On Fri, 2023-06-23 at 21:43 +0800, WANG Xuerui wrote:Hmm, now I remember (still vaguely) about the discussion... I'll have to check how to minimize churn around FPU-touching code though if -msoft-float is to be kept.
-cflags-y += -G0 -pipe -msoft-float-msoft-float should not be removed. Our consensus (made when I was
developing https://gcc.gnu.org/r13-6500) is -mabi=lp64s does *not*
disable floating point instructions, but only disable FPRs for passing
arguments and return values. So w/o -msoft-float (or -mfpu=none) GCC is
allowed to generate FP instructions everywhere in kernel and it may
cause kernel FPD exception in the future.