Pull xpc-disengage into release branch
[linux-2.6] / arch / ppc64 / kernel / vdso.c
1 /*
2  *  linux/arch/ppc64/kernel/vdso.c
3  *
4  *    Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5  *                       <benh@kernel.crashing.org>
6  *
7  *  This program is free software; you can redistribute it and/or
8  *  modify it under the terms of the GNU General Public License
9  *  as published by the Free Software Foundation; either version
10  *  2 of the License, or (at your option) any later version.
11  */
12
13 #include <linux/config.h>
14 #include <linux/module.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/smp.h>
20 #include <linux/smp_lock.h>
21 #include <linux/stddef.h>
22 #include <linux/unistd.h>
23 #include <linux/slab.h>
24 #include <linux/user.h>
25 #include <linux/elf.h>
26 #include <linux/security.h>
27 #include <linux/bootmem.h>
28
29 #include <asm/pgtable.h>
30 #include <asm/system.h>
31 #include <asm/processor.h>
32 #include <asm/mmu.h>
33 #include <asm/mmu_context.h>
34 #include <asm/machdep.h>
35 #include <asm/cputable.h>
36 #include <asm/sections.h>
37 #include <asm/vdso.h>
38
39 #undef DEBUG
40
41 #ifdef DEBUG
42 #define DBG(fmt...) printk(fmt)
43 #else
44 #define DBG(fmt...)
45 #endif
46
47
48 /*
49  * The vDSOs themselves are here
50  */
51 extern char vdso64_start, vdso64_end;
52 extern char vdso32_start, vdso32_end;
53
54 static void *vdso64_kbase = &vdso64_start;
55 static void *vdso32_kbase = &vdso32_start;
56
57 unsigned int vdso64_pages;
58 unsigned int vdso32_pages;
59
60 /* Signal trampolines user addresses */
61
62 unsigned long vdso64_rt_sigtramp;
63 unsigned long vdso32_sigtramp;
64 unsigned long vdso32_rt_sigtramp;
65
66 /* Format of the patch table */
67 struct vdso_patch_def
68 {
69         u32             pvr_mask, pvr_value;
70         const char      *gen_name;
71         const char      *fix_name;
72 };
73
74 /* Table of functions to patch based on the CPU type/revision
75  *
76  * TODO: Improve by adding whole lists for each entry
77  */
78 static struct vdso_patch_def vdso_patches[] = {
79         {
80                 0xffff0000, 0x003a0000,         /* POWER5 */
81                 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
82         },
83         {
84                 0xffff0000, 0x003b0000,         /* POWER5 */
85                 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
86         },
87 };
88
89 /*
90  * Some infos carried around for each of them during parsing at
91  * boot time.
92  */
93 struct lib32_elfinfo
94 {
95         Elf32_Ehdr      *hdr;           /* ptr to ELF */
96         Elf32_Sym       *dynsym;        /* ptr to .dynsym section */
97         unsigned long   dynsymsize;     /* size of .dynsym section */
98         char            *dynstr;        /* ptr to .dynstr section */
99         unsigned long   text;           /* offset of .text section in .so */
100 };
101
102 struct lib64_elfinfo
103 {
104         Elf64_Ehdr      *hdr;
105         Elf64_Sym       *dynsym;
106         unsigned long   dynsymsize;
107         char            *dynstr;
108         unsigned long   text;
109 };
110
111
112 #ifdef __DEBUG
113 static void dump_one_vdso_page(struct page *pg, struct page *upg)
114 {
115         printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
116                page_count(pg),
117                pg->flags);
118         if (upg/* && pg != upg*/) {
119                 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg) << PAGE_SHIFT),
120                        page_count(upg),
121                        upg->flags);
122         }
123         printk("\n");
124 }
125
126 static void dump_vdso_pages(struct vm_area_struct * vma)
127 {
128         int i;
129
130         if (!vma || test_thread_flag(TIF_32BIT)) {
131                 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
132                 for (i=0; i<vdso32_pages; i++) {
133                         struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
134                         struct page *upg = (vma && vma->vm_mm) ?
135                                 follow_page(vma->vm_mm, vma->vm_start + i*PAGE_SIZE, 0)
136                                 : NULL;
137                         dump_one_vdso_page(pg, upg);
138                 }
139         }
140         if (!vma || !test_thread_flag(TIF_32BIT)) {
141                 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
142                 for (i=0; i<vdso64_pages; i++) {
143                         struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
144                         struct page *upg = (vma && vma->vm_mm) ?
145                                 follow_page(vma->vm_mm, vma->vm_start + i*PAGE_SIZE, 0)
146                                 : NULL;
147                         dump_one_vdso_page(pg, upg);
148                 }
149         }
150 }
151 #endif /* DEBUG */
152
153 /*
154  * Keep a dummy vma_close for now, it will prevent VMA merging.
155  */
156 static void vdso_vma_close(struct vm_area_struct * vma)
157 {
158 }
159
160 /*
161  * Our nopage() function, maps in the actual vDSO kernel pages, they will
162  * be mapped read-only by do_no_page(), and eventually COW'ed, either
163  * right away for an initial write access, or by do_wp_page().
164  */
165 static struct page * vdso_vma_nopage(struct vm_area_struct * vma,
166                                      unsigned long address, int *type)
167 {
168         unsigned long offset = address - vma->vm_start;
169         struct page *pg;
170         void *vbase = test_thread_flag(TIF_32BIT) ? vdso32_kbase : vdso64_kbase;
171
172         DBG("vdso_vma_nopage(current: %s, address: %016lx, off: %lx)\n",
173             current->comm, address, offset);
174
175         if (address < vma->vm_start || address > vma->vm_end)
176                 return NOPAGE_SIGBUS;
177
178         /*
179          * Last page is systemcfg, special handling here, no get_page() a
180          * this is a reserved page
181          */
182         if ((vma->vm_end - address) <= PAGE_SIZE)
183                 return virt_to_page(systemcfg);
184
185         pg = virt_to_page(vbase + offset);
186         get_page(pg);
187         DBG(" ->page count: %d\n", page_count(pg));
188
189         return pg;
190 }
191
192 static struct vm_operations_struct vdso_vmops = {
193         .close  = vdso_vma_close,
194         .nopage = vdso_vma_nopage,
195 };
196
197 /*
198  * This is called from binfmt_elf, we create the special vma for the
199  * vDSO and insert it into the mm struct tree
200  */
201 int arch_setup_additional_pages(struct linux_binprm *bprm, int executable_stack)
202 {
203         struct mm_struct *mm = current->mm;
204         struct vm_area_struct *vma;
205         unsigned long vdso_pages;
206         unsigned long vdso_base;
207
208         if (test_thread_flag(TIF_32BIT)) {
209                 vdso_pages = vdso32_pages;
210                 vdso_base = VDSO32_MBASE;
211         } else {
212                 vdso_pages = vdso64_pages;
213                 vdso_base = VDSO64_MBASE;
214         }
215
216         current->thread.vdso_base = 0;
217
218         /* vDSO has a problem and was disabled, just don't "enable" it for the
219          * process
220          */
221         if (vdso_pages == 0)
222                 return 0;
223
224         vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
225         if (vma == NULL)
226                 return -ENOMEM;
227
228         memset(vma, 0, sizeof(*vma));
229
230         /*
231          * pick a base address for the vDSO in process space. We try to put it
232          * at vdso_base which is the "natural" base for it, but we might fail
233          * and end up putting it elsewhere.
234          */
235         vdso_base = get_unmapped_area(NULL, vdso_base,
236                                       vdso_pages << PAGE_SHIFT, 0, 0);
237         if (vdso_base & ~PAGE_MASK) {
238                 kmem_cache_free(vm_area_cachep, vma);
239                 return (int)vdso_base;
240         }
241
242         current->thread.vdso_base = vdso_base;
243
244         vma->vm_mm = mm;
245         vma->vm_start = current->thread.vdso_base;
246
247         /*
248          * the VMA size is one page more than the vDSO since systemcfg
249          * is mapped in the last one
250          */
251         vma->vm_end = vma->vm_start + ((vdso_pages + 1) << PAGE_SHIFT);
252
253         /*
254          * our vma flags don't have VM_WRITE so by default, the process isn't allowed
255          * to write those pages.
256          * gdb can break that with ptrace interface, and thus trigger COW on those
257          * pages but it's then your responsibility to never do that on the "data" page
258          * of the vDSO or you'll stop getting kernel updates and your nice userland
259          * gettimeofday will be totally dead. It's fine to use that for setting
260          * breakpoints in the vDSO code pages though
261          */
262         vma->vm_flags = VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
263         vma->vm_flags |= mm->def_flags;
264         vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
265         vma->vm_ops = &vdso_vmops;
266
267         down_write(&mm->mmap_sem);
268         if (insert_vm_struct(mm, vma)) {
269                 up_write(&mm->mmap_sem);
270                 kmem_cache_free(vm_area_cachep, vma);
271                 return -ENOMEM;
272         }
273         mm->total_vm += (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
274         up_write(&mm->mmap_sem);
275
276         return 0;
277 }
278
279 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
280                                   unsigned long *size)
281 {
282         Elf32_Shdr *sechdrs;
283         unsigned int i;
284         char *secnames;
285
286         /* Grab section headers and strings so we can tell who is who */
287         sechdrs = (void *)ehdr + ehdr->e_shoff;
288         secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
289
290         /* Find the section they want */
291         for (i = 1; i < ehdr->e_shnum; i++) {
292                 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
293                         if (size)
294                                 *size = sechdrs[i].sh_size;
295                         return (void *)ehdr + sechdrs[i].sh_offset;
296                 }
297         }
298         *size = 0;
299         return NULL;
300 }
301
302 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
303                                   unsigned long *size)
304 {
305         Elf64_Shdr *sechdrs;
306         unsigned int i;
307         char *secnames;
308
309         /* Grab section headers and strings so we can tell who is who */
310         sechdrs = (void *)ehdr + ehdr->e_shoff;
311         secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
312
313         /* Find the section they want */
314         for (i = 1; i < ehdr->e_shnum; i++) {
315                 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
316                         if (size)
317                                 *size = sechdrs[i].sh_size;
318                         return (void *)ehdr + sechdrs[i].sh_offset;
319                 }
320         }
321         if (size)
322                 *size = 0;
323         return NULL;
324 }
325
326 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib, const char *symname)
327 {
328         unsigned int i;
329         char name[32], *c;
330
331         for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
332                 if (lib->dynsym[i].st_name == 0)
333                         continue;
334                 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name, 32);
335                 c = strchr(name, '@');
336                 if (c)
337                         *c = 0;
338                 if (strcmp(symname, name) == 0)
339                         return &lib->dynsym[i];
340         }
341         return NULL;
342 }
343
344 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib, const char *symname)
345 {
346         unsigned int i;
347         char name[32], *c;
348
349         for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
350                 if (lib->dynsym[i].st_name == 0)
351                         continue;
352                 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name, 32);
353                 c = strchr(name, '@');
354                 if (c)
355                         *c = 0;
356                 if (strcmp(symname, name) == 0)
357                         return &lib->dynsym[i];
358         }
359         return NULL;
360 }
361
362 /* Note that we assume the section is .text and the symbol is relative to
363  * the library base
364  */
365 static unsigned long __init find_function32(struct lib32_elfinfo *lib, const char *symname)
366 {
367         Elf32_Sym *sym = find_symbol32(lib, symname);
368
369         if (sym == NULL) {
370                 printk(KERN_WARNING "vDSO32: function %s not found !\n", symname);
371                 return 0;
372         }
373         return sym->st_value - VDSO32_LBASE;
374 }
375
376 /* Note that we assume the section is .text and the symbol is relative to
377  * the library base
378  */
379 static unsigned long __init find_function64(struct lib64_elfinfo *lib, const char *symname)
380 {
381         Elf64_Sym *sym = find_symbol64(lib, symname);
382
383         if (sym == NULL) {
384                 printk(KERN_WARNING "vDSO64: function %s not found !\n", symname);
385                 return 0;
386         }
387 #ifdef VDS64_HAS_DESCRIPTORS
388         return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) - VDSO64_LBASE;
389 #else
390         return sym->st_value - VDSO64_LBASE;
391 #endif
392 }
393
394
395 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
396                                         struct lib64_elfinfo *v64)
397 {
398         void *sect;
399
400         /*
401          * Locate symbol tables & text section
402          */
403
404         v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
405         v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
406         if (v32->dynsym == NULL || v32->dynstr == NULL) {
407                 printk(KERN_ERR "vDSO32: a required symbol section was not found\n");
408                 return -1;
409         }
410         sect = find_section32(v32->hdr, ".text", NULL);
411         if (sect == NULL) {
412                 printk(KERN_ERR "vDSO32: the .text section was not found\n");
413                 return -1;
414         }
415         v32->text = sect - vdso32_kbase;
416
417         v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
418         v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
419         if (v64->dynsym == NULL || v64->dynstr == NULL) {
420                 printk(KERN_ERR "vDSO64: a required symbol section was not found\n");
421                 return -1;
422         }
423         sect = find_section64(v64->hdr, ".text", NULL);
424         if (sect == NULL) {
425                 printk(KERN_ERR "vDSO64: the .text section was not found\n");
426                 return -1;
427         }
428         v64->text = sect - vdso64_kbase;
429
430         return 0;
431 }
432
433 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
434                                           struct lib64_elfinfo *v64)
435 {
436         /*
437          * Find signal trampolines
438          */
439
440         vdso64_rt_sigtramp      = find_function64(v64, "__kernel_sigtramp_rt64");
441         vdso32_sigtramp         = find_function32(v32, "__kernel_sigtramp32");
442         vdso32_rt_sigtramp      = find_function32(v32, "__kernel_sigtramp_rt32");
443 }
444
445 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
446                                        struct lib64_elfinfo *v64)
447 {
448         Elf32_Sym *sym32;
449         Elf64_Sym *sym64;
450
451         sym32 = find_symbol32(v32, "__kernel_datapage_offset");
452         if (sym32 == NULL) {
453                 printk(KERN_ERR "vDSO32: Can't find symbol __kernel_datapage_offset !\n");
454                 return -1;
455         }
456         *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
457                 (vdso32_pages << PAGE_SHIFT) - (sym32->st_value - VDSO32_LBASE);
458
459         sym64 = find_symbol64(v64, "__kernel_datapage_offset");
460         if (sym64 == NULL) {
461                 printk(KERN_ERR "vDSO64: Can't find symbol __kernel_datapage_offset !\n");
462                 return -1;
463         }
464         *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
465                 (vdso64_pages << PAGE_SHIFT) - (sym64->st_value - VDSO64_LBASE);
466
467         return 0;
468 }
469
470 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
471                                 struct lib64_elfinfo *v64,
472                                 const char *orig, const char *fix)
473 {
474         Elf32_Sym *sym32_gen, *sym32_fix;
475
476         sym32_gen = find_symbol32(v32, orig);
477         if (sym32_gen == NULL) {
478                 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
479                 return -1;
480         }
481         sym32_fix = find_symbol32(v32, fix);
482         if (sym32_fix == NULL) {
483                 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
484                 return -1;
485         }
486         sym32_gen->st_value = sym32_fix->st_value;
487         sym32_gen->st_size = sym32_fix->st_size;
488         sym32_gen->st_info = sym32_fix->st_info;
489         sym32_gen->st_other = sym32_fix->st_other;
490         sym32_gen->st_shndx = sym32_fix->st_shndx;
491
492         return 0;
493 }
494
495 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
496                                 struct lib64_elfinfo *v64,
497                                 const char *orig, const char *fix)
498 {
499         Elf64_Sym *sym64_gen, *sym64_fix;
500
501         sym64_gen = find_symbol64(v64, orig);
502         if (sym64_gen == NULL) {
503                 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
504                 return -1;
505         }
506         sym64_fix = find_symbol64(v64, fix);
507         if (sym64_fix == NULL) {
508                 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
509                 return -1;
510         }
511         sym64_gen->st_value = sym64_fix->st_value;
512         sym64_gen->st_size = sym64_fix->st_size;
513         sym64_gen->st_info = sym64_fix->st_info;
514         sym64_gen->st_other = sym64_fix->st_other;
515         sym64_gen->st_shndx = sym64_fix->st_shndx;
516
517         return 0;
518 }
519
520 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
521                                        struct lib64_elfinfo *v64)
522 {
523         u32 pvr;
524         int i;
525
526         pvr = mfspr(SPRN_PVR);
527         for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
528                 struct vdso_patch_def *patch = &vdso_patches[i];
529                 int match = (pvr & patch->pvr_mask) == patch->pvr_value;
530
531                 DBG("patch %d (mask: %x, pvr: %x) : %s\n",
532                     i, patch->pvr_mask, patch->pvr_value, match ? "match" : "skip");
533
534                 if (!match)
535                         continue;
536
537                 DBG("replacing %s with %s...\n", patch->gen_name, patch->fix_name);
538
539                 /*
540                  * Patch the 32 bits and 64 bits symbols. Note that we do not patch
541                  * the "." symbol on 64 bits. It would be easy to do, but doesn't
542                  * seem to be necessary, patching the OPD symbol is enough.
543                  */
544                 vdso_do_func_patch32(v32, v64, patch->gen_name, patch->fix_name);
545                 vdso_do_func_patch64(v32, v64, patch->gen_name, patch->fix_name);
546         }
547
548         return 0;
549 }
550
551
552 static __init int vdso_setup(void)
553 {
554         struct lib32_elfinfo    v32;
555         struct lib64_elfinfo    v64;
556
557         v32.hdr = vdso32_kbase;
558         v64.hdr = vdso64_kbase;
559
560         if (vdso_do_find_sections(&v32, &v64))
561                 return -1;
562
563         if (vdso_fixup_datapage(&v32, &v64))
564                 return -1;
565
566         if (vdso_fixup_alt_funcs(&v32, &v64))
567                 return -1;
568
569         vdso_setup_trampolines(&v32, &v64);
570
571         return 0;
572 }
573
574 void __init vdso_init(void)
575 {
576         int i;
577
578         vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
579         vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
580
581         DBG("vdso64_kbase: %p, 0x%x pages, vdso32_kbase: %p, 0x%x pages\n",
582                vdso64_kbase, vdso64_pages, vdso32_kbase, vdso32_pages);
583
584         /*
585          * Initialize the vDSO images in memory, that is do necessary
586          * fixups of vDSO symbols, locate trampolines, etc...
587          */
588         if (vdso_setup()) {
589                 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
590                 /* XXX should free pages here ? */
591                 vdso64_pages = vdso32_pages = 0;
592                 return;
593         }
594
595         /* Make sure pages are in the correct state */
596         for (i = 0; i < vdso64_pages; i++) {
597                 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
598                 ClearPageReserved(pg);
599                 get_page(pg);
600         }
601         for (i = 0; i < vdso32_pages; i++) {
602                 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
603                 ClearPageReserved(pg);
604                 get_page(pg);
605         }
606 }
607
608 int in_gate_area_no_task(unsigned long addr)
609 {
610         return 0;
611 }
612
613 int in_gate_area(struct task_struct *task, unsigned long addr)
614 {
615         return 0;
616 }
617
618 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
619 {
620         return NULL;
621 }
622