Intel backtracks on benchmark clause of microcode license agreement

Yesterday there was a story about Intel forbidding clients to publish details about the performance hit of the latest microcode updates, which mitigate the L1TF security vulnerability. The gag order received a lot of criticism but now Intel backtracks and reworked the license to no longer prohibit benchmarking:

[Open source pioneer Bruce] Perens, in a phone interview with The Register, approved of the change.

"This is a relatively innocuous license for proprietary software and it can be distributed in the non-free section of Debian, which is where is used to be, and it should be distributable by other Linux distributions," he said.

As to how Intel managed to shoot itself in the foot, Perens speculates that whoever wrote the text did not consider where the microcode was going and what the implications could be.

"You can't expect every lawyer to understand CPUs," he said. "Sometimes they have to have a deep conversation with their technical people."

The Register writes Intel's plan backfired, as this drew a lot more attention to the issue. The reason for the insistence on silence is that the latest mitigation hurts performance real bad in certain use cases. At the moment, you need to turn hyper-threading off to protect virtual machines against Foreshadow attacks. Most HT testing shows losses in the 0-30 percent range with current fixes.