Re: [PATCH v5 4/9] cxl/events: Remove passing a UUID to known event traces

[Jonathan Cameron]

On Tue, 9 Jan 2024 15:38:28 -0800
Dan Williams <dan.j.williams@xxxxxxxxx> wrote:

> Jonathan Cameron wrote:
> > On Wed, 20 Dec 2023 16:17:31 -0800
> > Ira Weiny <ira.weiny@xxxxxxxxx> wrote:
> >   
> > > The UUID data is redundant in the known event trace types.  The addition
> > > of static defines allows the trace macros to create the UUID data inside
> > > the trace thus removing unnecessary code.
> > > 
> > > Have well known trace events use static data to set the uuid field based
> > > on the event type.
> > > 
> > > Suggested-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
> > > Signed-off-by: Ira Weiny <ira.weiny@xxxxxxxxx>
> > >   
> >   
> > >  	TP_STRUCT__entry(
> > >  		CXL_EVT_TP_entry
> > > @@ -422,7 +424,8 @@ TRACE_EVENT(cxl_dram,
> > >  	),
> > >  
> > >  	TP_fast_assign(
> > > -		CXL_EVT_TP_fast_assign(cxlmd, log, uuid, rec->hdr);
> > > +		CXL_EVT_TP_fast_assign(cxlmd, log, rec->hdr);
> > > +		memcpy(&__entry->hdr_uuid, &CXL_EVENT_DRAM_UUID, sizeof(uuid_t));  
> > 
> > Hmm. Why not
> > 
> > 		__entry->hdr_uuid = CXL_EVENT_DRAM_UUID;
> > ?
> > 
> > Compiler should be able to squish the stuff in the define down to data as as the
> > UUID generation logic is pretty simple.
> > 
> > I've not emulated the cper records for these yet, so not tested that works beyond
> > compiling.  
> 
> We can follow on with this conversion later as I see other usage of uuid
> copying in trace events (bcache for instance). Although I probably would
> not replace it with straight assignment and instead use the uuid_copy()
> helper. Otherwise, why do {uuid,guid}_copy() helpers exist?

To copy unknown uuids and guids where the compiler can't optimize things
nearly as well because it can't see the values. 

Jonathan