MemoryMappingModule/source/memory_mapping.cpp

824 lines
32 KiB
C++

#include "memory_mapping.h"
#include <coreinit/memorymap.h>
#include <coreinit/memdefaultheap.h>
#include <coreinit/cache.h>
#include <coreinit/memexpheap.h>
#include <coreinit/thread.h>
#include <vector>
#include "memory.h"
#include "logger.h"
#include "CThread.h"
#include <stdio.h>
#include <string.h>
// #define DEBUG_FUNCTION_LINE(x,...)
void runOnAllCores(CThread::Callback callback, void *callbackArg, int32_t iAttr = 0, int32_t iPriority = 16, int32_t iStackSize = 0x8000) {
int32_t aff[] = {CThread::eAttributeAffCore2, CThread::eAttributeAffCore1, CThread::eAttributeAffCore0};
for (uint32_t i = 0; i < (sizeof(aff) / sizeof(aff[0])); i++) {
CThread *thread = CThread::create(callback, callbackArg, iAttr | aff[i], iPriority, iStackSize);
thread->resumeThread();
delete thread;
}
}
void writeSegmentRegister(CThread *thread, void *arg) {
sr_table_t *table = (sr_table_t *) arg;
uint16_t core = OSGetThreadAffinity(OSGetCurrentThread());
DEBUG_FUNCTION_LINE("Writing segment register to core %d\n", core);
DCFlushRange(table, sizeof(sr_table_t));
KernelWriteSRs(table);
}
void readAndPrintSegmentRegister(CThread *thread, void *arg) {
uint16_t core = OSGetThreadAffinity(OSGetCurrentThread());
DEBUG_FUNCTION_LINE("Reading segment register and page table from core %d", core);
sr_table_t srTable;
memset(&srTable, 0, sizeof(srTable));
KernelReadSRs(&srTable);
DCFlushRange(&srTable, sizeof(srTable));
for (int32_t i = 0; i < 16; i++) {
DEBUG_FUNCTION_LINE("[%d] SR[%d]=%08X", core, i, srTable.value[i]);
}
uint32_t pageTable[0x8000];
memset(pageTable, 0, sizeof(pageTable));
DEBUG_FUNCTION_LINE("Reading pageTable now.");
KernelReadPTE((uint32_t) pageTable, sizeof(pageTable));
DCFlushRange(pageTable, sizeof(pageTable));
DEBUG_FUNCTION_LINE("Reading pageTable done");
MemoryMapping_printPageTableTranslation(srTable, pageTable);
DEBUG_FUNCTION_LINE("-----------------------------");
}
bool MemoryMapping_isMemoryMapped() {
sr_table_t srTable;
memset(&srTable, 0, sizeof(srTable));
KernelReadSRs(&srTable);
if ((srTable.value[MEMORY_START_BASE >> 28] & 0x00FFFFFF) == SEGMENT_UNIQUE_ID) {
return true;
}
return false;
}
void MemoryMapping_searchEmptyMemoryRegions() {
DEBUG_FUNCTION_LINE("Searching for empty memory.");
for (int32_t i = 0;; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
uint32_t ea_start_address = mem_mapping[i].effective_start_address;
const memory_values_t *mem_vals = mem_mapping[i].physical_addresses;
uint32_t ea_size = 0;
for (uint32_t j = 0;; j++) {
uint32_t pa_start_address = mem_vals[j].start_address;
uint32_t pa_end_address = mem_vals[j].end_address;
if (pa_end_address == 0 && pa_start_address == 0) {
break;
}
ea_size += pa_end_address - pa_start_address;
}
uint32_t *flush_start = (uint32_t *) ea_start_address;
uint32_t flush_size = ea_size;
DEBUG_FUNCTION_LINE("Flushing %08X (%d kB) at %08X.", flush_size, flush_size / 1024, flush_start);
DCFlushRange(flush_start, flush_size);
DEBUG_FUNCTION_LINE("Searching in memory region %d. 0x%08X - 0x%08X. Size 0x%08X (%d KBytes).", i + 1, ea_start_address, ea_start_address + ea_size, ea_size, ea_size / 1024);
bool success = true;
uint32_t *memory_ptr = (uint32_t *) ea_start_address;
bool inFailRange = false;
uint32_t startFailing = 0;
uint32_t startGood = ea_start_address;
for (uint32_t j = 0; j < ea_size / 4; j++) {
if (memory_ptr[j] != 0) {
success = false;
if (!success && !inFailRange) {
if ((((uint32_t) &memory_ptr[j]) - (uint32_t) startGood) / 1024 > 512) {
uint32_t start_addr = startGood & 0xFFFE0000;
if (start_addr != startGood) {
start_addr += 0x20000;
}
uint32_t end_addr = ((uint32_t) &memory_ptr[j]) - MEMORY_START_BASE;
end_addr = (end_addr & 0xFFFE0000);
DEBUG_FUNCTION_LINE("+ Free between 0x%08X and 0x%08X size: %u kB", start_addr - MEMORY_START_BASE, end_addr, (((uint32_t) end_addr) - ((uint32_t) startGood - MEMORY_START_BASE)) / 1024);
}
startFailing = (uint32_t) &memory_ptr[j];
inFailRange = true;
startGood = 0;
j = ((j & 0xFFFF8000) + 0x00008000) - 1;
}
//break;
} else {
if (inFailRange) {
//DEBUG_FUNCTION_LINE("- Error between 0x%08X and 0x%08X size: %u kB",startFailing,&memory_ptr[j],(((uint32_t)&memory_ptr[j])-(uint32_t)startFailing)/1024);
startFailing = 0;
startGood = (uint32_t) &memory_ptr[j];
inFailRange = false;
}
}
}
if (startGood != 0 && (startGood != ea_start_address + ea_size)) {
DEBUG_FUNCTION_LINE("+ Good between 0x%08X and 0x%08X size: %u kB", startGood - MEMORY_START_BASE, ((uint32_t) (ea_start_address + ea_size) - (uint32_t) MEMORY_START_BASE),
((uint32_t) (ea_start_address + ea_size) - (uint32_t) startGood) / 1024);
} else if (inFailRange) {
DEBUG_FUNCTION_LINE("- Used between 0x%08X and 0x%08X size: %u kB", startFailing, ea_start_address + ea_size, ((uint32_t) (ea_start_address + ea_size) - (uint32_t) startFailing) / 1024);
}
if (success) {
DEBUG_FUNCTION_LINE("Test %d was successful!", i + 1);
}
}
DEBUG_FUNCTION_LINE("All tests done.");
}
void MemoryMapping_writeTestValuesToMemory() {
//don't smash the stack.
uint32_t chunk_size = 0x1000;
uint32_t testBuffer[chunk_size];
for (int32_t i = 0;; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
uint32_t cur_ea_start_address = mem_mapping[i].effective_start_address;
DEBUG_FUNCTION_LINE("Preparing memory test for region %d. Region start at effective address %08X.", i + 1, cur_ea_start_address);
const memory_values_t *mem_vals = mem_mapping[i].physical_addresses;
uint32_t counter = 0;
for (uint32_t j = 0;; j++) {
uint32_t pa_start_address = mem_vals[j].start_address;
uint32_t pa_end_address = mem_vals[j].end_address;
if (pa_end_address == 0 && pa_start_address == 0) {
break;
}
uint32_t pa_size = pa_end_address - pa_start_address;
DEBUG_FUNCTION_LINE("Writing region %d of mapping %d. From %08X to %08X Size: %d KBytes...", j + 1, i + 1, pa_start_address, pa_end_address, pa_size / 1024);
for (uint32_t k = 0; k <= pa_size / 4; k++) {
if (k > 0 && (k % chunk_size) == 0) {
DCFlushRange(&testBuffer, sizeof(testBuffer));
DCInvalidateRange(&testBuffer, sizeof(testBuffer));
uint32_t destination = pa_start_address + ((k * 4) - sizeof(testBuffer));
KernelCopyData(destination, (uint32_t) OSEffectiveToPhysical((uint32_t) testBuffer), sizeof(testBuffer));
//DEBUG_FUNCTION_LINE("Copy testBuffer into %08X",destination);
}
if (k != pa_size / 4) {
testBuffer[k % chunk_size] = counter++;
}
//DEBUG_FUNCTION_LINE("testBuffer[%d] = %d",i % chunk_size,i);
}
uint32_t *flush_start = (uint32_t *) cur_ea_start_address;
uint32_t flush_size = pa_size;
cur_ea_start_address += pa_size;
DEBUG_FUNCTION_LINE("Flushing %08X (%d kB) at %08X to map memory.", flush_size, flush_size / 1024, flush_start);
DCFlushRange(flush_start, flush_size);
}
DEBUG_FUNCTION_LINE("Done writing region %d", i + 1);
}
}
void MemoryMapping_readTestValuesFromMemory() {
DEBUG_FUNCTION_LINE("Testing reading the written values.");
for (int32_t i = 0;; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
uint32_t ea_start_address = mem_mapping[i].effective_start_address;
const memory_values_t *mem_vals = mem_mapping[i].physical_addresses;
//uint32_t counter = 0;
uint32_t ea_size = 0;
for (uint32_t j = 0;; j++) {
uint32_t pa_start_address = mem_vals[j].start_address;
uint32_t pa_end_address = mem_vals[j].end_address;
if (pa_end_address == 0 && pa_start_address == 0) {
break;
}
ea_size += pa_end_address - pa_start_address;
}
uint32_t *flush_start = (uint32_t *) ea_start_address;
uint32_t flush_size = ea_size;
DEBUG_FUNCTION_LINE("Flushing %08X (%d kB) at %08X to map memory.", flush_size, flush_size / 1024, flush_start);
DCFlushRange(flush_start, flush_size);
DEBUG_FUNCTION_LINE("Testing memory region %d. 0x%08X - 0x%08X. Size 0x%08X (%d KBytes).", i + 1, ea_start_address, ea_start_address + ea_size, ea_size, ea_size / 1024);
bool success = true;
uint32_t *memory_ptr = (uint32_t *) ea_start_address;
bool inFailRange = false;
uint32_t startFailing = 0;
uint32_t startGood = ea_start_address;
for (uint32_t j = 0; j < ea_size / 4; j++) {
if (memory_ptr[j] != j) {
success = false;
if (!success && !inFailRange) {
DEBUG_FUNCTION_LINE("+ Good between 0x%08X and 0x%08X size: %u kB", startGood, &memory_ptr[j], (((uint32_t) &memory_ptr[j]) - (uint32_t) startGood) / 1024);
startFailing = (uint32_t) &memory_ptr[j];
inFailRange = true;
startGood = 0;
j = ((j & 0xFFFF8000) + 0x00008000) - 1;
}
//break;
} else {
if (inFailRange) {
DEBUG_FUNCTION_LINE("- Error between 0x%08X and 0x%08X size: %u kB", startFailing, &memory_ptr[j], (((uint32_t) &memory_ptr[j]) - (uint32_t) startFailing) / 1024);
startFailing = 0;
startGood = (uint32_t) &memory_ptr[j];
inFailRange = false;
}
}
}
if (startGood != 0 && (startGood != ea_start_address + ea_size)) {
DEBUG_FUNCTION_LINE("+ Good between 0x%08X and 0x%08X size: %u kB", startGood, ea_start_address + ea_size, ((uint32_t) (ea_start_address + ea_size) - (uint32_t) startGood) / 1024);
} else if (inFailRange) {
DEBUG_FUNCTION_LINE("- Error between 0x%08X and 0x%08X size: %u kB", startFailing, ea_start_address + ea_size, ((uint32_t) (ea_start_address + ea_size) - (uint32_t) startFailing) / 1024);
}
if (success) {
DEBUG_FUNCTION_LINE("Test %d was successful!", i + 1);
}
}
DEBUG_FUNCTION_LINE("All tests done.");
}
void MemoryMapping_memoryMappingForRegions(const memory_mapping_t *memory_mapping, sr_table_t SRTable, uint32_t *translation_table) {
for (int32_t i = 0; /* waiting for a break */; i++) {
//DEBUG_FUNCTION_LINE("In loop %d",i);
if (memory_mapping[i].physical_addresses == NULL) {
//DEBUG_FUNCTION_LINE("break %d",i);
break;
}
uint32_t cur_ea_start_address = memory_mapping[i].effective_start_address;
DEBUG_FUNCTION_LINE("Mapping area %d. effective address %08X...", i + 1, cur_ea_start_address);
const memory_values_t *mem_vals = memory_mapping[i].physical_addresses;
for (uint32_t j = 0;; j++) {
//DEBUG_FUNCTION_LINE("In inner loop %d",j);
uint32_t pa_start_address = mem_vals[j].start_address;
uint32_t pa_end_address = mem_vals[j].end_address;
if (pa_end_address == 0 && pa_start_address == 0) {
//DEBUG_FUNCTION_LINE("inner break %d",j);
// Break if entry was empty.
break;
}
uint32_t pa_size = pa_end_address - pa_start_address;
DEBUG_FUNCTION_LINE("Adding page table entry %d for mapping area %d. %08X-%08X => %08X-%08X...", j + 1, i + 1, cur_ea_start_address, memory_mapping[i].effective_start_address + pa_size, pa_start_address, pa_end_address);
if (!MemoryMapping_mapMemory(pa_start_address, pa_end_address, cur_ea_start_address, SRTable, translation_table)) {
//log_print("error =(");
DEBUG_FUNCTION_LINE("Failed to map memory.");
//OSFatal("Failed to map memory.");
return;
break;
}
cur_ea_start_address += pa_size;
//log_print("done");
}
}
}
void MemoryMapping_setupMemoryMapping() {
// Override all writes to SR8 with nops.
KernelNOPAtPhysicalAddress(0xFFF1D754);
KernelNOPAtPhysicalAddress(0xFFF1D64C);
KernelNOPAtPhysicalAddress(0xFFE00638);
//runOnAllCores(readAndPrintSegmentRegister,NULL,0,16,0x80000);
sr_table_t srTableCpy;
uint32_t pageTableCpy[0x8000];
KernelReadSRs(&srTableCpy);
KernelReadPTE((uint32_t) pageTableCpy, sizeof(pageTableCpy));
DCFlushRange(&srTableCpy, sizeof(srTableCpy));
DCFlushRange(pageTableCpy, sizeof(pageTableCpy));
for (int32_t i = 0; i < 16; i++) {
DEBUG_FUNCTION_LINE("SR[%d]=%08X", i, srTableCpy.value[i]);
}
//printPageTableTranslation(srTableCpy,pageTableCpy);
// According to
// http://wiiubrew.org/wiki/Cafe_OS#Virtual_Memory_Map 0x80000000
// is currently unmapped.
// This is nice because it leads to SR[8] which also seems to be unused (was set to 0x30FFFFFF)
// The content of the segment was chosen randomly.
uint32_t segment_index = MEMORY_START_BASE >> 28;
uint32_t segment_content = 0x00000000 | SEGMENT_UNIQUE_ID;
DEBUG_FUNCTION_LINE("Setting SR[%d] to %08X", segment_index, segment_content);
srTableCpy.value[segment_index] = segment_content;
DCFlushRange(&srTableCpy, sizeof(srTableCpy));
DEBUG_FUNCTION_LINE("Writing segment registers...", segment_index, segment_content);
// Writing the segment registers to ALL cores.
//
//writeSegmentRegister(NULL, &srTableCpy);
runOnAllCores(writeSegmentRegister, &srTableCpy);
MemoryMapping_memoryMappingForRegions(mem_mapping, srTableCpy, pageTableCpy);
//printPageTableTranslation(srTableCpy,pageTableCpy);
DEBUG_FUNCTION_LINE("Writing PageTable... ");
DCFlushRange(pageTableCpy, sizeof(pageTableCpy));
KernelWritePTE((uint32_t) pageTableCpy, sizeof(pageTableCpy));
DCFlushRange(pageTableCpy, sizeof(pageTableCpy));
DEBUG_FUNCTION_LINE("done");
//printPageTableTranslation(srTableCpy,pageTableCpy);
//runOnAllCores(readAndPrintSegmentRegister,NULL,0,16,0x80000);
//searchEmptyMemoryRegions();
//writeTestValuesToMemory();
//readTestValuesFromMemory();
//runOnAllCores(writeSegmentRegister,&srTableCpy);
}
void *MemoryMapping_alloc(uint32_t size, uint32_t align) {
void *res = NULL;
for (int32_t i = 0; /* waiting for a break */; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
res = MEMAllocFromExpHeapEx((MEMHeapHandle) mem_mapping[i].effective_start_address, size, align);
if (res != NULL) {
break;
}
}
return res;
}
void MemoryMapping_free(void *ptr) {
if (ptr == NULL) {
return;
}
uint32_t ptr_val = (uint32_t) ptr;
for (int32_t i = 0; /* waiting for a break */; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
if (ptr_val > mem_mapping[i].effective_start_address && ptr_val < mem_mapping[i].effective_end_address) {
MEMFreeToExpHeap((MEMHeapHandle) mem_mapping[i].effective_start_address, ptr);
break;
}
}
}
uint32_t MemoryMapping_GetFreeSpace() {
uint32_t res = 0;
for (int32_t i = 0; /* waiting for a break */; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
uint32_t curRes = MEMGetTotalFreeSizeForExpHeap((MEMHeapHandle) mem_mapping[i].effective_start_address);
DEBUG_FUNCTION_LINE("heap at %08X MEMGetTotalFreeSizeForExpHeap: %d KiB", mem_mapping[i].effective_start_address, curRes / 1024);
res += curRes;
}
return res;
}
void MemoryMapping_CreateHeaps() {
for (int32_t i = 0; /* waiting for a break */; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
void *address = (void *) (mem_mapping[i].effective_start_address);
uint32_t size = mem_mapping[i].effective_end_address - mem_mapping[i].effective_start_address;
MEMCreateExpHeapEx(address, size, MEM_HEAP_FLAG_USE_LOCK);
DEBUG_FUNCTION_LINE("Created heap @%08X, size %d KiB", address, size / 1024);
}
}
void MemoryMapping_DestroyHeaps() {
for (int32_t i = 0; /* waiting for a break */; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
void *address = (void *) (mem_mapping[i].effective_start_address);
uint32_t size = mem_mapping[i].effective_end_address - mem_mapping[i].effective_start_address;
MEMDestroyExpHeap((MEMHeapHandle) address);
memset(address, 0, size);
DEBUG_FUNCTION_LINE("Destroyed heap @%08X", address);
}
}
uint32_t MemoryMapping_getAreaSizeFromPageTable(uint32_t start, uint32_t maxSize) {
sr_table_t srTable;
uint32_t pageTable[0x8000];
KernelReadSRs(&srTable);
KernelReadPTE((uint32_t) pageTable, sizeof(pageTable));
uint32_t sr_start = start >> 28;
uint32_t sr_end = (start + maxSize) >> 28;
if (sr_end < sr_start) {
return 0;
}
uint32_t cur_address = start;
uint32_t end_address = start + maxSize;
uint32_t memSize = 0;
for (uint32_t segment = sr_start; segment <= sr_end; segment++) {
uint32_t sr = srTable.value[segment];
if (sr >> 31) {
DEBUG_FUNCTION_LINE("Direct access not supported");
} else {
uint32_t vsid = sr & 0xFFFFFF;
uint32_t pageSize = 1 << PAGE_INDEX_SHIFT;
uint32_t cur_end_addr = 0;
if (segment == sr_end) {
cur_end_addr = end_address;
} else {
cur_end_addr = (segment + 1) * 0x10000000;
}
if (segment != sr_start) {
cur_address = (segment) * 0x10000000;
}
bool success = true;
for (uint32_t addr = cur_address; addr < cur_end_addr; addr += pageSize) {
uint32_t PTEH = 0;
uint32_t PTEL = 0;
if (MemoryMapping_getPageEntryForAddress(srTable.sdr1, addr, vsid, pageTable, &PTEH, &PTEL, false)) {
memSize += pageSize;
} else {
success = false;
break;
}
}
if (!success) {
break;
}
}
}
return memSize;
}
bool MemoryMapping_getPageEntryForAddress(uint32_t SDR1, uint32_t addr, uint32_t vsid, uint32_t *translation_table, uint32_t *oPTEH, uint32_t *oPTEL, bool checkSecondHash) {
uint32_t pageMask = SDR1 & 0x1FF;
uint32_t pageIndex = (addr >> PAGE_INDEX_SHIFT) & PAGE_INDEX_MASK;
uint32_t primaryHash = (vsid & 0x7FFFF) ^pageIndex;
if (MemoryMapping_getPageEntryForAddressEx(SDR1, addr, vsid, primaryHash, translation_table, oPTEH, oPTEL, 0)) {
return true;
}
if (checkSecondHash) {
if (MemoryMapping_getPageEntryForAddressEx(pageMask, addr, vsid, ~primaryHash, translation_table, oPTEH, oPTEL, 1)) {
return true;
}
}
return false;
}
bool MemoryMapping_getPageEntryForAddressEx(uint32_t pageMask, uint32_t addr, uint32_t vsid, uint32_t primaryHash, uint32_t *translation_table, uint32_t *oPTEH, uint32_t *oPTEL, uint32_t H) {
uint32_t maskedHash = primaryHash & ((pageMask << 10) | 0x3FF);
uint32_t api = (addr >> 22) & 0x3F;
uint32_t pteAddrOffset = (maskedHash << 6);
for (int32_t j = 0; j < 8; j++, pteAddrOffset += 8) {
uint32_t PTEH = 0;
uint32_t PTEL = 0;
uint32_t pteh_index = pteAddrOffset / 4;
uint32_t ptel_index = pteh_index + 1;
PTEH = translation_table[pteh_index];
PTEL = translation_table[ptel_index];
//Check validity
if (!(PTEH >> 31)) {
//printf("PTE is not valid ");
continue;
}
//DEBUG_FUNCTION_LINE("in");
// the H bit indicated if the PTE was found using the second hash.
if (((PTEH >> 6) & 1) != H) {
//DEBUG_FUNCTION_LINE("Secondary hash is used",((PTEH >> 6) & 1));
continue;
}
// Check if the VSID matches, otherwise this is a PTE for another SR
// This is the place where collision could happen.
// Hopefully no collision happen and only the PTEs of the SR will match.
if (((PTEH >> 7) & 0xFFFFFF) != vsid) {
//DEBUG_FUNCTION_LINE("VSID mismatch");
continue;
}
// Check the API (Abbreviated Page Index)
if ((PTEH & 0x3F) != api) {
//DEBUG_FUNCTION_LINE("API mismatch");
continue;
}
*oPTEH = PTEH;
*oPTEL = PTEL;
return true;
}
return false;
}
void MemoryMapping_printPageTableTranslation(sr_table_t srTable, uint32_t *translation_table) {
uint32_t SDR1 = srTable.sdr1;
pageInformation current;
memset(&current, 0, sizeof(current));
std::vector<pageInformation> pageInfos;
for (uint32_t segment = 0; segment < 16; segment++) {
uint32_t sr = srTable.value[segment];
if (sr >> 31) {
DEBUG_FUNCTION_LINE("Direct access not supported");
} else {
uint32_t ks = (sr >> 30) & 1;
uint32_t kp = (sr >> 29) & 1;
uint32_t nx = (sr >> 28) & 1;
uint32_t vsid = sr & 0xFFFFFF;
DEBUG_FUNCTION_LINE("ks %08X kp %08X nx %08X vsid %08X", ks, kp, nx, vsid);
uint32_t pageSize = 1 << PAGE_INDEX_SHIFT;
for (uint32_t addr = segment * 0x10000000; addr < (segment + 1) * 0x10000000; addr += pageSize) {
uint32_t PTEH = 0;
uint32_t PTEL = 0;
if (MemoryMapping_getPageEntryForAddress(SDR1, addr, vsid, translation_table, &PTEH, &PTEL, false)) {
uint32_t pp = PTEL & 3;
uint32_t phys = PTEL & 0xFFFFF000;
//DEBUG_FUNCTION_LINE("current.phys == phys - current.size ( %08X %08X)",current.phys, phys - current.size);
if (current.ks == ks &&
current.kp == kp &&
current.nx == nx &&
current.pp == pp &&
current.phys == phys - current.size
) {
current.size += pageSize;
//DEBUG_FUNCTION_LINE("New size of %08X is %08X",current.addr,current.size);
} else {
if (current.addr != 0 && current.size != 0) {
/*DEBUG_FUNCTION_LINE("Saving old block from %08X",current.addr);
DEBUG_FUNCTION_LINE("ks %08X new %08X",current.ks,ks);
DEBUG_FUNCTION_LINE("kp %08X new %08X",current.kp,kp);
DEBUG_FUNCTION_LINE("nx %08X new %08X",current.nx,nx);
DEBUG_FUNCTION_LINE("pp %08X new %08X",current.pp,pp);*/
pageInfos.push_back(current);
memset(&current, 0, sizeof(current));
}
//DEBUG_FUNCTION_LINE("Found new block at %08X",addr);
current.addr = addr;
current.size = pageSize;
current.kp = kp;
current.ks = ks;
current.nx = nx;
current.pp = pp;
current.phys = phys;
}
} else {
if (current.addr != 0 && current.size != 0) {
pageInfos.push_back(current);
memset(&current, 0, sizeof(current));
}
}
}
}
}
const char *access1[] = {"read/write", "read/write", "read/write", "read only"};
const char *access2[] = {"no access", "read only", "read/write", "read only"};
for (std::vector<pageInformation>::iterator it = pageInfos.begin(); it != pageInfos.end(); ++it) {
pageInformation cur = *it;
DEBUG_FUNCTION_LINE("%08X %08X -> %08X %08X. user access %s. supervisor access %s. %s", cur.addr, cur.addr + cur.size, cur.phys, cur.phys + cur.size, cur.kp ? access2[cur.pp] : access1[cur.pp],
cur.ks ? access2[cur.pp] : access1[cur.pp], cur.nx ? "not executable" : "executable");
}
}
bool MemoryMapping_mapMemory(uint32_t pa_start_address, uint32_t pa_end_address, uint32_t ea_start_address, sr_table_t SRTable, uint32_t *translation_table) {
// Based on code from dimok. Thanks!
//uint32_t byteOffsetMask = (1 << PAGE_INDEX_SHIFT) - 1;
//uint32_t apiShift = 22 - PAGE_INDEX_SHIFT;
// Information on page 5.
// https://www.nxp.com/docs/en/application-note/AN2794.pdf
uint32_t HTABORG = SRTable.sdr1 >> 16;
uint32_t HTABMASK = SRTable.sdr1 & 0x1FF;
// Iterate to all possible pages. Each page is 1<<(PAGE_INDEX_SHIFT) big.
uint32_t pageSize = 1 << (PAGE_INDEX_SHIFT);
for (uint32_t i = 0; i < pa_end_address - pa_start_address; i += pageSize) {
// Calculate the current effective address.
uint32_t ea_addr = ea_start_address + i;
// Calculate the segement.
uint32_t segment = SRTable.value[ea_addr >> 28];
// Unique ID from the segment which is the input for the hash function.
// Change it to prevent collisions.
uint32_t VSID = segment & 0x00FFFFFF;
uint32_t V = 1;
//Indicated if second hash is used.
uint32_t H = 0;
// Abbreviated Page Index
// Real page number
uint32_t RPN = (pa_start_address + i) >> 12;
uint32_t RC = 3;
uint32_t WIMG = 0x02;
uint32_t PP = 0x02;
uint32_t page_index = (ea_addr >> PAGE_INDEX_SHIFT) & PAGE_INDEX_MASK;
uint32_t API = (ea_addr >> 22) & 0x3F;
uint32_t PTEH = (V << 31) | (VSID << 7) | (H << 6) | API;
uint32_t PTEL = (RPN << 12) | (RC << 7) | (WIMG << 3) | PP;
//unsigned long long virtual_address = ((unsigned long long)VSID << 28UL) | (page_index << PAGE_INDEX_SHIFT) | (ea_addr & 0xFFF);
uint32_t primary_hash = (VSID & 0x7FFFF);
uint32_t hashvalue1 = primary_hash ^page_index;
// hashvalue 2 is the complement of the first hash.
uint32_t hashvalue2 = ~hashvalue1;
//uint32_t pageMask = SRTable.sdr1 & 0x1FF;
// calculate the address of the PTE groups.
// PTEs are saved in a group of 8 PTEs
// When PTEGaddr1 is full (all 8 PTEs set), PTEGaddr2 is used.
// Then H in PTEH needs to be set to 1.
uint32_t PTEGaddr1 = (HTABORG << 16) | (((hashvalue1 >> 10) & HTABMASK) << 16) | ((hashvalue1 & 0x3FF) << 6);
uint32_t PTEGaddr2 = (HTABORG << 16) | (((hashvalue2 >> 10) & HTABMASK) << 16) | ((hashvalue2 & 0x3FF) << 6);
//offset of the group inside the PTE Table.
uint32_t PTEGoffset = PTEGaddr1 - (HTABORG << 16);
bool setSuccessfully = false;
PTEGoffset += 7 * 8;
// Lets iterate through the PTE group where out PTE should be saved.
for (int32_t j = 7; j > 0; PTEGoffset -= 8) {
int32_t index = (PTEGoffset / 4);
uint32_t pteh = translation_table[index];
// Check if it's already taken. The first bit indicates if the PTE-slot inside
// this group is already taken.
if ((pteh == 0)) {
// If we found a free slot, set the PTEH and PTEL value.
//DEBUG_FUNCTION_LINE("Used slot %d. PTEGaddr1 %08X addr %08X",j+1,PTEGaddr1 - (HTABORG << 16),PTEGoffset);
translation_table[index] = PTEH;
translation_table[index + 1] = PTEL;
setSuccessfully = true;
break;
} else {
//printf("PTEGoffset %08X was taken",PTEGoffset);
}
j--;
}
// Check if we already found a slot.
if (!setSuccessfully) {
//DEBUG_FUNCTION_LINE("-------------- Using second slot -----------------------");
// We still have a chance to find a slot in the PTEGaddr2 using the complement of the hash.
// We need to set the H flag in PTEH and use PTEGaddr2.
// (Not well tested)
H = 1;
PTEH = (V << 31) | (VSID << 7) | (H << 6) | API;
PTEGoffset = PTEGaddr2 - (HTABORG << 16);
PTEGoffset += 7 * 8;
// Same as before.
for (int32_t j = 7; j > 0; PTEGoffset -= 8) {
int32_t index = (PTEGoffset / 4);
uint32_t pteh = translation_table[index];
//Check if it's already taken.
if ((pteh == 0)) {
translation_table[index] = PTEH;
translation_table[index + 1] = PTEL;
setSuccessfully = true;
break;
} else {
//printf("PTEGoffset %08X was taken",PTEGoffset);
}
j--;
}
if (!setSuccessfully) {
// Fail if we couldn't find a free slot.
DEBUG_FUNCTION_LINE("-------------- No more free PTE -----------------------");
return false;
}
}
}
return true;
}
uint32_t MemoryMapping_PhysicalToEffective(uint32_t phyiscalAddress) {
if (phyiscalAddress >= 0x30800000 && phyiscalAddress < 0x31000000) {
return phyiscalAddress - (0x30800000 - 0x00800000);
}
uint32_t result = 0;
const memory_values_t *curMemValues = NULL;
int32_t curOffset = 0;
//iterate through all own mapped memory regions
for (int32_t i = 0; true; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
curMemValues = mem_mapping[i].physical_addresses;
uint32_t curOffsetInEA = 0;
// iterate through all memory values of this region
for (int32_t j = 0; true; j++) {
if (curMemValues[j].end_address == 0) {
break;
}
if (phyiscalAddress >= curMemValues[j].start_address && phyiscalAddress < curMemValues[j].end_address) {
// calculate the EA
result = (phyiscalAddress - curMemValues[j].start_address) + (mem_mapping[i].effective_start_address + curOffsetInEA);
return result;
}
curOffsetInEA += curMemValues[j].end_address - curMemValues[j].start_address;
}
}
return result;
}
uint32_t MemoryMapping_EffectiveToPhysical(uint32_t effectiveAddress) {
if (effectiveAddress >= 0x00800000 && effectiveAddress < 0x01000000) {
return effectiveAddress + (0x30800000 - 0x00800000);
}
uint32_t result = 0;
// CAUTION: The data may be fragmented between multiple areas in PA.
const memory_values_t *curMemValues = NULL;
uint32_t curOffset = 0;
for (int32_t i = 0; true; i++) {
if (mem_mapping[i].physical_addresses == NULL) {
break;
}
if (effectiveAddress >= mem_mapping[i].effective_start_address && effectiveAddress < mem_mapping[i].effective_end_address) {
curMemValues = mem_mapping[i].physical_addresses;
curOffset = mem_mapping[i].effective_start_address;
break;
}
}
if (curMemValues == NULL) {
return result;
}
for (int32_t i = 0; true; i++) {
if (curMemValues[i].end_address == 0) {
break;
}
uint32_t curChunkSize = curMemValues[i].end_address - curMemValues[i].start_address;
if (effectiveAddress < (curOffset + curChunkSize)) {
result = (effectiveAddress - curOffset) + curMemValues[i].start_address;
break;
}
curOffset += curChunkSize;
}
return result;
}