xref: /linux/drivers/gpu/drm/msm/msm_gpu.h (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /*
2  * Copyright (C) 2013 Red Hat
3  * Author: Rob Clark <robdclark@gmail.com>
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published by
7  * the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  *
14  * You should have received a copy of the GNU General Public License along with
15  * this program.  If not, see <http://www.gnu.org/licenses/>.
16  */
17 
18 #ifndef __MSM_GPU_H__
19 #define __MSM_GPU_H__
20 
21 #include <linux/clk.h>
22 #include <linux/regulator/consumer.h>
23 
24 #include "msm_drv.h"
25 #include "msm_fence.h"
26 #include "msm_ringbuffer.h"
27 
28 struct msm_gem_submit;
29 struct msm_gpu_perfcntr;
30 
31 /* So far, with hardware that I've seen to date, we can have:
32  *  + zero, one, or two z180 2d cores
33  *  + a3xx or a2xx 3d core, which share a common CP (the firmware
34  *    for the CP seems to implement some different PM4 packet types
35  *    but the basics of cmdstream submission are the same)
36  *
37  * Which means that the eventual complete "class" hierarchy, once
38  * support for all past and present hw is in place, becomes:
39  *  + msm_gpu
40  *    + adreno_gpu
41  *      + a3xx_gpu
42  *      + a2xx_gpu
43  *    + z180_gpu
44  */
45 struct msm_gpu_funcs {
46 	int (*get_param)(struct msm_gpu *gpu, uint32_t param, uint64_t *value);
47 	int (*hw_init)(struct msm_gpu *gpu);
48 	int (*pm_suspend)(struct msm_gpu *gpu);
49 	int (*pm_resume)(struct msm_gpu *gpu);
50 	void (*submit)(struct msm_gpu *gpu, struct msm_gem_submit *submit,
51 			struct msm_file_private *ctx);
52 	void (*flush)(struct msm_gpu *gpu);
53 	bool (*idle)(struct msm_gpu *gpu);
54 	irqreturn_t (*irq)(struct msm_gpu *irq);
55 	uint32_t (*last_fence)(struct msm_gpu *gpu);
56 	void (*recover)(struct msm_gpu *gpu);
57 	void (*destroy)(struct msm_gpu *gpu);
58 #ifdef CONFIG_DEBUG_FS
59 	/* show GPU status in debugfs: */
60 	void (*show)(struct msm_gpu *gpu, struct seq_file *m);
61 #endif
62 };
63 
64 struct msm_gpu {
65 	const char *name;
66 	struct drm_device *dev;
67 	const struct msm_gpu_funcs *funcs;
68 
69 	/* performance counters (hw & sw): */
70 	spinlock_t perf_lock;
71 	bool perfcntr_active;
72 	struct {
73 		bool active;
74 		ktime_t time;
75 	} last_sample;
76 	uint32_t totaltime, activetime;    /* sw counters */
77 	uint32_t last_cntrs[5];            /* hw counters */
78 	const struct msm_gpu_perfcntr *perfcntrs;
79 	uint32_t num_perfcntrs;
80 
81 	/* ringbuffer: */
82 	struct msm_ringbuffer *rb;
83 	uint64_t rb_iova;
84 
85 	/* list of GEM active objects: */
86 	struct list_head active_list;
87 
88 	/* fencing: */
89 	struct msm_fence_context *fctx;
90 
91 	/* is gpu powered/active? */
92 	int active_cnt;
93 	bool inactive;
94 
95 	/* worker for handling active-list retiring: */
96 	struct work_struct retire_work;
97 
98 	void __iomem *mmio;
99 	int irq;
100 
101 	struct msm_gem_address_space *aspace;
102 	int id;
103 
104 	/* Power Control: */
105 	struct regulator *gpu_reg, *gpu_cx;
106 	struct clk *ebi1_clk, *grp_clks[6];
107 	uint32_t fast_rate, slow_rate, bus_freq;
108 
109 #ifdef DOWNSTREAM_CONFIG_MSM_BUS_SCALING
110 	struct msm_bus_scale_pdata *bus_scale_table;
111 	uint32_t bsc;
112 #endif
113 
114 	/* Hang and Inactivity Detection:
115 	 */
116 #define DRM_MSM_INACTIVE_PERIOD   66 /* in ms (roughly four frames) */
117 #define DRM_MSM_INACTIVE_JIFFIES  msecs_to_jiffies(DRM_MSM_INACTIVE_PERIOD)
118 	struct timer_list inactive_timer;
119 	struct work_struct inactive_work;
120 #define DRM_MSM_HANGCHECK_PERIOD 500 /* in ms */
121 #define DRM_MSM_HANGCHECK_JIFFIES msecs_to_jiffies(DRM_MSM_HANGCHECK_PERIOD)
122 	struct timer_list hangcheck_timer;
123 	uint32_t hangcheck_fence;
124 	struct work_struct recover_work;
125 
126 	struct list_head submit_list;
127 };
128 
129 static inline bool msm_gpu_active(struct msm_gpu *gpu)
130 {
131 	return gpu->fctx->last_fence > gpu->funcs->last_fence(gpu);
132 }
133 
134 /* Perf-Counters:
135  * The select_reg and select_val are just there for the benefit of the child
136  * class that actually enables the perf counter..  but msm_gpu base class
137  * will handle sampling/displaying the counters.
138  */
139 
140 struct msm_gpu_perfcntr {
141 	uint32_t select_reg;
142 	uint32_t sample_reg;
143 	uint32_t select_val;
144 	const char *name;
145 };
146 
147 static inline void gpu_write(struct msm_gpu *gpu, u32 reg, u32 data)
148 {
149 	msm_writel(data, gpu->mmio + (reg << 2));
150 }
151 
152 static inline u32 gpu_read(struct msm_gpu *gpu, u32 reg)
153 {
154 	return msm_readl(gpu->mmio + (reg << 2));
155 }
156 
157 static inline void gpu_rmw(struct msm_gpu *gpu, u32 reg, u32 mask, u32 or)
158 {
159 	uint32_t val = gpu_read(gpu, reg);
160 
161 	val &= ~mask;
162 	gpu_write(gpu, reg, val | or);
163 }
164 
165 static inline u64 gpu_read64(struct msm_gpu *gpu, u32 lo, u32 hi)
166 {
167 	u64 val;
168 
169 	/*
170 	 * Why not a readq here? Two reasons: 1) many of the LO registers are
171 	 * not quad word aligned and 2) the GPU hardware designers have a bit
172 	 * of a history of putting registers where they fit, especially in
173 	 * spins. The longer a GPU family goes the higher the chance that
174 	 * we'll get burned.  We could do a series of validity checks if we
175 	 * wanted to, but really is a readq() that much better? Nah.
176 	 */
177 
178 	/*
179 	 * For some lo/hi registers (like perfcounters), the hi value is latched
180 	 * when the lo is read, so make sure to read the lo first to trigger
181 	 * that
182 	 */
183 	val = (u64) msm_readl(gpu->mmio + (lo << 2));
184 	val |= ((u64) msm_readl(gpu->mmio + (hi << 2)) << 32);
185 
186 	return val;
187 }
188 
189 static inline void gpu_write64(struct msm_gpu *gpu, u32 lo, u32 hi, u64 val)
190 {
191 	/* Why not a writeq here? Read the screed above */
192 	msm_writel(lower_32_bits(val), gpu->mmio + (lo << 2));
193 	msm_writel(upper_32_bits(val), gpu->mmio + (hi << 2));
194 }
195 
196 int msm_gpu_pm_suspend(struct msm_gpu *gpu);
197 int msm_gpu_pm_resume(struct msm_gpu *gpu);
198 
199 void msm_gpu_perfcntr_start(struct msm_gpu *gpu);
200 void msm_gpu_perfcntr_stop(struct msm_gpu *gpu);
201 int msm_gpu_perfcntr_sample(struct msm_gpu *gpu, uint32_t *activetime,
202 		uint32_t *totaltime, uint32_t ncntrs, uint32_t *cntrs);
203 
204 void msm_gpu_retire(struct msm_gpu *gpu);
205 void msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
206 		struct msm_file_private *ctx);
207 
208 int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,
209 		struct msm_gpu *gpu, const struct msm_gpu_funcs *funcs,
210 		const char *name, const char *ioname, const char *irqname, int ringsz);
211 void msm_gpu_cleanup(struct msm_gpu *gpu);
212 
213 struct msm_gpu *adreno_load_gpu(struct drm_device *dev);
214 void __init adreno_register(void);
215 void __exit adreno_unregister(void);
216 
217 #endif /* __MSM_GPU_H__ */
218