From owner-freebsd-stable@FreeBSD.ORG Sat May 22 06:19:35 2010 Return-Path: Delivered-To: freebsd-stable@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id A94E6106564A for ; Sat, 22 May 2010 06:19:35 +0000 (UTC) (envelope-from rihad@mail.ru) Received: from wombat.diezmil.com (aa.81.b6.static.xlhost.com [207.182.129.170]) by mx1.freebsd.org (Postfix) with ESMTP id 609A18FC16 for ; Sat, 22 May 2010 06:19:35 +0000 (UTC) Received: from wombat (wombat [127.0.0.1]) by wombat.diezmil.com (8.14.2/8.14.2) with ESMTP id o4M5i9OU019303 for ; Sat, 22 May 2010 01:44:09 -0400 Date: Sat, 22 May 2010 01:44:09 -0400 From: rihad@mail.ru To: freebsd-stable@freebsd.org Message-ID: <6852241.21274507049390.JavaMail.root@wombat> In-Reply-To: <20090806125454.07fe563c.stas@FreeBSD.org> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Subject: Re: Re: em driver input errors X-BeenThere: freebsd-stable@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list Reply-To: rihad@mail.ru List-Id: Production branch of FreeBSD source code List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 22 May 2010 06:19:35 -0000 -- Stanislav Sedov wrote : On Wed, 5 Aug 2009 00:30:20 -0700 (PDT) alexpalias-bsdstable@yaho... mentioned: >> dev.em.0.rx_processing_limit=300 >> dev.em.1.rx_processing_limit=300 >> dev.em.2.rx_processing_limit=300 >> dev.em.3.rx_processing_limit=300 > This tunables only affects polling mode. Do you use polling with this > adapters or just standard interrupt-based mode? Why? In if_em.c, em_rxeof() is used both in legacy polling and in fast-interrupts / MSIX interrupts modes. I have the same problem of many Ierrs on input on em0 at a load of 500-600 mbps, and increasing rx_processing_limit (or setting it to -1) doesn't help at all. em0: Receive No Buffers = 4442943 em0: RX overruns = 83367 Next thing I'll try will be hw.em.rxd=4096. I don't think polling would help as it is not the interrupt load causing the drops (as evidenced by top -HS), but rather insufficiently large input buffers. -- This message was sent on behalf of rihad@mail.ru at openSubscriber.com http://www.opensubscriber.com/message/freebsd-stable@freebsd.org/12647494.html