1 /* MC interface: definitions that non-MC modules must see, but not the user */
3 /* Copyright (c) 2014-2015. The SimGrid Team. All rights reserved. */
5 /* This program is free software; you can redistribute it and/or modify it
6 * under the terms of the license (GNU LGPL) which comes with this package. */
8 #include "mc_page_store.h"
10 #include "mc_private.h"
11 #include "mc_snapshot.h"
13 #include <xbt/mmalloc.h>
15 #define SOFT_DIRTY_BIT_NUMBER 55
16 #define SOFT_DIRTY (((uint64_t)1) << SOFT_DIRTY_BIT_NUMBER)
20 // ***** Region management:
22 /** @brief Take a per-page snapshot of a region
24 * @param data The start of the region (must be at the beginning of a page)
25 * @param pag_count Number of pages of the region
26 * @param pagemap Linux kernel pagemap values fot this region (or NULL)
27 * @param reference_pages Snapshot page numbers of the previous soft_dirty_reset (or NULL)
28 * @return Snapshot page numbers of this new snapshot
30 size_t* mc_take_page_snapshot_region(void* data, size_t page_count, uint64_t* pagemap, size_t* reference_pages)
32 size_t* pagenos = (size_t*) malloc(page_count * sizeof(size_t));
34 for (size_t i=0; i!=page_count; ++i) {
35 bool softclean = pagemap && !(pagemap[i] & SOFT_DIRTY);
36 if (softclean && reference_pages) {
37 // The page is softclean, it is the same page as the reference page:
38 pagenos[i] = reference_pages[i];
39 mc_model_checker->pages->ref_page(reference_pages[i]);
41 // Otherwise, we need to store the page the hard way
42 // (by reading its content):
43 void* page = (char*) data + (i << xbt_pagebits);
44 pagenos[i] = mc_model_checker->pages->store_page(page);
51 void mc_free_page_snapshot_region(size_t* pagenos, size_t page_count)
53 for (size_t i=0; i!=page_count; ++i) {
54 mc_model_checker->pages->unref_page(pagenos[i]);
58 /** @brief Restore a snapshot of a region
60 * If possible, the restoration will be incremental
61 * (the modified pages will not be touched).
64 * @param page_count Number of pages of the region
66 * @param pagemap Linux kernel pagemap values fot this region (or NULL)
67 * @param reference_pages Snapshot page numbers of the previous soft_dirty_reset (or NULL)
69 void mc_restore_page_snapshot_region(void* start_addr, size_t page_count, size_t* pagenos, uint64_t* pagemap, size_t* reference_pagenos)
71 for (size_t i=0; i!=page_count; ++i) {
73 bool softclean = pagemap && !(pagemap[i] & SOFT_DIRTY);
74 if (softclean && reference_pagenos && pagenos[i] == reference_pagenos[i]) {
75 // The page is softclean and is the same as the reference one:
76 // the page is already in the target state.
80 // Otherwise, copy the page:
81 void* target_page = mc_page_from_number(start_addr, i);
82 const void* source_page = mc_model_checker->pages->get_page(pagenos[i]);
83 memcpy(target_page, source_page, xbt_pagesize);
87 // ***** Soft dirty tracking
89 /** @brief Like pread() but without partial reads */
90 static size_t pread_whole(int fd, void* buf, size_t count, off_t offset) {
93 char* data = (char*) buf;
95 ssize_t n = pread(fd, buf, count, offset);
108 // It might be a partial read:
118 static inline __attribute__ ((always_inline))
119 void mc_ensure_fd(int* fd, const char* path, int flags) {
122 *fd = open(path, flags);
124 xbt_die("Could not open file %s", path);
128 /** @brief Reset the soft-dirty bits
130 * This is done after checkpointing and after checkpoint restoration
131 * (if per page checkpoiting is used) in order to know which pages were
134 * See https://www.kernel.org/doc/Documentation/vm/soft-dirty.txt
136 void mc_softdirty_reset() {
137 mc_ensure_fd(&mc_model_checker->fd_clear_refs, "/proc/self/clear_refs", O_WRONLY|O_CLOEXEC);
138 if( ::write(mc_model_checker->fd_clear_refs, "4\n", 2) != 2) {
139 xbt_die("Could not reset softdirty bits");
143 /** @brief Read memory page informations
145 * For each virtual memory page of the process,
146 * /proc/self/pagemap provides a 64 bit field of information.
147 * We are interested in the soft-dirty bit: with this we can track which
148 * pages were modified between snapshots/restorations and avoid
149 * copying data which was not modified.
151 * See https://www.kernel.org/doc/Documentation/vm/pagemap.txt
153 * @param pagemap Output buffer for pagemap informations
154 * @param start_addr Address of the first page
155 * @param page_count Number of pages
157 static void mc_read_pagemap(uint64_t* pagemap, size_t page_start, size_t page_count)
159 mc_ensure_fd(&mc_model_checker->fd_pagemap, "/proc/self/pagemap", O_RDONLY|O_CLOEXEC);
160 size_t bytesize = sizeof(uint64_t) * page_count;
161 off_t offset = sizeof(uint64_t) * page_start;
162 if (pread_whole(mc_model_checker->fd_pagemap, pagemap, bytesize, offset) != bytesize) {
163 xbt_die("Could not read pagemap");
167 // ***** High level API
169 mc_mem_region_t mc_region_new_sparse(int type, void *start_addr, void* permanent_addr, size_t size, mc_mem_region_t ref_reg)
171 mc_mem_region_t new_reg = xbt_new(s_mc_mem_region_t, 1);
173 new_reg->start_addr = start_addr;
174 new_reg->permanent_addr = permanent_addr;
175 new_reg->data = NULL;
176 new_reg->size = size;
177 new_reg->page_numbers = NULL;
179 xbt_assert((((uintptr_t)start_addr) & (xbt_pagesize-1)) == 0,
180 "Not at the beginning of a page");
181 xbt_assert((((uintptr_t)permanent_addr) & (xbt_pagesize-1)) == 0,
182 "Not at the beginning of a page");
183 size_t page_count = mc_page_count(size);
185 uint64_t* pagemap = NULL;
186 if (_sg_mc_soft_dirty && mc_model_checker->parent_snapshot) {
187 pagemap = (uint64_t*) mmalloc_no_memset(mc_heap, sizeof(uint64_t) * page_count);
188 mc_read_pagemap(pagemap, mc_page_number(NULL, permanent_addr), page_count);
191 // Take incremental snapshot:
192 new_reg->page_numbers = mc_take_page_snapshot_region(permanent_addr, page_count, pagemap,
193 ref_reg==NULL ? NULL : ref_reg->page_numbers);
196 mfree(mc_heap, pagemap);
201 void mc_region_restore_sparse(mc_mem_region_t reg, mc_mem_region_t ref_reg)
203 xbt_assert((((uintptr_t)reg->permanent_addr) & (xbt_pagesize-1)) == 0,
204 "Not at the beginning of a page");
205 size_t page_count = mc_page_count(reg->size);
207 uint64_t* pagemap = NULL;
209 // Read soft-dirty bits if necessary in order to know which pages have changed:
210 if (_sg_mc_soft_dirty && mc_model_checker->parent_snapshot) {
211 pagemap = (uint64_t*) mmalloc_no_memset(mc_heap, sizeof(uint64_t) * page_count);
212 mc_read_pagemap(pagemap, mc_page_number(NULL, reg->permanent_addr), page_count);
215 // Incremental per-page snapshot restoration:
216 mc_restore_page_snapshot_region(reg->permanent_addr, page_count, reg->page_numbers,
217 pagemap, ref_reg ? ref_reg->page_numbers : NULL);