mirror of https://github.com/acidanthera/audk.git
542 lines
21 KiB
Python
Executable File
542 lines
21 KiB
Python
Executable File
#!/usr/bin/python
|
|
|
|
#
|
|
# Copyright 2014 Apple Inc. All righes reserved.
|
|
#
|
|
# SPDX-License-Identifier: BSD-2-Clause-Patent
|
|
#
|
|
|
|
import lldb
|
|
import os
|
|
import uuid
|
|
import string
|
|
import commands
|
|
import optparse
|
|
import shlex
|
|
|
|
guid_dict = {}
|
|
|
|
|
|
def EFI_GUID_TypeSummary (valobj,internal_dict):
|
|
""" Type summary for EFI GUID, print C Name if known
|
|
"""
|
|
# typedef struct {
|
|
# UINT32 Data1;
|
|
# UINT16 Data2;
|
|
# UINT16 Data3;
|
|
# UINT8 Data4[8];
|
|
# } EFI_GUID;
|
|
SBError = lldb.SBError()
|
|
|
|
data1_val = valobj.GetChildMemberWithName('Data1')
|
|
data1 = data1_val.GetValueAsUnsigned(0)
|
|
data2_val = valobj.GetChildMemberWithName('Data2')
|
|
data2 = data2_val.GetValueAsUnsigned(0)
|
|
data3_val = valobj.GetChildMemberWithName('Data3')
|
|
data3 = data3_val.GetValueAsUnsigned(0)
|
|
str = "%x-%x-%x-" % (data1, data2, data3)
|
|
|
|
data4_val = valobj.GetChildMemberWithName('Data4')
|
|
for i in range (data4_val.num_children):
|
|
if i == 2:
|
|
str +='-'
|
|
str += "%02x" % data4_val.GetChildAtIndex(i).data.GetUnsignedInt8(SBError, 0)
|
|
|
|
return guid_dict.get (str.upper(), '')
|
|
|
|
|
|
|
|
EFI_STATUS_Dict = {
|
|
(0x8000000000000000 | 1): "Load Error",
|
|
(0x8000000000000000 | 2): "Invalid Parameter",
|
|
(0x8000000000000000 | 3): "Unsupported",
|
|
(0x8000000000000000 | 4): "Bad Buffer Size",
|
|
(0x8000000000000000 | 5): "Buffer Too Small",
|
|
(0x8000000000000000 | 6): "Not Ready",
|
|
(0x8000000000000000 | 7): "Device Error",
|
|
(0x8000000000000000 | 8): "Write Protected",
|
|
(0x8000000000000000 | 9): "Out of Resources",
|
|
(0x8000000000000000 | 10): "Volume Corrupt",
|
|
(0x8000000000000000 | 11): "Volume Full",
|
|
(0x8000000000000000 | 12): "No Media",
|
|
(0x8000000000000000 | 13): "Media changed",
|
|
(0x8000000000000000 | 14): "Not Found",
|
|
(0x8000000000000000 | 15): "Access Denied",
|
|
(0x8000000000000000 | 16): "No Response",
|
|
(0x8000000000000000 | 17): "No mapping",
|
|
(0x8000000000000000 | 18): "Time out",
|
|
(0x8000000000000000 | 19): "Not started",
|
|
(0x8000000000000000 | 20): "Already started",
|
|
(0x8000000000000000 | 21): "Aborted",
|
|
(0x8000000000000000 | 22): "ICMP Error",
|
|
(0x8000000000000000 | 23): "TFTP Error",
|
|
(0x8000000000000000 | 24): "Protocol Error",
|
|
|
|
0 : "Success",
|
|
1 : "Warning Unknown Glyph",
|
|
2 : "Warning Delete Failure",
|
|
3 : "Warning Write Failure",
|
|
4 : "Warning Buffer Too Small",
|
|
|
|
(0x80000000 | 1): "Load Error",
|
|
(0x80000000 | 2): "Invalid Parameter",
|
|
(0x80000000 | 3): "Unsupported",
|
|
(0x80000000 | 4): "Bad Buffer Size",
|
|
(0x80000000 | 5): "Buffer Too Small",
|
|
(0x80000000 | 6): "Not Ready",
|
|
(0x80000000 | 7): "Device Error",
|
|
(0x80000000 | 8): "Write Protected",
|
|
(0x80000000 | 9): "Out of Resources",
|
|
(0x80000000 | 10): "Volume Corrupt",
|
|
(0x80000000 | 11): "Volume Full",
|
|
(0x80000000 | 12): "No Media",
|
|
(0x80000000 | 13): "Media changed",
|
|
(0x80000000 | 14): "Not Found",
|
|
(0x80000000 | 15): "Access Denied",
|
|
(0x80000000 | 16): "No Response",
|
|
(0x80000000 | 17): "No mapping",
|
|
(0x80000000 | 18): "Time out",
|
|
(0x80000000 | 19): "Not started",
|
|
(0x80000000 | 20): "Already started",
|
|
(0x80000000 | 21): "Aborted",
|
|
(0x80000000 | 22): "ICMP Error",
|
|
(0x80000000 | 23): "TFTP Error",
|
|
(0x80000000 | 24): "Protocol Error",
|
|
}
|
|
|
|
def EFI_STATUS_TypeSummary (valobj,internal_dict):
|
|
#
|
|
# Return summary string for EFI_STATUS from dictionary
|
|
#
|
|
Status = valobj.GetValueAsUnsigned(0)
|
|
return EFI_STATUS_Dict.get (Status, '')
|
|
|
|
|
|
def EFI_TPL_TypeSummary (valobj,internal_dict):
|
|
#
|
|
# Return TPL values
|
|
#
|
|
|
|
if valobj.TypeIsPointerType():
|
|
return ""
|
|
|
|
Tpl = valobj.GetValueAsUnsigned(0)
|
|
if Tpl < 4:
|
|
Str = "%d" % Tpl
|
|
elif Tpl == 6:
|
|
Str = "TPL_DRIVER (Obsolete Concept in edk2)"
|
|
elif Tpl < 8:
|
|
Str = "TPL_APPLICATION"
|
|
if Tpl - 4 > 0:
|
|
Str += " + " + "%d" % (Tpl - 4)
|
|
elif Tpl < 16:
|
|
Str = "TPL_CALLBACK"
|
|
if Tpl - 8 > 0:
|
|
Str += " + " + "%d" % (Tpl - 4)
|
|
elif Tpl < 31:
|
|
Str = "TPL_NOTIFY"
|
|
if Tpl - 16 > 0:
|
|
Str += " + " + "%d" % (Tpl - 4)
|
|
elif Tpl == 31:
|
|
Str = "TPL_HIGH_LEVEL"
|
|
else:
|
|
Str = "Invalid TPL"
|
|
|
|
return Str
|
|
|
|
|
|
def CHAR16_TypeSummary (valobj,internal_dict):
|
|
#
|
|
# Display EFI CHAR16 'unsigned short' as string
|
|
#
|
|
SBError = lldb.SBError()
|
|
Str = ''
|
|
if valobj.TypeIsPointerType():
|
|
if valobj.GetValueAsUnsigned () == 0:
|
|
return "NULL"
|
|
|
|
# CHAR16 * max string size 1024
|
|
for i in range (1024):
|
|
Char = valobj.GetPointeeData(i,1).GetUnsignedInt16(SBError, 0)
|
|
if SBError.fail or Char == 0:
|
|
break
|
|
Str += unichr (Char)
|
|
Str = 'L"' + Str + '"'
|
|
return Str.encode ('utf-8', 'replace')
|
|
|
|
if valobj.num_children == 0:
|
|
# CHAR16
|
|
if chr (valobj.unsigned) in string.printable:
|
|
Str = "L'" + unichr (valobj.unsigned) + "'"
|
|
return Str.encode ('utf-8', 'replace')
|
|
else:
|
|
# CHAR16 []
|
|
for i in range (valobj.num_children):
|
|
Char = valobj.GetChildAtIndex(i).data.GetUnsignedInt16(SBError, 0)
|
|
if Char == 0:
|
|
break
|
|
Str += unichr (Char)
|
|
Str = 'L"' + Str + '"'
|
|
return Str.encode ('utf-8', 'replace')
|
|
|
|
return Str
|
|
|
|
def CHAR8_TypeSummary (valobj,internal_dict):
|
|
#
|
|
# Display EFI CHAR8 'signed char' as string
|
|
# unichr() is used as a junk string can produce an error message like this:
|
|
# UnicodeEncodeError: 'ascii' codec can't encode character u'\x90' in position 1: ordinal not in range(128)
|
|
#
|
|
SBError = lldb.SBError()
|
|
Str = ''
|
|
if valobj.TypeIsPointerType():
|
|
if valobj.GetValueAsUnsigned () == 0:
|
|
return "NULL"
|
|
|
|
# CHAR8 * max string size 1024
|
|
for i in range (1024):
|
|
Char = valobj.GetPointeeData(i,1).GetUnsignedInt8(SBError, 0)
|
|
if SBError.fail or Char == 0:
|
|
break
|
|
Str += unichr (Char)
|
|
Str = '"' + Str + '"'
|
|
return Str.encode ('utf-8', 'replace')
|
|
|
|
if valobj.num_children == 0:
|
|
# CHAR8
|
|
if chr (valobj.unsigned) in string.printable:
|
|
Str = '"' + unichr (valobj.unsigned) + '"'
|
|
return Str.encode ('utf-8', 'replace')
|
|
else:
|
|
# CHAR8 []
|
|
for i in range (valobj.num_children):
|
|
Char = valobj.GetChildAtIndex(i).data.GetUnsignedInt8(SBError, 0)
|
|
if Char == 0:
|
|
break
|
|
Str += unichr (Char)
|
|
Str = '"' + Str + '"'
|
|
return Str.encode ('utf-8', 'replace')
|
|
|
|
return Str
|
|
|
|
device_path_dict = {
|
|
(0x01, 0x01): "PCI_DEVICE_PATH",
|
|
(0x01, 0x02): "PCCARD_DEVICE_PATH",
|
|
(0x01, 0x03): "MEMMAP_DEVICE_PATH",
|
|
(0x01, 0x04): "VENDOR_DEVICE_PATH",
|
|
(0x01, 0x05): "CONTROLLER_DEVICE_PATH",
|
|
(0x02, 0x01): "ACPI_HID_DEVICE_PATH",
|
|
(0x02, 0x02): "ACPI_EXTENDED_HID_DEVICE_PATH",
|
|
(0x02, 0x03): "ACPI_ADR_DEVICE_PATH",
|
|
(0x03, 0x01): "ATAPI_DEVICE_PATH",
|
|
(0x03, 0x12): "SATA_DEVICE_PATH",
|
|
(0x03, 0x02): "SCSI_DEVICE_PATH",
|
|
(0x03, 0x03): "FIBRECHANNEL_DEVICE_PATH",
|
|
(0x03, 0x04): "F1394_DEVICE_PATH",
|
|
(0x03, 0x05): "USB_DEVICE_PATH",
|
|
(0x03, 0x0f): "USB_CLASS_DEVICE_PATH",
|
|
(0x03, 0x10): "FW_SBP2_UNIT_LUN_DEVICE_PATH",
|
|
(0x03, 0x11): "DEVICE_LOGICAL_UNIT_DEVICE_PATH",
|
|
(0x03, 0x06): "I2O_DEVICE_PATH",
|
|
(0x03, 0x0b): "MAC_ADDR_DEVICE_PATH",
|
|
(0x03, 0x0c): "IPv4_DEVICE_PATH",
|
|
(0x03, 0x09): "INFINIBAND_DEVICE_PATH",
|
|
(0x03, 0x0e): "UART_DEVICE_PATH",
|
|
(0x03, 0x0a): "VENDOR_DEVICE_PATH",
|
|
(0x03, 0x13): "ISCSI_DEVICE_PATH",
|
|
(0x04, 0x01): "HARDDRIVE_DEVICE_PATH",
|
|
(0x04, 0x02): "CDROM_DEVICE_PATH",
|
|
(0x04, 0x03): "VENDOR_DEVICE_PATH",
|
|
(0x04, 0x04): "FILEPATH_DEVICE_PATH",
|
|
(0x04, 0x05): "MEDIA_PROTOCOL_DEVICE_PATH",
|
|
(0x05, 0x01): "BBS_BBS_DEVICE_PATH",
|
|
(0x7F, 0xFF): "EFI_DEVICE_PATH_PROTOCOL",
|
|
(0xFF, 0xFF): "EFI_DEVICE_PATH_PROTOCOL",
|
|
}
|
|
|
|
def EFI_DEVICE_PATH_PROTOCOL_TypeSummary (valobj,internal_dict):
|
|
#
|
|
#
|
|
#
|
|
if valobj.TypeIsPointerType():
|
|
# EFI_DEVICE_PATH_PROTOCOL *
|
|
return ""
|
|
|
|
Str = ""
|
|
if valobj.num_children == 3:
|
|
# EFI_DEVICE_PATH_PROTOCOL
|
|
Type = valobj.GetChildMemberWithName('Type').unsigned
|
|
SubType = valobj.GetChildMemberWithName('SubType').unsigned
|
|
if (Type, SubType) in device_path_dict:
|
|
TypeStr = device_path_dict[Type, SubType]
|
|
else:
|
|
TypeStr = ""
|
|
|
|
LenLow = valobj.GetChildMemberWithName('Length').GetChildAtIndex(0).unsigned
|
|
LenHigh = valobj.GetChildMemberWithName('Length').GetChildAtIndex(1).unsigned
|
|
Len = LenLow + (LenHigh >> 8)
|
|
|
|
Address = long ("%d" % valobj.addr)
|
|
if (Address == lldb.LLDB_INVALID_ADDRESS):
|
|
# Need to reserach this, it seems to be the nested struct case
|
|
ExprStr = ""
|
|
elif (Type & 0x7f == 0x7f):
|
|
ExprStr = "End Device Path" if SubType == 0xff else "End This Instance"
|
|
else:
|
|
ExprStr = "expr *(%s *)0x%08x" % (TypeStr, Address)
|
|
|
|
Str = " {\n"
|
|
Str += " (UINT8) Type = 0x%02x // %s\n" % (Type, "END" if (Type & 0x7f == 0x7f) else "")
|
|
Str += " (UINT8) SubType = 0x%02x // %s\n" % (SubType, ExprStr)
|
|
Str += " (UINT8 [2]) Length = { // 0x%04x (%d) bytes\n" % (Len, Len)
|
|
Str += " (UINT8) [0] = 0x%02x\n" % LenLow
|
|
Str += " (UINT8) [1] = 0x%02x\n" % LenHigh
|
|
Str += " }\n"
|
|
if (Type & 0x7f == 0x7f) and (SubType == 0xff):
|
|
pass
|
|
elif ExprStr != "":
|
|
NextNode = Address + Len
|
|
Str += "// Next node 'expr *(EFI_DEVICE_PATH_PROTOCOL *)0x%08x'\n" % NextNode
|
|
|
|
return Str
|
|
|
|
|
|
|
|
def TypePrintFormating(debugger):
|
|
#
|
|
# Set the default print formating for EFI types in lldb.
|
|
# seems lldb defaults to decimal.
|
|
#
|
|
category = debugger.GetDefaultCategory()
|
|
FormatBool = lldb.SBTypeFormat(lldb.eFormatBoolean)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("BOOLEAN"), FormatBool)
|
|
|
|
FormatHex = lldb.SBTypeFormat(lldb.eFormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("UINT64"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("INT64"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("UINT32"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("INT32"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("UINT16"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("INT16"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("UINT8"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("INT8"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("UINTN"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("INTN"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("CHAR8"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("CHAR16"), FormatHex)
|
|
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("EFI_PHYSICAL_ADDRESS"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("PHYSICAL_ADDRESS"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("EFI_STATUS"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("EFI_TPL"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("EFI_LBA"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("EFI_BOOT_MODE"), FormatHex)
|
|
category.AddTypeFormat(lldb.SBTypeNameSpecifier("EFI_FV_FILETYPE"), FormatHex)
|
|
|
|
#
|
|
# Smart type printing for EFI
|
|
#
|
|
debugger.HandleCommand("type summary add EFI_GUID --python-function lldbefi.EFI_GUID_TypeSummary")
|
|
debugger.HandleCommand("type summary add EFI_STATUS --python-function lldbefi.EFI_STATUS_TypeSummary")
|
|
debugger.HandleCommand("type summary add EFI_TPL --python-function lldbefi.EFI_TPL_TypeSummary")
|
|
debugger.HandleCommand("type summary add EFI_DEVICE_PATH_PROTOCOL --python-function lldbefi.EFI_DEVICE_PATH_PROTOCOL_TypeSummary")
|
|
|
|
debugger.HandleCommand("type summary add CHAR16 --python-function lldbefi.CHAR16_TypeSummary")
|
|
debugger.HandleCommand('type summary add --regex "CHAR16 \[[0-9]+\]" --python-function lldbefi.CHAR16_TypeSummary')
|
|
debugger.HandleCommand("type summary add CHAR8 --python-function lldbefi.CHAR8_TypeSummary")
|
|
debugger.HandleCommand('type summary add --regex "CHAR8 \[[0-9]+\]" --python-function lldbefi.CHAR8_TypeSummary')
|
|
|
|
debugger.HandleCommand(
|
|
'setting set frame-format "frame #${frame.index}: ${frame.pc}'
|
|
'{ ${module.file.basename}{:${function.name}()${function.pc-offset}}}'
|
|
'{ at ${line.file.fullpath}:${line.number}}\n"'
|
|
)
|
|
|
|
gEmulatorBreakWorkaroundNeeded = True
|
|
|
|
def LoadEmulatorEfiSymbols(frame, bp_loc , internal_dict):
|
|
#
|
|
# This is an lldb breakpoint script, and assumes the breakpoint is on a
|
|
# function with the same prototype as SecGdbScriptBreak(). The
|
|
# argument names are important as lldb looks them up.
|
|
#
|
|
# VOID
|
|
# SecGdbScriptBreak (
|
|
# char *FileName,
|
|
# int FileNameLength,
|
|
# long unsigned int LoadAddress,
|
|
# int AddSymbolFlag
|
|
# )
|
|
# {
|
|
# return;
|
|
# }
|
|
#
|
|
# When the emulator loads a PE/COFF image, it calls the stub function with
|
|
# the filename of the symbol file, the length of the FileName, the
|
|
# load address and a flag to indicate if this is a load or unload operation
|
|
#
|
|
global gEmulatorBreakWorkaroundNeeded
|
|
|
|
if gEmulatorBreakWorkaroundNeeded:
|
|
# turn off lldb debug prints on SIGALRM (EFI timer tick)
|
|
frame.thread.process.target.debugger.HandleCommand("process handle SIGALRM -n false")
|
|
gEmulatorBreakWorkaroundNeeded = False
|
|
|
|
# Convert C string to Python string
|
|
Error = lldb.SBError()
|
|
FileNamePtr = frame.FindVariable ("FileName").GetValueAsUnsigned()
|
|
FileNameLen = frame.FindVariable ("FileNameLength").GetValueAsUnsigned()
|
|
|
|
FileName = frame.thread.process.ReadCStringFromMemory (FileNamePtr, FileNameLen, Error)
|
|
if not Error.Success():
|
|
print "!ReadCStringFromMemory() did not find a %d byte C string at %x" % (FileNameLen, FileNamePtr)
|
|
# make breakpoint command contiue
|
|
return False
|
|
|
|
debugger = frame.thread.process.target.debugger
|
|
if frame.FindVariable ("AddSymbolFlag").GetValueAsUnsigned() == 1:
|
|
LoadAddress = frame.FindVariable ("LoadAddress").GetValueAsUnsigned() - 0x240
|
|
|
|
debugger.HandleCommand ("target modules add %s" % FileName)
|
|
print "target modules load --slid 0x%x %s" % (LoadAddress, FileName)
|
|
debugger.HandleCommand ("target modules load --slide 0x%x --file %s" % (LoadAddress, FileName))
|
|
else:
|
|
target = debugger.GetSelectedTarget()
|
|
for SBModule in target.module_iter():
|
|
ModuleName = SBModule.GetFileSpec().GetDirectory() + '/'
|
|
ModuleName += SBModule.GetFileSpec().GetFilename()
|
|
if FileName == ModuleName or FileName == SBModule.GetFileSpec().GetFilename():
|
|
target.ClearModuleLoadAddress (SBModule)
|
|
if not target.RemoveModule (SBModule):
|
|
print "!lldb.target.RemoveModule (%s) FAILED" % SBModule
|
|
|
|
# make breakpoint command contiue
|
|
return False
|
|
|
|
def GuidToCStructStr (guid, Name=False):
|
|
#
|
|
# Convert a 16-byte bytesarry (or bytearray compat object) to C guid string
|
|
# { 0xB402621F, 0xA940, 0x1E4A, { 0x86, 0x6B, 0x4D, 0xC9, 0x16, 0x2B, 0x34, 0x7C } }
|
|
#
|
|
# Name=True means lookup name in GuidNameDict and us it if you find it
|
|
#
|
|
|
|
if not isinstance (guid, bytearray):
|
|
# convert guid object to UUID, and UUID to bytearray
|
|
Uuid = uuid.UUID(guid)
|
|
guid = bytearray (Uuid.bytes_le)
|
|
|
|
return "{ 0x%02.2X%02.2X%02.2X%02.2X, 0x%02.2X%02.2X, 0x%02.2X%02.2X, { 0x%02.2X, 0x%02.2X, 0x%02.2X, 0x%02.2X, 0x%02.2X, 0x%02.2X, 0x%02.2X, 0x%02.2X } }" % \
|
|
(guid[3], guid[2], guid[1], guid[0], guid[5], guid[4], guid[7], guid[6], guid[8], guid[9], guid[10], guid[11], guid[12], guid[13], guid[14], guid[15])
|
|
|
|
def ParseGuidString(GuidStr):
|
|
#
|
|
# Error check and convert C Guid init to string
|
|
# ParseGuidString("49152E77-1ADA-4764-B7A2-7AFEFED95E8B")
|
|
# ParseGuidString("{ 0xBA24B391, 0x73FD, 0xC54C, { 0x9E, 0xAF, 0x0C, 0xA7, 0x8A, 0x35, 0x46, 0xD1 } }")
|
|
#
|
|
|
|
if "{" in GuidStr :
|
|
# convert C form "{ 0xBA24B391, 0x73FD, 0xC54C, { 0x9E, 0xAF, 0x0C, 0xA7, 0x8A, 0x35, 0x46, 0xD1 } }"
|
|
# to string form BA24B391-73FD-C54C-9EAF-0CA78A3546D1
|
|
# make a list of Hex numbers like: ['0xBA24B391', '0x73FD', '0xC54C', '0x9E', '0xAF', '0x0C', '0xA7', '0x8A', '0x35', '0x46', '0xD1']
|
|
Hex = ''.join(x for x in GuidStr if x not in '{,}').split()
|
|
Str = "%08X-%04X-%04X-%02.2X%02.2X-%02.2X%02.2X%02.2X%02.2X%02.2X%02.2X" % \
|
|
(int(Hex[0], 0), int(Hex[1], 0), int(Hex[2], 0), int(Hex[3], 0), int(Hex[4], 0), \
|
|
int(Hex[5], 0), int(Hex[6], 0), int(Hex[7], 0), int(Hex[8], 0), int(Hex[9], 0), int(Hex[10], 0))
|
|
elif GuidStr.count('-') == 4:
|
|
# validate "49152E77-1ADA-4764-B7A2-7AFEFED95E8B" form
|
|
Check = "%s" % str(uuid.UUID(GuidStr)).upper()
|
|
if GuidStr.upper() == Check:
|
|
Str = GuidStr.upper()
|
|
else:
|
|
Ste = ""
|
|
else:
|
|
Str = ""
|
|
|
|
return Str
|
|
|
|
|
|
def create_guid_options():
|
|
usage = "usage: %prog [data]"
|
|
description='''lookup EFI_GUID by CName, C struct, or GUID string and print out all three.
|
|
'''
|
|
parser = optparse.OptionParser(description=description, prog='guid',usage=usage)
|
|
return parser
|
|
|
|
def efi_guid_command(debugger, command, result, dict):
|
|
# Use the Shell Lexer to properly parse up command options just like a
|
|
# shell would
|
|
command_args = shlex.split(command)
|
|
parser = create_guid_options()
|
|
try:
|
|
(options, args) = parser.parse_args(command_args)
|
|
if len(args) >= 1:
|
|
if args[0] == "{":
|
|
# caller forgot to quote the string"
|
|
# mark arg[0] a string containing all args[n]
|
|
args[0] = ' '.join(args)
|
|
GuidStr = ParseGuidString (args[0])
|
|
if GuidStr == "":
|
|
# return Key of GuidNameDict for value args[0]
|
|
GuidStr = [Key for Key, Value in guid_dict.iteritems() if Value == args[0]][0]
|
|
GuidStr = GuidStr.upper()
|
|
except:
|
|
# if you don't handle exceptions, passing an incorrect argument to the OptionParser will cause LLDB to exit
|
|
# (courtesy of OptParse dealing with argument errors by throwing SystemExit)
|
|
result.SetError ("option parsing failed")
|
|
return
|
|
|
|
|
|
if len(args) >= 1:
|
|
if GuidStr in guid_dict:
|
|
print "%s = %s" % (guid_dict[GuidStr], GuidStr)
|
|
print "%s = %s" % (guid_dict[GuidStr], GuidToCStructStr (GuidStr))
|
|
else:
|
|
print GuidStr
|
|
else:
|
|
# dump entire dictionary
|
|
width = max(len(v) for k,v in guid_dict.iteritems())
|
|
for value in sorted(guid_dict, key=guid_dict.get):
|
|
print '%-*s %s %s' % (width, guid_dict[value], value, GuidToCStructStr(value))
|
|
|
|
return
|
|
|
|
|
|
#
|
|
########## Code that runs when this script is imported into LLDB ###########
|
|
#
|
|
def __lldb_init_module (debugger, internal_dict):
|
|
# This initializer is being run from LLDB in the embedded command interpreter
|
|
# Make the options so we can generate the help text for the new LLDB
|
|
# command line command prior to registering it with LLDB below
|
|
|
|
global guid_dict
|
|
|
|
# Source Guid.xref file if we can find it
|
|
inputfile = os.getcwd()
|
|
inputfile += os.sep + os.pardir + os.sep + 'FV' + os.sep + 'Guid.xref'
|
|
with open(inputfile) as f:
|
|
for line in f:
|
|
data = line.split(' ')
|
|
if len(data) >= 2:
|
|
guid_dict[data[0].upper()] = data[1].strip('\n')
|
|
|
|
# init EFI specific type formaters
|
|
TypePrintFormating (debugger)
|
|
|
|
|
|
# add guid command
|
|
parser = create_guid_options()
|
|
efi_guid_command.__doc__ = parser.format_help()
|
|
debugger.HandleCommand('command script add -f lldbefi.efi_guid_command guid')
|
|
|
|
|
|
Target = debugger.GetTargetAtIndex(0)
|
|
if Target:
|
|
Breakpoint = Target.BreakpointCreateByName('SecGdbScriptBreak')
|
|
if Breakpoint.GetNumLocations() == 1:
|
|
# Set the emulator breakpoints, if we are in the emulator
|
|
debugger.HandleCommand("breakpoint command add -s python -F lldbefi.LoadEmulatorEfiSymbols {id}".format(id=Breakpoint.GetID()))
|
|
print 'Type r to run emulator. SecLldbScriptBreak armed. EFI modules should now get source level debugging in the emulator.'
|