linux-kernel - Re: [PATCH v2] KVM: x86/pt: Ignore all unknown Intel PT capabilities

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <80b40829-0d25-eb84-7bd7-f21685daeb20@gmail.com>
Date:   Tue, 11 Jan 2022 12:20:29 +0800
From:   Like Xu <like.xu.linux@...il.com>
To:     Sean Christopherson <seanjc@...gle.com>
Cc:     Paolo Bonzini <pbonzini@...hat.com>,
        Jim Mattson <jmattson@...gle.com>,
        Wanpeng Li <wanpengli@...cent.com>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Joerg Roedel <joro@...tes.org>, x86@...nel.org,
        kvm@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] KVM: x86/pt: Ignore all unknown Intel PT capabilities

On 11/1/2022 8:57 am, Sean Christopherson wrote:
> On Mon, Jan 10, 2022, Like Xu wrote:
>> From: Like Xu <likexu@...cent.com>
>>
>> Some of the new Intel PT capabilities (e.g. SDM Vol3, 32.2.4 Event
>> Tracing, it exposes details about the asynchronous events, when they are
>> generated, and when their corresponding software event handler completes
>> execution) cannot be safely and fully emulated by the KVM, especially
>> emulating the simultaneous writing of guest PT packets generated by
>> the KVM to the guest PT buffer.
>>
>> For KVM, it's better to advertise currently supported features based on
>> the "static struct pt_cap_desc" implemented in the host PT driver and
>> ignore _all_ unknown features before they have been investigated one by
>> one and supported in a safe manner, leaving the rest as system-wide-only
>> tracing capabilities.
>>
>> Suggested-by: Paolo Bonzini <pbonzini@...hat.com>
>> Signed-off-by: Like Xu <likexu@...cent.com>
>> ---
>> v1 -> v2 Changelog:
>> - Be safe and ignore _all_ unknown capabilities. (Paolo)
>>
>> Previous:
>> https://lore.kernel.org/kvm/20220106085533.84356-1-likexu@tencent.com/
>>
>>   arch/x86/kvm/cpuid.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
>> index 0b920e12bb6d..439b93359848 100644
>> --- a/arch/x86/kvm/cpuid.c
>> +++ b/arch/x86/kvm/cpuid.c
>> @@ -901,6 +901,8 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
>>   			break;
>>   		}
>>   
>> +		/* It's better to be safe and ignore _all_ unknown capabilities. */
> 
> No need to justify why unknown capabilities are hidden as that's very much (supposed
> to be) standard KVM behavior.
> 
>> +		entry->ebx &= GENMASK(5, 0);
> 
> Please add a #define somewhere so that this is self-documenting, e.g. see
> KVM_SUPPORTED_XCR0.

How about we define this macro in the <asm/intel_pt.h> so that the next PT 
capability
enabler can update the mask with minimal effort, considering that many pure kernel
developers don't care about KVM code ?

> 
> And why just EBX?  ECX appears to enumerate features too, and EDX is presumably
> reserved to enumerate yet more features when EBX/ECX run out of bits.

Yes, how about this version:

diff --git a/arch/x86/include/asm/intel_pt.h b/arch/x86/include/asm/intel_pt.h
index ebe8d2ea44fe..da94d0eeb9df 100644
--- a/arch/x86/include/asm/intel_pt.h
+++ b/arch/x86/include/asm/intel_pt.h
@@ -24,6 +24,12 @@ enum pt_capabilities {
  	PT_CAP_psb_periods,
  };

+#define GUEST_SUPPORTED_CPUID_14_EBX	\
+	(BIT(0) | BIT(1) | BIT(2) | BIT(3) | BIT(4) | BIT(5))
+
+#define GUEST_SUPPORTED_CPUID_14_ECX	\
+	(BIT(0) | BIT(1) | BIT(2) | BIT(3) | BIT(31))
+
  #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
  void cpu_emergency_stop_pt(void);
  extern u32 intel_pt_validate_hw_cap(enum pt_capabilities cap);
diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
index 0b920e12bb6d..be8c9170f98e 100644
--- a/arch/x86/kvm/cpuid.c
+++ b/arch/x86/kvm/cpuid.c
@@ -19,6 +19,7 @@
  #include <asm/user.h>
  #include <asm/fpu/xstate.h>
  #include <asm/sgx.h>
+#include <asm/intel_pt.h>
  #include "cpuid.h"
  #include "lapic.h"
  #include "mmu.h"
@@ -900,7 +901,10 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array 
*array, u32 function)
  			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
  			break;
  		}
-
+		entry->eax = min(entry->eax, 1u);
+		entry->ebx &= GUEST_SUPPORTED_CPUID_14_EBX;
+		entry->ecx &= GUEST_SUPPORTED_CPUID_14_ECX;
+		entry->edx = 0;
  		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
  			if (!do_host_cpuid(array, function, i))
  				goto out;

> 
> And is there any possibility of a malicious user/guest using features to cause
> problems in the host?  I.e. does KVM need to enforce that the guest can't enable
> any unsupported features?

If a user space is set up with features not supported by KVM, it owns the risk 
itself.

AFAI, the guest Intel PT introduces a great attack interface for the host and
we only use the guest supported PT features in a highly trusted environment.

I agree that more uncertainty and fixes can be triggered in the security motive,
not expecting too much from this patch. :D

> 
>>   		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
>>   			if (!do_host_cpuid(array, function, i))
>>   				goto out;
>> -- 
>> 2.33.1
>>