1 // SPDX-License-Identifier: GPL-2.0 OR MIT
2 /**************************************************************************
3 *
4 * Copyright 2009-2015 VMware, Inc., Palo Alto, CA., USA
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 * USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include <linux/pci.h>
29 #include <linux/sched/signal.h>
30
31 #include "vmwgfx_drv.h"
32
33 #define VMW_FENCE_WRAP (1 << 24)
34
vmw_irqflag_fence_goal(struct vmw_private * vmw)35 static u32 vmw_irqflag_fence_goal(struct vmw_private *vmw)
36 {
37 if ((vmw->capabilities2 & SVGA_CAP2_EXTRA_REGS) != 0)
38 return SVGA_IRQFLAG_REG_FENCE_GOAL;
39 else
40 return SVGA_IRQFLAG_FENCE_GOAL;
41 }
42
43 /**
44 * vmw_thread_fn - Deferred (process context) irq handler
45 *
46 * @irq: irq number
47 * @arg: Closure argument. Pointer to a struct drm_device cast to void *
48 *
49 * This function implements the deferred part of irq processing.
50 * The function is guaranteed to run at least once after the
51 * vmw_irq_handler has returned with IRQ_WAKE_THREAD.
52 *
53 */
vmw_thread_fn(int irq,void * arg)54 static irqreturn_t vmw_thread_fn(int irq, void *arg)
55 {
56 struct drm_device *dev = (struct drm_device *)arg;
57 struct vmw_private *dev_priv = vmw_priv(dev);
58 irqreturn_t ret = IRQ_NONE;
59
60 if (test_and_clear_bit(VMW_IRQTHREAD_FENCE,
61 dev_priv->irqthread_pending)) {
62 vmw_fences_update(dev_priv->fman);
63 wake_up_all(&dev_priv->fence_queue);
64 ret = IRQ_HANDLED;
65 }
66
67 if (test_and_clear_bit(VMW_IRQTHREAD_CMDBUF,
68 dev_priv->irqthread_pending)) {
69 vmw_cmdbuf_irqthread(dev_priv->cman);
70 ret = IRQ_HANDLED;
71 }
72
73 return ret;
74 }
75
76 /**
77 * vmw_irq_handler: irq handler
78 *
79 * @irq: irq number
80 * @arg: Closure argument. Pointer to a struct drm_device cast to void *
81 *
82 * This function implements the quick part of irq processing.
83 * The function performs fast actions like clearing the device interrupt
84 * flags and also reasonably quick actions like waking processes waiting for
85 * FIFO space. Other IRQ actions are deferred to the IRQ thread.
86 */
vmw_irq_handler(int irq,void * arg)87 static irqreturn_t vmw_irq_handler(int irq, void *arg)
88 {
89 struct drm_device *dev = (struct drm_device *)arg;
90 struct vmw_private *dev_priv = vmw_priv(dev);
91 uint32_t status, masked_status;
92 irqreturn_t ret = IRQ_HANDLED;
93
94 status = vmw_irq_status_read(dev_priv);
95 masked_status = status & READ_ONCE(dev_priv->irq_mask);
96
97 if (likely(status))
98 vmw_irq_status_write(dev_priv, status);
99
100 if (!status)
101 return IRQ_NONE;
102
103 if (masked_status & SVGA_IRQFLAG_FIFO_PROGRESS)
104 wake_up_all(&dev_priv->fifo_queue);
105
106 if ((masked_status & (SVGA_IRQFLAG_ANY_FENCE |
107 vmw_irqflag_fence_goal(dev_priv))) &&
108 !test_and_set_bit(VMW_IRQTHREAD_FENCE, dev_priv->irqthread_pending))
109 ret = IRQ_WAKE_THREAD;
110
111 if ((masked_status & (SVGA_IRQFLAG_COMMAND_BUFFER |
112 SVGA_IRQFLAG_ERROR)) &&
113 !test_and_set_bit(VMW_IRQTHREAD_CMDBUF,
114 dev_priv->irqthread_pending))
115 ret = IRQ_WAKE_THREAD;
116
117 return ret;
118 }
119
vmw_fifo_idle(struct vmw_private * dev_priv,uint32_t seqno)120 static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t seqno)
121 {
122
123 return (vmw_read(dev_priv, SVGA_REG_BUSY) == 0);
124 }
125
vmw_seqno_passed(struct vmw_private * dev_priv,uint32_t seqno)126 bool vmw_seqno_passed(struct vmw_private *dev_priv,
127 uint32_t seqno)
128 {
129 bool ret;
130 u32 last_read_seqno = atomic_read_acquire(&dev_priv->last_read_seqno);
131
132 if (last_read_seqno - seqno < VMW_FENCE_WRAP)
133 return true;
134
135 last_read_seqno = vmw_fences_update(dev_priv->fman);
136 if (last_read_seqno - seqno < VMW_FENCE_WRAP)
137 return true;
138
139 if (!vmw_has_fences(dev_priv) && vmw_fifo_idle(dev_priv, seqno))
140 return true;
141
142 /**
143 * Then check if the seqno is higher than what we've actually
144 * emitted. Then the fence is stale and signaled.
145 */
146
147 ret = ((atomic_read(&dev_priv->marker_seq) - seqno)
148 > VMW_FENCE_WRAP);
149
150 return ret;
151 }
152
vmw_fallback_wait(struct vmw_private * dev_priv,bool lazy,bool fifo_idle,uint32_t seqno,bool interruptible,unsigned long timeout)153 int vmw_fallback_wait(struct vmw_private *dev_priv,
154 bool lazy,
155 bool fifo_idle,
156 uint32_t seqno,
157 bool interruptible,
158 unsigned long timeout)
159 {
160 struct vmw_fifo_state *fifo_state = dev_priv->fifo;
161 bool fifo_down = false;
162
163 uint32_t count = 0;
164 uint32_t signal_seq;
165 int ret;
166 unsigned long end_jiffies = jiffies + timeout;
167 bool (*wait_condition)(struct vmw_private *, uint32_t);
168 DEFINE_WAIT(__wait);
169
170 wait_condition = (fifo_idle) ? &vmw_fifo_idle :
171 &vmw_seqno_passed;
172
173 /**
174 * Block command submission while waiting for idle.
175 */
176
177 if (fifo_idle) {
178 if (dev_priv->cman) {
179 ret = vmw_cmdbuf_idle(dev_priv->cman, interruptible,
180 10*HZ);
181 if (ret)
182 goto out_err;
183 } else if (fifo_state) {
184 down_read(&fifo_state->rwsem);
185 fifo_down = true;
186 }
187 }
188
189 signal_seq = atomic_read(&dev_priv->marker_seq);
190 ret = 0;
191
192 for (;;) {
193 prepare_to_wait(&dev_priv->fence_queue, &__wait,
194 (interruptible) ?
195 TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE);
196 if (wait_condition(dev_priv, seqno))
197 break;
198 if (time_after_eq(jiffies, end_jiffies)) {
199 DRM_ERROR("SVGA device lockup.\n");
200 break;
201 }
202 if (lazy)
203 schedule_timeout(1);
204 else if ((++count & 0x0F) == 0) {
205 /**
206 * FIXME: Use schedule_hr_timeout here for
207 * newer kernels and lower CPU utilization.
208 */
209
210 __set_current_state(TASK_RUNNING);
211 schedule();
212 __set_current_state((interruptible) ?
213 TASK_INTERRUPTIBLE :
214 TASK_UNINTERRUPTIBLE);
215 }
216 if (interruptible && signal_pending(current)) {
217 ret = -ERESTARTSYS;
218 break;
219 }
220 }
221 finish_wait(&dev_priv->fence_queue, &__wait);
222 if (ret == 0 && fifo_idle && fifo_state)
223 vmw_fence_write(dev_priv, signal_seq);
224
225 wake_up_all(&dev_priv->fence_queue);
226 out_err:
227 if (fifo_down)
228 up_read(&fifo_state->rwsem);
229
230 return ret;
231 }
232
vmw_generic_waiter_add(struct vmw_private * dev_priv,u32 flag,int * waiter_count)233 bool vmw_generic_waiter_add(struct vmw_private *dev_priv,
234 u32 flag, int *waiter_count)
235 {
236 bool hw_programmed = false;
237
238 spin_lock(&dev_priv->waiter_lock);
239 if ((*waiter_count)++ == 0) {
240 vmw_irq_status_write(dev_priv, flag);
241 dev_priv->irq_mask |= flag;
242 vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
243 hw_programmed = true;
244 }
245 spin_unlock(&dev_priv->waiter_lock);
246 return hw_programmed;
247 }
248
vmw_generic_waiter_remove(struct vmw_private * dev_priv,u32 flag,int * waiter_count)249 bool vmw_generic_waiter_remove(struct vmw_private *dev_priv,
250 u32 flag, int *waiter_count)
251 {
252 bool hw_programmed = false;
253
254 spin_lock(&dev_priv->waiter_lock);
255 if (--(*waiter_count) == 0) {
256 dev_priv->irq_mask &= ~flag;
257 vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
258 hw_programmed = true;
259 }
260 spin_unlock(&dev_priv->waiter_lock);
261 return hw_programmed;
262 }
263
vmw_seqno_waiter_add(struct vmw_private * dev_priv)264 bool vmw_seqno_waiter_add(struct vmw_private *dev_priv)
265 {
266 return vmw_generic_waiter_add(dev_priv, SVGA_IRQFLAG_ANY_FENCE,
267 &dev_priv->fence_queue_waiters);
268 }
269
vmw_seqno_waiter_remove(struct vmw_private * dev_priv)270 bool vmw_seqno_waiter_remove(struct vmw_private *dev_priv)
271 {
272 return vmw_generic_waiter_remove(dev_priv, SVGA_IRQFLAG_ANY_FENCE,
273 &dev_priv->fence_queue_waiters);
274 }
275
vmw_goal_waiter_add(struct vmw_private * dev_priv)276 bool vmw_goal_waiter_add(struct vmw_private *dev_priv)
277 {
278 return vmw_generic_waiter_add(dev_priv, vmw_irqflag_fence_goal(dev_priv),
279 &dev_priv->goal_queue_waiters);
280 }
281
vmw_goal_waiter_remove(struct vmw_private * dev_priv)282 bool vmw_goal_waiter_remove(struct vmw_private *dev_priv)
283 {
284 return vmw_generic_waiter_remove(dev_priv, vmw_irqflag_fence_goal(dev_priv),
285 &dev_priv->goal_queue_waiters);
286 }
287
vmw_irq_preinstall(struct drm_device * dev)288 static void vmw_irq_preinstall(struct drm_device *dev)
289 {
290 struct vmw_private *dev_priv = vmw_priv(dev);
291 uint32_t status;
292
293 status = vmw_irq_status_read(dev_priv);
294 vmw_irq_status_write(dev_priv, status);
295 }
296
vmw_irq_uninstall(struct drm_device * dev)297 void vmw_irq_uninstall(struct drm_device *dev)
298 {
299 struct vmw_private *dev_priv = vmw_priv(dev);
300 struct pci_dev *pdev = to_pci_dev(dev->dev);
301 uint32_t status;
302 u32 i;
303
304 if (!(dev_priv->capabilities & SVGA_CAP_IRQMASK))
305 return;
306
307 vmw_write(dev_priv, SVGA_REG_IRQMASK, 0);
308
309 status = vmw_irq_status_read(dev_priv);
310 vmw_irq_status_write(dev_priv, status);
311
312 for (i = 0; i < dev_priv->num_irq_vectors; ++i)
313 free_irq(dev_priv->irqs[i], dev);
314
315 pci_free_irq_vectors(pdev);
316 dev_priv->num_irq_vectors = 0;
317 }
318
319 /**
320 * vmw_irq_install - Install the irq handlers
321 *
322 * @dev_priv: Pointer to the vmw_private device.
323 * Return: Zero if successful. Negative number otherwise.
324 */
vmw_irq_install(struct vmw_private * dev_priv)325 int vmw_irq_install(struct vmw_private *dev_priv)
326 {
327 struct pci_dev *pdev = to_pci_dev(dev_priv->drm.dev);
328 struct drm_device *dev = &dev_priv->drm;
329 int ret;
330 int nvec;
331 int i = 0;
332
333 BUILD_BUG_ON((SVGA_IRQFLAG_MAX >> VMWGFX_MAX_NUM_IRQS) != 1);
334 BUG_ON(VMWGFX_MAX_NUM_IRQS != get_count_order(SVGA_IRQFLAG_MAX));
335
336 nvec = pci_alloc_irq_vectors(pdev, 1, VMWGFX_MAX_NUM_IRQS,
337 PCI_IRQ_ALL_TYPES);
338
339 if (nvec <= 0) {
340 drm_err(&dev_priv->drm,
341 "IRQ's are unavailable, nvec: %d\n", nvec);
342 ret = nvec;
343 goto done;
344 }
345
346 vmw_irq_preinstall(dev);
347
348 for (i = 0; i < nvec; ++i) {
349 ret = pci_irq_vector(pdev, i);
350 if (ret < 0) {
351 drm_err(&dev_priv->drm,
352 "failed getting irq vector: %d\n", ret);
353 goto done;
354 }
355 dev_priv->irqs[i] = ret;
356
357 ret = request_threaded_irq(dev_priv->irqs[i], vmw_irq_handler, vmw_thread_fn,
358 IRQF_SHARED, VMWGFX_DRIVER_NAME, dev);
359 if (ret != 0) {
360 drm_err(&dev_priv->drm,
361 "Failed installing irq(%d): %d\n",
362 dev_priv->irqs[i], ret);
363 goto done;
364 }
365 }
366
367 done:
368 dev_priv->num_irq_vectors = i;
369 return ret;
370 }
371