Re: [PATCH v5 2/9] iommu/arm-smmu-v3: Replace s1_cfg with cdtab_cfg

[Jason Gunthorpe <jgg@nvidia.com>]

On Fri, Aug 11, 2023 at 01:15:14AM +0800, Michael Shavit wrote:
> > > > What dead code? Is the deal here that we keep the field, but still
> > > > infer the value to write from (cd_table->l1_desc == null) in
> > > > arm_smmu_write_strtab_ent??
> > >
> > > Keep the field and write it directly when populating the ste (i.e. don't
> > > infer anything), but the field moves into 'struct arm_smmu_ctx_desc_cfg'.
> >
> > Yes - the 'dead code' is that we introduce storage for a field that is
> > always a known constant (STRTAB_STE_0_S1FMT_64K_L2).
> 
> I'm not sure we're on the same page here. s1fmt could contain either
> `STRTAB_STE_0_S1FMT_64K_L2` or `STRTAB_STE_0_S1FMT_LINEAR`, and this
> value will be directly copied in arm_smmu_write_strtab_ent.

Ah, I did not check this closely, Will said:

 > But the computation isn't happening -- the STRTAB_STE_0_S1FMT_64K_L2
 > constant is hardcoded here.

So the nuanced answer is that computation is happening because today
the format of the CD table (linear vs 64k) is encoded in l1_desc:

 +				  cd_table->l1_desc ?
 +					  STRTAB_STE_0_S1FMT_64K_L2 :
 +					  STRTAB_STE_0_S1FMT_LINEAR);

So I would suggest that along with adding s1fmt to
arm_smmu_ctx_desc_cfg you also go and normalize the usage:

@@ -1030,7 +1030,7 @@ static __le64 *arm_smmu_get_cd_ptr(struct arm_smmu_master *master, u32 ssid)
        struct arm_smmu_device *smmu = master->smmu;
        struct arm_smmu_ctx_desc_cfg *cd_table = &master->cd_table;
 
-       if (!cd_table->l1_desc)
+       if (cd_table->s1fmt == STRTAB_STE_0_S1FMT_LINEAR)
                return cd_table->cdtab + ssid * CTXDESC_CD_DWORDS;

Jason