[tracing] Fix PartitionAlloc dumper reporting
- Fix virtual vs resident size reporting in the "size" column: Prior to this change the PartitionAlloc dumper was reporting the mmap-ed (virtual address spave) size in the "size" column. This is inconsistent with the rest of the dumpers, which strive to report an estimation of actually resident memory in the "size" column. This change moves the PA resident size into "size". - Remove the "thread_1234" sub-node: turns out that there can be at most one PartitionAlloc instance per-process, even in single-process mode. - Improve reporting of allocated_objects: - Each bucket gets a column "allocated_objects_size", previously called "active_size" (confusing), which reports the size of bytes requested by PA clients. - The per-bucket "allocated_objects" sub-rows are removed: they just added visual clutter and didn't add any extra information than an extra column. - A global allocated_objects node is introduced. This will be required by subsequent CLs to properly account memory reporter by other pieces of blink. This will be to avoid that WebCache memory that is allocated via PA gets double-counted in the effective_size. - Improve other column names for sake of clarity: - committed -> virtual_committed - num_{active,full,...} -> {active,full,...}_pages. The previous name didn't make it clear that they count pages and not objects. BUG=488472 Review URL: https://codereview.chromium.org/1312843010 git-svn-id: svn://svn.chromium.org/blink/trunk@201972 bbb929c8-8fbe-4397-9dbb-9b2b20218538
Showing
Please register or sign in to comment