Re: [PATCH v6 1/4] iommu/arm-smmu: add NVIDIA implementation for dual ARM MMU-500 usage

From: Thierry Reding
Date: Tue Jun 23 2020 - 06:29:37 EST


On Thu, Jun 04, 2020 at 04:44:11PM -0700, Krishna Reddy wrote:
> NVIDIA's Tegra194 soc uses two ARM MMU-500s together to interleave

s/soc/SoC/

> IOVA accesses across them.
> Add NVIDIA implementation for dual ARM MMU-500s and add new compatible
> string for Tegra194 soc.

Same here.

>
> Signed-off-by: Krishna Reddy <vdumpa@xxxxxxxxxx>
> ---
> MAINTAINERS | 2 +
> drivers/iommu/Makefile | 2 +-
> drivers/iommu/arm-smmu-impl.c | 3 +
> drivers/iommu/arm-smmu-nvidia.c | 161 ++++++++++++++++++++++++++++++++
> drivers/iommu/arm-smmu.h | 1 +
> 5 files changed, 168 insertions(+), 1 deletion(-)
> create mode 100644 drivers/iommu/arm-smmu-nvidia.c
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 50659d76976b7..118da0893c964 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -16572,9 +16572,11 @@ F: drivers/i2c/busses/i2c-tegra.c
>
> TEGRA IOMMU DRIVERS
> M: Thierry Reding <thierry.reding@xxxxxxxxx>
> +R: Krishna Reddy <vdumpa@xxxxxxxxxx>
> L: linux-tegra@xxxxxxxxxxxxxxx
> S: Supported
> F: drivers/iommu/tegra*
> +F: drivers/iommu/arm-smmu-nvidia.c
>
> TEGRA KBC DRIVER
> M: Laxman Dewangan <ldewangan@xxxxxxxxxx>
> diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile
> index 57cf4ba5e27cb..35542df00da72 100644
> --- a/drivers/iommu/Makefile
> +++ b/drivers/iommu/Makefile
> @@ -15,7 +15,7 @@ obj-$(CONFIG_AMD_IOMMU) += amd_iommu.o amd_iommu_init.o amd_iommu_quirks.o
> obj-$(CONFIG_AMD_IOMMU_DEBUGFS) += amd_iommu_debugfs.o
> obj-$(CONFIG_AMD_IOMMU_V2) += amd_iommu_v2.o
> obj-$(CONFIG_ARM_SMMU) += arm_smmu.o
> -arm_smmu-objs += arm-smmu.o arm-smmu-impl.o arm-smmu-qcom.o
> +arm_smmu-objs += arm-smmu.o arm-smmu-impl.o arm-smmu-qcom.o arm-smmu-nvidia.o
> obj-$(CONFIG_ARM_SMMU_V3) += arm-smmu-v3.o
> obj-$(CONFIG_DMAR_TABLE) += dmar.o
> obj-$(CONFIG_INTEL_IOMMU) += intel-iommu.o intel-pasid.o
> diff --git a/drivers/iommu/arm-smmu-impl.c b/drivers/iommu/arm-smmu-impl.c
> index c75b9d957b702..52c84c30f83e4 100644
> --- a/drivers/iommu/arm-smmu-impl.c
> +++ b/drivers/iommu/arm-smmu-impl.c
> @@ -160,6 +160,9 @@ struct arm_smmu_device *arm_smmu_impl_init(struct arm_smmu_device *smmu)
> */
> switch (smmu->model) {
> case ARM_MMU500:
> + if (of_device_is_compatible(smmu->dev->of_node,
> + "nvidia,tegra194-smmu-500"))
> + return nvidia_smmu_impl_init(smmu);

Should NVIDIA_TEGRA194_SMMU be a separate value for smmu->model,
perhaps? That way we avoid this somewhat odd check here.

> smmu->impl = &arm_mmu500_impl;
> break;
> case CAVIUM_SMMUV2:
> diff --git a/drivers/iommu/arm-smmu-nvidia.c b/drivers/iommu/arm-smmu-nvidia.c

I wonder if it would be better to name this arm-smmu-tegra.c to make it
clearer that this is for a Tegra chip. We do have regular expressions in
MAINTAINERS that catch anything with "tegra" in it to make this easier.

> new file mode 100644
> index 0000000000000..dafc293a45217
> --- /dev/null
> +++ b/drivers/iommu/arm-smmu-nvidia.c
> @@ -0,0 +1,161 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +// Nvidia ARM SMMU v2 implementation quirks

s/Nvidia/NVIDIA/

> +// Copyright (C) 2019 NVIDIA CORPORATION. All rights reserved.

I suppose this should now also include 2020.

> +
> +#define pr_fmt(fmt) "nvidia-smmu: " fmt

Same here. Might be worth making this "tegra-smmu: " for consistency.

> +
> +#include <linux/bitfield.h>
> +#include <linux/delay.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/slab.h>
> +
> +#include "arm-smmu.h"
> +
> +/* Tegra194 has three ARM MMU-500 Instances.
> + * Two of them are used together for Interleaved IOVA accesses and
> + * used by Non-Isochronous Hw devices for SMMU translations.

"non-isochronous", s/Hw/HW/

> + * Third one is used for SMMU translations from Isochronous HW devices.

"isochronous"

> + * It is possible to use this Implementation to program either

"implementation"

> + * all three or two of the instances identically as desired through
> + * DT node.
> + *
> + * Programming all the three instances identically comes with redundant tlb

s/tlb/TLB/

> + * invalidations as all three never need to be tlb invalidated for a HW device.

Same here.

> + *
> + * When Linux Kernel supports multiple SMMU devices, The SMMU device used for

"kernel" and "..., the SMMU device"

> + * Isochornous HW devices should be added as a separate ARM MMU-500 device

"isochronous"

> + * in DT and be programmed independently for efficient tlb invalidates.

"TLB"

> + *
> + */
> +#define MAX_SMMU_INSTANCES 3
> +
> +#define TLB_LOOP_TIMEOUT 1000000 /* 1s! */

USEC_PER_SEC?

> +#define TLB_SPIN_COUNT 10
> +
> +struct nvidia_smmu {
> + struct arm_smmu_device smmu;
> + unsigned int num_inst;
> + void __iomem *bases[MAX_SMMU_INSTANCES];
> +};
> +
> +#define to_nvidia_smmu(s) container_of(s, struct nvidia_smmu, smmu)

Making this static inline can make error messages more readable.

> +
> +#define nsmmu_page(smmu, inst, page) \
> + (((inst) ? to_nvidia_smmu(smmu)->bases[(inst)] : smmu->base) + \
> + ((page) << smmu->pgshift))

Can we simply define to_nvidia_smmu(smmu)->bases[0] = smmu->base in
nvidia_smmu_impl_init()? Then this would become just:

to_nvidia_smmu(smmu)->bases[inst] + ((page) << (smmu)->pgshift)

Also, the nsmmu_ prefix looks somewhat odd here. You already use struct
nvidia_smmu as the name of the structure, so why not be consistent and
continue to use nvidia_smmu_ as the prefix for function names?

Or perhaps even use tegra_smmu_ as the prefix to match the filename
change I suggested earlier.

> +
> +static u32 nsmmu_read_reg(struct arm_smmu_device *smmu,
> + int page, int offset)
> +{
> + return readl_relaxed(nsmmu_page(smmu, 0, page) + offset);
> +}
> +
> +static void nsmmu_write_reg(struct arm_smmu_device *smmu,
> + int page, int offset, u32 val)
> +{
> + unsigned int i;
> +
> + for (i = 0; i < to_nvidia_smmu(smmu)->num_inst; i++)
> + writel_relaxed(val, nsmmu_page(smmu, i, page) + offset);
> +}
> +
> +static u64 nsmmu_read_reg64(struct arm_smmu_device *smmu,
> + int page, int offset)
> +{
> + return readq_relaxed(nsmmu_page(smmu, 0, page) + offset);
> +}
> +
> +static void nsmmu_write_reg64(struct arm_smmu_device *smmu,
> + int page, int offset, u64 val)
> +{
> + unsigned int i;
> +
> + for (i = 0; i < to_nvidia_smmu(smmu)->num_inst; i++)
> + writeq_relaxed(val, nsmmu_page(smmu, i, page) + offset);
> +}
> +
> +static void nsmmu_tlb_sync(struct arm_smmu_device *smmu, int page,
> + int sync, int status)
> +{
> + u32 reg;

I see this is being used to store the value read from a register. I find
it clearer to call this "value" or "val" (or in this case perhaps even
"status") because whenever I read "reg" I immediately think this is
meant to be a register offset, which can then be confusing when I see it
used in I/O accessors because it is in the wrong position.

But anyway, that's just my opinion and this is a bit subjective, so feel
free to ignore.

> + unsigned int i;
> + unsigned int spin_cnt, delay;
> +
> + arm_smmu_writel(smmu, page, sync, 0);
> +
> + for (delay = 1; delay < TLB_LOOP_TIMEOUT; delay *= 2) {
> + for (spin_cnt = TLB_SPIN_COUNT; spin_cnt > 0; spin_cnt--) {
> + reg = 0;

You may want to declare the variable at this scope since you never need
it outside. Also, use a space between variable initialization and the
for block below for better readability.

> + for (i = 0; i < to_nvidia_smmu(smmu)->num_inst; i++) {
> + reg |= readl_relaxed(
> + nsmmu_page(smmu, i, page) + status);
> + }

Maybe add a separate variable for the page address so this can be a bit
uncluttered. Also, I'd prefer a blank line after the block for
readability.

> + if (!(reg & ARM_SMMU_sTLBGSTATUS_GSACTIVE))
> + return;
> + cpu_relax();

Blank line above cpu_relax() for readability.

> + }
> + udelay(delay);

Again, a blank line between blocks and subsequent statements can help
readability.

> + }
> + dev_err_ratelimited(smmu->dev,
> + "TLB sync timed out -- SMMU may be deadlocked\n");

Same here.

Also, is there anything we can do when this happens?

> +}
> +
> +static int nsmmu_reset(struct arm_smmu_device *smmu)
> +{
> + u32 reg;
> + unsigned int i;
> +
> + for (i = 0; i < to_nvidia_smmu(smmu)->num_inst; i++) {
> + /* clear global FSR */
> + reg = readl_relaxed(nsmmu_page(smmu, i, ARM_SMMU_GR0) +
> + ARM_SMMU_GR0_sGFSR);
> + writel_relaxed(reg, nsmmu_page(smmu, i, ARM_SMMU_GR0) +
> + ARM_SMMU_GR0_sGFSR);
> + }
> +
> + return 0;
> +}
> +
> +static const struct arm_smmu_impl nvidia_smmu_impl = {
> + .read_reg = nsmmu_read_reg,
> + .write_reg = nsmmu_write_reg,
> + .read_reg64 = nsmmu_read_reg64,
> + .write_reg64 = nsmmu_write_reg64,
> + .reset = nsmmu_reset,
> + .tlb_sync = nsmmu_tlb_sync,
> +};
> +
> +struct arm_smmu_device *nvidia_smmu_impl_init(struct arm_smmu_device *smmu)
> +{
> + unsigned int i;
> + struct nvidia_smmu *nsmmu;
> + struct resource *res;
> + struct device *dev = smmu->dev;
> + struct platform_device *pdev = to_platform_device(smmu->dev);
> +
> + nsmmu = devm_kzalloc(smmu->dev, sizeof(*nsmmu), GFP_KERNEL);
> + if (!nsmmu)
> + return ERR_PTR(-ENOMEM);
> +
> + nsmmu->smmu = *smmu;
> + /* Instance 0 is ioremapped by arm-smmu.c */
> + nsmmu->num_inst = 1;

Maybe add this here to simplify the nsmmu_page() macro above:

nsmmu->bases[0] = smmu->base;

> +
> + for (i = 1; i < MAX_SMMU_INSTANCES; i++) {
> + res = platform_get_resource(pdev, IORESOURCE_MEM, i);
> + if (!res)
> + break;
> + nsmmu->bases[i] = devm_ioremap_resource(dev, res);
> + if (IS_ERR(nsmmu->bases[i]))
> + return (struct arm_smmu_device *)nsmmu->bases[i];

ERR_CAST()?

> + nsmmu->num_inst++;
> + }

More blank lines would make this much easier to read, in my opinion.

> +
> + nsmmu->smmu.impl = &nvidia_smmu_impl;
> + devm_kfree(smmu->dev, smmu);

Maybe a comment here would be useful for readers to immediately
understand why you're doing this here.

> + pr_info("NVIDIA ARM SMMU Implementation, Instances=%d\n",
> + nsmmu->num_inst);

I think I'd just omit this. In general you should only let the user know
when things go wrong, but the above is printed when everything goes as
expected.

Thierry

Attachment: signature.asc
Description: PGP signature