xref: /openbmc/linux/arch/powerpc/kernel/vdso.c (revision 4f3db074)
1 
2 /*
3  *    Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4  *			 <benh@kernel.crashing.org>
5  *
6  *  This program is free software; you can redistribute it and/or
7  *  modify it under the terms of the GNU General Public License
8  *  as published by the Free Software Foundation; either version
9  *  2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/errno.h>
13 #include <linux/sched.h>
14 #include <linux/kernel.h>
15 #include <linux/mm.h>
16 #include <linux/smp.h>
17 #include <linux/stddef.h>
18 #include <linux/unistd.h>
19 #include <linux/slab.h>
20 #include <linux/user.h>
21 #include <linux/elf.h>
22 #include <linux/security.h>
23 #include <linux/memblock.h>
24 
25 #include <asm/pgtable.h>
26 #include <asm/processor.h>
27 #include <asm/mmu.h>
28 #include <asm/mmu_context.h>
29 #include <asm/prom.h>
30 #include <asm/machdep.h>
31 #include <asm/cputable.h>
32 #include <asm/sections.h>
33 #include <asm/firmware.h>
34 #include <asm/vdso.h>
35 #include <asm/vdso_datapage.h>
36 #include <asm/setup.h>
37 
38 #undef DEBUG
39 
40 #ifdef DEBUG
41 #define DBG(fmt...) printk(fmt)
42 #else
43 #define DBG(fmt...)
44 #endif
45 
46 /* Max supported size for symbol names */
47 #define MAX_SYMNAME	64
48 
49 /* The alignment of the vDSO */
50 #define VDSO_ALIGNMENT	(1 << 16)
51 
52 extern char vdso32_start, vdso32_end;
53 static void *vdso32_kbase = &vdso32_start;
54 static unsigned int vdso32_pages;
55 static struct page **vdso32_pagelist;
56 unsigned long vdso32_sigtramp;
57 unsigned long vdso32_rt_sigtramp;
58 
59 #ifdef CONFIG_PPC64
60 extern char vdso64_start, vdso64_end;
61 static void *vdso64_kbase = &vdso64_start;
62 static unsigned int vdso64_pages;
63 static struct page **vdso64_pagelist;
64 unsigned long vdso64_rt_sigtramp;
65 #endif /* CONFIG_PPC64 */
66 
67 static int vdso_ready;
68 
69 /*
70  * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
71  * Once the early boot kernel code no longer needs to muck around
72  * with it, it will become dynamically allocated
73  */
74 static union {
75 	struct vdso_data	data;
76 	u8			page[PAGE_SIZE];
77 } vdso_data_store __page_aligned_data;
78 struct vdso_data *vdso_data = &vdso_data_store.data;
79 
80 /* Format of the patch table */
81 struct vdso_patch_def
82 {
83 	unsigned long	ftr_mask, ftr_value;
84 	const char	*gen_name;
85 	const char	*fix_name;
86 };
87 
88 /* Table of functions to patch based on the CPU type/revision
89  *
90  * Currently, we only change sync_dicache to do nothing on processors
91  * with a coherent icache
92  */
93 static struct vdso_patch_def vdso_patches[] = {
94 	{
95 		CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
96 		"__kernel_sync_dicache", "__kernel_sync_dicache_p5"
97 	},
98 	{
99 		CPU_FTR_USE_TB, 0,
100 		"__kernel_gettimeofday", NULL
101 	},
102 	{
103 		CPU_FTR_USE_TB, 0,
104 		"__kernel_clock_gettime", NULL
105 	},
106 	{
107 		CPU_FTR_USE_TB, 0,
108 		"__kernel_clock_getres", NULL
109 	},
110 	{
111 		CPU_FTR_USE_TB, 0,
112 		"__kernel_get_tbfreq", NULL
113 	},
114 	{
115 		CPU_FTR_USE_TB, 0,
116 		"__kernel_time", NULL
117 	},
118 };
119 
120 /*
121  * Some infos carried around for each of them during parsing at
122  * boot time.
123  */
124 struct lib32_elfinfo
125 {
126 	Elf32_Ehdr	*hdr;		/* ptr to ELF */
127 	Elf32_Sym	*dynsym;	/* ptr to .dynsym section */
128 	unsigned long	dynsymsize;	/* size of .dynsym section */
129 	char		*dynstr;	/* ptr to .dynstr section */
130 	unsigned long	text;		/* offset of .text section in .so */
131 };
132 
133 struct lib64_elfinfo
134 {
135 	Elf64_Ehdr	*hdr;
136 	Elf64_Sym	*dynsym;
137 	unsigned long	dynsymsize;
138 	char		*dynstr;
139 	unsigned long	text;
140 };
141 
142 
143 #ifdef __DEBUG
144 static void dump_one_vdso_page(struct page *pg, struct page *upg)
145 {
146 	printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
147 	       page_count(pg),
148 	       pg->flags);
149 	if (upg && !IS_ERR(upg) /* && pg != upg*/) {
150 		printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
151 						       << PAGE_SHIFT),
152 		       page_count(upg),
153 		       upg->flags);
154 	}
155 	printk("\n");
156 }
157 
158 static void dump_vdso_pages(struct vm_area_struct * vma)
159 {
160 	int i;
161 
162 	if (!vma || is_32bit_task()) {
163 		printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
164 		for (i=0; i<vdso32_pages; i++) {
165 			struct page *pg = virt_to_page(vdso32_kbase +
166 						       i*PAGE_SIZE);
167 			struct page *upg = (vma && vma->vm_mm) ?
168 				follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
169 				: NULL;
170 			dump_one_vdso_page(pg, upg);
171 		}
172 	}
173 	if (!vma || !is_32bit_task()) {
174 		printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
175 		for (i=0; i<vdso64_pages; i++) {
176 			struct page *pg = virt_to_page(vdso64_kbase +
177 						       i*PAGE_SIZE);
178 			struct page *upg = (vma && vma->vm_mm) ?
179 				follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
180 				: NULL;
181 			dump_one_vdso_page(pg, upg);
182 		}
183 	}
184 }
185 #endif /* DEBUG */
186 
187 /*
188  * This is called from binfmt_elf, we create the special vma for the
189  * vDSO and insert it into the mm struct tree
190  */
191 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
192 {
193 	struct mm_struct *mm = current->mm;
194 	struct page **vdso_pagelist;
195 	unsigned long vdso_pages;
196 	unsigned long vdso_base;
197 	int rc;
198 
199 	if (!vdso_ready)
200 		return 0;
201 
202 #ifdef CONFIG_PPC64
203 	if (is_32bit_task()) {
204 		vdso_pagelist = vdso32_pagelist;
205 		vdso_pages = vdso32_pages;
206 		vdso_base = VDSO32_MBASE;
207 	} else {
208 		vdso_pagelist = vdso64_pagelist;
209 		vdso_pages = vdso64_pages;
210 		/*
211 		 * On 64bit we don't have a preferred map address. This
212 		 * allows get_unmapped_area to find an area near other mmaps
213 		 * and most likely share a SLB entry.
214 		 */
215 		vdso_base = 0;
216 	}
217 #else
218 	vdso_pagelist = vdso32_pagelist;
219 	vdso_pages = vdso32_pages;
220 	vdso_base = VDSO32_MBASE;
221 #endif
222 
223 	current->mm->context.vdso_base = 0;
224 
225 	/* vDSO has a problem and was disabled, just don't "enable" it for the
226 	 * process
227 	 */
228 	if (vdso_pages == 0)
229 		return 0;
230 	/* Add a page to the vdso size for the data page */
231 	vdso_pages ++;
232 
233 	/*
234 	 * pick a base address for the vDSO in process space. We try to put it
235 	 * at vdso_base which is the "natural" base for it, but we might fail
236 	 * and end up putting it elsewhere.
237 	 * Add enough to the size so that the result can be aligned.
238 	 */
239 	down_write(&mm->mmap_sem);
240 	vdso_base = get_unmapped_area(NULL, vdso_base,
241 				      (vdso_pages << PAGE_SHIFT) +
242 				      ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
243 				      0, 0);
244 	if (IS_ERR_VALUE(vdso_base)) {
245 		rc = vdso_base;
246 		goto fail_mmapsem;
247 	}
248 
249 	/* Add required alignment. */
250 	vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
251 
252 	/*
253 	 * Put vDSO base into mm struct. We need to do this before calling
254 	 * install_special_mapping or the perf counter mmap tracking code
255 	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
256 	 */
257 	current->mm->context.vdso_base = vdso_base;
258 
259 	/*
260 	 * our vma flags don't have VM_WRITE so by default, the process isn't
261 	 * allowed to write those pages.
262 	 * gdb can break that with ptrace interface, and thus trigger COW on
263 	 * those pages but it's then your responsibility to never do that on
264 	 * the "data" page of the vDSO or you'll stop getting kernel updates
265 	 * and your nice userland gettimeofday will be totally dead.
266 	 * It's fine to use that for setting breakpoints in the vDSO code
267 	 * pages though.
268 	 */
269 	rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
270 				     VM_READ|VM_EXEC|
271 				     VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
272 				     vdso_pagelist);
273 	if (rc) {
274 		current->mm->context.vdso_base = 0;
275 		goto fail_mmapsem;
276 	}
277 
278 	up_write(&mm->mmap_sem);
279 	return 0;
280 
281  fail_mmapsem:
282 	up_write(&mm->mmap_sem);
283 	return rc;
284 }
285 
286 const char *arch_vma_name(struct vm_area_struct *vma)
287 {
288 	if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
289 		return "[vdso]";
290 	return NULL;
291 }
292 
293 
294 
295 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
296 				  unsigned long *size)
297 {
298 	Elf32_Shdr *sechdrs;
299 	unsigned int i;
300 	char *secnames;
301 
302 	/* Grab section headers and strings so we can tell who is who */
303 	sechdrs = (void *)ehdr + ehdr->e_shoff;
304 	secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
305 
306 	/* Find the section they want */
307 	for (i = 1; i < ehdr->e_shnum; i++) {
308 		if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
309 			if (size)
310 				*size = sechdrs[i].sh_size;
311 			return (void *)ehdr + sechdrs[i].sh_offset;
312 		}
313 	}
314 	*size = 0;
315 	return NULL;
316 }
317 
318 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
319 					const char *symname)
320 {
321 	unsigned int i;
322 	char name[MAX_SYMNAME], *c;
323 
324 	for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
325 		if (lib->dynsym[i].st_name == 0)
326 			continue;
327 		strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
328 			MAX_SYMNAME);
329 		c = strchr(name, '@');
330 		if (c)
331 			*c = 0;
332 		if (strcmp(symname, name) == 0)
333 			return &lib->dynsym[i];
334 	}
335 	return NULL;
336 }
337 
338 /* Note that we assume the section is .text and the symbol is relative to
339  * the library base
340  */
341 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
342 					    const char *symname)
343 {
344 	Elf32_Sym *sym = find_symbol32(lib, symname);
345 
346 	if (sym == NULL) {
347 		printk(KERN_WARNING "vDSO32: function %s not found !\n",
348 		       symname);
349 		return 0;
350 	}
351 	return sym->st_value - VDSO32_LBASE;
352 }
353 
354 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
355 				       struct lib64_elfinfo *v64,
356 				       const char *orig, const char *fix)
357 {
358 	Elf32_Sym *sym32_gen, *sym32_fix;
359 
360 	sym32_gen = find_symbol32(v32, orig);
361 	if (sym32_gen == NULL) {
362 		printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
363 		return -1;
364 	}
365 	if (fix == NULL) {
366 		sym32_gen->st_name = 0;
367 		return 0;
368 	}
369 	sym32_fix = find_symbol32(v32, fix);
370 	if (sym32_fix == NULL) {
371 		printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
372 		return -1;
373 	}
374 	sym32_gen->st_value = sym32_fix->st_value;
375 	sym32_gen->st_size = sym32_fix->st_size;
376 	sym32_gen->st_info = sym32_fix->st_info;
377 	sym32_gen->st_other = sym32_fix->st_other;
378 	sym32_gen->st_shndx = sym32_fix->st_shndx;
379 
380 	return 0;
381 }
382 
383 
384 #ifdef CONFIG_PPC64
385 
386 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
387 				  unsigned long *size)
388 {
389 	Elf64_Shdr *sechdrs;
390 	unsigned int i;
391 	char *secnames;
392 
393 	/* Grab section headers and strings so we can tell who is who */
394 	sechdrs = (void *)ehdr + ehdr->e_shoff;
395 	secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
396 
397 	/* Find the section they want */
398 	for (i = 1; i < ehdr->e_shnum; i++) {
399 		if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
400 			if (size)
401 				*size = sechdrs[i].sh_size;
402 			return (void *)ehdr + sechdrs[i].sh_offset;
403 		}
404 	}
405 	if (size)
406 		*size = 0;
407 	return NULL;
408 }
409 
410 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
411 					const char *symname)
412 {
413 	unsigned int i;
414 	char name[MAX_SYMNAME], *c;
415 
416 	for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
417 		if (lib->dynsym[i].st_name == 0)
418 			continue;
419 		strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
420 			MAX_SYMNAME);
421 		c = strchr(name, '@');
422 		if (c)
423 			*c = 0;
424 		if (strcmp(symname, name) == 0)
425 			return &lib->dynsym[i];
426 	}
427 	return NULL;
428 }
429 
430 /* Note that we assume the section is .text and the symbol is relative to
431  * the library base
432  */
433 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
434 					    const char *symname)
435 {
436 	Elf64_Sym *sym = find_symbol64(lib, symname);
437 
438 	if (sym == NULL) {
439 		printk(KERN_WARNING "vDSO64: function %s not found !\n",
440 		       symname);
441 		return 0;
442 	}
443 #ifdef VDS64_HAS_DESCRIPTORS
444 	return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
445 		VDSO64_LBASE;
446 #else
447 	return sym->st_value - VDSO64_LBASE;
448 #endif
449 }
450 
451 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
452 				       struct lib64_elfinfo *v64,
453 				       const char *orig, const char *fix)
454 {
455 	Elf64_Sym *sym64_gen, *sym64_fix;
456 
457 	sym64_gen = find_symbol64(v64, orig);
458 	if (sym64_gen == NULL) {
459 		printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
460 		return -1;
461 	}
462 	if (fix == NULL) {
463 		sym64_gen->st_name = 0;
464 		return 0;
465 	}
466 	sym64_fix = find_symbol64(v64, fix);
467 	if (sym64_fix == NULL) {
468 		printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
469 		return -1;
470 	}
471 	sym64_gen->st_value = sym64_fix->st_value;
472 	sym64_gen->st_size = sym64_fix->st_size;
473 	sym64_gen->st_info = sym64_fix->st_info;
474 	sym64_gen->st_other = sym64_fix->st_other;
475 	sym64_gen->st_shndx = sym64_fix->st_shndx;
476 
477 	return 0;
478 }
479 
480 #endif /* CONFIG_PPC64 */
481 
482 
483 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
484 					struct lib64_elfinfo *v64)
485 {
486 	void *sect;
487 
488 	/*
489 	 * Locate symbol tables & text section
490 	 */
491 
492 	v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
493 	v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
494 	if (v32->dynsym == NULL || v32->dynstr == NULL) {
495 		printk(KERN_ERR "vDSO32: required symbol section not found\n");
496 		return -1;
497 	}
498 	sect = find_section32(v32->hdr, ".text", NULL);
499 	if (sect == NULL) {
500 		printk(KERN_ERR "vDSO32: the .text section was not found\n");
501 		return -1;
502 	}
503 	v32->text = sect - vdso32_kbase;
504 
505 #ifdef CONFIG_PPC64
506 	v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
507 	v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
508 	if (v64->dynsym == NULL || v64->dynstr == NULL) {
509 		printk(KERN_ERR "vDSO64: required symbol section not found\n");
510 		return -1;
511 	}
512 	sect = find_section64(v64->hdr, ".text", NULL);
513 	if (sect == NULL) {
514 		printk(KERN_ERR "vDSO64: the .text section was not found\n");
515 		return -1;
516 	}
517 	v64->text = sect - vdso64_kbase;
518 #endif /* CONFIG_PPC64 */
519 
520 	return 0;
521 }
522 
523 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
524 					  struct lib64_elfinfo *v64)
525 {
526 	/*
527 	 * Find signal trampolines
528 	 */
529 
530 #ifdef CONFIG_PPC64
531 	vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
532 #endif
533 	vdso32_sigtramp	   = find_function32(v32, "__kernel_sigtramp32");
534 	vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
535 }
536 
537 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
538 				       struct lib64_elfinfo *v64)
539 {
540 	Elf32_Sym *sym32;
541 #ifdef CONFIG_PPC64
542 	Elf64_Sym *sym64;
543 
544        	sym64 = find_symbol64(v64, "__kernel_datapage_offset");
545 	if (sym64 == NULL) {
546 		printk(KERN_ERR "vDSO64: Can't find symbol "
547 		       "__kernel_datapage_offset !\n");
548 		return -1;
549 	}
550 	*((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
551 		(vdso64_pages << PAGE_SHIFT) -
552 		(sym64->st_value - VDSO64_LBASE);
553 #endif /* CONFIG_PPC64 */
554 
555 	sym32 = find_symbol32(v32, "__kernel_datapage_offset");
556 	if (sym32 == NULL) {
557 		printk(KERN_ERR "vDSO32: Can't find symbol "
558 		       "__kernel_datapage_offset !\n");
559 		return -1;
560 	}
561 	*((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
562 		(vdso32_pages << PAGE_SHIFT) -
563 		(sym32->st_value - VDSO32_LBASE);
564 
565 	return 0;
566 }
567 
568 
569 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
570 				      struct lib64_elfinfo *v64)
571 {
572 	void *start32;
573 	unsigned long size32;
574 
575 #ifdef CONFIG_PPC64
576 	void *start64;
577 	unsigned long size64;
578 
579 	start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
580 	if (start64)
581 		do_feature_fixups(cur_cpu_spec->cpu_features,
582 				  start64, start64 + size64);
583 
584 	start64 = find_section64(v64->hdr, "__mmu_ftr_fixup", &size64);
585 	if (start64)
586 		do_feature_fixups(cur_cpu_spec->mmu_features,
587 				  start64, start64 + size64);
588 
589 	start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
590 	if (start64)
591 		do_feature_fixups(powerpc_firmware_features,
592 				  start64, start64 + size64);
593 
594 	start64 = find_section64(v64->hdr, "__lwsync_fixup", &size64);
595 	if (start64)
596 		do_lwsync_fixups(cur_cpu_spec->cpu_features,
597 				 start64, start64 + size64);
598 #endif /* CONFIG_PPC64 */
599 
600 	start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
601 	if (start32)
602 		do_feature_fixups(cur_cpu_spec->cpu_features,
603 				  start32, start32 + size32);
604 
605 	start32 = find_section32(v32->hdr, "__mmu_ftr_fixup", &size32);
606 	if (start32)
607 		do_feature_fixups(cur_cpu_spec->mmu_features,
608 				  start32, start32 + size32);
609 
610 #ifdef CONFIG_PPC64
611 	start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
612 	if (start32)
613 		do_feature_fixups(powerpc_firmware_features,
614 				  start32, start32 + size32);
615 #endif /* CONFIG_PPC64 */
616 
617 	start32 = find_section32(v32->hdr, "__lwsync_fixup", &size32);
618 	if (start32)
619 		do_lwsync_fixups(cur_cpu_spec->cpu_features,
620 				 start32, start32 + size32);
621 
622 	return 0;
623 }
624 
625 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
626 				       struct lib64_elfinfo *v64)
627 {
628 	int i;
629 
630 	for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
631 		struct vdso_patch_def *patch = &vdso_patches[i];
632 		int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
633 			== patch->ftr_value;
634 		if (!match)
635 			continue;
636 
637 		DBG("replacing %s with %s...\n", patch->gen_name,
638 		    patch->fix_name ? "NONE" : patch->fix_name);
639 
640 		/*
641 		 * Patch the 32 bits and 64 bits symbols. Note that we do not
642 		 * patch the "." symbol on 64 bits.
643 		 * It would be easy to do, but doesn't seem to be necessary,
644 		 * patching the OPD symbol is enough.
645 		 */
646 		vdso_do_func_patch32(v32, v64, patch->gen_name,
647 				     patch->fix_name);
648 #ifdef CONFIG_PPC64
649 		vdso_do_func_patch64(v32, v64, patch->gen_name,
650 				     patch->fix_name);
651 #endif /* CONFIG_PPC64 */
652 	}
653 
654 	return 0;
655 }
656 
657 
658 static __init int vdso_setup(void)
659 {
660 	struct lib32_elfinfo	v32;
661 	struct lib64_elfinfo	v64;
662 
663 	v32.hdr = vdso32_kbase;
664 #ifdef CONFIG_PPC64
665 	v64.hdr = vdso64_kbase;
666 #endif
667 	if (vdso_do_find_sections(&v32, &v64))
668 		return -1;
669 
670 	if (vdso_fixup_datapage(&v32, &v64))
671 		return -1;
672 
673 	if (vdso_fixup_features(&v32, &v64))
674 		return -1;
675 
676 	if (vdso_fixup_alt_funcs(&v32, &v64))
677 		return -1;
678 
679 	vdso_setup_trampolines(&v32, &v64);
680 
681 	return 0;
682 }
683 
684 /*
685  * Called from setup_arch to initialize the bitmap of available
686  * syscalls in the systemcfg page
687  */
688 static void __init vdso_setup_syscall_map(void)
689 {
690 	unsigned int i;
691 	extern unsigned long *sys_call_table;
692 	extern unsigned long sys_ni_syscall;
693 
694 
695 	for (i = 0; i < __NR_syscalls; i++) {
696 #ifdef CONFIG_PPC64
697 		if (sys_call_table[i*2] != sys_ni_syscall)
698 			vdso_data->syscall_map_64[i >> 5] |=
699 				0x80000000UL >> (i & 0x1f);
700 		if (sys_call_table[i*2+1] != sys_ni_syscall)
701 			vdso_data->syscall_map_32[i >> 5] |=
702 				0x80000000UL >> (i & 0x1f);
703 #else /* CONFIG_PPC64 */
704 		if (sys_call_table[i] != sys_ni_syscall)
705 			vdso_data->syscall_map_32[i >> 5] |=
706 				0x80000000UL >> (i & 0x1f);
707 #endif /* CONFIG_PPC64 */
708 	}
709 }
710 
711 #ifdef CONFIG_PPC64
712 int vdso_getcpu_init(void)
713 {
714 	unsigned long cpu, node, val;
715 
716 	/*
717 	 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
718 	 * in the next 16 bits.  The VDSO uses this to implement getcpu().
719 	 */
720 	cpu = get_cpu();
721 	WARN_ON_ONCE(cpu > 0xffff);
722 
723 	node = cpu_to_node(cpu);
724 	WARN_ON_ONCE(node > 0xffff);
725 
726 	val = (cpu & 0xfff) | ((node & 0xffff) << 16);
727 	mtspr(SPRN_SPRG_VDSO_WRITE, val);
728 	get_paca()->sprg_vdso = val;
729 
730 	put_cpu();
731 
732 	return 0;
733 }
734 /* We need to call this before SMP init */
735 early_initcall(vdso_getcpu_init);
736 #endif
737 
738 static int __init vdso_init(void)
739 {
740 	int i;
741 
742 #ifdef CONFIG_PPC64
743 	/*
744 	 * Fill up the "systemcfg" stuff for backward compatibility
745 	 */
746 	strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
747 	vdso_data->version.major = SYSTEMCFG_MAJOR;
748 	vdso_data->version.minor = SYSTEMCFG_MINOR;
749 	vdso_data->processor = mfspr(SPRN_PVR);
750 	/*
751 	 * Fake the old platform number for pSeries and add
752 	 * in LPAR bit if necessary
753 	 */
754 	vdso_data->platform = 0x100;
755 	if (firmware_has_feature(FW_FEATURE_LPAR))
756 		vdso_data->platform |= 1;
757 	vdso_data->physicalMemorySize = memblock_phys_mem_size();
758 	vdso_data->dcache_size = ppc64_caches.dsize;
759 	vdso_data->dcache_line_size = ppc64_caches.dline_size;
760 	vdso_data->icache_size = ppc64_caches.isize;
761 	vdso_data->icache_line_size = ppc64_caches.iline_size;
762 
763 	/* XXXOJN: Blocks should be added to ppc64_caches and used instead */
764 	vdso_data->dcache_block_size = ppc64_caches.dline_size;
765 	vdso_data->icache_block_size = ppc64_caches.iline_size;
766 	vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
767 	vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
768 
769 	/*
770 	 * Calculate the size of the 64 bits vDSO
771 	 */
772 	vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
773 	DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
774 #else
775 	vdso_data->dcache_block_size = L1_CACHE_BYTES;
776 	vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
777 	vdso_data->icache_block_size = L1_CACHE_BYTES;
778 	vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
779 #endif /* CONFIG_PPC64 */
780 
781 
782 	/*
783 	 * Calculate the size of the 32 bits vDSO
784 	 */
785 	vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
786 	DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
787 
788 
789 	/*
790 	 * Setup the syscall map in the vDOS
791 	 */
792 	vdso_setup_syscall_map();
793 
794 	/*
795 	 * Initialize the vDSO images in memory, that is do necessary
796 	 * fixups of vDSO symbols, locate trampolines, etc...
797 	 */
798 	if (vdso_setup()) {
799 		printk(KERN_ERR "vDSO setup failure, not enabled !\n");
800 		vdso32_pages = 0;
801 #ifdef CONFIG_PPC64
802 		vdso64_pages = 0;
803 #endif
804 		return 0;
805 	}
806 
807 	/* Make sure pages are in the correct state */
808 	vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
809 				  GFP_KERNEL);
810 	BUG_ON(vdso32_pagelist == NULL);
811 	for (i = 0; i < vdso32_pages; i++) {
812 		struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
813 		ClearPageReserved(pg);
814 		get_page(pg);
815 		vdso32_pagelist[i] = pg;
816 	}
817 	vdso32_pagelist[i++] = virt_to_page(vdso_data);
818 	vdso32_pagelist[i] = NULL;
819 
820 #ifdef CONFIG_PPC64
821 	vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
822 				  GFP_KERNEL);
823 	BUG_ON(vdso64_pagelist == NULL);
824 	for (i = 0; i < vdso64_pages; i++) {
825 		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
826 		ClearPageReserved(pg);
827 		get_page(pg);
828 		vdso64_pagelist[i] = pg;
829 	}
830 	vdso64_pagelist[i++] = virt_to_page(vdso_data);
831 	vdso64_pagelist[i] = NULL;
832 #endif /* CONFIG_PPC64 */
833 
834 	get_page(virt_to_page(vdso_data));
835 
836 	smp_wmb();
837 	vdso_ready = 1;
838 
839 	return 0;
840 }
841 arch_initcall(vdso_init);
842