On Fri, 19 Dec 2014, Markos Chandras wrote: > >> The use of "add" instruction for immediate operations is wrong and > >> relies to gas being smart enough to notice that and replace it with > >> either addi or addui. However, MIPS R6 removed the addi instruction > >> so, fix this problem properly by using the correct instruction > >> directly. Not true, depending on the arguments the ADD assembly macro expands to either of the ADD and the ADDI hardware instructions; where an immediate outside the 16-bit signed range is used it also expands to a longer sequence involving LUI and the actual operation is ADD. It never expands to ADDIU (which I gather you meant). > > This is another case of the use of "add" being a real bug. We should > > never have faulting instructions like this in the kernel. > > > > Can you send all patches in this set that fix this bug as a separate > > patch? Since they are obviously correct, and really should be used by > > all non-R6 processors, we can get them in sooner that the entire R6 thing. > > > > Thanks, > > David Daney > > sure i will move these patches away from R6 and post them separately. I think using the ADDU macro is preferred here as it allows arbitrary 32-bit values for `off', just like with memory references in MIPS assembly instructions. Maciej
