From owner-freebsd-net Fri Oct 11 17:17:54 2002 Delivered-To: freebsd-net@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id D639237B404; Fri, 11 Oct 2002 17:17:49 -0700 (PDT) Received: from gull.mail.pas.earthlink.net (gull.mail.pas.earthlink.net [207.217.120.84]) by mx1.FreeBSD.org (Postfix) with ESMTP id 303C643E9C; Fri, 11 Oct 2002 17:17:49 -0700 (PDT) (envelope-from tlambert2@mindspring.com) Received: from pool0285.cvx40-bradley.dialup.earthlink.net ([216.244.43.30] helo=mindspring.com) by gull.mail.pas.earthlink.net with esmtp (Exim 3.33 #1) id 1809yS-0003T8-00; Fri, 11 Oct 2002 17:17:44 -0700 Message-ID: <3DA769DC.58E6BBF0@mindspring.com> Date: Fri, 11 Oct 2002 17:16:28 -0700 From: Terry Lambert X-Mailer: Mozilla 4.79 [en] (Win98; U) X-Accept-Language: en MIME-Version: 1.0 To: silby@silby.com Cc: csmith@its.uq.edu.au, hardware@freebsd.org, net@freebsd.org Subject: [PATCH: if_ti] Re: High interrupt load on firewalls Content-Type: multipart/mixed; boundary="------------0C5A8C380344382790784831" Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org This is a multi-part message in MIME format. --------------0C5A8C380344382790784831 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Mike Silbersack wrote: > On Wed, 9 Oct 2002, Christopher Smith wrote: > > The rule processing can't be done on the other CPU, can it ? Am I right in > > saying that at this point in time, buying a dual CPU (vs single CPU) machine > > for firewalling with FreeBSD is just a waste of money ? > > Even if it could be done, I doubt that would be the most cost effectively > solution to the problem. Try out different NICs, then move on to kernel > profiling if it's still a problem. > > Luigi can probably comment more on this, but one thing which comes to mind > is that the if_ti driver might not be updated to use the new m_getcl > function Luigi added. Luigi claimed a 10% increase in forwarding speed > for drivers using it, I believe. :) Please find attached a patch for the if_ti to support DEVICE_POLLING (the patch is against -current). -- Terry --------------0C5A8C380344382790784831 Content-Type: text/plain; charset=us-ascii; name="tipoll.diff" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="tipoll.diff" Index: if_ti.c =================================================================== RCS file: /cvs/src/sys/pci/if_ti.c,v retrieving revision 1.63 diff -c -r1.63 if_ti.c *** if_ti.c 24 Aug 2002 00:02:03 -0000 1.63 --- if_ti.c 11 Oct 2002 20:09:48 -0000 *************** *** 2520,2525 **** --- 2520,2533 ---- u_int16_t vlan_tag = 0; int have_tag = 0; + #ifdef DEVICE_POLLING + if (ifp->if_flags & IFF_POLLING) { + if (sc->rxcycles <= 0) + break; + sc->rxcycles--; + } + #endif /* DEVICE_POLLING */ + cur_rx = &sc->ti_rdata->ti_rx_return_ring[sc->ti_rx_saved_considx]; rxidx = cur_rx->ti_idx; *************** *** 2678,2683 **** --- 2686,2731 ---- return; } + #ifdef DEVICE_POLLING + static poll_handler_t ti_poll; + + static void + ti_poll(struct ifnet *ifp, enum poll_cmd cmd, int count) + { + struct ti_softc *sc = ifp->if_softc; + + TI_LOCK(sc); + if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */ + CSR_WRITE_4(sc, TI_MB_HOSTINTR, 0); + goto done; + } + + /* + * Before returning to intr mode we must make sure that all + * possible pending sources of interrupts have been served. + * In practice this means run to completion the *eof routines, + * and then call the interrupt routine + */ + sc->rxcycles = count; + if (ifp->if_snd.ifq_head != NULL) + ti_start(ifp); + + if (sc->rxcycles > 0 || cmd == POLL_AND_CHECK_STATUS) { + if (ifp->if_flags & IFF_RUNNING) { + /* Check RX return ring producer/consumer */ + ti_rxeof(sc); + + /* Check TX ring producer/consumer */ + ti_txeof(sc); + } + ti_handle_events(sc); + } + done: + TI_UNLOCK(sc); + return; + } + #endif /* DEVICE_POLLING */ + static void ti_intr(xsc) void *xsc; *************** *** 2688,2693 **** --- 2736,2749 ---- sc = xsc; TI_LOCK(sc); ifp = &sc->arpcom.ac_if; + #ifdef DEVICE_POLLING + if (ifp->if_flags & IFF_POLLING) + goto done; + if (ether_poll_register(ti_poll, ifp)) { /* ok, disable interrupts */ + CSR_WRITE_4(sc, TI_MB_HOSTINTR, 1); + goto done; + } + #endif /* DEVICE_POLLING */ /*#ifdef notdef*/ /* Avoid this for now -- checking this register is expensive. */ *************** *** 2716,2722 **** if (ifp->if_flags & IFF_RUNNING && ifp->if_snd.ifq_head != NULL) ti_start(ifp); ! TI_UNLOCK(sc); return; --- 2772,2780 ---- if (ifp->if_flags & IFF_RUNNING && ifp->if_snd.ifq_head != NULL) ti_start(ifp); ! #ifdef DEVICE_POLLING ! done: ! #endif /* DEVICE_POLLING */ TI_UNLOCK(sc); return; *************** *** 2999,3004 **** --- 3057,3071 ---- TI_DO_CMD(TI_CMD_HOST_STATE, TI_CMD_CODE_STACK_UP, 0); /* Enable host interrupts. */ + #ifdef DEVICE_POLLING + /* + * ... only enable interrupts if we are not polling, make sure + * they are off otherwise. + */ + if (ifp->if_flags & IFF_POLLING) + CSR_WRITE_4(sc, TI_MB_HOSTINTR, 1); + else + #endif /* DEVICE_POLLING */ CSR_WRITE_4(sc, TI_MB_HOSTINTR, 0); ifp->if_flags |= IFF_RUNNING; *************** *** 3613,3618 **** --- 3680,3688 ---- TI_LOCK(sc); ifp = &sc->arpcom.ac_if; + #ifdef DEVICE_POLLING + ether_poll_deregister(ifp); + #endif /* Disable host interrupts. */ CSR_WRITE_4(sc, TI_MB_HOSTINTR, 1); Index: if_tireg.h =================================================================== RCS file: /cvs/src/sys/pci/if_tireg.h,v retrieving revision 1.25 diff -c -r1.25 if_tireg.h *** if_tireg.h 26 Jun 2002 03:34:52 -0000 1.25 --- if_tireg.h 11 Oct 2002 19:25:04 -0000 *************** *** 1026,1031 **** --- 1026,1034 ---- struct mtx ti_mtx; ti_flag_vals ti_flags; dev_t dev; + #ifdef DEVICE_POLLING + int rxcycles; + #endif }; #define TI_LOCK(_sc) mtx_lock(&(_sc)->ti_mtx) --------------0C5A8C380344382790784831-- To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message