On 09/25, Pavel Begunkov wrote:
On 9/24/26 17:48, Stanislav Fomichev wrote:
quoted
On 09/24, Pavel Begunkov wrote:
quoted
On 9/24/26 16:02, Mina Almasry wrote:
quoted
On Tue, Sep 22, 2026 at 1:43 PM Stanislav Fomichev [off-list ref] wrote:
quoted
io_uring zero-copy receive and devmem both maintain a bounded LIFO for
net_iovs in a contiguous area. Store the freelist in struct net_iov_area
and provide common push and pop helpers.
Leave synchronization to area owners. Keep devmem's area adjacent to its
This could be a follow up change, but I think synchronization should
be provided by the netmem/niov infra, rather than the area owners. TBH
the infra providing an unsynchronized data structure and letting the
area owner use it and shoot themselves in the foot feels error prone.
For now we could use a comment.
I don't think we want it. The duplication is minor, but I'm not set
on the per area index array approach, and it'd make changing it
more difficult.
Do you want me to not touch iou in this patch at all? Or are you talking
about potential future synchronization part?
Sorry, I should've more specific. I meant that I don't think trying to
consolidate it at all makes much sense, and I'd just drop this patch.
quoted
I don't see how this is making changing it more difficult, the freelist is
now behind push/pop which you can freely change, and both UAPIs benefit
from a faster/better freelist.
There are several reasons. If there is any mismatch in how it's done
b/w io_uring and devmem it'd need to be split back, and then having it
in struct net_iov_area for devmem only wouldn't make sense. E.g.
Packaging of {area/offset} pair if converted to a ifq global list might
differ. And it moves one part of buffer management to another tree, and
there was already a precedent of patches being blocked for no technical
reasons; I'd rather minimise cross-tree changes. Synchronisation might
also be a bit difficult, zcrx uses it to protect more than just the
freelist modifications, patterns like lock(zcrx_area->common_area.lock)
in the zcrx code is usually not a great idea. In general, I just think
the upside is smaller comparing to losing the development flexibility,
it's not it makes it faster, nor the current code is tricky or complex.
Hope it explains it.
SG, none of these sound material to me :-p but I'll resubmit with io_uring
part removed. I do like the index approach (for halving the array memory
requirements), will switch devmem to it.