[PATCH 5/5] iommu/arm-smmu-v3: workaround for STE abort in kdump kernel

[Zhen Lei]

Some boards may not implement the STE.config=0b000 correctly, it also
reports event C_BAD_STE when a transaction incoming. To make kdump kernel
can be worked well in this situation, backup the strtab_base which is used
in the first kernel, to make the unexpected devices can reuse the old
mapping if we detected the STE.config=0b000 take no effect.

Signed-off-by: Zhen Lei <thunder.leizhen@xxxxxxxxxx>
---
 drivers/iommu/arm-smmu-v3.c | 100 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 100 insertions(+)

diff --git a/drivers/iommu/arm-smmu-v3.c b/drivers/iommu/arm-smmu-v3.c
index 84adecc..4e95710 100644
--- a/drivers/iommu/arm-smmu-v3.c
+++ b/drivers/iommu/arm-smmu-v3.c
@@ -335,6 +335,9 @@
 #define EVTQ_MAX_SZ_SHIFT		7

 #define EVTQ_0_ID			GENMASK_ULL(7, 0)
+#define EVTQ_0_ID_C_BAD_STE		0x4
+#define EVTQ_0_SSV			GENMASK_ULL(11, 11)
+#define EVTQ_0_SID			GENMASK_ULL(63, 32)

 /* PRI queue */
 #define PRIQ_ENT_DWORDS			2
@@ -525,6 +528,7 @@ struct arm_smmu_strtab_ent {
 struct arm_smmu_strtab_cfg {
 	__le64				*strtab;
 	dma_addr_t			strtab_dma;
+	dma_addr_t			former_strtab_dma;
 	struct arm_smmu_strtab_l1_desc	*l1_desc;
 	unsigned int			num_l1_ents;

@@ -1295,6 +1299,95 @@ static __le64 *arm_smmu_get_step_for_sid(struct arm_smmu_device *smmu, u32 sid)
 	return step;
 }

+/*
+ * This function is only called in the kdump kernel, and mainly because of the
+ * smmu hardware feature "ste abort" is not effective.
+ *
+ * The first kernel flushed all cache before start the secondary kernel, so
+ * it's safe base on ioremap() to access the former smmu tables.
+ *
+ * If any error detected, just simply give up the attempt, directly return
+ * without any error reported.
+ */
+static void arm_smmu_ste_abort_quirks(struct arm_smmu_device *smmu, u64 evt0)
+{
+	int i;
+	__le64 *dst, *src;
+	u64 val, paddr;
+	u32 sid = FIELD_GET(EVTQ_0_SID, evt0);
+	struct arm_smmu_strtab_cfg *cfg = &smmu->strtab_cfg;
+
+	/* SubStreamID is not support yet */
+	if (FIELD_GET(EVTQ_0_SSV, evt0))
+		return;
+
+	/*
+	 * If no device within this L2ST range has been added, the L1STD.L2Ptr
+	 * still point to the dummy L2ST, we should allocate one now.
+	 */
+	if (smmu->features & ARM_SMMU_FEAT_2_LVL_STRTAB) {
+		int idx, ret;
+
+		idx = (sid >> STRTAB_SPLIT) * STRTAB_L1_DESC_DWORDS;
+		if (!cfg->l1_desc[idx].l2ptr) {
+			ret = arm_smmu_init_l2_strtab(smmu, sid);
+			if (ret)
+				return;
+		}
+	}
+
+	dst = arm_smmu_get_step_for_sid(smmu, sid);
+	val = le64_to_cpu(dst[0]);
+	if (FIELD_GET(STRTAB_STE_0_CFG, val) != STRTAB_STE_0_CFG_ABORT)
+		return;
+
+	/* The value of SMMU_STRTAB_BASE maybe corrupted, sanity check it */
+	if (cfg->former_strtab_dma & ~(STRTAB_BASE_RA | STRTAB_BASE_ADDR_MASK))
+		return;
+
+	/* Find the STE base address of "sid" */
+	if (smmu->features & ARM_SMMU_FEAT_2_LVL_STRTAB) {
+		u32 span;
+
+		paddr = cfg->former_strtab_dma +
+			(sid >> STRTAB_SPLIT) * STRTAB_L1_DESC_SIZE;
+		src = ioremap(paddr, STRTAB_L1_DESC_SIZE);
+		if (!src)
+			return;
+
+		val   = le64_to_cpu(*src);
+		paddr = val & STRTAB_L1_DESC_L2PTR_MASK;
+		span  = val & STRTAB_L1_DESC_SPAN;
+		iounmap(src);
+
+		/* The content of L1STD maybe corrupted, sanity check it */
+		if (val & ~(STRTAB_L1_DESC_L2PTR_MASK | STRTAB_L1_DESC_SPAN))
+			return;
+		paddr += (sid & ((1 << STRTAB_SPLIT) - 1)) * STRTAB_STE_SIZE;
+	} else {
+		paddr = cfg->former_strtab_dma + (sid * STRTAB_STE_SIZE);
+	}
+
+	src = ioremap(paddr, STRTAB_STE_SIZE);
+	if (!src)
+		return;
+
+	/*
+	 * Copy the former STE content, so that the device can base the former
+	 * mapping to access "memory", and does not report any event again.
+	 *
+	 * Please note that, the "memory" is legally allocated in the first
+	 * kernel, so that it will not corrupt the memory of current secondary
+	 * kernel.
+	 */
+	for (i = 1; i < STRTAB_STE_DWORDS; i++)
+		dst[i] = src[i];
+	arm_smmu_sync_ste_for_sid(smmu, sid);
+	dst[0] = src[0];
+	arm_smmu_sync_ste_for_sid(smmu, sid);
+	iounmap(src);
+}
+
 /* IRQ and event handlers */
 static irqreturn_t arm_smmu_evtq_thread(int irq, void *dev)
 {
@@ -1312,6 +1405,8 @@ static irqreturn_t arm_smmu_evtq_thread(int irq, void *dev)
 				dev_info(smmu->dev, "\t0x%016llx\n",
 					 (unsigned long long)evt[i]);

+			if ((id == EVTQ_0_ID_C_BAD_STE) && is_kdump_kernel())
+				arm_smmu_ste_abort_quirks(smmu, evt[0]);
 		}

 		/*
@@ -2491,6 +2586,7 @@ static int arm_smmu_device_reset(struct arm_smmu_device *smmu, bool bypass)
 {
 	int ret;
 	u32 reg, enables;
+	u64 reg64;
 	struct arm_smmu_cmdq_ent cmd;

 	/* Clear CR0 and sync (disables SMMU and queue processing) */
@@ -2519,6 +2615,10 @@ static int arm_smmu_device_reset(struct arm_smmu_device *smmu, bool bypass)
 	reg = CR2_PTM | CR2_RECINVSID | CR2_E2H;
 	writel_relaxed(reg, smmu->base + ARM_SMMU_CR2);

+	/* save the former strtab base */
+	reg64 = readq_relaxed(smmu->base + ARM_SMMU_STRTAB_BASE);
+	smmu->strtab_cfg.former_strtab_dma = reg64 & STRTAB_BASE_ADDR_MASK;
+
 	/* Stream table */
 	writeq_relaxed(smmu->strtab_cfg.strtab_base,
 		       smmu->base + ARM_SMMU_STRTAB_BASE);
--
1.8.3