Make SET_ERRORX() an expression evaluating to the error number

[kib]

Summary

Suggested by: markj

[markj]

Thanks.

sys/sys/exterrvar.h

51

Perhaps this should be a proper function which returns error? I'm not sure. I wonder if inlining will result in significant .text bloat.

[kib]

sys/sys/exterrvar.h

51

Wouldn't setting up args for the call consume approximately the same amount of code?

I also wanted to make mmsg store a dead store if the BLOAT option is not enabled.

[markj]

sys/sys/exterrvar.h

51

Maybe, but I'd expect a function call to be a bit smaller: it will usually be a tail call, and then the inline code doesn't need to fetch curthread. We would need to try it and see.

[kib]

sys/sys/exterrvar.h

51

So for existing calls of vm/vm_map.c

Macro
     text     data       bss        dec        hex   filename
5873740   599651   3595968   10069359   0x99a56f   sys/amd64/compile/X/kernel

Function
     text     data       bss        dec        hex   filename
  5872671   599651   3595968   10068290   0x99a142   sys/amd64/compile/X/kernel

% rg SET_ERROR sys/vm sys/kern | wc -l
      25

The difference is 1K for 25 places.

[markj]

In D50836#1160355, @markj wrote:

Thanks.

Oops, I meant to accept.

sys/sys/exterrvar.h

51

Seems worthwhile? In either case it may be worthwhile to annotate the check with __predict_false so that the body of the macro is moved to the end of the function.

[kib]

The function variant.

[markj]

sys/kern/sys_generic.c

2291 ↗

(On Diff #156988)

Left over backslash here.