[Intel-gfx] Need help to root cause IGD driver issue on xen.

[Jesse Barnes]

On Sat, 8 Dec 2012 19:20:26 +0800
"G.R." <firemeteor at users.sourceforge.net> wrote:

> Hi all, I performed some diff on the logs.
> From the dmsg, I find that the failing log fails to enable transcoder.
> However, it appears to detect the panel more accurately (same LG TV is used
> for both case).
> From the register dump, it appears that PFA / PFB controls are zero in the
> failing log and the TRANSA related config are disabled too.
> Could you comment on which causes the panel no signal syndrome?
> 
> Thanks,
> Timothy
> 
> The dmsg diff, with timestamp stripped out:
> 
> $ diff  pass_lin3.2.31_nomsi_hvm_2.20/dmsg_notime
> fail_lin3.2.31_hvm_nomsi_2.20/dmsg_notime
> 508a509
> > [drm:intel_enable_transcoder] *ERROR* failed to enable transcoder 0
> 649,651c650,653
> < HDMI: detected monitor  at connection type HDMI
> < HDMI: available speakers: FL/FR
> < HDMI: supports coding type LPCM: channels = 2, rates = 32000 44100 48000,
> bits = 16
> ---
> > HDMI: detected monitor LG TV at connection type HDMI
> > HDMI: available speakers: FL/FR LFE FC RL/RR RC FLC/FRC RLC/RRC FLW/FRW
> FLH/FRH TC FCH
> > HDMI: supports coding type AC-3: channels = 6, rates = 32000 44100 48000,
> max bitrate = 640000
> > HDMI: supports coding type LPCM: channels = 2, rates = 32000 44100 48000
> 96000 192000, bits = 16 20 24
> 895d896
> < [drm:intel_prepare_page_flip], preparing flip with no unpin work?
> 1158,1164d1158
> < [drm:drm_mode_addfb], [FB:44]
> < [drm:intel_update_fbc],
> < [drm:drm_mode_addfb], [FB:42]
> < [drm:intel_update_fbc],
> < [drm:drm_mode_addfb], [FB:44]
> < [drm:drm_mode_addfb], [FB:42]
> 
> And here is the difference in register values:
> 
> $ diff  pass_lin3.2.31_nomsi_hvm_2.20/intel_reg_dumper.txt
> fail_lin3.2.31_hvm_nomsi_2.20/intel_reg_dumper.txt
> 34c34
> <                       DSPASURF: 0x0084d008
> ---
> >                       DSPASURF: 0x010af008
> 59,61c59,61
> <                      PFA_CTL_2: 0x00007de4 (vscale 0.983521)
> <                      PFA_CTL_3: 0x00003ef2 (vscale initial phase 0.491760)
> <                      PFA_CTL_4: 0x00007c40 (hscale 0.970703)
> ---
> >                      PFA_CTL_2: 0x00000000 (vscale 0.000000)
> >                      PFA_CTL_3: 0x00000000 (vscale initial phase 0.000000)
> >                      PFA_CTL_4: 0x00000000 (hscale 0.000000)
> 65,67c65,67
> <                      PFB_CTL_2: 0x00007de4 (vscale 0.983521)
> <                      PFB_CTL_3: 0x00003ef2 (vscale initial phase 0.491760)
> <                      PFB_CTL_4: 0x00007c40 (hscale 0.970703)
> ---
> >                      PFB_CTL_2: 0x00000000 (vscale 0.000000)
> >                      PFB_CTL_3: 0x00000000 (vscale initial phase 0.000000)
> >                      PFB_CTL_4: 0x00000000 (hscale 0.000000)
> 75c75
> <                   PCH_DPLL_SEL: 0x00000008 (TransA DPLL enable (DPLL A),
> TransB DPLL disable (DPLL (null)))
> ---
> >                   PCH_DPLL_SEL: 0x00000000 (TransA DPLL disable (DPLL
> (null)), TransB DPLL disable (DPLL (null)))
> 128c128
> <                     TRANSACONF: 0xc0000000 (enable, active)
> ---
> >                     TRANSACONF: 0x80000000 (enable, inactive)
> 174,175c174,175
> <             RC6_RESIDENCY_TIME: 0x00395d5a
> <            RC6p_RESIDENCY_TIME: 0x2a023fc3
> ---
> >             RC6_RESIDENCY_TIME: 0x0b5edc2c
> >            RC6p_RESIDENCY_TIME: 0xbc9fc8dc
> 
> 
> 
> 
> 
> On Sat, Dec 8, 2012 at 2:01 AM, G.R. <firemeteor at users.sourceforge.net>wrote:
> 
> > Hi all,
> > I'm running IGD pass through on xen and only get partial success.
> > Basically PVHVM guest works while HVM guest (windows, linux without pvops
> > compiled in) does not.
> > I'm trying to figure out what's the missing piece in xen and need help
> > from you experts.
> > I managed to capture some logs for both passing && failing system.
> >
> > not working
> > reg dump: http://pastebin.com/7vyG0biu
> > dmsg: http://pastebin.com/svWG0LdY
> > working
> > reg dump: http://pastebin.com/ELsabQEp
> > dmsg: http://pastebin.com/CVrkh8S0
> >
> > Both are linux 3.2.31 kernel with 2.20.14 display driver, running on i7
> > 3770. drm.debug=0x6.
> > I noticed some difference on register values and some failure on dmsg logs.
> > Could you help to identify the fatal ones so that I can dig into xen
> > source?

Hm, yeah in the full HVM case we don't bring up the panel fitter and
the transcoder fails to come up, which would definitely cause trouble.

3.2.31 is pretty old though, can you reproduce this with a newer
kernel?  Tons of fixes have landed since then, and many of them aren't
suitable for 3.2.x due to their invasiveness.

And can I ask you to file a bug for this issue so it doesn't get lost?
bugs.freedesktop.org (see intellinuxgraphics.org for bug filing
instructions).

Thanks,
-- 
Jesse Barnes, Intel Open Source Technology Center