[dpdk-dev] [PATCH v3 0/3] enable uio_pci_generic support

Bruce Richardson bruce.richardson at intel.com
Fri Feb 20 17:59:14 CET 2015


v3 changes
- made processing of uio devices identical, irrespective of igb_uio or uio_pci_generic
- removed storage of kernel driver name from dev node as now unneeded.

v2 changes:
- Change variable name of kernel driver with precise comment
- Fix a union definition error in v1 patchset
- Move redefined macro IORESOURCE_MEM to rte_pci.h with comment

Linux kernel provides UIO as well as VFIO mechanism to support writing user
space device driver. Comparing to UIO which is available since 2.6.32 kernel,
the VFIO is introduced into kernel since version 3.6.0 with better interrupt
and memory protection (build on top of Intel VT-d technology) supports.
Basically, UIO and VFIO do two common things below:
1) Map PCIe device's I/O memory space to user space driver
2) Support device interrupt notification mechanism that notifies user space
   driver/application when a device interrupt triggers.

To run an DPDK application and make use of VFIO, two in_kernel modules
vfio and vfio-pci module must be loaded. To use UIO, a DPDK kernel
module igb_uio, which was there since DPDK is invented, must be loaded to
attach to in_kernel uio module. As an solution to deprecate igb_uio,
this patch serials leverage the uio_pci_generic in_kernel module to support
DPDK user space PMD in a generic fashion (similar to how VFIO works), to
remove user space DPDK dependency on GPL code igb_uio in kernel.

Example to bind Network Ports to uio_pci_generic:
           modprobe uio
           modprobe uio_pci_generic
           /* to bind device 08:00.0, to the uio_pci_generic driver */
           ./tools/dpdk_nic_bind.py -b uio_pci_generic 08:00.0

Note: this patch set does not remove igb_uio support due to igb_uio supports
creating maximum number of SR-IOV VFs (Virtual Functions) by using max_vfs
kernel parameter on older kernels (kernel 3.7.x and below).
Specifically, igb_uio explicitly calls pci_enable_sriov() to create VFs, while
it is not invoked in either uio or uio_pci_generic kernel modules. On kernel 3.8.x
and above, user can use the standard sysfs to enable VFs. For examples:

#echo $num_vf_enabled > /sys/class/net/$dev/device/sriov_numvfs  // enable VFs
#echo 0 > /sys/class/net/$dev/device/sriov_numvfs                // disable VFs


Zhou Danny (3):
  eal: enable uio_pci_generic support
  eal: add interrupt enable/disable routines for uio_pci_generic
  tools: enable binding NIC device to uio_pci_generic

 lib/librte_eal/common/include/rte_pci.h            |   3 +
 lib/librte_eal/linuxapp/eal/eal_interrupts.c       |  68 +++++++--
 lib/librte_eal/linuxapp/eal/eal_pci.c              |   4 +-
 lib/librte_eal/linuxapp/eal/eal_pci_uio.c          | 167 ++++++++++-----------
 .../linuxapp/eal/include/exec-env/rte_interrupts.h |   8 +-
 tools/dpdk_nic_bind.py                             |   2 +-
 6 files changed, 144 insertions(+), 108 deletions(-)

-- 
2.1.0



More information about the dev mailing list