xref: /openbmc/linux/arch/mips/kernel/vpe.c (revision f58437f1)
1 /*
2  * Copyright (C) 2004, 2005 MIPS Technologies, Inc.  All rights reserved.
3  *
4  *  This program is free software; you can distribute it and/or modify it
5  *  under the terms of the GNU General Public License (Version 2) as
6  *  published by the Free Software Foundation.
7  *
8  *  This program is distributed in the hope it will be useful, but WITHOUT
9  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
11  *  for more details.
12  *
13  *  You should have received a copy of the GNU General Public License along
14  *  with this program; if not, write to the Free Software Foundation, Inc.,
15  *  59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
16  */
17 
18 /*
19  * VPE support module
20  *
21  * Provides support for loading a MIPS SP program on VPE1.
22  * The SP environment is rather simple, no tlb's.  It needs to be relocatable
23  * (or partially linked). You should initialise your stack in the startup
24  * code. This loader looks for the symbol __start and sets up
25  * execution to resume from there. The MIPS SDE kit contains suitable examples.
26  *
27  * To load and run, simply cat a SP 'program file' to /dev/vpe1.
28  * i.e cat spapp >/dev/vpe1.
29  */
30 #include <linux/kernel.h>
31 #include <linux/device.h>
32 #include <linux/fs.h>
33 #include <linux/init.h>
34 #include <asm/uaccess.h>
35 #include <linux/slab.h>
36 #include <linux/list.h>
37 #include <linux/vmalloc.h>
38 #include <linux/elf.h>
39 #include <linux/seq_file.h>
40 #include <linux/syscalls.h>
41 #include <linux/moduleloader.h>
42 #include <linux/interrupt.h>
43 #include <linux/poll.h>
44 #include <linux/bootmem.h>
45 #include <asm/mipsregs.h>
46 #include <asm/mipsmtregs.h>
47 #include <asm/cacheflush.h>
48 #include <linux/atomic.h>
49 #include <asm/cpu.h>
50 #include <asm/mips_mt.h>
51 #include <asm/processor.h>
52 #include <asm/vpe.h>
53 
54 typedef void *vpe_handle;
55 
56 #ifndef ARCH_SHF_SMALL
57 #define ARCH_SHF_SMALL 0
58 #endif
59 
60 /* If this is set, the section belongs in the init part of the module */
61 #define INIT_OFFSET_MASK (1UL << (BITS_PER_LONG-1))
62 
63 /*
64  * The number of TCs and VPEs physically available on the core
65  */
66 static int hw_tcs, hw_vpes;
67 static char module_name[] = "vpe";
68 static int major;
69 static const int minor = 1;	/* fixed for now  */
70 
71 /* grab the likely amount of memory we will need. */
72 #ifdef CONFIG_MIPS_VPE_LOADER_TOM
73 #define P_SIZE (2 * 1024 * 1024)
74 #else
75 /* add an overhead to the max kmalloc size for non-striped symbols/etc */
76 #define P_SIZE (256 * 1024)
77 #endif
78 
79 extern unsigned long physical_memsize;
80 
81 #define MAX_VPES 16
82 #define VPE_PATH_MAX 256
83 
84 enum vpe_state {
85 	VPE_STATE_UNUSED = 0,
86 	VPE_STATE_INUSE,
87 	VPE_STATE_RUNNING
88 };
89 
90 enum tc_state {
91 	TC_STATE_UNUSED = 0,
92 	TC_STATE_INUSE,
93 	TC_STATE_RUNNING,
94 	TC_STATE_DYNAMIC
95 };
96 
97 struct vpe {
98 	enum vpe_state state;
99 
100 	/* (device) minor associated with this vpe */
101 	int minor;
102 
103 	/* elfloader stuff */
104 	void *load_addr;
105 	unsigned long len;
106 	char *pbuffer;
107 	unsigned long plen;
108 	char cwd[VPE_PATH_MAX];
109 
110 	unsigned long __start;
111 
112 	/* tc's associated with this vpe */
113 	struct list_head tc;
114 
115 	/* The list of vpe's */
116 	struct list_head list;
117 
118 	/* shared symbol address */
119 	void *shared_ptr;
120 
121 	/* the list of who wants to know when something major happens */
122 	struct list_head notify;
123 
124 	unsigned int ntcs;
125 };
126 
127 struct tc {
128 	enum tc_state state;
129 	int index;
130 
131 	struct vpe *pvpe;	/* parent VPE */
132 	struct list_head tc;	/* The list of TC's with this VPE */
133 	struct list_head list;	/* The global list of tc's */
134 };
135 
136 struct {
137 	spinlock_t vpe_list_lock;
138 	struct list_head vpe_list;	/* Virtual processing elements */
139 	spinlock_t tc_list_lock;
140 	struct list_head tc_list;	/* Thread contexts */
141 } vpecontrol = {
142 	.vpe_list_lock	= __SPIN_LOCK_UNLOCKED(vpe_list_lock),
143 	.vpe_list	= LIST_HEAD_INIT(vpecontrol.vpe_list),
144 	.tc_list_lock	= __SPIN_LOCK_UNLOCKED(tc_list_lock),
145 	.tc_list	= LIST_HEAD_INIT(vpecontrol.tc_list)
146 };
147 
148 static void release_progmem(void *ptr);
149 
150 /* get the vpe associated with this minor */
151 static struct vpe *get_vpe(int minor)
152 {
153 	struct vpe *res, *v;
154 
155 	if (!cpu_has_mipsmt)
156 		return NULL;
157 
158 	res = NULL;
159 	spin_lock(&vpecontrol.vpe_list_lock);
160 	list_for_each_entry(v, &vpecontrol.vpe_list, list) {
161 		if (v->minor == minor) {
162 			res = v;
163 			break;
164 		}
165 	}
166 	spin_unlock(&vpecontrol.vpe_list_lock);
167 
168 	return res;
169 }
170 
171 /* get the vpe associated with this minor */
172 static struct tc *get_tc(int index)
173 {
174 	struct tc *res, *t;
175 
176 	res = NULL;
177 	spin_lock(&vpecontrol.tc_list_lock);
178 	list_for_each_entry(t, &vpecontrol.tc_list, list) {
179 		if (t->index == index) {
180 			res = t;
181 			break;
182 		}
183 	}
184 	spin_unlock(&vpecontrol.tc_list_lock);
185 
186 	return res;
187 }
188 
189 /* allocate a vpe and associate it with this minor (or index) */
190 static struct vpe *alloc_vpe(int minor)
191 {
192 	struct vpe *v;
193 
194 	if ((v = kzalloc(sizeof(struct vpe), GFP_KERNEL)) == NULL)
195 		return NULL;
196 
197 	INIT_LIST_HEAD(&v->tc);
198 	spin_lock(&vpecontrol.vpe_list_lock);
199 	list_add_tail(&v->list, &vpecontrol.vpe_list);
200 	spin_unlock(&vpecontrol.vpe_list_lock);
201 
202 	INIT_LIST_HEAD(&v->notify);
203 	v->minor = minor;
204 
205 	return v;
206 }
207 
208 /* allocate a tc. At startup only tc0 is running, all other can be halted. */
209 static struct tc *alloc_tc(int index)
210 {
211 	struct tc *tc;
212 
213 	if ((tc = kzalloc(sizeof(struct tc), GFP_KERNEL)) == NULL)
214 		goto out;
215 
216 	INIT_LIST_HEAD(&tc->tc);
217 	tc->index = index;
218 
219 	spin_lock(&vpecontrol.tc_list_lock);
220 	list_add_tail(&tc->list, &vpecontrol.tc_list);
221 	spin_unlock(&vpecontrol.tc_list_lock);
222 
223 out:
224 	return tc;
225 }
226 
227 /* clean up and free everything */
228 static void release_vpe(struct vpe *v)
229 {
230 	list_del(&v->list);
231 	if (v->load_addr)
232 		release_progmem(v);
233 	kfree(v);
234 }
235 
236 static void __maybe_unused dump_mtregs(void)
237 {
238 	unsigned long val;
239 
240 	val = read_c0_config3();
241 	printk("config3 0x%lx MT %ld\n", val,
242 	       (val & CONFIG3_MT) >> CONFIG3_MT_SHIFT);
243 
244 	val = read_c0_mvpcontrol();
245 	printk("MVPControl 0x%lx, STLB %ld VPC %ld EVP %ld\n", val,
246 	       (val & MVPCONTROL_STLB) >> MVPCONTROL_STLB_SHIFT,
247 	       (val & MVPCONTROL_VPC) >> MVPCONTROL_VPC_SHIFT,
248 	       (val & MVPCONTROL_EVP));
249 
250 	val = read_c0_mvpconf0();
251 	printk("mvpconf0 0x%lx, PVPE %ld PTC %ld M %ld\n", val,
252 	       (val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT,
253 	       val & MVPCONF0_PTC, (val & MVPCONF0_M) >> MVPCONF0_M_SHIFT);
254 }
255 
256 /* Find some VPE program space	*/
257 static void *alloc_progmem(unsigned long len)
258 {
259 	void *addr;
260 
261 #ifdef CONFIG_MIPS_VPE_LOADER_TOM
262 	/*
263 	 * This means you must tell Linux to use less memory than you
264 	 * physically have, for example by passing a mem= boot argument.
265 	 */
266 	addr = pfn_to_kaddr(max_low_pfn);
267 	memset(addr, 0, len);
268 #else
269 	/* simple grab some mem for now */
270 	addr = kzalloc(len, GFP_KERNEL);
271 #endif
272 
273 	return addr;
274 }
275 
276 static void release_progmem(void *ptr)
277 {
278 #ifndef CONFIG_MIPS_VPE_LOADER_TOM
279 	kfree(ptr);
280 #endif
281 }
282 
283 /* Update size with this section: return offset. */
284 static long get_offset(unsigned long *size, Elf_Shdr * sechdr)
285 {
286 	long ret;
287 
288 	ret = ALIGN(*size, sechdr->sh_addralign ? : 1);
289 	*size = ret + sechdr->sh_size;
290 	return ret;
291 }
292 
293 /* Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
294    might -- code, read-only data, read-write data, small data.	Tally
295    sizes, and place the offsets into sh_entsize fields: high bit means it
296    belongs in init. */
297 static void layout_sections(struct module *mod, const Elf_Ehdr * hdr,
298 			    Elf_Shdr * sechdrs, const char *secstrings)
299 {
300 	static unsigned long const masks[][2] = {
301 		/* NOTE: all executable code must be the first section
302 		 * in this array; otherwise modify the text_size
303 		 * finder in the two loops below */
304 		{SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL},
305 		{SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL},
306 		{SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL},
307 		{ARCH_SHF_SMALL | SHF_ALLOC, 0}
308 	};
309 	unsigned int m, i;
310 
311 	for (i = 0; i < hdr->e_shnum; i++)
312 		sechdrs[i].sh_entsize = ~0UL;
313 
314 	for (m = 0; m < ARRAY_SIZE(masks); ++m) {
315 		for (i = 0; i < hdr->e_shnum; ++i) {
316 			Elf_Shdr *s = &sechdrs[i];
317 
318 			//  || strncmp(secstrings + s->sh_name, ".init", 5) == 0)
319 			if ((s->sh_flags & masks[m][0]) != masks[m][0]
320 			    || (s->sh_flags & masks[m][1])
321 			    || s->sh_entsize != ~0UL)
322 				continue;
323 			s->sh_entsize =
324 				get_offset((unsigned long *)&mod->core_size, s);
325 		}
326 
327 		if (m == 0)
328 			mod->core_text_size = mod->core_size;
329 
330 	}
331 }
332 
333 
334 /* from module-elf32.c, but subverted a little */
335 
336 struct mips_hi16 {
337 	struct mips_hi16 *next;
338 	Elf32_Addr *addr;
339 	Elf32_Addr value;
340 };
341 
342 static struct mips_hi16 *mips_hi16_list;
343 static unsigned int gp_offs, gp_addr;
344 
345 static int apply_r_mips_none(struct module *me, uint32_t *location,
346 			     Elf32_Addr v)
347 {
348 	return 0;
349 }
350 
351 static int apply_r_mips_gprel16(struct module *me, uint32_t *location,
352 				Elf32_Addr v)
353 {
354 	int rel;
355 
356 	if( !(*location & 0xffff) ) {
357 		rel = (int)v - gp_addr;
358 	}
359 	else {
360 		/* .sbss + gp(relative) + offset */
361 		/* kludge! */
362 		rel =  (int)(short)((int)v + gp_offs +
363 				    (int)(short)(*location & 0xffff) - gp_addr);
364 	}
365 
366 	if( (rel > 32768) || (rel < -32768) ) {
367 		printk(KERN_DEBUG "VPE loader: apply_r_mips_gprel16: "
368 		       "relative address 0x%x out of range of gp register\n",
369 		       rel);
370 		return -ENOEXEC;
371 	}
372 
373 	*location = (*location & 0xffff0000) | (rel & 0xffff);
374 
375 	return 0;
376 }
377 
378 static int apply_r_mips_pc16(struct module *me, uint32_t *location,
379 			     Elf32_Addr v)
380 {
381 	int rel;
382 	rel = (((unsigned int)v - (unsigned int)location));
383 	rel >>= 2;		// because the offset is in _instructions_ not bytes.
384 	rel -= 1;		// and one instruction less due to the branch delay slot.
385 
386 	if( (rel > 32768) || (rel < -32768) ) {
387 		printk(KERN_DEBUG "VPE loader: "
388 		       "apply_r_mips_pc16: relative address out of range 0x%x\n", rel);
389 		return -ENOEXEC;
390 	}
391 
392 	*location = (*location & 0xffff0000) | (rel & 0xffff);
393 
394 	return 0;
395 }
396 
397 static int apply_r_mips_32(struct module *me, uint32_t *location,
398 			   Elf32_Addr v)
399 {
400 	*location += v;
401 
402 	return 0;
403 }
404 
405 static int apply_r_mips_26(struct module *me, uint32_t *location,
406 			   Elf32_Addr v)
407 {
408 	if (v % 4) {
409 		printk(KERN_DEBUG "VPE loader: apply_r_mips_26 "
410 		       " unaligned relocation\n");
411 		return -ENOEXEC;
412 	}
413 
414 /*
415  * Not desperately convinced this is a good check of an overflow condition
416  * anyway. But it gets in the way of handling undefined weak symbols which
417  * we want to set to zero.
418  * if ((v & 0xf0000000) != (((unsigned long)location + 4) & 0xf0000000)) {
419  * printk(KERN_ERR
420  * "module %s: relocation overflow\n",
421  * me->name);
422  * return -ENOEXEC;
423  * }
424  */
425 
426 	*location = (*location & ~0x03ffffff) |
427 		((*location + (v >> 2)) & 0x03ffffff);
428 	return 0;
429 }
430 
431 static int apply_r_mips_hi16(struct module *me, uint32_t *location,
432 			     Elf32_Addr v)
433 {
434 	struct mips_hi16 *n;
435 
436 	/*
437 	 * We cannot relocate this one now because we don't know the value of
438 	 * the carry we need to add.  Save the information, and let LO16 do the
439 	 * actual relocation.
440 	 */
441 	n = kmalloc(sizeof *n, GFP_KERNEL);
442 	if (!n)
443 		return -ENOMEM;
444 
445 	n->addr = location;
446 	n->value = v;
447 	n->next = mips_hi16_list;
448 	mips_hi16_list = n;
449 
450 	return 0;
451 }
452 
453 static int apply_r_mips_lo16(struct module *me, uint32_t *location,
454 			     Elf32_Addr v)
455 {
456 	unsigned long insnlo = *location;
457 	Elf32_Addr val, vallo;
458 	struct mips_hi16 *l, *next;
459 
460 	/* Sign extend the addend we extract from the lo insn.	*/
461 	vallo = ((insnlo & 0xffff) ^ 0x8000) - 0x8000;
462 
463 	if (mips_hi16_list != NULL) {
464 
465 		l = mips_hi16_list;
466 		while (l != NULL) {
467 			unsigned long insn;
468 
469 			/*
470 			 * The value for the HI16 had best be the same.
471 			 */
472 			if (v != l->value) {
473 				printk(KERN_DEBUG "VPE loader: "
474 				       "apply_r_mips_lo16/hi16: \t"
475 				       "inconsistent value information\n");
476 				goto out_free;
477 			}
478 
479 			/*
480 			 * Do the HI16 relocation.  Note that we actually don't
481 			 * need to know anything about the LO16 itself, except
482 			 * where to find the low 16 bits of the addend needed
483 			 * by the LO16.
484 			 */
485 			insn = *l->addr;
486 			val = ((insn & 0xffff) << 16) + vallo;
487 			val += v;
488 
489 			/*
490 			 * Account for the sign extension that will happen in
491 			 * the low bits.
492 			 */
493 			val = ((val >> 16) + ((val & 0x8000) != 0)) & 0xffff;
494 
495 			insn = (insn & ~0xffff) | val;
496 			*l->addr = insn;
497 
498 			next = l->next;
499 			kfree(l);
500 			l = next;
501 		}
502 
503 		mips_hi16_list = NULL;
504 	}
505 
506 	/*
507 	 * Ok, we're done with the HI16 relocs.	 Now deal with the LO16.
508 	 */
509 	val = v + vallo;
510 	insnlo = (insnlo & ~0xffff) | (val & 0xffff);
511 	*location = insnlo;
512 
513 	return 0;
514 
515 out_free:
516 	while (l != NULL) {
517 		next = l->next;
518 		kfree(l);
519 		l = next;
520 	}
521 	mips_hi16_list = NULL;
522 
523 	return -ENOEXEC;
524 }
525 
526 static int (*reloc_handlers[]) (struct module *me, uint32_t *location,
527 				Elf32_Addr v) = {
528 	[R_MIPS_NONE]	= apply_r_mips_none,
529 	[R_MIPS_32]	= apply_r_mips_32,
530 	[R_MIPS_26]	= apply_r_mips_26,
531 	[R_MIPS_HI16]	= apply_r_mips_hi16,
532 	[R_MIPS_LO16]	= apply_r_mips_lo16,
533 	[R_MIPS_GPREL16] = apply_r_mips_gprel16,
534 	[R_MIPS_PC16] = apply_r_mips_pc16
535 };
536 
537 static char *rstrs[] = {
538 	[R_MIPS_NONE]	= "MIPS_NONE",
539 	[R_MIPS_32]	= "MIPS_32",
540 	[R_MIPS_26]	= "MIPS_26",
541 	[R_MIPS_HI16]	= "MIPS_HI16",
542 	[R_MIPS_LO16]	= "MIPS_LO16",
543 	[R_MIPS_GPREL16] = "MIPS_GPREL16",
544 	[R_MIPS_PC16] = "MIPS_PC16"
545 };
546 
547 static int apply_relocations(Elf32_Shdr *sechdrs,
548 		      const char *strtab,
549 		      unsigned int symindex,
550 		      unsigned int relsec,
551 		      struct module *me)
552 {
553 	Elf32_Rel *rel = (void *) sechdrs[relsec].sh_addr;
554 	Elf32_Sym *sym;
555 	uint32_t *location;
556 	unsigned int i;
557 	Elf32_Addr v;
558 	int res;
559 
560 	for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
561 		Elf32_Word r_info = rel[i].r_info;
562 
563 		/* This is where to make the change */
564 		location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
565 			+ rel[i].r_offset;
566 		/* This is the symbol it is referring to */
567 		sym = (Elf32_Sym *)sechdrs[symindex].sh_addr
568 			+ ELF32_R_SYM(r_info);
569 
570 		if (!sym->st_value) {
571 			printk(KERN_DEBUG "%s: undefined weak symbol %s\n",
572 			       me->name, strtab + sym->st_name);
573 			/* just print the warning, dont barf */
574 		}
575 
576 		v = sym->st_value;
577 
578 		res = reloc_handlers[ELF32_R_TYPE(r_info)](me, location, v);
579 		if( res ) {
580 			char *r = rstrs[ELF32_R_TYPE(r_info)];
581 			printk(KERN_WARNING "VPE loader: .text+0x%x "
582 			       "relocation type %s for symbol \"%s\" failed\n",
583 			       rel[i].r_offset, r ? r : "UNKNOWN",
584 			       strtab + sym->st_name);
585 			return res;
586 		}
587 	}
588 
589 	return 0;
590 }
591 
592 static inline void save_gp_address(unsigned int secbase, unsigned int rel)
593 {
594 	gp_addr = secbase + rel;
595 	gp_offs = gp_addr - (secbase & 0xffff0000);
596 }
597 /* end module-elf32.c */
598 
599 
600 
601 /* Change all symbols so that sh_value encodes the pointer directly. */
602 static void simplify_symbols(Elf_Shdr * sechdrs,
603 			    unsigned int symindex,
604 			    const char *strtab,
605 			    const char *secstrings,
606 			    unsigned int nsecs, struct module *mod)
607 {
608 	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
609 	unsigned long secbase, bssbase = 0;
610 	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
611 	int size;
612 
613 	/* find the .bss section for COMMON symbols */
614 	for (i = 0; i < nsecs; i++) {
615 		if (strncmp(secstrings + sechdrs[i].sh_name, ".bss", 4) == 0) {
616 			bssbase = sechdrs[i].sh_addr;
617 			break;
618 		}
619 	}
620 
621 	for (i = 1; i < n; i++) {
622 		switch (sym[i].st_shndx) {
623 		case SHN_COMMON:
624 			/* Allocate space for the symbol in the .bss section.
625 			   st_value is currently size.
626 			   We want it to have the address of the symbol. */
627 
628 			size = sym[i].st_value;
629 			sym[i].st_value = bssbase;
630 
631 			bssbase += size;
632 			break;
633 
634 		case SHN_ABS:
635 			/* Don't need to do anything */
636 			break;
637 
638 		case SHN_UNDEF:
639 			/* ret = -ENOENT; */
640 			break;
641 
642 		case SHN_MIPS_SCOMMON:
643 			printk(KERN_DEBUG "simplify_symbols: ignoring SHN_MIPS_SCOMMON "
644 			       "symbol <%s> st_shndx %d\n", strtab + sym[i].st_name,
645 			       sym[i].st_shndx);
646 			// .sbss section
647 			break;
648 
649 		default:
650 			secbase = sechdrs[sym[i].st_shndx].sh_addr;
651 
652 			if (strncmp(strtab + sym[i].st_name, "_gp", 3) == 0) {
653 				save_gp_address(secbase, sym[i].st_value);
654 			}
655 
656 			sym[i].st_value += secbase;
657 			break;
658 		}
659 	}
660 }
661 
662 #ifdef DEBUG_ELFLOADER
663 static void dump_elfsymbols(Elf_Shdr * sechdrs, unsigned int symindex,
664 			    const char *strtab, struct module *mod)
665 {
666 	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
667 	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
668 
669 	printk(KERN_DEBUG "dump_elfsymbols: n %d\n", n);
670 	for (i = 1; i < n; i++) {
671 		printk(KERN_DEBUG " i %d name <%s> 0x%x\n", i,
672 		       strtab + sym[i].st_name, sym[i].st_value);
673 	}
674 }
675 #endif
676 
677 /* We are prepared so configure and start the VPE... */
678 static int vpe_run(struct vpe * v)
679 {
680 	unsigned long flags, val, dmt_flag;
681 	struct vpe_notifications *n;
682 	unsigned int vpeflags;
683 	struct tc *t;
684 
685 	/* check we are the Master VPE */
686 	local_irq_save(flags);
687 	val = read_c0_vpeconf0();
688 	if (!(val & VPECONF0_MVP)) {
689 		printk(KERN_WARNING
690 		       "VPE loader: only Master VPE's are allowed to configure MT\n");
691 		local_irq_restore(flags);
692 
693 		return -1;
694 	}
695 
696 	dmt_flag = dmt();
697 	vpeflags = dvpe();
698 
699 	if (list_empty(&v->tc)) {
700 		evpe(vpeflags);
701 		emt(dmt_flag);
702 		local_irq_restore(flags);
703 
704 		printk(KERN_WARNING
705 		       "VPE loader: No TC's associated with VPE %d\n",
706 		       v->minor);
707 
708 		return -ENOEXEC;
709 	}
710 
711 	t = list_first_entry(&v->tc, struct tc, tc);
712 
713 	/* Put MVPE's into 'configuration state' */
714 	set_c0_mvpcontrol(MVPCONTROL_VPC);
715 
716 	settc(t->index);
717 
718 	/* should check it is halted, and not activated */
719 	if ((read_tc_c0_tcstatus() & TCSTATUS_A) || !(read_tc_c0_tchalt() & TCHALT_H)) {
720 		evpe(vpeflags);
721 		emt(dmt_flag);
722 		local_irq_restore(flags);
723 
724 		printk(KERN_WARNING "VPE loader: TC %d is already active!\n",
725 		       t->index);
726 
727 		return -ENOEXEC;
728 	}
729 
730 	/* Write the address we want it to start running from in the TCPC register. */
731 	write_tc_c0_tcrestart((unsigned long)v->__start);
732 	write_tc_c0_tccontext((unsigned long)0);
733 
734 	/*
735 	 * Mark the TC as activated, not interrupt exempt and not dynamically
736 	 * allocatable
737 	 */
738 	val = read_tc_c0_tcstatus();
739 	val = (val & ~(TCSTATUS_DA | TCSTATUS_IXMT)) | TCSTATUS_A;
740 	write_tc_c0_tcstatus(val);
741 
742 	write_tc_c0_tchalt(read_tc_c0_tchalt() & ~TCHALT_H);
743 
744 	/*
745 	 * The sde-kit passes 'memsize' to __start in $a3, so set something
746 	 * here...  Or set $a3 to zero and define DFLT_STACK_SIZE and
747 	 * DFLT_HEAP_SIZE when you compile your program
748 	 */
749 	mttgpr(6, v->ntcs);
750 	mttgpr(7, physical_memsize);
751 
752 	/* set up VPE1 */
753 	/*
754 	 * bind the TC to VPE 1 as late as possible so we only have the final
755 	 * VPE registers to set up, and so an EJTAG probe can trigger on it
756 	 */
757 	write_tc_c0_tcbind((read_tc_c0_tcbind() & ~TCBIND_CURVPE) | 1);
758 
759 	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~(VPECONF0_VPA));
760 
761 	back_to_back_c0_hazard();
762 
763 	/* Set up the XTC bit in vpeconf0 to point at our tc */
764 	write_vpe_c0_vpeconf0( (read_vpe_c0_vpeconf0() & ~(VPECONF0_XTC))
765 			      | (t->index << VPECONF0_XTC_SHIFT));
766 
767 	back_to_back_c0_hazard();
768 
769 	/* enable this VPE */
770 	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() | VPECONF0_VPA);
771 
772 	/* clear out any left overs from a previous program */
773 	write_vpe_c0_status(0);
774 	write_vpe_c0_cause(0);
775 
776 	/* take system out of configuration state */
777 	clear_c0_mvpcontrol(MVPCONTROL_VPC);
778 
779 	/*
780 	 * SMTC/SMVP kernels manage VPE enable independently,
781 	 * but uniprocessor kernels need to turn it on, even
782 	 * if that wasn't the pre-dvpe() state.
783 	 */
784 #ifdef CONFIG_SMP
785 	evpe(vpeflags);
786 #else
787 	evpe(EVPE_ENABLE);
788 #endif
789 	emt(dmt_flag);
790 	local_irq_restore(flags);
791 
792 	list_for_each_entry(n, &v->notify, list)
793 		n->start(minor);
794 
795 	return 0;
796 }
797 
798 static int find_vpe_symbols(struct vpe * v, Elf_Shdr * sechdrs,
799 				      unsigned int symindex, const char *strtab,
800 				      struct module *mod)
801 {
802 	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
803 	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
804 
805 	for (i = 1; i < n; i++) {
806 		if (strcmp(strtab + sym[i].st_name, "__start") == 0) {
807 			v->__start = sym[i].st_value;
808 		}
809 
810 		if (strcmp(strtab + sym[i].st_name, "vpe_shared") == 0) {
811 			v->shared_ptr = (void *)sym[i].st_value;
812 		}
813 	}
814 
815 	if ( (v->__start == 0) || (v->shared_ptr == NULL))
816 		return -1;
817 
818 	return 0;
819 }
820 
821 /*
822  * Allocates a VPE with some program code space(the load address), copies the
823  * contents of the program (p)buffer performing relocatations/etc, free's it
824  * when finished.
825  */
826 static int vpe_elfload(struct vpe * v)
827 {
828 	Elf_Ehdr *hdr;
829 	Elf_Shdr *sechdrs;
830 	long err = 0;
831 	char *secstrings, *strtab = NULL;
832 	unsigned int len, i, symindex = 0, strindex = 0, relocate = 0;
833 	struct module mod;	// so we can re-use the relocations code
834 
835 	memset(&mod, 0, sizeof(struct module));
836 	strcpy(mod.name, "VPE loader");
837 
838 	hdr = (Elf_Ehdr *) v->pbuffer;
839 	len = v->plen;
840 
841 	/* Sanity checks against insmoding binaries or wrong arch,
842 	   weird elf version */
843 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0
844 	    || (hdr->e_type != ET_REL && hdr->e_type != ET_EXEC)
845 	    || !elf_check_arch(hdr)
846 	    || hdr->e_shentsize != sizeof(*sechdrs)) {
847 		printk(KERN_WARNING
848 		       "VPE loader: program wrong arch or weird elf version\n");
849 
850 		return -ENOEXEC;
851 	}
852 
853 	if (hdr->e_type == ET_REL)
854 		relocate = 1;
855 
856 	if (len < hdr->e_shoff + hdr->e_shnum * sizeof(Elf_Shdr)) {
857 		printk(KERN_ERR "VPE loader: program length %u truncated\n",
858 		       len);
859 
860 		return -ENOEXEC;
861 	}
862 
863 	/* Convenience variables */
864 	sechdrs = (void *)hdr + hdr->e_shoff;
865 	secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
866 	sechdrs[0].sh_addr = 0;
867 
868 	/* And these should exist, but gcc whinges if we don't init them */
869 	symindex = strindex = 0;
870 
871 	if (relocate) {
872 		for (i = 1; i < hdr->e_shnum; i++) {
873 			if (sechdrs[i].sh_type != SHT_NOBITS
874 			    && len < sechdrs[i].sh_offset + sechdrs[i].sh_size) {
875 				printk(KERN_ERR "VPE program length %u truncated\n",
876 				       len);
877 				return -ENOEXEC;
878 			}
879 
880 			/* Mark all sections sh_addr with their address in the
881 			   temporary image. */
882 			sechdrs[i].sh_addr = (size_t) hdr + sechdrs[i].sh_offset;
883 
884 			/* Internal symbols and strings. */
885 			if (sechdrs[i].sh_type == SHT_SYMTAB) {
886 				symindex = i;
887 				strindex = sechdrs[i].sh_link;
888 				strtab = (char *)hdr + sechdrs[strindex].sh_offset;
889 			}
890 		}
891 		layout_sections(&mod, hdr, sechdrs, secstrings);
892 	}
893 
894 	v->load_addr = alloc_progmem(mod.core_size);
895 	if (!v->load_addr)
896 		return -ENOMEM;
897 
898 	pr_info("VPE loader: loading to %p\n", v->load_addr);
899 
900 	if (relocate) {
901 		for (i = 0; i < hdr->e_shnum; i++) {
902 			void *dest;
903 
904 			if (!(sechdrs[i].sh_flags & SHF_ALLOC))
905 				continue;
906 
907 			dest = v->load_addr + sechdrs[i].sh_entsize;
908 
909 			if (sechdrs[i].sh_type != SHT_NOBITS)
910 				memcpy(dest, (void *)sechdrs[i].sh_addr,
911 				       sechdrs[i].sh_size);
912 			/* Update sh_addr to point to copy in image. */
913 			sechdrs[i].sh_addr = (unsigned long)dest;
914 
915 			printk(KERN_DEBUG " section sh_name %s sh_addr 0x%x\n",
916 			       secstrings + sechdrs[i].sh_name, sechdrs[i].sh_addr);
917 		}
918 
919 		/* Fix up syms, so that st_value is a pointer to location. */
920 		simplify_symbols(sechdrs, symindex, strtab, secstrings,
921 				 hdr->e_shnum, &mod);
922 
923 		/* Now do relocations. */
924 		for (i = 1; i < hdr->e_shnum; i++) {
925 			const char *strtab = (char *)sechdrs[strindex].sh_addr;
926 			unsigned int info = sechdrs[i].sh_info;
927 
928 			/* Not a valid relocation section? */
929 			if (info >= hdr->e_shnum)
930 				continue;
931 
932 			/* Don't bother with non-allocated sections */
933 			if (!(sechdrs[info].sh_flags & SHF_ALLOC))
934 				continue;
935 
936 			if (sechdrs[i].sh_type == SHT_REL)
937 				err = apply_relocations(sechdrs, strtab, symindex, i,
938 							&mod);
939 			else if (sechdrs[i].sh_type == SHT_RELA)
940 				err = apply_relocate_add(sechdrs, strtab, symindex, i,
941 							 &mod);
942 			if (err < 0)
943 				return err;
944 
945 		}
946 	} else {
947 		struct elf_phdr *phdr = (struct elf_phdr *) ((char *)hdr + hdr->e_phoff);
948 
949 		for (i = 0; i < hdr->e_phnum; i++) {
950 			if (phdr->p_type == PT_LOAD) {
951 				memcpy((void *)phdr->p_paddr,
952 				       (char *)hdr + phdr->p_offset,
953 				       phdr->p_filesz);
954 				memset((void *)phdr->p_paddr + phdr->p_filesz,
955 				       0, phdr->p_memsz - phdr->p_filesz);
956 		    }
957 		    phdr++;
958 		}
959 
960 		for (i = 0; i < hdr->e_shnum; i++) {
961 			/* Internal symbols and strings. */
962 			if (sechdrs[i].sh_type == SHT_SYMTAB) {
963 				symindex = i;
964 				strindex = sechdrs[i].sh_link;
965 				strtab = (char *)hdr + sechdrs[strindex].sh_offset;
966 
967 				/* mark the symtab's address for when we try to find the
968 				   magic symbols */
969 				sechdrs[i].sh_addr = (size_t) hdr + sechdrs[i].sh_offset;
970 			}
971 		}
972 	}
973 
974 	/* make sure it's physically written out */
975 	flush_icache_range((unsigned long)v->load_addr,
976 			   (unsigned long)v->load_addr + v->len);
977 
978 	if ((find_vpe_symbols(v, sechdrs, symindex, strtab, &mod)) < 0) {
979 		if (v->__start == 0) {
980 			printk(KERN_WARNING "VPE loader: program does not contain "
981 			       "a __start symbol\n");
982 			return -ENOEXEC;
983 		}
984 
985 		if (v->shared_ptr == NULL)
986 			printk(KERN_WARNING "VPE loader: "
987 			       "program does not contain vpe_shared symbol.\n"
988 			       " Unable to use AMVP (AP/SP) facilities.\n");
989 	}
990 
991 	printk(" elf loaded\n");
992 	return 0;
993 }
994 
995 static void cleanup_tc(struct tc *tc)
996 {
997 	unsigned long flags;
998 	unsigned int mtflags, vpflags;
999 	int tmp;
1000 
1001 	local_irq_save(flags);
1002 	mtflags = dmt();
1003 	vpflags = dvpe();
1004 	/* Put MVPE's into 'configuration state' */
1005 	set_c0_mvpcontrol(MVPCONTROL_VPC);
1006 
1007 	settc(tc->index);
1008 	tmp = read_tc_c0_tcstatus();
1009 
1010 	/* mark not allocated and not dynamically allocatable */
1011 	tmp &= ~(TCSTATUS_A | TCSTATUS_DA);
1012 	tmp |= TCSTATUS_IXMT;	/* interrupt exempt */
1013 	write_tc_c0_tcstatus(tmp);
1014 
1015 	write_tc_c0_tchalt(TCHALT_H);
1016 	mips_ihb();
1017 
1018 	/* bind it to anything other than VPE1 */
1019 //	write_tc_c0_tcbind(read_tc_c0_tcbind() & ~TCBIND_CURVPE); // | TCBIND_CURVPE
1020 
1021 	clear_c0_mvpcontrol(MVPCONTROL_VPC);
1022 	evpe(vpflags);
1023 	emt(mtflags);
1024 	local_irq_restore(flags);
1025 }
1026 
1027 static int getcwd(char *buff, int size)
1028 {
1029 	mm_segment_t old_fs;
1030 	int ret;
1031 
1032 	old_fs = get_fs();
1033 	set_fs(KERNEL_DS);
1034 
1035 	ret = sys_getcwd(buff, size);
1036 
1037 	set_fs(old_fs);
1038 
1039 	return ret;
1040 }
1041 
1042 /* checks VPE is unused and gets ready to load program	*/
1043 static int vpe_open(struct inode *inode, struct file *filp)
1044 {
1045 	enum vpe_state state;
1046 	struct vpe_notifications *not;
1047 	struct vpe *v;
1048 	int ret;
1049 
1050 	if (minor != iminor(inode)) {
1051 		/* assume only 1 device at the moment. */
1052 		pr_warning("VPE loader: only vpe1 is supported\n");
1053 
1054 		return -ENODEV;
1055 	}
1056 
1057 	if ((v = get_vpe(tclimit)) == NULL) {
1058 		pr_warning("VPE loader: unable to get vpe\n");
1059 
1060 		return -ENODEV;
1061 	}
1062 
1063 	state = xchg(&v->state, VPE_STATE_INUSE);
1064 	if (state != VPE_STATE_UNUSED) {
1065 		printk(KERN_DEBUG "VPE loader: tc in use dumping regs\n");
1066 
1067 		list_for_each_entry(not, &v->notify, list) {
1068 			not->stop(tclimit);
1069 		}
1070 
1071 		release_progmem(v->load_addr);
1072 		cleanup_tc(get_tc(tclimit));
1073 	}
1074 
1075 	/* this of-course trashes what was there before... */
1076 	v->pbuffer = vmalloc(P_SIZE);
1077 	if (!v->pbuffer) {
1078 		pr_warning("VPE loader: unable to allocate memory\n");
1079 		return -ENOMEM;
1080 	}
1081 	v->plen = P_SIZE;
1082 	v->load_addr = NULL;
1083 	v->len = 0;
1084 
1085 	v->cwd[0] = 0;
1086 	ret = getcwd(v->cwd, VPE_PATH_MAX);
1087 	if (ret < 0)
1088 		printk(KERN_WARNING "VPE loader: open, getcwd returned %d\n", ret);
1089 
1090 	v->shared_ptr = NULL;
1091 	v->__start = 0;
1092 
1093 	return 0;
1094 }
1095 
1096 static int vpe_release(struct inode *inode, struct file *filp)
1097 {
1098 	struct vpe *v;
1099 	Elf_Ehdr *hdr;
1100 	int ret = 0;
1101 
1102 	v = get_vpe(tclimit);
1103 	if (v == NULL)
1104 		return -ENODEV;
1105 
1106 	hdr = (Elf_Ehdr *) v->pbuffer;
1107 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) == 0) {
1108 		if (vpe_elfload(v) >= 0) {
1109 			vpe_run(v);
1110 		} else {
1111 			printk(KERN_WARNING "VPE loader: ELF load failed.\n");
1112 			ret = -ENOEXEC;
1113 		}
1114 	} else {
1115 		printk(KERN_WARNING "VPE loader: only elf files are supported\n");
1116 		ret = -ENOEXEC;
1117 	}
1118 
1119 	/* It's good to be able to run the SP and if it chokes have a look at
1120 	   the /dev/rt?. But if we reset the pointer to the shared struct we
1121 	   lose what has happened. So perhaps if garbage is sent to the vpe
1122 	   device, use it as a trigger for the reset. Hopefully a nice
1123 	   executable will be along shortly. */
1124 	if (ret < 0)
1125 		v->shared_ptr = NULL;
1126 
1127 	vfree(v->pbuffer);
1128 	v->plen = 0;
1129 
1130 	return ret;
1131 }
1132 
1133 static ssize_t vpe_write(struct file *file, const char __user * buffer,
1134 			 size_t count, loff_t * ppos)
1135 {
1136 	size_t ret = count;
1137 	struct vpe *v;
1138 
1139 	if (iminor(file_inode(file)) != minor)
1140 		return -ENODEV;
1141 
1142 	v = get_vpe(tclimit);
1143 	if (v == NULL)
1144 		return -ENODEV;
1145 
1146 	if ((count + v->len) > v->plen) {
1147 		printk(KERN_WARNING
1148 		       "VPE loader: elf size too big. Perhaps strip uneeded symbols\n");
1149 		return -ENOMEM;
1150 	}
1151 
1152 	count -= copy_from_user(v->pbuffer + v->len, buffer, count);
1153 	if (!count)
1154 		return -EFAULT;
1155 
1156 	v->len += count;
1157 	return ret;
1158 }
1159 
1160 static const struct file_operations vpe_fops = {
1161 	.owner = THIS_MODULE,
1162 	.open = vpe_open,
1163 	.release = vpe_release,
1164 	.write = vpe_write,
1165 	.llseek = noop_llseek,
1166 };
1167 
1168 /* module wrapper entry points */
1169 /* give me a vpe */
1170 vpe_handle vpe_alloc(void)
1171 {
1172 	int i;
1173 	struct vpe *v;
1174 
1175 	/* find a vpe */
1176 	for (i = 1; i < MAX_VPES; i++) {
1177 		if ((v = get_vpe(i)) != NULL) {
1178 			v->state = VPE_STATE_INUSE;
1179 			return v;
1180 		}
1181 	}
1182 	return NULL;
1183 }
1184 
1185 EXPORT_SYMBOL(vpe_alloc);
1186 
1187 /* start running from here */
1188 int vpe_start(vpe_handle vpe, unsigned long start)
1189 {
1190 	struct vpe *v = vpe;
1191 
1192 	v->__start = start;
1193 	return vpe_run(v);
1194 }
1195 
1196 EXPORT_SYMBOL(vpe_start);
1197 
1198 /* halt it for now */
1199 int vpe_stop(vpe_handle vpe)
1200 {
1201 	struct vpe *v = vpe;
1202 	struct tc *t;
1203 	unsigned int evpe_flags;
1204 
1205 	evpe_flags = dvpe();
1206 
1207 	if ((t = list_entry(v->tc.next, struct tc, tc)) != NULL) {
1208 
1209 		settc(t->index);
1210 		write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~VPECONF0_VPA);
1211 	}
1212 
1213 	evpe(evpe_flags);
1214 
1215 	return 0;
1216 }
1217 
1218 EXPORT_SYMBOL(vpe_stop);
1219 
1220 /* I've done with it thank you */
1221 int vpe_free(vpe_handle vpe)
1222 {
1223 	struct vpe *v = vpe;
1224 	struct tc *t;
1225 	unsigned int evpe_flags;
1226 
1227 	if ((t = list_entry(v->tc.next, struct tc, tc)) == NULL) {
1228 		return -ENOEXEC;
1229 	}
1230 
1231 	evpe_flags = dvpe();
1232 
1233 	/* Put MVPE's into 'configuration state' */
1234 	set_c0_mvpcontrol(MVPCONTROL_VPC);
1235 
1236 	settc(t->index);
1237 	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~VPECONF0_VPA);
1238 
1239 	/* halt the TC */
1240 	write_tc_c0_tchalt(TCHALT_H);
1241 	mips_ihb();
1242 
1243 	/* mark the TC unallocated */
1244 	write_tc_c0_tcstatus(read_tc_c0_tcstatus() & ~TCSTATUS_A);
1245 
1246 	v->state = VPE_STATE_UNUSED;
1247 
1248 	clear_c0_mvpcontrol(MVPCONTROL_VPC);
1249 	evpe(evpe_flags);
1250 
1251 	return 0;
1252 }
1253 
1254 EXPORT_SYMBOL(vpe_free);
1255 
1256 void *vpe_get_shared(int index)
1257 {
1258 	struct vpe *v;
1259 
1260 	if ((v = get_vpe(index)) == NULL)
1261 		return NULL;
1262 
1263 	return v->shared_ptr;
1264 }
1265 
1266 EXPORT_SYMBOL(vpe_get_shared);
1267 
1268 int vpe_notify(int index, struct vpe_notifications *notify)
1269 {
1270 	struct vpe *v;
1271 
1272 	if ((v = get_vpe(index)) == NULL)
1273 		return -1;
1274 
1275 	list_add(&notify->list, &v->notify);
1276 	return 0;
1277 }
1278 
1279 EXPORT_SYMBOL(vpe_notify);
1280 
1281 char *vpe_getcwd(int index)
1282 {
1283 	struct vpe *v;
1284 
1285 	if ((v = get_vpe(index)) == NULL)
1286 		return NULL;
1287 
1288 	return v->cwd;
1289 }
1290 
1291 EXPORT_SYMBOL(vpe_getcwd);
1292 
1293 static ssize_t store_kill(struct device *dev, struct device_attribute *attr,
1294 			  const char *buf, size_t len)
1295 {
1296 	struct vpe *vpe = get_vpe(tclimit);
1297 	struct vpe_notifications *not;
1298 
1299 	list_for_each_entry(not, &vpe->notify, list) {
1300 		not->stop(tclimit);
1301 	}
1302 
1303 	release_progmem(vpe->load_addr);
1304 	cleanup_tc(get_tc(tclimit));
1305 	vpe_stop(vpe);
1306 	vpe_free(vpe);
1307 
1308 	return len;
1309 }
1310 static DEVICE_ATTR(kill, S_IWUSR, NULL, store_kill);
1311 
1312 static ssize_t ntcs_show(struct device *cd, struct device_attribute *attr,
1313 			 char *buf)
1314 {
1315 	struct vpe *vpe = get_vpe(tclimit);
1316 
1317 	return sprintf(buf, "%d\n", vpe->ntcs);
1318 }
1319 
1320 static ssize_t ntcs_store(struct device *dev, struct device_attribute *attr,
1321 			  const char *buf, size_t len)
1322 {
1323 	struct vpe *vpe = get_vpe(tclimit);
1324 	unsigned long new;
1325 	char *endp;
1326 
1327 	new = simple_strtoul(buf, &endp, 0);
1328 	if (endp == buf)
1329 		goto out_einval;
1330 
1331 	if (new == 0 || new > (hw_tcs - tclimit))
1332 		goto out_einval;
1333 
1334 	vpe->ntcs = new;
1335 
1336 	return len;
1337 
1338 out_einval:
1339 	return -EINVAL;
1340 }
1341 static DEVICE_ATTR_RW(ntcs);
1342 
1343 static struct attribute *vpe_attrs[] = {
1344 	&dev_attr_kill.attr,
1345 	&dev_attr_ntcs.attr,
1346 	NULL,
1347 };
1348 ATTRIBUTE_GROUPS(vpe);
1349 
1350 static void vpe_device_release(struct device *cd)
1351 {
1352 	kfree(cd);
1353 }
1354 
1355 struct class vpe_class = {
1356 	.name = "vpe",
1357 	.owner = THIS_MODULE,
1358 	.dev_release = vpe_device_release,
1359 	.dev_groups = vpe_groups,
1360 };
1361 
1362 struct device vpe_device;
1363 
1364 static int __init vpe_module_init(void)
1365 {
1366 	unsigned int mtflags, vpflags;
1367 	unsigned long flags, val;
1368 	struct vpe *v = NULL;
1369 	struct tc *t;
1370 	int tc, err;
1371 
1372 	if (!cpu_has_mipsmt) {
1373 		printk("VPE loader: not a MIPS MT capable processor\n");
1374 		return -ENODEV;
1375 	}
1376 
1377 	if (vpelimit == 0) {
1378 		printk(KERN_WARNING "No VPEs reserved for AP/SP, not "
1379 		       "initializing VPE loader.\nPass maxvpes=<n> argument as "
1380 		       "kernel argument\n");
1381 
1382 		return -ENODEV;
1383 	}
1384 
1385 	if (tclimit == 0) {
1386 		printk(KERN_WARNING "No TCs reserved for AP/SP, not "
1387 		       "initializing VPE loader.\nPass maxtcs=<n> argument as "
1388 		       "kernel argument\n");
1389 
1390 		return -ENODEV;
1391 	}
1392 
1393 	major = register_chrdev(0, module_name, &vpe_fops);
1394 	if (major < 0) {
1395 		printk("VPE loader: unable to register character device\n");
1396 		return major;
1397 	}
1398 
1399 	err = class_register(&vpe_class);
1400 	if (err) {
1401 		printk(KERN_ERR "vpe_class registration failed\n");
1402 		goto out_chrdev;
1403 	}
1404 
1405 	device_initialize(&vpe_device);
1406 	vpe_device.class	= &vpe_class,
1407 	vpe_device.parent	= NULL,
1408 	dev_set_name(&vpe_device, "vpe1");
1409 	vpe_device.devt = MKDEV(major, minor);
1410 	err = device_add(&vpe_device);
1411 	if (err) {
1412 		printk(KERN_ERR "Adding vpe_device failed\n");
1413 		goto out_class;
1414 	}
1415 
1416 	local_irq_save(flags);
1417 	mtflags = dmt();
1418 	vpflags = dvpe();
1419 
1420 	/* Put MVPE's into 'configuration state' */
1421 	set_c0_mvpcontrol(MVPCONTROL_VPC);
1422 
1423 	/* dump_mtregs(); */
1424 
1425 	val = read_c0_mvpconf0();
1426 	hw_tcs = (val & MVPCONF0_PTC) + 1;
1427 	hw_vpes = ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1;
1428 
1429 	for (tc = tclimit; tc < hw_tcs; tc++) {
1430 		/*
1431 		 * Must re-enable multithreading temporarily or in case we
1432 		 * reschedule send IPIs or similar we might hang.
1433 		 */
1434 		clear_c0_mvpcontrol(MVPCONTROL_VPC);
1435 		evpe(vpflags);
1436 		emt(mtflags);
1437 		local_irq_restore(flags);
1438 		t = alloc_tc(tc);
1439 		if (!t) {
1440 			err = -ENOMEM;
1441 			goto out;
1442 		}
1443 
1444 		local_irq_save(flags);
1445 		mtflags = dmt();
1446 		vpflags = dvpe();
1447 		set_c0_mvpcontrol(MVPCONTROL_VPC);
1448 
1449 		/* VPE's */
1450 		if (tc < hw_tcs) {
1451 			settc(tc);
1452 
1453 			if ((v = alloc_vpe(tc)) == NULL) {
1454 				printk(KERN_WARNING "VPE: unable to allocate VPE\n");
1455 
1456 				goto out_reenable;
1457 			}
1458 
1459 			v->ntcs = hw_tcs - tclimit;
1460 
1461 			/* add the tc to the list of this vpe's tc's. */
1462 			list_add(&t->tc, &v->tc);
1463 
1464 			/* deactivate all but vpe0 */
1465 			if (tc >= tclimit) {
1466 				unsigned long tmp = read_vpe_c0_vpeconf0();
1467 
1468 				tmp &= ~VPECONF0_VPA;
1469 
1470 				/* master VPE */
1471 				tmp |= VPECONF0_MVP;
1472 				write_vpe_c0_vpeconf0(tmp);
1473 			}
1474 
1475 			/* disable multi-threading with TC's */
1476 			write_vpe_c0_vpecontrol(read_vpe_c0_vpecontrol() & ~VPECONTROL_TE);
1477 
1478 			if (tc >= vpelimit) {
1479 				/*
1480 				 * Set config to be the same as vpe0,
1481 				 * particularly kseg0 coherency alg
1482 				 */
1483 				write_vpe_c0_config(read_c0_config());
1484 			}
1485 		}
1486 
1487 		/* TC's */
1488 		t->pvpe = v;	/* set the parent vpe */
1489 
1490 		if (tc >= tclimit) {
1491 			unsigned long tmp;
1492 
1493 			settc(tc);
1494 
1495 			/* Any TC that is bound to VPE0 gets left as is - in case
1496 			   we are running SMTC on VPE0. A TC that is bound to any
1497 			   other VPE gets bound to VPE0, ideally I'd like to make
1498 			   it homeless but it doesn't appear to let me bind a TC
1499 			   to a non-existent VPE. Which is perfectly reasonable.
1500 
1501 			   The (un)bound state is visible to an EJTAG probe so may
1502 			   notify GDB...
1503 			*/
1504 
1505 			if (((tmp = read_tc_c0_tcbind()) & TCBIND_CURVPE)) {
1506 				/* tc is bound >vpe0 */
1507 				write_tc_c0_tcbind(tmp & ~TCBIND_CURVPE);
1508 
1509 				t->pvpe = get_vpe(0);	/* set the parent vpe */
1510 			}
1511 
1512 			/* halt the TC */
1513 			write_tc_c0_tchalt(TCHALT_H);
1514 			mips_ihb();
1515 
1516 			tmp = read_tc_c0_tcstatus();
1517 
1518 			/* mark not activated and not dynamically allocatable */
1519 			tmp &= ~(TCSTATUS_A | TCSTATUS_DA);
1520 			tmp |= TCSTATUS_IXMT;	/* interrupt exempt */
1521 			write_tc_c0_tcstatus(tmp);
1522 		}
1523 	}
1524 
1525 out_reenable:
1526 	/* release config state */
1527 	clear_c0_mvpcontrol(MVPCONTROL_VPC);
1528 
1529 	evpe(vpflags);
1530 	emt(mtflags);
1531 	local_irq_restore(flags);
1532 
1533 	return 0;
1534 
1535 out_class:
1536 	class_unregister(&vpe_class);
1537 out_chrdev:
1538 	unregister_chrdev(major, module_name);
1539 
1540 out:
1541 	return err;
1542 }
1543 
1544 static void __exit vpe_module_exit(void)
1545 {
1546 	struct vpe *v, *n;
1547 
1548 	device_del(&vpe_device);
1549 	unregister_chrdev(major, module_name);
1550 
1551 	/* No locking needed here */
1552 	list_for_each_entry_safe(v, n, &vpecontrol.vpe_list, list) {
1553 		if (v->state != VPE_STATE_UNUSED)
1554 			release_vpe(v);
1555 	}
1556 }
1557 
1558 module_init(vpe_module_init);
1559 module_exit(vpe_module_exit);
1560 MODULE_DESCRIPTION("MIPS VPE Loader");
1561 MODULE_AUTHOR("Elizabeth Oldham, MIPS Technologies, Inc.");
1562 MODULE_LICENSE("GPL");
1563