High CPU of Windows Driver Foundation after a while in SXS mode

[FuJa0815]

• I have searched open and closed issues for duplicates
• I am submitting a bug report for existing functionality that does not work as intended
• I have read https://vigem.org/Community-Support/
• This isn't a feature request or a discussion topic

---

Bug description

I am using DsHidMini for a Dualshock 3 controller. After a while the CPU usage of Windows Driver Foundation - User-mode Driver Framework Host Process starts rising and rising. This leads to PCSX2 getting slower and slower. After reconnecting the controller the CPU usage is down again and everything runs smoothly. I have already reinstalled the drivers, did not work.

Screenshots

Machine info

CPU-Architecture: x64
Windows version: Windows 10.0.19043 Build 19043
Software/driver version(s): 2.2.282.0

Any other helpful information

DSHMC20221113.txt

I am aware that reproducing this error might be difficult. I also don't have any "proof" that DsHidMini is 100% at fault here, but reconnecting the controller fixes this issue again for a while, therefore it seems obvious to me.

[nefarius]

Are you using the XI proxy DLL?

[mrpenguinb]

I have the same issue, after a while the WDF process got to 10 or more percent CPU usage. I'm using the x64 XInput .dll from DSHidMini for PCSX2.

[englebert3rd]

I'm getting the same issue using the proxy DLL.

[nefarius]

To those affected feel free to test it with the complete overhaul of the DLL, it works with the old driver just fine, no need for any other upgrades.

[MoaMission]

I can confirm that the issue seems to be resolved. I haven't noticed a VRAM or CPU ramp-up after playing for ten minutes like before.

Thanks so much for addressing this!

[mrpenguinb]

The issue has been resolved on my end as well. Thank you!

So the DLL has been completely overhauled? Must have taken a while.

[nefarius]

Pretty much a complete rewrite, yeah. More sophisticated and thought through. You can see all the work in the finished pull requests and the commit history.

EDIT: see #361