xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h (revision 3027ce13e04eee76539ca65c2cb1028a01c8c508)
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 5
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 	int (*wait_reset)(struct amdgpu_device *adev);
92 	void (*trans_msg)(struct amdgpu_device *adev, enum idh_request req,
93 			  u32 data1, u32 data2, u32 data3);
94 	void (*ras_poison_handler)(struct amdgpu_device *adev,
95 					enum amdgpu_ras_block block);
96 };
97 
98 /*
99  * Firmware Reserve Frame buffer
100  */
101 struct amdgpu_virt_fw_reserve {
102 	struct amd_sriov_msg_pf2vf_info_header *p_pf2vf;
103 	struct amd_sriov_msg_vf2pf_info_header *p_vf2pf;
104 	unsigned int checksum_key;
105 };
106 
107 /*
108  * Legacy GIM header
109  *
110  * Defination between PF and VF
111  * Structures forcibly aligned to 4 to keep the same style as PF.
112  */
113 #define AMDGIM_DATAEXCHANGE_OFFSET		(64 * 1024)
114 
115 #define AMDGIM_GET_STRUCTURE_RESERVED_SIZE(total, u8, u16, u32, u64) \
116 		(total - (((u8)+3) / 4 + ((u16)+1) / 2 + (u32) + (u64)*2))
117 
118 enum AMDGIM_FEATURE_FLAG {
119 	/* GIM supports feature of Error log collecting */
120 	AMDGIM_FEATURE_ERROR_LOG_COLLECT = 0x1,
121 	/* GIM supports feature of loading uCodes */
122 	AMDGIM_FEATURE_GIM_LOAD_UCODES   = 0x2,
123 	/* VRAM LOST by GIM */
124 	AMDGIM_FEATURE_GIM_FLR_VRAMLOST = 0x4,
125 	/* MM bandwidth */
126 	AMDGIM_FEATURE_GIM_MM_BW_MGR = 0x8,
127 	/* PP ONE VF MODE in GIM */
128 	AMDGIM_FEATURE_PP_ONE_VF = (1 << 4),
129 	/* Indirect Reg Access enabled */
130 	AMDGIM_FEATURE_INDIRECT_REG_ACCESS = (1 << 5),
131 	/* AV1 Support MODE*/
132 	AMDGIM_FEATURE_AV1_SUPPORT = (1 << 6),
133 	/* VCN RB decouple */
134 	AMDGIM_FEATURE_VCN_RB_DECOUPLE = (1 << 7),
135 	/* MES info */
136 	AMDGIM_FEATURE_MES_INFO_ENABLE = (1 << 8),
137 };
138 
139 enum AMDGIM_REG_ACCESS_FLAG {
140 	/* Use PSP to program IH_RB_CNTL */
141 	AMDGIM_FEATURE_IH_REG_PSP_EN     = (1 << 0),
142 	/* Use RLC to program MMHUB regs */
143 	AMDGIM_FEATURE_MMHUB_REG_RLC_EN  = (1 << 1),
144 	/* Use RLC to program GC regs */
145 	AMDGIM_FEATURE_GC_REG_RLC_EN     = (1 << 2),
146 };
147 
148 struct amdgim_pf2vf_info_v1 {
149 	/* header contains size and version */
150 	struct amd_sriov_msg_pf2vf_info_header header;
151 	/* max_width * max_height */
152 	unsigned int uvd_enc_max_pixels_count;
153 	/* 16x16 pixels/sec, codec independent */
154 	unsigned int uvd_enc_max_bandwidth;
155 	/* max_width * max_height */
156 	unsigned int vce_enc_max_pixels_count;
157 	/* 16x16 pixels/sec, codec independent */
158 	unsigned int vce_enc_max_bandwidth;
159 	/* MEC FW position in kb from the start of visible frame buffer */
160 	unsigned int mecfw_kboffset;
161 	/* The features flags of the GIM driver supports. */
162 	unsigned int feature_flags;
163 	/* use private key from mailbox 2 to create chueksum */
164 	unsigned int checksum;
165 } __aligned(4);
166 
167 struct amdgim_vf2pf_info_v1 {
168 	/* header contains size and version */
169 	struct amd_sriov_msg_vf2pf_info_header header;
170 	/* driver version */
171 	char driver_version[64];
172 	/* driver certification, 1=WHQL, 0=None */
173 	unsigned int driver_cert;
174 	/* guest OS type and version: need a define */
175 	unsigned int os_info;
176 	/* in the unit of 1M */
177 	unsigned int fb_usage;
178 	/* guest gfx engine usage percentage */
179 	unsigned int gfx_usage;
180 	/* guest gfx engine health percentage */
181 	unsigned int gfx_health;
182 	/* guest compute engine usage percentage */
183 	unsigned int compute_usage;
184 	/* guest compute engine health percentage */
185 	unsigned int compute_health;
186 	/* guest vce engine usage percentage. 0xffff means N/A. */
187 	unsigned int vce_enc_usage;
188 	/* guest vce engine health percentage. 0xffff means N/A. */
189 	unsigned int vce_enc_health;
190 	/* guest uvd engine usage percentage. 0xffff means N/A. */
191 	unsigned int uvd_enc_usage;
192 	/* guest uvd engine usage percentage. 0xffff means N/A. */
193 	unsigned int uvd_enc_health;
194 	unsigned int checksum;
195 } __aligned(4);
196 
197 struct amdgim_vf2pf_info_v2 {
198 	/* header contains size and version */
199 	struct amd_sriov_msg_vf2pf_info_header header;
200 	uint32_t checksum;
201 	/* driver version */
202 	uint8_t driver_version[64];
203 	/* driver certification, 1=WHQL, 0=None */
204 	uint32_t driver_cert;
205 	/* guest OS type and version: need a define */
206 	uint32_t os_info;
207 	/* in the unit of 1M */
208 	uint32_t fb_usage;
209 	/* guest gfx engine usage percentage */
210 	uint32_t gfx_usage;
211 	/* guest gfx engine health percentage */
212 	uint32_t gfx_health;
213 	/* guest compute engine usage percentage */
214 	uint32_t compute_usage;
215 	/* guest compute engine health percentage */
216 	uint32_t compute_health;
217 	/* guest vce engine usage percentage. 0xffff means N/A. */
218 	uint32_t vce_enc_usage;
219 	/* guest vce engine health percentage. 0xffff means N/A. */
220 	uint32_t vce_enc_health;
221 	/* guest uvd engine usage percentage. 0xffff means N/A. */
222 	uint32_t uvd_enc_usage;
223 	/* guest uvd engine usage percentage. 0xffff means N/A. */
224 	uint32_t uvd_enc_health;
225 	uint32_t reserved[AMDGIM_GET_STRUCTURE_RESERVED_SIZE(256, 64, 0, (12 + sizeof(struct amd_sriov_msg_vf2pf_info_header)/sizeof(uint32_t)), 0)];
226 } __aligned(4);
227 
228 struct amdgpu_virt_ras_err_handler_data {
229 	/* point to bad page records array */
230 	struct eeprom_table_record *bps;
231 	/* point to reserved bo array */
232 	struct amdgpu_bo **bps_bo;
233 	/* the count of entries */
234 	int count;
235 	/* last reserved entry's index + 1 */
236 	int last_reserved;
237 };
238 
239 /* GPU virtualization */
240 struct amdgpu_virt {
241 	uint32_t			caps;
242 	struct amdgpu_bo		*csa_obj;
243 	void				*csa_cpu_addr;
244 	bool chained_ib_support;
245 	uint32_t			reg_val_offs;
246 	struct amdgpu_irq_src		ack_irq;
247 	struct amdgpu_irq_src		rcv_irq;
248 	struct work_struct		flr_work;
249 	struct amdgpu_mm_table		mm_table;
250 	const struct amdgpu_virt_ops	*ops;
251 	struct amdgpu_vf_error_buffer	vf_errors;
252 	struct amdgpu_virt_fw_reserve	fw_reserve;
253 	uint32_t gim_feature;
254 	uint32_t reg_access_mode;
255 	int req_init_data_ver;
256 	bool tdr_debug;
257 	struct amdgpu_virt_ras_err_handler_data *virt_eh_data;
258 	bool ras_init_done;
259 	uint32_t reg_access;
260 
261 	/* vf2pf message */
262 	struct delayed_work vf2pf_work;
263 	uint32_t vf2pf_update_interval_ms;
264 	int vf2pf_update_retry_cnt;
265 
266 	/* multimedia bandwidth config */
267 	bool     is_mm_bw_enabled;
268 	uint32_t decode_max_dimension_pixels;
269 	uint32_t decode_max_frame_pixels;
270 	uint32_t encode_max_dimension_pixels;
271 	uint32_t encode_max_frame_pixels;
272 
273 	/* the ucode id to signal the autoload */
274 	uint32_t autoload_ucode_id;
275 
276 	struct mutex rlcg_reg_lock;
277 };
278 
279 struct amdgpu_video_codec_info;
280 
281 #define amdgpu_sriov_enabled(adev) \
282 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_ENABLE_IOV)
283 
284 #define amdgpu_sriov_vf(adev) \
285 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_IS_VF)
286 
287 #define amdgpu_sriov_bios(adev) \
288 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS)
289 
290 #define amdgpu_sriov_runtime(adev) \
291 ((adev)->virt.caps & AMDGPU_SRIOV_CAPS_RUNTIME)
292 
293 #define amdgpu_sriov_fullaccess(adev) \
294 (amdgpu_sriov_vf((adev)) && !amdgpu_sriov_runtime((adev)))
295 
296 #define amdgpu_sriov_reg_indirect_en(adev) \
297 (amdgpu_sriov_vf((adev)) && \
298 	((adev)->virt.gim_feature & (AMDGIM_FEATURE_INDIRECT_REG_ACCESS)))
299 
300 #define amdgpu_sriov_reg_indirect_ih(adev) \
301 (amdgpu_sriov_vf((adev)) && \
302 	((adev)->virt.reg_access & (AMDGIM_FEATURE_IH_REG_PSP_EN)))
303 
304 #define amdgpu_sriov_reg_indirect_mmhub(adev) \
305 (amdgpu_sriov_vf((adev)) && \
306 	((adev)->virt.reg_access & (AMDGIM_FEATURE_MMHUB_REG_RLC_EN)))
307 
308 #define amdgpu_sriov_reg_indirect_gc(adev) \
309 (amdgpu_sriov_vf((adev)) && \
310 	((adev)->virt.reg_access & (AMDGIM_FEATURE_GC_REG_RLC_EN)))
311 
312 #define amdgpu_sriov_rlcg_error_report_enabled(adev) \
313         (amdgpu_sriov_reg_indirect_mmhub(adev) || amdgpu_sriov_reg_indirect_gc(adev))
314 
315 #define amdgpu_passthrough(adev) \
316 ((adev)->virt.caps & AMDGPU_PASSTHROUGH_MODE)
317 
318 #define amdgpu_sriov_vf_mmio_access_protection(adev) \
319 ((adev)->virt.caps & AMDGPU_VF_MMIO_ACCESS_PROTECT)
320 
321 static inline bool is_virtual_machine(void)
322 {
323 #if defined(CONFIG_X86)
324 	return boot_cpu_has(X86_FEATURE_HYPERVISOR);
325 #elif defined(CONFIG_ARM64)
326 	return !is_kernel_in_hyp_mode();
327 #else
328 	return false;
329 #endif
330 }
331 
332 #define amdgpu_sriov_is_pp_one_vf(adev) \
333 	((adev)->virt.gim_feature & AMDGIM_FEATURE_PP_ONE_VF)
334 #define amdgpu_sriov_is_debug(adev) \
335 	((!amdgpu_in_reset(adev)) && adev->virt.tdr_debug)
336 #define amdgpu_sriov_is_normal(adev) \
337 	((!amdgpu_in_reset(adev)) && (!adev->virt.tdr_debug))
338 #define amdgpu_sriov_is_av1_support(adev) \
339 	((adev)->virt.gim_feature & AMDGIM_FEATURE_AV1_SUPPORT)
340 #define amdgpu_sriov_is_vcn_rb_decouple(adev) \
341 	((adev)->virt.gim_feature & AMDGIM_FEATURE_VCN_RB_DECOUPLE)
342 #define amdgpu_sriov_is_mes_info_enable(adev) \
343 	((adev)->virt.gim_feature & AMDGIM_FEATURE_MES_INFO_ENABLE)
344 bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev);
345 void amdgpu_virt_init_setting(struct amdgpu_device *adev);
346 int amdgpu_virt_request_full_gpu(struct amdgpu_device *adev, bool init);
347 int amdgpu_virt_release_full_gpu(struct amdgpu_device *adev, bool init);
348 int amdgpu_virt_reset_gpu(struct amdgpu_device *adev);
349 void amdgpu_virt_request_init_data(struct amdgpu_device *adev);
350 int amdgpu_virt_wait_reset(struct amdgpu_device *adev);
351 int amdgpu_virt_alloc_mm_table(struct amdgpu_device *adev);
352 void amdgpu_virt_free_mm_table(struct amdgpu_device *adev);
353 void amdgpu_virt_release_ras_err_handler_data(struct amdgpu_device *adev);
354 void amdgpu_virt_init_data_exchange(struct amdgpu_device *adev);
355 void amdgpu_virt_exchange_data(struct amdgpu_device *adev);
356 void amdgpu_virt_fini_data_exchange(struct amdgpu_device *adev);
357 void amdgpu_detect_virtualization(struct amdgpu_device *adev);
358 
359 bool amdgpu_virt_can_access_debugfs(struct amdgpu_device *adev);
360 int amdgpu_virt_enable_access_debugfs(struct amdgpu_device *adev);
361 void amdgpu_virt_disable_access_debugfs(struct amdgpu_device *adev);
362 
363 enum amdgpu_sriov_vf_mode amdgpu_virt_get_sriov_vf_mode(struct amdgpu_device *adev);
364 
365 void amdgpu_virt_update_sriov_video_codec(struct amdgpu_device *adev,
366 			struct amdgpu_video_codec_info *encode, uint32_t encode_array_size,
367 			struct amdgpu_video_codec_info *decode, uint32_t decode_array_size);
368 void amdgpu_sriov_wreg(struct amdgpu_device *adev,
369 		       u32 offset, u32 value,
370 		       u32 acc_flags, u32 hwip, u32 xcc_id);
371 u32 amdgpu_sriov_rreg(struct amdgpu_device *adev,
372 		      u32 offset, u32 acc_flags, u32 hwip, u32 xcc_id);
373 bool amdgpu_virt_fw_load_skip_check(struct amdgpu_device *adev,
374 			uint32_t ucode_id);
375 void amdgpu_virt_post_reset(struct amdgpu_device *adev);
376 bool amdgpu_sriov_xnack_support(struct amdgpu_device *adev);
377 bool amdgpu_virt_get_rlcg_reg_access_flag(struct amdgpu_device *adev,
378 					  u32 acc_flags, u32 hwip,
379 					  bool write, u32 *rlcg_flag);
380 u32 amdgpu_virt_rlcg_reg_rw(struct amdgpu_device *adev, u32 offset, u32 v, u32 flag, u32 xcc_id);
381 #endif
382