blob: 1eed9ef708000c038827a219f9051138101f3276 [file] [log] [blame]
//===-- GDBRemoteRegisterContext.cpp ----------------------------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#include "GDBRemoteRegisterContext.h"
// C Includes
// C++ Includes
// Other libraries and framework includes
#include "lldb/Core/DataBufferHeap.h"
#include "lldb/Core/DataExtractor.h"
#include "lldb/Core/RegisterValue.h"
#include "lldb/Core/Scalar.h"
#include "lldb/Core/StreamString.h"
#ifndef LLDB_DISABLE_PYTHON
#include "lldb/Interpreter/PythonDataObjects.h"
#endif
#include "lldb/Target/ExecutionContext.h"
#include "lldb/Target/Target.h"
#include "lldb/Utility/Utils.h"
// Project includes
#include "Utility/StringExtractorGDBRemote.h"
#include "ProcessGDBRemote.h"
#include "ProcessGDBRemoteLog.h"
#include "ThreadGDBRemote.h"
#include "Utility/ARM_GCC_Registers.h"
#include "Utility/ARM_DWARF_Registers.h"
using namespace lldb;
using namespace lldb_private;
//----------------------------------------------------------------------
// GDBRemoteRegisterContext constructor
//----------------------------------------------------------------------
GDBRemoteRegisterContext::GDBRemoteRegisterContext
(
ThreadGDBRemote &thread,
uint32_t concrete_frame_idx,
GDBRemoteDynamicRegisterInfo &reg_info,
bool read_all_at_once
) :
RegisterContext (thread, concrete_frame_idx),
m_reg_info (reg_info),
m_reg_valid (),
m_reg_data (),
m_read_all_at_once (read_all_at_once)
{
// Resize our vector of bools to contain one bool for every register.
// We will use these boolean values to know when a register value
// is valid in m_reg_data.
m_reg_valid.resize (reg_info.GetNumRegisters());
// Make a heap based buffer that is big enough to store all registers
DataBufferSP reg_data_sp(new DataBufferHeap (reg_info.GetRegisterDataByteSize(), 0));
m_reg_data.SetData (reg_data_sp);
m_reg_data.SetByteOrder(thread.GetProcess()->GetByteOrder());
}
//----------------------------------------------------------------------
// Destructor
//----------------------------------------------------------------------
GDBRemoteRegisterContext::~GDBRemoteRegisterContext()
{
}
void
GDBRemoteRegisterContext::InvalidateAllRegisters ()
{
SetAllRegisterValid (false);
}
void
GDBRemoteRegisterContext::SetAllRegisterValid (bool b)
{
std::vector<bool>::iterator pos, end = m_reg_valid.end();
for (pos = m_reg_valid.begin(); pos != end; ++pos)
*pos = b;
}
size_t
GDBRemoteRegisterContext::GetRegisterCount ()
{
return m_reg_info.GetNumRegisters ();
}
const RegisterInfo *
GDBRemoteRegisterContext::GetRegisterInfoAtIndex (size_t reg)
{
return m_reg_info.GetRegisterInfoAtIndex (reg);
}
size_t
GDBRemoteRegisterContext::GetRegisterSetCount ()
{
return m_reg_info.GetNumRegisterSets ();
}
const RegisterSet *
GDBRemoteRegisterContext::GetRegisterSet (size_t reg_set)
{
return m_reg_info.GetRegisterSet (reg_set);
}
bool
GDBRemoteRegisterContext::ReadRegister (const RegisterInfo *reg_info, RegisterValue &value)
{
// Read the register
if (ReadRegisterBytes (reg_info, m_reg_data))
{
const bool partial_data_ok = false;
Error error (value.SetValueFromData(reg_info, m_reg_data, reg_info->byte_offset, partial_data_ok));
return error.Success();
}
return false;
}
bool
GDBRemoteRegisterContext::PrivateSetRegisterValue (uint32_t reg, StringExtractor &response)
{
const RegisterInfo *reg_info = GetRegisterInfoAtIndex (reg);
if (reg_info == NULL)
return false;
// Invalidate if needed
InvalidateIfNeeded(false);
const uint32_t reg_byte_size = reg_info->byte_size;
const size_t bytes_copied = response.GetHexBytes (const_cast<uint8_t*>(m_reg_data.PeekData(reg_info->byte_offset, reg_byte_size)), reg_byte_size, '\xcc');
bool success = bytes_copied == reg_byte_size;
if (success)
{
SetRegisterIsValid(reg, true);
}
else if (bytes_copied > 0)
{
// Only set register is valid to false if we copied some bytes, else
// leave it as it was.
SetRegisterIsValid(reg, false);
}
return success;
}
// Helper function for GDBRemoteRegisterContext::ReadRegisterBytes().
bool
GDBRemoteRegisterContext::GetPrimordialRegister(const lldb_private::RegisterInfo *reg_info,
GDBRemoteCommunicationClient &gdb_comm)
{
const uint32_t reg = reg_info->kinds[eRegisterKindLLDB];
StringExtractorGDBRemote response;
if (gdb_comm.ReadRegister(m_thread.GetProtocolID(), reg, response))
return PrivateSetRegisterValue (reg, response);
return false;
}
bool
GDBRemoteRegisterContext::ReadRegisterBytes (const RegisterInfo *reg_info, DataExtractor &data)
{
ExecutionContext exe_ctx (CalculateThread());
Process *process = exe_ctx.GetProcessPtr();
Thread *thread = exe_ctx.GetThreadPtr();
if (process == NULL || thread == NULL)
return false;
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *)process)->GetGDBRemote());
InvalidateIfNeeded(false);
const uint32_t reg = reg_info->kinds[eRegisterKindLLDB];
if (!GetRegisterIsValid(reg))
{
if (m_read_all_at_once)
{
StringExtractorGDBRemote response;
if (!gdb_comm.ReadAllRegisters(m_thread.GetProtocolID(), response))
return false;
if (response.IsNormalResponse())
if (response.GetHexBytes ((void *)m_reg_data.GetDataStart(), m_reg_data.GetByteSize(), '\xcc') == m_reg_data.GetByteSize())
SetAllRegisterValid (true);
}
else if (reg_info->value_regs)
{
// Process this composite register request by delegating to the constituent
// primordial registers.
// Index of the primordial register.
bool success = true;
for (uint32_t idx = 0; success; ++idx)
{
const uint32_t prim_reg = reg_info->value_regs[idx];
if (prim_reg == LLDB_INVALID_REGNUM)
break;
// We have a valid primordial register as our constituent.
// Grab the corresponding register info.
const RegisterInfo *prim_reg_info = GetRegisterInfoAtIndex(prim_reg);
if (prim_reg_info == NULL)
success = false;
else
{
// Read the containing register if it hasn't already been read
if (!GetRegisterIsValid(prim_reg))
success = GetPrimordialRegister(prim_reg_info, gdb_comm);
}
}
if (success)
{
// If we reach this point, all primordial register requests have succeeded.
// Validate this composite register.
SetRegisterIsValid (reg_info, true);
}
}
else
{
// Get each register individually
GetPrimordialRegister(reg_info, gdb_comm);
}
// Make sure we got a valid register value after reading it
if (!GetRegisterIsValid(reg))
return false;
}
if (&data != &m_reg_data)
{
// If we aren't extracting into our own buffer (which
// only happens when this function is called from
// ReadRegisterValue(uint32_t, Scalar&)) then
// we transfer bytes from our buffer into the data
// buffer that was passed in
data.SetByteOrder (m_reg_data.GetByteOrder());
data.SetData (m_reg_data, reg_info->byte_offset, reg_info->byte_size);
}
return true;
}
bool
GDBRemoteRegisterContext::WriteRegister (const RegisterInfo *reg_info,
const RegisterValue &value)
{
DataExtractor data;
if (value.GetData (data))
return WriteRegisterBytes (reg_info, data, 0);
return false;
}
// Helper function for GDBRemoteRegisterContext::WriteRegisterBytes().
bool
GDBRemoteRegisterContext::SetPrimordialRegister(const lldb_private::RegisterInfo *reg_info,
GDBRemoteCommunicationClient &gdb_comm)
{
StreamString packet;
StringExtractorGDBRemote response;
const uint32_t reg = reg_info->kinds[eRegisterKindLLDB];
packet.Printf ("P%x=", reg);
packet.PutBytesAsRawHex8 (m_reg_data.PeekData(reg_info->byte_offset, reg_info->byte_size),
reg_info->byte_size,
lldb::endian::InlHostByteOrder(),
lldb::endian::InlHostByteOrder());
if (gdb_comm.GetThreadSuffixSupported())
packet.Printf (";thread:%4.4" PRIx64 ";", m_thread.GetProtocolID());
// Invalidate just this register
SetRegisterIsValid(reg, false);
if (gdb_comm.SendPacketAndWaitForResponse(packet.GetString().c_str(),
packet.GetString().size(),
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
if (response.IsOKResponse())
return true;
}
return false;
}
void
GDBRemoteRegisterContext::SyncThreadState(Process *process)
{
// NB. We assume our caller has locked the sequence mutex.
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *) process)->GetGDBRemote());
if (!gdb_comm.GetSyncThreadStateSupported())
return;
StreamString packet;
StringExtractorGDBRemote response;
packet.Printf ("QSyncThreadState:%4.4" PRIx64 ";", m_thread.GetProtocolID());
if (gdb_comm.SendPacketAndWaitForResponse(packet.GetString().c_str(),
packet.GetString().size(),
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
if (response.IsOKResponse())
InvalidateAllRegisters();
}
}
bool
GDBRemoteRegisterContext::WriteRegisterBytes (const lldb_private::RegisterInfo *reg_info, DataExtractor &data, uint32_t data_offset)
{
ExecutionContext exe_ctx (CalculateThread());
Process *process = exe_ctx.GetProcessPtr();
Thread *thread = exe_ctx.GetThreadPtr();
if (process == NULL || thread == NULL)
return false;
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *)process)->GetGDBRemote());
// FIXME: This check isn't right because IsRunning checks the Public state, but this
// is work you need to do - for instance in ShouldStop & friends - before the public
// state has been changed.
// if (gdb_comm.IsRunning())
// return false;
// Grab a pointer to where we are going to put this register
uint8_t *dst = const_cast<uint8_t*>(m_reg_data.PeekData(reg_info->byte_offset, reg_info->byte_size));
if (dst == NULL)
return false;
if (data.CopyByteOrderedData (data_offset, // src offset
reg_info->byte_size, // src length
dst, // dst
reg_info->byte_size, // dst length
m_reg_data.GetByteOrder())) // dst byte order
{
Mutex::Locker locker;
if (gdb_comm.GetSequenceMutex (locker, "Didn't get sequence mutex for write register."))
{
const bool thread_suffix_supported = gdb_comm.GetThreadSuffixSupported();
ProcessSP process_sp (m_thread.GetProcess());
if (thread_suffix_supported || static_cast<ProcessGDBRemote *>(process_sp.get())->GetGDBRemote().SetCurrentThread(m_thread.GetProtocolID()))
{
StreamString packet;
StringExtractorGDBRemote response;
if (m_read_all_at_once)
{
// Set all registers in one packet
packet.PutChar ('G');
packet.PutBytesAsRawHex8 (m_reg_data.GetDataStart(),
m_reg_data.GetByteSize(),
lldb::endian::InlHostByteOrder(),
lldb::endian::InlHostByteOrder());
if (thread_suffix_supported)
packet.Printf (";thread:%4.4" PRIx64 ";", m_thread.GetProtocolID());
// Invalidate all register values
InvalidateIfNeeded (true);
if (gdb_comm.SendPacketAndWaitForResponse(packet.GetString().c_str(),
packet.GetString().size(),
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
SetAllRegisterValid (false);
if (response.IsOKResponse())
{
return true;
}
}
}
else
{
bool success = true;
if (reg_info->value_regs)
{
// This register is part of another register. In this case we read the actual
// register data for any "value_regs", and once all that data is read, we will
// have enough data in our register context bytes for the value of this register
// Invalidate this composite register first.
for (uint32_t idx = 0; success; ++idx)
{
const uint32_t reg = reg_info->value_regs[idx];
if (reg == LLDB_INVALID_REGNUM)
break;
// We have a valid primordial register as our constituent.
// Grab the corresponding register info.
const RegisterInfo *value_reg_info = GetRegisterInfoAtIndex(reg);
if (value_reg_info == NULL)
success = false;
else
success = SetPrimordialRegister(value_reg_info, gdb_comm);
}
}
else
{
// This is an actual register, write it
success = SetPrimordialRegister(reg_info, gdb_comm);
}
// Check if writing this register will invalidate any other register values?
// If so, invalidate them
if (reg_info->invalidate_regs)
{
for (uint32_t idx = 0, reg = reg_info->invalidate_regs[0];
reg != LLDB_INVALID_REGNUM;
reg = reg_info->invalidate_regs[++idx])
{
SetRegisterIsValid(reg, false);
}
}
return success;
}
}
}
else
{
Log *log (ProcessGDBRemoteLog::GetLogIfAnyCategoryIsSet (GDBR_LOG_THREAD | GDBR_LOG_PACKETS));
if (log)
{
if (log->GetVerbose())
{
StreamString strm;
gdb_comm.DumpHistory(strm);
log->Printf("error: failed to get packet sequence mutex, not sending write register for \"%s\":\n%s", reg_info->name, strm.GetData());
}
else
log->Printf("error: failed to get packet sequence mutex, not sending write register for \"%s\"", reg_info->name);
}
}
}
return false;
}
bool
GDBRemoteRegisterContext::ReadAllRegisterValues (lldb_private::RegisterCheckpoint &reg_checkpoint)
{
ExecutionContext exe_ctx (CalculateThread());
Process *process = exe_ctx.GetProcessPtr();
Thread *thread = exe_ctx.GetThreadPtr();
if (process == NULL || thread == NULL)
return false;
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *)process)->GetGDBRemote());
uint32_t save_id = 0;
if (gdb_comm.SaveRegisterState(thread->GetProtocolID(), save_id))
{
reg_checkpoint.SetID(save_id);
reg_checkpoint.GetData().reset();
return true;
}
else
{
reg_checkpoint.SetID(0); // Invalid save ID is zero
return ReadAllRegisterValues(reg_checkpoint.GetData());
}
}
bool
GDBRemoteRegisterContext::WriteAllRegisterValues (const lldb_private::RegisterCheckpoint &reg_checkpoint)
{
uint32_t save_id = reg_checkpoint.GetID();
if (save_id != 0)
{
ExecutionContext exe_ctx (CalculateThread());
Process *process = exe_ctx.GetProcessPtr();
Thread *thread = exe_ctx.GetThreadPtr();
if (process == NULL || thread == NULL)
return false;
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *)process)->GetGDBRemote());
return gdb_comm.RestoreRegisterState(m_thread.GetProtocolID(), save_id);
}
else
{
return WriteAllRegisterValues(reg_checkpoint.GetData());
}
}
bool
GDBRemoteRegisterContext::ReadAllRegisterValues (lldb::DataBufferSP &data_sp)
{
ExecutionContext exe_ctx (CalculateThread());
Process *process = exe_ctx.GetProcessPtr();
Thread *thread = exe_ctx.GetThreadPtr();
if (process == NULL || thread == NULL)
return false;
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *)process)->GetGDBRemote());
StringExtractorGDBRemote response;
const bool use_g_packet = gdb_comm.AvoidGPackets ((ProcessGDBRemote *)process) == false;
Mutex::Locker locker;
if (gdb_comm.GetSequenceMutex (locker, "Didn't get sequence mutex for read all registers."))
{
SyncThreadState(process);
char packet[32];
const bool thread_suffix_supported = gdb_comm.GetThreadSuffixSupported();
ProcessSP process_sp (m_thread.GetProcess());
if (thread_suffix_supported || static_cast<ProcessGDBRemote *>(process_sp.get())->GetGDBRemote().SetCurrentThread(m_thread.GetProtocolID()))
{
int packet_len = 0;
if (thread_suffix_supported)
packet_len = ::snprintf (packet, sizeof(packet), "g;thread:%4.4" PRIx64, m_thread.GetProtocolID());
else
packet_len = ::snprintf (packet, sizeof(packet), "g");
assert (packet_len < ((int)sizeof(packet) - 1));
if (use_g_packet && gdb_comm.SendPacketAndWaitForResponse(packet, packet_len, response, false) == GDBRemoteCommunication::PacketResult::Success)
{
int packet_len = 0;
if (thread_suffix_supported)
packet_len = ::snprintf (packet, sizeof(packet), "g;thread:%4.4" PRIx64, m_thread.GetProtocolID());
else
packet_len = ::snprintf (packet, sizeof(packet), "g");
assert (packet_len < ((int)sizeof(packet) - 1));
if (gdb_comm.SendPacketAndWaitForResponse(packet, packet_len, response, false) == GDBRemoteCommunication::PacketResult::Success)
{
if (response.IsErrorResponse())
return false;
std::string &response_str = response.GetStringRef();
if (isxdigit(response_str[0]))
{
response_str.insert(0, 1, 'G');
if (thread_suffix_supported)
{
char thread_id_cstr[64];
::snprintf (thread_id_cstr, sizeof(thread_id_cstr), ";thread:%4.4" PRIx64 ";", m_thread.GetProtocolID());
response_str.append (thread_id_cstr);
}
data_sp.reset (new DataBufferHeap (response_str.c_str(), response_str.size()));
return true;
}
}
}
else
{
// For the use_g_packet == false case, we're going to read each register
// individually and store them as binary data in a buffer instead of as ascii
// characters.
const RegisterInfo *reg_info;
// data_sp will take ownership of this DataBufferHeap pointer soon.
DataBufferSP reg_ctx(new DataBufferHeap(m_reg_info.GetRegisterDataByteSize(), 0));
for (uint32_t i = 0; (reg_info = GetRegisterInfoAtIndex (i)) != NULL; i++)
{
if (reg_info->value_regs) // skip registers that are slices of real registers
continue;
ReadRegisterBytes (reg_info, m_reg_data);
// ReadRegisterBytes saves the contents of the register in to the m_reg_data buffer
}
memcpy (reg_ctx->GetBytes(), m_reg_data.GetDataStart(), m_reg_info.GetRegisterDataByteSize());
data_sp = reg_ctx;
return true;
}
}
}
else
{
Log *log (ProcessGDBRemoteLog::GetLogIfAnyCategoryIsSet (GDBR_LOG_THREAD | GDBR_LOG_PACKETS));
if (log)
{
if (log->GetVerbose())
{
StreamString strm;
gdb_comm.DumpHistory(strm);
log->Printf("error: failed to get packet sequence mutex, not sending read all registers:\n%s", strm.GetData());
}
else
log->Printf("error: failed to get packet sequence mutex, not sending read all registers");
}
}
data_sp.reset();
return false;
}
bool
GDBRemoteRegisterContext::WriteAllRegisterValues (const lldb::DataBufferSP &data_sp)
{
if (!data_sp || data_sp->GetBytes() == NULL || data_sp->GetByteSize() == 0)
return false;
ExecutionContext exe_ctx (CalculateThread());
Process *process = exe_ctx.GetProcessPtr();
Thread *thread = exe_ctx.GetThreadPtr();
if (process == NULL || thread == NULL)
return false;
GDBRemoteCommunicationClient &gdb_comm (((ProcessGDBRemote *)process)->GetGDBRemote());
const bool use_g_packet = gdb_comm.AvoidGPackets ((ProcessGDBRemote *)process) == false;
StringExtractorGDBRemote response;
Mutex::Locker locker;
if (gdb_comm.GetSequenceMutex (locker, "Didn't get sequence mutex for write all registers."))
{
const bool thread_suffix_supported = gdb_comm.GetThreadSuffixSupported();
ProcessSP process_sp (m_thread.GetProcess());
if (thread_suffix_supported || static_cast<ProcessGDBRemote *>(process_sp.get())->GetGDBRemote().SetCurrentThread(m_thread.GetProtocolID()))
{
// The data_sp contains the entire G response packet including the
// G, and if the thread suffix is supported, it has the thread suffix
// as well.
const char *G_packet = (const char *)data_sp->GetBytes();
size_t G_packet_len = data_sp->GetByteSize();
if (use_g_packet
&& gdb_comm.SendPacketAndWaitForResponse (G_packet,
G_packet_len,
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
// The data_sp contains the entire G response packet including the
// G, and if the thread suffix is supported, it has the thread suffix
// as well.
const char *G_packet = (const char *)data_sp->GetBytes();
size_t G_packet_len = data_sp->GetByteSize();
if (gdb_comm.SendPacketAndWaitForResponse (G_packet,
G_packet_len,
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
if (response.IsOKResponse())
return true;
else if (response.IsErrorResponse())
{
uint32_t num_restored = 0;
// We need to manually go through all of the registers and
// restore them manually
response.GetStringRef().assign (G_packet, G_packet_len);
response.SetFilePos(1); // Skip the leading 'G'
// G_packet_len is hex-ascii characters plus prefix 'G' plus suffix thread specifier.
// This means buffer will be a little more than 2x larger than necessary but we resize
// it down once we've extracted all hex ascii chars from the packet.
DataBufferHeap buffer (G_packet_len, 0);
DataExtractor restore_data (buffer.GetBytes(),
buffer.GetByteSize(),
m_reg_data.GetByteOrder(),
m_reg_data.GetAddressByteSize());
const uint32_t bytes_extracted = response.GetHexBytes ((void *)restore_data.GetDataStart(),
restore_data.GetByteSize(),
'\xcc');
if (bytes_extracted < restore_data.GetByteSize())
restore_data.SetData(restore_data.GetDataStart(), bytes_extracted, m_reg_data.GetByteOrder());
const RegisterInfo *reg_info;
// The g packet contents may either include the slice registers (registers defined in
// terms of other registers, e.g. eax is a subset of rax) or not. The slice registers
// should NOT be in the g packet, but some implementations may incorrectly include them.
//
// If the slice registers are included in the packet, we must step over the slice registers
// when parsing the packet -- relying on the RegisterInfo byte_offset field would be incorrect.
// If the slice registers are not included, then using the byte_offset values into the
// data buffer is the best way to find individual register values.
uint64_t size_including_slice_registers = 0;
uint64_t size_not_including_slice_registers = 0;
uint64_t size_by_highest_offset = 0;
for (uint32_t reg_idx=0; (reg_info = GetRegisterInfoAtIndex (reg_idx)) != NULL; ++reg_idx)
{
size_including_slice_registers += reg_info->byte_size;
if (reg_info->value_regs == NULL)
size_not_including_slice_registers += reg_info->byte_size;
if (reg_info->byte_offset >= size_by_highest_offset)
size_by_highest_offset = reg_info->byte_offset + reg_info->byte_size;
}
bool use_byte_offset_into_buffer;
if (size_by_highest_offset == restore_data.GetByteSize())
{
// The size of the packet agrees with the highest offset: + size in the register file
use_byte_offset_into_buffer = true;
}
else if (size_not_including_slice_registers == restore_data.GetByteSize())
{
// The size of the packet is the same as concatenating all of the registers sequentially,
// skipping the slice registers
use_byte_offset_into_buffer = true;
}
else if (size_including_slice_registers == restore_data.GetByteSize())
{
// The slice registers are present in the packet (when they shouldn't be).
// Don't try to use the RegisterInfo byte_offset into the restore_data, it will
// point to the wrong place.
use_byte_offset_into_buffer = false;
}
else {
// None of our expected sizes match the actual g packet data we're looking at.
// The most conservative approach here is to use the running total byte offset.
use_byte_offset_into_buffer = false;
}
// In case our register definitions don't include the correct offsets,
// keep track of the size of each reg & compute offset based on that.
uint32_t running_byte_offset = 0;
for (uint32_t reg_idx=0; (reg_info = GetRegisterInfoAtIndex (reg_idx)) != NULL; ++reg_idx, running_byte_offset += reg_info->byte_size)
{
// Skip composite aka slice registers (e.g. eax is a slice of rax).
if (reg_info->value_regs)
continue;
const uint32_t reg = reg_info->kinds[eRegisterKindLLDB];
uint32_t register_offset;
if (use_byte_offset_into_buffer)
{
register_offset = reg_info->byte_offset;
}
else
{
register_offset = running_byte_offset;
}
// Only write down the registers that need to be written
// if we are going to be doing registers individually.
bool write_reg = true;
const uint32_t reg_byte_size = reg_info->byte_size;
const char *restore_src = (const char *)restore_data.PeekData(register_offset, reg_byte_size);
if (restore_src)
{
StreamString packet;
packet.Printf ("P%x=", reg);
packet.PutBytesAsRawHex8 (restore_src,
reg_byte_size,
lldb::endian::InlHostByteOrder(),
lldb::endian::InlHostByteOrder());
if (thread_suffix_supported)
packet.Printf (";thread:%4.4" PRIx64 ";", m_thread.GetProtocolID());
SetRegisterIsValid(reg, false);
if (gdb_comm.SendPacketAndWaitForResponse(packet.GetString().c_str(),
packet.GetString().size(),
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
const char *current_src = (const char *)m_reg_data.PeekData(register_offset, reg_byte_size);
if (current_src)
write_reg = memcmp (current_src, restore_src, reg_byte_size) != 0;
}
if (write_reg)
{
StreamString packet;
packet.Printf ("P%x=", reg);
packet.PutBytesAsRawHex8 (restore_src,
reg_byte_size,
lldb::endian::InlHostByteOrder(),
lldb::endian::InlHostByteOrder());
if (thread_suffix_supported)
packet.Printf (";thread:%4.4" PRIx64 ";", m_thread.GetProtocolID());
SetRegisterIsValid(reg, false);
if (gdb_comm.SendPacketAndWaitForResponse(packet.GetString().c_str(),
packet.GetString().size(),
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
if (response.IsOKResponse())
++num_restored;
}
}
}
}
return num_restored > 0;
}
}
}
else
{
// For the use_g_packet == false case, we're going to write each register
// individually. The data buffer is binary data in this case, instead of
// ascii characters.
bool arm64_debugserver = false;
if (m_thread.GetProcess().get())
{
const ArchSpec &arch = m_thread.GetProcess()->GetTarget().GetArchitecture();
if (arch.IsValid()
&& arch.GetMachine() == llvm::Triple::arm64
&& arch.GetTriple().getVendor() == llvm::Triple::Apple
&& arch.GetTriple().getOS() == llvm::Triple::IOS)
{
arm64_debugserver = true;
}
}
uint32_t num_restored = 0;
const RegisterInfo *reg_info;
for (uint32_t i = 0; (reg_info = GetRegisterInfoAtIndex (i)) != NULL; i++)
{
if (reg_info->value_regs) // skip registers that are slices of real registers
continue;
// Skip the fpsr and fpcr floating point status/control register writing to
// work around a bug in an older version of debugserver that would lead to
// register context corruption when writing fpsr/fpcr.
if (arm64_debugserver &&
(strcmp (reg_info->name, "fpsr") == 0 || strcmp (reg_info->name, "fpcr") == 0))
{
continue;
}
StreamString packet;
packet.Printf ("P%x=", reg_info->kinds[eRegisterKindLLDB]);
packet.PutBytesAsRawHex8 (data_sp->GetBytes() + reg_info->byte_offset, reg_info->byte_size, lldb::endian::InlHostByteOrder(), lldb::endian::InlHostByteOrder());
if (thread_suffix_supported)
packet.Printf (";thread:%4.4" PRIx64 ";", m_thread.GetProtocolID());
SetRegisterIsValid(reg_info, false);
if (gdb_comm.SendPacketAndWaitForResponse(packet.GetString().c_str(),
packet.GetString().size(),
response,
false) == GDBRemoteCommunication::PacketResult::Success)
{
if (response.IsOKResponse())
++num_restored;
}
}
return num_restored > 0;
}
}
}
else
{
Log *log (ProcessGDBRemoteLog::GetLogIfAnyCategoryIsSet (GDBR_LOG_THREAD | GDBR_LOG_PACKETS));
if (log)
{
if (log->GetVerbose())
{
StreamString strm;
gdb_comm.DumpHistory(strm);
log->Printf("error: failed to get packet sequence mutex, not sending write all registers:\n%s", strm.GetData());
}
else
log->Printf("error: failed to get packet sequence mutex, not sending write all registers");
}
}
return false;
}
uint32_t
GDBRemoteRegisterContext::ConvertRegisterKindToRegisterNumber (lldb::RegisterKind kind, uint32_t num)
{
return m_reg_info.ConvertRegisterKindToRegisterNumber (kind, num);
}
void
GDBRemoteDynamicRegisterInfo::HardcodeARMRegisters(bool from_scratch)
{
// For Advanced SIMD and VFP register mapping.
static uint32_t g_d0_regs[] = { 26, 27, LLDB_INVALID_REGNUM }; // (s0, s1)
static uint32_t g_d1_regs[] = { 28, 29, LLDB_INVALID_REGNUM }; // (s2, s3)
static uint32_t g_d2_regs[] = { 30, 31, LLDB_INVALID_REGNUM }; // (s4, s5)
static uint32_t g_d3_regs[] = { 32, 33, LLDB_INVALID_REGNUM }; // (s6, s7)
static uint32_t g_d4_regs[] = { 34, 35, LLDB_INVALID_REGNUM }; // (s8, s9)
static uint32_t g_d5_regs[] = { 36, 37, LLDB_INVALID_REGNUM }; // (s10, s11)
static uint32_t g_d6_regs[] = { 38, 39, LLDB_INVALID_REGNUM }; // (s12, s13)
static uint32_t g_d7_regs[] = { 40, 41, LLDB_INVALID_REGNUM }; // (s14, s15)
static uint32_t g_d8_regs[] = { 42, 43, LLDB_INVALID_REGNUM }; // (s16, s17)
static uint32_t g_d9_regs[] = { 44, 45, LLDB_INVALID_REGNUM }; // (s18, s19)
static uint32_t g_d10_regs[] = { 46, 47, LLDB_INVALID_REGNUM }; // (s20, s21)
static uint32_t g_d11_regs[] = { 48, 49, LLDB_INVALID_REGNUM }; // (s22, s23)
static uint32_t g_d12_regs[] = { 50, 51, LLDB_INVALID_REGNUM }; // (s24, s25)
static uint32_t g_d13_regs[] = { 52, 53, LLDB_INVALID_REGNUM }; // (s26, s27)
static uint32_t g_d14_regs[] = { 54, 55, LLDB_INVALID_REGNUM }; // (s28, s29)
static uint32_t g_d15_regs[] = { 56, 57, LLDB_INVALID_REGNUM }; // (s30, s31)
static uint32_t g_q0_regs[] = { 26, 27, 28, 29, LLDB_INVALID_REGNUM }; // (d0, d1) -> (s0, s1, s2, s3)
static uint32_t g_q1_regs[] = { 30, 31, 32, 33, LLDB_INVALID_REGNUM }; // (d2, d3) -> (s4, s5, s6, s7)
static uint32_t g_q2_regs[] = { 34, 35, 36, 37, LLDB_INVALID_REGNUM }; // (d4, d5) -> (s8, s9, s10, s11)
static uint32_t g_q3_regs[] = { 38, 39, 40, 41, LLDB_INVALID_REGNUM }; // (d6, d7) -> (s12, s13, s14, s15)
static uint32_t g_q4_regs[] = { 42, 43, 44, 45, LLDB_INVALID_REGNUM }; // (d8, d9) -> (s16, s17, s18, s19)
static uint32_t g_q5_regs[] = { 46, 47, 48, 49, LLDB_INVALID_REGNUM }; // (d10, d11) -> (s20, s21, s22, s23)
static uint32_t g_q6_regs[] = { 50, 51, 52, 53, LLDB_INVALID_REGNUM }; // (d12, d13) -> (s24, s25, s26, s27)
static uint32_t g_q7_regs[] = { 54, 55, 56, 57, LLDB_INVALID_REGNUM }; // (d14, d15) -> (s28, s29, s30, s31)
static uint32_t g_q8_regs[] = { 59, 60, LLDB_INVALID_REGNUM }; // (d16, d17)
static uint32_t g_q9_regs[] = { 61, 62, LLDB_INVALID_REGNUM }; // (d18, d19)
static uint32_t g_q10_regs[] = { 63, 64, LLDB_INVALID_REGNUM }; // (d20, d21)
static uint32_t g_q11_regs[] = { 65, 66, LLDB_INVALID_REGNUM }; // (d22, d23)
static uint32_t g_q12_regs[] = { 67, 68, LLDB_INVALID_REGNUM }; // (d24, d25)
static uint32_t g_q13_regs[] = { 69, 70, LLDB_INVALID_REGNUM }; // (d26, d27)
static uint32_t g_q14_regs[] = { 71, 72, LLDB_INVALID_REGNUM }; // (d28, d29)
static uint32_t g_q15_regs[] = { 73, 74, LLDB_INVALID_REGNUM }; // (d30, d31)
// This is our array of composite registers, with each element coming from the above register mappings.
static uint32_t *g_composites[] = {
g_d0_regs, g_d1_regs, g_d2_regs, g_d3_regs, g_d4_regs, g_d5_regs, g_d6_regs, g_d7_regs,
g_d8_regs, g_d9_regs, g_d10_regs, g_d11_regs, g_d12_regs, g_d13_regs, g_d14_regs, g_d15_regs,
g_q0_regs, g_q1_regs, g_q2_regs, g_q3_regs, g_q4_regs, g_q5_regs, g_q6_regs, g_q7_regs,
g_q8_regs, g_q9_regs, g_q10_regs, g_q11_regs, g_q12_regs, g_q13_regs, g_q14_regs, g_q15_regs
};
static RegisterInfo g_register_infos[] = {
// NAME ALT SZ OFF ENCODING FORMAT COMPILER DWARF GENERIC GDB LLDB VALUE REGS INVALIDATE REGS
// ====== ====== === === ============= ============ =================== =================== ====================== === ==== ========== ===============
{ "r0", "arg1", 4, 0, eEncodingUint, eFormatHex, { gcc_r0, dwarf_r0, LLDB_REGNUM_GENERIC_ARG1,0, 0 }, NULL, NULL},
{ "r1", "arg2", 4, 0, eEncodingUint, eFormatHex, { gcc_r1, dwarf_r1, LLDB_REGNUM_GENERIC_ARG2,1, 1 }, NULL, NULL},
{ "r2", "arg3", 4, 0, eEncodingUint, eFormatHex, { gcc_r2, dwarf_r2, LLDB_REGNUM_GENERIC_ARG3,2, 2 }, NULL, NULL},
{ "r3", "arg4", 4, 0, eEncodingUint, eFormatHex, { gcc_r3, dwarf_r3, LLDB_REGNUM_GENERIC_ARG4,3, 3 }, NULL, NULL},
{ "r4", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r4, dwarf_r4, LLDB_INVALID_REGNUM, 4, 4 }, NULL, NULL},
{ "r5", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r5, dwarf_r5, LLDB_INVALID_REGNUM, 5, 5 }, NULL, NULL},
{ "r6", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r6, dwarf_r6, LLDB_INVALID_REGNUM, 6, 6 }, NULL, NULL},
{ "r7", "fp", 4, 0, eEncodingUint, eFormatHex, { gcc_r7, dwarf_r7, LLDB_REGNUM_GENERIC_FP, 7, 7 }, NULL, NULL},
{ "r8", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r8, dwarf_r8, LLDB_INVALID_REGNUM, 8, 8 }, NULL, NULL},
{ "r9", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r9, dwarf_r9, LLDB_INVALID_REGNUM, 9, 9 }, NULL, NULL},
{ "r10", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r10, dwarf_r10, LLDB_INVALID_REGNUM, 10, 10 }, NULL, NULL},
{ "r11", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r11, dwarf_r11, LLDB_INVALID_REGNUM, 11, 11 }, NULL, NULL},
{ "r12", NULL, 4, 0, eEncodingUint, eFormatHex, { gcc_r12, dwarf_r12, LLDB_INVALID_REGNUM, 12, 12 }, NULL, NULL},
{ "sp", "r13", 4, 0, eEncodingUint, eFormatHex, { gcc_sp, dwarf_sp, LLDB_REGNUM_GENERIC_SP, 13, 13 }, NULL, NULL},
{ "lr", "r14", 4, 0, eEncodingUint, eFormatHex, { gcc_lr, dwarf_lr, LLDB_REGNUM_GENERIC_RA, 14, 14 }, NULL, NULL},
{ "pc", "r15", 4, 0, eEncodingUint, eFormatHex, { gcc_pc, dwarf_pc, LLDB_REGNUM_GENERIC_PC, 15, 15 }, NULL, NULL},
{ "f0", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 16, 16 }, NULL, NULL},
{ "f1", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 17, 17 }, NULL, NULL},
{ "f2", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 18, 18 }, NULL, NULL},
{ "f3", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 19, 19 }, NULL, NULL},
{ "f4", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 20, 20 }, NULL, NULL},
{ "f5", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 21, 21 }, NULL, NULL},
{ "f6", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 22, 22 }, NULL, NULL},
{ "f7", NULL, 12, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 23, 23 }, NULL, NULL},
{ "fps", NULL, 4, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 24, 24 }, NULL, NULL},
{ "cpsr","flags", 4, 0, eEncodingUint, eFormatHex, { gcc_cpsr, dwarf_cpsr, LLDB_INVALID_REGNUM, 25, 25 }, NULL, NULL},
{ "s0", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s0, LLDB_INVALID_REGNUM, 26, 26 }, NULL, NULL},
{ "s1", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s1, LLDB_INVALID_REGNUM, 27, 27 }, NULL, NULL},
{ "s2", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s2, LLDB_INVALID_REGNUM, 28, 28 }, NULL, NULL},
{ "s3", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s3, LLDB_INVALID_REGNUM, 29, 29 }, NULL, NULL},
{ "s4", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s4, LLDB_INVALID_REGNUM, 30, 30 }, NULL, NULL},
{ "s5", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s5, LLDB_INVALID_REGNUM, 31, 31 }, NULL, NULL},
{ "s6", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s6, LLDB_INVALID_REGNUM, 32, 32 }, NULL, NULL},
{ "s7", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s7, LLDB_INVALID_REGNUM, 33, 33 }, NULL, NULL},
{ "s8", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s8, LLDB_INVALID_REGNUM, 34, 34 }, NULL, NULL},
{ "s9", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s9, LLDB_INVALID_REGNUM, 35, 35 }, NULL, NULL},
{ "s10", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s10, LLDB_INVALID_REGNUM, 36, 36 }, NULL, NULL},
{ "s11", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s11, LLDB_INVALID_REGNUM, 37, 37 }, NULL, NULL},
{ "s12", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s12, LLDB_INVALID_REGNUM, 38, 38 }, NULL, NULL},
{ "s13", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s13, LLDB_INVALID_REGNUM, 39, 39 }, NULL, NULL},
{ "s14", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s14, LLDB_INVALID_REGNUM, 40, 40 }, NULL, NULL},
{ "s15", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s15, LLDB_INVALID_REGNUM, 41, 41 }, NULL, NULL},
{ "s16", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s16, LLDB_INVALID_REGNUM, 42, 42 }, NULL, NULL},
{ "s17", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s17, LLDB_INVALID_REGNUM, 43, 43 }, NULL, NULL},
{ "s18", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s18, LLDB_INVALID_REGNUM, 44, 44 }, NULL, NULL},
{ "s19", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s19, LLDB_INVALID_REGNUM, 45, 45 }, NULL, NULL},
{ "s20", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s20, LLDB_INVALID_REGNUM, 46, 46 }, NULL, NULL},
{ "s21", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s21, LLDB_INVALID_REGNUM, 47, 47 }, NULL, NULL},
{ "s22", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s22, LLDB_INVALID_REGNUM, 48, 48 }, NULL, NULL},
{ "s23", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s23, LLDB_INVALID_REGNUM, 49, 49 }, NULL, NULL},
{ "s24", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s24, LLDB_INVALID_REGNUM, 50, 50 }, NULL, NULL},
{ "s25", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s25, LLDB_INVALID_REGNUM, 51, 51 }, NULL, NULL},
{ "s26", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s26, LLDB_INVALID_REGNUM, 52, 52 }, NULL, NULL},
{ "s27", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s27, LLDB_INVALID_REGNUM, 53, 53 }, NULL, NULL},
{ "s28", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s28, LLDB_INVALID_REGNUM, 54, 54 }, NULL, NULL},
{ "s29", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s29, LLDB_INVALID_REGNUM, 55, 55 }, NULL, NULL},
{ "s30", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s30, LLDB_INVALID_REGNUM, 56, 56 }, NULL, NULL},
{ "s31", NULL, 4, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_s31, LLDB_INVALID_REGNUM, 57, 57 }, NULL, NULL},
{ "fpscr",NULL, 4, 0, eEncodingUint, eFormatHex, { LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, LLDB_INVALID_REGNUM, 58, 58 }, NULL, NULL},
{ "d16", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d16, LLDB_INVALID_REGNUM, 59, 59 }, NULL, NULL},
{ "d17", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d17, LLDB_INVALID_REGNUM, 60, 60 }, NULL, NULL},
{ "d18", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d18, LLDB_INVALID_REGNUM, 61, 61 }, NULL, NULL},
{ "d19", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d19, LLDB_INVALID_REGNUM, 62, 62 }, NULL, NULL},
{ "d20", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d20, LLDB_INVALID_REGNUM, 63, 63 }, NULL, NULL},
{ "d21", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d21, LLDB_INVALID_REGNUM, 64, 64 }, NULL, NULL},
{ "d22", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d22, LLDB_INVALID_REGNUM, 65, 65 }, NULL, NULL},
{ "d23", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d23, LLDB_INVALID_REGNUM, 66, 66 }, NULL, NULL},
{ "d24", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d24, LLDB_INVALID_REGNUM, 67, 67 }, NULL, NULL},
{ "d25", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d25, LLDB_INVALID_REGNUM, 68, 68 }, NULL, NULL},
{ "d26", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d26, LLDB_INVALID_REGNUM, 69, 69 }, NULL, NULL},
{ "d27", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d27, LLDB_INVALID_REGNUM, 70, 70 }, NULL, NULL},
{ "d28", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d28, LLDB_INVALID_REGNUM, 71, 71 }, NULL, NULL},
{ "d29", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d29, LLDB_INVALID_REGNUM, 72, 72 }, NULL, NULL},
{ "d30", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d30, LLDB_INVALID_REGNUM, 73, 73 }, NULL, NULL},
{ "d31", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d31, LLDB_INVALID_REGNUM, 74, 74 }, NULL, NULL},
{ "d0", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d0, LLDB_INVALID_REGNUM, 75, 75 }, g_d0_regs, NULL},
{ "d1", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d1, LLDB_INVALID_REGNUM, 76, 76 }, g_d1_regs, NULL},
{ "d2", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d2, LLDB_INVALID_REGNUM, 77, 77 }, g_d2_regs, NULL},
{ "d3", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d3, LLDB_INVALID_REGNUM, 78, 78 }, g_d3_regs, NULL},
{ "d4", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d4, LLDB_INVALID_REGNUM, 79, 79 }, g_d4_regs, NULL},
{ "d5", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d5, LLDB_INVALID_REGNUM, 80, 80 }, g_d5_regs, NULL},
{ "d6", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d6, LLDB_INVALID_REGNUM, 81, 81 }, g_d6_regs, NULL},
{ "d7", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d7, LLDB_INVALID_REGNUM, 82, 82 }, g_d7_regs, NULL},
{ "d8", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d8, LLDB_INVALID_REGNUM, 83, 83 }, g_d8_regs, NULL},
{ "d9", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d9, LLDB_INVALID_REGNUM, 84, 84 }, g_d9_regs, NULL},
{ "d10", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d10, LLDB_INVALID_REGNUM, 85, 85 }, g_d10_regs, NULL},
{ "d11", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d11, LLDB_INVALID_REGNUM, 86, 86 }, g_d11_regs, NULL},
{ "d12", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d12, LLDB_INVALID_REGNUM, 87, 87 }, g_d12_regs, NULL},
{ "d13", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d13, LLDB_INVALID_REGNUM, 88, 88 }, g_d13_regs, NULL},
{ "d14", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d14, LLDB_INVALID_REGNUM, 89, 89 }, g_d14_regs, NULL},
{ "d15", NULL, 8, 0, eEncodingIEEE754, eFormatFloat, { LLDB_INVALID_REGNUM, dwarf_d15, LLDB_INVALID_REGNUM, 90, 90 }, g_d15_regs, NULL},
{ "q0", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q0, LLDB_INVALID_REGNUM, 91, 91 }, g_q0_regs, NULL},
{ "q1", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q1, LLDB_INVALID_REGNUM, 92, 92 }, g_q1_regs, NULL},
{ "q2", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q2, LLDB_INVALID_REGNUM, 93, 93 }, g_q2_regs, NULL},
{ "q3", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q3, LLDB_INVALID_REGNUM, 94, 94 }, g_q3_regs, NULL},
{ "q4", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q4, LLDB_INVALID_REGNUM, 95, 95 }, g_q4_regs, NULL},
{ "q5", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q5, LLDB_INVALID_REGNUM, 96, 96 }, g_q5_regs, NULL},
{ "q6", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q6, LLDB_INVALID_REGNUM, 97, 97 }, g_q6_regs, NULL},
{ "q7", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q7, LLDB_INVALID_REGNUM, 98, 98 }, g_q7_regs, NULL},
{ "q8", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q8, LLDB_INVALID_REGNUM, 99, 99 }, g_q8_regs, NULL},
{ "q9", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q9, LLDB_INVALID_REGNUM, 100, 100 }, g_q9_regs, NULL},
{ "q10", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q10, LLDB_INVALID_REGNUM, 101, 101 }, g_q10_regs, NULL},
{ "q11", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q11, LLDB_INVALID_REGNUM, 102, 102 }, g_q11_regs, NULL},
{ "q12", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q12, LLDB_INVALID_REGNUM, 103, 103 }, g_q12_regs, NULL},
{ "q13", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q13, LLDB_INVALID_REGNUM, 104, 104 }, g_q13_regs, NULL},
{ "q14", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q14, LLDB_INVALID_REGNUM, 105, 105 }, g_q14_regs, NULL},
{ "q15", NULL, 16, 0, eEncodingVector, eFormatVectorOfUInt8, { LLDB_INVALID_REGNUM, dwarf_q15, LLDB_INVALID_REGNUM, 106, 106 }, g_q15_regs, NULL}
};
static const uint32_t num_registers = llvm::array_lengthof(g_register_infos);
static ConstString gpr_reg_set ("General Purpose Registers");
static ConstString sfp_reg_set ("Software Floating Point Registers");
static ConstString vfp_reg_set ("Floating Point Registers");
size_t i;
if (from_scratch)
{
// Calculate the offsets of the registers
// Note that the layout of the "composite" registers (d0-d15 and q0-q15) which comes after the
// "primordial" registers is important. This enables us to calculate the offset of the composite
// register by using the offset of its first primordial register. For example, to calculate the
// offset of q0, use s0's offset.
if (g_register_infos[2].byte_offset == 0)
{
uint32_t byte_offset = 0;
for (i=0; i<num_registers; ++i)
{
// For primordial registers, increment the byte_offset by the byte_size to arrive at the
// byte_offset for the next register. Otherwise, we have a composite register whose
// offset can be calculated by consulting the offset of its first primordial register.
if (!g_register_infos[i].value_regs)
{
g_register_infos[i].byte_offset = byte_offset;
byte_offset += g_register_infos[i].byte_size;
}
else
{
const uint32_t first_primordial_reg = g_register_infos[i].value_regs[0];
g_register_infos[i].byte_offset = g_register_infos[first_primordial_reg].byte_offset;
}
}
}
for (i=0; i<num_registers; ++i)
{
ConstString name;
ConstString alt_name;
if (g_register_infos[i].name && g_register_infos[i].name[0])
name.SetCString(g_register_infos[i].name);
if (g_register_infos[i].alt_name && g_register_infos[i].alt_name[0])
alt_name.SetCString(g_register_infos[i].alt_name);
if (i <= 15 || i == 25)
AddRegister (g_register_infos[i], name, alt_name, gpr_reg_set);
else if (i <= 24)
AddRegister (g_register_infos[i], name, alt_name, sfp_reg_set);
else
AddRegister (g_register_infos[i], name, alt_name, vfp_reg_set);
}
}
else
{
// Add composite registers to our primordial registers, then.
const size_t num_composites = llvm::array_lengthof(g_composites);
const size_t num_dynamic_regs = GetNumRegisters();
const size_t num_common_regs = num_registers - num_composites;
RegisterInfo *g_comp_register_infos = g_register_infos + num_common_regs;
// First we need to validate that all registers that we already have match the non composite regs.
// If so, then we can add the registers, else we need to bail
bool match = true;
if (num_dynamic_regs == num_common_regs)
{
for (i=0; match && i<num_dynamic_regs; ++i)
{
// Make sure all register names match
if (m_regs[i].name && g_register_infos[i].name)
{
if (strcmp(m_regs[i].name, g_register_infos[i].name))
{
match = false;
break;
}
}
// Make sure all register byte sizes match
if (m_regs[i].byte_size != g_register_infos[i].byte_size)
{
match = false;
break;
}
}
}
else
{
// Wrong number of registers.
match = false;
}
// If "match" is true, then we can add extra registers.
if (match)
{
for (i=0; i<num_composites; ++i)
{
ConstString name;
ConstString alt_name;
const uint32_t first_primordial_reg = g_comp_register_infos[i].value_regs[0];
const char *reg_name = g_register_infos[first_primordial_reg].name;
if (reg_name && reg_name[0])
{
for (uint32_t j = 0; j < num_dynamic_regs; ++j)
{
const RegisterInfo *reg_info = GetRegisterInfoAtIndex(j);
// Find a matching primordial register info entry.
if (reg_info && reg_info->name && ::strcasecmp(reg_info->name, reg_name) == 0)
{
// The name matches the existing primordial entry.
// Find and assign the offset, and then add this composite register entry.
g_comp_register_infos[i].byte_offset = reg_info->byte_offset;
name.SetCString(g_comp_register_infos[i].name);
AddRegister(g_comp_register_infos[i], name, alt_name, vfp_reg_set);
}
}
}
}
}
}
}