1/*
  2 * vdso setup for s390
  3 *
  4 *  Copyright IBM Corp. 2008
  5 *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
  6 *
  7 * This program is free software; you can redistribute it and/or modify
  8 * it under the terms of the GNU General Public License (version 2 only)
  9 * as published by the Free Software Foundation.
 10 */
 11
 12#include <linux/module.h>
 13#include <linux/errno.h>
 14#include <linux/sched.h>
 15#include <linux/kernel.h>
 16#include <linux/mm.h>
 17#include <linux/smp.h>
 18#include <linux/stddef.h>
 19#include <linux/unistd.h>
 20#include <linux/slab.h>
 21#include <linux/user.h>
 22#include <linux/elf.h>
 23#include <linux/security.h>
 24#include <linux/bootmem.h>
 25#include <linux/compat.h>
 26#include <asm/asm-offsets.h>
 27#include <asm/pgtable.h>
 28#include <asm/system.h>
 29#include <asm/processor.h>
 30#include <asm/mmu.h>
 31#include <asm/mmu_context.h>
 32#include <asm/sections.h>
 33#include <asm/vdso.h>
 
 34
 35#if defined(CONFIG_32BIT) || defined(CONFIG_COMPAT)
 36extern char vdso32_start, vdso32_end;
 37static void *vdso32_kbase = &vdso32_start;
 38static unsigned int vdso32_pages;
 39static struct page **vdso32_pagelist;
 40#endif
 41
 42#ifdef CONFIG_64BIT
 43extern char vdso64_start, vdso64_end;
 44static void *vdso64_kbase = &vdso64_start;
 45static unsigned int vdso64_pages;
 46static struct page **vdso64_pagelist;
 47#endif /* CONFIG_64BIT */
 48
 49/*
 50 * Should the kernel map a VDSO page into processes and pass its
 51 * address down to glibc upon exec()?
 52 */
 53unsigned int __read_mostly vdso_enabled = 1;
 54
 55static int __init vdso_setup(char *s)
 56{
 57	unsigned long val;
 58	int rc;
 59
 60	rc = 0;
 61	if (strncmp(s, "on", 3) == 0)
 62		vdso_enabled = 1;
 63	else if (strncmp(s, "off", 4) == 0)
 64		vdso_enabled = 0;
 65	else {
 66		rc = strict_strtoul(s, 0, &val);
 67		vdso_enabled = rc ? 0 : !!val;
 68	}
 69	return !rc;
 70}
 71__setup("vdso=", vdso_setup);
 72
 73/*
 74 * The vdso data page
 75 */
 76static union {
 77	struct vdso_data	data;
 78	u8			page[PAGE_SIZE];
 79} vdso_data_store __page_aligned_data;
 80struct vdso_data *vdso_data = &vdso_data_store.data;
 81
 82/*
 83 * Setup vdso data page.
 84 */
 85static void vdso_init_data(struct vdso_data *vd)
 86{
 87	vd->ectg_available = user_mode != HOME_SPACE_MODE && test_facility(31);
 88}
 89
 90#ifdef CONFIG_64BIT
 91/*
 92 * Setup per cpu vdso data page.
 93 */
 94static void vdso_init_per_cpu_data(int cpu, struct vdso_per_cpu_data *vpcd)
 95{
 
 96}
 97
 98/*
 99 * Allocate/free per cpu vdso data.
100 */
101#define SEGMENT_ORDER	2
102
103int vdso_alloc_per_cpu(int cpu, struct _lowcore *lowcore)
104{
105	unsigned long segment_table, page_table, page_frame;
 
106	u32 *psal, *aste;
107	int i;
108
109	lowcore->vdso_per_cpu_data = __LC_PASTE;
110
111	if (user_mode == HOME_SPACE_MODE || !vdso_enabled)
112		return 0;
113
114	segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
115	page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
116	page_frame = get_zeroed_page(GFP_KERNEL);
117	if (!segment_table || !page_table || !page_frame)
118		goto out;
119
 
 
 
 
 
 
120	clear_table((unsigned long *) segment_table, _SEGMENT_ENTRY_EMPTY,
121		    PAGE_SIZE << SEGMENT_ORDER);
122	clear_table((unsigned long *) page_table, _PAGE_TYPE_EMPTY,
123		    256*sizeof(unsigned long));
124
125	*(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
126	*(unsigned long *) page_table = _PAGE_RO + page_frame;
127
128	psal = (u32 *) (page_table + 256*sizeof(unsigned long));
129	aste = psal + 32;
130
131	for (i = 4; i < 32; i += 4)
132		psal[i] = 0x80000000;
133
134	lowcore->paste[4] = (u32)(addr_t) psal;
135	psal[0] = 0x20000000;
136	psal[2] = (u32)(addr_t) aste;
137	*(unsigned long *) (aste + 2) = segment_table +
138		_ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
139	aste[4] = (u32)(addr_t) psal;
140	lowcore->vdso_per_cpu_data = page_frame;
141
142	vdso_init_per_cpu_data(cpu, (struct vdso_per_cpu_data *) page_frame);
143	return 0;
144
145out:
146	free_page(page_frame);
147	free_page(page_table);
148	free_pages(segment_table, SEGMENT_ORDER);
149	return -ENOMEM;
150}
151
152void vdso_free_per_cpu(int cpu, struct _lowcore *lowcore)
153{
154	unsigned long segment_table, page_table, page_frame;
155	u32 *psal, *aste;
156
157	if (user_mode == HOME_SPACE_MODE || !vdso_enabled)
158		return;
159
160	psal = (u32 *)(addr_t) lowcore->paste[4];
161	aste = (u32 *)(addr_t) psal[2];
162	segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
163	page_table = *(unsigned long *) segment_table;
164	page_frame = *(unsigned long *) page_table;
165
166	free_page(page_frame);
167	free_page(page_table);
168	free_pages(segment_table, SEGMENT_ORDER);
169}
170
171static void __vdso_init_cr5(void *dummy)
172{
173	unsigned long cr5;
174
175	cr5 = offsetof(struct _lowcore, paste);
 
 
176	__ctl_load(cr5, 5, 5);
177}
178
179static void vdso_init_cr5(void)
180{
181	if (user_mode != HOME_SPACE_MODE && vdso_enabled)
182		on_each_cpu(__vdso_init_cr5, NULL, 1);
183}
184#endif /* CONFIG_64BIT */
185
186/*
187 * This is called from binfmt_elf, we create the special vma for the
188 * vDSO and insert it into the mm struct tree
189 */
190int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
191{
192	struct mm_struct *mm = current->mm;
193	struct page **vdso_pagelist;
194	unsigned long vdso_pages;
195	unsigned long vdso_base;
196	int rc;
197
198	if (!vdso_enabled)
199		return 0;
200	/*
201	 * Only map the vdso for dynamically linked elf binaries.
202	 */
203	if (!uses_interp)
204		return 0;
205
206#ifdef CONFIG_64BIT
207	vdso_pagelist = vdso64_pagelist;
208	vdso_pages = vdso64_pages;
209#ifdef CONFIG_COMPAT
210	if (is_compat_task()) {
211		vdso_pagelist = vdso32_pagelist;
212		vdso_pages = vdso32_pages;
213	}
214#endif
215#else
216	vdso_pagelist = vdso32_pagelist;
217	vdso_pages = vdso32_pages;
218#endif
219
220	/*
221	 * vDSO has a problem and was disabled, just don't "enable" it for
222	 * the process
223	 */
224	if (vdso_pages == 0)
225		return 0;
226
227	current->mm->context.vdso_base = 0;
228
229	/*
230	 * pick a base address for the vDSO in process space. We try to put
231	 * it at vdso_base which is the "natural" base for it, but we might
232	 * fail and end up putting it elsewhere.
233	 */
234	down_write(&mm->mmap_sem);
235	vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
236	if (IS_ERR_VALUE(vdso_base)) {
237		rc = vdso_base;
238		goto out_up;
239	}
240
241	/*
242	 * Put vDSO base into mm struct. We need to do this before calling
243	 * install_special_mapping or the perf counter mmap tracking code
244	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
245	 */
246	current->mm->context.vdso_base = vdso_base;
247
248	/*
249	 * our vma flags don't have VM_WRITE so by default, the process
250	 * isn't allowed to write those pages.
251	 * gdb can break that with ptrace interface, and thus trigger COW
252	 * on those pages but it's then your responsibility to never do that
253	 * on the "data" page of the vDSO or you'll stop getting kernel
254	 * updates and your nice userland gettimeofday will be totally dead.
255	 * It's fine to use that for setting breakpoints in the vDSO code
256	 * pages though
257	 *
258	 * Make sure the vDSO gets into every core dump.
259	 * Dumping its contents makes post-mortem fully interpretable later
260	 * without matching up the same kernel and hardware config to see
261	 * what PC values meant.
262	 */
263	rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
264				     VM_READ|VM_EXEC|
265				     VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
266				     VM_ALWAYSDUMP,
267				     vdso_pagelist);
268	if (rc)
269		current->mm->context.vdso_base = 0;
270out_up:
271	up_write(&mm->mmap_sem);
272	return rc;
273}
274
275const char *arch_vma_name(struct vm_area_struct *vma)
276{
277	if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
278		return "[vdso]";
279	return NULL;
280}
281
282static int __init vdso_init(void)
283{
284	int i;
285
286	if (!vdso_enabled)
287		return 0;
288	vdso_init_data(vdso_data);
289#if defined(CONFIG_32BIT) || defined(CONFIG_COMPAT)
290	/* Calculate the size of the 32 bit vDSO */
291	vdso32_pages = ((&vdso32_end - &vdso32_start
292			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
293
294	/* Make sure pages are in the correct state */
295	vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
296				  GFP_KERNEL);
297	BUG_ON(vdso32_pagelist == NULL);
298	for (i = 0; i < vdso32_pages - 1; i++) {
299		struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
300		ClearPageReserved(pg);
301		get_page(pg);
302		vdso32_pagelist[i] = pg;
303	}
304	vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
305	vdso32_pagelist[vdso32_pages] = NULL;
306#endif
307
308#ifdef CONFIG_64BIT
309	/* Calculate the size of the 64 bit vDSO */
310	vdso64_pages = ((&vdso64_end - &vdso64_start
311			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
312
313	/* Make sure pages are in the correct state */
314	vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
315				  GFP_KERNEL);
316	BUG_ON(vdso64_pagelist == NULL);
317	for (i = 0; i < vdso64_pages - 1; i++) {
318		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
319		ClearPageReserved(pg);
320		get_page(pg);
321		vdso64_pagelist[i] = pg;
322	}
323	vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
324	vdso64_pagelist[vdso64_pages] = NULL;
325#ifndef CONFIG_SMP
326	if (vdso_alloc_per_cpu(0, &S390_lowcore))
327		BUG();
328#endif
329	vdso_init_cr5();
330#endif /* CONFIG_64BIT */
331
332	get_page(virt_to_page(vdso_data));
333
334	smp_wmb();
335
336	return 0;
337}
338arch_initcall(vdso_init);
339
340int in_gate_area_no_mm(unsigned long addr)
341{
342	return 0;
343}
344
345int in_gate_area(struct mm_struct *mm, unsigned long addr)
346{
347	return 0;
348}
349
350struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
351{
352	return NULL;
353}

  1/*
  2 * vdso setup for s390
  3 *
  4 *  Copyright IBM Corp. 2008
  5 *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
  6 *
  7 * This program is free software; you can redistribute it and/or modify
  8 * it under the terms of the GNU General Public License (version 2 only)
  9 * as published by the Free Software Foundation.
 10 */
 11
 12#include <linux/module.h>
 13#include <linux/errno.h>
 14#include <linux/sched.h>
 15#include <linux/kernel.h>
 16#include <linux/mm.h>
 17#include <linux/smp.h>
 18#include <linux/stddef.h>
 19#include <linux/unistd.h>
 20#include <linux/slab.h>
 21#include <linux/user.h>
 22#include <linux/elf.h>
 23#include <linux/security.h>
 24#include <linux/bootmem.h>
 25#include <linux/compat.h>
 26#include <asm/asm-offsets.h>
 27#include <asm/pgtable.h>
 
 28#include <asm/processor.h>
 29#include <asm/mmu.h>
 30#include <asm/mmu_context.h>
 31#include <asm/sections.h>
 32#include <asm/vdso.h>
 33#include <asm/facility.h>
 34
 35#ifdef CONFIG_COMPAT
 36extern char vdso32_start, vdso32_end;
 37static void *vdso32_kbase = &vdso32_start;
 38static unsigned int vdso32_pages;
 39static struct page **vdso32_pagelist;
 40#endif
 41
 
 42extern char vdso64_start, vdso64_end;
 43static void *vdso64_kbase = &vdso64_start;
 44static unsigned int vdso64_pages;
 45static struct page **vdso64_pagelist;
 
 46
 47/*
 48 * Should the kernel map a VDSO page into processes and pass its
 49 * address down to glibc upon exec()?
 50 */
 51unsigned int __read_mostly vdso_enabled = 1;
 52
 53static int __init vdso_setup(char *s)
 54{
 55	unsigned long val;
 56	int rc;
 57
 58	rc = 0;
 59	if (strncmp(s, "on", 3) == 0)
 60		vdso_enabled = 1;
 61	else if (strncmp(s, "off", 4) == 0)
 62		vdso_enabled = 0;
 63	else {
 64		rc = kstrtoul(s, 0, &val);
 65		vdso_enabled = rc ? 0 : !!val;
 66	}
 67	return !rc;
 68}
 69__setup("vdso=", vdso_setup);
 70
 71/*
 72 * The vdso data page
 73 */
 74static union {
 75	struct vdso_data	data;
 76	u8			page[PAGE_SIZE];
 77} vdso_data_store __page_aligned_data;
 78struct vdso_data *vdso_data = &vdso_data_store.data;
 79
 80/*
 81 * Setup vdso data page.
 82 */
 83static void __init vdso_init_data(struct vdso_data *vd)
 
 
 
 
 
 
 
 
 
 84{
 85	vd->ectg_available = test_facility(31);
 86}
 87
 88/*
 89 * Allocate/free per cpu vdso data.
 90 */
 91#define SEGMENT_ORDER	2
 92
 93int vdso_alloc_per_cpu(struct lowcore *lowcore)
 94{
 95	unsigned long segment_table, page_table, page_frame;
 96	struct vdso_per_cpu_data *vd;
 97	u32 *psal, *aste;
 98	int i;
 99
100	lowcore->vdso_per_cpu_data = __LC_PASTE;
101
102	if (!vdso_enabled)
103		return 0;
104
105	segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
106	page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
107	page_frame = get_zeroed_page(GFP_KERNEL);
108	if (!segment_table || !page_table || !page_frame)
109		goto out;
110
111	/* Initialize per-cpu vdso data page */
112	vd = (struct vdso_per_cpu_data *) page_frame;
113	vd->cpu_nr = lowcore->cpu_nr;
114	vd->node_id = cpu_to_node(vd->cpu_nr);
115
116	/* Set up access register mode page table */
117	clear_table((unsigned long *) segment_table, _SEGMENT_ENTRY_EMPTY,
118		    PAGE_SIZE << SEGMENT_ORDER);
119	clear_table((unsigned long *) page_table, _PAGE_INVALID,
120		    256*sizeof(unsigned long));
121
122	*(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
123	*(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
124
125	psal = (u32 *) (page_table + 256*sizeof(unsigned long));
126	aste = psal + 32;
127
128	for (i = 4; i < 32; i += 4)
129		psal[i] = 0x80000000;
130
131	lowcore->paste[4] = (u32)(addr_t) psal;
132	psal[0] = 0x02000000;
133	psal[2] = (u32)(addr_t) aste;
134	*(unsigned long *) (aste + 2) = segment_table +
135		_ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
136	aste[4] = (u32)(addr_t) psal;
137	lowcore->vdso_per_cpu_data = page_frame;
138
 
139	return 0;
140
141out:
142	free_page(page_frame);
143	free_page(page_table);
144	free_pages(segment_table, SEGMENT_ORDER);
145	return -ENOMEM;
146}
147
148void vdso_free_per_cpu(struct lowcore *lowcore)
149{
150	unsigned long segment_table, page_table, page_frame;
151	u32 *psal, *aste;
152
153	if (!vdso_enabled)
154		return;
155
156	psal = (u32 *)(addr_t) lowcore->paste[4];
157	aste = (u32 *)(addr_t) psal[2];
158	segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
159	page_table = *(unsigned long *) segment_table;
160	page_frame = *(unsigned long *) page_table;
161
162	free_page(page_frame);
163	free_page(page_table);
164	free_pages(segment_table, SEGMENT_ORDER);
165}
166
167static void vdso_init_cr5(void)
168{
169	unsigned long cr5;
170
171	if (!vdso_enabled)
172		return;
173	cr5 = offsetof(struct lowcore, paste);
174	__ctl_load(cr5, 5, 5);
175}
176
 
 
 
 
 
 
 
177/*
178 * This is called from binfmt_elf, we create the special vma for the
179 * vDSO and insert it into the mm struct tree
180 */
181int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
182{
183	struct mm_struct *mm = current->mm;
184	struct page **vdso_pagelist;
185	unsigned long vdso_pages;
186	unsigned long vdso_base;
187	int rc;
188
189	if (!vdso_enabled)
190		return 0;
191	/*
192	 * Only map the vdso for dynamically linked elf binaries.
193	 */
194	if (!uses_interp)
195		return 0;
196
 
197	vdso_pagelist = vdso64_pagelist;
198	vdso_pages = vdso64_pages;
199#ifdef CONFIG_COMPAT
200	if (is_compat_task()) {
201		vdso_pagelist = vdso32_pagelist;
202		vdso_pages = vdso32_pages;
203	}
204#endif
 
 
 
 
 
205	/*
206	 * vDSO has a problem and was disabled, just don't "enable" it for
207	 * the process
208	 */
209	if (vdso_pages == 0)
210		return 0;
211
212	current->mm->context.vdso_base = 0;
213
214	/*
215	 * pick a base address for the vDSO in process space. We try to put
216	 * it at vdso_base which is the "natural" base for it, but we might
217	 * fail and end up putting it elsewhere.
218	 */
219	down_write(&mm->mmap_sem);
220	vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
221	if (IS_ERR_VALUE(vdso_base)) {
222		rc = vdso_base;
223		goto out_up;
224	}
225
226	/*
227	 * Put vDSO base into mm struct. We need to do this before calling
228	 * install_special_mapping or the perf counter mmap tracking code
229	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
230	 */
231	current->mm->context.vdso_base = vdso_base;
232
233	/*
234	 * our vma flags don't have VM_WRITE so by default, the process
235	 * isn't allowed to write those pages.
236	 * gdb can break that with ptrace interface, and thus trigger COW
237	 * on those pages but it's then your responsibility to never do that
238	 * on the "data" page of the vDSO or you'll stop getting kernel
239	 * updates and your nice userland gettimeofday will be totally dead.
240	 * It's fine to use that for setting breakpoints in the vDSO code
241	 * pages though.
 
 
 
 
 
242	 */
243	rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
244				     VM_READ|VM_EXEC|
245				     VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 
246				     vdso_pagelist);
247	if (rc)
248		current->mm->context.vdso_base = 0;
249out_up:
250	up_write(&mm->mmap_sem);
251	return rc;
252}
253
254const char *arch_vma_name(struct vm_area_struct *vma)
255{
256	if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
257		return "[vdso]";
258	return NULL;
259}
260
261static int __init vdso_init(void)
262{
263	int i;
264
265	if (!vdso_enabled)
266		return 0;
267	vdso_init_data(vdso_data);
268#ifdef CONFIG_COMPAT
269	/* Calculate the size of the 32 bit vDSO */
270	vdso32_pages = ((&vdso32_end - &vdso32_start
271			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
272
273	/* Make sure pages are in the correct state */
274	vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
275				  GFP_KERNEL);
276	BUG_ON(vdso32_pagelist == NULL);
277	for (i = 0; i < vdso32_pages - 1; i++) {
278		struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
279		ClearPageReserved(pg);
280		get_page(pg);
281		vdso32_pagelist[i] = pg;
282	}
283	vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
284	vdso32_pagelist[vdso32_pages] = NULL;
285#endif
286
 
287	/* Calculate the size of the 64 bit vDSO */
288	vdso64_pages = ((&vdso64_end - &vdso64_start
289			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
290
291	/* Make sure pages are in the correct state */
292	vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
293				  GFP_KERNEL);
294	BUG_ON(vdso64_pagelist == NULL);
295	for (i = 0; i < vdso64_pages - 1; i++) {
296		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
297		ClearPageReserved(pg);
298		get_page(pg);
299		vdso64_pagelist[i] = pg;
300	}
301	vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
302	vdso64_pagelist[vdso64_pages] = NULL;
303	if (vdso_alloc_per_cpu(&S390_lowcore))
 
304		BUG();
 
305	vdso_init_cr5();
 
306
307	get_page(virt_to_page(vdso_data));
308
 
 
 
 
 
 
 
 
 
 
 
 
 
309	return 0;
310}
311early_initcall(vdso_init);