[v2] x86_64: Update THREAD_SETMEM/THREAD_SETMEM_NC for IMM64

Message ID CAMe9rOoYT36HpmoBz9NpnOsK+Q_-T8v7TQpdBBNzwvvTnCCuEg@mail.gmail.com
State Superseded
Delegated to: Carlos O'Donell
Headers
Series [v2] x86_64: Update THREAD_SETMEM/THREAD_SETMEM_NC for IMM64 |

Commit Message

H.J. Lu March 2, 2021, 2:21 p.m. UTC
  On Mon, Mar 1, 2021 at 5:30 AM Carlos O'Donell <carlos@redhat.com> wrote:
>
> On 2/2/21 2:12 PM, H.J. Lu via Libc-alpha wrote:
> > Since there is only "movq imm32s, mem64" and no "movq imm64, mem64", use
> > "movq reg64, mem64" to store 64-bit constant in TCB.
> > ---
> >  sysdeps/x86_64/nptl/tls.h | 11 +++++++++++
> >  1 file changed, 11 insertions(+)
> >
> > diff --git a/sysdeps/x86_64/nptl/tls.h b/sysdeps/x86_64/nptl/tls.h
> > index 20f0958780..9ec8703e45 100644
> > --- a/sysdeps/x86_64/nptl/tls.h
> > +++ b/sysdeps/x86_64/nptl/tls.h
> > @@ -269,6 +269,11 @@ _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
> >         asm volatile ("movl %0,%%fs:%P1" :                                  \
> >                    : IMM_MODE (value),                                      \
> >                      "i" (offsetof (struct pthread, member)));              \
> > +     else if (__builtin_constant_p (value)                                 \
> > +           && (int64_t) (int32_t) (uintptr_t) value != (uintptr_t) value)  \
> > +       asm volatile ("movq %0,%%fs:%P1" :                                  \
> > +                  : "r" (value),                                           \
> > +                    "i" (offsetof (struct pthread, member)));              \
>
> (1) Move conditional into 'else /* 8 */' section.
>
> The blocks of conditionals are predicated on the size of the member we are
> about to write to.
>
> In the block below...
>
> >       else /* 8 */                                                          \
>
> ... here, we are about to write value into a member that is size 8.
>
> Your code changes the logical construction of the code, but in way that makes
> it more difficult to understand.
>
> We previously had:
>
> if (sizeof() == 1)
>
> else if (sizeof() == 4)
>
> else /* Assume 8 */
>
> In your case we must already be in the 'else /* Assume 8 */' because otherwise
> we've be writing a 64-bit constant into a < 64-bit structure member.
>
> I think we should put your code into the else clause.
>
> 272      else /* 8 */                                                             \
> 273        {                                                                      \
>
>              if (__builtin_constant_p (value)
>                  && ([the value is a >32-bit constant])
>                asm volatile ([use movq reg64, mem64]);
>              else
>
> 274          asm volatile ("movq %q0,%%fs:%P1" :                                  \
> 275                        : IMM_MODE ((uint64_t) cast_to_integer (value)),       \
> 276                          "i" (offsetof (struct pthread, member)));            \
> 277        }})
>
> (2) What if gcc can't prove it's constant?
>
> If the constant is >32-bit, but gcc can't prove it's constant, then don't we
> try to put a >32-bit constant into a 32-bit immediate?
>
> Shouldn't the code be structured the other way around?
>
> e.g.
>
> else /* 8 */
> {
>   if (__builtin_constant_p (value)
>       && ([the value is a <= 32-bit constant])
>     asm volatile ([use movq imm32, mem64]);
>   else
>     asm volatile ([use movq reg64, mem64]);
> }
>
> This way the code is always correct?

I changed it to

         if (!__builtin_constant_p (value)                                    \
             || (int64_t) (int32_t) (uintptr_t) value == (uintptr_t) value)   \
           asm volatile ("movq %q0,%%fs:%P1" :                                \
                         : IMM_MODE ((uint64_t) cast_to_integer (value)),     \
                           "i" (offsetof (struct pthread, member)));          \
         else                                                                 \
           asm volatile ("movq %0,%%fs:%P1" :                                 \
                         : "r" (value),                                       \
                           "i" (offsetof (struct pthread, member)));          \

> >         {                                                                   \
> >        asm volatile ("movq %q0,%%fs:%P1" :                                  \
> > @@ -294,6 +299,12 @@ _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
> >                    : IMM_MODE (value),                                      \
> >                      "i" (offsetof (struct pthread, member[0])),            \
> >                      "r" (idx));                                            \
> > +     else if (__builtin_constant_p (value)                                 \
> > +           && (int64_t) (int32_t) (uintptr_t) value != (uintptr_t) value)  \
> > +       asm volatile ("movq %0,%%fs:%P1(,%q2,8)" :                          \
> > +                  : "r" (value),                                           \
> > +                    "i" (offsetof (struct pthread, member[0])),            \
> > +                    "r" (idx));                                            \
> >       else /* 8 */                                                          \
> >         {                                                                   \
> >        asm volatile ("movq %q0,%%fs:%P1(,%q2,8)" :                          \
> >
>

Here is the v2 patch.  OK for master?

Thanks.
  

Comments

Carlos O'Donell March 8, 2021, 10:28 p.m. UTC | #1
On 3/2/21 9:21 AM, H.J. Lu wrote:
> On Mon, Mar 1, 2021 at 5:30 AM Carlos O'Donell <carlos@redhat.com> wrote:
>>
>> On 2/2/21 2:12 PM, H.J. Lu via Libc-alpha wrote:
>>> Since there is only "movq imm32s, mem64" and no "movq imm64, mem64", use
>>> "movq reg64, mem64" to store 64-bit constant in TCB.
>>> ---
>>>  sysdeps/x86_64/nptl/tls.h | 11 +++++++++++
>>>  1 file changed, 11 insertions(+)
>>>
>>> diff --git a/sysdeps/x86_64/nptl/tls.h b/sysdeps/x86_64/nptl/tls.h
>>> index 20f0958780..9ec8703e45 100644
>>> --- a/sysdeps/x86_64/nptl/tls.h
>>> +++ b/sysdeps/x86_64/nptl/tls.h
>>> @@ -269,6 +269,11 @@ _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
>>>         asm volatile ("movl %0,%%fs:%P1" :                                  \
>>>                    : IMM_MODE (value),                                      \
>>>                      "i" (offsetof (struct pthread, member)));              \
>>> +     else if (__builtin_constant_p (value)                                 \
>>> +           && (int64_t) (int32_t) (uintptr_t) value != (uintptr_t) value)  \
>>> +       asm volatile ("movq %0,%%fs:%P1" :                                  \
>>> +                  : "r" (value),                                           \
>>> +                    "i" (offsetof (struct pthread, member)));              \
>>
>> (1) Move conditional into 'else /* 8 */' section.
>>
>> The blocks of conditionals are predicated on the size of the member we are
>> about to write to.
>>
>> In the block below...
>>
>>>       else /* 8 */                                                          \
>>
>> ... here, we are about to write value into a member that is size 8.
>>
>> Your code changes the logical construction of the code, but in way that makes
>> it more difficult to understand.
>>
>> We previously had:
>>
>> if (sizeof() == 1)
>>
>> else if (sizeof() == 4)
>>
>> else /* Assume 8 */
>>
>> In your case we must already be in the 'else /* Assume 8 */' because otherwise
>> we've be writing a 64-bit constant into a < 64-bit structure member.
>>
>> I think we should put your code into the else clause.
>>
>> 272      else /* 8 */                                                             \
>> 273        {                                                                      \
>>
>>              if (__builtin_constant_p (value)
>>                  && ([the value is a >32-bit constant])
>>                asm volatile ([use movq reg64, mem64]);
>>              else
>>
>> 274          asm volatile ("movq %q0,%%fs:%P1" :                                  \
>> 275                        : IMM_MODE ((uint64_t) cast_to_integer (value)),       \
>> 276                          "i" (offsetof (struct pthread, member)));            \
>> 277        }})
>>
>> (2) What if gcc can't prove it's constant?
>>
>> If the constant is >32-bit, but gcc can't prove it's constant, then don't we
>> try to put a >32-bit constant into a 32-bit immediate?
>>
>> Shouldn't the code be structured the other way around?
>>
>> e.g.
>>
>> else /* 8 */
>> {
>>   if (__builtin_constant_p (value)
>>       && ([the value is a <= 32-bit constant])
>>     asm volatile ([use movq imm32, mem64]);
>>   else
>>     asm volatile ([use movq reg64, mem64]);
>> }
>>
>> This way the code is always correct?
> 
> I changed it to
> 
>          if (!__builtin_constant_p (value)                                    \
>              || (int64_t) (int32_t) (uintptr_t) value == (uintptr_t) value)   \
>            asm volatile ("movq %q0,%%fs:%P1" :                                \
>                          : IMM_MODE ((uint64_t) cast_to_integer (value)),     \
>                            "i" (offsetof (struct pthread, member)));          \
>          else                                                                 \
>            asm volatile ("movq %0,%%fs:%P1" :                                 \
>                          : "r" (value),                                       \
>                            "i" (offsetof (struct pthread, member)));          \
> 
>>>         {                                                                   \
>>>        asm volatile ("movq %q0,%%fs:%P1" :                                  \
>>> @@ -294,6 +299,12 @@ _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
>>>                    : IMM_MODE (value),                                      \
>>>                      "i" (offsetof (struct pthread, member[0])),            \
>>>                      "r" (idx));                                            \
>>> +     else if (__builtin_constant_p (value)                                 \
>>> +           && (int64_t) (int32_t) (uintptr_t) value != (uintptr_t) value)  \
>>> +       asm volatile ("movq %0,%%fs:%P1(,%q2,8)" :                          \
>>> +                  : "r" (value),                                           \
>>> +                    "i" (offsetof (struct pthread, member[0])),            \
>>> +                    "r" (idx));                                            \
>>>       else /* 8 */                                                          \
>>>         {                                                                   \
>>>        asm volatile ("movq %q0,%%fs:%P1(,%q2,8)" :                          \
>>>
>>
> 
> Here is the v2 patch.  OK for master?
> 
> Thanks.
> 

> From 4fbc9ab67933d662506a01658d4c81a922e50fdf Mon Sep 17 00:00:00 2001
> From: "H.J. Lu" <hjl.tools@gmail.com>
> Date: Fri, 8 Jan 2021 15:38:14 -0800
> Subject: [PATCH v2] x86_64: Update THREAD_SETMEM/THREAD_SETMEM_NC for IMM64
> 
> Since there is only "movq imm32s, mem64" and no "movq imm64, mem64", use
> "movq reg64, mem64" to store 64-bit constant.
> ---
>  sysdeps/x86_64/nptl/tls.h | 27 ++++++++++++++++++++-------
>  1 file changed, 20 insertions(+), 7 deletions(-)
> 
> diff --git a/sysdeps/x86_64/nptl/tls.h b/sysdeps/x86_64/nptl/tls.h
> index 20f0958780..0dbd66209c 100644
> --- a/sysdeps/x86_64/nptl/tls.h
> +++ b/sysdeps/x86_64/nptl/tls.h
> @@ -271,9 +271,15 @@ _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
>  		       "i" (offsetof (struct pthread, member)));	      \
>       else /* 8 */							      \
>         {								      \
> -	 asm volatile ("movq %q0,%%fs:%P1" :				      \
> -		       : IMM_MODE ((uint64_t) cast_to_integer (value)),	      \
> -			 "i" (offsetof (struct pthread, member)));	      \
> +	 if (!__builtin_constant_p (value)				      \

Whis is this a "!__builtin_constant_p?"

I would have expected:

if ([the value is a constant a therefore known quantity]
    || [it is a 32-bit value])

Perhaps a quick single line comment here explaining the conditionals would help?

> +	     || (int64_t) (int32_t) (uintptr_t) value == (uintptr_t) value)   \
> +	   asm volatile ("movq %q0,%%fs:%P1" :				      \
> +			 : IMM_MODE ((uint64_t) cast_to_integer (value)),     \
> +			   "i" (offsetof (struct pthread, member)));	      \
> +	 else								      \
> +	   asm volatile ("movq %0,%%fs:%P1" :				      \
> +			 : "r" (value),					      \
> +			   "i" (offsetof (struct pthread, member)));	      \
>         }})
>  
>  
> @@ -296,10 +302,17 @@ _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
>  		       "r" (idx));					      \
>       else /* 8 */							      \
>         {								      \
> -	 asm volatile ("movq %q0,%%fs:%P1(,%q2,8)" :			      \
> -		       : IMM_MODE ((uint64_t) cast_to_integer (value)),	      \
> -			 "i" (offsetof (struct pthread, member[0])),	      \
> -			 "r" (idx));					      \
> +	 if (!__builtin_constant_p (value)				      \
> +	     || (int64_t) (int32_t) (uintptr_t) value == (uintptr_t) value)   \
> +	   asm volatile ("movq %q0,%%fs:%P1(,%q2,8)" :			      \
> +			 : IMM_MODE ((uint64_t) cast_to_integer (value)),     \
> +			   "i" (offsetof (struct pthread, member[0])),	      \
> +			   "r" (idx));					      \
> +	 else								      \
> +	   asm volatile ("movq %0,%%fs:%P1(,%q2,8)" :			      \
> +			 : "r" (value),					      \
> +			   "i" (offsetof (struct pthread, member[0])),	      \
> +			   "r" (idx));					      \
>         }})
>  
>  
> -- 
> 2.29.2
>
  

Patch

From 4fbc9ab67933d662506a01658d4c81a922e50fdf Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Fri, 8 Jan 2021 15:38:14 -0800
Subject: [PATCH v2] x86_64: Update THREAD_SETMEM/THREAD_SETMEM_NC for IMM64

Since there is only "movq imm32s, mem64" and no "movq imm64, mem64", use
"movq reg64, mem64" to store 64-bit constant.
---
 sysdeps/x86_64/nptl/tls.h | 27 ++++++++++++++++++++-------
 1 file changed, 20 insertions(+), 7 deletions(-)

diff --git a/sysdeps/x86_64/nptl/tls.h b/sysdeps/x86_64/nptl/tls.h
index 20f0958780..0dbd66209c 100644
--- a/sysdeps/x86_64/nptl/tls.h
+++ b/sysdeps/x86_64/nptl/tls.h
@@ -271,9 +271,15 @@  _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
 		       "i" (offsetof (struct pthread, member)));	      \
      else /* 8 */							      \
        {								      \
-	 asm volatile ("movq %q0,%%fs:%P1" :				      \
-		       : IMM_MODE ((uint64_t) cast_to_integer (value)),	      \
-			 "i" (offsetof (struct pthread, member)));	      \
+	 if (!__builtin_constant_p (value)				      \
+	     || (int64_t) (int32_t) (uintptr_t) value == (uintptr_t) value)   \
+	   asm volatile ("movq %q0,%%fs:%P1" :				      \
+			 : IMM_MODE ((uint64_t) cast_to_integer (value)),     \
+			   "i" (offsetof (struct pthread, member)));	      \
+	 else								      \
+	   asm volatile ("movq %0,%%fs:%P1" :				      \
+			 : "r" (value),					      \
+			   "i" (offsetof (struct pthread, member)));	      \
        }})
 
 
@@ -296,10 +302,17 @@  _Static_assert (offsetof (tcbhead_t, __glibc_unused2) == 0x80,
 		       "r" (idx));					      \
      else /* 8 */							      \
        {								      \
-	 asm volatile ("movq %q0,%%fs:%P1(,%q2,8)" :			      \
-		       : IMM_MODE ((uint64_t) cast_to_integer (value)),	      \
-			 "i" (offsetof (struct pthread, member[0])),	      \
-			 "r" (idx));					      \
+	 if (!__builtin_constant_p (value)				      \
+	     || (int64_t) (int32_t) (uintptr_t) value == (uintptr_t) value)   \
+	   asm volatile ("movq %q0,%%fs:%P1(,%q2,8)" :			      \
+			 : IMM_MODE ((uint64_t) cast_to_integer (value)),     \
+			   "i" (offsetof (struct pthread, member[0])),	      \
+			   "r" (idx));					      \
+	 else								      \
+	   asm volatile ("movq %0,%%fs:%P1(,%q2,8)" :			      \
+			 : "r" (value),					      \
+			   "i" (offsetof (struct pthread, member[0])),	      \
+			   "r" (idx));					      \
        }})
 
 
-- 
2.29.2