| Message ID | 1461860781-4504-2-git-send-email-david.s.gordon@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Apr 28, 2016 at 05:26:21PM +0100, Dave Gordon wrote: > intel_pin_and_map_ringbuffer_obj() should have been setting the dirty > flag, but wasn't; and the LRC code in intel_lr_context_do_pin() is also > updated to use the new function. Technically it doesn't need to. Whilst there is any valid content in the ringbuffer, it is pinned for use by requests and the hadware. Unlike the context object whose content does need to persist when inactive. -Chris
On 28/04/16 17:36, Chris Wilson wrote: > On Thu, Apr 28, 2016 at 05:26:21PM +0100, Dave Gordon wrote: >> intel_pin_and_map_ringbuffer_obj() should have been setting the dirty >> flag, but wasn't; and the LRC code in intel_lr_context_do_pin() is also >> updated to use the new function. > > Technically it doesn't need to. Whilst there is any valid content in the > ringbuffer, it is pinned for use by requests and the hadware. Unlike the > context object whose content does need to persist when inactive. > -Chris What about across suspend and/or hibernate? Any chance of there being valid (new) content in a ringbuffer that gets thrown away? Though presumably we insist that all pending work is completed before allowing hibernation/suspend anyway? .Dave.
On Thu, Apr 28, 2016 at 06:20:30PM +0100, Dave Gordon wrote: > On 28/04/16 17:36, Chris Wilson wrote: > >On Thu, Apr 28, 2016 at 05:26:21PM +0100, Dave Gordon wrote: > >>intel_pin_and_map_ringbuffer_obj() should have been setting the dirty > >>flag, but wasn't; and the LRC code in intel_lr_context_do_pin() is also > >>updated to use the new function. > > > >Technically it doesn't need to. Whilst there is any valid content in the > >ringbuffer, it is pinned for use by requests and the hadware. Unlike the > >context object whose content does need to persist when inactive. > >-Chris > > What about across suspend and/or hibernate? Any chance of there > being valid (new) content in a ringbuffer that gets thrown away? > Though presumably we insist that all pending work is completed > before allowing hibernation/suspend anyway? Correct, before suspend we wait for all GPU work to be completed. We don't want the GPU writing into memory after the suspend/hibernation thinks we are ready. The pages are pinned by our activity as well. -Chris
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index 15d341e..707aec0 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -1098,17 +1098,17 @@ static int intel_lr_context_do_pin(struct intel_context *ctx, goto unpin_ctx_obj; } - lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; - ret = intel_pin_and_map_ringbuffer_obj(engine->dev, ringbuf); if (ret) goto unpin_map; - ctx->engine[engine->id].lrc_vma = i915_gem_obj_to_ggtt(ctx_obj); - intel_lr_context_descriptor_update(ctx, engine); + i915_gem_object_mark_dirty(ctx_obj); + + lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; lrc_reg_state[CTX_RING_BUFFER_START+1] = ringbuf->vma->node.start; ctx->engine[engine->id].lrc_reg_state = lrc_reg_state; - ctx_obj->dirty = true; + ctx->engine[engine->id].lrc_vma = i915_gem_obj_to_ggtt(ctx_obj); + intel_lr_context_descriptor_update(ctx, engine); /* Invalidate GuC TLB. */ if (i915.enable_guc_submission) diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c index 66f69cd..22a703e 100644 --- a/drivers/gpu/drm/i915/intel_ringbuffer.c +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c @@ -2145,6 +2145,7 @@ int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev, ringbuf->virtual_start = addr; ringbuf->vma = i915_gem_obj_to_ggtt(obj); + i915_gem_object_mark_dirty(obj); return 0; err_unpin:
intel_pin_and_map_ringbuffer_obj() should have been setting the dirty flag, but wasn't; and the LRC code in intel_lr_context_do_pin() is also updated to use the new function. Signed-off-by: Dave Gordon <david.s.gordon@intel.com> --- drivers/gpu/drm/i915/intel_lrc.c | 10 +++++----- drivers/gpu/drm/i915/intel_ringbuffer.c | 1 + 2 files changed, 6 insertions(+), 5 deletions(-)