[Intel-gfx] Missed IRQs on IVB while running intel_gpu_top

[Daniel Vetter]

Hi!

Hm, this sounds rather interesting ... Can you please try to reproduce
this on a recent 3.7-rc kernel, and if your reproduction method is
still effective, please file a bug on bugs.freedesktop.org agains DRI
-> DRM(Intel)? We've never really gotten a good handle on these missed
interrupts, only made them rather irrelevant in practive. So any
insights and new ways to hit this are good.

Thanks, Daniel

On Mon, Nov 26, 2012 at 1:48 PM, Joe Bloggsian <joebloggsian at gmail.com> wrote:
> Hi, I have an app that uses the IVB bitstream hardware (via
> libva/intel-driver), and I have noticed some reported missed bsd IRQs
> in dmesg.  After some testing I believe this only occurs when I run
> intel_gpu_top, so I'm assuming this is an issue with intel_gpu_top
> rather than the resolved bug 38862.
>
> I can reproduce the issue quickly by running intel_gpu_top at the same
> time as 8 instances of gem_dummy_reloc_loop on a loop (one instance is
> enough, but takes longer!).  I only see missed bsd/blt ring IRQs, not
> render.  I haven't seen any missed IRQs in the same test without
> intel_gpu_top.
>
> Of course a workaround is to only use intel_gpu_top while developing,
> and not on production devices since it's really a test tool, but it
> would be good to understand and it's definitely useful to know GPU
> load for real - as an aside, it's great that Intel releases such
> tools/source/documentation.
>
> Does anyone know whether this is expected behavior and/or how to fix it please?
>
> I am using a core i7 3612QE/HD4000 with a 3.5.0 kernel, libdrm 2.4.39
> (Ubuntu 12.10 x64) and intel_gpu_top/gem_dummy_reloc_loop built from
> the very recent commit 3149531 of intel-gpu-tools.
>
> Thanks,
> Joe.
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx



-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch