ULONG KeDcacheFlushCount = 0;
ULONG KeIcacheFlushCount = 0;
ULONG KiDmaIoCoherency = 0;
+ULONG KePrefetchNTAGranularity = 32;
CHAR KeNumberProcessors;
KAFFINITY KeActiveProcessors = 1;
BOOLEAN KiI386PentiumLockErrataPresent;
return __readmsr(Register);
}
+/* NSC/Cyrix CPU configuration register index */
+#define CX86_CCR1 0xc1
+
+/* NSC/Cyrix CPU indexed register access macros */
+static __inline
+ULONG
+getCx86(UCHAR reg)
+{
+ WRITE_PORT_UCHAR((PUCHAR)(ULONG_PTR)0x22, reg);
+ return READ_PORT_UCHAR((PUCHAR)(ULONG_PTR)0x23);
+}
+
+#define setCx86(reg, data) do { \
+ WRITE_PORT_UCHAR((PUCHAR)(ULONG_PTR)0x22,(reg)); \
+ WRITE_PORT_UCHAR((PUCHAR)(ULONG_PTR)0x23,(data)); \
+} while (0)
+
/* FUNCTIONS *****************************************************************/
VOID
NTAPI
+INIT_FUNCTION
KiSetProcessorType(VOID)
{
ULONG EFlags, NewEFlags;
ULONG
NTAPI
+INIT_FUNCTION
KiGetCpuVendor(VOID)
{
PKPRCB Prcb = KeGetCurrentPrcb();
ULONG
NTAPI
+INIT_FUNCTION
KiGetFeatureBits(VOID)
{
PKPRCB Prcb = KeGetCurrentPrcb();
ULONG Vendor;
ULONG FeatureBits = KF_WORKING_PTE;
- ULONG Reg[4], Dummy;
+ ULONG Reg[4], Dummy, Ccr1;
BOOLEAN ExtendedCPUID = TRUE;
ULONG CpuFeatures = 0;
/* Remove support for correct PTE support. */
FeatureBits &= ~KF_WORKING_PTE;
}
+
+ /* Virtualbox claims to have no SYSENTER support,
+ * which is false for processors >= Pentium Pro */
+ if(Prcb->CpuType >= 6)
+ {
+ Reg[3] |= 0x800;
+ }
- /* Check if the CPU is too old to support SYSENTER */
- if ((Prcb->CpuType < 6) ||
- ((Prcb->CpuType == 6) && (Prcb->CpuStep < 0x0303)))
+ /* Check if the CPU is too old to support SYSENTER,
+ * See Intel CPUID instruction manual for details*/
+ if ((Reg[0] & 0x0FFF3FFF) < 0x00000633)
{
/* Disable it */
Reg[3] &= ~0x800;
/* Cyrix CPUs */
case CPU_CYRIX:
- /* FIXME: CMPXCGH8B */
+ /* Workaround the "COMA" bug on 6x family of Cyrix CPUs */
+ if (Prcb->CpuType == 6 &&
+ Prcb->CpuStep <= 1)
+ {
+ /* Get CCR1 value */
+ Ccr1 = getCx86(CX86_CCR1);
+
+ /* Enable the NO_LOCK bit */
+ Ccr1 |= 0x10;
+
+ /* Set the new CCR1 value */
+ setCx86(CX86_CCR1, Ccr1);
+ }
+
+ /* Set the current features */
+ CpuFeatures = Reg[3];
break;
}
}
}
+
+ DPRINT1("Supported CPU features : %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s %s\n",
+#define print_supported(kf_value) \
+ FeatureBits & kf_value ? #kf_value : ""
+ print_supported(KF_V86_VIS),
+ print_supported(KF_RDTSC),
+ print_supported(KF_CR4),
+ print_supported(KF_CMOV),
+ print_supported(KF_GLOBAL_PAGE),
+ print_supported(KF_LARGE_PAGE),
+ print_supported(KF_MTRR),
+ print_supported(KF_CMPXCHG8B),
+ print_supported(KF_MMX),
+ print_supported(KF_WORKING_PTE),
+ print_supported(KF_PAT),
+ print_supported(KF_FXSR),
+ print_supported(KF_FAST_SYSCALL),
+ print_supported(KF_XMMI),
+ print_supported(KF_3DNOW),
+ print_supported(KF_AMDK6MTRR),
+ print_supported(KF_XMMI64),
+ print_supported(KF_DTS),
+ print_supported(KF_NX_BIT),
+ print_supported(KF_NX_DISABLED),
+ print_supported(KF_NX_ENABLED));
+#undef print_supported
/* Return the Feature Bits */
return FeatureBits;
VOID
NTAPI
+INIT_FUNCTION
KiGetCacheInformation(VOID)
{
PKIPCR Pcr = (PKIPCR)KeGetPcr();
ULONG CacheRequests = 0, i;
ULONG CurrentRegister;
UCHAR RegisterByte;
+ ULONG Size, Associativity = 0, CacheLine = 64, CurrentSize = 0;
BOOLEAN FirstPass = TRUE;
/* Set default L2 size */
* (32MB), or from 0x80 to 0x89 (same size but
* 8-way associative.
*/
- if (((RegisterByte > 0x40) &&
- (RegisterByte <= 0x49)) ||
- ((RegisterByte > 0x80) &&
- (RegisterByte <= 0x89)))
+ if (((RegisterByte > 0x40) && (RegisterByte <= 0x47)) ||
+ ((RegisterByte > 0x78) && (RegisterByte <= 0x7C)) ||
+ ((RegisterByte > 0x80) && (RegisterByte <= 0x85)))
{
+ /* Compute associativity */
+ Associativity = 4;
+ if (RegisterByte >= 0x79) Associativity = 8;
+
/* Mask out only the first nibble */
- RegisterByte &= 0x0F;
-
- /* Set the L2 Cache Size */
- Pcr->SecondLevelCacheSize = 0x10000 <<
- RegisterByte;
+ RegisterByte &= 0x07;
+
+ /* Check if this cache is bigger than the last */
+ Size = 0x10000 << RegisterByte;
+ if ((Size / Associativity) > CurrentSize)
+ {
+ /* Set the L2 Cache Size and Associativity */
+ CurrentSize = Size / Associativity;
+ Pcr->SecondLevelCacheSize = Size;
+ Pcr->SecondLevelCacheAssociativity = Associativity;
+ }
+ }
+ else if ((RegisterByte > 0x21) && (RegisterByte <= 0x29))
+ {
+ /* Set minimum cache line size */
+ if (CacheLine < 128) CacheLine = 128;
+
+ /* Hard-code size/associativity */
+ Associativity = 8;
+ switch (RegisterByte)
+ {
+ case 0x22:
+ Size = 512 * 1024;
+ Associativity = 4;
+ break;
+
+ case 0x23:
+ Size = 1024 * 1024;
+ break;
+
+ case 0x25:
+ Size = 2048 * 1024;
+ break;
+
+ case 0x29:
+ Size = 4096 * 1024;
+ break;
+
+ default:
+ Size = 0;
+ break;
+ }
+
+ /* Check if this cache is bigger than the last */
+ if ((Size / Associativity) > CurrentSize)
+ {
+ /* Set the L2 Cache Size and Associativity */
+ CurrentSize = Size / Associativity;
+ Pcr->SecondLevelCacheSize = Size;
+ Pcr->SecondLevelCacheAssociativity = Associativity;
+ }
+ }
+ else if (((RegisterByte > 0x65) && (RegisterByte < 0x69)) ||
+ (RegisterByte == 0x2C) || (RegisterByte == 0xF0))
+ {
+ /* Indicates L1 cache line of 64 bytes */
+ KePrefetchNTAGranularity = 64;
+ }
+ else if (RegisterByte == 0xF1)
+ {
+ /* Indicates L1 cache line of 128 bytes */
+ KePrefetchNTAGranularity = 128;
+ }
+ else if (((RegisterByte >= 0x4A) && (RegisterByte <= 0x4C)) ||
+ (RegisterByte == 0x78) ||
+ (RegisterByte == 0x7D) ||
+ (RegisterByte == 0x7F) ||
+ (RegisterByte == 0x86) ||
+ (RegisterByte == 0x87))
+ {
+ /* Set minimum cache line size */
+ if (CacheLine < 64) CacheLine = 64;
+
+ /* Hard-code size/associativity */
+ switch (RegisterByte)
+ {
+ case 0x4A:
+ Size = 4 * 1024 * 1024;
+ Associativity = 8;
+ break;
+
+ case 0x4B:
+ Size = 6 * 1024 * 1024;
+ Associativity = 12;
+ break;
+
+ case 0x4C:
+ Size = 8 * 1024 * 1024;
+ Associativity = 16;
+ break;
+
+ case 0x78:
+ Size = 1 * 1024 * 1024;
+ Associativity = 4;
+ break;
+
+ case 0x7D:
+ Size = 2 * 1024 * 1024;
+ Associativity = 8;
+ break;
+
+ case 0x7F:
+ Size = 512 * 1024;
+ Associativity = 2;
+ break;
+
+ case 0x86:
+ Size = 512 * 1024;
+ Associativity = 4;
+ break;
+
+ case 0x87:
+ Size = 1 * 1024 * 1024;
+ Associativity = 8;
+ break;
+
+ default:
+ Size = 0;
+ break;
+ }
+
+ /* Check if this cache is bigger than the last */
+ if ((Size / Associativity) > CurrentSize)
+ {
+ /* Set the L2 Cache Size and Associativity */
+ CurrentSize = Size / Associativity;
+ Pcr->SecondLevelCacheSize = Size;
+ Pcr->SecondLevelCacheAssociativity = Associativity;
+ }
}
}
}
case CPU_AMD:
- /* Check if we support CPUID 0x80000006 */
- CPUID(0x80000000, &Data[0], &Dummy, &Dummy, &Dummy);
- if (Data[0] >= 6)
+ /* Check if we support CPUID 0x80000005 */
+ CPUID(0x80000000, &Data[0], &Data[1], &Data[2], &Data[3]);
+ if (Data[0] >= 0x80000006)
{
- /* Get 2nd level cache and tlb size */
- CPUID(0x80000006, &Dummy, &Dummy, &Data[2], &Dummy);
+ /* Get L1 size first */
+ CPUID(0x80000005, &Data[0], &Data[1], &Data[2], &Data[3]);
+ KePrefetchNTAGranularity = Data[2] & 0xFF;
+
+ /* Check if we support CPUID 0x80000006 */
+ CPUID(0x80000000, &Data[0], &Data[1], &Data[2], &Data[3]);
+ if (Data[0] >= 0x80000006)
+ {
+ /* Get 2nd level cache and tlb size */
+ CPUID(0x80000006, &Data[0], &Data[1], &Data[2], &Data[3]);
+
+ /* Cache line size */
+ CacheLine = Data[2] & 0xFF;
+
+ /* Hardcode associativity */
+ RegisterByte = Data[2] >> 12;
+ switch (RegisterByte)
+ {
+ case 2:
+ Associativity = 2;
+ break;
+
+ case 4:
+ Associativity = 4;
+ break;
+
+ case 6:
+ Associativity = 8;
+ break;
+
+ case 8:
+ case 15:
+ Associativity = 16;
+ break;
+
+ default:
+ Associativity = 1;
+ break;
+ }
+
+ /* Compute size */
+ Size = (Data[2] >> 16) << 10;
+
+ /* Hack for Model 6, Steping 300 */
+ if ((KeGetCurrentPrcb()->CpuType == 6) &&
+ (KeGetCurrentPrcb()->CpuStep == 0x300))
+ {
+ /* Stick 64K in there */
+ Size = 64 * 1024;
+ }
- /* Set the L2 Cache Size */
- Pcr->SecondLevelCacheSize = (Data[2] & 0xFFFF0000) >> 6;
+ /* Set the L2 Cache Size and associativity */
+ Pcr->SecondLevelCacheSize = Size;
+ Pcr->SecondLevelCacheAssociativity = Associativity;
+ }
}
break;
/* FIXME */
break;
}
+
+ /* Set the cache line */
+ if (CacheLine > KeLargestCacheLine) KeLargestCacheLine = CacheLine;
+ DPRINT1("Prefetch Cache: %d bytes\tL2 Cache: %d bytes\tL2 Cache Line: %d bytes\tL2 Cache Associativity: %d\n",
+ KePrefetchNTAGranularity,
+ Pcr->SecondLevelCacheSize,
+ KeLargestCacheLine,
+ Pcr->SecondLevelCacheAssociativity);
}
VOID
NTAPI
+INIT_FUNCTION
KiSetCR0Bits(VOID)
{
ULONG Cr0;
VOID
NTAPI
+INIT_FUNCTION
KiInitializeTSS2(IN PKTSS Tss,
IN PKGDTENTRY TssEntry OPTIONAL)
{
VOID
FASTCALL
+INIT_FUNCTION
Ki386InitializeTss(IN PKTSS Tss,
IN PKIDTENTRY Idt,
IN PKGDTENTRY Gdt)
VOID
NTAPI
+INIT_FUNCTION
KiInitializeMachineType(VOID)
{
/* Set the Machine Type we got from NTLDR */
ULONG_PTR
NTAPI
+INIT_FUNCTION
KiLoadFastSyscallMachineSpecificRegisters(IN ULONG_PTR Context)
{
/* Set CS and ESP */
VOID
NTAPI
-KiDisableFastSyscallReturn(VOID)
-{
- /* Was it applied? */
- if (KiSystemCallExitAdjusted)
- {
- /* Restore the original value */
- KiSystemCallExitBranch[1] = KiSystemCallExitBranch[1] - KiSystemCallExitAdjusted;
-
- /* It's not adjusted anymore */
- KiSystemCallExitAdjusted = FALSE;
- }
-}
-
-VOID
-NTAPI
-KiEnableFastSyscallReturn(VOID)
-{
- /* Check if the patch has already been done */
- if ((KiSystemCallExitAdjusted == KiSystemCallExitAdjust) &&
- (KiFastCallCopyDoneOnce))
- {
- DPRINT1("SYSEXIT Code Patch was already done!\n");
- return;
- }
-
- /* Make sure the offset is within the distance of a Jxx SHORT */
- if ((KiSystemCallExitBranch[1] - KiSystemCallExitAdjust) < 0x80)
- {
- /* Remove any existing code patch */
- DPRINT1("Correct SHORT size found\n");
- KiDisableFastSyscallReturn();
-
- /* We should have a JNZ there */
- ASSERT(KiSystemCallExitBranch[0] == 0x75);
-
- /* Do the patch */
- DPRINT1("Current jump offset: %lx\n", KiSystemCallExitBranch[1]);
- KiSystemCallExitAdjusted = KiSystemCallExitAdjust;
- KiSystemCallExitBranch[1] -= KiSystemCallExitAdjusted;
- DPRINT1("New jump offset: %lx\n", KiSystemCallExitBranch[1]);
-
- /* Remember that we've done it */
- KiFastCallCopyDoneOnce = TRUE;
- }
- else
- {
- /* This shouldn't happen unless we've messed the macros up */
- DPRINT1("Your compiled kernel is broken!\n");
- DbgBreakPoint();
- }
-}
-
-VOID
-NTAPI
+INIT_FUNCTION
KiRestoreFastSyscallReturnState(VOID)
{
/* Check if the CPU Supports fast system call */
/* Check if it has been disabled */
if (!KiFastSystemCallDisable)
{
- /* KiSystemCallExit2 should come BEFORE KiSystemCallExit */
- DPRINT1("Exit2: %p Exit1: %p\n", KiSystemCallExit2, KiSystemCallExit);
- ASSERT(KiSystemCallExit2 < KiSystemCallExit);
-
- /* It's enabled, so we'll have to do a code patch */
- KiSystemCallExitAdjust = KiSystemCallExit - KiSystemCallExit2;
- DPRINT1("SYSENTER Capable Machine. Jump Offset Delta: %lx\n", KiSystemCallExitAdjust);
+ /* Do an IPI to enable it */
+ KeIpiGenericCall(KiLoadFastSyscallMachineSpecificRegisters, 0);
+
+ /* It's enabled, so use the proper exit stub */
+ KiFastCallExitHandler = KiSystemCallSysExitReturn;
+ DPRINT1("Support for SYSENTER detected.\n");
}
else
{
/* Disable fast system call */
KeFeatureBits &= ~KF_FAST_SYSCALL;
+ KiFastCallExitHandler = KiSystemCallTrapReturn;
+ DPRINT1("Support for SYSENTER disabled.\n");
}
}
-
- /* Now check if all CPUs support fast system call, and the registry allows it */
- if (KeFeatureBits & KF_FAST_SYSCALL)
+ else
{
- /* Do an IPI to enable it */
- KeIpiGenericCall(KiLoadFastSyscallMachineSpecificRegisters, 0);
+ /* Use the IRET handler */
+ KiFastCallExitHandler = KiSystemCallTrapReturn;
+ DPRINT1("No support for SYSENTER detected.\n");
}
-
- /* Perform the code patch that is required */
- KiEnableFastSyscallReturn();
}
ULONG_PTR
NTAPI
+INIT_FUNCTION
Ki386EnableDE(IN ULONG_PTR Context)
{
/* Enable DE */
ULONG_PTR
NTAPI
+INIT_FUNCTION
Ki386EnableFxsr(IN ULONG_PTR Context)
{
/* Enable FXSR */
ULONG_PTR
NTAPI
+INIT_FUNCTION
Ki386EnableXMMIExceptions(IN ULONG_PTR Context)
{
PKIDTENTRY IdtEntry;
VOID
NTAPI
+INIT_FUNCTION
KiI386PentiumLockErrataFixup(VOID)
{
KDESCRIPTOR IdtDescriptor;
BOOLEAN
NTAPI
+INIT_FUNCTION
KiIsNpxPresent(VOID)
{
ULONG Cr0;
Cr0 = __readcr0() & ~(CR0_MP | CR0_TS | CR0_EM | CR0_ET);
/* Store on FPU stack */
+#ifdef _MSC_VER
+ __asm fninit;
+ __asm fnstsw Magic;
+#else
asm volatile ("fninit;" "fnstsw %0" : "+m"(Magic));
+#endif
/* Magic should now be cleared */
if (Magic & 0xFF)
BOOLEAN
NTAPI
+INIT_FUNCTION
KiIsNpxErrataPresent(VOID)
{
BOOLEAN ErrataPresent;
__writecr0(Cr0 & ~(CR0_MP | CR0_TS | CR0_EM));
/* Initialize FPU state */
- asm volatile ("fninit");
+ Ke386FnInit();
/* Multiply the magic values and divide, we should get the result back */
Value1 = 4195835.0;
return ErrataPresent;
}
-NTAPI
VOID
+NTAPI
KiFlushNPXState(IN PFLOATING_SAVE_AREA SaveArea)
{
ULONG EFlags, Cr0;
projects / reactos.git / blobdiff