From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: [PATCH] libata: rewrite SCSI host scheme to be one per ATA host Date: Wed, 22 Apr 2009 12:48:05 -0400 Message-ID: <49EF4A45.8040901@garzik.org> References: <20090422090929.GA14928@havoc.gtf.org> <49EEE225.3010700@garzik.org> <49EF0A92.1070400@panasas.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <49EF0A92.1070400@panasas.com> Sender: linux-scsi-owner@vger.kernel.org To: Boaz Harrosh Cc: linux-ide@vger.kernel.org, linux-scsi@vger.kernel.org, LKML , Jens Axboe , Tejun Heo List-Id: linux-ide@vger.kernel.org Boaz Harrosh wrote: > On 04/22/2009 12:23 PM, Jeff Garzik wrote: >> Jeff Garzik wrote: >>> Currently, libata creates a Scsi_Host per port. This was originally >>> done to leverage SCSI's infrastructure to arbitrate among master/slave >>> devices, but is not needed for most modern SATA controllers. And I >>> _think_ it is not needed for master/slave if done properly, either. >> BTW note the above, with regards to the libata SCSI->block conversion. >> libata currently relies on SCSI for some amount of generic device >> arbitration, in several situations (see ->qc_defer, >> SCSI_MLQUEUE_.*_BUSY). libata expects SCSI to be intelligent and not >> starve devices, etc. >> >> >>> I was able to successfully boot the following patch on >>> AHCI/x86-64/Fedora. >>> >>> It may work with other controllers -- TRY AT YOUR OWN RISK. It will >>> probably fail for master/slave configurations, and SAS & PMP also >>> need looking at. It yielded this lsscsi output on my AHCI box: >>> >>> [0:0:0:0] disk ATA ST3500320AS SD15 /dev/sda >>> [0:2:0:0] disk ATA G.SKILL 128GB SS 02.1 /dev/sdb >>> [0:5:0:0] cd/dvd PIONEER BD-ROM BDC-202 1.04 /dev/sr0 >> For comparison, here is unmodified 2.6.30-rc3: >> >> [jgarzik@bd ~]$ lsscsi >> [0:0:0:0] disk ATA ST3500320AS SD15 /dev/sda >> [2:0:0:0] disk ATA G.SKILL 128GB SS 02.1 /dev/sdb >> [5:0:0:0] cd/dvd PIONEER BD-ROM BDC-202 1.04 /dev/sr0 >> > > Could the master/slave be simply solved by emulating a SCSI LUN > for example below is my machine today: > []$ lsscsi > [0:0:0:0] disk ATA ST3160023A 3.01 /dev/sda > [1:0:0:0] cd/dvd _NEC DVD_RW ND-3550A 1.05 /dev/sr0 > [4:0:0:0] disk ATA WDC WD1600JS-60M 10.0 /dev/sdb > > the /dev/sda and /dev/sr0 share a master/slave wide cable (sdb is sata) > > it could be made to scan as: > []$ lsscsi > [0:0:0:0] disk ATA ST3160023A 3.01 /dev/sda > [0:0:0:1] cd/dvd _NEC DVD_RW ND-3550A 1.05 /dev/sr0 > [1:0:0:0] disk ATA WDC WD1600JS-60M 10.0 /dev/sdb > > So we need to emulate the REPORT_LUN (or what ever else) to return > two LUNs. Or do you want to report a separate target for the master/slave? Mapping master/slave is not difficult -- each should be a separate target, just like with parallel SCSI. The issue with master/slave and simplex is guaranteeing that only _one_ command may be executing at a time, for a given set of targets, i.e. only one command per master/slave pair, only one command per pair of simplex ports (== 4 ATA devices max). Originally this was done by setting can_queue==1, cmd_per_lun==1, and assigning each master/slave pair to a different Scsi_Host. Jeff