Re: [PATCH v8 11/16] x86/virt/tdx: Designate reserved areas for all TDMRs

[Huang, Kai]

On Wed, 2023-01-11 at 08:16 -0800, Dave Hansen wrote:
> On 1/11/23 02:57, Huang, Kai wrote:
> > On Tue, 2023-01-10 at 07:19 -0800, Dave Hansen wrote:
> > > On 1/10/23 03:01, Huang, Kai wrote:
> > > > On Mon, 2023-01-09 at 17:22 -0800, Dave Hansen wrote:
> > > > > On 1/9/23 17:19, Huang, Kai wrote:
> > > > > > > It's probably also worth noting *somewhere* that there's a balance to be
> > > > > > > had between TDMRs and reserved areas.  A system that is running out of
> > > > > > > reserved areas in a TDMR could split a TDMR to get more reserved areas.
> > > > > > > A system that has run out of TDMRs could relatively easily coalesce two
> > > > > > > adjacent TDMRs (before the PAMTs are allocated) and use a reserved area
> > > > > > > if there was a gap between them.
> > > > > > We can add above to the changelog of this patch, or the patch 09 ("x86/virt/tdx:
> > > > > > Fill out TDMRs to cover all TDX memory regions").  The latter perhaps is better
> > > > > > since that patch is the first place where the balance of TDMRs and reserved
> > > > > > areas is related.
> > > > > > 
> > > > > > What is your suggestion?
> > > > > Just put it close to the code that actually hits the problem so the
> > > > > potential solution is close at hand to whoever hits the problem.
> > > > > 
> > > > Sorry to double check: the code which hits the problem is the 'if (idx >=
> > > > max_reserved_per_tdmr)' check in tdmr_add_rsvd_area(), so I think I can add
> > > > right before this check?
> > > 
> > > Please just hack together how you think it should look and either reply
> > > with an updated patch, or paste the relevant code snippet in your reply.
> > >  That'll keep me from having to go chase this code back down.
> > > 
> > 
> > Thanks for the tip.  How about below?
> > 
> > static int tdmr_add_rsvd_area(struct tdmr_info *tdmr, int *p_idx, u64 addr,
> >                               u64 size, u16 max_reserved_per_tdmr)
> > {
> >         struct tdmr_reserved_area *rsvd_areas = tdmr->reserved_areas;
> >         int idx = *p_idx;
> > 
> >         /* Reserved area must be 4K aligned in offset and size */
> >         if (WARN_ON(addr & ~PAGE_MASK || size & ~PAGE_MASK))
> >                 return -EINVAL;
> > 
> >         /*
> >          * The TDX module supports only limited number of TDMRs and
> >          * limited number of reserved areas for each TDMR.  There's a
> >          * balance to be had between TDMRs a2nd reserved areas.  A system
> >          * that is running out of reserved areas in a TDMR could split a
> >          * TDMR to get more reserved areas.  A system that has run out
> >          * of TDMRs could relatively easily coalesce two adjacent TDMRs
> >          * (before the PAMTs are allocated) and use a reserved area if
> >          * there was a gap between them.
> >          */
> >         if (idx >= max_reserved_per_tdmr) {
> >                 pr_warn("too many reserved areas for TDMR [0x%llx, 0x%llx)\n",
> >                                 tdmr->base, tdmr_end(tdmr));
> >                 return -ENOSPC;
> >         }
> 
> This isn't really converging on a solution.  At this point, I just see
> my verbatim text being copied and pasted into these functions without
> really anything additional.
> 
> This comment, for instance, just blathers about what could be done but
> doesn't actually explain what it is doing here.
> 
> But, again, this isn't converging.  It's just thrashing and not getting
> any better.  I guess I'll just fix it up best I can when I apply it.

Appreciate your help!