Re: [PATCH 4/7] iommu/arm-smmu-v3: Split struct arm_smmu_ctx_desc_cfg.cdtab

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Jun 04, 2024 at 04:07:21PM +0000, Mostafa Saleh wrote:
> > @@ -1236,7 +1235,7 @@ struct arm_smmu_cd *arm_smmu_alloc_cd_ptr(struct arm_smmu_master *master,
> >  			if (arm_smmu_alloc_cd_leaf_table(smmu, l1_desc))
> >  				return NULL;
> >  
> > -			l1ptr = cd_table->cdtab + idx * CTXDESC_L1_DESC_DWORDS;
> 
> Similarly to patch 1, I believe we can get rid of CTXDESC_L1_DESC_DWORDS entirely.

I gave CD the same treatment, it ends up like:

struct arm_smmu_cd {
	__le64 data[8];
};

struct arm_smmu_cdtab_l2 {
	struct arm_smmu_cd cds[CTXDESC_L2_ENTRIES];
};

struct arm_smmu_cdtab_l1 {
	__le64 l2ptr;
};

struct arm_smmu_ctx_desc_cfg {
	union {
		struct {
			struct arm_smmu_cd *table;
			unsigned int num_ents;
		} linear;
		struct {
			struct arm_smmu_cdtab_l1 *l1tab;
			struct arm_smmu_cdtab_l2 **l2ptrs;
			unsigned int num_l1_ents;
		} l2;
	};
	dma_addr_t			cdtab_dma;
	u8				s1fmt;
	/* log2 of the maximum number of CDs supported by this table */
	u8				s1cdmax;
};

And we can remove a bunch of the #defines in favour of sizeof()

> > -	dmam_free_coherent(smmu->dev, l1size, cd_table->cdtab, cd_table->cdtab_dma);
> >  	cd_table->cdtab_dma = 0;
> > -	cd_table->cdtab = NULL;
> > +	cd_table->cdtab.linear = NULL;
> 
> nit: AFAIU, arm_smmu_cdtab_allocated() was added to check the cd table existence
> regardless the config.
> So, I think we should be consistent, as here it is set as linear, while it can be 2lvl.

I deleted this code, the tables are allocated once and stay allocated
forever until release when the entire cd_table is deleted. No reason
to 0 memory we are about to kfree.

Thanks,
Jason