52350054
That sounds plausible, thanks.

>>52349837
>blog
>uses chrome and gmail
>porn
>no windows
>concerned about privacy on his own machine from his own eyes
confirmed underage

WHY IS THIS THREAD STILL ALIVE?

>>52350054
lol just lol. how much do you get? is it for posts or for threads?

>>52350129
on windows*

>>52349837
This isn't really a driver problem
Chrome could clear the memory after using it
And Diablo could clear the memory before drawing it
Both were too lazy to

File: Screen Shot 2016-01-10 at 19.13.51.png (51KB, 539x439px) Image search: [Google]

51KB, 539x439px

>>52350054
>AMD
>going to
highly unlikely

>>52350202
I disagree, because that would imply that *every* application should clear their memory pages after exiting. The better solution would be to handle this from the drivers, not the applications.

>>52350229
Work is already well underway. AMDGPU is already "out" technically it's a part of the kernel already the work just isn't finished on the catalyst side. In the long term you may see AMD bring all GCN GPUs under the unified driver

>>52350250
but muh static destructors and atexit()

>>52350250
Some applications share GPU memory so you can't simply clear it after the application that created them is exiting
Also these buffers are sometimes re-created every frame
Clearing the old one every time would lower performance

>chrom*

No surprises there.

>>52350296
That sounds reasonable, but then what about this: the drivers clear the memory pages after the program exits by default. Applications that need to share memory put locks in place after they exit so that the memory can still be used by others. Only when the last program has removed the lock the drivers clear the memory.

>>52350421
There's no command in OpenGL or DirectX to do this as far as I know

>>52350296
Sounds like we need two modes, one for secure programs that don't need the extra performance, and have sensitive data.
So, web browsers, document viewers, etc.

Games run in the old mode.

>using incognito
>leaving porn open in the background
Literally end yourself.

>>52349837
Because then they'd be out of a job. Also they have to maintain control as to not letting the cards perform too well by gimping them so people will buy their next line-up of GPU's.

>>52350250
It's like a driver-level GC.

>>52350852
Or maybe you should use your own private computer to fap

>>52349915

IIRC they licensed some shit from S3 many moons ago. S3 ain't lettin their properties go for free.

File: amd shill.png (729KB, 1100x1002px) Image search: [Google]

729KB, 1100x1002px

>>52350054

>>52349837
why on g u are all opensource fetishist?

>chrome from botnet
>incognito
toppest of lels, maximumest of lyls, bestest of löls.

>>52351022
Or maybe keeping applications separate from each other, like we do every where else, is a good idea.

GPU's being able to do insanely insecure things in the name of performance is crazy when the majority of time, we don't need that performance.
They should only be taking those shortcuts when they need too .

>>52350852
or just have the developers of the program be aware of what they are actually making.
if there's a reason to clear it then they clear it themselves and if there isn't then just free it.

>>52351400
free software is more sensible

>>52351383
Ironically, the 7970 beats those cards hands down today

Its not Nvidias fault. Chrome devs are shit drones.

>>52350515
That's a shame.

>>52352301
Blame game.

>>52352031
Not every developer should have to do that just so other programs can't get their screens.

It's the shame reason we don't use C for everything, as long as performance isn't important.

Make a "safe" mode that non graphically intensive programs can use.

Because Devs will make mistakes, that how we've gotten so many buffer overflow errors from c-strings.

>>52351070
All right, but you *can* publish the source code of which you do have the licenses. That's at least your own code.

>>52350202
>>52350250
>>52350296
>>52351876
>>52352701

Main memory modules do not check for this themselves, nor any CPU. It is also not about the driver nor applications, this is the operating systems failure, they are the ones responsible for memory management. This leak might go deeper than exit/open for all we know. If multiple programs need access to a shared memory there are IPC mechanisms for this, the real physical memory space must be guarded.

This is even more of a critical issue given global addressing merging main and GPU memory, specially so for computing (like opencl) and low level drivers (Vulkan).

There's a flag you can set to allocate buffers as confidential. It's used for the copy-protected video pathways. That wipes it on deallocation and won't let other processes access it - it's fairly hardcore, actually, albeit a bit buggy.

If you don't set that flag? (And Chrome incognito doesn't, because it's only being told not to store the web history, not actually take serious measures to protect your privacy like something like Tails?) Of course it doesn't wipe it. Graphics cards with proprietary drivers are for gamers, very performance-sensitive and highly-optimised. They don't do ANYTHING they don't have to do, and shortcut on everything they can, because that's what optimisation is.

They don't wipe on reboot unless told to, either. GDDR3 and GDDR5 have pretty good retention for cold-boot attacks, if you can stop the cores next to them from boiling your freezer spray off.

>>52353615
Yea, but doesn't the GPU encapsulate all this, forcing any kind of universal guard to be done in either drivers, or per program.
I guess the gpu makers could work with OSes to change this.

>>52352701
if they can't develop programs that protect sensitive information they shouldn't develop programs that handle sensitive information
i understand things can happen, but the smaller user space software should handle this rather than imposing restrictions or making potentially harmful changes to a larger piece of software that a lot of other programs depend on.
nulling memory is a waste of resources if the data is garbage anyway.
if anything the problem that this actually leaked without nobody trying to leak it is because some piece of software, be it the drivers or the game, tried to render a framebuffer they didn't know what was in.

>>52354684
The other program is buggy yes, but we take this approach with main memory, because it's a major risk not to, I don't see why gpu is that much of an exception, zeroing out ram hurts main memory performance, but we still do it, because it's pretty much universally considered worth it..

>>52354743
>>52354684
And as many times as we say, well, devs should be competent, it just doesn't happen.

A mode for more performance, where this doesn't happen, should be manually enabled if needed, rather than the default.

>>52350054
Just make a hardware chip that will comunicate with internal by simple opensource API

>incognito mode
>implying i'd let anyone near my porn

If shared and muttable memory didn't exist we wouldn't have half the problems we have today with software development.

>>52353115
that's what makes amd linux drivers bad, they have copyrighted material in their drivers that they don't own.

i forget the reason the official drivers are so broke but it relates to this.

>>52356549
AMD and Nvidia are indeed both crap in this regard.

>>52356549
No, it's more that because they keep them closed source, the drivers are a pita to intergrate.

Copyrighted material is a reason they stay closed source though.

File: pita.jpg (65KB, 820x498px) Image search: [Google]

65KB, 820x498px

>>52356785
This post made me hungry.