1 // SPDX-License-Identifier: GPL-2.0+
2 /* Copyright (C) 2014-2018 Broadcom */
5 * DOC: Broadcom V3D Graphics Driver
7 * This driver supports the Broadcom V3D 3.3 and 4.1 OpenGL ES GPUs.
8 * For V3D 2.x support, see the VC4 driver.
10 * The V3D GPU includes a tiled render (composed of a bin and render
11 * pipelines), the TFU (texture formatting unit), and the CSD (compute
15 #include <linux/clk.h>
16 #include <linux/device.h>
17 #include <linux/dma-mapping.h>
19 #include <linux/module.h>
20 #include <linux/of_platform.h>
21 #include <linux/platform_device.h>
22 #include <linux/sched/clock.h>
23 #include <linux/reset.h>
25 #include <drm/drm_drv.h>
26 #include <drm/drm_managed.h>
27 #include <uapi/drm/v3d_drm.h>
32 #define DRIVER_NAME "v3d"
33 #define DRIVER_DESC "Broadcom V3D graphics"
34 #define DRIVER_MAJOR 1
35 #define DRIVER_MINOR 0
36 #define DRIVER_PATCHLEVEL 0
38 /* Only expose the `super_pages` modparam if THP is enabled. */
39 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
40 bool super_pages = true;
41 module_param_named(super_pages, super_pages, bool, 0400);
42 MODULE_PARM_DESC(super_pages, "Enable/Disable Super Pages support.");
45 static int v3d_get_param_ioctl(struct drm_device *dev, void *data,
46 struct drm_file *file_priv)
48 struct v3d_dev *v3d = to_v3d_dev(dev);
49 struct drm_v3d_get_param *args = data;
50 static const u32 reg_map[] = {
51 [DRM_V3D_PARAM_V3D_UIFCFG] = V3D_HUB_UIFCFG,
52 [DRM_V3D_PARAM_V3D_HUB_IDENT1] = V3D_HUB_IDENT1,
53 [DRM_V3D_PARAM_V3D_HUB_IDENT2] = V3D_HUB_IDENT2,
54 [DRM_V3D_PARAM_V3D_HUB_IDENT3] = V3D_HUB_IDENT3,
55 [DRM_V3D_PARAM_V3D_CORE0_IDENT0] = V3D_CTL_IDENT0,
56 [DRM_V3D_PARAM_V3D_CORE0_IDENT1] = V3D_CTL_IDENT1,
57 [DRM_V3D_PARAM_V3D_CORE0_IDENT2] = V3D_CTL_IDENT2,
63 /* Note that DRM_V3D_PARAM_V3D_CORE0_IDENT0 is 0, so we need
64 * to explicitly allow it in the "the register in our
65 * parameter map" check.
67 if (args->param < ARRAY_SIZE(reg_map) &&
68 (reg_map[args->param] ||
69 args->param == DRM_V3D_PARAM_V3D_CORE0_IDENT0)) {
70 u32 offset = reg_map[args->param];
75 if (args->param >= DRM_V3D_PARAM_V3D_CORE0_IDENT0 &&
76 args->param <= DRM_V3D_PARAM_V3D_CORE0_IDENT2) {
77 args->value = V3D_CORE_READ(0, offset);
79 args->value = V3D_READ(offset);
84 switch (args->param) {
85 case DRM_V3D_PARAM_SUPPORTS_TFU:
88 case DRM_V3D_PARAM_SUPPORTS_CSD:
89 args->value = v3d_has_csd(v3d);
91 case DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH:
94 case DRM_V3D_PARAM_SUPPORTS_PERFMON:
95 args->value = (v3d->ver >= 40);
97 case DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT:
100 case DRM_V3D_PARAM_SUPPORTS_CPU_QUEUE:
103 case DRM_V3D_PARAM_MAX_PERF_COUNTERS:
104 args->value = v3d->perfmon_info.max_counters;
106 case DRM_V3D_PARAM_SUPPORTS_SUPER_PAGES:
107 args->value = !!v3d->gemfs;
110 DRM_DEBUG("Unknown parameter %d\n", args->param);
116 v3d_open(struct drm_device *dev, struct drm_file *file)
118 struct v3d_dev *v3d = to_v3d_dev(dev);
119 struct v3d_file_priv *v3d_priv;
120 struct drm_gpu_scheduler *sched;
123 v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
129 for (i = 0; i < V3D_MAX_QUEUES; i++) {
130 sched = &v3d->queue[i].sched;
131 drm_sched_entity_init(&v3d_priv->sched_entity[i],
132 DRM_SCHED_PRIORITY_NORMAL, &sched,
135 memset(&v3d_priv->stats[i], 0, sizeof(v3d_priv->stats[i]));
136 seqcount_init(&v3d_priv->stats[i].lock);
139 v3d_perfmon_open_file(v3d_priv);
140 file->driver_priv = v3d_priv;
146 v3d_postclose(struct drm_device *dev, struct drm_file *file)
148 struct v3d_file_priv *v3d_priv = file->driver_priv;
151 for (q = 0; q < V3D_MAX_QUEUES; q++)
152 drm_sched_entity_destroy(&v3d_priv->sched_entity[q]);
154 v3d_perfmon_close_file(v3d_priv);
158 void v3d_get_stats(const struct v3d_stats *stats, u64 timestamp,
159 u64 *active_runtime, u64 *jobs_completed)
164 seq = read_seqcount_begin(&stats->lock);
165 *active_runtime = stats->enabled_ns;
167 *active_runtime += timestamp - stats->start_ns;
168 *jobs_completed = stats->jobs_completed;
169 } while (read_seqcount_retry(&stats->lock, seq));
172 static void v3d_show_fdinfo(struct drm_printer *p, struct drm_file *file)
174 struct v3d_file_priv *file_priv = file->driver_priv;
175 u64 timestamp = local_clock();
176 enum v3d_queue queue;
178 for (queue = 0; queue < V3D_MAX_QUEUES; queue++) {
179 struct v3d_stats *stats = &file_priv->stats[queue];
180 u64 active_runtime, jobs_completed;
182 v3d_get_stats(stats, timestamp, &active_runtime, &jobs_completed);
184 /* Note that, in case of a GPU reset, the time spent during an
185 * attempt of executing the job is not computed in the runtime.
187 drm_printf(p, "drm-engine-%s: \t%llu ns\n",
188 v3d_queue_to_string(queue), active_runtime);
190 /* Note that we only count jobs that completed. Therefore, jobs
191 * that were resubmitted due to a GPU reset are not computed.
193 drm_printf(p, "v3d-jobs-%s: \t%llu jobs\n",
194 v3d_queue_to_string(queue), jobs_completed);
197 drm_show_memory_stats(p, file);
200 static const struct file_operations v3d_drm_fops = {
201 .owner = THIS_MODULE,
203 .show_fdinfo = drm_show_fdinfo,
206 /* DRM_AUTH is required on SUBMIT_CL for now, while we don't have GMP
207 * protection between clients. Note that render nodes would be
208 * able to submit CLs that could access BOs from clients authenticated
209 * with the master node. The TFU doesn't use the GMP, so it would
210 * need to stay DRM_AUTH until we do buffer size/offset validation.
212 static const struct drm_ioctl_desc v3d_drm_ioctls[] = {
213 DRM_IOCTL_DEF_DRV(V3D_SUBMIT_CL, v3d_submit_cl_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
214 DRM_IOCTL_DEF_DRV(V3D_WAIT_BO, v3d_wait_bo_ioctl, DRM_RENDER_ALLOW),
215 DRM_IOCTL_DEF_DRV(V3D_CREATE_BO, v3d_create_bo_ioctl, DRM_RENDER_ALLOW),
216 DRM_IOCTL_DEF_DRV(V3D_MMAP_BO, v3d_mmap_bo_ioctl, DRM_RENDER_ALLOW),
217 DRM_IOCTL_DEF_DRV(V3D_GET_PARAM, v3d_get_param_ioctl, DRM_RENDER_ALLOW),
218 DRM_IOCTL_DEF_DRV(V3D_GET_BO_OFFSET, v3d_get_bo_offset_ioctl, DRM_RENDER_ALLOW),
219 DRM_IOCTL_DEF_DRV(V3D_SUBMIT_TFU, v3d_submit_tfu_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
220 DRM_IOCTL_DEF_DRV(V3D_SUBMIT_CSD, v3d_submit_csd_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
221 DRM_IOCTL_DEF_DRV(V3D_PERFMON_CREATE, v3d_perfmon_create_ioctl, DRM_RENDER_ALLOW),
222 DRM_IOCTL_DEF_DRV(V3D_PERFMON_DESTROY, v3d_perfmon_destroy_ioctl, DRM_RENDER_ALLOW),
223 DRM_IOCTL_DEF_DRV(V3D_PERFMON_GET_VALUES, v3d_perfmon_get_values_ioctl, DRM_RENDER_ALLOW),
224 DRM_IOCTL_DEF_DRV(V3D_SUBMIT_CPU, v3d_submit_cpu_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
225 DRM_IOCTL_DEF_DRV(V3D_PERFMON_GET_COUNTER, v3d_perfmon_get_counter_ioctl, DRM_RENDER_ALLOW),
226 DRM_IOCTL_DEF_DRV(V3D_PERFMON_SET_GLOBAL, v3d_perfmon_set_global_ioctl, DRM_RENDER_ALLOW),
229 static const struct drm_driver v3d_drm_driver = {
230 .driver_features = (DRIVER_GEM |
235 .postclose = v3d_postclose,
237 #if defined(CONFIG_DEBUG_FS)
238 .debugfs_init = v3d_debugfs_init,
241 .gem_create_object = v3d_create_object,
242 .gem_prime_import_sg_table = v3d_prime_import_sg_table,
244 .ioctls = v3d_drm_ioctls,
245 .num_ioctls = ARRAY_SIZE(v3d_drm_ioctls),
246 .fops = &v3d_drm_fops,
247 .show_fdinfo = v3d_show_fdinfo,
251 .major = DRIVER_MAJOR,
252 .minor = DRIVER_MINOR,
253 .patchlevel = DRIVER_PATCHLEVEL,
256 static const struct of_device_id v3d_of_match[] = {
257 { .compatible = "brcm,2711-v3d" },
258 { .compatible = "brcm,2712-v3d" },
259 { .compatible = "brcm,7268-v3d" },
260 { .compatible = "brcm,7278-v3d" },
263 MODULE_DEVICE_TABLE(of, v3d_of_match);
266 map_regs(struct v3d_dev *v3d, void __iomem **regs, const char *name)
268 *regs = devm_platform_ioremap_resource_byname(v3d_to_pdev(v3d), name);
269 return PTR_ERR_OR_ZERO(*regs);
272 static int v3d_platform_drm_probe(struct platform_device *pdev)
274 struct device *dev = &pdev->dev;
275 struct drm_device *drm;
282 v3d = devm_drm_dev_alloc(dev, &v3d_drm_driver, struct v3d_dev, drm);
288 platform_set_drvdata(pdev, drm);
290 ret = map_regs(v3d, &v3d->hub_regs, "hub");
294 ret = map_regs(v3d, &v3d->core_regs[0], "core0");
298 mmu_debug = V3D_READ(V3D_MMU_DEBUG_INFO);
299 mask = DMA_BIT_MASK(30 + V3D_GET_FIELD(mmu_debug, V3D_MMU_PA_WIDTH));
300 ret = dma_set_mask_and_coherent(dev, mask);
304 v3d->va_width = 30 + V3D_GET_FIELD(mmu_debug, V3D_MMU_VA_WIDTH);
306 ident1 = V3D_READ(V3D_HUB_IDENT1);
307 v3d->ver = (V3D_GET_FIELD(ident1, V3D_HUB_IDENT1_TVER) * 10 +
308 V3D_GET_FIELD(ident1, V3D_HUB_IDENT1_REV));
309 v3d->cores = V3D_GET_FIELD(ident1, V3D_HUB_IDENT1_NCORES);
310 WARN_ON(v3d->cores > 1); /* multicore not yet implemented */
312 ident3 = V3D_READ(V3D_HUB_IDENT3);
313 v3d->rev = V3D_GET_FIELD(ident3, V3D_HUB_IDENT3_IPREV);
315 v3d_perfmon_init(v3d);
317 v3d->reset = devm_reset_control_get_exclusive(dev, NULL);
318 if (IS_ERR(v3d->reset)) {
319 ret = PTR_ERR(v3d->reset);
321 if (ret == -EPROBE_DEFER)
325 ret = map_regs(v3d, &v3d->bridge_regs, "bridge");
328 "Failed to get reset control or bridge regs\n");
334 ret = map_regs(v3d, &v3d->gca_regs, "gca");
339 v3d->mmu_scratch = dma_alloc_wc(dev, 4096, &v3d->mmu_scratch_paddr,
340 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
341 if (!v3d->mmu_scratch) {
342 dev_err(dev, "Failed to allocate MMU scratch page\n");
346 ret = v3d_gem_init(drm);
350 ret = v3d_irq_init(v3d);
354 ret = drm_dev_register(drm, 0);
358 ret = v3d_sysfs_init(dev);
365 drm_dev_unregister(drm);
367 v3d_irq_disable(v3d);
369 v3d_gem_destroy(drm);
371 dma_free_wc(dev, 4096, v3d->mmu_scratch, v3d->mmu_scratch_paddr);
375 static void v3d_platform_drm_remove(struct platform_device *pdev)
377 struct drm_device *drm = platform_get_drvdata(pdev);
378 struct v3d_dev *v3d = to_v3d_dev(drm);
379 struct device *dev = &pdev->dev;
381 v3d_sysfs_destroy(dev);
383 drm_dev_unregister(drm);
385 v3d_gem_destroy(drm);
387 dma_free_wc(v3d->drm.dev, 4096, v3d->mmu_scratch,
388 v3d->mmu_scratch_paddr);
391 static struct platform_driver v3d_platform_driver = {
392 .probe = v3d_platform_drm_probe,
393 .remove = v3d_platform_drm_remove,
396 .of_match_table = v3d_of_match,
400 module_platform_driver(v3d_platform_driver);
402 MODULE_ALIAS("platform:v3d-drm");
403 MODULE_DESCRIPTION("Broadcom V3D DRM Driver");
405 MODULE_LICENSE("GPL v2");