From mboxrd@z Thu Jan 1 00:00:00 1970 From: Florian Fainelli Subject: Re: [PATCH net] r8169: fix NAPI handling under high load Date: Tue, 16 Oct 2018 16:08:12 -0700 Message-ID: <92db72c9-1f8c-d6a5-bcf4-241fa4c5a310@gmail.com> References: <8f84fe39-3d8d-396d-3b97-027e0a83f8cb@gmail.com> <20181016160355.1cc0a2e9@xeon-e3> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit Cc: Heiner Kallweit , David Miller , Realtek linux nic maintainers , "netdev@vger.kernel.org" To: Stephen Hemminger , =?UTF-8?Q?Holger_Hoffst=c3=a4tte?= Return-path: Received: from mail-pf1-f193.google.com ([209.85.210.193]:37563 "EHLO mail-pf1-f193.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726703AbeJQHBH (ORCPT ); Wed, 17 Oct 2018 03:01:07 -0400 Received: by mail-pf1-f193.google.com with SMTP id j23-v6so12184877pfi.4 for ; Tue, 16 Oct 2018 16:08:27 -0700 (PDT) In-Reply-To: <20181016160355.1cc0a2e9@xeon-e3> Content-Language: en-US Sender: netdev-owner@vger.kernel.org List-ID: On 10/16/2018 04:03 PM, Stephen Hemminger wrote: > On Tue, 16 Oct 2018 23:17:31 +0200 > Holger Hoffstätte wrote: > >> On 10/16/18 22:37, Heiner Kallweit wrote: >>> rtl_rx() and rtl_tx() are called only if the respective bits are set >>> in the interrupt status register. Under high load NAPI may not be >>> able to process all data (work_done == budget) and it will schedule >>> subsequent calls to the poll callback. >>> rtl_ack_events() however resets the bits in the interrupt status >>> register, therefore subsequent calls to rtl8169_poll() won't call >>> rtl_rx() and rtl_tx() - chip interrupts are still disabled. >> >> Very interesting! Could this be the reason for the mysterious >> hangs & resets we experienced when enabling BQL for r8169? >> They happened more often with TSO/GSO enabled and several people >> attempted to fix those hangs unsuccessfully; it was later reverted >> and has been since then (#87cda7cb43). >> If this bug has been there "forever" it might be tempting to >> re-apply BQL and see what happens. Any chance you could give that >> a try? I'll gladly test patches, just like I'll run this one. >> >> cheers >> Holger > > Many drivers have buggy usage of napi_complete_done. > > Might even be worth forcing all network drivers to check the return > value. But fixing 150 broken drivers will be a nuisance. I had started doing that about a month ago in light of the ixbge ndo_poll_controller vs. napi problem, but have not had time to submit that series yet: https://github.com/ffainelli/linux/commits/napi-check feel free to piggy back on top of that series if you would like to address this. > > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h > index dc1d9ed33b31..c38bc66ffe74 100644 > --- a/include/linux/netdevice.h > +++ b/include/linux/netdevice.h > @@ -466,7 +466,8 @@ static inline bool napi_reschedule(struct napi_struct *napi) > return false; > } > > -bool napi_complete_done(struct napi_struct *n, int work_done); > +bool __must_check napi_complete_done(struct napi_struct *n, int work_done); > + > /** > * napi_complete - NAPI processing complete > * @n: NAPI context > -- Florian