drm/i915: Add provision to extend Golden context batch
authorArun Siluvery <arun.siluvery@linux.intel.com>
Mon, 20 Jul 2015 09:46:10 +0000 (10:46 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 21 Jul 2015 07:30:57 +0000 (09:30 +0200)
The Golden batch carries 3D state at the beginning so that HW starts with
a known state. It is carried as a binary blob which is auto-generated from
source. The idea was it would be easier to maintain and keep the complexity
out of the kernel which makes sense as we don't really touch it. However if
you really need to update it then you need to update generator source and
keep the binary blob in sync with it.

There is a need to patch this in bxt to send one additional command to enable
a feature. A solution was to patch the binary data with some additional
data structures (included as part of auto-generator source) but it was
unnecessarily complicated.

Chris suggested the idea of having a secondary batch and execute two batch
buffers. It has clear advantages as we needn't touch the base golden batch,
can customize secondary/auxiliary batch depending on Gen and can be carried
in the driver with no dependencies.

This patch adds support for this auxiliary batch which is inserted at the
end of golden batch and is completely independent from it. Thanks to Mika
for the preliminary review.

v2: Strictly conform to the batch size requirements to cover Gen2 and
add comments to clarify overflow check in macro (Chris, Mika).

v3: aux_batch_offset was declared as u64, change it to u32 (Chris)

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Armin Reese <armin.c.reese@intel.com>
Signed-off-by: Arun Siluvery <arun.siluvery@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_gem_render_state.c
drivers/gpu/drm/i915/i915_gem_render_state.h
drivers/gpu/drm/i915/intel_lrc.c

index b6492fe3722b56a180ea53b1cfecba7edf9fcb46..5026a6267a88034b3cae18acc0674952a711521a 100644 (file)
@@ -73,6 +73,24 @@ free_gem:
        return ret;
 }
 
+/*
+ * Macro to add commands to auxiliary batch.
+ * This macro only checks for page overflow before inserting the commands,
+ * this is sufficient as the null state generator makes the final batch
+ * with two passes to build command and state separately. At this point
+ * the size of both are known and it compacts them by relocating the state
+ * right after the commands taking care of aligment so we should sufficient
+ * space below them for adding new commands.
+ */
+#define OUT_BATCH(batch, i, val)                               \
+       do {                                                    \
+               if (WARN_ON((i) >= PAGE_SIZE / sizeof(u32))) {  \
+                       ret = -ENOSPC;                          \
+                       goto err_out;                           \
+               }                                               \
+               (batch)[(i)++] = (val);                         \
+       } while(0)
+
 static int render_state_setup(struct render_state *so)
 {
        const struct intel_renderstate_rodata *rodata = so->rodata;
@@ -110,6 +128,21 @@ static int render_state_setup(struct render_state *so)
 
                d[i++] = s;
        }
+
+       while (i % CACHELINE_DWORDS)
+               OUT_BATCH(d, i, MI_NOOP);
+
+       so->aux_batch_offset = i * sizeof(u32);
+
+       OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
+       so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;
+
+       /*
+        * Since we are sending length, we need to strictly conform to
+        * all requirements. For Gen2 this must be a multiple of 8.
+        */
+       so->aux_batch_size = ALIGN(so->aux_batch_size, 8);
+
        kunmap(page);
 
        ret = i915_gem_object_set_to_gtt_domain(so->obj, false);
@@ -128,6 +161,8 @@ err_out:
        return ret;
 }
 
+#undef OUT_BATCH
+
 void i915_gem_render_state_fini(struct render_state *so)
 {
        i915_gem_object_ggtt_unpin(so->obj);
@@ -176,6 +211,16 @@ int i915_gem_render_state_init(struct drm_i915_gem_request *req)
        if (ret)
                goto out;
 
+       if (so.aux_batch_size > 8) {
+               ret = req->ring->dispatch_execbuffer(req,
+                                                    (so.ggtt_offset +
+                                                     so.aux_batch_offset),
+                                                    so.aux_batch_size,
+                                                    I915_DISPATCH_SECURE);
+               if (ret)
+                       goto out;
+       }
+
        i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req);
 
 out:
index 7aa73728178ace3a5e03851ace0e753d0d21a4c8..e641bb093a903bba18e02cd0cd156e6070a40a62 100644 (file)
@@ -37,6 +37,8 @@ struct render_state {
        struct drm_i915_gem_object *obj;
        u64 ggtt_offset;
        int gen;
+       u32 aux_batch_size;
+       u32 aux_batch_offset;
 };
 
 int i915_gem_render_state_init(struct drm_i915_gem_request *req);
index 9faad82c42ecd05fe599072bcaf25d49d05ae13b..99bba8ece464bd5f9844e28e70fb96851797dd75 100644 (file)
@@ -1740,6 +1740,12 @@ static int intel_lr_context_render_state_init(struct drm_i915_gem_request *req)
        if (ret)
                goto out;
 
+       ret = req->ring->emit_bb_start(req,
+                                      (so.ggtt_offset + so.aux_batch_offset),
+                                      I915_DISPATCH_SECURE);
+       if (ret)
+               goto out;
+
        i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req);
 
 out: