Lots of Changes for Intel Graphics Coming in Linux 3.9
With the Linux 3.8 merge over, the Intel Linux graphics developers are looking toward 3.9. From a weblog entry by one of them: "Let's first look at bit at the drm core changes: The headline item this time around is the reworked kernel modeset locking. Finally the kernel doesn't stall for a few frames while probing outputs in the background! ... For general robustness of our GEM implementation we've clarified the various gpu reset state transitions. This should prevent applications from crashing while a gpu reset is going on due to the kernel leaking that transitory state to userspace. Ville Syrjälä also started to fix up our handling of pageflips across gpu hangs so that compositors no longer get stuck after a reset. Unfortunately not all of his patches made it into 3.9. Somewhat related is Mika Kuoppala's work to fix bugs across the seqnqo wrap-around. And to make sure that those bugs won't pop up again he also added some testing infrastructure. "
The thing I am most looking forward to is the gen4 relocation regression finally being fixed. No more GPU hangs when under heavy I/O load (the bane of my existence for a while now). The bug report is a good read if you think hunting for a tricky bug is fun.
The merge for 3.8 isn't just over, it has already been released.
New things are always on the horizon
Currently, when I look in store I really only see one vendor.
I think they'd take over the linux graphics-card market. Maybe not much of a market now, but potentially could be big enough to justify doing it.
But I comed into your mom's mouth last nite
I was busy crashing your mom's ass last night
(While you were busy in her basement resetting your GPU)
Maybe he meant something else, but it makes me nervous when somebody mentions "drm" and "Linux kernel" in the same sentence.
Since Vista, the Windows OS deals with this by chatting with the graphics,
if the graphics doesn't reply in a reasonable amount of time the OS will reset the program.
Playing an intensive game like Battle Field 3 this tends to happen a lot. The video
card is too busy and doesn't have the time to chat; so the game just goes away.
Your either at your desktop like nothing is wrong other than not playing a game anymore,
or staring at the last graphics frame shown with a hard reset in your future.
VIDEO_TDR_FAILURE is what one error calls itself and fixed by disabling the "Timeout Detection and Recovery (TDR)."
Key added here: HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\GraphicsDrivers then disabled.
I have a write up about this in my journal but it's not worth reading, it's wordy and doesn't flow a bit. .DMP file that managed to be written before the OS
A better write up is here http://mikemstech.blogspot.com/2011/12/troubleshooting-0x116-videotdrerror.html
a site I found after running a debug on the one (single)
became unresponsive.
GPU resets:
"Other ideas for troubleshooting:"
{list not quoted}
"I wouldn't pin this problem on Microsoft. Ultimately, this crash is due to game/software developers
and graphics card manufacturers (such as ATI/AMD and NVidia) developing buggy devices and software
and not playing by the rules and standards dictated for a specific platform like Windows. There are
many cases of similar events happening on UNIX/Linux systems, so this problem is not specifically isolated to Windows."
The above is a quote from http://mikemstech.blogspot.com/2011/12/troubleshooting-0x116-videotdrerror.html
a site I have nothing to do with; just a google result that helped me out one time.
Anyways...
For years and years on end I used to study linux kernel code and hack it a few times.
Not so much, anymore, these days.
GEM's, Syrjälä's, seqnqo's --- by golly, I just hope you kids know what yer doing...
You need to stop blaming Windows and/or the GPU vendor and troubleshoot your system. My GPU has reset precisely zero times when playing BF3, over a total of about 107 hours. I have seen GPU resets on my system on rare occasion, generally with broken software, but never in BF3 and I have more than a bit of testing with it.
You have something wrong on your system, you should figure out what.
>the kernel doesn't stall for a few frames while probing outputs in the background!
Is this why my Intel based XBMC box skips frames occasionally when watching videos? I was able to fix it by downgrading to a specific kernel version that did not have the problem.
I've been unable to upgrade this machine ever since, and am still running on that ancient install and kernel for this reason. It'd be nice if it was finally fixed!
My patience on the above subject ran out a loooong time ago....
Presumably the article refers to Direct Rendering Management and not Digital Rights Management. Any clue what GEM they are talking about? Unless it's a revival of Digital Research's GEM -- the alternate windowing platform that Ventura Publisher used to use. Article authors: Please expand acronyms with the first usage.
when the EXPORT_SYMBOL_GPL nonsense gets removed
Too bad the AMD sacked the vast majority of their software people with any sort of Linux skills/background at the end of October. They could have been relevant.
... was proposed back in 1984 by Andrew Tanenbaum (do a Google scholar search for his name and "immediate files"). Glad to see it made it into Linux a mere 29 years later.