2 * Copyright (c) 2014 The Linux Foundation. All rights reserved.
3 * Copyright (C) 2013 Red Hat
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program. If not, see <http://www.gnu.org/licenses/>.
19 #include <linux/clk.h>
20 #include <linux/component.h>
21 #include <linux/pm_runtime.h>
25 #ifdef CONFIG_DEBUG_FS
26 #define REGDEF(reg) { reg, #reg }
111 int vc4_v3d_debugfs_regs(struct seq_file *m, void *unused)
113 struct drm_info_node *node = (struct drm_info_node *)m->private;
114 struct drm_device *dev = node->minor->dev;
115 struct vc4_dev *vc4 = to_vc4_dev(dev);
118 for (i = 0; i < ARRAY_SIZE(vc4_reg_defs); i++) {
119 seq_printf(m, "%s (0x%04x): 0x%08x\n",
120 vc4_reg_defs[i].name, vc4_reg_defs[i].reg,
121 V3D_READ(vc4_reg_defs[i].reg));
127 int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused)
129 struct drm_info_node *node = (struct drm_info_node *)m->private;
130 struct drm_device *dev = node->minor->dev;
131 struct vc4_dev *vc4 = to_vc4_dev(dev);
132 uint32_t ident1 = V3D_READ(V3D_IDENT1);
133 uint32_t nslc = VC4_GET_FIELD(ident1, V3D_IDENT1_NSLC);
134 uint32_t tups = VC4_GET_FIELD(ident1, V3D_IDENT1_TUPS);
135 uint32_t qups = VC4_GET_FIELD(ident1, V3D_IDENT1_QUPS);
137 seq_printf(m, "Revision: %d\n",
138 VC4_GET_FIELD(ident1, V3D_IDENT1_REV));
139 seq_printf(m, "Slices: %d\n", nslc);
140 seq_printf(m, "TMUs: %d\n", nslc * tups);
141 seq_printf(m, "QPUs: %d\n", nslc * qups);
142 seq_printf(m, "Semaphores: %d\n",
143 VC4_GET_FIELD(ident1, V3D_IDENT1_NSEM));
147 #endif /* CONFIG_DEBUG_FS */
149 static void vc4_v3d_init_hw(struct drm_device *dev)
151 struct vc4_dev *vc4 = to_vc4_dev(dev);
153 /* Take all the memory that would have been reserved for user
154 * QPU programs, since we don't have an interface for running
157 V3D_WRITE(V3D_VPMBASE, 0);
160 int vc4_v3d_get_bin_slot(struct vc4_dev *vc4)
162 struct drm_device *dev = vc4->dev;
163 unsigned long irqflags;
166 struct vc4_exec_info *exec;
169 spin_lock_irqsave(&vc4->job_lock, irqflags);
170 slot = ffs(~vc4->bin_alloc_used);
172 /* Switch from ffs() bit index to a 0-based index. */
174 vc4->bin_alloc_used |= BIT(slot);
175 spin_unlock_irqrestore(&vc4->job_lock, irqflags);
179 /* Couldn't find an open slot. Wait for render to complete
182 exec = vc4_last_render_job(vc4);
185 spin_unlock_irqrestore(&vc4->job_lock, irqflags);
188 int ret = vc4_wait_for_seqno(dev, seqno, ~0ull, true);
200 * vc4_allocate_bin_bo() - allocates the memory that will be used for
203 * The binner has a limitation that the addresses in the tile state
204 * buffer that point into the tile alloc buffer or binner overflow
205 * memory only have 28 bits (256MB), and the top 4 on the bus for
206 * tile alloc references end up coming from the tile state buffer's
209 * To work around this, we allocate a single large buffer while V3D is
210 * in use, make sure that it has the top 4 bits constant across its
211 * entire extent, and then put the tile state, tile alloc, and binner
212 * overflow memory inside that buffer.
214 * This creates a limitation where we may not be able to execute a job
215 * if it doesn't fit within the buffer that we allocated up front.
216 * However, it turns out that 16MB is "enough for anybody", and
217 * real-world applications run into allocation failures from the
218 * overall CMA pool before they make scenes complicated enough to run
222 vc4_allocate_bin_bo(struct drm_device *drm)
224 struct vc4_dev *vc4 = to_vc4_dev(drm);
225 struct vc4_v3d *v3d = vc4->v3d;
226 uint32_t size = 16 * 1024 * 1024;
228 struct list_head list;
230 /* We may need to try allocating more than once to get a BO
231 * that doesn't cross 256MB. Track the ones we've allocated
232 * that failed so far, so that we can free them when we've got
233 * one that succeeded (if we freed them right away, our next
234 * allocation would probably be the same chunk of memory).
236 INIT_LIST_HEAD(&list);
239 struct vc4_bo *bo = vc4_bo_create(drm, size, true,
245 dev_err(&v3d->pdev->dev,
246 "Failed to allocate memory for tile binning: "
247 "%d. You may need to enable CMA or give it "
253 /* Check if this BO won't trigger the addressing bug. */
254 if ((bo->base.paddr & 0xf0000000) ==
255 ((bo->base.paddr + bo->base.base.size - 1) & 0xf0000000)) {
258 /* Set up for allocating 512KB chunks of
259 * binner memory. The biggest allocation we
260 * need to do is for the initial tile alloc +
261 * tile state buffer. We can render to a
262 * maximum of ((2048*2048) / (32*32) = 4096
263 * tiles in a frame (until we do floating
264 * point rendering, at which point it would be
265 * 8192). Tile state is 48b/tile (rounded to
266 * a page), and tile alloc is 32b/tile
267 * (rounded to a page), plus a page of extra,
268 * for a total of 320kb for our worst-case.
269 * We choose 512kb so that it divides evenly
270 * into our 16MB, and the rest of the 512kb
271 * will be used as storage for the overflow
272 * from the initial 32b CL per bin.
274 vc4->bin_alloc_size = 512 * 1024;
275 vc4->bin_alloc_used = 0;
276 vc4->bin_alloc_overflow = 0;
277 WARN_ON_ONCE(sizeof(vc4->bin_alloc_used) * 8 !=
278 bo->base.base.size / vc4->bin_alloc_size);
283 /* Put it on the list to free later, and try again. */
284 list_add(&bo->unref_head, &list);
287 /* Free all the BOs we allocated but didn't choose. */
288 while (!list_empty(&list)) {
289 struct vc4_bo *bo = list_last_entry(&list,
290 struct vc4_bo, unref_head);
292 list_del(&bo->unref_head);
293 drm_gem_object_put_unlocked(&bo->base.base);
300 static int vc4_v3d_runtime_suspend(struct device *dev)
302 struct vc4_v3d *v3d = dev_get_drvdata(dev);
303 struct vc4_dev *vc4 = v3d->vc4;
305 vc4_irq_uninstall(vc4->dev);
307 drm_gem_object_put_unlocked(&vc4->bin_bo->base.base);
310 clk_disable_unprepare(v3d->clk);
315 static int vc4_v3d_runtime_resume(struct device *dev)
317 struct vc4_v3d *v3d = dev_get_drvdata(dev);
318 struct vc4_dev *vc4 = v3d->vc4;
321 ret = vc4_allocate_bin_bo(vc4->dev);
325 ret = clk_prepare_enable(v3d->clk);
329 vc4_v3d_init_hw(vc4->dev);
330 vc4_irq_postinstall(vc4->dev);
336 static int vc4_v3d_bind(struct device *dev, struct device *master, void *data)
338 struct platform_device *pdev = to_platform_device(dev);
339 struct drm_device *drm = dev_get_drvdata(master);
340 struct vc4_dev *vc4 = to_vc4_dev(drm);
341 struct vc4_v3d *v3d = NULL;
344 v3d = devm_kzalloc(&pdev->dev, sizeof(*v3d), GFP_KERNEL);
348 dev_set_drvdata(dev, v3d);
352 v3d->regs = vc4_ioremap_regs(pdev, 0);
353 if (IS_ERR(v3d->regs))
354 return PTR_ERR(v3d->regs);
359 v3d->clk = devm_clk_get(dev, NULL);
360 if (IS_ERR(v3d->clk)) {
361 int ret = PTR_ERR(v3d->clk);
363 if (ret == -ENOENT) {
364 /* bcm2835 didn't have a clock reference in the DT. */
368 if (ret != -EPROBE_DEFER)
369 dev_err(dev, "Failed to get V3D clock: %d\n",
375 if (V3D_READ(V3D_IDENT0) != V3D_EXPECTED_IDENT0) {
376 DRM_ERROR("V3D_IDENT0 read 0x%08x instead of 0x%08x\n",
377 V3D_READ(V3D_IDENT0), V3D_EXPECTED_IDENT0);
381 ret = clk_prepare_enable(v3d->clk);
385 ret = vc4_allocate_bin_bo(drm);
387 clk_disable_unprepare(v3d->clk);
391 /* Reset the binner overflow address/size at setup, to be sure
392 * we don't reuse an old one.
394 V3D_WRITE(V3D_BPOA, 0);
395 V3D_WRITE(V3D_BPOS, 0);
397 vc4_v3d_init_hw(drm);
399 ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
401 DRM_ERROR("Failed to install IRQ handler\n");
405 pm_runtime_set_active(dev);
406 pm_runtime_use_autosuspend(dev);
407 pm_runtime_set_autosuspend_delay(dev, 40); /* a little over 2 frames. */
408 pm_runtime_enable(dev);
413 static void vc4_v3d_unbind(struct device *dev, struct device *master,
416 struct drm_device *drm = dev_get_drvdata(master);
417 struct vc4_dev *vc4 = to_vc4_dev(drm);
419 pm_runtime_disable(dev);
421 drm_irq_uninstall(drm);
423 /* Disable the binner's overflow memory address, so the next
424 * driver probe (if any) doesn't try to reuse our old
427 V3D_WRITE(V3D_BPOA, 0);
428 V3D_WRITE(V3D_BPOS, 0);
433 static const struct dev_pm_ops vc4_v3d_pm_ops = {
434 SET_RUNTIME_PM_OPS(vc4_v3d_runtime_suspend, vc4_v3d_runtime_resume, NULL)
437 static const struct component_ops vc4_v3d_ops = {
438 .bind = vc4_v3d_bind,
439 .unbind = vc4_v3d_unbind,
442 static int vc4_v3d_dev_probe(struct platform_device *pdev)
444 return component_add(&pdev->dev, &vc4_v3d_ops);
447 static int vc4_v3d_dev_remove(struct platform_device *pdev)
449 component_del(&pdev->dev, &vc4_v3d_ops);
453 static const struct of_device_id vc4_v3d_dt_match[] = {
454 { .compatible = "brcm,bcm2835-v3d" },
455 { .compatible = "brcm,cygnus-v3d" },
456 { .compatible = "brcm,vc4-v3d" },
460 struct platform_driver vc4_v3d_driver = {
461 .probe = vc4_v3d_dev_probe,
462 .remove = vc4_v3d_dev_remove,
465 .of_match_table = vc4_v3d_dt_match,
466 .pm = &vc4_v3d_pm_ops,