diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 2957404..c40c1a1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -912,7 +912,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 
 			/* each GFX command submit allows 0 or 1 IB preemptible for CE & DE */
 			if (ce_preempt > 1 || de_preempt > 1)
-				BUG();
+				return -EINVAL;
 		}
 
 		r = amdgpu_cs_get_ring(adev, chunk_ib->ip_type,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index c485708..6b8bb1b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -161,9 +161,6 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		return r;
 	}
 
-	if (ring->funcs->init_cond_exec)
-		patch_offset = amdgpu_ring_init_cond_exec(ring);
-
 	if (vm) {
 		r = amdgpu_vm_flush(ring, job);
 		if (r) {
@@ -172,7 +169,10 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		}
 	}
 
-	if (ring->funcs->emit_hdp_flush
+	if (ring->funcs->init_cond_exec)
+		patch_offset = amdgpu_ring_init_cond_exec(ring);
+
+		if (ring->funcs->emit_hdp_flush
 #ifdef CONFIG_X86_64
 	    && !(adev->flags & AMD_IS_APU)
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 10e8232..72bef22 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -577,42 +577,59 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job)
 		id->oa_size != job->oa_size);
 	int r;
 
-	if (ring->funcs->emit_pipeline_sync && (
-	    job->vm_needs_flush || gds_switch_needed ||
-	    amdgpu_vm_ring_has_compute_vm_bug(ring)))
-		amdgpu_ring_emit_pipeline_sync(ring);
+	if (job->vm_needs_flush || gds_switch_needed ||
+		amdgpu_vm_is_gpu_reset(adev, id) ||
+		amdgpu_vm_ring_has_compute_vm_bug(ring)) {
+		unsigned patch_offset = 0;
 
-	if (ring->funcs->emit_vm_flush && (job->vm_needs_flush ||
-	    amdgpu_vm_is_gpu_reset(adev, id))) {
-		struct dma_fence *fence;
-		u64 pd_addr = amdgpu_vm_adjust_mc_addr(adev, job->vm_pd_addr);
+		if (ring->funcs->init_cond_exec)
+			patch_offset = amdgpu_ring_init_cond_exec(ring);
 
-		trace_amdgpu_vm_flush(pd_addr, ring->idx, job->vm_id);
-		amdgpu_ring_emit_vm_flush(ring, job->vm_id, pd_addr);
+		if (ring->funcs->emit_pipeline_sync &&
+			(job->vm_needs_flush || gds_switch_needed ||
+			amdgpu_vm_ring_has_compute_vm_bug(ring)))
+			amdgpu_ring_emit_pipeline_sync(ring);
 
-		r = amdgpu_fence_emit(ring, &fence);
-		if (r)
-			return r;
+		if (ring->funcs->emit_vm_flush && (job->vm_needs_flush ||
+			amdgpu_vm_is_gpu_reset(adev, id))) {
+			struct dma_fence *fence;
+			u64 pd_addr = amdgpu_vm_adjust_mc_addr(adev, job->vm_pd_addr);
 
-		mutex_lock(&adev->vm_manager.lock);
-		dma_fence_put(id->last_flush);
-		id->last_flush = fence;
-		mutex_unlock(&adev->vm_manager.lock);
+			trace_amdgpu_vm_flush(pd_addr, ring->idx, job->vm_id);
+			amdgpu_ring_emit_vm_flush(ring, job->vm_id, pd_addr);
+
+			r = amdgpu_fence_emit(ring, &fence);
+			if (r)
+				return r;
+
+			mutex_lock(&adev->vm_manager.lock);
+			dma_fence_put(id->last_flush);
+			id->last_flush = fence;
+			mutex_unlock(&adev->vm_manager.lock);
+		}
+
+		if (gds_switch_needed) {
+			id->gds_base = job->gds_base;
+			id->gds_size = job->gds_size;
+			id->gws_base = job->gws_base;
+			id->gws_size = job->gws_size;
+			id->oa_base = job->oa_base;
+			id->oa_size = job->oa_size;
+			amdgpu_ring_emit_gds_switch(ring, job->vm_id,
+							job->gds_base, job->gds_size,
+							job->gws_base, job->gws_size,
+							job->oa_base, job->oa_size);
+		}
+
+		if (ring->funcs->patch_cond_exec)
+			amdgpu_ring_patch_cond_exec(ring, patch_offset);
+
+		/* the double SWITCH_BUFFER here *cannot* be skipped by COND_EXEC */
+		if (ring->funcs->emit_switch_buffer) {
+			amdgpu_ring_emit_switch_buffer(ring);
+			amdgpu_ring_emit_switch_buffer(ring);
+		}
 	}
-
-	if (gds_switch_needed) {
-		id->gds_base = job->gds_base;
-		id->gds_size = job->gds_size;
-		id->gws_base = job->gws_base;
-		id->gws_size = job->gws_size;
-		id->oa_base = job->oa_base;
-		id->oa_size = job->oa_size;
-		amdgpu_ring_emit_gds_switch(ring, job->vm_id,
-					    job->gds_base, job->gds_size,
-					    job->gws_base, job->gws_size,
-					    job->oa_base, job->oa_size);
-	}
-
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index c59bb38..e0fa0d3 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -6675,8 +6675,6 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
 		amdgpu_ring_write(ring, 0x0);
-		/* GFX8 emits 128 dw nop to prevent CE access VM before vm_flush finish */
-		amdgpu_ring_insert_nop(ring, 128);
 	}
 }
 
@@ -7078,15 +7076,24 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.get_rptr = gfx_v8_0_ring_get_rptr,
 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
-	.emit_frame_size =
-		20 + /* gfx_v8_0_ring_emit_gds_switch */
-		7 + /* gfx_v8_0_ring_emit_hdp_flush */
-		5 + /* gfx_v8_0_ring_emit_hdp_invalidate */
-		6 + 6 + 6 +/* gfx_v8_0_ring_emit_fence_gfx x3 for user fence, vm fence */
-		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
-		128 + 19 + /* gfx_v8_0_ring_emit_vm_flush */
-		2 + /* gfx_v8_ring_emit_sb */
-		3 + 4 + 29, /* gfx_v8_ring_emit_cntxcntl including vgt flush/meta-data */
+	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
+		5 +  /* COND_EXEC */
+		7 +  /* PIPELINE_SYNC */
+		19 + /* VM_FLUSH */
+		8 +  /* FENCE for VM_FLUSH */
+		20 + /* GDS switch */
+		4 + /* double SWITCH_BUFFER,
+		       the first COND_EXEC jump to the place just
+			   prior to this double SWITCH_BUFFER  */
+		5 + /* COND_EXEC */
+		7 +	 /*	HDP_flush */
+		4 +	 /*	VGT_flush */
+		14 + /*	CE_META */
+		31 + /*	DE_META */
+		3 + /* CNTX_CTRL */
+		5 + /* HDP_INVL */
+		8 + 8 + /* FENCE x2 */
+		2, /* SWITCH_BUFFER */
 	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index f124f6d..7666add 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -3186,8 +3186,6 @@ static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
 		amdgpu_ring_write(ring, 0x0);
-		/* Emits 128 dw nop to prevent CE access VM before vm_flush finish */
-		amdgpu_ring_insert_nop(ring, 128);
 	}
 }
 
@@ -3682,15 +3680,24 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
 	.get_rptr = gfx_v9_0_ring_get_rptr_gfx,
 	.get_wptr = gfx_v9_0_ring_get_wptr_gfx,
 	.set_wptr = gfx_v9_0_ring_set_wptr_gfx,
-	.emit_frame_size =
-		20 + /* gfx_v9_0_ring_emit_gds_switch */
-		7 + /* gfx_v9_0_ring_emit_hdp_flush */
-		5 + /* gfx_v9_0_ring_emit_hdp_invalidate */
-		8 + 8 + 8 +/* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
-		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
-		128 + 66 + /* gfx_v9_0_ring_emit_vm_flush */
-		2 + /* gfx_v9_ring_emit_sb */
-		3, /* gfx_v9_ring_emit_cntxcntl */
+	.emit_frame_size = /* totally 242 maximum if 16 IBs */
+		5 +  /* COND_EXEC */
+		7 +  /* PIPELINE_SYNC */
+		46 + /* VM_FLUSH */
+		8 +  /* FENCE for VM_FLUSH */
+		20 + /* GDS switch */
+		4 + /* double SWITCH_BUFFER,
+		       the first COND_EXEC jump to the place just
+			   prior to this double SWITCH_BUFFER  */
+		5 + /* COND_EXEC */
+		7 +	 /*	HDP_flush */
+		4 +	 /*	VGT_flush */
+		14 + /*	CE_META */
+		31 + /*	DE_META */
+		3 + /* CNTX_CTRL */
+		5 + /* HDP_INVL */
+		8 + 8 + /* FENCE x2 */
+		2, /* SWITCH_BUFFER */
 	.emit_ib_size =	4, /* gfx_v9_0_ring_emit_ib_gfx */
 	.emit_ib = gfx_v9_0_ring_emit_ib_gfx,
 	.emit_fence = gfx_v9_0_ring_emit_fence,