| Message ID | 20170321172137.GF30948@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 21/03/17 17:21, Will Deacon wrote: > On Tue, Mar 21, 2017 at 04:45:27PM +0100, Joerg Roedel wrote: >> On Fri, Mar 10, 2017 at 08:49:36PM +0000, Will Deacon wrote: >>> @@ -1014,8 +1027,8 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) >>> * IOMMU driver. >>> */ >>> if (!group->default_domain) { >>> - group->default_domain = __iommu_domain_alloc(dev->bus, >>> - IOMMU_DOMAIN_DMA); >>> + group->default_domain = >>> + __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >> >> It would be good to have a fall-back here if we are talking to an IOMMU >> driver that uses default domains, but does not support identity-mapped >> domains (yet). Exynos and Rockchip IOMMU drivers seem to fall into this >> category. A dev_warn() also makes sense in case allocating a identity >> domain fails. > > Sure, something like the diff below? > > Will > > --->8 > > > diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c > index 42a842e3f95f..f787626a745d 100644 > --- a/drivers/iommu/iommu.c > +++ b/drivers/iommu/iommu.c > @@ -1027,10 +1027,19 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) > * IOMMU driver. > */ > if (!group->default_domain) { > - group->default_domain = > - __iommu_domain_alloc(dev->bus, iommu_def_domain_type); > + struct iommu_domain *dom; > + > + dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type); > + if (!dom) { > + dev_warn(dev, > + "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA", > + iommu_def_domain_type); Conversely, that's going to be noisy if iommu_def_domain_type was IOMMU_DOMAIN_DMA to begin with. I think it makes sense to warn if the user asked for a specific default domain type on the command line and that didn't work, but maybe not to bother otherwise. Plus, if they asked for passthrough, then not allocating a default domain at all is probably closer to the desired result than installing a DMA ops domain would be. Robin. > + dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA); > + } > + > + group->default_domain = dom; > if (!group->domain) > - group->domain = group->default_domain; > + group->domain = dom; > } > > ret = iommu_group_add_device(group, dev); >
On Tue, Mar 21, 2017 at 05:46:29PM +0000, Robin Murphy wrote: > On 21/03/17 17:21, Will Deacon wrote: > > On Tue, Mar 21, 2017 at 04:45:27PM +0100, Joerg Roedel wrote: > >> On Fri, Mar 10, 2017 at 08:49:36PM +0000, Will Deacon wrote: > >>> @@ -1014,8 +1027,8 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) > >>> * IOMMU driver. > >>> */ > >>> if (!group->default_domain) { > >>> - group->default_domain = __iommu_domain_alloc(dev->bus, > >>> - IOMMU_DOMAIN_DMA); > >>> + group->default_domain = > >>> + __iommu_domain_alloc(dev->bus, iommu_def_domain_type); > >> > >> It would be good to have a fall-back here if we are talking to an IOMMU > >> driver that uses default domains, but does not support identity-mapped > >> domains (yet). Exynos and Rockchip IOMMU drivers seem to fall into this > >> category. A dev_warn() also makes sense in case allocating a identity > >> domain fails. > > > > Sure, something like the diff below? > > > > Will > > > > --->8 > > > > > > diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c > > index 42a842e3f95f..f787626a745d 100644 > > --- a/drivers/iommu/iommu.c > > +++ b/drivers/iommu/iommu.c > > @@ -1027,10 +1027,19 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) > > * IOMMU driver. > > */ > > if (!group->default_domain) { > > - group->default_domain = > > - __iommu_domain_alloc(dev->bus, iommu_def_domain_type); > > + struct iommu_domain *dom; > > + > > + dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type); > > + if (!dom) { > > + dev_warn(dev, > > + "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA", > > + iommu_def_domain_type); > > Conversely, that's going to be noisy if iommu_def_domain_type was > IOMMU_DOMAIN_DMA to begin with. I think it makes sense to warn if the > user asked for a specific default domain type on the command line and > that didn't work, but maybe not to bother otherwise. Plus, if they asked > for passthrough, then not allocating a default domain at all is probably > closer to the desired result than installing a DMA ops domain would be. You're right -- I'll hack this about to check if the default type isn't DOMAIN_DMA before warning about the allocation failure. Cheers, Will
On Tue, Mar 21, 2017 at 05:21:37PM +0000, Will Deacon wrote: > On Tue, Mar 21, 2017 at 04:45:27PM +0100, Joerg Roedel wrote: > > On Fri, Mar 10, 2017 at 08:49:36PM +0000, Will Deacon wrote: > > > @@ -1014,8 +1027,8 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) > > > * IOMMU driver. > > > */ > > > if (!group->default_domain) { > > > - group->default_domain = __iommu_domain_alloc(dev->bus, > > > - IOMMU_DOMAIN_DMA); > > > + group->default_domain = > > > + __iommu_domain_alloc(dev->bus, iommu_def_domain_type); > > > > It would be good to have a fall-back here if we are talking to an IOMMU > > driver that uses default domains, but does not support identity-mapped > > domains (yet). Exynos and Rockchip IOMMU drivers seem to fall into this > > category. A dev_warn() also makes sense in case allocating a identity > > domain fails. > > Sure, something like the diff below? Yes, this looks good. Thanks, Joerg
Hi, On 3/21/2017 11:47 PM, Will Deacon wrote: > On Tue, Mar 21, 2017 at 05:46:29PM +0000, Robin Murphy wrote: >> On 21/03/17 17:21, Will Deacon wrote: >>> On Tue, Mar 21, 2017 at 04:45:27PM +0100, Joerg Roedel wrote: >>>> On Fri, Mar 10, 2017 at 08:49:36PM +0000, Will Deacon wrote: >>>>> @@ -1014,8 +1027,8 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) >>>>> * IOMMU driver. >>>>> */ >>>>> if (!group->default_domain) { >>>>> - group->default_domain = __iommu_domain_alloc(dev->bus, >>>>> - IOMMU_DOMAIN_DMA); >>>>> + group->default_domain = >>>>> + __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >>>> >>>> It would be good to have a fall-back here if we are talking to an IOMMU >>>> driver that uses default domains, but does not support identity-mapped >>>> domains (yet). Exynos and Rockchip IOMMU drivers seem to fall into this >>>> category. A dev_warn() also makes sense in case allocating a identity >>>> domain fails. >>> >>> Sure, something like the diff below? >>> >>> Will >>> >>> --->8 >>> >>> >>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c >>> index 42a842e3f95f..f787626a745d 100644 >>> --- a/drivers/iommu/iommu.c >>> +++ b/drivers/iommu/iommu.c >>> @@ -1027,10 +1027,19 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) >>> * IOMMU driver. >>> */ >>> if (!group->default_domain) { >>> - group->default_domain = >>> - __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >>> + struct iommu_domain *dom; >>> + >>> + dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >>> + if (!dom) { >>> + dev_warn(dev, >>> + "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA", >>> + iommu_def_domain_type); >> >> Conversely, that's going to be noisy if iommu_def_domain_type was >> IOMMU_DOMAIN_DMA to begin with. I think it makes sense to warn if the >> user asked for a specific default domain type on the command line and >> that didn't work, but maybe not to bother otherwise. Plus, if they asked >> for passthrough, then not allocating a default domain at all is probably >> closer to the desired result than installing a DMA ops domain would be. > > You're right -- I'll hack this about to check if the default type isn't > DOMAIN_DMA before warning about the allocation failure. if some master devices want 'IDENTITY_DOMAIN' as default (because those devices do not want any iommu resources to be used/dma_ops to be set) and some 'DMA_DOMAIN' as default, then should the default be 'DMA_DOMAIN' and then masters needing IDENTITY_DOMAIN explicitly do an detach_dev later. This [1] was adding the support for detach_dev of the default DMA_DOMAINs. [1] https://patchwork.codeaurora.org/patch/164933/ Regards, Sricharan
Hi, Replying again, as there was some issue with mailer last time. On 3/21/2017 11:47 PM, Will Deacon wrote: > On Tue, Mar 21, 2017 at 05:46:29PM +0000, Robin Murphy wrote: >> On 21/03/17 17:21, Will Deacon wrote: >>> On Tue, Mar 21, 2017 at 04:45:27PM +0100, Joerg Roedel wrote: >>>> On Fri, Mar 10, 2017 at 08:49:36PM +0000, Will Deacon wrote: >>>>> @@ -1014,8 +1027,8 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) >>>>> * IOMMU driver. >>>>> */ >>>>> if (!group->default_domain) { >>>>> - group->default_domain = __iommu_domain_alloc(dev->bus, >>>>> - IOMMU_DOMAIN_DMA); >>>>> + group->default_domain = >>>>> + __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >>>> >>>> It would be good to have a fall-back here if we are talking to an IOMMU >>>> driver that uses default domains, but does not support identity-mapped >>>> domains (yet). Exynos and Rockchip IOMMU drivers seem to fall into this >>>> category. A dev_warn() also makes sense in case allocating a identity >>>> domain fails. >>> >>> Sure, something like the diff below? >>> >>> Will >>> >>> --->8 >>> >>> >>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c >>> index 42a842e3f95f..f787626a745d 100644 >>> --- a/drivers/iommu/iommu.c >>> +++ b/drivers/iommu/iommu.c >>> @@ -1027,10 +1027,19 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) >>> * IOMMU driver. >>> */ >>> if (!group->default_domain) { >>> - group->default_domain = >>> - __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >>> + struct iommu_domain *dom; >>> + >>> + dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type); >>> + if (!dom) { >>> + dev_warn(dev, >>> + "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA", >>> + iommu_def_domain_type); >> >> Conversely, that's going to be noisy if iommu_def_domain_type was >> IOMMU_DOMAIN_DMA to begin with. I think it makes sense to warn if the >> user asked for a specific default domain type on the command line and >> that didn't work, but maybe not to bother otherwise. Plus, if they asked >> for passthrough, then not allocating a default domain at all is probably >> closer to the desired result than installing a DMA ops domain would be. > > You're right -- I'll hack this about to check if the default type isn't > DOMAIN_DMA before warning about the allocation failure. if some master devices want 'IDENTITY_DOMAIN' as default (because those devices do not want any iommu resources to be used/dma_ops to be set) and some 'DMA_DOMAIN' as default, then should the default be 'DMA_DOMAIN' and then masters needing IDENTITY_DOMAIN explicitly do an detach_dev later. This [1] was adding the support for detach_dev of the default DMA_DOMAINs. [1] https://patchwork.codeaurora.org/patch/164933/ Regards, Sricharan
diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c index 42a842e3f95f..f787626a745d 100644 --- a/drivers/iommu/iommu.c +++ b/drivers/iommu/iommu.c @@ -1027,10 +1027,19 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev) * IOMMU driver. */ if (!group->default_domain) { - group->default_domain = - __iommu_domain_alloc(dev->bus, iommu_def_domain_type); + struct iommu_domain *dom; + + dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type); + if (!dom) { + dev_warn(dev, + "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA", + iommu_def_domain_type); + dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA); + } + + group->default_domain = dom; if (!group->domain) - group->domain = group->default_domain; + group->domain = dom; } ret = iommu_group_add_device(group, dev);