> -----Original Message----- > From: Jean Delvare [mailto:jdelvare@xxxxxxx] > Sent: Thursday, March 02, 2017 4:05 PM > To: Greg KH > Cc: stable@xxxxxxxxxxxxxxx; Cui, Flora; Zhang, Jerry; Deucher, Alexander > Subject: Re: [PATCH] Revert "drm/amdgpu: update tile table for > oland/hainan" > > Hi Greg, > > On Thu, 2 Mar 2017 20:32:56 +0100, Greg KH wrote: > > On Thu, Mar 02, 2017 at 06:21:35PM +0100, Jean Delvare wrote: > > > Revert commit f8d9422ef80c ("drm/amdgpu: update tile table for > > > oland/hainan") as it is causing ugly visual artifacts on at least > > > Oland. This is only an optimization so we can live without it. > > > > > > This fixes kernel bug #194761: > > > amdgpu driver breaks on Oland (SI) > > > https://bugzilla.kernel.org/show_bug.cgi?id=194761 > > > > > > Signed-off-by: Jean Delvare <jdelvare@xxxxxxx> > > > Fixes: f8d9422ef80c ("drm/amdgpu: update tile table for oland/hainan") > > > Cc: Flora Cui <Flora.Cui@xxxxxxx> > > > Cc: Junwei Zhang <Jerry.Zhang@xxxxxxx> > > > Cc: Alex Deucher <alexander.deucher@xxxxxxx> > > > --- > > > Note: This is for stable v4.10 branch only. v4.11 and later have a > > > different fix, but it's much larger and more intrusive so not suitable > > > for a stable branch. > > > > > > drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c | 330 ++++++++++++++------ > -------------- > > > 1 file changed, 139 insertions(+), 191 deletions(-) > > > > More intrusive than this? I'd much rather take what is in 4.11 than > > have you make a huge change like this if at all possible. > > I agree it's not small, but it is just a revert. So it gets you back to > 4.9 state, which is known good and well tested. > > If you insist on going into the other direction, Flora Cui says we need > to backport 7 commits, the combined diffstat of which looks like: > > drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c | 216 +++++++++++++------------ > --------- > drivers/gpu/drm/amd/amdgpu/si.c | 90 +++++++++++--- > drivers/gpu/drm/amd/amdgpu/si_enums.h | 4 > 3 files changed, 162 insertions(+), 148 deletions(-) > > So barely better IMHO, and I have not tested it, I don't know yet how > easy that will be to backport nor if it actually solves the problem. > > Personally I'd rather revert one bad commit than backport 7, but I'm > not the maintainer of the amdgpu driver and not responsible for the > 4.10 stable branch either, so obviously it's not my call. SI asic support in amdgpu is experimental. The radeon driver is still the preferred driver for SI asics as it is stable and supports more features. It would be nice to fix up the 4.10 branch for people that want to experiment with SI support in amdgpu, so the revert seems easier in my opinion. Alex