CloverBootloader/Library/OcCpuLib/OcCpuLib.c
SergeySlice cd23181296 move OcQuirks.plist into config.plist and into GUI menu
Signed-off-by: SergeySlice <sergey.slice@gmail.com>
2020-07-15 20:29:27 +03:00

1053 lines
30 KiB
C
Executable File

/** @file
Copyright (C) 2016 - 2017, The HermitCrabs Lab. 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 <PiDxe.h>
#include <IndustryStandard/AppleSmBios.h>
#include <Protocol/FrameworkMpService.h>
#include <Protocol/MpService.h>
#include <Library/BaseLib.h>
#include <Library/BaseMemoryLib.h>
#include <Library/DebugLib.h>
#include <Library/OcCpuLib.h>
#include <Library/UefiBootServicesTableLib.h>
#include <IndustryStandard/ProcessorInfo.h>
#include <Register/Microcode.h>
#include <Register/Msr.h>
#include <Register/Intel/Msr/SandyBridgeMsr.h>
#include <Register/Intel/Msr/NehalemMsr.h>
#include "OcCpuInternals.h"
STATIC
EFI_STATUS
ScanMpServices (
IN EFI_MP_SERVICES_PROTOCOL *MpServices,
OUT OC_CPU_INFO *Cpu,
OUT UINTN *NumberOfProcessors,
OUT UINTN *NumberOfEnabledProcessors
)
{
EFI_STATUS Status;
UINTN Index;
EFI_PROCESSOR_INFORMATION Info;
ASSERT (MpServices != NULL);
ASSERT (Cpu != NULL);
ASSERT (NumberOfProcessors != NULL);
ASSERT (NumberOfEnabledProcessors != NULL);
Status = MpServices->GetNumberOfProcessors (
MpServices,
NumberOfProcessors,
NumberOfEnabledProcessors
);
if (EFI_ERROR(Status)) {
return Status;
}
if (*NumberOfProcessors == 0) {
return EFI_NOT_FOUND;
}
//
// This code assumes that all CPUs have same amount of cores and threads.
//
for (Index = 0; Index < *NumberOfProcessors; ++Index) {
Status = MpServices->GetProcessorInfo (MpServices, Index, &Info);
if (EFI_ERROR(Status)) {
DEBUG ((
DEBUG_INFO,
"OCCPU: Failed to get info for processor %Lu - %r\n",
(UINT64) Index,
Status
));
continue;
}
if (Info.Location.Package + 1 >= Cpu->PackageCount) {
Cpu->PackageCount = (UINT16) (Info.Location.Package + 1);
}
if (Info.Location.Core + 1 >= Cpu->CoreCount) {
Cpu->CoreCount = (UINT16) (Info.Location.Core + 1);
}
if (Info.Location.Thread + 1 >= Cpu->ThreadCount) {
Cpu->ThreadCount = (UINT16) (Info.Location.Thread + 1);
}
}
return Status;
}
STATIC
EFI_STATUS
ScanFrameworkMpServices (
IN FRAMEWORK_EFI_MP_SERVICES_PROTOCOL *FrameworkMpServices,
OUT OC_CPU_INFO *Cpu,
OUT UINTN *NumberOfProcessors,
OUT UINTN *NumberOfEnabledProcessors
)
{
EFI_STATUS Status;
UINTN Index;
EFI_MP_PROC_CONTEXT Context;
UINTN ContextSize;
ASSERT (FrameworkMpServices != NULL);
ASSERT (Cpu != NULL);
ASSERT (NumberOfProcessors != NULL);
ASSERT (NumberOfEnabledProcessors != NULL);
Status = FrameworkMpServices->GetGeneralMPInfo (
FrameworkMpServices,
NumberOfProcessors,
NULL,
NumberOfEnabledProcessors,
NULL,
NULL
);
if (EFI_ERROR(Status)) {
return Status;
}
if (*NumberOfProcessors == 0) {
return EFI_NOT_FOUND;
}
//
// This code assumes that all CPUs have same amount of cores and threads.
//
for (Index = 0; Index < *NumberOfProcessors; ++Index) {
ContextSize = sizeof (Context);
Status = FrameworkMpServices->GetProcessorContext (
FrameworkMpServices,
Index,
&ContextSize,
&Context
);
if (EFI_ERROR(Status)) {
DEBUG ((
DEBUG_INFO,
"OCCPU: Failed to get context for processor %Lu - %r\n",
(UINT64) Index,
Status
));
continue;
}
if (Context.PackageNumber + 1 >= Cpu->PackageCount) {
Cpu->PackageCount = (UINT16) (Context.PackageNumber + 1);
}
//
// According to the FrameworkMpServices header, EFI_MP_PROC_CONTEXT.NumberOfCores is the
// zero-indexed physical core number for the current processor. However, Apple appears to
// set this to the total number of physical cores (observed on Xserve3,1 with 2x Xeon X5550).
//
// This number may not be accurate; on MacPro5,1 with 2x Xeon X5690, NumberOfCores is 16 when
// it should be 12 (even though NumberOfProcessors is correct). Regardless, CoreCount and
// ThreadCount will be corrected in ScanIntelProcessor.
//
// We will follow Apple's implementation, as the FrameworkMpServices fallback was added for
// legacy Macs.
//
if (Context.NumberOfCores >= Cpu->CoreCount) {
Cpu->CoreCount = (UINT16) (Context.NumberOfCores);
}
//
// Similarly, EFI_MP_PROC_CONTEXT.NumberOfThreads is supposed to be the zero-indexed logical
// thread number for the current processor. On Xserve3,1 and MacPro5,1 this was set to 2
// (presumably to indicate that there are 2 threads per physical core).
//
if (Context.NumberOfCores * Context.NumberOfThreads >= Cpu->ThreadCount) {
Cpu->ThreadCount = (UINT16) (Context.NumberOfCores * Context.NumberOfThreads);
}
}
return Status;
}
STATIC
EFI_STATUS
ScanThreadCount (
OUT OC_CPU_INFO *Cpu
)
{
EFI_STATUS Status;
EFI_MP_SERVICES_PROTOCOL *MpServices;
FRAMEWORK_EFI_MP_SERVICES_PROTOCOL *FrameworkMpServices;
UINTN NumberOfProcessors;
UINTN NumberOfEnabledProcessors;
Cpu->PackageCount = 1;
Cpu->CoreCount = 1;
Cpu->ThreadCount = 1;
NumberOfProcessors = 0;
NumberOfEnabledProcessors = 0;
Status = gBS->LocateProtocol (
&gEfiMpServiceProtocolGuid,
NULL,
(VOID **) &MpServices
);
if (EFI_ERROR(Status)) {
Status = gBS->LocateProtocol (
&gFrameworkEfiMpServiceProtocolGuid,
NULL,
(VOID **) &FrameworkMpServices
);
if (EFI_ERROR(Status)) {
DEBUG ((DEBUG_INFO, "OCCPU: No MP services - %r\n", Status));
return Status;
}
Status = ScanFrameworkMpServices (
FrameworkMpServices,
Cpu,
&NumberOfProcessors,
&NumberOfEnabledProcessors
);
} else {
Status = ScanMpServices (
MpServices,
Cpu,
&NumberOfProcessors,
&NumberOfEnabledProcessors
);
}
DEBUG ((
DEBUG_INFO,
"OCCPU: MP services threads %Lu (enabled %Lu) - %r\n",
(UINT64) NumberOfProcessors,
(UINT64) NumberOfEnabledProcessors,
Status
));
if (EFI_ERROR(Status)) {
return Status;
}
DEBUG ((
DEBUG_INFO,
"OCCPU: MP services Pkg %u Cores %u Threads %u - %r\n",
Cpu->PackageCount,
Cpu->CoreCount,
Cpu->ThreadCount,
Status
));
//
// Several implementations may not report virtual threads.
//
if (Cpu->ThreadCount < Cpu->CoreCount) {
Cpu->ThreadCount = Cpu->CoreCount;
}
return Status;
}
STATIC
VOID
ScanIntelProcessor (
IN OUT OC_CPU_INFO *Cpu
)
{
UINT64 Msr;
CPUID_CACHE_PARAMS_EAX CpuidCacheEax;
CPUID_CACHE_PARAMS_EBX CpuidCacheEbx;
UINT8 AppleMajorType;
MSR_SANDY_BRIDGE_PKG_CST_CONFIG_CONTROL_REGISTER PkgCstConfigControl;
MSR_IA32_PERF_STATUS_REGISTER PerfStatus;
MSR_NEHALEM_PLATFORM_INFO_REGISTER PlatformInfo;
OC_CPU_GENERATION CpuGeneration;
MSR_NEHALEM_TURBO_RATIO_LIMIT_REGISTER TurboLimit;
UINT16 CoreCount;
CONST CHAR8 *TimerSourceType;
UINTN TimerAddr;
BOOLEAN Recalculate;
AppleMajorType = InternalDetectAppleMajorType (Cpu->BrandString);
Cpu->AppleProcessorType = InternalDetectAppleProcessorType (Cpu->Model, Cpu->Stepping, AppleMajorType);
DEBUG ((DEBUG_INFO, "OCCPU: Detected Apple Processor Type: %02X -> %04X\n", AppleMajorType, Cpu->AppleProcessorType));
if ((Cpu->Family != 0x06 || Cpu->Model < 0x0c)
&& (Cpu->Family != 0x0f || Cpu->Model < 0x03)) {
return;
}
//
// Some virtual machines like QEMU 5.0 with KVM will fail to read this value.
// REF: https://github.com/acidanthera/bugtracker/issues/914
//
if (Cpu->Model >= CPU_MODEL_SANDYBRIDGE && !Cpu->Hypervisor) {
PkgCstConfigControl.Uint64 = AsmReadMsr64 (MSR_SANDY_BRIDGE_PKG_CST_CONFIG_CONTROL);
Cpu->CstConfigLock = PkgCstConfigControl.Bits.CFGLock == 1;
} else {
Cpu->CstConfigLock = FALSE;
}
DEBUG ((DEBUG_INFO, "OCCPU: EIST CFG Lock %d\n", Cpu->CstConfigLock));
//
// When the CPU is virtualized and cpuid invtsc is enabled, then we already get
// the information we want outside the function, skip anyway.
// Things may be different in other hypervisors, but should work with QEMU/VMWare for now.
//
if (Cpu->CPUFrequencyFromVMT == 0) {
//
// TODO: this may not be accurate on some older processors.
//
if (Cpu->Model >= CPU_MODEL_NEHALEM) {
PerfStatus.Uint64 = AsmReadMsr64 (MSR_IA32_PERF_STATUS);
PlatformInfo.Uint64 = AsmReadMsr64 (MSR_NEHALEM_PLATFORM_INFO);
Cpu->MinBusRatio = (UINT8) PlatformInfo.Bits.MaximumEfficiencyRatio;
Cpu->MaxBusRatio = (UINT8) PlatformInfo.Bits.MaximumNonTurboRatio;
CpuGeneration = OcCpuGetGeneration ();
if (CpuGeneration == OcCpuGenerationNehalem
|| CpuGeneration == OcCpuGenerationWestmere) {
Cpu->CurBusRatio = (UINT8) PerfStatus.Bits.State;
} else {
Cpu->CurBusRatio = (UINT8) (PerfStatus.Bits.State >> 8U);
}
} else {
PerfStatus.Uint64 = AsmReadMsr64 (MSR_IA32_PERF_STATUS);
Cpu->MaxBusRatio = (UINT8) (RShiftU64 (PerfStatus.Uint64, 8) & 0x1FU);
//
// Undocumented values:
// Non-integer bus ratio for the max-multi.
// Non-integer bus ratio for the current-multi.
//
// MaxBusRatioDiv = (UINT8) (RShiftU64 (PerfStatus.Uint64, 46) & 0x01U);
// CurrDiv = (UINT8) (RShiftU64 (PerfStatus.Uint64, 14) & 0x01U);
//
}
if (Cpu->Model >= CPU_MODEL_NEHALEM
&& Cpu->Model != CPU_MODEL_NEHALEM_EX
&& Cpu->Model != CPU_MODEL_WESTMERE_EX) {
TurboLimit.Uint64 = AsmReadMsr64 (MSR_NEHALEM_TURBO_RATIO_LIMIT);
Cpu->TurboBusRatio1 = (UINT8) TurboLimit.Bits.Maximum1C;
Cpu->TurboBusRatio2 = (UINT8) TurboLimit.Bits.Maximum2C;
Cpu->TurboBusRatio3 = (UINT8) TurboLimit.Bits.Maximum3C;
Cpu->TurboBusRatio4 = (UINT8) TurboLimit.Bits.Maximum4C;
}
DEBUG ((
DEBUG_INFO,
"OCCPU: Ratio Min %d Max %d Current %d Turbo %d %d %d %d\n",
Cpu->MinBusRatio,
Cpu->MaxBusRatio,
Cpu->CurBusRatio,
Cpu->TurboBusRatio1,
Cpu->TurboBusRatio2,
Cpu->TurboBusRatio3,
Cpu->TurboBusRatio4
));
//
// For logging purposes (the first call to these functions might happen
// before logging is fully initialised), do not use the cached results in
// DEBUG builds.
//
Recalculate = FALSE;
DEBUG_CODE_BEGIN ();
Recalculate = TRUE;
DEBUG_CODE_END ();
//
// Determine our core crystal clock frequency
//
Cpu->ARTFrequency = InternalCalculateARTFrequencyIntel (
&Cpu->CPUFrequencyFromART,
&Cpu->TscAdjust,
Recalculate
);
//
// Calculate the TSC frequency only if ART frequency is not available or we are in debug builds.
//
if (Cpu->CPUFrequencyFromART == 0 || Recalculate) {
DEBUG_CODE_BEGIN ();
TimerAddr = InternalGetPmTimerAddr (&TimerSourceType);
DEBUG ((DEBUG_INFO, "OCCPU: Timer address is %Lx from %a\n", (UINT64) TimerAddr, TimerSourceType));
DEBUG_CODE_END ();
Cpu->CPUFrequencyFromTSC = InternalCalculateTSCFromPMTimer (Recalculate);
}
//
// Calculate CPU frequency based on ART if present, otherwise TSC
//
Cpu->CPUFrequency = Cpu->CPUFrequencyFromART != 0 ? Cpu->CPUFrequencyFromART : Cpu->CPUFrequencyFromTSC;
//
// Verify that our two CPU frequency calculations do not differ substantially.
//
if (Cpu->CPUFrequencyFromART > 0 && Cpu->CPUFrequencyFromTSC > 0
&& ABS((INT64) Cpu->CPUFrequencyFromART - (INT64) Cpu->CPUFrequencyFromTSC) > OC_CPU_FREQUENCY_TOLERANCE) {
DEBUG ((
DEBUG_WARN,
"OCCPU: ART based CPU frequency differs substantially from TSC: %11LuHz != %11LuHz\n",
Cpu->CPUFrequencyFromART,
Cpu->CPUFrequencyFromTSC
));
}
//
// There may be some quirks with virtual CPUs (VMware is fine).
// Formerly we checked Cpu->MinBusRatio > 0, but we have no MinBusRatio on Penryn.
//
if (Cpu->CPUFrequency > 0 && Cpu->MaxBusRatio > Cpu->MinBusRatio) {
Cpu->FSBFrequency = DivU64x32 (Cpu->CPUFrequency, Cpu->MaxBusRatio);
} else {
//
// TODO: It seems to be possible that CPU frequency == 0 here...
//
Cpu->FSBFrequency = 100000000; // 100 Mhz
}
}
//
// Calculate number of cores
// If we are under virtualization, then we should get the topology from CPUID the same was as with Penryn.
//
if (Cpu->MaxId >= CPUID_CACHE_PARAMS && (Cpu->Model <= CPU_MODEL_PENRYN || Cpu->Hypervisor)) {
AsmCpuidEx (CPUID_CACHE_PARAMS, 0, &CpuidCacheEax.Uint32, &CpuidCacheEbx.Uint32, NULL, NULL);
if (CpuidCacheEax.Bits.CacheType != CPUID_CACHE_PARAMS_CACHE_TYPE_NULL) {
CoreCount = (UINT16)GetPowerOfTwo32 (CpuidCacheEax.Bits.MaximumAddressableIdsForProcessorCores + 1);
if (CoreCount < CpuidCacheEax.Bits.MaximumAddressableIdsForProcessorCores + 1) {
CoreCount *= 2;
}
Cpu->CoreCount = CoreCount;
//
// We should not be blindly relying on Cpu->Features & CPUID_FEATURE_HTT.
// On Penryn CPUs it is set even without Hyper Threading.
//
if (Cpu->ThreadCount < Cpu->CoreCount) {
Cpu->ThreadCount = Cpu->CoreCount;
}
}
} else if (Cpu->Model == CPU_MODEL_WESTMERE) {
Msr = AsmReadMsr64 (MSR_CORE_THREAD_COUNT);
Cpu->CoreCount = (UINT16)BitFieldRead64 (Msr, 16, 19);
Cpu->ThreadCount = (UINT16)BitFieldRead64 (Msr, 0, 15);
} else {
Msr = AsmReadMsr64 (MSR_CORE_THREAD_COUNT);
Cpu->CoreCount = (UINT16)BitFieldRead64 (Msr, 16, 31);
Cpu->ThreadCount = (UINT16)BitFieldRead64 (Msr, 0, 15);
}
if (Cpu->CoreCount == 0) {
Cpu->CoreCount = 1;
}
if (Cpu->ThreadCount == 0) {
Cpu->ThreadCount = 1;
}
}
STATIC
VOID
ScanAmdProcessor (
IN OUT OC_CPU_INFO *Cpu
)
{
UINT32 CpuidEbx;
UINT32 CpuidEcx;
UINT64 CofVid;
UINT8 CoreFrequencyID;
UINT8 CoreDivisorID;
UINT8 Divisor;
BOOLEAN Recalculate;
//
// For logging purposes (the first call to these functions might happen
// before logging is fully initialised), do not use the cached results in
// DEBUG builds.
//
Recalculate = FALSE;
DEBUG_CODE_BEGIN ();
Recalculate = TRUE;
DEBUG_CODE_END ();
//
// Faking an Intel Core i5 Processor.
// This value is purely cosmetic, but it makes sense to fake something
// that is somewhat representative of the kind of Processor that's actually
// in the system
//
Cpu->AppleProcessorType = AppleProcessorTypeCorei5Type5;
//
// get TSC Frequency calculated in OcTimerLib, unless we got it already from virtualization extensions.
// FIXME(1): This code assumes the CPU operates in P0. Either ensure it does
// and raise the mode on demand, or adapt the logic to consider
// both the operating and the nominal frequency, latter for
// the invariant TSC.
//
if (Cpu->CPUFrequencyFromVMT == 0) {
Cpu->CPUFrequencyFromTSC = InternalCalculateTSCFromPMTimer (Recalculate);
Cpu->CPUFrequency = Cpu->CPUFrequencyFromTSC;
}
//
// Get core and thread count from CPUID
//
if (Cpu->MaxExtId >= 0x80000008) {
AsmCpuid (0x80000008, NULL, NULL, &CpuidEcx, NULL);
Cpu->ThreadCount = (UINT16) (BitFieldRead32 (CpuidEcx, 0, 7) + 1);
}
if (Cpu->Family == AMD_CPU_FAMILY) {
Divisor = 0;
CoreFrequencyID = 0;
CoreDivisorID = 0;
switch (Cpu->ExtFamily) {
case AMD_CPU_EXT_FAMILY_17H:
if (Cpu->CPUFrequencyFromVMT == 0) {
CofVid = AsmReadMsr64 (K10_PSTATE_STATUS);
CoreFrequencyID = (UINT8)BitFieldRead64 (CofVid, 0, 7);
CoreDivisorID = (UINT8)BitFieldRead64 (CofVid, 8, 13);
if (CoreDivisorID > 0ULL) {
//
// Sometimes incorrect hypervisor configuration will lead to dividing by zero,
// but these variables will not be used under hypervisor, so just skip these.
//
Cpu->MaxBusRatio = (UINT8) (CoreFrequencyID / CoreDivisorID * 2);
}
}
//
// Get core count from CPUID
//
if (Cpu->MaxExtId >= 0x8000001E) {
AsmCpuid (0x8000001E, NULL, &CpuidEbx, NULL, NULL);
Cpu->CoreCount =
(UINT16) DivU64x32 (
Cpu->ThreadCount,
(BitFieldRead32 (CpuidEbx, 8, 15) + 1)
);
}
break;
case AMD_CPU_EXT_FAMILY_15H:
case AMD_CPU_EXT_FAMILY_16H:
if (Cpu->CPUFrequencyFromVMT == 0) {
// FIXME: Please refer to FIXME(1) for the MSR used here.
CofVid = AsmReadMsr64 (K10_COFVID_STATUS);
CoreFrequencyID = (UINT8)BitFieldRead64 (CofVid, 0, 5);
CoreDivisorID = (UINT8)BitFieldRead64 (CofVid, 6, 8);
Divisor = 1U << CoreDivisorID;
//
// BKDG for AMD Family 15h Models 10h-1Fh Processors (42300 Rev 3.12)
// Core current operating frequency in MHz. CoreCOF = 100 *
// (MSRC001_00[6B:64][CpuFid] + 10h) / (2 ^ MSRC001_00[6B:64][CpuDid]).
//
if (Divisor > 0ULL) {
//
// Sometimes incorrect hypervisor configuration will lead to dividing by zero,
// but these variables will not be used under hypervisor, so just skip these.
//
Cpu->MaxBusRatio = (UINT8)((CoreFrequencyID + 0x10) / Divisor);
}
}
//
// AMD 15h and 16h CPUs don't support hyperthreading,
// so the core count is equal to the thread count
//
Cpu->CoreCount = Cpu->ThreadCount;
break;
default:
return;
}
DEBUG ((
DEBUG_INFO,
"OCCPU: FID %u DID %u Divisor %u MaxBR %u\n",
CoreFrequencyID,
CoreDivisorID,
Divisor,
Cpu->MaxBusRatio
));
//
// When under virtualization this information is already available to us.
//
if (Cpu->CPUFrequencyFromVMT == 0) {
//
// Sometimes incorrect hypervisor configuration will lead to dividing by zero.
//
if (Cpu->MaxBusRatio == 0) {
Cpu->FSBFrequency = 100000000; // 100 Mhz like Intel part.
Cpu->MaxBusRatio = 1; // TODO: Maybe unsafe too, we need more investigation.
} else {
Cpu->FSBFrequency = DivU64x32 (Cpu->CPUFrequency, Cpu->MaxBusRatio);
}
//
// CPUPM is not supported on AMD, meaning the current
// and minimum bus ratio are equal to the maximum bus ratio
//
Cpu->CurBusRatio = Cpu->MaxBusRatio;
Cpu->MinBusRatio = Cpu->MaxBusRatio;
}
}
}
/** Scan the processor and fill the cpu info structure with results
@param[in] Cpu A pointer to the cpu info structure to fill with results
@retval EFI_SUCCESS The scan was completed successfully.
**/
VOID
OcCpuScanProcessor (
IN OUT OC_CPU_INFO *Cpu
)
{
UINT32 CpuidEax;
UINT32 CpuidEbx;
UINT32 CpuidEcx;
UINT32 CpuidEdx;
ASSERT (Cpu != NULL);
ZeroMem (Cpu, sizeof (*Cpu));
//
// Get vendor CPUID 0x00000000
//
AsmCpuid (CPUID_SIGNATURE, &CpuidEax, &Cpu->Vendor[0], &Cpu->Vendor[2], &Cpu->Vendor[1]);
Cpu->MaxId = CpuidEax;
//
// Get extended CPUID 0x80000000
//
AsmCpuid (CPUID_EXTENDED_FUNCTION, &CpuidEax, &CpuidEbx, &CpuidEcx, &CpuidEdx);
Cpu->MaxExtId = CpuidEax;
//
// Get brand string CPUID 0x80000002 - 0x80000004
//
if (Cpu->MaxExtId >= CPUID_BRAND_STRING3) {
//
// The brandstring 48 bytes max, guaranteed NULL terminated.
//
UINT32 *BrandString = (UINT32 *) Cpu->BrandString;
AsmCpuid (
CPUID_BRAND_STRING1,
BrandString,
(BrandString + 1),
(BrandString + 2),
(BrandString + 3)
);
AsmCpuid (
CPUID_BRAND_STRING2,
(BrandString + 4),
(BrandString + 5),
(BrandString + 6),
(BrandString + 7)
);
AsmCpuid (
CPUID_BRAND_STRING3,
(BrandString + 8),
(BrandString + 9),
(BrandString + 10),
(BrandString + 11)
);
}
ScanThreadCount (Cpu);
//
// Get processor signature and decode
//
if (Cpu->MaxId >= CPUID_VERSION_INFO) {
if (Cpu->Vendor[0] == CPUID_VENDOR_INTEL) {
Cpu->MicrocodeRevision = AsmReadIntelMicrocodeRevision ();
}
AsmCpuid (
CPUID_VERSION_INFO,
&Cpu->CpuidVerEax.Uint32,
&Cpu->CpuidVerEbx.Uint32,
&Cpu->CpuidVerEcx.Uint32,
&Cpu->CpuidVerEdx.Uint32
);
Cpu->Signature = Cpu->CpuidVerEax.Uint32;
Cpu->Stepping = (UINT8) Cpu->CpuidVerEax.Bits.SteppingId;
Cpu->ExtModel = (UINT8) Cpu->CpuidVerEax.Bits.ExtendedModelId;
Cpu->Model = (UINT8) Cpu->CpuidVerEax.Bits.Model | (UINT8) (Cpu->CpuidVerEax.Bits.ExtendedModelId << 4U);
Cpu->Family = (UINT8) Cpu->CpuidVerEax.Bits.FamilyId;
Cpu->Type = (UINT8) Cpu->CpuidVerEax.Bits.ProcessorType;
Cpu->ExtFamily = (UINT8) Cpu->CpuidVerEax.Bits.ExtendedFamilyId;
Cpu->Brand = (UINT8) Cpu->CpuidVerEbx.Bits.BrandIndex;
Cpu->Features = LShiftU64 (Cpu->CpuidVerEcx.Uint32, 32) | Cpu->CpuidVerEdx.Uint32;
if (Cpu->Features & CPUID_FEATURE_HTT) {
Cpu->ThreadCount = (UINT16) Cpu->CpuidVerEbx.Bits.MaximumAddressableIdsForLogicalProcessors;
}
}
DEBUG ((DEBUG_INFO, "OCCPU: Found %a\n", Cpu->BrandString));
DEBUG ((
DEBUG_INFO,
"OCCPU: Signature %0X Stepping %0X Model %0X Family %0X Type %0X ExtModel %0X ExtFamily %0X uCode %0X\n",
Cpu->Signature,
Cpu->Stepping,
Cpu->Model,
Cpu->Family,
Cpu->Type,
Cpu->ExtModel,
Cpu->ExtFamily,
Cpu->MicrocodeRevision
));
Cpu->CPUFrequencyFromVMT = InternalCalculateVMTFrequency (
&Cpu->FSBFrequency,
&Cpu->Hypervisor
);
if (Cpu->Hypervisor) {
DEBUG ((DEBUG_INFO, "OCCPU: Hypervisor detected\n"));
}
if (Cpu->CPUFrequencyFromVMT > 0) {
Cpu->CPUFrequency = Cpu->CPUFrequencyFromVMT;
//
// We can calculate Bus Ratio here
//
Cpu->MaxBusRatio = (UINT8) DivU64x32 (Cpu->CPUFrequency, (UINT32)Cpu->FSBFrequency);
//
// We don't have anything like turbo, so we just assign some variables here
//
Cpu->MinBusRatio = Cpu->MaxBusRatio;
Cpu->CurBusRatio = Cpu->MaxBusRatio;
DEBUG ((
DEBUG_INFO,
"OCCPU: VMWare TSC: %11LuHz, %5LuMHz; FSB: %11LuHz, %5LuMHz; BusRatio: %d\n",
Cpu->CPUFrequency,
DivU64x32 (Cpu->CPUFrequency, 1000000),
Cpu->FSBFrequency,
DivU64x32 (Cpu->FSBFrequency, 1000000),
Cpu->MaxBusRatio
));
}
if (Cpu->Vendor[0] == CPUID_VENDOR_INTEL) {
ScanIntelProcessor (Cpu);
} else if (Cpu->Vendor[0] == CPUID_VENDOR_AMD) {
ScanAmdProcessor (Cpu);
} else {
DEBUG ((DEBUG_WARN, "OCCPU: Found unsupported CPU vendor: %0X", Cpu->Vendor[0]));
return;
}
//
// SMBIOS Type4 ExternalClock field is assumed to have X*4 FSB frequency in MT/s.
// This value is cosmetics, but we still want to set it properly.
// Magic 4 value comes from 4 links in pretty much every modern Intel CPU.
// On modern CPUs this is now named Base clock (BCLK).
// Note, that this value was incorrect for most Macs since iMac12,x till iMac18,x inclusive.
// REF: https://github.com/acidanthera/bugtracker/issues/622#issuecomment-570811185
//
Cpu->ExternalClock = (UINT16) DivU64x32 (Cpu->FSBFrequency, 1000000);
//
// This is again cosmetics by errors in FSBFrequency calculation.
//
if (Cpu->ExternalClock >= 99 && Cpu->ExternalClock <= 101) {
Cpu->ExternalClock = 100;
} else if (Cpu->ExternalClock >= 132 && Cpu->ExternalClock <= 134) {
Cpu->ExternalClock = 133;
} else if (Cpu->ExternalClock >= 265 && Cpu->ExternalClock <= 267) {
Cpu->ExternalClock = 266;
}
DEBUG ((
DEBUG_INFO,
"OCCPU: CPUFrequencyFromTSC %11LuHz %5LuMHz\n",
Cpu->CPUFrequencyFromTSC,
DivU64x32 (Cpu->CPUFrequencyFromTSC, 1000000)
));
DEBUG ((
DEBUG_INFO,
"OCCPU: CPUFrequency %11LuHz %5LuMHz\n",
Cpu->CPUFrequency,
DivU64x32 (Cpu->CPUFrequency, 1000000)
));
DEBUG ((
DEBUG_INFO,
"OCCPU: FSBFrequency %11LuHz %5LuMHz\n",
Cpu->FSBFrequency,
DivU64x32 (Cpu->FSBFrequency, 1000000)
));
DEBUG ((
DEBUG_INFO,
"OCCPU: Pkg %u Cores %u Threads %u\n",
Cpu->PackageCount,
Cpu->CoreCount,
Cpu->ThreadCount
));
}
VOID
OcCpuCorrectFlexRatio (
IN OC_CPU_INFO *Cpu
)
{
UINT64 Msr;
UINT64 FlexRatio;
if (Cpu->Vendor[0] == CPUID_VENDOR_INTEL
&& Cpu->Model != CPU_MODEL_GOLDMONT
&& Cpu->Model != CPU_MODEL_AIRMONT
&& Cpu->Model != CPU_MODEL_AVOTON) {
Msr = AsmReadMsr64 (MSR_FLEX_RATIO);
if (Msr & FLEX_RATIO_EN) {
FlexRatio = BitFieldRead64 (Msr, 8, 15);
if (FlexRatio == 0) {
//
// Disable Flex Ratio if current value is 0.
//
AsmWriteMsr64 (MSR_FLEX_RATIO, Msr & ~((UINT64) FLEX_RATIO_EN));
}
}
}
}
STATIC
VOID
EFIAPI
SyncTscOnCpu (
IN VOID *Buffer
)
{
OC_CPU_TSC_SYNC *Sync;
Sync = Buffer;
AsmIncrementUint32 (&Sync->CurrentCount);
while (Sync->CurrentCount < Sync->APThreadCount) {
//
// Busy-wait on AP CPU cores.
//
}
AsmWriteMsr64 (MSR_IA32_TIME_STAMP_COUNTER, Sync->Tsc);
}
STATIC
VOID
EFIAPI
ResetAdjustTsc (
IN VOID *Buffer
)
{
OC_CPU_TSC_SYNC *Sync;
Sync = Buffer;
AsmIncrementUint32 (&Sync->CurrentCount);
while (Sync->CurrentCount < Sync->APThreadCount) {
//
// Busy-wait on AP CPU cores.
//
}
AsmWriteMsr64 (MSR_IA32_TSC_ADJUST, 0);
}
EFI_STATUS
OcCpuCorrectTscSync (
IN OC_CPU_INFO *Cpu,
IN UINTN Timeout
)
{
EFI_STATUS Status;
EFI_MP_SERVICES_PROTOCOL *MpServices;
FRAMEWORK_EFI_MP_SERVICES_PROTOCOL *FrameworkMpServices;
OC_CPU_TSC_SYNC Sync;
EFI_TPL OldTpl;
BOOLEAN InterruptState;
if (Cpu->ThreadCount <= 1) {
DEBUG ((DEBUG_INFO, "OCCPU: Thread count is too low for sync - %u\n", Cpu->ThreadCount));
return EFI_UNSUPPORTED;
}
Status = gBS->LocateProtocol (
&gEfiMpServiceProtocolGuid,
NULL,
(VOID **) &MpServices
);
if (EFI_ERROR(Status)) {
MpServices = NULL;
Status = gBS->LocateProtocol (
&gFrameworkEfiMpServiceProtocolGuid,
NULL,
(VOID **) &FrameworkMpServices
);
if (EFI_ERROR(Status)) {
DEBUG ((DEBUG_INFO, "OCCPU: Failed to find mp services - %r\n", Status));
return Status;
}
}
Sync.CurrentCount = 0;
Sync.APThreadCount = Cpu->ThreadCount - 1;
OldTpl = gBS->RaiseTPL (TPL_HIGH_LEVEL);
InterruptState = SaveAndDisableInterrupts ();
if (Cpu->TscAdjust > 0) {
if (MpServices != NULL) {
Status = MpServices->StartupAllAPs (MpServices, ResetAdjustTsc, FALSE, NULL, Timeout, &Sync, NULL);
} else {
Status = FrameworkMpServices->StartupAllAPs (FrameworkMpServices, ResetAdjustTsc, FALSE, NULL, Timeout, &Sync, NULL);
}
AsmWriteMsr64 (MSR_IA32_TSC_ADJUST, 0);
} else {
Sync.Tsc = AsmReadTsc ();
if (MpServices != NULL) {
Status = MpServices->StartupAllAPs (MpServices, SyncTscOnCpu, FALSE, NULL, Timeout, &Sync, NULL);
} else {
Status = FrameworkMpServices->StartupAllAPs (FrameworkMpServices, SyncTscOnCpu, FALSE, NULL, Timeout, &Sync, NULL);
}
AsmWriteMsr64 (MSR_IA32_TIME_STAMP_COUNTER, Sync.Tsc);
}
SetInterruptState (InterruptState);
gBS->RestoreTPL (OldTpl);
DEBUG ((DEBUG_INFO, "OCCPU: Completed TSC sync with code - %r\n", Status));
return Status;
}
OC_CPU_GENERATION
OcCpuGetGeneration (
VOID
)
{
CPU_MICROCODE_PROCESSOR_SIGNATURE Sig;
UINT32 CpuFamily;
UINT32 CpuModel;
OC_CPU_GENERATION CpuGeneration;
Sig.Uint32 = 0;
AsmCpuid (1, &Sig.Uint32, NULL, NULL, NULL);
CpuFamily = Sig.Bits.Family;
if (CpuFamily == 15) {
CpuFamily += Sig.Bits.ExtendedFamily;
}
CpuModel = Sig.Bits.Model;
if (CpuFamily == 15 || CpuFamily == 6) {
CpuModel |= Sig.Bits.ExtendedModel << 4;
}
CpuGeneration = OcCpuGenerationUnknown;
if (CpuFamily == 6) {
switch (CpuModel) {
case CPU_MODEL_PENRYN:
CpuGeneration = OcCpuGenerationPenryn;
break;
case CPU_MODEL_NEHALEM:
case CPU_MODEL_FIELDS:
case CPU_MODEL_DALES:
case CPU_MODEL_NEHALEM_EX:
CpuGeneration = OcCpuGenerationNehalem;
break;
case CPU_MODEL_DALES_32NM:
case CPU_MODEL_WESTMERE:
case CPU_MODEL_WESTMERE_EX:
CpuGeneration = OcCpuGenerationWestmere;
break;
case CPU_MODEL_SANDYBRIDGE:
case CPU_MODEL_JAKETOWN:
CpuGeneration = OcCpuGenerationSandyBridge;
break;
case CPU_MODEL_IVYBRIDGE:
case CPU_MODEL_IVYBRIDGE_EP:
CpuGeneration = OcCpuGenerationIvyBridge;
break;
case CPU_MODEL_HASWELL:
case CPU_MODEL_HASWELL_EP:
case CPU_MODEL_HASWELL_ULT:
case CPU_MODEL_CRYSTALWELL:
CpuGeneration = OcCpuGenerationHaswell;
break;
case CPU_MODEL_BROADWELL:
case CPU_MODEL_BROADWELL_EP:
case CPU_MODEL_BRYSTALWELL:
CpuGeneration = OcCpuGenerationBroadwell;
break;
case CPU_MODEL_SKYLAKE:
case CPU_MODEL_SKYLAKE_DT:
case CPU_MODEL_SKYLAKE_W:
CpuGeneration = OcCpuGenerationSkylake;
break;
case CPU_MODEL_KABYLAKE:
case CPU_MODEL_KABYLAKE_DT:
//
// Kaby has 0x9 stepping, and Coffee use 0xA / 0xB stepping.
//
if (Sig.Bits.Stepping == 9) {
CpuGeneration = OcCpuGenerationKabyLake;
} else {
CpuGeneration = OcCpuGenerationCoffeeLake;
}
break;
case CPU_MODEL_CANNONLAKE:
CpuGeneration = OcCpuGenerationCannonLake;
break;
}
}
DEBUG ((
DEBUG_VERBOSE,
"OCCPU: Discovered CpuFamily %d CpuModel %d CpuStepping %d CpuGeneration %d\n",
CpuFamily,
CpuModel,
Sig.Bits.Stepping,
CpuGeneration
));
return CpuGeneration;
}