toolchain: fix for building blu_spark kernel using google's aosp toolchains. addresses error: undefined symbol 'stack_chk_guard';

[micfogas]

details: this patch checks for CC_IS_AOSP_CLANG before
stackprotector. if it is, then global stack-
protector guard is used, instead of per-task
and sysreg (using a canary instead of TLS).

            blu_spark KCFLAGS changes from aosp kernel
            source tree in /Makefile cause final linking
            of vmlinux.o to fail with per-task stack-
            protector guard. this patch is designed to
            resolve the error without modifying blu_spark
            code.

    credit: micfogas

signed-off-by: micfogas tech@technotic.us

[engstk]

This doesn't make much sense outside of forks, not aiming to build with exp higher clang.
It builds fine with clang-r487747c, which is the stable ABI used across all production android14-6.1 trees.
Any clang 17.X will work fine and that should what we would expect for better API compatibility.
Also global vs. per-task(sysreg) lowers security a bit.

[micfogas]

I'm curious about your build environment. I determined that the CFLAG '-mcpu=cortex-a55' in your Makefile is the cause of this problem on both of my build systems. Both Almalinux 10 and Ubuntu 22.04.

I am aware that global vs. per_task is less secure. It was an older patch I made and yes, even 487747c has this issue. I build primarily with 19.0.1 (r536225) due to some optimizations with clang 19 over 17, particularly with CFI.

I am far from an advanced kernel dev, and I've been doing it for maybe only 6 months. I started learning to build for my Pixel 6 Pro (device target raven, kernel target raviole) from forking your kernel source. I then requested the source for 250916 (qpr2 b3) and then 251205 (qpr2 release) from Google, which gave me the full Bazel/kleaf systems, though I've stuck to keeping KMI compatibility creating only Image.lz4.

I provided the patch (I suppose simply omitting the problem CFLAG would be better, I'm not sure on all the optimizations that cortex-a55 adds) as a workaround without modifying your code any, including your Makefile optimizations.

Do you have any insight on why that CFLAG would cause an issue with stack guard? Thanks for your time.