| d88f3b1d | 25-Jun-2018 |
YoungGyoun <[email protected]> |
GetRSSCPUCore() now works correctly for N cores (where N is not 2 ^ n) with i40e (e.g., XL710)
-> the functions from the RSS hash value to the Rx queue number are different between NICs
1) ixgbe (e
GetRSSCPUCore() now works correctly for N cores (where N is not 2 ^ n) with i40e (e.g., XL710)
-> the functions from the RSS hash value to the Rx queue number are different between NICs
1) ixgbe (e.g., X520): (Rx queue number) = (7 LS bits of RSS hash value) mod N
* See Section 7.1.2.8 of https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/82599-10-gbe-controller-datasheet.pdf
2) i40e (e.g., XL710): (Rx queue number) = (9 LS bits of RSS hash value) mod N
* See Section 7.1.8 of https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xl710-10-40-controller-datasheet.pdf
-> fixed the GetRSSCPUCore() to handle each case correctly
show more ...
|
| 35f6bf22 | 28-Apr-2018 |
Tom Barbette <[email protected]> |
Fix issue introduced by 05e3289
I had a lot of lines in the NAT like
[translate_addr: 144] mtcp_getpeer() failed sock=61 side=1
In nat.c, MOS_SIDE_CLI has been changed by MOS_SIDE_BOTH, without
cha
Fix issue introduced by 05e3289
I had a lot of lines in the NAT like
[translate_addr: 144] mtcp_getpeer() failed sock=61 side=1
In nat.c, MOS_SIDE_CLI has been changed by MOS_SIDE_BOTH, without
changing the length of the sockaddr len.
Actually, by luck up to this patch the size of a pointer * 2 was equal
to the size of a sockaddr, but now the error in the sizeof appears.
I suspect this also happens in midstat.c, but I don't have time to push
this further.
show more ...
|