GCC doesnt work with -O0

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

When i compile my Kernel with no optimization flags or with -O0 (same thing), it randomly works/doesn't work or causes faults or print doesn't work or i get weird results from kmemcpy. it also causes the output to grow to 332k from 44k in size. I am clueless to what is going on... I'v tried "fixing" functions but nothing works, and all works fine with -O1, -O2, -O3, -Ofast and -Os.

Compile line from script:
~/opt/cross/bin/x86_64-elf-gcc -I./kheaders -c -s $1 -o $2 -mgeneral-regs-only -ffreestanding -O0 -mcmodel=large

Octocontrabass · **Joined:** Mon Mar 25, 2013 7:01 pm **Posts:** 5173

I see you're not passing "-mno-red-zone" to your cross-compiler. If you have interrupts enabled, they'll corrupt the red zone (unless you use the IST for all interrupts, which is a questionable design choice). Optimizations tend to reduce usage of the red zone, so that might be why you haven't seen problems when optimizations are enabled.

I see you're also using the large code model instead of the kernel code model. Why is that?

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

Funny thing it just kinda randomly does it... for some reason ld doesnt work with kernel
I will try the red zone flag, thank you

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

Ok thank you, that seems to be the problem...but it was weird if i ran it several times it would fail once in a while even with no code changes.

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

Probably from the clock interrupt.

Octocontrabass · **Joined:** Mon Mar 25, 2013 7:01 pm **Posts:** 5173

Xeno wrote:

for some reason ld doesnt work with kernel

The reason is usually that the kernel code model requires your linker script to place your kernel between 0xFFFFFFFF80000000 and 0xFFFFFFFFFFFFFFFF, and your linker script probably doesn't do that.

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

My kernel is loaded at 0xffffFA0000000000

Octocontrabass · **Joined:** Mon Mar 25, 2013 7:01 pm **Posts:** 5173

Why? If you load it at 0xFFFFFFFF80000000 you can use the kernel code model, which will make your kernel smaller and faster.

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

I have it maped there so that i can allocate the virtual memory above it for kernel data structures (page maps/memory bitmap etc) with all user data below
This is also for a method of memory allocation I designed, it uses a different design for virtual memory, then the typical idea

Octocontrabass · **Joined:** Mon Mar 25, 2013 7:01 pm **Posts:** 5173

Why can't you allocate those kernel data structures at 0xFFFFFA0000000000 and link your kernel at (or above) 0xFFFFFFFF80000000?

Xeno · **Joined:** Tue Oct 10, 2023 7:40 pm **Posts:** 52

I might but i think it would conflict with my goals...I'll look into it

OSDev.org

GCC doesnt work with -O0

Who is online