1 /* SPDX-License-Identifier: GPL-2.0-only or MIT */
2 /* Copyright 2025 Arm, Ltd. */
3
4 #ifndef __ETHOSU_DEVICE_H__
5 #define __ETHOSU_DEVICE_H__
6
7 #include <linux/bitfield.h>
8 #include <linux/bits.h>
9 #include <linux/types.h>
10
11 #include <drm/drm_device.h>
12 #include <drm/gpu_scheduler.h>
13
14 #include <drm/ethosu_accel.h>
15
16 struct clk;
17 struct gen_pool;
18
19 #define NPU_REG_ID 0x0000
20 #define NPU_REG_STATUS 0x0004
21 #define NPU_REG_CMD 0x0008
22 #define NPU_REG_RESET 0x000c
23 #define NPU_REG_QBASE 0x0010
24 #define NPU_REG_QBASE_HI 0x0014
25 #define NPU_REG_QREAD 0x0018
26 #define NPU_REG_QCONFIG 0x001c
27 #define NPU_REG_QSIZE 0x0020
28 #define NPU_REG_PROT 0x0024
29 #define NPU_REG_CONFIG 0x0028
30 #define NPU_REG_REGIONCFG 0x003c
31 #define NPU_REG_AXILIMIT0 0x0040 // U65
32 #define NPU_REG_AXILIMIT1 0x0044 // U65
33 #define NPU_REG_AXILIMIT2 0x0048 // U65
34 #define NPU_REG_AXILIMIT3 0x004c // U65
35 #define NPU_REG_MEM_ATTR0 0x0040 // U85
36 #define NPU_REG_MEM_ATTR1 0x0044 // U85
37 #define NPU_REG_MEM_ATTR2 0x0048 // U85
38 #define NPU_REG_MEM_ATTR3 0x004c // U85
39 #define NPU_REG_AXI_SRAM 0x0050 // U85
40 #define NPU_REG_AXI_EXT 0x0054 // U85
41
42 #define NPU_REG_BASEP(x) (0x0080 + (x) * 8)
43 #define NPU_REG_BASEP_HI(x) (0x0084 + (x) * 8)
44 #define NPU_BASEP_REGION_MAX 8
45
46 #define ID_ARCH_MAJOR_MASK GENMASK(31, 28)
47 #define ID_ARCH_MINOR_MASK GENMASK(27, 20)
48 #define ID_ARCH_PATCH_MASK GENMASK(19, 16)
49 #define ID_VER_MAJOR_MASK GENMASK(11, 8)
50 #define ID_VER_MINOR_MASK GENMASK(7, 4)
51
52 #define CONFIG_MACS_PER_CC_MASK GENMASK(3, 0)
53 #define CONFIG_CMD_STREAM_VER_MASK GENMASK(7, 4)
54
55 #define STATUS_STATE_RUNNING BIT(0)
56 #define STATUS_IRQ_RAISED BIT(1)
57 #define STATUS_BUS_STATUS BIT(2)
58 #define STATUS_RESET_STATUS BIT(3)
59 #define STATUS_CMD_PARSE_ERR BIT(4)
60 #define STATUS_CMD_END_REACHED BIT(5)
61
62 #define CMD_CLEAR_IRQ BIT(1)
63 #define CMD_TRANSITION_TO_RUN BIT(0)
64
65 #define RESET_PENDING_CSL BIT(1)
66 #define RESET_PENDING_CPL BIT(0)
67
68 #define PROT_ACTIVE_CSL BIT(1)
69
70 enum ethosu_cmds {
71 NPU_OP_CONV = 0x2,
72 NPU_OP_DEPTHWISE = 0x3,
73 NPU_OP_POOL = 0x5,
74 NPU_OP_ELEMENTWISE = 0x6,
75 NPU_OP_RESIZE = 0x7, // U85 only
76 NPU_OP_DMA_START = 0x10,
77 NPU_SET_IFM_PAD_TOP = 0x100,
78 NPU_SET_IFM_PAD_LEFT = 0x101,
79 NPU_SET_IFM_PAD_RIGHT = 0x102,
80 NPU_SET_IFM_PAD_BOTTOM = 0x103,
81 NPU_SET_IFM_DEPTH_M1 = 0x104,
82 NPU_SET_IFM_PRECISION = 0x105,
83 NPU_SET_IFM_BROADCAST = 0x108,
84 NPU_SET_IFM_WIDTH0_M1 = 0x10a,
85 NPU_SET_IFM_HEIGHT0_M1 = 0x10b,
86 NPU_SET_IFM_HEIGHT1_M1 = 0x10c,
87 NPU_SET_IFM_REGION = 0x10f,
88 NPU_SET_OFM_WIDTH_M1 = 0x111,
89 NPU_SET_OFM_HEIGHT_M1 = 0x112,
90 NPU_SET_OFM_DEPTH_M1 = 0x113,
91 NPU_SET_OFM_PRECISION = 0x114,
92 NPU_SET_OFM_WIDTH0_M1 = 0x11a,
93 NPU_SET_OFM_HEIGHT0_M1 = 0x11b,
94 NPU_SET_OFM_HEIGHT1_M1 = 0x11c,
95 NPU_SET_OFM_REGION = 0x11f,
96 NPU_SET_KERNEL_WIDTH_M1 = 0x120,
97 NPU_SET_KERNEL_HEIGHT_M1 = 0x121,
98 NPU_SET_KERNEL_STRIDE = 0x122,
99 NPU_SET_WEIGHT_REGION = 0x128,
100 NPU_SET_SCALE_REGION = 0x129,
101 NPU_SET_DMA0_SRC_REGION = 0x130,
102 NPU_SET_DMA0_DST_REGION = 0x131,
103 NPU_SET_DMA0_SIZE0 = 0x132,
104 NPU_SET_DMA0_SIZE1 = 0x133,
105 NPU_SET_IFM2_BROADCAST = 0x180,
106 NPU_SET_IFM2_PRECISION = 0x185,
107 NPU_SET_IFM2_WIDTH0_M1 = 0x18a,
108 NPU_SET_IFM2_HEIGHT0_M1 = 0x18b,
109 NPU_SET_IFM2_HEIGHT1_M1 = 0x18c,
110 NPU_SET_IFM2_REGION = 0x18f,
111 NPU_SET_IFM_BASE0 = 0x4000,
112 NPU_SET_IFM_BASE1 = 0x4001,
113 NPU_SET_IFM_BASE2 = 0x4002,
114 NPU_SET_IFM_BASE3 = 0x4003,
115 NPU_SET_IFM_STRIDE_X = 0x4004,
116 NPU_SET_IFM_STRIDE_Y = 0x4005,
117 NPU_SET_IFM_STRIDE_C = 0x4006,
118 NPU_SET_OFM_BASE0 = 0x4010,
119 NPU_SET_OFM_BASE1 = 0x4011,
120 NPU_SET_OFM_BASE2 = 0x4012,
121 NPU_SET_OFM_BASE3 = 0x4013,
122 NPU_SET_OFM_STRIDE_X = 0x4014,
123 NPU_SET_OFM_STRIDE_Y = 0x4015,
124 NPU_SET_OFM_STRIDE_C = 0x4016,
125 NPU_SET_WEIGHT_BASE = 0x4020,
126 NPU_SET_WEIGHT_LENGTH = 0x4021,
127 NPU_SET_SCALE_BASE = 0x4022,
128 NPU_SET_SCALE_LENGTH = 0x4023,
129 NPU_SET_DMA0_SRC = 0x4030,
130 NPU_SET_DMA0_DST = 0x4031,
131 NPU_SET_DMA0_LEN = 0x4032,
132 NPU_SET_DMA0_SRC_STRIDE0 = 0x4033,
133 NPU_SET_DMA0_SRC_STRIDE1 = 0x4034,
134 NPU_SET_DMA0_DST_STRIDE0 = 0x4035,
135 NPU_SET_DMA0_DST_STRIDE1 = 0x4036,
136 NPU_SET_IFM2_BASE0 = 0x4080,
137 NPU_SET_IFM2_BASE1 = 0x4081,
138 NPU_SET_IFM2_BASE2 = 0x4082,
139 NPU_SET_IFM2_BASE3 = 0x4083,
140 NPU_SET_IFM2_STRIDE_X = 0x4084,
141 NPU_SET_IFM2_STRIDE_Y = 0x4085,
142 NPU_SET_IFM2_STRIDE_C = 0x4086,
143 NPU_SET_WEIGHT1_BASE = 0x4090,
144 NPU_SET_WEIGHT1_LENGTH = 0x4091,
145 NPU_SET_SCALE1_BASE = 0x4092,
146 NPU_SET_WEIGHT2_BASE = 0x4092,
147 NPU_SET_SCALE1_LENGTH = 0x4093,
148 NPU_SET_WEIGHT2_LENGTH = 0x4093,
149 NPU_SET_WEIGHT3_BASE = 0x4094,
150 NPU_SET_WEIGHT3_LENGTH = 0x4095,
151 };
152
153 #define ETHOSU_SRAM_REGION 2 /* Matching Vela compiler */
154
155 /**
156 * struct ethosu_device - Ethosu device
157 */
158 struct ethosu_device {
159 /** @base: Base drm_device. */
160 struct drm_device base;
161
162 /** @iomem: CPU mapping of the registers. */
163 void __iomem *regs;
164
165 void __iomem *sram;
166 struct gen_pool *srampool;
167 dma_addr_t sramphys;
168
169 struct clk_bulk_data *clks;
170 int num_clks;
171 int irq;
172
173 struct drm_ethosu_npu_info npu_info;
174
175 struct ethosu_job *in_flight_job;
176 /* For in_flight_job and ethosu_job_hw_submit() */
177 struct mutex job_lock;
178
179 /* For dma_fence */
180 spinlock_t fence_lock;
181
182 struct drm_gpu_scheduler sched;
183 /* For ethosu_job_do_push() */
184 struct mutex sched_lock;
185 u64 fence_context;
186 u64 emit_seqno;
187 };
188
189 #define to_ethosu_device(drm_dev) \
190 ((struct ethosu_device *)container_of(drm_dev, struct ethosu_device, base))
191
ethosu_is_u65(const struct ethosu_device * ethosudev)192 static inline bool ethosu_is_u65(const struct ethosu_device *ethosudev)
193 {
194 return FIELD_GET(ID_ARCH_MAJOR_MASK, ethosudev->npu_info.id) == 1;
195 }
196
197 #endif
198