adapter speed: require init script setting and centralize activation from drivers...
[fw/openocd] / src / jtag / zy1000 / zy1000.c
index b50e96a8a00c2e9178be5111a77ff14f54a6baec..7a3a0f2eacc416a52ca016205b54d13afa21e658 100644 (file)
@@ -45,6 +45,8 @@
 #include "config.h"
 #endif
 
+#include <pthread.h>
+
 #include <target/embeddedice.h>
 #include <jtag/minidriver.h>
 #include <jtag/interface.h>
@@ -62,6 +64,9 @@
 #ifdef CYGPKG_HAL_NIOS2
 #include <cyg/hal/io.h>
 #include <cyg/firmwareutil/firmwareutil.h>
+#define ZYLIN_KHZ 60000
+#else
+#define ZYLIN_KHZ 64000
 #endif
 
 #define ZYLIN_VERSION GIT_ZY1000_VERSION
 #define ZYLIN_OPENOCD GIT_OPENOCD_VERSION
 #define ZYLIN_OPENOCD_VERSION "ZY1000 " ZYLIN_VERSION " " ZYLIN_DATE
 
+#else
+/* Assume we're connecting to a revc w/60MHz clock. */
+#define ZYLIN_KHZ 60000
 #endif
 
+
+/* The software needs to check if it's in RCLK mode or not */
+static bool zy1000_rclk = false;
+
 static int zy1000_khz(int khz, int *jtag_speed)
 {
        if (khz == 0)
@@ -80,7 +92,33 @@ static int zy1000_khz(int khz, int *jtag_speed)
        }
        else
        {
-               *jtag_speed = 64000/khz;
+               int speed;
+               /* Round speed up to nearest divisor.
+                *
+                * E.g. 16000kHz
+                * (64000 + 15999) / 16000 = 4
+                * (4 + 1) / 2 = 2
+                * 2 * 2 = 4
+                *
+                * 64000 / 4 = 16000
+                *
+                * E.g. 15999
+                * (64000 + 15998) / 15999 = 5
+                * (5 + 1) / 2 = 3
+                * 3 * 2 = 6
+                *
+                * 64000 / 6 = 10666
+                *
+                */
+               speed = (ZYLIN_KHZ + (khz -1)) / khz;
+               speed = (speed + 1 ) / 2;
+               speed *= 2;
+               if (speed > 8190)
+               {
+                       /* maximum dividend */
+                       speed = 8190;
+               }
+               *jtag_speed = speed;
        }
        return ERROR_OK;
 }
@@ -93,7 +131,7 @@ static int zy1000_speed_div(int speed, int *khz)
        }
        else
        {
-               *khz = 64000/speed;
+               *khz = ZYLIN_KHZ / speed;
        }
 
        return ERROR_OK;
@@ -134,6 +172,45 @@ static int zy1000_power_dropout(int *dropout)
        return ERROR_OK;
 }
 
+/* Wait for SRST to assert or deassert */
+static void waitSRST(bool asserted)
+{
+       bool first = true;
+       long long start = 0;
+       long total = 0;
+       const char *mode = asserted ? "assert" : "deassert";
+
+       for (;;)
+       {
+               bool srstAsserted = readSRST();
+               if ( (asserted && srstAsserted) || (!asserted && !srstAsserted) )
+               {
+                       if (total > 1)
+                       {
+                               LOG_USER("SRST took %dms to %s", (int)total, mode);
+                       }
+                       break;
+               }
+
+               if (first)
+               {
+                       first = false;
+                       start = timeval_ms();
+               }
+
+               total = timeval_ms() - start;
+
+               keep_alive();
+
+               if (total > 5000)
+               {
+                       LOG_ERROR("SRST took too long to %s: %dms", mode, (int)total);
+                       break;
+               }
+       }
+}
+
+
 void zy1000_reset(int trst, int srst)
 {
        LOG_DEBUG("zy1000 trst=%d, srst=%d", trst, srst);
@@ -154,6 +231,8 @@ void zy1000_reset(int trst, int srst)
                 * idle in TAP_IDLE, reset halt on str912 will fail.
                 */
                ZY1000_POKE(ZY1000_JTAG_BASE + 0x10, 0x00000001);
+
+               waitSRST(true);
        }
 
        if (!trst)
@@ -180,40 +259,7 @@ void zy1000_reset(int trst, int srst)
        if ((!srst && ((jtag_get_reset_config() & RESET_TRST_PULLS_SRST) == 0))||
                (!srst && !trst && (jtag_get_reset_config() & RESET_TRST_PULLS_SRST)))
        {
-               bool first = true;
-               long long start = 0;
-               long total = 0;
-               for (;;)
-               {       
-                       // We don't want to sense our own reset, so we clear here.
-                       // There is of course a timing hole where we could loose
-                       // a "real" reset.
-                       if (!readSRST())
-                       {
-                               if (total > 1)
-                               {
-                                 LOG_USER("SRST took %dms to deassert", (int)total);
-                               }
-                               break;
-                       }
-
-                       if (first)
-                       {
-                           first = false;
-                           start = timeval_ms();
-                       }
-
-                       total = timeval_ms() - start;
-
-                       keep_alive();
-
-                       if (total > 5000)
-                       {
-                               LOG_ERROR("SRST took too long to deassert: %dms", (int)total);
-                           break;
-                       }
-               }
-
+               waitSRST(false);
        }
 }
 
@@ -222,24 +268,30 @@ int zy1000_speed(int speed)
        /* flush JTAG master FIFO before setting speed */
        waitIdle();
 
+       zy1000_rclk = false;
+
        if (speed == 0)
        {
                /*0 means RCLK*/
-               speed = 0;
                ZY1000_POKE(ZY1000_JTAG_BASE + 0x10, 0x100);
+               zy1000_rclk = true;
                LOG_DEBUG("jtag_speed using RCLK");
        }
        else
        {
                if (speed > 8190 || speed < 2)
                {
-                       LOG_USER("valid ZY1000 jtag_speed=[8190,2]. Divisor is 64MHz / even values between 8190-2, i.e. min 7814Hz, max 32MHz");
+                       LOG_USER("valid ZY1000 jtag_speed=[8190,2]. With divisor is %dkHz / even values between 8190-2, i.e. min %dHz, max %dMHz",
+                                       ZYLIN_KHZ, (ZYLIN_KHZ * 1000) / 8190, ZYLIN_KHZ / (2 * 1000));
                        return ERROR_INVALID_ARGUMENTS;
                }
 
-               LOG_USER("jtag_speed %d => JTAG clk=%f", speed, 64.0/(float)speed);
+               int khz;
+               speed &= ~1;
+               zy1000_speed_div(speed, &khz);
+               LOG_USER("jtag_speed %d => JTAG clk=%d kHz", speed, khz);
                ZY1000_POKE(ZY1000_JTAG_BASE + 0x14, 0x100);
-               ZY1000_POKE(ZY1000_JTAG_BASE + 0x1c, speed&~1);
+               ZY1000_POKE(ZY1000_JTAG_BASE + 0x1c, speed);
        }
        return ERROR_OK;
 }
@@ -279,7 +331,7 @@ COMMAND_HANDLER(handle_power_command)
        return ERROR_OK;
 }
 
-#if !BUILD_ECOSBOARD
+#if !BUILD_ZY1000_MASTER
 static char *tcp_server = "notspecified";
 static int jim_zy1000_server(Jim_Interp *interp, int argc, Jim_Obj *const *argv)
 {
@@ -396,24 +448,10 @@ struct cyg_upgrade_info firmware_info =
                report_info,
 };
 
+// File written to /ram/firmware.phi before arriving at this fn
 static int jim_zy1000_writefirmware(Jim_Interp *interp, int argc, Jim_Obj *const *argv)
 {
-       if (argc != 2)
-               return JIM_ERR;
-
-       int length;
-       const char *str = Jim_GetString(argv[1], &length);
-
-       /* */
-       int tmpFile;
-       if ((tmpFile = open(firmware_info.file, O_RDWR | O_CREAT | O_TRUNC)) <= 0)
-       {
-               return JIM_ERR;
-       }
-       bool success;
-       success = write(tmpFile, str, length) == length;
-       close(tmpFile);
-       if (!success)
+       if (argc != 1)
                return JIM_ERR;
 
        if (!cyg_firmware_upgrade(NULL, firmware_info))
@@ -456,17 +494,32 @@ int interface_jtag_execute_queue(void)
        uint32_t empty;
 
        waitIdle();
-       ZY1000_PEEK(ZY1000_JTAG_BASE + 0x10, empty);
-       /* clear JTAG error register */
-       ZY1000_POKE(ZY1000_JTAG_BASE + 0x14, 0x400);
 
-       if ((empty&0x400) != 0)
+       /* We must make sure to write data read back to memory location before we return
+        * from this fn
+        */
+       zy1000_flush_readqueue();
+
+       /* and handle any callbacks... */
+       zy1000_flush_callbackqueue();
+
+       if (zy1000_rclk)
        {
-               LOG_WARNING("RCLK timeout");
-               /* the error is informative only as we don't want to break the firmware if there
-                * is a false positive.
+               /* Only check for errors when using RCLK to speed up
+                * jtag over TCP/IP
                 */
-//             return ERROR_FAIL;
+               ZY1000_PEEK(ZY1000_JTAG_BASE + 0x10, empty);
+               /* clear JTAG error register */
+               ZY1000_POKE(ZY1000_JTAG_BASE + 0x14, 0x400);
+
+               if ((empty&0x400) != 0)
+               {
+                       LOG_WARNING("RCLK timeout");
+                       /* the error is informative only as we don't want to break the firmware if there
+                        * is a false positive.
+                        */
+       //              return ERROR_FAIL;
+               }
        }
        return ERROR_OK;
 }
@@ -474,38 +527,7 @@ int interface_jtag_execute_queue(void)
 
 
 
-
-static uint32_t getShiftValue(void)
-{
-       uint32_t value;
-       waitIdle();
-       ZY1000_PEEK(ZY1000_JTAG_BASE + 0xc, value);
-       VERBOSE(LOG_INFO("getShiftValue %08x", value));
-       return value;
-}
-#if 0
-static uint32_t getShiftValueFlip(void)
-{
-       uint32_t value;
-       waitIdle();
-       ZY1000_PEEK(ZY1000_JTAG_BASE + 0x18, value);
-       VERBOSE(LOG_INFO("getShiftValue %08x (flipped)", value));
-       return value;
-}
-#endif
-
-#if 0
-static void shiftValueInnerFlip(const tap_state_t state, const tap_state_t endState, int repeat, uint32_t value)
-{
-       VERBOSE(LOG_INFO("shiftValueInner %s %s %d %08x (flipped)", tap_state_name(state), tap_state_name(endState), repeat, value));
-       uint32_t a,b;
-       a = state;
-       b = endState;
-       ZY1000_POKE(ZY1000_JTAG_BASE + 0xc, value);
-       ZY1000_POKE(ZY1000_JTAG_BASE + 0x8, (1 << 15) | (repeat << 8) | (a << 4) | b);
-       VERBOSE(getShiftValueFlip());
-}
-#endif
+static void writeShiftValue(uint8_t *data, int bits);
 
 // here we shuffle N bits out/in
 static __inline void scanBits(const uint8_t *out_value, uint8_t *in_value, int num_bits, bool pause_now, tap_state_t shiftState, tap_state_t end_state)
@@ -549,15 +571,7 @@ static __inline void scanBits(const uint8_t *out_value, uint8_t *in_value, int n
 
                if (in_value != NULL)
                {
-                       // data in, LSB to MSB
-                       value = getShiftValue();
-                       // we're shifting in data to MSB, shift data to be aligned for returning the value
-                       value >>= 32-k;
-
-                       for (int l = 0; l < k; l += 8)
-                       {
-                               in_value[(j + l)/8]=(value >> l)&0xff;
-                       }
+                       writeShiftValue(in_value + (j/8), k);
                }
        }
 }
@@ -604,6 +618,11 @@ int interface_jtag_add_ir_scan(struct jtag_tap *active, const struct scan_field
                        assert(scan_size <= 32);
                        shiftValueInner(TAP_IRSHIFT, pause_state, scan_size, 0xffffffff);
 
+                       /* Optimization code will check what the cur_instr is set to, so
+                        * we must set it to bypass value.
+                        */
+                       buf_set_ones(tap->cur_instr, tap->ir_length);
+
                        tap->bypass = 1;
                }
        }
@@ -834,7 +853,7 @@ static void jtag_pre_post_bits(struct jtag_tap *tap, int *pre, int *post)
                        TAP_IDLE);
 */
 
-void embeddedice_write_dcc(struct jtag_tap *tap, int reg_addr, uint8_t *buffer, int little, int count)
+void embeddedice_write_dcc(struct jtag_tap *tap, int reg_addr, const uint8_t *buffer, int little, int count)
 {
 #if 0
        int i;
@@ -876,103 +895,102 @@ 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)
+int arm11_run_instr_data_to_core_noack_inner(struct jtag_tap * tap, uint32_t opcode, const 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);
-
-               uint32_t value;
-               value = *t++;
-               value |= (*t++<<8);
-               value |= (*t++<<16);
-               value |= (*t++<<24);
+               int arm11_run_instr_data_to_core_noack_inner_default(struct jtag_tap *, uint32_t, const uint32_t *, size_t);
+               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};
 
-               shiftValueInner(TAP_DRSHIFT, TAP_DRSHIFT, 32, value);
-               /* minimum 2 bits */
-               shiftValueInner(TAP_DRSHIFT, TAP_DRPAUSE, post_bits, 0);
+               /* FIX!!!!!! the target_write_memory() API started this nasty problem
+                * with unaligned uint32_t * pointers... */
+               const uint8_t *t = (const uint8_t *)data;
 
+               while (--count > 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
+                       /* 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);
 
-               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);
+                       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
-               };
-
-               jtag_add_pathmove(ARRAY_SIZE(arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay),
-                       arm11_MOVE_DRPAUSE_IDLE_DRPAUSE_with_delay);
+                       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);
+               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);
+               /* 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();
-#endif
+               return jtag_execute_queue();
+       }
 }
 
 
@@ -985,6 +1003,7 @@ static const struct command_registration zy1000_commands[] = {
                        "With no arguments, prints status.",
                .usage = "('on'|'off)",
        },
+#if BUILD_ZY1000_MASTER
 #if BUILD_ECOSBOARD
        {
                .name = "zy1000_version",
@@ -993,6 +1012,7 @@ static const struct command_registration zy1000_commands[] = {
                .help = "Print version info for zy1000.",
                .usage = "['openocd'|'zy1000'|'date'|'time'|'pcb'|'fpga']",
        },
+#endif
 #else
        {
                .name = "zy1000_server",
@@ -1021,6 +1041,8 @@ static const struct command_registration zy1000_commands[] = {
 };
 
 
+#if !BUILD_ZY1000_MASTER
+
 static int tcp_ip = -1;
 
 /* Write large packets if we can */
@@ -1057,14 +1079,6 @@ static bool writeLong(uint32_t l)
 
 static bool readLong(uint32_t *out_data)
 {
-       if (out_pos > 0)
-       {
-               if (!flush_writes())
-               {
-                       return false;
-               }
-       }
-
        uint32_t data = 0;
        int i;
        for (i = 0; i < 4; i++)
@@ -1072,6 +1086,17 @@ static bool readLong(uint32_t *out_data)
                uint8_t c;
                if (in_pos == in_write)
                {
+                       /* If we have some data that we can send, send them before
+                        * we wait for more data
+                        */
+                       if (out_pos > 0)
+                       {
+                               if (!flush_writes())
+                               {
+                                       return false;
+                               }
+                       }
+
                        /* read more */
                        int t;
                        t = read(tcp_ip, in_buffer, sizeof(in_buffer));
@@ -1095,11 +1120,9 @@ enum ZY1000_CMD
        ZY1000_CMD_POKE = 0x0,
        ZY1000_CMD_PEEK = 0x8,
        ZY1000_CMD_SLEEP = 0x1,
+       ZY1000_CMD_WAITIDLE = 2
 };
 
-
-#if !BUILD_ECOSBOARD
-
 #include <sys/socket.h> /* for socket(), connect(), send(), and recv() */
 #include <arpa/inet.h>  /* for sockaddr_in and inet_addr() */
 
@@ -1155,9 +1178,26 @@ void zy1000_tcpout(uint32_t address, uint32_t data)
        }
 }
 
+/* By sending the wait to the server, we avoid a readback
+ * of status. Radically improves performance for this operation
+ * with long ping times.
+ */
+void waitIdle(void)
+{
+       tcpip_open();
+       if (!writeLong((ZY1000_CMD_WAITIDLE << 24)))
+       {
+               fprintf(stderr, "Could not write to zy1000 server\n");
+               exit(-1);
+       }
+}
+
 uint32_t zy1000_tcpin(uint32_t address)
 {
        tcpip_open();
+
+       zy1000_flush_readqueue();
+
        uint32_t data;
        if (!writeLong((ZY1000_CMD_PEEK << 24) | address)||
                        !readLong(&data))
@@ -1180,127 +1220,158 @@ int interface_jtag_add_sleep(uint32_t us)
        return ERROR_OK;
 }
 
+/* queue a readback */
+#define readqueue_size 16384
+static struct
+{
+       uint8_t *dest;
+       int bits;
+} readqueue[readqueue_size];
 
-#endif
-
-#if BUILD_ECOSBOARD
-static char tcpip_stack[2048];
-static cyg_thread tcpip_thread_object;
-static cyg_handle_t tcpip_thread_handle;
-
-static char watchdog_stack[2048];
-static cyg_thread watchdog_thread_object;
-static cyg_handle_t watchdog_thread_handle;
+static int readqueue_pos = 0;
 
-/* Infinite loop peeking & poking */
-static void tcpipserver(void)
+/* flush the readqueue, this means reading any data that
+ * we're expecting and store them into the final position
+ */
+void zy1000_flush_readqueue(void)
 {
-       for (;;)
+       if (readqueue_pos == 0)
+       {
+               /* simply debugging by allowing easy breakpoints when there
+                * is something to do. */
+               return;
+       }
+       int i;
+       tcpip_open();
+       for (i = 0; i < readqueue_pos; i++)
        {
-               uint32_t address;
-               if (!readLong(&address))
-                       return;
-               enum ZY1000_CMD c = (address >> 24) & 0xff;
-               address &= 0xffffff;
-               switch (c)
+               uint32_t value;
+               if (!readLong(&value))
                {
-                       case ZY1000_CMD_POKE:
-                       {
-                               uint32_t data;
-                               if (!readLong(&data))
-                                       return;
-                               address &= ~0x80000000;
-                               ZY1000_POKE(address + ZY1000_JTAG_BASE, data);
-                               break;
-                       }
-                       case ZY1000_CMD_PEEK:
-                       {
-                               uint32_t data;
-                               ZY1000_PEEK(address + ZY1000_JTAG_BASE, data);
-                               if (!writeLong(data))
-                                       return;
-                               break;
-                       }
-                       case ZY1000_CMD_SLEEP:
-                       {
-                               uint32_t data;
-                               if (!readLong(&data))
-                                       return;
-                               jtag_sleep(data);
-                               break;
-                       }
-                       default:
-                               return;
+                       fprintf(stderr, "Could not read from zy1000 server\n");
+                       exit(-1);
+               }
+
+               uint8_t *in_value = readqueue[i].dest;
+               int k = readqueue[i].bits;
+
+               // we're shifting in data to MSB, shift data to be aligned for returning the value
+               value >>= 32-k;
+
+               for (int l = 0; l < k; l += 8)
+               {
+                       in_value[l/8]=(value >> l)&0xff;
                }
        }
+       readqueue_pos = 0;
 }
 
-
-static void tcpip_server(cyg_addrword_t data)
+/* By queuing the callback's we avoid flushing the
+read queue until jtag_execute_queue(). This can
+reduce latency dramatically for cases where
+callbacks are used extensively.
+*/
+#define callbackqueue_size 128
+static struct callbackentry
 {
-       int so_reuseaddr_option = 1;
+       jtag_callback_t callback;
+       jtag_callback_data_t data0;
+       jtag_callback_data_t data1;
+       jtag_callback_data_t data2;
+       jtag_callback_data_t data3;
+} callbackqueue[callbackqueue_size];
 
-       int fd;
-       if ((fd = socket(AF_INET, SOCK_STREAM, 0)) == -1)
+static int callbackqueue_pos = 0;
+
+void zy1000_jtag_add_callback4(jtag_callback_t callback, jtag_callback_data_t data0, jtag_callback_data_t data1, jtag_callback_data_t data2, jtag_callback_data_t data3)
+{
+       if (callbackqueue_pos >= callbackqueue_size)
        {
-               LOG_ERROR("error creating socket: %s", strerror(errno));
-               exit(-1);
+               zy1000_flush_callbackqueue();
        }
 
-       setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (void*) &so_reuseaddr_option,
-                       sizeof(int));
+       callbackqueue[callbackqueue_pos].callback = callback;
+       callbackqueue[callbackqueue_pos].data0 = data0;
+       callbackqueue[callbackqueue_pos].data1 = data1;
+       callbackqueue[callbackqueue_pos].data2 = data2;
+       callbackqueue[callbackqueue_pos].data3 = data3;
+       callbackqueue_pos++;
+}
 
-       struct sockaddr_in sin;
-       unsigned int address_size;
-       address_size = sizeof(sin);
-       memset(&sin, 0, sizeof(sin));
-       sin.sin_family = AF_INET;
-       sin.sin_addr.s_addr = INADDR_ANY;
-       sin.sin_port = htons(7777);
+static int zy1000_jtag_convert_to_callback4(jtag_callback_data_t data0, jtag_callback_data_t data1, jtag_callback_data_t data2, jtag_callback_data_t data3)
+{
+       ((jtag_callback1_t)data1)(data0);
+       return ERROR_OK;
+}
 
-       if (bind(fd, (struct sockaddr *) &sin, sizeof(sin)) == -1)
+void zy1000_jtag_add_callback(jtag_callback1_t callback, jtag_callback_data_t data0)
+{
+       zy1000_jtag_add_callback4(zy1000_jtag_convert_to_callback4, data0, (jtag_callback_data_t)callback, 0, 0);
+}
+
+void zy1000_flush_callbackqueue(void)
+{
+       /* we have to flush the read queue so we have access to
+        the data the callbacks will use 
+       */
+       zy1000_flush_readqueue();
+       int i;
+       for (i = 0; i < callbackqueue_pos; i++)
        {
-               LOG_ERROR("couldn't bind to socket: %s", strerror(errno));
-               exit(-1);
+               struct callbackentry *entry = &callbackqueue[i];
+               jtag_set_error(entry->callback(entry->data0, entry->data1, entry->data2, entry->data3));
        }
+       callbackqueue_pos = 0;
+}
 
-       if (listen(fd, 1) == -1)
+static void writeShiftValue(uint8_t *data, int bits)
+{
+       waitIdle();
+
+       if (!writeLong((ZY1000_CMD_PEEK << 24) | (ZY1000_JTAG_BASE + 0xc)))
        {
-               LOG_ERROR("couldn't listen on socket: %s", strerror(errno));
+               fprintf(stderr, "Could not read from zy1000 server\n");
                exit(-1);
        }
 
-
-       for (;;)
+       if (readqueue_pos >= readqueue_size)
        {
-               tcp_ip = accept(fd, (struct sockaddr *) &sin, &address_size);
-               if (tcp_ip < 0)
-               {
-                       continue;
-               }
+               zy1000_flush_readqueue();
+       }
 
-               int flag = 1;
-               setsockopt(tcp_ip,      /* socket affected */
-                               IPPROTO_TCP,            /* set option at TCP level */
-                               TCP_NODELAY,            /* name of option */
-                               (char *)&flag,          /* the cast is historical cruft */
-                               sizeof(int));           /* length of option value */
+       readqueue[readqueue_pos].dest = data;
+       readqueue[readqueue_pos].bits = bits;
+       readqueue_pos++;
+}
 
-               bool save_poll = jtag_poll_get_enabled();
+#else
 
-               /* polling will screw up the "connection" */
-               jtag_poll_set_enabled(false);
+static void writeShiftValue(uint8_t *data, int bits)
+{
+       uint32_t value;
+       waitIdle();
+       ZY1000_PEEK(ZY1000_JTAG_BASE + 0xc, value);
+       VERBOSE(LOG_INFO("getShiftValue %08x", value));
 
-               tcpipserver();
+       // data in, LSB to MSB
+       // we're shifting in data to MSB, shift data to be aligned for returning the value
+       value >>= 32 - bits;
 
-               jtag_poll_set_enabled(save_poll);
+       for (int l = 0; l < bits; l += 8)
+       {
+               data[l/8]=(value >> l)&0xff;
+       }
+}
 
-               close(tcp_ip);
+#endif
 
-       }
-       close(fd);
+#if BUILD_ZY1000_MASTER
 
-}
+#if BUILD_ECOSBOARD
+static char watchdog_stack[2048];
+static cyg_thread watchdog_thread_object;
+static cyg_handle_t watchdog_thread_handle;
+#endif
 
 #ifdef WATCHDOG_BASE
 /* If we connect to port 8888 we must send a char every 10s or the board resets itself */
@@ -1384,20 +1455,48 @@ static void watchdog_server(cyg_addrword_t data)
 }
 #endif
 
+#endif
+
+#if BUILD_ZY1000_MASTER
 int interface_jtag_add_sleep(uint32_t us)
 {
        jtag_sleep(us);
        return ERROR_OK;
 }
-
 #endif
 
+#if BUILD_ZY1000_MASTER && !BUILD_ECOSBOARD
+volatile void *zy1000_jtag_master;
+#include <sys/mman.h>
+#endif
 
 int zy1000_init(void)
 {
 #if BUILD_ECOSBOARD
        LOG_USER("%s", ZYLIN_OPENOCD_VERSION);
+#elif BUILD_ZY1000_MASTER
+       int fd;
+       if((fd = open("/dev/mem", O_RDWR | O_SYNC)) == -1)
+       {
+               LOG_ERROR("No access to /dev/mem");
+               return ERROR_FAIL;
+       }
+#ifndef REGISTERS_BASE
+#define REGISTERS_BASE 0x9002000
+#define REGISTERS_SPAN 128
 #endif
+    
+    zy1000_jtag_master = mmap(0, REGISTERS_SPAN, PROT_READ | PROT_WRITE, MAP_SHARED, fd, REGISTERS_BASE);
+    
+    if(zy1000_jtag_master == (void *) -1) 
+    {
+           close(fd);
+               LOG_ERROR("No access to /dev/mem");
+               return ERROR_FAIL;
+    } 
+#endif
+
+
 
        ZY1000_POKE(ZY1000_JTAG_BASE + 0x10, 0x30); // Turn on LED1 & LED2
 
@@ -1406,20 +1505,16 @@ int zy1000_init(void)
 
         /* deassert resets. Important to avoid infinite loop waiting for SRST to deassert */
        zy1000_reset(0, 0);
-       zy1000_speed(jtag_get_speed());
-
 
+#if BUILD_ZY1000_MASTER
 #if BUILD_ECOSBOARD
-       cyg_thread_create(1, tcpip_server, (cyg_addrword_t) 0, "tcip/ip server",
-                       (void *) tcpip_stack, sizeof(tcpip_stack),
-                       &tcpip_thread_handle, &tcpip_thread_object);
-       cyg_thread_resume(tcpip_thread_handle);
 #ifdef WATCHDOG_BASE
        cyg_thread_create(1, watchdog_server, (cyg_addrword_t) 0, "watchdog tcip/ip server",
                        (void *) watchdog_stack, sizeof(watchdog_stack),
                        &watchdog_thread_handle, &watchdog_thread_object);
        cyg_thread_resume(watchdog_thread_handle);
 #endif
+#endif
 #endif
 
        return ERROR_OK;
@@ -1441,4 +1536,3 @@ struct jtag_interface zy1000_interface =
        .power_dropout = zy1000_power_dropout,
        .srst_asserted = zy1000_srst_asserted,
 };
-