On Sat, Aug 04, 2018 at 05:29:56PM +0300, Dmitry Osipenko wrote:
> In order to report clients name and access direction on GART's page fault,
> MC driver need to access GART registers. Add facility that provides access
> to the GART.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/memory/tegra/mc.c | 26 +++++++++++++++++++++++---
>  include/soc/tegra/mc.h    | 13 +++++++++++++
>  2 files changed, 36 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/memory/tegra/mc.c b/drivers/memory/tegra/mc.c
> index e56862495f36..4940d72b5263 100644
> --- a/drivers/memory/tegra/mc.c
> +++ b/drivers/memory/tegra/mc.c
> @@ -72,6 +72,8 @@ static const struct of_device_id tegra_mc_of_match[] = {
>  };
>  MODULE_DEVICE_TABLE(of, tegra_mc_of_match);
>  
> +static struct tegra_mc_gart_handle *gart_handle;
> +

Why the global variable? Can't this be part of struct tegra_mc? We
already do a very similar thing with the Tegra SMMU integration, why
invent something completely different? Can't we stick to a similar
mechanism?

Given that struct tegra_smmu is opaque at the memory controller level,
we could even go and store the GART related data in the same pointer.

How about the registration code goes into a struct tegra_gart_probe()
function that is called from tegra_mc_probe() right after the SMMU
block:

	if (IS_ENABLED(CONFIG_TEGRA_IOMMU_SMMU)) {
		mc->smmu = tegra_smmu_probe(...);
		...
	}

	if (IS_ENABLED(CONFIG_TEGRA_IOMMU_GART)) {
		mc->smmu = tegra_gart_probe(...);
		...
	}

?

Thierry