P4
The comment and the condition here are wrong:
void SpaceManager::verify() {
// If there are blocks in the dictionary, then
// verification of chunks does not work since
// being in the dictionary alters a chunk.
if (block_freelists() != NULL && block_freelists()->total_size() == 0) {
....
}
}
They used to make sense: in an earlier version of the Metaspace ("6964458: Reimplement class meta-data storage to use native memory") the Metablocks inside a metachunk always had headers and therefore were walkable. So, Metachunk::verify walked all the blocks - which was not possible if blocks were returned with Metaspace::deallocate() and added to the block dictionary.
This is not true anymore, now MetaBlocks generally do not have a header, so MetaChunk::verify() does not walk them and is generally oblivious to anything happening in the MetaChunk payload area. So, one can now always call MetaChunk::verify(), regardless if MetaBlocks are in the block dictionary or not.
void SpaceManager::verify() {
// If there are blocks in the dictionary, then
// verification of chunks does not work since
// being in the dictionary alters a chunk.
if (block_freelists() != NULL && block_freelists()->total_size() == 0) {
....
}
}
They used to make sense: in an earlier version of the Metaspace ("6964458: Reimplement class meta-data storage to use native memory") the Metablocks inside a metachunk always had headers and therefore were walkable. So, Metachunk::verify walked all the blocks - which was not possible if blocks were returned with Metaspace::deallocate() and added to the block dictionary.
This is not true anymore, now MetaBlocks generally do not have a header, so MetaChunk::verify() does not walk them and is generally oblivious to anything happening in the MetaChunk payload area. So, one can now always call MetaChunk::verify(), regardless if MetaBlocks are in the block dictionary or not.