linux-kernel - Re: [PATCH 1/1] KVM: use KVM_HVA_ERR

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 for Android: free password hash cracker in your pocket

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

Message-ID: <7fcfc226-0263-0364-bed7-fc95e6c945cb@oracle.com>
Date: Thu, 29 Feb 2024 14:13:49 -0800
From: Dongli Zhang <dongli.zhang@...cle.com>
To: "Huang, Kai" <kai.huang@...el.com>, kvm@...r.kernel.org
Cc: pbonzini@...hat.com, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/1] KVM: use KVM_HVA_ERR_BAD to check bad hva



On 2/29/24 13:53, Huang, Kai wrote:
> 
> 
> On 1/03/2024 10:25 am, Dongli Zhang wrote:
>> Replace PAGE_OFFSET with KVM_HVA_ERR_BAD, to facilitate the cscope when
>> looking for where KVM_HVA_ERR_BAD is used.
>>
>> Every time I use cscope to query the functions that are impacted by the
>> return value (KVM_HVA_ERR_BAD) of __gfn_to_hva_many(), I may miss
>> kvm_is_error_hva().
> 
> I am not sure "to facilitate cscope" could be a justification to do some code
> change in the kernel.
> 
>>
>> Signed-off-by: Dongli Zhang <dongli.zhang@...cle.com>
>> ---
>>   include/linux/kvm_host.h | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
>> index 7e7fd25b09b3..4dc0300e7766 100644
>> --- a/include/linux/kvm_host.h
>> +++ b/include/linux/kvm_host.h
>> @@ -143,7 +143,7 @@ static inline bool is_noslot_pfn(kvm_pfn_t pfn)
>>     static inline bool kvm_is_error_hva(unsigned long addr)
>>   {
>> -    return addr >= PAGE_OFFSET;
>> +    return addr >= KVM_HVA_ERR_BAD;
>>   }
>>     #endif
> 
> 
> Also, IIUC the KVM_HVA_ERR_BAD _theoretically_ can be any random value that can
> make kvm_is_error_hva() return false, while kvm_is_error_hva() must catch all
> error HVAs.
> 
> E.g., if we ever change KVM_HVA_ERR_BAD to use any other value (although I don't
> see why this could ever happen), then using KVM_HVA_ERR_BAD in
> kvm_is_error_hva() would be broken.
> 
> In other words, it seems to me we should just use PAGE_OFFSET in
> kvm_is_error_hva().
> 


At least so far PAGE_OFFSET is the same value as KVM_HVA_ERR_BAD (except
mips/s390), as line 141. Therefore, this is "No functional change".

It indicates the userspace VMM can never have hva in the range of kernel space.

 139 #ifndef KVM_HVA_ERR_BAD
 140
 141 #define KVM_HVA_ERR_BAD         (PAGE_OFFSET)
 142 #define KVM_HVA_ERR_RO_BAD      (PAGE_OFFSET + PAGE_SIZE)
 143
 144 static inline bool kvm_is_error_hva(unsigned long addr)
 145 {
 146         return addr >= PAGE_OFFSET;
 147 }
 148
 149 #endif


Regarding to "facilitate cscope", this happened since long time ago when I read
about ept_violation/mmio path.

1. The __gfn_to_hva_many() may return KVM_HVA_ERR_BAD for mmio.
2. Then I used cscope to find the location of KVM_HVA_ERR_BAD.
3. The kvm_is_error_hva() is not in the results.
4. It took me a while to figure out that the 'KVM_HVA_ERR_BAD' is indirectly
used by kvm_is_error_hva().

This is just based on my own experience when reading mmio code path. Thank you
very much!

Dongli Zhang