Re: [PATCH v2] x86/mce: set MCE_IN_KERNEL_COPYIN for all MC-Safe Copy

[Kefeng Wang]

On 2023/5/26 20:18, Kefeng Wang wrote:
> On 2023/5/26 15:09, Borislav Petkov wrote:
> > On Fri, May 26, 2023 at 02:32:42PM +0800, Kefeng Wang wrote:
> > > The best way to fix them is set MCE_IN_KERNEL_COPYIN for MC-Safe Copy,
> > > then let the core do_machine_check() to isolate corrupted page instead
> > > of doing it one-by-one.
> >
> > No, this whole thing is confused.
> >
> >   * Indicates an MCE that happened in kernel space while copying data
> >   * from user.
> >
> > #define MCE_IN_KERNEL_COPYIN
> >
> > This is a very specific exception type: EX_TYPE_COPY which got added by
> >
> >    278b917f8cb9 ("x86/mce: Add _ASM_EXTABLE_CPY for copy user access")
> >
> > but Linus then removed all such user copy exception points in
> >
> >    034ff37d3407 ("x86: rewrite '__copy_user_nocache' function")
> >
> > So now that EX_TYPE_COPY never happens.
>
> Is this broken the recover when kernel was copying from user space?
>
> + Youquan  could you help to check it?
>
> > And what you're doing is lumping the handling for
> > EX_TYPE_DEFAULT_MCE_SAFE and EX_TYPE_FAULT_MCE_SAFE together and saying
> > that the MCE happened while copying data from user.
> >
> > And XSTATE_OP() is one example where this is not really the case.
>
> Oh, for XSTATE_OP(), it uses EX_TYPE_DEFAULT_MCE_SAFE, but I'm focus on 
> EX_TYPE_DEFAULT_MCE_SAFE, which use copy_mc (arch/x86/lib/copy_mc_64.S),
> like I maintained in changelog, CoW/Coredump/nvdimm/dax, they use 
> copy_mc_xxx function,  sorry for mixed them up.
>
>
> > So no, this is not correct.
>
> so only add MCE_IN_KERNEL_COPYIN for EX_TYPE_DEFAULT_MCE_SAFE?
>
> diff --git a/arch/x86/kernel/cpu/mce/severity.c 
> b/arch/x86/kernel/cpu/mce/severity.c
> index c4477162c07d..6d2587994623 100644
> --- a/arch/x86/kernel/cpu/mce/severity.c
> +++ b/arch/x86/kernel/cpu/mce/severity.c
> @@ -293,11 +293,11 @@ static noinstr int error_context(struct mce *m, 
> struct pt_regs *regs)
>          case EX_TYPE_COPY:
>                  if (!copy_user)
>                          return IN_KERNEL;
> +               fallthrough;
> +       case EX_TYPE_DEFAULT_MCE_SAFE:
>                  m->kflags |= MCE_IN_KERNEL_COPYIN;
>                  fallthrough;

As mentioned above, I am focus on copy_mc_XXX calling, it will
abort if the exception fires when accessing the source, and we
want to isolate the corrupted src page, maybe we could a new flag
to indicate this scenario, the *Final Goals* is to let core
do_machine_check to deal with the corrupted src page.

Any suggestiong, thanks


> -
>          case EX_TYPE_FAULT_MCE_SAFE:
> -       case EX_TYPE_DEFAULT_MCE_SAFE:
>                  m->kflags |= MCE_IN_KERNEL_RECOV;
>                  return IN_KERNEL_RECOV;
>
> Correct me if I am wrong, thanks for you reviewing.
>
>
> >