You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
weston/libweston/linux-dmabuf.c

1079 lines
32 KiB

/*
* Copyright © 2014, 2015 Collabora, Ltd.
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice (including the
* next paragraph) shall be included in all copies or substantial
* portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "config.h"
#include <assert.h>
#include <errno.h>
#include <stdint.h>
#include <unistd.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <libweston/libweston.h>
#include "linux-dmabuf.h"
#include "shared/os-compatibility.h"
#include "libweston-internal.h"
#include "shared/weston-drm-fourcc.h"
static void
linux_dmabuf_buffer_destroy(struct linux_dmabuf_buffer *buffer)
{
int i;
for (i = 0; i < buffer->attributes.n_planes; i++) {
close(buffer->attributes.fd[i]);
buffer->attributes.fd[i] = -1;
}
buffer->attributes.n_planes = 0;
free(buffer);
}
static void
destroy_params(struct wl_resource *params_resource)
{
struct linux_dmabuf_buffer *buffer;
buffer = wl_resource_get_user_data(params_resource);
if (!buffer)
return;
linux_dmabuf_buffer_destroy(buffer);
}
static void
params_destroy(struct wl_client *client, struct wl_resource *resource)
{
wl_resource_destroy(resource);
}
static void
params_add(struct wl_client *client,
struct wl_resource *params_resource,
int32_t name_fd,
uint32_t plane_idx,
uint32_t offset,
uint32_t stride,
uint32_t modifier_hi,
uint32_t modifier_lo)
{
struct linux_dmabuf_buffer *buffer;
buffer = wl_resource_get_user_data(params_resource);
if (!buffer) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_ALREADY_USED,
"params was already used to create a wl_buffer");
close(name_fd);
return;
}
assert(buffer->params_resource == params_resource);
assert(!buffer->buffer_resource);
if (plane_idx >= MAX_DMABUF_PLANES) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_PLANE_IDX,
"plane index %u is too high", plane_idx);
close(name_fd);
return;
}
if (buffer->attributes.fd[plane_idx] != -1) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_PLANE_SET,
"a dmabuf has already been added for plane %u",
plane_idx);
close(name_fd);
return;
}
buffer->attributes.fd[plane_idx] = name_fd;
buffer->attributes.offset[plane_idx] = offset;
buffer->attributes.stride[plane_idx] = stride;
if (wl_resource_get_version(params_resource) < ZWP_LINUX_DMABUF_V1_MODIFIER_SINCE_VERSION)
buffer->attributes.modifier[plane_idx] = DRM_FORMAT_MOD_INVALID;
else
buffer->attributes.modifier[plane_idx] = ((uint64_t)modifier_hi << 32) |
modifier_lo;
buffer->attributes.n_planes++;
}
static void
linux_dmabuf_wl_buffer_destroy(struct wl_client *client,
struct wl_resource *resource)
{
wl_resource_destroy(resource);
}
static const struct wl_buffer_interface linux_dmabuf_buffer_implementation = {
linux_dmabuf_wl_buffer_destroy
};
static void
destroy_linux_dmabuf_wl_buffer(struct wl_resource *resource)
{
struct linux_dmabuf_buffer *buffer;
buffer = wl_resource_get_user_data(resource);
assert(buffer->buffer_resource == resource);
assert(!buffer->params_resource);
if (buffer->user_data_destroy_func)
buffer->user_data_destroy_func(buffer);
linux_dmabuf_buffer_destroy(buffer);
}
static void
params_create_common(struct wl_client *client,
struct wl_resource *params_resource,
uint32_t buffer_id,
int32_t width,
int32_t height,
uint32_t format,
uint32_t flags)
{
struct linux_dmabuf_buffer *buffer;
int i;
buffer = wl_resource_get_user_data(params_resource);
if (!buffer) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_ALREADY_USED,
"params was already used to create a wl_buffer");
return;
}
assert(buffer->params_resource == params_resource);
assert(!buffer->buffer_resource);
/* Switch the linux_dmabuf_buffer object from params resource to
* eventually wl_buffer resource.
*/
wl_resource_set_user_data(buffer->params_resource, NULL);
buffer->params_resource = NULL;
if (!buffer->attributes.n_planes) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_INCOMPLETE,
"no dmabuf has been added to the params");
goto err_out;
}
/* Check for holes in the dmabufs set (e.g. [0, 1, 3]) */
for (i = 0; i < buffer->attributes.n_planes; i++) {
if (buffer->attributes.fd[i] == -1) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_INCOMPLETE,
"no dmabuf has been added for plane %i", i);
goto err_out;
}
}
buffer->attributes.width = width;
buffer->attributes.height = height;
buffer->attributes.format = format;
buffer->attributes.flags = flags;
if (width < 1 || height < 1) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_INVALID_DIMENSIONS,
"invalid width %d or height %d", width, height);
goto err_out;
}
for (i = 0; i < buffer->attributes.n_planes; i++) {
off_t size;
if ((uint64_t) buffer->attributes.offset[i] + buffer->attributes.stride[i] > UINT32_MAX) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_OUT_OF_BOUNDS,
"size overflow for plane %i", i);
goto err_out;
}
if (i == 0 &&
(uint64_t) buffer->attributes.offset[i] +
(uint64_t) buffer->attributes.stride[i] * height > UINT32_MAX) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_OUT_OF_BOUNDS,
"size overflow for plane %i", i);
goto err_out;
}
/* Don't report an error as it might be caused
* by the kernel not supporting seeking on dmabuf */
size = lseek(buffer->attributes.fd[i], 0, SEEK_END);
if (size == -1)
continue;
if (buffer->attributes.offset[i] >= size) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_OUT_OF_BOUNDS,
"invalid offset %i for plane %i",
buffer->attributes.offset[i], i);
goto err_out;
}
if (buffer->attributes.offset[i] + buffer->attributes.stride[i] > size) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_OUT_OF_BOUNDS,
"invalid stride %i for plane %i",
buffer->attributes.stride[i], i);
goto err_out;
}
/* Only valid for first plane as other planes might be
* sub-sampled according to fourcc format */
if (i == 0 &&
buffer->attributes.offset[i] + buffer->attributes.stride[i] * height > size) {
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_OUT_OF_BOUNDS,
"invalid buffer stride or height for plane %i", i);
goto err_out;
}
}
if (buffer->direct_display) {
if (!weston_compositor_dmabuf_can_scanout(buffer->compositor,
buffer))
goto err_failed;
goto avoid_gpu_import;
}
if (!weston_compositor_import_dmabuf(buffer->compositor, buffer))
goto err_failed;
avoid_gpu_import:
buffer->buffer_resource = wl_resource_create(client,
&wl_buffer_interface,
1, buffer_id);
if (!buffer->buffer_resource) {
wl_resource_post_no_memory(params_resource);
goto err_buffer;
}
wl_resource_set_implementation(buffer->buffer_resource,
&linux_dmabuf_buffer_implementation,
buffer, destroy_linux_dmabuf_wl_buffer);
/* send 'created' event when the request is not for an immediate
* import, ie buffer_id is zero */
if (buffer_id == 0)
zwp_linux_buffer_params_v1_send_created(params_resource,
buffer->buffer_resource);
return;
err_buffer:
if (buffer->user_data_destroy_func)
buffer->user_data_destroy_func(buffer);
err_failed:
if (buffer_id == 0)
zwp_linux_buffer_params_v1_send_failed(params_resource);
else
/* since the behavior is left implementation defined by the
* protocol in case of create_immed failure due to an unknown cause,
* we choose to treat it as a fatal error and immediately kill the
* client instead of creating an invalid handle and waiting for it
* to be used.
*/
wl_resource_post_error(params_resource,
ZWP_LINUX_BUFFER_PARAMS_V1_ERROR_INVALID_WL_BUFFER,
"importing the supplied dmabufs failed");
err_out:
linux_dmabuf_buffer_destroy(buffer);
}
static void
params_create(struct wl_client *client,
struct wl_resource *params_resource,
int32_t width,
int32_t height,
uint32_t format,
uint32_t flags)
{
params_create_common(client, params_resource, 0, width, height, format,
flags);
}
static void
params_create_immed(struct wl_client *client,
struct wl_resource *params_resource,
uint32_t buffer_id,
int32_t width,
int32_t height,
uint32_t format,
uint32_t flags)
{
params_create_common(client, params_resource, buffer_id, width, height,
format, flags);
}
static const struct zwp_linux_buffer_params_v1_interface
zwp_linux_buffer_params_implementation = {
params_destroy,
params_add,
params_create,
params_create_immed
};
static void
linux_dmabuf_destroy(struct wl_client *client, struct wl_resource *resource)
{
wl_resource_destroy(resource);
}
static void
linux_dmabuf_create_params(struct wl_client *client,
struct wl_resource *linux_dmabuf_resource,
uint32_t params_id)
{
struct weston_compositor *compositor;
struct linux_dmabuf_buffer *buffer;
uint32_t version;
int i;
version = wl_resource_get_version(linux_dmabuf_resource);
compositor = wl_resource_get_user_data(linux_dmabuf_resource);
buffer = zalloc(sizeof *buffer);
if (!buffer)
goto err_out;
for (i = 0; i < MAX_DMABUF_PLANES; i++)
buffer->attributes.fd[i] = -1;
buffer->compositor = compositor;
buffer->params_resource =
wl_resource_create(client,
&zwp_linux_buffer_params_v1_interface,
version, params_id);
buffer->direct_display = false;
if (!buffer->params_resource)
goto err_dealloc;
wl_resource_set_implementation(buffer->params_resource,
&zwp_linux_buffer_params_implementation,
buffer, destroy_params);
return;
err_dealloc:
free(buffer);
err_out:
wl_resource_post_no_memory(linux_dmabuf_resource);
}
/** Creates dma-buf feedback tranche
*
* The tranche is added to dma-buf feedback's tranche list
*
* @param dmabuf_feedback The dma-buf feedback object to which the tranche is added
* @param format_table The dma-buf feedback formats table
* @param target_device The target device of the new tranche
* @param flags The flags of the new tranche
* @param preference The preference of the new tranche
* @return The tranche created, or NULL on failure
*/
WL_EXPORT struct weston_dmabuf_feedback_tranche *
weston_dmabuf_feedback_tranche_create(struct weston_dmabuf_feedback *dmabuf_feedback,
struct weston_dmabuf_feedback_format_table *format_table,
dev_t target_device, uint32_t flags,
enum weston_dmabuf_feedback_tranche_preference preference)
{
struct weston_dmabuf_feedback_tranche *tranche, *ptr;
struct wl_list *pos;
tranche = zalloc(sizeof(*tranche));
if (!tranche) {
weston_log("%s: out of memory\n", __func__);
return NULL;
}
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
tranche->active = true;
tranche->target_device = target_device;
tranche->flags = flags;
tranche->preference = preference;
/* Get the formats indices array */
if (flags == 0) {
if (wl_array_copy(&tranche->formats_indices,
&format_table->renderer_formats_indices) < 0) {
weston_log("%s: out of memory\n", __func__);
goto err;
}
} else if (flags & ZWP_LINUX_DMABUF_FEEDBACK_V1_TRANCHE_FLAGS_SCANOUT) {
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
if (wl_array_copy(&tranche->formats_indices,
&format_table->scanout_formats_indices) < 0) {
weston_log("%s: out of memory\n", __func__);
goto err;
}
} else {
weston_log("error: for now we just have renderer and scanout "
"tranches, can't create other type of tranche\n");
goto err;
}
/* The list of tranches is ordered by preference.
* Highest preference comes first. */
pos = &dmabuf_feedback->tranche_list;
wl_list_for_each(ptr, &dmabuf_feedback->tranche_list, link) {
pos = &ptr->link;
if (ptr->preference <= tranche->preference)
break;
}
wl_list_insert(pos->prev, &tranche->link);
return tranche;
err:
free(tranche);
return NULL;
}
static void
weston_dmabuf_feedback_tranche_destroy(struct weston_dmabuf_feedback_tranche *tranche)
{
wl_array_release(&tranche->formats_indices);
wl_list_remove(&tranche->link);
free(tranche);
}
static int
format_table_add_renderer_formats(struct weston_dmabuf_feedback_format_table *format_table,
const struct weston_drm_format_array *renderer_formats)
{
struct weston_drm_format *fmt;
unsigned int num_modifiers;
const uint64_t *modifiers;
uint16_t index, *index_ptr;
unsigned int size;
unsigned int i;
size = sizeof(index) *
weston_drm_format_array_count_pairs(renderer_formats);
if (!wl_array_add(&format_table->renderer_formats_indices, size)) {
weston_log("%s: out of memory\n", __func__);
return -1;
}
index = 0;
wl_array_for_each(fmt, &renderer_formats->arr) {
modifiers = weston_drm_format_get_modifiers(fmt, &num_modifiers);
for (i = 0; i < num_modifiers; i++) {
format_table->data[index].format = fmt->format;
format_table->data[index].modifier = modifiers[i];
index++;
}
}
index = 0;
wl_array_for_each(index_ptr, &format_table->renderer_formats_indices)
*index_ptr = index++;
return 0;
}
/** Creates dma-buf feedback format table
*
* @param renderer_formats The formats that compose the table
* @return The dma-buf feedback format table, or NULL on failure
*/
WL_EXPORT struct weston_dmabuf_feedback_format_table *
weston_dmabuf_feedback_format_table_create(const struct weston_drm_format_array *renderer_formats)
{
struct weston_dmabuf_feedback_format_table *format_table;
int ret;
format_table = zalloc(sizeof(*format_table));
if (!format_table) {
weston_log("%s: out of memory\n", __func__);
return NULL;
}
wl_array_init(&format_table->renderer_formats_indices);
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
wl_array_init(&format_table->scanout_formats_indices);
/* Creates formats file table and mmap it */
format_table->size = weston_drm_format_array_count_pairs(renderer_formats) *
sizeof(*format_table->data);
format_table->fd = os_create_anonymous_file(format_table->size);
if (format_table->fd < 0) {
weston_log("error: failed to create format table file: %s\n",
strerror(errno));
goto err_fd;
}
format_table->data = mmap(NULL, format_table->size, PROT_READ | PROT_WRITE,
MAP_SHARED, format_table->fd, 0);
if (format_table->data == MAP_FAILED) {
weston_log("error: mmap for format table failed: %s\n",
strerror(errno));
goto err_mmap;
}
/* Add renderer formats to file table */
ret = format_table_add_renderer_formats(format_table, renderer_formats);
if (ret < 0)
goto err_formats;
return format_table;
err_formats:
munmap(format_table->data, format_table->size);
err_mmap:
close(format_table->fd);
err_fd:
wl_array_release(&format_table->renderer_formats_indices);
free(format_table);
return NULL;
}
/** Destroys dma-buf feedback formats table
*
* @param format_table The dma-buf feedback format table to destroy
*/
WL_EXPORT void
weston_dmabuf_feedback_format_table_destroy(struct weston_dmabuf_feedback_format_table *format_table)
{
wl_array_release(&format_table->renderer_formats_indices);
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
wl_array_release(&format_table->scanout_formats_indices);
munmap(format_table->data, format_table->size);
close(format_table->fd);
free(format_table);
}
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
static int
format_table_get_format_index(struct weston_dmabuf_feedback_format_table *format_table,
uint32_t format, uint64_t modifier, uint16_t *index_out)
{
uint16_t index;
unsigned int num_elements = format_table->size / sizeof(index);
for (index = 0; index < num_elements; index++) {
if (format_table->data[index].format == format &&
format_table->data[index].modifier == modifier) {
*index_out = index;
return 0;
}
}
return -1;
}
/** Set scanout formats indices in the dma-buf feedback format table
*
* The table consists of the formats supported by the renderer. A dma-buf
* feedback scanout tranche consists of the union of the KMS plane's formats
* intersected with the renderer formats. With this function we compute the
* indices of these plane's formats in the table and save them in the
* table->scanout_formats_indices, allowing us to create scanout tranches.
*
* @param format_table The dma-buf feedback format table
* @param scanout_formats The scanout formats
* @return 0 on success, -1 on failure
*/
WL_EXPORT int
weston_dmabuf_feedback_format_table_set_scanout_indices(struct weston_dmabuf_feedback_format_table *format_table,
const struct weston_drm_format_array *scanout_formats)
{
struct weston_drm_format *fmt;
unsigned int num_modifiers;
const uint64_t *modifiers;
uint16_t index, *index_ptr;
unsigned int i;
int ret;
wl_array_for_each(fmt, &scanout_formats->arr) {
modifiers = weston_drm_format_get_modifiers(fmt, &num_modifiers);
for (i = 0; i < num_modifiers; i++) {
index_ptr =
wl_array_add(&format_table->scanout_formats_indices,
sizeof(index));
if (!index_ptr)
goto err;
ret = format_table_get_format_index(format_table, fmt->format,
modifiers[i], &index);
if (ret < 0)
goto err;
*index_ptr = index;
}
}
return 0;
err:
wl_array_release(&format_table->scanout_formats_indices);
wl_array_init(&format_table->scanout_formats_indices);
return -1;
}
/** Creates dma-buf feedback object
*
* @param main_device The main device of the dma-buf feedback
* @return The dma-buf feedback object created, or NULL on failure
*/
WL_EXPORT struct weston_dmabuf_feedback *
weston_dmabuf_feedback_create(dev_t main_device)
{
struct weston_dmabuf_feedback *dmabuf_feedback;
dmabuf_feedback = zalloc(sizeof(*dmabuf_feedback));
if (!dmabuf_feedback) {
weston_log("%s: out of memory\n", __func__);
return NULL;
}
dmabuf_feedback->main_device = main_device;
wl_list_init(&dmabuf_feedback->tranche_list);
wl_list_init(&dmabuf_feedback->resource_list);
return dmabuf_feedback;
}
/** Destroy dma-buf feedback object
*
* @param dmabuf_feedback The dma-buf feedback object to destroy
*/
WL_EXPORT void
weston_dmabuf_feedback_destroy(struct weston_dmabuf_feedback *dmabuf_feedback)
{
struct weston_dmabuf_feedback_tranche *tranche, *tranche_tmp;
struct wl_resource *res, *res_tmp;
wl_list_for_each_safe(tranche, tranche_tmp, &dmabuf_feedback->tranche_list, link)
weston_dmabuf_feedback_tranche_destroy(tranche);
wl_resource_for_each_safe(res, res_tmp, &dmabuf_feedback->resource_list) {
wl_list_remove(wl_resource_get_link(res));
wl_list_init(wl_resource_get_link(res));
}
free(dmabuf_feedback);
}
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
/** Find tranche in a dma-buf feedback object
*
* @param dmabuf_feedback The dma-buf feedback object where to look for
* @param target_device The target device of the tranche
* @param flags The flags of the tranche
* @param preference The preference of the tranche
* @return The tranche, or NULL if it was not found
*/
WL_EXPORT struct weston_dmabuf_feedback_tranche *
weston_dmabuf_feedback_find_tranche(struct weston_dmabuf_feedback *dmabuf_feedback,
dev_t target_device, uint32_t flags,
enum weston_dmabuf_feedback_tranche_preference preference)
{
struct weston_dmabuf_feedback_tranche *tranche;
wl_list_for_each(tranche, &dmabuf_feedback->tranche_list, link)
if (tranche->target_device == target_device &&
tranche->flags == flags && tranche->preference == preference)
return tranche;
return NULL;
}
static void
weston_dmabuf_feedback_send(struct weston_dmabuf_feedback *dmabuf_feedback,
struct weston_dmabuf_feedback_format_table *format_table,
struct wl_resource *res, bool advertise_format_table)
{
struct weston_dmabuf_feedback_tranche *tranche;
struct wl_array device;
dev_t *dev;
/* main_device and target_device events need a dev_t as parameter,
* but we can't use this directly to communicate with the Wayland
* client. The solution is to use a wl_array, which is supported by
* Wayland, and add the dev_t as an element of the array. */
wl_array_init(&device);
dev = wl_array_add(&device, sizeof(*dev));
if (!dev) {
wl_resource_post_no_memory(res);
return;
}
/* format_table event - In Weston, we never modify the dma-buf feedback
* format table. So we have this flag in order to advertise the format
* table only if the client has just subscribed to receive the events
* for this feedback object. When we need to re-send the feedback events
* for this client, the table event won't be sent. */
if (advertise_format_table)
zwp_linux_dmabuf_feedback_v1_send_format_table(res, format_table->fd,
format_table->size);
/* main_device event */
*dev = dmabuf_feedback->main_device;
zwp_linux_dmabuf_feedback_v1_send_main_device(res, &device);
/* send events for each tranche */
wl_list_for_each(tranche, &dmabuf_feedback->tranche_list, link) {
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
if (!tranche->active)
continue;
/* tranche_target_device event */
*dev = tranche->target_device;
zwp_linux_dmabuf_feedback_v1_send_tranche_target_device(res, &device);
/* tranche_flags event */
zwp_linux_dmabuf_feedback_v1_send_tranche_flags(res, tranche->flags);
/* tranche_formats event */
zwp_linux_dmabuf_feedback_v1_send_tranche_formats(res, &tranche->formats_indices);
/* tranche_done_event */
zwp_linux_dmabuf_feedback_v1_send_tranche_done(res);
}
/* compositor_done_event */
zwp_linux_dmabuf_feedback_v1_send_done(res);
wl_array_release(&device);
}
backend-drm: add KMS plane's formats to per-surface dma-buf feedback In commit "libweston: add initial dma-buf feedback implementation" we've added initial support to dma-buf feedback, but it was not using the feedback from DRM-backend to add useful per-surface feedback. In this patch we add this. The scanout tranche of the per-surface feedback is based on the union of formats/modifiers of the primary and overlay planes available. These are intersected with the formats/modifiers supported by the renderer device. Also, it's important to mention that the scene can change a lot and we can't predict much. So this patch also adds hysteresis to the dma-buf feedback. We wait a few seconds to be sure that we reached stability before adding or removing the scanout tranche from dma-buf feedback and resending them. This help us to avoid spamming clients and leading to unnecessary buffer reallocations on their end. Here's an example of what we want to avoid: 1. We detect that a view was not placed in a plane only because its format is not supported by the plane, so we add the scanout tranche to the feedback and send the events. 2. A few milliseconds after, the view gets occluded. So now the view can't be placed in a plane anymore. We need to remove the scanout tranche and resend the feedback with formats/modifiers optimal for the renderer device. The client will then reallocate its buffers. 3. A few milliseconds after, the view that was causing the occlusion gets minimized. So we got back to the first situation, in which the format of the view is not compatible with the plane. Then we need to add a scanout tranche and resend the feedback... This patch is based on previous work of Scott Anderson (@ascent). Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> Signed-off-by: Scott Anderson <scott.anderson@collabora.com> Reviewed-by: Daniel Stone <daniels@collabora.com>
3 years ago
/** Sends the feedback events for a dma-buf feedback object
*
* Given a dma-buf feedback object, this will send events to clients that are
* subscribed to it. This is useful for the per-surface dma-buf feedback, which
* is dynamic and can change throughout compositor's life. These changes results
* in the need to resend the feedback events to clients.
*
* @param dmabuf_feedback The weston_dmabuf_feedback object
* @param format_table The dma-buf feedback formats table
*/
WL_EXPORT void
weston_dmabuf_feedback_send_all(struct weston_dmabuf_feedback *dmabuf_feedback,
struct weston_dmabuf_feedback_format_table *format_table)
{
struct wl_resource *res;
wl_resource_for_each(res, &dmabuf_feedback->resource_list)
weston_dmabuf_feedback_send(dmabuf_feedback,
format_table, res, false);
}
static void
dmabuf_feedback_resource_destroy(struct wl_resource *resource)
{
wl_list_remove(wl_resource_get_link(resource));
}
static void
dmabuf_feedback_destroy(struct wl_client *client, struct wl_resource *resource)
{
wl_resource_destroy(resource);
}
static const struct zwp_linux_dmabuf_feedback_v1_interface
zwp_linux_dmabuf_feedback_implementation = {
dmabuf_feedback_destroy
};
static struct wl_resource *
dmabuf_feedback_resource_create(struct wl_resource *dmabuf_resource,
struct wl_client *client, uint32_t dmabuf_feedback_id)
{
struct wl_resource *dmabuf_feedback_res;
uint32_t version;
version = wl_resource_get_version(dmabuf_resource);
dmabuf_feedback_res =
wl_resource_create(client, &zwp_linux_dmabuf_feedback_v1_interface,
version, dmabuf_feedback_id);
if (!dmabuf_feedback_res)
return NULL;
wl_list_init(wl_resource_get_link(dmabuf_feedback_res));
wl_resource_set_implementation(dmabuf_feedback_res,
&zwp_linux_dmabuf_feedback_implementation,
NULL, dmabuf_feedback_resource_destroy);
return dmabuf_feedback_res;
}
static void
linux_dmabuf_get_default_feedback(struct wl_client *client,
struct wl_resource *dmabuf_resource,
uint32_t dmabuf_feedback_id)
{
struct weston_compositor *compositor =
wl_resource_get_user_data(dmabuf_resource);
struct wl_resource *dmabuf_feedback_resource;
dmabuf_feedback_resource =
dmabuf_feedback_resource_create(dmabuf_resource,
client, dmabuf_feedback_id);
if (!dmabuf_feedback_resource) {
wl_resource_post_no_memory(dmabuf_resource);
return;
}
weston_dmabuf_feedback_send(compositor->default_dmabuf_feedback,
compositor->dmabuf_feedback_format_table,
dmabuf_feedback_resource, true);
}
static void
linux_dmabuf_get_per_surface_feedback(struct wl_client *client,
struct wl_resource *dmabuf_resource,
uint32_t dmabuf_feedback_id,
struct wl_resource *surface_resource)
{
struct weston_surface *surface =
wl_resource_get_user_data(surface_resource);
struct wl_resource *dmabuf_feedback_resource;
dmabuf_feedback_resource =
dmabuf_feedback_resource_create(dmabuf_resource,
client, dmabuf_feedback_id);
if (!dmabuf_feedback_resource) {
wl_resource_post_no_memory(dmabuf_resource);
return;
}
/* Surface dma-buf feedback is dynamic and may need to be resent to
* clients when they change. So we need to keep the resources list */
wl_list_insert(&surface->dmabuf_feedback->resource_list,
wl_resource_get_link(dmabuf_feedback_resource));
weston_dmabuf_feedback_send(surface->dmabuf_feedback,
surface->compositor->dmabuf_feedback_format_table,
dmabuf_feedback_resource, true);
}
/** Get the linux_dmabuf_buffer from a wl_buffer resource
*
* If the given wl_buffer resource was created through the linux_dmabuf
* protocol interface, returns the linux_dmabuf_buffer object. This can
* be used as a type check for a wl_buffer.
*
* \param resource A wl_buffer resource.
* \return The linux_dmabuf_buffer if it exists, or NULL otherwise.
*/
WL_EXPORT struct linux_dmabuf_buffer *
linux_dmabuf_buffer_get(struct wl_resource *resource)
{
struct linux_dmabuf_buffer *buffer;
if (!resource)
return NULL;
if (!wl_resource_instance_of(resource, &wl_buffer_interface,
&linux_dmabuf_buffer_implementation))
return NULL;
buffer = wl_resource_get_user_data(resource);
assert(buffer);
assert(!buffer->params_resource);
assert(buffer->buffer_resource == resource);
return buffer;
}
/** Set renderer-private data
*
* Set the user data for the linux_dmabuf_buffer. It is invalid to overwrite
* a non-NULL user data with a new non-NULL pointer. This is meant to
* protect against renderers fighting over linux_dmabuf_buffer user data
* ownership.
*
* The renderer-private data is usually set from the
* weston_renderer::import_dmabuf hook.
*
* \param buffer The linux_dmabuf_buffer object to set for.
* \param data The new renderer-private data pointer.
* \param func Destructor function to be called for the renderer-private
* data when the linux_dmabuf_buffer gets destroyed.
*
* \sa weston_compositor_import_dmabuf
*/
WL_EXPORT void
linux_dmabuf_buffer_set_user_data(struct linux_dmabuf_buffer *buffer,
void *data,
dmabuf_user_data_destroy_func func)
{
assert(data == NULL || buffer->user_data == NULL);
buffer->user_data = data;
buffer->user_data_destroy_func = func;
}
/** Get renderer-private data
*
* Get the user data from the linux_dmabuf_buffer.
*
* \param buffer The linux_dmabuf_buffer to query.
* \return Renderer-private data pointer.
*
* \sa linux_dmabuf_buffer_set_user_data
*/
WL_EXPORT void *
linux_dmabuf_buffer_get_user_data(struct linux_dmabuf_buffer *buffer)
{
return buffer->user_data;
}
static const struct zwp_linux_dmabuf_v1_interface linux_dmabuf_implementation = {
linux_dmabuf_destroy,
linux_dmabuf_create_params,
linux_dmabuf_get_default_feedback,
linux_dmabuf_get_per_surface_feedback
};
static void
bind_linux_dmabuf(struct wl_client *client,
void *data, uint32_t version, uint32_t id)
{
struct weston_compositor *compositor = data;
const struct weston_drm_format_array *supported_formats;
struct wl_resource *resource;
struct weston_drm_format *fmt;
const uint64_t *modifiers;
unsigned int num_modifiers;
unsigned int i;
resource = wl_resource_create(client, &zwp_linux_dmabuf_v1_interface,
version, id);
if (resource == NULL) {
wl_client_post_no_memory(client);
return;
}
wl_resource_set_implementation(resource, &linux_dmabuf_implementation,
compositor, NULL);
/* Advertise formats/modifiers. From version 4 onwards, we should not send
* zwp_linux_dmabuf_v1_send_modifier and zwp_linux_dmabuf_v1_send_format
* events, instead we must send the dma-buf feedback events. */
if (version >= 4)
return;
/* If we got here, it means that the renderer is able to import dma-buf
* buffers, and so it must have get_supported_formats() set. */
assert(compositor->renderer->get_supported_formats != NULL);
supported_formats = compositor->renderer->get_supported_formats(compositor);
wl_array_for_each(fmt, &supported_formats->arr) {
modifiers = weston_drm_format_get_modifiers(fmt, &num_modifiers);
for (i = 0; i < num_modifiers; i++) {
if (version >= ZWP_LINUX_DMABUF_V1_MODIFIER_SINCE_VERSION) {
uint32_t modifier_lo = modifiers[i] & 0xFFFFFFFF;
uint32_t modifier_hi = modifiers[i] >> 32;
zwp_linux_dmabuf_v1_send_modifier(resource,
fmt->format,
modifier_hi,
modifier_lo);
} else if (modifiers[i] == DRM_FORMAT_MOD_LINEAR ||
modifiers[i] == DRM_FORMAT_MOD_INVALID) {
zwp_linux_dmabuf_v1_send_format(resource,
fmt->format);
}
}
}
}
/** Advertise linux_dmabuf support
*
* Calling this initializes the zwp_linux_dmabuf protocol support, so that
* the interface will be advertised to clients. Essentially it creates a
* global. Do not call this function multiple times in the compositor's
* lifetime. There is no way to deinit explicitly, globals will be reaped
* when the wl_display gets destroyed.
*
* \param compositor The compositor to init for.
* \return Zero on success, -1 on failure.
*/
WL_EXPORT int
linux_dmabuf_setup(struct weston_compositor *compositor)
{
int max_version;
/* If we were able to create the default dma-buf feedback for the
* compositor, that means that we are able to advertise dma-buf feedback
* events. In such case we support the version 4 of the protocol. */
max_version = compositor->default_dmabuf_feedback ? 4 : 3;
if (!wl_global_create(compositor->wl_display,
&zwp_linux_dmabuf_v1_interface,
max_version, compositor, bind_linux_dmabuf))
return -1;
return 0;
}
/** Resolve an internal compositor error by disconnecting the client.
*
* This function is used in cases when the dmabuf-based wl_buffer
* turns out unusable and there is no fallback path. This is used by
* renderers which are the fallback path in the first place.
*
* It is possible the fault is caused by a compositor bug, the underlying
* graphics stack bug or normal behaviour, or perhaps a client mistake.
* In any case, the options are to either composite garbage or nothing,
* or disconnect the client. This is a helper function for the latter.
*
* The error is sent as an INVALID_OBJECT error on the client's wl_display.
*
* \param buffer The linux_dmabuf_buffer that is unusable.
* \param msg A custom error message attached to the protocol error.
*/
WL_EXPORT void
linux_dmabuf_buffer_send_server_error(struct linux_dmabuf_buffer *buffer,
const char *msg)
{
struct wl_client *client;
struct wl_resource *display_resource;
uint32_t id;
assert(buffer->buffer_resource);
id = wl_resource_get_id(buffer->buffer_resource);
client = wl_resource_get_client(buffer->buffer_resource);
display_resource = wl_client_get_object(client, 1);
assert(display_resource);
wl_resource_post_error(display_resource,
WL_DISPLAY_ERROR_INVALID_OBJECT,
"linux_dmabuf server error with "
"wl_buffer@%u: %s", id, msg);
}