On Sun, Aug 28, 2022 at 05:19:33PM -0700, Sathyanarayanan Kuppuswamy wrote:
> Hi,
>
> On 8/28/22 1:14 PM, Dave Hansen wrote:
> > On 8/26/22 08:06, Kuppuswamy Sathyanarayanan wrote:
> >> +struct tdx_report_req {
> >> + __u8 subtype;
> >> + __u64 reportdata;
> >> + __u32 rpd_len;
> >> + __u64 tdreport;
> >> + __u32 tdr_len;
> >> +};
> >
> > Please do take a look at how the compiler ends up building that structure.
> >
> > If you sized things to "save space", the way the compiler treats that
> > structure may surprise you. You might also want to look at how a 32-bit
> > compile deals with it versus a 64-bit one.
>
> Since it is only used in user/kernel ABI, I did not consider the size
> issue.
That is _EXACTLY_ why you need to consider the size issues here.
> But I understand your point. The size of this struct in a 64-bit
> system is 40 bytes. So we did not gain anything with using different member
> sizes. In a 32-bit system, size due to padding is less compared to 64-bit.
That's not the issue here, please use a tool like pahole to see the
problems with this definition.
> I will re-arrange the struct as below. With this change, the size will come
> down to 32 bytes.
>
> struct tdx_report_req {
> __u64 reportdata;;
> __u64 tdreport;
> __u32 rpd_len
> __u32 tdr_len;
> __u8 subtype;
> };
That's better, but again, please use pahole.
thanks,
greg k-h
