2023-01-30 11:23 (UTC+0200), Ophir Munk:
> In current DPDK the RTE_MAX_MEMZONE definition is unconditionally hard
> coded as 2560. For applications requiring different values of this
> parameter – it is more convenient to set its value as part of the meson
> command line or to set the max value via an rte API - rather than
> changing the dpdk source code per application.
>
> An example would be of an application that uses the DPDK mempool library
> which is based on DPDK memzone library. The application may need to
> create a number of steering tables, each of which will require its own
> mempool allocation. This RFC is not about how to optimize the
> application usage of mempool nor about how to improve the mempool
> implementation based on memzone. It is about how to make the max
> memzone definition - build-time or run-time customized.
>
> I would like to suggest three options.
>
> Option 1
> ========
> Add a Meson option in meson options.txt and remove the
> RTE_MAX_MEMZONE definition from config/rte_config.h
>
[...]
>
> Option 2
> ========
> Use Meson setup -Dc_args="-DRTE_MAX_MEMZONE=XXX" and
> make RTE_MAX_MEMZONE conditional in config/rte_config.h
>
> For example, see the code of this commit.
>
> Option 3
> ========
> Add a function which must be called before rte_eal_init():
> void rte_memzone_set_max(int max) {memzone_max = max;}
> If not called, the default memzone (RTE_MAX_MEMZONE) is used.
>
> With this option there is no need to recompile DPDK and it allows
> using an in-box packaged DPDK.
[...]
Ideally, there should be no limitation at all.
I vote for some compile-time solution for now
with a plan to remove the restriction inside EAL later.
Option 2 does not expose a user-facing option that would be obsoleted anyway,
so it seems preferable to me, but Option 1 is also OK and consistent.
`RTE_MAX_MEMZONE` is needed currently, because `struct rte_memzone` are stored
in `rte_fbarray` (mapped file-backed array) with a fixed capacity.
Unlike e.g. `rte_eth_devices`, this is not used for efficient access by index
and is not even exposed to PMDs, so the storage can be changed painlessly.
In DPDK, only net/qede uses this constant and also for slow-path bookkeeping.
