Re: [GIT preview] for-6.3/cxl-ram-region

["Verma, Vishal L" <vishal.l.verma@intel.com>]

On Tue, 2023-01-31 at 14:03 -0500, Gregory Price wrote:
> > 
> > I've pushed a cxl-cli branch[1] that incorporates this flow, and allows
> > for: cxl create-region -t ram <other options>
> > 
> > Feel free to give it a spin!
> > 
> > [1]: https://github.com/pmem/ndctl/commits/vv/volatile-regions
> 
> Right now I believe this is failing due to the interleave and size not
> having default values
> 
> ./cxl create-region -m -t ram -d decoder0.0 -w 1 -g 4096 mem0
> cxl region: create_region: create_region: unable to determine region size
> cxl region: cmd_create_region: created 0 regions
> 
> 
> appears to be due to this code
> static int create_region(struct cxl_ctx *ctx, int *count,
>              struct parsed_params *p)
> {
> // ... snip ...
>     rc = create_region_validate_config(ctx, p);
>     if (rc)
>         return rc;
> 
>     if (p->size) {
>         size = p->size;
>         default_size = false;
>     } else if (p->ep_min_size) {
>         size = p->ep_min_size * p->ways;
> **    } else {
> **        log_err(&rl, "%s: unable to determine region size\n", __func__);
> **        return -ENXIO;
> **    }
> 
> So both size and ep_min_size are 0 here
> 
> echo region0 > /sys/bus/cxl/devices/decoder0.0/create_ram_region
> cat /sys/bus/cxl/devices/region0/interleave_ways
> 0
> cat /sys/bus/cxl/devices/region0/interleave_granularity
> 0
> cat /sys/bus/cxl/devices/region0/size
> 0

Ah - this revealed an actual bug in these commits - the size and
ep_min_size don't refer to the region's size, it is the capacity of the
component memdevs. Right after create_ram_region, the region size is
expected to be zero.

However the bug here was a pmem assumption I had missed. When
determining sizes, we only look at pmem capacity, which is wrong. It
happened to work in my testing because the memdevs I used had both pmem
and ram capacity. I'll update with a fix shortly. Thanks for trying it
out and reporting this!