On 11/1/2017 4:06 AM, santosh wrote:
>
> On Wednesday 01 November 2017 04:01 PM, Thomas Monjalon wrote:
>> 01/11/2017 02:07, Ferruh Yigit:
>>> Fix kernel crash with KNI because KNI requires physical addresses.
>>>
>>> A config option introduced to disable IOVA mode detection and to set it
>>> to physical address by default. Disabling config option will enable IOVA
>>> mode detection.
>>>
>>> When there is no intension to use KNI, it is safe to enable detection.
>>>
>>> Config option disable IOVA mode detection by default to be sure only who
>>> is aware of result enable it.
>>>
>>> Fixes: 72d013644bd6 ("mem: honor IOVA mode in malloc virt2phy")
>>>
>>> Signed-off-by: Ferruh Yigit <ferruh.yi...@intel.com>
>>> ---
>>> +#ifdef RTE_EAL_USE_PHYS_IOVA
>>> + rte_eal_get_configuration()->iova_mode = RTE_IOVA_PA;
>>> +#else
>>> /* autodetect the iova mapping mode (default is iova_pa) */
>>> rte_eal_get_configuration()->iova_mode = rte_bus_get_iommu_class();
>>> +#endif
>> I don't understand why you are adding a compile-time option.
>> I think it should be an EAL option --use-phys-addr.
>> The opposite option may be required to force VA: --use-virt-addr.
>> And if there is no option given, we fallback to autodetect.
>> We can improve the autodetect by checking whether rte_kni.ko is loaded.
>
> IMO, we could introduce a rule for kni in iova autodetection(/for linux case)
> such away that:
>
> if cat /proc/modules has rte_kni entry
> then use default iova=pa mode.
This sounds good as a workaround for this release [1].
For long term I believe KNI should be updated to work with virtual addresses.
[1]
This won't work if both IOVA VA mode only device and KNI used together. But I
assume this is not common use case.
>
> so +1 to your proposition Thomas.
>
>> Opinions?
>
