Arch: Arm: Slow MMU mapping

[urvashivs11]

I'm running Zephyr on an Arch ARM64 platform and attempting to map a 2GB memory region with the attribute (K_MEM_CACHE_WB | K_MEM_PERM_RW) & (~K_MEM_PERM_EXEC);.

As part of the mapping process, Zephyr first calls:
arch_mem_unmap((void *)address, (size_t)length);
followed by:

arch_mem_map((void *)address, (uintptr_t)address, (size_t)length, mmu_attr);

During this process, Zephyr performs a cache invalidation on the entire 2GB region using: sys_cache_data_invd_range(virt, size);

This step is significantly time-consuming and negatively impacts performance. Unfortunately, splitting the 2GB region into smaller chunks isn't a viable option due to the increased page fault overhead.

Since I, as the client, know that this is the first mapping of this memory region, cache invalidation should be unnecessary. Is there a way in Zephyr to skip the cache invalidation step in such cases?

[wearyzen]

duplicate of #91151