Re: [PATCH v13 12/12] x86/kvm: Add guest support for detecting and enabling SEV Live Migration feature.

[Paolo Bonzini <pbonzini@redhat.com>]

On 21/04/21 16:44, Borislav Petkov wrote:
> On Thu, Apr 15, 2021 at 04:01:16PM +0000, Ashish Kalra wrote:
>> From: Ashish Kalra <ashish.kalra@amd.com>
>>
>> The guest support for detecting and enabling SEV Live migration
>> feature uses the following logic :
>>
>>   - kvm_init_plaform() invokes check_kvm_sev_migration() which
>>     checks if its booted under the EFI
>>
>>     - If not EFI,
>>
>>       i) check for the KVM_FEATURE_CPUID
> 
> Where do you do that?
> 
> $ git grep KVM_FEATURE_CPUID
> $
> 
> Do you mean
> 
> 	kvm_para_has_feature(KVM_FEATURE_SEV_LIVE_MIGRATION)
> 
> per chance?

Yep.  Or KVM_CPUID_FEATURES perhaps.

> 
>> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
>> index 78bb0fae3982..94ef16d263a7 100644
>> --- a/arch/x86/kernel/kvm.c
>> +++ b/arch/x86/kernel/kvm.c
>> @@ -26,6 +26,7 @@
>>   #include <linux/kprobes.h>
>>   #include <linux/nmi.h>
>>   #include <linux/swait.h>
>> +#include <linux/efi.h>
>>   #include <asm/timer.h>
>>   #include <asm/cpu.h>
>>   #include <asm/traps.h>
>> @@ -429,6 +430,59 @@ static inline void __set_percpu_decrypted(void *ptr, unsigned long size)
>>   	early_set_memory_decrypted((unsigned long) ptr, size);
>>   }
>>   
>> +static int __init setup_kvm_sev_migration(void)
> 
> kvm_init_sev_migration() or so.
> 
> ...
> 
>> @@ -48,6 +50,8 @@ EXPORT_SYMBOL_GPL(sev_enable_key);
>>   
>>   bool sev_enabled __section(".data");
>>   
>> +bool sev_live_migration_enabled __section(".data");
> 
> Pls add a function called something like:
> 
> bool sev_feature_enabled(enum sev_feature)
> 
> and gets SEV_FEATURE_LIVE_MIGRATION and then use it instead of adding
> yet another boolean which contains whether some aspect of SEV has been
> enabled or not.
> 
> Then add a
> 
> static enum sev_feature sev_features;
> 
> in mem_encrypt.c and that function above will query that sev_features
> enum for set flags.

Even better: let's stop callings things SEV/SEV_ES.  Long term we want 
anyway to use things like mem_encrypt_enabled (SEV), 
guest_instruction_trap_enabled (SEV/ES), etc.

For this one we don't need a bool at all, we can simply check whether 
the pvop points to paravirt_nop.  Also keep everything but the BSS 
handling in arch/x86/kernel/kvm.c.  Only the BSS handling should be in 
arch/x86/mm/mem_encrypt.c.  This way all KVM paravirt hypercalls and 
MSRs are in kvm.c.

That is:

void kvm_init_platform(void)
{
	if (sev_active() &&
	    kvm_para_has_feature(KVM_FEATURE_SEV_LIVE_MIGRATION)) {
		pv_ops.mmu.notify_page_enc_status_changed =
			kvm_sev_hc_page_enc_status;
		/* this takes care of bss_decrypted */
		early_set_page_enc_status();
		if (!efi_enabled(EFI_BOOT))
			wrmsrl(MSR_KVM_SEV_LIVE_MIGRATION,
			       KVM_SEV_LIVE_MIGRATION_ENABLED);
	}
	/* existing kvm_init_platform code goes here */
}

// the pvop is changed to take the pfn, so that the vaddr loop
// is not KVM specific
static inline void notify_page_enc_status_changed(unsigned long pfn,
				int npages, bool enc)
{
	PVOP_VCALL3(mmu.page_encryption_changed, pfn, npages, enc);
}

static void notify_addr_enc_status_changed(unsigned long addr,
					   int numpages, bool enc)
{
#ifdef CONFIG_PARAVIRT
	if (pv_ops.mmu.notify_page_enc_status_changed == paravirt_nop)
		return;

	/* the body of set_memory_enc_dec_hypercall goes here */
	for (; vaddr < vaddr_end; vaddr = vaddr_next) {
		...
		notify_page_enc_status_changed(pfn, psize >> PAGE_SHIFT,
					       enc);
		vaddr_next = (vaddr & pmask) + psize;
	}
#endif
}

static int __set_memory_enc_dec(unsigned long addr,
				int numpages, bool enc)
{
	...
  	cpa_flush(&cpa, 0);
	notify_addr_enc_status_changed(addr, numpages, enc);
  	return ret;
}


> +static int __init setup_kvm_sev_migration(void)

Please rename this to include efi in the function name.

> 
> +		 */
> +		if (!efi_enabled(EFI_BOOT))
> +			wrmsrl(MSR_KVM_SEV_LIVE_MIGRATION,
> +			       KVM_SEV_LIVE_MIGRATION_ENABLED);
> +		} else {
> +			pr_info("KVM enable live migration feature unsupported\n");
> +		}
> +}

I think this pr_info is incorrect, because it can still be enabled in 
the late_initcall.  Just remove it as in the sketch above.

Paolo