Re: [Qemu-devel] [PATCH 2/3] vhost-scsi: new device supporting the tcm_vhost Linux kernel module

[Michael S. Tsirkin]

On Wed, Mar 27, 2013 at 11:28:56PM -0700, Nicholas A. Bellinger wrote:
> On Thu, 2013-03-28 at 08:19 +0200, Michael S. Tsirkin wrote:
> > On Wed, Mar 27, 2013 at 04:24:22PM -0700, Nicholas A. Bellinger wrote:
> > > On Thu, 2013-03-28 at 00:53 +0200, Michael S. Tsirkin wrote:
> > > > On Thu, Mar 28, 2013 at 12:50:21AM +0200, Michael S. Tsirkin wrote:
> > > > > On Wed, Mar 27, 2013 at 03:45:42PM -0700, Nicholas A. Bellinger wrote:
> > > > > > On Thu, 2013-03-28 at 00:28 +0200, Michael S. Tsirkin wrote:
> > > > > > > On Wed, Mar 27, 2013 at 09:59:45PM +0000, Nicholas A. Bellinger 
> > > > > > > wrote:
> > > > > > > > From: Paolo Bonzini <address@hidden>
> > > > > > > > 
> > > > > > > > The WWPN specified in configfs is passed to "-device 
> > > > > > > > vhost-scsi-pci".
> > > > > > > > The tgpt field of the SET_ENDPOINT ioctl is obsolete now, so it 
> > > > > > > > is not
> > > > > > > > available from the QEMU command-line.  Instead, I hardcode it 
> > > > > > > > to zero.
> > > > > > > > 
> > > > > > > > Changes in V4:
> > > > > > > >    - Set event_idx=off by default (nab, thanks asias)
> > > > > > > 
> > > > > > > Why? What's going on here?
> > > > > > > 
> > > > > > 
> > > > > > Not disabling event_idx by default, or disabling from the command 
> > > > > > line
> > > > > > ends up resulting in ->handle_kick() not getting called for 
> > > > > > subsequent
> > > > > > commands..
> > > > > > 
> > > > > > I spent some time trying to track this down recently with no luck, 
> > > > > > and
> > > > > > AFAICT it's always been required in order for vhost-scsi to 
> > > > > > function.
> > > > > 
> > > > > Hmm this is a bug in kernel then. A better work-around is
> > > > > to disable EVENT_IDX in kernel. Let's do it for 3.9?
> > > > 
> > > > But before we do, can you check that SET_FEATURES is
> > > > called with this bit set if you enable event_idx?
> > > > If not that's your bug then ...
> > > > 
> > > 
> > > Ok, SET_FEATURES is currently not setting any bits at all based upon
> > > vhost_dev->features, so it looks like a vhost-scsi-pci bug..
> > > 
> > > 
> > 
> > If you call SET_FEATURES, does it start working even without
> > disabling EVENT_IDX in kernel?
> > 
> 
> Unfortunately, no.  The same strangeness persists unless event_idx=off
> is passed, or EVENT_IDX is disabled in the kernel and stripped off with
> the vhost_scsi_get_features() patch.
> 


I think it's important to fix it (usually exposes other bugs)
but I agree, lets get it working without first.