1 /*
2  * Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved.
3  *
4  * SPDX-License-Identifier: BSD-3-Clause
5  */
6 
7 #include <arch.h>
8 #include <assert.h>
9 #include <stdbool.h>
10 #include <string.h>
11 
12 #include <arch_helpers.h>
13 #include <bpmp_ipc.h>
14 #include <common/bl_common.h>
15 #include <common/debug.h>
16 #include <context.h>
17 #include <drivers/delay_timer.h>
18 #include <denver.h>
19 #include <lib/el3_runtime/context_mgmt.h>
20 #include <lib/psci/psci.h>
21 #include <mce.h>
22 #include <mce_private.h>
23 #include <memctrl_v2.h>
24 #include <plat/common/platform.h>
25 #include <se.h>
26 #include <smmu.h>
27 #include <t194_nvg.h>
28 #include <tegra194_private.h>
29 #include <tegra_platform.h>
30 #include <tegra_private.h>
31 
32 extern uint32_t __tegra194_cpu_reset_handler_data,
33 		__tegra194_cpu_reset_handler_end;
34 
35 /* TZDRAM offset for saving SMMU context */
36 #define TEGRA194_SMMU_CTX_OFFSET	16U
37 
38 /* state id mask */
39 #define TEGRA194_STATE_ID_MASK		0xFU
40 /* constants to get power state's wake time */
41 #define TEGRA194_WAKE_TIME_MASK		0x0FFFFFF0U
42 #define TEGRA194_WAKE_TIME_SHIFT	4U
43 /* default core wake mask for CPU_SUSPEND */
44 #define TEGRA194_CORE_WAKE_MASK		0x180cU
45 
46 static struct t19x_psci_percpu_data {
47 	uint32_t wake_time;
48 } __aligned(CACHE_WRITEBACK_GRANULE) t19x_percpu_data[PLATFORM_CORE_COUNT];
49 
tegra_soc_validate_power_state(uint32_t power_state,psci_power_state_t * req_state)50 int32_t tegra_soc_validate_power_state(uint32_t power_state,
51 					psci_power_state_t *req_state)
52 {
53 	uint8_t state_id = (uint8_t)psci_get_pstate_id(power_state) &
54 			   TEGRA194_STATE_ID_MASK;
55 	uint32_t cpu = plat_my_core_pos();
56 	int32_t ret = PSCI_E_SUCCESS;
57 
58 	/* save the core wake time (in TSC ticks)*/
59 	t19x_percpu_data[cpu].wake_time = (power_state & TEGRA194_WAKE_TIME_MASK)
60 			<< TEGRA194_WAKE_TIME_SHIFT;
61 
62 	/*
63 	 * Clean t19x_percpu_data[cpu] to DRAM. This needs to be done to ensure
64 	 * that the correct value is read in tegra_soc_pwr_domain_suspend(),
65 	 * which is called with caches disabled. It is possible to read a stale
66 	 * value from DRAM in that function, because the L2 cache is not flushed
67 	 * unless the cluster is entering CC6/CC7.
68 	 */
69 	clean_dcache_range((uint64_t)&t19x_percpu_data[cpu],
70 			sizeof(t19x_percpu_data[cpu]));
71 
72 	/* Sanity check the requested state id */
73 	switch (state_id) {
74 	case PSTATE_ID_CORE_IDLE:
75 
76 		if (psci_get_pstate_type(power_state) != PSTATE_TYPE_STANDBY) {
77 			ret = PSCI_E_INVALID_PARAMS;
78 			break;
79 		}
80 
81 		/* Core idle request */
82 		req_state->pwr_domain_state[MPIDR_AFFLVL0] = PLAT_MAX_RET_STATE;
83 		req_state->pwr_domain_state[MPIDR_AFFLVL1] = PSCI_LOCAL_STATE_RUN;
84 		break;
85 
86 	default:
87 		ERROR("%s: unsupported state id (%d)\n", __func__, state_id);
88 		ret = PSCI_E_INVALID_PARAMS;
89 		break;
90 	}
91 
92 	return ret;
93 }
94 
tegra_soc_cpu_standby(plat_local_state_t cpu_state)95 int32_t tegra_soc_cpu_standby(plat_local_state_t cpu_state)
96 {
97 	uint32_t cpu = plat_my_core_pos();
98 	mce_cstate_info_t cstate_info = { 0 };
99 
100 	/* Program default wake mask */
101 	cstate_info.wake_mask = TEGRA194_CORE_WAKE_MASK;
102 	cstate_info.update_wake_mask = 1;
103 	mce_update_cstate_info(&cstate_info);
104 
105 	/* Enter CPU idle */
106 	(void)mce_command_handler((uint64_t)MCE_CMD_ENTER_CSTATE,
107 				  (uint64_t)TEGRA_NVG_CORE_C6,
108 				  t19x_percpu_data[cpu].wake_time,
109 				  0U);
110 
111 	return PSCI_E_SUCCESS;
112 }
113 
tegra_soc_pwr_domain_suspend(const psci_power_state_t * target_state)114 int32_t tegra_soc_pwr_domain_suspend(const psci_power_state_t *target_state)
115 {
116 	const plat_local_state_t *pwr_domain_state;
117 	uint8_t stateid_afflvl2;
118 	plat_params_from_bl2_t *params_from_bl2 = bl31_get_plat_params();
119 	uint64_t mc_ctx_base;
120 	uint32_t val;
121 	mce_cstate_info_t sc7_cstate_info = {
122 		.cluster = (uint32_t)TEGRA_NVG_CLUSTER_CC6,
123 		.ccplex = (uint32_t)TEGRA_NVG_CG_CG7,
124 		.system = (uint32_t)TEGRA_NVG_SYSTEM_SC7,
125 		.system_state_force = 1U,
126 		.update_wake_mask = 1U,
127 	};
128 	int32_t ret = 0;
129 
130 	/* get the state ID */
131 	pwr_domain_state = target_state->pwr_domain_state;
132 	stateid_afflvl2 = pwr_domain_state[PLAT_MAX_PWR_LVL] &
133 		TEGRA194_STATE_ID_MASK;
134 
135 	if (stateid_afflvl2 == PSTATE_ID_SOC_POWERDN) {
136 
137 		/* save 'Secure Boot' Processor Feature Config Register */
138 		val = mmio_read_32(TEGRA_MISC_BASE + MISCREG_PFCFG);
139 		mmio_write_32(TEGRA_SCRATCH_BASE + SCRATCH_SECURE_BOOTP_FCFG, val);
140 
141 		/* save MC context */
142 		mc_ctx_base = params_from_bl2->tzdram_base +
143 				tegra194_get_mc_ctx_offset();
144 		tegra_mc_save_context((uintptr_t)mc_ctx_base);
145 
146 		/*
147 		 * Suspend SE, RNG1 and PKA1 only on silcon and fpga,
148 		 * since VDK does not support atomic se ctx save
149 		 */
150 		if (tegra_platform_is_silicon() || tegra_platform_is_fpga()) {
151 			ret = tegra_se_suspend();
152 			assert(ret == 0);
153 		}
154 
155 		/* Prepare for system suspend */
156 		mce_update_cstate_info(&sc7_cstate_info);
157 
158 		do {
159 			val = (uint32_t)mce_command_handler(
160 					(uint32_t)MCE_CMD_IS_SC7_ALLOWED,
161 					(uint32_t)TEGRA_NVG_CORE_C7,
162 					MCE_CORE_SLEEP_TIME_INFINITE,
163 					0U);
164 		} while (val == 0U);
165 
166 		/* Instruct the MCE to enter system suspend state */
167 		ret = mce_command_handler(
168 				(uint64_t)MCE_CMD_ENTER_CSTATE,
169 				(uint64_t)TEGRA_NVG_CORE_C7,
170 				MCE_CORE_SLEEP_TIME_INFINITE,
171 				0U);
172 		assert(ret == 0);
173 
174 		/* set system suspend state for house-keeping */
175 		tegra194_set_system_suspend_entry();
176 	}
177 
178 	return PSCI_E_SUCCESS;
179 }
180 
181 /*******************************************************************************
182  * Helper function to check if this is the last ON CPU in the cluster
183  ******************************************************************************/
tegra_last_on_cpu_in_cluster(const plat_local_state_t * states,uint32_t ncpu)184 static bool tegra_last_on_cpu_in_cluster(const plat_local_state_t *states,
185 			uint32_t ncpu)
186 {
187 	plat_local_state_t target;
188 	bool last_on_cpu = true;
189 	uint32_t num_cpus = ncpu, pos = 0;
190 
191 	do {
192 		target = states[pos];
193 		if (target != PLAT_MAX_OFF_STATE) {
194 			last_on_cpu = false;
195 		}
196 		--num_cpus;
197 		pos++;
198 	} while (num_cpus != 0U);
199 
200 	return last_on_cpu;
201 }
202 
203 /*******************************************************************************
204  * Helper function to get target power state for the cluster
205  ******************************************************************************/
tegra_get_afflvl1_pwr_state(const plat_local_state_t * states,uint32_t ncpu)206 static plat_local_state_t tegra_get_afflvl1_pwr_state(const plat_local_state_t *states,
207 			uint32_t ncpu)
208 {
209 	uint32_t core_pos = (uint32_t)read_mpidr() & (uint32_t)MPIDR_CPU_MASK;
210 	plat_local_state_t target = states[core_pos];
211 	mce_cstate_info_t cstate_info = { 0 };
212 
213 	/* CPU off */
214 	if (target == PLAT_MAX_OFF_STATE) {
215 
216 		/* Enable cluster powerdn from last CPU in the cluster */
217 		if (tegra_last_on_cpu_in_cluster(states, ncpu)) {
218 
219 			/* Enable CC6 state and turn off wake mask */
220 			cstate_info.cluster = (uint32_t)TEGRA_NVG_CLUSTER_CC6;
221 			cstate_info.ccplex = (uint32_t)TEGRA_NVG_CG_CG7;
222 			cstate_info.system_state_force = 1;
223 			cstate_info.update_wake_mask = 1U;
224 			mce_update_cstate_info(&cstate_info);
225 
226 		} else {
227 
228 			/* Turn off wake_mask */
229 			cstate_info.update_wake_mask = 1U;
230 			mce_update_cstate_info(&cstate_info);
231 			target = PSCI_LOCAL_STATE_RUN;
232 		}
233 	}
234 
235 	return target;
236 }
237 
238 /*******************************************************************************
239  * Platform handler to calculate the proper target power level at the
240  * specified affinity level
241  ******************************************************************************/
tegra_soc_get_target_pwr_state(uint32_t lvl,const plat_local_state_t * states,uint32_t ncpu)242 plat_local_state_t tegra_soc_get_target_pwr_state(uint32_t lvl,
243 					     const plat_local_state_t *states,
244 					     uint32_t ncpu)
245 {
246 	plat_local_state_t target = PSCI_LOCAL_STATE_RUN;
247 	uint32_t cpu = plat_my_core_pos();
248 
249 	/* System Suspend */
250 	if ((lvl == (uint32_t)MPIDR_AFFLVL2) && (states[cpu] == PSTATE_ID_SOC_POWERDN)) {
251 		target = PSTATE_ID_SOC_POWERDN;
252 	}
253 
254 	/* CPU off, CPU suspend */
255 	if (lvl == (uint32_t)MPIDR_AFFLVL1) {
256 		target = tegra_get_afflvl1_pwr_state(states, ncpu);
257 	}
258 
259 	/* target cluster/system state */
260 	return target;
261 }
262 
tegra_soc_pwr_domain_power_down_wfi(const psci_power_state_t * target_state)263 int32_t tegra_soc_pwr_domain_power_down_wfi(const psci_power_state_t *target_state)
264 {
265 	const plat_local_state_t *pwr_domain_state =
266 		target_state->pwr_domain_state;
267 	plat_params_from_bl2_t *params_from_bl2 = bl31_get_plat_params();
268 	uint8_t stateid_afflvl2 = pwr_domain_state[PLAT_MAX_PWR_LVL] &
269 		TEGRA194_STATE_ID_MASK;
270 	uint64_t src_len_in_bytes = (uintptr_t)&__BL31_END__ - (uintptr_t)BL31_BASE;
271 	uint64_t val;
272 	int32_t ret = PSCI_E_SUCCESS;
273 
274 	if (stateid_afflvl2 == PSTATE_ID_SOC_POWERDN) {
275 		val = params_from_bl2->tzdram_base +
276 		      tegra194_get_cpu_reset_handler_size();
277 
278 		/* initialise communication channel with BPMP */
279 		ret = tegra_bpmp_ipc_init();
280 		assert(ret == 0);
281 
282 		/* Enable SE clock before SE context save */
283 		ret = tegra_bpmp_ipc_enable_clock(TEGRA194_CLK_SE);
284 		assert(ret == 0);
285 
286 		/*
287 		 * It is very unlikely that the BL31 image would be
288 		 * bigger than 2^32 bytes
289 		 */
290 		assert(src_len_in_bytes < UINT32_MAX);
291 
292 		if (tegra_se_calculate_save_sha256(BL31_BASE,
293 					(uint32_t)src_len_in_bytes) != 0) {
294 			ERROR("Hash calculation failed. Reboot\n");
295 			(void)tegra_soc_prepare_system_reset();
296 		}
297 
298 		/*
299 		 * The TZRAM loses power when we enter system suspend. To
300 		 * allow graceful exit from system suspend, we need to copy
301 		 * BL3-1 over to TZDRAM.
302 		 */
303 		val = params_from_bl2->tzdram_base +
304 		      tegra194_get_cpu_reset_handler_size();
305 		memcpy((void *)(uintptr_t)val, (void *)(uintptr_t)BL31_BASE,
306 		       src_len_in_bytes);
307 
308 		/* Disable SE clock after SE context save */
309 		ret = tegra_bpmp_ipc_disable_clock(TEGRA194_CLK_SE);
310 		assert(ret == 0);
311 	}
312 
313 	return ret;
314 }
315 
tegra_soc_pwr_domain_suspend_pwrdown_early(const psci_power_state_t * target_state)316 int32_t tegra_soc_pwr_domain_suspend_pwrdown_early(const psci_power_state_t *target_state)
317 {
318 	return PSCI_E_NOT_SUPPORTED;
319 }
320 
tegra_soc_pwr_domain_on(u_register_t mpidr)321 int32_t tegra_soc_pwr_domain_on(u_register_t mpidr)
322 {
323 	uint64_t target_cpu = mpidr & MPIDR_CPU_MASK;
324 	uint64_t target_cluster = (mpidr & MPIDR_CLUSTER_MASK) >>
325 			MPIDR_AFFINITY_BITS;
326 	int32_t ret = 0;
327 
328 	if (target_cluster > ((uint32_t)PLATFORM_CLUSTER_COUNT - 1U)) {
329 		ERROR("%s: unsupported CPU (0x%lx)\n", __func__ , mpidr);
330 		return PSCI_E_NOT_PRESENT;
331 	}
332 
333 	/* construct the target CPU # */
334 	target_cpu += (target_cluster << 1U);
335 
336 	ret = mce_command_handler((uint64_t)MCE_CMD_ONLINE_CORE, target_cpu, 0U, 0U);
337 	if (ret < 0) {
338 		return PSCI_E_DENIED;
339 	}
340 
341 	return PSCI_E_SUCCESS;
342 }
343 
tegra_soc_pwr_domain_on_finish(const psci_power_state_t * target_state)344 int32_t tegra_soc_pwr_domain_on_finish(const psci_power_state_t *target_state)
345 {
346 	const plat_params_from_bl2_t *params_from_bl2 = bl31_get_plat_params();
347 	uint8_t enable_ccplex_lock_step = params_from_bl2->enable_ccplex_lock_step;
348 	uint8_t stateid_afflvl2 = target_state->pwr_domain_state[PLAT_MAX_PWR_LVL];
349 	cpu_context_t *ctx = cm_get_context(NON_SECURE);
350 	uint64_t actlr_elx;
351 
352 	/*
353 	 * Reset power state info for CPUs when onlining, we set
354 	 * deepest power when offlining a core but that may not be
355 	 * requested by non-secure sw which controls idle states. It
356 	 * will re-init this info from non-secure software when the
357 	 * core come online.
358 	 */
359 	actlr_elx = read_ctx_reg((get_el1_sysregs_ctx(ctx)), (CTX_ACTLR_EL1));
360 	actlr_elx &= ~DENVER_CPU_PMSTATE_MASK;
361 	actlr_elx |= DENVER_CPU_PMSTATE_C1;
362 	write_ctx_reg((get_el1_sysregs_ctx(ctx)), (CTX_ACTLR_EL1), (actlr_elx));
363 
364 	/*
365 	 * Check if we are exiting from deep sleep and restore SE
366 	 * context if we are.
367 	 */
368 	if (stateid_afflvl2 == PSTATE_ID_SOC_POWERDN) {
369 
370 #if ENABLE_STRICT_CHECKING_MODE
371 		/*
372 		 * Enable strict checking after programming the GSC for
373 		 * enabling TZSRAM and TZDRAM
374 		 */
375 		mce_enable_strict_checking();
376 #endif
377 
378 		/* Init SMMU */
379 		tegra_smmu_init();
380 
381 		/* Resume SE, RNG1 and PKA1 */
382 		tegra_se_resume();
383 
384 		/*
385 		 * Program XUSB STREAMIDs
386 		 * ======================
387 		 * T19x XUSB has support for XUSB virtualization. It will
388 		 * have one physical function (PF) and four Virtual functions
389 		 * (VF)
390 		 *
391 		 * There were below two SIDs for XUSB until T186.
392 		 * 1) #define TEGRA_SID_XUSB_HOST    0x1bU
393 		 * 2) #define TEGRA_SID_XUSB_DEV    0x1cU
394 		 *
395 		 * We have below four new SIDs added for VF(s)
396 		 * 3) #define TEGRA_SID_XUSB_VF0    0x5dU
397 		 * 4) #define TEGRA_SID_XUSB_VF1    0x5eU
398 		 * 5) #define TEGRA_SID_XUSB_VF2    0x5fU
399 		 * 6) #define TEGRA_SID_XUSB_VF3    0x60U
400 		 *
401 		 * When virtualization is enabled then we have to disable SID
402 		 * override and program above SIDs in below newly added SID
403 		 * registers in XUSB PADCTL MMIO space. These registers are
404 		 * TZ protected and so need to be done in ATF.
405 		 *
406 		 * a) #define XUSB_PADCTL_HOST_AXI_STREAMID_PF_0 (0x136cU)
407 		 * b) #define XUSB_PADCTL_DEV_AXI_STREAMID_PF_0  (0x139cU)
408 		 * c) #define XUSB_PADCTL_HOST_AXI_STREAMID_VF_0 (0x1370U)
409 		 * d) #define XUSB_PADCTL_HOST_AXI_STREAMID_VF_1 (0x1374U)
410 		 * e) #define XUSB_PADCTL_HOST_AXI_STREAMID_VF_2 (0x1378U)
411 		 * f) #define XUSB_PADCTL_HOST_AXI_STREAMID_VF_3 (0x137cU)
412 		 *
413 		 * This change disables SID override and programs XUSB SIDs
414 		 * in above registers to support both virtualization and
415 		 * non-virtualization platforms
416 		 */
417 		if (tegra_platform_is_silicon() || tegra_platform_is_fpga()) {
418 
419 			mmio_write_32(TEGRA_XUSB_PADCTL_BASE +
420 				XUSB_PADCTL_HOST_AXI_STREAMID_PF_0, TEGRA_SID_XUSB_HOST);
421 			assert(mmio_read_32(TEGRA_XUSB_PADCTL_BASE +
422 				XUSB_PADCTL_HOST_AXI_STREAMID_PF_0) == TEGRA_SID_XUSB_HOST);
423 			mmio_write_32(TEGRA_XUSB_PADCTL_BASE +
424 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_0, TEGRA_SID_XUSB_VF0);
425 			assert(mmio_read_32(TEGRA_XUSB_PADCTL_BASE +
426 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_0) == TEGRA_SID_XUSB_VF0);
427 			mmio_write_32(TEGRA_XUSB_PADCTL_BASE +
428 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_1, TEGRA_SID_XUSB_VF1);
429 			assert(mmio_read_32(TEGRA_XUSB_PADCTL_BASE +
430 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_1) == TEGRA_SID_XUSB_VF1);
431 			mmio_write_32(TEGRA_XUSB_PADCTL_BASE +
432 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_2, TEGRA_SID_XUSB_VF2);
433 			assert(mmio_read_32(TEGRA_XUSB_PADCTL_BASE +
434 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_2) == TEGRA_SID_XUSB_VF2);
435 			mmio_write_32(TEGRA_XUSB_PADCTL_BASE +
436 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_3, TEGRA_SID_XUSB_VF3);
437 			assert(mmio_read_32(TEGRA_XUSB_PADCTL_BASE +
438 				XUSB_PADCTL_HOST_AXI_STREAMID_VF_3) == TEGRA_SID_XUSB_VF3);
439 			mmio_write_32(TEGRA_XUSB_PADCTL_BASE +
440 				XUSB_PADCTL_DEV_AXI_STREAMID_PF_0, TEGRA_SID_XUSB_DEV);
441 			assert(mmio_read_32(TEGRA_XUSB_PADCTL_BASE +
442 				XUSB_PADCTL_DEV_AXI_STREAMID_PF_0) == TEGRA_SID_XUSB_DEV);
443 		}
444 	}
445 
446 	/*
447 	 * Enable dual execution optimized translations for all ELx.
448 	 */
449 	if (enable_ccplex_lock_step != 0U) {
450 		actlr_elx = read_actlr_el3();
451 		actlr_elx |= DENVER_CPU_ENABLE_DUAL_EXEC_EL3;
452 		write_actlr_el3(actlr_elx);
453 
454 		actlr_elx = read_actlr_el2();
455 		actlr_elx |= DENVER_CPU_ENABLE_DUAL_EXEC_EL2;
456 		write_actlr_el2(actlr_elx);
457 
458 		actlr_elx = read_actlr_el1();
459 		actlr_elx |= DENVER_CPU_ENABLE_DUAL_EXEC_EL1;
460 		write_actlr_el1(actlr_elx);
461 	}
462 
463 	return PSCI_E_SUCCESS;
464 }
465 
tegra_soc_pwr_domain_off(const psci_power_state_t * target_state)466 int32_t tegra_soc_pwr_domain_off(const psci_power_state_t *target_state)
467 {
468 	uint64_t impl = (read_midr() >> MIDR_IMPL_SHIFT) & MIDR_IMPL_MASK;
469 	int32_t ret = 0;
470 
471 	(void)target_state;
472 
473 	/* Disable Denver's DCO operations */
474 	if (impl == DENVER_IMPL) {
475 		denver_disable_dco();
476 	}
477 
478 	/* Turn off CPU */
479 	ret = mce_command_handler((uint64_t)MCE_CMD_ENTER_CSTATE,
480 			(uint64_t)TEGRA_NVG_CORE_C7, MCE_CORE_SLEEP_TIME_INFINITE, 0U);
481 	assert(ret == 0);
482 
483 	return PSCI_E_SUCCESS;
484 }
485 
tegra_soc_prepare_system_off(void)486 __dead2 void tegra_soc_prepare_system_off(void)
487 {
488 	/* System power off */
489 	mce_system_shutdown();
490 
491 	wfi();
492 
493 	/* wait for the system to power down */
494 	for (;;) {
495 		;
496 	}
497 }
498 
tegra_soc_prepare_system_reset(void)499 int32_t tegra_soc_prepare_system_reset(void)
500 {
501 	/* System reboot */
502 	mce_system_reboot();
503 
504 	return PSCI_E_SUCCESS;
505 }
506