> -----Original Message----- > From: Horatiu Vultur <horatiu.vultur@xxxxxxxxxxxxx> > Sent: Thursday, May 25, 2023 2:49 AM > To: Haiyang Zhang <haiyangz@xxxxxxxxxxxxx> > Cc: linux-hyperv@xxxxxxxxxxxxxxx; netdev@xxxxxxxxxxxxxxx; Dexuan Cui > <decui@xxxxxxxxxxxxx>; KY Srinivasan <kys@xxxxxxxxxxxxx>; Paul Rosswurm > <paulros@xxxxxxxxxxxxx>; olaf@xxxxxxxxx; vkuznets@xxxxxxxxxx; > davem@xxxxxxxxxxxxx; wei.liu@xxxxxxxxxx; edumazet@xxxxxxxxxx; > kuba@xxxxxxxxxx; pabeni@xxxxxxxxxx; leon@xxxxxxxxxx; Long Li > <longli@xxxxxxxxxxxxx>; ssengar@xxxxxxxxxxxxxxxxxxx; linux- > rdma@xxxxxxxxxxxxxxx; daniel@xxxxxxxxxxxxx; john.fastabend@xxxxxxxxx; > bpf@xxxxxxxxxxxxxxx; ast@xxxxxxxxxx; Ajay Sharma > <sharmaajay@xxxxxxxxxxxxx>; hawk@xxxxxxxxxx; linux- > kernel@xxxxxxxxxxxxxxx; stable@xxxxxxxxxxxxxxx > Subject: Re: [PATCH net] net: mana: Fix perf regression: remove rx_cqes, > tx_cqes counters > > [Some people who received this message don't often get email from > horatiu.vultur@xxxxxxxxxxxxx. Learn why this is important at > https://aka.ms/LearnAboutSenderIdentification ] > > The 05/24/2023 14:22, Haiyang Zhang wrote: > > Hi Haiyang, > > > > > The apc->eth_stats.rx_cqes is one per NIC (vport), and it's on the > > frequent and parallel code path of all queues. So, r/w into this > > single shared variable by many threads on different CPUs creates a > > lot caching and memory overhead, hence perf regression. And, it's > > not accurate due to the high volume concurrent r/w. > > Do you have any numbers to show the improvement of this change? The numbers are not published. The perf regression of the previous patch is very significant, and this patch eliminates the regression. > > > > > Since the error path of mana_poll_rx_cq() already has warnings, so > > keeping the counter and convert it to a per-queue variable is not > > necessary. So, just remove this counter from this high frequency > > code path. > > > > Also, remove the tx_cqes counter for the same reason. We have > > warnings & other counters for errors on that path, and don't need > > to count every normal cqe processing. > > Will you not have problems with the counter 'apc->eth_stats.tx_cqe_err'? > It is not in the hot path but you will have concurrent access to it. Yes, but that error happens rarely, so a shared variable is good enough. So, I don't change it in this patch. Thanks, - Haiyang
