blob: c7d4f0c21b76a24cf75be98ff8a994b56399335c [file] [log] [blame]
/*
* This file is part of the flashrom project.
*
* Copyright (C) 2007, 2008, 2009, 2010 Carl-Daniel Hailfinger
* Copyright (C) 2008 coresystems GmbH
*
* 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; version 2 of the License.
*
* 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.
*
*/
/*
* Contains the common SPI chip driver functions
*/
#include <stddef.h>
#include <string.h>
#include "flash.h"
#include "flashchips.h"
#include "chipdrivers.h"
#include "programmer.h"
#include "spi.h"
#include "spi4ba.h"
enum id_type {
RDID,
RDID4,
REMS,
// RES1, /* TODO */
RES2,
NUM_ID_TYPES,
};
static struct {
int is_cached;
unsigned char bytes[4]; /* enough to hold largest ID type */
} id_cache[NUM_ID_TYPES];
void clear_spi_id_cache(void)
{
memset(id_cache, 0, sizeof(id_cache));
return;
}
static int spi_rdid(struct flashctx *flash, unsigned char *readarr, int bytes)
{
static const unsigned char cmd[JEDEC_RDID_OUTSIZE] = { JEDEC_RDID };
int ret;
int i;
ret = spi_send_command(flash, sizeof(cmd), bytes, cmd, readarr);
if (ret)
return ret;
msg_cspew("RDID returned");
for (i = 0; i < bytes; i++)
msg_cspew(" 0x%02x", readarr[i]);
msg_cspew(". ");
return 0;
}
static int spi_rems(struct flashctx *flash, unsigned char *readarr)
{
unsigned char cmd[JEDEC_REMS_OUTSIZE] = { JEDEC_REMS, 0, 0, 0 };
uint32_t readaddr;
int ret;
ret = spi_send_command(flash, sizeof(cmd), JEDEC_REMS_INSIZE, cmd, readarr);
if (ret == SPI_INVALID_ADDRESS) {
/* Find the lowest even address allowed for reads. */
readaddr = (spi_get_valid_read_addr(flash) + 1) & ~1;
cmd[1] = (readaddr >> 16) & 0xff,
cmd[2] = (readaddr >> 8) & 0xff,
cmd[3] = (readaddr >> 0) & 0xff,
ret = spi_send_command(flash, sizeof(cmd), JEDEC_REMS_INSIZE, cmd, readarr);
}
if (ret)
return ret;
msg_cspew("REMS returned 0x%02x 0x%02x. ", readarr[0], readarr[1]);
return 0;
}
static int spi_res(struct flashctx *flash, unsigned char *readarr, int bytes)
{
unsigned char cmd[JEDEC_RES_OUTSIZE] = { JEDEC_RES, 0, 0, 0 };
uint32_t readaddr;
int ret;
int i;
ret = spi_send_command(flash, sizeof(cmd), bytes, cmd, readarr);
if (ret == SPI_INVALID_ADDRESS) {
/* Find the lowest even address allowed for reads. */
readaddr = (spi_get_valid_read_addr(flash) + 1) & ~1;
cmd[1] = (readaddr >> 16) & 0xff,
cmd[2] = (readaddr >> 8) & 0xff,
cmd[3] = (readaddr >> 0) & 0xff,
ret = spi_send_command(flash, sizeof(cmd), bytes, cmd, readarr);
}
if (ret)
return ret;
msg_cspew("RES returned");
for (i = 0; i < bytes; i++)
msg_cspew(" 0x%02x", readarr[i]);
msg_cspew(". ");
return 0;
}
int spi_write_enable(struct flashctx *flash)
{
static const unsigned char cmd[JEDEC_WREN_OUTSIZE] = { JEDEC_WREN };
int result;
/* Send WREN (Write Enable) */
result = spi_send_command(flash, sizeof(cmd), 0, cmd, NULL);
if (result)
msg_cerr("%s failed\n", __func__);
return result;
}
int spi_write_disable(struct flashctx *flash)
{
static const unsigned char cmd[JEDEC_WRDI_OUTSIZE] = { JEDEC_WRDI };
/* Send WRDI (Write Disable) */
return spi_send_command(flash, sizeof(cmd), 0, cmd, NULL);
}
static void rdid_get_ids(unsigned char *readarr,
int bytes, uint32_t *id1, uint32_t *id2)
{
if (!oddparity(readarr[0]))
msg_cdbg("RDID byte 0 parity violation. ");
/* Check if this is a continuation vendor ID.
* FIXME: Handle continuation device IDs.
*/
if (readarr[0] == 0x7f) {
if (!oddparity(readarr[1]))
msg_cdbg("RDID byte 1 parity violation. ");
*id1 = (readarr[0] << 8) | readarr[1];
*id2 = readarr[2];
if (bytes > 3) {
*id2 <<= 8;
*id2 |= readarr[3];
}
} else {
*id1 = readarr[0];
*id2 = (readarr[1] << 8) | readarr[2];
}
}
static int compare_id(struct flashctx *flash, uint32_t id1, uint32_t id2)
{
msg_cdbg("id1 0x%02x, id2 0x%02x\n", id1, id2);
if (id1 == flash->chip->manufacture_id && id2 == flash->chip->model_id)
return 1;
/* Test if this is a pure vendor match. */
if (id1 == flash->chip->manufacture_id &&
GENERIC_DEVICE_ID == flash->chip->model_id)
return 1;
/* Test if there is any vendor ID. */
if (GENERIC_MANUF_ID == flash->chip->manufacture_id &&
id1 != 0xff)
return 1;
return 0;
}
int probe_spi_rdid(struct flashctx *flash)
{
uint32_t id1, id2;
if (!id_cache[RDID].is_cached) {
if (spi_rdid(flash, id_cache[RDID].bytes, 3))
return 0;
id_cache[RDID].is_cached = 1;
}
rdid_get_ids(id_cache[RDID].bytes, 3, &id1, &id2);
return compare_id(flash, id1, id2);
}
int probe_spi_rdid4(struct flashctx *flash)
{
uint32_t id1, id2;
/* Some SPI controllers do not support commands with writecnt=1 and
* readcnt=4.
*/
switch (spi_master->type) {
#if CONFIG_INTERNAL == 1
#if defined(__i386__) || defined(__x86_64__)
case SPI_CONTROLLER_IT87XX:
case SPI_CONTROLLER_WBSIO:
msg_cinfo("4 byte RDID not supported on this SPI controller\n");
break;
#endif
#endif
default:
break;
}
if (!id_cache[RDID4].is_cached) {
if (spi_rdid(flash, id_cache[RDID4].bytes, 4))
return 0;
id_cache[RDID4].is_cached = 1;
}
rdid_get_ids(id_cache[RDID4].bytes, 4, &id1, &id2);
return compare_id(flash, id1, id2);
}
int probe_spi_rems(struct flashctx *flash)
{
uint32_t id1, id2;
if (!id_cache[REMS].is_cached) {
if (spi_rems(flash, id_cache[REMS].bytes))
return 0;
id_cache[REMS].is_cached = 1;
}
id1 = id_cache[REMS].bytes[0];
id2 = id_cache[REMS].bytes[1];
return compare_id(flash, id1, id2);
}
int probe_spi_res1(struct flashctx *flash)
{
static const unsigned char allff[] = {0xff, 0xff, 0xff};
static const unsigned char all00[] = {0x00, 0x00, 0x00};
unsigned char readarr[3];
uint32_t id2;
/* We only want one-byte RES if RDID and REMS are unusable. */
/* Check if RDID is usable and does not return 0xff 0xff 0xff or
* 0x00 0x00 0x00. In that case, RES is pointless.
*/
if (!spi_rdid(flash, readarr, 3) && memcmp(readarr, allff, 3) &&
memcmp(readarr, all00, 3)) {
msg_cdbg("Ignoring RES in favour of RDID.\n");
return 0;
}
/* Check if REMS is usable and does not return 0xff 0xff or
* 0x00 0x00. In that case, RES is pointless.
*/
if (!spi_rems(flash, readarr) && memcmp(readarr, allff, JEDEC_REMS_INSIZE) &&
memcmp(readarr, all00, JEDEC_REMS_INSIZE)) {
msg_cdbg("Ignoring RES in favour of REMS.\n");
return 0;
}
if (spi_res(flash, readarr, 1)) {
return 0;
}
id2 = readarr[0];
msg_cdbg("%s: id 0x%x\n", __func__, id2);
if (id2 != flash->chip->model_id)
return 0;
return 1;
}
int probe_spi_res2(struct flashctx *flash)
{
uint32_t id1, id2;
if (!id_cache[RES2].is_cached) {
if (spi_res(flash, id_cache[RES2].bytes, 2))
return 0;
id_cache[RES2].is_cached = 1;
}
id1 = id_cache[RES2].bytes[0];
id2 = id_cache[RES2].bytes[1];
msg_cdbg("%s: id1 0x%x, id2 0x%x\n", __func__, id1, id2);
if (id1 != flash->chip->manufacture_id || id2 != flash->chip->model_id)
return 0;
return 1;
}
/**
* Execute WREN plus another one byte `op`, optionally poll WIP afterwards.
*
* @param flash the flash chip's context
* @param op the operation to execute
* @param poll_delay interval in us for polling WIP, don't poll if zero
* @return 0 on success, non-zero otherwise
*/
static int spi_simple_write_cmd(struct flashctx *const flash, const uint8_t op, const unsigned int poll_delay)
{
struct spi_command cmds[] = {
{
.writecnt = 1,
.writearr = (const unsigned char[]){ JEDEC_WREN },
}, {
.writecnt = 1,
.writearr = (const unsigned char[]){ op },
},
NULL_SPI_CMD,
};
const int result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution\n", __func__);
return result;
}
/* Wait until the Write-In-Progress bit is cleared.
* This usually takes 1-85 s, so wait in 1 s steps.
*/
/* FIXME: We can't tell if spi_read_status_register() failed. */
/* FIXME: We don't time out. */
while (poll_delay && spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(poll_delay);
/* FIXME: Check the status register for errors. */
return result;
}
int spi_chip_erase_60(struct flashctx *flash)
{
/* This usually takes 1-85s, so wait in 1s steps. */
return spi_simple_write_cmd(flash, 0x60, 1000 * 1000);
}
int spi_chip_erase_62(struct flashctx *flash)
{
/* This usually takes 2-5s, so wait in 100ms steps. */
return spi_simple_write_cmd(flash, 0x62, 100 * 1000);
}
int spi_chip_erase_c7(struct flashctx *flash)
{
/* This usually takes 1-85s, so wait in 1s steps. */
return spi_simple_write_cmd(flash, 0xc7, 1000 * 1000);
}
int spi_block_erase_52(struct flashctx *flash, unsigned int addr, unsigned int blocklen)
{
int result;
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_BE_52_OUTSIZE,
.writearr = (const unsigned char[]){
JEDEC_BE_52,
(addr >> 16) & 0xff,
(addr >> 8) & 0xff,
(addr & 0xff)
},
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution at address 0x%x\n",
__func__, addr);
return result;
}
/* Wait until the Write-In-Progress bit is cleared.
* This usually takes 100-4000 ms, so wait in 100 ms steps.
*/
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(100 * 1000);
/* FIXME: Check the status register for errors. */
return 0;
}
/* Block size is usually
* 64k for Macronix
* 32k for SST
* 4-32k non-uniform for EON
*/
int spi_block_erase_d8(struct flashctx *flash, unsigned int addr, unsigned int blocklen)
{
int result;
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_BE_D8_OUTSIZE,
.writearr = (const unsigned char[]){
JEDEC_BE_D8,
(addr >> 16) & 0xff,
(addr >> 8) & 0xff,
(addr & 0xff)
},
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution at address 0x%x\n",
__func__, addr);
return result;
}
/* Wait until the Write-In-Progress bit is cleared.
* This usually takes 100-4000 ms, so wait in 100 ms steps.
*/
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(100 * 1000);
/* FIXME: Check the status register for errors. */
return 0;
}
/* Block size is usually
* 4k for PMC
*/
int spi_block_erase_d7(struct flashctx *flash, unsigned int addr, unsigned int blocklen)
{
int result;
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_BE_D7_OUTSIZE,
.writearr = (const unsigned char[]){
JEDEC_BE_D7,
(addr >> 16) & 0xff,
(addr >> 8) & 0xff,
(addr & 0xff)
},
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution at address 0x%x\n",
__func__, addr);
return result;
}
/* Wait until the Write-In-Progress bit is cleared.
* This usually takes 100-4000 ms, so wait in 100 ms steps.
*/
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(100 * 1000);
/* FIXME: Check the status register for errors. */
return 0;
}
/* Sector size is usually 4k, though Macronix eliteflash has 64k */
int spi_block_erase_20(struct flashctx *flash, unsigned int addr, unsigned int blocklen)
{
int result;
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_SE_OUTSIZE,
.writearr = (const unsigned char[]){
JEDEC_SE,
(addr >> 16) & 0xff,
(addr >> 8) & 0xff,
(addr & 0xff)
},
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution at address 0x%x\n",
__func__, addr);
return result;
}
/* Wait until the Write-In-Progress bit is cleared.
* This usually takes 15-800 ms, so wait in 10 ms steps.
*/
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(10 * 1000);
/* FIXME: Check the status register for errors. */
return 0;
}
int spi_block_erase_60(struct flashctx *flash, unsigned int addr, unsigned int blocklen)
{
if ((addr != 0) || (blocklen != flash->chip->total_size * 1024)) {
msg_cerr("%s called with incorrect arguments\n",
__func__);
return -1;
}
return spi_chip_erase_60(flash);
}
int spi_block_erase_c7(struct flashctx *flash, unsigned int addr, unsigned int blocklen)
{
if ((addr != 0) || (blocklen != flash->chip->total_size * 1024)) {
msg_cerr("%s called with incorrect arguments\n",
__func__);
return -1;
}
return spi_chip_erase_c7(flash);
}
int spi_write_status_register_wren(const struct flashctx *flash, int status)
{
int result;
int i = 0;
struct spi_command cmds[] = {
{
/* WRSR requires either EWSR or WREN depending on chip type. */
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_WRSR_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WRSR, (unsigned char) status },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution\n",
__func__);
/* No point in waiting for the command to complete if execution
* failed.
*/
return result;
}
/* WRSR performs a self-timed erase before the changes take effect.
* This may take 50-85 ms in most cases, and some chips apparently
* allow running RDSR only once. Therefore pick an initial delay of
* 100 ms, then wait in 10 ms steps until a total of 5 s have elapsed.
*/
programmer_delay(100 * 1000);
while (spi_read_status_register(flash) & SPI_SR_WIP) {
if (++i > 490) {
msg_cerr("Error: WIP bit after WRSR never cleared\n");
return TIMEOUT_ERROR;
}
programmer_delay(10 * 1000);
}
return 0;
}
int spi_byte_program(struct flashctx *flash, unsigned int addr, uint8_t databyte)
{
int result;
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_BYTE_PROGRAM_OUTSIZE,
.writearr = (const unsigned char[]){
JEDEC_BYTE_PROGRAM,
(addr >> 16) & 0xff,
(addr >> 8) & 0xff,
(addr & 0xff),
databyte
},
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during command execution at address 0x%x\n",
__func__, addr);
}
return result;
}
int spi_nbyte_program(struct flashctx *flash, unsigned int addr, const uint8_t *bytes, unsigned int len)
{
int result;
/* FIXME: Switch to malloc based on len unless that kills speed. */
unsigned char cmd[JEDEC_BYTE_PROGRAM_OUTSIZE - 1 + 256] = {
JEDEC_BYTE_PROGRAM,
(addr >> 16) & 0xff,
(addr >> 8) & 0xff,
(addr >> 0) & 0xff,
};
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_BYTE_PROGRAM_OUTSIZE - 1 + len,
.writearr = cmd,
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
if (!len) {
msg_cerr("%s called for zero-length write\n", __func__);
return 1;
}
if (len > 256) {
msg_cerr("%s called for too long a write\n", __func__);
return 1;
}
memcpy(&cmd[4], bytes, len);
result = spi_send_multicommand(flash, cmds);
if (result) {
if (result != SPI_ACCESS_DENIED) {
msg_cerr("%s failed during command execution at address 0x%x\n",
__func__, addr);
}
}
return result;
}
int spi_restore_status(struct flashctx *flash, uint8_t status)
{
msg_cdbg("restoring chip status (0x%02x)\n", status);
return spi_write_status_register(flash, status);
}
int spi_nbyte_read(struct flashctx *flash, unsigned int address, uint8_t *bytes, unsigned int len)
{
const unsigned char cmd[JEDEC_READ_OUTSIZE] = {
JEDEC_READ,
(address >> 16) & 0xff,
(address >> 8) & 0xff,
(address >> 0) & 0xff,
};
/* Send Read */
return spi_send_command(flash, sizeof(cmd), len, cmd, bytes);
}
/*
* Read a part of the flash chip.
* FIXME: Use the chunk code from Michael Karcher instead.
* Each page is read separately in chunks with a maximum size of chunksize.
*/
int spi_read_chunked(struct flashctx *flash, uint8_t *buf, unsigned int start, unsigned int len, unsigned int chunksize)
{
int rc = 0, chunk_status = 0;
unsigned int i, j, starthere, lenhere, toread;
unsigned int page_size = flash->chip->page_size;
/* Warning: This loop has a very unusual condition and body.
* The loop needs to go through each page with at least one affected
* byte. The lowest page number is (start / page_size) since that
* division rounds down. The highest page number we want is the page
* where the last byte of the range lives. That last byte has the
* address (start + len - 1), thus the highest page number is
* (start + len - 1) / page_size. Since we want to include that last
* page as well, the loop condition uses <=.
*/
for (i = start / page_size; i <= (start + len - 1) / page_size; i++) {
/* Byte position of the first byte in the range in this page. */
/* starthere is an offset to the base address of the chip. */
starthere = max(start, i * page_size);
/* Length of bytes in the range in this page. */
lenhere = min(start + len, (i + 1) * page_size) - starthere;
for (j = 0; j < lenhere; j += chunksize) {
toread = min(chunksize, lenhere - j);
chunk_status = (flash->chip->feature_bits & FEATURE_4BA_SUPPORT) == 0
? spi_nbyte_read(flash, starthere + j, buf + starthere - start + j, toread)
: flash->chip->four_bytes_addr_funcs.read_nbyte(flash, starthere + j,
buf + starthere - start + j, toread);
if (chunk_status) {
if (ignore_error(chunk_status)) {
/* fill this chunk with 0xff bytes and
let caller know about the error */
memset(buf + starthere - start + j, 0xff, toread);
rc = chunk_status;
chunk_status = 0;
continue;
} else {
rc = chunk_status;
break;
}
}
}
if (chunk_status)
break;
}
return rc;
}
/*
* Read a part of the flash chip.
* Ignore pages and read the data continuously, the only bound is the chunksize.
*/
int spi_read_unbound(struct flashctx *flash, uint8_t *buf, unsigned int start, unsigned int len, unsigned int chunksize)
{
int rc = 0;
unsigned int i;
for (i = start; i < (start + len); i += chunksize) {
int chunk_status = 0;
unsigned int toread = min(chunksize, start + len - i);
if (flash->chip->feature_bits & FEATURE_4BA_SUPPORT) {
chunk_status = flash->chip->four_bytes_addr_funcs.read_nbyte(
flash, i, buf + (i - start), toread);
} else {
chunk_status = spi_nbyte_read(flash, i, buf + (i - start), toread);
}
if (chunk_status) {
if (ignore_error(chunk_status)) {
/* fill this chunk with 0xff bytes and
let caller know about the error */
memset(buf + (i - start), 0xff, toread);
rc = chunk_status;
continue;
} else {
rc = chunk_status;
break;
}
}
}
return rc;
}
/*
* Write a part of the flash chip.
* FIXME: Use the chunk code from Michael Karcher instead.
* Each page is written separately in chunks with a maximum size of chunksize.
*/
int spi_write_chunked(struct flashctx *flash, const uint8_t *buf, unsigned int start, unsigned int len, unsigned int chunksize)
{
int rc = 0;
unsigned int i, j, starthere, lenhere, towrite;
/* FIXME: page_size is the wrong variable. We need max_writechunk_size
* in struct flashctx to do this properly. All chips using
* spi_chip_write_256 have page_size set to max_writechunk_size, so
* we're OK for now.
*/
unsigned int page_size = flash->chip->page_size;
/* Warning: This loop has a very unusual condition and body.
* The loop needs to go through each page with at least one affected
* byte. The lowest page number is (start / page_size) since that
* division rounds down. The highest page number we want is the page
* where the last byte of the range lives. That last byte has the
* address (start + len - 1), thus the highest page number is
* (start + len - 1) / page_size. Since we want to include that last
* page as well, the loop condition uses <=.
*/
for (i = start / page_size; i <= (start + len - 1) / page_size; i++) {
/* Byte position of the first byte in the range in this page. */
/* starthere is an offset to the base address of the chip. */
starthere = max(start, i * page_size);
/* Length of bytes in the range in this page. */
lenhere = min(start + len, (i + 1) * page_size) - starthere;
for (j = 0; j < lenhere; j += chunksize) {
towrite = min(chunksize, lenhere - j);
rc = (flash->chip->feature_bits & FEATURE_4BA_SUPPORT) == 0
? spi_nbyte_program(flash, starthere + j, buf + starthere - start + j, towrite)
: flash->chip->four_bytes_addr_funcs.program_nbyte(flash, starthere + j,
buf + starthere - start + j, towrite);
if (rc)
break;
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(10);
}
if (rc)
break;
}
return rc;
}
/*
* Program chip using byte programming. (SLOW!)
* This is for chips which can only handle one byte writes
* and for chips where memory mapped programming is impossible
* (e.g. due to size constraints in IT87* for over 512 kB)
*/
/* real chunksize is 1, logical chunksize is 1 */
int spi_chip_write_1(struct flashctx *flash, const uint8_t *buf, unsigned int start, unsigned int len)
{
unsigned int i;
int result = 0;
for (i = start; i < start + len; i++) {
result = (flash->chip->feature_bits & FEATURE_4BA_SUPPORT) == 0
? spi_byte_program(flash, i, buf[i - start])
: flash->chip->four_bytes_addr_funcs.program_byte(flash, i, buf[i - start]);
if (result)
return 1;
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(10);
}
return 0;
}
int spi_aai_write(struct flashctx *flash, const uint8_t *buf, unsigned int start, unsigned int len)
{
uint32_t pos = start;
int result;
unsigned char cmd[JEDEC_AAI_WORD_PROGRAM_CONT_OUTSIZE] = {
JEDEC_AAI_WORD_PROGRAM,
};
struct spi_command cmds[] = {
{
.writecnt = JEDEC_WREN_OUTSIZE,
.writearr = (const unsigned char[]){ JEDEC_WREN },
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = JEDEC_AAI_WORD_PROGRAM_OUTSIZE,
.writearr = (const unsigned char[]){
JEDEC_AAI_WORD_PROGRAM,
(start >> 16) & 0xff,
(start >> 8) & 0xff,
(start & 0xff),
buf[0],
buf[1]
},
.readcnt = 0,
.readarr = NULL,
}, {
.writecnt = 0,
.writearr = NULL,
.readcnt = 0,
.readarr = NULL,
}};
switch (spi_master->type) {
#if CONFIG_INTERNAL == 1
#if defined(__i386__) || defined(__x86_64__)
case SPI_CONTROLLER_IT87XX:
case SPI_CONTROLLER_WBSIO:
msg_perr("%s: impossible with this SPI controller,"
" degrading to byte program\n", __func__);
return spi_chip_write_1(flash, buf, start, len);
#endif
#endif
default:
break;
}
/* The even start address and even length requirements can be either
* honored outside this function, or we can call spi_byte_program
* for the first and/or last byte and use AAI for the rest.
* FIXME: Move this to generic code.
*/
/* The data sheet requires a start address with the low bit cleared. */
if (start % 2) {
msg_cerr("%s: start address not even! Please report a bug at "
"flashrom@flashrom.org\n", __func__);
if (spi_chip_write_1(flash, buf, start, start % 2))
return SPI_GENERIC_ERROR;
pos += start % 2;
cmds[1].writearr = (const unsigned char[]){
JEDEC_AAI_WORD_PROGRAM,
(pos >> 16) & 0xff,
(pos >> 8) & 0xff,
(pos & 0xff),
buf[pos - start],
buf[pos - start + 1]
};
/* Do not return an error for now. */
//return SPI_GENERIC_ERROR;
}
/* The data sheet requires total AAI write length to be even. */
if (len % 2) {
msg_cerr("%s: total write length not even! Please report a "
"bug at flashrom@flashrom.org\n", __func__);
/* Do not return an error for now. */
//return SPI_GENERIC_ERROR;
}
result = spi_send_multicommand(flash, cmds);
if (result) {
msg_cerr("%s failed during start command execution\n",
__func__);
/* FIXME: Should we send WRDI here as well to make sure the chip
* is not in AAI mode?
*/
return result;
}
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(10);
/* We already wrote 2 bytes in the multicommand step. */
pos += 2;
/* Are there at least two more bytes to write? */
while (pos < start + len - 1) {
cmd[1] = buf[pos++ - start];
cmd[2] = buf[pos++ - start];
spi_send_command(flash, JEDEC_AAI_WORD_PROGRAM_CONT_OUTSIZE, 0, cmd, NULL);
while (spi_read_status_register(flash) & SPI_SR_WIP)
programmer_delay(10);
}
/* Use WRDI to exit AAI mode. This needs to be done before issuing any
* other non-AAI command.
*/
spi_write_disable(flash);
/* Write remaining byte (if any). */
if (pos < start + len) {
if (spi_chip_write_1(flash, buf + pos - start, pos, pos % 2))
return SPI_GENERIC_ERROR;
pos += pos % 2;
}
return 0;
}