Building hypervisor from scratch part 1
In this blog series, we’ll be creating a type 1 (bare metal) hypervisor from scratch. We’ll be creating a Linux kernel module that will function as a VMM (Virtual Machine monitor) for the AMD CPU’s native SVM functionality. For basic concepts of virtualization, refer this excellent blog post by rayanfam.com. In this post, we’ll understand how to detect and enable AMD’s virtualization operation. Let’s go!
Checking if CPU supports virtualization
Let’s start by checking the CPU model - we need to make sure we know the underlying CPU as there are differences between the virtualization technologies used by different processors.
// Get the vendor name in string format
cpu = cpu_data(0);
vendor_name = cpu.x86_vendor_id;
// Get the model number of the processor
pr_warn("%d", boot_cpu_data.x86_model);
Now that we have the vendor name, we can continue if the CPU is AMD and abort otherwise. Let’s also check if SVM is supported in the CPU.
pr_warn("SVM supported? : %s\n", boot_cpu_has(X86_FEATURE_SVM) == 1?"Yes":"No");
In the above line, boot_cpu_has() is a macro which in our case checks if the CPU has the SVM flag set.
We know now if SVM is supported, so the next step is to check if the SVME bit is set. For this, we can use MSRs (Model specific Registers). In our case, we can use AMD’s Extended Feature enable register (EFER) in which the 12th bit (starting from 0) indicates if SVM is enabled. The Linux kernel does provide APIs for this, although we can also get this information from running assembly instructions. Let’s read this register and see if the SVME bit is enabled.
rdmsrl(MSR_EFER, efer);
if (efer & EFER_SVME)
pr_warn("EFER_SVME is already set\n");
else {
wrmsrl(MSR_EFER, efer | EFER_SVME);
pr_debug("SVM bit is enabled\n", efer);
}
In the above code, we first check if the SVME bit is enabled and if it is not enabled, we set it so that we put the CPU in SVM operation and can use other SVM operations.
In the next post, we’ll learn how to setup our own VM. Stay tuned.