1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2019, The Linux Foundation. All rights reserved.
4  */
5 
6 #include <linux/acpi.h>
7 #include <linux/adreno-smmu-priv.h>
8 #include <linux/delay.h>
9 #include <linux/of_device.h>
10 #include <linux/qcom_scm.h>
11 
12 #include "arm-smmu.h"
13 #include "arm-smmu-qcom.h"
14 
15 #define QCOM_DUMMY_VAL	-1
16 
17 static struct qcom_smmu *to_qcom_smmu(struct arm_smmu_device *smmu)
18 {
19 	return container_of(smmu, struct qcom_smmu, smmu);
20 }
21 
22 static void qcom_smmu_tlb_sync(struct arm_smmu_device *smmu, int page,
23 				int sync, int status)
24 {
25 	unsigned int spin_cnt, delay;
26 	u32 reg;
27 
28 	arm_smmu_writel(smmu, page, sync, QCOM_DUMMY_VAL);
29 	for (delay = 1; delay < TLB_LOOP_TIMEOUT; delay *= 2) {
30 		for (spin_cnt = TLB_SPIN_COUNT; spin_cnt > 0; spin_cnt--) {
31 			reg = arm_smmu_readl(smmu, page, status);
32 			if (!(reg & ARM_SMMU_sTLBGSTATUS_GSACTIVE))
33 				return;
34 			cpu_relax();
35 		}
36 		udelay(delay);
37 	}
38 
39 	qcom_smmu_tlb_sync_debug(smmu);
40 }
41 
42 static void qcom_adreno_smmu_write_sctlr(struct arm_smmu_device *smmu, int idx,
43 		u32 reg)
44 {
45 	struct qcom_smmu *qsmmu = to_qcom_smmu(smmu);
46 
47 	/*
48 	 * On the GPU device we want to process subsequent transactions after a
49 	 * fault to keep the GPU from hanging
50 	 */
51 	reg |= ARM_SMMU_SCTLR_HUPCF;
52 
53 	if (qsmmu->stall_enabled & BIT(idx))
54 		reg |= ARM_SMMU_SCTLR_CFCFG;
55 
56 	arm_smmu_cb_write(smmu, idx, ARM_SMMU_CB_SCTLR, reg);
57 }
58 
59 static void qcom_adreno_smmu_get_fault_info(const void *cookie,
60 		struct adreno_smmu_fault_info *info)
61 {
62 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
63 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
64 	struct arm_smmu_device *smmu = smmu_domain->smmu;
65 
66 	info->fsr = arm_smmu_cb_read(smmu, cfg->cbndx, ARM_SMMU_CB_FSR);
67 	info->fsynr0 = arm_smmu_cb_read(smmu, cfg->cbndx, ARM_SMMU_CB_FSYNR0);
68 	info->fsynr1 = arm_smmu_cb_read(smmu, cfg->cbndx, ARM_SMMU_CB_FSYNR1);
69 	info->far = arm_smmu_cb_readq(smmu, cfg->cbndx, ARM_SMMU_CB_FAR);
70 	info->cbfrsynra = arm_smmu_gr1_read(smmu, ARM_SMMU_GR1_CBFRSYNRA(cfg->cbndx));
71 	info->ttbr0 = arm_smmu_cb_readq(smmu, cfg->cbndx, ARM_SMMU_CB_TTBR0);
72 	info->contextidr = arm_smmu_cb_read(smmu, cfg->cbndx, ARM_SMMU_CB_CONTEXTIDR);
73 }
74 
75 static void qcom_adreno_smmu_set_stall(const void *cookie, bool enabled)
76 {
77 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
78 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
79 	struct qcom_smmu *qsmmu = to_qcom_smmu(smmu_domain->smmu);
80 
81 	if (enabled)
82 		qsmmu->stall_enabled |= BIT(cfg->cbndx);
83 	else
84 		qsmmu->stall_enabled &= ~BIT(cfg->cbndx);
85 }
86 
87 static void qcom_adreno_smmu_resume_translation(const void *cookie, bool terminate)
88 {
89 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
90 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
91 	struct arm_smmu_device *smmu = smmu_domain->smmu;
92 	u32 reg = 0;
93 
94 	if (terminate)
95 		reg |= ARM_SMMU_RESUME_TERMINATE;
96 
97 	arm_smmu_cb_write(smmu, cfg->cbndx, ARM_SMMU_CB_RESUME, reg);
98 }
99 
100 #define QCOM_ADRENO_SMMU_GPU_SID 0
101 
102 static bool qcom_adreno_smmu_is_gpu_device(struct device *dev)
103 {
104 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
105 	int i;
106 
107 	/*
108 	 * The GPU will always use SID 0 so that is a handy way to uniquely
109 	 * identify it and configure it for per-instance pagetables
110 	 */
111 	for (i = 0; i < fwspec->num_ids; i++) {
112 		u16 sid = FIELD_GET(ARM_SMMU_SMR_ID, fwspec->ids[i]);
113 
114 		if (sid == QCOM_ADRENO_SMMU_GPU_SID)
115 			return true;
116 	}
117 
118 	return false;
119 }
120 
121 static const struct io_pgtable_cfg *qcom_adreno_smmu_get_ttbr1_cfg(
122 		const void *cookie)
123 {
124 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
125 	struct io_pgtable *pgtable =
126 		io_pgtable_ops_to_pgtable(smmu_domain->pgtbl_ops);
127 	return &pgtable->cfg;
128 }
129 
130 /*
131  * Local implementation to configure TTBR0 with the specified pagetable config.
132  * The GPU driver will call this to enable TTBR0 when per-instance pagetables
133  * are active
134  */
135 
136 static int qcom_adreno_smmu_set_ttbr0_cfg(const void *cookie,
137 		const struct io_pgtable_cfg *pgtbl_cfg)
138 {
139 	struct arm_smmu_domain *smmu_domain = (void *)cookie;
140 	struct io_pgtable *pgtable = io_pgtable_ops_to_pgtable(smmu_domain->pgtbl_ops);
141 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
142 	struct arm_smmu_cb *cb = &smmu_domain->smmu->cbs[cfg->cbndx];
143 
144 	/* The domain must have split pagetables already enabled */
145 	if (cb->tcr[0] & ARM_SMMU_TCR_EPD1)
146 		return -EINVAL;
147 
148 	/* If the pagetable config is NULL, disable TTBR0 */
149 	if (!pgtbl_cfg) {
150 		/* Do nothing if it is already disabled */
151 		if ((cb->tcr[0] & ARM_SMMU_TCR_EPD0))
152 			return -EINVAL;
153 
154 		/* Set TCR to the original configuration */
155 		cb->tcr[0] = arm_smmu_lpae_tcr(&pgtable->cfg);
156 		cb->ttbr[0] = FIELD_PREP(ARM_SMMU_TTBRn_ASID, cb->cfg->asid);
157 	} else {
158 		u32 tcr = cb->tcr[0];
159 
160 		/* Don't call this again if TTBR0 is already enabled */
161 		if (!(cb->tcr[0] & ARM_SMMU_TCR_EPD0))
162 			return -EINVAL;
163 
164 		tcr |= arm_smmu_lpae_tcr(pgtbl_cfg);
165 		tcr &= ~(ARM_SMMU_TCR_EPD0 | ARM_SMMU_TCR_EPD1);
166 
167 		cb->tcr[0] = tcr;
168 		cb->ttbr[0] = pgtbl_cfg->arm_lpae_s1_cfg.ttbr;
169 		cb->ttbr[0] |= FIELD_PREP(ARM_SMMU_TTBRn_ASID, cb->cfg->asid);
170 	}
171 
172 	arm_smmu_write_context_bank(smmu_domain->smmu, cb->cfg->cbndx);
173 
174 	return 0;
175 }
176 
177 static int qcom_adreno_smmu_alloc_context_bank(struct arm_smmu_domain *smmu_domain,
178 					       struct arm_smmu_device *smmu,
179 					       struct device *dev, int start)
180 {
181 	int count;
182 
183 	/*
184 	 * Assign context bank 0 to the GPU device so the GPU hardware can
185 	 * switch pagetables
186 	 */
187 	if (qcom_adreno_smmu_is_gpu_device(dev)) {
188 		start = 0;
189 		count = 1;
190 	} else {
191 		start = 1;
192 		count = smmu->num_context_banks;
193 	}
194 
195 	return __arm_smmu_alloc_bitmap(smmu->context_map, start, count);
196 }
197 
198 static bool qcom_adreno_can_do_ttbr1(struct arm_smmu_device *smmu)
199 {
200 	const struct device_node *np = smmu->dev->of_node;
201 
202 	if (of_device_is_compatible(np, "qcom,msm8996-smmu-v2"))
203 		return false;
204 
205 	return true;
206 }
207 
208 static int qcom_adreno_smmu_init_context(struct arm_smmu_domain *smmu_domain,
209 		struct io_pgtable_cfg *pgtbl_cfg, struct device *dev)
210 {
211 	struct adreno_smmu_priv *priv;
212 
213 	smmu_domain->cfg.flush_walk_prefer_tlbiasid = true;
214 
215 	/* Only enable split pagetables for the GPU device (SID 0) */
216 	if (!qcom_adreno_smmu_is_gpu_device(dev))
217 		return 0;
218 
219 	/*
220 	 * All targets that use the qcom,adreno-smmu compatible string *should*
221 	 * be AARCH64 stage 1 but double check because the arm-smmu code assumes
222 	 * that is the case when the TTBR1 quirk is enabled
223 	 */
224 	if (qcom_adreno_can_do_ttbr1(smmu_domain->smmu) &&
225 	    (smmu_domain->stage == ARM_SMMU_DOMAIN_S1) &&
226 	    (smmu_domain->cfg.fmt == ARM_SMMU_CTX_FMT_AARCH64))
227 		pgtbl_cfg->quirks |= IO_PGTABLE_QUIRK_ARM_TTBR1;
228 
229 	/*
230 	 * Initialize private interface with GPU:
231 	 */
232 
233 	priv = dev_get_drvdata(dev);
234 	priv->cookie = smmu_domain;
235 	priv->get_ttbr1_cfg = qcom_adreno_smmu_get_ttbr1_cfg;
236 	priv->set_ttbr0_cfg = qcom_adreno_smmu_set_ttbr0_cfg;
237 	priv->get_fault_info = qcom_adreno_smmu_get_fault_info;
238 	priv->set_stall = qcom_adreno_smmu_set_stall;
239 	priv->resume_translation = qcom_adreno_smmu_resume_translation;
240 
241 	return 0;
242 }
243 
244 static const struct of_device_id qcom_smmu_client_of_match[] __maybe_unused = {
245 	{ .compatible = "qcom,adreno" },
246 	{ .compatible = "qcom,mdp4" },
247 	{ .compatible = "qcom,mdss" },
248 	{ .compatible = "qcom,sc7180-mdss" },
249 	{ .compatible = "qcom,sc7180-mss-pil" },
250 	{ .compatible = "qcom,sc7280-mdss" },
251 	{ .compatible = "qcom,sc7280-mss-pil" },
252 	{ .compatible = "qcom,sc8180x-mdss" },
253 	{ .compatible = "qcom,sm8250-mdss" },
254 	{ .compatible = "qcom,sdm845-mdss" },
255 	{ .compatible = "qcom,sdm845-mss-pil" },
256 	{ }
257 };
258 
259 static int qcom_smmu_init_context(struct arm_smmu_domain *smmu_domain,
260 		struct io_pgtable_cfg *pgtbl_cfg, struct device *dev)
261 {
262 	smmu_domain->cfg.flush_walk_prefer_tlbiasid = true;
263 
264 	return 0;
265 }
266 
267 static int qcom_smmu_cfg_probe(struct arm_smmu_device *smmu)
268 {
269 	unsigned int last_s2cr = ARM_SMMU_GR0_S2CR(smmu->num_mapping_groups - 1);
270 	struct qcom_smmu *qsmmu = to_qcom_smmu(smmu);
271 	u32 reg;
272 	u32 smr;
273 	int i;
274 
275 	/*
276 	 * With some firmware versions writes to S2CR of type FAULT are
277 	 * ignored, and writing BYPASS will end up written as FAULT in the
278 	 * register. Perform a write to S2CR to detect if this is the case and
279 	 * if so reserve a context bank to emulate bypass streams.
280 	 */
281 	reg = FIELD_PREP(ARM_SMMU_S2CR_TYPE, S2CR_TYPE_BYPASS) |
282 	      FIELD_PREP(ARM_SMMU_S2CR_CBNDX, 0xff) |
283 	      FIELD_PREP(ARM_SMMU_S2CR_PRIVCFG, S2CR_PRIVCFG_DEFAULT);
284 	arm_smmu_gr0_write(smmu, last_s2cr, reg);
285 	reg = arm_smmu_gr0_read(smmu, last_s2cr);
286 	if (FIELD_GET(ARM_SMMU_S2CR_TYPE, reg) != S2CR_TYPE_BYPASS) {
287 		qsmmu->bypass_quirk = true;
288 		qsmmu->bypass_cbndx = smmu->num_context_banks - 1;
289 
290 		set_bit(qsmmu->bypass_cbndx, smmu->context_map);
291 
292 		arm_smmu_cb_write(smmu, qsmmu->bypass_cbndx, ARM_SMMU_CB_SCTLR, 0);
293 
294 		reg = FIELD_PREP(ARM_SMMU_CBAR_TYPE, CBAR_TYPE_S1_TRANS_S2_BYPASS);
295 		arm_smmu_gr1_write(smmu, ARM_SMMU_GR1_CBAR(qsmmu->bypass_cbndx), reg);
296 	}
297 
298 	for (i = 0; i < smmu->num_mapping_groups; i++) {
299 		smr = arm_smmu_gr0_read(smmu, ARM_SMMU_GR0_SMR(i));
300 
301 		if (FIELD_GET(ARM_SMMU_SMR_VALID, smr)) {
302 			/* Ignore valid bit for SMR mask extraction. */
303 			smr &= ~ARM_SMMU_SMR_VALID;
304 			smmu->smrs[i].id = FIELD_GET(ARM_SMMU_SMR_ID, smr);
305 			smmu->smrs[i].mask = FIELD_GET(ARM_SMMU_SMR_MASK, smr);
306 			smmu->smrs[i].valid = true;
307 
308 			smmu->s2crs[i].type = S2CR_TYPE_BYPASS;
309 			smmu->s2crs[i].privcfg = S2CR_PRIVCFG_DEFAULT;
310 			smmu->s2crs[i].cbndx = 0xff;
311 		}
312 	}
313 
314 	return 0;
315 }
316 
317 static void qcom_smmu_write_s2cr(struct arm_smmu_device *smmu, int idx)
318 {
319 	struct arm_smmu_s2cr *s2cr = smmu->s2crs + idx;
320 	struct qcom_smmu *qsmmu = to_qcom_smmu(smmu);
321 	u32 cbndx = s2cr->cbndx;
322 	u32 type = s2cr->type;
323 	u32 reg;
324 
325 	if (qsmmu->bypass_quirk) {
326 		if (type == S2CR_TYPE_BYPASS) {
327 			/*
328 			 * Firmware with quirky S2CR handling will substitute
329 			 * BYPASS writes with FAULT, so point the stream to the
330 			 * reserved context bank and ask for translation on the
331 			 * stream
332 			 */
333 			type = S2CR_TYPE_TRANS;
334 			cbndx = qsmmu->bypass_cbndx;
335 		} else if (type == S2CR_TYPE_FAULT) {
336 			/*
337 			 * Firmware with quirky S2CR handling will ignore FAULT
338 			 * writes, so trick it to write FAULT by asking for a
339 			 * BYPASS.
340 			 */
341 			type = S2CR_TYPE_BYPASS;
342 			cbndx = 0xff;
343 		}
344 	}
345 
346 	reg = FIELD_PREP(ARM_SMMU_S2CR_TYPE, type) |
347 	      FIELD_PREP(ARM_SMMU_S2CR_CBNDX, cbndx) |
348 	      FIELD_PREP(ARM_SMMU_S2CR_PRIVCFG, s2cr->privcfg);
349 	arm_smmu_gr0_write(smmu, ARM_SMMU_GR0_S2CR(idx), reg);
350 }
351 
352 static int qcom_smmu_def_domain_type(struct device *dev)
353 {
354 	const struct of_device_id *match =
355 		of_match_device(qcom_smmu_client_of_match, dev);
356 
357 	return match ? IOMMU_DOMAIN_IDENTITY : 0;
358 }
359 
360 static int qcom_sdm845_smmu500_reset(struct arm_smmu_device *smmu)
361 {
362 	int ret;
363 
364 	/*
365 	 * To address performance degradation in non-real time clients,
366 	 * such as USB and UFS, turn off wait-for-safe on sdm845 based boards,
367 	 * such as MTP and db845, whose firmwares implement secure monitor
368 	 * call handlers to turn on/off the wait-for-safe logic.
369 	 */
370 	ret = qcom_scm_qsmmu500_wait_safe_toggle(0);
371 	if (ret)
372 		dev_warn(smmu->dev, "Failed to turn off SAFE logic\n");
373 
374 	return ret;
375 }
376 
377 static int qcom_smmu500_reset(struct arm_smmu_device *smmu)
378 {
379 	const struct device_node *np = smmu->dev->of_node;
380 
381 	arm_mmu500_reset(smmu);
382 
383 	if (of_device_is_compatible(np, "qcom,sdm845-smmu-500"))
384 		return qcom_sdm845_smmu500_reset(smmu);
385 
386 	return 0;
387 }
388 
389 static const struct arm_smmu_impl qcom_smmu_impl = {
390 	.init_context = qcom_smmu_init_context,
391 	.cfg_probe = qcom_smmu_cfg_probe,
392 	.def_domain_type = qcom_smmu_def_domain_type,
393 	.reset = qcom_smmu500_reset,
394 	.write_s2cr = qcom_smmu_write_s2cr,
395 	.tlb_sync = qcom_smmu_tlb_sync,
396 };
397 
398 static const struct arm_smmu_impl qcom_adreno_smmu_impl = {
399 	.init_context = qcom_adreno_smmu_init_context,
400 	.def_domain_type = qcom_smmu_def_domain_type,
401 	.reset = qcom_smmu500_reset,
402 	.alloc_context_bank = qcom_adreno_smmu_alloc_context_bank,
403 	.write_sctlr = qcom_adreno_smmu_write_sctlr,
404 	.tlb_sync = qcom_smmu_tlb_sync,
405 };
406 
407 static struct arm_smmu_device *qcom_smmu_create(struct arm_smmu_device *smmu,
408 		const struct arm_smmu_impl *impl)
409 {
410 	struct qcom_smmu *qsmmu;
411 
412 	/* Check to make sure qcom_scm has finished probing */
413 	if (!qcom_scm_is_available())
414 		return ERR_PTR(-EPROBE_DEFER);
415 
416 	qsmmu = devm_krealloc(smmu->dev, smmu, sizeof(*qsmmu), GFP_KERNEL);
417 	if (!qsmmu)
418 		return ERR_PTR(-ENOMEM);
419 
420 	qsmmu->smmu.impl = impl;
421 	qsmmu->cfg = qcom_smmu_impl_data(smmu);
422 
423 	return &qsmmu->smmu;
424 }
425 
426 static const struct of_device_id __maybe_unused qcom_smmu_impl_of_match[] = {
427 	{ .compatible = "qcom,msm8998-smmu-v2" },
428 	{ .compatible = "qcom,qcm2290-smmu-500" },
429 	{ .compatible = "qcom,sc7180-smmu-500" },
430 	{ .compatible = "qcom,sc7280-smmu-500" },
431 	{ .compatible = "qcom,sc8180x-smmu-500" },
432 	{ .compatible = "qcom,sc8280xp-smmu-500" },
433 	{ .compatible = "qcom,sdm630-smmu-v2" },
434 	{ .compatible = "qcom,sdm845-smmu-500" },
435 	{ .compatible = "qcom,sm6125-smmu-500" },
436 	{ .compatible = "qcom,sm6350-smmu-500" },
437 	{ .compatible = "qcom,sm6375-smmu-500" },
438 	{ .compatible = "qcom,sm8150-smmu-500" },
439 	{ .compatible = "qcom,sm8250-smmu-500" },
440 	{ .compatible = "qcom,sm8350-smmu-500" },
441 	{ .compatible = "qcom,sm8450-smmu-500" },
442 	{ }
443 };
444 
445 #ifdef CONFIG_ACPI
446 static struct acpi_platform_list qcom_acpi_platlist[] = {
447 	{ "LENOVO", "CB-01   ", 0x8180, ACPI_SIG_IORT, equal, "QCOM SMMU" },
448 	{ "QCOM  ", "QCOMEDK2", 0x8180, ACPI_SIG_IORT, equal, "QCOM SMMU" },
449 	{ }
450 };
451 #endif
452 
453 struct arm_smmu_device *qcom_smmu_impl_init(struct arm_smmu_device *smmu)
454 {
455 	const struct device_node *np = smmu->dev->of_node;
456 
457 #ifdef CONFIG_ACPI
458 	if (np == NULL) {
459 		/* Match platform for ACPI boot */
460 		if (acpi_match_platform_list(qcom_acpi_platlist) >= 0)
461 			return qcom_smmu_create(smmu, &qcom_smmu_impl);
462 	}
463 #endif
464 
465 	/*
466 	 * Do not change this order of implementation, i.e., first adreno
467 	 * smmu impl and then apss smmu since we can have both implementing
468 	 * arm,mmu-500 in which case we will miss setting adreno smmu specific
469 	 * features if the order is changed.
470 	 */
471 	if (of_device_is_compatible(np, "qcom,adreno-smmu"))
472 		return qcom_smmu_create(smmu, &qcom_adreno_smmu_impl);
473 
474 	if (of_match_node(qcom_smmu_impl_of_match, np))
475 		return qcom_smmu_create(smmu, &qcom_smmu_impl);
476 
477 	return smmu;
478 }
479