https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81715
--- Comment #8 from Arnd Bergmann <arnd at linaro dot org> --- (In reply to Martin Liška from comment #7) > Ok, I'm quite opened for changes that will make smaller red zones for > smaller variables. However, in case of sanitization-aware inlining, it's > probably too complicated and I would rather use no_inline attribute on > places where needed. Ok, makes sense. What do you think of a possible optimization of the way that the stack variables get allocated (regardless of asan-stack), to allow gcc to reuse the stack locations for multiple instances of inlining the same function? I guess we don't want to do that for -fsanitize-address-use-after-scope, but for all other cases it sounds like a useful optimization that would drastically reduce the frame size of nl80211_send_wiphy() and many other functions. When I looked at this in the past, I found that clang does this more often than gcc already, and it did not seem to be impacted by enabling or disabling -fsanitize=kernel-address or asan-stack=1.
