Isolating GPU access in its own process
summary
Chromium moves all GPU calls into a separate GPU process and uses IPC over shared memory with a command buffer ring and IPC channels to serialize GLES2 calls; software video frames are decoded by ffmpeg in renderers, copied into shared memory, uploaded as GL images and synchronized with sync token / mailbox fencing. The talk shows the Linux composition pipeline that stitches many render processes into one window, describes Chromium-specific GLES2 extensions, and explains security, robustness and dependency-separation tradeoffs.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:27Isolating Gpu Access in Its Own Process
- 3:08Composition
- 4:54Css Demo
- 8:58Process Relationships
- 14:14Sink Token
- 17:07Software Decoding
- 25:45Security
- 29:29Dependency Separation
related talks
01h 00mIsolating GPU access in its own process
55mLinux Security APIs and the Chromium Sandbox
33mWe are all good with browsers made by others
32mReasonably Secure GPU Acceleration
58mUntangling Browser Threads
42mChrome Internals and Rendering Performance
24mMigration from WebGL to WebGPU
50mLinux Security and How Web Browser Sandboxes Really Work