In RFC rev2, Matt Brost requested for a comparison of the error capture from execlist vs guc-capture. I've added that data into the following links: gem_exec_capture_errordump_ADLS_execlist : https://pastebin.com/RBwkHFNq gem_exec_capture_errordump_ADLS_gucsubmission: https://pastebin.com/8k5p3kSZ This result is obtained after an additional fix reported below. I dont think i can make them an exact match, but its close enough and gem_exec_capture-capture passes. ...alan On Tue, 2022-01-11 at 01:30 -0800, Alan Previn wrote: > Print the GuC captured error state register list (string names > and values) when gpu_coredump_state printout is invoked via > the i915 debugfs for flushing the gpu error-state that was > captured prior. > > > > diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_capture.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_capture.c > index 048b1b7b9259..04b6d25abd47 100644 > --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_capture.c > +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_capture.c > + } > + for (ee = gt->engine; ee; ee = ee->next) { > + const struct i915_vma_coredump *vma; > + > + if (ee->engine == eng && ee->gucinfo.eng_id == guc_enginst && > + ee->gucinfo.guc_id == guc_gucid && > + (ee->gucinfo.lrca & CTX_GTT_ADDRESS_MASK) == > + (guc_lrca & CTX_GTT_ADDRESS_MASK)) { > There is a bug in above code - discovered this morning after additional debug of certain subtest failures: - if (ee->engine == eng && ee->gucinfo.eng_id == guc_enginst && + if (ee->engine == eng && + guc_enginst == GUC_ID_TO_ENGINE_INSTANCE(ee->gucinfo.eng_id) && + guc_engclss == GUC_ID_TO_ENGINE_CLASS(ee->gucinfo.eng_id) && ee->gucinfo.guc_id == guc_gucid && (ee->gucinfo.lrca & CTX_GTT_ADDRESS_MASK) == (guc_lrca & CTX_GTT_ADDRESS_MASK)) { > + PRINT(&i915->drm, ebuf, "i915-Ctx-VMA-Matched:\n"); > + GCAP_PRINT_BATCH(i915, ebuf, ee, batch); > + PRINT(&i915->drm, ebuf, " engine reset count: %u\n", > + ee->reset_count); > + ctx = &ee->context; > + GCAP_PRINT_CONTEXT(i915, ebuf, ctx); > + > + for (vma = ee->vma; vma; vma = vma->next) > + intel_gpu_error_print_vma(ebuf, ee->engine, vma); > + } > + }
