drm/i915/gt: Relocate Gen7 context-specific workarounds

CACHE_MODE_1 and CACHE_MODE_0 register should be saved and restored
as part of the context, not during engine reset. Move the related
workarounds (RC_OP_FLUSH_ENABLE, PIXEL_SUBSPAN_COLLECT_OPT_DISABLE)
from rcs_engine_wa_init() to gen7_ctx_workarounds_init() for
Gen7 platforms. This ensures the WA is applied during context
initialisation.

BSPEC: 11322, 11323

Signed-off-by: Sebastian Brzezinka <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
Reviewed-by: Krzysztof Karas <[email protected]>
Signed-off-by: Andi Shyti <[email protected]>
Link: https://lore.kernel.org/r/06cf152803ab0050e09c521ac2fc3637549860b3.1754902406.git.sebastian.brzezinka@intel.com

Author: sberbz

drivers/gpu/drm/i915/gt/intel_workarounds.c

@@ -343,6 +343,17 @@ static void gen7_ctx_workarounds_init(struct intel_engine_cs *engine,
 				      struct i915_wa_list *wal)
 {
 	wa_masked_en(wal, INSTPM, INSTPM_FORCE_ORDERING);
+	/* WaDisable_RenderCache_OperationalFlush:ivb,vlv,hsw */
+	wa_masked_dis(wal, CACHE_MODE_0_GEN7, RC_OP_FLUSH_ENABLE);
+
+	/*
+	 * BSpec says this must be set, even though
+	 * WaDisable4x2SubspanOptimization:ivb,hsw
+	 * WaDisable4x2SubspanOptimization isn't listed for VLV.
+	 */
+	wa_masked_en(wal,
+		     CACHE_MODE_1,
+		     PIXEL_SUBSPAN_COLLECT_OPT_DISABLE);
 }
 
 static void gen8_ctx_workarounds_init(struct intel_engine_cs *engine,
@@ -2567,18 +2578,6 @@ rcs_engine_wa_init(struct intel_engine_cs *engine, struct i915_wa_list *wal)
 			     RING_MODE_GEN7(RENDER_RING_BASE),
 			     GFX_TLB_INVALIDATE_EXPLICIT | GFX_REPLAY_MODE);
 
-		/* WaDisable_RenderCache_OperationalFlush:ivb,vlv,hsw */
-		wa_masked_dis(wal, CACHE_MODE_0_GEN7, RC_OP_FLUSH_ENABLE);
-
-		/*
-		 * BSpec says this must be set, even though
-		 * WaDisable4x2SubspanOptimization:ivb,hsw
-		 * WaDisable4x2SubspanOptimization isn't listed for VLV.
-		 */
-		wa_masked_en(wal,
-			     CACHE_MODE_1,
-			     PIXEL_SUBSPAN_COLLECT_OPT_DISABLE);
-
 		/*
 		 * BSpec recommends 8x4 when MSAA is used,
 		 * however in practice 16x4 seems fastest.