1 /*
2 * Copyright © 2014 Broadcom
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 /**
25 * DOC: Interrupt management for the V3D engine
26 *
27 * We have an interrupt status register (V3D_INTCTL) which reports
28 * interrupts, and where writing 1 bits clears those interrupts.
29 * There are also a pair of interrupt registers
30 * (V3D_INTENA/V3D_INTDIS) where writing a 1 to their bits enables or
31 * disables that specific interrupt, and 0s written are ignored
32 * (reading either one returns the set of enabled interrupts).
33 *
34 * When we take a binning flush done interrupt, we need to submit the
35 * next frame for binning and move the finished frame to the render
36 * thread.
37 *
38 * When we take a render frame interrupt, we need to wake the
39 * processes waiting for some frame to be done, and get the next frame
40 * submitted ASAP (so the hardware doesn't sit idle when there's work
41 * to do).
42 *
43 * When we take the binner out of memory interrupt, we need to
44 * allocate some new memory and pass it to the binner so that the
45 * current job can make progress.
46 */
47
48 #include <linux/platform_device.h>
49
50 #include <drm/drm_drv.h>
51 #include <drm/drm_print.h>
52
53 #include "vc4_drv.h"
54 #include "vc4_regs.h"
55 #include "vc4_trace.h"
56
57 #define V3D_DRIVER_IRQS (V3D_INT_OUTOMEM | \
58 V3D_INT_FLDONE | \
59 V3D_INT_FRDONE)
60
61 static void
vc4_overflow_mem_work(struct work_struct * work)62 vc4_overflow_mem_work(struct work_struct *work)
63 {
64 struct vc4_dev *vc4 =
65 container_of(work, struct vc4_dev, overflow_mem_work);
66 struct vc4_bo *bo;
67 int bin_bo_slot;
68 struct vc4_exec_info *exec;
69 unsigned long irqflags;
70
71 mutex_lock(&vc4->bin_bo_lock);
72
73 if (!vc4->bin_bo)
74 goto complete;
75
76 bo = vc4->bin_bo;
77
78 bin_bo_slot = vc4_v3d_get_bin_slot(vc4);
79 if (bin_bo_slot < 0) {
80 drm_err(&vc4->base, "Couldn't allocate binner overflow mem\n");
81 goto complete;
82 }
83
84 spin_lock_irqsave(&vc4->job_lock, irqflags);
85
86 if (vc4->bin_alloc_overflow) {
87 /* If we had overflow memory allocated previously,
88 * then that chunk will free when the current bin job
89 * is done. If we don't have a bin job running, then
90 * the chunk will be done whenever the list of render
91 * jobs has drained.
92 */
93 exec = vc4_first_bin_job(vc4);
94 if (!exec)
95 exec = vc4_last_render_job(vc4);
96 if (exec) {
97 exec->bin_slots |= vc4->bin_alloc_overflow;
98 } else {
99 /* There's nothing queued in the hardware, so
100 * the old slot is free immediately.
101 */
102 vc4->bin_alloc_used &= ~vc4->bin_alloc_overflow;
103 }
104 }
105 vc4->bin_alloc_overflow = BIT(bin_bo_slot);
106
107 V3D_WRITE(V3D_BPOA, bo->base.dma_addr + bin_bo_slot * vc4->bin_alloc_size);
108 V3D_WRITE(V3D_BPOS, bo->base.base.size);
109 V3D_WRITE(V3D_INTCTL, V3D_INT_OUTOMEM);
110 V3D_WRITE(V3D_INTENA, V3D_INT_OUTOMEM);
111 spin_unlock_irqrestore(&vc4->job_lock, irqflags);
112
113 complete:
114 mutex_unlock(&vc4->bin_bo_lock);
115 }
116
117 static void
vc4_irq_finish_bin_job(struct drm_device * dev)118 vc4_irq_finish_bin_job(struct drm_device *dev)
119 {
120 struct vc4_dev *vc4 = to_vc4_dev(dev);
121 struct vc4_exec_info *next, *exec = vc4_first_bin_job(vc4);
122
123 if (!exec)
124 return;
125
126 trace_vc4_bcl_end_irq(dev, exec->seqno);
127
128 vc4_move_job_to_render(dev, exec);
129 next = vc4_first_bin_job(vc4);
130
131 /* Only submit the next job in the bin list if it matches the perfmon
132 * attached to the one that just finished (or if both jobs don't have
133 * perfmon attached to them).
134 */
135 if (next && next->perfmon == exec->perfmon)
136 vc4_submit_next_bin_job(dev);
137 }
138
139 static void
vc4_cancel_bin_job(struct drm_device * dev)140 vc4_cancel_bin_job(struct drm_device *dev)
141 {
142 struct vc4_dev *vc4 = to_vc4_dev(dev);
143 struct vc4_exec_info *exec = vc4_first_bin_job(vc4);
144
145 if (!exec)
146 return;
147
148 /* Stop the perfmon so that the next bin job can be started. */
149 if (exec->perfmon)
150 vc4_perfmon_stop(vc4, exec->perfmon, false);
151
152 list_move_tail(&exec->head, &vc4->bin_job_list);
153 vc4_submit_next_bin_job(dev);
154 }
155
156 static void
vc4_irq_finish_render_job(struct drm_device * dev)157 vc4_irq_finish_render_job(struct drm_device *dev)
158 {
159 struct vc4_dev *vc4 = to_vc4_dev(dev);
160 struct vc4_exec_info *exec = vc4_first_render_job(vc4);
161 struct vc4_exec_info *nextbin, *nextrender;
162
163 if (!exec)
164 return;
165
166 trace_vc4_rcl_end_irq(dev, exec->seqno);
167
168 vc4->finished_seqno++;
169 list_move_tail(&exec->head, &vc4->job_done_list);
170
171 nextbin = vc4_first_bin_job(vc4);
172 nextrender = vc4_first_render_job(vc4);
173
174 /* Only stop the perfmon if following jobs in the queue don't expect it
175 * to be enabled.
176 */
177 if (exec->perfmon && !nextrender &&
178 (!nextbin || nextbin->perfmon != exec->perfmon))
179 vc4_perfmon_stop(vc4, exec->perfmon, true);
180
181 /* If there's a render job waiting, start it. If this is not the case
182 * we may have to unblock the binner if it's been stalled because of
183 * perfmon (this can be checked by comparing the perfmon attached to
184 * the finished renderjob to the one attached to the next bin job: if
185 * they don't match, this means the binner is stalled and should be
186 * restarted).
187 */
188 if (nextrender)
189 vc4_submit_next_render_job(dev);
190 else if (nextbin && nextbin->perfmon != exec->perfmon)
191 vc4_submit_next_bin_job(dev);
192
193 if (exec->fence) {
194 dma_fence_signal_locked(exec->fence);
195 dma_fence_put(exec->fence);
196 exec->fence = NULL;
197 }
198
199 wake_up_all(&vc4->job_wait_queue);
200 schedule_work(&vc4->job_done_work);
201 }
202
203 static irqreturn_t
vc4_irq(int irq,void * arg)204 vc4_irq(int irq, void *arg)
205 {
206 struct drm_device *dev = arg;
207 struct vc4_dev *vc4 = to_vc4_dev(dev);
208 uint32_t intctl;
209 irqreturn_t status = IRQ_NONE;
210
211 barrier();
212 intctl = V3D_READ(V3D_INTCTL);
213
214 /* Acknowledge the interrupts we're handling here. The binner
215 * last flush / render frame done interrupt will be cleared,
216 * while OUTOMEM will stay high until the underlying cause is
217 * cleared.
218 */
219 V3D_WRITE(V3D_INTCTL, intctl);
220
221 if (intctl & V3D_INT_OUTOMEM) {
222 /* Disable OUTOMEM until the work is done. */
223 V3D_WRITE(V3D_INTDIS, V3D_INT_OUTOMEM);
224 schedule_work(&vc4->overflow_mem_work);
225 status = IRQ_HANDLED;
226 }
227
228 if (intctl & V3D_INT_FLDONE) {
229 spin_lock(&vc4->job_lock);
230 vc4_irq_finish_bin_job(dev);
231 spin_unlock(&vc4->job_lock);
232 status = IRQ_HANDLED;
233 }
234
235 if (intctl & V3D_INT_FRDONE) {
236 spin_lock(&vc4->job_lock);
237 vc4_irq_finish_render_job(dev);
238 spin_unlock(&vc4->job_lock);
239 status = IRQ_HANDLED;
240 }
241
242 return status;
243 }
244
245 static void
vc4_irq_prepare(struct drm_device * dev)246 vc4_irq_prepare(struct drm_device *dev)
247 {
248 struct vc4_dev *vc4 = to_vc4_dev(dev);
249
250 if (!vc4->v3d)
251 return;
252
253 init_waitqueue_head(&vc4->job_wait_queue);
254 INIT_WORK(&vc4->overflow_mem_work, vc4_overflow_mem_work);
255
256 /* Clear any pending interrupts someone might have left around
257 * for us.
258 */
259 V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
260 }
261
262 void
vc4_irq_enable(struct drm_device * dev)263 vc4_irq_enable(struct drm_device *dev)
264 {
265 struct vc4_dev *vc4 = to_vc4_dev(dev);
266
267 if (WARN_ON_ONCE(vc4->gen > VC4_GEN_4))
268 return;
269
270 if (!vc4->v3d)
271 return;
272
273 /* Enable the render done interrupts. The out-of-memory interrupt is
274 * enabled as soon as we have a binner BO allocated.
275 */
276 V3D_WRITE(V3D_INTENA, V3D_INT_FLDONE | V3D_INT_FRDONE);
277 }
278
279 void
vc4_irq_disable(struct drm_device * dev)280 vc4_irq_disable(struct drm_device *dev)
281 {
282 struct vc4_dev *vc4 = to_vc4_dev(dev);
283
284 if (WARN_ON_ONCE(vc4->gen > VC4_GEN_4))
285 return;
286
287 if (!vc4->v3d)
288 return;
289
290 /* Disable sending interrupts for our driver's IRQs. */
291 V3D_WRITE(V3D_INTDIS, V3D_DRIVER_IRQS);
292
293 /* Clear any pending interrupts we might have left. */
294 V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
295
296 /* Finish any interrupt handler still in flight. */
297 synchronize_irq(vc4->irq);
298
299 cancel_work_sync(&vc4->overflow_mem_work);
300 }
301
vc4_irq_install(struct drm_device * dev,int irq)302 int vc4_irq_install(struct drm_device *dev, int irq)
303 {
304 struct vc4_dev *vc4 = to_vc4_dev(dev);
305 int ret;
306
307 if (WARN_ON_ONCE(vc4->gen > VC4_GEN_4))
308 return -ENODEV;
309
310 if (irq == IRQ_NOTCONNECTED)
311 return -ENOTCONN;
312
313 vc4_irq_prepare(dev);
314
315 ret = request_irq(irq, vc4_irq, 0, dev->driver->name, dev);
316 if (ret)
317 return ret;
318
319 vc4_irq_enable(dev);
320
321 return 0;
322 }
323
vc4_irq_uninstall(struct drm_device * dev)324 void vc4_irq_uninstall(struct drm_device *dev)
325 {
326 struct vc4_dev *vc4 = to_vc4_dev(dev);
327
328 if (WARN_ON_ONCE(vc4->gen > VC4_GEN_4))
329 return;
330
331 vc4_irq_disable(dev);
332 free_irq(vc4->irq, dev);
333 }
334
335 /** Reinitializes interrupt registers when a GPU reset is performed. */
vc4_irq_reset(struct drm_device * dev)336 void vc4_irq_reset(struct drm_device *dev)
337 {
338 struct vc4_dev *vc4 = to_vc4_dev(dev);
339 unsigned long irqflags;
340
341 if (WARN_ON_ONCE(vc4->gen > VC4_GEN_4))
342 return;
343
344 /* Acknowledge any stale IRQs. */
345 V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
346
347 /*
348 * Turn all our interrupts on. Binner out of memory is the
349 * only one we expect to trigger at this point, since we've
350 * just come from poweron and haven't supplied any overflow
351 * memory yet.
352 */
353 V3D_WRITE(V3D_INTENA, V3D_DRIVER_IRQS);
354
355 spin_lock_irqsave(&vc4->job_lock, irqflags);
356 vc4_cancel_bin_job(dev);
357 vc4_irq_finish_render_job(dev);
358 spin_unlock_irqrestore(&vc4->job_lock, irqflags);
359 }
360