xref: /OK3568_Linux_fs/kernel/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2019, The Linux Foundation. All rights reserved.
4  */
5 
6 #include <linux/adreno-smmu-priv.h>
7 #include <linux/bitfield.h>
8 #include <linux/of_device.h>
9 #include <linux/qcom_scm.h>
10 
11 #include "arm-smmu.h"
12 
13 struct qcom_smmu {
14 	struct arm_smmu_device smmu;
15 	bool bypass_quirk;
16 	u8 bypass_cbndx;
17 };
18 
qcom_sdm845_smmu500_cfg_probe(struct arm_smmu_device * smmu)19 static int qcom_sdm845_smmu500_cfg_probe(struct arm_smmu_device *smmu)
20 {
21 	u32 s2cr;
22 	u32 smr;
23 	int i;
24 
25 	for (i = 0; i < smmu->num_mapping_groups; i++) {
26 		smr = arm_smmu_gr0_read(smmu, ARM_SMMU_GR0_SMR(i));
27 		s2cr = arm_smmu_gr0_read(smmu, ARM_SMMU_GR0_S2CR(i));
28 
29 		smmu->smrs[i].mask = FIELD_GET(ARM_SMMU_SMR_MASK, smr);
30 		smmu->smrs[i].id = FIELD_GET(ARM_SMMU_SMR_ID, smr);
31 		if (smmu->features & ARM_SMMU_FEAT_EXIDS)
32 			smmu->smrs[i].valid = FIELD_GET(
33 						ARM_SMMU_S2CR_EXIDVALID,
34 						s2cr);
35 		else
36 			smmu->smrs[i].valid = FIELD_GET(
37 						ARM_SMMU_SMR_VALID,
38 						smr);
39 
40 		smmu->s2crs[i].group = NULL;
41 		smmu->s2crs[i].count = 0;
42 		smmu->s2crs[i].type = FIELD_GET(ARM_SMMU_S2CR_TYPE, s2cr);
43 		smmu->s2crs[i].privcfg = FIELD_GET(ARM_SMMU_S2CR_PRIVCFG, s2cr);
44 		smmu->s2crs[i].cbndx = FIELD_GET(ARM_SMMU_S2CR_CBNDX, s2cr);
45 
46 		if (!smmu->smrs[i].valid)
47 			continue;
48 
49 		smmu->s2crs[i].pinned = true;
50 		bitmap_set(smmu->context_map, smmu->s2crs[i].cbndx, 1);
51 	}
52 
53 	return 0;
54 }
55 
56 #define QCOM_ADRENO_SMMU_GPU_SID 0
57 
qcom_adreno_smmu_is_gpu_device(struct device * dev)58 static bool qcom_adreno_smmu_is_gpu_device(struct device *dev)
59 {
60 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
61 	int i;
62 
63 	/*
64 	 * The GPU will always use SID 0 so that is a handy way to uniquely
65 	 * identify it and configure it for per-instance pagetables
66 	 */
67 	for (i = 0; i < fwspec->num_ids; i++) {
68 		u16 sid = FIELD_GET(ARM_SMMU_SMR_ID, fwspec->ids[i]);
69 
70 		if (sid == QCOM_ADRENO_SMMU_GPU_SID)
71 			return true;
72 	}
73 
74 	return false;
75 }
76 
qcom_adreno_smmu_get_ttbr1_cfg(const void * cookie)77 static const struct io_pgtable_cfg *qcom_adreno_smmu_get_ttbr1_cfg(
78 		const void *cookie)
79 {
80 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
81 	struct io_pgtable *pgtable =
82 		io_pgtable_ops_to_pgtable(smmu_domain->pgtbl_ops);
83 	return &pgtable->cfg;
84 }
85 
86 /*
87  * Local implementation to configure TTBR0 with the specified pagetable config.
88  * The GPU driver will call this to enable TTBR0 when per-instance pagetables
89  * are active
90  */
91 
qcom_adreno_smmu_set_ttbr0_cfg(const void * cookie,const struct io_pgtable_cfg * pgtbl_cfg)92 static int qcom_adreno_smmu_set_ttbr0_cfg(const void *cookie,
93 		const struct io_pgtable_cfg *pgtbl_cfg)
94 {
95 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
96 	struct io_pgtable *pgtable = io_pgtable_ops_to_pgtable(smmu_domain->pgtbl_ops);
97 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
98 	struct arm_smmu_cb *cb = &smmu_domain->smmu->cbs[cfg->cbndx];
99 
100 	/* The domain must have split pagetables already enabled */
101 	if (cb->tcr[0] & ARM_SMMU_TCR_EPD1)
102 		return -EINVAL;
103 
104 	/* If the pagetable config is NULL, disable TTBR0 */
105 	if (!pgtbl_cfg) {
106 		/* Do nothing if it is already disabled */
107 		if ((cb->tcr[0] & ARM_SMMU_TCR_EPD0))
108 			return -EINVAL;
109 
110 		/* Set TCR to the original configuration */
111 		cb->tcr[0] = arm_smmu_lpae_tcr(&pgtable->cfg);
112 		cb->ttbr[0] = FIELD_PREP(ARM_SMMU_TTBRn_ASID, cb->cfg->asid);
113 	} else {
114 		u32 tcr = cb->tcr[0];
115 
116 		/* Don't call this again if TTBR0 is already enabled */
117 		if (!(cb->tcr[0] & ARM_SMMU_TCR_EPD0))
118 			return -EINVAL;
119 
120 		tcr |= arm_smmu_lpae_tcr(pgtbl_cfg);
121 		tcr &= ~(ARM_SMMU_TCR_EPD0 | ARM_SMMU_TCR_EPD1);
122 
123 		cb->tcr[0] = tcr;
124 		cb->ttbr[0] = pgtbl_cfg->arm_lpae_s1_cfg.ttbr;
125 		cb->ttbr[0] |= FIELD_PREP(ARM_SMMU_TTBRn_ASID, cb->cfg->asid);
126 	}
127 
128 	arm_smmu_write_context_bank(smmu_domain->smmu, cb->cfg->cbndx);
129 
130 	return 0;
131 }
132 
qcom_adreno_smmu_alloc_context_bank(struct arm_smmu_domain * smmu_domain,struct arm_smmu_device * smmu,struct device * dev,int start)133 static int qcom_adreno_smmu_alloc_context_bank(struct arm_smmu_domain *smmu_domain,
134 					       struct arm_smmu_device *smmu,
135 					       struct device *dev, int start)
136 {
137 	int count;
138 
139 	/*
140 	 * Assign context bank 0 to the GPU device so the GPU hardware can
141 	 * switch pagetables
142 	 */
143 	if (qcom_adreno_smmu_is_gpu_device(dev)) {
144 		start = 0;
145 		count = 1;
146 	} else {
147 		start = 1;
148 		count = smmu->num_context_banks;
149 	}
150 
151 	return __arm_smmu_alloc_bitmap(smmu->context_map, start, count);
152 }
153 
qcom_adreno_smmu_init_context(struct arm_smmu_domain * smmu_domain,struct io_pgtable_cfg * pgtbl_cfg,struct device * dev)154 static int qcom_adreno_smmu_init_context(struct arm_smmu_domain *smmu_domain,
155 		struct io_pgtable_cfg *pgtbl_cfg, struct device *dev)
156 {
157 	struct adreno_smmu_priv *priv;
158 
159 	/* Only enable split pagetables for the GPU device (SID 0) */
160 	if (!qcom_adreno_smmu_is_gpu_device(dev))
161 		return 0;
162 
163 	/*
164 	 * All targets that use the qcom,adreno-smmu compatible string *should*
165 	 * be AARCH64 stage 1 but double check because the arm-smmu code assumes
166 	 * that is the case when the TTBR1 quirk is enabled
167 	 */
168 	if ((smmu_domain->stage == ARM_SMMU_DOMAIN_S1) &&
169 	    (smmu_domain->cfg.fmt == ARM_SMMU_CTX_FMT_AARCH64))
170 		pgtbl_cfg->quirks |= IO_PGTABLE_QUIRK_ARM_TTBR1;
171 
172 	/*
173 	 * Initialize private interface with GPU:
174 	 */
175 
176 	priv = dev_get_drvdata(dev);
177 	priv->cookie = smmu_domain;
178 	priv->get_ttbr1_cfg = qcom_adreno_smmu_get_ttbr1_cfg;
179 	priv->set_ttbr0_cfg = qcom_adreno_smmu_set_ttbr0_cfg;
180 
181 	return 0;
182 }
183 
to_qcom_smmu(struct arm_smmu_device * smmu)184 static struct qcom_smmu *to_qcom_smmu(struct arm_smmu_device *smmu)
185 {
186 	return container_of(smmu, struct qcom_smmu, smmu);
187 }
188 
189 static const struct of_device_id qcom_smmu_client_of_match[] __maybe_unused = {
190 	{ .compatible = "qcom,adreno" },
191 	{ .compatible = "qcom,mdp4" },
192 	{ .compatible = "qcom,mdss" },
193 	{ .compatible = "qcom,sc7180-mdss" },
194 	{ .compatible = "qcom,sc7180-mss-pil" },
195 	{ .compatible = "qcom,sdm845-mdss" },
196 	{ .compatible = "qcom,sdm845-mss-pil" },
197 	{ }
198 };
199 
qcom_smmu_cfg_probe(struct arm_smmu_device * smmu)200 static int qcom_smmu_cfg_probe(struct arm_smmu_device *smmu)
201 {
202 	unsigned int last_s2cr = ARM_SMMU_GR0_S2CR(smmu->num_mapping_groups - 1);
203 	struct qcom_smmu *qsmmu = to_qcom_smmu(smmu);
204 	u32 reg;
205 	u32 smr;
206 	int i;
207 
208 	/*
209 	 * With some firmware versions writes to S2CR of type FAULT are
210 	 * ignored, and writing BYPASS will end up written as FAULT in the
211 	 * register. Perform a write to S2CR to detect if this is the case and
212 	 * if so reserve a context bank to emulate bypass streams.
213 	 */
214 	reg = FIELD_PREP(ARM_SMMU_S2CR_TYPE, S2CR_TYPE_BYPASS) |
215 	      FIELD_PREP(ARM_SMMU_S2CR_CBNDX, 0xff) |
216 	      FIELD_PREP(ARM_SMMU_S2CR_PRIVCFG, S2CR_PRIVCFG_DEFAULT);
217 	arm_smmu_gr0_write(smmu, last_s2cr, reg);
218 	reg = arm_smmu_gr0_read(smmu, last_s2cr);
219 	if (FIELD_GET(ARM_SMMU_S2CR_TYPE, reg) != S2CR_TYPE_BYPASS) {
220 		qsmmu->bypass_quirk = true;
221 		qsmmu->bypass_cbndx = smmu->num_context_banks - 1;
222 
223 		set_bit(qsmmu->bypass_cbndx, smmu->context_map);
224 
225 		arm_smmu_cb_write(smmu, qsmmu->bypass_cbndx, ARM_SMMU_CB_SCTLR, 0);
226 
227 		reg = FIELD_PREP(ARM_SMMU_CBAR_TYPE, CBAR_TYPE_S1_TRANS_S2_BYPASS);
228 		arm_smmu_gr1_write(smmu, ARM_SMMU_GR1_CBAR(qsmmu->bypass_cbndx), reg);
229 	}
230 
231 	for (i = 0; i < smmu->num_mapping_groups; i++) {
232 		smr = arm_smmu_gr0_read(smmu, ARM_SMMU_GR0_SMR(i));
233 
234 		if (FIELD_GET(ARM_SMMU_SMR_VALID, smr)) {
235 			/* Ignore valid bit for SMR mask extraction. */
236 			smr &= ~ARM_SMMU_SMR_VALID;
237 			smmu->smrs[i].id = FIELD_GET(ARM_SMMU_SMR_ID, smr);
238 			smmu->smrs[i].mask = FIELD_GET(ARM_SMMU_SMR_MASK, smr);
239 			smmu->smrs[i].valid = true;
240 
241 			smmu->s2crs[i].type = S2CR_TYPE_BYPASS;
242 			smmu->s2crs[i].privcfg = S2CR_PRIVCFG_DEFAULT;
243 			smmu->s2crs[i].cbndx = 0xff;
244 		}
245 	}
246 
247 	return 0;
248 }
249 
qcom_smmu_write_s2cr(struct arm_smmu_device * smmu,int idx)250 static void qcom_smmu_write_s2cr(struct arm_smmu_device *smmu, int idx)
251 {
252 	struct arm_smmu_s2cr *s2cr = smmu->s2crs + idx;
253 	struct qcom_smmu *qsmmu = to_qcom_smmu(smmu);
254 	u32 cbndx = s2cr->cbndx;
255 	u32 type = s2cr->type;
256 	u32 reg;
257 
258 	if (qsmmu->bypass_quirk) {
259 		if (type == S2CR_TYPE_BYPASS) {
260 			/*
261 			 * Firmware with quirky S2CR handling will substitute
262 			 * BYPASS writes with FAULT, so point the stream to the
263 			 * reserved context bank and ask for translation on the
264 			 * stream
265 			 */
266 			type = S2CR_TYPE_TRANS;
267 			cbndx = qsmmu->bypass_cbndx;
268 		} else if (type == S2CR_TYPE_FAULT) {
269 			/*
270 			 * Firmware with quirky S2CR handling will ignore FAULT
271 			 * writes, so trick it to write FAULT by asking for a
272 			 * BYPASS.
273 			 */
274 			type = S2CR_TYPE_BYPASS;
275 			cbndx = 0xff;
276 		}
277 	}
278 
279 	reg = FIELD_PREP(ARM_SMMU_S2CR_TYPE, type) |
280 	      FIELD_PREP(ARM_SMMU_S2CR_CBNDX, cbndx) |
281 	      FIELD_PREP(ARM_SMMU_S2CR_PRIVCFG, s2cr->privcfg);
282 	arm_smmu_gr0_write(smmu, ARM_SMMU_GR0_S2CR(idx), reg);
283 }
284 
qcom_smmu_def_domain_type(struct device * dev)285 static int qcom_smmu_def_domain_type(struct device *dev)
286 {
287 	const struct of_device_id *match =
288 		of_match_device(qcom_smmu_client_of_match, dev);
289 
290 	return match ? IOMMU_DOMAIN_IDENTITY : 0;
291 }
292 
qcom_sdm845_smmu500_reset(struct arm_smmu_device * smmu)293 static int qcom_sdm845_smmu500_reset(struct arm_smmu_device *smmu)
294 {
295 	int ret;
296 
297 	/*
298 	 * To address performance degradation in non-real time clients,
299 	 * such as USB and UFS, turn off wait-for-safe on sdm845 based boards,
300 	 * such as MTP and db845, whose firmwares implement secure monitor
301 	 * call handlers to turn on/off the wait-for-safe logic.
302 	 */
303 	ret = qcom_scm_qsmmu500_wait_safe_toggle(0);
304 	if (ret)
305 		dev_warn(smmu->dev, "Failed to turn off SAFE logic\n");
306 
307 	return ret;
308 }
309 
qcom_smmu500_reset(struct arm_smmu_device * smmu)310 static int qcom_smmu500_reset(struct arm_smmu_device *smmu)
311 {
312 	const struct device_node *np = smmu->dev->of_node;
313 
314 	arm_mmu500_reset(smmu);
315 
316 	if (of_device_is_compatible(np, "qcom,sdm845-smmu-500"))
317 		return qcom_sdm845_smmu500_reset(smmu);
318 
319 	return 0;
320 }
321 
322 static const struct arm_smmu_impl qcom_smmu_impl = {
323 	.cfg_probe = qcom_smmu_cfg_probe,
324 	.def_domain_type = qcom_smmu_def_domain_type,
325 	.cfg_probe = qcom_sdm845_smmu500_cfg_probe,
326 	.reset = qcom_smmu500_reset,
327 	.write_s2cr = qcom_smmu_write_s2cr,
328 };
329 
330 static const struct arm_smmu_impl qcom_adreno_smmu_impl = {
331 	.init_context = qcom_adreno_smmu_init_context,
332 	.def_domain_type = qcom_smmu_def_domain_type,
333 	.reset = qcom_smmu500_reset,
334 	.alloc_context_bank = qcom_adreno_smmu_alloc_context_bank,
335 };
336 
qcom_smmu_create(struct arm_smmu_device * smmu,const struct arm_smmu_impl * impl)337 static struct arm_smmu_device *qcom_smmu_create(struct arm_smmu_device *smmu,
338 		const struct arm_smmu_impl *impl)
339 {
340 	struct qcom_smmu *qsmmu;
341 
342 	/* Check to make sure qcom_scm has finished probing */
343 	if (!qcom_scm_is_available())
344 		return ERR_PTR(-EPROBE_DEFER);
345 
346 	qsmmu = devm_kzalloc(smmu->dev, sizeof(*qsmmu), GFP_KERNEL);
347 	if (!qsmmu)
348 		return ERR_PTR(-ENOMEM);
349 
350 	qsmmu->smmu = *smmu;
351 
352 	qsmmu->smmu.impl = impl;
353 	devm_kfree(smmu->dev, smmu);
354 
355 	return &qsmmu->smmu;
356 }
357 
qcom_smmu_impl_init(struct arm_smmu_device * smmu)358 struct arm_smmu_device *qcom_smmu_impl_init(struct arm_smmu_device *smmu)
359 {
360 	return qcom_smmu_create(smmu, &qcom_smmu_impl);
361 }
362 
qcom_adreno_smmu_impl_init(struct arm_smmu_device * smmu)363 struct arm_smmu_device *qcom_adreno_smmu_impl_init(struct arm_smmu_device *smmu)
364 {
365 	return qcom_smmu_create(smmu, &qcom_adreno_smmu_impl);
366 }
367