01.02.2019 16:28, Thierry Reding пишет:
> From: Thierry Reding <treding@xxxxxxxxxx>
>
> On Tegra186 and later, the ARM SMMU provides an input address space that
> is 48 bits wide. However, memory clients can only address up to 40 bits.
> If the geometry is used as-is, allocations of IOVA space can end up in a
> region that is not addressable by the memory clients.
>
> To fix this, restrict the IOVA space to the DMA mask of the host1x
> device.
>
> Signed-off-by: Thierry Reding <treding@xxxxxxxxxx>
> ---
> drivers/gpu/host1x/dev.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/host1x/dev.c b/drivers/gpu/host1x/dev.c
> index 4c044ee54fe6..544b67f2b3ff 100644
> --- a/drivers/gpu/host1x/dev.c
> +++ b/drivers/gpu/host1x/dev.c
> @@ -283,6 +283,8 @@ static int host1x_probe(struct platform_device *pdev)
> host->group = iommu_group_get(&pdev->dev);
> if (host->group) {
> struct iommu_domain_geometry *geometry;
> + u64 mask = dma_get_mask(host->dev);
> + dma_addr_t start, end;
> unsigned long order;
>
> err = iova_cache_get();
> @@ -310,11 +312,12 @@ static int host1x_probe(struct platform_device *pdev)
> }
>
> geometry = &host->domain->geometry;
> + start = geometry->aperture_start & mask;
> + end = geometry->aperture_end & mask;
>
> order = __ffs(host->domain->pgsize_bitmap);
> - init_iova_domain(&host->iova, 1UL << order,
> - geometry->aperture_start >> order);
> - host->iova_end = geometry->aperture_end;
> + init_iova_domain(&host->iova, 1UL << order, start >> order);
> + host->iova_end = end;
> }
>
> skip_iommu:
>
For older Tegra's:
Reviewed-by: Dmitry Osipenko <digetx@xxxxxxxxx>
Tested-by: Dmitry Osipenko <digetx@xxxxxxxxx>
