Re: [Xen-devel] [PATCH 1/4] x86/svm: Clean up construct_vmcb()

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 04/12/2019 10:06, Jan Beulich wrote:
> On 04.12.2019 10:43, Andrew Cooper wrote:
>> The vmcb is zeroed on allocate - drop all explicit writes of 0.  Move
>> hvm_update_guest_efer() to co-locate it with the other control register
>> updates.
>>
>> Move the BUILD_BUG_ON() into build_assertions(), and add some offset checks
>> for fields after the large blocks of reserved fields (as these are the most
>> likely to trigger from a mis-edit).  Take the opportunity to fold 6 adjacent
>> res* fields into one.
>>
>> Finally, drop all trailing whitespace in the file.
>>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
> Reviewed-by: Jan Beulich <jbeulich@suse.com>
> albeit with two (optional) suggestions:
>
>> @@ -297,14 +258,26 @@ void __init setup_vmcb_dump(void)
>>  
>>  static void __init __maybe_unused build_assertions(void)
>>  {
>> -    struct segment_register sreg;
>> +    struct vmcb_struct vmcb;
>> +
>> +    /* Build-time check of the VMCB layout. */
>> +    BUILD_BUG_ON(sizeof(vmcb) != PAGE_SIZE);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, _pause_filter_thresh) != 0x03c);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, _vintr)               != 0x060);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, eventinj)             != 0x0a8);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, es)                   != 0x400);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, _cpl)                 != 0x4cb);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, _cr4)                 != 0x548);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, rsp)                  != 0x5d8);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, rax)                  != 0x5f8);
>> +    BUILD_BUG_ON(offsetof(struct vmcb_struct, _g_pat)               != 0x668);
>>  
>>      /* Check struct segment_register against the VMCB segment layout. */
>> -    BUILD_BUG_ON(sizeof(sreg)       != 16);
>> -    BUILD_BUG_ON(sizeof(sreg.sel)   != 2);
>> -    BUILD_BUG_ON(sizeof(sreg.attr)  != 2);
>> -    BUILD_BUG_ON(sizeof(sreg.limit) != 4);
>> -    BUILD_BUG_ON(sizeof(sreg.base)  != 8);
>> +    BUILD_BUG_ON(sizeof(vmcb.es)       != 16);
>> +    BUILD_BUG_ON(sizeof(vmcb.es.sel)   != 2);
>> +    BUILD_BUG_ON(sizeof(vmcb.es.attr)  != 2);
>> +    BUILD_BUG_ON(sizeof(vmcb.es.limit) != 4);
>> +    BUILD_BUG_ON(sizeof(vmcb.es.base)  != 8);
>>      BUILD_BUG_ON(offsetof(struct segment_register, sel)   != 0);
>>      BUILD_BUG_ON(offsetof(struct segment_register, attr)  != 2);
>>      BUILD_BUG_ON(offsetof(struct segment_register, limit) != 4);
> For the ones only supplying context here, how about using the
> shorter offsetof(typeof(vmcb.es), ...), also tying things better
> to the prior sizeof() checks? The same, albeit to a lesser degree,
> might then go for the earlier block, which could use the shorter
> typeof(vmcb).

Fixed.

>
>> --- a/xen/include/asm-x86/hvm/svm/vmcb.h
>> +++ b/xen/include/asm-x86/hvm/svm/vmcb.h
>> @@ -406,12 +406,7 @@ struct vmcb_struct {
>>      u32 _exception_intercepts;  /* offset 0x08 - cleanbit 0 */
>>      u32 _general1_intercepts;   /* offset 0x0C - cleanbit 0 */
>>      u32 _general2_intercepts;   /* offset 0x10 - cleanbit 0 */
>> -    u32 res01;                  /* offset 0x14 */
>> -    u64 res02;                  /* offset 0x18 */
>> -    u64 res03;                  /* offset 0x20 */
>> -    u64 res04;                  /* offset 0x28 */
>> -    u64 res05;                  /* offset 0x30 */
>> -    u32 res06;                  /* offset 0x38 */
>> +    u32 res01[10];
> Was it intentional for the comment to be lost altogether?

Yes.  The offset is trivial (0x10 + sizeof(u32)) and of no interest.

Omitting it increases readability by helping to highlight where the
reserved blocks are.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel