I would like to share some preliminary findings regarding this issue. According to the customer's description, the problem persists in mainline v6.12.16 but appears to be resolved in v6.12.17. On the Ubuntu side, a noticeable performance degradation has been observed in kernel version v6.8.0-40.
Upon reviewing the commits between v6.12.16 and the stable v6.12.17, one commit stands out as potentially relevant: 997ef6117efc tcp: adjust rcvq_space after updating scaling ratio This commit addresses the commit a2cbb1603943 (tcp: Update window clamping condition). The performance degradation may be linked to the commit 7d9aff859ddc (tcp: increase the default TCP scaling ratio), which is present in v6.8.0-40 but not in v6.8.0-39. It may be worthwhile to either revert this commit from mainline v6.12.17 or backport it to v6.8.0-40 for testing purposes. If the issue persists after testing, we could proceed with bisecting the commits between v6.12.16 and v6.12.17 to gather more insight. -- You received this bug notification because you are a member of Kernel Packages, which is subscribed to linux in Ubuntu. https://bugs.launchpad.net/bugs/2107347 Title: Severe Network Performance Regression with BCM57800/BCM57810 on Kernel 6.8.0-40 through 6.12.16 Status in linux package in Ubuntu: Confirmed Bug description: Significant network performance degradation when using Broadcom BCM57800/BCM57810 (bnx2x) NICs with jumbo frames on Ubuntu Jammy with kernels starting from 6.8.0-38, with the issue becoming severe and rendering the system unusable from 6.8.0-40 onwards. There is a similar report here: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2098961 However, the proposed fix in that report does not resolve the issue for BCM57800 using the bnx2x driver. Environment: OS: Ubuntu 22.04(Jammy Jellyfish) Current HWE Kernel: 6.8.0-57.59~22.04.1 NIC: Broadcom BCM57800 / BCM57810 Driver: bnx2x Observed Behavior: Noticeable network degradation begins in kernel 6.8.0-38 Becomes unusable in kernel 6.8.0-40 and beyond Still reproducible in latest HWE kernel 6.8.0-57 Lowering the MTU to 1500 increases performance Expected Behavior: Stable and performant network performance as seen in previous kernel versions Additional Notes: As mentioned in LP#2098961, the issue appears to be fixed in kernel 6.12.17[0]. Request that this fix be backported or that further investigation be made into kernel regressions affecting bnx2x on these NICs. [0] https://kernel.ubuntu.com/mainline/v6.12.17/ To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2107347/+subscriptions -- Mailing list: https://launchpad.net/~kernel-packages Post to : kernel-packages@lists.launchpad.net Unsubscribe : https://launchpad.net/~kernel-packages More help : https://help.launchpad.net/ListHelp
