Skip to contentSkip to conversation

changing format/size of data in TRACE_EVENT(extlog_mem_event)

DISCUSSION Luck, Tony 3 participants 3 messages Jun 24, 2015 – Jun 25, 2015

lore
Share & tools
lore

Luck, Tony

author
#0 lore raw

In <ras/ras_event.h> we define a trace event for memory errors. The last field is:

__field_struct(struct cper_mem_err_compact, data)

where the structure is defined in <linux/cper.h> as:

struct cper_mem_err_compact {
        __u64   validation_bits;
        __u16   node;
        __u16   card;
        __u16   module;
        __u16   bank;
        __u16   device;
        __u16   row;
        __u16   column;
        __u16   bit_pos;
        __u64   requestor_id;
        __u64   responder_id;
        __u64   target_id;
        __u16   rank;
        __u16   mem_array_handle;
        __u16   mem_dev_handle;
};

This structure was defined based on the useful bits in the UEFI 2.4 spec appendix N, section 2.5 "Memory Error Section".

But UEFI have released a new version of the spec ... 2.5

http://www.uefi.org/sites/default/files/resources/UEFI%202_5.pdf

and things have been updated to cope with ever increasing memory sizes thanks to Moore's law. The old structure got a couple of tweaks as a quick band-aid to handle current problems (__u16 isn't big enough for the "row" entry for some 64GB DIMMs, so they squeezed bits 16:17 into a reserved field). But looking to the future they added a whole new GUID record "Memory Error Section 2" that increases the width of the device, row, column, rank and bit_pos fields from u16 to u32 and adds a couple of completely new fields.

So the question is - how can we update the trace event to include these new wider fields with the minimum pain to applications that look at it? I don't know if there are any other consumers besides rasdaemon at the moment ... but we don't want ugly transitions where you have to guess which version of the application you need to run to work with a given kernel version.

-Tony

#1 lore raw
Quoted 6 lines · Tony

So the question is - how can we update the trace event to include these new wider fields with the minimum pain to applications that look at it? I don't know if there are any other consumers besides rasdaemon at the moment ... but we don't want ugly transitions where you have to guess which version of the application you need to run to work with a given kernel version.

Well, can we issue an UEFI version and a binary blob which gets unpacked in userspace depending on that version? Can tracepoints even do that?

Because the next time they change the spec, we're going to have to deal with that PITA all over again... And if we're going to change the tracepoint now, we better change it only once but in a way that can accomodate future changes...

Signature · 5 lines
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--
#2 lore raw
In reply to Luck, Tony
Quoted 6 lines · Tony

So the question is - how can we update the trace event to include these new wider fields with the minimum pain to applications that look at it? I don't know if there are any other consumers besides rasdaemon at the moment ... but we don't want ugly transitions where you have to guess which version of the application you need to run to work with a given kernel version.

It comes down to if the rasdaemon (and any other user) included the event_parse.c "library" (it's not a public library yet, and we really should make it one). Because if it did, it doesn't matter what the field is, the event descriptions will give the size, and as long as the name of a field exists, and it doesn't change type (that is, from integer to string), it should be fine.

-- Steve