1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2020 Stefan Roese <sr@denx.de>
4  */
5 
6 #include <command.h>
7 #include <config.h>
8 #include <cpu_func.h>
9 #include <dm.h>
10 #include <elf.h>
11 #include <env.h>
12 #include <asm/global_data.h>
13 
14 #include <asm/io.h>
15 #include <linux/compat.h>
16 #include <linux/ctype.h>
17 #include <linux/delay.h>
18 #include <linux/io.h>
19 
20 #include <mach/cvmx-coremask.h>
21 #include <mach/cvmx-bootinfo.h>
22 #include <mach/cvmx-bootmem.h>
23 #include <mach/cvmx-regs.h>
24 #include <mach/cvmx-fuse.h>
25 #include <mach/octeon-model.h>
26 #include <mach/octeon-feature.h>
27 #include <mach/bootoct_cmd.h>
28 #include <mach/cvmx-ciu-defs.h>
29 
30 DECLARE_GLOBAL_DATA_PTR;
31 
32 /* ToDo: Revisit these settings */
33 #define OCTEON_RESERVED_LOW_MEM_SIZE		(512 * 1024)
34 #define OCTEON_RESERVED_LOW_BOOT_MEM_SIZE	(1024 * 1024)
35 #define BOOTLOADER_BOOTMEM_DESC_SPACE		(1024 * 1024)
36 
37 /* Default stack and heap sizes, in bytes */
38 #define DEFAULT_STACK_SIZE			(1 * 1024 * 1024)
39 #define DEFAULT_HEAP_SIZE			(3 * 1024 * 1024)
40 
41 /**
42  * NOTE: This must duplicate octeon_boot_descriptor_t in the toolchain
43  * octeon-app-init.h file.
44  */
45 enum {
46 	/* If set, core should do app-wide init, only one core per app will have
47 	 * this flag set.
48 	 */
49 	BOOT_FLAG_INIT_CORE     = 1,
50 	OCTEON_BL_FLAG_DEBUG    = 1 << 1,
51 	OCTEON_BL_FLAG_NO_MAGIC = 1 << 2,
52 	/* If set, use uart1 for console */
53 	OCTEON_BL_FLAG_CONSOLE_UART1 = 1 << 3,
54 	OCTEON_BL_FLAG_CONSOLE_PCI = 1 << 4,	/* If set, use PCI console */
55 	/* Call exit on break on serial port */
56 	OCTEON_BL_FLAG_BREAK    = 1 << 5,
57 	/*
58 	 * Be sure to update OCTEON_APP_INIT_H_VERSION when new fields are added
59 	 * and to conditionalize the new flag's usage based on the version.
60 	 */
61 } octeon_boot_descriptor_flag;
62 
63 /**
64  * NOTE: This must duplicate octeon_boot_descriptor_t in the toolchain
65  * octeon-app-init.h file.
66  */
67 #ifndef OCTEON_CURRENT_DESC_VERSION
68 # define OCTEON_CURRENT_DESC_VERSION	7
69 #endif
70 /**
71  * NOTE: This must duplicate octeon_boot_descriptor_t in the toolchain
72  * octeon-app-init.h file.
73  */
74 /* Version 7 changes: Change names of deprecated fields */
75 #ifndef OCTEON_ARGV_MAX_ARGS
76 # define OCTEON_ARGV_MAX_ARGS		64
77 #endif
78 
79 /**
80  * NOTE: This must duplicate octeon_boot_descriptor_t in the toolchain
81  * octeon-app-init.h file.
82  */
83 #ifndef OCTEON_SERIAL_LEN
84 # define OCTEON_SERIAL_LEN		20
85 #endif
86 
87 /**
88  * Bootloader structure used to pass info to Octeon executive startup code.
89  * NOTE: all fields are deprecated except for:
90  *  * desc_version
91  *  * desc_size,
92  *  * heap_base
93  *  * heap_end
94  *  * eclock_hz
95  *  * flags
96  *  * argc
97  *  * argv
98  *  * cvmx_desc_vaddr
99  *  * debugger_flags_base_addr
100  *
101  *  All other fields have been moved to the cvmx_descriptor, and the new
102  *  fields should be added there. They are left as placeholders in this
103  *  structure for binary compatibility.
104  *
105  * NOTE: This structure must match what is in the toolchain octeon-app-init.h
106  * file.
107  */
108 struct octeon_boot_descriptor {
109 	/* Start of block referenced by assembly code - do not change! */
110 	u32 desc_version;
111 	u32 desc_size;
112 	u64 stack_top;
113 	u64 heap_base;
114 	u64 heap_end;
115 	u64 deprecated17;
116 	u64 deprecated16;
117 	/* End of block referenced by assembly code - do not change! */
118 	u32 deprecated18;
119 	u32 deprecated15;
120 	u32 deprecated14;
121 	u32 argc;  /* argc for main() */
122 	u32 argv[OCTEON_ARGV_MAX_ARGS];  /* argv for main() */
123 	u32 flags;   /* Flags for application */
124 	u32 core_mask;   /* Coremask running this image */
125 	u32 dram_size;  /* DEPRECATED, DRAM size in megabyes. Used up to SDK 1.8.1 */
126 	u32 phy_mem_desc_addr;
127 	u32 debugger_flags_base_addr;  /* used to pass flags from app to debugger. */
128 	u32 eclock_hz;  /* CPU clock speed, in hz. */
129 	u32 deprecated10;
130 	u32 deprecated9;
131 	u16 deprecated8;
132 	u8  deprecated7;
133 	u8  deprecated6;
134 	u16 deprecated5;
135 	u8  deprecated4;
136 	u8  deprecated3;
137 	char deprecated2[OCTEON_SERIAL_LEN];
138 	u8  deprecated1[6];
139 	u8  deprecated0;
140 	u64 cvmx_desc_vaddr;  /* Address of cvmx descriptor */
141 };
142 
143 static struct octeon_boot_descriptor boot_desc[CVMX_MIPS_MAX_CORES];
144 static struct cvmx_bootinfo cvmx_bootinfo_array[CVMX_MIPS_MAX_CORES];
145 
146 /**
147  * Programs the boot bus moveable region
148  * @param	base	base address to place the boot bus moveable region
149  *			(bits [31:7])
150  * @param	region_num	Selects which region, 0 or 1 for node 0,
151  *				2 or 3 for node 1
152  * @param	enable		Set true to enable, false to disable
153  * @param	data		Pointer to data to put in the region, up to
154  *				16 dwords.
155  * @param	num_words	Number of data dwords (up to 32)
156  *
157  * Return:	0 for success, -1 on error
158  */
octeon_set_moveable_region(u32 base,int region_num,bool enable,const u64 * data,unsigned int num_words)159 static int octeon_set_moveable_region(u32 base, int region_num,
160 				      bool enable, const u64 *data,
161 				      unsigned int num_words)
162 {
163 	int node = region_num >> 1;
164 	u64 val;
165 	int i;
166 	u8 node_mask = 0x01;	/* ToDo: Currently only one node is supported */
167 
168 	debug("%s(0x%x, %d, %d, %p, %u)\n", __func__, base, region_num, enable,
169 	      data, num_words);
170 
171 	if (num_words > 32) {
172 		printf("%s: Too many words (%d) for region %d\n", __func__,
173 		       num_words, region_num);
174 		return -1;
175 	}
176 
177 	if (base & 0x7f) {
178 		printf("%s: Error: base address 0x%x must be 128 byte aligned\n",
179 		       __func__, base);
180 		return -1;
181 	}
182 
183 	if (region_num > (node_mask > 1 ? 3 : 1)) {
184 		printf("%s: Region number %d out of range\n",
185 		       __func__, region_num);
186 		return -1;
187 	}
188 
189 	if (!data && num_words > 0) {
190 		printf("%s: Error: NULL data\n", __func__);
191 		return -1;
192 	}
193 
194 	region_num &= 1;
195 
196 	val = MIO_BOOT_LOC_CFG_EN |
197 		FIELD_PREP(MIO_BOOT_LOC_CFG_BASE, base >> 7);
198 	debug("%s: Setting MIO_BOOT_LOC_CFG(%d) on node %d to 0x%llx\n",
199 	      __func__, region_num, node, val);
200 	csr_wr(CVMX_MIO_BOOT_LOC_CFGX(region_num & 1), val);
201 
202 	val = FIELD_PREP(MIO_BOOT_LOC_ADR_ADR, (region_num ? 0x80 : 0x00) >> 3);
203 	debug("%s: Setting MIO_BOOT_LOC_ADR start to 0x%llx\n", __func__, val);
204 	csr_wr(CVMX_MIO_BOOT_LOC_ADR, val);
205 
206 	for (i = 0; i < num_words; i++) {
207 		debug("  0x%02llx: 0x%016llx\n",
208 		      csr_rd(CVMX_MIO_BOOT_LOC_ADR), data[i]);
209 		csr_wr(CVMX_MIO_BOOT_LOC_DAT, data[i]);
210 	}
211 
212 	return 0;
213 }
214 
215 /**
216  * Parse comma separated numbers into an array
217  *
218  * @param[out] values values read for each node
219  * @param[in] str string to parse
220  * @param base 0 for auto, otherwise 8, 10 or 16 for the number base
221  *
222  * Return: number of values read.
223  */
octeon_parse_nodes(u64 values[CVMX_MAX_NODES],const char * str,int base)224 static int octeon_parse_nodes(u64 values[CVMX_MAX_NODES],
225 			      const char *str, int base)
226 {
227 	int node = 0;
228 	char *sep;
229 
230 	do {
231 		debug("Parsing node %d: \"%s\"\n", node, str);
232 		values[node] = simple_strtoull(str, &sep, base);
233 		debug("  node %d: 0x%llx\n", node, values[node]);
234 		str = sep + 1;
235 	} while (++node < CVMX_MAX_NODES && *sep == ',');
236 
237 	debug("%s: returning %d\n", __func__, node);
238 	return node;
239 }
240 
241 /**
242  * Parse command line arguments
243  *
244  * @param argc			number of arguments
245  * @param[in] argv		array of argument strings
246  * @param cmd			command type
247  * @param[out] boot_args	parsed values
248  *
249  * Return: number of arguments parsed
250  */
octeon_parse_bootopts(int argc,char * const argv[],enum octeon_boot_cmd_type cmd,struct octeon_boot_args * boot_args)251 int octeon_parse_bootopts(int argc, char *const argv[],
252 			  enum octeon_boot_cmd_type cmd,
253 			  struct octeon_boot_args *boot_args)
254 {
255 	u64 node_values[CVMX_MAX_NODES];
256 	int arg, j;
257 	int num_values;
258 	int node;
259 	u8 node_mask = 0x01;	/* ToDo: Currently only one node is supported */
260 
261 	debug("%s(%d, %p, %d, %p)\n", __func__, argc, argv, cmd, boot_args);
262 	memset(boot_args, 0, sizeof(*boot_args));
263 	boot_args->stack_size = DEFAULT_STACK_SIZE;
264 	boot_args->heap_size = DEFAULT_HEAP_SIZE;
265 	boot_args->node_mask = 0;
266 
267 	for (arg = 0; arg < argc; arg++) {
268 		debug("  argv[%d]: %s\n", arg, argv[arg]);
269 		if (cmd == BOOTOCT && !strncmp(argv[arg], "stack=", 6)) {
270 			boot_args->stack_size = simple_strtoul(argv[arg] + 6,
271 							       NULL, 0);
272 		} else if (cmd == BOOTOCT && !strncmp(argv[arg], "heap=", 5)) {
273 			boot_args->heap_size = simple_strtoul(argv[arg] + 5,
274 							      NULL, 0);
275 		} else if (!strncmp(argv[arg], "debug", 5)) {
276 			puts("setting debug flag!\n");
277 			boot_args->boot_flags |= OCTEON_BL_FLAG_DEBUG;
278 		} else if (cmd == BOOTOCT && !strncmp(argv[arg], "break", 5)) {
279 			puts("setting break flag!\n");
280 			boot_args->boot_flags |= OCTEON_BL_FLAG_BREAK;
281 		} else if (!strncmp(argv[arg], "forceboot", 9)) {
282 			boot_args->forceboot = true;
283 		} else if (!strncmp(argv[arg], "nodemask=", 9)) {
284 			boot_args->node_mask = hextoul(argv[arg] + 9, NULL);
285 		} else if (!strncmp(argv[arg], "numcores=", 9)) {
286 			memset(node_values, 0, sizeof(node_values));
287 			num_values = octeon_parse_nodes(node_values,
288 							argv[arg] + 9, 0);
289 			for (j = 0; j < num_values; j++)
290 				boot_args->num_cores[j] = node_values[j];
291 			boot_args->num_cores_set = true;
292 		} else if (!strncmp(argv[arg], "skipcores=", 10)) {
293 			memset(node_values, 0, sizeof(node_values));
294 			num_values = octeon_parse_nodes(node_values,
295 							argv[arg] + 10, 0);
296 			for (j = 0; j < num_values; j++)
297 				boot_args->num_skipped[j] = node_values[j];
298 			boot_args->num_skipped_set = true;
299 		} else if (!strncmp(argv[arg], "console_uart=", 13)) {
300 			boot_args->console_uart = simple_strtoul(argv[arg] + 13,
301 								 NULL, 0);
302 			if (boot_args->console_uart == 1) {
303 				boot_args->boot_flags |=
304 					OCTEON_BL_FLAG_CONSOLE_UART1;
305 			} else if (!boot_args->console_uart) {
306 				boot_args->boot_flags &=
307 					~OCTEON_BL_FLAG_CONSOLE_UART1;
308 			}
309 		} else if (!strncmp(argv[arg], "coremask=", 9)) {
310 			memset(node_values, 0, sizeof(node_values));
311 			num_values = octeon_parse_nodes(node_values,
312 							argv[arg] + 9, 16);
313 			for (j = 0; j < num_values; j++)
314 				cvmx_coremask_set64_node(&boot_args->coremask,
315 							 j, node_values[j]);
316 			boot_args->coremask_set = true;
317 		} else if (cmd == BOOTOCTLINUX &&
318 			   !strncmp(argv[arg], "namedblock=", 11)) {
319 			boot_args->named_block = argv[arg] + 11;
320 		} else if (!strncmp(argv[arg], "endbootargs", 11)) {
321 			boot_args->endbootargs = 1;
322 			arg++;
323 			if (argc >= arg && cmd != BOOTOCTLINUX)
324 				boot_args->app_name = argv[arg];
325 			break;
326 		} else {
327 			debug(" Unknown argument \"%s\"\n", argv[arg]);
328 		}
329 	}
330 
331 	if (boot_args->coremask_set && boot_args->num_cores_set) {
332 		puts("Warning: both coremask and numcores are set, using coremask.\n");
333 	} else if (!boot_args->coremask_set && !boot_args->num_cores_set) {
334 		cvmx_coremask_set_core(&boot_args->coremask, 0);
335 		boot_args->coremask_set = true;
336 	} else if ((!boot_args->coremask_set) && boot_args->num_cores_set) {
337 		cvmx_coremask_for_each_node(node, node_mask)
338 			cvmx_coremask_set64_node(&boot_args->coremask, node,
339 				((1ull << boot_args->num_cores[node]) - 1) <<
340 					boot_args->num_skipped[node]);
341 		boot_args->coremask_set = true;
342 	}
343 
344 	/* Update the node mask based on the coremask or the number of cores */
345 	for (j = 0; j < CVMX_MAX_NODES; j++) {
346 		if (cvmx_coremask_get64_node(&boot_args->coremask, j))
347 			boot_args->node_mask |= 1 << j;
348 	}
349 
350 	debug("%s: return %d\n", __func__, arg);
351 	return arg;
352 }
353 
do_bootoctlinux(struct cmd_tbl * cmdtp,int flag,int argc,char * const argv[])354 int do_bootoctlinux(struct cmd_tbl *cmdtp, int flag, int argc,
355 		    char *const argv[])
356 {
357 	typedef void __noreturn (*kernel_entry_t)(int, ulong, ulong, ulong);
358 	kernel_entry_t kernel;
359 	struct octeon_boot_args boot_args;
360 	int arg_start = 1;
361 	int arg_count;
362 	u64 addr = 0;		/* Address of the ELF image     */
363 	int arg0;
364 	u64 arg1;
365 	u64 arg2;
366 	u64 arg3;
367 	int ret;
368 	struct cvmx_coremask core_mask;
369 	struct cvmx_coremask coremask_to_run;
370 	struct cvmx_coremask avail_coremask;
371 	int first_core;
372 	int core;
373 	const u64 *nmi_code;
374 	int num_dwords;
375 	u8 node_mask = 0x01;
376 	int i;
377 
378 	cvmx_coremask_clear_all(&core_mask);
379 	cvmx_coremask_clear_all(&coremask_to_run);
380 
381 	if (argc >= 2 && (isxdigit(argv[1][0]) && (isxdigit(argv[1][1]) ||
382 						   argv[1][1] == 'x' ||
383 						   argv[1][1] == 'X' ||
384 						   argv[1][1] == '\0'))) {
385 		addr = hextoul(argv[1], NULL);
386 		if (!addr)
387 			addr = CONFIG_SYS_LOAD_ADDR;
388 		arg_start++;
389 	}
390 	if (addr == 0)
391 		addr = CONFIG_SYS_LOAD_ADDR;
392 
393 	debug("%s: arg start: %d\n", __func__, arg_start);
394 	arg_count = octeon_parse_bootopts(argc - arg_start, argv + arg_start,
395 					  BOOTOCTLINUX, &boot_args);
396 
397 	debug("%s:\n"
398 	      " named block: %s\n"
399 	      " node mask: 0x%x\n"
400 	      " stack size: 0x%x\n"
401 	      " heap size: 0x%x\n"
402 	      " boot flags: 0x%x\n"
403 	      " force boot: %s\n"
404 	      " coremask set: %s\n"
405 	      " num cores set: %s\n"
406 	      " num skipped set: %s\n"
407 	      " endbootargs: %s\n",
408 	      __func__,
409 	      boot_args.named_block ? boot_args.named_block : "none",
410 	      boot_args.node_mask,
411 	      boot_args.stack_size,
412 	      boot_args.heap_size,
413 	      boot_args.boot_flags,
414 	      boot_args.forceboot ? "true" : "false",
415 	      boot_args.coremask_set ? "true" : "false",
416 	      boot_args.num_cores_set ? "true" : "false",
417 	      boot_args.num_skipped_set ? "true" : "false",
418 	      boot_args.endbootargs ? "true" : "false");
419 	debug(" num cores: ");
420 	for (i = 0; i < CVMX_MAX_NODES; i++)
421 		debug("%s%d", i > 0 ? ", " : "", boot_args.num_cores[i]);
422 	debug("\n num skipped: ");
423 	for (i = 0; i < CVMX_MAX_NODES; i++) {
424 		debug("%s%d", i > 0 ? ", " : "", boot_args.num_skipped[i]);
425 		debug("\n coremask:\n");
426 		cvmx_coremask_dprint(&boot_args.coremask);
427 	}
428 
429 	if (boot_args.endbootargs) {
430 		debug("endbootargs set, adjusting argc from %d to %d, arg_count: %d, arg_start: %d\n",
431 		      argc, argc - (arg_count + arg_start), arg_count,
432 		      arg_start);
433 		argc -= (arg_count + arg_start);
434 		argv += (arg_count + arg_start);
435 	}
436 
437 	/*
438 	 * numcores specification overrides a coremask on the same command line
439 	 */
440 	cvmx_coremask_copy(&core_mask, &boot_args.coremask);
441 
442 	/*
443 	 * Remove cores from coremask based on environment variable stored in
444 	 * flash
445 	 */
446 	if (validate_coremask(&core_mask) != 0) {
447 		puts("Invalid coremask.\n");
448 		return 1;
449 	} else if (cvmx_coremask_is_empty(&core_mask)) {
450 		puts("Coremask is empty after coremask_override mask.  Nothing to do.\n");
451 		return 0;
452 	}
453 
454 	if (cvmx_coremask_intersects(&core_mask, &coremask_to_run)) {
455 		puts("ERROR: Can't load code on core twice!  Provided coremask:\n");
456 		cvmx_coremask_print(&core_mask);
457 		puts("overlaps previously loaded coremask:\n");
458 		cvmx_coremask_print(&coremask_to_run);
459 		return -1;
460 	}
461 
462 	debug("Setting up boot descriptor block with core mask:\n");
463 	cvmx_coremask_dprint(&core_mask);
464 
465 	/*
466 	 * Add coremask to global mask of cores that have been set up and are
467 	 * runable
468 	 */
469 	cvmx_coremask_or(&coremask_to_run, &coremask_to_run, &core_mask);
470 
471 	/*
472 	 * Load kernel ELF image, or try binary if ELF is not detected.
473 	 * This way the much smaller vmlinux.bin can also be started but
474 	 * has to be loaded at the correct address (ep as parameter).
475 	 */
476 	if (!valid_elf_image(addr))
477 		printf("Booting binary image instead (vmlinux.bin)...\n");
478 	else
479 		addr = load_elf_image_shdr(addr);
480 
481 	/* Set kernel entry point */
482 	kernel = (kernel_entry_t)addr;
483 
484 	/* Init bootmem list for Linux kernel booting */
485 	if (!cvmx_bootmem_phy_mem_list_init(
486 		    gd->ram_size, OCTEON_RESERVED_LOW_MEM_SIZE,
487 		    (void *)CKSEG0ADDR(BOOTLOADER_BOOTMEM_DESC_SPACE))) {
488 		printf("FATAL: Error initializing free memory list\n");
489 		return 0;
490 	}
491 
492 	first_core = cvmx_coremask_get_first_core(&coremask_to_run);
493 
494 	cvmx_coremask_for_each_core(core, &coremask_to_run) {
495 		debug("%s: Activating core %d\n",  __func__, core);
496 
497 		cvmx_bootinfo_array[core].core_mask =
498 			cvmx_coremask_get32(&coremask_to_run);
499 		cvmx_coremask_copy(&cvmx_bootinfo_array[core].ext_core_mask,
500 				   &coremask_to_run);
501 
502 		if (core == first_core)
503 			cvmx_bootinfo_array[core].flags |= BOOT_FLAG_INIT_CORE;
504 
505 		cvmx_bootinfo_array[core].dram_size = gd->ram_size /
506 			(1024 * 1024);
507 
508 		cvmx_bootinfo_array[core].dclock_hz = gd->mem_clk * 1000000;
509 		cvmx_bootinfo_array[core].eclock_hz = gd->cpu_clk;
510 
511 		cvmx_bootinfo_array[core].led_display_base_addr = 0;
512 		cvmx_bootinfo_array[core].phy_mem_desc_addr =
513 			((u32)(u64)__cvmx_bootmem_internal_get_desc_ptr()) &
514 			0x7ffffff;
515 
516 		cvmx_bootinfo_array[core].major_version = CVMX_BOOTINFO_MAJ_VER;
517 		cvmx_bootinfo_array[core].minor_version = CVMX_BOOTINFO_MIN_VER;
518 		cvmx_bootinfo_array[core].fdt_addr = virt_to_phys(gd->fdt_blob);
519 
520 		boot_desc[core].dram_size = gd->ram_size / (1024 * 1024);
521 		boot_desc[core].cvmx_desc_vaddr =
522 			virt_to_phys(&cvmx_bootinfo_array[core]);
523 
524 		boot_desc[core].desc_version = OCTEON_CURRENT_DESC_VERSION;
525 		boot_desc[core].desc_size = sizeof(boot_desc[0]);
526 
527 		boot_desc[core].flags = cvmx_bootinfo_array[core].flags;
528 		boot_desc[core].eclock_hz = cvmx_bootinfo_array[core].eclock_hz;
529 
530 		boot_desc[core].argc = argc;
531 		for (i = 0; i < argc; i++)
532 			boot_desc[core].argv[i] = (u32)virt_to_phys(argv[i]);
533 	}
534 
535 	core = 0;
536 	arg0 = argc;
537 	arg1 = (u64)argv;
538 	arg2 = 0x1;	/* Core 0 sets init core for Linux */
539 	arg3 = XKPHYS | virt_to_phys(&boot_desc[core]);
540 
541 	debug("## Transferring control to Linux (at address %p) ...\n", kernel);
542 
543 	/*
544 	 * Flush cache before jumping to application. Let's flush the
545 	 * whole SDRAM area, since we don't know the size of the image
546 	 * that was loaded.
547 	 */
548 	flush_cache(gd->ram_base, gd->ram_top - gd->ram_base);
549 
550 	/* Take all cores out of reset */
551 	csr_wr(CVMX_CIU_PP_RST, 0);
552 	sync();
553 
554 	/* Wait a short while for the other cores... */
555 	mdelay(100);
556 
557 	/* Install boot code into moveable bus for NMI (other cores) */
558 	nmi_code = (const u64 *)nmi_bootvector;
559 	num_dwords = (((u64)&nmi_handler_para[0] - (u64)nmi_code) + 7) / 8;
560 
561 	ret = octeon_set_moveable_region(0x1fc00000, 0, true, nmi_code,
562 					 num_dwords);
563 	if (ret) {
564 		printf("Error installing NMI handler for SMP core startup\n");
565 		return 0;
566 	}
567 
568 	/* Write NMI handler parameters for Linux kernel booting */
569 	nmi_handler_para[0] = (u64)kernel;
570 	nmi_handler_para[1] = arg0;
571 	nmi_handler_para[2] = arg1;
572 	nmi_handler_para[3] = 0; /* Don't set init core for secondary cores */
573 	nmi_handler_para[4] = arg3;
574 	sync();
575 
576 	/* Wait a short while for the other cores... */
577 	mdelay(100);
578 
579 	/*
580 	 * Cores have already been taken out of reset to conserve power.
581 	 * We need to send a NMI to get the cores out of their wait loop
582 	 */
583 	octeon_get_available_coremask(&avail_coremask);
584 	debug("Available coremask:\n");
585 	cvmx_coremask_dprint(&avail_coremask);
586 	debug("Starting coremask:\n");
587 	cvmx_coremask_dprint(&coremask_to_run);
588 	debug("Sending NMIs to other cores\n");
589 	if (octeon_has_feature(OCTEON_FEATURE_CIU3)) {
590 		u64 avail_cm;
591 		int node;
592 
593 		cvmx_coremask_for_each_node(node, node_mask) {
594 			avail_cm = cvmx_coremask_get64_node(&avail_coremask,
595 							    node);
596 
597 			if (avail_cm != 0) {
598 				debug("Sending NMI  to node %d, coremask=0x%llx, CIU3_NMI=0x%llx\n",
599 				      node, avail_cm,
600 				      (node > 0 ? -1ull : -2ull) & avail_cm);
601 				csr_wr(CVMX_CIU3_NMI,
602 				       (node > 0 ? -1ull : -2ull) & avail_cm);
603 			}
604 		}
605 	} else {
606 		csr_wr(CVMX_CIU_NMI,
607 		       -2ull & cvmx_coremask_get64(&avail_coremask));
608 	}
609 	debug("Done sending NMIs\n");
610 
611 	/* Wait a short while for the other cores... */
612 	mdelay(100);
613 
614 	/*
615 	 * pass address parameter as argv[0] (aka command name),
616 	 * and all remaining args
617 	 * a0 = argc
618 	 * a1 = argv (32 bit physical addresses, not pointers)
619 	 * a2 = init core
620 	 * a3 = boot descriptor address
621 	 * a4/t0 = entry point (only used by assembly stub)
622 	 */
623 	kernel(arg0, arg1, arg2, arg3);
624 
625 	return 0;
626 }
627 
628 U_BOOT_CMD(bootoctlinux, 32, 0, do_bootoctlinux,
629 	   "Boot from a linux ELF image in memory",
630 	   "elf_address [coremask=mask_to_run | numcores=core_cnt_to_run] "
631 	   "[forceboot] [skipcores=core_cnt_to_skip] [namedblock=name] [endbootargs] [app_args ...]\n"
632 	   "elf_address - address of ELF image to load. If 0, default load address\n"
633 	   "              is  used.\n"
634 	   "coremask    - mask of cores to run on.  Anded with coremask_override\n"
635 	   "              environment variable to ensure only working cores are used\n"
636 	   "numcores    - number of cores to run on.  Runs on specified number of cores,\n"
637 	   "              taking into account the coremask_override.\n"
638 	   "skipcores   - only meaningful with numcores.  Skips this many cores\n"
639 	   "              (starting from 0) when loading the numcores cores.\n"
640 	   "              For example, setting skipcores to 1 will skip core 0\n"
641 	   "              and load the application starting at the next available core.\n"
642 	   "forceboot   - if set, boots application even if core 0 is not in mask\n"
643 	   "namedblock	- specifies a named block to load the kernel\n"
644 	   "endbootargs - if set, bootloader does not process any further arguments and\n"
645 	   "              only passes the arguments that follow to the kernel.\n"
646 	   "              If not set, the kernel gets the entire commnad line as\n"
647 	   "              arguments.\n" "\n");
648