Re: [PATCH v4 1/2] acpi/ghes: Process CXL Component Events

[Ira Weiny]

Tony Luck wrote:
> On Fri, Apr 26, 2024 at 08:34:00PM -0700, Ira Weiny wrote:
> > @@ -707,6 +805,18 @@ static bool ghes_do_proc(struct ghes *ghes,
> >  		}
> >  		else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
> >  			queued = ghes_handle_arm_hw_error(gdata, sev, sync);
> > +		} else if (guid_equal(sec_type, &CPER_SEC_CXL_GEN_MEDIA_GUID)) {
> > +			struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
> > +
> > +			cxl_cper_post_event(CXL_CPER_EVENT_GEN_MEDIA, rec);
> > +		} else if (guid_equal(sec_type, &CPER_SEC_CXL_DRAM_GUID)) {
> > +			struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
> > +
> > +			cxl_cper_post_event(CXL_CPER_EVENT_DRAM, rec);
> > +		} else if (guid_equal(sec_type, &CPER_SEC_CXL_MEM_MODULE_GUID)) {
> > +			struct cxl_cper_event_rec *rec = acpi_hest_get_payload(gdata);
> > +
> > +			cxl_cper_post_event(CXL_CPER_EVENT_MEM_MODULE, rec);
> >  		} else {
> >  			void *err = acpi_hest_get_payload(gdata);
> 
> You pass "rec" to cxl_cper_post_event() in all these cases for later
> processing in context where you can sleep to get locks. But that's
> just a pointer somewhere into the "gdata" error record received from
> BIOS.
> 
> What's the lifetime of that record?
>

The lifetime is contained to the cxl_cper_post_envent() which does not
block.  The kfifo_put() copies the data to a cxl specific record to be
used by the CXL thread for processing.

>
> Can it be re-used/overwritten
> before that other kernel thread gets around to looking at it?

Yes because the CXL kernel thread has its own copy in the kfifo.

Ira