Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2
3/*
4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5 * <benh@kernel.crashing.org>
6 */
7
8#include <linux/errno.h>
9#include <linux/sched.h>
10#include <linux/kernel.h>
11#include <linux/mm.h>
12#include <linux/smp.h>
13#include <linux/stddef.h>
14#include <linux/unistd.h>
15#include <linux/slab.h>
16#include <linux/user.h>
17#include <linux/elf.h>
18#include <linux/security.h>
19#include <linux/memblock.h>
20
21#include <asm/processor.h>
22#include <asm/mmu.h>
23#include <asm/mmu_context.h>
24#include <asm/prom.h>
25#include <asm/machdep.h>
26#include <asm/cputable.h>
27#include <asm/sections.h>
28#include <asm/firmware.h>
29#include <asm/vdso.h>
30#include <asm/vdso_datapage.h>
31#include <asm/setup.h>
32
33#undef DEBUG
34
35#ifdef DEBUG
36#define DBG(fmt...) printk(fmt)
37#else
38#define DBG(fmt...)
39#endif
40
41/* Max supported size for symbol names */
42#define MAX_SYMNAME 64
43
44/* The alignment of the vDSO */
45#define VDSO_ALIGNMENT (1 << 16)
46
47static unsigned int vdso32_pages;
48static void *vdso32_kbase;
49static struct page **vdso32_pagelist;
50unsigned long vdso32_sigtramp;
51unsigned long vdso32_rt_sigtramp;
52
53#ifdef CONFIG_VDSO32
54extern char vdso32_start, vdso32_end;
55#endif
56
57#ifdef CONFIG_PPC64
58extern char vdso64_start, vdso64_end;
59static void *vdso64_kbase = &vdso64_start;
60static unsigned int vdso64_pages;
61static struct page **vdso64_pagelist;
62unsigned long vdso64_rt_sigtramp;
63#endif /* CONFIG_PPC64 */
64
65static int vdso_ready;
66
67/*
68 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
69 * Once the early boot kernel code no longer needs to muck around
70 * with it, it will become dynamically allocated
71 */
72static union {
73 struct vdso_data data;
74 u8 page[PAGE_SIZE];
75} vdso_data_store __page_aligned_data;
76struct vdso_data *vdso_data = &vdso_data_store.data;
77
78/* Format of the patch table */
79struct vdso_patch_def
80{
81 unsigned long ftr_mask, ftr_value;
82 const char *gen_name;
83 const char *fix_name;
84};
85
86/* Table of functions to patch based on the CPU type/revision
87 *
88 * Currently, we only change sync_dicache to do nothing on processors
89 * with a coherent icache
90 */
91static struct vdso_patch_def vdso_patches[] = {
92 {
93 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
94 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
95 },
96};
97
98/*
99 * Some infos carried around for each of them during parsing at
100 * boot time.
101 */
102struct lib32_elfinfo
103{
104 Elf32_Ehdr *hdr; /* ptr to ELF */
105 Elf32_Sym *dynsym; /* ptr to .dynsym section */
106 unsigned long dynsymsize; /* size of .dynsym section */
107 char *dynstr; /* ptr to .dynstr section */
108 unsigned long text; /* offset of .text section in .so */
109};
110
111struct lib64_elfinfo
112{
113 Elf64_Ehdr *hdr;
114 Elf64_Sym *dynsym;
115 unsigned long dynsymsize;
116 char *dynstr;
117 unsigned long text;
118};
119
120
121/*
122 * This is called from binfmt_elf, we create the special vma for the
123 * vDSO and insert it into the mm struct tree
124 */
125int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
126{
127 struct mm_struct *mm = current->mm;
128 struct page **vdso_pagelist;
129 unsigned long vdso_pages;
130 unsigned long vdso_base;
131 int rc;
132
133 if (!vdso_ready)
134 return 0;
135
136#ifdef CONFIG_PPC64
137 if (is_32bit_task()) {
138 vdso_pagelist = vdso32_pagelist;
139 vdso_pages = vdso32_pages;
140 vdso_base = VDSO32_MBASE;
141 } else {
142 vdso_pagelist = vdso64_pagelist;
143 vdso_pages = vdso64_pages;
144 /*
145 * On 64bit we don't have a preferred map address. This
146 * allows get_unmapped_area to find an area near other mmaps
147 * and most likely share a SLB entry.
148 */
149 vdso_base = 0;
150 }
151#else
152 vdso_pagelist = vdso32_pagelist;
153 vdso_pages = vdso32_pages;
154 vdso_base = VDSO32_MBASE;
155#endif
156
157 current->mm->context.vdso_base = 0;
158
159 /* vDSO has a problem and was disabled, just don't "enable" it for the
160 * process
161 */
162 if (vdso_pages == 0)
163 return 0;
164 /* Add a page to the vdso size for the data page */
165 vdso_pages ++;
166
167 /*
168 * pick a base address for the vDSO in process space. We try to put it
169 * at vdso_base which is the "natural" base for it, but we might fail
170 * and end up putting it elsewhere.
171 * Add enough to the size so that the result can be aligned.
172 */
173 if (mmap_write_lock_killable(mm))
174 return -EINTR;
175 vdso_base = get_unmapped_area(NULL, vdso_base,
176 (vdso_pages << PAGE_SHIFT) +
177 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
178 0, 0);
179 if (IS_ERR_VALUE(vdso_base)) {
180 rc = vdso_base;
181 goto fail_mmapsem;
182 }
183
184 /* Add required alignment. */
185 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
186
187 /*
188 * Put vDSO base into mm struct. We need to do this before calling
189 * install_special_mapping or the perf counter mmap tracking code
190 * will fail to recognise it as a vDSO (since arch_vma_name fails).
191 */
192 current->mm->context.vdso_base = vdso_base;
193
194 /*
195 * our vma flags don't have VM_WRITE so by default, the process isn't
196 * allowed to write those pages.
197 * gdb can break that with ptrace interface, and thus trigger COW on
198 * those pages but it's then your responsibility to never do that on
199 * the "data" page of the vDSO or you'll stop getting kernel updates
200 * and your nice userland gettimeofday will be totally dead.
201 * It's fine to use that for setting breakpoints in the vDSO code
202 * pages though.
203 */
204 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
205 VM_READ|VM_EXEC|
206 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
207 vdso_pagelist);
208 if (rc) {
209 current->mm->context.vdso_base = 0;
210 goto fail_mmapsem;
211 }
212
213 mmap_write_unlock(mm);
214 return 0;
215
216 fail_mmapsem:
217 mmap_write_unlock(mm);
218 return rc;
219}
220
221const char *arch_vma_name(struct vm_area_struct *vma)
222{
223 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
224 return "[vdso]";
225 return NULL;
226}
227
228
229
230#ifdef CONFIG_VDSO32
231static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
232 unsigned long *size)
233{
234 Elf32_Shdr *sechdrs;
235 unsigned int i;
236 char *secnames;
237
238 /* Grab section headers and strings so we can tell who is who */
239 sechdrs = (void *)ehdr + ehdr->e_shoff;
240 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
241
242 /* Find the section they want */
243 for (i = 1; i < ehdr->e_shnum; i++) {
244 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
245 if (size)
246 *size = sechdrs[i].sh_size;
247 return (void *)ehdr + sechdrs[i].sh_offset;
248 }
249 }
250 *size = 0;
251 return NULL;
252}
253
254static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
255 const char *symname)
256{
257 unsigned int i;
258 char name[MAX_SYMNAME], *c;
259
260 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
261 if (lib->dynsym[i].st_name == 0)
262 continue;
263 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
264 MAX_SYMNAME);
265 c = strchr(name, '@');
266 if (c)
267 *c = 0;
268 if (strcmp(symname, name) == 0)
269 return &lib->dynsym[i];
270 }
271 return NULL;
272}
273
274/* Note that we assume the section is .text and the symbol is relative to
275 * the library base
276 */
277static unsigned long __init find_function32(struct lib32_elfinfo *lib,
278 const char *symname)
279{
280 Elf32_Sym *sym = find_symbol32(lib, symname);
281
282 if (sym == NULL) {
283 printk(KERN_WARNING "vDSO32: function %s not found !\n",
284 symname);
285 return 0;
286 }
287 return sym->st_value - VDSO32_LBASE;
288}
289
290static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
291 struct lib64_elfinfo *v64,
292 const char *orig, const char *fix)
293{
294 Elf32_Sym *sym32_gen, *sym32_fix;
295
296 sym32_gen = find_symbol32(v32, orig);
297 if (sym32_gen == NULL) {
298 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
299 return -1;
300 }
301 if (fix == NULL) {
302 sym32_gen->st_name = 0;
303 return 0;
304 }
305 sym32_fix = find_symbol32(v32, fix);
306 if (sym32_fix == NULL) {
307 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
308 return -1;
309 }
310 sym32_gen->st_value = sym32_fix->st_value;
311 sym32_gen->st_size = sym32_fix->st_size;
312 sym32_gen->st_info = sym32_fix->st_info;
313 sym32_gen->st_other = sym32_fix->st_other;
314 sym32_gen->st_shndx = sym32_fix->st_shndx;
315
316 return 0;
317}
318#else /* !CONFIG_VDSO32 */
319static unsigned long __init find_function32(struct lib32_elfinfo *lib,
320 const char *symname)
321{
322 return 0;
323}
324
325static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
326 struct lib64_elfinfo *v64,
327 const char *orig, const char *fix)
328{
329 return 0;
330}
331#endif /* CONFIG_VDSO32 */
332
333
334#ifdef CONFIG_PPC64
335
336static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
337 unsigned long *size)
338{
339 Elf64_Shdr *sechdrs;
340 unsigned int i;
341 char *secnames;
342
343 /* Grab section headers and strings so we can tell who is who */
344 sechdrs = (void *)ehdr + ehdr->e_shoff;
345 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
346
347 /* Find the section they want */
348 for (i = 1; i < ehdr->e_shnum; i++) {
349 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
350 if (size)
351 *size = sechdrs[i].sh_size;
352 return (void *)ehdr + sechdrs[i].sh_offset;
353 }
354 }
355 if (size)
356 *size = 0;
357 return NULL;
358}
359
360static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
361 const char *symname)
362{
363 unsigned int i;
364 char name[MAX_SYMNAME], *c;
365
366 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
367 if (lib->dynsym[i].st_name == 0)
368 continue;
369 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
370 MAX_SYMNAME);
371 c = strchr(name, '@');
372 if (c)
373 *c = 0;
374 if (strcmp(symname, name) == 0)
375 return &lib->dynsym[i];
376 }
377 return NULL;
378}
379
380/* Note that we assume the section is .text and the symbol is relative to
381 * the library base
382 */
383static unsigned long __init find_function64(struct lib64_elfinfo *lib,
384 const char *symname)
385{
386 Elf64_Sym *sym = find_symbol64(lib, symname);
387
388 if (sym == NULL) {
389 printk(KERN_WARNING "vDSO64: function %s not found !\n",
390 symname);
391 return 0;
392 }
393 return sym->st_value - VDSO64_LBASE;
394}
395
396static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
397 struct lib64_elfinfo *v64,
398 const char *orig, const char *fix)
399{
400 Elf64_Sym *sym64_gen, *sym64_fix;
401
402 sym64_gen = find_symbol64(v64, orig);
403 if (sym64_gen == NULL) {
404 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
405 return -1;
406 }
407 if (fix == NULL) {
408 sym64_gen->st_name = 0;
409 return 0;
410 }
411 sym64_fix = find_symbol64(v64, fix);
412 if (sym64_fix == NULL) {
413 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
414 return -1;
415 }
416 sym64_gen->st_value = sym64_fix->st_value;
417 sym64_gen->st_size = sym64_fix->st_size;
418 sym64_gen->st_info = sym64_fix->st_info;
419 sym64_gen->st_other = sym64_fix->st_other;
420 sym64_gen->st_shndx = sym64_fix->st_shndx;
421
422 return 0;
423}
424
425#endif /* CONFIG_PPC64 */
426
427
428static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
429 struct lib64_elfinfo *v64)
430{
431 void *sect;
432
433 /*
434 * Locate symbol tables & text section
435 */
436
437#ifdef CONFIG_VDSO32
438 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
439 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
440 if (v32->dynsym == NULL || v32->dynstr == NULL) {
441 printk(KERN_ERR "vDSO32: required symbol section not found\n");
442 return -1;
443 }
444 sect = find_section32(v32->hdr, ".text", NULL);
445 if (sect == NULL) {
446 printk(KERN_ERR "vDSO32: the .text section was not found\n");
447 return -1;
448 }
449 v32->text = sect - vdso32_kbase;
450#endif
451
452#ifdef CONFIG_PPC64
453 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
454 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
455 if (v64->dynsym == NULL || v64->dynstr == NULL) {
456 printk(KERN_ERR "vDSO64: required symbol section not found\n");
457 return -1;
458 }
459 sect = find_section64(v64->hdr, ".text", NULL);
460 if (sect == NULL) {
461 printk(KERN_ERR "vDSO64: the .text section was not found\n");
462 return -1;
463 }
464 v64->text = sect - vdso64_kbase;
465#endif /* CONFIG_PPC64 */
466
467 return 0;
468}
469
470static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
471 struct lib64_elfinfo *v64)
472{
473 /*
474 * Find signal trampolines
475 */
476
477#ifdef CONFIG_PPC64
478 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
479#endif
480 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
481 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
482}
483
484static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
485 struct lib64_elfinfo *v64)
486{
487#ifdef CONFIG_VDSO32
488 Elf32_Sym *sym32;
489#endif
490#ifdef CONFIG_PPC64
491 Elf64_Sym *sym64;
492
493 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
494 if (sym64 == NULL) {
495 printk(KERN_ERR "vDSO64: Can't find symbol "
496 "__kernel_datapage_offset !\n");
497 return -1;
498 }
499 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
500 (vdso64_pages << PAGE_SHIFT) -
501 (sym64->st_value - VDSO64_LBASE);
502#endif /* CONFIG_PPC64 */
503
504#ifdef CONFIG_VDSO32
505 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
506 if (sym32 == NULL) {
507 printk(KERN_ERR "vDSO32: Can't find symbol "
508 "__kernel_datapage_offset !\n");
509 return -1;
510 }
511 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
512 (vdso32_pages << PAGE_SHIFT) -
513 (sym32->st_value - VDSO32_LBASE);
514#endif
515
516 return 0;
517}
518
519
520static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
521 struct lib64_elfinfo *v64)
522{
523 unsigned long size;
524 void *start;
525
526#ifdef CONFIG_PPC64
527 start = find_section64(v64->hdr, "__ftr_fixup", &size);
528 if (start)
529 do_feature_fixups(cur_cpu_spec->cpu_features,
530 start, start + size);
531
532 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
533 if (start)
534 do_feature_fixups(cur_cpu_spec->mmu_features,
535 start, start + size);
536
537 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
538 if (start)
539 do_feature_fixups(powerpc_firmware_features,
540 start, start + size);
541
542 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
543 if (start)
544 do_lwsync_fixups(cur_cpu_spec->cpu_features,
545 start, start + size);
546#endif /* CONFIG_PPC64 */
547
548#ifdef CONFIG_VDSO32
549 start = find_section32(v32->hdr, "__ftr_fixup", &size);
550 if (start)
551 do_feature_fixups(cur_cpu_spec->cpu_features,
552 start, start + size);
553
554 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
555 if (start)
556 do_feature_fixups(cur_cpu_spec->mmu_features,
557 start, start + size);
558
559#ifdef CONFIG_PPC64
560 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
561 if (start)
562 do_feature_fixups(powerpc_firmware_features,
563 start, start + size);
564#endif /* CONFIG_PPC64 */
565
566 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
567 if (start)
568 do_lwsync_fixups(cur_cpu_spec->cpu_features,
569 start, start + size);
570#endif
571
572 return 0;
573}
574
575static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
576 struct lib64_elfinfo *v64)
577{
578 int i;
579
580 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
581 struct vdso_patch_def *patch = &vdso_patches[i];
582 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
583 == patch->ftr_value;
584 if (!match)
585 continue;
586
587 DBG("replacing %s with %s...\n", patch->gen_name,
588 patch->fix_name ? "NONE" : patch->fix_name);
589
590 /*
591 * Patch the 32 bits and 64 bits symbols. Note that we do not
592 * patch the "." symbol on 64 bits.
593 * It would be easy to do, but doesn't seem to be necessary,
594 * patching the OPD symbol is enough.
595 */
596 vdso_do_func_patch32(v32, v64, patch->gen_name,
597 patch->fix_name);
598#ifdef CONFIG_PPC64
599 vdso_do_func_patch64(v32, v64, patch->gen_name,
600 patch->fix_name);
601#endif /* CONFIG_PPC64 */
602 }
603
604 return 0;
605}
606
607
608static __init int vdso_setup(void)
609{
610 struct lib32_elfinfo v32;
611 struct lib64_elfinfo v64;
612
613 v32.hdr = vdso32_kbase;
614#ifdef CONFIG_PPC64
615 v64.hdr = vdso64_kbase;
616#endif
617 if (vdso_do_find_sections(&v32, &v64))
618 return -1;
619
620 if (vdso_fixup_datapage(&v32, &v64))
621 return -1;
622
623 if (vdso_fixup_features(&v32, &v64))
624 return -1;
625
626 if (vdso_fixup_alt_funcs(&v32, &v64))
627 return -1;
628
629 vdso_setup_trampolines(&v32, &v64);
630
631 return 0;
632}
633
634/*
635 * Called from setup_arch to initialize the bitmap of available
636 * syscalls in the systemcfg page
637 */
638static void __init vdso_setup_syscall_map(void)
639{
640 unsigned int i;
641 extern unsigned long *sys_call_table;
642#ifdef CONFIG_PPC64
643 extern unsigned long *compat_sys_call_table;
644#endif
645 extern unsigned long sys_ni_syscall;
646
647
648 for (i = 0; i < NR_syscalls; i++) {
649#ifdef CONFIG_PPC64
650 if (sys_call_table[i] != sys_ni_syscall)
651 vdso_data->syscall_map_64[i >> 5] |=
652 0x80000000UL >> (i & 0x1f);
653 if (IS_ENABLED(CONFIG_COMPAT) &&
654 compat_sys_call_table[i] != sys_ni_syscall)
655 vdso_data->syscall_map_32[i >> 5] |=
656 0x80000000UL >> (i & 0x1f);
657#else /* CONFIG_PPC64 */
658 if (sys_call_table[i] != sys_ni_syscall)
659 vdso_data->syscall_map_32[i >> 5] |=
660 0x80000000UL >> (i & 0x1f);
661#endif /* CONFIG_PPC64 */
662 }
663}
664
665#ifdef CONFIG_PPC64
666int vdso_getcpu_init(void)
667{
668 unsigned long cpu, node, val;
669
670 /*
671 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
672 * in the next 16 bits. The VDSO uses this to implement getcpu().
673 */
674 cpu = get_cpu();
675 WARN_ON_ONCE(cpu > 0xffff);
676
677 node = cpu_to_node(cpu);
678 WARN_ON_ONCE(node > 0xffff);
679
680 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
681 mtspr(SPRN_SPRG_VDSO_WRITE, val);
682 get_paca()->sprg_vdso = val;
683
684 put_cpu();
685
686 return 0;
687}
688/* We need to call this before SMP init */
689early_initcall(vdso_getcpu_init);
690#endif
691
692static int __init vdso_init(void)
693{
694 int i;
695
696#ifdef CONFIG_PPC64
697 /*
698 * Fill up the "systemcfg" stuff for backward compatibility
699 */
700 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
701 vdso_data->version.major = SYSTEMCFG_MAJOR;
702 vdso_data->version.minor = SYSTEMCFG_MINOR;
703 vdso_data->processor = mfspr(SPRN_PVR);
704 /*
705 * Fake the old platform number for pSeries and add
706 * in LPAR bit if necessary
707 */
708 vdso_data->platform = 0x100;
709 if (firmware_has_feature(FW_FEATURE_LPAR))
710 vdso_data->platform |= 1;
711 vdso_data->physicalMemorySize = memblock_phys_mem_size();
712 vdso_data->dcache_size = ppc64_caches.l1d.size;
713 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
714 vdso_data->icache_size = ppc64_caches.l1i.size;
715 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
716 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
717 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
718 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
719 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
720
721 /*
722 * Calculate the size of the 64 bits vDSO
723 */
724 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
725 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
726#endif /* CONFIG_PPC64 */
727
728
729#ifdef CONFIG_VDSO32
730 vdso32_kbase = &vdso32_start;
731
732 /*
733 * Calculate the size of the 32 bits vDSO
734 */
735 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
736 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
737#endif
738
739
740 /*
741 * Setup the syscall map in the vDOS
742 */
743 vdso_setup_syscall_map();
744
745 /*
746 * Initialize the vDSO images in memory, that is do necessary
747 * fixups of vDSO symbols, locate trampolines, etc...
748 */
749 if (vdso_setup()) {
750 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
751 vdso32_pages = 0;
752#ifdef CONFIG_PPC64
753 vdso64_pages = 0;
754#endif
755 return 0;
756 }
757
758#ifdef CONFIG_VDSO32
759 /* Make sure pages are in the correct state */
760 vdso32_pagelist = kcalloc(vdso32_pages + 2, sizeof(struct page *),
761 GFP_KERNEL);
762 BUG_ON(vdso32_pagelist == NULL);
763 for (i = 0; i < vdso32_pages; i++) {
764 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
765 get_page(pg);
766 vdso32_pagelist[i] = pg;
767 }
768 vdso32_pagelist[i++] = virt_to_page(vdso_data);
769 vdso32_pagelist[i] = NULL;
770#endif
771
772#ifdef CONFIG_PPC64
773 vdso64_pagelist = kcalloc(vdso64_pages + 2, sizeof(struct page *),
774 GFP_KERNEL);
775 BUG_ON(vdso64_pagelist == NULL);
776 for (i = 0; i < vdso64_pages; i++) {
777 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
778 get_page(pg);
779 vdso64_pagelist[i] = pg;
780 }
781 vdso64_pagelist[i++] = virt_to_page(vdso_data);
782 vdso64_pagelist[i] = NULL;
783#endif /* CONFIG_PPC64 */
784
785 get_page(virt_to_page(vdso_data));
786
787 smp_wmb();
788 vdso_ready = 1;
789
790 return 0;
791}
792arch_initcall(vdso_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2
3/*
4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5 * <benh@kernel.crashing.org>
6 */
7
8#include <linux/errno.h>
9#include <linux/sched.h>
10#include <linux/kernel.h>
11#include <linux/mm.h>
12#include <linux/smp.h>
13#include <linux/stddef.h>
14#include <linux/unistd.h>
15#include <linux/slab.h>
16#include <linux/user.h>
17#include <linux/elf.h>
18#include <linux/security.h>
19#include <linux/memblock.h>
20#include <linux/syscalls.h>
21#include <linux/time_namespace.h>
22#include <vdso/datapage.h>
23
24#include <asm/syscall.h>
25#include <asm/processor.h>
26#include <asm/mmu.h>
27#include <asm/mmu_context.h>
28#include <asm/machdep.h>
29#include <asm/cputable.h>
30#include <asm/sections.h>
31#include <asm/firmware.h>
32#include <asm/vdso.h>
33#include <asm/vdso_datapage.h>
34#include <asm/setup.h>
35
36/* The alignment of the vDSO */
37#define VDSO_ALIGNMENT (1 << 16)
38
39extern char vdso32_start, vdso32_end;
40extern char vdso64_start, vdso64_end;
41
42long sys_ni_syscall(void);
43
44/*
45 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
46 * Once the early boot kernel code no longer needs to muck around
47 * with it, it will become dynamically allocated
48 */
49static union {
50 struct vdso_arch_data data;
51 u8 page[PAGE_SIZE];
52} vdso_data_store __page_aligned_data;
53struct vdso_arch_data *vdso_data = &vdso_data_store.data;
54
55enum vvar_pages {
56 VVAR_DATA_PAGE_OFFSET,
57 VVAR_TIMENS_PAGE_OFFSET,
58 VVAR_NR_PAGES,
59};
60
61static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma,
62 unsigned long text_size)
63{
64 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
65
66 if (new_size != text_size)
67 return -EINVAL;
68
69 current->mm->context.vdso = (void __user *)new_vma->vm_start;
70
71 return 0;
72}
73
74static int vdso32_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
75{
76 return vdso_mremap(sm, new_vma, &vdso32_end - &vdso32_start);
77}
78
79static int vdso64_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
80{
81 return vdso_mremap(sm, new_vma, &vdso64_end - &vdso64_start);
82}
83
84static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
85 struct vm_area_struct *vma, struct vm_fault *vmf);
86
87static struct vm_special_mapping vvar_spec __ro_after_init = {
88 .name = "[vvar]",
89 .fault = vvar_fault,
90};
91
92static struct vm_special_mapping vdso32_spec __ro_after_init = {
93 .name = "[vdso]",
94 .mremap = vdso32_mremap,
95};
96
97static struct vm_special_mapping vdso64_spec __ro_after_init = {
98 .name = "[vdso]",
99 .mremap = vdso64_mremap,
100};
101
102#ifdef CONFIG_TIME_NS
103struct vdso_data *arch_get_vdso_data(void *vvar_page)
104{
105 return ((struct vdso_arch_data *)vvar_page)->data;
106}
107
108/*
109 * The vvar mapping contains data for a specific time namespace, so when a task
110 * changes namespace we must unmap its vvar data for the old namespace.
111 * Subsequent faults will map in data for the new namespace.
112 *
113 * For more details see timens_setup_vdso_data().
114 */
115int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
116{
117 struct mm_struct *mm = task->mm;
118 VMA_ITERATOR(vmi, mm, 0);
119 struct vm_area_struct *vma;
120
121 mmap_read_lock(mm);
122 for_each_vma(vmi, vma) {
123 unsigned long size = vma->vm_end - vma->vm_start;
124
125 if (vma_is_special_mapping(vma, &vvar_spec))
126 zap_page_range(vma, vma->vm_start, size);
127 }
128 mmap_read_unlock(mm);
129
130 return 0;
131}
132#endif
133
134static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
135 struct vm_area_struct *vma, struct vm_fault *vmf)
136{
137 struct page *timens_page = find_timens_vvar_page(vma);
138 unsigned long pfn;
139
140 switch (vmf->pgoff) {
141 case VVAR_DATA_PAGE_OFFSET:
142 if (timens_page)
143 pfn = page_to_pfn(timens_page);
144 else
145 pfn = virt_to_pfn(vdso_data);
146 break;
147#ifdef CONFIG_TIME_NS
148 case VVAR_TIMENS_PAGE_OFFSET:
149 /*
150 * If a task belongs to a time namespace then a namespace
151 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
152 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
153 * offset.
154 * See also the comment near timens_setup_vdso_data().
155 */
156 if (!timens_page)
157 return VM_FAULT_SIGBUS;
158 pfn = virt_to_pfn(vdso_data);
159 break;
160#endif /* CONFIG_TIME_NS */
161 default:
162 return VM_FAULT_SIGBUS;
163 }
164
165 return vmf_insert_pfn(vma, vmf->address, pfn);
166}
167
168/*
169 * This is called from binfmt_elf, we create the special vma for the
170 * vDSO and insert it into the mm struct tree
171 */
172static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
173{
174 unsigned long vdso_size, vdso_base, mappings_size;
175 struct vm_special_mapping *vdso_spec;
176 unsigned long vvar_size = VVAR_NR_PAGES * PAGE_SIZE;
177 struct mm_struct *mm = current->mm;
178 struct vm_area_struct *vma;
179
180 if (is_32bit_task()) {
181 vdso_spec = &vdso32_spec;
182 vdso_size = &vdso32_end - &vdso32_start;
183 } else {
184 vdso_spec = &vdso64_spec;
185 vdso_size = &vdso64_end - &vdso64_start;
186 }
187
188 mappings_size = vdso_size + vvar_size;
189 mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK;
190
191 /*
192 * Pick a base address for the vDSO in process space.
193 * Add enough to the size so that the result can be aligned.
194 */
195 vdso_base = get_unmapped_area(NULL, 0, mappings_size, 0, 0);
196 if (IS_ERR_VALUE(vdso_base))
197 return vdso_base;
198
199 /* Add required alignment. */
200 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
201
202 /*
203 * Put vDSO base into mm struct. We need to do this before calling
204 * install_special_mapping or the perf counter mmap tracking code
205 * will fail to recognise it as a vDSO.
206 */
207 mm->context.vdso = (void __user *)vdso_base + vvar_size;
208
209 vma = _install_special_mapping(mm, vdso_base, vvar_size,
210 VM_READ | VM_MAYREAD | VM_IO |
211 VM_DONTDUMP | VM_PFNMAP, &vvar_spec);
212 if (IS_ERR(vma))
213 return PTR_ERR(vma);
214
215 /*
216 * our vma flags don't have VM_WRITE so by default, the process isn't
217 * allowed to write those pages.
218 * gdb can break that with ptrace interface, and thus trigger COW on
219 * those pages but it's then your responsibility to never do that on
220 * the "data" page of the vDSO or you'll stop getting kernel updates
221 * and your nice userland gettimeofday will be totally dead.
222 * It's fine to use that for setting breakpoints in the vDSO code
223 * pages though.
224 */
225 vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size,
226 VM_READ | VM_EXEC | VM_MAYREAD |
227 VM_MAYWRITE | VM_MAYEXEC, vdso_spec);
228 if (IS_ERR(vma))
229 do_munmap(mm, vdso_base, vvar_size, NULL);
230
231 return PTR_ERR_OR_ZERO(vma);
232}
233
234int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
235{
236 struct mm_struct *mm = current->mm;
237 int rc;
238
239 mm->context.vdso = NULL;
240
241 if (mmap_write_lock_killable(mm))
242 return -EINTR;
243
244 rc = __arch_setup_additional_pages(bprm, uses_interp);
245 if (rc)
246 mm->context.vdso = NULL;
247
248 mmap_write_unlock(mm);
249 return rc;
250}
251
252#define VDSO_DO_FIXUPS(type, value, bits, sec) do { \
253 void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start); \
254 void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end); \
255 \
256 do_##type##_fixups((value), __start, __end); \
257} while (0)
258
259static void __init vdso_fixup_features(void)
260{
261#ifdef CONFIG_PPC64
262 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup);
263 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup);
264 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup);
265 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup);
266#endif /* CONFIG_PPC64 */
267
268#ifdef CONFIG_VDSO32
269 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup);
270 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup);
271#ifdef CONFIG_PPC64
272 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup);
273#endif /* CONFIG_PPC64 */
274 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup);
275#endif
276}
277
278/*
279 * Called from setup_arch to initialize the bitmap of available
280 * syscalls in the systemcfg page
281 */
282static void __init vdso_setup_syscall_map(void)
283{
284 unsigned int i;
285
286 for (i = 0; i < NR_syscalls; i++) {
287 if (sys_call_table[i] != (void *)&sys_ni_syscall)
288 vdso_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
289 if (IS_ENABLED(CONFIG_COMPAT) &&
290 compat_sys_call_table[i] != (void *)&sys_ni_syscall)
291 vdso_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
292 }
293}
294
295#ifdef CONFIG_PPC64
296int vdso_getcpu_init(void)
297{
298 unsigned long cpu, node, val;
299
300 /*
301 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
302 * in the next 16 bits. The VDSO uses this to implement getcpu().
303 */
304 cpu = get_cpu();
305 WARN_ON_ONCE(cpu > 0xffff);
306
307 node = cpu_to_node(cpu);
308 WARN_ON_ONCE(node > 0xffff);
309
310 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
311 mtspr(SPRN_SPRG_VDSO_WRITE, val);
312 get_paca()->sprg_vdso = val;
313
314 put_cpu();
315
316 return 0;
317}
318/* We need to call this before SMP init */
319early_initcall(vdso_getcpu_init);
320#endif
321
322static struct page ** __init vdso_setup_pages(void *start, void *end)
323{
324 int i;
325 struct page **pagelist;
326 int pages = (end - start) >> PAGE_SHIFT;
327
328 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
329 if (!pagelist)
330 panic("%s: Cannot allocate page list for VDSO", __func__);
331
332 for (i = 0; i < pages; i++)
333 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
334
335 return pagelist;
336}
337
338static int __init vdso_init(void)
339{
340#ifdef CONFIG_PPC64
341 /*
342 * Fill up the "systemcfg" stuff for backward compatibility
343 */
344 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
345 vdso_data->version.major = SYSTEMCFG_MAJOR;
346 vdso_data->version.minor = SYSTEMCFG_MINOR;
347 vdso_data->processor = mfspr(SPRN_PVR);
348 /*
349 * Fake the old platform number for pSeries and add
350 * in LPAR bit if necessary
351 */
352 vdso_data->platform = 0x100;
353 if (firmware_has_feature(FW_FEATURE_LPAR))
354 vdso_data->platform |= 1;
355 vdso_data->physicalMemorySize = memblock_phys_mem_size();
356 vdso_data->dcache_size = ppc64_caches.l1d.size;
357 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
358 vdso_data->icache_size = ppc64_caches.l1i.size;
359 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
360 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
361 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
362 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
363 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
364#endif /* CONFIG_PPC64 */
365
366 vdso_setup_syscall_map();
367
368 vdso_fixup_features();
369
370 if (IS_ENABLED(CONFIG_VDSO32))
371 vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end);
372
373 if (IS_ENABLED(CONFIG_PPC64))
374 vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end);
375
376 smp_wmb();
377
378 return 0;
379}
380arch_initcall(vdso_init);