Re: [PATCH v2 -tip] x86/percpu: Use C for arch_raw_cpu_ptr()

[Nadav Amit]

> On Oct 18, 2023, at 12:53 AM, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> 
> 
> but none of this should even matter, because by the time we actually
> *schedule* that thread, we'll set active_mm to the right thing.
> 
> Can anybody see what's up?

Could it be related to exec_mmap() -> exec_mm_release() -> mm_release() -> deactivate_mm() ?

#define deactivate_mm(tsk, mm)                  \
do {                                            \
        if (!tsk->vfork_done)                   \
                shstk_free(tsk);                \
        load_gs_index(0);                       \
        loadsegment(fs, 0);                     \
} while (0)

We change gs_index(), so perhaps it affects later GS reads. There is also this
X86_BUG_NULL_SEG. Need to dive deeper; just initial thoughts though (i.e., I might be
completely off).