Re: [Xen-devel] [PATCH v5 3/4] xen: refactor debugtrace data

[Juergen Gross <jgross@suse.com>]

On 05.09.19 14:46, Juergen Gross wrote:
> On 05.09.19 14:37, Jan Beulich wrote:
>> On 05.09.2019 14:27, Juergen Gross wrote:
>>> On 05.09.19 14:22, Jan Beulich wrote:
>>>> On 05.09.2019 14:12, Juergen Gross wrote:
>>>>> On 05.09.19 14:01, Jan Beulich wrote:
>>>>>> On 05.09.2019 13:39, Juergen Gross wrote:
>>>>>>> As a preparation for per-cpu buffers do a little refactoring of the
>>>>>>> debugtrace data: put the needed buffer admin data into the buffer as
>>>>>>> it will be needed for each buffer. In order not to limit buffer size
>>>>>>> switch the related fields from unsigned int to unsigned long, as on
>>>>>>> huge machines with RAM in the TB range it might be interesting to
>>>>>>> support buffers >4GB.
>>>>>>
>>>>>> Just as a further remark in this regard:
>>>>>>
>>>>>>>     void debugtrace_printk(const char *fmt, ...)
>>>>>>>     {
>>>>>>>         static char buf[DEBUG_TRACE_ENTRY_SIZE];
>>>>>>> -    static unsigned int count, last_count, last_prd;
>>>>>>> +    static unsigned int count, last_count;
>>>>>>
>>>>>> How long do we think will it take until their wrapping will become
>>>>>> an issue with such huge buffers?
>>>>>
>>>>> Count wrapping will not result in any misbehavior of tracing. With
>>>>> per-cpu buffers it might result in ambiguity regarding sorting the
>>>>> entries, but I guess chances are rather low this being a real issue.
>>>>>
>>>>> BTW: wrapping of count is not related to buffer size, but to the
>>>>> amount of trace data written.
>>>>
>>>> Sure, but the chance of ambiguity increases with larger buffer sizes.
>>>
>>> Well, better safe than sorry. Switching to unsigned long will rarely
>>> hurt, so I'm going to do just that. The only downside will be some waste
>>> of buffer space for very long running traces with huge amounts of
>>> entries.
>>
>> Hmm, true. Maybe we could get someone else's opinion on this - anyone?
> 
> TBH, I wouldn't be concerned about the buffer space. In case there are
> really billions of entries, the difference between a 10-digit count
> value and maybe a 15 digit one (and that is already a massive amount)
> isn't that large. The average printed size of count with about
> 4 billion entries is 9.75 digits (and not just 5 :-) ).

Another option would be to let count wrap at e.g. 4 billion (or even 1
million) and add a wrap count. Each buffer struct would contain the
wrap count of the last entry, and when hitting a higher wrap count a
new entry like ("wrap %u->%u", old_wrap, new_wrap) would be printed.
This saves buffer space without loosing information.


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel