On 13.06.2018 15:03, Igor Mammedov wrote: > On Mon, 11 Jun 2018 14:16:51 +0200 > David Hildenbrand <da...@redhat.com> wrote: > >> We already verify when realizing that the memdev property has been >> set. We have no more accesses to get_memory_region() before the device >> is realized. > this stems from assumption that get_memory_region shouldn't be called > before devices realize is executed, which I don't agree to begin with. > > However wrt error handling, we should probably leave device internal error > up to devices and make check for error only in pre_plug handler > (since pre_plug was successful, the rest machine helpers would have > access to the same region until device is removed. >
Something like a generic Device "validate()"/"pre_realize()" function, that can be called before realize() and validates all properties (+initializes derived properties) would be something I could agree to. Then we could drop all error handling from access functions (as they have been validated early during pre_plug()) Moving all checks out of realize into pre_plug() looks ugly, because we have implementation details split across c-files. > >> So this function will never fail. Remove the stale check and the >> error variable. Add a comment to the functions stating that they should >> never be called on uninitialized devices. >> -- Thanks, David / dhildenb
