2 * Copyright (C) 2013 Red Hat
3 * Author: Rob Clark <robdclark@gmail.com>
5 * Copyright (c) 2014 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 #include "adreno_gpu.h"
24 #define RB_SIZE SZ_32K
27 int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value)
29 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
32 case MSM_PARAM_GPU_ID:
33 *value = adreno_gpu->info->revn;
35 case MSM_PARAM_GMEM_SIZE:
36 *value = adreno_gpu->gmem;
38 case MSM_PARAM_GMEM_BASE:
41 case MSM_PARAM_CHIP_ID:
42 *value = adreno_gpu->rev.patchid |
43 (adreno_gpu->rev.minor << 8) |
44 (adreno_gpu->rev.major << 16) |
45 (adreno_gpu->rev.core << 24);
47 case MSM_PARAM_MAX_FREQ:
48 *value = adreno_gpu->base.fast_rate;
50 case MSM_PARAM_TIMESTAMP:
51 if (adreno_gpu->funcs->get_timestamp) {
54 pm_runtime_get_sync(&gpu->pdev->dev);
55 ret = adreno_gpu->funcs->get_timestamp(gpu, value);
56 pm_runtime_put_autosuspend(&gpu->pdev->dev);
62 DBG("%s: invalid param: %u", gpu->name, param);
67 const struct firmware *
68 adreno_request_fw(struct adreno_gpu *adreno_gpu, const char *fwname)
70 struct drm_device *drm = adreno_gpu->base.dev;
71 const struct firmware *fw = NULL;
74 ret = request_firmware(&fw, fwname, drm->dev);
76 dev_err(drm->dev, "failed to load %s: %d\n", fwname, ret);
83 static int adreno_load_fw(struct adreno_gpu *adreno_gpu)
85 const struct firmware *fw;
90 fw = adreno_request_fw(adreno_gpu, adreno_gpu->info->pm4fw);
95 fw = adreno_request_fw(adreno_gpu, adreno_gpu->info->pfpfw);
97 release_firmware(adreno_gpu->pm4);
98 adreno_gpu->pm4 = NULL;
101 adreno_gpu->pfp = fw;
106 int adreno_hw_init(struct msm_gpu *gpu)
108 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
111 DBG("%s", gpu->name);
113 ret = adreno_load_fw(adreno_gpu);
117 ret = msm_gem_get_iova(gpu->rb->bo, gpu->aspace, &gpu->rb_iova);
120 dev_err(gpu->dev->dev, "could not map ringbuffer: %d\n", ret);
124 /* reset ringbuffer: */
125 gpu->rb->cur = gpu->rb->start;
127 /* reset completed fence seqno: */
128 adreno_gpu->memptrs->fence = gpu->fctx->completed_fence;
129 adreno_gpu->memptrs->rptr = 0;
131 /* Setup REG_CP_RB_CNTL: */
132 adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_CNTL,
133 /* size is log2(quad-words): */
134 AXXX_CP_RB_CNTL_BUFSZ(ilog2(gpu->rb->size / 8)) |
135 AXXX_CP_RB_CNTL_BLKSZ(ilog2(RB_BLKSIZE / 8)) |
136 (adreno_is_a430(adreno_gpu) ? AXXX_CP_RB_CNTL_NO_UPDATE : 0));
138 /* Setup ringbuffer address: */
139 adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_BASE,
140 REG_ADRENO_CP_RB_BASE_HI, gpu->rb_iova);
142 if (!adreno_is_a430(adreno_gpu)) {
143 adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_RPTR_ADDR,
144 REG_ADRENO_CP_RB_RPTR_ADDR_HI,
145 rbmemptr(adreno_gpu, rptr));
151 static uint32_t get_wptr(struct msm_ringbuffer *ring)
153 return ring->cur - ring->start;
156 /* Use this helper to read rptr, since a430 doesn't update rptr in memory */
157 static uint32_t get_rptr(struct adreno_gpu *adreno_gpu)
159 if (adreno_is_a430(adreno_gpu))
160 return adreno_gpu->memptrs->rptr = adreno_gpu_read(
161 adreno_gpu, REG_ADRENO_CP_RB_RPTR);
163 return adreno_gpu->memptrs->rptr;
166 uint32_t adreno_last_fence(struct msm_gpu *gpu)
168 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
169 return adreno_gpu->memptrs->fence;
172 void adreno_recover(struct msm_gpu *gpu)
174 struct drm_device *dev = gpu->dev;
177 // XXX pm-runtime?? we *need* the device to be off after this
178 // so maybe continuing to call ->pm_suspend/resume() is better?
180 gpu->funcs->pm_suspend(gpu);
181 gpu->funcs->pm_resume(gpu);
183 ret = msm_gpu_hw_init(gpu);
185 dev_err(dev->dev, "gpu hw init failed: %d\n", ret);
190 void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
191 struct msm_file_private *ctx)
193 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
194 struct msm_drm_private *priv = gpu->dev->dev_private;
195 struct msm_ringbuffer *ring = gpu->rb;
198 for (i = 0; i < submit->nr_cmds; i++) {
199 switch (submit->cmd[i].type) {
200 case MSM_SUBMIT_CMD_IB_TARGET_BUF:
201 /* ignore IB-targets */
203 case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
204 /* ignore if there has not been a ctx switch: */
205 if (priv->lastctx == ctx)
207 case MSM_SUBMIT_CMD_BUF:
208 OUT_PKT3(ring, adreno_is_a430(adreno_gpu) ?
209 CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2);
210 OUT_RING(ring, submit->cmd[i].iova);
211 OUT_RING(ring, submit->cmd[i].size);
217 OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG2, 1);
218 OUT_RING(ring, submit->fence->seqno);
220 if (adreno_is_a3xx(adreno_gpu) || adreno_is_a4xx(adreno_gpu)) {
221 /* Flush HLSQ lazy updates to make sure there is nothing
222 * pending for indirect loads after the timestamp has
225 OUT_PKT3(ring, CP_EVENT_WRITE, 1);
226 OUT_RING(ring, HLSQ_FLUSH);
228 OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
229 OUT_RING(ring, 0x00000000);
232 OUT_PKT3(ring, CP_EVENT_WRITE, 3);
233 OUT_RING(ring, CACHE_FLUSH_TS);
234 OUT_RING(ring, rbmemptr(adreno_gpu, fence));
235 OUT_RING(ring, submit->fence->seqno);
237 /* we could maybe be clever and only CP_COND_EXEC the interrupt: */
238 OUT_PKT3(ring, CP_INTERRUPT, 1);
239 OUT_RING(ring, 0x80000000);
241 /* Workaround for missing irq issue on 8x16/a306. Unsure if the
242 * root cause is a platform issue or some a306 quirk, but this
243 * keeps things humming along:
245 if (adreno_is_a306(adreno_gpu)) {
246 OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
247 OUT_RING(ring, 0x00000000);
248 OUT_PKT3(ring, CP_INTERRUPT, 1);
249 OUT_RING(ring, 0x80000000);
253 if (adreno_is_a3xx(adreno_gpu)) {
254 /* Dummy set-constant to trigger context rollover */
255 OUT_PKT3(ring, CP_SET_CONSTANT, 2);
256 OUT_RING(ring, CP_REG(REG_A3XX_HLSQ_CL_KERNEL_GROUP_X_REG));
257 OUT_RING(ring, 0x00000000);
261 gpu->funcs->flush(gpu);
264 void adreno_flush(struct msm_gpu *gpu)
266 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
270 * Mask wptr value that we calculate to fit in the HW range. This is
271 * to account for the possibility that the last command fit exactly into
272 * the ringbuffer and rb->next hasn't wrapped to zero yet
274 wptr = get_wptr(gpu->rb) & ((gpu->rb->size / 4) - 1);
276 /* ensure writes to ringbuffer have hit system memory: */
279 adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_WPTR, wptr);
282 bool adreno_idle(struct msm_gpu *gpu)
284 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
285 uint32_t wptr = get_wptr(gpu->rb);
287 /* wait for CP to drain ringbuffer: */
288 if (!spin_until(get_rptr(adreno_gpu) == wptr))
291 /* TODO maybe we need to reset GPU here to recover from hang? */
292 DRM_ERROR("%s: timeout waiting to drain ringbuffer!\n", gpu->name);
296 #ifdef CONFIG_DEBUG_FS
297 void adreno_show(struct msm_gpu *gpu, struct seq_file *m)
299 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
302 seq_printf(m, "revision: %d (%d.%d.%d.%d)\n",
303 adreno_gpu->info->revn, adreno_gpu->rev.core,
304 adreno_gpu->rev.major, adreno_gpu->rev.minor,
305 adreno_gpu->rev.patchid);
307 seq_printf(m, "fence: %d/%d\n", adreno_gpu->memptrs->fence,
308 gpu->fctx->last_fence);
309 seq_printf(m, "rptr: %d\n", get_rptr(adreno_gpu));
310 seq_printf(m, "rb wptr: %d\n", get_wptr(gpu->rb));
312 /* dump these out in a form that can be parsed by demsm: */
313 seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name);
314 for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
315 uint32_t start = adreno_gpu->registers[i];
316 uint32_t end = adreno_gpu->registers[i+1];
319 for (addr = start; addr <= end; addr++) {
320 uint32_t val = gpu_read(gpu, addr);
321 seq_printf(m, "IO:R %08x %08x\n", addr<<2, val);
327 /* Dump common gpu status and scratch registers on any hang, to make
328 * the hangcheck logs more useful. The scratch registers seem always
329 * safe to read when GPU has hung (unlike some other regs, depending
330 * on how the GPU hung), and they are useful to match up to cmdstream
331 * dumps when debugging hangs:
333 void adreno_dump_info(struct msm_gpu *gpu)
335 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
337 printk("revision: %d (%d.%d.%d.%d)\n",
338 adreno_gpu->info->revn, adreno_gpu->rev.core,
339 adreno_gpu->rev.major, adreno_gpu->rev.minor,
340 adreno_gpu->rev.patchid);
342 printk("fence: %d/%d\n", adreno_gpu->memptrs->fence,
343 gpu->fctx->last_fence);
344 printk("rptr: %d\n", get_rptr(adreno_gpu));
345 printk("rb wptr: %d\n", get_wptr(gpu->rb));
348 /* would be nice to not have to duplicate the _show() stuff with printk(): */
349 void adreno_dump(struct msm_gpu *gpu)
351 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
354 /* dump these out in a form that can be parsed by demsm: */
355 printk("IO:region %s 00000000 00020000\n", gpu->name);
356 for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
357 uint32_t start = adreno_gpu->registers[i];
358 uint32_t end = adreno_gpu->registers[i+1];
361 for (addr = start; addr <= end; addr++) {
362 uint32_t val = gpu_read(gpu, addr);
363 printk("IO:R %08x %08x\n", addr<<2, val);
368 static uint32_t ring_freewords(struct msm_gpu *gpu)
370 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
371 uint32_t size = gpu->rb->size / 4;
372 uint32_t wptr = get_wptr(gpu->rb);
373 uint32_t rptr = get_rptr(adreno_gpu);
374 return (rptr + (size - 1) - wptr) % size;
377 void adreno_wait_ring(struct msm_gpu *gpu, uint32_t ndwords)
379 if (spin_until(ring_freewords(gpu) >= ndwords))
380 DRM_ERROR("%s: timeout waiting for ringbuffer space\n", gpu->name);
383 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
384 struct adreno_gpu *adreno_gpu, const struct adreno_gpu_funcs *funcs)
386 struct adreno_platform_config *config = pdev->dev.platform_data;
387 struct msm_gpu_config adreno_gpu_config = { 0 };
388 struct msm_gpu *gpu = &adreno_gpu->base;
391 adreno_gpu->funcs = funcs;
392 adreno_gpu->info = adreno_info(config->rev);
393 adreno_gpu->gmem = adreno_gpu->info->gmem;
394 adreno_gpu->revn = adreno_gpu->info->revn;
395 adreno_gpu->rev = config->rev;
397 gpu->fast_rate = config->fast_rate;
398 gpu->bus_freq = config->bus_freq;
399 #ifdef DOWNSTREAM_CONFIG_MSM_BUS_SCALING
400 gpu->bus_scale_table = config->bus_scale_table;
403 DBG("fast_rate=%u, slow_rate=27000000, bus_freq=%u",
404 gpu->fast_rate, gpu->bus_freq);
406 adreno_gpu_config.ioname = "kgsl_3d0_reg_memory";
407 adreno_gpu_config.irqname = "kgsl_3d0_irq";
409 adreno_gpu_config.va_start = SZ_16M;
410 adreno_gpu_config.va_end = 0xffffffff;
412 adreno_gpu_config.ringsz = RB_SIZE;
414 pm_runtime_set_autosuspend_delay(&pdev->dev, DRM_MSM_INACTIVE_PERIOD);
415 pm_runtime_use_autosuspend(&pdev->dev);
416 pm_runtime_enable(&pdev->dev);
418 ret = msm_gpu_init(drm, pdev, &adreno_gpu->base, &funcs->base,
419 adreno_gpu->info->name, &adreno_gpu_config);
423 adreno_gpu->memptrs = msm_gem_kernel_new(drm,
424 sizeof(*adreno_gpu->memptrs), MSM_BO_UNCACHED, gpu->aspace,
425 &adreno_gpu->memptrs_bo, &adreno_gpu->memptrs_iova);
427 if (IS_ERR(adreno_gpu->memptrs)) {
428 ret = PTR_ERR(adreno_gpu->memptrs);
429 adreno_gpu->memptrs = NULL;
430 dev_err(drm->dev, "could not allocate memptrs: %d\n", ret);
436 void adreno_gpu_cleanup(struct adreno_gpu *adreno_gpu)
438 struct msm_gpu *gpu = &adreno_gpu->base;
440 if (adreno_gpu->memptrs_bo) {
441 if (adreno_gpu->memptrs)
442 msm_gem_put_vaddr(adreno_gpu->memptrs_bo);
444 if (adreno_gpu->memptrs_iova)
445 msm_gem_put_iova(adreno_gpu->memptrs_bo, gpu->aspace);
447 drm_gem_object_unreference_unlocked(adreno_gpu->memptrs_bo);
449 release_firmware(adreno_gpu->pm4);
450 release_firmware(adreno_gpu->pfp);
452 msm_gpu_cleanup(gpu);