Re: [PATCH v3] cxl/events: Use a common struct for DRAM and General Media events

[Alison Schofield <alison.schofield@intel.com>]

On Mon, May 20, 2024 at 07:55:17PM +0200, Fabio M. De Francesco wrote:
> On Saturday, May 18, 2024 1:26:21 PM GMT+2 Fabio M. De Francesco wrote:
> > cxl_event_common was an unfortunate naming choice and caused confusion with
> > the existing Common Event Record. Furthermore, its fields didn't map all
> > the common information between DRAM and General Media Events.
> > 
> > Remove cxl_event_common and introduce cxl_event_media_hdr to record common
> > information between DRAM and General Media events.
> > 
> > cxl_event_media_hdr, which is embedded in both cxl_event_gen_media and
> > cxl_event_dram, leverages the commonalities between the two events to
> > simplify their respective handling.
> > 
> > Suggested-by: Dan Williams <dan.j.williams@intel.com>
> > Signed-off-by: Fabio M. De Francesco <fabio.m.de.francesco@linux.intel.com>
> > ---
> > 
> > [...]
> > -
> >  union cxl_event {
> >  	struct cxl_event_generic generic;
> >  	struct cxl_event_gen_media gen_media;
> >  	struct cxl_event_dram dram;
> >  	struct cxl_event_mem_module mem_module;
> > -	struct cxl_event_common common;
> > +	struct cxl_event_media_hdr media_hdr;
> >  } __packed;
> 
> Today I was thinking about a comment from Ira. He didn't like the addition of 
> an event that is not in the specs.[0] (Notice that the other issues have been 
> already addressed). 
> 
> I dislike the addition of an artificial event for the very same reasons Ira 
> expressed.

Agree it should not appear like an artifical event. It should appear
as a special member of the union. A union's purpose is to allow access
like this. Perhaps it's a naming thing.

How about s/cxl_event_media_hdr/cxl_media_hdr and add a comment:

-	struct cxl_event_common common;
+	struct cxl_media_hdr media_hdr; /* dram & gen_media event header */

> 
> This additional event could be easily removed by something simple like the 
> following diff.

Some typos below and also you'll need to look at the larger chunk of
code and reconsider the if/else flow. Perhaps revert to an earlier
version of the event patchset.

-- Alison

> 
> Fabio
> 
> [0] https://lore.kernel.org/linux-cxl/66467b8b47170_8c79294b3@iweiny-mobl.notmuch/
> 
> diff --git a/drivers/cxl/core/mbox.c b/drivers/cxl/core/mbox.c
> index a08f050cc1ca..05de8836adea 100644
> --- a/drivers/cxl/core/mbox.c
> +++ b/drivers/cxl/core/mbox.c
> @@ -875,7 +875,13 @@ void cxl_event_trace_record(const struct cxl_memdev 
> *cxlmd,
>                 guard(rwsem_read)(&cxl_region_rwsem);
>                 guard(rwsem_read)(&cxl_dpa_rwsem);
>  
> -               dpa = le64_to_cpu(evt->media_hdr.phys_addr) & CXL_DPA_MASK;
> +               if (event_type == CXL_CPER_EVENT_GEN_MEDIA)
> +                       dpa = le64_to_cpu(evt->gen_media.media_hdr.phys_addr)
> +                             & CXL_DPA_MASK;
> +               else if (event_type == CXL_CPER_EVENT_GEN_MEDIA)
> +                       dpa = le64_to_cpu(evt->dram.media_hdr.phys_addr)
> +                             & CXL_DPA_MASK;
> +
>                 cxlr = cxl_dpa_to_region(cxlmd, dpa);
>                 if (cxlr)
>                         hpa = cxl_trace_hpa(cxlr, cxlmd, dpa);
> diff --git a/include/linux/cxl-event.h b/include/linux/cxl-event.h
> index 6562663a036d..f0a5be131e6a 100644
> --- a/include/linux/cxl-event.h
> +++ b/include/linux/cxl-event.h
> @@ -97,7 +97,6 @@ union cxl_event {
>         struct cxl_event_gen_media gen_media;
>         struct cxl_event_dram dram;
>         struct cxl_event_mem_module mem_module;
> -       struct cxl_event_media_hdr media_hdr;
>  } __packed;
>  
>  /*
> 
> 
>