Refactor CPUID to make it thread safe

Needed for #1819 and unfortunately Windows does not allow thread local
data to be stored as a member of a DLL exported class. So hide it
behind an accessor function instead.

This slows down CPUID test somewhat and I would like to address that
but it seems hard without breaking the CPUID API, which is for better
or worse public.

6 files changed, 76 insertions, 48 deletions

diff --git a/src/lib/utils/compiler.h b/src/lib/utils/compiler.h
index fa8e1cfbf..9dfbd9fa2 100644
--- a/src/lib/utils/compiler.h
+++ b/src/lib/utils/compiler.h
@@ -131,6 +131,15 @@
 #endif
 
 /*
+* Define BOTAN_THREAD_LOCAL
+*/
+#if defined(BOTAN_TARGET_OS_HAS_THREADS)
+   #define BOTAN_THREAD_LOCAL thread_local
+#else
+   #define BOTAN_THREAD_LOCAL /**/
+#endif
+
+/*
 * Define BOTAN_IF_CONSTEXPR
 */
 
diff --git a/src/lib/utils/cpuid/cpuid.cpp b/src/lib/utils/cpuid/cpuid.cpp
index d2d08dd43..e890af061 100644
--- a/src/lib/utils/cpuid/cpuid.cpp
+++ b/src/lib/utils/cpuid/cpuid.cpp
@@ -13,9 +13,12 @@
 
 namespace Botan {
 
-uint64_t CPUID::g_processor_features = 0;
-size_t CPUID::g_cache_line_size = BOTAN_TARGET_CPU_DEFAULT_CACHE_LINE_SIZE;
-CPUID::Endian_status CPUID::g_endian_status = ENDIAN_UNKNOWN;
+//static
+CPUID::CPUID_Data& CPUID::state()
+   {
+   static BOTAN_THREAD_LOCAL CPUID::CPUID_Data g_cpuid;
+   return g_cpuid;
+   }
 
 bool CPUID::has_simd_32()
    {
@@ -90,36 +93,44 @@ void CPUID::print(std::ostream& o)
 //static
 void CPUID::initialize()
    {
-   g_processor_features = 0;
+   state() = CPUID_Data();
+   }
 
+CPUID::CPUID_Data::CPUID_Data()
+   {
 #if defined(BOTAN_TARGET_CPU_IS_PPC_FAMILY) || \
     defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY) || \
     defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
 
-   g_processor_features = CPUID::detect_cpu_features(&g_cache_line_size);
+   m_cache_line_size = 0;
+   m_processor_features = detect_cpu_features(&m_cache_line_size);
 
 #endif
 
-   g_endian_status = runtime_check_endian();
-   g_processor_features |= CPUID::CPUID_INITIALIZED_BIT;
+   m_processor_features |= CPUID::CPUID_INITIALIZED_BIT;
+
+   if(m_cache_line_size == 0)
+      m_cache_line_size = BOTAN_TARGET_CPU_DEFAULT_CACHE_LINE_SIZE;
+
+   m_endian_status = runtime_check_endian();
    }
 
 //static
-CPUID::Endian_status CPUID::runtime_check_endian()
+CPUID::Endian_Status CPUID::CPUID_Data::runtime_check_endian()
    {
    // Check runtime endian
    const uint32_t endian32 = 0x01234567;
    const uint8_t* e8 = reinterpret_cast<const uint8_t*>(&endian32);
 
-   Endian_status endian = ENDIAN_UNKNOWN;
+   CPUID::Endian_Status endian = CPUID::Endian_Status::Unknown;
 
    if(e8[0] == 0x01 && e8[1] == 0x23 && e8[2] == 0x45 && e8[3] == 0x67)
       {
-      endian = ENDIAN_BIG;
+      endian = CPUID::Endian_Status::Big;
       }
    else if(e8[0] == 0x67 && e8[1] == 0x45 && e8[2] == 0x23 && e8[3] == 0x01)
       {
-      endian = ENDIAN_LITTLE;
+      endian = CPUID::Endian_Status::Little;
       }
    else
       {
@@ -128,9 +139,9 @@ CPUID::Endian_status CPUID::runtime_check_endian()
 
    // If we were compiled with a known endian, verify it matches at runtime
 #if defined(BOTAN_TARGET_CPU_IS_LITTLE_ENDIAN)
-   BOTAN_ASSERT(endian == ENDIAN_LITTLE, "Build and runtime endian match");
+   BOTAN_ASSERT(endian == CPUID::Endian_Status::Little, "Build and runtime endian match");
 #elif defined(BOTAN_TARGET_CPU_IS_BIG_ENDIAN)
-   BOTAN_ASSERT(endian == ENDIAN_BIG, "Build and runtime endian match");
+   BOTAN_ASSERT(endian == CPUID::Endian_Status::Big, "Build and runtime endian match");
 #endif
 
    return endian;
diff --git a/src/lib/utils/cpuid/cpuid.h b/src/lib/utils/cpuid/cpuid.h
index 95f6d687b..46eef2095 100644
--- a/src/lib/utils/cpuid/cpuid.h
+++ b/src/lib/utils/cpuid/cpuid.h
@@ -65,21 +65,17 @@ class BOTAN_PUBLIC_API(2,1) CPUID final
       */
       static size_t cache_line_size()
          {
-         if(g_processor_features == 0)
-            {
-            initialize();
-            }
-         return g_cache_line_size;
+         return state().cache_line_size();
          }
 
       static bool is_little_endian()
          {
-         return endian_status() == ENDIAN_LITTLE;
+         return state().endian_status() == Endian_Status::Little;
          }
 
       static bool is_big_endian()
          {
-         return endian_status() == ENDIAN_BIG;
+         return state().endian_status() == Endian_Status::Big;
          }
 
       enum CPUID_bits : uint64_t {
@@ -309,8 +305,7 @@ class BOTAN_PUBLIC_API(2,1) CPUID final
       */
       static void clear_cpuid_bit(CPUID_bits bit)
          {
-         const uint64_t mask = ~(static_cast<uint64_t>(bit));
-         g_processor_features &= mask;
+         state().clear_cpuid_bit(static_cast<uint64_t>(bit));
          }
 
       /*
@@ -319,43 +314,56 @@ class BOTAN_PUBLIC_API(2,1) CPUID final
       */
       static bool has_cpuid_bit(CPUID_bits elem)
          {
-         if(g_processor_features == 0)
-            initialize();
-
          const uint64_t elem64 = static_cast<uint64_t>(elem);
-         return ((g_processor_features & elem64) == elem64);
+         return state().has_bit(elem64);
          }
 
       static std::vector<CPUID::CPUID_bits> bit_from_string(const std::string& tok);
    private:
-      enum Endian_status : uint32_t {
-         ENDIAN_UNKNOWN = 0x00000000,
-         ENDIAN_BIG     = 0x01234567,
-         ENDIAN_LITTLE  = 0x67452301,
+      enum class Endian_Status : uint32_t {
+         Unknown = 0x00000000,
+         Big     = 0x01234567,
+         Little  = 0x67452301,
       };
 
+      struct CPUID_Data
+         {
+         public:
+            CPUID_Data();
+
+            CPUID_Data(const CPUID_Data& other) = default;
+            CPUID_Data& operator=(const CPUID_Data& other) = default;
+
+            void clear_cpuid_bit(uint64_t bit)
+               {
+               m_processor_features &= ~bit;
+               }
+
+            bool has_bit(uint64_t bit) const
+               {
+               return (m_processor_features & bit) == bit;
+               }
+
+            uint64_t processor_features() const { return m_processor_features; }
+            Endian_Status endian_status() const { return m_endian_status; }
+            size_t cache_line_size() const { return m_cache_line_size; }
+
+         private:
+            static Endian_Status runtime_check_endian();
+
 #if defined(BOTAN_TARGET_CPU_IS_PPC_FAMILY) || \
     defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY) || \
     defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
 
-      static uint64_t detect_cpu_features(size_t* cache_line_size);
+            static uint64_t detect_cpu_features(size_t* cache_line_size);
 
 #endif
+            uint64_t m_processor_features;
+            size_t m_cache_line_size;
+            Endian_Status m_endian_status;
+         };
 
-      static Endian_status runtime_check_endian();
-
-      static Endian_status endian_status()
-         {
-         if(g_endian_status == ENDIAN_UNKNOWN)
-            {
-            g_endian_status = runtime_check_endian();
-            }
-         return g_endian_status;
-         }
-
-      static uint64_t g_processor_features;
-      static size_t g_cache_line_size;
-      static Endian_status g_endian_status;
+      static CPUID_Data& state();
    };
 
 }
diff --git a/src/lib/utils/cpuid/cpuid_arm.cpp b/src/lib/utils/cpuid/cpuid_arm.cpp
index beb80f7d0..b5626f96b 100644
--- a/src/lib/utils/cpuid/cpuid_arm.cpp
+++ b/src/lib/utils/cpuid/cpuid_arm.cpp
@@ -102,7 +102,7 @@ uint64_t flags_by_ios_machine_type(const std::string& machine)
 
 #endif
 
-uint64_t CPUID::detect_cpu_features(size_t* cache_line_size)
+uint64_t CPUID::CPUID_Data::detect_cpu_features(size_t* cache_line_size)
    {
    uint64_t detected_features = 0;
 
diff --git a/src/lib/utils/cpuid/cpuid_ppc.cpp b/src/lib/utils/cpuid/cpuid_ppc.cpp
index 80a6241df..3afca539c 100644
--- a/src/lib/utils/cpuid/cpuid_ppc.cpp
+++ b/src/lib/utils/cpuid/cpuid_ppc.cpp
@@ -33,7 +33,7 @@ namespace Botan {
 * PowerPC specific block: check for AltiVec using either
 * sysctl or by reading processor version number register.
 */
-uint64_t CPUID::detect_cpu_features(size_t* cache_line_size)
+uint64_t CPUID::CPUID_Data::detect_cpu_features(size_t* cache_line_size)
    {
    BOTAN_UNUSED(cache_line_size);
 
diff --git a/src/lib/utils/cpuid/cpuid_x86.cpp b/src/lib/utils/cpuid/cpuid_x86.cpp
index 5387a801e..97c17da86 100644
--- a/src/lib/utils/cpuid/cpuid_x86.cpp
+++ b/src/lib/utils/cpuid/cpuid_x86.cpp
@@ -25,7 +25,7 @@ namespace Botan {
 
 #if defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)
 
-uint64_t CPUID::detect_cpu_features(size_t* cache_line_size)
+uint64_t CPUID::CPUID_Data::detect_cpu_features(size_t* cache_line_size)
    {
 #if defined(BOTAN_BUILD_COMPILER_IS_MSVC)
   #define X86_CPUID(type, out) do { __cpuid((int*)out, type); } while(0)