Re: [PATCH] vfio: allow to map other memory regions

-----Original Message-----
From: Thomas Monjalon [mailto:thomas@monjalon.net]
Sent: Monday, June 19, 2017 11:04 PM
To: Wodkowski, PawelX <redacted>
Cc: dev@dpdk.org
Subject: Re: [dpdk-dev] [PATCH] vfio: allow to map other memory regions

Hi,
Some comments below

24/05/2017 13:17, Pawel Wodkowski:

quoted

Currently it is not possible to use memory that is not owned by DPDK to
perform DMA. This scenarion might be used in vhost applications (like
SPDK) where guest send its own memory table. To fill this gap provide
API to allow registering arbitrary address in VFIO container.

Signed-off-by: Pawel Wodkowski <redacted>
---
 lib/librte_eal/linuxapp/eal/Makefile            |   3 +
 lib/librte_eal/linuxapp/eal/eal_vfio.c          | 142

+++++++++++++++++++++---

quoted

 lib/librte_eal/linuxapp/eal/eal_vfio.h          |  10 ++
 lib/librte_eal/linuxapp/eal/include/rte_iommu.h |  78 +++++++++++++
 lib/librte_eal/linuxapp/eal/rte_eal_version.map |   8 ++
 5 files changed, 224 insertions(+), 17 deletions(-)
 create mode 100644 lib/librte_eal/linuxapp/eal/include/rte_iommu.h

VFIO is not referenced in the doxygen of these functions.
Could we use this API for something else than VFIO?

This is for any IOMMU hw/module/driver used in host which require special
care about memory regions used for DMA. It is not restricted to VFIO even though
only VFIO is implemented.

Any API should be declared in common directory, even if it is not
implemented for FreeBSD (returning -ENOTSUP).

I think those function should be NOP for FreeBSD (like RTE_VFIO_NOIOMMU do)
or be conditionally compiled/included (like it is now). I decide to take second way.
Do you think that I should move rte_iommu.h to common directory and use #ifdef
there?

Pawel