On 12 March 2018 at 15:01, Auger Eric <[email protected]> wrote: > Hi Peter, > > On 12/03/18 13:46, Peter Maydell wrote: >> On 17 February 2018 at 18:46, Eric Auger <[email protected]> wrote: >>> + qemu_fdt_setprop(vms->fdt, node, "dma-coherent", NULL, 0); >> >> Is this definitely the right setting for dma-coherent? My (possibly >> naive) thought is that for an emulated SMMU our page-table walks are >> going to be coherent with the CPU's view. > > Indeed we add the dma-capable property to make the DMA ops of the SMMU > cache coherent with the CPU. We pass NULL/O since there is just > dma-coherent in the dt property. What am I missing?
My mistake, I thought this was the equivalent of "dma-coherent = 0;". thanks -- PMM
