diff mbox

[V3] x86/cpuid: AVX-512 Feature Detection

Message ID 1467199664-13676-1-git-send-email-luwei.kang@intel.com (mailing list archive)
State New, archived
Headers show

Commit Message

Luwei Kang June 29, 2016, 11:27 a.m. UTC
AVX-512 is an extention of AVX2. Its spec can be found at:
https://software.intel.com/sites/default/files/managed/b4/3a/319433-024.pdf
This patch detects AVX-512 features by CPUID.

Signed-off-by: Luwei Kang <luwei.kang@intel.com>
---
[V3]
1.adjust dependencies between features.
[V2]
1.one per bit, change from
    > + xstate_size = max(xstate_size,
	> +                 xstate_offsets[_XSTATE_HI_ZMM] +
	> +                 xstate_sizes[_XSTATE_HI_ZMM]);
  to
    xstate_size = max(xstate_size,
	                  xstate_offsets[_XSTATE_OPMASK] +
                      xstate_sizes[_XSTATE_OPMASK]);
    xstate_size = max(xstate_size,
                      xstate_offsets[_XSTATE_ZMM] +
                      xstate_sizes[_XSTATE_ZMM]);
    xstate_size = max(xstate_size,
                      xstate_offsets[_XSTATE_HI_ZMM] +
                      xstate_sizes[_XSTATE_HI_ZMM]);
2.change form
    domain_cpuid(currd, 0x07, 0, &tmp, &_ebx, &tmp, &tmp);
  to
    domain_cpuid(currd, 7, 0, &tmp, &_ebx, &tmp, &tmp);
3.add dependencies between features in xen/tools/gen-cpuid.py
4.split the cpuid call just like the way the hvm_cpuid() side works.

 xen/arch/x86/hvm/hvm.c                      | 14 ++++++++++++++
 xen/arch/x86/traps.c                        | 22 +++++++++++++++++++++-
 xen/include/public/arch-x86/cpufeatureset.h |  9 +++++++++
 xen/tools/gen-cpuid.py                      | 11 +++++++++++
 4 files changed, 55 insertions(+), 1 deletion(-)

Comments

Jan Beulich June 29, 2016, 2:29 p.m. UTC | #1
>>> On 29.06.16 at 13:27, <luwei.kang@intel.com> wrote:
> AVX-512 is an extention of AVX2. Its spec can be found at:
> https://software.intel.com/sites/default/files/managed/b4/3a/319433-024.pdf 
> This patch detects AVX-512 features by CPUID.
> 
> Signed-off-by: Luwei Kang <luwei.kang@intel.com>
> ---
> [V3]
> 1.adjust dependencies between features.
> [V2]
> 1.one per bit, change from
>     > + xstate_size = max(xstate_size,
> 	> +                 xstate_offsets[_XSTATE_HI_ZMM] +
> 	> +                 xstate_sizes[_XSTATE_HI_ZMM]);
>   to
>     xstate_size = max(xstate_size,
> 	                  xstate_offsets[_XSTATE_OPMASK] +
>                       xstate_sizes[_XSTATE_OPMASK]);
>     xstate_size = max(xstate_size,
>                       xstate_offsets[_XSTATE_ZMM] +
>                       xstate_sizes[_XSTATE_ZMM]);
>     xstate_size = max(xstate_size,
>                       xstate_offsets[_XSTATE_HI_ZMM] +
>                       xstate_sizes[_XSTATE_HI_ZMM]);
> 2.change form
>     domain_cpuid(currd, 0x07, 0, &tmp, &_ebx, &tmp, &tmp);
>   to
>     domain_cpuid(currd, 7, 0, &tmp, &_ebx, &tmp, &tmp);
> 3.add dependencies between features in xen/tools/gen-cpuid.py
> 4.split the cpuid call just like the way the hvm_cpuid() side works.

Especially item 1 is clearly to verbose; note how I said "brief" when
I asked for the revision log.

> --- a/xen/tools/gen-cpuid.py
> +++ b/xen/tools/gen-cpuid.py
> @@ -243,6 +243,17 @@ def crunch_numbers(state):
>          # AMD K6-2+ and K6-III processors shipped with 3DNow+, beyond the
>          # standard 3DNow in the earlier K6 processors.
>          _3DNOW: [_3DNOWEXT],
> +
> +        # AVX2 is an extension to AVX, providing mainly new integer instructions.
> +        # In principle, AVX512 only depends on YMM register state, but many AVX2
> +        # instructions are extended by AVX512F to 512-bit forms.

I realize you used the wording as suggested by Andrew, and while
his reply to my question about it meanwhile clarified what is meant,
I continue to think that the mentioning of YMM registers above is
misleading. May I suggest something like "AVX512 only takes YMM
register state as a prerequisite", subject to further improvement by
Andrew or another native speaker?

Apart from that the patch looks fine to me now.

Jan
diff mbox

Patch

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index c89ab6e..7696b1e 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -3474,6 +3474,20 @@  void hvm_cpuid(unsigned int input, unsigned int *eax, unsigned int *ebx,
                                   xstate_sizes[_XSTATE_BNDCSR]);
             }
 
+            if ( _ebx & cpufeat_mask(X86_FEATURE_AVX512F) )
+            {
+                xfeature_mask |= XSTATE_OPMASK | XSTATE_ZMM | XSTATE_HI_ZMM;
+                xstate_size = max(xstate_size,
+                                  xstate_offsets[_XSTATE_OPMASK] +
+                                  xstate_sizes[_XSTATE_OPMASK]);
+                xstate_size = max(xstate_size,
+                                  xstate_offsets[_XSTATE_ZMM] +
+                                  xstate_sizes[_XSTATE_ZMM]);
+                xstate_size = max(xstate_size,
+                                  xstate_offsets[_XSTATE_HI_ZMM] +
+                                  xstate_sizes[_XSTATE_HI_ZMM]);
+            }
+
             if ( _ecx & cpufeat_mask(X86_FEATURE_PKU) )
             {
                 xfeature_mask |= XSTATE_PKRU;
diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
index 767d0b0..8fb697b 100644
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -975,7 +975,7 @@  void pv_cpuid(struct cpu_user_regs *regs)
 
     switch ( leaf )
     {
-        uint32_t tmp, _ecx;
+        uint32_t tmp, _ecx, _ebx;
 
     case 0x00000001:
         c &= pv_featureset[FEATURESET_1c];
@@ -1157,6 +1157,26 @@  void pv_cpuid(struct cpu_user_regs *regs)
                                xstate_sizes[_XSTATE_YMM]);
             }
 
+            if ( !is_control_domain(currd) && !is_hardware_domain(currd) )
+                domain_cpuid(currd, 7, 0, &tmp, &_ebx, &tmp, &tmp);
+            else
+                cpuid_count(7, 0, &tmp, &_ebx, &tmp, &tmp);
+            _ebx &= pv_featureset[FEATURESET_7b0];
+
+            if ( _ebx & cpufeat_mask(X86_FEATURE_AVX512F) )
+            {
+                xfeature_mask |= XSTATE_OPMASK | XSTATE_ZMM | XSTATE_HI_ZMM;
+                xstate_size = max(xstate_size,
+                                  xstate_offsets[_XSTATE_OPMASK] +
+                                  xstate_sizes[_XSTATE_OPMASK]);
+                xstate_size = max(xstate_size,
+                                  xstate_offsets[_XSTATE_ZMM] +
+                                  xstate_sizes[_XSTATE_ZMM]);
+                xstate_size = max(xstate_size,
+                                  xstate_offsets[_XSTATE_HI_ZMM] +
+                                  xstate_sizes[_XSTATE_HI_ZMM]);
+            }
+
             a = (uint32_t)xfeature_mask;
             d = (uint32_t)(xfeature_mask >> 32);
             c = xstate_size;
diff --git a/xen/include/public/arch-x86/cpufeatureset.h b/xen/include/public/arch-x86/cpufeatureset.h
index 39acf8c..9320c9e 100644
--- a/xen/include/public/arch-x86/cpufeatureset.h
+++ b/xen/include/public/arch-x86/cpufeatureset.h
@@ -206,15 +206,24 @@  XEN_CPUFEATURE(PQM,           5*32+12) /*   Platform QoS Monitoring */
 XEN_CPUFEATURE(NO_FPU_SEL,    5*32+13) /*!  FPU CS/DS stored as zero */
 XEN_CPUFEATURE(MPX,           5*32+14) /*S  Memory Protection Extensions */
 XEN_CPUFEATURE(PQE,           5*32+15) /*   Platform QoS Enforcement */
+XEN_CPUFEATURE(AVX512F,       5*32+16) /*A  AVX-512 Foundation Instructions */
+XEN_CPUFEATURE(AVX512DQ,      5*32+17) /*A  AVX-512 Doubleword & Quadword Instrs */
 XEN_CPUFEATURE(RDSEED,        5*32+18) /*A  RDSEED instruction */
 XEN_CPUFEATURE(ADX,           5*32+19) /*A  ADCX, ADOX instructions */
 XEN_CPUFEATURE(SMAP,          5*32+20) /*S  Supervisor Mode Access Prevention */
+XEN_CPUFEATURE(AVX512IFMA,    5*32+21) /*A  AVX-512 Integer Fused Multiply Add */
 XEN_CPUFEATURE(CLFLUSHOPT,    5*32+23) /*A  CLFLUSHOPT instruction */
 XEN_CPUFEATURE(CLWB,          5*32+24) /*A  CLWB instruction */
+XEN_CPUFEATURE(AVX512PF,      5*32+26) /*A  AVX-512 Prefetch Instructions */
+XEN_CPUFEATURE(AVX512ER,      5*32+27) /*A  AVX-512 Exponent & Reciprocal Instrs */
+XEN_CPUFEATURE(AVX512CD,      5*32+28) /*A  AVX-512 Conflict Detection Instrs */
 XEN_CPUFEATURE(SHA,           5*32+29) /*A  SHA1 & SHA256 instructions */
+XEN_CPUFEATURE(AVX512BW,      5*32+30) /*A  AVX-512 Byte and Word Instructions */
+XEN_CPUFEATURE(AVX512VL,      5*32+31) /*A  AVX-512 Vector Length Extensions */
 
 /* Intel-defined CPU features, CPUID level 0x00000007:0.ecx, word 6 */
 XEN_CPUFEATURE(PREFETCHWT1,   6*32+ 0) /*A  PREFETCHWT1 instruction */
+XEN_CPUFEATURE(AVX512VBMI,    6*32+ 1) /*A  AVX-512 Vector Byte Manipulation Instrs */
 XEN_CPUFEATURE(PKU,           6*32+ 3) /*H  Protection Keys for Userspace */
 XEN_CPUFEATURE(OSPKE,         6*32+ 4) /*!  OS Protection Keys Enable */
 
diff --git a/xen/tools/gen-cpuid.py b/xen/tools/gen-cpuid.py
index 7c45eca..cf7bb67 100755
--- a/xen/tools/gen-cpuid.py
+++ b/xen/tools/gen-cpuid.py
@@ -243,6 +243,17 @@  def crunch_numbers(state):
         # AMD K6-2+ and K6-III processors shipped with 3DNow+, beyond the
         # standard 3DNow in the earlier K6 processors.
         _3DNOW: [_3DNOWEXT],
+
+        # AVX2 is an extension to AVX, providing mainly new integer instructions.
+        # In principle, AVX512 only depends on YMM register state, but many AVX2
+        # instructions are extended by AVX512F to 512-bit forms.
+        AVX2: [AVX512F],
+
+        # AVX512F is taken to mean hardware support for EVEX encoded instructions,
+        # 512bit registers, and the instructions themselves. All further AVX512 features
+        # are built on top of AVX512F
+        AVX512F: [AVX512DQ, AVX512IFMA, AVX512PF, AVX512ER, AVX512CD,
+                    AVX512BW, AVX512VL, AVX512VBMI],
     }
 
     deep_features = tuple(sorted(deps.keys()))