Files
Project64-Legacy/WatchPoints.cpp
Jay Oster 680b9f849d Fix race condition when stopping the CPU while a watchpoint is trapped (#52)
This uses strong (read and write) memory barriers to prevent instruction
reordering when trying to stop the CPU thread while a watchpoint has
trapped it.

The barrier strength can probably be reduced, but I'm a bit paranoid
because the way this whole code base uses threads really has me on edge.
2023-08-19 16:31:59 +02:00

184 lines
4.8 KiB
C++

/*
* Project 64 - A Nintendo 64 emulator.
*
* (c) Copyright 2022 parasyte (jay@kodewerx.org)
*
* pj64 homepage: www.pj64.net
*
* Permission to use, copy, modify and distribute Project64 in both binary and
* source form, for non-commercial purposes, is hereby granted without fee,
* providing that this license information and copyright notice appear with
* all copies and any derived work.
*
* This software is provided 'as-is', without any express or implied
* warranty. In no event shall the authors be held liable for any damages
* arising from the use of this software.
*
* Project64 is freeware for PERSONAL USE only. Commercial users should
* seek permission of the copyright holders first. Commercial use includes
* charging money for Project64 or software derived from Project64.
*
* The copyright holders request that bug fixes and improvements to the code
* should be forwarded to them so if they want them.
*
*/
#include <unordered_map>
#include <windows.h>
extern "C" {
#include "WatchPoints.h"
#include "Interpreter CPU.h"
#include "main.h"
#include "cpu.h"
}
std::unordered_map<DWORD, int[8]>* WatchPoints = NULL;
void InitWatchPoints(void) {
// Default size is implementation-defined (may as well be unknown).
//
// Since we are using the hash table to quickly avoid false positives with a large
// keyspace (32-bit virtual addresses) the hash table size should be selected so that
// more than 99% of address lookups hit an empty bucket.
//
// Assuming the average number of watchpoints enabled is 5 and we want a 1% hit-rate
// (everything else being equal) the optimal size should be 5 / 0.01 = 500.
const UINT HASH_SIZE = 500;
WatchPoints = new std::unordered_map<DWORD, int[8]>({}, HASH_SIZE);
}
void AddWatchPoint(DWORD Location, WATCH_TYPE Type) {
auto search = WatchPoints->find(Location & ~7);
if (search == WatchPoints->end()) {
int* wp = (*WatchPoints)[Location & ~7];
for (int i = 0; i < 8; i++) {
wp[i] = WP_NONE;
}
}
(*WatchPoints)[Location & ~7][Location & 7] = (int)Type | WP_ENABLED;
}
void RemoveWatchPoint(DWORD Location) {
auto search = WatchPoints->find(Location & ~7);
if (search != WatchPoints->end()) {
int *wp = search->second;
wp[Location & 7] = WP_NONE;
int i;
for (i = 0; i < 8; i++) {
if (wp[i] != WP_NONE) {
break;
}
}
if (i == 8) {
WatchPoints->erase(Location & ~7);
}
}
}
void RemoveAllWatchPoints(void) {
WatchPoints->clear();
}
void ToggleWatchPoint(DWORD Location) {
WATCH_TYPE Type = HasWatchPoint(Location);
if (Type != WP_NONE) {
(*WatchPoints)[Location & ~7][Location & 7] = (int)Type ^ WP_ENABLED;
}
}
WATCH_TYPE HasWatchPoint(DWORD Location) {
auto search = WatchPoints->find(Location & ~7);
if (search == WatchPoints->end()) {
return WP_NONE;
}
return (WATCH_TYPE)search->second[Location & 7];
}
BOOL CheckForWatchPoint(DWORD Location, WATCH_TYPE Type, int Size) {
// The memory barrier here is a precaution for inlining this function in tight sequences.
// E.g. DMAs will need to check for watchpoints on read and write accesses
// between the source and destination with two calls.
//
// The barrier ensures that if a watchpoint has trapped the CPU thread and is then released
// by stopping the CPU, the second funtion call will not attempt to trap the CPU again.
// Thus avoiding a potential deadlock.
MemoryBarrier();
if (!HaveDebugger || CPU_Action.CloseCPU || CPU_Action.Stepping || WatchPoints->empty()) {
return FALSE;
}
auto search = WatchPoints->find(Location & ~7);
if (search == WatchPoints->end()) {
return FALSE;
}
int *wp = search->second;
int start = Location & 7;
for (int i = start; i < start + Size; i++) {
int value = wp[i];
if ((value & WP_ENABLED) && (value & (int)Type)) {
TriggerDebugger();
// Block the CPU thread until resumed by the debugger
WaitForSingleObject(CPU_Action.hStepping, INFINITE);
return TRUE;
}
}
return FALSE;
}
int CountWatchPoints(void) {
int count = 0;
for (auto &iter : *WatchPoints) {
int* wp = iter.second;
for (int i = 0; i < 8; i++) {
if (wp[i] != WP_NONE) {
count++;
}
}
}
return count;
}
void RefreshWatchPoints(HWND hList) {
char message[100];
for (auto &iter : *WatchPoints) {
int key = iter.first;
int *wp = iter.second;
for (int i = 0; i < 8; i++) {
int value = wp[i];
if (value == WP_NONE) {
continue;
}
char flags[5] = "----";
if (value & WP_ENABLED) {
flags[0] = 'e';
}
if (value & WP_READ) {
flags[2] = 'r';
}
if (value & WP_WRITE) {
flags[3] = 'w';
}
int location = key | i;
sprintf(message, " at 0x%08X (r4300i %s)", location, flags);
SendMessage(hList, LB_ADDSTRING, 0, (LPARAM)message);
int index = SendMessage(hList, LB_GETCOUNT, 0, 0) - 1;
SendMessage(hList, LB_SETITEMDATA, index, (LPARAM)location);
}
}
}