xref: /linux/drivers/gpu/drm/i915/gt/uc/intel_guc_fw.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2014-2019 Intel Corporation
4  *
5  * Authors:
6  *    Vinit Azad <vinit.azad@intel.com>
7  *    Ben Widawsky <ben@bwidawsk.net>
8  *    Dave Gordon <david.s.gordon@intel.com>
9  *    Alex Dai <yu.dai@intel.com>
10  */
11 
12 #include "gt/intel_gt.h"
13 #include "gt/intel_gt_mcr.h"
14 #include "gt/intel_gt_regs.h"
15 #include "gt/intel_rps.h"
16 #include "intel_guc_fw.h"
17 #include "intel_guc_print.h"
18 #include "i915_drv.h"
19 
20 static void guc_prepare_xfer(struct intel_gt *gt)
21 {
22 	struct intel_uncore *uncore = gt->uncore;
23 
24 	u32 shim_flags = GUC_ENABLE_READ_CACHE_LOGIC |
25 			 GUC_ENABLE_READ_CACHE_FOR_SRAM_DATA |
26 			 GUC_ENABLE_READ_CACHE_FOR_WOPCM_DATA |
27 			 GUC_ENABLE_MIA_CLOCK_GATING;
28 
29 	if (GRAPHICS_VER_FULL(uncore->i915) < IP_VER(12, 55))
30 		shim_flags |= GUC_DISABLE_SRAM_INIT_TO_ZEROES |
31 			      GUC_ENABLE_MIA_CACHING;
32 
33 	/* Must program this register before loading the ucode with DMA */
34 	intel_uncore_write(uncore, GUC_SHIM_CONTROL, shim_flags);
35 
36 	if (IS_GEN9_LP(uncore->i915))
37 		intel_uncore_write(uncore, GEN9LP_GT_PM_CONFIG, GT_DOORBELL_ENABLE);
38 	else
39 		intel_uncore_write(uncore, GEN9_GT_PM_CONFIG, GT_DOORBELL_ENABLE);
40 
41 	if (GRAPHICS_VER(uncore->i915) == 9) {
42 		/* DOP Clock Gating Enable for GuC clocks */
43 		intel_uncore_rmw(uncore, GEN7_MISCCPCTL, 0,
44 				 GEN8_DOP_CLOCK_GATE_GUC_ENABLE);
45 
46 		/* allows for 5us (in 10ns units) before GT can go to RC6 */
47 		intel_uncore_write(uncore, GUC_ARAT_C6DIS, 0x1FF);
48 	}
49 }
50 
51 static int guc_xfer_rsa_mmio(struct intel_uc_fw *guc_fw,
52 			     struct intel_uncore *uncore)
53 {
54 	u32 rsa[UOS_RSA_SCRATCH_COUNT];
55 	size_t copied;
56 	int i;
57 
58 	copied = intel_uc_fw_copy_rsa(guc_fw, rsa, sizeof(rsa));
59 	if (copied < sizeof(rsa))
60 		return -ENOMEM;
61 
62 	for (i = 0; i < UOS_RSA_SCRATCH_COUNT; i++)
63 		intel_uncore_write(uncore, UOS_RSA_SCRATCH(i), rsa[i]);
64 
65 	return 0;
66 }
67 
68 static int guc_xfer_rsa_vma(struct intel_uc_fw *guc_fw,
69 			    struct intel_uncore *uncore)
70 {
71 	struct intel_guc *guc = container_of(guc_fw, struct intel_guc, fw);
72 
73 	intel_uncore_write(uncore, UOS_RSA_SCRATCH(0),
74 			   intel_guc_ggtt_offset(guc, guc_fw->rsa_data));
75 
76 	return 0;
77 }
78 
79 /* Copy RSA signature from the fw image to HW for verification */
80 static int guc_xfer_rsa(struct intel_uc_fw *guc_fw,
81 			struct intel_uncore *uncore)
82 {
83 	if (guc_fw->rsa_data)
84 		return guc_xfer_rsa_vma(guc_fw, uncore);
85 	else
86 		return guc_xfer_rsa_mmio(guc_fw, uncore);
87 }
88 
89 /*
90  * Read the GuC status register (GUC_STATUS) and store it in the
91  * specified location; then return a boolean indicating whether
92  * the value matches either completion or a known failure code.
93  *
94  * This is used for polling the GuC status in a wait_for()
95  * loop below.
96  */
97 static inline bool guc_load_done(struct intel_uncore *uncore, u32 *status, bool *success)
98 {
99 	u32 val = intel_uncore_read(uncore, GUC_STATUS);
100 	u32 uk_val = REG_FIELD_GET(GS_UKERNEL_MASK, val);
101 	u32 br_val = REG_FIELD_GET(GS_BOOTROM_MASK, val);
102 
103 	*status = val;
104 	switch (uk_val) {
105 	case INTEL_GUC_LOAD_STATUS_READY:
106 		*success = true;
107 		return true;
108 
109 	case INTEL_GUC_LOAD_STATUS_ERROR_DEVID_BUILD_MISMATCH:
110 	case INTEL_GUC_LOAD_STATUS_GUC_PREPROD_BUILD_MISMATCH:
111 	case INTEL_GUC_LOAD_STATUS_ERROR_DEVID_INVALID_GUCTYPE:
112 	case INTEL_GUC_LOAD_STATUS_HWCONFIG_ERROR:
113 	case INTEL_GUC_LOAD_STATUS_DPC_ERROR:
114 	case INTEL_GUC_LOAD_STATUS_EXCEPTION:
115 	case INTEL_GUC_LOAD_STATUS_INIT_DATA_INVALID:
116 	case INTEL_GUC_LOAD_STATUS_MPU_DATA_INVALID:
117 	case INTEL_GUC_LOAD_STATUS_INIT_MMIO_SAVE_RESTORE_INVALID:
118 	case INTEL_GUC_LOAD_STATUS_KLV_WORKAROUND_INIT_ERROR:
119 		*success = false;
120 		return true;
121 	}
122 
123 	switch (br_val) {
124 	case INTEL_BOOTROM_STATUS_NO_KEY_FOUND:
125 	case INTEL_BOOTROM_STATUS_RSA_FAILED:
126 	case INTEL_BOOTROM_STATUS_PAVPC_FAILED:
127 	case INTEL_BOOTROM_STATUS_WOPCM_FAILED:
128 	case INTEL_BOOTROM_STATUS_LOADLOC_FAILED:
129 	case INTEL_BOOTROM_STATUS_JUMP_FAILED:
130 	case INTEL_BOOTROM_STATUS_RC6CTXCONFIG_FAILED:
131 	case INTEL_BOOTROM_STATUS_MPUMAP_INCORRECT:
132 	case INTEL_BOOTROM_STATUS_EXCEPTION:
133 	case INTEL_BOOTROM_STATUS_PROD_KEY_CHECK_FAILURE:
134 		*success = false;
135 		return true;
136 	}
137 
138 	return false;
139 }
140 
141 /*
142  * Use a longer timeout for debug builds so that problems can be detected
143  * and analysed. But a shorter timeout for releases so that user's don't
144  * wait forever to find out there is a problem. Note that the only reason
145  * an end user should hit the timeout is in case of extreme thermal throttling.
146  * And a system that is that hot during boot is probably dead anyway!
147  */
148 #if defined(CONFIG_DRM_I915_DEBUG_GEM)
149 #define GUC_LOAD_RETRY_LIMIT	20
150 #else
151 #define GUC_LOAD_RETRY_LIMIT	3
152 #endif
153 
154 static int guc_wait_ucode(struct intel_guc *guc)
155 {
156 	struct intel_gt *gt = guc_to_gt(guc);
157 	struct intel_uncore *uncore = gt->uncore;
158 	ktime_t before, after, delta;
159 	bool success;
160 	u32 status;
161 	int ret, count;
162 	u64 delta_ms;
163 	u32 before_freq;
164 
165 	/*
166 	 * Wait for the GuC to start up.
167 	 *
168 	 * Measurements indicate this should take no more than 20ms
169 	 * (assuming the GT clock is at maximum frequency). So, a
170 	 * timeout here indicates that the GuC has failed and is unusable.
171 	 * (Higher levels of the driver may decide to reset the GuC and
172 	 * attempt the ucode load again if this happens.)
173 	 *
174 	 * FIXME: There is a known (but exceedingly unlikely) race condition
175 	 * where the asynchronous frequency management code could reduce
176 	 * the GT clock while a GuC reload is in progress (during a full
177 	 * GT reset). A fix is in progress but there are complex locking
178 	 * issues to be resolved. In the meantime bump the timeout to
179 	 * 200ms. Even at slowest clock, this should be sufficient. And
180 	 * in the working case, a larger timeout makes no difference.
181 	 *
182 	 * IFWI updates have also been seen to cause sporadic failures due to
183 	 * the requested frequency not being granted and thus the firmware
184 	 * load is attempted at minimum frequency. That can lead to load times
185 	 * in the seconds range. However, there is a limit on how long an
186 	 * individual wait_for() can wait. So wrap it in a loop.
187 	 */
188 	before_freq = intel_rps_read_actual_frequency(&gt->rps);
189 	before = ktime_get();
190 	for (count = 0; count < GUC_LOAD_RETRY_LIMIT; count++) {
191 		ret = wait_for(guc_load_done(uncore, &status, &success), 1000);
192 		if (!ret || !success)
193 			break;
194 
195 		guc_dbg(guc, "load still in progress, count = %d, freq = %dMHz, status = 0x%08X [0x%02X/%02X]\n",
196 			count, intel_rps_read_actual_frequency(&gt->rps), status,
197 			REG_FIELD_GET(GS_BOOTROM_MASK, status),
198 			REG_FIELD_GET(GS_UKERNEL_MASK, status));
199 	}
200 	after = ktime_get();
201 	delta = ktime_sub(after, before);
202 	delta_ms = ktime_to_ms(delta);
203 	if (ret || !success) {
204 		u32 ukernel = REG_FIELD_GET(GS_UKERNEL_MASK, status);
205 		u32 bootrom = REG_FIELD_GET(GS_BOOTROM_MASK, status);
206 
207 		guc_info(guc, "load failed: status = 0x%08X, time = %lldms, freq = %dMHz, ret = %d\n",
208 			 status, delta_ms, intel_rps_read_actual_frequency(&gt->rps), ret);
209 		guc_info(guc, "load failed: status: Reset = %d, BootROM = 0x%02X, UKernel = 0x%02X, MIA = 0x%02X, Auth = 0x%02X\n",
210 			 REG_FIELD_GET(GS_MIA_IN_RESET, status),
211 			 bootrom, ukernel,
212 			 REG_FIELD_GET(GS_MIA_MASK, status),
213 			 REG_FIELD_GET(GS_AUTH_STATUS_MASK, status));
214 
215 		switch (bootrom) {
216 		case INTEL_BOOTROM_STATUS_NO_KEY_FOUND:
217 			guc_info(guc, "invalid key requested, header = 0x%08X\n",
218 				 intel_uncore_read(uncore, GUC_HEADER_INFO));
219 			ret = -ENOEXEC;
220 			break;
221 
222 		case INTEL_BOOTROM_STATUS_RSA_FAILED:
223 			guc_info(guc, "firmware signature verification failed\n");
224 			ret = -ENOEXEC;
225 			break;
226 
227 		case INTEL_BOOTROM_STATUS_PROD_KEY_CHECK_FAILURE:
228 			guc_info(guc, "firmware production part check failure\n");
229 			ret = -ENOEXEC;
230 			break;
231 		}
232 
233 		switch (ukernel) {
234 		case INTEL_GUC_LOAD_STATUS_EXCEPTION:
235 			guc_info(guc, "firmware exception. EIP: %#x\n",
236 				 intel_uncore_read(uncore, SOFT_SCRATCH(13)));
237 			ret = -ENXIO;
238 			break;
239 
240 		case INTEL_GUC_LOAD_STATUS_INIT_MMIO_SAVE_RESTORE_INVALID:
241 			guc_info(guc, "illegal register in save/restore workaround list\n");
242 			ret = -EPERM;
243 			break;
244 
245 		case INTEL_GUC_LOAD_STATUS_KLV_WORKAROUND_INIT_ERROR:
246 			guc_info(guc, "invalid w/a KLV entry\n");
247 			ret = -EINVAL;
248 			break;
249 
250 		case INTEL_GUC_LOAD_STATUS_HWCONFIG_START:
251 			guc_info(guc, "still extracting hwconfig table.\n");
252 			ret = -ETIMEDOUT;
253 			break;
254 		}
255 
256 		/* Uncommon/unexpected error, see earlier status code print for details */
257 		if (ret == 0)
258 			ret = -ENXIO;
259 	} else if (delta_ms > 200) {
260 		guc_warn(guc, "excessive init time: %lldms! [status = 0x%08X, count = %d, ret = %d]\n",
261 			 delta_ms, status, count, ret);
262 		guc_warn(guc, "excessive init time: [freq = %dMHz, before = %dMHz, perf_limit_reasons = 0x%08X]\n",
263 			 intel_rps_read_actual_frequency(&gt->rps), before_freq,
264 			 intel_uncore_read(uncore, intel_gt_perf_limit_reasons_reg(gt)));
265 	} else {
266 		guc_dbg(guc, "init took %lldms, freq = %dMHz, before = %dMHz, status = 0x%08X, count = %d, ret = %d\n",
267 			delta_ms, intel_rps_read_actual_frequency(&gt->rps),
268 			before_freq, status, count, ret);
269 	}
270 
271 	return ret;
272 }
273 
274 /**
275  * intel_guc_fw_upload() - load GuC uCode to device
276  * @guc: intel_guc structure
277  *
278  * Called from intel_uc_init_hw() during driver load, resume from sleep and
279  * after a GPU reset.
280  *
281  * The firmware image should have already been fetched into memory, so only
282  * check that fetch succeeded, and then transfer the image to the h/w.
283  *
284  * Return:	non-zero code on error
285  */
286 int intel_guc_fw_upload(struct intel_guc *guc)
287 {
288 	struct intel_gt *gt = guc_to_gt(guc);
289 	struct intel_uncore *uncore = gt->uncore;
290 	int ret;
291 
292 	guc_prepare_xfer(gt);
293 
294 	/*
295 	 * Note that GuC needs the CSS header plus uKernel code to be copied
296 	 * by the DMA engine in one operation, whereas the RSA signature is
297 	 * loaded separately, either by copying it to the UOS_RSA_SCRATCH
298 	 * register (if key size <= 256) or through a ggtt-pinned vma (if key
299 	 * size > 256). The RSA size and therefore the way we provide it to the
300 	 * HW is fixed for each platform and hard-coded in the bootrom.
301 	 */
302 	ret = guc_xfer_rsa(&guc->fw, uncore);
303 	if (ret)
304 		goto out;
305 
306 	/*
307 	 * Current uCode expects the code to be loaded at 8k; locations below
308 	 * this are used for the stack.
309 	 */
310 	ret = intel_uc_fw_upload(&guc->fw, 0x2000, UOS_MOVE);
311 	if (ret)
312 		goto out;
313 
314 	ret = guc_wait_ucode(guc);
315 	if (ret)
316 		goto out;
317 
318 	intel_uc_fw_change_status(&guc->fw, INTEL_UC_FIRMWARE_RUNNING);
319 	return 0;
320 
321 out:
322 	intel_uc_fw_change_status(&guc->fw, INTEL_UC_FIRMWARE_LOAD_FAIL);
323 	return ret;
324 }
325