On Thursday 26 September 2013, Kishon Vijay Abraham I wrote: > On Wednesday 25 September 2013 02:53 AM, Arnd Bergmann wrote: > > On Monday 23 September 2013, Kishon Vijay Abraham I wrote: > >> Btw if we hadn't programmed inbound translation table, the address will go > >> untranslated (according to the data book). I guess that's how it was working > >> for Jingoo Han. > >> > >> ** > >> 3.10.4 > >> Inbound iATU Operation > >> > >> When there is no match, then the address is untranslated > >> ** > >> > > > > Well, that should work just as well, since you have a 1:1 translation anyway. > > Do you get the same error without the translation? > > Yes. I get the same non-fatal error interrupt in RC. Ok, then I guess the translation is actually not at fault here but something else. I would recommend looking at the IOMMU as the potential culprit. Maybe having it disabled means that no DMA is going through, rather than all DMA going through untranslated. Another possibility is that the IOMMU is set up so that when disabled, it maps DMA address 0 to the start of RAM, rather than identity mapping DMA address 0x80000000 there. If that's the case, you either have to use the IOMMU, or set up the mapping in the root complex to revert it. Arnd -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html