| Message ID | 1573243090-2721-1-git-send-email-magnus.karlsson@intel.com |
|---|---|
| State | Awaiting Upstream |
| Delegated to: | David Miller |
| Headers | show |
| Series | [net,1/2] i40e: need_wakeup flag might not be set for Tx | expand |
Jeff, please pick these up. Thanks.
On Fri, 2019-11-08 at 14:11 -0800, David Miller wrote: > Jeff, please pick these up. > > Thanks. Yep already have them queued up. Just waiting confirmation from validation. I will have a 6 patch series of fixes for you in the next day or two.
> -----Original Message----- > From: Intel-wired-lan [mailto:intel-wired-lan-bounces@osuosl.org] On > Behalf Of Magnus Karlsson > Sent: Friday, November 8, 2019 11:58 AM > To: Karlsson, Magnus <magnus.karlsson@intel.com>; Topel, Bjorn > <bjorn.topel@intel.com>; intel-wired-lan@lists.osuosl.org > Cc: maciejromanfijalkowski@gmail.com; Fijalkowski, Maciej > <maciej.fijalkowski@intel.com>; netdev@vger.kernel.org > Subject: [Intel-wired-lan] [PATCH net 1/2] i40e: need_wakeup flag might not > be set for Tx > > The need_wakeup flag for Tx might not be set for AF_XDP sockets that are > only used to send packets. This happens if there is at least one outstanding > packet that has not been completed by the hardware and we get that > corresponding completion (which will not generate an interrupt since > interrupts are disabled in the napi poll loop) between the time we stopped > processing the Tx completions and interrupts are enabled again. In this case, > the need_wakeup flag will have been cleared at the end of the Tx > completion processing as we believe we will get an interrupt from the > outstanding completion at a later point in time. But if this completion > interrupt occurs before interrupts are enable, we lose it and should at that > point really have set the need_wakeup flag since there are no more > outstanding completions that can generate an interrupt to continue the > processing. When this happens, user space will see a Tx queue > need_wakeup of 0 and skip issuing a syscall, which means will never get into > the Tx processing again and we have a deadlock. > > This patch introduces a quick fix for this issue by just setting the > need_wakeup flag for Tx to 1 all the time. I am working on a proper fix for > this that will toggle the flag appropriately, but it is more challenging than I > anticipated and I am afraid that this patch will not be completed before the > merge window closes, therefore this easier fix for now. This fix has a > negative performance impact in the range of 0% to 4%. Towards the higher > end of the scale if you have driver and application on the same core and issue > a lot of packets, and towards no negative impact if you use two cores, lower > transmission speeds and/or a workload that also receives packets. > > Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com> > --- > drivers/net/ethernet/intel/i40e/i40e_xsk.c | 10 ++-------- > 1 file changed, 2 insertions(+), 8 deletions(-) Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Hi Magnus, On 2019-11-08 21:58, Magnus Karlsson wrote: > This happens if there is at least one > outstanding packet that has not been completed by the hardware and we > get that corresponding completion (which will not generate an > interrupt since interrupts are disabled in the napi poll loop) between > the time we stopped processing the Tx completions and interrupts are > enabled again. > But if this completion interrupt occurs before interrupts > are enable, we lose it Why can't it happen for regular traffic? From your description it looks to me as if you can miss a completion for non-AF_XDP traffic, too. Is there any detail that makes this issue AF_XDP-specific? Thanks, Max
On Mon, Nov 25, 2019 at 4:40 PM Maxim Mikityanskiy <maximmi@mellanox.com> wrote: > > Hi Magnus, > > On 2019-11-08 21:58, Magnus Karlsson wrote: > > This happens if there is at least one > > outstanding packet that has not been completed by the hardware and we > > get that corresponding completion (which will not generate an > > interrupt since interrupts are disabled in the napi poll loop) between > > the time we stopped processing the Tx completions and interrupts are > > enabled again. > > > But if this completion interrupt occurs before interrupts > > are enable, we lose it > Why can't it happen for regular traffic? From your description it looks > to me as if you can miss a completion for non-AF_XDP traffic, too. Is > there any detail that makes this issue AF_XDP-specific? It can happen for regular traffic too, but it does not matter in that case since the application is not dependent on getting a notification on completion. The only thing that will happen is that there is some memory and HW descriptors being used for longer than necessary. It will get completed and reused next time there is some Tx action. In the cases where it matters, the skb path code has feature that arms a later interrupt. I need to introduce something similar for the AF_XDP ZC path. /Magnus > Thanks, > Max > _______________________________________________ > Intel-wired-lan mailing list > Intel-wired-lan@osuosl.org > https://lists.osuosl.org/mailman/listinfo/intel-wired-lan
diff --git a/drivers/net/ethernet/intel/i40e/i40e_xsk.c b/drivers/net/ethernet/intel/i40e/i40e_xsk.c index a05dfec..d07e1a8 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_xsk.c +++ b/drivers/net/ethernet/intel/i40e/i40e_xsk.c @@ -689,8 +689,6 @@ static bool i40e_xmit_zc(struct i40e_ring *xdp_ring, unsigned int budget) i40e_xdp_ring_update_tail(xdp_ring); xsk_umem_consume_tx_done(xdp_ring->xsk_umem); - if (xsk_umem_uses_need_wakeup(xdp_ring->xsk_umem)) - xsk_clear_tx_need_wakeup(xdp_ring->xsk_umem); } return !!budget && work_done; @@ -769,12 +767,8 @@ bool i40e_clean_xdp_tx_irq(struct i40e_vsi *vsi, i40e_update_tx_stats(tx_ring, completed_frames, total_bytes); out_xmit: - if (xsk_umem_uses_need_wakeup(tx_ring->xsk_umem)) { - if (tx_ring->next_to_clean == tx_ring->next_to_use) - xsk_set_tx_need_wakeup(tx_ring->xsk_umem); - else - xsk_clear_tx_need_wakeup(tx_ring->xsk_umem); - } + if (xsk_umem_uses_need_wakeup(tx_ring->xsk_umem)) + xsk_set_tx_need_wakeup(tx_ring->xsk_umem); xmit_done = i40e_xmit_zc(tx_ring, budget);
The need_wakeup flag for Tx might not be set for AF_XDP sockets that are only used to send packets. This happens if there is at least one outstanding packet that has not been completed by the hardware and we get that corresponding completion (which will not generate an interrupt since interrupts are disabled in the napi poll loop) between the time we stopped processing the Tx completions and interrupts are enabled again. In this case, the need_wakeup flag will have been cleared at the end of the Tx completion processing as we believe we will get an interrupt from the outstanding completion at a later point in time. But if this completion interrupt occurs before interrupts are enable, we lose it and should at that point really have set the need_wakeup flag since there are no more outstanding completions that can generate an interrupt to continue the processing. When this happens, user space will see a Tx queue need_wakeup of 0 and skip issuing a syscall, which means will never get into the Tx processing again and we have a deadlock. This patch introduces a quick fix for this issue by just setting the need_wakeup flag for Tx to 1 all the time. I am working on a proper fix for this that will toggle the flag appropriately, but it is more challenging than I anticipated and I am afraid that this patch will not be completed before the merge window closes, therefore this easier fix for now. This fix has a negative performance impact in the range of 0% to 4%. Towards the higher end of the scale if you have driver and application on the same core and issue a lot of packets, and towards no negative impact if you use two cores, lower transmission speeds and/or a workload that also receives packets. Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com> --- drivers/net/ethernet/intel/i40e/i40e_xsk.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)