mips: optimize mips32_pracc_read_regs() code
[fw/openocd] / src / target / mips32_pracc.c
1 /***************************************************************************
2  *   Copyright (C) 2008 by Spencer Oliver                                  *
3  *   spen@spen-soft.co.uk                                                  *
4  *                                                                         *
5  *   Copyright (C) 2008 by David T.L. Wong                                 *
6  *                                                                         *
7  *   Copyright (C) 2009 by David N. Claffey <dnclaffey@gmail.com>          *
8  *                                                                         *
9  *   Copyright (C) 2011 by Drasko DRASKOVIC                                *
10  *   drasko.draskovic@gmail.com                                            *
11  *                                                                         *
12  *   This program is free software; you can redistribute it and/or modify  *
13  *   it under the terms of the GNU General Public License as published by  *
14  *   the Free Software Foundation; either version 2 of the License, or     *
15  *   (at your option) any later version.                                   *
16  *                                                                         *
17  *   This program is distributed in the hope that it will be useful,       *
18  *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
19  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
20  *   GNU General Public License for more details.                          *
21  *                                                                         *
22  *   You should have received a copy of the GNU General Public License     *
23  *   along with this program; if not, write to the                         *
24  *   Free Software Foundation, Inc.,                                       *
25  *   59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.             *
26  ***************************************************************************/
27
28 /*
29  * This version has optimized assembly routines for 32 bit operations:
30  * - read word
31  * - write word
32  * - write array of words
33  *
34  * One thing to be aware of is that the MIPS32 cpu will execute the
35  * instruction after a branch instruction (one delay slot).
36  *
37  * For example:
38  *  LW $2, ($5 +10)
39  *  B foo
40  *  LW $1, ($2 +100)
41  *
42  * The LW $1, ($2 +100) instruction is also executed. If this is
43  * not wanted a NOP can be inserted:
44  *
45  *  LW $2, ($5 +10)
46  *  B foo
47  *  NOP
48  *  LW $1, ($2 +100)
49  *
50  * or the code can be changed to:
51  *
52  *  B foo
53  *  LW $2, ($5 +10)
54  *  LW $1, ($2 +100)
55  *
56  * The original code contained NOPs. I have removed these and moved
57  * the branches.
58  *
59  * I also moved the PRACC_STACK to 0xFF204000. This allows
60  * the use of 16 bits offsets to get pointers to the input
61  * and output area relative to the stack. Note that the stack
62  * isn't really a stack (the stack pointer is not 'moving')
63  * but a FIFO simulated in software.
64  *
65  * These changes result in a 35% speed increase when programming an
66  * external flash.
67  *
68  * More improvement could be gained if the registers do no need
69  * to be preserved but in that case the routines should be aware
70  * OpenOCD is used as a flash programmer or as a debug tool.
71  *
72  * Nico Coesel
73  */
74
75 #ifdef HAVE_CONFIG_H
76 #include "config.h"
77 #endif
78
79 #include <helper/time_support.h>
80
81 #include "mips32.h"
82 #include "mips32_pracc.h"
83
84 struct mips32_pracc_context {
85         uint32_t *local_iparam;
86         int num_iparam;
87         uint32_t *local_oparam;
88         int num_oparam;
89         const uint32_t *code;
90         int code_len;
91         uint32_t stack[32];
92         int stack_offset;
93         struct mips_ejtag *ejtag_info;
94 };
95
96 static int mips32_pracc_sync_cache(struct mips_ejtag *ejtag_info,
97                 uint32_t start_addr, uint32_t end_addr);
98 static int mips32_pracc_clean_invalidate_cache(struct mips_ejtag *ejtag_info,
99                 uint32_t start_addr, uint32_t end_addr);
100
101 static int wait_for_pracc_rw(struct mips_ejtag *ejtag_info, uint32_t *ctrl)
102 {
103         uint32_t ejtag_ctrl;
104         long long then = timeval_ms();
105         int timeout;
106         int retval;
107
108         /* wait for the PrAcc to become "1" */
109         mips_ejtag_set_instr(ejtag_info, EJTAG_INST_CONTROL);
110
111         while (1) {
112                 ejtag_ctrl = ejtag_info->ejtag_ctrl;
113                 retval = mips_ejtag_drscan_32(ejtag_info, &ejtag_ctrl);
114                 if (retval != ERROR_OK)
115                         return retval;
116
117                 if (ejtag_ctrl & EJTAG_CTRL_PRACC)
118                         break;
119
120                 timeout = timeval_ms() - then;
121                 if (timeout > 1000) {
122                         LOG_DEBUG("DEBUGMODULE: No memory access in progress!");
123                         return ERROR_JTAG_DEVICE_ERROR;
124                 }
125         }
126
127         *ctrl = ejtag_ctrl;
128         return ERROR_OK;
129 }
130
131 static int mips32_pracc_exec_read(struct mips32_pracc_context *ctx, uint32_t address)
132 {
133         struct mips_ejtag *ejtag_info = ctx->ejtag_info;
134         int offset;
135         uint32_t ejtag_ctrl, data;
136
137         if ((address >= MIPS32_PRACC_PARAM_IN)
138                 && (address < MIPS32_PRACC_PARAM_IN + ctx->num_iparam * 4)) {
139                 offset = (address - MIPS32_PRACC_PARAM_IN) / 4;
140                 data = ctx->local_iparam[offset];
141         } else if ((address >= MIPS32_PRACC_PARAM_OUT)
142                 && (address < MIPS32_PRACC_PARAM_OUT + ctx->num_oparam * 4)) {
143                 offset = (address - MIPS32_PRACC_PARAM_OUT) / 4;
144                 data = ctx->local_oparam[offset];
145         } else if ((address >= MIPS32_PRACC_TEXT)
146                 && (address < MIPS32_PRACC_TEXT + ctx->code_len * 4)) {
147                 offset = (address - MIPS32_PRACC_TEXT) / 4;
148                 data = ctx->code[offset];
149         } else if (address == MIPS32_PRACC_STACK) {
150                 if (ctx->stack_offset <= 0) {
151                         LOG_ERROR("Error: Pracc stack out of bounds");
152                         return ERROR_JTAG_DEVICE_ERROR;
153                 }
154                 /* save to our debug stack */
155                 data = ctx->stack[--ctx->stack_offset];
156         } else {
157                 /* TODO: send JMP 0xFF200000 instruction. Hopefully processor jump back
158                  * to start of debug vector */
159
160                 LOG_ERROR("Error reading unexpected address 0x%8.8" PRIx32 "", address);
161                 return ERROR_JTAG_DEVICE_ERROR;
162         }
163
164         /* Send the data out */
165         mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_DATA);
166         mips_ejtag_drscan_32_out(ctx->ejtag_info, data);
167
168         /* Clear the access pending bit (let the processor eat!) */
169         ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
170         mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_CONTROL);
171         mips_ejtag_drscan_32_out(ctx->ejtag_info, ejtag_ctrl);
172
173         return jtag_execute_queue();
174 }
175
176 static int mips32_pracc_exec_write(struct mips32_pracc_context *ctx, uint32_t address)
177 {
178         uint32_t ejtag_ctrl, data;
179         int offset;
180         struct mips_ejtag *ejtag_info = ctx->ejtag_info;
181         int retval;
182
183         mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_DATA);
184         retval = mips_ejtag_drscan_32(ctx->ejtag_info, &data);
185         if (retval != ERROR_OK)
186                 return retval;
187
188         /* Clear access pending bit */
189         ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
190         mips_ejtag_set_instr(ctx->ejtag_info, EJTAG_INST_CONTROL);
191         mips_ejtag_drscan_32_out(ctx->ejtag_info, ejtag_ctrl);
192
193         retval = jtag_execute_queue();
194         if (retval != ERROR_OK)
195                 return retval;
196
197         if ((address >= MIPS32_PRACC_PARAM_IN)
198                 && (address < MIPS32_PRACC_PARAM_IN + ctx->num_iparam * 4)) {
199                 offset = (address - MIPS32_PRACC_PARAM_IN) / 4;
200                 ctx->local_iparam[offset] = data;
201         } else if ((address >= MIPS32_PRACC_PARAM_OUT)
202                 && (address < MIPS32_PRACC_PARAM_OUT + ctx->num_oparam * 4)) {
203                 offset = (address - MIPS32_PRACC_PARAM_OUT) / 4;
204                 ctx->local_oparam[offset] = data;
205         } else if (address == MIPS32_PRACC_STACK) {
206                 if (ctx->stack_offset >= 32) {
207                         LOG_ERROR("Error: Pracc stack out of bounds");
208                         return ERROR_JTAG_DEVICE_ERROR;
209                 }
210                 /* save data onto our stack */
211                 ctx->stack[ctx->stack_offset++] = data;
212         } else {
213                 LOG_ERROR("Error writing unexpected address 0x%8.8" PRIx32 "", address);
214                 return ERROR_JTAG_DEVICE_ERROR;
215         }
216
217         return ERROR_OK;
218 }
219
220 int mips32_pracc_exec(struct mips_ejtag *ejtag_info, int code_len, const uint32_t *code,
221                 int num_param_in, uint32_t *param_in, int num_param_out, uint32_t *param_out, int cycle)
222 {
223         uint32_t ejtag_ctrl;
224         uint32_t address;
225         struct mips32_pracc_context ctx;
226         int retval;
227         int pass = 0;
228
229         ctx.local_iparam = param_in;
230         ctx.local_oparam = param_out;
231         ctx.num_iparam = num_param_in;
232         ctx.num_oparam = num_param_out;
233         ctx.code = code;
234         ctx.code_len = code_len;
235         ctx.ejtag_info = ejtag_info;
236         ctx.stack_offset = 0;
237
238         while (1) {
239                 retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
240                 if (retval != ERROR_OK)
241                         return retval;
242
243                 address = 0;
244                 mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ADDRESS);
245                 retval = mips_ejtag_drscan_32(ejtag_info, &address);
246                 if (retval != ERROR_OK)
247                         return retval;
248
249                 /* Check for read or write */
250                 if (ejtag_ctrl & EJTAG_CTRL_PRNW) {
251                         retval = mips32_pracc_exec_write(&ctx, address);
252                         if (retval != ERROR_OK)
253                                 return retval;
254                 } else {
255                         /* Check to see if its reading at the debug vector. The first pass through
256                          * the module is always read at the vector, so the first one we allow.  When
257                          * the second read from the vector occurs we are done and just exit. */
258                         if ((address == MIPS32_PRACC_TEXT) && (pass++))
259                                 break;
260
261                         retval = mips32_pracc_exec_read(&ctx, address);
262                         if (retval != ERROR_OK)
263                                 return retval;
264                 }
265
266                 if (cycle == 0)
267                         break;
268         }
269
270         /* stack sanity check */
271         if (ctx.stack_offset != 0)
272                 LOG_DEBUG("Pracc Stack not zero");
273
274         return ERROR_OK;
275 }
276
277 static int mips32_pracc_read_u32(struct mips_ejtag *ejtag_info, uint32_t addr, uint32_t *buf)
278 {
279         uint32_t code[] = {
280                                                                                                                 /* start: */
281                 MIPS32_MTC0(15, 31, 0),                                         /* move $15 to COP0 DeSave */
282                 MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR),                          /* $15 = MIPS32_PRACC_BASE_ADDR */
283                 MIPS32_SW(8, PRACC_STACK_OFFSET, 15),                           /* sw $8,PRACC_STACK_OFFSET($15) */
284
285                 MIPS32_LUI(8, UPPER16((addr + 0x8000))),                        /* load  $8 with modified upper address */
286                 MIPS32_LW(8, LOWER16(addr), 8),                                 /* lw $8, LOWER16(addr)($8) */
287                 MIPS32_SW(8, PRACC_OUT_OFFSET, 15),                             /* sw $8,PRACC_OUT_OFFSET($15) */
288
289                 MIPS32_LW(8, PRACC_STACK_OFFSET, 15),                           /* lw $8,PRACC_STACK_OFFSET($15) */
290                 MIPS32_B(NEG16(8)),                                                     /* b start */
291                 MIPS32_MFC0(15, 31, 0),                                         /* move COP0 DeSave to $15 */
292         };
293
294         return mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 0, NULL, 1, buf, 1);
295 }
296
297 int mips32_pracc_read_mem(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
298 {
299         if (count == 1 && size == 4)
300                 return mips32_pracc_read_u32(ejtag_info, addr, (uint32_t *)buf);
301
302         int retval = ERROR_FAIL;
303
304         uint32_t *code = NULL;
305         uint32_t *data = NULL;
306
307         code = malloc((256 * 2 + 10) * sizeof(uint32_t));
308         if (code == NULL) {
309                 LOG_ERROR("Out of memory");
310                 goto exit;
311         }
312
313         if (size != 4) {
314                 data = malloc(256 * sizeof(uint32_t));
315                 if (data == NULL) {
316                         LOG_ERROR("Out of memory");
317                         goto exit;
318                 }
319         }
320
321         uint32_t *buf32 = buf;
322         uint16_t *buf16 = buf;
323         uint8_t *buf8 = buf;
324
325         int i;
326         uint32_t upper_base_addr, last_upper_base_addr;
327         int this_round_count;
328         int code_len;
329
330         while (count) {
331                 this_round_count = (count > 256) ? 256 : count;
332                 last_upper_base_addr = UPPER16((addr + 0x8000));
333                 uint32_t *code_p = code;
334
335                 *code_p++ = MIPS32_MTC0(15, 31, 0);                                     /* save $15 in DeSave */
336                 *code_p++ = MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR);                      /* $15 = MIPS32_PRACC_BASE_ADDR */
337                 *code_p++ = MIPS32_SW(8, PRACC_STACK_OFFSET, 15);                       /* save $8 and $9 to pracc stack */
338                 *code_p++ = MIPS32_SW(9, PRACC_STACK_OFFSET, 15);
339                 *code_p++ = MIPS32_LUI(9, last_upper_base_addr);                        /* load the upper memory address in $9*/
340                 code_len = 5;
341
342                 for (i = 0; i != this_round_count; i++) {               /* Main code loop */
343                         upper_base_addr = UPPER16((addr + 0x8000));
344                         if (last_upper_base_addr != upper_base_addr) {
345                                 *code_p++ = MIPS32_LUI(9, upper_base_addr);             /* if needed, change upper address in $9*/
346                                 code_len++;
347                                 last_upper_base_addr = upper_base_addr;
348                         }
349
350                         if (size == 4)
351                                 *code_p++ = MIPS32_LW(8, LOWER16(addr), 9);             /* load from memory to $8 */
352                         else if (size == 2)
353                                 *code_p++ = MIPS32_LHU(8, LOWER16(addr), 9);
354                         else
355                                 *code_p++ = MIPS32_LBU(8, LOWER16(addr), 9);
356
357                         *code_p++ = MIPS32_SW(8, PRACC_OUT_OFFSET + i * 4, 15);         /* store $8 at param out */
358
359                         code_len += 2;
360                         addr += size;
361                 }
362
363                 *code_p++ = MIPS32_LW(9, PRACC_STACK_OFFSET, 15);                       /* restore $8 and $9 from pracc stack */
364                 *code_p++ = MIPS32_LW(8, PRACC_STACK_OFFSET, 15);
365
366                 code_len += 4;
367                 *code_p++ = MIPS32_B(NEG16(code_len - 1));                                      /* jump to start */
368                 *code_p = MIPS32_MFC0(15, 31, 0);                                       /* restore $15 from DeSave */
369
370                 if (size == 4) {
371                         retval = mips32_pracc_exec(ejtag_info, code_len, code, 0, NULL, this_round_count, buf32, 1);
372                         if (retval != ERROR_OK)
373                                 goto exit;
374                         buf32 += this_round_count;
375                 } else {
376                         retval = mips32_pracc_exec(ejtag_info, code_len, code, 0, NULL, this_round_count, data, 1);
377                         if (retval != ERROR_OK)
378                                 goto exit;
379                         uint32_t *data_p = data;
380                         for (i = 0; i != this_round_count; i++) {
381                                 if (size == 2)
382                                         *buf16++ = *data_p++;
383                                 else
384                                         *buf8++ = *data_p++;
385                         }
386                 }
387                 count -= this_round_count;
388         }
389
390 exit:
391         if (code)
392                 free(code);
393         if (data)
394                 free(data);
395         return retval;
396 }
397
398 int mips32_cp0_read(struct mips_ejtag *ejtag_info, uint32_t *val, uint32_t cp0_reg, uint32_t cp0_sel)
399 {
400         /**
401          * Do not make this code static, but regenerate it every time,
402          * as 3th element has to be changed to add parameters
403          */
404         uint32_t code[] = {
405                                                                                                                 /* start: */
406                 MIPS32_MTC0(15, 31, 0),                                                 /* move $15 to COP0 DeSave */
407                 MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR),                                  /* $15 = MIPS32_PRACC_BASE_ADDR */
408                 MIPS32_SW(8, PRACC_STACK_OFFSET, 15),                                   /* sw $8,PRACC_STACK_OFFSET($15) */
409
410                 /* 3 */ MIPS32_MFC0(8, 0, 0),                                           /* move COP0 [cp0_reg select] to $8 */
411                 MIPS32_SW(8, PRACC_OUT_OFFSET, 15),                                     /* sw $8,PRACC_OUT_OFFSET($15) */
412
413                 MIPS32_LW(8, PRACC_STACK_OFFSET, 15),                                   /* lw $8,PRACC_STACK_OFFSET($15) */
414                 MIPS32_B(NEG16(7)),                                                     /* b start */
415                 MIPS32_MFC0(15, 31, 0),                                                 /* move COP0 DeSave to $15 */
416         };
417
418         /**
419          * Note that our input parametes cp0_reg and cp0_sel
420          * are numbers (not gprs) which make part of mfc0 instruction opcode.
421          *
422          * These are not fix, but can be different for each mips32_cp0_read() function call,
423          * and that is why we must insert them directly into opcode,
424          * i.e. we can not pass it on EJTAG microprogram stack (via param_in),
425          * and put them into the gprs later from MIPS32_PRACC_STACK
426          * because mfc0 do not use gpr as a parameter for the cp0_reg and select part,
427          * but plain (immediate) number.
428          *
429          * MIPS32_MTC0 is implemented via MIPS32_R_INST macro.
430          * In order to insert our parameters, we must change rd and funct fields.
431          */
432         code[3] |= (cp0_reg << 11) | cp0_sel;  /* change rd and funct of MIPS32_R_INST macro */
433
434         return mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 0, NULL, 1, val, 1);
435 }
436
437 int mips32_cp0_write(struct mips_ejtag *ejtag_info, uint32_t val, uint32_t cp0_reg, uint32_t cp0_sel)
438 {
439         uint32_t code[] = {
440                                                                                                                         /* start: */
441                 MIPS32_MTC0(15, 31, 0),                                                 /* move $15 to COP0 DeSave */
442                 MIPS32_LUI(15, UPPER16(val)),                                           /* Load val to $15 */
443                 MIPS32_ORI(15, 15, LOWER16(val)),
444
445                 /* 3 */ MIPS32_MTC0(15, 0, 0),                                          /* move $15 to COP0 [cp0_reg select] */
446
447                 MIPS32_B(NEG16(5)),                                                     /* b start */
448                 MIPS32_MFC0(15, 31, 0),                                                 /* move COP0 DeSave to $15 */
449         };
450
451         /**
452          * Note that MIPS32_MTC0 macro is implemented via MIPS32_R_INST macro.
453          * In order to insert our parameters, we must change rd and funct fields.
454          */
455         code[3] |= (cp0_reg << 11) | cp0_sel;  /* change rd and funct fields of MIPS32_R_INST macro */
456
457         return mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 0, NULL, 0, NULL, 1);
458 }
459
460 /**
461  * \b mips32_pracc_sync_cache
462  *
463  * Synchronize Caches to Make Instruction Writes Effective
464  * (ref. doc. MIPS32 Architecture For Programmers Volume II: The MIPS32 Instruction Set,
465  *  Document Number: MD00086, Revision 2.00, June 9, 2003)
466  *
467  * When the instruction stream is written, the SYNCI instruction should be used
468  * in conjunction with other instructions to make the newly-written instructions effective.
469  *
470  * Explanation :
471  * A program that loads another program into memory is actually writing the D- side cache.
472  * The instructions it has loaded can't be executed until they reach the I-cache.
473  *
474  * After the instructions have been written, the loader should arrange
475  * to write back any containing D-cache line and invalidate any locations
476  * already in the I-cache.
477  *
478  * You can do that with cache instructions, but those instructions are only available in kernel mode,
479  * and a loader writing instructions for the use of its own process need not be privileged software.
480  *
481  * In the latest MIPS32/64 CPUs, MIPS provides the synci instruction,
482  * which does the whole job for a cache-line-sized chunk of the memory you just loaded:
483  * That is, it arranges a D-cache write-back and an I-cache invalidate.
484  *
485  * To employ synci at user level, you need to know the size of a cache line,
486  * and that can be obtained with a rdhwr SYNCI_Step
487  * from one of the standard “hardware registers”.
488  */
489 static int mips32_pracc_sync_cache(struct mips_ejtag *ejtag_info,
490                 uint32_t start_addr, uint32_t end_addr)
491 {
492         static const uint32_t code[] = {
493                                                                                                                         /* start: */
494                 MIPS32_MTC0(15, 31, 0),                                                         /* move $15 to COP0 DeSave */
495                 MIPS32_LUI(15, UPPER16(MIPS32_PRACC_STACK)),            /* $15 = MIPS32_PRACC_STACK */
496                 MIPS32_ORI(15, 15, LOWER16(MIPS32_PRACC_STACK)),
497                 MIPS32_SW(8, 0, 15),                                                            /* sw $8,($15) */
498                 MIPS32_SW(9, 0, 15),                                                            /* sw $9,($15) */
499                 MIPS32_SW(10, 0, 15),                                                           /* sw $10,($15) */
500                 MIPS32_SW(11, 0, 15),                                                           /* sw $11,($15) */
501
502                 MIPS32_LUI(8, UPPER16(MIPS32_PRACC_PARAM_IN)),          /* $8 = MIPS32_PRACC_PARAM_IN */
503                 MIPS32_ORI(8, 8, LOWER16(MIPS32_PRACC_PARAM_IN)),
504                 MIPS32_LW(9, 0, 8),                                                                     /* Load write start_addr to $9 */
505                 MIPS32_LW(10, 4, 8),                                                            /* Load write end_addr to $10 */
506
507                 MIPS32_RDHWR(11, MIPS32_SYNCI_STEP),                            /* $11 = MIPS32_SYNCI_STEP */
508                 MIPS32_BEQ(11, 0, 6),                                                           /* beq $11, $0, end */
509                 MIPS32_NOP,
510                                                                                                                         /* synci_loop : */
511                 MIPS32_SYNCI(0, 9),                                                                     /* synci 0($9) */
512                 MIPS32_SLTU(8, 10, 9),                                                          /* sltu $8, $10, $9  # $8 = $10 < $9 ? 1 : 0 */
513                 MIPS32_BNE(8, 0, NEG16(3)),                                                     /* bne $8, $0, synci_loop */
514                 MIPS32_ADDU(9, 9, 11),                                                          /* $9 += MIPS32_SYNCI_STEP */
515                 MIPS32_SYNC,
516                                                                                                                         /* end: */
517                 MIPS32_LW(11, 0, 15),                                                           /* lw $11,($15) */
518                 MIPS32_LW(10, 0, 15),                                                           /* lw $10,($15) */
519                 MIPS32_LW(9, 0, 15),                                                            /* lw $9,($15) */
520                 MIPS32_LW(8, 0, 15),                                                            /* lw $8,($15) */
521                 MIPS32_B(NEG16(24)),                                                            /* b start */
522                 MIPS32_MFC0(15, 31, 0),                                                         /* move COP0 DeSave to $15 */
523         };
524
525         /* TODO remove array */
526         uint32_t *param_in = malloc(2 * sizeof(uint32_t));
527         int retval;
528         param_in[0] = start_addr;
529         param_in[1] = end_addr;
530
531         retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 2, param_in, 0, NULL, 1);
532
533         free(param_in);
534
535         return retval;
536 }
537
538 /**
539  * \b mips32_pracc_clean_invalidate_cache
540  *
541  * Writeback D$ and Invalidate I$
542  * so that the instructions written can be visible to CPU
543  */
544 static int mips32_pracc_clean_invalidate_cache(struct mips_ejtag *ejtag_info,
545                                                                                                         uint32_t start_addr, uint32_t end_addr)
546 {
547         static const uint32_t code[] = {
548                                                                                                                         /* start: */
549                 MIPS32_MTC0(15, 31, 0),                                                         /* move $15 to COP0 DeSave */
550                 MIPS32_LUI(15, UPPER16(MIPS32_PRACC_STACK)),            /* $15 = MIPS32_PRACC_STACK */
551                 MIPS32_ORI(15, 15, LOWER16(MIPS32_PRACC_STACK)),
552                 MIPS32_SW(8, 0, 15),                                                            /* sw $8,($15) */
553                 MIPS32_SW(9, 0, 15),                                                            /* sw $9,($15) */
554                 MIPS32_SW(10, 0, 15),                                                           /* sw $10,($15) */
555                 MIPS32_SW(11, 0, 15),                                                           /* sw $11,($15) */
556
557                 MIPS32_LUI(8, UPPER16(MIPS32_PRACC_PARAM_IN)),          /* $8 = MIPS32_PRACC_PARAM_IN */
558                 MIPS32_ORI(8, 8, LOWER16(MIPS32_PRACC_PARAM_IN)),
559                 MIPS32_LW(9, 0, 8),                                                                     /* Load write start_addr to $9 */
560                 MIPS32_LW(10, 4, 8),                                                            /* Load write end_addr to $10 */
561                 MIPS32_LW(11, 8, 8),                                                            /* Load write clsiz to $11 */
562
563                                                                                                                         /* cache_loop: */
564                 MIPS32_SLTU(8, 10, 9),                                                          /* sltu $8, $10, $9  :  $8 <- $10 < $9 ? */
565                 MIPS32_BGTZ(8, 6),                                                                      /* bgtz $8, end */
566                 MIPS32_NOP,
567
568                 MIPS32_CACHE(MIPS32_CACHE_D_HIT_WRITEBACK, 0, 9),               /* cache Hit_Writeback_D, 0($9) */
569                 MIPS32_CACHE(MIPS32_CACHE_I_HIT_INVALIDATE, 0, 9),      /* cache Hit_Invalidate_I, 0($9) */
570
571                 MIPS32_ADDU(9, 9, 11),                                                          /* $9 += $11 */
572
573                 MIPS32_B(NEG16(7)),                                                                     /* b cache_loop */
574                 MIPS32_NOP,
575                                                                                                                         /* end: */
576                 MIPS32_LW(11, 0, 15),                                                           /* lw $11,($15) */
577                 MIPS32_LW(10, 0, 15),                                                           /* lw $10,($15) */
578                 MIPS32_LW(9, 0, 15),                                                            /* lw $9,($15) */
579                 MIPS32_LW(8, 0, 15),                                                            /* lw $8,($15) */
580                 MIPS32_B(NEG16(25)),                                                            /* b start */
581                 MIPS32_MFC0(15, 31, 0),                                                         /* move COP0 DeSave to $15 */
582         };
583
584         /**
585          * Find cache line size in bytes
586          */
587         uint32_t conf;
588         uint32_t dl, clsiz;
589
590         mips32_cp0_read(ejtag_info, &conf, 16, 1);
591         dl = (conf & MIPS32_CONFIG1_DL_MASK) >> MIPS32_CONFIG1_DL_SHIFT;
592
593         /* dl encoding : dl=1 => 4 bytes, dl=2 => 8 bytes, etc... */
594         clsiz = 0x2 << dl;
595
596         /* TODO remove array */
597         uint32_t *param_in = malloc(3 * sizeof(uint32_t));
598         int retval;
599         param_in[0] = start_addr;
600         param_in[1] = end_addr;
601         param_in[2] = clsiz;
602
603         retval = mips32_pracc_exec(ejtag_info, ARRAY_SIZE(code), code, 3, param_in, 0, NULL, 1);
604
605         free(param_in);
606
607         return retval;
608 }
609
610 static int mips32_pracc_write_mem_generic(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
611 {
612         uint32_t *code;
613         code = malloc((128 * 3 + 9) * sizeof(uint32_t));        /* alloc memory for the worst case */
614         if (code == NULL) {
615                 LOG_ERROR("Out of memory");
616                 return ERROR_FAIL;
617         }
618
619         uint32_t *buf32 = buf;
620         uint16_t *buf16 = buf;
621         uint8_t *buf8 = buf;
622
623         int i;
624         int retval = ERROR_FAIL;
625         uint32_t *code_p;
626         uint32_t upper_base_addr, last_upper_base_addr;
627         int this_round_count;
628         int code_len;
629
630         while (count) {
631                 this_round_count = (count > 128) ? 128 : count;
632                 last_upper_base_addr = UPPER16((addr + 0x8000));
633                 code_p = code;
634
635                 *code_p++ = MIPS32_MTC0(15, 31, 0);                                     /* save $15 in DeSave */
636                 *code_p++ = MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR);                      /* $15 = MIPS32_PRACC_BASE_ADDR */
637                 *code_p++ = MIPS32_SW(8, PRACC_STACK_OFFSET, 15);                       /* save $8 to pracc stack */
638                 *code_p++ = MIPS32_LUI(15, last_upper_base_addr);                       /* reuse $15 as memory base address */
639                 code_len = 4;
640
641                 for (i = 0; i != this_round_count; i++) {
642                         upper_base_addr = UPPER16((addr + 0x8000));
643                         if (last_upper_base_addr != upper_base_addr) {
644                                 *code_p++ = MIPS32_LUI(15, upper_base_addr);            /* if needed, change upper address in $15*/
645                                 code_len++;
646                                 last_upper_base_addr = upper_base_addr;
647                         }
648
649                         if (size == 4) {        /* for word write check if one half word is 0 and load it accordingly */
650                                 if (LOWER16(*buf32) == 0) {
651                                         *code_p++ = MIPS32_LUI(8, UPPER16(*buf32));             /* load only upper value */
652                                         code_len++;
653                                 } else if (UPPER16(*buf32) == 0) {
654                                         *code_p++ = MIPS32_ORI(8, 0, LOWER16(*buf32));          /* load only lower value */
655                                         code_len++;
656                                 } else {
657                                         *code_p++ = MIPS32_LUI(8, UPPER16(*buf32));             /* load upper and lower */
658                                         *code_p++ = MIPS32_ORI(8, 8, LOWER16(*buf32));
659                                         code_len += 2;
660                                 }
661                                 *code_p++ = MIPS32_SW(8, LOWER16(addr), 15);                    /* store word to memory */
662                                 code_len++;
663                                 buf32++;
664
665                         } else if (size == 2) {
666                                 *code_p++ = MIPS32_ORI(8, 0, *buf16);                           /* load lower value */
667                                 *code_p++ = MIPS32_SH(8, LOWER16(addr), 15);            /* store half word to memory */
668                                 code_len += 2;
669                                 buf16++;
670
671                         } else {
672                                 *code_p++ = MIPS32_ORI(8, 0, *buf8);                            /* load lower value */
673                                 *code_p++ = MIPS32_SB(8, LOWER16(addr), 15);                    /* store byte to memory */
674                                 code_len += 2;
675                                 buf8++;
676                         }
677
678                         addr += size;
679                 }
680
681                 *code_p++ = MIPS32_LUI(15, PRACC_UPPER_BASE_ADDR);                      /* $15 = MIPS32_PRACC_BASE_ADDR */
682                 *code_p++ = MIPS32_LW(8, PRACC_STACK_OFFSET, 15);                       /* restore $8 from pracc stack */
683
684                 code_len += 4;
685                 *code_p++ = MIPS32_B(NEG16(code_len - 1));                                      /* jump to start */
686                 *code_p = MIPS32_MFC0(15, 31, 0);                                       /* restore $15 from DeSave */
687
688                 retval = mips32_pracc_exec(ejtag_info, code_len, code, 0, NULL, 0, NULL, 1);
689                 if (retval != ERROR_OK)
690                         goto exit;
691
692                 count -= this_round_count;
693         }
694
695 exit:
696         free(code);
697         return retval;
698 }
699
700 int mips32_pracc_write_mem(struct mips_ejtag *ejtag_info, uint32_t addr, int size, int count, void *buf)
701 {
702         int retval = mips32_pracc_write_mem_generic(ejtag_info, addr, size, count, buf);
703         if (retval != ERROR_OK)
704                 return retval;
705
706         /**
707          * If we are in the cachable regoion and cache is activated,
708          * we must clean D$ + invalidate I$ after we did the write,
709          * so that changes do not continue to live only in D$, but to be
710          * replicated in I$ also (maybe we wrote the istructions)
711          */
712         uint32_t conf = 0;
713         int cached = 0;
714
715         if ((KSEGX(addr) == KSEG1) || ((addr >= 0xff200000) && (addr <= 0xff3fffff)))
716                 return retval; /*Nothing to do*/
717
718         mips32_cp0_read(ejtag_info, &conf, 16, 0);
719
720         switch (KSEGX(addr)) {
721                 case KUSEG:
722                         cached = (conf & MIPS32_CONFIG0_KU_MASK) >> MIPS32_CONFIG0_KU_SHIFT;
723                         break;
724                 case KSEG0:
725                         cached = (conf & MIPS32_CONFIG0_K0_MASK) >> MIPS32_CONFIG0_K0_SHIFT;
726                         break;
727                 case KSEG2:
728                 case KSEG3:
729                         cached = (conf & MIPS32_CONFIG0_K23_MASK) >> MIPS32_CONFIG0_K23_SHIFT;
730                         break;
731                 default:
732                         /* what ? */
733                         break;
734         }
735
736         /**
737          * Check cachablitiy bits coherency algorithm -
738          * is the region cacheable or uncached.
739          * If cacheable we have to synchronize the cache
740          */
741         if (cached == 0x3) {
742                 uint32_t start_addr, end_addr;
743                 uint32_t rel;
744
745                 start_addr = addr;
746                 end_addr = addr + count * size;
747
748                 /** select cache synchronisation mechanism based on Architecture Release */
749                 rel = (conf & MIPS32_CONFIG0_AR_MASK) >> MIPS32_CONFIG0_AR_SHIFT;
750                 switch (rel) {
751                         case MIPS32_ARCH_REL1:
752                                 /* MIPS32/64 Release 1 - we must use cache instruction */
753                                 mips32_pracc_clean_invalidate_cache(ejtag_info, start_addr, end_addr);
754                                 break;
755                         case MIPS32_ARCH_REL2:
756                                 /* MIPS32/64 Release 2 - we can use synci instruction */
757                                 mips32_pracc_sync_cache(ejtag_info, start_addr, end_addr);
758                                 break;
759                         default:
760                                 /* what ? */
761                                 break;
762                 }
763         }
764
765         return retval;
766 }
767
768 int mips32_pracc_write_regs(struct mips_ejtag *ejtag_info, uint32_t *regs)
769 {
770         static const uint32_t cp0_write_code[] = {
771                 MIPS32_MTC0(1, 12, 0),                                                  /* move $1 to status */
772                 MIPS32_MTLO(1),                                                                 /* move $1 to lo */
773                 MIPS32_MTHI(1),                                                                 /* move $1 to hi */
774                 MIPS32_MTC0(1, 8, 0),                                                   /* move $1 to badvaddr */
775                 MIPS32_MTC0(1, 13, 0),                                                  /* move $1 to cause*/
776                 MIPS32_MTC0(1, 24, 0),                                                  /* move $1 to depc (pc) */
777         };
778
779         uint32_t *code;
780         code = malloc((37 * 2 + 6 + 1) * sizeof(uint32_t));     /* alloc memory for the worst case */
781         if (code == NULL) {
782                 LOG_ERROR("Out of memory");
783                 return ERROR_FAIL;
784         }
785
786         uint32_t *code_p = code;
787         int code_len = 0;
788         /* load registers 2 to 31 with lui an ori instructions, check if same instructions can be saved */
789         for (int i = 2; i < 32; i++) {
790                 if (LOWER16((regs[i])) == 0) {
791                         *code_p++ = MIPS32_LUI(i, UPPER16((regs[i])));          /* if lower half word is 0, lui instruction only */
792                         code_len++;
793                 } else if (UPPER16((regs[i])) == 0) {
794                         *code_p++ = MIPS32_ORI(i, 0, LOWER16((regs[i])));       /* if upper half word is 0, ori with $0 only*/
795                         code_len++;
796                 } else {
797                         *code_p++ = MIPS32_LUI(i, UPPER16((regs[i])));          /* default, load with lui and ori instructions */
798                         *code_p++ = MIPS32_ORI(i, i, LOWER16((regs[i])));
799                         code_len += 2;
800                 }
801         }
802
803         for (int i = 0; i != 6; i++) {
804                 *code_p++ = MIPS32_LUI(1, UPPER16((regs[i + 32])));             /* load CPO value in $1, with lui and ori */
805                 *code_p++ = MIPS32_ORI(1, 1, LOWER16((regs[i + 32])));
806                 *code_p++ = cp0_write_code[i];                                  /* write value from $1 to CPO register */
807                 code_len += 3;
808         }
809
810         *code_p++ = MIPS32_LUI(1, UPPER16((regs[1])));                          /* load upper half word in $1 */
811         code_len += 3;
812         *code_p++ = MIPS32_B(NEG16(code_len - 1)),                                                      /* b start */
813         *code_p = MIPS32_ORI(1, 1, LOWER16((regs[1])));                         /* load lower half word in $1 */
814
815         int retval = mips32_pracc_exec(ejtag_info, code_len, code, 0, NULL, 0, NULL, 1);
816         free(code);
817         return retval;
818 }
819
820 int mips32_pracc_read_regs(struct mips_ejtag *ejtag_info, uint32_t *regs)
821 {
822         static int cp0_read_code[] = {
823                 MIPS32_MFC0(2, 12, 0),                                                  /* move status to $2 */
824                 MIPS32_MFLO(2),                                                                 /* move lo to $2 */
825                 MIPS32_MFHI(2),                                                                 /* move hi to $2 */
826                 MIPS32_MFC0(2, 8, 0),                                                   /* move badvaddr to $2 */
827                 MIPS32_MFC0(2, 13, 0),                                                  /* move cause to $2 */
828                 MIPS32_MFC0(2, 24, 0),                                                  /* move depc (pc) to $2 */
829         };
830
831         uint32_t *code;
832         code = malloc(49 * sizeof(uint32_t));
833         if (code == NULL) {
834                 LOG_ERROR("Out of memory");
835                 return ERROR_FAIL;
836         }
837
838         uint32_t *code_p = code;
839
840         *code_p++ = MIPS32_MTC0(1, 31, 0),                                              /* move $1 to COP0 DeSave */
841         *code_p++ = MIPS32_LUI(1, PRACC_UPPER_BASE_ADDR);                               /* $1 = MIP32_PRACC_BASE_ADDR */
842
843         for (int i = 2; i != 32; i++)
844                 *code_p++ = MIPS32_SW(i, PRACC_OUT_OFFSET + (i * 4), 1);                /* store GPR's 2 to 31 */
845
846         for (int i = 0; i != 6; i++) {
847                 *code_p++ = cp0_read_code[i];                                           /* load COP0 needed registers to $2 */
848                 *code_p++ = MIPS32_SW(2, PRACC_OUT_OFFSET + (i + 32) * 4, 1);   /* store COP0 registers from $2 to param out */
849         }
850
851         *code_p++ = MIPS32_MFC0(2, 31, 0),                                              /* move DeSave to $2, reg1 value */
852         *code_p++ = MIPS32_SW(2, PRACC_OUT_OFFSET + 4, 1);                              /* store reg1 value from $2 to param out */
853
854         *code_p++ = MIPS32_LW(2, PRACC_OUT_OFFSET + 8, 1);                              /* restore $2 from param out (singularity) */
855         *code_p++ = MIPS32_B(NEG16(48));                                                /* b start */
856         *code_p = MIPS32_MFC0(1, 31, 0);                                                /* move COP0 DeSave to $1 */
857
858         int retval = mips32_pracc_exec(ejtag_info, 49, code, 0, NULL, MIPS32NUMCOREREGS, regs, 1);
859
860         free(code);
861         return retval;
862 }
863
864 /* fastdata upload/download requires an initialized working area
865  * to load the download code; it should not be called otherwise
866  * fetch order from the fastdata area
867  * 1. start addr
868  * 2. end addr
869  * 3. data ...
870  */
871 int mips32_pracc_fastdata_xfer(struct mips_ejtag *ejtag_info, struct working_area *source,
872                 int write_t, uint32_t addr, int count, uint32_t *buf)
873 {
874         uint32_t handler_code[] = {
875                 /* caution when editing, table is modified below */
876                 /* r15 points to the start of this code */
877                 MIPS32_SW(8, MIPS32_FASTDATA_HANDLER_SIZE - 4, 15),
878                 MIPS32_SW(9, MIPS32_FASTDATA_HANDLER_SIZE - 8, 15),
879                 MIPS32_SW(10, MIPS32_FASTDATA_HANDLER_SIZE - 12, 15),
880                 MIPS32_SW(11, MIPS32_FASTDATA_HANDLER_SIZE - 16, 15),
881                 /* start of fastdata area in t0 */
882                 MIPS32_LUI(8, UPPER16(MIPS32_PRACC_FASTDATA_AREA)),
883                 MIPS32_ORI(8, 8, LOWER16(MIPS32_PRACC_FASTDATA_AREA)),
884                 MIPS32_LW(9, 0, 8),                                                             /* start addr in t1 */
885                 MIPS32_LW(10, 0, 8),                                                    /* end addr to t2 */
886                                                                                                                 /* loop: */
887                 /* 8 */ MIPS32_LW(11, 0, 0),                                    /* lw t3,[t8 | r9] */
888                 /* 9 */ MIPS32_SW(11, 0, 0),                                    /* sw t3,[r9 | r8] */
889                 MIPS32_BNE(10, 9, NEG16(3)),                                    /* bne $t2,t1,loop */
890                 MIPS32_ADDI(9, 9, 4),                                                   /* addi t1,t1,4 */
891
892                 MIPS32_LW(8, MIPS32_FASTDATA_HANDLER_SIZE - 4, 15),
893                 MIPS32_LW(9, MIPS32_FASTDATA_HANDLER_SIZE - 8, 15),
894                 MIPS32_LW(10, MIPS32_FASTDATA_HANDLER_SIZE - 12, 15),
895                 MIPS32_LW(11, MIPS32_FASTDATA_HANDLER_SIZE - 16, 15),
896
897                 MIPS32_LUI(15, UPPER16(MIPS32_PRACC_TEXT)),
898                 MIPS32_ORI(15, 15, LOWER16(MIPS32_PRACC_TEXT)),
899                 MIPS32_JR(15),                                                          /* jr start */
900                 MIPS32_MFC0(15, 31, 0),                                         /* move COP0 DeSave to $15 */
901         };
902
903         uint32_t jmp_code[] = {
904                 MIPS32_MTC0(15, 31, 0),                 /* move $15 to COP0 DeSave */
905                 /* 1 */ MIPS32_LUI(15, 0),              /* addr of working area added below */
906                 /* 2 */ MIPS32_ORI(15, 15, 0),  /* addr of working area added below */
907                 MIPS32_JR(15),                                  /* jump to ram program */
908                 MIPS32_NOP,
909         };
910
911         int retval, i;
912         uint32_t val, ejtag_ctrl, address;
913
914         if (source->size < MIPS32_FASTDATA_HANDLER_SIZE)
915                 return ERROR_TARGET_RESOURCE_NOT_AVAILABLE;
916
917         if (write_t) {
918                 handler_code[8] = MIPS32_LW(11, 0, 8);  /* load data from probe at fastdata area */
919                 handler_code[9] = MIPS32_SW(11, 0, 9);  /* store data to RAM @ r9 */
920         } else {
921                 handler_code[8] = MIPS32_LW(11, 0, 9);  /* load data from RAM @ r9 */
922                 handler_code[9] = MIPS32_SW(11, 0, 8);  /* store data to probe at fastdata area */
923         }
924
925         /* write program into RAM */
926         if (write_t != ejtag_info->fast_access_save) {
927                 mips32_pracc_write_mem_generic(ejtag_info, source->address, 4, ARRAY_SIZE(handler_code), handler_code);
928                 /* save previous operation to speed to any consecutive read/writes */
929                 ejtag_info->fast_access_save = write_t;
930         }
931
932         LOG_DEBUG("%s using 0x%.8" PRIx32 " for write handler", __func__, source->address);
933
934         jmp_code[1] |= UPPER16(source->address);
935         jmp_code[2] |= LOWER16(source->address);
936
937         for (i = 0; i < (int) ARRAY_SIZE(jmp_code); i++) {
938                 retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
939                 if (retval != ERROR_OK)
940                         return retval;
941
942                 mips_ejtag_set_instr(ejtag_info, EJTAG_INST_DATA);
943                 mips_ejtag_drscan_32_out(ejtag_info, jmp_code[i]);
944
945                 /* Clear the access pending bit (let the processor eat!) */
946                 ejtag_ctrl = ejtag_info->ejtag_ctrl & ~EJTAG_CTRL_PRACC;
947                 mips_ejtag_set_instr(ejtag_info, EJTAG_INST_CONTROL);
948                 mips_ejtag_drscan_32_out(ejtag_info, ejtag_ctrl);
949         }
950
951         retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
952         if (retval != ERROR_OK)
953                 return retval;
954
955         /* next fetch to dmseg should be in FASTDATA_AREA, check */
956         address = 0;
957         mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ADDRESS);
958         retval = mips_ejtag_drscan_32(ejtag_info, &address);
959         if (retval != ERROR_OK)
960                 return retval;
961
962         if (address != MIPS32_PRACC_FASTDATA_AREA)
963                 return ERROR_FAIL;
964
965         /* wait PrAcc pending bit for FASTDATA write */
966         retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
967         if (retval != ERROR_OK)
968                 return retval;
969
970         /* Send the load start address */
971         val = addr;
972         mips_ejtag_set_instr(ejtag_info, EJTAG_INST_FASTDATA);
973         mips_ejtag_fastdata_scan(ejtag_info, 1, &val);
974
975         /* Send the load end address */
976         val = addr + (count - 1) * 4;
977         mips_ejtag_fastdata_scan(ejtag_info, 1, &val);
978
979         for (i = 0; i < count; i++) {
980                 retval = mips_ejtag_fastdata_scan(ejtag_info, write_t, buf++);
981                 if (retval != ERROR_OK)
982                         return retval;
983         }
984
985         retval = jtag_execute_queue();
986         if (retval != ERROR_OK) {
987                 LOG_ERROR("fastdata load failed");
988                 return retval;
989         }
990
991         retval = wait_for_pracc_rw(ejtag_info, &ejtag_ctrl);
992         if (retval != ERROR_OK)
993                 return retval;
994
995         address = 0;
996         mips_ejtag_set_instr(ejtag_info, EJTAG_INST_ADDRESS);
997         retval = mips_ejtag_drscan_32(ejtag_info, &address);
998         if (retval != ERROR_OK)
999                 return retval;
1000
1001         if (address != MIPS32_PRACC_TEXT)
1002                 LOG_ERROR("mini program did not return to start");
1003
1004         return retval;
1005 }