iommu: arm-smmu: Merge for msm-kona kernel upgrade

Merge all iommu/smmu changes from msm-4.14 to msm-kona
as of:
'commit 72ac9228fe40 ("iommu: arm-smmu: dump additional
smmu registers in fault handler")'.

Change-Id: I14b697f63d786032b31b490f2102393094762ff3
Signed-off-by: Swathi Sridhar <swatsrid@codeaurora.org>

drivers/iommu/Kconfig

@@ -405,6 +405,16 @@ config ARM_SMMU_SELFTEST
 
 	  If unsure, say N here.
 
+config IOMMU_TLBSYNC_DEBUG
+	bool "TLB sync timeout debug"
+	depends on ARM_SMMU
+	help
+	  Enables to collect the SMMU system state information right
+	  after the first TLB sync timeout failure by calling BUG().
+	  Note to use this only on debug builds.
+
+	  If unsure, say N here.
+
 config QCOM_LAZY_MAPPING
 	bool "Reference counted iommu-mapping support"
 	depends on ION

drivers/iommu/arm-smmu-regs.h

@@ -191,6 +191,8 @@ enum arm_smmu_s2cr_privcfg {
 #define TLBSTATUS_SACTIVE		(1 << 0)
 #define ARM_SMMU_CB_ATS1PR		0x800
 #define ARM_SMMU_CB_ATSR		0x8f0
+#define ARM_SMMU_STATS_SYNC_INV_TBU_ACK 0x25dc
+#define ARM_SMMU_TBU_PWR_STATUS         0x2204
 
 #define SCTLR_MEM_ATTR_SHIFT		16
 #define SCTLR_SHCFG_SHIFT		22

drivers/iommu/arm-smmu.c

@@ -693,6 +693,20 @@ static int arm_smmu_arch_device_group(struct device *dev,
 	return smmu->arch_ops->device_group(dev, group);
 }
 
+static void arm_smmu_arch_write_sync(struct arm_smmu_device *smmu)
+{
+	u32 id;
+
+	if (!smmu)
+		return;
+
+	/* Read to complete prior write transcations */
+	id = readl_relaxed(ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_ID0);
+
+	/* Wait for read to complete before off */
+	rmb();
+}
+
 static struct device_node *dev_get_dev_node(struct device *dev)
 {
 	if (dev_is_pci(dev)) {
@@ -940,6 +954,9 @@ static int arm_smmu_power_on_atomic(struct arm_smmu_power_resources *pwr)
 static void arm_smmu_power_off_atomic(struct arm_smmu_power_resources *pwr)
 {
 	unsigned long flags;
+	struct arm_smmu_device *smmu = pwr->dev->driver_data;
+
+	arm_smmu_arch_write_sync(smmu);
 
 	spin_lock_irqsave(&pwr->clock_refs_lock, flags);
 	if (pwr->clock_refs_count == 0) {
@@ -1080,6 +1097,7 @@ static int __arm_smmu_tlb_sync(struct arm_smmu_device *smmu,
 				void __iomem *sync, void __iomem *status)
 {
 	unsigned int spin_cnt, delay;
+	u32 sync_inv_ack, tbu_pwr_status;
 
 	writel_relaxed(0, sync);
 	for (delay = 1; delay < TLB_LOOP_TIMEOUT; delay *= 2) {
@@ -1090,9 +1108,15 @@ static int __arm_smmu_tlb_sync(struct arm_smmu_device *smmu,
 		}
 		udelay(delay);
 	}
+	sync_inv_ack = scm_io_read((unsigned long)(smmu->phys_addr +
+				     ARM_SMMU_STATS_SYNC_INV_TBU_ACK));
+	tbu_pwr_status = scm_io_read((unsigned long)(smmu->phys_addr +
+				     ARM_SMMU_TBU_PWR_STATUS));
 	trace_tlbsync_timeout(smmu->dev, 0);
 	dev_err_ratelimited(smmu->dev,
-			    "TLB sync timed out -- SMMU may be deadlocked\n");
+			    "TLB sync timed out -- SMMU may be deadlocked ack 0x%x pwr 0x%x\n",
+			    sync_inv_ack, tbu_pwr_status);
+	BUG_ON(IS_ENABLED(CONFIG_IOMMU_TLBSYNC_DEBUG));
 	return -EINVAL;
 }
 
@@ -1376,6 +1400,62 @@ static struct iommu_gather_ops msm_smmu_gather_ops = {
 	.free_pages_exact = arm_smmu_free_pages_exact,
 };
 
+static void print_ctx_regs(struct arm_smmu_device *smmu, struct arm_smmu_cfg
+			   *cfg, unsigned int fsr)
+{
+	u32 fsynr0;
+	void __iomem *cb_base = ARM_SMMU_CB(smmu, cfg->cbndx);
+	void __iomem *gr1_base = ARM_SMMU_GR1(smmu);
+	bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
+
+	fsynr0 = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
+
+	dev_err(smmu->dev, "FAR    = 0x%016llx\n",
+		readq_relaxed(cb_base + ARM_SMMU_CB_FAR));
+	dev_err(smmu->dev, "PAR    = 0x%pK\n",
+		readq_relaxed(cb_base + ARM_SMMU_CB_PAR));
+
+	dev_err(smmu->dev,
+		"FSR    = 0x%08x [%s%s%s%s%s%s%s%s%s%s]\n",
+		fsr,
+		(fsr & 0x02) ?  (fsynr0 & 0x10 ?
+				 "TF W " : "TF R ") : "",
+		(fsr & 0x04) ? "AFF " : "",
+		(fsr & 0x08) ? (fsynr0 & 0x10 ?
+				"PF W " : "PF R ") : "",
+		(fsr & 0x10) ? "EF " : "",
+		(fsr & 0x20) ? "TLBMCF " : "",
+		(fsr & 0x40) ? "TLBLKF " : "",
+		(fsr & 0x80) ? "MHF " : "",
+		(fsr & 0x100) ? "UUT " : "",
+		(fsr & 0x40000000) ? "SS " : "",
+		(fsr & 0x80000000) ? "MULTI " : "");
+
+	if (cfg->fmt == ARM_SMMU_CTX_FMT_AARCH32_S) {
+		dev_err(smmu->dev, "TTBR0  = 0x%pK\n",
+			readl_relaxed(cb_base + ARM_SMMU_CB_TTBR0));
+		dev_err(smmu->dev, "TTBR1  = 0x%pK\n",
+			readl_relaxed(cb_base + ARM_SMMU_CB_TTBR1));
+	} else {
+		dev_err(smmu->dev, "TTBR0  = 0x%pK\n",
+			readq_relaxed(cb_base + ARM_SMMU_CB_TTBR0));
+		if (stage1)
+			dev_err(smmu->dev, "TTBR1  = 0x%pK\n",
+				readq_relaxed(cb_base + ARM_SMMU_CB_TTBR1));
+	}
+
+
+	dev_err(smmu->dev, "SCTLR  = 0x%08x ACTLR  = 0x%08x\n",
+	       readl_relaxed(cb_base + ARM_SMMU_CB_SCTLR),
+	       readl_relaxed(cb_base + ARM_SMMU_CB_ACTLR));
+	dev_err(smmu->dev, "CBAR  = 0x%08x\n",
+	       readl_relaxed(gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx)));
+	dev_err(smmu->dev, "MAIR0   = 0x%08x MAIR1   = 0x%08x\n",
+	       readl_relaxed(cb_base + ARM_SMMU_CB_S1_MAIR0),
+	       readl_relaxed(cb_base + ARM_SMMU_CB_S1_MAIR1));
+
+}
+
 static phys_addr_t arm_smmu_verify_fault(struct iommu_domain *domain,
 					 dma_addr_t iova, u32 fsr)
 {
@@ -1463,29 +1543,17 @@ static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
 		ret = IRQ_HANDLED;
 		resume = RESUME_TERMINATE;
 	} else {
-		phys_addr_t phys_atos = arm_smmu_verify_fault(domain, iova,
-							      fsr);
 		if (__ratelimit(&_rs)) {
+			phys_addr_t phys_atos = arm_smmu_verify_fault(domain,
+								      iova,
+								      fsr);
+
 			dev_err(smmu->dev,
 				"Unhandled context fault: iova=0x%08lx, cb=%d, fsr=0x%x, fsynr0=0x%x, fsynr1=0x%x\n",
 				iova, cfg->cbndx, fsr, fsynr0, fsynr1);
-			dev_err(smmu->dev, "FAR    = %016lx\n",
-				(unsigned long)iova);
-			dev_err(smmu->dev,
-				"FSR    = %08x [%s%s%s%s%s%s%s%s%s%s]\n",
-				fsr,
-				(fsr & 0x02) ?  (fsynr0 & 0x10 ?
-						"TF W " : "TF R ") : "",
-				(fsr & 0x04) ? "AFF " : "",
-				(fsr & 0x08) ? (fsynr0 & 0x10 ?
-						"PF W " : "PF R ") : "",
-				(fsr & 0x10) ? "EF " : "",
-				(fsr & 0x20) ? "TLBMCF " : "",
-				(fsr & 0x40) ? "TLBLKF " : "",
-				(fsr & 0x80) ? "MHF " : "",
-				(fsr & 0x100) ? "UUT " : "",
-				(fsr & 0x40000000) ? "SS " : "",
-				(fsr & 0x80000000) ? "MULTI " : "");
+
+			print_ctx_regs(smmu, cfg, fsr);
+
 			dev_err(smmu->dev,
 				"soft iova-to-phys=%pa\n", &phys_soft);
 			if (!phys_soft)
@@ -2426,7 +2494,10 @@ static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
 
 	mutex_lock(&smmu->stream_map_mutex);
 	for_each_cfg_sme(fwspec, i, idx) {
-		WARN_ON(s2cr[idx].attach_count == 0);
+		if (WARN_ON(s2cr[idx].attach_count == 0)) {
+			mutex_unlock(&smmu->stream_map_mutex);
+			return;
+		}
 		s2cr[idx].attach_count -= 1;
 
 		if (s2cr[idx].attach_count > 0)
@@ -4324,7 +4395,7 @@ static int arm_smmu_init_bus_scaling(struct arm_smmu_power_resources *pwr)
 	pwr->bus_client = msm_bus_scale_register_client(pwr->bus_dt_data);
 	if (!pwr->bus_client) {
 		dev_err(dev, "Bus client registration failed\n");
-		return -EINVAL;
+		return -EPROBE_DEFER;
 	}
 
 	return 0;
@@ -5018,6 +5089,11 @@ module_exit(arm_smmu_exit);
 #define DEBUG_PAR_PA_SHIFT		12
 #define DEBUG_PAR_FAULT_VAL		0x1
 
+#define DEBUG_AXUSER_REG		0x30
+#define DEBUG_AXUSER_CDMID_MASK         0xff
+#define DEBUG_AXUSER_CDMID_SHIFT        36
+#define DEBUG_AXUSER_CDMID_VAL          255
+
 #define TBU_DBG_TIMEOUT_US		100
 
 struct actlr_setting {
@@ -5277,9 +5353,13 @@ static phys_addr_t qsmmuv500_iova_to_phys(
 redo:
 	/* Set address and stream-id */
 	val = readq_relaxed(tbu->base + DEBUG_SID_HALT_REG);
+	val &= ~DEBUG_SID_HALT_SID_MASK;
 	val |= sid & DEBUG_SID_HALT_SID_MASK;
 	writeq_relaxed(val, tbu->base + DEBUG_SID_HALT_REG);
 	writeq_relaxed(iova, tbu->base + DEBUG_VA_ADDR_REG);
+	val = (u64)(DEBUG_AXUSER_CDMID_VAL & DEBUG_AXUSER_CDMID_MASK) <<
+		DEBUG_AXUSER_CDMID_SHIFT;
+	writeq_relaxed(val, tbu->base + DEBUG_AXUSER_REG);
 
 	/*
 	 * Write-back Read and Write-Allocate
@@ -5336,6 +5416,9 @@ static phys_addr_t qsmmuv500_iova_to_phys(
 	/* Reset hardware */
 	writeq_relaxed(0, tbu->base + DEBUG_TXN_TRIGG_REG);
 	writeq_relaxed(0, tbu->base + DEBUG_VA_ADDR_REG);
+	val = readl_relaxed(tbu->base + DEBUG_SID_HALT_REG);
+	val &= ~DEBUG_SID_HALT_SID_MASK;
+	writel_relaxed(val, tbu->base + DEBUG_SID_HALT_REG);
 
 	/*
 	 * After a failed translation, the next successful translation will
@@ -5351,6 +5434,12 @@ static phys_addr_t qsmmuv500_iova_to_phys(
 	qsmmuv500_tbu_resume(tbu);
 
 out_power_off:
+	/* Read to complete prior write transcations */
+	val = readl_relaxed(tbu->base + DEBUG_SR_HALT_ACK_REG);
+
+	/* Wait for read to complete before off */
+	rmb();
+
 	arm_smmu_power_off(tbu->pwr);
 
 	return phys;

drivers/iommu/dma-mapping-fast.c

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0-only
 /*
- * Copyright (c) 2016-2018, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2016-2019, The Linux Foundation. All rights reserved.
  */
 
 #include <linux/dma-contiguous.h>
@@ -540,12 +540,22 @@ static void *fast_smmu_alloc(struct device *dev, size_t size,
 	void *addr;
 	unsigned long flags;
 	struct sg_mapping_iter miter;
-	unsigned int count = ALIGN(size, SZ_4K) >> PAGE_SHIFT;
+	size_t count = ALIGN(size, SZ_4K) >> PAGE_SHIFT;
 	bool is_coherent = is_dma_coherent(dev, attrs);
 	int prot = dma_info_to_prot(DMA_BIDIRECTIONAL, is_coherent, attrs);
 	pgprot_t remap_prot = __get_dma_pgprot(attrs, PAGE_KERNEL, is_coherent);
 	struct page **pages;
 
+	/*
+	 * sg_alloc_table_from_pages accepts unsigned int value for count
+	 * so check count doesn't exceed UINT_MAX.
+	 */
+
+	if (count > UINT_MAX) {
+		dev_err(dev, "count: %zx exceeds UNIT_MAX\n", count);
+		return NULL;
+	}
+
 	*handle = DMA_ERROR_CODE;
 
 	pages = __fast_smmu_alloc_pages(count, gfp);

drivers/iommu/iommu-debug.c

@@ -289,20 +289,20 @@ static int iommu_debug_dma_reconfigure(struct iommu_debug_device *ddev,
 	bool coherent;
 
 	if (ddev->domain) {
-		dev_err(dev, "Already attached.\n");
+		dev_err_ratelimited(dev, "Already attached.\n");
 		return -EBUSY;
 	}
 
 	iommu = of_iommu_configure(dev, dev->of_node);
 	if (!iommu) {
-		dev_err(dev, "Is not associated with an iommu\n");
+		dev_err_ratelimited(dev, "Is not associated with an iommu\n");
 		return -EINVAL;
 	}
 
 	coherent = of_dma_is_coherent(dev->of_node);
 
 	if (!dev->iommu_group) {
-		dev_err(dev, "Does not have an iommu group\n");
+		dev_err_ratelimited(dev, "Does not have an iommu group\n");
 		return -EINVAL;
 	}
 
@@ -310,7 +310,7 @@ static int iommu_debug_dma_reconfigure(struct iommu_debug_device *ddev,
 	domain = iommu_get_domain_for_dev(dev);
 	if (domain) {
 		if (domain->type != IOMMU_DOMAIN_DMA) {
-			dev_err(dev, "Attached, but its not a default domain?\n");
+			dev_err_ratelimited(dev, "Attached, but its not a default domain?\n");
 			return -EINVAL;
 		}
 		iommu_detach_group(domain, dev->iommu_group);
@@ -318,19 +318,19 @@ static int iommu_debug_dma_reconfigure(struct iommu_debug_device *ddev,
 
 	domain = iommu_domain_alloc(dev->bus);
 	if (!domain) {
-		dev_err(dev, "Allocating iommu domain failed\n");
+		dev_err_ratelimited(dev, "Allocating iommu domain failed\n");
 		return -EINVAL;
 	}
 
 	domain->is_debug_domain = true;
 
 	if (iommu_debug_set_attrs(ddev, domain, attrs)) {
-		dev_err(dev, "Setting attrs failed\n");
+		dev_err_ratelimited(dev, "Setting attrs failed\n");
 		goto out_free_domain;
 	}
 
 	if (iommu_attach_group(domain, dev->iommu_group)) {
-		dev_err(dev, "attach group failed\n");
+		dev_err_ratelimited(dev, "attach group failed\n");
 		goto out_free_domain;
 	}
 
@@ -341,7 +341,7 @@ static int iommu_debug_dma_reconfigure(struct iommu_debug_device *ddev,
 	set_dma_ops(dev, NULL);
 	arch_setup_dma_ops(dev, dma_base, size, iommu, coherent);
 	if (!get_dma_ops(dev)) {
-		dev_err(dev, "arch_setup_dma_ops failed, dma ops are null.\n");
+		dev_err_ratelimited(dev, "arch_setup_dma_ops failed, dma ops are null.\n");
 		goto out_detach_group;
 	}
 
@@ -362,13 +362,13 @@ static void iommu_debug_dma_deconfigure(struct iommu_debug_device *ddev)
 	struct device *dev = ddev->dev;
 
 	if (!dev->iommu_group) {
-		dev_err(dev, "Does not have an iommu group\n");
+		dev_err_ratelimited(dev, "Does not have an iommu group\n");
 		return;
 	}
 
 	domain = ddev->domain;
 	if (!domain) {
-		dev_err(dev, "Is not attached\n");
+		dev_err_ratelimited(dev, "Is not attached\n");
 		return;
 	}
 
@@ -770,14 +770,14 @@ static int __tlb_stress_sweep(struct device *dev, struct seq_file *s)
 	for (iova = 0, i = 0; iova < max; iova += SZ_8K, ++i) {
 		dma_addr = dma_map_single(dev, virt, SZ_8K, DMA_TO_DEVICE);
 		if (dma_addr == DMA_ERROR_CODE) {
-			dev_err(dev, "Failed map on iter %d\n", i);
+			dev_err_ratelimited(dev, "Failed map on iter %d\n", i);
 			ret = -EINVAL;
 			goto out;
 		}
 	}
 
 	if (dma_map_single(dev, virt, SZ_4K, DMA_TO_DEVICE) != DMA_ERROR_CODE) {
-		dev_err(dev,
+		dev_err_ratelimited(dev,
 			"dma_map_single unexpectedly (VA should have been exhausted)\n");
 		ret = -EINVAL;
 		goto out;
@@ -797,7 +797,7 @@ static int __tlb_stress_sweep(struct device *dev, struct seq_file *s)
 	if (dma_addr != SZ_8K) {
 		dma_addr_t expected = SZ_8K;
 
-		dev_err(dev, "Unexpected dma_addr. got: %pa expected: %pa\n",
+		dev_err_ratelimited(dev, "Unexpected dma_addr. got: %pa expected: %pa\n",
 			&dma_addr, &expected);
 		ret = -EINVAL;
 		goto out;
@@ -812,14 +812,14 @@ static int __tlb_stress_sweep(struct device *dev, struct seq_file *s)
 	if (dma_addr != 0) {
 		dma_addr_t expected = 0;
 
-		dev_err(dev, "Unexpected dma_addr. got: %pa expected: %pa\n",
+		dev_err_ratelimited(dev, "Unexpected dma_addr. got: %pa expected: %pa\n",
 			&dma_addr, &expected);
 		ret = -EINVAL;
 		goto out;
 	}
 
 	if (dma_map_single(dev, virt, SZ_4K, DMA_TO_DEVICE) != DMA_ERROR_CODE) {
-		dev_err(dev,
+		dev_err_ratelimited(dev,
 			"dma_map_single unexpectedly after remaps (VA should have been exhausted)\n");
 		ret = -EINVAL;
 		goto out;
@@ -869,7 +869,7 @@ static int __rand_va_sweep(struct device *dev, struct seq_file *s,
 	virt = (void *)__get_free_pages(GFP_KERNEL, get_order(size));
 	if (!virt) {
 		if (size > SZ_8K) {
-			dev_err(dev,
+			dev_err_ratelimited(dev,
 				"Failed to allocate %s of memory, which is a lot. Skipping test for this size\n",
 				_size_to_string(size));
 			return 0;
@@ -881,7 +881,7 @@ static int __rand_va_sweep(struct device *dev, struct seq_file *s,
 	for (iova = 0, i = 0; iova < max; iova += size, ++i) {
 		dma_addr = dma_map_single(dev, virt, size, DMA_TO_DEVICE);
 		if (dma_addr == DMA_ERROR_CODE) {
-			dev_err(dev, "Failed map on iter %d\n", i);
+			dev_err_ratelimited(dev, "Failed map on iter %d\n", i);
 			ret = -EINVAL;
 			goto out;
 		}
@@ -914,7 +914,7 @@ static int __rand_va_sweep(struct device *dev, struct seq_file *s,
 	}
 
 	if (unmapped != remapped) {
-		dev_err(dev,
+		dev_err_ratelimited(dev,
 			"Unexpected random remap count! Unmapped %d but remapped %d\n",
 			unmapped, remapped);
 		ret = -EINVAL;
@@ -959,7 +959,7 @@ static int __full_va_sweep(struct device *dev, struct seq_file *s,
 	virt = (void *)__get_free_pages(GFP_KERNEL, get_order(size));
 	if (!virt) {
 		if (size > SZ_8K) {
-			dev_err(dev,
+			dev_err_ratelimited(dev,
 				"Failed to allocate %s of memory, which is a lot. Skipping test for this size\n",
 				_size_to_string(size));
 			return 0;
@@ -988,7 +988,7 @@ static int __full_va_sweep(struct device *dev, struct seq_file *s,
 			phys_addr_t expected = phys;
 
 			if (__check_mapping(dev, domain, iova, expected)) {
-				dev_err(dev, "iter: %d\n", i);
+				dev_err_ratelimited(dev, "iter: %d\n", i);
 				ret = -EINVAL;
 				goto out;
 			}
@@ -999,7 +999,7 @@ static int __full_va_sweep(struct device *dev, struct seq_file *s,
 			unsigned long theiova = ((SZ_1G * 4ULL) - size) - iova;
 
 			if (__check_mapping(dev, domain, theiova, expected)) {
-				dev_err(dev, "iter: %d\n", i);
+				dev_err_ratelimited(dev, "iter: %d\n", i);
 				ret = -EINVAL;
 				goto out;
 			}
@@ -1158,7 +1158,7 @@ static int __functional_dma_api_basic_test(struct device *dev,
 		pa = iommu_iova_to_phys(domain, iova);
 		pa2 = iommu_iova_to_phys_hard(domain, iova);
 		if (pa != pa2) {
-			dev_err(dev,
+			dev_err_ratelimited(dev,
 				"iova_to_phys doesn't match iova_to_phys_hard: %pa != %pa\n",
 				&pa, &pa2);
 			ret = -EINVAL;
@@ -1166,7 +1166,7 @@ static int __functional_dma_api_basic_test(struct device *dev,
 		}
 		pa2 = virt_to_phys(data);
 		if (pa != pa2) {
-			dev_err(dev,
+			dev_err_ratelimited(dev,
 				"iova_to_phys doesn't match virt_to_phys: %pa != %pa\n",
 				&pa, &pa2);
 			ret = -EINVAL;
@@ -1175,7 +1175,8 @@ static int __functional_dma_api_basic_test(struct device *dev,
 		dma_unmap_single(dev, iova, size, DMA_TO_DEVICE);
 		for (j = 0; j < size; ++j) {
 			if (data[j] != 0xa5) {
-				dev_err(dev, "data[%d] != 0xa5\n", data[j]);
+				dev_err_ratelimited(dev,
+					       "data[%d] != 0xa5\n", data[j]);
 				ret = -EINVAL;
 				goto out;
 			}
@@ -1230,7 +1231,7 @@ static int __functional_dma_api_map_sg_test(struct device *dev,
 			pa = iommu_iova_to_phys(domain, iova);
 			pa2 = iommu_iova_to_phys_hard(domain, iova);
 			if (pa != pa2) {
-				dev_err(dev,
+				dev_err_ratelimited(dev,
 					"iova_to_phys doesn't match iova_to_phys_hard: %pa != %pa\n",
 					&pa, &pa2);
 				ret = -EINVAL;
@@ -1241,7 +1242,7 @@ static int __functional_dma_api_map_sg_test(struct device *dev,
 			pa = iommu_iova_to_phys(domain, iova);
 			pa2 = iommu_iova_to_phys_hard(domain, iova);
 			if (pa != pa2) {
-				dev_err(dev,
+				dev_err_ratelimited(dev,
 					"iova_to_phys doesn't match iova_to_phys_hard: %pa != %pa\n",
 					&pa, &pa2);
 				ret = -EINVAL;
@@ -1289,7 +1290,7 @@ static int __apply_to_new_mapping(struct seq_file *s,
 		goto out_release_mapping;
 	}
 
-	dev_err(dev, "testing with pgtables at %pa\n", &pt_phys);
+	dev_err_ratelimited(dev, "testing with pgtables at %pa\n", &pt_phys);
 	if (iommu_enable_config_clocks(domain)) {
 		ds_printf(dev, s, "Couldn't enable clocks\n");
 		goto out_release_mapping;
@@ -1378,7 +1379,7 @@ static ssize_t __iommu_debug_attach_write(struct file *file,
 	int val, ret;
 
 	if (kstrtoint_from_user(ubuf, count, 0, &val)) {
-		pr_err("Invalid format. Expected a hex or decimal integer");
+		pr_err_ratelimited("Invalid format. Expected a hex or decimal integer");
 		return -EFAULT;
 	}
 
@@ -1386,10 +1387,10 @@ static ssize_t __iommu_debug_attach_write(struct file *file,
 	if (val) {
 		ret = iommu_debug_dma_reconfigure(ddev, attrs, 0, SZ_1G * 4ULL);
 		if (!ret)
-			pr_err("Attached\n");
+			pr_err_ratelimited("Attached\n");
 	} else {
 		iommu_debug_dma_deconfigure(ddev);
-		pr_err("Detached\n");
+		pr_err_ratelimited("Detached\n");
 	}
 	mutex_unlock(&ddev->state_lock);
 	retval = count;
@@ -1424,7 +1425,7 @@ static ssize_t iommu_debug_attach_read(struct file *file, char __user *ubuf,
 	c[0] = ddev->domain ? '1' : '0';
 	c[1] = '\n';
 	if (copy_to_user(ubuf, &c, 2)) {
-		pr_err("copy_to_user failed\n");
+		pr_err_ratelimited("copy_to_user failed\n");
 		return -EFAULT;
 	}
 	*offset = 1;		/* non-zero means we're done */
@@ -1459,7 +1460,7 @@ static ssize_t iommu_debug_test_virt_addr_read(struct file *file,
 
 	buflen = strlen(buf);
 	if (copy_to_user(ubuf, buf, buflen)) {
-		pr_err("Couldn't copy_to_user\n");
+		pr_err_ratelimited("Couldn't copy_to_user\n");
 		retval = -EFAULT;
 	} else {
 		*offset = 1;	/* non-zero means we're done */
@@ -1494,13 +1495,13 @@ static ssize_t iommu_debug_pte_write(struct file *file,
 	dma_addr_t iova;
 
 	if (kstrtox_from_user(ubuf, count, 0, &iova)) {
-		pr_err("Invalid format for iova\n");
+		pr_err_ratelimited("Invalid format for iova\n");
 		ddev->iova = 0;
 		return -EINVAL;
 	}
 
 	ddev->iova = iova;
-	pr_err("Saved iova=%pa for future PTE commands\n", &iova);
+	pr_err_ratelimited("Saved iova=%pa for future PTE commands\n", &iova);
 	return count;
 }
 
@@ -1515,7 +1516,7 @@ static ssize_t iommu_debug_pte_read(struct file *file, char __user *ubuf,
 	size_t buflen;
 
 	if (kptr_restrict != 0) {
-		pr_err("kptr_restrict needs to be disabled.\n");
+		pr_err_ratelimited("kptr_restrict needs to be disabled.\n");
 		return -EPERM;
 	}
 
@@ -1524,7 +1525,7 @@ static ssize_t iommu_debug_pte_read(struct file *file, char __user *ubuf,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
@@ -1540,7 +1541,7 @@ static ssize_t iommu_debug_pte_read(struct file *file, char __user *ubuf,
 
 	buflen = strlen(buf);
 	if (copy_to_user(ubuf, buf, buflen)) {
-		pr_err("Couldn't copy_to_user\n");
+		pr_err_ratelimited("Couldn't copy_to_user\n");
 		retval = -EFAULT;
 	} else {
 		*offset = 1;	/* non-zero means we're done */
@@ -1565,13 +1566,13 @@ static ssize_t iommu_debug_atos_write(struct file *file,
 	dma_addr_t iova;
 
 	if (kstrtox_from_user(ubuf, count, 0, &iova)) {
-		pr_err("Invalid format for iova\n");
+		pr_err_ratelimited("Invalid format for iova\n");
 		ddev->iova = 0;
 		return -EINVAL;
 	}
 
 	ddev->iova = iova;
-	pr_err("Saved iova=%pa for future ATOS commands\n", &iova);
+	pr_err_ratelimited("Saved iova=%pa for future ATOS commands\n", &iova);
 	return count;
 }
 
@@ -1585,7 +1586,7 @@ static ssize_t iommu_debug_atos_read(struct file *file, char __user *ubuf,
 	size_t buflen;
 
 	if (kptr_restrict != 0) {
-		pr_err("kptr_restrict needs to be disabled.\n");
+		pr_err_ratelimited("kptr_restrict needs to be disabled.\n");
 		return -EPERM;
 	}
 
@@ -1594,7 +1595,7 @@ static ssize_t iommu_debug_atos_read(struct file *file, char __user *ubuf,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
@@ -1605,7 +1606,7 @@ static ssize_t iommu_debug_atos_read(struct file *file, char __user *ubuf,
 	if (!phys) {
 		strlcpy(buf, "FAIL\n", 100);
 		phys = iommu_iova_to_phys(ddev->domain, ddev->iova);
-		dev_err(ddev->dev, "ATOS for %pa failed. Software walk returned: %pa\n",
+		dev_err_ratelimited(ddev->dev, "ATOS for %pa failed. Software walk returned: %pa\n",
 			&ddev->iova, &phys);
 	} else {
 		snprintf(buf, 100, "%pa\n", &phys);
@@ -1613,7 +1614,7 @@ static ssize_t iommu_debug_atos_read(struct file *file, char __user *ubuf,
 
 	buflen = strlen(buf);
 	if (copy_to_user(ubuf, buf, buflen)) {
-		pr_err("Couldn't copy_to_user\n");
+		pr_err_ratelimited("Couldn't copy_to_user\n");
 		retval = -EFAULT;
 	} else {
 		*offset = 1;	/* non-zero means we're done */
@@ -1640,7 +1641,7 @@ static ssize_t iommu_debug_dma_atos_read(struct file *file, char __user *ubuf,
 	size_t buflen;
 
 	if (kptr_restrict != 0) {
-		pr_err("kptr_restrict needs to be disabled.\n");
+		pr_err_ratelimited("kptr_restrict needs to be disabled.\n");
 		return -EPERM;
 	}
 	if (*offset)
@@ -1648,7 +1649,7 @@ static ssize_t iommu_debug_dma_atos_read(struct file *file, char __user *ubuf,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
@@ -1664,7 +1665,7 @@ static ssize_t iommu_debug_dma_atos_read(struct file *file, char __user *ubuf,
 
 	buflen = strlen(buf);
 	if (copy_to_user(ubuf, buf, buflen)) {
-		pr_err("Couldn't copy_to_user\n");
+		pr_err_ratelimited("Couldn't copy_to_user\n");
 		retval = -EFAULT;
 	} else {
 		*offset = 1;	/* non-zero means we're done */
@@ -1695,14 +1696,14 @@ static ssize_t iommu_debug_map_write(struct file *file, const char __user *ubuf,
 	struct iommu_debug_device *ddev = file->private_data;
 
 	if (count >= 100) {
-		pr_err("Value too large\n");
+		pr_err_ratelimited("Value too large\n");
 		return -EINVAL;
 	}
 
 	memset(buf, 0, 100);
 
 	if (copy_from_user(buf, ubuf, count)) {
-		pr_err("Couldn't copy from user\n");
+		pr_err_ratelimited("Couldn't copy from user\n");
 		retval = -EFAULT;
 	}
 
@@ -1735,27 +1736,27 @@ static ssize_t iommu_debug_map_write(struct file *file, const char __user *ubuf,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
 
 	ret = iommu_map(ddev->domain, iova, phys, size, prot);
 	if (ret) {
-		pr_err("iommu_map failed with %d\n", ret);
+		pr_err_ratelimited("iommu_map failed with %d\n", ret);
 		retval = -EIO;
 		goto out;
 	}
 
 	retval = count;
-	pr_err("Mapped %pa to %pa (len=0x%zx, prot=0x%x)\n",
+	pr_err_ratelimited("Mapped %pa to %pa (len=0x%zx, prot=0x%x)\n",
 	       &iova, &phys, size, prot);
 out:
 	mutex_unlock(&ddev->state_lock);
 	return retval;
 
 invalid_format:
-	pr_err("Invalid format. Expected: iova,phys,len,prot where `prot' is the bitwise OR of IOMMU_READ, IOMMU_WRITE, etc.\n");
+	pr_err_ratelimited("Invalid format. Expected: iova,phys,len,prot where `prot' is the bitwise OR of IOMMU_READ, IOMMU_WRITE, etc.\n");
 	return -EINVAL;
 }
 
@@ -1789,14 +1790,14 @@ static ssize_t iommu_debug_dma_map_write(struct file *file,
 	struct device *dev = ddev->dev;
 
 	if (count >= sizeof(buf)) {
-		pr_err("Value too large\n");
+		pr_err_ratelimited("Value too large\n");
 		return -EINVAL;
 	}
 
 	memset(buf, 0, sizeof(buf));
 
 	if (copy_from_user(buf, ubuf, count)) {
-		pr_err("Couldn't copy from user\n");
+		pr_err_ratelimited("Couldn't copy from user\n");
 		return -EFAULT;
 	}
 
@@ -1836,7 +1837,7 @@ static ssize_t iommu_debug_dma_map_write(struct file *file,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
@@ -1845,26 +1846,27 @@ static ssize_t iommu_debug_dma_map_write(struct file *file,
 					DMA_TO_DEVICE, dma_attrs);
 
 	if (dma_mapping_error(dev, iova)) {
-		pr_err("Failed to perform dma_map_single\n");
+		pr_err_ratelimited("Failed to perform dma_map_single\n");
 		ret = -EINVAL;
 		goto out;
 	}
 
 	retval = count;
-	pr_err("Mapped 0x%p to %pa (len=0x%zx)\n",
+	pr_err_ratelimited("Mapped 0x%p to %pa (len=0x%zx)\n",
 			v_addr, &iova, size);
 	ddev->iova = iova;
-		pr_err("Saved iova=%pa for future PTE commands\n", &iova);
+		pr_err_ratelimited("Saved iova=%pa for future PTE commands\n",
+				&iova);
 out:
 	mutex_unlock(&ddev->state_lock);
 	return retval;
 
 invalid_format:
-	pr_err("Invalid format. Expected: addr,len,dma attr where 'dma attr' is\n0: normal mapping\n1: force coherent\n2: force non-cohernet\n3: use system cache\n");
+	pr_err_ratelimited("Invalid format. Expected: addr,len,dma attr where 'dma attr' is\n0: normal mapping\n1: force coherent\n2: force non-cohernet\n3: use system cache\n");
 	return retval;
 
 invalid_addr:
-	pr_err("Invalid addr given! Address should be within 1MB size from start addr returned by doing 'cat test_virt_addr'.\n");
+	pr_err_ratelimited("Invalid addr given! Address should be within 1MB size from start addr returned by doing 'cat test_virt_addr'.\n");
 	return retval;
 }
 
@@ -1887,7 +1889,7 @@ static ssize_t iommu_debug_dma_map_read(struct file *file, char __user *ubuf,
 
 	buflen = strlen(buf);
 	if (copy_to_user(ubuf, buf, buflen)) {
-		pr_err("Couldn't copy_to_user\n");
+		pr_err_ratelimited("Couldn't copy_to_user\n");
 		retval = -EFAULT;
 	} else {
 		*offset = 1;	/* non-zero means we're done */
@@ -1916,19 +1918,19 @@ static ssize_t iommu_debug_unmap_write(struct file *file,
 	struct iommu_debug_device *ddev = file->private_data;
 
 	if (count >= 100) {
-		pr_err("Value too large\n");
+		pr_err_ratelimited("Value too large\n");
 		return -EINVAL;
 	}
 
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		return -EINVAL;
 	}
 
 	memset(buf, 0, 100);
 
 	if (copy_from_user(buf, ubuf, count)) {
-		pr_err("Couldn't copy from user\n");
+		pr_err_ratelimited("Couldn't copy from user\n");
 		retval = -EFAULT;
 		goto out;
 	}
@@ -1948,27 +1950,27 @@ static ssize_t iommu_debug_unmap_write(struct file *file,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
 
 	unmapped = iommu_unmap(ddev->domain, iova, size);
 	if (unmapped != size) {
-		pr_err("iommu_unmap failed. Expected to unmap: 0x%zx, unmapped: 0x%zx",
+		pr_err_ratelimited("iommu_unmap failed. Expected to unmap: 0x%zx, unmapped: 0x%zx",
 		       size, unmapped);
 		retval = -EIO;
 		goto out;
 	}
 
 	retval = count;
-	pr_err("Unmapped %pa (len=0x%zx)\n", &iova, size);
+	pr_err_ratelimited("Unmapped %pa (len=0x%zx)\n", &iova, size);
 out:
 	mutex_unlock(&ddev->state_lock);
 	return retval;
 
 invalid_format:
-	pr_err("Invalid format. Expected: iova,len\n");
+	pr_err_ratelimited("Invalid format. Expected: iova,len\n");
 	return -EINVAL;
 }
 
@@ -1992,14 +1994,14 @@ static ssize_t iommu_debug_dma_unmap_write(struct file *file,
 	struct device *dev = ddev->dev;
 
 	if (count >= sizeof(buf)) {
-		pr_err("Value too large\n");
+		pr_err_ratelimited("Value too large\n");
 		return -EINVAL;
 	}
 
 	memset(buf, 0, sizeof(buf));
 
 	if (copy_from_user(buf, ubuf, count)) {
-		pr_err("Couldn't copy from user\n");
+		pr_err_ratelimited("Couldn't copy from user\n");
 		retval = -EFAULT;
 		goto out;
 	}
@@ -2036,20 +2038,20 @@ static ssize_t iommu_debug_dma_unmap_write(struct file *file,
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
 	dma_unmap_single_attrs(dev, iova, size, DMA_TO_DEVICE, dma_attrs);
 
 	retval = count;
-	pr_err("Unmapped %pa (len=0x%zx)\n", &iova, size);
+	pr_err_ratelimited("Unmapped %pa (len=0x%zx)\n", &iova, size);
 out:
 	mutex_unlock(&ddev->state_lock);
 	return retval;
 
 invalid_format:
-	pr_err("Invalid format. Expected: iova,len, dma attr\n");
+	pr_err_ratelimited("Invalid format. Expected: iova,len, dma attr\n");
 	return retval;
 }
 
@@ -2068,17 +2070,17 @@ static ssize_t iommu_debug_config_clocks_write(struct file *file,
 
 	/* we're expecting a single character plus (optionally) a newline */
 	if (count > 2) {
-		dev_err(dev, "Invalid value\n");
+		dev_err_ratelimited(dev, "Invalid value\n");
 		return -EINVAL;
 	}
 
 	if (!ddev->domain) {
-		dev_err(dev, "No domain. Did you already attach?\n");
+		dev_err_ratelimited(dev, "No domain. Did you already attach?\n");
 		return -EINVAL;
 	}
 
 	if (copy_from_user(&buf, ubuf, 1)) {
-		dev_err(dev, "Couldn't copy from user\n");
+		dev_err_ratelimited(dev, "Couldn't copy from user\n");
 		return -EFAULT;
 	}
 
@@ -2086,26 +2088,26 @@ static ssize_t iommu_debug_config_clocks_write(struct file *file,
 	switch (buf) {
 	case '0':
 		if (ddev->clk_count == 0) {
-			dev_err(dev, "Config clocks already disabled\n");
+			dev_err_ratelimited(dev, "Config clocks already disabled\n");
 			break;
 		}
 
 		if (--ddev->clk_count > 0)
 			break;
 
-		dev_err(dev, "Disabling config clocks\n");
+		dev_err_ratelimited(dev, "Disabling config clocks\n");
 		iommu_disable_config_clocks(ddev->domain);
 		break;
 	case '1':
 		if (ddev->clk_count++ > 0)
 			break;
 
-		dev_err(dev, "Enabling config clocks\n");
+		dev_err_ratelimited(dev, "Enabling config clocks\n");
 		if (iommu_enable_config_clocks(ddev->domain))
-			dev_err(dev, "Failed!\n");
+			dev_err_ratelimited(dev, "Failed!\n");
 		break;
 	default:
-		dev_err(dev, "Invalid value. Should be 0 or 1.\n");
+		dev_err_ratelimited(dev, "Invalid value. Should be 0 or 1.\n");
 		mutex_unlock(&ddev->clk_lock);
 		return -EINVAL;
 	}
@@ -2127,13 +2129,13 @@ static ssize_t iommu_debug_trigger_fault_write(
 	unsigned long flags;
 
 	if (kstrtoul_from_user(ubuf, count, 0, &flags)) {
-		pr_err("Invalid flags format\n");
+		pr_err_ratelimited("Invalid flags format\n");
 		return -EFAULT;
 	}
 
 	mutex_lock(&ddev->state_lock);
 	if (!ddev->domain) {
-		pr_err("No domain. Did you already attach?\n");
+		pr_err_ratelimited("No domain. Did you already attach?\n");
 		mutex_unlock(&ddev->state_lock);
 		return -EINVAL;
 	}
@@ -2177,147 +2179,147 @@ static int snarf_iommu_devices(struct device *dev, void *ignored)
 	ddev->dev = dev;
 	dir = debugfs_create_dir(dev_name(dev), debugfs_tests_dir);
 	if (!dir) {
-		pr_err("Couldn't create iommu/devices/%s debugfs dir\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s debugfs dir\n",
 		       dev_name(dev));
 		goto err;
 	}
 
 	if (!debugfs_create_file("nr_iters", 0400, dir, &iters_per_op,
 				&iommu_debug_nr_iters_ops)) {
-		pr_err("Couldn't create iommu/devices/%s/nr_iters debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/nr_iters debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("test_virt_addr", 0400, dir, ddev,
 				&iommu_debug_test_virt_addr_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/test_virt_addr debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/test_virt_addr debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("profiling", 0400, dir, ddev,
 				 &iommu_debug_profiling_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/profiling debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/profiling debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("secure_profiling", 0400, dir, ddev,
 				 &iommu_debug_secure_profiling_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/secure_profiling debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/secure_profiling debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("profiling_fast", 0400, dir, ddev,
 				 &iommu_debug_profiling_fast_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/profiling_fast debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/profiling_fast debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("profiling_fast_dma_api", 0400, dir, ddev,
 				 &iommu_debug_profiling_fast_dma_api_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/profiling_fast_dma_api debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/profiling_fast_dma_api debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("functional_fast_dma_api", 0400, dir, ddev,
 				 &iommu_debug_functional_fast_dma_api_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/functional_fast_dma_api debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/functional_fast_dma_api debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("functional_arm_dma_api", 0400, dir, ddev,
 				 &iommu_debug_functional_arm_dma_api_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/functional_arm_dma_api debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/functional_arm_dma_api debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("dma_attach", 0600, dir, ddev,
 				 &iommu_debug_dma_attach_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/dma_attach debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/dma_attach debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("attach", 0400, dir, ddev,
 				 &iommu_debug_attach_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/attach debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/attach debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("secure_attach", 0400, dir, ddev,
 				 &iommu_debug_secure_attach_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/secure_attach debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/secure_attach debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("atos", 0200, dir, ddev,
 				 &iommu_debug_atos_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/atos debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/atos debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("dma_atos", 0600, dir, ddev,
 				 &iommu_debug_dma_atos_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/dma_atos debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/dma_atos debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("map", 0200, dir, ddev,
 				 &iommu_debug_map_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/map debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/map debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("dma_map", 0600, dir, ddev,
 					 &iommu_debug_dma_map_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/dma_map debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/dma_map debugfs file\n",
 		       dev_name(dev));
 			goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("unmap", 0200, dir, ddev,
 				 &iommu_debug_unmap_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/unmap debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/unmap debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("dma_unmap", 0200, dir, ddev,
 					 &iommu_debug_dma_unmap_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/dma_unmap debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/dma_unmap debugfs file\n",
 		       dev_name(dev));
 			goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("pte", 0600, dir, ddev,
 			&iommu_debug_pte_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/pte debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/pte debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("config_clocks", 0200, dir, ddev,
 				 &iommu_debug_config_clocks_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/config_clocks debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/config_clocks debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
 
 	if (!debugfs_create_file("trigger-fault", 0200, dir, ddev,
 				 &iommu_debug_trigger_fault_fops)) {
-		pr_err("Couldn't create iommu/devices/%s/trigger-fault debugfs file\n",
+		pr_err_ratelimited("Couldn't create iommu/devices/%s/trigger-fault debugfs file\n",
 		       dev_name(dev));
 		goto err_rmdir;
 	}
@@ -2337,7 +2339,7 @@ static int iommu_debug_init_tests(void)
 	debugfs_tests_dir = debugfs_create_dir("tests",
 					       iommu_debugfs_top);
 	if (!debugfs_tests_dir) {
-		pr_err("Couldn't create iommu/tests debugfs directory\n");
+		pr_err_ratelimited("Couldn't create iommu/tests debugfs directory\n");
 		return -ENODEV;
 	}
 

drivers/iommu/of_iommu.c

@@ -103,6 +103,14 @@ static int of_iommu_xlate(struct device *dev,
 	int err;
 
 	ops = iommu_ops_from_fwnode(fwnode);
+	/*
+	 * Return -EPROBE_DEFER for the platform devices which are dependent
+	 * on the SMMU driver registration. Deferring from here helps in adding
+	 * the clients in proper iommu groups.
+	 */
+	if (!dev_is_pci(dev) && of_device_is_available(iommu_spec->np) && !ops)
+		return -EPROBE_DEFER;
+
 	if ((ops && !ops->of_xlate) ||
 	    !of_device_is_available(iommu_spec->np))
 		return NO_IOMMU;

include/linux/dma-mapping.h

@@ -823,6 +823,10 @@ int dma_declare_coherent_memory(struct device *dev, phys_addr_t phys_addr,
 void dma_release_declared_memory(struct device *dev);
 void *dma_mark_declared_memory_occupied(struct device *dev,
 					dma_addr_t device_addr, size_t size);
+dma_addr_t dma_get_device_base(struct device *dev,
+			       struct dma_coherent_mem *mem);
+unsigned long dma_get_size(struct dma_coherent_mem *mem);
+
 #else
 static inline int
 dma_declare_coherent_memory(struct device *dev, phys_addr_t phys_addr,
@@ -842,6 +846,17 @@ dma_mark_declared_memory_occupied(struct device *dev,
 {
 	return ERR_PTR(-EBUSY);
 }
+static inline dma_addr_t
+dma_get_device_base(struct device *dev, struct dma_coherent_mem *mem)
+{
+	return 0;
+}
+
+static inline unsigned long dma_get_size(struct dma_coherent_mem *mem)
+{
+	return 0;
+}
+
 #endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */
 
 #ifdef CONFIG_HAS_DMA

kernel/dma/coherent.c

@@ -29,14 +29,21 @@ static inline struct dma_coherent_mem *dev_get_coherent_memory(struct device *de
 	return NULL;
 }
 
-static inline dma_addr_t dma_get_device_base(struct device *dev,
-					     struct dma_coherent_mem * mem)
+dma_addr_t dma_get_device_base(struct device *dev,
+			       struct dma_coherent_mem *mem)
 {
 	if (mem->use_dev_dma_pfn_offset)
 		return (mem->pfn_base - dev->dma_pfn_offset) << PAGE_SHIFT;
 	else
 		return mem->device_base;
 }
+EXPORT_SYMBOL(dma_get_device_base);
+
+unsigned long dma_get_size(struct dma_coherent_mem *mem)
+{
+	return mem->size << PAGE_SHIFT;
+}
+EXPORT_SYMBOL(dma_get_size);
 
 static int dma_init_coherent_memory(
 	phys_addr_t phys_addr, dma_addr_t device_addr, size_t size, int flags,

kernel/dma/removed.c

@@ -75,7 +75,7 @@ void *removed_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 	bool skip_zeroing = attrs & DMA_ATTR_SKIP_ZEROING;
 	int pageno;
 	unsigned long order;
-	void *addr = NULL;
+	void __iomem *addr = NULL;
 	struct removed_region *dma_mem = dev->removed_mem;
 	int nbits;
 	unsigned int align;
@@ -108,7 +108,7 @@ void *removed_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 			goto out;
 		}
 
-		addr = ioremap(base, size);
+		addr = ioremap_wc(base, size);
 		if (WARN_ON(!addr)) {
 			bitmap_clear(dma_mem->bitmap, pageno, nbits);
 		} else {
@@ -202,10 +202,10 @@ void removed_sync_sg_for_device(struct device *dev,
 {
 }
 
-void *removed_remap(struct device *dev, void *cpu_addr, dma_addr_t handle,
-			size_t size, unsigned long attrs)
+static void __iomem *removed_remap(struct device *dev, void *cpu_addr,
+			dma_addr_t handle, size_t size, unsigned long attrs)
 {
-	return ioremap(handle, size);
+	return ioremap_wc(handle, size);
 }
 
 void removed_unremap(struct device *dev, void *remapped_address, size_t size)