Radeon R9 390X regression - No output (CVT) or corruption (CVT-R) at 4096x2160@60Hz over DisplayPort

Submitted by fre..@..ne.com

Assigned to Default DRI bug account

Description

I had thought I had filed this bug a long time ago on here, but I can't find it...

Graphics card: Radeon R9 390X (Saphhire 11241-04-20G)
Display: LG 31MU97 (via DisplayPort)

With Linux 4.4.x, 4096 x 2160 @ 60 Hz progressive output was working with standard CVT timing.

With Linux 4.5.x onwards, standard CVT timing at that resolution fails (no signal) with the following dmesg output:

kernel: [drm:radeon_dp_link_train [radeon]] ERROR channel eq failed
kernel: [drm:radeon_dp_link_train [radeon]] ERROR channel eq failed: 5 tries

CVT-R timing results in output, but with a flickering/corrupted full-height band down the right hand side (it approximates the intended output, but pixels seem be to be corrupt in single-pixel horizontal bands and/or flicker in horizontal single-pixel bands) It could be complete coincidence, but the width of the corrupt band seems to be the difference in width vs. 3840x2160, i.e. aobut 256 pixels. With this timing there are no messages in dmesg.

There's a downstream Fedora bug report here: https://bugzilla.redhat.com/show_bug.cgi?id=1353341