Linux kernel 6.13-rc3 includes a change that eliminates a performance regression in the KVM hypervisor related to the slow processing of CPUID calls on new CPUs. For example, on Intel Emerald Rapids CPUs, CPUID operations are performed 3-4 times slower than on Intel Skylake CPUs. This characteristic has led to reduced performance in the KVM hypervisor, which uses CPUID during the save and restore of processor states on each handover to the virtual machine when nested virtualization is in use. To address the issue, a reduced patch has been accepted into the 6.13 kernel branch, which allows for up to a 40% reduction in operation time even on SkyLake CPUs, thanks to CPUID caching. A complete version of the patch will be presented in kernel 6.14, further improving performance.
Additionally, there is discussion in the Linux kernel developers' mailing list about ending support for 32-bit systems in the hypervisor. KVMOriginally, a patch was proposed for the kernel that removes support for KVM virtualization on the x86-32 platform. Concerns were raised in response that support for x86-32 might be useful for testing host mode operation in general on 32-bit systems since the kernel still supports three other 32-bit architectures in KVM besides x86-32. In response, another patch was presented, this time removing support for all 32-bit systems from KVM.
The desire to eliminate support for 32-bit systems is explained by the fact that, even on mass-produced 32-bit ARM processors such as the ARM Cortex A7, virtualization has not developed and usage virtuaalmasinate on 32-bit ARM systems has been so rare that support for these systems in KVM was removed in 2020 without objection. It is noted that for other 32-bit systems, the changes will only affect exotic platforms that have long been obsolete or never were popular. Among the processors affected by the patch are: Intel Core Duo/Silverthorne, PowerPC e300/e500/e600, and MIPS P5600. The patch has been published as an RFC, meaning it is open for discussion.
Allikas: opennet.ru
