Re: [PATCH] iommu: Flow ERR_PTR out from __iommu_domain_alloc()

[Jason Gunthorpe <jgg@nvidia.com>]

On Thu, Nov 02, 2023 at 08:22:13AM +0300, Dan Carpenter wrote:
> On Wed, Nov 01, 2023 at 10:09:18PM -0700, Jerry Snitselaar wrote:
> > On Thu, Nov 02, 2023 at 07:10:57AM +0300, Dan Carpenter wrote:
> > > On Wed, Nov 01, 2023 at 08:17:27PM -0300, Jason Gunthorpe wrote:
> > > >         /* The driver gave no guidance on what type to use, try the default */
> > > >         dom = __iommu_group_alloc_default_domain(group, iommu_def_domain_type);
> > > > -       if (dom)
> > > > +       if (!IS_ERR(dom))
> > > >                 return dom;
> > > >  
> > > >         /* Otherwise IDENTITY and DMA_FQ defaults will try DMA */
> > > >         if (iommu_def_domain_type == IOMMU_DOMAIN_DMA)
> > > > -               return ERR_PTR(-ENODEV);
> > > > +               return ERR_PTR(-EINVAL);
> > > >         dom = __iommu_group_alloc_default_domain(group, IOMMU_DOMAIN_DMA);
> > > > -       if (!dom)
> > > > +       if (IS_ERR(dom))
> > > >                 return dom;
> > > >  
> > > >         pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
> > > 
> > > Not related to your patch, but we should delete/move this warning as
> > > well.  It's on a success path.
> > > 
> > > regards,
> > > dan carpenter
> > > 
> > 
> > It seems worthwhile to note due to the performance implications, if
> > nothing else, of having to fall back to the dma domain type from the
> > two types mentioned in the comment. I'm not sure where would be a
> > better place since this is first spot where it knows that is going to
> > be the case.
> 
> Maybe it should be a pr_info() instead of a pr_warn()?  Maybe I'm just
> wrong.

TBH, I'm not entirely sure what it is for, it has been there a long
time.. I think it means that kernel config and command line options
are incompatible with this iommu driver

But at this moment there are very few drivers that don't support
IDENTITY (s390 notably, I suppose) and the DMA_FQ stuff is no longer
driver visible, so it probably never triggers..

Jason