Re: [PATCH v10 05/16] x86/virt/tdx: Add skeleton to enable TDX on demand

[Huang, Kai]

On Thu, 2023-03-23 at 06:49 -0700, Hansen, Dave wrote:
> On 3/15/23 04:10, Huang, Kai wrote:
> > I can do.  Just want to make sure do you want to retry TDX_SYS_BUSY, or retry
> > TDX_RND_NO_ENTROPY (if we want to ask TDX module guys to change to return this
> > value)?
> 
> I'll put it this way:
> 
> 	Linux is going to treat TDX_SYS_BUSY like a Linux bug and assume
> 	Linux is doing something wrong.  It'll mostly mean that
> 	users will see something nasty and may even cause Linux to give
> 	up on TDX.  In other words, the TDX module shouldn't use
> 	TDX_SYS_BUSY for things that aren't Linux's fault.
> 
> > Also, even we retry either TDX_SYS_BUSY or TDX_RND_NO_ENTROPY in common
> > seamcall() code, it doesn't handle the TDH.SYS.KEY.CONFIG, because sadly this
> > SEAMCALL returns a different error code:
> > 
> > TDX_KEY_GENERATION_FAILED       Failed to generate a random key. This is
> >                                 typically caused by an entropy error of the
> >                                 CPU's random number generator, and may
> >                                 be impacted by RDSEED, RDRAND or PCONFIG
> >                                 executing on other LPs. The operation should be
> >                                 retried.
> 
> Sounds like we should just replace TDX_KEY_GENERATION_FAILED with
> TDX_RND_NO_ENTROPY in cases where key generation fails because of a lack
> of entropy.

Thanks for feedback.

I'll do following, please let me know for any comments in case I have any
misunderstanding.

1) In TDH.SYS.INIT, ask TDX module team to return TDX_RND_NO_ENTROPY instead of
TDX_SYS_BUSY when running out of entropy. 

2) In TDH.SYS.KEY.CONFIG, ask TDX module to return TDX_RND_NO_ENTROPY instead of
TDX_KEY_GENERATION_FAILED when running out of entropy.  Whether
TDX_KEY_GENERATION_FAILED should be still kept is  up to TDX module team
(because it looks running concurrent PCONFIGs is also related).

3) Ask TDX module to always return TDX_RND_NO_ENTROPY in _ALL_ SEAMCALLs and
keep this behaviour for future TDX modules too.

4) In the common seamcall(), retry on TDX_RND_NO_ENTROPY.

In terms of how many times to retry, I will use a fixed value for now, similar
to the kernel code below:

#define RDRAND_RETRY_LOOPS      10                                             
                                                                                                                                                   
/* Unconditional execution of RDRAND and RDSEED */                             
                                                                                                                                                   
static inline bool __must_check rdrand_long(unsigned long *v)                  
{                                                                              
        bool ok;                                                               
        unsigned int retry = RDRAND_RETRY_LOOPS;                               
        do {                                                                   
                asm volatile("rdrand %[out]"                                   
                             CC_SET(c)                                         
                             : CC_OUT(c) (ok), [out] "=r" (*v));               
                if (ok)                                                        
                        return true;                                           
        } while (--retry);                                                     
        return false;                                                          
}