[mc] pread_whole must restart in cas of EINTR, not EAGAIN

[simgrid.git] / src / mc / mc_page_snapshot.cpp

diff --git a/src/mc/mc_page_snapshot.cpp b/src/mc/mc_page_snapshot.cpp
index 245fc13..833d70c 100644 (file)
--- a/src/mc/mc_page_snapshot.cpp
+++ b/src/mc/mc_page_snapshot.cpp
@@ -13,7 +13,8 @@ size_t* mc_take_page_snapshot_region(void* data, size_t page_count, uint64_t* pa
   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) {
       // 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]);
@@ -57,7 +58,7 @@ 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 +67,15 @@ 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;
@@ -92,11 +94,13 @@ 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 +109,16 @@ 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
@@ -122,7 +129,7 @@ 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");
   }
 }
 
@@ -144,11 +151,12 @@ mc_mem_region_t mc_region_new_sparse(int type, void *start_addr, size_t size, mc
   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(start_addr), page_count);
+      mc_read_pagemap(pagemap, mc_page_number(NULL, start_addr), page_count);
   }
 
   // Take incremental snapshot:
-  new_reg->page_numbers = mc_take_page_snapshot_region(start_addr, page_count, pagemap, ref_reg->page_numbers);
+  new_reg->page_numbers = mc_take_page_snapshot_region(start_addr, page_count, pagemap,
+    ref_reg==NULL ? NULL : ref_reg->page_numbers);
 
   return new_reg;
 }
@@ -164,7 +172,7 @@ void mc_region_restore_sparse(mc_mem_region_t reg, mc_mem_region_t ref_reg)
   // 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(reg->start_addr), page_count);
+    mc_read_pagemap(pagemap, mc_page_number(NULL, reg->start_addr), page_count);
   }
 
   // Incremental per-page snapshot restoration: