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  * Authors: Christian König
23  */
24 #ifndef __AMDGPU_RING_H__
25 #define __AMDGPU_RING_H__
26 
27 #include "gpu_scheduler.h"
28 
29 /* max number of rings */
30 #define AMDGPU_MAX_RINGS		18
31 #define AMDGPU_MAX_GFX_RINGS		1
32 #define AMDGPU_MAX_COMPUTE_RINGS	8
33 #define AMDGPU_MAX_VCE_RINGS		3
34 #define AMDGPU_MAX_UVD_ENC_RINGS	2
35 
36 /* some special values for the owner field */
37 #define AMDGPU_FENCE_OWNER_UNDEFINED	((void*)0ul)
38 #define AMDGPU_FENCE_OWNER_VM		((void*)1ul)
39 
40 #define AMDGPU_FENCE_FLAG_64BIT         (1 << 0)
41 #define AMDGPU_FENCE_FLAG_INT           (1 << 1)
42 
43 enum amdgpu_ring_type {
44 	AMDGPU_RING_TYPE_GFX,
45 	AMDGPU_RING_TYPE_COMPUTE,
46 	AMDGPU_RING_TYPE_SDMA,
47 	AMDGPU_RING_TYPE_UVD,
48 	AMDGPU_RING_TYPE_VCE,
49 	AMDGPU_RING_TYPE_KIQ,
50 	AMDGPU_RING_TYPE_UVD_ENC
51 };
52 
53 struct amdgpu_device;
54 struct amdgpu_ring;
55 struct amdgpu_ib;
56 struct amdgpu_cs_parser;
57 
58 /*
59  * Fences.
60  */
61 struct amdgpu_fence_driver {
62 	uint64_t			gpu_addr;
63 	volatile uint32_t		*cpu_addr;
64 	/* sync_seq is protected by ring emission lock */
65 	uint32_t			sync_seq;
66 	atomic_t			last_seq;
67 	bool				initialized;
68 	struct amdgpu_irq_src		*irq_src;
69 	unsigned			irq_type;
70 	struct timer_list		fallback_timer;
71 	unsigned			num_fences_mask;
72 	spinlock_t			lock;
73 	struct dma_fence		**fences;
74 };
75 
76 int amdgpu_fence_driver_init(struct amdgpu_device *adev);
77 void amdgpu_fence_driver_fini(struct amdgpu_device *adev);
78 void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev);
79 
80 int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
81 				  unsigned num_hw_submission);
82 int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
83 				   struct amdgpu_irq_src *irq_src,
84 				   unsigned irq_type);
85 void amdgpu_fence_driver_suspend(struct amdgpu_device *adev);
86 void amdgpu_fence_driver_resume(struct amdgpu_device *adev);
87 int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence);
88 void amdgpu_fence_process(struct amdgpu_ring *ring);
89 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
90 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
91 
92 /*
93  * Rings.
94  */
95 
96 /* provided by hw blocks that expose a ring buffer for commands */
97 struct amdgpu_ring_funcs {
98 	enum amdgpu_ring_type	type;
99 	uint32_t		align_mask;
100 	u32			nop;
101 	bool			support_64bit_ptrs;
102 	unsigned		vmhub;
103 
104 	/* ring read/write ptr handling */
105 	u64 (*get_rptr)(struct amdgpu_ring *ring);
106 	u64 (*get_wptr)(struct amdgpu_ring *ring);
107 	void (*set_wptr)(struct amdgpu_ring *ring);
108 	/* validating and patching of IBs */
109 	int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx);
110 	/* constants to calculate how many DW are needed for an emit */
111 	unsigned emit_frame_size;
112 	unsigned emit_ib_size;
113 	/* command emit functions */
114 	void (*emit_ib)(struct amdgpu_ring *ring,
115 			struct amdgpu_ib *ib,
116 			unsigned vm_id, bool ctx_switch);
117 	void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr,
118 			   uint64_t seq, unsigned flags);
119 	void (*emit_pipeline_sync)(struct amdgpu_ring *ring);
120 	void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vm_id,
121 			      uint64_t pd_addr);
122 	void (*emit_hdp_flush)(struct amdgpu_ring *ring);
123 	void (*emit_hdp_invalidate)(struct amdgpu_ring *ring);
124 	void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid,
125 				uint32_t gds_base, uint32_t gds_size,
126 				uint32_t gws_base, uint32_t gws_size,
127 				uint32_t oa_base, uint32_t oa_size);
128 	/* testing functions */
129 	int (*test_ring)(struct amdgpu_ring *ring);
130 	int (*test_ib)(struct amdgpu_ring *ring, long timeout);
131 	/* insert NOP packets */
132 	void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count);
133 	void (*insert_end)(struct amdgpu_ring *ring);
134 	/* pad the indirect buffer to the necessary number of dw */
135 	void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
136 	unsigned (*init_cond_exec)(struct amdgpu_ring *ring);
137 	void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset);
138 	/* note usage for clock and power gating */
139 	void (*begin_use)(struct amdgpu_ring *ring);
140 	void (*end_use)(struct amdgpu_ring *ring);
141 	void (*emit_switch_buffer) (struct amdgpu_ring *ring);
142 	void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags);
143 	void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg);
144 	void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val);
145 };
146 
147 struct amdgpu_ring {
148 	struct amdgpu_device		*adev;
149 	const struct amdgpu_ring_funcs	*funcs;
150 	struct amdgpu_fence_driver	fence_drv;
151 	struct amd_gpu_scheduler	sched;
152 
153 	struct amdgpu_bo	*ring_obj;
154 	volatile uint32_t	*ring;
155 	unsigned		rptr_offs;
156 	u64			wptr;
157 	u64			wptr_old;
158 	unsigned		ring_size;
159 	unsigned		max_dw;
160 	int			count_dw;
161 	uint64_t		gpu_addr;
162 	uint64_t		ptr_mask;
163 	uint32_t		buf_mask;
164 	bool			ready;
165 	u32			idx;
166 	u32			me;
167 	u32			pipe;
168 	u32			queue;
169 	struct amdgpu_bo	*mqd_obj;
170 	uint64_t                mqd_gpu_addr;
171 	void                    *mqd_ptr;
172 	uint64_t                eop_gpu_addr;
173 	u32			doorbell_index;
174 	bool			use_doorbell;
175 	unsigned		wptr_offs;
176 	unsigned		fence_offs;
177 	uint64_t		current_ctx;
178 	char			name[16];
179 	unsigned		cond_exe_offs;
180 	u64			cond_exe_gpu_addr;
181 	volatile u32		*cond_exe_cpu_addr;
182 	unsigned		vm_inv_eng;
183 #if defined(CONFIG_DEBUG_FS)
184 	struct dentry *ent;
185 #endif
186 };
187 
188 int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
189 void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
190 void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
191 void amdgpu_ring_commit(struct amdgpu_ring *ring);
192 void amdgpu_ring_undo(struct amdgpu_ring *ring);
193 int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
194 		     unsigned ring_size, struct amdgpu_irq_src *irq_src,
195 		     unsigned irq_type);
196 void amdgpu_ring_fini(struct amdgpu_ring *ring);
197 static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring)
198 {
199 	int i = 0;
200 	while (i <= ring->buf_mask)
201 		ring->ring[i++] = ring->funcs->nop;
202 
203 }
204 
205 #endif
206