On 3/19/21 1:53 PM, Stefano Stabellini wrote: > On Fri, 19 Mar 2021, Boris Ostrovsky wrote: >> On 3/18/21 7:28 PM, Stefano Stabellini wrote: >>> So, I'll follow you suggestion, keep the x86 side named as it is today, >>> and provide a tiny wrapper so that we can still have an arch-neutral >>> xen_swiotlb_detect function (on x86 just calls pci_xen_swiotlb_detect.) >> >> But now on x86 side we end up with a routine that noone calls. And >> pci_xen_swiotlb_detect() becomes not __init and so it will have to stick >> around after boot for no good reason. (You could have made it __ref btw). >> >> >> I think we should leave x86 alone. And if there is a declaration in >> include/xen/swiotlb-xen.h that's only relevant to ARM --- well, so be it. Or >> perhaps you can create arch/arm/include/asm/xen/swiotlb-xen.h > Yeah I wanted to avoid creating arch/arm/include/asm/xen/swiotlb-xen.h > because I would have to do one of the following: > > 1) add one more #include <asm/xen/swiotlb-xen.h> to > arch/arm64/mm/dma-mapping.c > (#include <xen/swiotlb-xen.h> is there already) > 2) add #include <asm/xen/swiotlb-xen.h> to include/xen/swiotlb-xen.h > > What's your preference? If I have to create > arch/arm/include/asm/xen/swiotlb-xen.h, I would go with 2).
Yes, I agree --- #2 is the better choice. -boris
