From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754129Ab1JLUlh (ORCPT ); Wed, 12 Oct 2011 16:41:37 -0400 Received: from cantor2.suse.de ([195.135.220.15]:41573 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752171Ab1JLUlf (ORCPT ); Wed, 12 Oct 2011 16:41:35 -0400 Date: Wed, 12 Oct 2011 14:36:14 -0600 From: Greg KH To: Mihai Moldovan Cc: LKML Subject: Re: [Patch] Increase USBFS Bulk Transfer size Message-ID: <20111012203614.GB14572@suse.de> References: <20111012141724.GA17414@suse.de> <4E95D5C6.8040404@ionic.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <4E95D5C6.8040404@ionic.de> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Oct 12, 2011 at 08:00:38PM +0200, Mihai Moldovan wrote: > Hi Greg, > > * On 12.10.2011 04:17 PM, Greg KH wrote: > > As stated before, this patch is not acceptable. Please work to figure > > out the real reason for your device problems here, this is not the > > correct solution at all. > > I've gone through this whole (and previous) thread, but couldn't find a > real argument why this is so wrong. Because it does nothing except increase kernel memory pressure. Why not increase it to 1Gb while we are at it, it realistically makes no difference. And so, because it makes no difference, we should not change the existing value as there seems to be something else wrong with what is going on here in the userspace code that can't handle smaller buffer sizes. Within the kernel, and the host controller driver, the memory is split up into much smaller pieces anyway, so there should not be any difference at all, if you write your userspace code correctly, that the USB device would see anything different with a bigger (or even smaller) usbfs buffer size. > So far everybody has argued that it's 'wrong' and may break older user > code. The latter argument even is wrong, as drivers not requiring a > higher bulk transfer size just aren't affected. > > This being said, I agree that allocating more memory than needed is > wasting memory and bad, if it can be avoided. On the other hand, we're > talking about very few devices here and not several ten of MB system > memory being wasted by all bulk transfers in total. > I basically see two cases: > - systems with a few MB of RAM. I highly doubt those use usbfs anyway > (usually other stuff like usb-storage) Not true at all, I know of some using usbfs quite well, and have been since the 2.2 kernel days. greg k-h