Re: [PATCH v1 1/1] app/test-gpudev: introduce ethdev to rx/tx packets using GPU memory

[Elena Agostini <eagostini@nvidia.com>]

[-- Attachment #1: Type: text/plain, Size: 2239 bytes --]

> From: Jerin Jacob <jerinjacobk@gmail.com>
> Date: Thursday, 18 November 2021 at 07:17
> To: Elena Agostini <eagostini@nvidia.com>
> Cc: dpdk-dev <dev@dpdk.org>
> Subject: Re: [PATCH v1 1/1] app/test-gpudev: introduce ethdev to rx/tx packets using GPU memory
> External email: Use caution opening links or attachments>
>

> On Thu, Nov 18, 2021 at 12:28 AM <eagostini@nvidia.com> wrote:
> >
> > From: Elena Agostini <eagostini@nvidia.com>
> >
> > This patch introduces ethdev in test-gpudev app to provide:
> > - an example to show how GPU memory can be used to send and receive packets
> > - an useful tool to measure network metrics when using GPU memory with
> > io forwarding
> >
> > With this feature test-gpudev can:
> > - RX packets in CPU or GPU memory
> > - Store packets in the gpudev communication list
> > - TX receive packets from the communication list
> >
> > It's a simulation of a multi-core application.
> >
> > Signed-off-by: Elena Agostini <eagostini@nvidia.com>
> > ---
> >  app/test-gpudev/main.c | 471 +++++++++++++++++++++++++++++++++++++++--
> >  1 file changed, 452 insertions(+), 19 deletions(-)
> >
> > diff --git a/app/test-gpudev/main.c b/app/test-gpudev/main.c
> > index 250fba6427..daa586c64e 100644
> > --- a/app/test-gpudev/main.c
> > +++ b/app/test-gpudev/main.c
> > @@ -10,6 +10,8 @@
> >  #include <stdarg.h>
> >  #include <errno.h>
> >  #include <getopt.h>
> > +#include <stdbool.h>
> > +#include <signal.h>
> >
> >  #include <rte_common.h>
> >  #include <rte_malloc.h>
> > @@ -19,22 +21,98 @@
> >  #include <rte_ethdev.h>
> >  #include <rte_mempool.h>
> >  #include <rte_mbuf.h>
> > +#include <rte_launch.h>
> > +#include <rte_lcore.h>
> > +#include <rte_per_lcore.h>
> >
> >  #include <rte_gpudev.h>
> >
> > +#ifndef ACCESS_ONCE
> > +#define ACCESS_ONCE(x) (*(volatile typeof(x) *)&x)
> > +#endif
> > +
> > +#ifndef WRITE_ONCE
> > +#define WRITE_ONCE(x, v) (ACCESS_ONCE(x) = (v))
> > +#endif>

> Better to have a public version of this macro as it uses just in this
> test application.

Thanks for taking time to review this patch.
I can actually use the RTE_GPU_VOLATILE macro exposed in the gpudev library
to replace both of them.

[-- Attachment #2: Type: text/html, Size: 8813 bytes --]