Michel Thierry <michel.thierry@xxxxxxxxx> writes:
> From: Ben Widawsky <benjamin.widawsky@xxxxxxxxx>
>
> This patch just breaks out the logic of context switch skip.
>
> It also adds pd load pre, and pd load post logic (for GEN8).
>
I dont think this patch just breaks out the logic but it changes
it. And the reasons remains a mystery.
Could you please add justification why the logic changes are
required?
> v2: Use new functions to replace the logic right away (Daniel)
> v3: Add missing pd load logic.
>
> Cc: Daniel Vetter <daniel@xxxxxxxx>
> Signed-off-by: Ben Widawsky <ben@xxxxxxxxxxxx>
> Signed-off-by: Michel Thierry <michel.thierry@xxxxxxxxx> (v2+)
> ---
> drivers/gpu/drm/i915/i915_gem_context.c | 48 +++++++++++++++++++++++++++++++--
> 1 file changed, 46 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
> index 70346b0..8474e2c 100644
> --- a/drivers/gpu/drm/i915/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/i915_gem_context.c
> @@ -569,6 +569,33 @@ mi_set_context(struct intel_engine_cs *ring,
> return ret;
> }
>
> +static inline bool should_skip_switch(struct intel_engine_cs *ring,
> + struct intel_context *from,
> + struct intel_context *to)
> +{
> + if (from == to && !to->remap_slice)
> + return true;
> +
> + return false;
> +}
> +
> +static bool
> +needs_pd_load_pre(struct intel_engine_cs *ring, struct intel_context *to)
> +{
> + struct drm_i915_private *dev_priv = ring->dev->dev_private;
> +
> + return ((INTEL_INFO(ring->dev)->gen < 8) ||
> + (ring != &dev_priv->ring[RCS])) && to->ppgtt;
> +}
> +
> +static bool
> +needs_pd_load_post(struct intel_engine_cs *ring, struct intel_context *to)
> +{
> + return (!to->legacy_hw_ctx.initialized ||
> + i915_gem_context_is_default(to)) &&
> + to->ppgtt && IS_GEN8(ring->dev);
> +}
> +
> static int do_switch(struct intel_engine_cs *ring,
> struct intel_context *to)
> {
> @@ -584,7 +611,7 @@ static int do_switch(struct intel_engine_cs *ring,
> BUG_ON(!i915_gem_obj_is_pinned(from->legacy_hw_ctx.rcs_state));
> }
>
> - if (from == to && !to->remap_slice)
> + if (should_skip_switch(ring, from, to))
> return 0;
>
> /* Trying to pin first makes error handling easier. */
> @@ -602,7 +629,11 @@ static int do_switch(struct intel_engine_cs *ring,
> */
> from = ring->last_context;
>
> - if (to->ppgtt) {
> + if (needs_pd_load_pre(ring, to)) {
> + /* Older GENs and non render rings still want the load first,
> + * "PP_DCLV followed by PP_DIR_BASE register through Load
> + * Register Immediate commands in Ring Buffer before submitting
> + * a context."*/
> trace_switch_mm(ring, to);
> ret = to->ppgtt->switch_mm(to->ppgtt, ring);
> if (ret)
> @@ -644,6 +675,19 @@ static int do_switch(struct intel_engine_cs *ring,
> if (ret)
> goto unpin_out;
>
> + if (needs_pd_load_post(ring, to)) {
> + ret = to->ppgtt->switch_mm(to->ppgtt, ring);
> + /* The hardware context switch is emitted, but we haven't
> + * actually changed the state - so it's probably safe to bail
> + * here. Still, let the user know something dangerous has
> + * happened.
> + */
> + if (ret) {
> + DRM_ERROR("Failed to change address space on context switch\n");
> + goto unpin_out;
> + }
> + }
> +
Can there be a context where both pd load pre and pd load post is
needed? If not, please consider adding WARN_ON which trigger is
when we emit switch_mm twice due to both being true.
Thanks,
-Mika
> for (i = 0; i < MAX_L3_SLICES; i++) {
> if (!(to->remap_slice & (1<<i)))
> continue;
> --
> 2.1.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
