Linux Audio

Check our new training course

Loading...
v6.2
  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);
v4.10.11
 
  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/cpu_has_feature.h>
 26#include <asm/pgtable.h>
 27#include <asm/processor.h>
 28#include <asm/mmu.h>
 29#include <asm/mmu_context.h>
 30#include <asm/prom.h>
 31#include <asm/machdep.h>
 32#include <asm/cputable.h>
 33#include <asm/sections.h>
 34#include <asm/firmware.h>
 35#include <asm/vdso.h>
 36#include <asm/vdso_datapage.h>
 37#include <asm/setup.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/* Max supported size for symbol names */
 48#define MAX_SYMNAME	64
 49
 50/* The alignment of the vDSO */
 51#define VDSO_ALIGNMENT	(1 << 16)
 52
 53static unsigned int vdso32_pages;
 54static void *vdso32_kbase;
 55static struct page **vdso32_pagelist;
 56unsigned long vdso32_sigtramp;
 57unsigned long vdso32_rt_sigtramp;
 58
 59#ifdef CONFIG_VDSO32
 60extern char vdso32_start, vdso32_end;
 61#endif
 62
 63#ifdef CONFIG_PPC64
 64extern char vdso64_start, vdso64_end;
 65static void *vdso64_kbase = &vdso64_start;
 66static unsigned int vdso64_pages;
 67static struct page **vdso64_pagelist;
 68unsigned long vdso64_rt_sigtramp;
 69#endif /* CONFIG_PPC64 */
 70
 71static int vdso_ready;
 72
 73/*
 74 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
 75 * Once the early boot kernel code no longer needs to muck around
 76 * with it, it will become dynamically allocated
 77 */
 78static union {
 79	struct vdso_data	data;
 80	u8			page[PAGE_SIZE];
 81} vdso_data_store __page_aligned_data;
 82struct vdso_data *vdso_data = &vdso_data_store.data;
 83
 84/* Format of the patch table */
 85struct vdso_patch_def
 86{
 87	unsigned long	ftr_mask, ftr_value;
 88	const char	*gen_name;
 89	const char	*fix_name;
 90};
 91
 92/* Table of functions to patch based on the CPU type/revision
 93 *
 94 * Currently, we only change sync_dicache to do nothing on processors
 95 * with a coherent icache
 96 */
 97static struct vdso_patch_def vdso_patches[] = {
 98	{
 99		CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
100		"__kernel_sync_dicache", "__kernel_sync_dicache_p5"
101	},
102	{
103		CPU_FTR_USE_TB, 0,
104		"__kernel_gettimeofday", NULL
105	},
106	{
107		CPU_FTR_USE_TB, 0,
108		"__kernel_clock_gettime", NULL
109	},
110	{
111		CPU_FTR_USE_TB, 0,
112		"__kernel_clock_getres", NULL
113	},
114	{
115		CPU_FTR_USE_TB, 0,
116		"__kernel_get_tbfreq", NULL
117	},
118	{
119		CPU_FTR_USE_TB, 0,
120		"__kernel_time", NULL
121	},
122};
123
124/*
125 * Some infos carried around for each of them during parsing at
126 * boot time.
127 */
128struct lib32_elfinfo
129{
130	Elf32_Ehdr	*hdr;		/* ptr to ELF */
131	Elf32_Sym	*dynsym;	/* ptr to .dynsym section */
132	unsigned long	dynsymsize;	/* size of .dynsym section */
133	char		*dynstr;	/* ptr to .dynstr section */
134	unsigned long	text;		/* offset of .text section in .so */
135};
136
137struct lib64_elfinfo
138{
139	Elf64_Ehdr	*hdr;
140	Elf64_Sym	*dynsym;
141	unsigned long	dynsymsize;
142	char		*dynstr;
143	unsigned long	text;
144};
145
146
147/*
148 * This is called from binfmt_elf, we create the special vma for the
149 * vDSO and insert it into the mm struct tree
150 */
151int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
152{
153	struct mm_struct *mm = current->mm;
154	struct page **vdso_pagelist;
155	unsigned long vdso_pages;
156	unsigned long vdso_base;
157	int rc;
158
159	if (!vdso_ready)
160		return 0;
161
162#ifdef CONFIG_PPC64
163	if (is_32bit_task()) {
164		vdso_pagelist = vdso32_pagelist;
165		vdso_pages = vdso32_pages;
166		vdso_base = VDSO32_MBASE;
167	} else {
168		vdso_pagelist = vdso64_pagelist;
169		vdso_pages = vdso64_pages;
170		/*
171		 * On 64bit we don't have a preferred map address. This
172		 * allows get_unmapped_area to find an area near other mmaps
173		 * and most likely share a SLB entry.
174		 */
175		vdso_base = 0;
176	}
177#else
178	vdso_pagelist = vdso32_pagelist;
179	vdso_pages = vdso32_pages;
180	vdso_base = VDSO32_MBASE;
181#endif
182
183	current->mm->context.vdso_base = 0;
184
185	/* vDSO has a problem and was disabled, just don't "enable" it for the
186	 * process
187	 */
188	if (vdso_pages == 0)
189		return 0;
190	/* Add a page to the vdso size for the data page */
191	vdso_pages ++;
192
193	/*
194	 * pick a base address for the vDSO in process space. We try to put it
195	 * at vdso_base which is the "natural" base for it, but we might fail
196	 * and end up putting it elsewhere.
197	 * Add enough to the size so that the result can be aligned.
198	 */
199	if (down_write_killable(&mm->mmap_sem))
200		return -EINTR;
201	vdso_base = get_unmapped_area(NULL, vdso_base,
202				      (vdso_pages << PAGE_SHIFT) +
203				      ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
204				      0, 0);
205	if (IS_ERR_VALUE(vdso_base)) {
206		rc = vdso_base;
207		goto fail_mmapsem;
208	}
209
210	/* Add required alignment. */
211	vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
212
213	/*
214	 * Put vDSO base into mm struct. We need to do this before calling
215	 * install_special_mapping or the perf counter mmap tracking code
216	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
217	 */
218	current->mm->context.vdso_base = vdso_base;
219
220	/*
221	 * our vma flags don't have VM_WRITE so by default, the process isn't
222	 * allowed to write those pages.
223	 * gdb can break that with ptrace interface, and thus trigger COW on
224	 * those pages but it's then your responsibility to never do that on
225	 * the "data" page of the vDSO or you'll stop getting kernel updates
226	 * and your nice userland gettimeofday will be totally dead.
227	 * It's fine to use that for setting breakpoints in the vDSO code
228	 * pages though.
229	 */
230	rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
231				     VM_READ|VM_EXEC|
232				     VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
233				     vdso_pagelist);
234	if (rc) {
235		current->mm->context.vdso_base = 0;
236		goto fail_mmapsem;
237	}
238
239	up_write(&mm->mmap_sem);
240	return 0;
 
241
242 fail_mmapsem:
243	up_write(&mm->mmap_sem);
244	return rc;
245}
246
247const char *arch_vma_name(struct vm_area_struct *vma)
248{
249	if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
250		return "[vdso]";
251	return NULL;
252}
253
 
 
254
 
 
 
 
255
256#ifdef CONFIG_VDSO32
257static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
258				  unsigned long *size)
259{
260	Elf32_Shdr *sechdrs;
261	unsigned int i;
262	char *secnames;
263
264	/* Grab section headers and strings so we can tell who is who */
265	sechdrs = (void *)ehdr + ehdr->e_shoff;
266	secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
267
268	/* Find the section they want */
269	for (i = 1; i < ehdr->e_shnum; i++) {
270		if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
271			if (size)
272				*size = sechdrs[i].sh_size;
273			return (void *)ehdr + sechdrs[i].sh_offset;
274		}
275	}
276	*size = 0;
277	return NULL;
278}
279
280static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
281					const char *symname)
282{
283	unsigned int i;
284	char name[MAX_SYMNAME], *c;
285
286	for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
287		if (lib->dynsym[i].st_name == 0)
288			continue;
289		strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
290			MAX_SYMNAME);
291		c = strchr(name, '@');
292		if (c)
293			*c = 0;
294		if (strcmp(symname, name) == 0)
295			return &lib->dynsym[i];
296	}
297	return NULL;
298}
299
300/* Note that we assume the section is .text and the symbol is relative to
301 * the library base
 
 
 
 
302 */
303static unsigned long __init find_function32(struct lib32_elfinfo *lib,
304					    const char *symname)
305{
306	Elf32_Sym *sym = find_symbol32(lib, symname);
 
 
307
308	if (sym == NULL) {
309		printk(KERN_WARNING "vDSO32: function %s not found !\n",
310		       symname);
311		return 0;
312	}
313	return sym->st_value - VDSO32_LBASE;
314}
315
316static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
317				       struct lib64_elfinfo *v64,
318				       const char *orig, const char *fix)
319{
320	Elf32_Sym *sym32_gen, *sym32_fix;
321
322	sym32_gen = find_symbol32(v32, orig);
323	if (sym32_gen == NULL) {
324		printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
325		return -1;
326	}
327	if (fix == NULL) {
328		sym32_gen->st_name = 0;
329		return 0;
330	}
331	sym32_fix = find_symbol32(v32, fix);
332	if (sym32_fix == NULL) {
333		printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
334		return -1;
335	}
336	sym32_gen->st_value = sym32_fix->st_value;
337	sym32_gen->st_size = sym32_fix->st_size;
338	sym32_gen->st_info = sym32_fix->st_info;
339	sym32_gen->st_other = sym32_fix->st_other;
340	sym32_gen->st_shndx = sym32_fix->st_shndx;
341
342	return 0;
343}
344#else /* !CONFIG_VDSO32 */
345static unsigned long __init find_function32(struct lib32_elfinfo *lib,
346					    const char *symname)
347{
348	return 0;
349}
350
351static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
352				       struct lib64_elfinfo *v64,
353				       const char *orig, const char *fix)
354{
355	return 0;
356}
357#endif /* CONFIG_VDSO32 */
358
359
360#ifdef CONFIG_PPC64
361
362static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
363				  unsigned long *size)
364{
365	Elf64_Shdr *sechdrs;
366	unsigned int i;
367	char *secnames;
368
369	/* Grab section headers and strings so we can tell who is who */
370	sechdrs = (void *)ehdr + ehdr->e_shoff;
371	secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
372
373	/* Find the section they want */
374	for (i = 1; i < ehdr->e_shnum; i++) {
375		if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
376			if (size)
377				*size = sechdrs[i].sh_size;
378			return (void *)ehdr + sechdrs[i].sh_offset;
379		}
 
 
380	}
381	if (size)
382		*size = 0;
383	return NULL;
384}
385
386static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
387					const char *symname)
388{
389	unsigned int i;
390	char name[MAX_SYMNAME], *c;
391
392	for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
393		if (lib->dynsym[i].st_name == 0)
394			continue;
395		strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
396			MAX_SYMNAME);
397		c = strchr(name, '@');
398		if (c)
399			*c = 0;
400		if (strcmp(symname, name) == 0)
401			return &lib->dynsym[i];
402	}
403	return NULL;
404}
405
406/* Note that we assume the section is .text and the symbol is relative to
407 * the library base
 
408 */
409static unsigned long __init find_function64(struct lib64_elfinfo *lib,
410					    const char *symname)
411{
412	Elf64_Sym *sym = find_symbol64(lib, symname);
 
 
 
 
413
414	if (sym == NULL) {
415		printk(KERN_WARNING "vDSO64: function %s not found !\n",
416		       symname);
417		return 0;
 
 
418	}
419#ifdef VDS64_HAS_DESCRIPTORS
420	return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
421		VDSO64_LBASE;
422#else
423	return sym->st_value - VDSO64_LBASE;
424#endif
425}
426
427static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
428				       struct lib64_elfinfo *v64,
429				       const char *orig, const char *fix)
430{
431	Elf64_Sym *sym64_gen, *sym64_fix;
432
433	sym64_gen = find_symbol64(v64, orig);
434	if (sym64_gen == NULL) {
435		printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
436		return -1;
437	}
438	if (fix == NULL) {
439		sym64_gen->st_name = 0;
440		return 0;
441	}
442	sym64_fix = find_symbol64(v64, fix);
443	if (sym64_fix == NULL) {
444		printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
445		return -1;
446	}
447	sym64_gen->st_value = sym64_fix->st_value;
448	sym64_gen->st_size = sym64_fix->st_size;
449	sym64_gen->st_info = sym64_fix->st_info;
450	sym64_gen->st_other = sym64_fix->st_other;
451	sym64_gen->st_shndx = sym64_fix->st_shndx;
452
453	return 0;
454}
 
 
 
 
 
455
456#endif /* CONFIG_PPC64 */
457
458
459static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
460					struct lib64_elfinfo *v64)
461{
462	void *sect;
463
464	/*
465	 * Locate symbol tables & text section
 
 
466	 */
 
467
468#ifdef CONFIG_VDSO32
469	v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
470	v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
471	if (v32->dynsym == NULL || v32->dynstr == NULL) {
472		printk(KERN_ERR "vDSO32: required symbol section not found\n");
473		return -1;
474	}
475	sect = find_section32(v32->hdr, ".text", NULL);
476	if (sect == NULL) {
477		printk(KERN_ERR "vDSO32: the .text section was not found\n");
478		return -1;
479	}
480	v32->text = sect - vdso32_kbase;
481#endif
482
483#ifdef CONFIG_PPC64
484	v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
485	v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
486	if (v64->dynsym == NULL || v64->dynstr == NULL) {
487		printk(KERN_ERR "vDSO64: required symbol section not found\n");
488		return -1;
489	}
490	sect = find_section64(v64->hdr, ".text", NULL);
491	if (sect == NULL) {
492		printk(KERN_ERR "vDSO64: the .text section was not found\n");
493		return -1;
494	}
495	v64->text = sect - vdso64_kbase;
496#endif /* CONFIG_PPC64 */
497
498	return 0;
499}
500
501static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
502					  struct lib64_elfinfo *v64)
503{
504	/*
505	 * Find signal trampolines
 
 
 
 
 
 
 
506	 */
 
 
 
 
 
507
508#ifdef CONFIG_PPC64
509	vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
510#endif
511	vdso32_sigtramp	   = find_function32(v32, "__kernel_sigtramp32");
512	vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
513}
514
515static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
516				       struct lib64_elfinfo *v64)
517{
518#ifdef CONFIG_VDSO32
519	Elf32_Sym *sym32;
520#endif
521#ifdef CONFIG_PPC64
522	Elf64_Sym *sym64;
523
524       	sym64 = find_symbol64(v64, "__kernel_datapage_offset");
525	if (sym64 == NULL) {
526		printk(KERN_ERR "vDSO64: Can't find symbol "
527		       "__kernel_datapage_offset !\n");
528		return -1;
529	}
530	*((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
531		(vdso64_pages << PAGE_SHIFT) -
532		(sym64->st_value - VDSO64_LBASE);
533#endif /* CONFIG_PPC64 */
534
535#ifdef CONFIG_VDSO32
536	sym32 = find_symbol32(v32, "__kernel_datapage_offset");
537	if (sym32 == NULL) {
538		printk(KERN_ERR "vDSO32: Can't find symbol "
539		       "__kernel_datapage_offset !\n");
540		return -1;
541	}
542	*((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
543		(vdso32_pages << PAGE_SHIFT) -
544		(sym32->st_value - VDSO32_LBASE);
545#endif
546
547	return 0;
 
548}
549
550
551static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
552				      struct lib64_elfinfo *v64)
 
 
 
 
 
553{
554	unsigned long size;
555	void *start;
556
557#ifdef CONFIG_PPC64
558	start = find_section64(v64->hdr, "__ftr_fixup", &size);
559	if (start)
560		do_feature_fixups(cur_cpu_spec->cpu_features,
561				  start, start + size);
562
563	start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
564	if (start)
565		do_feature_fixups(cur_cpu_spec->mmu_features,
566				  start, start + size);
567
568	start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
569	if (start)
570		do_feature_fixups(powerpc_firmware_features,
571				  start, start + size);
572
573	start = find_section64(v64->hdr, "__lwsync_fixup", &size);
574	if (start)
575		do_lwsync_fixups(cur_cpu_spec->cpu_features,
576				 start, start + size);
577#endif /* CONFIG_PPC64 */
578
579#ifdef CONFIG_VDSO32
580	start = find_section32(v32->hdr, "__ftr_fixup", &size);
581	if (start)
582		do_feature_fixups(cur_cpu_spec->cpu_features,
583				  start, start + size);
584
585	start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
586	if (start)
587		do_feature_fixups(cur_cpu_spec->mmu_features,
588				  start, start + size);
589
590#ifdef CONFIG_PPC64
591	start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
592	if (start)
593		do_feature_fixups(powerpc_firmware_features,
594				  start, start + size);
595#endif /* CONFIG_PPC64 */
596
597	start = find_section32(v32->hdr, "__lwsync_fixup", &size);
598	if (start)
599		do_lwsync_fixups(cur_cpu_spec->cpu_features,
600				 start, start + size);
601#endif
602
603	return 0;
604}
605
606static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
607				       struct lib64_elfinfo *v64)
608{
609	int i;
610
611	for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
612		struct vdso_patch_def *patch = &vdso_patches[i];
613		int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
614			== patch->ftr_value;
615		if (!match)
616			continue;
617
618		DBG("replacing %s with %s...\n", patch->gen_name,
619		    patch->fix_name ? "NONE" : patch->fix_name);
620
621		/*
622		 * Patch the 32 bits and 64 bits symbols. Note that we do not
623		 * patch the "." symbol on 64 bits.
624		 * It would be easy to do, but doesn't seem to be necessary,
625		 * patching the OPD symbol is enough.
626		 */
627		vdso_do_func_patch32(v32, v64, patch->gen_name,
628				     patch->fix_name);
629#ifdef CONFIG_PPC64
630		vdso_do_func_patch64(v32, v64, patch->gen_name,
631				     patch->fix_name);
632#endif /* CONFIG_PPC64 */
633	}
634
635	return 0;
636}
637
638
639static __init int vdso_setup(void)
640{
641	struct lib32_elfinfo	v32;
642	struct lib64_elfinfo	v64;
643
644	v32.hdr = vdso32_kbase;
645#ifdef CONFIG_PPC64
646	v64.hdr = vdso64_kbase;
647#endif
648	if (vdso_do_find_sections(&v32, &v64))
649		return -1;
650
651	if (vdso_fixup_datapage(&v32, &v64))
652		return -1;
653
654	if (vdso_fixup_features(&v32, &v64))
655		return -1;
656
657	if (vdso_fixup_alt_funcs(&v32, &v64))
658		return -1;
659
660	vdso_setup_trampolines(&v32, &v64);
661
662	return 0;
663}
664
665/*
666 * Called from setup_arch to initialize the bitmap of available
667 * syscalls in the systemcfg page
668 */
669static void __init vdso_setup_syscall_map(void)
670{
671	unsigned int i;
672	extern unsigned long *sys_call_table;
673	extern unsigned long sys_ni_syscall;
674
675
676	for (i = 0; i < NR_syscalls; i++) {
677#ifdef CONFIG_PPC64
678		if (sys_call_table[i*2] != sys_ni_syscall)
679			vdso_data->syscall_map_64[i >> 5] |=
680				0x80000000UL >> (i & 0x1f);
681		if (sys_call_table[i*2+1] != sys_ni_syscall)
682			vdso_data->syscall_map_32[i >> 5] |=
683				0x80000000UL >> (i & 0x1f);
684#else /* CONFIG_PPC64 */
685		if (sys_call_table[i] != sys_ni_syscall)
686			vdso_data->syscall_map_32[i >> 5] |=
687				0x80000000UL >> (i & 0x1f);
688#endif /* CONFIG_PPC64 */
689	}
690}
691
692#ifdef CONFIG_PPC64
693int vdso_getcpu_init(void)
694{
695	unsigned long cpu, node, val;
696
697	/*
698	 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
699	 * in the next 16 bits.  The VDSO uses this to implement getcpu().
700	 */
701	cpu = get_cpu();
702	WARN_ON_ONCE(cpu > 0xffff);
703
704	node = cpu_to_node(cpu);
705	WARN_ON_ONCE(node > 0xffff);
706
707	val = (cpu & 0xfff) | ((node & 0xffff) << 16);
708	mtspr(SPRN_SPRG_VDSO_WRITE, val);
709	get_paca()->sprg_vdso = val;
710
711	put_cpu();
712
713	return 0;
714}
715/* We need to call this before SMP init */
716early_initcall(vdso_getcpu_init);
717#endif
718
719static int __init vdso_init(void)
720{
721	int i;
 
 
 
 
 
 
 
 
 
722
 
 
 
 
 
723#ifdef CONFIG_PPC64
724	/*
725	 * Fill up the "systemcfg" stuff for backward compatibility
726	 */
727	strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
728	vdso_data->version.major = SYSTEMCFG_MAJOR;
729	vdso_data->version.minor = SYSTEMCFG_MINOR;
730	vdso_data->processor = mfspr(SPRN_PVR);
731	/*
732	 * Fake the old platform number for pSeries and add
733	 * in LPAR bit if necessary
734	 */
735	vdso_data->platform = 0x100;
736	if (firmware_has_feature(FW_FEATURE_LPAR))
737		vdso_data->platform |= 1;
738	vdso_data->physicalMemorySize = memblock_phys_mem_size();
739	vdso_data->dcache_size = ppc64_caches.dsize;
740	vdso_data->dcache_line_size = ppc64_caches.dline_size;
741	vdso_data->icache_size = ppc64_caches.isize;
742	vdso_data->icache_line_size = ppc64_caches.iline_size;
743
744	/* XXXOJN: Blocks should be added to ppc64_caches and used instead */
745	vdso_data->dcache_block_size = ppc64_caches.dline_size;
746	vdso_data->icache_block_size = ppc64_caches.iline_size;
747	vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
748	vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
749
750	/*
751	 * Calculate the size of the 64 bits vDSO
752	 */
753	vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
754	DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
755#else
756	vdso_data->dcache_block_size = L1_CACHE_BYTES;
757	vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
758	vdso_data->icache_block_size = L1_CACHE_BYTES;
759	vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
760#endif /* CONFIG_PPC64 */
761
762
763#ifdef CONFIG_VDSO32
764	vdso32_kbase = &vdso32_start;
765
766	/*
767	 * Calculate the size of the 32 bits vDSO
768	 */
769	vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
770	DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
771#endif
772
773
774	/*
775	 * Setup the syscall map in the vDOS
776	 */
777	vdso_setup_syscall_map();
778
779	/*
780	 * Initialize the vDSO images in memory, that is do necessary
781	 * fixups of vDSO symbols, locate trampolines, etc...
782	 */
783	if (vdso_setup()) {
784		printk(KERN_ERR "vDSO setup failure, not enabled !\n");
785		vdso32_pages = 0;
786#ifdef CONFIG_PPC64
787		vdso64_pages = 0;
788#endif
789		return 0;
790	}
791
792#ifdef CONFIG_VDSO32
793	/* Make sure pages are in the correct state */
794	vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
795				  GFP_KERNEL);
796	BUG_ON(vdso32_pagelist == NULL);
797	for (i = 0; i < vdso32_pages; i++) {
798		struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
799		ClearPageReserved(pg);
800		get_page(pg);
801		vdso32_pagelist[i] = pg;
802	}
803	vdso32_pagelist[i++] = virt_to_page(vdso_data);
804	vdso32_pagelist[i] = NULL;
805#endif
806
807#ifdef CONFIG_PPC64
808	vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
809				  GFP_KERNEL);
810	BUG_ON(vdso64_pagelist == NULL);
811	for (i = 0; i < vdso64_pages; i++) {
812		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
813		ClearPageReserved(pg);
814		get_page(pg);
815		vdso64_pagelist[i] = pg;
816	}
817	vdso64_pagelist[i++] = virt_to_page(vdso_data);
818	vdso64_pagelist[i] = NULL;
819#endif /* CONFIG_PPC64 */
820
821	get_page(virt_to_page(vdso_data));
 
822
823	smp_wmb();
824	vdso_ready = 1;
825
826	return 0;
827}
828arch_initcall(vdso_init);