1/*
  2 * Adapted from arm64 version.
  3 *
  4 * Copyright (C) 2012 ARM Limited
  5 * Copyright (C) 2015 Mentor Graphics Corporation.
  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 as
  9 * published by the Free Software Foundation.
 10 *
 11 * This program is distributed in the hope that it will be useful,
 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 14 * GNU General Public License for more details.
 15 *
 16 * You should have received a copy of the GNU General Public License
 17 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 18 */
 19
 20#include <linux/cache.h>
 21#include <linux/elf.h>
 22#include <linux/err.h>
 23#include <linux/kernel.h>
 24#include <linux/mm.h>
 25#include <linux/of.h>
 26#include <linux/printk.h>
 27#include <linux/slab.h>
 28#include <linux/timekeeper_internal.h>
 29#include <linux/vmalloc.h>
 30#include <asm/arch_timer.h>
 31#include <asm/barrier.h>
 32#include <asm/cacheflush.h>
 33#include <asm/page.h>
 34#include <asm/vdso.h>
 35#include <asm/vdso_datapage.h>
 36#include <clocksource/arm_arch_timer.h>
 37
 38#define MAX_SYMNAME	64
 39
 40static struct page **vdso_text_pagelist;
 41
 42extern char vdso_start[], vdso_end[];
 43
 44/* Total number of pages needed for the data and text portions of the VDSO. */
 45unsigned int vdso_total_pages __ro_after_init;
 46
 47/*
 48 * The VDSO data page.
 49 */
 50static union vdso_data_store vdso_data_store __page_aligned_data;
 51static struct vdso_data *vdso_data = &vdso_data_store.data;
 52
 53static struct page *vdso_data_page __ro_after_init;
 54static const struct vm_special_mapping vdso_data_mapping = {
 55	.name = "[vvar]",
 56	.pages = &vdso_data_page,
 57};
 58
 59static int vdso_mremap(const struct vm_special_mapping *sm,
 60		struct vm_area_struct *new_vma)
 61{
 62	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
 63	unsigned long vdso_size;
 64
 65	/* without VVAR page */
 66	vdso_size = (vdso_total_pages - 1) << PAGE_SHIFT;
 67
 68	if (vdso_size != new_size)
 69		return -EINVAL;
 70
 71	current->mm->context.vdso = new_vma->vm_start;
 72
 73	return 0;
 74}
 75
 76static struct vm_special_mapping vdso_text_mapping __ro_after_init = {
 77	.name = "[vdso]",
 78	.mremap = vdso_mremap,
 79};
 80
 81struct elfinfo {
 82	Elf32_Ehdr	*hdr;		/* ptr to ELF */
 83	Elf32_Sym	*dynsym;	/* ptr to .dynsym section */
 84	unsigned long	dynsymsize;	/* size of .dynsym section */
 85	char		*dynstr;	/* ptr to .dynstr section */
 86};
 87
 88/* Cached result of boot-time check for whether the arch timer exists,
 89 * and if so, whether the virtual counter is useable.
 90 */
 91static bool cntvct_ok __ro_after_init;
 92
 93static bool __init cntvct_functional(void)
 94{
 95	struct device_node *np;
 96	bool ret = false;
 97
 98	if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
 99		goto out;
100
101	/* The arm_arch_timer core should export
102	 * arch_timer_use_virtual or similar so we don't have to do
103	 * this.
104	 */
105	np = of_find_compatible_node(NULL, NULL, "arm,armv7-timer");
106	if (!np)
107		goto out_put;
108
109	if (of_property_read_bool(np, "arm,cpu-registers-not-fw-configured"))
110		goto out_put;
111
112	ret = true;
113
114out_put:
115	of_node_put(np);
116out:
117	return ret;
118}
119
120static void * __init find_section(Elf32_Ehdr *ehdr, const char *name,
121				  unsigned long *size)
122{
123	Elf32_Shdr *sechdrs;
124	unsigned int i;
125	char *secnames;
126
127	/* Grab section headers and strings so we can tell who is who */
128	sechdrs = (void *)ehdr + ehdr->e_shoff;
129	secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
130
131	/* Find the section they want */
132	for (i = 1; i < ehdr->e_shnum; i++) {
133		if (strcmp(secnames + sechdrs[i].sh_name, name) == 0) {
134			if (size)
135				*size = sechdrs[i].sh_size;
136			return (void *)ehdr + sechdrs[i].sh_offset;
137		}
138	}
139
140	if (size)
141		*size = 0;
142	return NULL;
143}
144
145static Elf32_Sym * __init find_symbol(struct elfinfo *lib, const char *symname)
146{
147	unsigned int i;
148
149	for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
150		char name[MAX_SYMNAME], *c;
151
152		if (lib->dynsym[i].st_name == 0)
153			continue;
154		strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
155			MAX_SYMNAME);
156		c = strchr(name, '@');
157		if (c)
158			*c = 0;
159		if (strcmp(symname, name) == 0)
160			return &lib->dynsym[i];
161	}
162	return NULL;
163}
164
165static void __init vdso_nullpatch_one(struct elfinfo *lib, const char *symname)
166{
167	Elf32_Sym *sym;
168
169	sym = find_symbol(lib, symname);
170	if (!sym)
171		return;
172
173	sym->st_name = 0;
174}
175
176static void __init patch_vdso(void *ehdr)
177{
178	struct elfinfo einfo;
179
180	einfo = (struct elfinfo) {
181		.hdr = ehdr,
182	};
183
184	einfo.dynsym = find_section(einfo.hdr, ".dynsym", &einfo.dynsymsize);
185	einfo.dynstr = find_section(einfo.hdr, ".dynstr", NULL);
186
187	/* If the virtual counter is absent or non-functional we don't
188	 * want programs to incur the slight additional overhead of
189	 * dispatching through the VDSO only to fall back to syscalls.
190	 */
191	if (!cntvct_ok) {
192		vdso_nullpatch_one(&einfo, "__vdso_gettimeofday");
193		vdso_nullpatch_one(&einfo, "__vdso_clock_gettime");
194	}
195}
196
197static int __init vdso_init(void)
198{
199	unsigned int text_pages;
200	int i;
201
202	if (memcmp(vdso_start, "\177ELF", 4)) {
203		pr_err("VDSO is not a valid ELF object!\n");
204		return -ENOEXEC;
205	}
206
207	text_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
208	pr_debug("vdso: %i text pages at base %p\n", text_pages, vdso_start);
209
210	/* Allocate the VDSO text pagelist */
211	vdso_text_pagelist = kcalloc(text_pages, sizeof(struct page *),
212				     GFP_KERNEL);
213	if (vdso_text_pagelist == NULL)
214		return -ENOMEM;
215
216	/* Grab the VDSO data page. */
217	vdso_data_page = virt_to_page(vdso_data);
218
219	/* Grab the VDSO text pages. */
220	for (i = 0; i < text_pages; i++) {
221		struct page *page;
222
223		page = virt_to_page(vdso_start + i * PAGE_SIZE);
224		vdso_text_pagelist[i] = page;
225	}
226
227	vdso_text_mapping.pages = vdso_text_pagelist;
228
229	vdso_total_pages = 1; /* for the data/vvar page */
230	vdso_total_pages += text_pages;
231
232	cntvct_ok = cntvct_functional();
233
234	patch_vdso(vdso_start);
235
236	return 0;
237}
238arch_initcall(vdso_init);
239
240static int install_vvar(struct mm_struct *mm, unsigned long addr)
241{
242	struct vm_area_struct *vma;
243
244	vma = _install_special_mapping(mm, addr, PAGE_SIZE,
245				       VM_READ | VM_MAYREAD,
246				       &vdso_data_mapping);
247
248	return PTR_ERR_OR_ZERO(vma);
249}
250
251/* assumes mmap_sem is write-locked */
252void arm_install_vdso(struct mm_struct *mm, unsigned long addr)
253{
254	struct vm_area_struct *vma;
255	unsigned long len;
256
257	mm->context.vdso = 0;
258
259	if (vdso_text_pagelist == NULL)
260		return;
261
262	if (install_vvar(mm, addr))
263		return;
264
265	/* Account for vvar page. */
266	addr += PAGE_SIZE;
267	len = (vdso_total_pages - 1) << PAGE_SHIFT;
268
269	vma = _install_special_mapping(mm, addr, len,
270		VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
271		&vdso_text_mapping);
272
273	if (!IS_ERR(vma))
274		mm->context.vdso = addr;
275}
276
277static void vdso_write_begin(struct vdso_data *vdata)
278{
279	++vdso_data->seq_count;
280	smp_wmb(); /* Pairs with smp_rmb in vdso_read_retry */
281}
282
283static void vdso_write_end(struct vdso_data *vdata)
284{
285	smp_wmb(); /* Pairs with smp_rmb in vdso_read_begin */
286	++vdso_data->seq_count;
287}
288
289static bool tk_is_cntvct(const struct timekeeper *tk)
290{
291	if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
292		return false;
293
294	if (!tk->tkr_mono.clock->archdata.vdso_direct)
295		return false;
296
297	return true;
298}
299
300/**
301 * update_vsyscall - update the vdso data page
302 *
303 * Increment the sequence counter, making it odd, indicating to
304 * userspace that an update is in progress.  Update the fields used
305 * for coarse clocks and, if the architected system timer is in use,
306 * the fields used for high precision clocks.  Increment the sequence
307 * counter again, making it even, indicating to userspace that the
308 * update is finished.
309 *
310 * Userspace is expected to sample seq_count before reading any other
311 * fields from the data page.  If seq_count is odd, userspace is
312 * expected to wait until it becomes even.  After copying data from
313 * the page, userspace must sample seq_count again; if it has changed
314 * from its previous value, userspace must retry the whole sequence.
315 *
316 * Calls to update_vsyscall are serialized by the timekeeping core.
317 */
318void update_vsyscall(struct timekeeper *tk)
319{
320	struct timespec64 *wtm = &tk->wall_to_monotonic;
321
322	if (!cntvct_ok) {
323		/* The entry points have been zeroed, so there is no
324		 * point in updating the data page.
325		 */
326		return;
327	}
328
329	vdso_write_begin(vdso_data);
330
331	vdso_data->tk_is_cntvct			= tk_is_cntvct(tk);
332	vdso_data->xtime_coarse_sec		= tk->xtime_sec;
333	vdso_data->xtime_coarse_nsec		= (u32)(tk->tkr_mono.xtime_nsec >>
334							tk->tkr_mono.shift);
335	vdso_data->wtm_clock_sec		= wtm->tv_sec;
336	vdso_data->wtm_clock_nsec		= wtm->tv_nsec;
337
338	if (vdso_data->tk_is_cntvct) {
339		vdso_data->cs_cycle_last	= tk->tkr_mono.cycle_last;
340		vdso_data->xtime_clock_sec	= tk->xtime_sec;
341		vdso_data->xtime_clock_snsec	= tk->tkr_mono.xtime_nsec;
342		vdso_data->cs_mult		= tk->tkr_mono.mult;
343		vdso_data->cs_shift		= tk->tkr_mono.shift;
344		vdso_data->cs_mask		= tk->tkr_mono.mask;
345	}
346
347	vdso_write_end(vdso_data);
348
349	flush_dcache_page(virt_to_page(vdso_data));
350}
351
352void update_vsyscall_tz(void)
353{
354	vdso_data->tz_minuteswest	= sys_tz.tz_minuteswest;
355	vdso_data->tz_dsttime		= sys_tz.tz_dsttime;
356	flush_dcache_page(virt_to_page(vdso_data));
357}