From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Wed, 15 Aug 2001 07:03:48 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Wed, 15 Aug 2001 07:03:39 -0400 Received: from customers.imt.ru ([212.16.0.33]:28164 "HELO smtp.direct.ru") by vger.kernel.org with SMTP id ; Wed, 15 Aug 2001 07:03:24 -0400 Message-ID: <20010815035800.A24565@saw.sw.com.sg> Date: Wed, 15 Aug 2001 03:58:00 -0700 From: Andrey Savochkin To: Ime Smits Cc: linux-kernel@vger.kernel.org Subject: Re: Camino 2 (82815/82820) v2.4.x eth/sound related lockups In-Reply-To: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 0.93.2i In-Reply-To: ; from "Ime Smits" on Tue, Aug 14, 2001 at 02:01:31PM Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org All these wait_for_cmd_done timeout issues look very much like new (or previously unknown) timing constraints in new onboard chips. It isn't completely power management-related. Donald fixed one of such issues several months ago, which I primarily observed on i82559ER. Apparently, it's another one. The chip gets stuck and stops to process commands. For a lot of people inserting `udelay(1);' inside wait_for_cmd_done loop helped, not because it's right to make the delay, but because it changes timing and introduces minimal 1usec delay even if the chip is ready almost instantly :-( Best regards Andrey On Tue, Aug 14, 2001 at 02:01:31PM +0200, Ime Smits wrote: > > INDIVIDVVS VOCATVR Alan Cox DIE 14/8/2001 12:40 VERE SCRIPSIT: > > | Those are not so good. I was having similar problems on an i810 box with > | onboard eepro100 until I disabled the pm stuff in 2.4.8ac2, but you > | seem to be running that one > > Already figured that out. Oh, and I forget to mention that the same lockups > happen with all 2.4 versions I was able to find on my boxen, including > 2.4.0, -ac7, 2.4.2, 2.4.5, 2.4.5-ac3, -ac5, 2.4.6, 2.4.7, 2.4.8 and -ac2, > so this is not something introduced in recent kernels. With > 2.4.8-ac2 I played with enabling/disabling ACPI & PM stuff and also APIC > irq stuff as mentioned in the eepro100 thread. No go.