[PATCH] arm64: mte: Document that user PSTATE.TCO is ignored by kernel uaccess

[Catalin Marinas]

On Mon, Oct 26, 2020 at 01:14:15PM +0000, Will Deacon wrote:
> On Mon, Oct 26, 2020 at 12:16:56PM +0000, Catalin Marinas wrote:
> > On exception entry, the kernel explicitly resets the PSTATE.TCO (tag
> > check override) so that any kernel memory accesses will be checked (the
> > bit is restored on exception return). This has the side-effect that the
> > uaccess routines will not honour the PSTATE.TCO that may have been set
> > by the user prior to a syscall.
> > 
> > There is no issue in practice since PSTATE.TCO is expected to be used
> > only for brief periods in specific routines (e.g. garbage collection).
> > To control the tag checking mode of the uaccess routines, the user will
> > have to invoke a corresponding prctl() call.
> > 
> > Document the kernel behaviour w.r.t. PSTATE.TCO accordingly.
> > 
> > Signed-off-by: Catalin Marinas <catalin.marinas at arm.com>
> > Fixes: df9d7a22dd21 ("arm64: mte: Add Memory Tagging Extension documentation")
> > Cc: Will Deacon <will at kernel.org>
> > Cc: Szabolcs Nagy <szabolcs.nagy at arm.com>
> > ---
> >  Documentation/arm64/memory-tagging-extension.rst | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/Documentation/arm64/memory-tagging-extension.rst b/Documentation/arm64/memory-tagging-extension.rst
> > index 034d37c605e8..648f8e8d480b 100644
> > --- a/Documentation/arm64/memory-tagging-extension.rst
> > +++ b/Documentation/arm64/memory-tagging-extension.rst
> > @@ -102,7 +102,9 @@ applications.
> >  system call) are not checked if the user thread tag checking mode is
> >  ``PR_MTE_TCF_NONE`` or ``PR_MTE_TCF_ASYNC``. If the tag checking mode is
> >  ``PR_MTE_TCF_SYNC``, the kernel makes a best effort to check its user
> > -address accesses, however it cannot always guarantee it.
> > +address accesses, however it cannot always guarantee it. The user
> > +``PSTATE.TCO`` bit is ignored by the kernel accesses to user addresses,
> > +its value assumed 0.
> 
> nit, but the wording feels a bit odd to me here. How about:
> 
>   Kernel accesses to user addresses are always performed with an effective
>   PSTATE.TCO value of zero, regardless of the user configuration.

That's better. Thanks.

-- 
Catalin