I have completed the bisection, and it points to: commit 2a9e851cb5c132f0cba9bcc25822f0a25af23179 Author: Jiri Slaby Date: Wed Aug 22 11:34:05 2018 +0200 drm/drivers: add support for using the arch wc mapping API (git-fixes). which corresponds to patches.suse/drm-drivers-add-support-for-using-the-arch-wc-mappin.patch, which is upstream commit 7cf321d118a825c1541b43ca45294126fd474efa. That is the same patch that Takashi reverted in comment #52. The kernel from comment #52 _does_ solve the performance regression I'm seeing in the console. However Edward says in comment #53 that it does not solve his problem. I have to admit I'm rather confused at this point. One possibility is that we in fact have 2 different bugs, one affecting the console, and one affecting X.org. But I have a hard time believing that. If anyone has a better explanation, I'm all ears. My test case is: I switch to a text console (Ctrl+Alt+F2), log in as root, and measure the time it takes to "cat" a relatively large text file. Without the patch mentioned above, it takes around 68 seconds; with the patch, it takes around 128 seconds. Edward, what tests are you doing exactly to determine which kernels are good and which are bad?