Yes, this is the solution I was looking for. Thank you. Nitpicking...
Could you use with a more meaningful name than dynfield3, e.g. dynfield_nc or dynfield_no_copy. A few minor clarifications suggested inline below. With or without suggested changes, Reviewed-by: Morten Brørup <[email protected]> > --- a/doc/guides/prog_guide/mbuf_lib.rst > +++ b/doc/guides/prog_guide/mbuf_lib.rst > @@ -234,6 +234,15 @@ The dynamic fields and flags are managed with the > functions ``rte_mbuf_dyn*``. > > It is not possible to unregister fields or flags. > > +The build option ``mbuf_dynfield3_size`` can add extra cache-line- > aligned > +dynamic field storage to ``struct rte_mbuf``. This increases every > mbuf by "increases every mbuf" -> "increases the size of every mbuf" > +the configured amount and changes the mbuf layout, so applications and > +secondary processes must be built with the same value as the primary > process. > +The option defaults to ``0``. The extra storage is reserved for > dynamic > +fields registered with ``RTE_MBUF_DYNFIELD_F_NO_COPY``. These fields > are not > +copied by mbuf copy and clone operations. Dynamic fields registered > without > +this flag continue to use the existing copied dynamic-field storage. > + > .. _direct_indirect_buffer: > > Direct and Indirect Buffers > diff --git a/doc/guides/rel_notes/release_26_11.rst > b/doc/guides/rel_notes/release_26_11.rst > index dec96ccbc7..cff68d7034 100644 > --- a/doc/guides/rel_notes/release_26_11.rst > +++ b/doc/guides/rel_notes/release_26_11.rst > @@ -60,6 +60,22 @@ New Features > Added the experimental ``rte_cpu_socket_id()`` function > to map an OS logical CPU ID to the NUMA socket containing that CPU. > > +* **Added optional extra mbuf dynamic field storage.** > + > + Added ``mbuf_dynfield3_size`` build option to enable a > + cache-line-aligned ``dynfield3`` area in ``struct rte_mbuf``. > + The configured size is defined as ``RTE_MBUF_DYNFIELD3_SIZE`` > + in ``rte_build_config.h``. > + The extra storage is reserved for dynamic fields registered with > + ``RTE_MBUF_DYNFIELD_F_NO_COPY``. > + These fields are not copied by generic mbuf copy and clone > operations. > + Dynamic fields registered without this flag continue to use the > existing > + copied dynamic-field storage. > + Applications and secondary processes must be built with the same > + non-zero value as the primary process. "with the same non-zero value" ->"with the same value" > --- a/meson_options.txt > +++ b/meson_options.txt > @@ -44,6 +44,8 @@ option('max_numa_nodes', type: 'string', value: > 'default', description: > 'Set the highest NUMA node supported by EAL; "default" is > different per-arch, "detect" detects the highest NUMA node on the build > machine.') > option('enable_iova_as_pa', type: 'boolean', value: true, description: > 'Support the use of physical addresses for IO addresses, such > as used by UIO or VFIO in no-IOMMU mode. When disabled, DPDK can only > run with IOMMU support for address mappings, but will have more space > available in the mbuf structure.') > +option('mbuf_dynfield3_size', type: 'integer', min: 0, value: 0, > description: > + 'Size of optional extra mbuf dynamic field area, in bytes.') The description of this option should mention that the area is for mbuf dynamic fields with the no-copy property. > option('mbuf_refcnt_atomic', type: 'boolean', value: true, > description: > 'Atomically access the mbuf refcnt.') > option('platform', type: 'string', value: 'native', description: > -- > 2.35.6
