1 // SPDX-License-Identifier: GPL-2.0-only
2 // Copyright (C) 2019-2020 NVIDIA CORPORATION.  All rights reserved.
3 
4 #include <linux/bitfield.h>
5 #include <linux/delay.h>
6 #include <linux/of.h>
7 #include <linux/platform_device.h>
8 #include <linux/slab.h>
9 
10 #include "arm-smmu.h"
11 
12 /*
13  * Tegra194 has three ARM MMU-500 Instances.
14  * Two of them are used together and must be programmed identically for
15  * interleaved IOVA accesses across them and translates accesses from
16  * non-isochronous HW devices.
17  * Third one is used for translating accesses from isochronous HW devices.
18  * This implementation supports programming of the two instances that must
19  * be programmed identically.
20  * The third instance usage is through standard arm-smmu driver itself and
21  * is out of scope of this implementation.
22  */
23 #define NUM_SMMU_INSTANCES 2
24 
25 struct nvidia_smmu {
26 	struct arm_smmu_device	smmu;
27 	void __iomem		*bases[NUM_SMMU_INSTANCES];
28 };
29 
30 static inline void __iomem *nvidia_smmu_page(struct arm_smmu_device *smmu,
31 					     unsigned int inst, int page)
32 {
33 	struct nvidia_smmu *nvidia_smmu;
34 
35 	nvidia_smmu = container_of(smmu, struct nvidia_smmu, smmu);
36 	return nvidia_smmu->bases[inst] + (page << smmu->pgshift);
37 }
38 
39 static u32 nvidia_smmu_read_reg(struct arm_smmu_device *smmu,
40 				int page, int offset)
41 {
42 	void __iomem *reg = nvidia_smmu_page(smmu, 0, page) + offset;
43 
44 	return readl_relaxed(reg);
45 }
46 
47 static void nvidia_smmu_write_reg(struct arm_smmu_device *smmu,
48 				  int page, int offset, u32 val)
49 {
50 	unsigned int i;
51 
52 	for (i = 0; i < NUM_SMMU_INSTANCES; i++) {
53 		void __iomem *reg = nvidia_smmu_page(smmu, i, page) + offset;
54 
55 		writel_relaxed(val, reg);
56 	}
57 }
58 
59 static u64 nvidia_smmu_read_reg64(struct arm_smmu_device *smmu,
60 				  int page, int offset)
61 {
62 	void __iomem *reg = nvidia_smmu_page(smmu, 0, page) + offset;
63 
64 	return readq_relaxed(reg);
65 }
66 
67 static void nvidia_smmu_write_reg64(struct arm_smmu_device *smmu,
68 				    int page, int offset, u64 val)
69 {
70 	unsigned int i;
71 
72 	for (i = 0; i < NUM_SMMU_INSTANCES; i++) {
73 		void __iomem *reg = nvidia_smmu_page(smmu, i, page) + offset;
74 
75 		writeq_relaxed(val, reg);
76 	}
77 }
78 
79 static void nvidia_smmu_tlb_sync(struct arm_smmu_device *smmu, int page,
80 				 int sync, int status)
81 {
82 	unsigned int delay;
83 
84 	arm_smmu_writel(smmu, page, sync, 0);
85 
86 	for (delay = 1; delay < TLB_LOOP_TIMEOUT; delay *= 2) {
87 		unsigned int spin_cnt;
88 
89 		for (spin_cnt = TLB_SPIN_COUNT; spin_cnt > 0; spin_cnt--) {
90 			u32 val = 0;
91 			unsigned int i;
92 
93 			for (i = 0; i < NUM_SMMU_INSTANCES; i++) {
94 				void __iomem *reg;
95 
96 				reg = nvidia_smmu_page(smmu, i, page) + status;
97 				val |= readl_relaxed(reg);
98 			}
99 
100 			if (!(val & ARM_SMMU_sTLBGSTATUS_GSACTIVE))
101 				return;
102 
103 			cpu_relax();
104 		}
105 
106 		udelay(delay);
107 	}
108 
109 	dev_err_ratelimited(smmu->dev,
110 			    "TLB sync timed out -- SMMU may be deadlocked\n");
111 }
112 
113 static int nvidia_smmu_reset(struct arm_smmu_device *smmu)
114 {
115 	unsigned int i;
116 
117 	for (i = 0; i < NUM_SMMU_INSTANCES; i++) {
118 		u32 val;
119 		void __iomem *reg = nvidia_smmu_page(smmu, i, ARM_SMMU_GR0) +
120 				    ARM_SMMU_GR0_sGFSR;
121 
122 		/* clear global FSR */
123 		val = readl_relaxed(reg);
124 		writel_relaxed(val, reg);
125 	}
126 
127 	return 0;
128 }
129 
130 static irqreturn_t nvidia_smmu_global_fault_inst(int irq,
131 						 struct arm_smmu_device *smmu,
132 						 int inst)
133 {
134 	u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
135 	void __iomem *gr0_base = nvidia_smmu_page(smmu, inst, 0);
136 
137 	gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
138 	if (!gfsr)
139 		return IRQ_NONE;
140 
141 	gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
142 	gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
143 	gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
144 
145 	dev_err_ratelimited(smmu->dev,
146 			    "Unexpected global fault, this could be serious\n");
147 	dev_err_ratelimited(smmu->dev,
148 			    "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
149 			    gfsr, gfsynr0, gfsynr1, gfsynr2);
150 
151 	writel_relaxed(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
152 	return IRQ_HANDLED;
153 }
154 
155 static irqreturn_t nvidia_smmu_global_fault(int irq, void *dev)
156 {
157 	unsigned int inst;
158 	irqreturn_t ret = IRQ_NONE;
159 	struct arm_smmu_device *smmu = dev;
160 
161 	for (inst = 0; inst < NUM_SMMU_INSTANCES; inst++) {
162 		irqreturn_t irq_ret;
163 
164 		irq_ret = nvidia_smmu_global_fault_inst(irq, smmu, inst);
165 		if (irq_ret == IRQ_HANDLED)
166 			ret = IRQ_HANDLED;
167 	}
168 
169 	return ret;
170 }
171 
172 static irqreturn_t nvidia_smmu_context_fault_bank(int irq,
173 						  struct arm_smmu_device *smmu,
174 						  int idx, int inst)
175 {
176 	u32 fsr, fsynr, cbfrsynra;
177 	unsigned long iova;
178 	void __iomem *gr1_base = nvidia_smmu_page(smmu, inst, 1);
179 	void __iomem *cb_base = nvidia_smmu_page(smmu, inst, smmu->numpage + idx);
180 
181 	fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
182 	if (!(fsr & ARM_SMMU_FSR_FAULT))
183 		return IRQ_NONE;
184 
185 	fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
186 	iova = readq_relaxed(cb_base + ARM_SMMU_CB_FAR);
187 	cbfrsynra = readl_relaxed(gr1_base + ARM_SMMU_GR1_CBFRSYNRA(idx));
188 
189 	dev_err_ratelimited(smmu->dev,
190 			    "Unhandled context fault: fsr=0x%x, iova=0x%08lx, fsynr=0x%x, cbfrsynra=0x%x, cb=%d\n",
191 			    fsr, iova, fsynr, cbfrsynra, idx);
192 
193 	writel_relaxed(fsr, cb_base + ARM_SMMU_CB_FSR);
194 	return IRQ_HANDLED;
195 }
196 
197 static irqreturn_t nvidia_smmu_context_fault(int irq, void *dev)
198 {
199 	int idx;
200 	unsigned int inst;
201 	irqreturn_t ret = IRQ_NONE;
202 	struct arm_smmu_device *smmu;
203 	struct iommu_domain *domain = dev;
204 	struct arm_smmu_domain *smmu_domain;
205 
206 	smmu_domain = container_of(domain, struct arm_smmu_domain, domain);
207 	smmu = smmu_domain->smmu;
208 
209 	for (inst = 0; inst < NUM_SMMU_INSTANCES; inst++) {
210 		irqreturn_t irq_ret;
211 
212 		/*
213 		 * Interrupt line is shared between all contexts.
214 		 * Check for faults across all contexts.
215 		 */
216 		for (idx = 0; idx < smmu->num_context_banks; idx++) {
217 			irq_ret = nvidia_smmu_context_fault_bank(irq, smmu,
218 								 idx, inst);
219 			if (irq_ret == IRQ_HANDLED)
220 				ret = IRQ_HANDLED;
221 		}
222 	}
223 
224 	return ret;
225 }
226 
227 static const struct arm_smmu_impl nvidia_smmu_impl = {
228 	.read_reg = nvidia_smmu_read_reg,
229 	.write_reg = nvidia_smmu_write_reg,
230 	.read_reg64 = nvidia_smmu_read_reg64,
231 	.write_reg64 = nvidia_smmu_write_reg64,
232 	.reset = nvidia_smmu_reset,
233 	.tlb_sync = nvidia_smmu_tlb_sync,
234 	.global_fault = nvidia_smmu_global_fault,
235 	.context_fault = nvidia_smmu_context_fault,
236 };
237 
238 struct arm_smmu_device *nvidia_smmu_impl_init(struct arm_smmu_device *smmu)
239 {
240 	struct resource *res;
241 	struct device *dev = smmu->dev;
242 	struct nvidia_smmu *nvidia_smmu;
243 	struct platform_device *pdev = to_platform_device(dev);
244 
245 	nvidia_smmu = devm_krealloc(dev, smmu, sizeof(*nvidia_smmu), GFP_KERNEL);
246 	if (!nvidia_smmu)
247 		return ERR_PTR(-ENOMEM);
248 
249 	/* Instance 0 is ioremapped by arm-smmu.c. */
250 	nvidia_smmu->bases[0] = smmu->base;
251 
252 	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
253 	if (!res)
254 		return ERR_PTR(-ENODEV);
255 
256 	nvidia_smmu->bases[1] = devm_ioremap_resource(dev, res);
257 	if (IS_ERR(nvidia_smmu->bases[1]))
258 		return ERR_CAST(nvidia_smmu->bases[1]);
259 
260 	nvidia_smmu->smmu.impl = &nvidia_smmu_impl;
261 
262 	return &nvidia_smmu->smmu;
263 }
264