On Sat, Apr 15, 2023 at 03:52:27PM -0700, Stephen Hemminger wrote: > On Sun, 16 Apr 2023 00:41:54 +0200 > Morten Brørup <m...@smartsharesystems.com> wrote: > > > > > > > > /** > > > > > > > * Force a structure to be packed > > > > > > > */ > > > > > > > +#ifndef RTE_TOOLCHAIN_MSVC > > > > > > > #define __rte_packed __attribute__((__packed__)) > > > > > > > +#else > > > > > > > +#define __rte_packed > > > > > > > +#endif > > Could there be cases this gets used for protocol headers or interacting with > HW memory map, And if not packed then the code will not function correctly?
yes, that's one of the valid / correct use cases and it can't be avoided if the structs are nested in recursively composed layout. there are a few instances where we don't need compiler generated static-stride. that is we don't need to force packing to get sizeof(T) == offsetof(T.lastfield) + sizeof(T.lastfield). anyway, this is more of an evaluate on a case by case basis for candidates that aren't needed. ty
