X-Git-Url: https://review.openocd.org/gitweb?p=openocd.git;a=blobdiff_plain;f=src%2Ftarget%2Fmips32_pracc.c;h=cab5555e563689ed3199d90bf37c4636f5276fcf;hp=0ee5715dd299811cbc7d991053068f709f9635d6;hb=d5e564625f72e1018f56a707b60d47a5ca9fece9;hpb=e1f5055bb089d1f854777da1d200e7551cdd3a4f

diff --git a/src/target/mips32_pracc.c b/src/target/mips32_pracc.c
index 0ee5715dd2..cab5555e56 100644
--- a/src/target/mips32_pracc.c
+++ b/src/target/mips32_pracc.c
@@ -6,6 +6,9 @@
  *                                                                         *
  *   Copyright (C) 2009 by David N. Claffey <dnclaffey@gmail.com>          *
  *                                                                         *
+ *   Copyright (C) 2011 by Drasko DRASKOVIC                                *
+ *   drasko.draskovic@gmail.com                                            *
+ *                                                                         *
  *   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 2 of the License, or     *
@@ -23,62 +26,65 @@
  ***************************************************************************/
 
 /*
-This version has optimized assembly routines for 32 bit operations:
-- read word
-- write word
-- write array of words
-
-One thing to be aware of is that the MIPS32 cpu will execute the
-instruction after a branch instruction (one delay slot).
-
-For example:
-
-    LW $2, ($5 +10)
-    B foo
-    LW $1, ($2 +100)
-
-The LW $1, ($2 +100) instruction is also executed. If this is
-not wanted a NOP can be inserted:
-
-    LW $2, ($5 +10)
-    B foo
-    NOP
-    LW $1, ($2 +100)
-
-or the code can be changed to:
-
-    B foo
-    LW $2, ($5 +10)
-    LW $1, ($2 +100)
-
-The original code contained NOPs. I have removed these and moved
-the branches.
-
-I also moved the PRACC_STACK to 0xFF204000. This allows
-the use of 16 bits offsets to get pointers to the input
-and output area relative to the stack. Note that the stack
-isn't really a stack (the stack pointer is not 'moving')
-but a FIFO simulated in software.
-
-These changes result in a 35% speed increase when programming an
-external flash.
-
-More improvement could be gained if the registers do no need
-to be preserved but in that case the routines should be aware
-OpenOCD is used as a flash programmer or as a debug tool.
-
-Nico Coesel
-*/
+ * This version has optimized assembly routines for 32 bit operations:
+ * - read word
+ * - write word
+ * - write array of words
+ *
+ * One thing to be aware of is that the MIPS32 cpu will execute the
+ * instruction after a branch instruction (one delay slot).
+ *
+ * For example:
+ *  LW $2, ($5 +10)
+ *  B foo
+ *  LW $1, ($2 +100)
+ *
+ * The LW $1, ($2 +100) instruction is also executed. If this is
+ * not wanted a NOP can be inserted:
+ *
+ *  LW $2, ($5 +10)
+ *  B foo
+ *  NOP
+ *  LW $1, ($2 +100)
+ *
+ * or the code can be changed to:
+ *
+ *  B foo
+ *  LW $2, ($5 +10)
+ *  LW $1, ($2 +100)
+ *
+ * The original code contained NOPs. I have removed these and moved
+ * the branches.
+ *
+ * I also moved the PRACC_STACK to 0xFF204000. This allows
+ * the use of 16 bits offsets to get pointers to the input
+ * and output area relative to the stack. Note that the stack
+ * isn't really a stack (the stack pointer is not 'moving')
+ * but a FIFO simulated in software.
+ *
+ * These changes result in a 35% speed increase when programming an
+ * external flash.
+ *
+ * More improvement could be gained if the registers do no need
+ * to be preserved but in that case the routines should be aware
+ * OpenOCD is used as a flash programmer or as a debug tool.
+ *
+ * Nico Coesel
+ */
 
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 
+#include <helper/time_support.h>
+
 #include "mips32.h"
 #include "mips32_pracc.h"
 
-struct mips32_pracc_context
-{
+#define PRACC_FETCH 0
+#define PRACC_STORE 1
+
+struct mips32_pracc_context {
 	uint32_t *local_iparam;
 	int num_iparam;
 	uint32_t *local_oparam;
@@ -90,37 +96,35 @@ struct mips32_pracc_context
 	struct mips_ejtag *ejtag_info;
 };
 
-static int mips32_pracc_read_mem8(struct mips_ejtag *ejtag_info,
-		uint32_t addr, int count, uint8_t *buf);
-static int mips32_pracc_read_mem16(struct mips_ejtag *ejtag_info,
-		uint32_t addr, int count, uint16_t *buf);
-static int mips32_pracc_read_mem32(struct mips_ejtag *ejtag_info,
-		uint32_t addr, int count, uint32_t *buf);
-static int mips32_pracc_read_u32(struct mips_ejtag *ejtag_info,
-		uint32_t addr, uint32_t *buf);
-
-static int mips32_pracc_write_mem8(struct mips_ejtag *ejtag_info,
-		uint32_t addr, int count, uint8_t *buf);
-static int mips32_pracc_write_mem16(struct mips_ejtag *ejtag_info,
-		uint32_t addr, int count, uint16_t *buf);
-static int mips32_pracc_write_mem32(struct mips_ejtag *ejtag_info,
-		uint32_t addr, int count, uint32_t *buf);
-static int mips32_pracc_write_u32(struct mips_ejtag *ejtag_info,
-		uint32_t addr, uint32_t *buf);
+static int mips32_pracc_sync_cache(struct mips_ejtag *ejtag_info,
+		uint32_t start_addr, uint32_t end_addr);
+static int mips32_pracc_clean_invalidate_cache(struct mips_ejtag *ejtag_info,
+		uint32_t start_addr, uint32_t end_addr);
 
 static int wait_for_pracc_rw(struct mips_ejtag *ejtag_info, uint32_t *ctrl)
 {
 	uint32_t ejtag_ctrl;
+	long long then = timeval_ms();
+	int timeout;
+	int retval;
 
-	while (1)
-	{
-		mips_ejtag_set_instr(ejtag_info, EJTAG_INST_CONTROL);
+	/* wait for the PrAcc to become "1" */
+	mips_ejtag_set_instr(ejtag_info, EJTAG_INST_CONTROL);
+
+	while (1) {
 		ejtag_ctrl = ejtag_info->ejtag_ctrl;
-		mips_ejtag_drscan_32(ejtag_info, &ejtag_ctrl);
+		retval = mips_ejtag_drscan_32(ejtag_info, &ejtag_ctrl);
+		if (retval != ERROR_OK)
+			return retval;
+
 		if (ejtag_ctrl & EJTAG_CTRL_PRACC)
 			break;
-		LOG_DEBUG("DEBUGMODULE: No memory access in progress!");
-		return ERROR_JTAG_DEVICE_ERROR;
+
+		timeout = timeval_ms() - then;
+		if (timeout > 1000) {
+			LOG_DEBUG("DEBUGMODULE: No memory access in progress!");
+			return ERROR_JTAG_DEVICE_ERROR;
+		}
 	}
 
 	*ctrl = ejtag_ctrl;
@@ -134,87 +138,77 @@ static int mips32_pracc_exec_read(struct mips32_pracc_context *ctx, uint32_t add
 	uint32_t ejtag_ctrl, data;
 
 	if ((address >= MIPS32_PRACC_PARAM_IN)
-		&& (address <= MIPS32_PRACC_PARAM_IN + ctx->num_iparam * 4))
-	{
+		&& (address < MIPS32_PRACC_PARAM_IN + ctx->num_iparam * 4)) {
 		offset = (address - MIPS32_PRACC_PARAM_IN) / 4;
 		data = ctx->local_iparam[offset];
-	}
-	else if ((address >= MIPS32_PRACC_PARAM_OUT)
-		&& (address <= MIPS32_PRACC_PARAM_OUT + ctx->num_oparam * 4))
-	{
+	} else if ((address >= MIPS32_PRACC_PARAM_OUT)
+		&& (address < MIPS32_PRACC_PARAM_OUT + ctx->num_oparam * 4)) {
 		offset = (address - MIPS32_PRACC_PARAM_OUT) / 4;
 		data = ctx->local_oparam[offset];
-	}
-	else if ((address >= MIPS32_PRACC_TEXT)
-		&& (address <= MIPS32_PRACC_TEXT + ctx->code_len * 4))
-	{
+	} else if ((address >= MIPS32_PRACC_TEXT)
+		&& (address < MIPS32_PRACC_TEXT + ctx->code_len * 4)) {
 		offset = (address - MIPS32_PRACC_TEXT) / 4;
 		data = ctx->code[offset];
-	}
-	else if (address == MIPS32_PRACC_STACK)
-	{
+	} else if (address == MIPS32_PRACC_STACK) {
+		if (ctx->stack_offset <= 0) {
+			LOG_ERROR("Error: Pracc stack out of bounds");
+			return ERROR_JTAG_DEVICE_ERROR;
+		}
 		/* save to our debug stack */
 		data = ctx->stack[--ctx->stack_offset];
-	}
-	else
-	{
+	} else {
 		/* TODO: send JMP 0xFF200000 instruction. Hopefully processor jump back
 		 * to start of debug vector */
 
-		data = 0;
 		LOG_ERROR("Error reading unexpected address 0x%8.8" PRIx32 "", address);
 		return ERROR_JTAG_DEVICE_ERROR;
 	}
 
 	/* Send the data out */
 	mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_DATA);
-	mips_ejtag_drscan_32(ctx->ejtag_info, &data);
+	mips_ejtag_drscan_32_out(ctx->ejtag_info, data);
 
 	/* Clear the access pending bit (let the processor eat!) */
 	ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
 	mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_CONTROL);
-	mips_ejtag_drscan_32(ctx->ejtag_info, &ejtag_ctrl);
+	mips_ejtag_drscan_32_out(ctx->ejtag_info, ejtag_ctrl);
 
-	jtag_add_clocks(5);
 	return jtag_execute_queue();
 }
 
 static int mips32_pracc_exec_write(struct mips32_pracc_context *ctx, uint32_t address)
 {
-	uint32_t ejtag_ctrl,data;
+	uint32_t ejtag_ctrl, data;
 	int offset;
 	struct mips_ejtag *ejtag_info = ctx->ejtag_info;
+	int retval;
 
 	mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_DATA);
-	mips_ejtag_drscan_32(ctx->ejtag_info, &data);
+	retval = mips_ejtag_drscan_32(ctx->ejtag_info, &data);
+	if (retval != ERROR_OK)
+		return retval;
 
 	/* Clear access pending bit */
 	ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
 	mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_CONTROL);
-	mips_ejtag_drscan_32(ctx->ejtag_info, &ejtag_ctrl);
+	mips_ejtag_drscan_32_out(ctx->ejtag_info, ejtag_ctrl);
 
-	jtag_add_clocks(5);
-	return jtag_execute_queue();
+	retval = jtag_execute_queue();
+	if (retval != ERROR_OK)
+		return retval;
 
-	if ((address >= MIPS32_PRACC_PARAM_IN)
-		&& (address <= MIPS32_PRACC_PARAM_IN + ctx->num_iparam * 4))
-	{
-		offset = (address - MIPS32_PRACC_PARAM_IN) / 4;
-		ctx->local_iparam[offset] = data;
-	}
-	else if ((address >= MIPS32_PRACC_PARAM_OUT)
-		&& (address <= MIPS32_PRACC_PARAM_OUT + ctx->num_oparam * 4))
-	{
+	if ((address >= MIPS32_PRACC_PARAM_OUT)
+		&& (address < MIPS32_PRACC_PARAM_OUT + ctx->num_oparam * 4)) {
 		offset = (address - MIPS32_PRACC_PARAM_OUT) / 4;
 		ctx->local_oparam[offset] = data;
-	}
-	else if (address == MIPS32_PRACC_STACK)
-	{
+	} else if (address == MIPS32_PRACC_STACK) {
+		if (ctx->stack_offset >= 32) {
+			LOG_ERROR("Error: Pracc stack out of bounds");
+			return ERROR_JTAG_DEVICE_ERROR;
+		}
 		/* save data onto our stack */
 		ctx->stack[ctx->stack_offset++] = data;
-	}
-	else
-	{
+	} else {
 		LOG_ERROR("Error writing unexpected address 0x%8.8" PRIx32 "", address);
 		return ERROR_JTAG_DEVICE_ERROR;
 	}
@@ -226,7 +220,7 @@ int mips32_pracc_exec(struct mips_ejtag *ejtag_info, int code_len, const uint32_
 		int num_param_in, uint32_t *param_in, int num_param_out, uint32_t *param_out, int cycle)
 {
 	uint32_t ejtag_ctrl;
-	uint32_t address, data;
+	uint32_t address;
 	struct mips32_pracc_context ctx;
 	int retval;
 	int pass = 0;
@@ -240,32 +234,31 @@ int mips32_pracc_exec(struct mips_ejtag *ejtag_info, int code_len, const uint32_
 	ctx.ejtag_info = ejtag_info;
 	ctx.stack_offset = 0;
 
-	while (1)
-	{
-		if ((retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl)) != ERROR_OK)
+	while (1) {
+		retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
+		if (retval != ERROR_OK)
 			return retval;
 
-		address = data = 0;
+		address = 0;
 		mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ADDRESS);
-		mips_ejtag_drscan_32(ejtag_info, &address);
+		retval = mips_ejtag_drscan_32(ejtag_info, &address);
+		if (retval != ERROR_OK)
+			return retval;
 
 		/* Check for read or write */
-		if (ejtag_ctrl & EJTAG_CTRL_PRNW)
-		{
-			if ((retval = mips32_pracc_exec_write(&ctx, address)) != ERROR_OK)
+		if (ejtag_ctrl & EJTAG_CTRL_PRNW) {
+			retval = mips32_pracc_exec_write(&ctx, address);
+			if (retval != ERROR_OK)
 				return retval;
-		}
-		else
-		{
+		} else {
 			/* Check to see if its reading at the debug vector. The first pass through
 			 * the module is always read at the vector, so the first one we allow.  When
 			 * the second read from the vector occurs we are done and just exit. */
 			if ((address == MIPS32_PRACC_TEXT) && (pass++))
-			{
 				break;
-			}
 
-			if ((retval = mips32_pracc_exec_read(&ctx, address)) != ERROR_OK)
+			retval = mips32_pracc_exec_read(&ctx, address);
+			if (retval != ERROR_OK)
 				return retval;
 		}
 
@@ -275,647 +268,716 @@ int mips32_pracc_exec(struct mips_ejtag *ejtag_info, int code_len, const uint32_
 
 	/* stack sanity check */
 	if (ctx.stack_offset != 0)
-	{
 		LOG_DEBUG("Pracc Stack not zero");
-	}
 
 	return ERROR_OK;
 }
 
-int mips32_pracc_read_mem(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
+inline void pracc_queue_init(struct pracc_queue_info *ctx)
 {
-	switch (size)
-	{
-		case 1:
-			return mips32_pracc_read_mem8(ejtag_info, addr, count, (uint8_t*)buf);
-		case 2:
-			return mips32_pracc_read_mem16(ejtag_info, addr, count, (uint16_t*)buf);
-		case 4:
-			if (count == 1)
-				return mips32_pracc_read_u32(ejtag_info, addr, (uint32_t*)buf);
-			else
-				return mips32_pracc_read_mem32(ejtag_info, addr, count, (uint32_t*)buf);
+	ctx->retval = ERROR_OK;
+	ctx->code_count = 0;
+	ctx->store_count = 0;
+
+	ctx->pracc_list = malloc(2 * ctx->max_code * sizeof(uint32_t));
+	if (ctx->pracc_list == NULL) {
+		LOG_ERROR("Out of memory");
+		ctx->retval = ERROR_FAIL;
 	}
-
-	return ERROR_OK;
 }
 
-static int mips32_pracc_read_mem32(struct mips_ejtag *ejtag_info, uint32_t addr, int count, uint32_t *buf)
+inline void pracc_add(struct pracc_queue_info *ctx, uint32_t addr, uint32_t instr)
 {
-	static const uint32_t code[] = {
-															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-		MIPS32_SW(10,0,15),									/* sw $10,($15) */
-		MIPS32_SW(11,0,15),									/* sw $11,($15) */
-
-		MIPS32_LUI(8,UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(8,8,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(9,0,8),									/* $9 = mem[$8]; read addr */
-		MIPS32_LW(10,4,8),									/* $10 = mem[$8 + 4]; read count */
-		MIPS32_LUI(11,UPPER16(MIPS32_PRACC_PARAM_OUT)),		/* $11 = MIPS32_PRACC_PARAM_OUT */
-		MIPS32_ORI(11,11,LOWER16(MIPS32_PRACC_PARAM_OUT)),
-															/* loop: */
-		MIPS32_BEQ(0,10,8),									/* beq 0, $10, end */
-		MIPS32_NOP,
+	ctx->pracc_list[ctx->max_code + ctx->code_count] = addr;
+	ctx->pracc_list[ctx->code_count++] = instr;
+	if (addr)
+		ctx->store_count++;
+}
 
-		MIPS32_LW(8,0,9),									/* lw $8,0($9), Load $8 with the word @mem[$9] */
-		MIPS32_SW(8,0,11),									/* sw $8,0($11) */
+inline void pracc_queue_free(struct pracc_queue_info *ctx)
+{
+	if (ctx->code_count > ctx->max_code)	/* Only for internal check, will be erased */
+		LOG_ERROR("Internal error, code count: %d > max code: %d", ctx->code_count, ctx->max_code);
+	if (ctx->pracc_list != NULL)
+		free(ctx->pracc_list);
+}
 
-		MIPS32_ADDI(10,10,NEG16(1)),						/* $10-- */
-		MIPS32_ADDI(9,9,4),									/* $1 += 4 */
-		MIPS32_ADDI(11,11,4),								/* $11 += 4 */
+int mips32_pracc_queue_exec(struct mips_ejtag *ejtag_info, struct pracc_queue_info *ctx, uint32_t *buf)
+{
+	if (ejtag_info->mode == 0)
+		return mips32_pracc_exec(ejtag_info, ctx->code_count, ctx->pracc_list, 0, NULL,
+				  ctx->store_count, buf, ctx->code_count - 1);
+
+	union scan_in {
+		uint8_t scan_96[12];
+		struct {
+			uint8_t ctrl[4];
+			uint8_t data[4];
+			uint8_t addr[4];
+		} scan_32;
+
+	} *scan_in = malloc(sizeof(union scan_in) * (ctx->code_count + ctx->store_count));
+	if (scan_in == NULL) {
+		LOG_ERROR("Out of memory");
+		return ERROR_FAIL;
+	}
 
-		MIPS32_B(NEG16(8)),									/* b loop */
-		MIPS32_NOP,
-															/* end: */
-		MIPS32_LW(11,0,15),									/* lw $11,($15) */
-		MIPS32_LW(10,0,15),									/* lw $10,($15) */
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(27)),								/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
-	};
+	unsigned num_clocks =
+		((uint64_t)(ejtag_info->scan_delay) * jtag_get_speed_khz() + 500000) / 1000000;
 
-	int retval = ERROR_OK;
-	int blocksize;
-	int bytesread;
-	uint32_t param_in[2];
+	uint32_t ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
+	mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ALL);
 
-	bytesread = 0;
+	int scan_count = 0;
+	for (int i = 0; i != 2 * ctx->code_count; i++) {
+		uint32_t data = 0;
+		if (i & 1u) {			/* Check store address from previous instruction, if not the first */
+			if (i < 2 || 0 == ctx->pracc_list[ctx->max_code + (i / 2) - 1])
+				continue;
+		} else
+			data = ctx->pracc_list[i / 2];
 
-	while (count > 0)
-	{
-		blocksize = count;
-		if (count > 0x400)
-			blocksize = 0x400;
+		jtag_add_clocks(num_clocks);
+		mips_ejtag_add_scan_96(ejtag_info, ejtag_ctrl, data, scan_in[scan_count++].scan_96);
+	}
 
-		param_in[0] = addr;
-		param_in[1] = blocksize;
+	int retval = jtag_execute_queue();		/* execute queued scans */
+	if (retval != ERROR_OK)
+		goto exit;
+
+	uint32_t fetch_addr = MIPS32_PRACC_TEXT;		/* start address */
+	scan_count = 0;
+	for (int i = 0; i != 2 * ctx->code_count; i++) {				/* verify every pracc access */
+		uint32_t store_addr = 0;
+		if (i & 1u) {			/* Read store addres from previous instruction, if not the first */
+			store_addr = ctx->pracc_list[ctx->max_code + (i / 2) - 1];
+			if (i < 2 || 0 == store_addr)
+				continue;
+		}
 
-		if ((retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code,
-			ARRAY_SIZE(param_in), param_in, blocksize, &buf[bytesread], 1)) != ERROR_OK)
-		{
-			return retval;
+		ejtag_ctrl = buf_get_u32(scan_in[scan_count].scan_32.ctrl, 0, 32);
+		if (!(ejtag_ctrl & EJTAG_CTRL_PRACC)) {
+			LOG_ERROR("Error: access not pending  count: %d", scan_count);
+			retval = ERROR_FAIL;
+			goto exit;
 		}
 
-		count -= blocksize;
-		addr += blocksize;
-		bytesread += blocksize;
-	}
+		uint32_t addr = buf_get_u32(scan_in[scan_count].scan_32.addr, 0, 32);
 
+		if (store_addr != 0) {
+			if (!(ejtag_ctrl & EJTAG_CTRL_PRNW)) {
+				LOG_ERROR("Not a store/write access, count: %d", scan_count);
+				retval = ERROR_FAIL;
+				goto exit;
+			}
+			if (addr != store_addr) {
+				LOG_ERROR("Store address mismatch, read: %x expected: %x count: %d",
+						addr, store_addr, scan_count);
+				retval = ERROR_FAIL;
+				goto exit;
+			}
+			int buf_index = (addr - MIPS32_PRACC_PARAM_OUT) / 4;
+			buf[buf_index] = buf_get_u32(scan_in[scan_count].scan_32.data, 0, 32);
+
+		} else {
+			if (ejtag_ctrl & EJTAG_CTRL_PRNW) {
+				LOG_ERROR("Not a fetch/read access, count: %d", scan_count);
+				retval = ERROR_FAIL;
+				goto exit;
+			}
+			if (addr != fetch_addr) {
+				LOG_ERROR("Fetch addr mismatch, read: %x expected: %x count: %d", addr, fetch_addr, scan_count);
+				retval = ERROR_FAIL;
+				goto exit;
+			}
+			fetch_addr += 4;
+		}
+		scan_count++;
+	}
+exit:
+	free(scan_in);
 	return retval;
 }
 
-static int mips32_pracc_read_u32(struct mips_ejtag *ejtag_info, uint32_t addr, uint32_t *buf)
+int mips32_pracc_read_u32(struct mips_ejtag *ejtag_info, uint32_t addr, uint32_t *buf)
 {
-	static const uint32_t code[] = {
-															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-
-		MIPS32_LW(8,NEG16(MIPS32_PRACC_STACK-MIPS32_PRACC_PARAM_IN),15), /* load R8 @ param_in[0] = address */
-
-		MIPS32_LW(8,0,8),									/* lw $8,0($8), Load $8 with the word @mem[$8] */
-		MIPS32_SW(8,NEG16(MIPS32_PRACC_STACK-MIPS32_PRACC_PARAM_OUT),15), /* store R8 @ param_out[0] */
+	struct pracc_queue_info ctx = {.max_code = 9};
+	pracc_queue_init(&ctx);
+	if (ctx.retval != ERROR_OK)
+		goto exit;
+
+	pracc_add(&ctx, 0, MIPS32_MTC0(15, 31, 0));					/* move $15 to COP0 DeSave */
+	pracc_add(&ctx, 0, MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR));			/* $15 = MIPS32_PRACC_BASE_ADDR */
+	pracc_add(&ctx, 0, MIPS32_LUI(8, UPPER16((addr + 0x8000))));		/* load  $8 with modified upper address */
+	pracc_add(&ctx, 0, MIPS32_LW(8, LOWER16(addr), 8));				/* lw $8, LOWER16(addr)($8) */
+	pracc_add(&ctx, MIPS32_PRACC_PARAM_OUT,
+				MIPS32_SW(8, PRACC_OUT_OFFSET, 15));			/* sw $8,PRACC_OUT_OFFSET($15) */
+	pracc_add(&ctx, 0, MIPS32_LUI(8, UPPER16(ejtag_info->reg8)));		/* restore upper 16 of $8 */
+	pracc_add(&ctx, 0, MIPS32_ORI(8, 8, LOWER16(ejtag_info->reg8)));		/* restore lower 16 of $8 */
+	pracc_add(&ctx, 0, MIPS32_B(NEG16(ctx.code_count + 1)));					/* jump to start */
+	pracc_add(&ctx, 0, MIPS32_MFC0(15, 31, 0));					/* move COP0 DeSave to $15 */
+
+	ctx.retval = mips32_pracc_queue_exec(ejtag_info, &ctx, buf);
+exit:
+	pracc_queue_free(&ctx);
+	return ctx.retval;
+}
 
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(9)),									/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
-	};
+int mips32_pracc_read_mem(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
+{
+	if (count == 1 && size == 4)
+		return mips32_pracc_read_u32(ejtag_info, addr, (uint32_t *)buf);
+
+	uint32_t *data = NULL;
+	struct pracc_queue_info ctx = {.max_code = 256 * 3 + 9 + 1};	/* alloc memory for the worst case */
+	pracc_queue_init(&ctx);
+	if (ctx.retval != ERROR_OK)
+		goto exit;
+
+	if (size != 4) {
+		data = malloc(256 * sizeof(uint32_t));
+		if (data == NULL) {
+			LOG_ERROR("Out of memory");
+			goto exit;
+		}
+	}
 
-	int retval = ERROR_OK;
-	uint32_t param_in[1];
+	uint32_t *buf32 = buf;
+	uint16_t *buf16 = buf;
+	uint8_t *buf8 = buf;
+
+	while (count) {
+		ctx.code_count = 0;
+		ctx.store_count = 0;
+		int this_round_count = (count > 256) ? 256 : count;
+		uint32_t last_upper_base_addr = UPPER16((addr + 0x8000));
+
+		pracc_add(&ctx, 0, MIPS32_MTC0(15, 31, 0));					/* save $15 in DeSave */
+		pracc_add(&ctx, 0, MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR));			/* $15 = MIPS32_PRACC_BASE_ADDR */
+		pracc_add(&ctx, 0, MIPS32_LUI(9, last_upper_base_addr));		/* load the upper memory address in $9 */
+
+		for (int i = 0; i != this_round_count; i++) {			/* Main code loop */
+			uint32_t upper_base_addr = UPPER16((addr + 0x8000));
+			if (last_upper_base_addr != upper_base_addr) {			/* if needed, change upper address in $9 */
+				pracc_add(&ctx, 0, MIPS32_LUI(9, upper_base_addr));
+				last_upper_base_addr = upper_base_addr;
+			}
 
-	param_in[0] = addr;
+			if (size == 4)
+				pracc_add(&ctx, 0, MIPS32_LW(8, LOWER16(addr), 9));		/* load from memory to $8 */
+			else if (size == 2)
+				pracc_add(&ctx, 0, MIPS32_LHU(8, LOWER16(addr), 9));
+			else
+				pracc_add(&ctx, 0, MIPS32_LBU(8, LOWER16(addr), 9));
 
-	if ((retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code,
-		ARRAY_SIZE(param_in), param_in, 1, buf, 1)) != ERROR_OK)
-	{
-		return retval;
+			pracc_add(&ctx, MIPS32_PRACC_PARAM_OUT + i * 4,
+					  MIPS32_SW(8, PRACC_OUT_OFFSET + i * 4, 15));		/* store $8 at param out */
+			addr += size;
+		}
+		pracc_add(&ctx, 0, MIPS32_LUI(8, UPPER16(ejtag_info->reg8)));		/* restore upper 16 bits of reg 8 */
+		pracc_add(&ctx, 0, MIPS32_ORI(8, 8, LOWER16(ejtag_info->reg8)));	/* restore lower 16 bits of reg 8 */
+		pracc_add(&ctx, 0, MIPS32_LUI(9, UPPER16(ejtag_info->reg9)));		/* restore upper 16 bits of reg 9 */
+		pracc_add(&ctx, 0, MIPS32_ORI(9, 9, LOWER16(ejtag_info->reg9)));	/* restore lower 16 bits of reg 9 */
+
+		pracc_add(&ctx, 0, MIPS32_B(NEG16(ctx.code_count + 1)));				/* jump to start */
+		pracc_add(&ctx, 0, MIPS32_MFC0(15, 31, 0));					/* restore $15 from DeSave */
+
+		if (size == 4) {
+			ctx.retval = mips32_pracc_queue_exec(ejtag_info, &ctx, buf32);
+			if (ctx.retval != ERROR_OK)
+				goto exit;
+			buf32 += this_round_count;
+		} else {
+			ctx.retval = mips32_pracc_queue_exec(ejtag_info, &ctx, data);
+			if (ctx.retval != ERROR_OK)
+				goto exit;
+
+			uint32_t *data_p = data;
+			for (int i = 0; i != this_round_count; i++) {
+				if (size == 2)
+					*buf16++ = *data_p++;
+				else
+					*buf8++ = *data_p++;
+			}
+		}
+		count -= this_round_count;
 	}
+exit:
+	pracc_queue_free(&ctx);
+	if (data != NULL)
+		free(data);
+	return ctx.retval;
+}
 
-	return retval;
+int mips32_cp0_read(struct mips_ejtag *ejtag_info, uint32_t *val, uint32_t cp0_reg, uint32_t cp0_sel)
+{
+	struct pracc_queue_info ctx = {.max_code = 8};
+	pracc_queue_init(&ctx);
+	if (ctx.retval != ERROR_OK)
+		goto exit;
+
+	pracc_add(&ctx, 0, MIPS32_MTC0(15, 31, 0));					/* move $15 to COP0 DeSave */
+	pracc_add(&ctx, 0, MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR));			/* $15 = MIPS32_PRACC_BASE_ADDR */
+	pracc_add(&ctx, 0, MIPS32_MFC0(8, 0, 0) | (cp0_reg << 11) | cp0_sel);	/* move COP0 [cp0_reg select] to $8 */
+	pracc_add(&ctx, MIPS32_PRACC_PARAM_OUT,
+				MIPS32_SW(8, PRACC_OUT_OFFSET, 15));			/* store $8 to pracc_out */
+	pracc_add(&ctx, 0, MIPS32_MFC0(15, 31, 0));					/* move COP0 DeSave to $15 */
+	pracc_add(&ctx, 0, MIPS32_LUI(8, UPPER16(ejtag_info->reg8)));		/* restore upper 16 bits  of $8 */
+	pracc_add(&ctx, 0, MIPS32_B(NEG16(ctx.code_count + 1)));					/* jump to start */
+	pracc_add(&ctx, 0, MIPS32_ORI(8, 8, LOWER16(ejtag_info->reg8)));		/* restore lower 16 bits of $8 */
+
+	ctx.retval = mips32_pracc_queue_exec(ejtag_info, &ctx, val);
+exit:
+	pracc_queue_free(&ctx);
+	return ctx.retval;
+
+	/**
+	 * Note that our input parametes cp0_reg and cp0_sel
+	 * are numbers (not gprs) which make part of mfc0 instruction opcode.
+	 *
+	 * These are not fix, but can be different for each mips32_cp0_read() function call,
+	 * and that is why we must insert them directly into opcode,
+	 * i.e. we can not pass it on EJTAG microprogram stack (via param_in),
+	 * and put them into the gprs later from MIPS32_PRACC_STACK
+	 * because mfc0 do not use gpr as a parameter for the cp0_reg and select part,
+	 * but plain (immediate) number.
+	 *
+	 * MIPS32_MTC0 is implemented via MIPS32_R_INST macro.
+	 * In order to insert our parameters, we must change rd and funct fields.
+	 *
+	 * code[2] |= (cp0_reg << 11) | cp0_sel;   change rd and funct of MIPS32_R_INST macro
+	 **/
 }
 
-static int mips32_pracc_read_mem16(struct mips_ejtag *ejtag_info, uint32_t addr, int count, uint16_t *buf)
+int mips32_cp0_write(struct mips_ejtag *ejtag_info, uint32_t val, uint32_t cp0_reg, uint32_t cp0_sel)
 {
-	static const uint32_t code[] = {
+	uint32_t code[] = {
 															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-		MIPS32_SW(10,0,15),									/* sw $10,($15) */
-		MIPS32_SW(11,0,15),									/* sw $11,($15) */
-
-		MIPS32_LUI(8,UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(8,8,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(9,0,8),									/* $9 = mem[$8]; read addr */
-		MIPS32_LW(10,4,8),									/* $10 = mem[$8 + 4]; read count */
-		MIPS32_LUI(11,UPPER16(MIPS32_PRACC_PARAM_OUT)),		/* $11 = MIPS32_PRACC_PARAM_OUT */
-		MIPS32_ORI(11,11,LOWER16(MIPS32_PRACC_PARAM_OUT)),
-															/* loop: */
-		MIPS32_BEQ(0,10,8),									/* beq 0, $10, end */
-		MIPS32_NOP,
+		MIPS32_MTC0(15, 31, 0),							/* move $15 to COP0 DeSave */
+		MIPS32_LUI(15, UPPER16(val)),						/* Load val to $15 */
+		MIPS32_ORI(15, 15, LOWER16(val)),
 
-		MIPS32_LHU(8,0,9),									/* lw $8,0($9), Load $8 with the halfword @mem[$9] */
-		MIPS32_SW(8,0,11),									/* sw $8,0($11) */
+		/* 3 */ MIPS32_MTC0(15, 0, 0),						/* move $15 to COP0 [cp0_reg select] */
 
-		MIPS32_ADDI(10,10,NEG16(1)),						/* $10-- */
-		MIPS32_ADDI(9,9,2),									/* $9 += 2 */
-		MIPS32_ADDI(11,11,4),								/* $11 += 4 */
-		MIPS32_B(NEG16(8)),									/* b loop */
-		MIPS32_NOP,
-															/* end: */
-		MIPS32_LW(11,0,15),									/* lw $11,($15) */
-		MIPS32_LW(10,0,15),									/* lw $10,($15) */
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(27)),								/* b start */
-		MIPS32_MFC0(15,30,0),								/* move COP0 DeSave to $15 */
+		MIPS32_B(NEG16(5)),							/* b start */
+		MIPS32_MFC0(15, 31, 0),							/* move COP0 DeSave to $15 */
 	};
 
-	/* TODO remove array */
-	uint32_t *param_out = malloc(count * sizeof(uint32_t));
-	int i;
-
-//	int retval;
-	int blocksize;
-	int bytesread;
-	uint32_t param_in[2];
-
-	bytesread = 0;
+	/**
+	 * Note that MIPS32_MTC0 macro is implemented via MIPS32_R_INST macro.
+	 * In order to insert our parameters, we must change rd and funct fields.
+	 */
+	code[3] |= (cp0_reg << 11) | cp0_sel;  /* change rd and funct fields of MIPS32_R_INST macro */
 
-	//while (count > 0)
-	{
-		blocksize = count;
-		if (count > 0x400)
-			blocksize = 0x400;
-
-		param_in[0] = addr;
-		param_in[1] = blocksize;
-
-		mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-			ARRAY_SIZE(param_in), param_in, count, param_out, 1);
-
-//		count -= blocksize;
-//		addr += blocksize;
-//		bytesread += blocksize;
-	}
-
-	for (i = 0; i < count; i++)
-	{
-		buf[i] = param_out[i];
-	}
-
-	free(param_out);
-
-	return ERROR_OK;
+	return mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 0, NULL, 0, NULL, 1);
 }
 
-static int mips32_pracc_read_mem8(struct mips_ejtag *ejtag_info, uint32_t addr, int count, uint8_t *buf)
+/**
+ * \b mips32_pracc_sync_cache
+ *
+ * Synchronize Caches to Make Instruction Writes Effective
+ * (ref. doc. MIPS32 Architecture For Programmers Volume II: The MIPS32 Instruction Set,
+ *  Document Number: MD00086, Revision 2.00, June 9, 2003)
+ *
+ * When the instruction stream is written, the SYNCI instruction should be used
+ * in conjunction with other instructions to make the newly-written instructions effective.
+ *
+ * Explanation :
+ * A program that loads another program into memory is actually writing the D- side cache.
+ * The instructions it has loaded can't be executed until they reach the I-cache.
+ *
+ * After the instructions have been written, the loader should arrange
+ * to write back any containing D-cache line and invalidate any locations
+ * already in the I-cache.
+ *
+ * You can do that with cache instructions, but those instructions are only available in kernel mode,
+ * and a loader writing instructions for the use of its own process need not be privileged software.
+ *
+ * In the latest MIPS32/64 CPUs, MIPS provides the synci instruction,
+ * which does the whole job for a cache-line-sized chunk of the memory you just loaded:
+ * That is, it arranges a D-cache write-back and an I-cache invalidate.
+ *
+ * To employ synci at user level, you need to know the size of a cache line,
+ * and that can be obtained with a rdhwr SYNCI_Step
+ * from one of the standard “hardware registers”.
+ */
+static int mips32_pracc_sync_cache(struct mips_ejtag *ejtag_info,
+		uint32_t start_addr, uint32_t end_addr)
 {
 	static const uint32_t code[] = {
 															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-		MIPS32_SW(10,0,15),									/* sw $10,($15) */
-		MIPS32_SW(11,0,15),									/* sw $11,($15) */
-
-		MIPS32_LUI(8,UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(8,8,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(9,0,8),									/* $9 = mem[$8]; read addr */
-		MIPS32_LW(10,4,8),									/* $10 = mem[$8 + 4]; read count */
-		MIPS32_LUI(11,UPPER16(MIPS32_PRACC_PARAM_OUT)),		/* $11 = MIPS32_PRACC_PARAM_OUT */
-		MIPS32_ORI(11,11,LOWER16(MIPS32_PRACC_PARAM_OUT)),
-															/* loop: */
-		MIPS32_BEQ(0,10,8),									/* beq 0, $10, end */
-		MIPS32_NOP,
-
-		MIPS32_LBU(8,0,9),									/* lw $8,0($9), Load t4 with the byte @mem[t1] */
-		MIPS32_SW(8,0,11),									/* sw $8,0($11) */
-
-		MIPS32_ADDI(10,10,NEG16(1)),						/* $10-- */
-		MIPS32_ADDI(9,9,1),									/* $9 += 1 */
-		MIPS32_ADDI(11,11,4),								/* $11 += 4 */
-		MIPS32_B(NEG16(8)),									/* b loop */
+		MIPS32_MTC0(15, 31, 0),								/* move $15 to COP0 DeSave */
+		MIPS32_LUI(15, UPPER16(MIPS32_PRACC_STACK)),		/* $15 = MIPS32_PRACC_STACK */
+		MIPS32_ORI(15, 15, LOWER16(MIPS32_PRACC_STACK)),
+		MIPS32_SW(8, 0, 15),								/* sw $8,($15) */
+		MIPS32_SW(9, 0, 15),								/* sw $9,($15) */
+		MIPS32_SW(10, 0, 15),								/* sw $10,($15) */
+		MIPS32_SW(11, 0, 15),								/* sw $11,($15) */
+
+		MIPS32_LUI(8, UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
+		MIPS32_ORI(8, 8, LOWER16(MIPS32_PRACC_PARAM_IN)),
+		MIPS32_LW(9, 0, 8),									/* Load write start_addr to $9 */
+		MIPS32_LW(10, 4, 8),								/* Load write end_addr to $10 */
+
+		MIPS32_RDHWR(11, MIPS32_SYNCI_STEP),				/* $11 = MIPS32_SYNCI_STEP */
+		MIPS32_BEQ(11, 0, 6),								/* beq $11, $0, end */
 		MIPS32_NOP,
+															/* synci_loop : */
+		MIPS32_SYNCI(0, 9),									/* synci 0($9) */
+		MIPS32_SLTU(8, 10, 9),								/* sltu $8, $10, $9  # $8 = $10 < $9 ? 1 : 0 */
+		MIPS32_BNE(8, 0, NEG16(3)),							/* bne $8, $0, synci_loop */
+		MIPS32_ADDU(9, 9, 11),								/* $9 += MIPS32_SYNCI_STEP */
+		MIPS32_SYNC,
 															/* end: */
-		MIPS32_LW(11,0,15),									/* lw $11,($15) */
-		MIPS32_LW(10,0,15),									/* lw $10,($15) */
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(27)),								/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
+		MIPS32_LW(11, 0, 15),								/* lw $11,($15) */
+		MIPS32_LW(10, 0, 15),								/* lw $10,($15) */
+		MIPS32_LW(9, 0, 15),								/* lw $9,($15) */
+		MIPS32_LW(8, 0, 15),								/* lw $8,($15) */
+		MIPS32_B(NEG16(24)),								/* b start */
+		MIPS32_MFC0(15, 31, 0),								/* move COP0 DeSave to $15 */
 	};
 
 	/* TODO remove array */
-	uint32_t *param_out = malloc(count * sizeof(uint32_t));
-	int i;
-
-//	int retval;
-	int blocksize;
-	int bytesread;
-	uint32_t param_in[2];
-
-	bytesread = 0;
-
-//	while (count > 0)
-	{
-		blocksize = count;
-		if (count > 0x400)
-			blocksize = 0x400;
-
-		param_in[0] = addr;
-		param_in[1] = blocksize;
-
-		mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-			ARRAY_SIZE(param_in), param_in, count, param_out, 1);
-
-//		count -= blocksize;
-//		addr += blocksize;
-//		bytesread += blocksize;
-	}
-
-	for (i = 0; i < count; i++)
-	{
-		buf[i] = param_out[i];
-	}
-
-	free(param_out);
+	uint32_t *param_in = malloc(2 * sizeof(uint32_t));
+	int retval;
+	param_in[0] = start_addr;
+	param_in[1] = end_addr;
 
-	return ERROR_OK;
-}
+	retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 2, param_in, 0, NULL, 1);
 
-int mips32_pracc_write_mem(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
-{
-	switch (size)
-	{
-		case 1:
-			return mips32_pracc_write_mem8(ejtag_info, addr, count, (uint8_t*)buf);
-		case 2:
-			return mips32_pracc_write_mem16(ejtag_info, addr, count,(uint16_t*)buf);
-		case 4:
-			if (count == 1)
-				return mips32_pracc_write_u32(ejtag_info, addr, (uint32_t*)buf);
-			else
-				return mips32_pracc_write_mem32(ejtag_info, addr, count, (uint32_t*)buf);
-	}
+	free(param_in);
 
-	return ERROR_OK;
+	return retval;
 }
 
-static int mips32_pracc_write_mem32(struct mips_ejtag *ejtag_info, uint32_t addr, int count, uint32_t *buf)
+/**
+ * \b mips32_pracc_clean_invalidate_cache
+ *
+ * Writeback D$ and Invalidate I$
+ * so that the instructions written can be visible to CPU
+ */
+static int mips32_pracc_clean_invalidate_cache(struct mips_ejtag *ejtag_info,
+													uint32_t start_addr, uint32_t end_addr)
 {
 	static const uint32_t code[] = {
 															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-		MIPS32_SW(10,0,15),									/* sw $10,($15) */
-		MIPS32_SW(11,0,15),									/* sw $11,($15) */
-
-		MIPS32_ADDI(8,15,NEG16(MIPS32_PRACC_STACK-MIPS32_PRACC_PARAM_IN)),  /* $8= MIPS32_PRACC_PARAM_IN */
-		MIPS32_LW(9,0,8),									/* Load write addr to $9 */
-		MIPS32_LW(10,4,8),									/* Load write count to $10 */
-		MIPS32_ADDI(8,8,8),									/* $8 += 8 beginning of data */
-
-															/* loop: */
-		MIPS32_LW(11,0,8),									/* lw $11,0($8), Load $11 with the word @mem[$8] */
-		MIPS32_SW(11,0,9),									/* sw $11,0($9) */
-
-		MIPS32_ADDI(9,9,4),									/* $9 += 4 */
-		MIPS32_BNE(10,9,NEG16(4)),							/* bne $10, $9, loop */
-		MIPS32_ADDI(8,8,4),									/* $8 += 4 */
+		MIPS32_MTC0(15, 31, 0),								/* move $15 to COP0 DeSave */
+		MIPS32_LUI(15, UPPER16(MIPS32_PRACC_STACK)),		/* $15 = MIPS32_PRACC_STACK */
+		MIPS32_ORI(15, 15, LOWER16(MIPS32_PRACC_STACK)),
+		MIPS32_SW(8, 0, 15),								/* sw $8,($15) */
+		MIPS32_SW(9, 0, 15),								/* sw $9,($15) */
+		MIPS32_SW(10, 0, 15),								/* sw $10,($15) */
+		MIPS32_SW(11, 0, 15),								/* sw $11,($15) */
+
+		MIPS32_LUI(8, UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
+		MIPS32_ORI(8, 8, LOWER16(MIPS32_PRACC_PARAM_IN)),
+		MIPS32_LW(9, 0, 8),									/* Load write start_addr to $9 */
+		MIPS32_LW(10, 4, 8),								/* Load write end_addr to $10 */
+		MIPS32_LW(11, 8, 8),								/* Load write clsiz to $11 */
+
+															/* cache_loop: */
+		MIPS32_SLTU(8, 10, 9),								/* sltu $8, $10, $9  :  $8 <- $10 < $9 ? */
+		MIPS32_BGTZ(8, 6),									/* bgtz $8, end */
+		MIPS32_NOP,
+
+		MIPS32_CACHE(MIPS32_CACHE_D_HIT_WRITEBACK, 0, 9),		/* cache Hit_Writeback_D, 0($9) */
+		MIPS32_CACHE(MIPS32_CACHE_I_HIT_INVALIDATE, 0, 9),	/* cache Hit_Invalidate_I, 0($9) */
+
+		MIPS32_ADDU(9, 9, 11),								/* $9 += $11 */
 
+		MIPS32_B(NEG16(7)),									/* b cache_loop */
+		MIPS32_NOP,
 															/* end: */
-		MIPS32_LW(11,0,15),									/* lw $11,($15) */
-		MIPS32_LW(10,0,15),									/* lw $10,($15) */
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(21)),								/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
+		MIPS32_LW(11, 0, 15),								/* lw $11,($15) */
+		MIPS32_LW(10, 0, 15),								/* lw $10,($15) */
+		MIPS32_LW(9, 0, 15),								/* lw $9,($15) */
+		MIPS32_LW(8, 0, 15),								/* lw $8,($15) */
+		MIPS32_B(NEG16(25)),								/* b start */
+		MIPS32_MFC0(15, 31, 0),								/* move COP0 DeSave to $15 */
 	};
 
-	/* TODO remove array */
-	uint32_t *param_in = malloc((count + 2) * sizeof(uint32_t));
-	param_in[0] = addr;
-	param_in[1] = addr + (count * sizeof(uint32_t));	/* last address */
+	/**
+	 * Find cache line size in bytes
+	 */
+	uint32_t conf;
+	uint32_t dl, clsiz;
 
-	memcpy(&param_in[2], buf, count * sizeof(uint32_t));
+	mips32_cp0_read(ejtag_info, &conf, 16, 1);
+	dl = (conf & MIPS32_CONFIG1_DL_MASK) >> MIPS32_CONFIG1_DL_SHIFT;
 
-	mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-		count + 2, param_in, 0, NULL, 1);
+	/* dl encoding : dl=1 => 4 bytes, dl=2 => 8 bytes, etc... */
+	clsiz = 0x2 << dl;
+
+	/* TODO remove array */
+	uint32_t *param_in = malloc(3 * sizeof(uint32_t));
+	int retval;
+	param_in[0] = start_addr;
+	param_in[1] = end_addr;
+	param_in[2] = clsiz;
+
+	retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 3, param_in, 0, NULL, 1);
 
 	free(param_in);
 
-	return ERROR_OK;
+	return retval;
 }
 
-static int mips32_pracc_write_u32(struct mips_ejtag *ejtag_info, uint32_t addr, uint32_t *buf)
+static int mips32_pracc_write_mem_generic(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
 {
-	static const uint32_t code[] = {
-															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-
-		MIPS32_LW(8,NEG16((MIPS32_PRACC_STACK-MIPS32_PRACC_PARAM_IN)-4), 15),	/* load R8 @ param_in[1] = data */
-		MIPS32_LW(9,NEG16(MIPS32_PRACC_STACK-MIPS32_PRACC_PARAM_IN), 15),		/* load R9 @ param_in[0] = address */
-
-		MIPS32_SW(8,0,9),									/* sw $8,0($9) */
-
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(11)),								/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
-	};
+	uint32_t *code;
+	code = malloc((128 * 3 + 9) * sizeof(uint32_t));	/* alloc memory for the worst case */
+	if (code == NULL) {
+		LOG_ERROR("Out of memory");
+		return ERROR_FAIL;
+	}
 
-	/* TODO remove array */
-	uint32_t param_in[1 + 1];
-	param_in[0] = addr;
-	param_in[1] = *buf;
+	uint32_t *buf32 = buf;
+	uint16_t *buf16 = buf;
+	uint8_t *buf8 = buf;
 
-	mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-		ARRAY_SIZE(param_in), param_in, 0, NULL, 1);
+	int i;
+	int retval = ERROR_FAIL;
+	uint32_t *code_p;
+	uint32_t upper_base_addr, last_upper_base_addr;
+	int this_round_count;
+	int code_len;
 
-	return ERROR_OK;
-}
+	while (count) {
+		this_round_count = (count > 128) ? 128 : count;
+		last_upper_base_addr = UPPER16((addr + 0x8000));
+		code_p = code;
+
+		*code_p++ = MIPS32_MTC0(15, 31, 0);					/* save $15 in DeSave */
+		*code_p++ = MIPS32_LUI(15, last_upper_base_addr);			/* load $15 with memory base address */
+		code_len = 2;
+
+		for (i = 0; i != this_round_count; i++) {
+			upper_base_addr = UPPER16((addr + 0x8000));
+			if (last_upper_base_addr != upper_base_addr) {
+				*code_p++ = MIPS32_LUI(15, upper_base_addr);		/* if needed, change upper address in $15*/
+				code_len++;
+				last_upper_base_addr = upper_base_addr;
+			}
 
-static int mips32_pracc_write_mem16(struct mips_ejtag *ejtag_info, uint32_t addr, int count, uint16_t *buf)
-{
-	static const uint32_t code[] = {
-															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-		MIPS32_SW(10,0,15),									/* sw $10,($15) */
-		MIPS32_SW(11,0,15),									/* sw $11,($15) */
-
-		MIPS32_LUI(8,UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(8,8,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(9,0,8),									/* Load write addr to $9 */
-		MIPS32_LW(10,4,8),									/* Load write count to $10 */
-		MIPS32_ADDI(8,8,8),									/* $8 += 8 */
-															/* loop: */
-		MIPS32_BEQ(0,10,8),									/* beq $0, $10, end */
-		MIPS32_NOP,
+			if (size == 4) {	/* for word write check if one half word is 0 and load it accordingly */
+				if (LOWER16(*buf32) == 0) {
+					*code_p++ = MIPS32_LUI(8, UPPER16(*buf32));		/* load only upper value */
+					code_len++;
+				} else if (UPPER16(*buf32) == 0) {
+					*code_p++ = MIPS32_ORI(8, 0, LOWER16(*buf32));		/* load only lower value */
+					code_len++;
+				} else {
+					*code_p++ = MIPS32_LUI(8, UPPER16(*buf32));		/* load upper and lower */
+					*code_p++ = MIPS32_ORI(8, 8, LOWER16(*buf32));
+					code_len += 2;
+				}
+				*code_p++ = MIPS32_SW(8, LOWER16(addr), 15);			/* store word to memory */
+				code_len++;
+				buf32++;
+
+			} else if (size == 2) {
+				*code_p++ = MIPS32_ORI(8, 0, *buf16);				/* load lower value */
+				*code_p++ = MIPS32_SH(8, LOWER16(addr), 15);		/* store half word to memory */
+				code_len += 2;
+				buf16++;
+
+			} else {
+				*code_p++ = MIPS32_ORI(8, 0, *buf8);				/* load lower value */
+				*code_p++ = MIPS32_SB(8, LOWER16(addr), 15);			/* store byte to memory */
+				code_len += 2;
+				buf8++;
+			}
 
-		MIPS32_LW(11,0,8),									/* lw $11,0($8), Load $11 with the word @mem[$8] */
-		MIPS32_SH(11,0,9),									/* sh $11,0($9) */
+			addr += size;
+		}
 
-		MIPS32_ADDI(10,10,NEG16(1)),						/* $10-- */
-		MIPS32_ADDI(9,9,2),									/* $9 += 2 */
-		MIPS32_ADDI(8,8,4),									/* $8 += 4 */
+		*code_p++ = MIPS32_LUI(8, UPPER16(ejtag_info->reg8)),			/* restore upper 16 bits of reg 8 */
+		*code_p++ = MIPS32_ORI(8, 8, LOWER16(ejtag_info->reg8)),		/* restore lower 16 bits of reg 8 */
 
-		MIPS32_B(NEG16(8)),									/* b loop */
-		MIPS32_NOP,
-															/* end: */
-		MIPS32_LW(11,0,15),									/* lw $11,($15) */
-		MIPS32_LW(10,0,15),									/* lw $10,($15) */
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(26)),								/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
-	};
+		code_len += 4;
+		*code_p++ = MIPS32_B(NEG16(code_len - 1));					/* jump to start */
+		*code_p = MIPS32_MFC0(15, 31, 0);					/* restore $15 from DeSave */
 
-	/* TODO remove array */
-	uint32_t *param_in = malloc((count + 2) * sizeof(uint32_t));
-	int i;
-	param_in[0] = addr;
-	param_in[1] = count;
+		retval = mips32_pracc_exec(ejtag_info, code_len, code, 0, NULL, 0, NULL, 1);
+		if (retval != ERROR_OK)
+			goto exit;
 
-	for (i = 0; i < count; i++)
-	{
-		param_in[i + 2] = buf[i];
+		count -= this_round_count;
 	}
 
-	mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-		count + 2, param_in, 0, NULL, 1);
-
-	free(param_in);
-
-	return ERROR_OK;
+exit:
+	free(code);
+	return retval;
 }
 
-static int mips32_pracc_write_mem8(struct mips_ejtag *ejtag_info, uint32_t addr, int count, uint8_t *buf)
+int mips32_pracc_write_mem(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
 {
-	static const uint32_t code[] = {
-															/* start: */
-		MIPS32_MTC0(15,31,0),								/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),			/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(8,0,15),									/* sw $8,($15) */
-		MIPS32_SW(9,0,15),									/* sw $9,($15) */
-		MIPS32_SW(10,0,15),									/* sw $10,($15) */
-		MIPS32_SW(11,0,15),									/* sw $11,($15) */
-
-		MIPS32_LUI(8,UPPER16(MIPS32_PRACC_PARAM_IN)),		/* $8 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(8,8,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(9,0,8),									/* Load write addr to $9 */
-		MIPS32_LW(10,4,8),									/* Load write count to $10 */
-		MIPS32_ADDI(8,8,8),									/* $8 += 8 */
-															/* loop: */
-		MIPS32_BEQ(0,10,8),									/* beq $0, $10, end */
-		MIPS32_NOP,
-
-		MIPS32_LW(11,0,8),									/* lw $11,0($8), Load $11 with the word @mem[$8] */
-		MIPS32_SB(11,0,9),									/* sb $11,0($9) */
+	int retval = mips32_pracc_write_mem_generic(ejtag_info, addr, size, count, buf);
+	if (retval != ERROR_OK)
+		return retval;
 
-		MIPS32_ADDI(10,10,NEG16(1)),						/* $10-- */
-		MIPS32_ADDI(9,9,1),									/* $9 += 1 */
-		MIPS32_ADDI(8,8,4),									/* $8 += 4 */
+	/**
+	 * If we are in the cachable regoion and cache is activated,
+	 * we must clean D$ + invalidate I$ after we did the write,
+	 * so that changes do not continue to live only in D$, but to be
+	 * replicated in I$ also (maybe we wrote the istructions)
+	 */
+	uint32_t conf = 0;
+	int cached = 0;
 
-		MIPS32_B(NEG16(8)),									/* b loop */
-		MIPS32_NOP,
-															/* end: */
-		MIPS32_LW(11,0,15),									/* lw $11,($15) */
-		MIPS32_LW(10,0,15),									/* lw $10,($15) */
-		MIPS32_LW(9,0,15),									/* lw $9,($15) */
-		MIPS32_LW(8,0,15),									/* lw $8,($15) */
-		MIPS32_B(NEG16(26)),								/* b start */
-		MIPS32_MFC0(15,31,0),								/* move COP0 DeSave to $15 */
-	};
+	if ((KSEGX(addr) == KSEG1) || ((addr >= 0xff200000) && (addr <= 0xff3fffff)))
+		return retval; /*Nothing to do*/
 
-	/* TODO remove array */
-	uint32_t *param_in = malloc((count + 2) * sizeof(uint32_t));
-	int retval;
-	int i;
-	param_in[0] = addr;
-	param_in[1] = count;
+	mips32_cp0_read(ejtag_info, &conf, 16, 0);
 
-	for (i = 0; i < count; i++)
-	{
-		param_in[i + 2] = buf[i];
+	switch (KSEGX(addr)) {
+		case KUSEG:
+			cached = (conf & MIPS32_CONFIG0_KU_MASK) >> MIPS32_CONFIG0_KU_SHIFT;
+			break;
+		case KSEG0:
+			cached = (conf & MIPS32_CONFIG0_K0_MASK) >> MIPS32_CONFIG0_K0_SHIFT;
+			break;
+		case KSEG2:
+		case KSEG3:
+			cached = (conf & MIPS32_CONFIG0_K23_MASK) >> MIPS32_CONFIG0_K23_SHIFT;
+			break;
+		default:
+			/* what ? */
+			break;
 	}
 
-	retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-		count + 2, param_in, 0, NULL, 1);
-
-	free(param_in);
+	/**
+	 * Check cachablitiy bits coherency algorithm -
+	 * is the region cacheable or uncached.
+	 * If cacheable we have to synchronize the cache
+	 */
+	if (cached == 0x3) {
+		uint32_t start_addr, end_addr;
+		uint32_t rel;
+
+		start_addr = addr;
+		end_addr = addr + count * size;
+
+		/** select cache synchronisation mechanism based on Architecture Release */
+		rel = (conf & MIPS32_CONFIG0_AR_MASK) >> MIPS32_CONFIG0_AR_SHIFT;
+		switch (rel) {
+			case MIPS32_ARCH_REL1:
+				/* MIPS32/64 Release 1 - we must use cache instruction */
+				mips32_pracc_clean_invalidate_cache(ejtag_info, start_addr, end_addr);
+				break;
+			case MIPS32_ARCH_REL2:
+				/* MIPS32/64 Release 2 - we can use synci instruction */
+				mips32_pracc_sync_cache(ejtag_info, start_addr, end_addr);
+				break;
+			default:
+				/* what ? */
+				break;
+		}
+	}
 
 	return retval;
 }
 
 int mips32_pracc_write_regs(struct mips_ejtag *ejtag_info, uint32_t *regs)
 {
-	static const uint32_t code[] = {
-														/* start: */
-		MIPS32_LUI(2,UPPER16(MIPS32_PRACC_PARAM_IN)),	/* $2 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(2,2,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(1,1*4,2),								/* lw $1,1*4($2) */
-		MIPS32_LW(15,15*4,2),							/* lw $15,15*4($2) */
-		MIPS32_MTC0(15,31,0),							/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),		/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(1,0,15),								/* sw $1,($15) */
-		MIPS32_LUI(1,UPPER16(MIPS32_PRACC_PARAM_IN)),	/* $1 = MIPS32_PRACC_PARAM_IN */
-		MIPS32_ORI(1,1,LOWER16(MIPS32_PRACC_PARAM_IN)),
-		MIPS32_LW(3,3*4,1),								/* lw $3,3*4($1) */
-		MIPS32_LW(4,4*4,1),								/* lw $4,4*4($1) */
-		MIPS32_LW(5,5*4,1),								/* lw $5,5*4($1) */
-		MIPS32_LW(6,6*4,1),								/* lw $6,6*4($1) */
-		MIPS32_LW(7,7*4,1),								/* lw $7,7*4($1) */
-		MIPS32_LW(8,8*4,1),								/* lw $8,8*4($1) */
-		MIPS32_LW(9,9*4,1),								/* lw $9,9*4($1) */
-		MIPS32_LW(10,10*4,1),							/* lw $10,10*4($1) */
-		MIPS32_LW(11,11*4,1),							/* lw $11,11*4($1) */
-		MIPS32_LW(12,12*4,1),							/* lw $12,12*4($1) */
-		MIPS32_LW(13,13*4,1),							/* lw $13,13*4($1) */
-		MIPS32_LW(14,14*4,1),							/* lw $14,14*4($1) */
-		MIPS32_LW(16,16*4,1),							/* lw $16,16*4($1) */
-		MIPS32_LW(17,17*4,1),							/* lw $17,17*4($1) */
-		MIPS32_LW(18,18*4,1),							/* lw $18,18*4($1) */
-		MIPS32_LW(19,19*4,1),							/* lw $19,19*4($1) */
-		MIPS32_LW(20,20*4,1),							/* lw $20,20*4($1) */
-		MIPS32_LW(21,21*4,1),							/* lw $21,21*4($1) */
-		MIPS32_LW(22,22*4,1),							/* lw $22,22*4($1) */
-		MIPS32_LW(23,23*4,1),							/* lw $23,23*4($1) */
-		MIPS32_LW(24,24*4,1),							/* lw $24,24*4($1) */
-		MIPS32_LW(25,25*4,1),							/* lw $25,25*4($1) */
-		MIPS32_LW(26,26*4,1),							/* lw $26,26*4($1) */
-		MIPS32_LW(27,27*4,1),							/* lw $27,27*4($1) */
-		MIPS32_LW(28,28*4,1),							/* lw $28,28*4($1) */
-		MIPS32_LW(29,29*4,1),							/* lw $29,29*4($1) */
-		MIPS32_LW(30,30*4,1),							/* lw $30,30*4($1) */
-		MIPS32_LW(31,31*4,1),							/* lw $31,31*4($1) */
-
-		MIPS32_LW(2,32*4,1),							/* lw $2,32*4($1) */
-		MIPS32_MTC0(2,12,0),							/* move $2 to status */
-		MIPS32_LW(2,33*4,1),							/* lw $2,33*4($1) */
-		MIPS32_MTLO(2),									/* move $2 to lo */
-		MIPS32_LW(2,34*4,1),							/* lw $2,34*4($1) */
-		MIPS32_MTHI(2),									/* move $2 to hi */
-		MIPS32_LW(2,35*4,1),							/* lw $2,35*4($1) */
-		MIPS32_MTC0(2,8,0),								/* move $2 to badvaddr */
-		MIPS32_LW(2,36*4,1),							/* lw $2,36*4($1) */
-		MIPS32_MTC0(2,13,0),							/* move $2 to cause*/
-		MIPS32_LW(2,37*4,1),							/* lw $2,37*4($1) */
-		MIPS32_MTC0(2,24,0),							/* move $2 to depc (pc) */
-
-		MIPS32_LW(2,2*4,1),								/* lw $2,2*4($1) */
-		MIPS32_LW(1,0,15),								/* lw $1,($15) */
-		MIPS32_B(NEG16(53)),							/* b start */
-		MIPS32_MFC0(15,31,0),							/* move COP0 DeSave to $15 */
+	static const uint32_t cp0_write_code[] = {
+		MIPS32_MTC0(1, 12, 0),							/* move $1 to status */
+		MIPS32_MTLO(1),									/* move $1 to lo */
+		MIPS32_MTHI(1),									/* move $1 to hi */
+		MIPS32_MTC0(1, 8, 0),							/* move $1 to badvaddr */
+		MIPS32_MTC0(1, 13, 0),							/* move $1 to cause*/
+		MIPS32_MTC0(1, 24, 0),							/* move $1 to depc (pc) */
 	};
 
-	int retval;
+	uint32_t *code;
+	code = malloc((37 * 2 + 6 + 1) * sizeof(uint32_t));	/* alloc memory for the worst case */
+	if (code == NULL) {
+		LOG_ERROR("Out of memory");
+		return ERROR_FAIL;
+	}
 
-	retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-			MIPS32NUMCOREREGS, regs, 0, NULL, 1);
+	uint32_t *code_p = code;
+	int code_len = 0;
+	/* load registers 2 to 31 with lui an ori instructions, check if same instructions can be saved */
+	for (int i = 2; i < 32; i++) {
+		if (LOWER16((regs[i])) == 0) {
+			*code_p++ = MIPS32_LUI(i, UPPER16((regs[i])));		/* if lower half word is 0, lui instruction only */
+			code_len++;
+		} else if (UPPER16((regs[i])) == 0) {
+			*code_p++ = MIPS32_ORI(i, 0, LOWER16((regs[i])));	/* if upper half word is 0, ori with $0 only*/
+			code_len++;
+		} else {
+			*code_p++ = MIPS32_LUI(i, UPPER16((regs[i])));		/* default, load with lui and ori instructions */
+			*code_p++ = MIPS32_ORI(i, i, LOWER16((regs[i])));
+			code_len += 2;
+		}
+	}
+
+	for (int i = 0; i != 6; i++) {
+		*code_p++ = MIPS32_LUI(1, UPPER16((regs[i + 32])));		/* load CPO value in $1, with lui and ori */
+		*code_p++ = MIPS32_ORI(1, 1, LOWER16((regs[i + 32])));
+		*code_p++ = cp0_write_code[i];					/* write value from $1 to CPO register */
+		code_len += 3;
+	}
 
+	*code_p++ = MIPS32_LUI(1, UPPER16((regs[1])));				/* load upper half word in $1 */
+	code_len += 3;
+	*code_p++ = MIPS32_B(NEG16(code_len - 1)),							/* b start */
+	*code_p = MIPS32_ORI(1, 1, LOWER16((regs[1])));				/* load lower half word in $1 */
+
+	int retval = mips32_pracc_exec(ejtag_info, code_len, code, 0, NULL, 0, NULL, 1);
+	free(code);
+
+	ejtag_info->reg8 = regs[8];
+	ejtag_info->reg9 = regs[9];
 	return retval;
 }
 
 int mips32_pracc_read_regs(struct mips_ejtag *ejtag_info, uint32_t *regs)
 {
-	static const uint32_t code[] = {
-														/* start: */
-		MIPS32_MTC0(2,31,0),							/* move $2 to COP0 DeSave */
-		MIPS32_LUI(2,UPPER16(MIPS32_PRACC_PARAM_OUT)),	/* $2 = MIPS32_PRACC_PARAM_OUT */
-		MIPS32_ORI(2,2,LOWER16(MIPS32_PRACC_PARAM_OUT)),
-		MIPS32_SW(0,0*4,2),								/* sw $0,0*4($2) */
-		MIPS32_SW(1,1*4,2),								/* sw $1,1*4($2) */
-		MIPS32_SW(15,15*4,2),							/* sw $15,15*4($2) */
-		MIPS32_MFC0(2,31,0),							/* move COP0 DeSave to $2 */
-		MIPS32_MTC0(15,31,0),							/* move $15 to COP0 DeSave */
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_STACK)),		/* $15 = MIPS32_PRACC_STACK */
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_STACK)),
-		MIPS32_SW(1,0,15),								/* sw $1,($15) */
-		MIPS32_SW(2,0,15),								/* sw $2,($15) */
-		MIPS32_LUI(1,UPPER16(MIPS32_PRACC_PARAM_OUT)),	/* $1 = MIPS32_PRACC_PARAM_OUT */
-		MIPS32_ORI(1,1,LOWER16(MIPS32_PRACC_PARAM_OUT)),
-		MIPS32_SW(2,2*4,1),								/* sw $2,2*4($1) */
-		MIPS32_SW(3,3*4,1),								/* sw $3,3*4($1) */
-		MIPS32_SW(4,4*4,1),								/* sw $4,4*4($1) */
-		MIPS32_SW(5,5*4,1),								/* sw $5,5*4($1) */
-		MIPS32_SW(6,6*4,1),								/* sw $6,6*4($1) */
-		MIPS32_SW(7,7*4,1),								/* sw $7,7*4($1) */
-		MIPS32_SW(8,8*4,1),								/* sw $8,8*4($1) */
-		MIPS32_SW(9,9*4,1),								/* sw $9,9*4($1) */
-		MIPS32_SW(10,10*4,1),							/* sw $10,10*4($1) */
-		MIPS32_SW(11,11*4,1),							/* sw $11,11*4($1) */
-		MIPS32_SW(12,12*4,1),							/* sw $12,12*4($1) */
-		MIPS32_SW(13,13*4,1),							/* sw $13,13*4($1) */
-		MIPS32_SW(14,14*4,1),							/* sw $14,14*4($1) */
-		MIPS32_SW(16,16*4,1),							/* sw $16,16*4($1) */
-		MIPS32_SW(17,17*4,1),							/* sw $17,17*4($1) */
-		MIPS32_SW(18,18*4,1),							/* sw $18,18*4($1) */
-		MIPS32_SW(19,19*4,1),							/* sw $19,19*4($1) */
-		MIPS32_SW(20,20*4,1),							/* sw $20,20*4($1) */
-		MIPS32_SW(21,21*4,1),							/* sw $21,21*4($1) */
-		MIPS32_SW(22,22*4,1),							/* sw $22,22*4($1) */
-		MIPS32_SW(23,23*4,1),							/* sw $23,23*4($1) */
-		MIPS32_SW(24,24*4,1),							/* sw $24,24*4($1) */
-		MIPS32_SW(25,25*4,1),							/* sw $25,25*4($1) */
-		MIPS32_SW(26,26*4,1),							/* sw $26,26*4($1) */
-		MIPS32_SW(27,27*4,1),							/* sw $27,27*4($1) */
-		MIPS32_SW(28,28*4,1),							/* sw $28,28*4($1) */
-		MIPS32_SW(29,29*4,1),							/* sw $29,29*4($1) */
-		MIPS32_SW(30,30*4,1),							/* sw $30,30*4($1) */
-		MIPS32_SW(31,31*4,1),							/* sw $31,31*4($1) */
-
-		MIPS32_MFC0(2,12,0),							/* move status to $2 */
-		MIPS32_SW(2,32*4,1),							/* sw $2,32*4($1) */
-		MIPS32_MFLO(2),									/* move lo to $2 */
-		MIPS32_SW(2,33*4,1),							/* sw $2,33*4($1) */
-		MIPS32_MFHI(2),									/* move hi to $2 */
-		MIPS32_SW(2,34*4,1),							/* sw $2,34*4($1) */
-		MIPS32_MFC0(2,8,0),								/* move badvaddr to $2 */
-		MIPS32_SW(2,35*4,1),							/* sw $2,35*4($1) */
-		MIPS32_MFC0(2,13,0),							/* move cause to $2 */
-		MIPS32_SW(2,36*4,1),							/* sw $2,36*4($1) */
-		MIPS32_MFC0(2,24,0),							/* move depc (pc) to $2 */
-		MIPS32_SW(2,37*4,1),							/* sw $2,37*4($1) */
-
-		MIPS32_LW(2,0,15),								/* lw $2,($15) */
-		MIPS32_LW(1,0,15),								/* lw $1,($15) */
-		MIPS32_B(NEG16(58)),							/* b start */
-		MIPS32_MFC0(15,31,0),							/* move COP0 DeSave to $15 */
+	static int cp0_read_code[] = {
+		MIPS32_MFC0(8, 12, 0),							/* move status to $8 */
+		MIPS32_MFLO(8),									/* move lo to $8 */
+		MIPS32_MFHI(8),									/* move hi to $8 */
+		MIPS32_MFC0(8, 8, 0),							/* move badvaddr to $8 */
+		MIPS32_MFC0(8, 13, 0),							/* move cause to $8 */
+		MIPS32_MFC0(8, 24, 0),							/* move depc (pc) to $8 */
 	};
 
-	int retval;
+	struct pracc_queue_info ctx = {.max_code = 48};
+	pracc_queue_init(&ctx);
+	if (ctx.retval != ERROR_OK)
+		goto exit;
 
-	retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, \
-		0, NULL, MIPS32NUMCOREREGS, regs, 1);
+	pracc_add(&ctx, 0, MIPS32_MTC0(1, 31, 0));						/* move $1 to COP0 DeSave */
+	pracc_add(&ctx, 0, MIPS32_LUI(1, PRACC_UPPER_BASE_ADDR));				/* $1 = MIP32_PRACC_BASE_ADDR */
 
-	return retval;
+	for (int i = 2; i != 32; i++)					/* store GPR's 2 to 31 */
+		pracc_add(&ctx, MIPS32_PRACC_PARAM_OUT + (i * 4),
+				  MIPS32_SW(i, PRACC_OUT_OFFSET + (i * 4), 1));
+
+	for (int i = 0; i != 6; i++) {
+		pracc_add(&ctx, 0, cp0_read_code[i]);				/* load COP0 needed registers to $8 */
+		pracc_add(&ctx, MIPS32_PRACC_PARAM_OUT + (i + 32) * 4,			/* store $8 at PARAM OUT */
+				  MIPS32_SW(8, PRACC_OUT_OFFSET + (i + 32) * 4, 1));
+	}
+	pracc_add(&ctx, 0, MIPS32_MFC0(8, 31, 0));					/* move DeSave to $8, reg1 value */
+	pracc_add(&ctx, MIPS32_PRACC_PARAM_OUT + 4,					/* store reg1 value from $8 to param out */
+			  MIPS32_SW(8, PRACC_OUT_OFFSET + 4, 1));
+
+	pracc_add(&ctx, 0, MIPS32_B(NEG16(ctx.code_count + 1)));					/* jump to start */
+	pracc_add(&ctx, 0, MIPS32_MFC0(1, 31, 0));					/* move COP0 DeSave to $1, restore reg1 */
+
+	if (ejtag_info->mode == 0)
+		ctx.store_count++;	/* Needed by legacy code, due to offset from reg0 */
+
+	ctx.retval = mips32_pracc_queue_exec(ejtag_info, &ctx, regs);
+
+	ejtag_info->reg8 = regs[8];	/* reg8 is saved but not restored, next called function should restore it */
+	ejtag_info->reg9 = regs[9];
+exit:
+	pracc_queue_free(&ctx);
+	return ctx.retval;
 }
 
 /* fastdata upload/download requires an initialized working area
@@ -926,41 +988,41 @@ int mips32_pracc_read_regs(struct mips_ejtag *ejtag_info, uint32_t *regs)
  * 3. data ...
  */
 int mips32_pracc_fastdata_xfer(struct mips_ejtag *ejtag_info, struct working_area *source,
-								int write_t, uint32_t addr, int count, uint32_t *buf)
+		int write_t, uint32_t addr, int count, uint32_t *buf)
 {
 	uint32_t handler_code[] = {
 		/* caution when editing, table is modified below */
 		/* r15 points to the start of this code */
-		MIPS32_SW(8,MIPS32_FASTDATA_HANDLER_SIZE - 4,15),
-		MIPS32_SW(9,MIPS32_FASTDATA_HANDLER_SIZE - 8,15),
-		MIPS32_SW(10,MIPS32_FASTDATA_HANDLER_SIZE - 12,15),
-		MIPS32_SW(11,MIPS32_FASTDATA_HANDLER_SIZE - 16,15),
+		MIPS32_SW(8, MIPS32_FASTDATA_HANDLER_SIZE - 4, 15),
+		MIPS32_SW(9, MIPS32_FASTDATA_HANDLER_SIZE - 8, 15),
+		MIPS32_SW(10, MIPS32_FASTDATA_HANDLER_SIZE - 12, 15),
+		MIPS32_SW(11, MIPS32_FASTDATA_HANDLER_SIZE - 16, 15),
 		/* start of fastdata area in t0 */
-		MIPS32_LUI(8,UPPER16(MIPS32_PRACC_FASTDATA_AREA)),
-		MIPS32_ORI(8,8,LOWER16(MIPS32_PRACC_FASTDATA_AREA)),
-		MIPS32_LW(9,0,8),								/* start addr in t1 */
-		MIPS32_LW(10,0,8),								/* end addr to t2 */
+		MIPS32_LUI(8, UPPER16(MIPS32_PRACC_FASTDATA_AREA)),
+		MIPS32_ORI(8, 8, LOWER16(MIPS32_PRACC_FASTDATA_AREA)),
+		MIPS32_LW(9, 0, 8),								/* start addr in t1 */
+		MIPS32_LW(10, 0, 8),							/* end addr to t2 */
 														/* loop: */
-		/* 8 */ MIPS32_LW(11,0,0),						/* lw t3,[t8 | r9] */
-		/* 9 */ MIPS32_SW(11,0,0),						/* sw t3,[r9 | r8] */
-		MIPS32_BNE(10,9,NEG16(3)),						/* bne $t2,t1,loop */
-		MIPS32_ADDI(9,9,4),								/* addi t1,t1,4 */
-
-		MIPS32_LW(8,MIPS32_FASTDATA_HANDLER_SIZE - 4,15),
-		MIPS32_LW(9,MIPS32_FASTDATA_HANDLER_SIZE - 8,15),
-		MIPS32_LW(10,MIPS32_FASTDATA_HANDLER_SIZE - 12,15),
-		MIPS32_LW(11,MIPS32_FASTDATA_HANDLER_SIZE - 16,15),
-
-		MIPS32_LUI(15,UPPER16(MIPS32_PRACC_TEXT)),
-		MIPS32_ORI(15,15,LOWER16(MIPS32_PRACC_TEXT)),
-		MIPS32_JR(15),									/* jr start */
-		MIPS32_MFC0(15,31,0),							/* move COP0 DeSave to $15 */
+		/* 8 */ MIPS32_LW(11, 0, 0),					/* lw t3,[t8 | r9] */
+		/* 9 */ MIPS32_SW(11, 0, 0),					/* sw t3,[r9 | r8] */
+		MIPS32_BNE(10, 9, NEG16(3)),					/* bne $t2,t1,loop */
+		MIPS32_ADDI(9, 9, 4),							/* addi t1,t1,4 */
+
+		MIPS32_LW(8, MIPS32_FASTDATA_HANDLER_SIZE - 4, 15),
+		MIPS32_LW(9, MIPS32_FASTDATA_HANDLER_SIZE - 8, 15),
+		MIPS32_LW(10, MIPS32_FASTDATA_HANDLER_SIZE - 12, 15),
+		MIPS32_LW(11, MIPS32_FASTDATA_HANDLER_SIZE - 16, 15),
+
+		MIPS32_LUI(15, UPPER16(MIPS32_PRACC_TEXT)),
+		MIPS32_ORI(15, 15, LOWER16(MIPS32_PRACC_TEXT)),
+		MIPS32_JR(15),								/* jr start */
+		MIPS32_MFC0(15, 31, 0),						/* move COP0 DeSave to $15 */
 	};
 
 	uint32_t jmp_code[] = {
-		MIPS32_MTC0(15,31,0),			/* move $15 to COP0 DeSave */
-		/* 1 */ MIPS32_LUI(15,0),		/* addr of working area added below */
-		/* 2 */ MIPS32_ORI(15,15,0),	/* addr of working area added below */
+		MIPS32_MTC0(15, 31, 0),			/* move $15 to COP0 DeSave */
+		/* 1 */ MIPS32_LUI(15, 0),		/* addr of working area added below */
+		/* 2 */ MIPS32_ORI(15, 15, 0),	/* addr of working area added below */
 		MIPS32_JR(15),					/* jump to ram program */
 		MIPS32_NOP,
 	};
@@ -971,21 +1033,17 @@ int mips32_pracc_fastdata_xfer(struct mips_ejtag *ejtag_info, struct working_are
 	if (source->size < MIPS32_FASTDATA_HANDLER_SIZE)
 		return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
 
-	if (write_t)
-	{
-		handler_code[8] = MIPS32_LW(11,0,8);	/* load data from probe at fastdata area */
-		handler_code[9] = MIPS32_SW(11,0,9);	/* store data to RAM @ r9 */
-	}
-	else
-	{
-		handler_code[8] = MIPS32_LW(11,0,9);	/* load data from RAM @ r9 */
-		handler_code[9] = MIPS32_SW(11,0,8);	/* store data to probe at fastdata area */
+	if (write_t) {
+		handler_code[8] = MIPS32_LW(11, 0, 8);	/* load data from probe at fastdata area */
+		handler_code[9] = MIPS32_SW(11, 0, 9);	/* store data to RAM @ r9 */
+	} else {
+		handler_code[8] = MIPS32_LW(11, 0, 9);	/* load data from RAM @ r9 */
+		handler_code[9] = MIPS32_SW(11, 0, 8);	/* store data to probe at fastdata area */
 	}
 
 	/* write program into RAM */
-	if (write_t != ejtag_info->fast_access_save)
-	{
-		mips32_pracc_write_mem32(ejtag_info, source->address, ARRAY_SIZE(handler_code), handler_code);
+	if (write_t != ejtag_info->fast_access_save) {
+		mips32_pracc_write_mem_generic(ejtag_info, source->address, 4, ARRAY_SIZE(handler_code), handler_code);
 		/* save previous operation to speed to any consecutive read/writes */
 		ejtag_info->fast_access_save = write_t;
 	}
@@ -995,27 +1053,31 @@ int mips32_pracc_fastdata_xfer(struct mips_ejtag *ejtag_info, struct working_are
 	jmp_code[1] |= UPPER16(source->address);
 	jmp_code[2] |= LOWER16(source->address);
 
-	for (i = 0; i < (int) ARRAY_SIZE(jmp_code); i++)
-	{
-		if ((retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl)) != ERROR_OK)
+	for (i = 0; i < (int) ARRAY_SIZE(jmp_code); i++) {
+		retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
+		if (retval != ERROR_OK)
 			return retval;
 
 		mips_ejtag_set_instr(ejtag_info, EJTAG_INST_DATA);
-		mips_ejtag_drscan_32(ejtag_info, &jmp_code[i]);
+		mips_ejtag_drscan_32_out(ejtag_info, jmp_code[i]);
 
 		/* Clear the access pending bit (let the processor eat!) */
 		ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
 		mips_ejtag_set_instr(ejtag_info, EJTAG_INST_CONTROL);
-		mips_ejtag_drscan_32(ejtag_info, &ejtag_ctrl);
+		mips_ejtag_drscan_32_out(ejtag_info, ejtag_ctrl);
 	}
 
-	if ((retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl)) != ERROR_OK)
+	/* wait PrAcc pending bit for FASTDATA write */
+	retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
+	if (retval != ERROR_OK)
 		return retval;
 
 	/* next fetch to dmseg should be in FASTDATA_AREA, check */
 	address = 0;
 	mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ADDRESS);
-	mips_ejtag_drscan_32(ejtag_info, &address);
+	retval = mips_ejtag_drscan_32(ejtag_info, &address);
+	if (retval != ERROR_OK)
+		return retval;
 
 	if (address != MIPS32_PRACC_FASTDATA_AREA)
 		return ERROR_FAIL;
@@ -1025,30 +1087,36 @@ int mips32_pracc_fastdata_xfer(struct mips_ejtag *ejtag_info, struct working_are
 	mips_ejtag_set_instr(ejtag_info, EJTAG_INST_FASTDATA);
 	mips_ejtag_fastdata_scan(ejtag_info, 1, &val);
 
+	retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
+	if (retval != ERROR_OK)
+		return retval;
+
 	/* Send the load end address */
 	val = addr + (count - 1) * 4;
 	mips_ejtag_set_instr(ejtag_info, EJTAG_INST_FASTDATA);
 	mips_ejtag_fastdata_scan(ejtag_info, 1, &val);
 
-	for (i = 0; i < count; i++)
-	{
-		/* Send the data out using fastdata (clears the access pending bit) */
-		if ((retval = mips_ejtag_fastdata_scan(ejtag_info, write_t, buf++)) != ERROR_OK)
+	for (i = 0; i < count; i++) {
+		retval = mips_ejtag_fastdata_scan(ejtag_info, write_t, buf++);
+		if (retval != ERROR_OK)
 			return retval;
 	}
 
-	if ((retval = jtag_execute_queue()) != ERROR_OK)
-	{
+	retval = jtag_execute_queue();
+	if (retval != ERROR_OK) {
 		LOG_ERROR("fastdata load failed");
 		return retval;
 	}
 
-	if ((retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl)) != ERROR_OK)
+	retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
+	if (retval != ERROR_OK)
 		return retval;
 
 	address = 0;
 	mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ADDRESS);
-	mips_ejtag_drscan_32(ejtag_info, &address);
+	retval = mips_ejtag_drscan_32(ejtag_info, &address);
+	if (retval != ERROR_OK)
+		return retval;
 
 	if (address != MIPS32_PRACC_TEXT)
 		LOG_ERROR("mini program did not return to start");