xref: /freebsd-12.1/sys/arm/arm/elf_trampoline.c (revision 59c9a224)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2005 Olivier Houchard.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /*
28  * Since we are compiled outside of the normal kernel build process, we
29  * need to include opt_global.h manually.
30  */
31 #include "opt_global.h"
32 #include "opt_kernname.h"
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 #include <machine/asm.h>
37 #include <sys/param.h>
38 #include <sys/elf32.h>
39 #include <sys/inflate.h>
40 #include <machine/elf.h>
41 #include <machine/pte-v4.h>
42 #include <machine/cpufunc.h>
43 #include <machine/armreg.h>
44 #include <machine/cpu.h>
45 #include <machine/vmparam.h>	/* For KERNVIRTADDR */
46 
47 #if __ARM_ARCH >= 6
48 #error "elf_trampline is not supported on ARMv6/v7 platforms"
49 #endif
50 extern char kernel_start[];
51 extern char kernel_end[];
52 
53 extern void *_end;
54 
55 void _start(void);
56 void __start(void);
57 void __startC(unsigned r0, unsigned r1, unsigned r2, unsigned r3);
58 
59 extern void do_call(void *, void *, void *, int);
60 
61 #define GZ_HEAD	0xa
62 
63 #if defined(CPU_ARM9E)
64 #define cpu_idcache_wbinv_all	armv5_ec_idcache_wbinv_all
65 extern void armv5_ec_idcache_wbinv_all(void);
66 #endif
67 #if defined(SOC_MV_KIRKWOOD) || defined(SOC_MV_DISCOVERY)
68 #define cpu_l2cache_wbinv_all	sheeva_l2cache_wbinv_all
69 extern void sheeva_l2cache_wbinv_all(void);
70 #else
71 #define cpu_l2cache_wbinv_all()
72 #endif
73 
74 /*
75  * Boot parameters
76  */
77 static struct arm_boot_params s_boot_params;
78 
79 static __inline void *
memcpy(void * dst,const void * src,int len)80 memcpy(void *dst, const void *src, int len)
81 {
82 	const char *s = src;
83     	char *d = dst;
84 
85 	while (len) {
86 		if (0 && len >= 4 && !((vm_offset_t)d & 3) &&
87 		    !((vm_offset_t)s & 3)) {
88 			*(uint32_t *)d = *(uint32_t *)s;
89 			s += 4;
90 			d += 4;
91 			len -= 4;
92 		} else {
93 			*d++ = *s++;
94 			len--;
95 		}
96 	}
97 	return (dst);
98 }
99 
100 static __inline void
bzero(void * addr,int count)101 bzero(void *addr, int count)
102 {
103 	char *tmp = (char *)addr;
104 
105 	while (count > 0) {
106 		if (count >= 4 && !((vm_offset_t)tmp & 3)) {
107 			*(uint32_t *)tmp = 0;
108 			tmp += 4;
109 			count -= 4;
110 		} else {
111 			*tmp = 0;
112 			tmp++;
113 			count--;
114 		}
115 	}
116 }
117 
118 void
_startC(unsigned r0,unsigned r1,unsigned r2,unsigned r3)119 _startC(unsigned r0, unsigned r1, unsigned r2, unsigned r3)
120 {
121 	int tmp1;
122 	unsigned int sp = ((unsigned int)&_end & ~3) + 4;
123 	unsigned int pc, kernphysaddr;
124 
125 	s_boot_params.abp_r0 = r0;
126 	s_boot_params.abp_r1 = r1;
127 	s_boot_params.abp_r2 = r2;
128 	s_boot_params.abp_r3 = r3;
129 
130 	/*
131 	 * Figure out the physical address the kernel was loaded at.  This
132 	 * assumes the entry point (this code right here) is in the first page,
133 	 * which will always be the case for this trampoline code.
134 	 */
135 	__asm __volatile("mov %0, pc\n"
136 	    : "=r" (pc));
137 	kernphysaddr = pc & ~PAGE_MASK;
138 
139 #if defined(FLASHADDR) && defined(PHYSADDR) && defined(LOADERRAMADDR)
140 	if ((FLASHADDR > LOADERRAMADDR && pc >= FLASHADDR) ||
141 	    (FLASHADDR < LOADERRAMADDR && pc < LOADERRAMADDR)) {
142 		/*
143 		 * We're running from flash, so just copy the whole thing
144 		 * from flash to memory.
145 		 * This is far from optimal, we could do the relocation or
146 		 * the unzipping directly from flash to memory to avoid this
147 		 * needless copy, but it would require to know the flash
148 		 * physical address.
149 		 */
150 		unsigned int target_addr;
151 		unsigned int tmp_sp;
152 		uint32_t src_addr = (uint32_t)&_start - PHYSADDR + FLASHADDR
153 		    + (pc - FLASHADDR - ((uint32_t)&_startC - PHYSADDR)) & 0xfffff000;
154 
155 		target_addr = (unsigned int)&_start - PHYSADDR + LOADERRAMADDR;
156 		tmp_sp = target_addr + 0x100000 +
157 		    (unsigned int)&_end - (unsigned int)&_start;
158 		memcpy((char *)target_addr, (char *)src_addr,
159 		    (unsigned int)&_end - (unsigned int)&_start);
160 		/* Temporary set the sp and jump to the new location. */
161 		__asm __volatile(
162 		    "mov sp, %1\n"
163 		    "mov r0, %2\n"
164 		    "mov r1, %3\n"
165 		    "mov r2, %4\n"
166 		    "mov r3, %5\n"
167 		    "mov pc, %0\n"
168 		    : : "r" (target_addr), "r" (tmp_sp),
169 		    "r" (s_boot_params.abp_r0), "r" (s_boot_params.abp_r1),
170 		    "r" (s_boot_params.abp_r2), "r" (s_boot_params.abp_r3)
171 		    : "r0", "r1", "r2", "r3");
172 
173 	}
174 #endif
175 #ifdef KZIP
176 	sp += KERNSIZE + 0x100;
177 	sp &= ~(L1_TABLE_SIZE - 1);
178 	sp += 2 * L1_TABLE_SIZE;
179 #endif
180 	sp += 1024 * 1024; /* Should be enough for a stack */
181 
182 	__asm __volatile("adr %0, 2f\n"
183 	    		 "bic %0, %0, #0xff000000\n"
184 			 "and %1, %1, #0xff000000\n"
185 			 "orr %0, %0, %1\n"
186 			 "mrc p15, 0, %1, c1, c0, 0\n" /* CP15_SCTLR(%1)*/
187 			 "bic %1, %1, #1\n" /* Disable MMU */
188 			 "orr %1, %1, #(4 | 8)\n" /* Add DC enable,
189 						     WBUF enable */
190 			 "orr %1, %1, #0x1000\n" /* Add IC enable */
191 			 "orr %1, %1, #(0x800)\n" /* BPRD enable */
192 
193 			 "mcr p15, 0, %1, c1, c0, 0\n" /* CP15_SCTLR(%1)*/
194 			 "nop\n"
195 			 "nop\n"
196 			 "nop\n"
197 			 "mov pc, %0\n"
198 			 "2: nop\n"
199 			 "mov sp, %2\n"
200 			 : "=r" (tmp1), "+r" (kernphysaddr), "+r" (sp));
201 	__start();
202 }
203 
204 #ifdef KZIP
205 static  unsigned char *orig_input, *i_input, *i_output;
206 
207 
208 static u_int memcnt;		/* Memory allocated: blocks */
209 static size_t memtot;		/* Memory allocated: bytes */
210 /*
211  * Library functions required by inflate().
212  */
213 
214 #define MEMSIZ 0x8000
215 
216 /*
217  * Allocate memory block.
218  */
219 unsigned char *
kzipmalloc(int size)220 kzipmalloc(int size)
221 {
222 	void *ptr;
223 	static u_char mem[MEMSIZ];
224 
225 	if (memtot + size > MEMSIZ)
226 		return NULL;
227 	ptr = mem + memtot;
228 	memtot += size;
229 	memcnt++;
230 	return ptr;
231 }
232 
233 /*
234  * Free allocated memory block.
235  */
236 void
kzipfree(void * ptr)237 kzipfree(void *ptr)
238 {
239 	memcnt--;
240 	if (!memcnt)
241 		memtot = 0;
242 }
243 
244 void
putstr(char * dummy)245 putstr(char *dummy)
246 {
247 }
248 
249 static int
input(void * dummy)250 input(void *dummy)
251 {
252 	if ((size_t)(i_input - orig_input) >= KERNCOMPSIZE) {
253 		return (GZ_EOF);
254 	}
255 	return *i_input++;
256 }
257 
258 static int
output(void * dummy,unsigned char * ptr,unsigned long len)259 output(void *dummy, unsigned char *ptr, unsigned long len)
260 {
261 
262 
263 	memcpy(i_output, ptr, len);
264 	i_output += len;
265 	return (0);
266 }
267 
268 static void *
inflate_kernel(void * kernel,void * startaddr)269 inflate_kernel(void *kernel, void *startaddr)
270 {
271 	struct inflate infl;
272 	unsigned char slide[GZ_WSIZE];
273 
274 	orig_input = kernel;
275 	memcnt = memtot = 0;
276 	i_input = (unsigned char *)kernel + GZ_HEAD;
277 	if (((char *)kernel)[3] & 0x18) {
278 		while (*i_input)
279 			i_input++;
280 		i_input++;
281 	}
282 	i_output = startaddr;
283 	bzero(&infl, sizeof(infl));
284 	infl.gz_input = input;
285 	infl.gz_output = output;
286 	infl.gz_slide = slide;
287 	inflate(&infl);
288 	return ((char *)(((vm_offset_t)i_output & ~3) + 4));
289 }
290 
291 #endif
292 
293 void *
load_kernel(unsigned int kstart,unsigned int curaddr,unsigned int func_end,int d)294 load_kernel(unsigned int kstart, unsigned int curaddr,unsigned int func_end,
295     int d)
296 {
297 	Elf32_Ehdr *eh;
298 	Elf32_Phdr phdr[64] /* XXX */, *php;
299 	Elf32_Shdr shdr[64] /* XXX */;
300 	int i,j;
301 	void *entry_point;
302 	int symtabindex = -1;
303 	int symstrindex = -1;
304 	vm_offset_t lastaddr = 0;
305 	Elf_Addr ssym = 0;
306 	Elf_Dyn *dp;
307 	struct arm_boot_params local_boot_params;
308 
309 	eh = (Elf32_Ehdr *)kstart;
310 	ssym = 0;
311 	entry_point = (void*)eh->e_entry;
312 	memcpy(phdr, (void *)(kstart + eh->e_phoff ),
313 	    eh->e_phnum * sizeof(phdr[0]));
314 
315 	/* Determine lastaddr. */
316 	for (i = 0; i < eh->e_phnum; i++) {
317 		if (lastaddr < (phdr[i].p_vaddr - KERNVIRTADDR + curaddr
318 		    + phdr[i].p_memsz))
319 			lastaddr = phdr[i].p_vaddr - KERNVIRTADDR +
320 			    curaddr + phdr[i].p_memsz;
321 	}
322 
323 	/* Save the symbol tables, as there're about to be scratched. */
324 	memcpy(shdr, (void *)(kstart + eh->e_shoff),
325 	    sizeof(*shdr) * eh->e_shnum);
326 	if (eh->e_shnum * eh->e_shentsize != 0 &&
327 	    eh->e_shoff != 0) {
328 		for (i = 0; i < eh->e_shnum; i++) {
329 			if (shdr[i].sh_type == SHT_SYMTAB) {
330 				for (j = 0; j < eh->e_phnum; j++) {
331 					if (phdr[j].p_type == PT_LOAD &&
332 					    shdr[i].sh_offset >=
333 					    phdr[j].p_offset &&
334 					    (shdr[i].sh_offset +
335 					     shdr[i].sh_size <=
336 					     phdr[j].p_offset +
337 					     phdr[j].p_filesz)) {
338 						shdr[i].sh_offset = 0;
339 						shdr[i].sh_size = 0;
340 						j = eh->e_phnum;
341 					}
342 				}
343 				if (shdr[i].sh_offset != 0 &&
344 				    shdr[i].sh_size != 0) {
345 					symtabindex = i;
346 					symstrindex = shdr[i].sh_link;
347 				}
348 			}
349 		}
350 		func_end = roundup(func_end, sizeof(long));
351 		if (symtabindex >= 0 && symstrindex >= 0) {
352 			ssym = lastaddr;
353 			if (d) {
354 				memcpy((void *)func_end, (void *)(
355 				    shdr[symtabindex].sh_offset + kstart),
356 				    shdr[symtabindex].sh_size);
357 				memcpy((void *)(func_end +
358 				    shdr[symtabindex].sh_size),
359 				    (void *)(shdr[symstrindex].sh_offset +
360 				    kstart), shdr[symstrindex].sh_size);
361 			} else {
362 				lastaddr += shdr[symtabindex].sh_size;
363 				lastaddr = roundup(lastaddr,
364 				    sizeof(shdr[symtabindex].sh_size));
365 				lastaddr += sizeof(shdr[symstrindex].sh_size);
366 				lastaddr += shdr[symstrindex].sh_size;
367 				lastaddr = roundup(lastaddr,
368 				    sizeof(shdr[symstrindex].sh_size));
369 			}
370 
371 		}
372 	}
373 	if (!d)
374 		return ((void *)lastaddr);
375 
376 	/*
377 	 * Now the stack is fixed, copy boot params
378 	 * before it's overrided
379 	 */
380 	memcpy(&local_boot_params, &s_boot_params, sizeof(local_boot_params));
381 
382 	j = eh->e_phnum;
383 	for (i = 0; i < j; i++) {
384 		volatile char c;
385 
386 		if (phdr[i].p_type != PT_LOAD)
387 			continue;
388 		memcpy((void *)(phdr[i].p_vaddr - KERNVIRTADDR + curaddr),
389 		    (void*)(kstart + phdr[i].p_offset), phdr[i].p_filesz);
390 		/* Clean space from oversized segments, eg: bss. */
391 		if (phdr[i].p_filesz < phdr[i].p_memsz)
392 			bzero((void *)(phdr[i].p_vaddr - KERNVIRTADDR +
393 			    curaddr + phdr[i].p_filesz), phdr[i].p_memsz -
394 			    phdr[i].p_filesz);
395 	}
396 	/* Now grab the symbol tables. */
397 	if (symtabindex >= 0 && symstrindex >= 0) {
398 		*(Elf_Size *)lastaddr =
399 		    shdr[symtabindex].sh_size;
400 		lastaddr += sizeof(shdr[symtabindex].sh_size);
401 		memcpy((void*)lastaddr,
402 		    (void *)func_end,
403 		    shdr[symtabindex].sh_size);
404 		lastaddr += shdr[symtabindex].sh_size;
405 		lastaddr = roundup(lastaddr,
406 		    sizeof(shdr[symtabindex].sh_size));
407 		*(Elf_Size *)lastaddr =
408 		    shdr[symstrindex].sh_size;
409 		lastaddr += sizeof(shdr[symstrindex].sh_size);
410 		memcpy((void*)lastaddr,
411 		    (void*)(func_end +
412 			    shdr[symtabindex].sh_size),
413 		    shdr[symstrindex].sh_size);
414 		lastaddr += shdr[symstrindex].sh_size;
415 		lastaddr = roundup(lastaddr,
416    		    sizeof(shdr[symstrindex].sh_size));
417 		*(Elf_Addr *)curaddr = MAGIC_TRAMP_NUMBER;
418 		*((Elf_Addr *)curaddr + 1) = ssym - curaddr + KERNVIRTADDR;
419 		*((Elf_Addr *)curaddr + 2) = lastaddr - curaddr + KERNVIRTADDR;
420 	} else
421 		*(Elf_Addr *)curaddr = 0;
422 	/* Invalidate the instruction cache. */
423 	__asm __volatile("mcr p15, 0, %0, c7, c5, 0\n"
424 	    		 "mcr p15, 0, %0, c7, c10, 4\n"
425 			 : : "r" (curaddr));
426 	__asm __volatile("mrc p15, 0, %0, c1, c0, 0\n" /* CP15_SCTLR(%0)*/
427 	    "bic %0, %0, #1\n" /* MMU_ENABLE */
428 	    "mcr p15, 0, %0, c1, c0, 0\n" /* CP15_SCTLR(%0)*/
429 	    : "=r" (ssym));
430 	/* Jump to the entry point. */
431 	((void(*)(unsigned, unsigned, unsigned, unsigned))
432 	(entry_point - KERNVIRTADDR + curaddr))
433 	(local_boot_params.abp_r0, local_boot_params.abp_r1,
434 	local_boot_params.abp_r2, local_boot_params.abp_r3);
435 	__asm __volatile(".globl func_end\n"
436 	    "func_end:");
437 
438 	/* NOTREACHED */
439 	return NULL;
440 }
441 
442 extern char func_end[];
443 
444 
445 #define PMAP_DOMAIN_KERNEL	0 /*
446 				    * Just define it instead of including the
447 				    * whole VM headers set.
448 				    */
449 int __hack;
450 static __inline void
setup_pagetables(unsigned int pt_addr,vm_paddr_t physstart,vm_paddr_t physend,int write_back)451 setup_pagetables(unsigned int pt_addr, vm_paddr_t physstart, vm_paddr_t physend,
452     int write_back)
453 {
454 	unsigned int *pd = (unsigned int *)pt_addr;
455 	vm_paddr_t addr;
456 	int domain = (DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2)) | DOMAIN_CLIENT;
457 	int tmp;
458 
459 	bzero(pd, L1_TABLE_SIZE);
460 	for (addr = physstart; addr < physend; addr += L1_S_SIZE) {
461 		pd[addr >> L1_S_SHIFT] = L1_TYPE_S|L1_S_C|L1_S_AP(AP_KRW)|
462 		    L1_S_DOM(PMAP_DOMAIN_KERNEL) | addr;
463 		if (write_back && 0)
464 			pd[addr >> L1_S_SHIFT] |= L1_S_B;
465 	}
466 	/* XXX: See below */
467 	if (0xfff00000 < physstart || 0xfff00000 > physend)
468 		pd[0xfff00000 >> L1_S_SHIFT] = L1_TYPE_S|L1_S_AP(AP_KRW)|
469 		    L1_S_DOM(PMAP_DOMAIN_KERNEL)|physstart;
470 	__asm __volatile("mcr p15, 0, %1, c2, c0, 0\n" /* set TTB */
471 	    		 "mcr p15, 0, %1, c8, c7, 0\n" /* Flush TTB */
472 			 "mcr p15, 0, %2, c3, c0, 0\n" /* Set DAR */
473 			 "mrc p15, 0, %0, c1, c0, 0\n" /* CP15_SCTLR(%0)*/
474 			 "orr %0, %0, #1\n" /* MMU_ENABLE */
475 			 "mcr p15, 0, %0, c1, c0, 0\n" /* CP15_SCTLR(%0)*/
476 			 "mrc p15, 0, %0, c2, c0, 0\n" /* CPWAIT */
477 			 "mov r0, r0\n"
478 			 "sub pc, pc, #4\n" :
479 			 "=r" (tmp) : "r" (pd), "r" (domain));
480 
481 	/*
482 	 * XXX: This is the most stupid workaround I've ever wrote.
483 	 * For some reason, the KB9202 won't boot the kernel unless
484 	 * we access an address which is not in the
485 	 * 0x20000000 - 0x20ffffff range. I hope I'll understand
486 	 * what's going on later.
487 	 */
488 	__hack = *(volatile int *)0xfffff21c;
489 }
490 
491 void
__start(void)492 __start(void)
493 {
494 	void *curaddr;
495 	void *dst, *altdst;
496 	char *kernel = (char *)&kernel_start;
497 	int sp;
498 	int pt_addr;
499 
500 	__asm __volatile("mov %0, pc"  :
501 	    "=r" (curaddr));
502 	curaddr = (void*)((unsigned int)curaddr & 0xfff00000);
503 #ifdef KZIP
504 	if (*kernel == 0x1f && kernel[1] == 0x8b) {
505 		pt_addr = L1_TABLE_SIZE +
506 		    rounddown2((int)&_end + KERNSIZE + 0x100, L1_TABLE_SIZE);
507 
508 		setup_pagetables(pt_addr, (vm_paddr_t)curaddr,
509 		    (vm_paddr_t)curaddr + 0x10000000, 1);
510 		/* Gzipped kernel */
511 		dst = inflate_kernel(kernel, &_end);
512 		kernel = (char *)&_end;
513 		altdst = 4 + load_kernel((unsigned int)kernel,
514 		    (unsigned int)curaddr,
515 		    (unsigned int)&func_end + 800 , 0);
516 		if (altdst > dst)
517 			dst = altdst;
518 
519 		/*
520 		 * Disable MMU.  Otherwise, setup_pagetables call below
521 		 * might overwrite the L1 table we are currently using.
522 		 */
523 		cpu_idcache_wbinv_all();
524 		cpu_l2cache_wbinv_all();
525 		__asm __volatile("mrc p15, 0, %0, c1, c0, 0\n" /* CP15_SCTLR(%0)*/
526 		  "bic %0, %0, #1\n" /* MMU_DISABLE */
527 		  "mcr p15, 0, %0, c1, c0, 0\n" /* CP15_SCTLR(%0)*/
528 		  :"=r" (pt_addr));
529 	} else
530 #endif
531 		dst = 4 + load_kernel((unsigned int)&kernel_start,
532 	    (unsigned int)curaddr,
533 	    (unsigned int)&func_end, 0);
534 	dst = (void *)(((vm_offset_t)dst & ~3));
535 	pt_addr = L1_TABLE_SIZE + rounddown2((unsigned int)dst, L1_TABLE_SIZE);
536 	setup_pagetables(pt_addr, (vm_paddr_t)curaddr,
537 	    (vm_paddr_t)curaddr + 0x10000000, 0);
538 	sp = pt_addr + L1_TABLE_SIZE + 8192;
539 	sp = sp &~3;
540 	dst = (void *)(sp + 4);
541 	memcpy((void *)dst, (void *)&load_kernel, (unsigned int)&func_end -
542 	    (unsigned int)&load_kernel + 800);
543 	do_call(dst, kernel, dst + (unsigned int)(&func_end) -
544 	    (unsigned int)(&load_kernel) + 800, sp);
545 }
546 
547 /* We need to provide these functions but never call them */
548 void __aeabi_unwind_cpp_pr0(void);
549 void __aeabi_unwind_cpp_pr1(void);
550 void __aeabi_unwind_cpp_pr2(void);
551 
552 __strong_reference(__aeabi_unwind_cpp_pr0, __aeabi_unwind_cpp_pr1);
553 __strong_reference(__aeabi_unwind_cpp_pr0, __aeabi_unwind_cpp_pr2);
554 void
__aeabi_unwind_cpp_pr0(void)555 __aeabi_unwind_cpp_pr0(void)
556 {
557 }
558