Re: [PATCHv8 00/11] Linear Address Masking enabling

[Kirill A. Shutemov]

On Mon, Sep 05, 2022 at 04:30:25PM +0200, Peter Zijlstra wrote:
> On Mon, Sep 05, 2022 at 04:44:57PM +0300, Kirill A. Shutemov wrote:
> > On Mon, Sep 05, 2022 at 10:35:44AM +0530, Bharata B Rao wrote:
> > > Hi Kirill,
> > > 
> > > On 9/4/2022 6:30 AM, Kirill A. Shutemov wrote:
> > > > On Tue, Aug 30, 2022 at 04:00:53AM +0300, Kirill A. Shutemov wrote:
> > > >> Linear Address Masking[1] (LAM) modifies the checking that is applied to
> > > >> 64-bit linear addresses, allowing software to use of the untranslated
> > > >> address bits for metadata.
> > > >>
> > > >> The patchset brings support for LAM for userspace addresses. Only LAM_U57 at
> > > >> this time.
> > > >>
> > > >> Please review and consider applying.
> > > >>
> > > >> git://git.kernel.org/pub/scm/linux/kernel/git/kas/linux.git lam
> > > > 
> > > > +Bharata, Ananth.
> > > > 
> > > > Do you folks have any feedback on the patchset?
> > > > 
> > > > Looks like AMD version of the tagged pointers feature does not get
> > > > traction as of now, but I want to be sure that the interface introduced
> > > > here can be suitable for your future plans.
> > > > 
> > > > Do you see anything in the interface that can prevent it to be extended to
> > > > the AMD feature?
> > > 
> > > The arch_prctl() extensions is generic enough that it should be good.
> > > 
> > > The untagged_addr() macro looks like this from one of the callers:
> > > 
> > >         start = untagged_addr(mm, start);
> > > ffffffff814d39bb:       48 8b 8d 40 ff ff ff    mov    -0xc0(%rbp),%rcx
> > > ffffffff814d39c2:       48 89 f2                mov    %rsi,%rdx
> > > ffffffff814d39c5:       48 c1 fa 3f             sar    $0x3f,%rdx
> > > ffffffff814d39c9:       48 0b 91 50 03 00 00    or     0x350(%rcx),%rdx
> > > ffffffff814d39d0:       48 21 f2                and    %rsi,%rdx
> > > ffffffff814d39d3:       49 89 d6                mov    %rdx,%r14
> > > 
> > > Can this overhead of a few additional instructions be removed for
> > > platforms that don't have LAM feature? I haven't measured how much
> > > overhead this effectively contributes to in real, but wonder if it is
> > > worth optimizing for non-LAM platforms.
> > 
> > I'm not sure how the optimization should look like. I guess we can stick
> > static_cpu_has() there, but I'm not convinced that adding jumps there will
> > be beneficial.
> 
> I suppose the critical bit is the memory load. That can stall and then
> you're sad. A jump_label is easy enough to add.

What about something like this?

diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
index 803241dfc473..868d2730884b 100644
--- a/arch/x86/include/asm/uaccess.h
+++ b/arch/x86/include/asm/uaccess.h
@@ -30,8 +30,10 @@ static inline bool pagefault_disabled(void);
  */
 #define untagged_addr(mm, addr)	({					\
 	u64 __addr = (__force u64)(addr);				\
-	s64 sign = (s64)__addr >> 63;					\
-	__addr &= (mm)->context.untag_mask | sign;			\
+	if (static_cpu_has(X86_FEATURE_LAM)) {				\
+		s64 sign = (s64)__addr >> 63;				\
+		__addr &= (mm)->context.untag_mask | sign;		\
+	}								\
 	(__force __typeof__(addr))__addr;				\
 })
-- 
  Kiryl Shutsemau / Kirill A. Shutemov