diff options
author | Quincey Koziol <koziol@hdfgroup.org> | 2009-03-24 17:52:17 (GMT) |
---|---|---|
committer | Quincey Koziol <koziol@hdfgroup.org> | 2009-03-24 17:52:17 (GMT) |
commit | 667b34c713318b7a70af5c9b63fccb9ac653a201 (patch) | |
tree | b887d45b649be5e87326786056e046f967bc01bf /src | |
parent | adf47d11b821bbdd40ec7ebc0a4e6a69781e6ca7 (diff) | |
download | hdf5-667b34c713318b7a70af5c9b63fccb9ac653a201.zip hdf5-667b34c713318b7a70af5c9b63fccb9ac653a201.tar.gz hdf5-667b34c713318b7a70af5c9b63fccb9ac653a201.tar.bz2 |
[svn-r16600] Description:
Ensure that metadata cache entries with flush dependencies are written
to the file in the correct order (for serial I/O, at least).
Tested on:
FreeBSD/32 6.3 (duty) in debug mode
FreeBSD/64 6.3 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (jam) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
Mac OS X/32 10.5.6 (amazon) in debug mode
Mac OS X/32 10.5.6 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
Diffstat (limited to 'src')
-rw-r--r-- | src/H5C.c | 984 | ||||
-rw-r--r-- | src/H5Cprivate.h | 2 |
2 files changed, 528 insertions, 458 deletions
@@ -3808,216 +3808,244 @@ H5C_flush_cache(H5F_t * f, ( protected_entries == 0 ) && ( flushed_entries_last_pass ) ) { - flushed_entries_last_pass = FALSE; - node_ptr = H5SL_first(cache_ptr->slist_ptr); + unsigned curr_flush_dep_height = 0; + unsigned flush_dep_passes = 0; - if ( node_ptr != NULL ) { + flushed_entries_last_pass = FALSE; - next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); + /* Loop over all flush dependency heights of entries */ + while((curr_flush_dep_height <= H5C__NUM_FLUSH_DEP_HEIGHTS) && + (cache_ptr->slist_len != 0) && + (flush_dep_passes < H5C__MAX_PASSES_ON_FLUSH) ) + { + hbool_t flushed_during_dep_loop = FALSE; - if ( next_entry_ptr == NULL ) { + /* Start at beginning of skip list each time */ + node_ptr = H5SL_first(cache_ptr->slist_ptr); + HDassert( node_ptr != NULL ); - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "next_entry_ptr == NULL 1 ?!?!"); - } - HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); - HDassert( next_entry_ptr->is_dirty ); + /* Get cache entry for this node */ + next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); + if ( NULL == next_entry_ptr ) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "next_entry_ptr == NULL ?!?!") + HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); + HDassert( next_entry_ptr->is_dirty ); HDassert( next_entry_ptr->in_slist ); - } else { - - next_entry_ptr = NULL; - - } - - HDassert( node_ptr != NULL ); - #if H5C_DO_SANITY_CHECKS - /* For sanity checking, try to verify that the skip list has - * the expected size and number of entries at the end of each - * internal while loop (see below). - * - * Doing this get a bit tricky, as depending on flags, we may - * or may not flush all the entries in the slist. - * - * To make things more entertaining, with the advent of the - * fractal heap, the entry flush callback can cause entries - * to be dirtied, resized, and/or renamed. - * - * To deal with this, we first make note of the initial - * skip list length and size: - */ - initial_slist_len = cache_ptr->slist_len; - initial_slist_size = cache_ptr->slist_size; + /* For sanity checking, try to verify that the skip list has + * the expected size and number of entries at the end of each + * internal while loop (see below). + * + * Doing this get a bit tricky, as depending on flags, we may + * or may not flush all the entries in the slist. + * + * To make things more entertaining, with the advent of the + * fractal heap, the entry flush callback can cause entries + * to be dirtied, resized, and/or renamed. + * + * To deal with this, we first make note of the initial + * skip list length and size: + */ + initial_slist_len = cache_ptr->slist_len; + initial_slist_size = cache_ptr->slist_size; - /* We then zero counters that we use to track the number - * and total size of entries flushed: - */ - flushed_entries_count = 0; - flushed_entries_size = 0; - - /* As mentioned above, there is the possibility that - * entries will be dirtied, resized, and/or flushed during - * our pass through the skip list. To capture the number - * of entries added, and the skip list size delta, - * zero the slist_len_increase and slist_size_increase of - * the cache's instance of H5C_t. These fields will be - * updated elsewhere to account for slist insertions and/or - * dirty entry size changes. - */ - cache_ptr->slist_len_increase = 0; - cache_ptr->slist_size_increase = 0; + /* We then zero counters that we use to track the number + * and total size of entries flushed: + */ + flushed_entries_count = 0; + flushed_entries_size = 0; + + /* As mentioned above, there is the possibility that + * entries will be dirtied, resized, and/or flushed during + * our pass through the skip list. To capture the number + * of entries added, and the skip list size delta, + * zero the slist_len_increase and slist_size_increase of + * the cache's instance of H5C_t. These fields will be + * updated elsewhere to account for slist insertions and/or + * dirty entry size changes. + */ + cache_ptr->slist_len_increase = 0; + cache_ptr->slist_size_increase = 0; - /* at the end of the loop, use these values to compute the - * expected slist length and size and compare this with the - * value recorded in the cache's instance of H5C_t. - */ + /* at the end of the loop, use these values to compute the + * expected slist length and size and compare this with the + * value recorded in the cache's instance of H5C_t. + */ #endif /* H5C_DO_SANITY_CHECKS */ - while ( node_ptr != NULL ) - { - entry_ptr = next_entry_ptr; - - /* With the advent of the fractal heap, it is possible - * that the flush callback will dirty and/or resize - * other entries in the cache. In particular, while - * Quincey has promised me that this will never happen, - * it is possible that the flush callback for an - * entry may protect an entry that is not in the cache, - * perhaps causing the cache to flush and possibly - * evict the entry associated with node_ptr to make - * space for the new entry. - * - * Thus we do a bit of extra sanity checking on entry_ptr, - * and break out of this scan of the skip list if we - * detect minor problems. We have a bit of leaway on the - * number of passes though the skip list, so this shouldn't - * be an issue in the flush in and of itself, as it should - * be all but impossible for this to happen more than once - * in any flush. - * - * Observe that that breaking out of the scan early - * shouldn't break the sanity checks just after the end - * of this while loop. - * - * If an entry has merely been marked clean and removed from - * the s-list, we simply break out of the scan. - * - * If the entry has been evicted, we flag an error and - * exit. - */ + while ( node_ptr != NULL ) + { + entry_ptr = next_entry_ptr; + + /* With the advent of the fractal heap, it is possible + * that the flush callback will dirty and/or resize + * other entries in the cache. In particular, while + * Quincey has promised me that this will never happen, + * it is possible that the flush callback for an + * entry may protect an entry that is not in the cache, + * perhaps causing the cache to flush and possibly + * evict the entry associated with node_ptr to make + * space for the new entry. + * + * Thus we do a bit of extra sanity checking on entry_ptr, + * and break out of this scan of the skip list if we + * detect minor problems. We have a bit of leaway on the + * number of passes though the skip list, so this shouldn't + * be an issue in the flush in and of itself, as it should + * be all but impossible for this to happen more than once + * in any flush. + * + * Observe that that breaking out of the scan early + * shouldn't break the sanity checks just after the end + * of this while loop. + * + * If an entry has merely been marked clean and removed from + * the s-list, we simply break out of the scan. + * + * If the entry has been evicted, we flag an error and + * exit. + */ #ifndef NDEBUG - if ( entry_ptr->magic != H5C__H5C_CACHE_ENTRY_T_MAGIC ) { + if ( entry_ptr->magic != H5C__H5C_CACHE_ENTRY_T_MAGIC ) { - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "entry_ptr->magic invalid ?!?!"); + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ + "entry_ptr->magic invalid ?!?!"); - } else + } else #endif /* NDEBUG */ - if ( ( ! entry_ptr->is_dirty ) || - ( ! entry_ptr->in_slist ) ) { + if ( ( ! entry_ptr->is_dirty ) || + ( ! entry_ptr->in_slist ) ) { - /* the s-list has been modified out from under us. - * set node_ptr to NULL and break out of the loop. - */ - node_ptr = NULL; - break; - } + /* the s-list has been modified out from under us. + * set node_ptr to NULL and break out of the inner loop. + */ + node_ptr = NULL; + goto end_of_inner_loop;; + } - /* increment node pointer now, before we delete its target - * from the slist. - */ - node_ptr = H5SL_next(node_ptr); + /* increment node pointer now, before we delete its target + * from the slist. + */ + node_ptr = H5SL_next(node_ptr); - if ( node_ptr != NULL ) { - next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); + if ( node_ptr != NULL ) { + next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); - if ( next_entry_ptr == NULL ) { - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "next_entry_ptr == NULL 2 ?!?!"); + if ( next_entry_ptr == NULL ) { + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ + "next_entry_ptr == NULL 2 ?!?!"); + } + HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); + HDassert( next_entry_ptr->is_dirty ); + HDassert( next_entry_ptr->in_slist ); + } else { + next_entry_ptr = NULL; } - HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); - HDassert( next_entry_ptr->is_dirty ); - HDassert( next_entry_ptr->in_slist ); - } else { - next_entry_ptr = NULL; - } - HDassert( entry_ptr != NULL ); - HDassert( entry_ptr->in_slist ); + HDassert( entry_ptr != NULL ); + HDassert( entry_ptr->in_slist ); - if ( ( ! flush_marked_entries ) || - ( entry_ptr->flush_marker ) ) { + if ( ( ! flush_marked_entries ) || + ( entry_ptr->flush_marker ) ) { - if ( entry_ptr->is_protected ) { + if ( entry_ptr->is_protected ) { - /* we probably have major problems -- but lets flush - * everything we can before we decide whether to flag - * an error. - */ - tried_to_flush_protected_entry = TRUE; - protected_entries++; - - } else if ( entry_ptr->is_pinned ) { - /* Test to see if we are can flush the entry now. - * If we can, go ahead and flush. Note that we - * aren't trying to do a destroy here, so that - * is not an issue. - */ - if ( TRUE ) { /* When we get to multithreaded cache, - * we will need either locking code, - * and/or a test to see if the entry - * is in flushable condition here. - */ + /* we probably have major problems -- but lets flush + * everything we can before we decide whether to flag + * an error. + */ + tried_to_flush_protected_entry = TRUE; + protected_entries++; + + } else if ( entry_ptr->is_pinned ) { + /* Test to see if we are can flush the entry now. + * If we can, go ahead and flush. Note that we + * aren't trying to do a destroy here, so that + * is not an issue. + */ + if(entry_ptr->flush_dep_height == curr_flush_dep_height ) { #if H5C_DO_SANITY_CHECKS - flushed_entries_count++; - flushed_entries_size += entry_ptr->size; + flushed_entries_count++; + flushed_entries_size += entry_ptr->size; #endif /* H5C_DO_SANITY_CHECKS */ - status = H5C_flush_single_entry(f, - primary_dxpl_id, - secondary_dxpl_id, - cache_ptr, - NULL, - entry_ptr->addr, - flags, - &first_flush, - FALSE); - if ( status < 0 ) { - - /* This shouldn't happen -- if it does, we are - * toast so just scream and die. - */ - HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ - "dirty pinned entry flush failed.") - } - flushed_entries_last_pass = TRUE; - } - } else { + status = H5C_flush_single_entry(f, + primary_dxpl_id, + secondary_dxpl_id, + cache_ptr, + NULL, + entry_ptr->addr, + flags, + &first_flush, + FALSE); + if ( status < 0 ) { + + /* This shouldn't happen -- if it does, we are + * toast so just scream and die. + */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ + "dirty pinned entry flush failed.") + } + flushed_during_dep_loop = TRUE; + } /* end if */ + else if(entry_ptr->flush_dep_height < curr_flush_dep_height) + /* This shouldn't happen -- if it does, just scream and die. */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "dirty entry below current flush dep. height.") + } else { + /* Test to see if we are can flush the entry now. + * If we can, go ahead and flush. Note that we + * aren't trying to do a destroy here, so that + * is not an issue. + */ + if(entry_ptr->flush_dep_height == curr_flush_dep_height ){ #if H5C_DO_SANITY_CHECKS - flushed_entries_count++; - flushed_entries_size += entry_ptr->size; + flushed_entries_count++; + flushed_entries_size += entry_ptr->size; #endif /* H5C_DO_SANITY_CHECKS */ - status = H5C_flush_single_entry(f, - primary_dxpl_id, - secondary_dxpl_id, - cache_ptr, - NULL, - entry_ptr->addr, - flags, - &first_flush, - FALSE); - if ( status < 0 ) { + status = H5C_flush_single_entry(f, + primary_dxpl_id, + secondary_dxpl_id, + cache_ptr, + NULL, + entry_ptr->addr, + flags, + &first_flush, + FALSE); + if ( status < 0 ) { + + /* This shouldn't happen -- if it does, we are + * toast so just scream and die. + */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ + "Can't flush entry.") + } + flushed_during_dep_loop = TRUE; + } /* end if */ + else if(entry_ptr->flush_dep_height < curr_flush_dep_height) + /* This shouldn't happen -- if it does, just scream and die. */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "dirty entry below current flush dep. height.") + } /* end else */ + } /* end if */ + } /* while ( node_ptr != NULL ) */ + + /* Check for incrementing flush dependency height */ + if(flushed_during_dep_loop) { + /* If we flushed an entry at this flush dependency height + * start over at the bottom level of the flush dependencies + */ + curr_flush_dep_height = 0; - /* This shouldn't happen -- if it does, we are - * toast so just scream and die. - */ - HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ - "Can't flush entry.") - } - flushed_entries_last_pass = TRUE; - } - } - } /* while ( node_ptr != NULL ) */ + /* Make certain we don't get stuck in an infinite loop */ + flush_dep_passes++; + + /* Set flag for outer loop */ + flushed_entries_last_pass = TRUE; + } /* end if */ + else + curr_flush_dep_height++; + + } /* while ( curr_flush_dep_height <= H5C__NUM_FLUSH_DEP_HEIGHTS) */ +end_of_inner_loop: #if H5C_DO_SANITY_CHECKS /* Verify that the slist size and length are as expected. */ @@ -8736,8 +8764,6 @@ H5C_destroy_flush_dependency(H5C_t * cache_ptr, void *parent_thing, HDassert(H5F_addr_defined(child_entry->addr)); /* Usage checks */ - if(!parent_entry->is_protected) - HGOTO_ERROR(H5E_CACHE, H5E_CANTUNDEPEND, FAIL, "Parent entry isn't pinned") if(!parent_entry->is_pinned) HGOTO_ERROR(H5E_CACHE, H5E_CANTUNDEPEND, FAIL, "Parent entry isn't pinned") if(0 == parent_entry->flush_dep_height) @@ -8785,6 +8811,9 @@ H5C_destroy_flush_dependency(H5C_t * cache_ptr, void *parent_thing, * parent of _any_ child flush dependencies). */ if(0 == parent_entry->flush_dep_height) { + if(!parent_entry->is_protected) + H5C__UPDATE_RP_FOR_UNPIN(cache_ptr, parent_entry, FAIL) + /* Unpin parent entry */ parent_entry->is_pinned = FALSE; H5C__UPDATE_STATS_FOR_UNPIN(cache_ptr, parent_entry) @@ -10263,335 +10292,376 @@ H5C_flush_invalidate_cache(H5F_t * f, while ( ! done ) { - /* first, try to flush-destroy any dirty entries. Do this by - * making a scan through the slist. Note that new dirty entries - * may be created by the flush call backs. Thus it is possible - * that the slist will not be empty after we finish the scan. - */ - - if ( cache_ptr->slist_len == 0 ) { - - node_ptr = NULL; - HDassert( cache_ptr->slist_size == 0 ); - - } else { - - node_ptr = H5SL_first(cache_ptr->slist_ptr); - - if ( node_ptr == NULL ) { - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "slist_len != 0 && node_ptr == NULL"); - } - - next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); - - if ( next_entry_ptr == NULL ) { - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "next_entry_ptr == NULL 1 ?!?!"); - } - HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); - HDassert( next_entry_ptr->is_dirty ); - HDassert( next_entry_ptr->in_slist ); - - } -#if H5C_DO_SANITY_CHECKS - /* Depending on circumstances, H5C_flush_single_entry() will - * remove dirty entries from the slist as it flushes them. - * Thus for sanity checks we must make note of the initial - * slist length and size before we do any flushes. - */ - initial_slist_len = cache_ptr->slist_len; - initial_slist_size = cache_ptr->slist_size; - - /* There is also the possibility that entries will be - * dirtied, resized, and/or renamed as the result of - * calls to the flush callbacks. We use the slist_len_increase - * and slist_size_increase increase fields in struct H5C_t - * to track these changes for purpose of sanity checking. - * To this end, we must zero these fields before we start - * the pass through the slist. - */ - cache_ptr->slist_len_increase = 0; - cache_ptr->slist_size_increase = 0; - - /* Finally, reset the actual_slist_len and actual_slist_size - * fields to zero, as these fields are used to accumulate - * the slist lenght and size that we see as we scan through - * the slist. - */ - actual_slist_len = 0; - actual_slist_size = 0; -#endif /* H5C_DO_SANITY_CHECKS */ + unsigned curr_flush_dep_height = 0; + unsigned flush_dep_passes = 0; - while ( node_ptr != NULL ) + /* Loop over all flush dependency heights of entries */ + while((curr_flush_dep_height <= H5C__NUM_FLUSH_DEP_HEIGHTS) && + (cache_ptr->index_len > 0 ) && + (flush_dep_passes < H5C__MAX_PASSES_ON_FLUSH) ) { - entry_ptr = next_entry_ptr; - - /* With the advent of the fractal heap, it is possible - * that the flush callback will dirty and/or resize - * other entries in the cache. In particular, while - * Quincey has promised me that this will never happen, - * it is possible that the flush callback for an - * entry may protect an entry that is not in the cache, - * perhaps causing the cache to flush and possibly - * evict the entry associated with node_ptr to make - * space for the new entry. - * - * Thus we do a bit of extra sanity checking on entry_ptr, - * and break out of this scan of the skip list if we - * detect major problems. We have a bit of leaway on the - * number of passes though the skip list, so this shouldn't - * be an issue in the flush in and of itself, as it should - * be all but impossible for this to happen more than once - * in any flush. - * - * Observe that that breaking out of the scan early - * shouldn't break the sanity checks just after the end - * of this while loop. - * - * If an entry has merely been marked clean and removed from - * the s-list, we simply break out of the scan. - * - * If the entry has been evicted, we flag an error and - * exit. + hbool_t flushed_during_dep_loop = FALSE; + + /* first, try to flush-destroy any dirty entries. Do this by + * making a scan through the slist. Note that new dirty entries + * may be created by the flush call backs. Thus it is possible + * that the slist will not be empty after we finish the scan. */ -#ifndef NDEBUG - if ( entry_ptr->magic != H5C__H5C_CACHE_ENTRY_T_MAGIC ) { - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "entry_ptr->magic is invalid ?!?!"); + if ( cache_ptr->slist_len == 0 ) { - } else -#endif /* NDEBUG */ - if ( ( ! entry_ptr->is_dirty ) || - ( ! entry_ptr->in_slist ) ) { + node_ptr = NULL; + HDassert( cache_ptr->slist_size == 0 ); - /* the s-list has been modified out from under us. - * break out of the loop. - */ - break; - } + } else { - /* increment node pointer now, before we delete its target - * from the slist. - */ + node_ptr = H5SL_first(cache_ptr->slist_ptr); - node_ptr = H5SL_next(node_ptr); - if ( node_ptr != NULL ) { + if ( node_ptr == NULL ) { + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ + "slist_len != 0 && node_ptr == NULL"); + } next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); if ( next_entry_ptr == NULL ) { HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "next_entry_ptr == NULL 2 ?!?!"); + "next_entry_ptr == NULL 1 ?!?!"); } - HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); + HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); HDassert( next_entry_ptr->is_dirty ); HDassert( next_entry_ptr->in_slist ); - } else { + } +#if H5C_DO_SANITY_CHECKS + /* Depending on circumstances, H5C_flush_single_entry() will + * remove dirty entries from the slist as it flushes them. + * Thus for sanity checks we must make note of the initial + * slist length and size before we do any flushes. + */ + initial_slist_len = cache_ptr->slist_len; + initial_slist_size = cache_ptr->slist_size; - next_entry_ptr = NULL; - } + /* There is also the possibility that entries will be + * dirtied, resized, and/or renamed as the result of + * calls to the flush callbacks. We use the slist_len_increase + * and slist_size_increase increase fields in struct H5C_t + * to track these changes for purpose of sanity checking. + * To this end, we must zero these fields before we start + * the pass through the slist. + */ + cache_ptr->slist_len_increase = 0; + cache_ptr->slist_size_increase = 0; - /* Note that we now remove nodes from the slist as we flush - * the associated entries, instead of leaving them there - * until we are done, and then destroying all nodes in + /* Finally, reset the actual_slist_len and actual_slist_size + * fields to zero, as these fields are used to accumulate + * the slist lenght and size that we see as we scan through * the slist. - * - * While this optimization used to be easy, with the possibility - * of new entries being added to the slist in the midst of the - * flush, we must keep the slist in cannonical form at all - * times. */ - - HDassert( entry_ptr != NULL ); - HDassert( entry_ptr->in_slist ); - -#if H5C_DO_SANITY_CHECKS - /* update actual_slist_len & actual_slist_size before - * the flush. Note that the entry will be removed - * from the slist after the flush, and thus may be - * resized by the flush callback. This is OK, as - * we will catch the size delta in - * cache_ptr->slist_size_increase. - * - * Note that we include pinned entries in this count, even - * though we will not actually flush them. - */ - actual_slist_len++; - actual_slist_size += entry_ptr->size; + actual_slist_len = 0; + actual_slist_size = 0; #endif /* H5C_DO_SANITY_CHECKS */ - if ( entry_ptr->is_protected ) { + while ( node_ptr != NULL ) + { + entry_ptr = next_entry_ptr; - /* we have major problems -- but lets flush - * everything we can before we flag an error. + /* With the advent of the fractal heap, it is possible + * that the flush callback will dirty and/or resize + * other entries in the cache. In particular, while + * Quincey has promised me that this will never happen, + * it is possible that the flush callback for an + * entry may protect an entry that is not in the cache, + * perhaps causing the cache to flush and possibly + * evict the entry associated with node_ptr to make + * space for the new entry. + * + * Thus we do a bit of extra sanity checking on entry_ptr, + * and break out of this scan of the skip list if we + * detect major problems. We have a bit of leaway on the + * number of passes though the skip list, so this shouldn't + * be an issue in the flush in and of itself, as it should + * be all but impossible for this to happen more than once + * in any flush. + * + * Observe that that breaking out of the scan early + * shouldn't break the sanity checks just after the end + * of this while loop. + * + * If an entry has merely been marked clean and removed from + * the s-list, we simply break out of the scan. + * + * If the entry has been evicted, we flag an error and + * exit. */ - protected_entries++; +#ifndef NDEBUG + if ( entry_ptr->magic != H5C__H5C_CACHE_ENTRY_T_MAGIC ) { - } else if ( entry_ptr->is_pinned ) { + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ + "entry_ptr->magic is invalid ?!?!"); + + } else +#endif /* NDEBUG */ + if ( ( ! entry_ptr->is_dirty ) || + ( ! entry_ptr->in_slist ) ) { - /* Test to see if we are can flush the entry now. - * If we can, go ahead and flush, but don't tell - * H5C_flush_single_entry() to destroy the entry - * as pinned entries can't be evicted. + /* the s-list has been modified out from under us. + * break out of the loop. + */ + goto end_of_inner_loop;; + } + + /* increment node pointer now, before we delete its target + * from the slist. */ - if ( TRUE ) { /* When we get to multithreaded cache, - * we will need either locking code, and/or - * a test to see if the entry is in flushable - * condition here. - */ - status = H5C_flush_single_entry(f, - primary_dxpl_id, - secondary_dxpl_id, - cache_ptr, - NULL, - entry_ptr->addr, - H5C__NO_FLAGS_SET, - &first_flush, - FALSE); - if ( status < 0 ) { + node_ptr = H5SL_next(node_ptr); + if ( node_ptr != NULL ) { - /* This shouldn't happen -- if it does, we are toast - * so just scream and die. - */ + next_entry_ptr = (H5C_cache_entry_t *)H5SL_item(node_ptr); - HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ - "dirty pinned entry flush failed.") + if ( next_entry_ptr == NULL ) { + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ + "next_entry_ptr == NULL 2 ?!?!"); } + HDassert( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ); + HDassert( next_entry_ptr->is_dirty ); + HDassert( next_entry_ptr->in_slist ); + + } else { + + next_entry_ptr = NULL; } - } else { - status = H5C_flush_single_entry(f, - primary_dxpl_id, - secondary_dxpl_id, - cache_ptr, - NULL, - entry_ptr->addr, - (cooked_flags | - H5C__FLUSH_INVALIDATE_FLAG), - &first_flush, - TRUE); - if ( status < 0 ) { + /* Note that we now remove nodes from the slist as we flush + * the associated entries, instead of leaving them there + * until we are done, and then destroying all nodes in + * the slist. + * + * While this optimization used to be easy, with the possibility + * of new entries being added to the slist in the midst of the + * flush, we must keep the slist in cannonical form at all + * times. + */ + + HDassert( entry_ptr != NULL ); + HDassert( entry_ptr->in_slist ); - /* This shouldn't happen -- if it does, we are toast so - * just scream and die. +#if H5C_DO_SANITY_CHECKS + /* update actual_slist_len & actual_slist_size before + * the flush. Note that the entry will be removed + * from the slist after the flush, and thus may be + * resized by the flush callback. This is OK, as + * we will catch the size delta in + * cache_ptr->slist_size_increase. + * + * Note that we include pinned entries in this count, even + * though we will not actually flush them. + */ + actual_slist_len++; + actual_slist_size += entry_ptr->size; +#endif /* H5C_DO_SANITY_CHECKS */ + + if ( entry_ptr->is_protected ) { + + /* we have major problems -- but lets flush + * everything we can before we flag an error. */ + protected_entries++; - HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ - "dirty entry flush destroy failed.") - } - } - } /* end while loop scanning skip list */ + } else if ( entry_ptr->is_pinned ) { + + /* Test to see if we are can flush the entry now. + * If we can, go ahead and flush, but don't tell + * H5C_flush_single_entry() to destroy the entry + * as pinned entries can't be evicted. + */ + if(entry_ptr->flush_dep_height == curr_flush_dep_height ) { + status = H5C_flush_single_entry(f, + primary_dxpl_id, + secondary_dxpl_id, + cache_ptr, + NULL, + entry_ptr->addr, + H5C__NO_FLAGS_SET, + &first_flush, + FALSE); + if ( status < 0 ) { + + /* This shouldn't happen -- if it does, we are toast + * so just scream and die. + */ + + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ + "dirty pinned entry flush failed.") + } /* end if */ + flushed_during_dep_loop = TRUE; + } /* end if */ + else if(entry_ptr->flush_dep_height < curr_flush_dep_height) + /* This shouldn't happen -- if it does, just scream and die. */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "dirty entry below current flush dep. height.") + } /* end if */ + else { + if(entry_ptr->flush_dep_height == curr_flush_dep_height ){ + + status = H5C_flush_single_entry(f, + primary_dxpl_id, + secondary_dxpl_id, + cache_ptr, + NULL, + entry_ptr->addr, + (cooked_flags | + H5C__FLUSH_INVALIDATE_FLAG), + &first_flush, + TRUE); + if ( status < 0 ) { + + /* This shouldn't happen -- if it does, we are toast so + * just scream and die. + */ + + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ + "dirty entry flush destroy failed.") + } /* end if */ + flushed_during_dep_loop = TRUE; + } /* end if */ + else if(entry_ptr->flush_dep_height < curr_flush_dep_height) + /* This shouldn't happen -- if it does, just scream and die. */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "dirty entry below current flush dep. height.") + } /* end else */ + } /* end while loop scanning skip list */ #if H5C_DO_SANITY_CHECKS - /* It is possible that entries were added to the slist during - * the scan, either before or after scan pointer. The following - * asserts take this into account. - * - * Don't bother with the sanity checks if node_ptr != NULL, as - * in this case we broke out of the loop because it got changed - * out from under us. - */ + /* It is possible that entries were added to the slist during + * the scan, either before or after scan pointer. The following + * asserts take this into account. + * + * Don't bother with the sanity checks if node_ptr != NULL, as + * in this case we broke out of the loop because it got changed + * out from under us. + */ - if ( node_ptr == NULL ) { + if ( node_ptr == NULL ) { - HDassert( (actual_slist_len + cache_ptr->slist_len) == - (initial_slist_len + cache_ptr->slist_len_increase) ); - HDassert( (actual_slist_size + cache_ptr->slist_size) == - (initial_slist_size + cache_ptr->slist_size_increase) ); - } + HDassert( (actual_slist_len + cache_ptr->slist_len) == + (initial_slist_len + cache_ptr->slist_len_increase) ); + HDassert( (actual_slist_size + cache_ptr->slist_size) == + (initial_slist_size + cache_ptr->slist_size_increase) ); + } #endif /* H5C_DO_SANITY_CHECKS */ - /* Since we are doing a destroy, we must make a pass through - * the hash table and try to flush - destroy all entries that - * remain. - * - * It used to be that all entries remaining in the cache at - * this point had to be clean, but with the fractal heap mods - * this may not be the case. If so, we will flush entries out - * of increasing address order. - * - * Writes to disk are possible here. - */ - for ( i = 0; i < H5C__HASH_TABLE_LEN; i++ ) - { - next_entry_ptr = cache_ptr->index[i]; - - while ( next_entry_ptr != NULL ) + /* Since we are doing a destroy, we must make a pass through + * the hash table and try to flush - destroy all entries that + * remain. + * + * It used to be that all entries remaining in the cache at + * this point had to be clean, but with the fractal heap mods + * this may not be the case. If so, we will flush entries out + * of increasing address order. + * + * Writes to disk are possible here. + */ + for ( i = 0; i < H5C__HASH_TABLE_LEN; i++ ) { - entry_ptr = next_entry_ptr; + next_entry_ptr = cache_ptr->index[i]; - next_entry_ptr = entry_ptr->ht_next; - HDassert ( ( next_entry_ptr == NULL ) || - ( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ) ); - if ( entry_ptr->is_protected ) { + while ( next_entry_ptr != NULL ) + { + entry_ptr = next_entry_ptr; - /* we have major problems -- but lets flush and destroy - * everything we can before we flag an error. - */ - protected_entries++; + next_entry_ptr = entry_ptr->ht_next; + HDassert ( ( next_entry_ptr == NULL ) || + ( next_entry_ptr->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC ) ); + if ( entry_ptr->is_protected ) { - if ( ! entry_ptr->in_slist ) { + /* we have major problems -- but lets flush and destroy + * everything we can before we flag an error. + */ + protected_entries++; - HDassert( !(entry_ptr->is_dirty) ); - } - } else if ( ! ( entry_ptr->is_pinned ) ) { + if ( ! entry_ptr->in_slist ) { - status = H5C_flush_single_entry(f, - primary_dxpl_id, - secondary_dxpl_id, - cache_ptr, - NULL, - entry_ptr->addr, - (cooked_flags | - H5C__FLUSH_INVALIDATE_FLAG), - &first_flush, - TRUE); - if ( status < 0 ) { + HDassert( !(entry_ptr->is_dirty) ); + } + } else if ( ! ( entry_ptr->is_pinned ) ) { - /* This shouldn't happen -- if it does, we are toast so - * just scream and die. + /* Test to see if we are can flush the entry now. + * If we can, go ahead and flush. */ + if(entry_ptr->flush_dep_height == curr_flush_dep_height ){ + status = H5C_flush_single_entry(f, + primary_dxpl_id, + secondary_dxpl_id, + cache_ptr, + NULL, + entry_ptr->addr, + (cooked_flags | + H5C__FLUSH_INVALIDATE_FLAG), + &first_flush, + TRUE); + if ( status < 0 ) { - HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ - "Entry flush destroy failed.") - } - } - /* We can't do anything if the entry is pinned. The - * hope is that the entry will be unpinned as the - * result of destroys of entries that reference it. - * - * We detect this by noting the change in the number - * of pinned entries from pass to pass. If it stops - * shrinking before it hits zero, we scream and die. - */ - /* if the flush function on the entry we last evicted - * loaded an entry into cache (as Quincey has promised me - * it never will), and if the cache was full, it is - * possible that *next_entry_ptr was flushed or evicted. - * - * Test to see if this happened here. Note that if this - * test is triggred, we are accessing a deallocated piece - * of dynamically allocated memory, so we just scream and - * die. - */ -#ifndef NDEBUG - if ( ( next_entry_ptr != NULL ) && - ( next_entry_ptr->magic != - H5C__H5C_CACHE_ENTRY_T_MAGIC ) ) { + /* This shouldn't happen -- if it does, we are toast so + * just scream and die. + */ - /* Something horrible has happened to - * *next_entry_ptr -- scream and die. + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, \ + "Entry flush destroy failed.") + } + flushed_during_dep_loop = TRUE; + } /* end if */ + else if(entry_ptr->flush_dep_height < curr_flush_dep_height) + /* This shouldn't happen -- if it does, just scream and die. */ + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "dirty entry below current flush dep. height.") + } /* end if */ + /* We can't do anything if the entry is pinned. The + * hope is that the entry will be unpinned as the + * result of destroys of entries that reference it. + * + * We detect this by noting the change in the number + * of pinned entries from pass to pass. If it stops + * shrinking before it hits zero, we scream and die. */ - HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ - "next_entry_ptr->magic is invalid?!?!?.") - } + /* if the flush function on the entry we last evicted + * loaded an entry into cache (as Quincey has promised me + * it never will), and if the cache was full, it is + * possible that *next_entry_ptr was flushed or evicted. + * + * Test to see if this happened here. Note that if this + * test is triggred, we are accessing a deallocated piece + * of dynamically allocated memory, so we just scream and + * die. + */ +#ifndef NDEBUG + if ( ( next_entry_ptr != NULL ) && + ( next_entry_ptr->magic != + H5C__H5C_CACHE_ENTRY_T_MAGIC ) ) { + + /* Something horrible has happened to + * *next_entry_ptr -- scream and die. + */ + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, \ + "next_entry_ptr->magic is invalid?!?!?.") + } #endif /* NDEBUG */ - } /* end while loop scanning hash table bin */ - } /* end for loop scanning hash table */ + } /* end while loop scanning hash table bin */ + } /* end for loop scanning hash table */ + + /* Check for incrementing flush dependency height */ + if(flushed_during_dep_loop) { + /* If we flushed an entry at this flush dependency height + * start over at the bottom level of the flush dependencies + */ + curr_flush_dep_height = 0; + + /* Make certain we don't get stuck in an infinite loop */ + flush_dep_passes++; + } /* end if */ + else + curr_flush_dep_height++; + + } /* end while loop over flush dependency heights */ +end_of_inner_loop: old_pel_len = cur_pel_len; cur_pel_len = cache_ptr->pel_len; diff --git a/src/H5Cprivate.h b/src/H5Cprivate.h index 65e6f6c..855c861 100644 --- a/src/H5Cprivate.h +++ b/src/H5Cprivate.h @@ -185,7 +185,7 @@ typedef herr_t (*H5C_log_flush_func_t)(H5C_t * cache_ptr, /* Maximum height of flush dependency relationships between entries. This is * currently tuned to the extensible array (H5EA) data structure, which only - * requires 4 levels of dependency (i.e. heights 0-3). + * requires 4 levels of dependency (i.e. heights 0-4). */ #define H5C__NUM_FLUSH_DEP_HEIGHTS 4 |