A Dummy Wait Op Has Been Unnecessarily Slowing AMD Chips
When you’ve run into workloads in your AMD powered Linux field which can be slower than anticipated, a repair is on the best way to make your life higher. There’s an historic workaround which was wanted to make sure some chipsets had time for STPCLK# assertions, which dates again to 2002 when Linux first included ACPI assist. This was achieved with a dummy wait op that gave previous chipsets, like these from VIA, sufficient time to correctly end operations earlier than transferring on.
The issue is that it hasn’t been mandatory in fairly a while, however it was by no means eliminated. This doesn’t have an enormous impact on a small Ryzen system, nevertheless it additionally applies to EPYC processors. That signifies that even a small delay can add up once you deploy these chips at scale. The repair has simply been merged into Linux 6.0 which signifies that Phoronix must get busy benchmarking.
You may learn particulars about why this repair was mandatory for AMD ACPI right here, with benchmarks to observe quickly.