Hi Paul, > And thanks for your review :) You're welcome! > For the record, the reason I went with placing the EXPORT_SYMBOL invocations > at the start of the functions rather than the end is that the end isn't always > the end of the code in question. For example (until another patch of mine) > memcpy ends part way through user_copy, with code continuing afterwards. We > would then need to place a .insn after the use of EXPORT_SYMBOL if any code > may branch there. Containing the need for .insn to the start of the function > seems neater since a function should always begin with an instruction, which > after this patch will be marked as such, and users of the macros will just get > behaviour that seems natural & expected. Well, the real end is where END (`.end') is, and I think placing `.end' midway-through a code block is really bad style, partly because it affects frame annotation (all the stuff from `.frame', `.mask', etc.) recorded in the object assembled. A LEAF/NESTED (`.ent') and END (`.end') pair should really only mark the physical beginning and ending of a whole single function. If a function has alternative entry points, then FEXPORT should be used for those, which on the MIPS target should include a `.aent' pseudo-op as a part of the symbol definition and I can't really tell why why we don't (trivial patch now in the works). NB in the MIPS assembly dialect `.type symbol, @function' is redundant in the presence of either `.ent symbol' or `.aent symbol', as is `.size symbol, . - symbol' where `.end symbol' is also present. There's no harm either though. > Of course another alternative would be to place EXPORT_SYMBOL before LEAF/ > NESTED/FEXPORT, but I don't think that would really make any difference > presuming people agree that this patch is a good idea regardless. TBH I do maintain keeping all the EXPORT_SYMBOL stuff outside function bodies would be the cleanest approach, grouping the handling for all given function's intended entry points after its final END, arranged as I described above. Do you find this proposal problematic for some reason? Maciej
