1 /**************************************************************************
3 * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
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:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
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.
26 **************************************************************************/
28 #include "vmwgfx_drv.h"
30 #include <drm/ttm/ttm_placement.h>
32 bool vmw_fifo_have_3d(struct vmw_private
*dev_priv
)
34 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
35 uint32_t fifo_min
, hwversion
;
36 const struct vmw_fifo_state
*fifo
= &dev_priv
->fifo
;
38 if (!(dev_priv
->capabilities
& SVGA_CAP_3D
))
41 if (dev_priv
->capabilities
& SVGA_CAP_GBOBJECTS
) {
44 if (!dev_priv
->has_mob
)
47 spin_lock(&dev_priv
->cap_lock
);
48 vmw_write(dev_priv
, SVGA_REG_DEV_CAP
, SVGA3D_DEVCAP_3D
);
49 result
= vmw_read(dev_priv
, SVGA_REG_DEV_CAP
);
50 spin_unlock(&dev_priv
->cap_lock
);
55 if (!(dev_priv
->capabilities
& SVGA_CAP_EXTENDED_FIFO
))
58 fifo_min
= ioread32(fifo_mem
+ SVGA_FIFO_MIN
);
59 if (fifo_min
<= SVGA_FIFO_3D_HWVERSION
* sizeof(unsigned int))
62 hwversion
= ioread32(fifo_mem
+
63 ((fifo
->capabilities
&
64 SVGA_FIFO_CAP_3D_HWVERSION_REVISED
) ?
65 SVGA_FIFO_3D_HWVERSION_REVISED
:
66 SVGA_FIFO_3D_HWVERSION
));
71 if (hwversion
< SVGA3D_HWVERSION_WS8_B1
)
74 /* Non-Screen Object path does not support surfaces */
75 if (!dev_priv
->sou_priv
)
81 bool vmw_fifo_have_pitchlock(struct vmw_private
*dev_priv
)
83 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
86 if (!(dev_priv
->capabilities
& SVGA_CAP_EXTENDED_FIFO
))
89 caps
= ioread32(fifo_mem
+ SVGA_FIFO_CAPABILITIES
);
90 if (caps
& SVGA_FIFO_CAP_PITCHLOCK
)
96 int vmw_fifo_init(struct vmw_private
*dev_priv
, struct vmw_fifo_state
*fifo
)
98 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
103 fifo
->static_buffer_size
= VMWGFX_FIFO_STATIC_SIZE
;
104 fifo
->static_buffer
= vmalloc(fifo
->static_buffer_size
);
105 if (unlikely(fifo
->static_buffer
== NULL
))
108 fifo
->dynamic_buffer
= NULL
;
109 fifo
->reserved_size
= 0;
110 fifo
->using_bounce_buffer
= false;
112 mutex_init(&fifo
->fifo_mutex
);
113 init_rwsem(&fifo
->rwsem
);
116 * Allow mapping the first page read-only to user-space.
119 DRM_INFO("width %d\n", vmw_read(dev_priv
, SVGA_REG_WIDTH
));
120 DRM_INFO("height %d\n", vmw_read(dev_priv
, SVGA_REG_HEIGHT
));
121 DRM_INFO("bpp %d\n", vmw_read(dev_priv
, SVGA_REG_BITS_PER_PIXEL
));
123 dev_priv
->enable_state
= vmw_read(dev_priv
, SVGA_REG_ENABLE
);
124 dev_priv
->config_done_state
= vmw_read(dev_priv
, SVGA_REG_CONFIG_DONE
);
125 dev_priv
->traces_state
= vmw_read(dev_priv
, SVGA_REG_TRACES
);
126 vmw_write(dev_priv
, SVGA_REG_ENABLE
, 1);
129 if (dev_priv
->capabilities
& SVGA_CAP_EXTENDED_FIFO
)
130 min
= vmw_read(dev_priv
, SVGA_REG_MEM_REGS
);
136 iowrite32(min
, fifo_mem
+ SVGA_FIFO_MIN
);
137 iowrite32(dev_priv
->mmio_size
, fifo_mem
+ SVGA_FIFO_MAX
);
139 iowrite32(min
, fifo_mem
+ SVGA_FIFO_NEXT_CMD
);
140 iowrite32(min
, fifo_mem
+ SVGA_FIFO_STOP
);
141 iowrite32(0, fifo_mem
+ SVGA_FIFO_BUSY
);
144 vmw_write(dev_priv
, SVGA_REG_CONFIG_DONE
, 1);
146 max
= ioread32(fifo_mem
+ SVGA_FIFO_MAX
);
147 min
= ioread32(fifo_mem
+ SVGA_FIFO_MIN
);
148 fifo
->capabilities
= ioread32(fifo_mem
+ SVGA_FIFO_CAPABILITIES
);
150 DRM_INFO("Fifo max 0x%08x min 0x%08x cap 0x%08x\n",
153 (unsigned int) fifo
->capabilities
);
155 atomic_set(&dev_priv
->marker_seq
, dev_priv
->last_read_seqno
);
156 iowrite32(dev_priv
->last_read_seqno
, fifo_mem
+ SVGA_FIFO_FENCE
);
157 vmw_marker_queue_init(&fifo
->marker_queue
);
158 return vmw_fifo_send_fence(dev_priv
, &dummy
);
161 void vmw_fifo_ping_host(struct vmw_private
*dev_priv
, uint32_t reason
)
163 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
164 static DEFINE_SPINLOCK(ping_lock
);
165 unsigned long irq_flags
;
168 * The ping_lock is needed because we don't have an atomic
169 * test-and-set of the SVGA_FIFO_BUSY register.
171 spin_lock_irqsave(&ping_lock
, irq_flags
);
172 if (unlikely(ioread32(fifo_mem
+ SVGA_FIFO_BUSY
) == 0)) {
173 iowrite32(1, fifo_mem
+ SVGA_FIFO_BUSY
);
174 vmw_write(dev_priv
, SVGA_REG_SYNC
, reason
);
176 spin_unlock_irqrestore(&ping_lock
, irq_flags
);
179 void vmw_fifo_release(struct vmw_private
*dev_priv
, struct vmw_fifo_state
*fifo
)
181 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
183 vmw_write(dev_priv
, SVGA_REG_SYNC
, SVGA_SYNC_GENERIC
);
184 while (vmw_read(dev_priv
, SVGA_REG_BUSY
) != 0)
187 dev_priv
->last_read_seqno
= ioread32(fifo_mem
+ SVGA_FIFO_FENCE
);
189 vmw_write(dev_priv
, SVGA_REG_CONFIG_DONE
,
190 dev_priv
->config_done_state
);
191 vmw_write(dev_priv
, SVGA_REG_ENABLE
,
192 dev_priv
->enable_state
);
193 vmw_write(dev_priv
, SVGA_REG_TRACES
,
194 dev_priv
->traces_state
);
196 vmw_marker_queue_takedown(&fifo
->marker_queue
);
198 if (likely(fifo
->static_buffer
!= NULL
)) {
199 vfree(fifo
->static_buffer
);
200 fifo
->static_buffer
= NULL
;
203 if (likely(fifo
->dynamic_buffer
!= NULL
)) {
204 vfree(fifo
->dynamic_buffer
);
205 fifo
->dynamic_buffer
= NULL
;
209 static bool vmw_fifo_is_full(struct vmw_private
*dev_priv
, uint32_t bytes
)
211 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
212 uint32_t max
= ioread32(fifo_mem
+ SVGA_FIFO_MAX
);
213 uint32_t next_cmd
= ioread32(fifo_mem
+ SVGA_FIFO_NEXT_CMD
);
214 uint32_t min
= ioread32(fifo_mem
+ SVGA_FIFO_MIN
);
215 uint32_t stop
= ioread32(fifo_mem
+ SVGA_FIFO_STOP
);
217 return ((max
- next_cmd
) + (stop
- min
) <= bytes
);
220 static int vmw_fifo_wait_noirq(struct vmw_private
*dev_priv
,
221 uint32_t bytes
, bool interruptible
,
222 unsigned long timeout
)
225 unsigned long end_jiffies
= jiffies
+ timeout
;
228 DRM_INFO("Fifo wait noirq.\n");
231 prepare_to_wait(&dev_priv
->fifo_queue
, &__wait
,
233 TASK_INTERRUPTIBLE
: TASK_UNINTERRUPTIBLE
);
234 if (!vmw_fifo_is_full(dev_priv
, bytes
))
236 if (time_after_eq(jiffies
, end_jiffies
)) {
238 DRM_ERROR("SVGA device lockup.\n");
242 if (interruptible
&& signal_pending(current
)) {
247 finish_wait(&dev_priv
->fifo_queue
, &__wait
);
248 wake_up_all(&dev_priv
->fifo_queue
);
249 DRM_INFO("Fifo noirq exit.\n");
253 static int vmw_fifo_wait(struct vmw_private
*dev_priv
,
254 uint32_t bytes
, bool interruptible
,
255 unsigned long timeout
)
258 unsigned long irq_flags
;
260 if (likely(!vmw_fifo_is_full(dev_priv
, bytes
)))
263 vmw_fifo_ping_host(dev_priv
, SVGA_SYNC_FIFOFULL
);
264 if (!(dev_priv
->capabilities
& SVGA_CAP_IRQMASK
))
265 return vmw_fifo_wait_noirq(dev_priv
, bytes
,
266 interruptible
, timeout
);
268 spin_lock(&dev_priv
->waiter_lock
);
269 if (atomic_add_return(1, &dev_priv
->fifo_queue_waiters
) > 0) {
270 spin_lock_irqsave(&dev_priv
->irq_lock
, irq_flags
);
271 outl(SVGA_IRQFLAG_FIFO_PROGRESS
,
272 dev_priv
->io_start
+ VMWGFX_IRQSTATUS_PORT
);
273 dev_priv
->irq_mask
|= SVGA_IRQFLAG_FIFO_PROGRESS
;
274 vmw_write(dev_priv
, SVGA_REG_IRQMASK
, dev_priv
->irq_mask
);
275 spin_unlock_irqrestore(&dev_priv
->irq_lock
, irq_flags
);
277 spin_unlock(&dev_priv
->waiter_lock
);
280 ret
= wait_event_interruptible_timeout
281 (dev_priv
->fifo_queue
,
282 !vmw_fifo_is_full(dev_priv
, bytes
), timeout
);
284 ret
= wait_event_timeout
285 (dev_priv
->fifo_queue
,
286 !vmw_fifo_is_full(dev_priv
, bytes
), timeout
);
288 if (unlikely(ret
== 0))
290 else if (likely(ret
> 0))
293 spin_lock(&dev_priv
->waiter_lock
);
294 if (atomic_dec_and_test(&dev_priv
->fifo_queue_waiters
)) {
295 spin_lock_irqsave(&dev_priv
->irq_lock
, irq_flags
);
296 dev_priv
->irq_mask
&= ~SVGA_IRQFLAG_FIFO_PROGRESS
;
297 vmw_write(dev_priv
, SVGA_REG_IRQMASK
, dev_priv
->irq_mask
);
298 spin_unlock_irqrestore(&dev_priv
->irq_lock
, irq_flags
);
300 spin_unlock(&dev_priv
->waiter_lock
);
306 * Reserve @bytes number of bytes in the fifo.
308 * This function will return NULL (error) on two conditions:
309 * If it timeouts waiting for fifo space, or if @bytes is larger than the
310 * available fifo space.
313 * Pointer to the fifo, or null on error (possible hardware hang).
315 void *vmw_fifo_reserve(struct vmw_private
*dev_priv
, uint32_t bytes
)
317 struct vmw_fifo_state
*fifo_state
= &dev_priv
->fifo
;
318 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
322 uint32_t reserveable
= fifo_state
->capabilities
& SVGA_FIFO_CAP_RESERVE
;
325 mutex_lock(&fifo_state
->fifo_mutex
);
326 max
= ioread32(fifo_mem
+ SVGA_FIFO_MAX
);
327 min
= ioread32(fifo_mem
+ SVGA_FIFO_MIN
);
328 next_cmd
= ioread32(fifo_mem
+ SVGA_FIFO_NEXT_CMD
);
330 if (unlikely(bytes
>= (max
- min
)))
333 BUG_ON(fifo_state
->reserved_size
!= 0);
334 BUG_ON(fifo_state
->dynamic_buffer
!= NULL
);
336 fifo_state
->reserved_size
= bytes
;
339 uint32_t stop
= ioread32(fifo_mem
+ SVGA_FIFO_STOP
);
340 bool need_bounce
= false;
341 bool reserve_in_place
= false;
343 if (next_cmd
>= stop
) {
344 if (likely((next_cmd
+ bytes
< max
||
345 (next_cmd
+ bytes
== max
&& stop
> min
))))
346 reserve_in_place
= true;
348 else if (vmw_fifo_is_full(dev_priv
, bytes
)) {
349 ret
= vmw_fifo_wait(dev_priv
, bytes
,
351 if (unlikely(ret
!= 0))
358 if (likely((next_cmd
+ bytes
< stop
)))
359 reserve_in_place
= true;
361 ret
= vmw_fifo_wait(dev_priv
, bytes
,
363 if (unlikely(ret
!= 0))
368 if (reserve_in_place
) {
369 if (reserveable
|| bytes
<= sizeof(uint32_t)) {
370 fifo_state
->using_bounce_buffer
= false;
373 iowrite32(bytes
, fifo_mem
+
375 return fifo_mem
+ (next_cmd
>> 2);
382 fifo_state
->using_bounce_buffer
= true;
383 if (bytes
< fifo_state
->static_buffer_size
)
384 return fifo_state
->static_buffer
;
386 fifo_state
->dynamic_buffer
= vmalloc(bytes
);
387 return fifo_state
->dynamic_buffer
;
392 fifo_state
->reserved_size
= 0;
393 mutex_unlock(&fifo_state
->fifo_mutex
);
397 static void vmw_fifo_res_copy(struct vmw_fifo_state
*fifo_state
,
398 __le32 __iomem
*fifo_mem
,
400 uint32_t max
, uint32_t min
, uint32_t bytes
)
402 uint32_t chunk_size
= max
- next_cmd
;
404 uint32_t *buffer
= (fifo_state
->dynamic_buffer
!= NULL
) ?
405 fifo_state
->dynamic_buffer
: fifo_state
->static_buffer
;
407 if (bytes
< chunk_size
)
410 iowrite32(bytes
, fifo_mem
+ SVGA_FIFO_RESERVED
);
412 memcpy_toio(fifo_mem
+ (next_cmd
>> 2), buffer
, chunk_size
);
413 rest
= bytes
- chunk_size
;
415 memcpy_toio(fifo_mem
+ (min
>> 2), buffer
+ (chunk_size
>> 2),
419 static void vmw_fifo_slow_copy(struct vmw_fifo_state
*fifo_state
,
420 __le32 __iomem
*fifo_mem
,
422 uint32_t max
, uint32_t min
, uint32_t bytes
)
424 uint32_t *buffer
= (fifo_state
->dynamic_buffer
!= NULL
) ?
425 fifo_state
->dynamic_buffer
: fifo_state
->static_buffer
;
428 iowrite32(*buffer
++, fifo_mem
+ (next_cmd
>> 2));
429 next_cmd
+= sizeof(uint32_t);
430 if (unlikely(next_cmd
== max
))
433 iowrite32(next_cmd
, fifo_mem
+ SVGA_FIFO_NEXT_CMD
);
435 bytes
-= sizeof(uint32_t);
439 void vmw_fifo_commit(struct vmw_private
*dev_priv
, uint32_t bytes
)
441 struct vmw_fifo_state
*fifo_state
= &dev_priv
->fifo
;
442 __le32 __iomem
*fifo_mem
= dev_priv
->mmio_virt
;
443 uint32_t next_cmd
= ioread32(fifo_mem
+ SVGA_FIFO_NEXT_CMD
);
444 uint32_t max
= ioread32(fifo_mem
+ SVGA_FIFO_MAX
);
445 uint32_t min
= ioread32(fifo_mem
+ SVGA_FIFO_MIN
);
446 bool reserveable
= fifo_state
->capabilities
& SVGA_FIFO_CAP_RESERVE
;
448 BUG_ON((bytes
& 3) != 0);
449 BUG_ON(bytes
> fifo_state
->reserved_size
);
451 fifo_state
->reserved_size
= 0;
453 if (fifo_state
->using_bounce_buffer
) {
455 vmw_fifo_res_copy(fifo_state
, fifo_mem
,
456 next_cmd
, max
, min
, bytes
);
458 vmw_fifo_slow_copy(fifo_state
, fifo_mem
,
459 next_cmd
, max
, min
, bytes
);
461 if (fifo_state
->dynamic_buffer
) {
462 vfree(fifo_state
->dynamic_buffer
);
463 fifo_state
->dynamic_buffer
= NULL
;
468 down_write(&fifo_state
->rwsem
);
469 if (fifo_state
->using_bounce_buffer
|| reserveable
) {
472 next_cmd
-= max
- min
;
474 iowrite32(next_cmd
, fifo_mem
+ SVGA_FIFO_NEXT_CMD
);
478 iowrite32(0, fifo_mem
+ SVGA_FIFO_RESERVED
);
480 up_write(&fifo_state
->rwsem
);
481 vmw_fifo_ping_host(dev_priv
, SVGA_SYNC_GENERIC
);
482 mutex_unlock(&fifo_state
->fifo_mutex
);
485 int vmw_fifo_send_fence(struct vmw_private
*dev_priv
, uint32_t *seqno
)
487 struct vmw_fifo_state
*fifo_state
= &dev_priv
->fifo
;
488 struct svga_fifo_cmd_fence
*cmd_fence
;
491 uint32_t bytes
= sizeof(__le32
) + sizeof(*cmd_fence
);
493 fm
= vmw_fifo_reserve(dev_priv
, bytes
);
494 if (unlikely(fm
== NULL
)) {
495 *seqno
= atomic_read(&dev_priv
->marker_seq
);
497 (void)vmw_fallback_wait(dev_priv
, false, true, *seqno
,
503 *seqno
= atomic_add_return(1, &dev_priv
->marker_seq
);
504 } while (*seqno
== 0);
506 if (!(fifo_state
->capabilities
& SVGA_FIFO_CAP_FENCE
)) {
509 * Don't request hardware to send a fence. The
510 * waiting code in vmwgfx_irq.c will emulate this.
513 vmw_fifo_commit(dev_priv
, 0);
517 *(__le32
*) fm
= cpu_to_le32(SVGA_CMD_FENCE
);
518 cmd_fence
= (struct svga_fifo_cmd_fence
*)
519 ((unsigned long)fm
+ sizeof(__le32
));
521 iowrite32(*seqno
, &cmd_fence
->fence
);
522 vmw_fifo_commit(dev_priv
, bytes
);
523 (void) vmw_marker_push(&fifo_state
->marker_queue
, *seqno
);
524 vmw_update_seqno(dev_priv
, fifo_state
);
531 * vmw_fifo_emit_dummy_legacy_query - emits a dummy query to the fifo using
532 * legacy query commands.
534 * @dev_priv: The device private structure.
535 * @cid: The hardware context id used for the query.
537 * See the vmw_fifo_emit_dummy_query documentation.
539 static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private
*dev_priv
,
543 * A query wait without a preceding query end will
544 * actually finish all queries for this cid
545 * without writing to the query result structure.
548 struct ttm_buffer_object
*bo
= dev_priv
->dummy_query_bo
;
550 SVGA3dCmdHeader header
;
551 SVGA3dCmdWaitForQuery body
;
554 cmd
= vmw_fifo_reserve(dev_priv
, sizeof(*cmd
));
556 if (unlikely(cmd
== NULL
)) {
557 DRM_ERROR("Out of fifo space for dummy query.\n");
561 cmd
->header
.id
= SVGA_3D_CMD_WAIT_FOR_QUERY
;
562 cmd
->header
.size
= sizeof(cmd
->body
);
564 cmd
->body
.type
= SVGA3D_QUERYTYPE_OCCLUSION
;
566 if (bo
->mem
.mem_type
== TTM_PL_VRAM
) {
567 cmd
->body
.guestResult
.gmrId
= SVGA_GMR_FRAMEBUFFER
;
568 cmd
->body
.guestResult
.offset
= bo
->offset
;
570 cmd
->body
.guestResult
.gmrId
= bo
->mem
.start
;
571 cmd
->body
.guestResult
.offset
= 0;
574 vmw_fifo_commit(dev_priv
, sizeof(*cmd
));
580 * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
581 * guest-backed resource query commands.
583 * @dev_priv: The device private structure.
584 * @cid: The hardware context id used for the query.
586 * See the vmw_fifo_emit_dummy_query documentation.
588 static int vmw_fifo_emit_dummy_gb_query(struct vmw_private
*dev_priv
,
592 * A query wait without a preceding query end will
593 * actually finish all queries for this cid
594 * without writing to the query result structure.
597 struct ttm_buffer_object
*bo
= dev_priv
->dummy_query_bo
;
599 SVGA3dCmdHeader header
;
600 SVGA3dCmdWaitForGBQuery body
;
603 cmd
= vmw_fifo_reserve(dev_priv
, sizeof(*cmd
));
605 if (unlikely(cmd
== NULL
)) {
606 DRM_ERROR("Out of fifo space for dummy query.\n");
610 cmd
->header
.id
= SVGA_3D_CMD_WAIT_FOR_GB_QUERY
;
611 cmd
->header
.size
= sizeof(cmd
->body
);
613 cmd
->body
.type
= SVGA3D_QUERYTYPE_OCCLUSION
;
614 BUG_ON(bo
->mem
.mem_type
!= VMW_PL_MOB
);
615 cmd
->body
.mobid
= bo
->mem
.start
;
616 cmd
->body
.offset
= 0;
618 vmw_fifo_commit(dev_priv
, sizeof(*cmd
));
625 * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
626 * appropriate resource query commands.
628 * @dev_priv: The device private structure.
629 * @cid: The hardware context id used for the query.
631 * This function is used to emit a dummy occlusion query with
632 * no primitives rendered between query begin and query end.
633 * It's used to provide a query barrier, in order to know that when
634 * this query is finished, all preceding queries are also finished.
636 * A Query results structure should have been initialized at the start
637 * of the dev_priv->dummy_query_bo buffer object. And that buffer object
638 * must also be either reserved or pinned when this function is called.
640 * Returns -ENOMEM on failure to reserve fifo space.
642 int vmw_fifo_emit_dummy_query(struct vmw_private
*dev_priv
,
645 if (dev_priv
->has_mob
)
646 return vmw_fifo_emit_dummy_gb_query(dev_priv
, cid
);
648 return vmw_fifo_emit_dummy_legacy_query(dev_priv
, cid
);