RE: [PATCH] libxl: tooling expects wrong errno

[Paul Durrant <xadimgnik@xxxxxxxxx>]

> -----Original Message-----
> From: Ian Jackson <ian.jackson@xxxxxxxxxx>
> Sent: 15 June 2020 17:38
> To: Roger Pau Monne <roger.pau@xxxxxxxxxx>
> Cc: Grzegorz Uriasz <gorbak25@xxxxxxxxx>; Jan Beulich <jbeulich@xxxxxxxx>; 
> Andrew Cooper
> <Andrew.Cooper3@xxxxxxxxxx>; Kevin Tian <kevin.tian@xxxxxxxxx>; Paul Durrant 
> <paul@xxxxxxx>; Wei Liu
> <wl@xxxxxxx>; jakub@xxxxxxxxxxx; marmarek@xxxxxxxxxxxxxxxxxxxxxx; 
> j.nowak26@xxxxxxxxxxxxxxxxx; Anthony
> Perard <anthony.perard@xxxxxxxxxx>; xen-devel@xxxxxxxxxxxxxxxxxxxx; 
> contact@xxxxxxxxxxxx
> Subject: Re: [PATCH] libxl: tooling expects wrong errno
> 
> Roger Pau Monne writes ("Re: [PATCH] libxl: tooling expects wrong errno"):
> > On Mon, Jun 15, 2020 at 03:59:50PM +0100, Ian Jackson wrote:
> > > Grzegorz Uriasz writes ("[PATCH] libxl: tooling expects wrong errno"):
> > > > When iommu is not enabled for a given domain then pci passthrough
> > > > hypercalls such as xc_test_assign_device return EOPNOTSUPP.
> > > > The code responsible for this is in "iommu_do_domctl" inside
> > > > xen/drivers/passthrough/iommu.c
> > > > This patch fixes the error message reported by libxl when assigning
> > > > pci devices to domains without iommu.
> > > >
> > > > Signed-off-by: Grzegorz Uriasz <gorbak25@xxxxxxxxx>
> > > > Tested-by: Grzegorz Uriasz <gorbak25@xxxxxxxxx>
> > > > ---
> > > >  tools/libxl/libxl_pci.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/tools/libxl/libxl_pci.c b/tools/libxl/libxl_pci.c
> > > > index 957ff5c8e9..bc5843b137 100644
> > > > --- a/tools/libxl/libxl_pci.c
> > > > +++ b/tools/libxl/libxl_pci.c
> > > > @@ -1561,7 +1561,7 @@ void libxl__device_pci_add(libxl__egc *egc, 
> > > > uint32_t domid,
> > > >              LOGD(ERROR, domid,
> > > >                   "PCI device %04x:%02x:%02x.%u %s?",
> > > >                   pcidev->domain, pcidev->bus, pcidev->dev, 
> > > > pcidev->func,
> > > > -                 errno == ENOSYS ? "cannot be assigned - no IOMMU"
> > > > +                 errno == EOPNOTSUPP ? "cannot be assigned - no IOMMU"
> > > >                   : "already assigned to a different guest");
> > > >              goto out;
> > > >          }
> > >
> > > Thanks.  I have addressed some Xen IOMMU maintainers.  Can you confirm
> > > whether this is right ?
> >
> > Not an IOMMU maintainer myself, but I've taken a look at the code and
> > I think Grzegorz is right. iommu_do_domctl will return -EOPNOTSUPP if
> > the IOMMU is not enabled for the domain. Another option would be to
> > check for EBUSY (which will certainly be returned when the device is
> > busy) and log the error code with a message when it's different than
> > EBUSY?
> >
> > There are many possible error here, for example the device itself
> > might not be behind an IOMMU, in which case Xen will return -ENODEV at
> > least on the Intel case.
> 
> Thanks for the analysis.  So:
> 
> Acked-by: Ian Jackson <ian.jackson@xxxxxxxxxxxxx>
> 
> This would seem to be a backport candidate.  AFAICT check has been
> there, looking for ENOSYS, since this code was introduced in
>    826eb17271d3c647516d9944c47b0779afedea25
>    libxl: suppress device assignment to HVM guest when there is no IOMMU
> ?
> 
> But that commit has a Tested-by.  Maybe Xen changed its error return
> at some point ?
> 

Yes, it happened in 71e617a6b8f69849c70eda1b3c58f1ff6b244e5a
use is_iommu_enabled() where appropriate...

  Paul

> Ian.