Re: [PATCH v9 1/2] x86, powerpc: Rename memcpy_mcsafe() to copy_mc_to_{user, kernel}()

[Dan Williams]

On Tue, Sep 29, 2020 at 3:25 AM Borislav Petkov <bp@xxxxxxxxx> wrote:
>
> On Wed, Sep 23, 2020 at 09:41:33AM -0700, Dan Williams wrote:
> > The rename replaces a single top-level memcpy_mcsafe() with either
> > copy_mc_to_user(), or copy_mc_to_kernel().
>
> What is "copy_mc" supposed to mean? Especially if it is called that on
> two arches...
>
> > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > index 7101ac64bb20..e876b3a087f9 100644
> > --- a/arch/x86/Kconfig
> > +++ b/arch/x86/Kconfig
> > @@ -75,7 +75,7 @@ config X86
> >       select ARCH_HAS_PTE_DEVMAP              if X86_64
> >       select ARCH_HAS_PTE_SPECIAL
> >       select ARCH_HAS_UACCESS_FLUSHCACHE      if X86_64
> > -     select ARCH_HAS_UACCESS_MCSAFE          if X86_64 && X86_MCE
> > +     select ARCH_HAS_COPY_MC                 if X86_64
>
> X86_MCE is dropped here. So if I have a build which has
>
> # CONFIG_X86_MCE is not set
>
> One of those quirks like:
>
>         /*
>          * CAPID0{7:6} indicate whether this is an advanced RAS SKU
>          * CAPID5{8:5} indicate that various NVDIMM usage modes are
>          * enabled, so memory machine check recovery is also enabled.
>          */
>         if ((capid0 & 0xc0) == 0xc0 || (capid5 & 0x1e0))
>                 enable_copy_mc_fragile();
>
> will still call enable_copy_mc_fragile() and none of those platforms
> need MCE functionality?
>
> But there's a hunk in here which sets it in the MCE code:
>
>         if (mca_cfg.recovery)
>                 enable_copy_mc_fragile();
>
> So which is it? They need it or they don't?
>
> The comment over copy_mc_to_kernel() says:
>
>  * Call into the 'fragile' version on systems that have trouble
>  * actually do machine check recovery
>
> If CONFIG_X86_MCE is not set, I'll say. :)

True, without CONFIG_X86_MCE there's no point in attempting the
fragile copy because the #MC will go unhandled. At the same time the
point of the new copy_mc_generic() is that it is suitable to use
without CONFIG_X86_MCE as it's just a typical fast string copy
instrumented for exception handling. So, I still think
CONFIG_ARCH_HAS_COPY_MC is independent of CONFIG_X86_MCE, but
enable_copy_mc_fragile() should be stubbed out by CONFIG_X86_MCE=n,
will re-spin.

>
> > +++ b/arch/x86/lib/copy_mc.c
> > @@ -0,0 +1,66 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/* Copyright(c) 2016-2020 Intel Corporation. All rights reserved. */
> > +
> > +#include <linux/jump_label.h>
> > +#include <linux/uaccess.h>
> > +#include <linux/export.h>
> > +#include <linux/string.h>
> > +#include <linux/types.h>
> > +
> > +static DEFINE_STATIC_KEY_FALSE(copy_mc_fragile_key);
> > +
> > +void enable_copy_mc_fragile(void)
> > +{
> > +     static_branch_inc(&copy_mc_fragile_key);
> > +}
> > +
> > +/**
> > + * copy_mc_to_kernel - memory copy that that handles source exceptions
>
> One "that" is enough.

Yup.