Re: [PATCH 18/19] cxl/region: Define a driver interface for region creation

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

On Sun, 04 Jun 2023 16:33:18 -0700
Dan Williams <dan.j.williams@intel.com> wrote:

> Scenarios like recreating persistent memory regions from label data and
> establishing new regions for CXL attached accelerators with local memory
> need a kernel internal facility to establish new regions.

Could probably make the label data one a userspace problem, but I agree
that it 'might' be done entirely in kernel.

> 
> Introduce cxl_create_region() that takes an array of endpoint decoders
> with reserved capacity and a root decoder object to establish a new
> region.
> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

> ---
>  drivers/cxl/core/region.c |  107 +++++++++++++++++++++++++++++++++++++++++++++
>  drivers/cxl/cxlmem.h      |    3 +
>  2 files changed, 110 insertions(+)
> 
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index a41756249f8d..543c4499379e 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
> @@ -2878,6 +2878,104 @@ construct_region_begin(struct cxl_root_decoder *cxlrd,
>  	return cxlr;
>  }
>  
> +static struct cxl_region *
> +__construct_new_region(struct cxl_root_decoder *cxlrd,
> +		       struct cxl_endpoint_decoder **cxled, int ways)
> +{
> +	struct cxl_decoder *cxld = &cxlrd->cxlsd.cxld;
> +	struct cxl_region_params *p;
> +	resource_size_t size = 0;
> +	struct cxl_region *cxlr;
> +	int rc, i;
> +
> +	if (ways < 1)
> +		return ERR_PTR(-EINVAL);
> +
> +	cxlr = construct_region_begin(cxlrd, cxled[0]);
> +	if (IS_ERR(cxlr))
> +		return cxlr;
> +
> +	rc = set_interleave_ways(cxlr, ways);
> +	if (rc)
> +		goto out;
> +
> +	rc = set_interleave_granularity(cxlr, cxld->interleave_granularity);
> +	if (rc)
> +		goto out;
> +
> +	down_read(&cxl_dpa_rwsem);
> +	for (i = 0; i < ways; i++) {
> +		if (!cxled[i]->dpa_res)
> +			break;
> +		size += resource_size(cxled[i]->dpa_res);
> +	}
> +	up_read(&cxl_dpa_rwsem);
> +
> +	if (i < ways)
> +		goto out;
> +
> +	rc = alloc_hpa(cxlr, size);
> +	if (rc)
> +		goto out;
> +
> +	down_read(&cxl_dpa_rwsem);
> +	for (i = 0; i < ways; i++) {
> +		rc = cxl_region_attach(cxlr, cxled[i], i);
> +		if (rc)
> +			break;
> +	}
> +	up_read(&cxl_dpa_rwsem);
> +
> +	if (rc)
> +		goto out;
> +
> +	rc = cxl_region_decode_commit(cxlr);
> +	if (rc)
> +		goto out;
> +
> +	p = &cxlr->params;
> +	p->state = CXL_CONFIG_COMMIT;
> +out:
> +	construct_region_end();
> +	if (rc) {
> +		drop_region(cxlr);
> +		return ERR_PTR(rc);
> +	}
> +	return cxlr;
> +}
> +

>  /* Establish an empty region covering the given HPA range */
>  static struct cxl_region *construct_region(struct cxl_root_decoder *cxlrd,
>  					   struct cxl_endpoint_decoder *cxled)
> @@ -3085,6 +3183,15 @@ static int cxl_region_probe(struct device *dev)
>  					p->res->start, p->res->end, cxlr,
>  					is_system_ram) > 0)
>  			return 0;
> +
> +		/*
> +		 * HDM-D[B] (device-memory) regions have accelerator
> +		 * specific usage, skip device-dax registration.
> +		 */

As before - I'm not yet convinced that is always the case for HDM-DB
Particularly given you support interleaving which may never make sense
for accelerators.

> +		if (cxlr->type == CXL_DECODER_DEVMEM)
> +			return 0;
> +
> +		/* HDM-H routes to device-dax */
>  		return devm_cxl_add_dax_region(cxlr);