mac-rom-simm-programmer/drivers/parallel_flash.c

509 lines
16 KiB
C

/*
* parallel_flash.c
*
* Created on: Nov 25, 2011
* Author: Doug
*
* Copyright (C) 2011-2023 Doug Brown
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <https://www.gnu.org/licenses/>.
*
*/
#include "parallel_flash.h"
#include "../util.h"
/// Erasable sector size in SST39SF040
#define SECTOR_SIZE_SST39SF040 (4*1024UL)
/// Erasable sector size in M29F160FB5AN6E2, 8-bit mode
#define SECTOR_SIZE_M29F160FB5AN6E2_8 (64*1024UL)
static uint32_t ParallelFlash_MaskForChips(uint8_t chips);
static ALWAYS_INLINE void ParallelFlash_WaitForCompletion(void);
static ALWAYS_INLINE uint32_t ParallelFlash_UnlockAddress1(void);
/// The type/arrangement of parallel flash chips we are talking to
static ParallelFlashChipType curChipType = ParallelFlash_SST39SF040_x4;
/** Sets the type/arrangement of parallel flash chips we are talking to
*
* @param type The type/arrangement of flash chips
*/
void ParallelFlash_SetChipType(ParallelFlashChipType type)
{
curChipType = type;
}
/** Gets the type/arrangement of parallel flash chips we are talking to
*
* @return The current type/arrangement of flash chips
*/
ParallelFlashChipType ParallelFlash_ChipType(void)
{
return curChipType;
}
/** Reads data from the flash chip
*
* @param startAddress The address for reading
* @param buf The buffer to read to
* @param len The number of bytes to read
*/
void ParallelFlash_Read(uint32_t startAddress, uint32_t *buf, uint16_t len)
{
// Just forward this request directly onto the parallel bus. Nothing
// special is required for reading a chunk of data.
ParallelBus_Read(startAddress, buf, len);
}
/** Unlocks the flash chips using the special write sequence
*
* @param chipsMask The mask of which chips to unlock
*/
void ParallelFlash_UnlockChips(uint8_t chipsMask)
{
// Use a mask so we don't unlock chips we don't want to talk with
uint32_t mask = ParallelFlash_MaskForChips(chipsMask);
uint32_t unlockAddress = ParallelFlash_UnlockAddress1();
// First part of unlock sequence:
// Write 0x55555555 to the address bus and 0xAA to the data bus
// (Some datasheets may only say 0x555 or 0x5555, but they ignore
// the upper bits, so writing the alternating pattern to all address lines
// should make it compatible with larger chips).
ParallelBus_WriteCycle(unlockAddress, 0xAAAAAAAAUL & mask);
// Second part of unlock sequence is the same thing, but reversed.
ParallelBus_WriteCycle(~unlockAddress, 0x55555555UL & mask);
}
/** Reads the ID of the chips
*
* @param Pointer to variable for storing ID info about each chip
*/
void ParallelFlash_IdentifyChips(ParallelFlashChipID *chips)
{
// Start by writing the unlock sequence to ALL chips
ParallelFlash_UnlockChips(ALL_CHIPS);
// Write 0x90 to the first unlock address for the identify command...
ParallelBus_WriteCycle(ParallelFlash_UnlockAddress1(), 0x90909090UL);
// Now we can read the vendor and product ID from addresses 0 and 1
// (or 1 and 2 if we're using the M29F160FB5AN6E2 in 8-bit mode).
// Note: The Micron datasheet says it requires 12V to be applied to A9
// in order for the identification command to work properly, but in
// practice the identification process works fine without it.
uint32_t vendorAddress = curChipType != ParallelFlash_M29F160FB5AN6E2_x4 ?
0 : 1;
uint32_t manufacturers = ParallelBus_ReadCycle(vendorAddress);
uint32_t devices = ParallelBus_ReadCycle(vendorAddress + 1);
for (int8_t i = 0; i < PARALLEL_FLASH_NUM_CHIPS; i++)
{
uint8_t manufacturer = (uint8_t)(manufacturers >> (8 * i));
uint8_t device = (uint8_t)(devices >> (8 * i));
chips[PARALLEL_FLASH_NUM_CHIPS - i - 1].manufacturer = manufacturer;
chips[PARALLEL_FLASH_NUM_CHIPS - i - 1].device = device;
}
// Exit software ID mode
ParallelBus_WriteCycle(0, 0xF0F0F0F0UL);
}
/** Erases the specified chips
*
* @param chipsMask The mask of which chips to erase
*/
void ParallelFlash_EraseChips(uint8_t chipsMask)
{
uint32_t unlockAddress = ParallelFlash_UnlockAddress1();
ParallelFlash_UnlockChips(chipsMask);
ParallelBus_WriteCycle(unlockAddress, 0x80808080UL);
ParallelFlash_UnlockChips(chipsMask);
ParallelBus_WriteCycle(unlockAddress, 0x10101010UL);
ParallelFlash_WaitForCompletion();
}
/** Erases only the range of sectors specified in the specified chips
*
* @param address The start address to erase (must be aligned to a sector boundary)
* @param length The number of bytes to erase (must be aligned to a sector boundary)
* @param chipsMask The mask of which chips to erase
* @param numEraseSectorGroups The number of erase sector groups we know about
* @param eraseSectorGroups The erase sector groups
* @return True on success, false on failure
*/
bool ParallelFlash_EraseSectors(uint32_t address, uint32_t length, uint8_t chipsMask, uint8_t numEraseSectorGroups, ParallelFlashEraseSectorGroup const *eraseSectorGroups)
{
// Choose a default sector group if we don't have the info
static const ParallelFlashEraseSectorGroup defaultSST39SF040Sectors[] = {
{0xFFFFFFFFUL, SECTOR_SIZE_SST39SF040}
};
static const ParallelFlashEraseSectorGroup defaultM29F160FBSectors[] = {
{1, 0x4000},
{2, 0x2000},
{1, 0x8000},
{0xFFFFFFFFUL, SECTOR_SIZE_M29F160FB5AN6E2_8}
};
// If we don't know the sector info (older programmer or unknown chips)
// then fall back to the previous hardcoded sector maps.
// Note that "chip type" isn't really accurate anymore; this is more about
// whether or not it has shifted unlock addresses. But these are the hardcoded
// defaults that seemed to work okay for people previously.
if (numEraseSectorGroups == 0)
{
switch (curChipType)
{
case ParallelFlash_SST39SF040_x4:
default:
eraseSectorGroups = defaultSST39SF040Sectors;
numEraseSectorGroups = sizeof(defaultSST39SF040Sectors)/sizeof(defaultSST39SF040Sectors[0]);
break;
case ParallelFlash_M29F160FB5AN6E2_x4:
eraseSectorGroups = defaultM29F160FBSectors;
numEraseSectorGroups = sizeof(defaultM29F160FBSectors)/sizeof(defaultM29F160FBSectors[0]);
break;
}
}
bool result = false;
// The first sector group and index in that group to erase
uint32_t firstSectorGroup = 0;
uint32_t firstSectorInGroup = 0;
// Temporary counters for matching up sector locations
uint32_t curSectorGroup = 0;
uint32_t curSectorInGroup = 0;
// Find the first sector we need to erase. Keep searching until we've
// 1) found it or gone past it, or
// 2) exhausted our list of erase sector groups
uint32_t curAddress = 0;
while (curAddress < address &&
curSectorGroup < numEraseSectorGroups)
{
curAddress += eraseSectorGroups[curSectorGroup].size;
curSectorInGroup++;
if (curSectorInGroup >= eraseSectorGroups[curSectorGroup].count)
{
curSectorGroup++;
curSectorInGroup = 0;
}
}
// If the start address wasn't on a sector boundary, bail
if (curAddress != address)
{
return false;
}
// OK, we've found our first sector to erase.
firstSectorGroup = curSectorGroup;
firstSectorInGroup = curSectorInGroup;
// Now, locate our last sector to erase.
uint32_t curLength = 0;
while (curLength < length &&
curSectorGroup < numEraseSectorGroups)
{
curLength += eraseSectorGroups[curSectorGroup].size;
// If we still haven't handled the entire requested space,
// go to the next sector
if (curLength < length)
{
curSectorInGroup++;
if (curSectorInGroup >= eraseSectorGroups[curSectorGroup].count)
{
curSectorGroup++;
curSectorInGroup = 0;
}
}
}
// If the length wasn't on a sector boundary, bail
if (curLength != length)
{
return false;
}
// We've now verified that everything is on a sector boundary, so we can
// go ahead with the erase operation!
curSectorGroup = firstSectorGroup;
curSectorInGroup = firstSectorInGroup;
// We're good to go. Let's do it. The process varies based on the chip type
if (curChipType == ParallelFlash_SST39SF040_x4)
{
// This chip sucks because you have to erase each sector with its own
// complete erase unlock command, which can take a while. At least
// individual erase operations are much faster on this chip...
while (length)
{
// Start the erase command
ParallelFlash_UnlockChips(chipsMask);
ParallelBus_WriteCycle(ParallelFlash_UnlockAddress1(), 0x80808080UL);
ParallelFlash_UnlockChips(chipsMask);
// Now provide a sector address, but only one. Then the whole
// unlock sequence has to be done again after this sector is done.
ParallelBus_WriteCycle(address, 0x30303030UL);
// Move our counters in preparation for the next sector
address += eraseSectorGroups[curSectorGroup].size;
length -= eraseSectorGroups[curSectorGroup].size;
curSectorInGroup++;
if (curSectorInGroup >= eraseSectorGroups[curSectorGroup].count)
{
curSectorGroup++;
curSectorInGroup = 0;
}
// Wait for completion of this individual erase operation before
// we can start a new erase operation.
ParallelFlash_WaitForCompletion();
}
result = true;
}
else if (curChipType == ParallelFlash_M29F160FB5AN6E2_x4)
{
// This chip is nicer because it can take all the sector addresses at
// once and then do the final erase operation in one fell swoop.
// Start the erase command
ParallelFlash_UnlockChips(chipsMask);
ParallelBus_WriteCycle(ParallelFlash_UnlockAddress1(), 0x80808080UL);
ParallelFlash_UnlockChips(chipsMask);
while (length)
{
ParallelBus_WriteCycle(address, 0x30303030UL);
// Move our counters in preparation for the next sector
address += eraseSectorGroups[curSectorGroup].size;
length -= eraseSectorGroups[curSectorGroup].size;
curSectorInGroup++;
if (curSectorInGroup >= eraseSectorGroups[curSectorGroup].count)
{
curSectorGroup++;
curSectorInGroup = 0;
}
}
// Wait for completion of the entire erase operation
ParallelFlash_WaitForCompletion();
result = true;
}
return result;
}
/** Writes a buffer of data to all 4 chips simultaneously
*
* @param startAddress The starting address to write in flash
* @param buf The buffer to write
* @param len The length of data to write
*
* The API may look silly to have broken into different functions like this, but
* it's a performance optimization. It means we don't have to check during every
* byte write to see the chip unlock mask. It saves a bunch of time.
*/
void ParallelFlash_WriteAllChips(uint32_t startAddress, uint32_t const *buf, uint16_t len)
{
uint32_t unlockAddress = ParallelFlash_UnlockAddress1();
// Normal write process used by most parallel flashes
if (curChipType != ParallelFlash_M29F160FB5AN6E2_x4)
{
while (len--)
{
// Write this byte.
// Unlock...and don't use the unlock function because this one
// is more efficient knowing the mask is 0xFFFFFFFF
ParallelBus_WriteCycle(unlockAddress, 0xAAAAAAAAUL);
ParallelBus_WriteCycle(~unlockAddress, 0x55555555UL);
ParallelBus_WriteCycle(unlockAddress, 0xA0A0A0A0UL);
ParallelBus_WriteCycle(startAddress, *buf);
ParallelFlash_WaitForCompletion();
startAddress++;
buf++;
}
}
// Optimized write process available on the M29F160FB5AN6E2, requires
// fewer write cycles per byte if you know you're writing multiple bytes.
else
{
// Do an unlock bypass command so that we can write bytes faster.
// Writes will only require 2 write cycles instead of 4
ParallelBus_WriteCycle(unlockAddress, 0xAAAAAAAAUL);
ParallelBus_WriteCycle(~unlockAddress, 0x55555555UL);
ParallelBus_WriteCycle(unlockAddress, 0x20202020UL);
while (len--)
{
// Write this byte.
ParallelBus_WriteCycle(0, 0xA0A0A0A0UL);
ParallelBus_WriteCycle(startAddress, *buf);
ParallelFlash_WaitForCompletion();
startAddress++;
buf++;
}
// When we're all done, do "unlock bypass reset" to exit from
// programming mode
ParallelBus_WriteCycle(0, 0x90909090UL);
ParallelBus_WriteCycle(0, 0x00000000UL);
}
}
/** Writes a buffer of data to the specified chips simultaneously
*
* @param startAddress The starting address to write in flash
* @param buf The buffer to write
* @param len The length of data to write
* @param chipsMask The mask of which chips to write
*/
void ParallelFlash_WriteSomeChips(uint32_t startAddress, uint32_t const *buf, uint16_t len, uint8_t chipsMask)
{
uint32_t unlockAddress = ParallelFlash_UnlockAddress1();
// Normal write process used by most parallel flashes
if (curChipType != ParallelFlash_M29F160FB5AN6E2_x4)
{
while (len--)
{
// Write this byte.
ParallelFlash_UnlockChips(chipsMask);
ParallelBus_WriteCycle(unlockAddress, 0xA0A0A0A0UL);
ParallelBus_WriteCycle(startAddress, *buf);
ParallelFlash_WaitForCompletion();
startAddress++;
buf++;
}
}
// Optimized write process available on the M29F160FB5AN6E2, requires
// fewer write cycles per byte if you know you're writing multiple bytes.
else
{
// Do an unlock bypass command so that we can write bytes faster.
// Writes will only require 2 write cycles instead of 4
ParallelFlash_UnlockChips(chipsMask);
ParallelBus_WriteCycle(unlockAddress, 0x20202020UL);
while (len--)
{
// Write this byte.
ParallelBus_WriteCycle(0, 0xA0A0A0A0UL);
ParallelBus_WriteCycle(startAddress, *buf);
ParallelFlash_WaitForCompletion();
startAddress++;
buf++;
}
// When we're all done, do "unlock bypass reset" to exit from
// programming mode
ParallelBus_WriteCycle(0, 0x90909090UL);
ParallelBus_WriteCycle(0, 0x00000000UL);
}
}
/** Calculates a 32-bit mask to use with the unlock process when unlocking chips
*
* @param chipsMask The mask of which chips to write
* @return A 32-bit mask that can be used on the data bus to filter out the unlock sequence
*
* For clarity, chipsMask has 1 bit per chip. The return value has 1 byte per
* chip. The return value masks the unlock sequence so we only supply a valid
* unlock sequence to the chips that we want to unlock.
*/
static uint32_t ParallelFlash_MaskForChips(uint8_t chips)
{
// Calculates a mask so we can filter out chips we don't care about.
// Optimization because we typically don't mask the chips
if (chips == 0x0F)
{
return 0xFFFFFFFFUL;
}
// This probably looks dumb not doing this as a loop...but AVR GCC is
// terrible. This approach results in more optimal generated instructions.
uint32_t mask = 0;
if (chips & (1 << 0))
{
mask |= 0x000000FFUL;
}
if (chips & (1 << 1))
{
mask |= 0x0000FF00UL;
}
if (chips & (1 << 2))
{
mask |= 0x00FF0000UL;
}
if (chips & (1 << 3))
{
mask |= 0xFF000000UL;
}
return mask;
}
/** Waits for an erase or write operation on the flash chip to complete.
*
* We know we're done when the value we read from the chip stops changing. There
* is a "toggle" status bit that will stop toggling when the op is complete.
*/
static ALWAYS_INLINE void ParallelFlash_WaitForCompletion(void)
{
uint32_t readback = ParallelBus_ReadCycle(0);
uint32_t next = ParallelBus_ReadCycle(0);
while (next != readback)
{
readback = next;
next = ParallelBus_ReadCycle(0);
}
}
/** Gets the first unlock address to use when unlocking writes on this chip
*
* @return The first unlock address.
*
* Note: The second unlock address is the bitwise NOT of this address.
*/
static ALWAYS_INLINE uint32_t ParallelFlash_UnlockAddress1(void)
{
// Most chips use alternating bits, with A0 being a 1 bit
if (curChipType != ParallelFlash_M29F160FB5AN6E2_x4)
{
return 0x55555555UL;
}
// The M29F160FB5AN6E2 is weird because it's an 8-/16-bit chip. In 8-bit
// mode it has an "A-1" pin that we treat as A0, then the chip's A0 pin is
// really our A1, and so on. The unlock sequence still starts on the chip's
// physical pin A0, so effectively the unlock address is inverted.
else
{
return 0xAAAAAAAAUL;
}
}