4 ms·

Said current patches haven't been merged. Just posted to the mailing list for review. Retpolines for Spectre, and PTI for Meltdown, are still in the kernel. Li

by Valmar 9y ago

Said current patches haven't been merged. Just posted to the mailing list for review. Retpolines for Spectre, and PTI for Meltdown, are still in the kernel. Linus doesn't have too much disagreement with them. It's the IBRS / IBPB code for Intel's CPUs that Linus has a problem with. Also, not everything can be mitigated in microcode, for many reasons, not all of which I understand myself. Sometimes, only a fundamental hardware-redesign can fix the issues proper.

Animats 9y ago
Well, for one thing, modern CPUs aren't microcoded much. Real microcoded CPUs went out with the 1980s. The microcode in today's fast CPUs is mostly initialization, and maybe emulation support for some instructions nobody uses any more, like decimal arithmetic or 16-bit mode. The microcode can maybe turn some stuff off and tweak some tuning parameters. Anyone know what Intel's "fixes" really do?
- StudentStuff 9y ago
  Wat? All the major hardware manufacturers from Intel to the hundreds of no name ARM licensees fix and cover up major issues in silicon with fixes in microcode (unless most chips from a particular mask/process are too screwed up, see Intel's 7nm process). Doing otherwise would throw the onus onto kernel maintainers for Windows, Linux, and the various BSD distros, and they don't want a crapton of one off fixes for the plethora of CPU errata that Intel and others patch in microcode.
- Sephr 9y ago
  Microcode is a little more capable than that. See https://media.ccc.de/v/34c3-9058-everything_you_want_to_know_about_x86_microcode_but_might_have_been_afraid_to_ask https://media.ccc.de/v/34c3-9058-everything_you_want_to_know... Edit: s/Intel m/M/
  - peoplewindow 9y ago
    That talk is about AMD.
- nabla9 9y ago
  All instructions that decode into more than four micro-ops still go trough microcode. Typically 5% of micro-ops are still coming from the microcode sequencer even in modern architectures. Compiler has large effect. Common operations like CALL and PUSH go trough microcode even in m64 but there are alternatives. Many floating point operations have assists. Nonoptimal memory alignment might go trough microcode.

by Valmar 9y ago

Animats 9y ago

Well, for one thing, modern CPUs aren't microcoded much. Real microcoded CPUs went out with the 1980s. The microcode in today's fast CPUs is mostly initialization, and maybe emulation support for some instructions nobody uses any more, like decimal arithmetic or 16-bit mode. The microcode can maybe turn some stuff off and tweak some tuning parameters. Anyone know what Intel's "fixes" really do?

StudentStuff 9y ago
Wat? All the major hardware manufacturers from Intel to the hundreds of no name ARM licensees fix and cover up major issues in silicon with fixes in microcode (unless most chips from a particular mask/process are too screwed up, see Intel's 7nm process). Doing otherwise would throw the onus onto kernel maintainers for Windows, Linux, and the various BSD distros, and they don't want a crapton of one off fixes for the plethora of CPU errata that Intel and others patch in microcode.
Sephr 9y ago
Microcode is a little more capable than that. See https://media.ccc.de/v/34c3-9058-everything_you_want_to_know_about_x86_microcode_but_might_have_been_afraid_to_ask https://media.ccc.de/v/34c3-9058-everything_you_want_to_know... Edit: s/Intel m/M/
- peoplewindow 9y ago
  That talk is about AMD.
nabla9 9y ago
All instructions that decode into more than four micro-ops still go trough microcode. Typically 5% of micro-ops are still coming from the microcode sequencer even in modern architectures. Compiler has large effect. Common operations like CALL and PUSH go trough microcode even in m64 but there are alternatives. Many floating point operations have assists. Nonoptimal memory alignment might go trough microcode.