2 * Copyright (C) 2013 Red Hat
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 <linux/pm_opp.h>
21 #include "adreno_gpu.h"
25 int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value)
27 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
30 case MSM_PARAM_GPU_ID:
31 *value = adreno_gpu->info->revn;
33 case MSM_PARAM_GMEM_SIZE:
34 *value = adreno_gpu->gmem;
36 case MSM_PARAM_GMEM_BASE:
39 case MSM_PARAM_CHIP_ID:
40 *value = adreno_gpu->rev.patchid |
41 (adreno_gpu->rev.minor << 8) |
42 (adreno_gpu->rev.major << 16) |
43 (adreno_gpu->rev.core << 24);
45 case MSM_PARAM_MAX_FREQ:
46 *value = adreno_gpu->base.fast_rate;
48 case MSM_PARAM_TIMESTAMP:
49 if (adreno_gpu->funcs->get_timestamp) {
52 pm_runtime_get_sync(&gpu->pdev->dev);
53 ret = adreno_gpu->funcs->get_timestamp(gpu, value);
54 pm_runtime_put_autosuspend(&gpu->pdev->dev);
59 case MSM_PARAM_NR_RINGS:
60 *value = gpu->nr_rings;
63 DBG("%s: invalid param: %u", gpu->name, param);
68 const struct firmware *
69 adreno_request_fw(struct adreno_gpu *adreno_gpu, const char *fwname)
71 struct drm_device *drm = adreno_gpu->base.dev;
72 const struct firmware *fw = NULL;
73 char newname[strlen("qcom/") + strlen(fwname) + 1];
76 sprintf(newname, "qcom/%s", fwname);
79 * Try first to load from qcom/$fwfile using a direct load (to avoid
80 * a potential timeout waiting for usermode helper)
82 if ((adreno_gpu->fwloc == FW_LOCATION_UNKNOWN) ||
83 (adreno_gpu->fwloc == FW_LOCATION_NEW)) {
85 ret = request_firmware_direct(&fw, newname, drm->dev);
87 dev_info(drm->dev, "loaded %s from new location\n",
89 adreno_gpu->fwloc = FW_LOCATION_NEW;
91 } else if (adreno_gpu->fwloc != FW_LOCATION_UNKNOWN) {
92 dev_err(drm->dev, "failed to load %s: %d\n",
99 * Then try the legacy location without qcom/ prefix
101 if ((adreno_gpu->fwloc == FW_LOCATION_UNKNOWN) ||
102 (adreno_gpu->fwloc == FW_LOCATION_LEGACY)) {
104 ret = request_firmware_direct(&fw, fwname, drm->dev);
106 dev_info(drm->dev, "loaded %s from legacy location\n",
108 adreno_gpu->fwloc = FW_LOCATION_LEGACY;
110 } else if (adreno_gpu->fwloc != FW_LOCATION_UNKNOWN) {
111 dev_err(drm->dev, "failed to load %s: %d\n",
118 * Finally fall back to request_firmware() for cases where the
119 * usermode helper is needed (I think mainly android)
121 if ((adreno_gpu->fwloc == FW_LOCATION_UNKNOWN) ||
122 (adreno_gpu->fwloc == FW_LOCATION_HELPER)) {
124 ret = request_firmware(&fw, newname, drm->dev);
126 dev_info(drm->dev, "loaded %s with helper\n",
128 adreno_gpu->fwloc = FW_LOCATION_HELPER;
130 } else if (adreno_gpu->fwloc != FW_LOCATION_UNKNOWN) {
131 dev_err(drm->dev, "failed to load %s: %d\n",
137 dev_err(drm->dev, "failed to load %s\n", fwname);
138 return ERR_PTR(-ENOENT);
141 static int adreno_load_fw(struct adreno_gpu *adreno_gpu)
145 for (i = 0; i < ARRAY_SIZE(adreno_gpu->info->fw); i++) {
146 const struct firmware *fw;
148 if (!adreno_gpu->info->fw[i])
151 /* Skip if the firmware has already been loaded */
152 if (adreno_gpu->fw[i])
155 fw = adreno_request_fw(adreno_gpu, adreno_gpu->info->fw[i]);
159 adreno_gpu->fw[i] = fw;
165 struct drm_gem_object *adreno_fw_create_bo(struct msm_gpu *gpu,
166 const struct firmware *fw, u64 *iova)
168 struct drm_gem_object *bo;
171 ptr = msm_gem_kernel_new_locked(gpu->dev, fw->size - 4,
172 MSM_BO_UNCACHED | MSM_BO_GPU_READONLY, gpu->aspace, &bo, iova);
175 return ERR_CAST(ptr);
177 memcpy(ptr, &fw->data[4], fw->size - 4);
179 msm_gem_put_vaddr(bo);
184 int adreno_hw_init(struct msm_gpu *gpu)
186 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
189 DBG("%s", gpu->name);
191 ret = adreno_load_fw(adreno_gpu);
195 for (i = 0; i < gpu->nr_rings; i++) {
196 struct msm_ringbuffer *ring = gpu->rb[i];
201 ret = msm_gem_get_iova(ring->bo, gpu->aspace, &ring->iova);
204 dev_err(gpu->dev->dev,
205 "could not map ringbuffer %d: %d\n", i, ret);
209 ring->cur = ring->start;
210 ring->next = ring->start;
212 /* reset completed fence seqno: */
213 ring->memptrs->fence = ring->seqno;
214 ring->memptrs->rptr = 0;
218 * Setup REG_CP_RB_CNTL. The same value is used across targets (with
219 * the excpetion of A430 that disables the RPTR shadow) - the cacluation
220 * for the ringbuffer size and block size is moved to msm_gpu.h for the
221 * pre-processor to deal with and the A430 variant is ORed in here
223 adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_CNTL,
224 MSM_GPU_RB_CNTL_DEFAULT |
225 (adreno_is_a430(adreno_gpu) ? AXXX_CP_RB_CNTL_NO_UPDATE : 0));
227 /* Setup ringbuffer address - use ringbuffer[0] for GPU init */
228 adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_BASE,
229 REG_ADRENO_CP_RB_BASE_HI, gpu->rb[0]->iova);
231 if (!adreno_is_a430(adreno_gpu)) {
232 adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_RPTR_ADDR,
233 REG_ADRENO_CP_RB_RPTR_ADDR_HI,
234 rbmemptr(gpu->rb[0], rptr));
240 /* Use this helper to read rptr, since a430 doesn't update rptr in memory */
241 static uint32_t get_rptr(struct adreno_gpu *adreno_gpu,
242 struct msm_ringbuffer *ring)
244 if (adreno_is_a430(adreno_gpu))
245 return ring->memptrs->rptr = adreno_gpu_read(
246 adreno_gpu, REG_ADRENO_CP_RB_RPTR);
248 return ring->memptrs->rptr;
251 struct msm_ringbuffer *adreno_active_ring(struct msm_gpu *gpu)
256 void adreno_recover(struct msm_gpu *gpu)
258 struct drm_device *dev = gpu->dev;
261 // XXX pm-runtime?? we *need* the device to be off after this
262 // so maybe continuing to call ->pm_suspend/resume() is better?
264 gpu->funcs->pm_suspend(gpu);
265 gpu->funcs->pm_resume(gpu);
267 ret = msm_gpu_hw_init(gpu);
269 dev_err(dev->dev, "gpu hw init failed: %d\n", ret);
274 void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
275 struct msm_file_private *ctx)
277 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
278 struct msm_drm_private *priv = gpu->dev->dev_private;
279 struct msm_ringbuffer *ring = submit->ring;
282 for (i = 0; i < submit->nr_cmds; i++) {
283 switch (submit->cmd[i].type) {
284 case MSM_SUBMIT_CMD_IB_TARGET_BUF:
285 /* ignore IB-targets */
287 case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
288 /* ignore if there has not been a ctx switch: */
289 if (priv->lastctx == ctx)
291 case MSM_SUBMIT_CMD_BUF:
292 OUT_PKT3(ring, adreno_is_a430(adreno_gpu) ?
293 CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2);
294 OUT_RING(ring, lower_32_bits(submit->cmd[i].iova));
295 OUT_RING(ring, submit->cmd[i].size);
301 OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG2, 1);
302 OUT_RING(ring, submit->seqno);
304 if (adreno_is_a3xx(adreno_gpu) || adreno_is_a4xx(adreno_gpu)) {
305 /* Flush HLSQ lazy updates to make sure there is nothing
306 * pending for indirect loads after the timestamp has
309 OUT_PKT3(ring, CP_EVENT_WRITE, 1);
310 OUT_RING(ring, HLSQ_FLUSH);
312 OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
313 OUT_RING(ring, 0x00000000);
316 /* BIT(31) of CACHE_FLUSH_TS triggers CACHE_FLUSH_TS IRQ from GPU */
317 OUT_PKT3(ring, CP_EVENT_WRITE, 3);
318 OUT_RING(ring, CACHE_FLUSH_TS | BIT(31));
319 OUT_RING(ring, rbmemptr(ring, fence));
320 OUT_RING(ring, submit->seqno);
323 if (adreno_is_a3xx(adreno_gpu)) {
324 /* Dummy set-constant to trigger context rollover */
325 OUT_PKT3(ring, CP_SET_CONSTANT, 2);
326 OUT_RING(ring, CP_REG(REG_A3XX_HLSQ_CL_KERNEL_GROUP_X_REG));
327 OUT_RING(ring, 0x00000000);
331 gpu->funcs->flush(gpu, ring);
334 void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
336 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
339 /* Copy the shadow to the actual register */
340 ring->cur = ring->next;
343 * Mask wptr value that we calculate to fit in the HW range. This is
344 * to account for the possibility that the last command fit exactly into
345 * the ringbuffer and rb->next hasn't wrapped to zero yet
347 wptr = get_wptr(ring);
349 /* ensure writes to ringbuffer have hit system memory: */
352 adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_WPTR, wptr);
355 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
357 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
358 uint32_t wptr = get_wptr(ring);
360 /* wait for CP to drain ringbuffer: */
361 if (!spin_until(get_rptr(adreno_gpu, ring) == wptr))
364 /* TODO maybe we need to reset GPU here to recover from hang? */
365 DRM_ERROR("%s: timeout waiting to drain ringbuffer %d rptr/wptr = %X/%X\n",
366 gpu->name, ring->id, get_rptr(adreno_gpu, ring), wptr);
371 #ifdef CONFIG_DEBUG_FS
372 void adreno_show(struct msm_gpu *gpu, struct seq_file *m)
374 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
377 seq_printf(m, "revision: %d (%d.%d.%d.%d)\n",
378 adreno_gpu->info->revn, adreno_gpu->rev.core,
379 adreno_gpu->rev.major, adreno_gpu->rev.minor,
380 adreno_gpu->rev.patchid);
382 for (i = 0; i < gpu->nr_rings; i++) {
383 struct msm_ringbuffer *ring = gpu->rb[i];
385 seq_printf(m, "rb %d: fence: %d/%d\n", i,
386 ring->memptrs->fence, ring->seqno);
388 seq_printf(m, " rptr: %d\n",
389 get_rptr(adreno_gpu, ring));
390 seq_printf(m, "rb wptr: %d\n", get_wptr(ring));
393 /* dump these out in a form that can be parsed by demsm: */
394 seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name);
395 for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
396 uint32_t start = adreno_gpu->registers[i];
397 uint32_t end = adreno_gpu->registers[i+1];
400 for (addr = start; addr <= end; addr++) {
401 uint32_t val = gpu_read(gpu, addr);
402 seq_printf(m, "IO:R %08x %08x\n", addr<<2, val);
408 /* Dump common gpu status and scratch registers on any hang, to make
409 * the hangcheck logs more useful. The scratch registers seem always
410 * safe to read when GPU has hung (unlike some other regs, depending
411 * on how the GPU hung), and they are useful to match up to cmdstream
412 * dumps when debugging hangs:
414 void adreno_dump_info(struct msm_gpu *gpu)
416 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
419 printk("revision: %d (%d.%d.%d.%d)\n",
420 adreno_gpu->info->revn, adreno_gpu->rev.core,
421 adreno_gpu->rev.major, adreno_gpu->rev.minor,
422 adreno_gpu->rev.patchid);
424 for (i = 0; i < gpu->nr_rings; i++) {
425 struct msm_ringbuffer *ring = gpu->rb[i];
427 printk("rb %d: fence: %d/%d\n", i,
428 ring->memptrs->fence,
431 printk("rptr: %d\n", get_rptr(adreno_gpu, ring));
432 printk("rb wptr: %d\n", get_wptr(ring));
436 /* would be nice to not have to duplicate the _show() stuff with printk(): */
437 void adreno_dump(struct msm_gpu *gpu)
439 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
442 /* dump these out in a form that can be parsed by demsm: */
443 printk("IO:region %s 00000000 00020000\n", gpu->name);
444 for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
445 uint32_t start = adreno_gpu->registers[i];
446 uint32_t end = adreno_gpu->registers[i+1];
449 for (addr = start; addr <= end; addr++) {
450 uint32_t val = gpu_read(gpu, addr);
451 printk("IO:R %08x %08x\n", addr<<2, val);
456 static uint32_t ring_freewords(struct msm_ringbuffer *ring)
458 struct adreno_gpu *adreno_gpu = to_adreno_gpu(ring->gpu);
459 uint32_t size = MSM_GPU_RINGBUFFER_SZ >> 2;
460 /* Use ring->next to calculate free size */
461 uint32_t wptr = ring->next - ring->start;
462 uint32_t rptr = get_rptr(adreno_gpu, ring);
463 return (rptr + (size - 1) - wptr) % size;
466 void adreno_wait_ring(struct msm_ringbuffer *ring, uint32_t ndwords)
468 if (spin_until(ring_freewords(ring) >= ndwords))
469 DRM_DEV_ERROR(ring->gpu->dev->dev,
470 "timeout waiting for space in ringbuffer %d\n",
474 /* Get legacy powerlevels from qcom,gpu-pwrlevels and populate the opp table */
475 static int adreno_get_legacy_pwrlevels(struct device *dev)
477 struct device_node *child, *node;
480 node = of_find_compatible_node(dev->of_node, NULL,
481 "qcom,gpu-pwrlevels");
483 dev_err(dev, "Could not find the GPU powerlevels\n");
487 for_each_child_of_node(node, child) {
490 ret = of_property_read_u32(child, "qcom,gpu-freq", &val);
495 * Skip the intentionally bogus clock value found at the bottom
496 * of most legacy frequency tables
499 dev_pm_opp_add(dev, val, 0);
505 static int adreno_get_pwrlevels(struct device *dev,
508 unsigned long freq = ULONG_MAX;
509 struct dev_pm_opp *opp;
514 /* You down with OPP? */
515 if (!of_find_property(dev->of_node, "operating-points-v2", NULL))
516 ret = adreno_get_legacy_pwrlevels(dev);
518 ret = dev_pm_opp_of_add_table(dev);
520 dev_err(dev, "Unable to set the OPP table\n");
524 /* Find the fastest defined rate */
525 opp = dev_pm_opp_find_freq_floor(dev, &freq);
527 gpu->fast_rate = freq;
532 if (!gpu->fast_rate) {
534 "Could not find a clock rate. Using a reasonable default\n");
535 /* Pick a suitably safe clock speed for any target */
536 gpu->fast_rate = 200000000;
539 DBG("fast_rate=%u, slow_rate=27000000", gpu->fast_rate);
544 int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
545 struct adreno_gpu *adreno_gpu,
546 const struct adreno_gpu_funcs *funcs, int nr_rings)
548 struct adreno_platform_config *config = pdev->dev.platform_data;
549 struct msm_gpu_config adreno_gpu_config = { 0 };
550 struct msm_gpu *gpu = &adreno_gpu->base;
552 adreno_gpu->funcs = funcs;
553 adreno_gpu->info = adreno_info(config->rev);
554 adreno_gpu->gmem = adreno_gpu->info->gmem;
555 adreno_gpu->revn = adreno_gpu->info->revn;
556 adreno_gpu->rev = config->rev;
558 adreno_gpu_config.ioname = "kgsl_3d0_reg_memory";
559 adreno_gpu_config.irqname = "kgsl_3d0_irq";
561 adreno_gpu_config.va_start = SZ_16M;
562 adreno_gpu_config.va_end = 0xffffffff;
564 adreno_gpu_config.nr_rings = nr_rings;
566 adreno_get_pwrlevels(&pdev->dev, gpu);
568 pm_runtime_set_autosuspend_delay(&pdev->dev, DRM_MSM_INACTIVE_PERIOD);
569 pm_runtime_use_autosuspend(&pdev->dev);
570 pm_runtime_enable(&pdev->dev);
572 return msm_gpu_init(drm, pdev, &adreno_gpu->base, &funcs->base,
573 adreno_gpu->info->name, &adreno_gpu_config);
576 void adreno_gpu_cleanup(struct adreno_gpu *adreno_gpu)
580 for (i = 0; i < ARRAY_SIZE(adreno_gpu->info->fw); i++)
581 release_firmware(adreno_gpu->fw[i]);
583 msm_gpu_cleanup(&adreno_gpu->base);