2 * Copyright (C) 2013 Red Hat
3 * Author: Rob Clark <robdclark@gmail.com>
5 * Copyright (c) 2014,2017 The Linux Foundation. All rights reserved.
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 as published by
9 * the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along with
17 * this program. If not, see <http://www.gnu.org/licenses/>.
20 #ifndef __ADRENO_GPU_H__
21 #define __ADRENO_GPU_H__
23 #include <linux/firmware.h>
24 #include <linux/iopoll.h>
28 #include "adreno_common.xml.h"
29 #include "adreno_pm4.xml.h"
31 #define REG_ADRENO_DEFINE(_offset, _reg) [_offset] = (_reg) + 1
33 #define REG_ADRENO_SKIP(_offset) [_offset] = REG_SKIP
36 * adreno_regs: List of registers that are used in across all
37 * 3D devices. Each device type has different offset value for the same
38 * register, so an array of register offsets are declared for every device
39 * and are indexed by the enumeration values defined in this enum
42 REG_ADRENO_CP_RB_BASE,
43 REG_ADRENO_CP_RB_BASE_HI,
44 REG_ADRENO_CP_RB_RPTR_ADDR,
45 REG_ADRENO_CP_RB_RPTR_ADDR_HI,
46 REG_ADRENO_CP_RB_RPTR,
47 REG_ADRENO_CP_RB_WPTR,
48 REG_ADRENO_CP_RB_CNTL,
49 REG_ADRENO_REGISTER_MAX,
54 ADRENO_FW_SQE = 0, /* a6xx */
56 ADRENO_FW_GMU = 1, /* a6xx */
62 ADRENO_QUIRK_TWO_PASS_USE_WFI = 1,
63 ADRENO_QUIRK_FAULT_DETECT_MASK = 2,
64 ADRENO_QUIRK_LMLOADKILL_DISABLE = 3,
74 #define ADRENO_REV(core, major, minor, patchid) \
75 ((struct adreno_rev){ core, major, minor, patchid })
77 struct adreno_gpu_funcs {
78 struct msm_gpu_funcs base;
79 int (*get_timestamp)(struct msm_gpu *gpu, uint64_t *value);
83 struct adreno_rev rev;
86 const char *fw[ADRENO_FW_MAX];
88 enum adreno_quirks quirks;
89 struct msm_gpu *(*init)(struct drm_device *dev);
94 const struct adreno_info *adreno_info(struct adreno_rev rev);
98 struct adreno_rev rev;
99 const struct adreno_info *info;
100 uint32_t gmem; /* actual gmem size */
101 uint32_t revn; /* numeric revision name */
102 const struct adreno_gpu_funcs *funcs;
104 /* interesting register offsets to dump: */
105 const unsigned int *registers;
108 * Are we loading fw from legacy path? Prior to addition
109 * of gpu firmware to linux-firmware, the fw files were
110 * placed in toplevel firmware directory, following qcom's
111 * android kernel. But linux-firmware preferred they be
112 * placed in a 'qcom' subdirectory.
114 * For backwards compatibility, we try first to load from
115 * the new path, using request_firmware_direct() to avoid
116 * any potential timeout waiting for usermode helper, then
117 * fall back to the old path (with direct load). And
118 * finally fall back to request_firmware() with the new
119 * path to allow the usermode helper.
122 FW_LOCATION_UNKNOWN = 0,
123 FW_LOCATION_NEW, /* /lib/firmware/qcom/$fwfile */
124 FW_LOCATION_LEGACY, /* /lib/firmware/$fwfile */
129 const struct firmware *fw[ADRENO_FW_MAX];
132 * Register offsets are different between some GPUs.
133 * GPU specific offsets will be exported by GPU specific
134 * code (a3xx_gpu.c) and stored in this common location.
136 const unsigned int *reg_offsets;
138 #define to_adreno_gpu(x) container_of(x, struct adreno_gpu, base)
140 /* platform config data (ie. from DT, or pdata) */
141 struct adreno_platform_config {
142 struct adreno_rev rev;
145 #define ADRENO_IDLE_TIMEOUT msecs_to_jiffies(1000)
147 #define spin_until(X) ({ \
148 int __ret = -ETIMEDOUT; \
149 unsigned long __t = jiffies + ADRENO_IDLE_TIMEOUT; \
155 } while (time_before(jiffies, __t)); \
159 static inline bool adreno_is_a2xx(struct adreno_gpu *gpu)
161 return (gpu->revn < 300);
164 static inline bool adreno_is_a20x(struct adreno_gpu *gpu)
166 return (gpu->revn < 210);
169 static inline bool adreno_is_a225(struct adreno_gpu *gpu)
171 return gpu->revn == 225;
174 static inline bool adreno_is_a3xx(struct adreno_gpu *gpu)
176 return (gpu->revn >= 300) && (gpu->revn < 400);
179 static inline bool adreno_is_a305(struct adreno_gpu *gpu)
181 return gpu->revn == 305;
184 static inline bool adreno_is_a306(struct adreno_gpu *gpu)
186 /* yes, 307, because a305c is 306 */
187 return gpu->revn == 307;
190 static inline bool adreno_is_a320(struct adreno_gpu *gpu)
192 return gpu->revn == 320;
195 static inline bool adreno_is_a330(struct adreno_gpu *gpu)
197 return gpu->revn == 330;
200 static inline bool adreno_is_a330v2(struct adreno_gpu *gpu)
202 return adreno_is_a330(gpu) && (gpu->rev.patchid > 0);
205 static inline bool adreno_is_a4xx(struct adreno_gpu *gpu)
207 return (gpu->revn >= 400) && (gpu->revn < 500);
210 static inline int adreno_is_a420(struct adreno_gpu *gpu)
212 return gpu->revn == 420;
215 static inline int adreno_is_a430(struct adreno_gpu *gpu)
217 return gpu->revn == 430;
220 static inline int adreno_is_a530(struct adreno_gpu *gpu)
222 return gpu->revn == 530;
225 static inline int adreno_is_a540(struct adreno_gpu *gpu)
227 return gpu->revn == 540;
230 int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value);
231 const struct firmware *adreno_request_fw(struct adreno_gpu *adreno_gpu,
233 struct drm_gem_object *adreno_fw_create_bo(struct msm_gpu *gpu,
234 const struct firmware *fw, u64 *iova);
235 int adreno_hw_init(struct msm_gpu *gpu);
236 void adreno_recover(struct msm_gpu *gpu);
237 void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
238 struct msm_file_private *ctx);
239 void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
240 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
241 #if defined(CONFIG_DEBUG_FS) || defined(CONFIG_DEV_COREDUMP)
242 void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
243 struct drm_printer *p);
245 void adreno_dump_info(struct msm_gpu *gpu);
246 void adreno_dump(struct msm_gpu *gpu);
247 void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords);
248 struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu);
250 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
251 struct adreno_gpu *gpu, const struct adreno_gpu_funcs *funcs,
253 void adreno_gpu_cleanup(struct adreno_gpu *gpu);
254 int adreno_load_fw(struct adreno_gpu *adreno_gpu);
256 void adreno_gpu_state_destroy(struct msm_gpu_state *state);
258 int adreno_gpu_state_get(struct msm_gpu *gpu, struct msm_gpu_state *state);
259 int adreno_gpu_state_put(struct msm_gpu_state *state);
262 * For a5xx and a6xx targets load the zap shader that is used to pull the GPU
265 int adreno_zap_shader_load(struct msm_gpu *gpu, u32 pasid);
267 /* ringbuffer helpers (the parts that are adreno specific) */
270 OUT_PKT0(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
272 adreno_wait_ring(ring, cnt+1);
273 OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF));
278 OUT_PKT2(struct msm_ringbuffer *ring)
280 adreno_wait_ring(ring, 1);
281 OUT_RING(ring, CP_TYPE2_PKT);
285 OUT_PKT3(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
287 adreno_wait_ring(ring, cnt+1);
288 OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8));
291 static inline u32 PM4_PARITY(u32 val)
293 return (0x9669 >> (0xF & (val ^
294 (val >> 4) ^ (val >> 8) ^ (val >> 12) ^
295 (val >> 16) ^ ((val) >> 20) ^ (val >> 24) ^
299 /* Maximum number of values that can be executed for one opcode */
300 #define TYPE4_MAX_PAYLOAD 127
302 #define PKT4(_reg, _cnt) \
303 (CP_TYPE4_PKT | ((_cnt) << 0) | (PM4_PARITY((_cnt)) << 7) | \
304 (((_reg) & 0x3FFFF) << 8) | (PM4_PARITY((_reg)) << 27))
307 OUT_PKT4(struct msm_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
309 adreno_wait_ring(ring, cnt + 1);
310 OUT_RING(ring, PKT4(regindx, cnt));
314 OUT_PKT7(struct msm_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
316 adreno_wait_ring(ring, cnt + 1);
317 OUT_RING(ring, CP_TYPE7_PKT | (cnt << 0) | (PM4_PARITY(cnt) << 15) |
318 ((opcode & 0x7F) << 16) | (PM4_PARITY(opcode) << 23));
322 * adreno_reg_check() - Checks the validity of a register enum
323 * @gpu: Pointer to struct adreno_gpu
324 * @offset_name: The register enum that is checked
326 static inline bool adreno_reg_check(struct adreno_gpu *gpu,
327 enum adreno_regs offset_name)
329 if (offset_name >= REG_ADRENO_REGISTER_MAX ||
330 !gpu->reg_offsets[offset_name]) {
335 * REG_SKIP is a special value that tell us that the register in
336 * question isn't implemented on target but don't trigger a BUG(). This
337 * is used to cleanly implement adreno_gpu_write64() and
338 * adreno_gpu_read64() in a generic fashion
340 if (gpu->reg_offsets[offset_name] == REG_SKIP)
346 static inline u32 adreno_gpu_read(struct adreno_gpu *gpu,
347 enum adreno_regs offset_name)
349 u32 reg = gpu->reg_offsets[offset_name];
351 if(adreno_reg_check(gpu,offset_name))
352 val = gpu_read(&gpu->base, reg - 1);
356 static inline void adreno_gpu_write(struct adreno_gpu *gpu,
357 enum adreno_regs offset_name, u32 data)
359 u32 reg = gpu->reg_offsets[offset_name];
360 if(adreno_reg_check(gpu, offset_name))
361 gpu_write(&gpu->base, reg - 1, data);
364 struct msm_gpu *a2xx_gpu_init(struct drm_device *dev);
365 struct msm_gpu *a3xx_gpu_init(struct drm_device *dev);
366 struct msm_gpu *a4xx_gpu_init(struct drm_device *dev);
367 struct msm_gpu *a5xx_gpu_init(struct drm_device *dev);
368 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev);
370 static inline void adreno_gpu_write64(struct adreno_gpu *gpu,
371 enum adreno_regs lo, enum adreno_regs hi, u64 data)
373 adreno_gpu_write(gpu, lo, lower_32_bits(data));
374 adreno_gpu_write(gpu, hi, upper_32_bits(data));
377 static inline uint32_t get_wptr(struct msm_ringbuffer *ring)
379 return (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >> 2);
383 * Given a register and a count, return a value to program into
384 * REG_CP_PROTECT_REG(n) - this will block both reads and writes for _len
385 * registers starting at _reg.
387 * The register base needs to be a multiple of the length. If it is not, the
388 * hardware will quietly mask off the bits for you and shift the size. For
389 * example, if you intend the protection to start at 0x07 for a length of 4
390 * (0x07-0x0A) the hardware will actually protect (0x04-0x07) which might
391 * expose registers you intended to protect!
393 #define ADRENO_PROTECT_RW(_reg, _len) \
394 ((1 << 30) | (1 << 29) | \
395 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF))
398 * Same as above, but allow reads over the range. For areas of mixed use (such
399 * as performance counters) this allows us to protect a much larger range with a
402 #define ADRENO_PROTECT_RDONLY(_reg, _len) \
404 ((ilog2((_len)) & 0x1F) << 24) | (((_reg) << 2) & 0xFFFFF))
407 #define gpu_poll_timeout(gpu, addr, val, cond, interval, timeout) \
408 readl_poll_timeout((gpu)->mmio + ((addr) << 2), val, cond, \
411 #endif /* __ADRENO_GPU_H__ */