interesting
/
riscv-openocd
mirror of https://github.com/riscv/riscv-openocd.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

zy1000: fix arm11 optimisation 

copy & paste error + added FIFO throttling to work around
lockup bug in FPGA.

The arm11 optimisation was introduced post v0.4.0, so this
is not a regression compared to previous release.

Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
This commit is contained in:
Øyvind Harboe 2010-06-17 09:04:46 +02:00
parent b89462d7dc
commit 80eb7e2a78
1 changed files with 81 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

163
src/jtag/zy1000/zy1000.c
View File

@ -878,101 +878,100 @@ void embeddedice_write_dcc(struct jtag_tap *tap, int reg_addr, uint8_t *buffer,
int arm11_run_instr_data_to_core_noack_inner(struct jtag_tap * tap, uint32_t opcode, uint32_t * data, size_t count)
{
#if 0
int arm11_run_instr_data_to_core_noack_inner_default(struct jtag_tap * tap, uint32_t opcode, uint32_t * data, size_t count);
return arm11_run_instr_data_to_core_noack_inner_default(tap, opcode, data, count);
#else
static const int bits[] = {32, 2};
uint32_t values[] = {0, 0};
/* FIX!!!!!! the target_write_memory() API started this nasty problem
* with unaligned uint32_t * pointers... */
const uint8_t *t = (const uint8_t *)data;
/* bypass bits before and after */
int pre_bits;
int post_bits;
jtag_pre_post_bits(tap, &pre_bits, &post_bits);
bool found = false;
struct jtag_tap *cur_tap, *nextTap;
for (cur_tap = jtag_tap_next_enabled(NULL); cur_tap!= NULL; cur_tap = nextTap)
{
nextTap = jtag_tap_next_enabled(cur_tap);
if (cur_tap == tap)
{
found = true;
} else
{
if (found)
{
post_bits++;
} else
{
pre_bits++;
}
}
}
post_bits+=2;
while (--count > 0)
if ((pre_bits > 32) || (post_bits > 32))
{
shiftValueInner(TAP_DRSHIFT, TAP_DRSHIFT, pre_bits, 0);
int arm11_run_instr_data_to_core_noack_inner_default(struct jtag_tap * tap, uint32_t opcode, uint32_t * data, size_t count);
return arm11_run_instr_data_to_core_noack_inner_default(tap, opcode, data, count);
} else
{
static const int bits[] = {32, 2};
uint32_t values[] = {0, 0};
uint32_t value;
value = *t++;
value |= (*t++<<8);
value |= (*t++<<16);
value |= (*t++<<24);
/* FIX!!!!!! the target_write_memory() API started this nasty problem
* with unaligned uint32_t * pointers... */
const uint8_t *t = (const uint8_t *)data;
shiftValueInner(TAP_DRSHIFT, TAP_DRSHIFT, 32, value);
/* minimum 2 bits */
shiftValueInner(TAP_DRSHIFT, TAP_DRPAUSE, post_bits, 0);
#if 1
/* copy & paste from arm11_dbgtap.c */
//TAP_DREXIT2, TAP_DRUPDATE, TAP_IDLE, TAP_IDLE, TAP_IDLE, TAP_DRSELECT, TAP_DRCAPTURE, TAP_DRSHIFT
waitIdle();
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 1);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 1);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 1);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
/* we don't have to wait for the queue to empty here. waitIdle(); */
ZY1000_POKE(ZY1000_JTAG_BASE + 0x20, TAP_DRSHIFT);
#else
static const tap_state_t arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay[] =
while (--count > 0)
{
TAP_DREXIT2, TAP_DRUPDATE, TAP_IDLE, TAP_IDLE, TAP_IDLE, TAP_DRSELECT, TAP_DRCAPTURE, TAP_DRSHIFT
};
#if 1
/* Danger! This code doesn't update cmd_queue_cur_state, so
* invoking jtag_add_pathmove() before jtag_add_dr_out() after
* this loop would fail!
*/
shiftValueInner(TAP_DRSHIFT, TAP_DRSHIFT, pre_bits, 0);
jtag_add_pathmove(ARRAY_SIZE(arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay),
arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay);
uint32_t value;
value = *t++;
value |= (*t++<<8);
value |= (*t++<<16);
value |= (*t++<<24);
shiftValueInner(TAP_DRSHIFT, TAP_DRSHIFT, 32, value);
/* minimum 2 bits */
shiftValueInner(TAP_DRSHIFT, TAP_DRPAUSE, post_bits, 0);
/* copy & paste from arm11_dbgtap.c */
//TAP_DREXIT2, TAP_DRUPDATE, TAP_IDLE, TAP_IDLE, TAP_IDLE, TAP_DRSELECT, TAP_DRCAPTURE, TAP_DRSHIFT
/* KLUDGE! we have to flush the fifo or the Nios CPU locks up.
* This is probably a bug in the Avalon bus(cross clocking bridge?)
* or in the jtag registers module.
*/
waitIdle();
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 1);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 1);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 1);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
ZY1000_POKE(ZY1000_JTAG_BASE + 0x28, 0);
/* we don't have to wait for the queue to empty here */
ZY1000_POKE(ZY1000_JTAG_BASE + 0x20, TAP_DRSHIFT);
waitIdle();
#else
static const tap_state_t arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay[] =
{
TAP_DREXIT2, TAP_DRUPDATE, TAP_IDLE, TAP_IDLE, TAP_IDLE, TAP_DRSELECT, TAP_DRCAPTURE, TAP_DRSHIFT
};
values[0] = *t++;
values[0] |= (*t++<<8);
values[0] |= (*t++<<16);
values[0] |= (*t++<<24);
jtag_add_dr_out(tap,
2,
bits,
values,
TAP_IDLE);
jtag_add_pathmove(ARRAY_SIZE(arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay),
arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay);
#endif
}
values[0] = *t++;
values[0] |= (*t++<<8);
values[0] |= (*t++<<16);
values[0] |= (*t++<<24);
/* This will happen on the last iteration updating cmd_queue_cur_state
* so we don't have to track it during the common code path
*/
jtag_add_dr_out(tap,
2,
bits,
values,
TAP_IDLE);
return jtag_execute_queue();
}
values[0] = *t++;
values[0] |= (*t++<<8);
values[0] |= (*t++<<16);
values[0] |= (*t++<<24);
/* This will happen on the last iteration updating the current tap state
* so we don't have to track it during the common code path */
jtag_add_dr_out(tap,
2,
bits,
values,
TAP_IDLE);
return jtag_execute_queue();
#endif
}