| Message ID | 1422316305-19216-3-git-send-email-joro@8bytes.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Joerg, Thanks for posting this! On Mon, Jan 26, 2015 at 11:51:32PM +0000, Joerg Roedel wrote: > From: Joerg Roedel <jroedel@suse.de> > > This allows to handle domains differently based on their > type in the future. An IOMMU driver can implement certain > optimizations for DMA-API domains for example. > > Signed-off-by: Joerg Roedel <jroedel@suse.de> > --- > drivers/iommu/iommu.c | 5 +++-- > include/linux/iommu.h | 11 ++++++++++- > 2 files changed, 13 insertions(+), 3 deletions(-) > > diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c > index 684efc0..ab24d77 100644 > --- a/drivers/iommu/iommu.c > +++ b/drivers/iommu/iommu.c > @@ -909,14 +909,15 @@ struct iommu_domain *iommu_domain_alloc(struct bus_type *bus) > ops = bus->iommu_ops; > > if (ops->domain_alloc) > - domain = ops->domain_alloc(); > + domain = ops->domain_alloc(IOMMU_DOMAIN_UNMANAGED); > else > domain = kzalloc(sizeof(*domain), GFP_KERNEL); > > if (!domain) > return NULL; > > - domain->ops = bus->iommu_ops; > + domain->ops = bus->iommu_ops; > + domain->type = IOMMU_DOMAIN_UNMANAGED; > > if (ops->domain_init && domain->ops->domain_init(domain)) > goto out_free; > diff --git a/include/linux/iommu.h b/include/linux/iommu.h > index 69d1d12..0b67f65 100644 > --- a/include/linux/iommu.h > +++ b/include/linux/iommu.h > @@ -51,7 +51,16 @@ struct iommu_domain_geometry { > bool force_aperture; /* DMA only allowed in mappable range? */ > }; > > +/* This are the possible domain-types */ > +enum iommu_domain_type { > + IOMMU_DOMAIN_DMA, /* Domain used for DMA-API */ > + IOMMU_DOMAIN_IDENTITY, /* Identity mapped domain */ What happens if somebody calls map or unmap on an identity-mapping domain? Can we catch that in the IOMMU core before calling the IOMMU driver? That also implies we need something extra to parameterise the attributes for the mapping (e.g. cacheable, read-only) and also potentially the address range. > + IOMMU_DOMAIN_UNMANAGED, /* Domain mappings are managed by a third party > + user (like KVM or VFIO) */ We already have the domain attributes (iommu_attr) to describe features of a domain. Is there really a need for this extra type, or can we extend the attribute set and allow for domain allocation with attributes? Will
Hi Will, On Wed, Jan 28, 2015 at 02:19:34PM +0000, Will Deacon wrote: > > +/* This are the possible domain-types */ > > +enum iommu_domain_type { > > + IOMMU_DOMAIN_DMA, /* Domain used for DMA-API */ > > + IOMMU_DOMAIN_IDENTITY, /* Identity mapped domain */ > > What happens if somebody calls map or unmap on an identity-mapping domain? > Can we catch that in the IOMMU core before calling the IOMMU driver? That > also implies we need something extra to parameterise the attributes for > the mapping (e.g. cacheable, read-only) and also potentially the address > range. The domain type is only used by the IOMMU drivers to do be able to enter special allocation paths (the AMD-Vi driver for example could allocate a special internal domain type for IOMMU_DOMAIN_DMA). But I think you are right, a couple of the current domain attributes we have could be moved here, turning the domain-type into a bit-field. So we could have bits for PAGING, CACHABLE, DMA_API and NESTING and build the domain types above from those bits. > > + IOMMU_DOMAIN_UNMANAGED, /* Domain mappings are managed by a third party > > + user (like KVM or VFIO) */ > > We already have the domain attributes (iommu_attr) to describe features > of a domain. Is there really a need for this extra type, or can we extend > the attribute set and allow for domain allocation with attributes? I don't think that domains attributed will become obsolete with this domain-type field. Some of the attributes should stay there for now, like all the PAMU specific stuff. But I agree, some could be moved over. Joerg
diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c index 684efc0..ab24d77 100644 --- a/drivers/iommu/iommu.c +++ b/drivers/iommu/iommu.c @@ -909,14 +909,15 @@ struct iommu_domain *iommu_domain_alloc(struct bus_type *bus) ops = bus->iommu_ops; if (ops->domain_alloc) - domain = ops->domain_alloc(); + domain = ops->domain_alloc(IOMMU_DOMAIN_UNMANAGED); else domain = kzalloc(sizeof(*domain), GFP_KERNEL); if (!domain) return NULL; - domain->ops = bus->iommu_ops; + domain->ops = bus->iommu_ops; + domain->type = IOMMU_DOMAIN_UNMANAGED; if (ops->domain_init && domain->ops->domain_init(domain)) goto out_free; diff --git a/include/linux/iommu.h b/include/linux/iommu.h index 69d1d12..0b67f65 100644 --- a/include/linux/iommu.h +++ b/include/linux/iommu.h @@ -51,7 +51,16 @@ struct iommu_domain_geometry { bool force_aperture; /* DMA only allowed in mappable range? */ }; +/* This are the possible domain-types */ +enum iommu_domain_type { + IOMMU_DOMAIN_DMA, /* Domain used for DMA-API */ + IOMMU_DOMAIN_IDENTITY, /* Identity mapped domain */ + IOMMU_DOMAIN_UNMANAGED, /* Domain mappings are managed by a third party + user (like KVM or VFIO) */ +}; + struct iommu_domain { + enum iommu_domain_type type; const struct iommu_ops *ops; void *priv; iommu_fault_handler_t handler; @@ -117,7 +126,7 @@ struct iommu_ops { void (*domain_destroy)(struct iommu_domain *domain); /* Domain allocation and freeing by the iommu driver */ - struct iommu_domain *(*domain_alloc)(void); + struct iommu_domain *(*domain_alloc)(enum iommu_domain_type); void (*domain_free)(struct iommu_domain *); int (*attach_dev)(struct iommu_domain *domain, struct device *dev);