From b1e6a85443ccaf47c6a2c11b18b090bb0f97eea0 Mon Sep 17 00:00:00 2001
From: Jesse Selover <jselover@google.com>
Date: Thu, 19 Apr 2018 16:34:35 -0400
Subject: [PATCH] Change OPENSSL_cpuid_setup to reserve more extended feature
 space.

Copy of openssl change https://git.openssl.org/gitweb/?p=openssl.git;h=d6ee8f3dc4414cd97bd63b801f8644f0ff8a1f17

OPENSSL_ia32cap: reserve for new extensions.
Change-Id: I96b43c82ba6568bae848449972d3ad9d20f6d063
Reviewed-on: https://boringssl-review.googlesource.com/27564
Reviewed-by: David Benjamin <davidben@google.com>
Reviewed-by: Steven Valdez <svaldez@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
---
 crypto/cpu-intel.c    | 15 ++++++++-------
 include/openssl/cpu.h |  3 ++-
 2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/crypto/cpu-intel.c b/crypto/cpu-intel.c
index 1ac280c8..701ebedb 100644
--- a/crypto/cpu-intel.c
+++ b/crypto/cpu-intel.c
@@ -170,10 +170,11 @@ void OPENSSL_cpuid_setup(void) {
     }
   }
 
-  uint32_t extended_features = 0;
+  uint32_t extended_features[2] = {0};
   if (num_ids >= 7) {
     OPENSSL_cpuid(&eax, &ebx, &ecx, &edx, 7);
-    extended_features = ebx;
+    extended_features[0] = ebx;
+    extended_features[1] = ecx;
   }
 
   // Determine the number of cores sharing an L1 data cache to adjust the
@@ -241,26 +242,26 @@ void OPENSSL_cpuid_setup(void) {
     //
     // TODO(davidben): Should bits 17 and 26-28 also be cleared? Upstream
     // doesn't clear those.
-    extended_features &=
+    extended_features[0] &=
         ~((1u << 5) | (1u << 16) | (1u << 21) | (1u << 30) | (1u << 31));
   }
   // See Intel manual, volume 1, section 15.2.
   if ((xcr0 & 0xe6) != 0xe6) {
     // Clear AVX512F. Note we don't touch other AVX512 extensions because they
     // can be used with YMM.
-    extended_features &= ~(1u << 16);
+    extended_features[0] &= ~(1u << 16);
   }
 
   // Disable ADX instructions on Knights Landing. See OpenSSL commit
   // 64d92d74985ebb3d0be58a9718f9e080a14a8e7f.
   if ((ecx & (1u << 26)) == 0) {
-    extended_features &= ~(1u << 19);
+    extended_features[0] &= ~(1u << 19);
   }
 
   OPENSSL_ia32cap_P[0] = edx;
   OPENSSL_ia32cap_P[1] = ecx;
-  OPENSSL_ia32cap_P[2] = extended_features;
-  OPENSSL_ia32cap_P[3] = 0;
+  OPENSSL_ia32cap_P[2] = extended_features[0];
+  OPENSSL_ia32cap_P[3] = extended_features[1];
 
   const char *env1, *env2;
   env1 = getenv("OPENSSL_ia32cap");
diff --git a/include/openssl/cpu.h b/include/openssl/cpu.h
index dd95ddc6..bb847f96 100644
--- a/include/openssl/cpu.h
+++ b/include/openssl/cpu.h
@@ -86,7 +86,8 @@ extern "C" {
 //     Bit 11 is used to indicate AMD XOP support, not SDBG
 //   Index 2:
 //     EBX for CPUID where EAX = 7
-//   Index 3 is set to zero.
+//   Index 3:
+//     ECX for CPUID where EAX = 7
 //
 // Note: the CPUID bits are pre-adjusted for the OSXSAVE bit and the YMM and XMM
 // bits in XCR0, so it is not necessary to check those.