On Wed, Mar 06, 2024 at 03:03:40PM -0800, Dan Williams wrote:
> alison.schofield@ wrote:
> > From: Alison Schofield <alison.schofi...@intel.com>
> >
> > Changes since v9:
> > - Replace the multi-use 'name' var, with multiple descriptive
> > flavors: memdev_name, region_name, decoder_name (DaveJ)
> > - Use a static string table for poison source lookup (DaveJ)
> > - Rebased on latest pending
> > Link to v9:
> > https://lore.kernel.org/r/cover.1709253898.git.alison.schofi...@intel.com/
> >
> >
> > Add the option to add a memory devices poison list to the cxl-list
> > json output. Offer the option by memdev and by region. Sample usage:
> >
> > # cxl list -m mem1 --media-errors
> > [
> > {
> > "memdev":"mem1",
> > "pmem_size":1073741824,
> > "ram_size":1073741824,
> > "serial":1,
> > "numa_node":1,
> > "host":"cxl_mem.1",
> > "media_errors":[
> > {
> > "dpa":0,
> > "length":64,
> > "source":"Internal"
> > },
> > {
> > "decoder":"decoder10.0",
> > "hpa":1035355557888,
> > "dpa":1073741824,
> > "length":64,
> > "source":"External"
> > },
> > {
> > "decoder":"decoder10.0",
> > "hpa":1035355566080,
> > "dpa":1073745920,
> > "length":64,
> > "source":"Injected"
> > }
Dan,
In cleaning up the man pages, I need to follow up on this offset, length
notation.
A default by memdev list now looks like this-
{
"offset" :
"length" :
"source" :
}
Which means dropping the 'decoder' even if it can be discovered from
the trace event. Recall previously a region was listed if present,
then we changed that to a decoder is listed if present. Now, with
no 'hpa' listing I've dropped the decoder too. That leaves no hint
in the by memdev listing that this poison is in a region.
Decoders will only be included in the by region list:
{
"decoder":
"offset" :
"length" :
"source" :
}
OK ?
> > }
> > ]
> >
> > # cxl list -r region5 --media-errors
> > [
> > {
> > "region":"region5",
> > "resource":1035355553792,
> > "size":2147483648,
> > "type":"pmem",
> > "interleave_ways":2,
> > "interleave_granularity":4096,
> > "decode_state":"commit",
> > "media_errors":[
> > {
> > "decoder":"decoder10.0",
> > "hpa":1035355557888,
> > "dpa":1073741824,
> > "length":64,
>
> I notice that the ndctl --media-errors records are:
>
> { offset, length }
>
> ...it is not clear to me that "dpa" and "hpa" have much meaning to
> userspace by default. Physical address information is privileged, so if
> these records were { offset, length } tuples there is the possibility
> that they can be provided to non-root.
>
> "Offset" is region relative "hpa" when listing region media errors, and
> "offset" is memdev relative "dpa" while listing memdev relative media
> errors.
