Yesterday a discussion on LKML raised the question if it is possible to disable optimization for kernel build. Sometimes you want to debug the kernel and the required function is optimized out. You are free to disassembly the source code at this place and set the break point manually. But this procedure is time consuming and error prone. I raised another use case: perf probes. Dynamic probepoints placed in the Kernel at runtime. For example you can check if a function is probeable via:

perf probe -k /usr/src/build-net-next/vmlinux -s /usr/src/linux --line tcp_rcv_established

But then if you place a probepoint you will eventually realize that this particular code is optimized out. Sometimes I used probepoint in kernel to understand time-critical intrinsic behavior. So there are use cases where the hard optimization is not required or even wanted! The reader may now note to disable kernel optimization and voilà. But there are some obstacles. The Linux Kernel per default is build with O2 or if OPTIMIZE_FOR_SIZE is selected with 0s. Both include a lot of gcc switches to optmize code. Among other flags -finline-small-functions, -fpartial-inlining and -findirect-inlining. Hypothetical: if build with no optimization at all, some function are always enabled because some of the Linux code is annotated with always inline attribute.

http://www.delorie.com/gnu/docs/gcc/gcc_81.html

A last warning: unoptimized builds are not tested (or at least only a few people have ever tested such a kernel)! So please do not use this kernel where sensitive data is stored!