From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: [PATCH] QStor SATA/RAID driver for 2.6.9-rc3 Date: Wed, 20 Oct 2004 16:44:32 +0100 Sender: linux-kernel-owner@vger.kernel.org Message-ID: <20041020154432.GE21985@infradead.org> References: <20041007150709.B12688@infradead.org> <4165624C.5060405@rtr.ca> <416565DB.4050006@pobox.com> <4165A45D.2090200@rtr.ca> <4165A766.1040104@pobox.com> <4165A85D.7080704@rtr.ca> <4165AB1B.8000204@pobox.com> <4165ACF8.8060208@rtr.ca> <20041007221537.A17712@infradead.org> <4165BD38.4020403@rtr.ca> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <4165BD38.4020403@rtr.ca> To: Mark Lord Cc: Christoph Hellwig , Jeff Garzik , Mark Lord , Linux Kernel , linux-scsi@vger.kernel.org List-Id: linux-scsi@vger.kernel.org On Thu, Oct 07, 2004 at 06:03:36PM -0400, Mark Lord wrote: > (Christoph Hellwig wrote: > > > > - the !dev case in qs_scsi_queuecomman can't happen > > Are you sure? > I have seen it occur immediately after hot-removal > of a drive. There have been other structural changes > since then, so perhaps it is no longer possible, > but I'd rather have the test there than have the > kernel ooops again. If you feel strongly about it, > then away it goes. Hmm. It's freed in ->slave_destroy and the scsi state model shouldn't allow new command submission long before. If you still see it happen please send a bugreport to linux-scsi. > > - never mess with eh_timeout from inside a driver > > Give us an interface for it, please. > In the meanwhile, gone! set scsi_device->timeout in ->slave_alloc to the value you want. > > - please don't implemente the HDIO_ ioctls, Jeff said this can > > be done via SG_IO > > SG_IO is incompatible with current user-mode toolsets. > Once that interface becomes more mature, and the distributions > gradually get updated with newer versions of the tools, > then the HDIO_ stuff can go (as per the comments in the source). > For now, it is essential for hdparm and smartmontools, among others. > > Alternatively, as Jeff has suggested, we may be able to implement > a generic HDIO_ mechanism in libata that re-issues the commands > through SG_IO (perhaps that is what you meant). Is that there now? So please fixup userspace. New hardware will require new system tools once in a while. > > - if ->info return a static string you can just store it into ->name > > So just nuke the _info() proc, and use .name = QS_DESC ? > Okay, done. Yes.