[RFC] iommu: arm-smmu: stall support
robdclark at gmail.com
Tue Sep 19 07:23:43 PDT 2017
On Tue, Sep 19, 2017 at 8:30 AM, Joerg Roedel <joro at 8bytes.org> wrote:
> Hi Rob,
> thanks for the RFC patch. I have some comments about the interface to
> the IOMMU-API below.
> On Thu, Sep 14, 2017 at 03:44:33PM -0400, Rob Clark wrote:
>> + * iommu_domain_resume - Resume translations for a domain after a fault.
>> + *
>> + * This can be called at some point after the fault handler is called,
>> + * allowing the user of the IOMMU to (for example) handle the fault
>> + * from a task context. It is illegal to call this if
>> + * iommu_domain_set_attr(STALL) failed.
>> + *
>> + * @domain: the domain to resume
>> + * @terminate: if true, the translation that triggered the fault should
>> + * be terminated, else it should be retried.
>> + */
>> +void iommu_domain_resume(struct iommu_domain *domain, bool terminate)
>> + /* invalid to call if iommu_domain_set_attr(STALL) failed: */
>> + if (WARN_ON(!domain->ops->domain_resume))
>> + return;
>> + domain->ops->domain_resume(domain, terminate);
> So this function is being called by the device driver owning the domain,
yes, this was my plan
> I don't think that the resume call-back you added needs to be exposed
> like this. It is better to do the page-fault handling completly in the
> iommu-code, including calling the resume call-back and just let the
> device-driver provide a per-domain call-back to let it handle the fault
> and map in the required pages.
I would like to decide in the IRQ whether or not to queue work or not,
because when we get a gpu fault, we tend to get 1000's of gpu faults
all at once (and I really only need to handle the first one). I
suppose that could also be achieved by having a special return value
from the fault handler to say "call me again from a wq"..
Note that in the drm driver I already have a suitable wq to queue the
work, so it really doesn't buy me anything to have the iommu driver
toss things off to a wq for me. Might be a different situation for
other drivers (but I guess mostly other drivers are using iommu API
indirectly via dma-mapping?)
> The interface could look like this:
> * New function iommu_domain_enable_stalls(domain) - When
> this function returns the domain is in stall-handling mode. A
> iommu_domain_disable_stalls() might make sense too, not sure
> about that.
I don't particularly see a use-case for disabling stalls, fwiw
> * When stalls are enabled for a domain, report_iommu_fault()
> queues the fault to a workqueue (so that its handler can
> block) and in the workqueue you call ->resume() based on the
> return value of the handler.
> As a side-note, as there has been discussion on this: For now it doesn't
> make sense to merge this with the SVM page-fault handling efforts, as
> this path is different enough (SVM will call handle_mm_fault() as the
> handler, for example).
More information about the linux-arm-kernel