Re: [PATCH v8 3/3] iommu/arm-smmu: Add global/context fault implementation hooks

From: Robin Murphy
Date: Tue Jun 30 2020 - 08:14:01 EST


On 2020-06-30 09:37, Jon Hunter wrote:

On 30/06/2020 01:10, Krishna Reddy wrote:
Add global/context fault hooks to allow NVIDIA SMMU implementation
handle faults across multiple SMMUs.

Nit ... this is not just for NVIDIA, but this allows anyone to add
custom global/context and fault hooks. So I think that the changelog
should be clear that this change permits custom fault hooks and that
custom fault hooks are needed for the Tegra194 SMMU. You may also want
to say why.


Signed-off-by: Krishna Reddy <vdumpa@xxxxxxxxxx>
---
drivers/iommu/arm-smmu-nvidia.c | 98 +++++++++++++++++++++++++++++++++
drivers/iommu/arm-smmu.c | 17 +++++-
drivers/iommu/arm-smmu.h | 3 +
3 files changed, 116 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/arm-smmu-nvidia.c b/drivers/iommu/arm-smmu-nvidia.c
index 1124f0ac1823a..c9423b4199c65 100644
--- a/drivers/iommu/arm-smmu-nvidia.c
+++ b/drivers/iommu/arm-smmu-nvidia.c
@@ -147,6 +147,102 @@ static int nvidia_smmu_reset(struct arm_smmu_device *smmu)
return 0;
}
+static struct arm_smmu_domain *to_smmu_domain(struct iommu_domain *dom)
+{
+ return container_of(dom, struct arm_smmu_domain, domain);
+}
+
+static irqreturn_t nvidia_smmu_global_fault_inst(int irq,
+ struct arm_smmu_device *smmu,
+ int inst)
+{
+ u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
+ void __iomem *gr0_base = nvidia_smmu_page(smmu, inst, 0);
+
+ gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
+ if (!gfsr)
+ return IRQ_NONE;
+
+ gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
+ gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
+ gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
+
+ dev_err_ratelimited(smmu->dev,
+ "Unexpected global fault, this could be serious\n");
+ dev_err_ratelimited(smmu->dev,
+ "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
+ gfsr, gfsynr0, gfsynr1, gfsynr2);
+
+ writel_relaxed(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
+ return IRQ_HANDLED;
+}
+
+static irqreturn_t nvidia_smmu_global_fault(int irq, void *dev)
+{
+ int inst;

Should be unsigned

+ irqreturn_t irq_ret = IRQ_NONE;
+ struct arm_smmu_device *smmu = dev;
+ struct nvidia_smmu *nvidia_smmu = to_nvidia_smmu(smmu);
+
+ for (inst = 0; inst < nvidia_smmu->num_inst; inst++) {
+ irq_ret = nvidia_smmu_global_fault_inst(irq, smmu, inst);
+ if (irq_ret == IRQ_HANDLED)
+ return irq_ret;

Any chance there could be more than one SMMU faulting by the time we
service the interrupt?

It certainly seems plausible if the interconnect is automatically load-balancing requests across the SMMU instances - say a driver bug caused a buffer to be unmapped too early, there could be many in-flight accesses to parts of that buffer that aren't all taking the same path and thus could now fault in parallel.

[ And anyone inclined to nitpick global vs. context faults, s/unmap a buffer/tear down a domain/ ;) ]

Either way I think it would be easier to reason about if we just handled these like a typical shared interrupt and always checked all the instances.

+ }
+
+ return irq_ret;
+}
+
+static irqreturn_t nvidia_smmu_context_fault_bank(int irq,
+ struct arm_smmu_device *smmu,
+ int idx, int inst)
+{
+ u32 fsr, fsynr, cbfrsynra;
+ unsigned long iova;
+ void __iomem *gr1_base = nvidia_smmu_page(smmu, inst, 1);
+ void __iomem *cb_base = nvidia_smmu_page(smmu, inst, smmu->numpage + idx);
+
+ fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
+ if (!(fsr & ARM_SMMU_FSR_FAULT))
+ return IRQ_NONE;
+
+ fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
+ iova = readq_relaxed(cb_base + ARM_SMMU_CB_FAR);
+ cbfrsynra = readl_relaxed(gr1_base + ARM_SMMU_GR1_CBFRSYNRA(idx));
+
+ dev_err_ratelimited(smmu->dev,
+ "Unhandled context fault: fsr=0x%x, iova=0x%08lx, fsynr=0x%x, cbfrsynra=0x%x, cb=%d\n",
+ fsr, iova, fsynr, cbfrsynra, idx);
+
+ writel_relaxed(fsr, cb_base + ARM_SMMU_CB_FSR);
+ return IRQ_HANDLED;
+}
+
+static irqreturn_t nvidia_smmu_context_fault(int irq, void *dev)
+{
+ int inst, idx;

Unsigned

+ irqreturn_t irq_ret = IRQ_NONE;
+ struct iommu_domain *domain = dev;
+ struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
+ struct arm_smmu_device *smmu = smmu_domain->smmu;
+
+ for (inst = 0; inst < to_nvidia_smmu(smmu)->num_inst; inst++) {
+ /*
+ * Interrupt line shared between all context faults.
+ * Check for faults across all contexts.
+ */
+ for (idx = 0; idx < smmu->num_context_banks; idx++) {
+ irq_ret = nvidia_smmu_context_fault_bank(irq, smmu,
+ idx, inst);
+
+ if (irq_ret == IRQ_HANDLED)
+ return irq_ret;

Any reason why we don't check all banks?

As above, we certainly shouldn't bail out without checking the bank for the offending domain across all of its instances, and I guess the way this works means that we would have to iterate all the banks to achieve that.

+ }
+ }
+
+ return irq_ret;
+}
+
static const struct arm_smmu_impl nvidia_smmu_impl = {
.read_reg = nvidia_smmu_read_reg,
.write_reg = nvidia_smmu_write_reg,
@@ -154,6 +250,8 @@ static const struct arm_smmu_impl nvidia_smmu_impl = {
.write_reg64 = nvidia_smmu_write_reg64,
.reset = nvidia_smmu_reset,
.tlb_sync = nvidia_smmu_tlb_sync,
+ .global_fault = nvidia_smmu_global_fault,
+ .context_fault = nvidia_smmu_context_fault,
};
struct arm_smmu_device *nvidia_smmu_impl_init(struct arm_smmu_device *smmu)
diff --git a/drivers/iommu/arm-smmu.c b/drivers/iommu/arm-smmu.c
index 243bc4cb2705b..3bb0aba15a356 100644
--- a/drivers/iommu/arm-smmu.c
+++ b/drivers/iommu/arm-smmu.c
@@ -673,6 +673,7 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
enum io_pgtable_fmt fmt;
struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
+ irqreturn_t (*context_fault)(int irq, void *dev);
mutex_lock(&smmu_domain->init_mutex);
if (smmu_domain->smmu)
@@ -835,7 +836,13 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
* handler seeing a half-initialised domain state.
*/
irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
- ret = devm_request_irq(smmu->dev, irq, arm_smmu_context_fault,
+
+ if (smmu->impl && smmu->impl->context_fault)
+ context_fault = smmu->impl->context_fault;
+ else
+ context_fault = arm_smmu_context_fault;

Why not see the default smmu->impl->context_fault to
arm_smmu_context_fault in arm_smmu_impl_init() and then allow the
various implementations to override as necessary? Then you can get rid
of this context_fault variable here and just use
smmu->impl->context_fault below.

Because the default smmu->impl is NULL. And as I've said before, NAK to forcing the common case to allocate a set of "quirks" purely to override the default IRQ handler with the default IRQ handler ;)

Robin.

+
+ ret = devm_request_irq(smmu->dev, irq, context_fault,
IRQF_SHARED, "arm-smmu-context-fault", domain);
if (ret < 0) {
dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
@@ -2107,6 +2114,7 @@ static int arm_smmu_device_probe(struct platform_device *pdev)
struct arm_smmu_device *smmu;
struct device *dev = &pdev->dev;
int num_irqs, i, err;
+ irqreturn_t (*global_fault)(int irq, void *dev);
smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
if (!smmu) {
@@ -2193,9 +2201,14 @@ static int arm_smmu_device_probe(struct platform_device *pdev)
smmu->num_context_irqs = smmu->num_context_banks;
}
+ if (smmu->impl && smmu->impl->global_fault)
+ global_fault = smmu->impl->global_fault;
+ else
+ global_fault = arm_smmu_global_fault;
+

Same here.

Jon