Re: [dpdk-dev] [PATCH v3 1/8] eal: introduce DMA memory barriers

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

-----Original Message-----
> Date: Mon, 22 Jan 2018 18:29:31 +0000
> From: Yongseok Koh <yskoh@mellanox.com>
> To: Andrew Rybchenko <arybchenko@solarflare.com>, Thomas Monjalon
>  <thomas@monjalon.net>, "jianbo.liu@arm.com" <jianbo.liu@arm.com>, Jerin
>  Jacob <jerin.jacob@caviumnetworks.com>
> CC: Adrien Mazarguil <adrien.mazarguil@6wind.com>, Nélio Laranjeiro
>  <nelio.laranjeiro@6wind.com>, "bruce.richardson@intel.com"
>  <bruce.richardson@intel.com>, "Ananyev, Konstantin"
>  <konstantin.ananyev@intel.com>, Chao Zhu <chaozhu@linux.vnet.ibm.com>,
>  "dev@dpdk.org" <dev@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH v3 1/8] eal: introduce DMA memory barriers
> 
> 
> > On Jan 18, 2018, at 11:16 PM, Andrew Rybchenko <arybchenko@solarflare.com> wrote:
> > 
> > On 01/19/2018 03:44 AM, Yongseok Koh wrote:
> >> This commit introduces rte_dma_wmb() and rte_dma_rmb(), in order to
> >> guarantee the ordering of coherent shared memory between the CPU and a DMA
> >> capable device.
> >> 
> >> Signed-off-by: Yongseok Koh 
> >> <yskoh@mellanox.com>
> > 
> > Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
> > 
> > It is already really good. Many thanks.
> 
> Thank you!
> 
> > Maybe it would be useful to:
> >  - avoid duplication of so long explanations (put in in one place and add reference?)
> 
> May have to ask Thomas how to do this. Thomas?
> 
> >  - explain why it is bound to DMA or call it in a different way, since right now it is bound
> >    to coherent-mapped IO (rte_cio_rmb() ?). Yes, I see benefits to follow Linux
> >    terminology, but may be DPDK can do better :) I just add my concerns, but let
> >    EAL code maintainers to decide
> 
> Good idea. Like to hear from other people. But, following linux terms sometime
> could be good to welcome developers from kernel community to DPDK world. :-)
> 
> To people in the cc list, any other concerns?
> Especially ARM users - Jianbo and Jerin?

I like Andrew's suggestion. IMO, rte_cio_?mb() makes more sense.

> 
> Thanks,
> Yongseok