655638 - Negative values for "heap-used/other" in about:memory due to overlap between image and canvas memory reporters

Status: RESOLVED FIXED

(Toolkit :: General, defect)

Description

[Nicholas Nethercote [inactive]]

While running www.thewildernessdowntown.com I saw this about:memory?verbose report:

Used Heap Memory
318,096,000 B (100.0%) -- heap-used
├──143,534,760 B (45.12%) -- content
│  └──143,534,760 B (45.12%) -- canvas
│     └──143,534,760 B (45.12%) -- 2d-pixel-bytes
├──117,853,666 B (37.05%) -- images
│  ├──117,632,046 B (36.98%) -- content
│  │  ├──117,632,046 B (36.98%) -- used
│  │  │  ├──109,251,088 B (34.35%) -- uncompressed
│  │  │  └────8,380,958 B (02.63%) -- raw
│  │  └────────────0 B (00.00%) -- unused
│  │               ├──0 B (00.00%) -- raw
│  │               └──0 B (00.00%) -- uncompressed
│  └──────221,620 B (00.07%) -- chrome
│         ├──221,620 B (00.07%) -- used
│         │  ├──221,620 B (00.07%) -- uncompressed
│         │  └────────0 B (00.00%) -- raw
│         │  ├──221,620 B (00.07%) -- uncompressed
│         │  └────────0 B (00.00%) -- raw
│         └────────0 B (00.00%) -- unused
│                  ├──0 B (00.00%) -- raw
│                  └──0 B (00.00%) -- uncompressed
├───82,941,384 B (26.07%) -- js
│   ├──80,740,352 B (25.38%) -- gc-heap
│   ├───1,188,336 B (00.37%) -- tjit-data
│   │   └──1,188,336 B (00.37%) -- allocators
│   │      ├────888,000 B (00.28%) -- reserve
│   │      └────300,336 B (00.09%) -- main
│   └───1,012,696 B (00.32%) -- mjit-data
├────3,208,944 B (01.01%) -- storage
│    └──3,208,944 B (01.01%) -- sqlite
│       ├──2,150,264 B (00.68%) -- places.sqlite
│       │  ├──1,884,568 B (00.59%) -- cache-used
│       │  ├────188,104 B (00.06%) -- stmt-used
│       │  └─────77,592 B (00.02%) -- schema-used
│       ├────610,576 B (00.19%) -- other
│       ├────214,304 B (00.07%) -- urlclassifier3.sqlite
│       │    ├──109,120 B (00.03%) -- cache-used
│       │    ├──100,944 B (00.03%) -- stmt-used
│       │    └────4,240 B (00.00%) -- schema-used
│       ├────146,992 B (00.05%) -- cookies.sqlite
│       │    ├──131,520 B (00.04%) -- cache-used
│       │    ├───12,528 B (00.00%) -- stmt-used
│       │    └────2,944 B (00.00%) -- schema-used
│       ├─────29,720 B (00.01%) -- formhistory.sqlite
│       │     ├──22,736 B (00.01%) -- stmt-used
│       │     ├───4,328 B (00.00%) -- cache-used
│       │     └───2,656 B (00.00%) -- schema-used
│       ├─────21,808 B (00.01%) -- content-prefs.sqlite
│       │     └───2,656 B (00.00%) -- schema-used
│       ├─────21,808 B (00.01%) -- content-prefs.sqlite
│       │     ├───9,632 B (00.00%) -- stmt-used
│       │     ├───8,152 B (00.00%) -- cache-used
│       │     └───4,024 B (00.00%) -- schema-used
│       ├─────13,848 B (00.00%) -- downloads.sqlite
│       │     ├───6,704 B (00.00%) -- stmt-used
│       │     ├───4,320 B (00.00%) -- cache-used
│       │     └───2,824 B (00.00%) -- schema-used
│       ├─────13,160 B (00.00%) -- permissions.sqlite
│       │     ├───6,776 B (00.00%) -- stmt-used
│       │     ├───4,328 B (00.00%) -- cache-used
│       │     └───2,056 B (00.00%) -- schema-used
│       └──────8,272 B (00.00%) -- search.sqlite
│              ├──4,312 B (00.00%) -- cache-used
│              ├──2,000 B (00.00%) -- schema-used
│              └──1,960 B (00.00%) -- stmt-used
├────1,236,113 B (00.39%) -- layout
│    ├──1,236,113 B (00.39%) -- all
│    └──────────0 B (00.00%) -- bidi
├──────177,276 B (00.06%) -- gfx
│      └──177,276 B (00.06%) -- surface
│         └──177,276 B (00.06%) -- image
└──-30,856,143 B (-9.70%) -- other

"heap-used/other" is negative, which indicates that some of the heap-used reporters are overlapping.  I bet that "content/canvas/2d-pixel-bytes" and "images/content/used", the two biggest numbers here, are the culprits.

Comment 1

[Nicholas Nethercote [inactive]]

I talked with Bas about this on IRC.  He thinks that "images/content/used" and "content/canvas/2d-pixel-bytes" can overlap.  He also thinks that "content/canvas/2d-pixel-bytes" can be in VRAM, on Windows at least.  Both of those points sound like arguments for moving content/canvas/2d-pixel-bytes out of the "Used Heap" tree into the "Other Measurements" list.

Comment 2

[Nicholas Nethercote [inactive]]

Attachment: patch

Trivial patch.

Comment 3

[Bas Schouten (:bas.schouten)]

Comment on attachment 531237 [details] [diff] [review]
patch

Review of attachment 531237 [details] [diff] [review]:
-----------------------------------------------------------------

Looks good in general, only a small comment.

::: content/canvas/src/nsCanvasRenderingContext2D.cpp
@@ +173,5 @@
>  }
>  
> +// This isn't "heap-used/content/canvas/2d-pixel-bytes" because it overlaps
> +// with "heap-used/images/content/used", and so has to go in the "Other
> +// Measurements" section of about:memory.  See bug 655638 for details.

Are we sure that's actually what it overlaps with? It can also simply not be on the heap for example (but in VRAM), I'd rather say something like: "This isn't "heap-used/content/canvas/2d-pixel-bytes" because the pixels of a canvas do not necessarily exist on the heap. And if they do, should be tracked by underlying surface implementations. See bug 655638 for details."

Comment 4

[Nicholas Nethercote [inactive]]

(In reply to comment #3)
>
> "This
> isn't "heap-used/content/canvas/2d-pixel-bytes" because the pixels of a
> canvas do not necessarily exist on the heap. And if they do, should be
> tracked by underlying surface implementations. See bug 655638 for details."

I'll update, thanks!

Comment 5

[Nicholas Nethercote [inactive]]

http://hg.mozilla.org/mozilla-central/rev/5ba52131c1a2