--- /dev/null
+/*++\r
+\r
+Copyright (c) 2006, Intel Corporation \r
+All rights reserved. This program and the accompanying materials \r
+are licensed and made available under the terms and conditions of the BSD License \r
+which accompanies this distribution. The full text of the license may be found at \r
+http://opensource.org/licenses/bsd-license.php \r
+ \r
+THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS, \r
+WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED. \r
+\r
+Module Name:\r
+\r
+ EbcSupport.c\r
+\r
+Abstract:\r
+\r
+ This module contains EBC support routines that are customized based on\r
+ the target processor.\r
+\r
+--*/\r
+\r
+#include "EbcInt.h"\r
+#include "EbcExecute.h"\r
+#include "EbcSupport.h"\r
+\r
+STATIC\r
+EFI_STATUS\r
+WriteBundle (\r
+ IN VOID *MemPtr,\r
+ IN UINT8 Template,\r
+ IN UINT64 Slot0,\r
+ IN UINT64 Slot1,\r
+ IN UINT64 Slot2\r
+ );\r
+\r
+STATIC\r
+VOID\r
+PushU64 (\r
+ VM_CONTEXT *VmPtr,\r
+ UINT64 Arg\r
+ )\r
+{\r
+ //\r
+ // Advance the VM stack down, and then copy the argument to the stack.\r
+ // Hope it's aligned.\r
+ //\r
+ VmPtr->R[0] -= sizeof (UINT64);\r
+ *(UINT64 *) VmPtr->R[0] = Arg;\r
+}\r
+\r
+STATIC\r
+UINT64\r
+EbcInterpret (\r
+ UINT64 Arg1,\r
+ ...\r
+ )\r
+{\r
+ //\r
+ // Create a new VM context on the stack\r
+ //\r
+ VM_CONTEXT VmContext;\r
+ UINTN Addr;\r
+ EFI_STATUS Status;\r
+ UINTN StackIndex;\r
+ VA_LIST List;\r
+ UINT64 Arg2;\r
+ UINT64 Arg3;\r
+ UINT64 Arg4;\r
+ UINT64 Arg5;\r
+ UINT64 Arg6;\r
+ UINT64 Arg7;\r
+ UINT64 Arg8;\r
+ UINT64 Arg9;\r
+ UINT64 Arg10;\r
+ UINT64 Arg11;\r
+ UINT64 Arg12;\r
+ UINT64 Arg13;\r
+ UINT64 Arg14;\r
+ UINT64 Arg15;\r
+ UINT64 Arg16;\r
+ //\r
+ // Get the EBC entry point from the processor register. Make sure you don't\r
+ // call any functions before this or you could mess up the register the\r
+ // entry point is passed in.\r
+ //\r
+ Addr = EbcLLGetEbcEntryPoint ();\r
+ //\r
+ // Need the args off the stack.\r
+ //\r
+ VA_START (List, Arg1);\r
+ Arg2 = VA_ARG (List, UINT64);\r
+ Arg3 = VA_ARG (List, UINT64);\r
+ Arg4 = VA_ARG (List, UINT64);\r
+ Arg5 = VA_ARG (List, UINT64);\r
+ Arg6 = VA_ARG (List, UINT64);\r
+ Arg7 = VA_ARG (List, UINT64);\r
+ Arg8 = VA_ARG (List, UINT64);\r
+ Arg9 = VA_ARG (List, UINT64);\r
+ Arg10 = VA_ARG (List, UINT64);\r
+ Arg11 = VA_ARG (List, UINT64);\r
+ Arg12 = VA_ARG (List, UINT64);\r
+ Arg13 = VA_ARG (List, UINT64);\r
+ Arg14 = VA_ARG (List, UINT64);\r
+ Arg15 = VA_ARG (List, UINT64);\r
+ Arg16 = VA_ARG (List, UINT64);\r
+ //\r
+ // Now clear out our context\r
+ //\r
+ ZeroMem ((VOID *) &VmContext, sizeof (VM_CONTEXT));\r
+ //\r
+ // Set the VM instruction pointer to the correct location in memory.\r
+ //\r
+ VmContext.Ip = (VMIP) Addr;\r
+ //\r
+ // Initialize the stack pointer for the EBC. Get the current system stack\r
+ // pointer and adjust it down by the max needed for the interpreter.\r
+ //\r
+ //\r
+ // NOTE: Eventually we should have the interpreter allocate memory\r
+ // for stack space which it will use during its execution. This\r
+ // would likely improve performance because the interpreter would\r
+ // no longer be required to test each memory access and adjust\r
+ // those reading from the stack gap.\r
+ //\r
+ // For IPF, the stack looks like (assuming 10 args passed)\r
+ // arg10\r
+ // arg9 (Bottom of high stack)\r
+ // [ stack gap for interpreter execution ]\r
+ // [ magic value for detection of stack corruption ]\r
+ // arg8 (Top of low stack)\r
+ // arg7....\r
+ // arg1\r
+ // [ 64-bit return address ]\r
+ // [ ebc stack ]\r
+ // If the EBC accesses memory in the stack gap, then we assume that it's\r
+ // actually trying to access args9 and greater. Therefore we need to\r
+ // adjust memory accesses in this region to point above the stack gap.\r
+ //\r
+ //\r
+ // Now adjust the EBC stack pointer down to leave a gap for interpreter\r
+ // execution. Then stuff a magic value there.\r
+ //\r
+ \r
+ Status = GetEBCStack((EFI_HANDLE)(UINTN)-1, &VmContext.StackPool, &StackIndex);\r
+ if (EFI_ERROR(Status)) {\r
+ return Status;\r
+ }\r
+ VmContext.StackTop = (UINT8*)VmContext.StackPool + (STACK_REMAIN_SIZE);\r
+ VmContext.R[0] = (UINT64) ((UINT8*)VmContext.StackPool + STACK_POOL_SIZE);\r
+ VmContext.HighStackBottom = (UINTN) VmContext.R[0];\r
+ VmContext.R[0] -= sizeof (UINTN);\r
+\r
+ \r
+ PushU64 (&VmContext, (UINT64) VM_STACK_KEY_VALUE);\r
+ VmContext.StackMagicPtr = (UINTN *) VmContext.R[0];\r
+ VmContext.LowStackTop = (UINTN) VmContext.R[0];\r
+ //\r
+ // Push the EBC arguments on the stack. Does not matter that they may not\r
+ // all be valid.\r
+ //\r
+ PushU64 (&VmContext, Arg16);\r
+ PushU64 (&VmContext, Arg15);\r
+ PushU64 (&VmContext, Arg14);\r
+ PushU64 (&VmContext, Arg13);\r
+ PushU64 (&VmContext, Arg12);\r
+ PushU64 (&VmContext, Arg11);\r
+ PushU64 (&VmContext, Arg10);\r
+ PushU64 (&VmContext, Arg9);\r
+ PushU64 (&VmContext, Arg8);\r
+ PushU64 (&VmContext, Arg7);\r
+ PushU64 (&VmContext, Arg6);\r
+ PushU64 (&VmContext, Arg5);\r
+ PushU64 (&VmContext, Arg4);\r
+ PushU64 (&VmContext, Arg3);\r
+ PushU64 (&VmContext, Arg2);\r
+ PushU64 (&VmContext, Arg1);\r
+ //\r
+ // Push a bogus return address on the EBC stack because the\r
+ // interpreter expects one there. For stack alignment purposes on IPF,\r
+ // EBC return addresses are always 16 bytes. Push a bogus value as well.\r
+ //\r
+ PushU64 (&VmContext, 0);\r
+ PushU64 (&VmContext, 0xDEADBEEFDEADBEEF);\r
+ VmContext.StackRetAddr = (UINT64) VmContext.R[0];\r
+ //\r
+ // Begin executing the EBC code\r
+ //\r
+ EbcExecute (&VmContext);\r
+ //\r
+ // Return the value in R[7] unless there was an error\r
+ //\r
+ ReturnEBCStack(StackIndex);\r
+ return (UINT64) VmContext.R[7];\r
+}\r
+\r
+STATIC\r
+UINT64\r
+ExecuteEbcImageEntryPoint (\r
+ IN EFI_HANDLE ImageHandle,\r
+ IN EFI_SYSTEM_TABLE *SystemTable\r
+ )\r
+/*++\r
+\r
+Routine Description:\r
+\r
+ IPF implementation.\r
+\r
+ Begin executing an EBC image. The address of the entry point is passed\r
+ in via a processor register, so we'll need to make a call to get the\r
+ value.\r
+ \r
+Arguments:\r
+\r
+ ImageHandle - image handle for the EBC application we're executing\r
+ SystemTable - standard system table passed into an driver's entry point\r
+\r
+Returns:\r
+\r
+ The value returned by the EBC application we're going to run.\r
+\r
+--*/\r
+{\r
+ //\r
+ // Create a new VM context on the stack\r
+ //\r
+ VM_CONTEXT VmContext;\r
+ UINTN Addr;\r
+ EFI_STATUS Status;\r
+ UINTN StackIndex;\r
+\r
+ //\r
+ // Get the EBC entry point from the processor register. Make sure you don't\r
+ // call any functions before this or you could mess up the register the\r
+ // entry point is passed in.\r
+ //\r
+ Addr = EbcLLGetEbcEntryPoint ();\r
+\r
+ //\r
+ // Now clear out our context\r
+ //\r
+ ZeroMem ((VOID *) &VmContext, sizeof (VM_CONTEXT));\r
+\r
+ //\r
+ // Save the image handle so we can track the thunks created for this image\r
+ //\r
+ VmContext.ImageHandle = ImageHandle;\r
+ VmContext.SystemTable = SystemTable;\r
+\r
+ //\r
+ // Set the VM instruction pointer to the correct location in memory.\r
+ //\r
+ VmContext.Ip = (VMIP) Addr;\r
+\r
+ //\r
+ // Get the stack pointer. This is the bottom of the upper stack.\r
+ //\r
+ Addr = EbcLLGetStackPointer ();\r
+ \r
+ Status = GetEBCStack(ImageHandle, &VmContext.StackPool, &StackIndex);\r
+ if (EFI_ERROR(Status)) {\r
+ return Status;\r
+ }\r
+ VmContext.StackTop = (UINT8*)VmContext.StackPool + (STACK_REMAIN_SIZE);\r
+ VmContext.R[0] = (UINT64) ((UINT8*)VmContext.StackPool + STACK_POOL_SIZE);\r
+ VmContext.HighStackBottom = (UINTN) VmContext.R[0];\r
+ VmContext.R[0] -= sizeof (UINTN);\r
+\r
+ \r
+ //\r
+ // Allocate stack space for the interpreter. Then put a magic value\r
+ // at the bottom so we can detect stack corruption.\r
+ //\r
+ PushU64 (&VmContext, (UINT64) VM_STACK_KEY_VALUE);\r
+ VmContext.StackMagicPtr = (UINTN *) (UINTN) VmContext.R[0];\r
+\r
+ //\r
+ // When we thunk to external native code, we copy the last 8 qwords from\r
+ // the EBC stack into the processor registers, and adjust the stack pointer\r
+ // up. If the caller is not passing 8 parameters, then we've moved the\r
+ // stack pointer up into the stack gap. If this happens, then the caller\r
+ // can mess up the stack gap contents (in particular our magic value).\r
+ // Therefore, leave another gap below the magic value. Pick 10 qwords down,\r
+ // just as a starting point.\r
+ //\r
+ VmContext.R[0] -= 10 * sizeof (UINT64);\r
+\r
+ //\r
+ // Align the stack pointer such that after pushing the system table,\r
+ // image handle, and return address on the stack, it's aligned on a 16-byte\r
+ // boundary as required for IPF.\r
+ //\r
+ VmContext.R[0] &= (INT64)~0x0f;\r
+ VmContext.LowStackTop = (UINTN) VmContext.R[0];\r
+ //\r
+ // Simply copy the image handle and system table onto the EBC stack.\r
+ // Greatly simplifies things by not having to spill the args\r
+ //\r
+ PushU64 (&VmContext, (UINT64) SystemTable);\r
+ PushU64 (&VmContext, (UINT64) ImageHandle);\r
+\r
+ //\r
+ // Interpreter assumes 64-bit return address is pushed on the stack.\r
+ // IPF does not do this so pad the stack accordingly. Also, a\r
+ // "return address" is 16 bytes as required for IPF stack alignments.\r
+ //\r
+ PushU64 (&VmContext, (UINT64) 0);\r
+ PushU64 (&VmContext, (UINT64) 0x1234567887654321);\r
+ VmContext.StackRetAddr = (UINT64) VmContext.R[0];\r
+\r
+ //\r
+ // Begin executing the EBC code\r
+ //\r
+ EbcExecute (&VmContext);\r
+\r
+ //\r
+ // Return the value in R[7] unless there was an error\r
+ //\r
+ ReturnEBCStack(StackIndex);\r
+ return (UINT64) VmContext.R[7];\r
+}\r
+\r
+EFI_STATUS\r
+EbcCreateThunks (\r
+ IN EFI_HANDLE ImageHandle,\r
+ IN VOID *EbcEntryPoint,\r
+ OUT VOID **Thunk,\r
+ IN UINT32 Flags\r
+ )\r
+/*++\r
+\r
+Routine Description:\r
+\r
+ Create thunks for an EBC image entry point, or an EBC protocol service.\r
+ \r
+Arguments:\r
+\r
+ ImageHandle - Image handle for the EBC image. If not null, then we're\r
+ creating a thunk for an image entry point.\r
+ EbcEntryPoint - Address of the EBC code that the thunk is to call\r
+ Thunk - Returned thunk we create here\r
+ Flags - Flags indicating options for creating the thunk\r
+ \r
+Returns:\r
+\r
+ Standard EFI status.\r
+ \r
+--*/\r
+{\r
+ UINT8 *Ptr;\r
+ UINT8 *ThunkBase;\r
+ UINT64 Addr;\r
+ UINT64 Code[3]; // Code in a bundle\r
+ UINT64 RegNum; // register number for MOVL\r
+ UINT64 I; // bits of MOVL immediate data\r
+ UINT64 Ic; // bits of MOVL immediate data\r
+ UINT64 Imm5c; // bits of MOVL immediate data\r
+ UINT64 Imm9d; // bits of MOVL immediate data\r
+ UINT64 Imm7b; // bits of MOVL immediate data\r
+ UINT64 Br; // branch register for loading and jumping\r
+ UINT64 *Data64Ptr;\r
+ UINT32 ThunkSize;\r
+ UINT32 Size;\r
+\r
+ //\r
+ // Check alignment of pointer to EBC code, which must always be aligned\r
+ // on a 2-byte boundary.\r
+ //\r
+ if ((UINT32) (UINTN) EbcEntryPoint & 0x01) {\r
+ return EFI_INVALID_PARAMETER;\r
+ }\r
+ //\r
+ // Allocate memory for the thunk. Make the (most likely incorrect) assumption\r
+ // that the returned buffer is not aligned, so round up to the next\r
+ // alignment size.\r
+ //\r
+ Size = EBC_THUNK_SIZE + EBC_THUNK_ALIGNMENT - 1;\r
+ ThunkSize = Size;\r
+ Ptr = AllocatePool (Size);\r
+\r
+ if (Ptr == NULL) {\r
+ return EFI_OUT_OF_RESOURCES;\r
+ }\r
+ //\r
+ // Save the start address of the buffer.\r
+ //\r
+ ThunkBase = Ptr;\r
+\r
+ //\r
+ // Make sure it's aligned for code execution. If not, then\r
+ // round up.\r
+ //\r
+ if ((UINT32) (UINTN) Ptr & (EBC_THUNK_ALIGNMENT - 1)) {\r
+ Ptr = (UINT8 *) (((UINTN) Ptr + (EBC_THUNK_ALIGNMENT - 1)) &~ (UINT64) (EBC_THUNK_ALIGNMENT - 1));\r
+ }\r
+ //\r
+ // Return the pointer to the thunk to the caller to user as the\r
+ // image entry point.\r
+ //\r
+ *Thunk = (VOID *) Ptr;\r
+\r
+ //\r
+ // Clear out the thunk entry\r
+ // ZeroMem(Ptr, Size);\r
+ //\r
+ // For IPF, when you do a call via a function pointer, the function pointer\r
+ // actually points to a function descriptor which consists of a 64-bit\r
+ // address of the function, followed by a 64-bit gp for the function being\r
+ // called. See the the Software Conventions and Runtime Architecture Guide\r
+ // for details.\r
+ // So first off in our thunk, create a descriptor for our actual thunk code.\r
+ // This means we need to create a pointer to the thunk code (which follows\r
+ // the descriptor we're going to create), followed by the gp of the Vm\r
+ // interpret function we're going to eventually execute.\r
+ //\r
+ Data64Ptr = (UINT64 *) Ptr;\r
+\r
+ //\r
+ // Write the function's entry point (which is our thunk code that follows\r
+ // this descriptor we're creating).\r
+ //\r
+ *Data64Ptr = (UINT64) (Data64Ptr + 2);\r
+ //\r
+ // Get the gp from the descriptor for EbcInterpret and stuff it in our thunk\r
+ // descriptor.\r
+ //\r
+ *(Data64Ptr + 1) = *(UINT64 *) ((UINT64 *) (UINTN) EbcInterpret + 1);\r
+ //\r
+ // Advance our thunk data pointer past the descriptor. Since the\r
+ // descriptor consists of 16 bytes, the pointer is still aligned for\r
+ // IPF code execution (on 16-byte boundary).\r
+ //\r
+ Ptr += sizeof (UINT64) * 2;\r
+\r
+ //\r
+ // *************************** MAGIC BUNDLE ********************************\r
+ //\r
+ // Write magic code bundle for: movl r8 = 0xca112ebcca112ebc to help the VM\r
+ // to recognize it is a thunk.\r
+ //\r
+ Addr = (UINT64) 0xCA112EBCCA112EBC;\r
+\r
+ //\r
+ // Now generate the code bytes. First is nop.m 0x0\r
+ //\r
+ Code[0] = OPCODE_NOP;\r
+\r
+ //\r
+ // Next is simply Addr[62:22] (41 bits) of the address\r
+ //\r
+ Code[1] = RShiftU64 (Addr, 22) & 0x1ffffffffff;\r
+\r
+ //\r
+ // Extract bits from the address for insertion into the instruction\r
+ // i = Addr[63:63]\r
+ //\r
+ I = RShiftU64 (Addr, 63) & 0x01;\r
+ //\r
+ // ic = Addr[21:21]\r
+ //\r
+ Ic = RShiftU64 (Addr, 21) & 0x01;\r
+ //\r
+ // imm5c = Addr[20:16] for 5 bits\r
+ //\r
+ Imm5c = RShiftU64 (Addr, 16) & 0x1F;\r
+ //\r
+ // imm9d = Addr[15:7] for 9 bits\r
+ //\r
+ Imm9d = RShiftU64 (Addr, 7) & 0x1FF;\r
+ //\r
+ // imm7b = Addr[6:0] for 7 bits\r
+ //\r
+ Imm7b = Addr & 0x7F;\r
+\r
+ //\r
+ // The EBC entry point will be put into r8, so r8 can be used here\r
+ // temporary. R8 is general register and is auto-serialized.\r
+ //\r
+ RegNum = 8;\r
+\r
+ //\r
+ // Next is jumbled data, including opcode and rest of address\r
+ //\r
+ Code[2] = LShiftU64 (Imm7b, 13);\r
+ Code[2] = Code[2] | LShiftU64 (0x00, 20); // vc\r
+ Code[2] = Code[2] | LShiftU64 (Ic, 21);\r
+ Code[2] = Code[2] | LShiftU64 (Imm5c, 22);\r
+ Code[2] = Code[2] | LShiftU64 (Imm9d, 27);\r
+ Code[2] = Code[2] | LShiftU64 (I, 36);\r
+ Code[2] = Code[2] | LShiftU64 ((UINT64)MOVL_OPCODE, 37);\r
+ Code[2] = Code[2] | LShiftU64 ((RegNum & 0x7F), 6);\r
+\r
+ WriteBundle ((VOID *) Ptr, 0x05, Code[0], Code[1], Code[2]);\r
+\r
+ //\r
+ // *************************** FIRST BUNDLE ********************************\r
+ //\r
+ // Write code bundle for: movl r8 = EBC_ENTRY_POINT so we pass\r
+ // the ebc entry point in to the interpreter function via a processor\r
+ // register.\r
+ // Note -- we could easily change this to pass in a pointer to a structure\r
+ // that contained, among other things, the EBC image's entry point. But\r
+ // for now pass it directly.\r
+ //\r
+ Ptr += 16;\r
+ Addr = (UINT64) EbcEntryPoint;\r
+\r
+ //\r
+ // Now generate the code bytes. First is nop.m 0x0\r
+ //\r
+ Code[0] = OPCODE_NOP;\r
+\r
+ //\r
+ // Next is simply Addr[62:22] (41 bits) of the address\r
+ //\r
+ Code[1] = RShiftU64 (Addr, 22) & 0x1ffffffffff;\r
+\r
+ //\r
+ // Extract bits from the address for insertion into the instruction\r
+ // i = Addr[63:63]\r
+ //\r
+ I = RShiftU64 (Addr, 63) & 0x01;\r
+ //\r
+ // ic = Addr[21:21]\r
+ //\r
+ Ic = RShiftU64 (Addr, 21) & 0x01;\r
+ //\r
+ // imm5c = Addr[20:16] for 5 bits\r
+ //\r
+ Imm5c = RShiftU64 (Addr, 16) & 0x1F;\r
+ //\r
+ // imm9d = Addr[15:7] for 9 bits\r
+ //\r
+ Imm9d = RShiftU64 (Addr, 7) & 0x1FF;\r
+ //\r
+ // imm7b = Addr[6:0] for 7 bits\r
+ //\r
+ Imm7b = Addr & 0x7F;\r
+\r
+ //\r
+ // Put the EBC entry point in r8, which is the location of the return value\r
+ // for functions.\r
+ //\r
+ RegNum = 8;\r
+\r
+ //\r
+ // Next is jumbled data, including opcode and rest of address\r
+ //\r
+ Code[2] = LShiftU64 (Imm7b, 13);\r
+ Code[2] = Code[2] | LShiftU64 (0x00, 20); // vc\r
+ Code[2] = Code[2] | LShiftU64 (Ic, 21);\r
+ Code[2] = Code[2] | LShiftU64 (Imm5c, 22);\r
+ Code[2] = Code[2] | LShiftU64 (Imm9d, 27);\r
+ Code[2] = Code[2] | LShiftU64 (I, 36);\r
+ Code[2] = Code[2] | LShiftU64 ((UINT64)MOVL_OPCODE, 37);\r
+ Code[2] = Code[2] | LShiftU64 ((RegNum & 0x7F), 6);\r
+\r
+ WriteBundle ((VOID *) Ptr, 0x05, Code[0], Code[1], Code[2]);\r
+\r
+ //\r
+ // *************************** NEXT BUNDLE *********************************\r
+ //\r
+ // Write code bundle for:\r
+ // movl rx = offset_of(EbcInterpret|ExecuteEbcImageEntryPoint)\r
+ //\r
+ // Advance pointer to next bundle, then compute the offset from this bundle\r
+ // to the address of the entry point of the interpreter.\r
+ //\r
+ Ptr += 16;\r
+ if (Flags & FLAG_THUNK_ENTRY_POINT) {\r
+ Addr = (UINT64) ExecuteEbcImageEntryPoint;\r
+ } else {\r
+ Addr = (UINT64) EbcInterpret;\r
+ }\r
+ //\r
+ // Indirection on Itanium-based systems\r
+ //\r
+ Addr = *(UINT64 *) Addr;\r
+\r
+ //\r
+ // Now write the code to load the offset into a register\r
+ //\r
+ Code[0] = OPCODE_NOP;\r
+\r
+ //\r
+ // Next is simply Addr[62:22] (41 bits) of the address\r
+ //\r
+ Code[1] = RShiftU64 (Addr, 22) & 0x1ffffffffff;\r
+\r
+ //\r
+ // Extract bits from the address for insertion into the instruction\r
+ // i = Addr[63:63]\r
+ //\r
+ I = RShiftU64 (Addr, 63) & 0x01;\r
+ //\r
+ // ic = Addr[21:21]\r
+ //\r
+ Ic = RShiftU64 (Addr, 21) & 0x01;\r
+ //\r
+ // imm5c = Addr[20:16] for 5 bits\r
+ //\r
+ Imm5c = RShiftU64 (Addr, 16) & 0x1F;\r
+ //\r
+ // imm9d = Addr[15:7] for 9 bits\r
+ //\r
+ Imm9d = RShiftU64 (Addr, 7) & 0x1FF;\r
+ //\r
+ // imm7b = Addr[6:0] for 7 bits\r
+ //\r
+ Imm7b = Addr & 0x7F;\r
+\r
+ //\r
+ // Put it in r31, a scratch register\r
+ //\r
+ RegNum = 31;\r
+\r
+ //\r
+ // Next is jumbled data, including opcode and rest of address\r
+ //\r
+ Code[2] = LShiftU64(Imm7b, 13);\r
+ Code[2] = Code[2] | LShiftU64 (0x00, 20); // vc\r
+ Code[2] = Code[2] | LShiftU64 (Ic, 21);\r
+ Code[2] = Code[2] | LShiftU64 (Imm5c, 22);\r
+ Code[2] = Code[2] | LShiftU64 (Imm9d, 27);\r
+ Code[2] = Code[2] | LShiftU64 (I, 36);\r
+ Code[2] = Code[2] | LShiftU64 ((UINT64)MOVL_OPCODE, 37);\r
+ Code[2] = Code[2] | LShiftU64 ((RegNum & 0x7F), 6);\r
+\r
+ WriteBundle ((VOID *) Ptr, 0x05, Code[0], Code[1], Code[2]);\r
+\r
+ //\r
+ // *************************** NEXT BUNDLE *********************************\r
+ //\r
+ // Load branch register with EbcInterpret() function offset from the bundle\r
+ // address: mov b6 = RegNum\r
+ //\r
+ // See volume 3 page 4-29 of the Arch. Software Developer's Manual.\r
+ //\r
+ // Advance pointer to next bundle\r
+ //\r
+ Ptr += 16;\r
+ Code[0] = OPCODE_NOP;\r
+ Code[1] = OPCODE_NOP;\r
+ Code[2] = OPCODE_MOV_BX_RX;\r
+\r
+ //\r
+ // Pick a branch register to use. Then fill in the bits for the branch\r
+ // register and user register (same user register as previous bundle).\r
+ //\r
+ Br = 6;\r
+ Code[2] |= LShiftU64 (Br, 6);\r
+ Code[2] |= LShiftU64 (RegNum, 13);\r
+ WriteBundle ((VOID *) Ptr, 0x0d, Code[0], Code[1], Code[2]);\r
+\r
+ //\r
+ // *************************** NEXT BUNDLE *********************************\r
+ //\r
+ // Now do the branch: (p0) br.cond.sptk.few b6\r
+ //\r
+ // Advance pointer to next bundle.\r
+ // Fill in the bits for the branch register (same reg as previous bundle)\r
+ //\r
+ Ptr += 16;\r
+ Code[0] = OPCODE_NOP;\r
+ Code[1] = OPCODE_NOP;\r
+ Code[2] = OPCODE_BR_COND_SPTK_FEW;\r
+ Code[2] |= LShiftU64 (Br, 13);\r
+ WriteBundle ((VOID *) Ptr, 0x1d, Code[0], Code[1], Code[2]);\r
+\r
+ //\r
+ // Add the thunk to our list of allocated thunks so we can do some cleanup\r
+ // when the image is unloaded. Do this last since the Add function flushes\r
+ // the instruction cache for us.\r
+ //\r
+ EbcAddImageThunk (ImageHandle, (VOID *) ThunkBase, ThunkSize);\r
+\r
+ //\r
+ // Done\r
+ //\r
+ return EFI_SUCCESS;\r
+}\r
+\r
+STATIC\r
+EFI_STATUS\r
+WriteBundle (\r
+ IN VOID *MemPtr,\r
+ IN UINT8 Template,\r
+ IN UINT64 Slot0,\r
+ IN UINT64 Slot1,\r
+ IN UINT64 Slot2\r
+ )\r
+/*++\r
+\r
+Routine Description:\r
+\r
+ Given raw bytes of Itanium based code, format them into a bundle and\r
+ write them out.\r
+ \r
+Arguments:\r
+\r
+ MemPtr - pointer to memory location to write the bundles to\r
+ Template - 5-bit template\r
+ Slot0-2 - instruction slot data for the bundle\r
+\r
+Returns:\r
+\r
+ EFI_INVALID_PARAMETER - Pointer is not aligned\r
+ - No more than 5 bits in template\r
+ - More than 41 bits used in code\r
+ EFI_SUCCESS - All data is written.\r
+\r
+--*/\r
+{\r
+ UINT8 *BPtr;\r
+ UINT32 Index;\r
+ UINT64 Low64;\r
+ UINT64 High64;\r
+\r
+ //\r
+ // Verify pointer is aligned\r
+ //\r
+ if ((UINT64) MemPtr & 0xF) {\r
+ return EFI_INVALID_PARAMETER;\r
+ }\r
+ //\r
+ // Verify no more than 5 bits in template\r
+ //\r
+ if (Template &~0x1F) {\r
+ return EFI_INVALID_PARAMETER;\r
+ }\r
+ //\r
+ // Verify max of 41 bits used in code\r
+ //\r
+ if ((Slot0 | Slot1 | Slot2) &~0x1ffffffffff) {\r
+ return EFI_INVALID_PARAMETER;\r
+ }\r
+\r
+ Low64 = LShiftU64 (Slot1, 46);\r
+ Low64 = Low64 | LShiftU64 (Slot0, 5) | Template;\r
+\r
+ High64 = RShiftU64 (Slot1, 18);\r
+ High64 = High64 | LShiftU64 (Slot2, 23);\r
+\r
+ //\r
+ // Now write it all out\r
+ //\r
+ BPtr = (UINT8 *) MemPtr;\r
+ for (Index = 0; Index < 8; Index++) {\r
+ *BPtr = (UINT8) Low64;\r
+ Low64 = RShiftU64 (Low64, 8);\r
+ BPtr++;\r
+ }\r
+\r
+ for (Index = 0; Index < 8; Index++) {\r
+ *BPtr = (UINT8) High64;\r
+ High64 = RShiftU64 (High64, 8);\r
+ BPtr++;\r
+ }\r
+\r
+ return EFI_SUCCESS;\r
+}\r
+\r
+VOID\r
+EbcLLCALLEX (\r
+ IN VM_CONTEXT *VmPtr,\r
+ IN UINTN FuncAddr,\r
+ IN UINTN NewStackPointer,\r
+ IN VOID *FramePtr,\r
+ IN UINT8 Size\r
+ )\r
+/*++\r
+\r
+Routine Description:\r
+\r
+ This function is called to execute an EBC CALLEX instruction. \r
+ The function check the callee's content to see whether it is common native\r
+ code or a thunk to another piece of EBC code.\r
+ If the callee is common native code, use EbcLLCAllEXASM to manipulate,\r
+ otherwise, set the VM->IP to target EBC code directly to avoid another VM\r
+ be startup which cost time and stack space.\r
+ \r
+Arguments:\r
+\r
+ VmPtr - Pointer to a VM context.\r
+ FuncAddr - Callee's address\r
+ NewStackPointer - New stack pointer after the call\r
+ FramePtr - New frame pointer after the call\r
+ Size - The size of call instruction\r
+\r
+Returns:\r
+\r
+ None.\r
+ \r
+--*/\r
+{\r
+ UINTN IsThunk;\r
+ UINTN TargetEbcAddr;\r
+ UINTN CodeOne18;\r
+ UINTN CodeOne23;\r
+ UINTN CodeTwoI;\r
+ UINTN CodeTwoIc;\r
+ UINTN CodeTwo7b;\r
+ UINTN CodeTwo5c;\r
+ UINTN CodeTwo9d;\r
+ UINTN CalleeAddr;\r
+\r
+ IsThunk = 1;\r
+ TargetEbcAddr = 0;\r
+\r
+ //\r
+ // FuncAddr points to the descriptor of the target instructions.\r
+ //\r
+ CalleeAddr = *((UINT64 *)FuncAddr);\r
+\r
+ //\r
+ // Processor specific code to check whether the callee is a thunk to EBC.\r
+ //\r
+ if (*((UINT64 *)CalleeAddr) != 0xBCCA000100000005) {\r
+ IsThunk = 0;\r
+ goto Action;\r
+ }\r
+ if (*((UINT64 *)CalleeAddr + 1) != 0x697623C1004A112E) {\r
+ IsThunk = 0;\r
+ goto Action;\r
+ }\r
+\r
+ CodeOne18 = RShiftU64 (*((UINT64 *)CalleeAddr + 2), 46) & 0x3FFFF;\r
+ CodeOne23 = (*((UINT64 *)CalleeAddr + 3)) & 0x7FFFFF;\r
+ CodeTwoI = RShiftU64 (*((UINT64 *)CalleeAddr + 3), 59) & 0x1;\r
+ CodeTwoIc = RShiftU64 (*((UINT64 *)CalleeAddr + 3), 44) & 0x1;\r
+ CodeTwo7b = RShiftU64 (*((UINT64 *)CalleeAddr + 3), 36) & 0x7F;\r
+ CodeTwo5c = RShiftU64 (*((UINT64 *)CalleeAddr + 3), 45) & 0x1F;\r
+ CodeTwo9d = RShiftU64 (*((UINT64 *)CalleeAddr + 3), 50) & 0x1FF;\r
+\r
+ TargetEbcAddr = CodeTwo7b;\r
+ TargetEbcAddr = TargetEbcAddr | LShiftU64 (CodeTwo9d, 7);\r
+ TargetEbcAddr = TargetEbcAddr | LShiftU64 (CodeTwo5c, 16);\r
+ TargetEbcAddr = TargetEbcAddr | LShiftU64 (CodeTwoIc, 21);\r
+ TargetEbcAddr = TargetEbcAddr | LShiftU64 (CodeOne18, 22);\r
+ TargetEbcAddr = TargetEbcAddr | LShiftU64 (CodeOne23, 40);\r
+ TargetEbcAddr = TargetEbcAddr | LShiftU64 (CodeTwoI, 63);\r
+\r
+Action:\r
+ if (IsThunk == 1){\r
+ //\r
+ // The callee is a thunk to EBC, adjust the stack pointer down 16 bytes and\r
+ // put our return address and frame pointer on the VM stack.\r
+ // Then set the VM's IP to new EBC code.\r
+ //\r
+ VmPtr->R[0] -= 8;\r
+ VmWriteMemN (VmPtr, (UINTN) VmPtr->R[0], (UINTN) FramePtr);\r
+ VmPtr->FramePtr = (VOID *) (UINTN) VmPtr->R[0];\r
+ VmPtr->R[0] -= 8;\r
+ VmWriteMem64 (VmPtr, (UINTN) VmPtr->R[0], (UINT64) (VmPtr->Ip + Size));\r
+\r
+ VmPtr->Ip = (VMIP) (UINTN) TargetEbcAddr;\r
+ } else {\r
+ //\r
+ // The callee is not a thunk to EBC, call native code.\r
+ //\r
+ EbcLLCALLEXNative (FuncAddr, NewStackPointer, FramePtr);\r
+\r
+ //\r
+ // Get return value and advance the IP.\r
+ //\r
+ VmPtr->R[7] = EbcLLGetReturnValue ();\r
+ VmPtr->Ip += Size;\r
+ }\r
+}\r