return value;
 }
 
+static u32 pipe_reg_read(struct etnaviv_gpu *gpu,
+       const struct etnaviv_pm_domain *domain,
+       const struct etnaviv_pm_signal *signal)
+{
+       u32 clock = gpu_read(gpu, VIVS_HI_CLOCK_CONTROL);
+       u32 value = 0;
+       unsigned i;
+
+       for (i = 0; i < gpu->identity.pixel_pipes; i++) {
+               clock &= ~(VIVS_HI_CLOCK_CONTROL_DEBUG_PIXEL_PIPE__MASK);
+               clock |= VIVS_HI_CLOCK_CONTROL_DEBUG_PIXEL_PIPE(i);
+               gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, clock);
+               value += gpu_read(gpu, signal->data);
+       }
+
+       /* switch back to pixel pipe 0 to prevent GPU hang */
+       clock &= ~(VIVS_HI_CLOCK_CONTROL_DEBUG_PIXEL_PIPE__MASK);
+       clock |= VIVS_HI_CLOCK_CONTROL_DEBUG_PIXEL_PIPE(0);
+       gpu_write(gpu, VIVS_HI_CLOCK_CONTROL, clock);
+
+       return value;
+}
+
 static u32 hi_total_cycle_read(struct etnaviv_gpu *gpu,
        const struct etnaviv_pm_domain *domain,
        const struct etnaviv_pm_signal *signal)
                .name = "HI",
                .profile_read = VIVS_MC_PROFILE_HI_READ,
                .profile_config = VIVS_MC_PROFILE_CONFIG2,
-               .nr_signals = 5,
+               .nr_signals = 7,
                .signal = (const struct etnaviv_pm_signal[]) {
+                       {
+                               "TOTAL_READ_BYTES8",
+                               VIVS_HI_PROFILE_READ_BYTES8,
+                               &pipe_reg_read,
+                       },
+                       {
+                               "TOTAL_WRITE_BYTES8",
+                               VIVS_HI_PROFILE_WRITE_BYTES8,
+                               &pipe_reg_read,
+                       },
                        {
                                "TOTAL_CYCLES",
                                0,