Created attachment 80593 [details] Photo of banding effect Bug description: After a cold boot of my MacBook Pro 9,2 using EFI (the problem does not appear using BIOS emulation), there is a disjoint (band) between lighter and darker colors (see attached photo). The problem disappears after suspending and resuming the system. System environment: -- chipset: Intel(R) Ivybridge Mobile (GT2) -- system architecture: 64-bit -- xf86-video-intel: 2.21.9 -- xserver: 1.14.1 -- mesa: 9.1.3 -- libdrm: 2.4.45 -- kernel: 3.9.5 -- Linux distribution: Arch Linux (with [testing]) -- Machine or mobo model: MacBookPro9,2 -- Display connector: FPD-Link (Internal display) Reproducing steps: -- Boot Linux on a MacBook Pro 9,2 using any UEFI bootloader. -- Display a gradient on the screen using any method. Additional info: When the issue is present, gamma adjustments I make using xgamma only apply to the darker colors. At a gamma of 1.1, the band disappears (photo is of gamma 1.0). After a suspend and resume, gamma adjustments apply to the entire screen. The issue appears both inside and outside of X.
Created attachment 80594 [details] Xorg.0.log
Created attachment 80595 [details] intel_reg_dumper output (before suspend/resume)
Created attachment 80596 [details] intel_reg_dumper output (after suspend/resume)
Created attachment 80597 [details] dmesg
-PIPEACONF: 0xc1000050 (enabled, active, pf-pd, rotate 0, 6bpc) +PIPEACONF: 0xc0000050 (enabled, active, pf-pd, rotate 0, 6bpc) -TRANSACONF: 0xc0000400 (enable, active, progressive) +TRANSACONF: 0xc0000000 (enable, active, progressive)
Do I get a price for guessing the right bug without even checking the register dump? The below hack should help, please test: diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c index 56746dc..992e0b4 100644 --- a/drivers/gpu/drm/i915/intel_display.c +++ b/drivers/gpu/drm/i915/intel_display.c @@ -5136,7 +5136,7 @@ static void ironlake_set_pipeconf(struct drm_crtc *crtc, int pipe = intel_crtc->pipe; uint32_t val; - val = I915_READ(PIPECONF(pipe)); + val = 0; val &= ~PIPECONF_BPC_MASK; switch (intel_crtc->config.pipe_bpp) {
That bit in PIPECONF selects the 10-bit gamma ramp, we only supply an 8-bit ramp...
Daniel: You're right. The banding does not occur with that patch. xgamma works correctly both before and after suspend/resume as well.
Daniel: *poke*. I presume you are going to submit a patch to compute pipeconf from scratch and remove all the rmw sequences...
Patch merged to -next as commit 2638cf2a81964ad7d6555c88cbc8be33857356a3 Author: Daniel Vetter <daniel.vetter@ffwll.ch> Date: Thu Jun 13 00:54:59 2013 +0200 drm/i915: explicitly set up PIPECONF (and gamma table) on haswell Will get backported once 3.11-rc1 comes out to stable kernels. The patch bears some regression risk, hence the extended testing time.
Is there any news on this? Is there any currently released (stable or rc) kernel that includes this patch?
It was included in v3.11-rc1, but not tagged for immediate inclusion into stable.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.