On Mon, Jul 15, 2013 at 09:33:25AM +0200, Laszlo Ersek wrote: > On 07/14/13 13:41, Michael S. Tsirkin wrote: > > On Fri, Jul 12, 2013 at 04:17:28PM +0200, Laszlo Ersek wrote: > >> On 07/10/13 15:51, Michael S. Tsirkin wrote: > > >>> +struct BiosLinkerLoaderEntry { > >>> + uint32_t command; > >>> + union { > >>> + /* > >>> + * COMMAND_ALLOCATE - allocate a table from @alloc_file > >>> + * subject to @alloc_align alignment (must be power of 2) > >>> + * and @alloc_zone (can be HIGH or FSEG) requirements. > >>> + * > >>> + * Must appear exactly once for each file, and before > >>> + * this file is referenced by any other command. > >>> + */ > >>> + struct { > >>> + char alloc_file[BIOS_LINKER_LOADER_FILESZ]; > >>> + uint32_t alloc_align; > >>> + uint8_t alloc_zone; > >>> + }; > >> > >> I think in OVMF we won't rely on the alloc_zone / alloc_align members, > >> but that's OVMF's private business. > > > > RSDP must be in FSEG though > > I didn't express myself clearly, sorry. The default edk2 ACPI table > protocol that OVMF uses should allocate RSDP and the like automatically > in correct regions. (Allocating reserved memory for > External(XXXX,OpRegionObj) needs a different call though.) > > >>> + > >>> + /* padding */ > >>> + char pad[124]; > >>> + }; > >> > >> The unnamed union member is a gcc-ism. I'd give it a short name (like > >> "u"), but feel free to ignore this. > >> > > > > This isn't a gcc-ism. It's in C1x: > > > > An unnamed member whose type specifier is a structure specifier with no > > tag is called an anonymous structure; an unnamed member whose type > > specifier is a union specifier with no tag is called an anonymous union. > > The members of an anonymous structure or union are considered to be > > members of the containing structure or union. This applies recursively > > if the containing structure or union is also anonymous. > > This part of the discussion is academic, but the unnamed union member is > a gcc-ism in the qemu source, because AFAIK qemu uses the gnu89 dialect > by default, and gnu99 on Solaris. >
The important thing is that it's part of the standard now, so won't go away or interfere with porting to a new compiler if we ever try to do it. > >>> +}; > >>> +typedef struct BiosLinkerLoaderEntry BiosLinkerLoaderEntry; > >> > >> Probably not needed in practice, but for documentation purposes I > >> suggest QEMU_PACKED from "include/qemu/compiler.h". > > > > It's not required in practice but I can add this though I'm not sure - > > what does this document? > > To me it documents that we rely on the absence of inter-member padding. > > Thanks > Laszlo