On Wed, Feb 19, 2025 at 07:09:26PM +0100, Greg Kroah-Hartman wrote:
On Wed, Feb 19, 2025 at 05:18:24PM +0000, Catalin Marinas wrote:
The problem is in the arm64 arch_calc_vm_flag_bits() as it returns VM_MTE_ALLOWED for any MAP_ANONYMOUS ignoring MAP_HUGETLB (it's been doing this since day 1 of MTE). The implementation does handle the hugetlb file mmap() correctly but not the MAP_ANONYMOUS case.
The fix would be something like below:
-----------------8<-------------------------- diff --git a/arch/arm64/include/asm/mman.h b/arch/arm64/include/asm/mman.h index 5966ee4a6154..8ff5d88c9f12 100644 --- a/arch/arm64/include/asm/mman.h +++ b/arch/arm64/include/asm/mman.h @@ -28,7 +28,8 @@ static inline unsigned long arch_calc_vm_flag_bits(unsigned long flags) * backed by tags-capable memory. The vm_flags may be overridden by a * filesystem supporting MTE (RAM-based). */
- if (system_supports_mte() && (flags & MAP_ANONYMOUS))
if (system_supports_mte() &&
((flags & MAP_ANONYMOUS) && !(flags & MAP_HUGETLB)))return VM_MTE_ALLOWED;
return 0;
-------------------8<-----------------------
This fix won't make sense for mainline since it supports MAP_HUGETLB already.
Greg, are you ok with a stable-only fix as above or you'd rather see the full 25c17c4b55de ("hugetlb: arm64: add mte support") backported?
A stable-only fix for this is fine, thanks! Can you send it with a changelog and I'll queue it up. Does it also need to go to older kernels as well?
5.10, that's when we got MTE support in. There may be some slight variations depending on how far 5baf8b037deb ("mm: refactor arch_calc_vm_flag_bits() and arm64 MTE handling") got backported. This goes together with deb0f6562884 ("mm/mmap: undo ->mmap() when arch_validate_flags() fails"), so they may actually go all the way to 5.10.
I'll prepare the patches tomorrow, give them a try and send to stable.
Thanks.