On 06.08.2020 13:08, Julien Grall wrote: > On 05/08/2020 20:30, Oleksandr wrote: >> I was thinking how to split handle_hvm_io_completion() >> gracefully but I failed find a good solution for that, so decided to add >> two stubs (msix_write_completion and handle_realmode_completion) on Arm. >> I could add a comment describing why they are here if appropriate. But >> if you think they shouldn't be called from the common code in any way, I >> will try to split it. > > I am not entirely sure what msix_write_completion is meant to do on x86. > Is it dealing with virtual MSIx? Maybe Jan, Roger or Paul could help?
Due to the split brain model of handling PCI pass-through (between Xen and qemu), a guest writing to an MSI-X entry needs this write handed to qemu, and upon completion of the write there Xen also needs to take some extra action. Jan
