On 2018-11-07 4:03 pm, Joerg Roedel wrote:
On Wed, Nov 07, 2018 at 04:50:40PM +0100, Geert Uytterhoeven wrote:It only got triggered by the combination of commits 6c2fb2ea76361da9 ("of/device: Set bus DMA mask as appropriate") and b4ebe6063204da58 ("dma-direct: implement complete bus_dma_mask handling"), which is being fixed by "of/device: Really only set bus DMA mask when appropriate" (https://patchwork.kernel.org/patch/10670177/).Okay, but the bug is triggered since 6c2fb2ea76361da9, so I took this one for the fixes-tag.
FWIW it looks like it *has* always been possible to hit this crash by allocating a domain and freeing it again without attaching any devices, it's just highly improbable for any sane code to do that explicitly, so the real latent triggers are failure paths in external callers (which in this case are themselves only being reached thanks to my bug elsewhere).
Robin.
