Re: [PATCH 2/3] mce: acpi/apei: trace: Add trace event for ghes memory error

["Naveen N. Rao" <naveen.n.rao@xxxxxxxxxxxxxxxxxx>]

On 08/09/2013 12:47 AM, Borislav Petkov wrote:
> On Thu, Aug 08, 2013 at 11:57:50PM +0530, Naveen N. Rao wrote:
> > +TRACE_EVENT(ghes_platform_memory_event,
> > +	TP_PROTO(const struct acpi_hest_generic_status *estatus,
> > +		 const struct acpi_hest_generic_data *gdata,
> > +		 const struct cper_sec_mem_err *mem),
> > +
> > +	TP_ARGS(estatus, gdata, mem),
> > +
> > +	TP_STRUCT__entry(
> > +		__field(	u32,	estatus_block_status		)
> > +		__field(	u32,	estatus_raw_data_offset		)
> > +		__field(	u32,	estatus_raw_data_length		)
> > +		__field(	u32,	estatus_data_length		)
> > +		__field(	u32,	estatus_error_severity		)
> > +		__array(	u8,	gdata_section_type,	16	)
> > +		__field(	u32,	gdata_error_severity		)
> > +		__field(	u16,	gdata_revision			)
> > +		__field(	u8,	gdata_validation_bits		)
> > +		__field(	u8,	gdata_flags			)
> > +		__field(	u32,	gdata_error_data_length		)
> > +		__array(	u8,	gdata_fru_id,		16	)
> > +		__array(	u8,	gdata_fru_text,		20	)
> > +		__field(	u64,	mem_validation_bits		)
> > +		__field(	u64,	mem_error_status		)
> > +		__field(	u64,	mem_physical_addr		)
> > +		__field(	u64,	mem_physical_addr_mask		)
> > +		__field(	u16,	mem_node			)
> > +		__field(	u16,	mem_card			)
> > +		__field(	u16,	mem_module			)
> > +		__field(	u16,	mem_bank			)
> > +		__field(	u16,	mem_device			)
> > +		__field(	u16,	mem_row				)
> > +		__field(	u16,	mem_column			)
> > +		__field(	u16,	mem_bit_pos			)
> > +		__field(	u64,	mem_requestor_id		)
> > +		__field(	u64,	mem_responder_id		)
> > +		__field(	u64,	mem_target_id			)
> > +		__field(	u8,	mem_error_type			)
> > +	),
>
> Without looking at the rest, a trace record from this tracepoint is
> going to be 160 bytes IINM, which looks kinda fat to me. And during an
> error storm we're probably not going to be able to log them all, maybe?
> Yes, no, maybe I'm off base...
>
> In any case, are we sure we want all those fields above? Can we make
> them smaller, drop some of them from the tracepoint, etc, etc? Can we
> compute some of them in userspace with information we already have?

Good idea - I hadn't thought from that perspective. I think we can drop 
a few fields there, especially the length/offset fields and perhaps the 
section_type since we know this is a memory error. Will get back with a 
new revision.

Thanks,
Naveen

--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html