On 7/8/24 5:04 AM, Jürgen Groß wrote:
On 06.07.24 00:36, boris.ostrovsky@xxxxxxxxxx wrote:
Also, would it be better to keep these fields as a struct of scalars
and instead have the percpu array of this struct? Otherwise there is a
whole bunch of [MC_BATCH] arrays, all of them really indexed by the
same value. (And while at it, there is no reason to have
callbacks[MC_BATCH] sized like that -- it has nothing to do with batch
size and can probably be made smaller)
As today the mc_buffer's entries are copied via a single memcpy(), there
are 3 options:
Ah yes, it's memcpy, I didn't think of that. Then leaving it as is is
the best.
- make mc_debug_data a percpu pointer to a single array, requiring to
copy the mc_buffer's entries in a loop
- let struct mc_debug_data contain two arrays (entries[] and struct foo
{}[],
with struct foo containing the other pointers/values)
- keep the layout as in my patch
Regarding the callbacks: I think the max number of callbacks is indeed
MC_BATCH,
as for each batch member one callback might be requested. So I'd rather
keep it
the way it is today.
Right, I was trying to point out that it's the max number but I suspect
it usually is smaller -- we currently ask for a callback in fewer than
half of the cases where we submit a request.
-boris
