Is anyone working on SCSI passthrough using the ''SCSI Generic'' support under Linux, eg /dev/sgX? This is how VMWare allows a VM to use SCSI devices. Thanks James _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> > Is anyone working on SCSI passthrough using the ''SCSI Generic'' support > under Linux, eg /dev/sgX? This is how VMWare allows a VM to use SCSI > devices. >I assume that the lack of response means ''no''. I''ve just been having a look through the documentation for ''SCSI Generic'' under Linux, and it looks pretty straightforward... It could be done much like the existing blkdev interface, the ''input'' half of the sg_io_hdr (or something like it), would go on the ring as the request, and the ''output'' half, would come back on the ring as the response. Buffers would be passed like the block interface does now, although I''d probably prefer that we allow single byte aligned rather than 512 byte aligned buffers as the block device currently does, as Windows tends to use the former a bit. I''m not sure if any other IOCTL''s apart from SG_IO (which is the equivalent of a write + read operation anyway) would be required as part of normal IO, they could just be performed at initialisation and written to xenstore. This would allow single SCSI devices or pretty much any type to be exported to domains. I''m interested in writing the Linux backend, and the Windows front end as I''m pretty desperate for tape drive passthrough... Comments? Thanks James _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
There are some folks (from Fujitsu? I think?) who are working on front/back-based SCSI passthrough. Patches have been submitted several times but not merged yet. They''ll be in the list archives... I don''t think they''ve done a Windows frontend though ;-) Cheers, Mark On Tuesday 04 December 2007, James Harper wrote:> > Is anyone working on SCSI passthrough using the ''SCSI Generic'' support > > under Linux, eg /dev/sgX? This is how VMWare allows a VM to use SCSI > > devices. > > I assume that the lack of response means ''no''. > > I''ve just been having a look through the documentation for ''SCSI > Generic'' under Linux, and it looks pretty straightforward... > > It could be done much like the existing blkdev interface, the ''input'' > half of the sg_io_hdr (or something like it), would go on the ring as > the request, and the ''output'' half, would come back on the ring as the > response. Buffers would be passed like the block interface does now, > although I''d probably prefer that we allow single byte aligned rather > than 512 byte aligned buffers as the block device currently does, as > Windows tends to use the former a bit. I''m not sure if any other IOCTL''s > apart from SG_IO (which is the equivalent of a write + read operation > anyway) would be required as part of normal IO, they could just be > performed at initialisation and written to xenstore. > > This would allow single SCSI devices or pretty much any type to be > exported to domains. > > I''m interested in writing the Linux backend, and the Windows front end > as I''m pretty desperate for tape drive passthrough... > > Comments? > > Thanks > > James > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel-- Dave: Just a question. What use is a unicyle with no seat? And no pedals! Mark: To answer a question with a question: What use is a skateboard? Dave: Skateboards have wheels. Mark: My wheel has a wheel! _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> > There are some folks (from Fujitsu? I think?) who are working on > front/back-based SCSI passthrough. Patches have been submittedseveral> times > but not merged yet. They''ll be in the list archives... > > I don''t think they''ve done a Windows frontend though ;-) >I''ve seen those patches posted on the list and have had a look through them, but I think they are passing through the whole scsi bus attached to a single HBA. I only want to pass through a single device. James _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hi. We are not developing SCSI of Windows. We have no plan to depvelop Windows SCSI pass-through currently. Our SCSI pass-through doesn''t use SCSI Generic. Does using SCSI Generic in backend mean user-space access ? Mark Williamson san wrote:---------------------- Sent: Tue, 4 Dec 2007 02:12:21 +0000 Subject: Re: [Xen-devel] SCSI passthrough using /dev/sgX> There are some folks (from Fujitsu? I think?) who are working on > front/back-based SCSI passthrough. Patches have been submitted several times > but not merged yet. They''ll be in the list archives... > > I don''t think they''ve done a Windows frontend though ;-) > > Cheers, > Mark > > On Tuesday 04 December 2007, James Harper wrote: > > > Is anyone working on SCSI passthrough using the ''SCSI Generic'' support > > > under Linux, eg /dev/sgX? This is how VMWare allows a VM to use SCSI > > > devices. > > > > I assume that the lack of response means ''no''. > > > > I''ve just been having a look through the documentation for ''SCSI > > Generic'' under Linux, and it looks pretty straightforward... > > > > It could be done much like the existing blkdev interface, the ''input'' > > half of the sg_io_hdr (or something like it), would go on the ring as > > the request, and the ''output'' half, would come back on the ring as the > > response. Buffers would be passed like the block interface does now, > > although I''d probably prefer that we allow single byte aligned rather > > than 512 byte aligned buffers as the block device currently does, as > > Windows tends to use the former a bit. I''m not sure if any other IOCTL''s > > apart from SG_IO (which is the equivalent of a write + read operation > > anyway) would be required as part of normal IO, they could just be > > performed at initialisation and written to xenstore. > > > > This would allow single SCSI devices or pretty much any type to be > > exported to domains. > > > > I''m interested in writing the Linux backend, and the Windows front end > > as I''m pretty desperate for tape drive passthrough... > > > > Comments? > > > > Thanks > > > > James > > > > > > _______________________________________________ > > Xen-devel mailing list > > Xen-devel@lists.xensource.com > > http://lists.xensource.com/xen-devel > > > > -- > Dave: Just a question. What use is a unicyle with no seat? And no pedals! > Mark: To answer a question with a question: What use is a skateboard? > Dave: Skateboards have wheels. > Mark: My wheel has a wheel! > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> We are not developing SCSI of Windows. > We have no plan to depvelop Windows SCSI pass-through currently. > > Our SCSI pass-through doesn''t use SCSI Generic.Are you planning on allowing the export of only a single device, eg if the SCSI bus was like this: #0 - SCSI Disk #1 - SCSI Disk #5 - SCSI CDROM #6 - SCSI Tape #7 - SCSI HBA Could I export #5 to one domain, #6 to another domain, and use #0 and #1 in Dom0? Maybe your code already can do this?> Does using SCSI Generic in backend mean user-space access ?/dev/sgX is the SCSI Generic interface, and it is available in userspace. Thanks James _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
James-san.> Are you planning on allowing the export of only a single device, eg if > the SCSI bus was like this: > > #0 - SCSI Disk > #1 - SCSI Disk > #5 - SCSI CDROM > #6 - SCSI Tape > #7 - SCSI HBA > > Could I export #5 to one domain, #6 to another domain, and use #0 and #1 > in Dom0? Maybe your code already can do this? >We also thought about singleLUN in the following plan. but, It has not existed yet.> > Does using SCSI Generic in backend mean user-space access ? > > /dev/sgX is the SCSI Generic interface, and it is available in > userspace. >I am interested in the use of user-space. Before, the following patches posted... http://lists.xensource.com/archives/html/xen-devel/2007-01/msg00022.html Thanks. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
hi does SCSI passthrough need hardware support ,such as VT-d, or not?if there are two device and two VMs, does it need VT-d support, or not ? can it run in PVmode or have to run in HVM mode? Thanks in advance Mark Williamson 写道:> There are some folks (from Fujitsu? I think?) who are working on > front/back-based SCSI passthrough. Patches have been submitted several times > but not merged yet. They''ll be in the list archives... > > I don''t think they''ve done a Windows frontend though ;-) > > Cheers, > Mark > > On Tuesday 04 December 2007, James Harper wrote: > >>> Is anyone working on SCSI passthrough using the ''SCSI Generic'' support >>> under Linux, eg /dev/sgX? This is how VMWare allows a VM to use SCSI >>> devices. >>> >> I assume that the lack of response means ''no''. >> >> I''ve just been having a look through the documentation for ''SCSI >> Generic'' under Linux, and it looks pretty straightforward... >> >> It could be done much like the existing blkdev interface, the ''input'' >> half of the sg_io_hdr (or something like it), would go on the ring as >> the request, and the ''output'' half, would come back on the ring as the >> response. Buffers would be passed like the block interface does now, >> although I''d probably prefer that we allow single byte aligned rather >> than 512 byte aligned buffers as the block device currently does, as >> Windows tends to use the former a bit. I''m not sure if any other IOCTL''s >> apart from SG_IO (which is the equivalent of a write + read operation >> anyway) would be required as part of normal IO, they could just be >> performed at initialisation and written to xenstore. >> >> This would allow single SCSI devices or pretty much any type to be >> exported to domains. >> >> I''m interested in writing the Linux backend, and the Windows front end >> as I''m pretty desperate for tape drive passthrough... >> >> Comments? >> >> Thanks >> >> James >> >> >> _______________________________________________ >> Xen-devel mailing list >> Xen-devel@lists.xensource.com >> http://lists.xensource.com/xen-devel >> > > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> does SCSI passthrough need hardware support ,such as VT-d, or not?if > there are two device and two VMs, does it need VT-d support, or not ? > can it run in PVmode or have to run in HVM mode?The SCSI passthrough code that I was talking about is a paravirtualised front / back driver, so it doesn''t use VT-d. I don''t think it is currently supported under HVM guests. Cheers, Mark> Thanks in advance > > Mark Williamson 写道: > > There are some folks (from Fujitsu? I think?) who are working on > > front/back-based SCSI passthrough. Patches have been submitted several > > times but not merged yet. They''ll be in the list archives... > > > > I don''t think they''ve done a Windows frontend though ;-) > > > > Cheers, > > Mark > > > > On Tuesday 04 December 2007, James Harper wrote: > >>> Is anyone working on SCSI passthrough using the ''SCSI Generic'' support > >>> under Linux, eg /dev/sgX? This is how VMWare allows a VM to use SCSI > >>> devices. > >> > >> I assume that the lack of response means ''no''. > >> > >> I''ve just been having a look through the documentation for ''SCSI > >> Generic'' under Linux, and it looks pretty straightforward... > >> > >> It could be done much like the existing blkdev interface, the ''input'' > >> half of the sg_io_hdr (or something like it), would go on the ring as > >> the request, and the ''output'' half, would come back on the ring as the > >> response. Buffers would be passed like the block interface does now, > >> although I''d probably prefer that we allow single byte aligned rather > >> than 512 byte aligned buffers as the block device currently does, as > >> Windows tends to use the former a bit. I''m not sure if any other IOCTL''s > >> apart from SG_IO (which is the equivalent of a write + read operation > >> anyway) would be required as part of normal IO, they could just be > >> performed at initialisation and written to xenstore. > >> > >> This would allow single SCSI devices or pretty much any type to be > >> exported to domains. > >> > >> I''m interested in writing the Linux backend, and the Windows front end > >> as I''m pretty desperate for tape drive passthrough... > >> > >> Comments? > >> > >> Thanks > >> > >> James > >> > >> > >> _______________________________________________ > >> Xen-devel mailing list > >> Xen-devel@lists.xensource.com > >> http://lists.xensource.com/xen-devel-- Dave: Just a question. What use is a unicyle with no seat? And no pedals! Mark: To answer a question with a question: What use is a skateboard? Dave: Skateboards have wheels. Mark: My wheel has a wheel! _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel