False warning about absence of KVM on Arch Linux

scripts/templates/qemu/run_qemu checks for presence of KVM in the system with which kvm:

Lines 12 to 20 in 3cde30b

    
           if ! which kvm &> /dev/null; then 
        
              echo "WARNING: it looks like KVM is not installed on this machine." 
        
              echo "Please install qemu-kvm and add the current user to the kvm group " 
        
              echo "OR just run the *nokvm* version of this script." 
        
              echo "The launch of QEMU with the -enable-kvm option will likely fail." 
        
              echo 
        
           else

I have KVM installed and the KVM module is loaded into the kernel:

$ lsmod | grep kvm
kvm_intel             335872  0
kvm                  1036288  1 kvm_intel
irqbypass              16384  1 kvm

But there is no kvm binary, so the check above fails.

The script also actually runs QEMU with -enable-kvm successfully and info kvm in the QEMU monitor outputs kvm support: enabled.

I understand that this is only a false warning, not a false error, but maybe there is another way of checking for absence of KVM that works on Arch Linux too?

Good point.
What about checking KVM's presence with: if lsmod | grep kvm &> /dev/null; then ?

EDIT: the problem is: can this new approach fail because the KVM module is loaded on-demand the first time QEMU need it?

Grepping only kvm is not enough, because for KVM to work properly either kvm_intel or kvm_amd has to be loaded. For example, on my machine QEMU errors with failed to initialize kvm without kvm_intel. So it's reasonable to write:

if ! (lsmod | grep kvm_amd &> /dev/null) && ! (lsmod | grep kvm_intel &> /dev/null); then

Regarding your last question, that's not my case, at least. I have just rebooted, run lsmod | grep kvm before running QEMU and the modules are there. I don't have any /etc/modprobe.d rules or any modprobe-ing scripts, hence the modules were loaded during boot thanks to some hardware detection.

EDIT: I have not fully answered your question actually. I think QEMU does not try to load the modules itself on a failure because otherwise I would not get failed to initialize kvm error I mentioned above. And I don't know if Linux can automatically modprobe kvm in this case either.

Mmm... isn't checking for either kvm_amd or kvm_intel a bit too tight as a requirement? If tomorrow the KVM guys create a new kvm_generic_XYZ module, those checks will fail and we'll see a false positive again. On the other hand, if we are in the corner case described here: https://wiki.archlinux.org/title/KVM:

Tip: If modprobing kvm_intel or kvm_amd fails but modprobing kvm succeeds, and lscpu claims that hardware acceleration is supported, check the BIOS settings. Some vendors, especially laptop vendors, disable these processor extensions by default. To determine whether there is no hardware support or whether the extensions are disabled in BIOS, the output from dmesg after having failed to modprobe will tell.

We'll have kvm loaded but neither kvm_amd or kvm_intel, because KVM is installed but the loading of kvm_amd or kvm_intel failed, we won't see a warning and QEMU -enable-kvm will fail with a different error. I believe that's acceptable as it's unlikely to happen and is not related with the installation of KVM per se, but with other things like BIOS settings.

Merged. Thanks for the fix!

	if ! which kvm &> /dev/null; then

	echo "WARNING: it looks like KVM is not installed on this machine."
	echo "Please install qemu-kvm and add the current user to the kvm group "
	echo "OR just run the nokvm version of this script."
	echo "The launch of QEMU with the -enable-kvm option will likely fail."
	echo

	else