amdgpu_ring_write(ring, 0);
 }
 
+static void gfx_v11_0_ring_emit_gfx_shadow(struct amdgpu_ring *ring,
+                                          u64 shadow_va, u64 csa_va,
+                                          u64 gds_va, bool init_shadow,
+                                          int vmid)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (!adev->gfx.cp_gfx_shadow)
+               return;
+
+       amdgpu_ring_write(ring, PACKET3(PACKET3_SET_Q_PREEMPTION_MODE, 7));
+       amdgpu_ring_write(ring, lower_32_bits(shadow_va));
+       amdgpu_ring_write(ring, upper_32_bits(shadow_va));
+       amdgpu_ring_write(ring, lower_32_bits(gds_va));
+       amdgpu_ring_write(ring, upper_32_bits(gds_va));
+       amdgpu_ring_write(ring, lower_32_bits(csa_va));
+       amdgpu_ring_write(ring, upper_32_bits(csa_va));
+       amdgpu_ring_write(ring, shadow_va ?
+                         PACKET3_SET_Q_PREEMPTION_MODE_IB_VMID(vmid) : 0);
+       amdgpu_ring_write(ring, init_shadow ?
+                         PACKET3_SET_Q_PREEMPTION_MODE_INIT_SHADOW_MEM : 0);
+}
+
 static unsigned gfx_v11_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
 {
        unsigned ret;
        .set_wptr = gfx_v11_0_ring_set_wptr_gfx,
        .emit_frame_size = /* totally 242 maximum if 16 IBs */
                5 + /* COND_EXEC */
+               9 + /* SET_Q_PREEMPTION_MODE */
                7 + /* PIPELINE_SYNC */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
        .insert_nop = amdgpu_ring_insert_nop,
        .pad_ib = amdgpu_ring_generic_pad_ib,
        .emit_cntxcntl = gfx_v11_0_ring_emit_cntxcntl,
+       .emit_gfx_shadow = gfx_v11_0_ring_emit_gfx_shadow,
        .init_cond_exec = gfx_v11_0_ring_emit_init_cond_exec,
        .patch_cond_exec = gfx_v11_0_ring_emit_patch_cond_exec,
        .preempt_ib = gfx_v11_0_ring_preempt_ib,