Message ID | de331296-a914-d8c7-2f25-a180724c5489@canonical.com |
---|---|
Headers | show |
Series | UBUNTU: SAUCE: bnxt_en_bpo: Fix TX timeout during netpoll | expand |
On Fri, 22 Feb 2019 11:20:18 +0100 Nivedita Singhvi <nivedita.singhvi@canonical.com> wrote: > BugLink: http://bugs.launchpad.net/bugs/1814095 > > > [Impact] > > The bnxt_en_bpo driver experienced tx timeouts causing the system to experience > network stalls and fail to send data and heartbeat packets. > > The following 25Gb Broadcom NIC error was seen on Xenial running the > 4.4.0-141-generic kernel on an amd64 host seeing moderate-heavy network > traffic (just once): > > * The bnxt_en_po driver froze on a "TX timed out" error and triggered the > Netdev Watchdog timer under load. > > * From kernel log: > "NETDEV WATCHDOG: eno2d1 (bnxt_en_bpo): transmit queue 0 timed out" > See attached kern.log excerpt file for full excerpt of error log. > > * Release = Xenial > Kernel = 4.4.0-141-generic #167 > eno2d1 = Product Name: Broadcom Adv. Dual 25Gb Ethernet > > * This caused the driver to reset in order to recover: > > "bnxt_en_bpo 0000:19:00.1 eno2d1: TX timeout detected, starting reset task!" > > driver: bnxt_en_bpo > version: 1.8.1 > source: ubuntu/bnxt/bnxt.c: bnxt_tx_timeout() > > * The loss of connectivity and softirq stall caused other cascading failures > on the system. > > * The bnxt_en_po driver is the imported Broadcom driver pulled in to support > newer Broadcom HW (specific boards) while the bnx_en module continues to > support the older HW. The current Linux upstream driver does not compile > easily with the 4.4 kernel (too many changes). > > * This upstream and bnxt_en driver fix is a likely solution: > "bnxt_en: Fix TX timeout during netpoll" > commit: 73f21c653f930f438d53eed29b5e4c65c8a0f906 > > This fix has not been applied to the bnxt_en_po driver version, but review of > the code indicates that it is susceptible to the bug, and the fix would be > reasonable. > > > [Test Case] > > * Unfortunately, this is not easy to reproduce. Also, it is only seen on > 4.4 kernels with newer Broadcom NICs supported by the bnxt_en_bpo driver. > > > [Regression Potential] > > * The patch is restricted to the bpo driver, with very constrained scope > - just the newest Broadcom NICs being used by the Xenial 4.4 kernel (as > opposed to the hwe 4.15 etc. kernels, which would have the in-tree fixed > driver). 4.15 doesn't have this patch. > * The patch is very small and backport is fairly minimal and simple. > > * The fix has been running on the in-tree driver in upstream mainline as well > as the Ubuntu Linux in-tree driver, although the Broadcom driver has a lot of > lower level code that is different, this piece is still the same. I'm a little reluctant to ACK this given that a) it's an upstream patch that is applied to an out-of-tree vendor driver and b) the problem can't be reproduced (or can it)? If I gave you a test kernel with a newer Broadcom driver, would you be able to do some testing? The latest Broadcom driver is version 1.9.2 from 01/02/2019 but it doesn't have the patch that you want. ...Juerg
On 2/26/19 3:07 PM, Juerg Haefliger wrote: > On Fri, 22 Feb 2019 11:20:18 +0100 > Nivedita Singhvi <nivedita.singhvi@canonical.com> wrote: > >> BugLink: http://bugs.launchpad.net/bugs/1814095 >> >> >> [Impact] >> >> The bnxt_en_bpo driver experienced tx timeouts causing the system to experience >> network stalls and fail to send data and heartbeat packets. >> >> The following 25Gb Broadcom NIC error was seen on Xenial running the >> 4.4.0-141-generic kernel on an amd64 host seeing moderate-heavy network >> traffic (just once): >> >> * The bnxt_en_po driver froze on a "TX timed out" error and triggered the >> Netdev Watchdog timer under load. >> >> * From kernel log: >> "NETDEV WATCHDOG: eno2d1 (bnxt_en_bpo): transmit queue 0 timed out" >> See attached kern.log excerpt file for full excerpt of error log. >> >> * Release = Xenial >> Kernel = 4.4.0-141-generic #167 >> eno2d1 = Product Name: Broadcom Adv. Dual 25Gb Ethernet >> >> * This caused the driver to reset in order to recover: >> >> "bnxt_en_bpo 0000:19:00.1 eno2d1: TX timeout detected, starting reset task!" >> >> driver: bnxt_en_bpo >> version: 1.8.1 >> source: ubuntu/bnxt/bnxt.c: bnxt_tx_timeout() >> >> * The loss of connectivity and softirq stall caused other cascading failures >> on the system. >> >> * The bnxt_en_po driver is the imported Broadcom driver pulled in to support >> newer Broadcom HW (specific boards) while the bnx_en module continues to >> support the older HW. The current Linux upstream driver does not compile >> easily with the 4.4 kernel (too many changes). >> >> * This upstream and bnxt_en driver fix is a likely solution: >> "bnxt_en: Fix TX timeout during netpoll" >> commit: 73f21c653f930f438d53eed29b5e4c65c8a0f906 >> >> This fix has not been applied to the bnxt_en_po driver version, but review of >> the code indicates that it is susceptible to the bug, and the fix would be >> reasonable. >> >> >> [Test Case] >> >> * Unfortunately, this is not easy to reproduce. Also, it is only seen on >> 4.4 kernels with newer Broadcom NICs supported by the bnxt_en_bpo driver. >> >> >> [Regression Potential] >> >> * The patch is restricted to the bpo driver, with very constrained scope >> - just the newest Broadcom NICs being used by the Xenial 4.4 kernel (as >> opposed to the hwe 4.15 etc. kernels, which would have the in-tree fixed >> driver). > > 4.15 doesn't have this patch. Which also does need patching (earlier thought had B only needing the hwe 4.18 kernel). I will submit that against this same bug. >> * The patch is very small and backport is fairly minimal and simple. >> >> * The fix has been running on the in-tree driver in upstream mainline as well >> as the Ubuntu Linux in-tree driver, although the Broadcom driver has a lot of >> lower level code that is different, this piece is still the same. > > I'm a little reluctant to ACK this given that a) it's an upstream patch that > is applied to an out-of-tree vendor driver and b) the problem can't be > reproduced (or can it)? If I gave you a test kernel with a newer Broadcom > driver, would you be able to do some testing? Sadly, as I don't have the specific Broadcom NICs yet which this driver supports, it's moot, and we're very dependent on getting users who do have the HW to test it. I would imagine that getting testing of the full new driver will be difficult and take more time. It is certainly a very infrequent to rare occurrence, but I'm reluctant to quantify that as we don't have sufficient details regarding the workload itself and how long that had been running. On the one occurrence of the issue, bad things happened with their Kafka server outage etc. As there is a fix and it is upstream this seemed the safer option. There are not too many people yet who have the newer Broadcom card and are also running 4.4 kernels (and hence the bnxt_en_bpo driver). Hence it's unclear how often this might happen in future (more likely, I would presume, with more people getting new HW). > > The latest Broadcom driver is version 1.9.2 from 01/02/2019 but it doesn't > have the patch that you want. Right, so we'd need to add the same fix to that as well, so it would not be any safer to pull in the full new Broadcom driver. I agree that we need a path forward with respect to updating this driver and I'll ask again if I can get hold of the right HW, but I suspect that it will not be very soon. The thinking was that if our options are to a) pull in full new driver, b) minimally patch existing driver, c) make no change, (b) would be least evil of the options. Let me know how you want to proceed, and believe me, I fully understand the reluctance ;).
I've had several discussions with Nivedita and Jay about this patch and I think it is worth pulling in. The issue being hit can not be reproduced. Let's go ahead and take this patch. Brad On Fri, Feb 22, 2019 at 11:20:18AM +0100, Nivedita Singhvi wrote: > BugLink: http://bugs.launchpad.net/bugs/1814095 > > > [Impact] > > The bnxt_en_bpo driver experienced tx timeouts causing the system to experience > network stalls and fail to send data and heartbeat packets. > > The following 25Gb Broadcom NIC error was seen on Xenial running the > 4.4.0-141-generic kernel on an amd64 host seeing moderate-heavy network > traffic (just once): > > * The bnxt_en_po driver froze on a "TX timed out" error and triggered the > Netdev Watchdog timer under load. > > * From kernel log: > "NETDEV WATCHDOG: eno2d1 (bnxt_en_bpo): transmit queue 0 timed out" > See attached kern.log excerpt file for full excerpt of error log. > > * Release = Xenial > Kernel = 4.4.0-141-generic #167 > eno2d1 = Product Name: Broadcom Adv. Dual 25Gb Ethernet > > * This caused the driver to reset in order to recover: > > "bnxt_en_bpo 0000:19:00.1 eno2d1: TX timeout detected, starting reset task!" > > driver: bnxt_en_bpo > version: 1.8.1 > source: ubuntu/bnxt/bnxt.c: bnxt_tx_timeout() > > * The loss of connectivity and softirq stall caused other cascading failures > on the system. > > * The bnxt_en_po driver is the imported Broadcom driver pulled in to support > newer Broadcom HW (specific boards) while the bnx_en module continues to > support the older HW. The current Linux upstream driver does not compile > easily with the 4.4 kernel (too many changes). > > * This upstream and bnxt_en driver fix is a likely solution: > "bnxt_en: Fix TX timeout during netpoll" > commit: 73f21c653f930f438d53eed29b5e4c65c8a0f906 > > This fix has not been applied to the bnxt_en_po driver version, but review of > the code indicates that it is susceptible to the bug, and the fix would be > reasonable. > > > [Test Case] > > * Unfortunately, this is not easy to reproduce. Also, it is only seen on > 4.4 kernels with newer Broadcom NICs supported by the bnxt_en_bpo driver. > > > [Regression Potential] > > * The patch is restricted to the bpo driver, with very constrained scope > - just the newest Broadcom NICs being used by the Xenial 4.4 kernel (as > opposed to the hwe 4.15 etc. kernels, which would have the in-tree fixed > driver). > > * The patch is very small and backport is fairly minimal and simple. > > * The fix has been running on the in-tree driver in upstream mainline as well > as the Ubuntu Linux in-tree driver, although the Broadcom driver has a lot of > lower level code that is different, this piece is still the same. > > > -- > kernel-team mailing list > kernel-team@lists.ubuntu.com > https://lists.ubuntu.com/mailman/listinfo/kernel-team