OcCpuLib: Implement ASM CPU frequency calculation for HSW and lower

Changelog.md

@@ -3,6 +3,7 @@ OpenCore Changelog
 #### v0.6.7
 - Fixed ocvalidate return code to be non-zero when issues are found
 - Added `OEM` values to `PlatformInfo` in `Automatic` mode
+- Improved CPU frequency calculation on Haswell and earlier
 
 #### v0.6.6
 - Added keyboard and pointer entry scroll support in OpenCanopy

Library/OcCpuLib/FrequencyDetect.c

@@ -129,15 +129,14 @@ InternalCalculateTSCFromPMTimer (
   //
   STATIC UINT64 TSCFrequency = 0;
 
-  UINTN       TimerAddr;
+  UINT16      TimerAddr;
   UINTN       VariableSize;
-  UINT64      Tsc0;
-  UINT64      Tsc1;
+  UINT64      TscTicksDelta;
   UINT32      AcpiTick0;
   UINT32      AcpiTick1;
   UINT32      AcpiTicksDelta;
-  UINT32      AcpiTicksTarget;
-  UINT32      TimerResolution;
+  UINT32      AcpiTicksDuration;
+  BOOLEAN     HasInterrupts;
   EFI_TPL     PrevTpl;
   EFI_STATUS  Status;
 
@@ -169,8 +168,7 @@ InternalCalculateTSCFromPMTimer (
   }
 
   if (TSCFrequency == 0) {
-    TimerAddr       = InternalGetPmTimerAddr (NULL);
-    TimerResolution = 10;
+    TimerAddr       = (UINT16) InternalGetPmTimerAddr (NULL);
 
     if (TimerAddr != 0) {
       //
@@ -187,63 +185,24 @@ InternalCalculateTSCFromPMTimer (
         // The code below can handle overflow with AcpiTicksTarget of up to 24-bit size,
         // on both available sizes of ACPI PM Timers (24-bit and 32-bit).
         //
-        // 357954 clocks of ACPI timer (100ms)
+        // 357954 clocks of ACPI timer (200ms)
         //
-        AcpiTicksTarget = V_ACPI_TMR_FREQUENCY / TimerResolution;
+        AcpiTicksDuration = V_ACPI_TMR_FREQUENCY / 10;
 
         //
         // Disable all events to ensure that nobody interrupts us.
         //
         PrevTpl   = gBS->RaiseTPL (TPL_HIGH_LEVEL);
+        HasInterrupts = SaveAndDisableInterrupts ();
+        AsmMeasureTicks (AcpiTicksDuration, TimerAddr, &AcpiTicksDelta, &TscTicksDelta);
+        if (HasInterrupts) {
+          EnableInterrupts ();
+        }
+        gBS->RestoreTPL (PrevTpl);
 
-        AcpiTick0 = IoRead32 (TimerAddr);
-        Tsc0      = AsmReadTsc ();
-
-        do {
-          CpuPause ();
-
-          //
-          // Check how many AcpiTicks have passed since we started.
-          //
-          AcpiTick1 = IoRead32 (TimerAddr);
-
-          if (AcpiTick0 <= AcpiTick1) {
-            //
-            // No overflow.
-            //
-            AcpiTicksDelta = AcpiTick1 - AcpiTick0;
-          } else if (AcpiTick0 - AcpiTick1 <= 0x00FFFFFF) {
-            //
-            // Overflow, 24-bit timer.
-            //
-            AcpiTicksDelta = 0x00FFFFFF - AcpiTick0 + AcpiTick1;
-          } else {
-            //
-            // Overflow, 32-bit timer.
-            //
-            AcpiTicksDelta = MAX_UINT32 - AcpiTick0 + AcpiTick1;
-          }
-
-          //
-          // Keep checking AcpiTicks until target is reached.
-          //
-        } while (AcpiTicksDelta < AcpiTicksTarget);
-
-        Tsc1 = AsmReadTsc ();
-
-        //
-        // On some systems we may end up waiting for notably longer than 100ms,
-        // despite disabling all events. Divide by actual time passed as suggested
-        // by asava's Clover patch r2668.
-        //
         TSCFrequency = DivU64x32 (
-          MultU64x32 (Tsc1 - Tsc0, V_ACPI_TMR_FREQUENCY), AcpiTicksDelta
+          MultU64x32 (TscTicksDelta, V_ACPI_TMR_FREQUENCY), AcpiTicksDelta
           );
-
-        //
-        // Restore to normal TPL.
-        //
-        gBS->RestoreTPL (PrevTpl);
       }
     }
 

Library/OcCpuLib/Ia32/MeasureTicks.c

+/** @file
+  Copyright (C) 2020, vit9696. All rights reserved.
+
+  All rights reserved.
+
+  This program and the accompanying materials
+  are licensed and made available under the terms and conditions of the BSD License
+  which accompanies this distribution.  The full text of the license may be found at
+  http://opensource.org/licenses/bsd-license.php
+
+  THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,
+  WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
+**/
+
+#include "OcCpuInternals.h"
+
+#include <Library/BaseLib.h>
+#include <Library/IoLib.h>
+
+VOID
+EFIAPI
+AsmMeasureTicks (
+  IN  UINT32  AcpiTicksDuration,
+  IN  UINT16  TimerAddr,
+  OUT UINT32  *AcpiTicksDelta,
+  OUT UINT64  *TscTicksDelta
+  )
+{
+  UINT32      AcpiTick0;
+  UINT32      AcpiTick1;
+  UINT32      AcpiCurrentDelta;
+  UINT64      Tsc0;
+  UINT64      Tsc1;
+
+  AcpiTick0 = IoRead32 (TimerAddr);
+  Tsc0      = AsmReadTsc ();
+
+  do {
+    CpuPause ();
+
+    //
+    // Check how many AcpiTicks have passed since we started.
+    //
+    AcpiTick1 = IoRead32 (TimerAddr);
+
+    if (AcpiTick0 <= AcpiTick1) {
+      //
+      // No overflow.
+      //
+      AcpiCurrentDelta = AcpiTick1 - AcpiTick0;
+    } else if (AcpiTick0 - AcpiTick1 <= 0x00FFFFFF) {
+      //
+      // Overflow, 24-bit timer.
+      //
+      AcpiCurrentDelta = 0x00FFFFFF - AcpiTick0 + AcpiTick1;
+    } else {
+      //
+      // Overflow, 32-bit timer.
+      //
+      AcpiCurrentDelta = MAX_UINT32 - AcpiTick0 + AcpiTick1;
+    }
+
+    //
+    // Keep checking AcpiTicks until target is reached.
+    //
+  } while (AcpiCurrentDelta < AcpiTicksDuration);
+
+  Tsc1 = AsmReadTsc ();
+
+  //
+  // On some systems we may end up waiting for notably longer than 100ms,
+  // despite disabling all events. Divide by actual time passed as suggested
+  // by asava's Clover patch r2668.
+  //
+  *TscTicksDelta = Tsc1 - Tsc0;
+  *AcpiTicksDelta = AcpiCurrentDelta;
+}

Library/OcCpuLib/OcCpuInternals.h

@@ -41,6 +41,23 @@ AsmReadIntelMicrocodeRevision (
   VOID
   );
 
+/**
+  Measures TSC and ACPI ticks over specified ACPI tick amount.
+
+  @param[in]  AcpiTicksDuration  Number of ACPI ticks for calculation.
+  @param[in]  TimerAddr          ACPI timer address.
+  @param[out] AcpiTicksDelta     Reported ACPI ticks delta.
+  @param[out] TscTicksDelta      Reported TSC ticks delta.
+**/
+VOID
+EFIAPI
+AsmMeasureTicks (
+  IN  UINT32  AcpiTicksDuration,
+  IN  UINT16  TimerAddr,
+  OUT UINT32  *AcpiTicksDelta,
+  OUT UINT64  *TscTicksDelta
+  );
+
 /**
   Detect Apple CPU major type.
 

Library/OcCpuLib/OcCpuLib.inf

@@ -52,8 +52,10 @@
 
 [Sources.Ia32]
   Ia32/Atomic.nasm
+  Ia32/MeasureTicks.c
   Ia32/Microcode.nasm
 
 [Sources.X64]
   X64/Atomic.nasm
+  X64/MeasureTicks.nasm
   X64/Microcode.nasm

Library/OcCpuLib/X64/MeasureTicks.nasm

+;------------------------------------------------------------------------------
+;  @file
+;  Copyright (C) 2020, vit9696. All rights reserved.
+;
+;  All rights reserved.
+;
+;  This program and the accompanying materials
+;  are licensed and made available under the terms and conditions of the BSD License
+;  which accompanies this distribution.  The full text of the license may be found at
+;  http://opensource.org/licenses/bsd-license.php
+;
+;  THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,
+;  WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
+;------------------------------------------------------------------------------
+
+BITS     64
+DEFAULT  REL
+
+SECTION  .text
+
+;------------------------------------------------------------------------------
+; VOID
+; EFIAPI
+; AsmMeasureTicks (
+;   IN  UINT32  AcpiTicksDuration,   ///< ecx
+;   IN  UINT16  TimerAddr,           ///< dx
+;   OUT UINT32  *AcpiTicksDelta,     ///< r8
+;   OUT UINT64  *TscTicksDelta       ///< r9
+;   );
+;------------------------------------------------------------------------------
+global ASM_PFX(AsmMeasureTicks)
+ASM_PFX(AsmMeasureTicks):
+  ; Push preserved registers.
+  push       rsi
+  push       rdi
+  push       rbp
+  push       rbx
+
+  ; Preserve TimerAddr in r11d as rdtsc overrides edx.
+  mov        r11d, edx
+
+  ; Read AcpiTicks0 into ebx.
+  in         eax, dx
+  mov        ebx, eax
+
+  ; Read Tsc0 into r10.
+  rdtsc
+  shl        rdx, 32
+  or         rax, rdx
+  mov        r10, rax
+
+  ; Store MAX_UINT32 - AcpiTick0 into ebp.
+  mov        ebp, ebx
+  not        ebp
+
+  ; Store MAX_UINT24 - AcpiTick0 into edi.
+  mov        edi, 0x00ffffff
+  sub        edi, ebx
+
+CalculationLoop:
+  nop ; or pause.
+
+  ; Read AcpiTicks1 into esi.
+  mov        edx, r11d
+  in         eax, dx
+  mov        esi, eax
+
+  ; Calculate AcpiTicks0 - AcpiTicks1 into eax.
+  mov        eax, ebx
+  sub        eax, esi
+  jbe        NoOverflow
+
+  ; Check for 32-bit overflow.
+  cmp        eax, 0x00ffffff
+  ja         Overflow32
+
+Overflow24:
+  ; AcpiCurrentDelta (esi) = AcpiTicks1 (esi) + MAX_UINT24 - AcpiTick0 (edi)
+  add        esi, edi
+  jmp        ContinueLoop
+
+NoOverflow:
+  ; AcpiCurrentDelta (esi) = AcpiTicks1 (esi) - AcpiTicks0 (ebx)
+  sub        esi, ebx
+  jmp        ContinueLoop
+
+Overflow32:
+  ; AcpiCurrentDelta (esi) = AcpiTicks1 (esi) + MAX_UINT32 - AcpiTick0 (ebp)
+  add        esi, ebp
+
+ContinueLoop:
+  cmp        esi, ecx
+  jb         CalculationLoop
+
+  ; Read Tsc1 into rax.
+  rdtsc
+  shl        rdx, 32
+  or         rax, rdx
+
+  ; Calculate TscDelta into rax.
+  sub        rax, r10
+
+  ; Store TscDelta and AcpiDelta.
+  mov        qword [r9], rax
+  mov        dword [r8], esi
+
+  ; Pop preserved registers & return.
+  pop        rbx
+  pop        rbp
+  pop        rdi
+  pop        rsi
+  ret