* [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
@ 2017-12-15 16:28 Samuel Li
[not found] ` <1513355282-20248-1-git-send-email-Samuel.Li-5C7GfCeVMHo@public.gmane.org>
` (2 more replies)
0 siblings, 3 replies; 7+ messages in thread
From: Samuel Li @ 2017-12-15 16:28 UTC (permalink / raw)
To: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW,
amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Christian König
From: Christian König <christian.koenig@amd.com>
Allow drivers to implement their own begin_cpu_access callback.
Change-Id: I97709b42b9351a04ee7e01106107a87bc56ea258
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/drm_prime.c | 13 +++++++++++++
include/drm/drm_drv.h | 2 ++
2 files changed, 15 insertions(+)
diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
index 8de93a2..b4b0e64 100644
--- a/drivers/gpu/drm/drm_prime.c
+++ b/drivers/gpu/drm/drm_prime.c
@@ -346,6 +346,18 @@ void drm_gem_dmabuf_release(struct dma_buf *dma_buf)
}
EXPORT_SYMBOL(drm_gem_dmabuf_release);
+static int drm_gem_dmabuf_begin_cpu_access(struct dma_buf *dma_buf,
+ enum dma_data_direction direction)
+{
+ struct drm_gem_object *obj = dma_buf->priv;
+ struct drm_device *dev = obj->dev;
+
+ if (!dev->driver->gem_prime_begin_cpu_access)
+ return 0;
+
+ return dev->driver->gem_prime_begin_cpu_access(obj, direction);
+}
+
static void *drm_gem_dmabuf_vmap(struct dma_buf *dma_buf)
{
struct drm_gem_object *obj = dma_buf->priv;
@@ -403,6 +415,7 @@ static const struct dma_buf_ops drm_gem_prime_dmabuf_ops = {
.map_dma_buf = drm_gem_map_dma_buf,
.unmap_dma_buf = drm_gem_unmap_dma_buf,
.release = drm_gem_dmabuf_release,
+ .begin_cpu_access = drm_gem_dmabuf_begin_cpu_access,
.map = drm_gem_dmabuf_kmap,
.map_atomic = drm_gem_dmabuf_kmap_atomic,
.unmap = drm_gem_dmabuf_kunmap,
diff --git a/include/drm/drm_drv.h b/include/drm/drm_drv.h
index 412e83a..1fbf298 100644
--- a/include/drm/drm_drv.h
+++ b/include/drm/drm_drv.h
@@ -475,6 +475,8 @@ struct drm_driver {
struct drm_device *dev,
struct dma_buf_attachment *attach,
struct sg_table *sgt);
+ int (*gem_prime_begin_cpu_access)(struct drm_gem_object *obj,
+ enum dma_data_direction direction);
void *(*gem_prime_vmap)(struct drm_gem_object *obj);
void (*gem_prime_vunmap)(struct drm_gem_object *obj, void *vaddr);
int (*gem_prime_mmap)(struct drm_gem_object *obj,
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
[not found] ` <1513355282-20248-1-git-send-email-Samuel.Li-5C7GfCeVMHo@public.gmane.org>
@ 2017-12-16 5:40 ` kbuild test robot
0 siblings, 0 replies; 7+ messages in thread
From: kbuild test robot @ 2017-12-16 5:40 UTC (permalink / raw)
To: Samuel Li
Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, kbuild-all-JC7UmRfGjtg,
dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Christian König
[-- Attachment #1: Type: text/plain, Size: 22971 bytes --]
Hi Christian,
I love your patch! Perhaps something to improve:
[auto build test WARNING on drm/drm-next]
[also build test WARNING on v4.15-rc3 next-20171215]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Samuel-Li/drm-prime-forward-begin_cpu_access-callback-to-drivers/20171216-125056
base: git://people.freedesktop.org/~airlied/linux.git drm-next
config: i386-randconfig-x016-201750 (attached as .config)
compiler: gcc-7 (Debian 7.2.0-12) 7.2.1 20171025
reproduce:
# save the attached .config to linux build tree
make ARCH=i386
All warnings (new ones prefixed by >>):
In file included from drivers/gpu/drm/drm_drv.c:36:0:
>> include/drm/drm_drv.h:494:14: warning: 'enum dma_data_direction' declared inside parameter list will not be visible outside of this definition or declaration
enum dma_data_direction direction);
^~~~~~~~~~~~~~~~~~
vim +494 include/drm/drm_drv.h
59
60 /**
61 * struct drm_driver - DRM driver structure
62 *
63 * This structure represent the common code for a family of cards. There will
64 * one drm_device for each card present in this family. It contains lots of
65 * vfunc entries, and a pile of those probably should be moved to more
66 * appropriate places like &drm_mode_config_funcs or into a new operations
67 * structure for GEM drivers.
68 */
69 struct drm_driver {
70 /**
71 * @load:
72 *
73 * Backward-compatible driver callback to complete
74 * initialization steps after the driver is registered. For
75 * this reason, may suffer from race conditions and its use is
76 * deprecated for new drivers. It is therefore only supported
77 * for existing drivers not yet converted to the new scheme.
78 * See drm_dev_init() and drm_dev_register() for proper and
79 * race-free way to set up a &struct drm_device.
80 *
81 * This is deprecated, do not use!
82 *
83 * Returns:
84 *
85 * Zero on success, non-zero value on failure.
86 */
87 int (*load) (struct drm_device *, unsigned long flags);
88
89 /**
90 * @open:
91 *
92 * Driver callback when a new &struct drm_file is opened. Useful for
93 * setting up driver-private data structures like buffer allocators,
94 * execution contexts or similar things. Such driver-private resources
95 * must be released again in @postclose.
96 *
97 * Since the display/modeset side of DRM can only be owned by exactly
98 * one &struct drm_file (see &drm_file.is_master and &drm_device.master)
99 * there should never be a need to set up any modeset related resources
100 * in this callback. Doing so would be a driver design bug.
101 *
102 * Returns:
103 *
104 * 0 on success, a negative error code on failure, which will be
105 * promoted to userspace as the result of the open() system call.
106 */
107 int (*open) (struct drm_device *, struct drm_file *);
108
109 /**
110 * @postclose:
111 *
112 * One of the driver callbacks when a new &struct drm_file is closed.
113 * Useful for tearing down driver-private data structures allocated in
114 * @open like buffer allocators, execution contexts or similar things.
115 *
116 * Since the display/modeset side of DRM can only be owned by exactly
117 * one &struct drm_file (see &drm_file.is_master and &drm_device.master)
118 * there should never be a need to tear down any modeset related
119 * resources in this callback. Doing so would be a driver design bug.
120 */
121 void (*postclose) (struct drm_device *, struct drm_file *);
122
123 /**
124 * @lastclose:
125 *
126 * Called when the last &struct drm_file has been closed and there's
127 * currently no userspace client for the &struct drm_device.
128 *
129 * Modern drivers should only use this to force-restore the fbdev
130 * framebuffer using drm_fb_helper_restore_fbdev_mode_unlocked().
131 * Anything else would indicate there's something seriously wrong.
132 * Modern drivers can also use this to execute delayed power switching
133 * state changes, e.g. in conjunction with the :ref:`vga_switcheroo`
134 * infrastructure.
135 *
136 * This is called after @postclose hook has been called.
137 *
138 * NOTE:
139 *
140 * All legacy drivers use this callback to de-initialize the hardware.
141 * This is purely because of the shadow-attach model, where the DRM
142 * kernel driver does not really own the hardware. Instead ownershipe is
143 * handled with the help of userspace through an inheritedly racy dance
144 * to set/unset the VT into raw mode.
145 *
146 * Legacy drivers initialize the hardware in the @firstopen callback,
147 * which isn't even called for modern drivers.
148 */
149 void (*lastclose) (struct drm_device *);
150
151 /**
152 * @unload:
153 *
154 * Reverse the effects of the driver load callback. Ideally,
155 * the clean up performed by the driver should happen in the
156 * reverse order of the initialization. Similarly to the load
157 * hook, this handler is deprecated and its usage should be
158 * dropped in favor of an open-coded teardown function at the
159 * driver layer. See drm_dev_unregister() and drm_dev_put()
160 * for the proper way to remove a &struct drm_device.
161 *
162 * The unload() hook is called right after unregistering
163 * the device.
164 *
165 */
166 void (*unload) (struct drm_device *);
167
168 /**
169 * @release:
170 *
171 * Optional callback for destroying device data after the final
172 * reference is released, i.e. the device is being destroyed. Drivers
173 * using this callback are responsible for calling drm_dev_fini()
174 * to finalize the device and then freeing the struct themselves.
175 */
176 void (*release) (struct drm_device *);
177
178 /**
179 * @get_vblank_counter:
180 *
181 * Driver callback for fetching a raw hardware vblank counter for the
182 * CRTC specified with the pipe argument. If a device doesn't have a
183 * hardware counter, the driver can simply leave the hook as NULL.
184 * The DRM core will account for missed vblank events while interrupts
185 * where disabled based on system timestamps.
186 *
187 * Wraparound handling and loss of events due to modesetting is dealt
188 * with in the DRM core code, as long as drivers call
189 * drm_crtc_vblank_off() and drm_crtc_vblank_on() when disabling or
190 * enabling a CRTC.
191 *
192 * This is deprecated and should not be used by new drivers.
193 * Use &drm_crtc_funcs.get_vblank_counter instead.
194 *
195 * Returns:
196 *
197 * Raw vblank counter value.
198 */
199 u32 (*get_vblank_counter) (struct drm_device *dev, unsigned int pipe);
200
201 /**
202 * @enable_vblank:
203 *
204 * Enable vblank interrupts for the CRTC specified with the pipe
205 * argument.
206 *
207 * This is deprecated and should not be used by new drivers.
208 * Use &drm_crtc_funcs.enable_vblank instead.
209 *
210 * Returns:
211 *
212 * Zero on success, appropriate errno if the given @crtc's vblank
213 * interrupt cannot be enabled.
214 */
215 int (*enable_vblank) (struct drm_device *dev, unsigned int pipe);
216
217 /**
218 * @disable_vblank:
219 *
220 * Disable vblank interrupts for the CRTC specified with the pipe
221 * argument.
222 *
223 * This is deprecated and should not be used by new drivers.
224 * Use &drm_crtc_funcs.disable_vblank instead.
225 */
226 void (*disable_vblank) (struct drm_device *dev, unsigned int pipe);
227
228 /**
229 * @get_scanout_position:
230 *
231 * Called by vblank timestamping code.
232 *
233 * Returns the current display scanout position from a crtc, and an
234 * optional accurate ktime_get() timestamp of when position was
235 * measured. Note that this is a helper callback which is only used if a
236 * driver uses drm_calc_vbltimestamp_from_scanoutpos() for the
237 * @get_vblank_timestamp callback.
238 *
239 * Parameters:
240 *
241 * dev:
242 * DRM device.
243 * pipe:
244 * Id of the crtc to query.
245 * in_vblank_irq:
246 * True when called from drm_crtc_handle_vblank(). Some drivers
247 * need to apply some workarounds for gpu-specific vblank irq quirks
248 * if flag is set.
249 * vpos:
250 * Target location for current vertical scanout position.
251 * hpos:
252 * Target location for current horizontal scanout position.
253 * stime:
254 * Target location for timestamp taken immediately before
255 * scanout position query. Can be NULL to skip timestamp.
256 * etime:
257 * Target location for timestamp taken immediately after
258 * scanout position query. Can be NULL to skip timestamp.
259 * mode:
260 * Current display timings.
261 *
262 * Returns vpos as a positive number while in active scanout area.
263 * Returns vpos as a negative number inside vblank, counting the number
264 * of scanlines to go until end of vblank, e.g., -1 means "one scanline
265 * until start of active scanout / end of vblank."
266 *
267 * Returns:
268 *
269 * True on success, false if a reliable scanout position counter could
270 * not be read out.
271 *
272 * FIXME:
273 *
274 * Since this is a helper to implement @get_vblank_timestamp, we should
275 * move it to &struct drm_crtc_helper_funcs, like all the other
276 * helper-internal hooks.
277 */
278 bool (*get_scanout_position) (struct drm_device *dev, unsigned int pipe,
279 bool in_vblank_irq, int *vpos, int *hpos,
280 ktime_t *stime, ktime_t *etime,
281 const struct drm_display_mode *mode);
282
283 /**
284 * @get_vblank_timestamp:
285 *
286 * Called by drm_get_last_vbltimestamp(). Should return a precise
287 * timestamp when the most recent VBLANK interval ended or will end.
288 *
289 * Specifically, the timestamp in @vblank_time should correspond as
290 * closely as possible to the time when the first video scanline of
291 * the video frame after the end of VBLANK will start scanning out,
292 * the time immediately after end of the VBLANK interval. If the
293 * @crtc is currently inside VBLANK, this will be a time in the future.
294 * If the @crtc is currently scanning out a frame, this will be the
295 * past start time of the current scanout. This is meant to adhere
296 * to the OpenML OML_sync_control extension specification.
297 *
298 * Paramters:
299 *
300 * dev:
301 * dev DRM device handle.
302 * pipe:
303 * crtc for which timestamp should be returned.
304 * max_error:
305 * Maximum allowable timestamp error in nanoseconds.
306 * Implementation should strive to provide timestamp
307 * with an error of at most max_error nanoseconds.
308 * Returns true upper bound on error for timestamp.
309 * vblank_time:
310 * Target location for returned vblank timestamp.
311 * in_vblank_irq:
312 * True when called from drm_crtc_handle_vblank(). Some drivers
313 * need to apply some workarounds for gpu-specific vblank irq quirks
314 * if flag is set.
315 *
316 * Returns:
317 *
318 * True on success, false on failure, which means the core should
319 * fallback to a simple timestamp taken in drm_crtc_handle_vblank().
320 *
321 * FIXME:
322 *
323 * We should move this hook to &struct drm_crtc_funcs like all the other
324 * vblank hooks.
325 */
326 bool (*get_vblank_timestamp) (struct drm_device *dev, unsigned int pipe,
327 int *max_error,
328 ktime_t *vblank_time,
329 bool in_vblank_irq);
330
331 /**
332 * @irq_handler:
333 *
334 * Interrupt handler called when using drm_irq_install(). Not used by
335 * drivers which implement their own interrupt handling.
336 */
337 irqreturn_t(*irq_handler) (int irq, void *arg);
338
339 /**
340 * @irq_preinstall:
341 *
342 * Optional callback used by drm_irq_install() which is called before
343 * the interrupt handler is registered. This should be used to clear out
344 * any pending interrupts (from e.g. firmware based drives) and reset
345 * the interrupt handling registers.
346 */
347 void (*irq_preinstall) (struct drm_device *dev);
348
349 /**
350 * @irq_postinstall:
351 *
352 * Optional callback used by drm_irq_install() which is called after
353 * the interrupt handler is registered. This should be used to enable
354 * interrupt generation in the hardware.
355 */
356 int (*irq_postinstall) (struct drm_device *dev);
357
358 /**
359 * @irq_uninstall:
360 *
361 * Optional callback used by drm_irq_uninstall() which is called before
362 * the interrupt handler is unregistered. This should be used to disable
363 * interrupt generation in the hardware.
364 */
365 void (*irq_uninstall) (struct drm_device *dev);
366
367 /**
368 * @master_create:
369 *
370 * Called whenever a new master is created. Only used by vmwgfx.
371 */
372 int (*master_create)(struct drm_device *dev, struct drm_master *master);
373
374 /**
375 * @master_destroy:
376 *
377 * Called whenever a master is destroyed. Only used by vmwgfx.
378 */
379 void (*master_destroy)(struct drm_device *dev, struct drm_master *master);
380
381 /**
382 * @master_set:
383 *
384 * Called whenever the minor master is set. Only used by vmwgfx.
385 */
386 int (*master_set)(struct drm_device *dev, struct drm_file *file_priv,
387 bool from_open);
388 /**
389 * @master_drop:
390 *
391 * Called whenever the minor master is dropped. Only used by vmwgfx.
392 */
393 void (*master_drop)(struct drm_device *dev, struct drm_file *file_priv);
394
395 /**
396 * @debugfs_init:
397 *
398 * Allows drivers to create driver-specific debugfs files.
399 */
400 int (*debugfs_init)(struct drm_minor *minor);
401
402 /**
403 * @gem_free_object: deconstructor for drm_gem_objects
404 *
405 * This is deprecated and should not be used by new drivers. Use
406 * @gem_free_object_unlocked instead.
407 */
408 void (*gem_free_object) (struct drm_gem_object *obj);
409
410 /**
411 * @gem_free_object_unlocked: deconstructor for drm_gem_objects
412 *
413 * This is for drivers which are not encumbered with &drm_device.struct_mutex
414 * legacy locking schemes. Use this hook instead of @gem_free_object.
415 */
416 void (*gem_free_object_unlocked) (struct drm_gem_object *obj);
417
418 /**
419 * @gem_open_object:
420 *
421 * Driver hook called upon gem handle creation
422 */
423 int (*gem_open_object) (struct drm_gem_object *, struct drm_file *);
424
425 /**
426 * @gem_close_object:
427 *
428 * Driver hook called upon gem handle release
429 */
430 void (*gem_close_object) (struct drm_gem_object *, struct drm_file *);
431
432 /**
433 * @gem_print_info:
434 *
435 * If driver subclasses struct &drm_gem_object, it can implement this
436 * optional hook for printing additional driver specific info.
437 *
438 * drm_printf_indent() should be used in the callback passing it the
439 * indent argument.
440 *
441 * This callback is called from drm_gem_print_info().
442 */
443 void (*gem_print_info)(struct drm_printer *p, unsigned int indent,
444 const struct drm_gem_object *obj);
445
446 /**
447 * @gem_create_object: constructor for gem objects
448 *
449 * Hook for allocating the GEM object struct, for use by core
450 * helpers.
451 */
452 struct drm_gem_object *(*gem_create_object)(struct drm_device *dev,
453 size_t size);
454
455 /* prime: */
456 /**
457 * @prime_handle_to_fd:
458 *
459 * export handle -> fd (see drm_gem_prime_handle_to_fd() helper)
460 */
461 int (*prime_handle_to_fd)(struct drm_device *dev, struct drm_file *file_priv,
462 uint32_t handle, uint32_t flags, int *prime_fd);
463 /**
464 * @prime_fd_to_handle:
465 *
466 * import fd -> handle (see drm_gem_prime_fd_to_handle() helper)
467 */
468 int (*prime_fd_to_handle)(struct drm_device *dev, struct drm_file *file_priv,
469 int prime_fd, uint32_t *handle);
470 /**
471 * @gem_prime_export:
472 *
473 * export GEM -> dmabuf
474 */
475 struct dma_buf * (*gem_prime_export)(struct drm_device *dev,
476 struct drm_gem_object *obj, int flags);
477 /**
478 * @gem_prime_import:
479 *
480 * import dmabuf -> GEM
481 */
482 struct drm_gem_object * (*gem_prime_import)(struct drm_device *dev,
483 struct dma_buf *dma_buf);
484 int (*gem_prime_pin)(struct drm_gem_object *obj);
485 void (*gem_prime_unpin)(struct drm_gem_object *obj);
486 struct reservation_object * (*gem_prime_res_obj)(
487 struct drm_gem_object *obj);
488 struct sg_table *(*gem_prime_get_sg_table)(struct drm_gem_object *obj);
489 struct drm_gem_object *(*gem_prime_import_sg_table)(
490 struct drm_device *dev,
491 struct dma_buf_attachment *attach,
492 struct sg_table *sgt);
493 int (*gem_prime_begin_cpu_access)(struct drm_gem_object *obj,
> 494 enum dma_data_direction direction);
495 void *(*gem_prime_vmap)(struct drm_gem_object *obj);
496 void (*gem_prime_vunmap)(struct drm_gem_object *obj, void *vaddr);
497 int (*gem_prime_mmap)(struct drm_gem_object *obj,
498 struct vm_area_struct *vma);
499
500 /**
501 * @dumb_create:
502 *
503 * This creates a new dumb buffer in the driver's backing storage manager (GEM,
504 * TTM or something else entirely) and returns the resulting buffer handle. This
505 * handle can then be wrapped up into a framebuffer modeset object.
506 *
507 * Note that userspace is not allowed to use such objects for render
508 * acceleration - drivers must create their own private ioctls for such a use
509 * case.
510 *
511 * Width, height and depth are specified in the &drm_mode_create_dumb
512 * argument. The callback needs to fill the handle, pitch and size for
513 * the created buffer.
514 *
515 * Called by the user via ioctl.
516 *
517 * Returns:
518 *
519 * Zero on success, negative errno on failure.
520 */
521 int (*dumb_create)(struct drm_file *file_priv,
522 struct drm_device *dev,
523 struct drm_mode_create_dumb *args);
524 /**
525 * @dumb_map_offset:
526 *
527 * Allocate an offset in the drm device node's address space to be able to
528 * memory map a dumb buffer. GEM-based drivers must use
529 * drm_gem_create_mmap_offset() to implement this.
530 *
531 * Called by the user via ioctl.
532 *
533 * Returns:
534 *
535 * Zero on success, negative errno on failure.
536 */
537 int (*dumb_map_offset)(struct drm_file *file_priv,
538 struct drm_device *dev, uint32_t handle,
539 uint64_t *offset);
540 /**
541 * @dumb_destroy:
542 *
543 * This destroys the userspace handle for the given dumb backing storage buffer.
544 * Since buffer objects must be reference counted in the kernel a buffer object
545 * won't be immediately freed if a framebuffer modeset object still uses it.
546 *
547 * Called by the user via ioctl.
548 *
549 * Returns:
550 *
551 * Zero on success, negative errno on failure.
552 */
553 int (*dumb_destroy)(struct drm_file *file_priv,
554 struct drm_device *dev,
555 uint32_t handle);
556
557 /**
558 * @gem_vm_ops: Driver private ops for this object
559 */
560 const struct vm_operations_struct *gem_vm_ops;
561
562 /** @major: driver major number */
563 int major;
564 /** @minor: driver minor number */
565 int minor;
566 /** @patchlevel: driver patch level */
567 int patchlevel;
568 /** @name: driver name */
569 char *name;
570 /** @desc: driver description */
571 char *desc;
572 /** @date: driver date */
573 char *date;
574
575 /** @driver_features: driver features */
576 u32 driver_features;
577
578 /**
579 * @ioctls:
580 *
581 * Array of driver-private IOCTL description entries. See the chapter on
582 * :ref:`IOCTL support in the userland interfaces
583 * chapter<drm_driver_ioctl>` for the full details.
584 */
585
586 const struct drm_ioctl_desc *ioctls;
587 /** @num_ioctls: Number of entries in @ioctls. */
588 int num_ioctls;
589
590 /**
591 * @fops:
592 *
593 * File operations for the DRM device node. See the discussion in
594 * :ref:`file operations<drm_driver_fops>` for in-depth coverage and
595 * some examples.
596 */
597 const struct file_operations *fops;
598
599 /* Everything below here is for legacy driver, never use! */
600 /* private: */
601
602 /* List of devices hanging off this driver with stealth attach. */
603 struct list_head legacy_dev_list;
604 int (*firstopen) (struct drm_device *);
605 void (*preclose) (struct drm_device *, struct drm_file *file_priv);
606 int (*dma_ioctl) (struct drm_device *dev, void *data, struct drm_file *file_priv);
607 int (*dma_quiescent) (struct drm_device *);
608 int (*context_dtor) (struct drm_device *dev, int context);
609 int dev_priv_size;
610 };
611
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 24047 bytes --]
[-- Attachment #3: Type: text/plain, Size: 154 bytes --]
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
2017-12-15 16:28 [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers Samuel Li
[not found] ` <1513355282-20248-1-git-send-email-Samuel.Li-5C7GfCeVMHo@public.gmane.org>
@ 2017-12-16 6:29 ` kbuild test robot
2017-12-20 17:20 ` Li, Samuel
2 siblings, 0 replies; 7+ messages in thread
From: kbuild test robot @ 2017-12-16 6:29 UTC (permalink / raw)
To: Samuel Li; +Cc: amd-gfx, kbuild-all, dri-devel, Christian König
[-- Attachment #1: Type: text/plain, Size: 23010 bytes --]
Hi Christian,
I love your patch! Perhaps something to improve:
[auto build test WARNING on drm/drm-next]
[also build test WARNING on v4.15-rc3 next-20171215]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Samuel-Li/drm-prime-forward-begin_cpu_access-callback-to-drivers/20171216-125056
base: git://people.freedesktop.org/~airlied/linux.git drm-next
config: i386-randconfig-a0-201750 (attached as .config)
compiler: gcc-4.9 (Debian 4.9.4-2) 4.9.4
reproduce:
# save the attached .config to linux build tree
make ARCH=i386
All warnings (new ones prefixed by >>):
In file included from drivers/gpu/drm/drm_drv.c:36:0:
>> include/drm/drm_drv.h:494:14: warning: 'enum dma_data_direction' declared inside parameter list
enum dma_data_direction direction);
^
>> include/drm/drm_drv.h:494:14: warning: its scope is only this definition or declaration, which is probably not what you want
vim +494 include/drm/drm_drv.h
59
60 /**
61 * struct drm_driver - DRM driver structure
62 *
63 * This structure represent the common code for a family of cards. There will
64 * one drm_device for each card present in this family. It contains lots of
65 * vfunc entries, and a pile of those probably should be moved to more
66 * appropriate places like &drm_mode_config_funcs or into a new operations
67 * structure for GEM drivers.
68 */
69 struct drm_driver {
70 /**
71 * @load:
72 *
73 * Backward-compatible driver callback to complete
74 * initialization steps after the driver is registered. For
75 * this reason, may suffer from race conditions and its use is
76 * deprecated for new drivers. It is therefore only supported
77 * for existing drivers not yet converted to the new scheme.
78 * See drm_dev_init() and drm_dev_register() for proper and
79 * race-free way to set up a &struct drm_device.
80 *
81 * This is deprecated, do not use!
82 *
83 * Returns:
84 *
85 * Zero on success, non-zero value on failure.
86 */
87 int (*load) (struct drm_device *, unsigned long flags);
88
89 /**
90 * @open:
91 *
92 * Driver callback when a new &struct drm_file is opened. Useful for
93 * setting up driver-private data structures like buffer allocators,
94 * execution contexts or similar things. Such driver-private resources
95 * must be released again in @postclose.
96 *
97 * Since the display/modeset side of DRM can only be owned by exactly
98 * one &struct drm_file (see &drm_file.is_master and &drm_device.master)
99 * there should never be a need to set up any modeset related resources
100 * in this callback. Doing so would be a driver design bug.
101 *
102 * Returns:
103 *
104 * 0 on success, a negative error code on failure, which will be
105 * promoted to userspace as the result of the open() system call.
106 */
107 int (*open) (struct drm_device *, struct drm_file *);
108
109 /**
110 * @postclose:
111 *
112 * One of the driver callbacks when a new &struct drm_file is closed.
113 * Useful for tearing down driver-private data structures allocated in
114 * @open like buffer allocators, execution contexts or similar things.
115 *
116 * Since the display/modeset side of DRM can only be owned by exactly
117 * one &struct drm_file (see &drm_file.is_master and &drm_device.master)
118 * there should never be a need to tear down any modeset related
119 * resources in this callback. Doing so would be a driver design bug.
120 */
121 void (*postclose) (struct drm_device *, struct drm_file *);
122
123 /**
124 * @lastclose:
125 *
126 * Called when the last &struct drm_file has been closed and there's
127 * currently no userspace client for the &struct drm_device.
128 *
129 * Modern drivers should only use this to force-restore the fbdev
130 * framebuffer using drm_fb_helper_restore_fbdev_mode_unlocked().
131 * Anything else would indicate there's something seriously wrong.
132 * Modern drivers can also use this to execute delayed power switching
133 * state changes, e.g. in conjunction with the :ref:`vga_switcheroo`
134 * infrastructure.
135 *
136 * This is called after @postclose hook has been called.
137 *
138 * NOTE:
139 *
140 * All legacy drivers use this callback to de-initialize the hardware.
141 * This is purely because of the shadow-attach model, where the DRM
142 * kernel driver does not really own the hardware. Instead ownershipe is
143 * handled with the help of userspace through an inheritedly racy dance
144 * to set/unset the VT into raw mode.
145 *
146 * Legacy drivers initialize the hardware in the @firstopen callback,
147 * which isn't even called for modern drivers.
148 */
149 void (*lastclose) (struct drm_device *);
150
151 /**
152 * @unload:
153 *
154 * Reverse the effects of the driver load callback. Ideally,
155 * the clean up performed by the driver should happen in the
156 * reverse order of the initialization. Similarly to the load
157 * hook, this handler is deprecated and its usage should be
158 * dropped in favor of an open-coded teardown function at the
159 * driver layer. See drm_dev_unregister() and drm_dev_put()
160 * for the proper way to remove a &struct drm_device.
161 *
162 * The unload() hook is called right after unregistering
163 * the device.
164 *
165 */
166 void (*unload) (struct drm_device *);
167
168 /**
169 * @release:
170 *
171 * Optional callback for destroying device data after the final
172 * reference is released, i.e. the device is being destroyed. Drivers
173 * using this callback are responsible for calling drm_dev_fini()
174 * to finalize the device and then freeing the struct themselves.
175 */
176 void (*release) (struct drm_device *);
177
178 /**
179 * @get_vblank_counter:
180 *
181 * Driver callback for fetching a raw hardware vblank counter for the
182 * CRTC specified with the pipe argument. If a device doesn't have a
183 * hardware counter, the driver can simply leave the hook as NULL.
184 * The DRM core will account for missed vblank events while interrupts
185 * where disabled based on system timestamps.
186 *
187 * Wraparound handling and loss of events due to modesetting is dealt
188 * with in the DRM core code, as long as drivers call
189 * drm_crtc_vblank_off() and drm_crtc_vblank_on() when disabling or
190 * enabling a CRTC.
191 *
192 * This is deprecated and should not be used by new drivers.
193 * Use &drm_crtc_funcs.get_vblank_counter instead.
194 *
195 * Returns:
196 *
197 * Raw vblank counter value.
198 */
199 u32 (*get_vblank_counter) (struct drm_device *dev, unsigned int pipe);
200
201 /**
202 * @enable_vblank:
203 *
204 * Enable vblank interrupts for the CRTC specified with the pipe
205 * argument.
206 *
207 * This is deprecated and should not be used by new drivers.
208 * Use &drm_crtc_funcs.enable_vblank instead.
209 *
210 * Returns:
211 *
212 * Zero on success, appropriate errno if the given @crtc's vblank
213 * interrupt cannot be enabled.
214 */
215 int (*enable_vblank) (struct drm_device *dev, unsigned int pipe);
216
217 /**
218 * @disable_vblank:
219 *
220 * Disable vblank interrupts for the CRTC specified with the pipe
221 * argument.
222 *
223 * This is deprecated and should not be used by new drivers.
224 * Use &drm_crtc_funcs.disable_vblank instead.
225 */
226 void (*disable_vblank) (struct drm_device *dev, unsigned int pipe);
227
228 /**
229 * @get_scanout_position:
230 *
231 * Called by vblank timestamping code.
232 *
233 * Returns the current display scanout position from a crtc, and an
234 * optional accurate ktime_get() timestamp of when position was
235 * measured. Note that this is a helper callback which is only used if a
236 * driver uses drm_calc_vbltimestamp_from_scanoutpos() for the
237 * @get_vblank_timestamp callback.
238 *
239 * Parameters:
240 *
241 * dev:
242 * DRM device.
243 * pipe:
244 * Id of the crtc to query.
245 * in_vblank_irq:
246 * True when called from drm_crtc_handle_vblank(). Some drivers
247 * need to apply some workarounds for gpu-specific vblank irq quirks
248 * if flag is set.
249 * vpos:
250 * Target location for current vertical scanout position.
251 * hpos:
252 * Target location for current horizontal scanout position.
253 * stime:
254 * Target location for timestamp taken immediately before
255 * scanout position query. Can be NULL to skip timestamp.
256 * etime:
257 * Target location for timestamp taken immediately after
258 * scanout position query. Can be NULL to skip timestamp.
259 * mode:
260 * Current display timings.
261 *
262 * Returns vpos as a positive number while in active scanout area.
263 * Returns vpos as a negative number inside vblank, counting the number
264 * of scanlines to go until end of vblank, e.g., -1 means "one scanline
265 * until start of active scanout / end of vblank."
266 *
267 * Returns:
268 *
269 * True on success, false if a reliable scanout position counter could
270 * not be read out.
271 *
272 * FIXME:
273 *
274 * Since this is a helper to implement @get_vblank_timestamp, we should
275 * move it to &struct drm_crtc_helper_funcs, like all the other
276 * helper-internal hooks.
277 */
278 bool (*get_scanout_position) (struct drm_device *dev, unsigned int pipe,
279 bool in_vblank_irq, int *vpos, int *hpos,
280 ktime_t *stime, ktime_t *etime,
281 const struct drm_display_mode *mode);
282
283 /**
284 * @get_vblank_timestamp:
285 *
286 * Called by drm_get_last_vbltimestamp(). Should return a precise
287 * timestamp when the most recent VBLANK interval ended or will end.
288 *
289 * Specifically, the timestamp in @vblank_time should correspond as
290 * closely as possible to the time when the first video scanline of
291 * the video frame after the end of VBLANK will start scanning out,
292 * the time immediately after end of the VBLANK interval. If the
293 * @crtc is currently inside VBLANK, this will be a time in the future.
294 * If the @crtc is currently scanning out a frame, this will be the
295 * past start time of the current scanout. This is meant to adhere
296 * to the OpenML OML_sync_control extension specification.
297 *
298 * Paramters:
299 *
300 * dev:
301 * dev DRM device handle.
302 * pipe:
303 * crtc for which timestamp should be returned.
304 * max_error:
305 * Maximum allowable timestamp error in nanoseconds.
306 * Implementation should strive to provide timestamp
307 * with an error of at most max_error nanoseconds.
308 * Returns true upper bound on error for timestamp.
309 * vblank_time:
310 * Target location for returned vblank timestamp.
311 * in_vblank_irq:
312 * True when called from drm_crtc_handle_vblank(). Some drivers
313 * need to apply some workarounds for gpu-specific vblank irq quirks
314 * if flag is set.
315 *
316 * Returns:
317 *
318 * True on success, false on failure, which means the core should
319 * fallback to a simple timestamp taken in drm_crtc_handle_vblank().
320 *
321 * FIXME:
322 *
323 * We should move this hook to &struct drm_crtc_funcs like all the other
324 * vblank hooks.
325 */
326 bool (*get_vblank_timestamp) (struct drm_device *dev, unsigned int pipe,
327 int *max_error,
328 ktime_t *vblank_time,
329 bool in_vblank_irq);
330
331 /**
332 * @irq_handler:
333 *
334 * Interrupt handler called when using drm_irq_install(). Not used by
335 * drivers which implement their own interrupt handling.
336 */
337 irqreturn_t(*irq_handler) (int irq, void *arg);
338
339 /**
340 * @irq_preinstall:
341 *
342 * Optional callback used by drm_irq_install() which is called before
343 * the interrupt handler is registered. This should be used to clear out
344 * any pending interrupts (from e.g. firmware based drives) and reset
345 * the interrupt handling registers.
346 */
347 void (*irq_preinstall) (struct drm_device *dev);
348
349 /**
350 * @irq_postinstall:
351 *
352 * Optional callback used by drm_irq_install() which is called after
353 * the interrupt handler is registered. This should be used to enable
354 * interrupt generation in the hardware.
355 */
356 int (*irq_postinstall) (struct drm_device *dev);
357
358 /**
359 * @irq_uninstall:
360 *
361 * Optional callback used by drm_irq_uninstall() which is called before
362 * the interrupt handler is unregistered. This should be used to disable
363 * interrupt generation in the hardware.
364 */
365 void (*irq_uninstall) (struct drm_device *dev);
366
367 /**
368 * @master_create:
369 *
370 * Called whenever a new master is created. Only used by vmwgfx.
371 */
372 int (*master_create)(struct drm_device *dev, struct drm_master *master);
373
374 /**
375 * @master_destroy:
376 *
377 * Called whenever a master is destroyed. Only used by vmwgfx.
378 */
379 void (*master_destroy)(struct drm_device *dev, struct drm_master *master);
380
381 /**
382 * @master_set:
383 *
384 * Called whenever the minor master is set. Only used by vmwgfx.
385 */
386 int (*master_set)(struct drm_device *dev, struct drm_file *file_priv,
387 bool from_open);
388 /**
389 * @master_drop:
390 *
391 * Called whenever the minor master is dropped. Only used by vmwgfx.
392 */
393 void (*master_drop)(struct drm_device *dev, struct drm_file *file_priv);
394
395 /**
396 * @debugfs_init:
397 *
398 * Allows drivers to create driver-specific debugfs files.
399 */
400 int (*debugfs_init)(struct drm_minor *minor);
401
402 /**
403 * @gem_free_object: deconstructor for drm_gem_objects
404 *
405 * This is deprecated and should not be used by new drivers. Use
406 * @gem_free_object_unlocked instead.
407 */
408 void (*gem_free_object) (struct drm_gem_object *obj);
409
410 /**
411 * @gem_free_object_unlocked: deconstructor for drm_gem_objects
412 *
413 * This is for drivers which are not encumbered with &drm_device.struct_mutex
414 * legacy locking schemes. Use this hook instead of @gem_free_object.
415 */
416 void (*gem_free_object_unlocked) (struct drm_gem_object *obj);
417
418 /**
419 * @gem_open_object:
420 *
421 * Driver hook called upon gem handle creation
422 */
423 int (*gem_open_object) (struct drm_gem_object *, struct drm_file *);
424
425 /**
426 * @gem_close_object:
427 *
428 * Driver hook called upon gem handle release
429 */
430 void (*gem_close_object) (struct drm_gem_object *, struct drm_file *);
431
432 /**
433 * @gem_print_info:
434 *
435 * If driver subclasses struct &drm_gem_object, it can implement this
436 * optional hook for printing additional driver specific info.
437 *
438 * drm_printf_indent() should be used in the callback passing it the
439 * indent argument.
440 *
441 * This callback is called from drm_gem_print_info().
442 */
443 void (*gem_print_info)(struct drm_printer *p, unsigned int indent,
444 const struct drm_gem_object *obj);
445
446 /**
447 * @gem_create_object: constructor for gem objects
448 *
449 * Hook for allocating the GEM object struct, for use by core
450 * helpers.
451 */
452 struct drm_gem_object *(*gem_create_object)(struct drm_device *dev,
453 size_t size);
454
455 /* prime: */
456 /**
457 * @prime_handle_to_fd:
458 *
459 * export handle -> fd (see drm_gem_prime_handle_to_fd() helper)
460 */
461 int (*prime_handle_to_fd)(struct drm_device *dev, struct drm_file *file_priv,
462 uint32_t handle, uint32_t flags, int *prime_fd);
463 /**
464 * @prime_fd_to_handle:
465 *
466 * import fd -> handle (see drm_gem_prime_fd_to_handle() helper)
467 */
468 int (*prime_fd_to_handle)(struct drm_device *dev, struct drm_file *file_priv,
469 int prime_fd, uint32_t *handle);
470 /**
471 * @gem_prime_export:
472 *
473 * export GEM -> dmabuf
474 */
475 struct dma_buf * (*gem_prime_export)(struct drm_device *dev,
476 struct drm_gem_object *obj, int flags);
477 /**
478 * @gem_prime_import:
479 *
480 * import dmabuf -> GEM
481 */
482 struct drm_gem_object * (*gem_prime_import)(struct drm_device *dev,
483 struct dma_buf *dma_buf);
484 int (*gem_prime_pin)(struct drm_gem_object *obj);
485 void (*gem_prime_unpin)(struct drm_gem_object *obj);
486 struct reservation_object * (*gem_prime_res_obj)(
487 struct drm_gem_object *obj);
488 struct sg_table *(*gem_prime_get_sg_table)(struct drm_gem_object *obj);
489 struct drm_gem_object *(*gem_prime_import_sg_table)(
490 struct drm_device *dev,
491 struct dma_buf_attachment *attach,
492 struct sg_table *sgt);
493 int (*gem_prime_begin_cpu_access)(struct drm_gem_object *obj,
> 494 enum dma_data_direction direction);
495 void *(*gem_prime_vmap)(struct drm_gem_object *obj);
496 void (*gem_prime_vunmap)(struct drm_gem_object *obj, void *vaddr);
497 int (*gem_prime_mmap)(struct drm_gem_object *obj,
498 struct vm_area_struct *vma);
499
500 /**
501 * @dumb_create:
502 *
503 * This creates a new dumb buffer in the driver's backing storage manager (GEM,
504 * TTM or something else entirely) and returns the resulting buffer handle. This
505 * handle can then be wrapped up into a framebuffer modeset object.
506 *
507 * Note that userspace is not allowed to use such objects for render
508 * acceleration - drivers must create their own private ioctls for such a use
509 * case.
510 *
511 * Width, height and depth are specified in the &drm_mode_create_dumb
512 * argument. The callback needs to fill the handle, pitch and size for
513 * the created buffer.
514 *
515 * Called by the user via ioctl.
516 *
517 * Returns:
518 *
519 * Zero on success, negative errno on failure.
520 */
521 int (*dumb_create)(struct drm_file *file_priv,
522 struct drm_device *dev,
523 struct drm_mode_create_dumb *args);
524 /**
525 * @dumb_map_offset:
526 *
527 * Allocate an offset in the drm device node's address space to be able to
528 * memory map a dumb buffer. GEM-based drivers must use
529 * drm_gem_create_mmap_offset() to implement this.
530 *
531 * Called by the user via ioctl.
532 *
533 * Returns:
534 *
535 * Zero on success, negative errno on failure.
536 */
537 int (*dumb_map_offset)(struct drm_file *file_priv,
538 struct drm_device *dev, uint32_t handle,
539 uint64_t *offset);
540 /**
541 * @dumb_destroy:
542 *
543 * This destroys the userspace handle for the given dumb backing storage buffer.
544 * Since buffer objects must be reference counted in the kernel a buffer object
545 * won't be immediately freed if a framebuffer modeset object still uses it.
546 *
547 * Called by the user via ioctl.
548 *
549 * Returns:
550 *
551 * Zero on success, negative errno on failure.
552 */
553 int (*dumb_destroy)(struct drm_file *file_priv,
554 struct drm_device *dev,
555 uint32_t handle);
556
557 /**
558 * @gem_vm_ops: Driver private ops for this object
559 */
560 const struct vm_operations_struct *gem_vm_ops;
561
562 /** @major: driver major number */
563 int major;
564 /** @minor: driver minor number */
565 int minor;
566 /** @patchlevel: driver patch level */
567 int patchlevel;
568 /** @name: driver name */
569 char *name;
570 /** @desc: driver description */
571 char *desc;
572 /** @date: driver date */
573 char *date;
574
575 /** @driver_features: driver features */
576 u32 driver_features;
577
578 /**
579 * @ioctls:
580 *
581 * Array of driver-private IOCTL description entries. See the chapter on
582 * :ref:`IOCTL support in the userland interfaces
583 * chapter<drm_driver_ioctl>` for the full details.
584 */
585
586 const struct drm_ioctl_desc *ioctls;
587 /** @num_ioctls: Number of entries in @ioctls. */
588 int num_ioctls;
589
590 /**
591 * @fops:
592 *
593 * File operations for the DRM device node. See the discussion in
594 * :ref:`file operations<drm_driver_fops>` for in-depth coverage and
595 * some examples.
596 */
597 const struct file_operations *fops;
598
599 /* Everything below here is for legacy driver, never use! */
600 /* private: */
601
602 /* List of devices hanging off this driver with stealth attach. */
603 struct list_head legacy_dev_list;
604 int (*firstopen) (struct drm_device *);
605 void (*preclose) (struct drm_device *, struct drm_file *file_priv);
606 int (*dma_ioctl) (struct drm_device *dev, void *data, struct drm_file *file_priv);
607 int (*dma_quiescent) (struct drm_device *);
608 int (*context_dtor) (struct drm_device *dev, int context);
609 int dev_priv_size;
610 };
611
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 29334 bytes --]
[-- Attachment #3: Type: text/plain, Size: 160 bytes --]
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* RE: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
2017-12-15 16:28 [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers Samuel Li
[not found] ` <1513355282-20248-1-git-send-email-Samuel.Li-5C7GfCeVMHo@public.gmane.org>
2017-12-16 6:29 ` kbuild test robot
@ 2017-12-20 17:20 ` Li, Samuel
2017-12-20 19:43 ` Daniel Vetter
2 siblings, 1 reply; 7+ messages in thread
From: Li, Samuel @ 2017-12-20 17:20 UTC (permalink / raw)
To: Li, Samuel, dri-devel, amd-gfx; +Cc: Koenig, Christian
Ping... can someone please review this patch?
Samuel Li
> -----Original Message-----
> From: amd-gfx [mailto:amd-gfx-bounces@lists.freedesktop.org] On Behalf
> Of Samuel Li
> Sent: Friday, December 15, 2017 11:28 AM
> To: dri-devel@lists.freedesktop.org; amd-gfx@lists.freedesktop.org
> Cc: Koenig, Christian <Christian.Koenig@amd.com>
> Subject: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to
> drivers
>
> From: Christian König <christian.koenig@amd.com>
>
> Allow drivers to implement their own begin_cpu_access callback.
>
> Change-Id: I97709b42b9351a04ee7e01106107a87bc56ea258
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
> drivers/gpu/drm/drm_prime.c | 13 +++++++++++++
> include/drm/drm_drv.h | 2 ++
> 2 files changed, 15 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
> index 8de93a2..b4b0e64 100644
> --- a/drivers/gpu/drm/drm_prime.c
> +++ b/drivers/gpu/drm/drm_prime.c
> @@ -346,6 +346,18 @@ void drm_gem_dmabuf_release(struct dma_buf
> *dma_buf) } EXPORT_SYMBOL(drm_gem_dmabuf_release);
>
> +static int drm_gem_dmabuf_begin_cpu_access(struct dma_buf *dma_buf,
> + enum dma_data_direction
> direction) {
> + struct drm_gem_object *obj = dma_buf->priv;
> + struct drm_device *dev = obj->dev;
> +
> + if (!dev->driver->gem_prime_begin_cpu_access)
> + return 0;
> +
> + return dev->driver->gem_prime_begin_cpu_access(obj, direction); }
> +
> static void *drm_gem_dmabuf_vmap(struct dma_buf *dma_buf) {
> struct drm_gem_object *obj = dma_buf->priv; @@ -403,6 +415,7
> @@ static const struct dma_buf_ops drm_gem_prime_dmabuf_ops = {
> .map_dma_buf = drm_gem_map_dma_buf,
> .unmap_dma_buf = drm_gem_unmap_dma_buf,
> .release = drm_gem_dmabuf_release,
> + .begin_cpu_access = drm_gem_dmabuf_begin_cpu_access,
> .map = drm_gem_dmabuf_kmap,
> .map_atomic = drm_gem_dmabuf_kmap_atomic,
> .unmap = drm_gem_dmabuf_kunmap,
> diff --git a/include/drm/drm_drv.h b/include/drm/drm_drv.h index
> 412e83a..1fbf298 100644
> --- a/include/drm/drm_drv.h
> +++ b/include/drm/drm_drv.h
> @@ -475,6 +475,8 @@ struct drm_driver {
> struct drm_device *dev,
> struct dma_buf_attachment *attach,
> struct sg_table *sgt);
> + int (*gem_prime_begin_cpu_access)(struct drm_gem_object *obj,
> + enum dma_data_direction
> direction);
> void *(*gem_prime_vmap)(struct drm_gem_object *obj);
> void (*gem_prime_vunmap)(struct drm_gem_object *obj, void
> *vaddr);
> int (*gem_prime_mmap)(struct drm_gem_object *obj,
> --
> 2.7.4
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
2017-12-20 17:20 ` Li, Samuel
@ 2017-12-20 19:43 ` Daniel Vetter
2017-12-20 19:54 ` Christian König
0 siblings, 1 reply; 7+ messages in thread
From: Daniel Vetter @ 2017-12-20 19:43 UTC (permalink / raw)
To: Li, Samuel; +Cc: amd-gfx, dri-devel, Koenig, Christian
On Wed, Dec 20, 2017 at 6:20 PM, Li, Samuel <Samuel.Li@amd.com> wrote:
> Ping... can someone please review this patch?
Might be simpler to implement your own dma-buf backend instead of
going through the drm_prime midlayer. That was mostly written to give
nvidia a set of non-EXPORT_GPL symbols to support dma-buf. Or
something like that.
Also don't expect people to look at patches when CI bots spot issues.
-Daniel
>
> Samuel Li
>
>
>
>> -----Original Message-----
>> From: amd-gfx [mailto:amd-gfx-bounces@lists.freedesktop.org] On Behalf
>> Of Samuel Li
>> Sent: Friday, December 15, 2017 11:28 AM
>> To: dri-devel@lists.freedesktop.org; amd-gfx@lists.freedesktop.org
>> Cc: Koenig, Christian <Christian.Koenig@amd.com>
>> Subject: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to
>> drivers
>>
>> From: Christian König <christian.koenig@amd.com>
>>
>> Allow drivers to implement their own begin_cpu_access callback.
>>
>> Change-Id: I97709b42b9351a04ee7e01106107a87bc56ea258
>> Signed-off-by: Christian König <christian.koenig@amd.com>
>> ---
>> drivers/gpu/drm/drm_prime.c | 13 +++++++++++++
>> include/drm/drm_drv.h | 2 ++
>> 2 files changed, 15 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
>> index 8de93a2..b4b0e64 100644
>> --- a/drivers/gpu/drm/drm_prime.c
>> +++ b/drivers/gpu/drm/drm_prime.c
>> @@ -346,6 +346,18 @@ void drm_gem_dmabuf_release(struct dma_buf
>> *dma_buf) } EXPORT_SYMBOL(drm_gem_dmabuf_release);
>>
>> +static int drm_gem_dmabuf_begin_cpu_access(struct dma_buf *dma_buf,
>> + enum dma_data_direction
>> direction) {
>> + struct drm_gem_object *obj = dma_buf->priv;
>> + struct drm_device *dev = obj->dev;
>> +
>> + if (!dev->driver->gem_prime_begin_cpu_access)
>> + return 0;
>> +
>> + return dev->driver->gem_prime_begin_cpu_access(obj, direction); }
>> +
>> static void *drm_gem_dmabuf_vmap(struct dma_buf *dma_buf) {
>> struct drm_gem_object *obj = dma_buf->priv; @@ -403,6 +415,7
>> @@ static const struct dma_buf_ops drm_gem_prime_dmabuf_ops = {
>> .map_dma_buf = drm_gem_map_dma_buf,
>> .unmap_dma_buf = drm_gem_unmap_dma_buf,
>> .release = drm_gem_dmabuf_release,
>> + .begin_cpu_access = drm_gem_dmabuf_begin_cpu_access,
>> .map = drm_gem_dmabuf_kmap,
>> .map_atomic = drm_gem_dmabuf_kmap_atomic,
>> .unmap = drm_gem_dmabuf_kunmap,
>> diff --git a/include/drm/drm_drv.h b/include/drm/drm_drv.h index
>> 412e83a..1fbf298 100644
>> --- a/include/drm/drm_drv.h
>> +++ b/include/drm/drm_drv.h
>> @@ -475,6 +475,8 @@ struct drm_driver {
>> struct drm_device *dev,
>> struct dma_buf_attachment *attach,
>> struct sg_table *sgt);
>> + int (*gem_prime_begin_cpu_access)(struct drm_gem_object *obj,
>> + enum dma_data_direction
>> direction);
>> void *(*gem_prime_vmap)(struct drm_gem_object *obj);
>> void (*gem_prime_vunmap)(struct drm_gem_object *obj, void
>> *vaddr);
>> int (*gem_prime_mmap)(struct drm_gem_object *obj,
>> --
>> 2.7.4
>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
2017-12-20 19:43 ` Daniel Vetter
@ 2017-12-20 19:54 ` Christian König
[not found] ` <4c0df435-913c-7569-591b-f15c212bc249-5C7GfCeVMHo@public.gmane.org>
0 siblings, 1 reply; 7+ messages in thread
From: Christian König @ 2017-12-20 19:54 UTC (permalink / raw)
To: Daniel Vetter, Li, Samuel; +Cc: amd-gfx, dri-devel
Am 20.12.2017 um 20:43 schrieb Daniel Vetter:
> On Wed, Dec 20, 2017 at 6:20 PM, Li, Samuel <Samuel.Li@amd.com> wrote:
>> Ping... can someone please review this patch?
> Might be simpler to implement your own dma-buf backend instead of
> going through the drm_prime midlayer. That was mostly written to give
> nvidia a set of non-EXPORT_GPL symbols to support dma-buf. Or
> something like that.
Ah, that explains that. Well the alternative Sam suggest was to export
most of the functions implementing that.
But then we have a hard time having a common detection logic if a
DMA-buf has a GEM object behind it or not.
> Also don't expect people to look at patches when CI bots spot issues.
That is either a false positive or a one liner.
Regards,
Christian.
> -Daniel
>
>> Samuel Li
>>
>>
>>
>>> -----Original Message-----
>>> From: amd-gfx [mailto:amd-gfx-bounces@lists.freedesktop.org] On Behalf
>>> Of Samuel Li
>>> Sent: Friday, December 15, 2017 11:28 AM
>>> To: dri-devel@lists.freedesktop.org; amd-gfx@lists.freedesktop.org
>>> Cc: Koenig, Christian <Christian.Koenig@amd.com>
>>> Subject: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to
>>> drivers
>>>
>>> From: Christian König <christian.koenig@amd.com>
>>>
>>> Allow drivers to implement their own begin_cpu_access callback.
>>>
>>> Change-Id: I97709b42b9351a04ee7e01106107a87bc56ea258
>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>> ---
>>> drivers/gpu/drm/drm_prime.c | 13 +++++++++++++
>>> include/drm/drm_drv.h | 2 ++
>>> 2 files changed, 15 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
>>> index 8de93a2..b4b0e64 100644
>>> --- a/drivers/gpu/drm/drm_prime.c
>>> +++ b/drivers/gpu/drm/drm_prime.c
>>> @@ -346,6 +346,18 @@ void drm_gem_dmabuf_release(struct dma_buf
>>> *dma_buf) } EXPORT_SYMBOL(drm_gem_dmabuf_release);
>>>
>>> +static int drm_gem_dmabuf_begin_cpu_access(struct dma_buf *dma_buf,
>>> + enum dma_data_direction
>>> direction) {
>>> + struct drm_gem_object *obj = dma_buf->priv;
>>> + struct drm_device *dev = obj->dev;
>>> +
>>> + if (!dev->driver->gem_prime_begin_cpu_access)
>>> + return 0;
>>> +
>>> + return dev->driver->gem_prime_begin_cpu_access(obj, direction); }
>>> +
>>> static void *drm_gem_dmabuf_vmap(struct dma_buf *dma_buf) {
>>> struct drm_gem_object *obj = dma_buf->priv; @@ -403,6 +415,7
>>> @@ static const struct dma_buf_ops drm_gem_prime_dmabuf_ops = {
>>> .map_dma_buf = drm_gem_map_dma_buf,
>>> .unmap_dma_buf = drm_gem_unmap_dma_buf,
>>> .release = drm_gem_dmabuf_release,
>>> + .begin_cpu_access = drm_gem_dmabuf_begin_cpu_access,
>>> .map = drm_gem_dmabuf_kmap,
>>> .map_atomic = drm_gem_dmabuf_kmap_atomic,
>>> .unmap = drm_gem_dmabuf_kunmap,
>>> diff --git a/include/drm/drm_drv.h b/include/drm/drm_drv.h index
>>> 412e83a..1fbf298 100644
>>> --- a/include/drm/drm_drv.h
>>> +++ b/include/drm/drm_drv.h
>>> @@ -475,6 +475,8 @@ struct drm_driver {
>>> struct drm_device *dev,
>>> struct dma_buf_attachment *attach,
>>> struct sg_table *sgt);
>>> + int (*gem_prime_begin_cpu_access)(struct drm_gem_object *obj,
>>> + enum dma_data_direction
>>> direction);
>>> void *(*gem_prime_vmap)(struct drm_gem_object *obj);
>>> void (*gem_prime_vunmap)(struct drm_gem_object *obj, void
>>> *vaddr);
>>> int (*gem_prime_mmap)(struct drm_gem_object *obj,
>>> --
>>> 2.7.4
>>>
>>> _______________________________________________
>>> amd-gfx mailing list
>>> amd-gfx@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers
[not found] ` <4c0df435-913c-7569-591b-f15c212bc249-5C7GfCeVMHo@public.gmane.org>
@ 2017-12-21 10:04 ` Daniel Vetter
0 siblings, 0 replies; 7+ messages in thread
From: Daniel Vetter @ 2017-12-21 10:04 UTC (permalink / raw)
To: Christian König
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW,
amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Li, Samuel,
Daniel Vetter
On Wed, Dec 20, 2017 at 08:54:33PM +0100, Christian König wrote:
> Am 20.12.2017 um 20:43 schrieb Daniel Vetter:
> > On Wed, Dec 20, 2017 at 6:20 PM, Li, Samuel <Samuel.Li@amd.com> wrote:
> > > Ping... can someone please review this patch?
> > Might be simpler to implement your own dma-buf backend instead of
> > going through the drm_prime midlayer. That was mostly written to give
> > nvidia a set of non-EXPORT_GPL symbols to support dma-buf. Or
> > something like that.
>
> Ah, that explains that. Well the alternative Sam suggest was to export most
> of the functions implementing that.
Yeah that sounds like a good idea, makes it feel less midlayer-y.
> But then we have a hard time having a common detection logic if a DMA-buf
> has a GEM object behind it or not.
Hm, why do you need that? I guess you'll care whether it's an
amdgpu-exported one, and you can still do that by checking against your
own dma_buf_ops table. We do the same in i915, without using the core
drm_prime helpers.
> > Also don't expect people to look at patches when CI bots spot issues.
>
> That is either a false positive or a one liner.
Hey it's holiday seasons, leave me some cheap excuses pls :-)
Cheers, Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2017-12-21 10:04 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-12-15 16:28 [PATCH 1/3] drm/prime: forward begin_cpu_access callback to drivers Samuel Li
[not found] ` <1513355282-20248-1-git-send-email-Samuel.Li-5C7GfCeVMHo@public.gmane.org>
2017-12-16 5:40 ` kbuild test robot
2017-12-16 6:29 ` kbuild test robot
2017-12-20 17:20 ` Li, Samuel
2017-12-20 19:43 ` Daniel Vetter
2017-12-20 19:54 ` Christian König
[not found] ` <4c0df435-913c-7569-591b-f15c212bc249-5C7GfCeVMHo@public.gmane.org>
2017-12-21 10:04 ` Daniel Vetter
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.