Re: [PATCH kernel v3 2/3] KVM: SEV: Enable data breakpoints in SEV-ES

From: Alexey Kardashevskiy <aik@amd.com>
To: Sean Christopherson <seanjc@google.com>
Cc: kvm@vger.kernel.org, x86@kernel.org,
	linux-kernel@vger.kernel.org, Yury Norov <yury.norov@gmail.com>,
	Venu Busireddy <venu.busireddy@oracle.com>,
	Tony Luck <tony.luck@intel.com>,
	Tom Lendacky <thomas.lendacky@amd.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Sandipan Das <sandipan.das@amd.com>,
	Pawan Gupta <pawan.kumar.gupta@linux.intel.com>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Michael Roth <michael.roth@amd.com>,
	Mario Limonciello <mario.limonciello@amd.com>,
	Kim Phillips <kim.phillips@amd.com>,
	Kees Cook <keescook@chromium.org>,
	Juergen Gross <jgross@suse.com>, Jakub Kicinski <kuba@kernel.org>,
	Ingo Molnar <mingo@redhat.com>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Daniel Sneddon <daniel.sneddon@linux.intel.com>,
	Brijesh Singh <brijesh.singh@amd.com>,
	Borislav Petkov <bp@alien8.de>,
	Arnaldo Carvalho de Melo <acme@redhat.com>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Alexander Shishkin <alexander.shishkin@linux.intel.com>,
	Adrian Hunter <adrian.hunter@intel.com>,
	"Peter Zijlstra (Intel)" <peterz@infradead.org>,
	"Jason A. Donenfeld" <Jason@zx2c4.com>,
	"H. Peter Anvin" <hpa@zytor.com>
Subject: Re: [PATCH kernel v3 2/3] KVM: SEV: Enable data breakpoints in SEV-ES
Date: Fri, 3 Feb 2023 14:37:28 +1100	[thread overview]
Message-ID: <3b3a9ebc-b02e-a365-7f68-3da9189d062a@amd.com> (raw)
In-Reply-To: <Y9nL8iqhiL5+ALa2@google.com>

On 01/02/2023 13:18, Sean Christopherson wrote:
> On Fri, Jan 20, 2023, Alexey Kardashevskiy wrote:
>> diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
>> index 4826e6cc611b..61f2cad1cbaf 100644
>> --- a/arch/x86/kvm/svm/svm.h
>> +++ b/arch/x86/kvm/svm/svm.h
>> @@ -389,6 +389,8 @@ static inline bool vmcb12_is_intercept(struct vmcb_ctrl_area_cached *control, u3
>>   	return test_bit(bit, (unsigned long *)&control->intercepts);
>>   }
>>   
>> +extern bool sev_es_is_debug_swap_enabled(void);
>> +
>>   static inline void set_dr_intercepts(struct vcpu_svm *svm)
>>   {
>>   	struct vmcb *vmcb = svm->vmcb01.ptr;
>> @@ -410,8 +412,10 @@ static inline void set_dr_intercepts(struct vcpu_svm *svm)
>>   		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR6_WRITE);
>>   	}
>>   
>> -	vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
>> -	vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
>> +	if (!sev_es_guest(svm->vcpu.kvm) || !sev_es_is_debug_swap_enabled()) {
> 
> Looking below, doesn't this do the wrong thing if set_dr_intercepts() is called
> before SVM_SEV_FEAT_DEBUG_SWAP is set?  I.e. when this is called before LAUNCH_UPDATE?
> Seems like this should check SVM_SEV_FEAT_DEBUG_SWAP in sev_features regardless
> of when SVM_SEV_FEAT_DEBUG_SWAP is set.
> 
> And if KVM checks sev_features, then I _think_ we can avoid having to expose
> sev_es_debug_swap_enabled to svm.{c,h} (though why on earth {set,clr}_dr_intercepts()
> is in svm.h is another question for the future).

883b0a91f41a ("KVM: SVM: Move Nested SVM Implementation to nested.c") 
did that. Makes sense for things like vmcb_set_intercept() but 
{set,clr}_dr_intercepts() are still only called from svm.c so I'll move 
them there (btw do I need a separate patch for that? usually yes)

> 
> Follow-up question: does KVM _have_ to wait until KVM_SEV_LAUNCH_UPDATE_VMSA to
> set the flag?

Nope. Will repost soon as a reply to this mail.

>> +		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
>> +		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
>> +	}
>>   
>>   	recalc_intercepts(svm);
>>   }
>> @@ -422,8 +426,12 @@ static inline void clr_dr_intercepts(struct vcpu_svm *svm)
>>   
>>   	vmcb->control.intercepts[INTERCEPT_DR] = 0;
>>   
>> -	/* DR7 access must remain intercepted for an SEV-ES guest */
>> -	if (sev_es_guest(svm->vcpu.kvm)) {
>> +	/*
>> +	 * DR7 access must remain intercepted for an SEV-ES guest unless DebugSwap
>> +	 * (depends on NO_NESTED_DATA_BP) is enabled as otherwise a VM writing to DR7
>> +	 * from the #DB handler may trigger infinite loop of #DB's.
>> +	 */
>> +	if (sev_es_guest(svm->vcpu.kvm) && !sev_es_is_debug_swap_enabled()) {
>>   		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
>>   		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
>>   	}
>>
>> @@ -52,11 +53,21 @@ module_param_named(sev, sev_enabled, bool, 0444);
>>   /* enable/disable SEV-ES support */
>>   static bool sev_es_enabled = true;
>>   module_param_named(sev_es, sev_es_enabled, bool, 0444);
>> +
>> +/* enable/disable SEV-ES DebugSwap support */
>> +static bool sev_es_debug_swap_enabled = true;
>> +module_param_named(debug_swap, sev_es_debug_swap_enabled, bool, 0644);
> 
> Module param needs 0444 permissions, i.e. shouldn't be writable after KVM is
> loaded.  Though I don't know that providing a module param is warranted in this
> case.

> KVM provides module params for SEV and SEV-ES because there are legitimate
> reasons to turn them off, but at a glance, I don't see why we'd want that for this
> feature.

/me confused. You suggested this in the first place for (I think) for 
the case if the feature is found to be broken later on so admins can 
disable it.

And I was using it to verify "x86/debug: Fix stack recursion caused by 
DR7 accesses" which is convenient but it is a minor thing.

>>   #else
>>   #define sev_enabled false
>>   #define sev_es_enabled false
>> +#define sev_es_debug_swap false
> 
> This needs to be sev_es_debug_swap_enabled, otherwise things fall apart with
> CONFIG_KVM_AMD_SEV=n.
> 
> arch/x86/kvm/svm/sev.c: In function ‘sev_es_is_debug_swap_enabled’:
> arch/x86/kvm/svm/sev.c:69:16: error: ‘sev_es_debug_swap_enabled’ undeclared (first use in this function); did you mean ‘sev_es_is_debug_swap_enabled’?
>     69 |         return sev_es_debug_swap_enabled;
>        |                ^~~~~~~~~~~~~~~~~~~~~~~~~
>        |                sev_es_is_debug_swap_enabled
> 
> 
>>   #endif /* CONFIG_KVM_AMD_SEV */
>>   
>> +bool sev_es_is_debug_swap_enabled(void)
>> +{
>> +	return sev_es_debug_swap_enabled;
>> +}
> 
> ...
> 
>> @@ -604,6 +615,9 @@ static int sev_es_sync_vmsa(struct vcpu_svm *svm)
>>   	save->xss  = svm->vcpu.arch.ia32_xss;
>>   	save->dr6  = svm->vcpu.arch.dr6;
>>   
>> +	if (sev_es_is_debug_swap_enabled())
>> +		save->sev_features |= SVM_SEV_FEAT_DEBUG_SWAP;
>> +
>>   	pr_debug("Virtual Machine Save Area (VMSA):\n");
>>   	print_hex_dump_debug("", DUMP_PREFIX_NONE, 16, 1, save, sizeof(*save), false);
>>   
>> @@ -2249,6 +2263,9 @@ void __init sev_hardware_setup(void)
>>   out:
>>   	sev_enabled = sev_supported;
>>   	sev_es_enabled = sev_es_supported;
>> +	if (sev_es_debug_swap_enabled)
>> +		sev_es_debug_swap_enabled = sev_es_enabled &&
>> +			cpu_feature_enabled(X86_FEATURE_NO_NESTED_DATA_BP);
> 
> Slight preference for:
> 
> 	if (!sev_es_enabled || !cpu_feature_enabled(X86_FEATURE_NO_NESTED_DATA_BP))
> 		sev_es_debug_swap_enabled = false;
> 
> KVM does short-circuit some checks on module param values, but usually only to
> avoid additional setup.
> 
>>   #endif
>>   }
>>   
>> @@ -3027,6 +3044,18 @@ void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa)
>>   
>>   	/* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
>>   	hostsa->xss = host_xss;
>> +
>> +	/* The DebugSwap SEV feature does Type B swaps of DR[0-3] */
>> +	if (sev_es_is_debug_swap_enabled()) {
>> +		hostsa->dr0 = native_get_debugreg(0);
>> +		hostsa->dr1 = native_get_debugreg(1);
>> +		hostsa->dr2 = native_get_debugreg(2);
>> +		hostsa->dr3 = native_get_debugreg(3);
>> +		hostsa->dr0_addr_mask = amd_get_dr_addr_mask(0);
>> +		hostsa->dr1_addr_mask = amd_get_dr_addr_mask(1);
>> +		hostsa->dr2_addr_mask = amd_get_dr_addr_mask(2);
>> +		hostsa->dr3_addr_mask = amd_get_dr_addr_mask(3);
>> +	}
>>   }
>>   
>>   void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
>> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
>> index 60c7c880266b..6c54a3c9d442 100644
>> --- a/arch/x86/kvm/svm/svm.c
>> +++ b/arch/x86/kvm/svm/svm.c
>> @@ -1190,7 +1190,8 @@ static void init_vmcb(struct kvm_vcpu *vcpu)
>>   	set_exception_intercept(svm, UD_VECTOR);
>>   	set_exception_intercept(svm, MC_VECTOR);
>>   	set_exception_intercept(svm, AC_VECTOR);
>> -	set_exception_intercept(svm, DB_VECTOR);
>> +	if (!sev_es_is_debug_swap_enabled())
>> +		set_exception_intercept(svm, DB_VECTOR);
> 
> This is wrong.  KVM needs to intercept #DBs when debugging non-SEV-ES VMs.

Sorry, not following. The #DB intercept for non-SEV-ES is enabled here. 
Thanks,

> This _could_ be tied to X86_FEATURE_NO_NESTED_DATA_BP, but the KVM would need to
> toggle the intercept depending on whether or not userspace wants to debug the
> guest.
> 
> Similar to the DR7 interception, can this check sev_features directly?

-- 
Alexey