[Bug sanitizer/111736] Address sanitizer is not compatible with named address spaces

["pchelkin at ispras dot ru" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111736

Fedor Pchelkin <pchelkin at ispras dot ru> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pchelkin at ispras dot ru

--- Comment #42 from Fedor Pchelkin <pchelkin at ispras dot ru> ---
Hi,

I'm facing upstream Linux kernel boot failures with KASAN enabled after recent
x86/percpu updates. HEAD kernel commit is
daa121128a2d2ac6006159e2c47676e4fcd21eab.

The kernel is built with GCC 14.1.1 20240507. It's Arch Linux, package details
here [1].

The failing Linux kernel commit is f61f02d1ff78 ("x86/percpu: Re-enable named
address spaces with KASAN for GCC 13.3+") [2].

You see that the above kernel commit states that GCC 14 must be fixed with
patches from this pull-request:

    Note that the patch considers GCC 14 to be fixed - if somebody is
    using snapshots of the GCC 14 before the fix, they should upgrade.

I'm not familiar with GCC stable branches model but it seems to me that GCC
13.3 was fixed with patches from this pull-request while GCC 14.1 didn't get
the fixes at the time. Or it did? frankly speaking, I haven't been able to
figure it out for sure (and would be surprised if latest GCC 14 release didn't
get the fixes before the kernel received an x86/percpu update regarding this
issue).

If GCC 14.1 is fixed then there is probably another issue found. However, I
still think that if it was a real widespread problem, then somebody would have
already reported it. So maybe my kernel-config is just weird :)

I've attached a dmesg output and kernel config I use for testing.

The failing code lines are decoded as following:
__text_poke+0x493/0xd50:
use_temporary_mm at arch/x86/kernel/alternative.c:1789 (discriminator 2)
(inlined by) __text_poke at arch/x86/kernel/alternative.c:1904 (discriminator
2)

This concerns a percpu variable access:

static inline temp_mm_state_t use_temporary_mm(struct mm_struct *mm)
{
[...]
        /*
         * Make sure not to be in TLB lazy mode, as otherwise we'll end up
         * with a stale address space WITHOUT being in lazy mode after
         * restoring the previous mm.
         */
        if (this_cpu_read(cpu_tlbstate_shared.is_lazy))
                leave_mm();


[1]:
https://gitlab.archlinux.org/archlinux/packaging/packages/gcc/-/blob/main/PKGBUILD?ref_type=heads
[2]:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f61f02d1ff788ae5ad485ef8edd88d9c93557994