issue: 1050049 Move poll_os logic to the internal thread #445
Conversation
|
Test FAILed. |
|
Test FAILed. |
|
Test FAILed. |
|
Test PASSed. |
| Similar to VMA_RX_SKIP_OS, but in select(), poll() or epoll_wait() this will | ||
| force the VMA to check the non offloaded fd even though an offloaded socket | ||
| has ready packets found while polling. | ||
| This will force the VMA to check the non offloaded fd even though an offloaded |
There was a problem hiding this comment.
Revert this change
This commit is not based on udp poll os commit (pr Mellanox#445). Signed-off-by: Liran Oz <[email protected]>
This commit is not based on udp poll os commit (pr Mellanox#445). * f0adf45 issue: 1117626 Move epoll poll_os logic to the internal thread Signed-off-by: Liran Oz <[email protected]>
This commit is not based on udp poll os commit (pr Mellanox#445). Fix review comments Remove boolean var from io_mux_call::handle_os_countdown() * 6a0b53a issue: 1117626 Move epoll poll_os logic to the internal thread * b1de6f5 issue: 1117626 Move epoll poll_os logic to the internal thread * f0adf45 issue: 1117626 Move epoll poll_os logic to the internal thread Signed-off-by: Liran Oz <[email protected]>
liranoz12
force-pushed
the
move_poll_os
branch
from
3f0742a to
b7e0ca4
Compare
|
Test FAILed. |
|
bot:retest |
|
Test FAILed. |
|
bot:retest |
|
Test PASSed. |
|
Test FAILed. |
|
Test PASSed. |
liranoz12
force-pushed
the
move_poll_os
branch
from
97bd221 to
310a745
Compare
|
Test FAILed. |
|
bot:retest |
|
Test FAILed. |
|
bot:retest |
|
Test FAILed. |
By default VMA checks the os fd once for 100 cq polls (VMA_RX_UDP_POLL_OS_RATIO) in order to receive UDP packets with size > MTU. This commit moves this logic to the internal thread which uses epoll_wait() to indicates that non-offloaded data is available. This commit removes VMA_RX_UDP_POLL_OS_RATIO parameter. Signed-off-by: Liran Oz <[email protected]>
liranoz12
force-pushed
the
move_poll_os
branch
from
310a745 to
a3e5b67
Compare
|
Test PASSed. |
|
Can one of the admins verify this patch? |
|
can be usefull |
By default VMA checks the os fd once for 100 cq polls (VMA_RX_UDP_POLL_OS_RATIO)
in order to receive UDP packets with size > MTU.
This commit moves this logic to the internal thread which uses
epoll_wait() to indicates that non-offloaded data is available.
This commit removes VMA_RX_UDP_POLL_OS_RATIO parameter.
Signed-off-by: Liran Oz [email protected]