[DTrace-devel] [PATCH] bpf: allocate the buffers BPF map to fit highest CPU id

[Eugene Loh]

This patch is clearly a step in the right direction (fixes a bug), and I 
do not want perfection to be the enemy of good here. So, if you're 
willing to add some remarks to the commit message that describe some of 
the issues we're discussing here -- and, quite frankly, even if you're 
not -- I'm good with this patch.

And...

On 12/12/25 00:34, Kris Van Hees wrote:

> On Fri, Dec 12, 2025 at 12:14:20AM -0500, Eugene Loh wrote:
>> On 12/11/25 23:36, Kris Van Hees wrote:
>>> I am not entirely sure we *can* test this properly specialized configs.  The
>>> problem came to light on a configuration that conveniently demonstrated the
>>> problem because of a large gap of unavailable CPUs (0-127 possible, but only
>>> 0-7,100-127 were online).  Because in that situation, it is quite likely that
>>> the CPU on which the events we are interested in occur falls beyond the end
>>> of the buffers BPF map.
>> I think at some level we agree that it is possible to test but that it can
>> be hard to do so (automated, regular, etc.).
> Yes, but quite difficult and very specialized.

Difficult?  It's outside of our normal routine, but we can get access to 
such systems and we could devise tests.  The problem is not that such 
testing is more challenging than other things we have to do, but only 
that we have too many other things to do.

Specialized?  The distinction between on-line and possible CPUs shows up 
in a number of places in the code.

Anyhow, yeah, let's try our best *and* not beat ourselves up on this one.

And, of course, thanks for fixing this problem.