Re: [PATCH v10 02/28] KVM: VMX: Initialize VM entry/exit FRED controls in vmcs_config

[Binbin Wu]

On 10/26/2026 1:00 PM, Sohil Mehta wrote:
> On 10/22/2026 8:10 PM, Sohil Mehta wrote:
> 
>> I couldn't find any precedent for this. I don't have a strong preference
>> here. Is it worth making future-proof?
> 
> 
> I did a bit of digging. The current entry in the table is the result of
> Sean's preference provided early on in the FRED series. See
> 
> https://lore.kernel.org/all/Z739wdGmk4ZuWJ8v@xxxxxxxxxx/
> https://lore.kernel.org/all/ZUz3cPmnqSq7Lol9@xxxxxxxxxx/
> 
> IIUC, we want the FRED entry/exit controls to be consistent without
> depending on the ordering.
> 
> The current vmx_check_entry_exit_pairs() doesn't cover the FRED case (a
> single entry bit mapping to two exit bits). So, the updated check would
> cover this scenario.

OK, thanks for the info!

> 
>> if (!(entry_controls & n_ctrl) == !(exit_controls & x_ctrl) &&
>>     !(entry_controls & n_ctrl) == ((exit_controls & x_ctrl) != x_ctrl))
>> 		continue;
>>
> 
> 
> 1st line: the entry control is set if and only if *any* exit bit is set.
> 2nd line: the entry control is set if and only if *all* exit bits are
> set (which is specific to FRED for now).
> 
> Also, to make it easier to read, planning to reformat it as:
> 
> struct {
> 	u32 entry_control;
> 	u64 exit_control;
> } const vmcs_entry_exit2_pairs[] = {
> 	{ VM_ENTRY_LOAD_IA32_FRED, SECONDARY_VM_EXIT_SAVE_IA32_FRED |
> 				   SECONDARY_VM_EXIT_LOAD_IA32_FRED },
> };
> 
> Please let me know if you disagree with the approach.

It looks good to me.