> I was trying to think of a graceful way that the SSE4.2 extensions could be checked for and an error reported instead of an illegal instruction.
That is possible, I'd wish that could be a single check at just one lib initializing.
How is the full lddtree look like of your example?
Usually even if you link to a particular sub-lib of dpdk that will in turn link to librte_eal.so eventually - so it could be done in just one place.
In fact I think that was already done for cpu features, e.g. the former (21.11) SIGILL on arm for CRC32 nowadays looks like:
HASH: Unsupported CRC32 algorithm requested using CRC32_SW
ERROR: This system does not support "CRC32".
Please check that RTE_MACHINE is set correctly.
EAL: FATAL: unsupported cpu type.
Could you try Ubuntu Lunar (23.04) which has DPDK 22.11.?
That will still not work, but I wonder if x86 - and sse4.2 on that arch - got a similar "now graceful" check.
> I guess that's about the best you can do. Anyway, thanks for the explanation.
yes, Unless we can convince for full runtime-detection sse4.2
> I've got an older machine ...
That is the reason why I think it might not be addressed at high-prio by anyone.
It was uncommon years ago to lack this. And it is even more uncommon nowadays and will be more uncommon in the future.
> I was trying to think of a graceful way that the SSE4.2 extensions could be checked for and an error reported instead of an illegal instruction.
That is possible, I'd wish that could be a single check at just one lib initializing.
How is the full lddtree look like of your example?
Usually even if you link to a particular sub-lib of dpdk that will in turn link to librte_eal.so eventually - so it could be done in just one place.
In fact I think that was already done for cpu features, e.g. the former (21.11) SIGILL on arm for CRC32 nowadays looks like:
HASH: Unsupported CRC32 algorithm requested using CRC32_SW
ERROR: This system does not support "CRC32".
Please check that RTE_MACHINE is set correctly.
EAL: FATAL: unsupported cpu type.
Could you try Ubuntu Lunar (23.04) which has DPDK 22.11.?
That will still not work, but I wonder if x86 - and sse4.2 on that arch - got a similar "now graceful" check.
> I guess that's about the best you can do. Anyway, thanks for the explanation.
yes, Unless we can convince for full runtime-detection sse4.2
> I've got an older machine ...
That is the reason why I think it might not be addressed at high-prio by anyone.
It was uncommon years ago to lack this. And it is even more uncommon nowadays and will be more uncommon in the future.