X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/d47e21d93998034c0f8fa4a9a6c750002f7642fa..e3f5ca3fd3726e67046afbd03e48e9cd7294b835:/src/mc/mc_page_snapshot.cpp

diff --git a/src/mc/mc_page_snapshot.cpp b/src/mc/mc_page_snapshot.cpp
index 92959adc98..dcf0d9bbc8 100644
--- a/src/mc/mc_page_snapshot.cpp
+++ b/src/mc/mc_page_snapshot.cpp
@@ -1,5 +1,9 @@
 #include "mc_page_store.h"
 #include "mc_mmu.h"
+#include "mc_private.h"
+#include "mc_snapshot.h"
+
+#include <xbt/mmalloc.h>
 
 #define SOFT_DIRTY_BIT_NUMBER 55
 #define SOFT_DIRTY (((uint64_t)1) << SOFT_DIRTY_BIT_NUMBER)
@@ -8,17 +12,26 @@ extern "C" {
 
 // ***** Region management:
 
+/** @brief Take a per-page snapshot of a region
+ *
+ *  @param data            The start of the region (must be at the beginning of a page)
+ *  @param pag_count       Number of pages of the region
+ *  @param pagemap         Linux kernel pagemap values fot this region (or NULL)
+ *  @param reference_pages Snapshot page numbers of the previous soft_dirty_reset (or NULL)
+ *  @return                Snapshot page numbers of this new snapshot
+ */
 size_t* mc_take_page_snapshot_region(void* data, size_t page_count, uint64_t* pagemap, size_t* reference_pages)
 {
   size_t* pagenos = (size_t*) malloc(page_count * sizeof(size_t));
 
   for (size_t i=0; i!=page_count; ++i) {
-    if (pagemap && (pagemap[i] & SOFT_DIRTY)) {
+    bool softclean = pagemap && !(pagemap[i] & SOFT_DIRTY);
+    if (softclean && reference_pages) {
       // The page is softclean, it is the same page as the reference page:
       pagenos[i] = reference_pages[i];
       mc_model_checker->pages->ref_page(reference_pages[i]);
     } else {
-      // Otherwise, we need to store the page the hard hard
+      // Otherwise, we need to store the page the hard way
       // (by reading its content):
       void* page = (char*) data + (i << xbt_pagebits);
       pagenos[i] = mc_model_checker->pages->store_page(page);
@@ -35,20 +48,31 @@ void mc_free_page_snapshot_region(size_t* pagenos, size_t page_count)
   }
 }
 
-void mc_restore_page_snapshot_region(mc_mem_region_t region, size_t page_count, uint64_t* pagemap, mc_mem_region_t reference_region)
+/** @brief Restore a snapshot of a region
+ *
+ *  If possible, the restoration will be incremental
+ *  (the modified pages will not be touched).
+ *
+ *  @param start_addr
+ *  @param page_count       Number of pages of the region
+ *  @param pagenos
+ *  @param pagemap         Linux kernel pagemap values fot this region (or NULL)
+ *  @param reference_pages Snapshot page numbers of the previous soft_dirty_reset (or NULL)
+ */
+void mc_restore_page_snapshot_region(void* start_addr, size_t page_count, size_t* pagenos, uint64_t* pagemap, size_t* reference_pagenos)
 {
   for (size_t i=0; i!=page_count; ++i) {
 
     bool softclean = pagemap && !(pagemap[i] & SOFT_DIRTY);
-    if (softclean && reference_region && reference_region->page_numbers[i] == region->page_numbers[i]) {
+    if (softclean && reference_pagenos && pagenos[i] == reference_pagenos[i]) {
       // The page is softclean and is the same as the reference one:
       // the page is already in the target state.
       continue;
     }
 
     // Otherwise, copy the page:
-    void* target_page = mc_page_from_number(region->start_addr, i);
-    const void* source_page = mc_model_checker->pages->get_page(region->page_numbers[i]);
+    void* target_page = mc_page_from_number(start_addr, i);
+    const void* source_page = mc_model_checker->pages->get_page(pagenos[i]);
     memcpy(target_page, source_page, xbt_pagesize);
   }
 }
@@ -57,7 +81,7 @@ void mc_restore_page_snapshot_region(mc_mem_region_t region, size_t page_count,
 
 /** @brief Like pread() but without partial reads */
 static size_t pread_whole(int fd, void* buf, size_t count, off_t offset) {
-  size_t res;
+  size_t res = 0;
 
   char* data = (char*) buf;
   while(count) {
@@ -66,14 +90,15 @@ static size_t pread_whole(int fd, void* buf, size_t count, off_t offset) {
     if (n==0)
       return res;
 
-    // Error (or EAGAIN):
+    // Error (or EINTR):
     if (n==-1) {
-      if (errno == EAGAIN)
+      if (errno == EINTR)
         continue;
       else
         return -1;
     }
 
+    // It might be a partial read:
     count -= n;
     data += n;
     offset += n;
@@ -83,7 +108,8 @@ static size_t pread_whole(int fd, void* buf, size_t count, off_t offset) {
   return res;
 }
 
-static inline void mc_ensure_fd(int* fd, const char* path, int flags) {
+static inline  __attribute__ ((always_inline))
+void mc_ensure_fd(int* fd, const char* path, int flags) {
   if (*fd != -1)
     return;
   *fd = open(path, flags);
@@ -92,11 +118,13 @@ static inline void mc_ensure_fd(int* fd, const char* path, int flags) {
   }
 }
 
-/** @brief Reset the softdirty bits
+/** @brief Reset the soft-dirty bits
  *
  *  This is done after checkpointing and after checkpoint restoration
  *  (if per page checkpoiting is used) in order to know which pages were
  *  modified.
+ *
+ *  See https://www.kernel.org/doc/Documentation/vm/soft-dirty.txt
  * */
 void mc_softdirty_reset() {
   mc_ensure_fd(&mc_model_checker->fd_clear_refs, "/proc/self/clear_refs", O_WRONLY|O_CLOEXEC);
@@ -105,13 +133,16 @@ void mc_softdirty_reset() {
   }
 }
 
-/** @brief Read /proc/self/pagemap informations in order to find properties on the pages
+/** @brief Read memory page informations
  *
- *  For each virtual memory page, this file provides informations.
+ *  For each virtual memory page of the process,
+ *  /proc/self/pagemap provides a 64 bit field of information.
  *  We are interested in the soft-dirty bit: with this we can track which
  *  pages were modified between snapshots/restorations and avoid
  *  copying data which was not modified.
  *
+ *  See https://www.kernel.org/doc/Documentation/vm/pagemap.txt
+ *
  *  @param pagemap    Output buffer for pagemap informations
  *  @param start_addr Address of the first page
  *  @param page_count Number of pages
@@ -122,53 +153,74 @@ static void mc_read_pagemap(uint64_t* pagemap, size_t page_start, size_t page_co
   size_t bytesize = sizeof(uint64_t) * page_count;
   off_t offset = sizeof(uint64_t) * page_start;
   if (pread_whole(mc_model_checker->fd_pagemap, pagemap, bytesize, offset) != bytesize) {
-    xbt_die("Coult not read pagemap");
+    xbt_die("Could not read pagemap");
   }
 }
 
 // ***** High level API
 
-mc_mem_region_t mc_region_new_sparse(int type, void *start_addr, size_t size, mc_mem_region_t ref_reg)
+mc_mem_region_t mc_region_new_sparse(mc_region_type_t region_type,
+  void *start_addr, void* permanent_addr, size_t size,
+  mc_mem_region_t ref_reg)
 {
-  mc_mem_region_t new_reg = xbt_new(s_mc_mem_region_t, 1);
-
-  new_reg->start_addr = start_addr;
-  new_reg->data = NULL;
-  new_reg->size = size;
-  new_reg->page_numbers = NULL;
+  mc_mem_region_t region = xbt_new(s_mc_mem_region_t, 1);
+  region->region_type = region_type;
+  region->storage_type = MC_REGION_STORAGE_TYPE_CHUNKED;
+  region->start_addr = start_addr;
+  region->permanent_addr = permanent_addr;
+  region->size = size;
 
   xbt_assert((((uintptr_t)start_addr) & (xbt_pagesize-1)) == 0,
     "Not at the beginning of a page");
+  xbt_assert((((uintptr_t)permanent_addr) & (xbt_pagesize-1)) == 0,
+    "Not at the beginning of a page");
   size_t page_count = mc_page_count(size);
 
   uint64_t* pagemap = NULL;
-  if (mc_model_checker->parent_snapshot) {
-      pagemap = (uint64_t*) alloca(sizeof(uint64_t) * page_count);
-      mc_read_pagemap(pagemap, mc_page_number(NULL, start_addr), page_count);
+  if (_sg_mc_soft_dirty && mc_model_checker->parent_snapshot) {
+      pagemap = (uint64_t*) mmalloc_no_memset(mc_heap, sizeof(uint64_t) * page_count);
+      mc_read_pagemap(pagemap, mc_page_number(NULL, permanent_addr), page_count);
   }
 
+  size_t* reg_page_numbers = NULL;
+  if (ref_reg!=NULL && ref_reg->storage_type == MC_REGION_STORAGE_TYPE_CHUNKED)
+    reg_page_numbers = ref_reg->chunked.page_numbers;
+
   // Take incremental snapshot:
-  new_reg->page_numbers = mc_take_page_snapshot_region(start_addr, page_count, pagemap, ref_reg->page_numbers);
+  region->chunked.page_numbers = mc_take_page_snapshot_region(
+    permanent_addr, page_count, pagemap, reg_page_numbers);
 
-  return new_reg;
+  if(pagemap) {
+    mfree(mc_heap, pagemap);
+  }
+  return region;
 }
 
 void mc_region_restore_sparse(mc_mem_region_t reg, mc_mem_region_t ref_reg)
 {
-  xbt_assert((((uintptr_t)reg->start_addr) & (xbt_pagesize-1)) == 0,
+  xbt_assert((((uintptr_t)reg->permanent_addr) & (xbt_pagesize-1)) == 0,
     "Not at the beginning of a page");
   size_t page_count = mc_page_count(reg->size);
 
   uint64_t* pagemap = NULL;
 
   // Read soft-dirty bits if necessary in order to know which pages have changed:
-  if (mc_model_checker->parent_snapshot) {
-    pagemap = (uint64_t*) alloca(sizeof(uint64_t) * page_count);
-    mc_read_pagemap(pagemap, mc_page_number(NULL, reg->start_addr), page_count);
+  if (_sg_mc_soft_dirty && mc_model_checker->parent_snapshot) {
+    pagemap = (uint64_t*) mmalloc_no_memset(mc_heap, sizeof(uint64_t) * page_count);
+    mc_read_pagemap(pagemap, mc_page_number(NULL, reg->permanent_addr), page_count);
   }
 
-  // Incremental per-page snapshot restoration:
-  mc_restore_page_snapshot_region(reg, page_count, pagemap, ref_reg);
+  // Incremental per-page snapshot restoration:s
+  size_t* reg_page_numbers = NULL;
+  if (ref_reg && ref_reg->storage_type == MC_REGION_STORAGE_TYPE_CHUNKED)
+    reg_page_numbers = ref_reg->chunked.page_numbers;
+
+  mc_restore_page_snapshot_region(reg->permanent_addr, page_count, reg->chunked.page_numbers,
+    pagemap, reg_page_numbers);
+
+  if(pagemap) {
+    free(pagemap);
+  }
 }
 
 }