Re: [dpdk-dev] [RFC 0/3] Make device mapping more reliable

["Burakov, Anatoly" <anatoly.burakov@intel.com>]

On 31-May-18 11:57 AM, Anatoly Burakov wrote:
> Currently, memory for device maps is allocated ad-hoc, by calculating
> end of VA space allocated for hugepages and crossing fingers in hopes that
> those addresses will be free in primary and secondary processes. This leads
> to situations such as this:
> 
> EAL: Detected 88 lcore(s)
> EAL: Detected 2 NUMA nodes
> EAL: Multi-process socket /var/run/dpdk/rte/mp_socket_178323_8af2229603de4
> EAL: Probing VFIO support...
> EAL: VFIO support initialized
> EAL: PCI device 0000:81:00.0 on NUMA socket 1
> EAL:   probe driver: 8086:1563 net_ixgbe
> EAL: Cannot mmap device resource file /sys/bus/pci/devices/0000:81:00.0/resource0 to address: 0x7ff7f5800000
> EAL: Requested device 0000:81:00.0 cannot be used
> EAL: Error - exiting with code: 1
>    Cause: No Ethernet ports - bye
> 
> As can be seen from the above log, secondary process has initialized
> successfully, but device BAR mapping has failed, which resulted in missing ports
> in the secondary process.
> 
> This patchset is an attempt to fix this problem once and for all, by using
> the same method we use for memory to do device mappings as well. That is,
> by preallocating all of the device memory in advance, so that initialization
> either succeeds and allows for device mappings, or it fails outright (whereas
> currently we may be in an in-between kind of situation, where init has
> succeeded but device mappings have failed).
> 
> This change breaks the ABI, so it is not for this release. However, i'd like
> to hear feedback on the approach and whether there are potential problems with
> other buses/use cases that i didn't think of.
> 

I would like to draw attention to the RFC, now that the release has gone 
out. We have a deprecation notice for ABI break due to external memory 
work, so we now have an opportunity to implement this for 18.11.

-- 
Thanks,
Anatoly