4b81a2d89a
This patch is about the bug that VMs can't idle. ACRN enables VM's C-state by extracting host's C-state table. The C-state table has two types of interfaces: system-IO and mwait. VMs just need one of them. ACRN can support both. Currently we are telling users to use the system-IO type. That is, by adding 'nomwait intel_idle.max_cstate=0' to host Linux's CMD line when using board_inspector. (The reaseon we were using system-IO is that mwait was buggy on Apollo Lake.) But recent tests show that system-IO is somehow buggy. Linux c-state driver(no matter intel_cstate or acpi_cstate) fails to enter idle state with system-IO. This can always be reproduced on native environments. MPERF counters show CPU cores are not in real idle state as expected. To enable C-state in VMs, we have to switch to mwait. As ACRN has already supported both system-IO and mwait, we don't have to modify any code. We just need to tell user to use mwait instead of system-IO. That is, don't add 'nomwait intel_idle.max_cstate=0' to host Linux's CMD line when using board_inspector. Just add 'intel_idle.max_cstate=0' Due to the Apollo Lake's mwait bug, 'nomwait' is still needed for Apollo Lake as an exception. Tracked-On: #7371 Reviewed-by: Junjie Mao <junjie.mao@intel.com> Signed-off-by: Zhou, Wu <wu.zhou@intel.com> |
||
|---|---|---|
| .. | ||
| acpi.py | ||
| board_parser.py | ||
| clos.py | ||
| dmar.py | ||
| dmi.py | ||
| misc.py | ||
| parser_lib.py | ||
| pci_dev.py | ||