Re: [PATCH v2 07/23 replacement] mips: add pte_unmap() to balance pte_offset_map()

[Yu Zhao <yuzhao@google.com>]

On Fri, Jun 16, 2023 at 9:54 PM Yu Zhao <yuzhao@google.com> wrote:
>
> On Thu, Jun 15, 2023 at 04:02:43PM -0700, Hugh Dickins wrote:
> > To keep balance in future, __update_tlb() remember to pte_unmap() after
> > pte_offset_map().  This is an odd case, since the caller has already done
> > pte_offset_map_lock(), then mips forgets the address and recalculates it;
> > but my two naive attempts to clean that up did more harm than good.
> >
> > Tested-by: Nathan Chancellor <nathan@kernel.org>
> > Signed-off-by: Hugh Dickins <hughd@google.com>
>
> FWIW: Tested-by: Yu Zhao <yuzhao@google.com>
>
> There is another problem, likely caused by khugepaged, happened multiple times. But I don't think it's related to your series, just FYI.
>
>   Got mcheck at ffffffff81134ef0
>   CPU: 3 PID: 36 Comm: khugepaged Not tainted 6.4.0-rc6-00049-g62d8779610bb-dirty #1

...

>   Kernel panic - not syncing: Caught Machine Check exception - caused by multiple matching entries in the TLB.

In case anyone plans to try to fix this - the problem goes back to at
least 5.15 stable. My (educated) guess is that nobody complained about
it because all the testing is done in QEMU, which does NOT detect
conflicting TLBs. This means the verification of the fix would need to
be on a real piece of h/w or an updated QEMU.

In target/mips/tcg/sysemu/tlb_helper.c:

static void r4k_fill_tlb(CPUMIPSState *env, int idx)
{
    r4k_tlb_t *tlb;
    uint64_t mask = env->CP0_PageMask >> (TARGET_PAGE_BITS + 1);

    /* XXX: detect conflicting TLBs and raise a MCHECK exception when needed */
...