On Tue, 15 Jan 2019 18:37:30 +0100
Pierre Morel <pmo...@linux.ibm.com> wrote:
> The s390 iommu can only allow DMA transactions between the zPCI device
> entries start_dma and end_dma.
>
> Let's declare the regions before start_dma and after end_dma as
> reserved regions using the appropriate callback in iommu_ops.
>
> Signed-off-by: Pierre Morel <pmo...@linux.ibm.com>
> ---
> drivers/iommu/s390-iommu.c | 29 +++++++++++++++++++++++++++++
> 1 file changed, 29 insertions(+)
>
> diff --git a/drivers/iommu/s390-iommu.c b/drivers/iommu/s390-iommu.c
> index 22d4db3..5ca91a1 100644
> --- a/drivers/iommu/s390-iommu.c
> +++ b/drivers/iommu/s390-iommu.c
> @@ -363,6 +363,33 @@ void zpci_destroy_iommu(struct zpci_dev *zdev)
> iommu_device_sysfs_remove(&zdev->iommu_dev);
> }
>
> +static void s390_get_resv_regions(struct device *dev, struct list_head *head)
> +{
> + struct iommu_resv_region *region;
> + struct zpci_dev *zdev = to_pci_dev(dev)->sysdata;
> +
> + region = iommu_alloc_resv_region(0, zdev->start_dma,
> + 0, IOMMU_RESV_RESERVED);
> + if (!region)
> + return;
> + list_add_tail(®ion->list, head);
> +
> + region = iommu_alloc_resv_region(zdev->end_dma + 1,
> + ~0UL - zdev->end_dma,
> + 0, IOMMU_RESV_RESERVED);
Can you guarantee that start_dma will never be 0 and end_dma never ~0UL,
even with future HW?
In any of these cases, your code would reserve strange ranges, and sysfs
would report broken reserved ranges.
Maybe add a check for start_dma > 0 and end_dma < ULONG_MAX?
> + if (!region)
> + return;
> + list_add_tail(®ion->list, head);
> +}
> +
> +static void s390_put_resv_regions(struct device *dev, struct list_head *head)
> +{
> + struct iommu_resv_region *entry, *next;
> +
> + list_for_each_entry_safe(entry, next, head, list)
> + kfree(entry);
> +}
It looks very wrong that there is no matching list_del() for the previous
list_add_tail(). However, it seems to be done like this everywhere else,
and the calling functions (currently) only use temporary list_heads as
far as I can see, so I guess it should be OK (for now).
Still, a list_del() would be nice :-)
> +
> static const struct iommu_ops s390_iommu_ops = {
> .capable = s390_iommu_capable,
> .domain_alloc = s390_domain_alloc,
> @@ -376,6 +403,8 @@ static const struct iommu_ops s390_iommu_ops = {
> .remove_device = s390_iommu_remove_device,
> .device_group = generic_device_group,
> .pgsize_bitmap = S390_IOMMU_PGSIZES,
> + .get_resv_regions = s390_get_resv_regions,
> + .put_resv_regions = s390_put_resv_regions,
> };
>
> static int __init s390_iommu_init(void)
With the start/end_dma issue addressed (if necessary):
Acked-by: Gerald Schaefer <gerald.schae...@de.ibm.com>
