xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h (revision 3fd6c59042dbba50391e30862beac979491145fe)
1 /*
2  * Copyright 2016 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  * Author: Monk.liu@amd.com
23  */
24 #ifndef AMDGPU_VIRT_H
25 #define AMDGPU_VIRT_H
26 
27 #include "amdgv_sriovmsg.h"
28 
29 #define AMDGPU_SRIOV_CAPS_SRIOV_VBIOS  (1 << 0) /* vBIOS is sr-iov ready */
30 #define AMDGPU_SRIOV_CAPS_ENABLE_IOV   (1 << 1) /* sr-iov is enabled on this GPU */
31 #define AMDGPU_SRIOV_CAPS_IS_VF        (1 << 2) /* this GPU is a virtual function */
32 #define AMDGPU_PASSTHROUGH_MODE        (1 << 3) /* thw whole GPU is pass through for VM */
33 #define AMDGPU_SRIOV_CAPS_RUNTIME      (1 << 4) /* is out of full access mode */
34 #define AMDGPU_VF_MMIO_ACCESS_PROTECT  (1 << 5) /* MMIO write access is not allowed in sriov runtime */
35 
36 /* flags for indirect register access path supported by rlcg for sriov */
37 #define AMDGPU_RLCG_GC_WRITE_LEGACY    (0x8 << 28)
38 #define AMDGPU_RLCG_GC_WRITE           (0x0 << 28)
39 #define AMDGPU_RLCG_GC_READ            (0x1 << 28)
40 #define AMDGPU_RLCG_MMHUB_WRITE        (0x2 << 28)
41 
42 /* error code for indirect register access path supported by rlcg for sriov */
43 #define AMDGPU_RLCG_VFGATE_DISABLED		0x4000000
44 #define AMDGPU_RLCG_WRONG_OPERATION_TYPE	0x2000000
45 #define AMDGPU_RLCG_REG_NOT_IN_RANGE		0x1000000
46 
47 #define AMDGPU_RLCG_SCRATCH1_ADDRESS_MASK	0xFFFFF
48 #define AMDGPU_RLCG_SCRATCH1_ERROR_MASK	0xF000000
49 
50 /* all asic after AI use this offset */
51 #define mmRCC_IOV_FUNC_IDENTIFIER 0xDE5
52 /* tonga/fiji use this offset */
53 #define mmBIF_IOV_FUNC_IDENTIFIER 0x1503
54 
55 #define AMDGPU_VF2PF_UPDATE_MAX_RETRY_LIMIT 2
56 
57 enum amdgpu_sriov_vf_mode {
58 	SRIOV_VF_MODE_BARE_METAL = 0,
59 	SRIOV_VF_MODE_ONE_VF,
60 	SRIOV_VF_MODE_MULTI_VF,
61 };
62 
63 struct amdgpu_mm_table {
64 	struct amdgpu_bo	*bo;
65 	uint32_t		*cpu_addr;
66 	uint64_t		gpu_addr;
67 };
68 
69 #define AMDGPU_VF_ERROR_ENTRY_SIZE    16
70 
71 /* struct error_entry - amdgpu VF error information. */
72 struct amdgpu_vf_error_buffer {
73 	struct mutex lock;
74 	int read_count;
75 	int write_count;
76 	uint16_t code[AMDGPU_VF_ERROR_ENTRY_SIZE];
77 	uint16_t flags[AMDGPU_VF_ERROR_ENTRY_SIZE];
78 	uint64_t data[AMDGPU_VF_ERROR_ENTRY_SIZE];
79 };
80 
81 enum idh_request;
82 
83 /**
84  * struct amdgpu_virt_ops - amdgpu device virt operations
85  */
86 struct amdgpu_virt_ops {
87 	int (*req_full_gpu)(struct amdgpu_device *adev, bool init);
88 	int (*rel_full_gpu)(struct amdgpu_device *adev, bool init);
89 	int (*req_init_data)(struct amdgpu_device *adev);
90 	int (*reset_gpu)(struct amdgpu_device *adev);
91 	void (*ready_to_reset)(struct amdgpu_device *adev);
92 	int (*wait_reset)(struct amdgpu_device *adev);
93 	void (*trans_msg)(struct amdgpu_device *adev, enum idh_request req,
94 			  u32 data1, u32 data2, u32 data3);
95 	void (*ras_poison_handler)(struct amdgpu_device *adev,
96 					enum amdgpu_ras_block block);
97 	bool (*rcvd_ras_intr)(struct amdgpu_device *adev);
98 	int (*req_ras_err_count)(struct amdgpu_device *adev);
99 };
100 
101 /*
102  * Firmware Reserve Frame buffer
103  */
104 struct amdgpu_virt_fw_reserve {
105 	struct amd_sriov_msg_pf2vf_info_header *p_pf2vf;
106 	struct amd_sriov_msg_vf2pf_info_header *p_vf2pf;
107 	void *ras_telemetry;
108 	unsigned int checksum_key;
109 };
110 
111 /*
112  * Legacy GIM header
113  *
114  * Defination between PF and VF
115  * Structures forcibly aligned to 4 to keep the same style as PF.
116  */
117 #define AMDGIM_DATAEXCHANGE_OFFSET		(64 * 1024)
118 
119 #define AMDGIM_GET_STRUCTURE_RESERVED_SIZE(total, u8, u16, u32, u64) \
120 		(total - (((u8)+3) / 4 + ((u16)+1) / 2 + (u32) + (u64)*2))
121 
122 enum AMDGIM_FEATURE_FLAG {
123 	/* GIM supports feature of Error log collecting */
124 	AMDGIM_FEATURE_ERROR_LOG_COLLECT = 0x1,
125 	/* GIM supports feature of loading uCodes */
126 	AMDGIM_FEATURE_GIM_LOAD_UCODES   = 0x2,
127 	/* VRAM LOST by GIM */
128 	AMDGIM_FEATURE_GIM_FLR_VRAMLOST = 0x4,
129 	/* MM bandwidth */
130 	AMDGIM_FEATURE_GIM_MM_BW_MGR = 0x8,
131 	/* PP ONE VF MODE in GIM */
132 	AMDGIM_FEATURE_PP_ONE_VF = (1 << 4),
133 	/* Indirect Reg Access enabled */
134 	AMDGIM_FEATURE_INDIRECT_REG_ACCESS = (1 << 5),
135 	/* AV1 Support MODE*/
136 	AMDGIM_FEATURE_AV1_SUPPORT = (1 << 6),
137 	/* VCN RB decouple */
138 	AMDGIM_FEATURE_VCN_RB_DECOUPLE = (1 << 7),
139 	/* MES info */
140 	AMDGIM_FEATURE_MES_INFO_ENABLE = (1 << 8),
141 	AMDGIM_FEATURE_RAS_CAPS = (1 << 9),
142 	AMDGIM_FEATURE_RAS_TELEMETRY = (1 << 10),
143 };
144 
145 enum AMDGIM_REG_ACCESS_FLAG {
146 	/* Use PSP to program IH_RB_CNTL */
147 	AMDGIM_FEATURE_IH_REG_PSP_EN     = (1 << 0),
148 	/* Use RLC to program MMHUB regs */
149 	AMDGIM_FEATURE_MMHUB_REG_RLC_EN  = (1 << 1),
150 	/* Use RLC to program GC regs */
151 	AMDGIM_FEATURE_GC_REG_RLC_EN     = (1 << 2),
152 };
153 
154 struct amdgim_pf2vf_info_v1 {
155 	/* header contains size and version */
156 	struct amd_sriov_msg_pf2vf_info_header header;
157 	/* max_width * max_height */
158 	unsigned int uvd_enc_max_pixels_count;
159 	/* 16x16 pixels/sec, codec independent */
160 	unsigned int uvd_enc_max_bandwidth;
161 	/* max_width * max_height */
162 	unsigned int vce_enc_max_pixels_count;
163 	/* 16x16 pixels/sec, codec independent */
164 	unsigned int vce_enc_max_bandwidth;
165 	/* MEC FW position in kb from the start of visible frame buffer */
166 	unsigned int mecfw_kboffset;
167 	/* The features flags of the GIM driver supports. */
168 	unsigned int feature_flags;
169 	/* use private key from mailbox 2 to create chueksum */
170 	unsigned int checksum;
171 } __aligned(4);
172 
173 struct amdgim_vf2pf_info_v1 {
174 	/* header contains size and version */
175 	struct amd_sriov_msg_vf2pf_info_header header;
176 	/* driver version */
177 	char driver_version[64];
178 	/* driver certification, 1=WHQL, 0=None */
179 	unsigned int driver_cert;
180 	/* guest OS type and version: need a define */
181 	unsigned int os_info;
182 	/* in the unit of 1M */
183 	unsigned int fb_usage;
184 	/* guest gfx engine usage percentage */
185 	unsigned int gfx_usage;
186 	/* guest gfx engine health percentage */
187 	unsigned int gfx_health;
188 	/* guest compute engine usage percentage */
189 	unsigned int compute_usage;
190 	/* guest compute engine health percentage */
191 	unsigned int compute_health;
192 	/* guest vce engine usage percentage. 0xffff means N/A. */
193 	unsigned int vce_enc_usage;
194 	/* guest vce engine health percentage. 0xffff means N/A. */
195 	unsigned int vce_enc_health;
196 	/* guest uvd engine usage percentage. 0xffff means N/A. */
197 	unsigned int uvd_enc_usage;
198 	/* guest uvd engine usage percentage. 0xffff means N/A. */
199 	unsigned int uvd_enc_health;
200 	unsigned int checksum;
201 } __aligned(4);
202 
203 struct amdgim_vf2pf_info_v2 {
204 	/* header contains size and version */
205 	struct amd_sriov_msg_vf2pf_info_header header;
206 	uint32_t checksum;
207 	/* driver version */
208 	uint8_t driver_version[64];
209 	/* driver certification, 1=WHQL, 0=None */
210 	uint32_t driver_cert;
211 	/* guest OS type and version: need a define */
212 	uint32_t os_info;
213 	/* in the unit of 1M */
214 	uint32_t fb_usage;
215 	/* guest gfx engine usage percentage */
216 	uint32_t gfx_usage;
217 	/* guest gfx engine health percentage */
218 	uint32_t gfx_health;
219 	/* guest compute engine usage percentage */
220 	uint32_t compute_usage;
221 	/* guest compute engine health percentage */
222 	uint32_t compute_health;
223 	/* guest vce engine usage percentage. 0xffff means N/A. */
224 	uint32_t vce_enc_usage;
225 	/* guest vce engine health percentage. 0xffff means N/A. */
226 	uint32_t vce_enc_health;
227 	/* guest uvd engine usage percentage. 0xffff means N/A. */
228 	uint32_t uvd_enc_usage;
229 	/* guest uvd engine usage percentage. 0xffff means N/A. */
230 	uint32_t uvd_enc_health;
231 	uint32_t reserved[AMDGIM_GET_STRUCTURE_RESERVED_SIZE(256, 64, 0, (12 + sizeof(struct amd_sriov_msg_vf2pf_info_header)/sizeof(uint32_t)), 0)];
232 } __aligned(4);
233 
234 struct amdgpu_virt_ras_err_handler_data {
235 	/* point to bad page records array */
236 	struct eeprom_table_record *bps;
237 	/* point to reserved bo array */
238 	struct amdgpu_bo **bps_bo;
239 	/* the count of entries */
240 	int count;
241 	/* last reserved entry's index + 1 */
242 	int last_reserved;
243 };
244 
245 /* GPU virtualization */
246 struct amdgpu_virt {
247 	uint32_t			caps;
248 	struct amdgpu_bo		*csa_obj;
249 	void				*csa_cpu_addr;
250 	bool chained_ib_support;
251 	uint32_t			reg_val_offs;
252 	struct amdgpu_irq_src		ack_irq;
253 	struct amdgpu_irq_src		rcv_irq;
254 	struct work_struct		flr_work;
255 	struct amdgpu_mm_table		mm_table;
256 	const struct amdgpu_virt_ops	*ops;
257 	struct amdgpu_vf_error_buffer	vf_errors;
258 	struct amdgpu_virt_fw_reserve	fw_reserve;
259 	uint32_t gim_feature;
260 	uint32_t reg_access_mode;
261 	int req_init_data_ver;
262 	bool tdr_debug;
263 	struct amdgpu_virt_ras_err_handler_data *virt_eh_data;
264 	bool ras_init_done;
265 	uint32_t reg_access;
266 
267 	/* vf2pf message */
268 	struct delayed_work vf2pf_work;
269 	uint32_t vf2pf_update_interval_ms;
270 	int vf2pf_update_retry_cnt;
271 
272 	/* multimedia bandwidth config */
273 	bool     is_mm_bw_enabled;
274 	uint32_t decode_max_dimension_pixels;
275 	uint32_t decode_max_frame_pixels;
276 	uint32_t encode_max_dimension_pixels;
277 	uint32_t encode_max_frame_pixels;
278 
279 	/* the ucode id to signal the autoload */
280 	uint32_t autoload_ucode_id;
281 
282 	struct mutex rlcg_reg_lock;
283 
284 	union amd_sriov_ras_caps ras_en_caps;
285 	union amd_sriov_ras_caps ras_telemetry_en_caps;
286 
287 	struct ratelimit_state ras_telemetry_rs;
288 	struct amd_sriov_ras_telemetry_error_count count_cache;
289 };
290 
291 struct amdgpu_video_codec_info;
292 
293 #define amdgpu_sriov_enabled(adev) \
294 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_ENABLE_IOV)
295 
296 #define amdgpu_sriov_vf(adev) \
297 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_IS_VF)
298 
299 #define amdgpu_sriov_bios(adev) \
300 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS)
301 
302 #define amdgpu_sriov_runtime(adev) \
303 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_RUNTIME)
304 
305 #define amdgpu_sriov_fullaccess(adev) \
306 (amdgpu_sriov_vf((adev)) && !amdgpu_sriov_runtime((adev)))
307 
308 #define amdgpu_sriov_reg_indirect_en(adev) \
309 (amdgpu_sriov_vf((adev)) && \
310 	((adev)->virt.gim_feature & (AMDGIM_FEATURE_INDIRECT_REG_ACCESS)))
311 
312 #define amdgpu_sriov_reg_indirect_ih(adev) \
313 (amdgpu_sriov_vf((adev)) && \
314 	((adev)->virt.reg_access & (AMDGIM_FEATURE_IH_REG_PSP_EN)))
315 
316 #define amdgpu_sriov_reg_indirect_mmhub(adev) \
317 (amdgpu_sriov_vf((adev)) && \
318 	((adev)->virt.reg_access & (AMDGIM_FEATURE_MMHUB_REG_RLC_EN)))
319 
320 #define amdgpu_sriov_reg_indirect_gc(adev) \
321 (amdgpu_sriov_vf((adev)) && \
322 	((adev)->virt.reg_access & (AMDGIM_FEATURE_GC_REG_RLC_EN)))
323 
324 #define amdgpu_sriov_rlcg_error_report_enabled(adev) \
325         (amdgpu_sriov_reg_indirect_mmhub(adev) || amdgpu_sriov_reg_indirect_gc(adev))
326 
327 #define amdgpu_passthrough(adev) \
328 ((adev)->virt.caps & AMDGPU_PASSTHROUGH_MODE)
329 
330 #define amdgpu_sriov_vf_mmio_access_protection(adev) \
331 ((adev)->virt.caps & AMDGPU_VF_MMIO_ACCESS_PROTECT)
332 
333 #define amdgpu_sriov_ras_caps_en(adev) \
334 ((adev)->virt.gim_feature & AMDGIM_FEATURE_RAS_CAPS)
335 
336 #define amdgpu_sriov_ras_telemetry_en(adev) \
337 (((adev)->virt.gim_feature & AMDGIM_FEATURE_RAS_TELEMETRY) && (adev)->virt.fw_reserve.ras_telemetry)
338 
339 #define amdgpu_sriov_ras_telemetry_block_en(adev, sriov_blk) \
340 (amdgpu_sriov_ras_telemetry_en((adev)) && (adev)->virt.ras_telemetry_en_caps.all & BIT(sriov_blk))
341 
is_virtual_machine(void)342 static inline bool is_virtual_machine(void)
343 {
344 #if defined(CONFIG_X86)
345 	return boot_cpu_has(X86_FEATURE_HYPERVISOR);
346 #elif defined(CONFIG_ARM64)
347 	return !is_kernel_in_hyp_mode();
348 #else
349 	return false;
350 #endif
351 }
352 
353 #define amdgpu_sriov_is_pp_one_vf(adev) \
354 	((adev)->virt.gim_feature & AMDGIM_FEATURE_PP_ONE_VF)
355 #define amdgpu_sriov_is_debug(adev) \
356 	((!amdgpu_in_reset(adev)) && adev->virt.tdr_debug)
357 #define amdgpu_sriov_is_normal(adev) \
358 	((!amdgpu_in_reset(adev)) && (!adev->virt.tdr_debug))
359 #define amdgpu_sriov_is_av1_support(adev) \
360 	((adev)->virt.gim_feature & AMDGIM_FEATURE_AV1_SUPPORT)
361 #define amdgpu_sriov_is_vcn_rb_decouple(adev) \
362 	((adev)->virt.gim_feature & AMDGIM_FEATURE_VCN_RB_DECOUPLE)
363 #define amdgpu_sriov_is_mes_info_enable(adev) \
364 	((adev)->virt.gim_feature & AMDGIM_FEATURE_MES_INFO_ENABLE)
365 bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev);
366 void amdgpu_virt_init_setting(struct amdgpu_device *adev);
367 int amdgpu_virt_request_full_gpu(struct amdgpu_device *adev, bool init);
368 int amdgpu_virt_release_full_gpu(struct amdgpu_device *adev, bool init);
369 int amdgpu_virt_reset_gpu(struct amdgpu_device *adev);
370 void amdgpu_virt_request_init_data(struct amdgpu_device *adev);
371 void amdgpu_virt_ready_to_reset(struct amdgpu_device *adev);
372 int amdgpu_virt_wait_reset(struct amdgpu_device *adev);
373 int amdgpu_virt_alloc_mm_table(struct amdgpu_device *adev);
374 void amdgpu_virt_free_mm_table(struct amdgpu_device *adev);
375 bool amdgpu_virt_rcvd_ras_interrupt(struct amdgpu_device *adev);
376 void amdgpu_virt_release_ras_err_handler_data(struct amdgpu_device *adev);
377 void amdgpu_virt_init_data_exchange(struct amdgpu_device *adev);
378 void amdgpu_virt_exchange_data(struct amdgpu_device *adev);
379 void amdgpu_virt_fini_data_exchange(struct amdgpu_device *adev);
380 void amdgpu_detect_virtualization(struct amdgpu_device *adev);
381 
382 bool amdgpu_virt_can_access_debugfs(struct amdgpu_device *adev);
383 int amdgpu_virt_enable_access_debugfs(struct amdgpu_device *adev);
384 void amdgpu_virt_disable_access_debugfs(struct amdgpu_device *adev);
385 
386 enum amdgpu_sriov_vf_mode amdgpu_virt_get_sriov_vf_mode(struct amdgpu_device *adev);
387 
388 void amdgpu_virt_update_sriov_video_codec(struct amdgpu_device *adev,
389 			struct amdgpu_video_codec_info *encode, uint32_t encode_array_size,
390 			struct amdgpu_video_codec_info *decode, uint32_t decode_array_size);
391 void amdgpu_sriov_wreg(struct amdgpu_device *adev,
392 		       u32 offset, u32 value,
393 		       u32 acc_flags, u32 hwip, u32 xcc_id);
394 u32 amdgpu_sriov_rreg(struct amdgpu_device *adev,
395 		      u32 offset, u32 acc_flags, u32 hwip, u32 xcc_id);
396 bool amdgpu_virt_fw_load_skip_check(struct amdgpu_device *adev,
397 			uint32_t ucode_id);
398 void amdgpu_virt_pre_reset(struct amdgpu_device *adev);
399 void amdgpu_virt_post_reset(struct amdgpu_device *adev);
400 bool amdgpu_sriov_xnack_support(struct amdgpu_device *adev);
401 bool amdgpu_virt_get_rlcg_reg_access_flag(struct amdgpu_device *adev,
402 					  u32 acc_flags, u32 hwip,
403 					  bool write, u32 *rlcg_flag);
404 u32 amdgpu_virt_rlcg_reg_rw(struct amdgpu_device *adev, u32 offset, u32 v, u32 flag, u32 xcc_id);
405 bool amdgpu_virt_get_ras_capability(struct amdgpu_device *adev);
406 int amdgpu_virt_req_ras_err_count(struct amdgpu_device *adev, enum amdgpu_ras_block block,
407 				  struct ras_err_data *err_data);
408 int amdgpu_virt_ras_telemetry_post_reset(struct amdgpu_device *adev);
409 #endif
410