Dealing with two-abi architectures
"Wichmann, Mats D" <[email protected]> Tue, 12 Aug 2003 19:56:06 -0700
| Newsgroups | gmane.linux.lsb.implementation |
|---|---|
| Message-ID | <[email protected]> |
Several of the LSB architectures require dual-ABI support. As of right now, I haven't read two of the three processor supplements referenced (hey, there's only so much time...) Itanium systems are delivered with LP64 Itanium binaries, and have a requirement to run IA32 binaries. This requirement is a "common usage" requirement and is not enshrined in the LSB. Thus this isn't a real dual-ABI system from the LSB viewpoint. AMD64 systems are delivered as LP64 AMD64 binaries, and must also run IA32 binaries. PPC64 systems are delivered as ILP32 PPC32 native binaries, and must also run LP64 PPC64 binaries. The lsbsi question is: how should these requirements be accommodated, and what changes to the lsbsi build procedure need to be made to implement that. In past discussions, the feeling has been that the lsbsi should be built consistent with what the Linux distributions are doing, and a separate build phase would build the libraries that support the "other ABI". That is, the AMD64 lsbsi should be a 64-bit build using the current lsbsi model, and then an extra phase is done to produce ILP32 libraries. And the PPC64 build should be essentially the PPC32 build (we think we could use the PPC32 lsbsi, but maybe prefer to build it in a PPC64 environment - open for discussion) and the extra phase builds the LP64 libraries. I propose to ignore the Itanium issue at this time. If this seems like the correct approach, I'll start looking into how to structure this within the lsbsi build process (perhaps after looking at the issues raised by uplifting the build tool to the current version, which has been on my todo list for a while but waiting for a moment when there isn't a great deal of time pressure). It's still expected that the "architecture champions" for PPC64 and AMD64 will actually take the process through to release. If it's not the right approach, now would be a good time to bring up why not, before time gets invested into going down the wrong path. Cheers, Mats