Last time, we speculated on how the buggy control panel extension truncated a value that it had right in front of it. When we sent our analysis to the vendor, they wrote back, “Can you check the driver version numbers on these crashes?”
When we checked the driver version numbers on all the crashing systems, they were something like “build 314”, when the current driver build number is something like “build 2718”. These users are running drivers that are ridiculously old! The vendor fixed that bug ages ago, but the user hasn’t gotten the fix. What’s going on?
My theory was that these users have turned off Windows Update or are otherwise declining to upgrade their video drivers. But I learned that my theory was probably wrong.
The deal here is that these are video drivers, which are a category of drivers where computer manufacturers have a lot of control. The manufacturer certifies the drivers for use on their PCs after performing their own acceptance testing on their specific hardware configurations. (Which are probably not hardware configurations that the video card vendors themselves are aware of.)
This responsibility carries forward post-sale. The computer manufacturer remains responsible for certifying driver updates, presumably by testing them against reference PCs that they maintain in their labs. Sometimes, manufacturers get customized versions of the video cards (all the better to differentiate your product with, my dear), which is why the video card vendor “driver downloads” sites often warn you to check with your computer manufacturer before installing a driver.
In practice, computer manufacturers are diligent about certifying drivers for a year, year and a half, two years tops.¹ After that, it’s not uncommon for them to abandon that model and not bother certifying drivers for it any more. All customers with that model of PC are just stuck with whatever video drivers were current as of the time the manufacturer stopped certifying drivers.
Microsoft maintains generic drivers for many classes of hardware, but intentionally sets them as low priority so that the PC manufacturer-provided drivers take precedence. The video drivers received directly from video card manufacturers are similarly deprioritized by the video card vendors. The computer manufacturer-certified drivers take precedence, even if that certification is horribly out of date.
¹ I wouldn’t be surprised if the length of time they certify drivers is somehow correlated with the length of the computer warranty.
The bug described here is not a security vulnerability so even in the brave new CRA world would more than likely not get fixed.
It’s going to take particularly noisy senior engineers (like me) to even get the conversations started about proper security support practices, never mind doing *gasp* sustaining engineering as well.
People like Nightmare Eclipse would say “Hold my beer…”
With video drivers in particular you often have to stick to a certain version that happens to work with all the programs that you use. Automatically updating graphics drivers is one of those things… you just don’t do it.
I remember around 10 years ago AMD introduced a display driver that caused the display to periodically freeze for one second every couple seconds. Tough luck if you happened to pull in that driver, especially if it is also the last one to be "supported". We had such a laptop using the generic VGA driver for this reason. If you don't blackhole...
This was probably happening back in the days of Windows Vista (cca 2008 or so).
I remember suffering through Sony VAIO ATI drivers and ATI not wanting to allow use of their generic display drivers on mobile GPUs — you had to hack the .inf file to install them. OEM display drivers were absolutely atrocious back then.
I’m guessing (happy to be corrected if I’m wrong) that this story is quite old – for a long time now, the two main GPU vendors have been providing universal Windows display drivers. Board manufacturer customisations (GPU and memory clock speeds, etc.) are handled using data stored in the option ROM.
I’m not 100% sure about the laptop GPU space, but I think the situation is the same over there.
Yes, I am pretty sure nobody in 2026 is being incommoded due to a programmer forgetting to change GWL_WNDPROC to GWLP_WNDPROC and LONG to LONG_PTR !!!
The story is from 2020. I went back and checked, and the hit rate is now extremely low. My guess is that nearly all of the machines that had been hitting the problem (which were already pretty old machines even in 2020) have since been decommissioned so they are no longer part of the population.