blob: cd89d41bbb386784161ca87f8961a2414072e128 [file] [log] [blame]
/*
* Copyright (c) 2011-2017 The Linux Foundation. All rights reserved.
* Not a Contribution
*
* Copyright (C) 2010 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#define DEBUG 0
#include <iomanip>
#include <utility>
#include <vector>
#include <sstream>
#include "qd_utils.h"
#include "gr_priv_handle.h"
#include "gr_buf_descriptor.h"
#include "gr_utils.h"
#include "gr_buf_mgr.h"
#include "qdMetaData.h"
namespace gralloc1 {
std::atomic<gralloc1_buffer_descriptor_t> BufferDescriptor::next_id_(1);
static BufferInfo GetBufferInfo(const BufferDescriptor &descriptor) {
return BufferInfo(descriptor.GetWidth(), descriptor.GetHeight(), descriptor.GetFormat(),
descriptor.GetProducerUsage(), descriptor.GetConsumerUsage());
}
BufferManager::BufferManager() : next_id_(0) {
char property[PROPERTY_VALUE_MAX];
// Map framebuffer memory
if ((property_get("debug.gralloc.map_fb_memory", property, NULL) > 0) &&
(!strncmp(property, "1", PROPERTY_VALUE_MAX) ||
(!strncasecmp(property, "true", PROPERTY_VALUE_MAX)))) {
map_fb_mem_ = true;
}
handles_map_.clear();
allocator_ = new Allocator();
allocator_->Init();
}
gralloc1_error_t BufferManager::CreateBufferDescriptor(
gralloc1_buffer_descriptor_t *descriptor_id) {
std::lock_guard<std::mutex> lock(descriptor_lock_);
auto descriptor = std::make_shared<BufferDescriptor>();
descriptors_map_.emplace(descriptor->GetId(), descriptor);
*descriptor_id = descriptor->GetId();
return GRALLOC1_ERROR_NONE;
}
gralloc1_error_t BufferManager::DestroyBufferDescriptor(
gralloc1_buffer_descriptor_t descriptor_id) {
std::lock_guard<std::mutex> lock(descriptor_lock_);
const auto descriptor = descriptors_map_.find(descriptor_id);
if (descriptor == descriptors_map_.end()) {
return GRALLOC1_ERROR_BAD_DESCRIPTOR;
}
descriptors_map_.erase(descriptor);
return GRALLOC1_ERROR_NONE;
}
BufferManager::~BufferManager() {
if (allocator_) {
delete allocator_;
}
}
gralloc1_error_t BufferManager::AllocateBuffers(uint32_t num_descriptors,
const gralloc1_buffer_descriptor_t *descriptor_ids,
buffer_handle_t *out_buffers) {
bool shared = true;
gralloc1_error_t status = GRALLOC1_ERROR_NONE;
// since GRALLOC1_CAPABILITY_TEST_ALLOCATE capability is supported
// client can ask to test the allocation by passing NULL out_buffers
bool test_allocate = !out_buffers;
// Validate descriptors
std::lock_guard<std::mutex> descriptor_lock(descriptor_lock_);
std::vector<std::shared_ptr<BufferDescriptor>> descriptors;
for (uint32_t i = 0; i < num_descriptors; i++) {
const auto map_descriptor = descriptors_map_.find(descriptor_ids[i]);
if (map_descriptor == descriptors_map_.end()) {
return GRALLOC1_ERROR_BAD_DESCRIPTOR;
} else {
descriptors.push_back(map_descriptor->second);
}
}
// Resolve implementation defined formats
for (auto &descriptor : descriptors) {
descriptor->SetColorFormat(allocator_->GetImplDefinedFormat(descriptor->GetProducerUsage(),
descriptor->GetConsumerUsage(),
descriptor->GetFormat()));
}
// Check if input descriptors can be supported AND
// Find out if a single buffer can be shared for all the given input descriptors
uint32_t i = 0;
ssize_t max_buf_index = -1;
shared = allocator_->CheckForBufferSharing(num_descriptors, descriptors, &max_buf_index);
if (test_allocate) {
status = shared ? GRALLOC1_ERROR_NOT_SHARED : status;
return status;
}
std::lock_guard<std::mutex> buffer_lock(buffer_lock_);
if (shared && (max_buf_index >= 0)) {
// Allocate one and duplicate/copy the handles for each descriptor
if (AllocateBufferLocked(*descriptors[UINT(max_buf_index)], &out_buffers[max_buf_index])) {
return GRALLOC1_ERROR_NO_RESOURCES;
}
for (i = 0; i < num_descriptors; i++) {
// Create new handle for a given descriptor.
// Current assumption is even MetaData memory would be same
// Need to revisit if there is a need for own metadata memory
if (i != UINT(max_buf_index)) {
CreateSharedHandle(out_buffers[max_buf_index], *descriptors[i], &out_buffers[i]);
}
}
} else {
// Buffer sharing is not feasible.
// Allocate separate buffer for each descriptor
for (i = 0; i < num_descriptors; i++) {
if (AllocateBufferLocked(*descriptors[i], &out_buffers[i])) {
return GRALLOC1_ERROR_NO_RESOURCES;
}
}
}
// Allocation is successful. If backstore is not shared inform the client.
if (!shared) {
return GRALLOC1_ERROR_NOT_SHARED;
}
return status;
}
void BufferManager::CreateSharedHandle(buffer_handle_t inbuffer, const BufferDescriptor &descriptor,
buffer_handle_t *outbuffer) {
// TODO(user): This path is not verified
private_handle_t const *input = reinterpret_cast<private_handle_t const *>(inbuffer);
// Get Buffer attributes or dimension
unsigned int alignedw = 0, alignedh = 0;
BufferInfo info = GetBufferInfo(descriptor);
GetAlignedWidthAndHeight(info, &alignedw, &alignedh);
// create new handle from input reference handle and given descriptor
int flags = GetHandleFlags(descriptor.GetFormat(), descriptor.GetProducerUsage(),
descriptor.GetConsumerUsage());
int buffer_type = GetBufferType(descriptor.GetFormat());
// Duplicate the fds
// TODO(user): Not sure what to do for fb_id. Use duped fd and new dimensions?
private_handle_t *out_hnd = new private_handle_t(dup(input->fd),
dup(input->fd_metadata),
flags,
INT(alignedw),
INT(alignedh),
descriptor.GetWidth(),
descriptor.GetHeight(),
descriptor.GetFormat(),
buffer_type,
input->size,
descriptor.GetProducerUsage(),
descriptor.GetConsumerUsage());
out_hnd->id = ++next_id_;
// TODO(user): Base address of shared handle and ion handles
RegisterHandleLocked(out_hnd, -1, -1);
*outbuffer = out_hnd;
}
gralloc1_error_t BufferManager::FreeBuffer(std::shared_ptr<Buffer> buf) {
auto hnd = buf->handle;
ALOGD_IF(DEBUG, "FreeBuffer handle:%p", hnd);
if (private_handle_t::validate(hnd) != 0) {
ALOGE("FreeBuffer: Invalid handle: %p", hnd);
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (allocator_->FreeBuffer(reinterpret_cast<void *>(hnd->base), hnd->size, hnd->offset,
hnd->fd, buf->ion_handle_main) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
unsigned int meta_size = ALIGN((unsigned int)sizeof(MetaData_t), PAGE_SIZE);
if (allocator_->FreeBuffer(reinterpret_cast<void *>(hnd->base_metadata), meta_size,
hnd->offset_metadata, hnd->fd_metadata, buf->ion_handle_meta) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
private_handle_t * handle = const_cast<private_handle_t *>(hnd);
handle->fd = -1;
handle->fd_metadata = -1;
if (!(handle->flags & private_handle_t::PRIV_FLAGS_CLIENT_ALLOCATED)) {
delete handle;
}
return GRALLOC1_ERROR_NONE;
}
void BufferManager::RegisterHandleLocked(const private_handle_t *hnd,
int ion_handle,
int ion_handle_meta) {
auto buffer = std::make_shared<Buffer>(hnd, ion_handle, ion_handle_meta);
handles_map_.emplace(std::make_pair(hnd, buffer));
}
gralloc1_error_t BufferManager::ImportHandleLocked(private_handle_t *hnd) {
ALOGD_IF(DEBUG, "Importing handle:%p id: %" PRIu64, hnd, hnd->id);
int ion_handle = allocator_->ImportBuffer(hnd->fd);
if (ion_handle < 0) {
ALOGE("Failed to import ion buffer: hnd: %p, fd:%d, id:%" PRIu64, hnd, hnd->fd, hnd->id);
return GRALLOC1_ERROR_BAD_HANDLE;
}
int ion_handle_meta = allocator_->ImportBuffer(hnd->fd_metadata);
if (ion_handle_meta < 0) {
ALOGE("Failed to import ion metadata buffer: hnd: %p, fd:%d, id:%" PRIu64, hnd,
hnd->fd, hnd->id);
return GRALLOC1_ERROR_BAD_HANDLE;
}
// Set base pointers to NULL since the data here was received over binder
hnd->base = 0;
hnd->base_metadata = 0;
RegisterHandleLocked(hnd, ion_handle, ion_handle_meta);
return GRALLOC1_ERROR_NONE;
}
std::shared_ptr<BufferManager::Buffer>
BufferManager::GetBufferFromHandleLocked(const private_handle_t *hnd) {
auto it = handles_map_.find(hnd);
if (it != handles_map_.end()) {
return it->second;
} else {
return nullptr;
}
}
gralloc1_error_t BufferManager::MapBuffer(private_handle_t const *handle) {
private_handle_t *hnd = const_cast<private_handle_t *>(handle);
ALOGD_IF(DEBUG, "Map buffer handle:%p id: %" PRIu64, hnd, hnd->id);
hnd->base = 0;
if (allocator_->MapBuffer(reinterpret_cast<void **>(&hnd->base), hnd->size, hnd->offset,
hnd->fd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
return GRALLOC1_ERROR_NONE;
}
gralloc1_error_t BufferManager::RetainBuffer(private_handle_t const *hnd) {
ALOGD_IF(DEBUG, "Retain buffer handle:%p id: %" PRIu64, hnd, hnd->id);
gralloc1_error_t err = GRALLOC1_ERROR_NONE;
std::lock_guard<std::mutex> lock(buffer_lock_);
auto buf = GetBufferFromHandleLocked(hnd);
if (buf != nullptr) {
buf->IncRef();
} else {
private_handle_t *handle = const_cast<private_handle_t *>(hnd);
err = ImportHandleLocked(handle);
}
return err;
}
gralloc1_error_t BufferManager::ReleaseBuffer(private_handle_t const *hnd) {
ALOGD_IF(DEBUG, "Release buffer handle:%p", hnd);
std::lock_guard<std::mutex> lock(buffer_lock_);
auto buf = GetBufferFromHandleLocked(hnd);
if (buf == nullptr) {
ALOGE("Could not find handle: %p id: %" PRIu64, hnd, hnd->id);
return GRALLOC1_ERROR_BAD_HANDLE;
} else {
if (buf->DecRef()) {
handles_map_.erase(hnd);
// Unmap, close ion handle and close fd
FreeBuffer(buf);
}
}
return GRALLOC1_ERROR_NONE;
}
gralloc1_error_t BufferManager::LockBuffer(const private_handle_t *hnd,
gralloc1_producer_usage_t prod_usage,
gralloc1_consumer_usage_t cons_usage) {
std::lock_guard<std::mutex> lock(buffer_lock_);
gralloc1_error_t err = GRALLOC1_ERROR_NONE;
ALOGD_IF(DEBUG, "LockBuffer buffer handle:%p id: %" PRIu64, hnd, hnd->id);
// If buffer is not meant for CPU return err
if (!CpuCanAccess(prod_usage, cons_usage)) {
return GRALLOC1_ERROR_BAD_VALUE;
}
auto buf = GetBufferFromHandleLocked(hnd);
if (buf == nullptr) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (hnd->base == 0) {
// we need to map for real
err = MapBuffer(hnd);
}
// Invalidate if CPU reads in software and there are non-CPU
// writers. No need to do this for the metadata buffer as it is
// only read/written in software.
// todo use handle here
if (!err && (hnd->flags & private_handle_t::PRIV_FLAGS_USES_ION) &&
(hnd->flags & private_handle_t::PRIV_FLAGS_CACHED)) {
if (allocator_->CleanBuffer(reinterpret_cast<void *>(hnd->base), hnd->size, hnd->offset,
buf->ion_handle_main, CACHE_INVALIDATE)) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
}
// Mark the buffer to be flushed after CPU write.
if (!err && CpuCanWrite(prod_usage)) {
private_handle_t *handle = const_cast<private_handle_t *>(hnd);
handle->flags |= private_handle_t::PRIV_FLAGS_NEEDS_FLUSH;
}
return err;
}
gralloc1_error_t BufferManager::UnlockBuffer(const private_handle_t *handle) {
std::lock_guard<std::mutex> lock(buffer_lock_);
gralloc1_error_t status = GRALLOC1_ERROR_NONE;
private_handle_t *hnd = const_cast<private_handle_t *>(handle);
auto buf = GetBufferFromHandleLocked(hnd);
if (buf == nullptr) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (hnd->flags & private_handle_t::PRIV_FLAGS_NEEDS_FLUSH) {
if (allocator_->CleanBuffer(reinterpret_cast<void *>(hnd->base), hnd->size, hnd->offset,
buf->ion_handle_main, CACHE_CLEAN) != 0) {
status = GRALLOC1_ERROR_BAD_HANDLE;
}
hnd->flags &= ~private_handle_t::PRIV_FLAGS_NEEDS_FLUSH;
}
return status;
}
uint32_t BufferManager::GetDataAlignment(int format, gralloc1_producer_usage_t prod_usage,
gralloc1_consumer_usage_t cons_usage) {
uint32_t align = UINT(getpagesize());
if (format == HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED) {
align = 8192;
}
if (prod_usage & GRALLOC1_PRODUCER_USAGE_PROTECTED) {
if ((prod_usage & GRALLOC1_PRODUCER_USAGE_CAMERA) ||
(cons_usage & GRALLOC1_CONSUMER_USAGE_PRIVATE_SECURE_DISPLAY)) {
// The alignment here reflects qsee mmu V7L/V8L requirement
align = SZ_2M;
} else {
align = SECURE_ALIGN;
}
}
return align;
}
int BufferManager::GetHandleFlags(int format, gralloc1_producer_usage_t prod_usage,
gralloc1_consumer_usage_t cons_usage) {
int flags = 0;
if (cons_usage & GRALLOC1_CONSUMER_USAGE_VIDEO_ENCODER) {
flags |= private_handle_t::PRIV_FLAGS_VIDEO_ENCODER;
}
if (prod_usage & GRALLOC1_PRODUCER_USAGE_CAMERA) {
flags |= private_handle_t::PRIV_FLAGS_CAMERA_WRITE;
}
if (prod_usage & GRALLOC1_CONSUMER_USAGE_CAMERA) {
flags |= private_handle_t::PRIV_FLAGS_CAMERA_READ;
}
if (cons_usage & GRALLOC1_CONSUMER_USAGE_HWCOMPOSER) {
flags |= private_handle_t::PRIV_FLAGS_HW_COMPOSER;
}
if (prod_usage & GRALLOC1_CONSUMER_USAGE_GPU_TEXTURE) {
flags |= private_handle_t::PRIV_FLAGS_HW_TEXTURE;
}
if (prod_usage & GRALLOC1_CONSUMER_USAGE_PRIVATE_SECURE_DISPLAY) {
flags |= private_handle_t::PRIV_FLAGS_SECURE_DISPLAY;
}
if (IsUBwcEnabled(format, prod_usage, cons_usage)) {
flags |= private_handle_t::PRIV_FLAGS_UBWC_ALIGNED;
}
if (prod_usage & (GRALLOC1_PRODUCER_USAGE_CPU_READ | GRALLOC1_PRODUCER_USAGE_CPU_WRITE)) {
flags |= private_handle_t::PRIV_FLAGS_CPU_RENDERED;
}
// TODO(user): is this correct???
if ((cons_usage &
(GRALLOC1_CONSUMER_USAGE_VIDEO_ENCODER | GRALLOC1_CONSUMER_USAGE_CLIENT_TARGET)) ||
(prod_usage & (GRALLOC1_PRODUCER_USAGE_CAMERA | GRALLOC1_PRODUCER_USAGE_GPU_RENDER_TARGET))) {
flags |= private_handle_t::PRIV_FLAGS_NON_CPU_WRITER;
}
if (cons_usage & GRALLOC1_CONSUMER_USAGE_HWCOMPOSER) {
flags |= private_handle_t::PRIV_FLAGS_DISP_CONSUMER;
}
if (!allocator_->UseUncached(prod_usage, cons_usage)) {
flags |= private_handle_t::PRIV_FLAGS_CACHED;
}
return flags;
}
int BufferManager::GetBufferType(int inputFormat) {
int buffer_type = BUFFER_TYPE_VIDEO;
if (IsUncompressedRGBFormat(inputFormat)) {
// RGB formats
buffer_type = BUFFER_TYPE_UI;
}
return buffer_type;
}
int BufferManager::AllocateBufferLocked(const BufferDescriptor &descriptor, buffer_handle_t *handle,
unsigned int bufferSize) {
if (!handle)
return -EINVAL;
gralloc1_producer_usage_t prod_usage = descriptor.GetProducerUsage();
gralloc1_consumer_usage_t cons_usage = descriptor.GetConsumerUsage();
int format = allocator_->GetImplDefinedFormat(prod_usage, cons_usage, descriptor.GetFormat());
uint32_t layer_count = descriptor.GetLayerCount();
unsigned int size;
unsigned int alignedw, alignedh;
int buffer_type = GetBufferType(format);
BufferInfo info = GetBufferInfo(descriptor);
GetBufferSizeAndDimensions(info, &size, &alignedw, &alignedh);
size = (bufferSize >= size) ? bufferSize : size;
int err = 0;
int flags = 0;
auto page_size = UINT(getpagesize());
AllocData data;
data.align = GetDataAlignment(format, prod_usage, cons_usage);
size = ALIGN(size, data.align) * layer_count;
data.size = size;
data.handle = (uintptr_t) handle;
data.uncached = allocator_->UseUncached(prod_usage, cons_usage);
// Allocate buffer memory
err = allocator_->AllocateMem(&data, prod_usage, cons_usage);
if (err) {
ALOGE("gralloc failed to allocate err=%s", strerror(-err));
return err;
}
// Allocate memory for MetaData
AllocData e_data;
e_data.size = ALIGN(UINT(sizeof(MetaData_t)), page_size);
e_data.handle = data.handle;
e_data.align = page_size;
err =
allocator_->AllocateMem(&e_data, GRALLOC1_PRODUCER_USAGE_NONE, GRALLOC1_CONSUMER_USAGE_NONE);
if (err) {
ALOGE("gralloc failed to allocate metadata error=%s", strerror(-err));
return err;
}
flags = GetHandleFlags(format, prod_usage, cons_usage);
flags |= data.alloc_type;
// Create handle
private_handle_t *hnd = new private_handle_t(data.fd,
e_data.fd,
flags,
INT(alignedw),
INT(alignedh),
descriptor.GetWidth(),
descriptor.GetHeight(),
format,
buffer_type,
data.size,
prod_usage,
cons_usage);
hnd->id = ++next_id_;
hnd->base = 0;
hnd->base_metadata = 0;
hnd->layer_count = layer_count;
ColorSpace_t colorSpace = ITU_R_601;
setMetaData(hnd, UPDATE_COLOR_SPACE, reinterpret_cast<void *>(&colorSpace));
*handle = hnd;
RegisterHandleLocked(hnd, data.ion_handle, e_data.ion_handle);
ALOGD_IF(DEBUG, "Allocated buffer handle: %p id: %" PRIu64, hnd, hnd->id);
if (DEBUG) {
private_handle_t::Dump(hnd);
}
return err;
}
gralloc1_error_t BufferManager::Perform(int operation, va_list args) {
switch (operation) {
case GRALLOC_MODULE_PERFORM_CREATE_HANDLE_FROM_BUFFER: {
int fd = va_arg(args, int);
unsigned int size = va_arg(args, unsigned int);
unsigned int offset = va_arg(args, unsigned int);
void *base = va_arg(args, void *);
int width = va_arg(args, int);
int height = va_arg(args, int);
int format = va_arg(args, int);
native_handle_t **handle = va_arg(args, native_handle_t **);
if (!handle) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
private_handle_t *hnd = reinterpret_cast<private_handle_t *>(
native_handle_create(private_handle_t::kNumFds, private_handle_t::NumInts()));
if (hnd) {
unsigned int alignedw = 0, alignedh = 0;
hnd->magic = private_handle_t::kMagic;
hnd->fd = fd;
hnd->flags = private_handle_t::PRIV_FLAGS_USES_ION;
hnd->size = size;
hnd->offset = offset;
hnd->base = uint64_t(base);
hnd->gpuaddr = 0;
BufferInfo info(width, height, format);
GetAlignedWidthAndHeight(info, &alignedw, &alignedh);
hnd->unaligned_width = width;
hnd->unaligned_height = height;
hnd->width = INT(alignedw);
hnd->height = INT(alignedh);
hnd->format = format;
*handle = reinterpret_cast<native_handle_t *>(hnd);
}
} break;
case GRALLOC_MODULE_PERFORM_GET_STRIDE: {
int width = va_arg(args, int);
int format = va_arg(args, int);
int *stride = va_arg(args, int *);
unsigned int alignedw = 0, alignedh = 0;
if (!stride) {
return GRALLOC1_ERROR_BAD_VALUE;
}
BufferInfo info(width, width, format);
GetAlignedWidthAndHeight(info, &alignedw, &alignedh);
*stride = INT(alignedw);
} break;
case GRALLOC_MODULE_PERFORM_GET_CUSTOM_STRIDE_FROM_HANDLE: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
int *stride = va_arg(args, int *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!stride) {
return GRALLOC1_ERROR_BAD_VALUE;
}
BufferDim_t buffer_dim;
if (getMetaData(hnd, GET_BUFFER_GEOMETRY, &buffer_dim) == 0) {
*stride = buffer_dim.sliceWidth;
} else {
*stride = hnd->width;
}
} break;
// TODO(user) : this alone should be sufficient, ask gfx to get rid of above
case GRALLOC_MODULE_PERFORM_GET_CUSTOM_STRIDE_AND_HEIGHT_FROM_HANDLE: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
int *stride = va_arg(args, int *);
int *height = va_arg(args, int *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!stride || !height) {
return GRALLOC1_ERROR_BAD_VALUE;
}
BufferDim_t buffer_dim;
int interlaced = 0;
*stride = hnd->width;
*height = hnd->height;
if (getMetaData(hnd, GET_BUFFER_GEOMETRY, &buffer_dim) == 0) {
*stride = buffer_dim.sliceWidth;
*height = buffer_dim.sliceHeight;
} else if (getMetaData(hnd, GET_PP_PARAM_INTERLACED, &interlaced) == 0) {
if (interlaced && IsUBwcFormat(hnd->format)) {
unsigned int alignedw = 0, alignedh = 0;
// Get re-aligned height for single ubwc interlaced field and
// multiple by 2 to get frame height.
BufferInfo info(hnd->width, ((hnd->height+1)>>1), hnd->format);
GetAlignedWidthAndHeight(info, &alignedw, &alignedh);
*stride = static_cast<int>(alignedw);
*height = static_cast<int>(alignedh * 2);
}
}
} break;
case GRALLOC_MODULE_PERFORM_GET_ATTRIBUTES: {
// TODO(user): Usage is split now. take care of it from Gfx client.
// see if we can directly expect descriptor from gfx client.
int width = va_arg(args, int);
int height = va_arg(args, int);
int format = va_arg(args, int);
uint64_t producer_usage = va_arg(args, uint64_t);
uint64_t consumer_usage = va_arg(args, uint64_t);
gralloc1_producer_usage_t prod_usage = static_cast<gralloc1_producer_usage_t>(producer_usage);
gralloc1_consumer_usage_t cons_usage = static_cast<gralloc1_consumer_usage_t>(consumer_usage);
int *aligned_width = va_arg(args, int *);
int *aligned_height = va_arg(args, int *);
int *tile_enabled = va_arg(args, int *);
if (!aligned_width || !aligned_height || !tile_enabled) {
return GRALLOC1_ERROR_BAD_VALUE;
}
unsigned int alignedw, alignedh;
BufferInfo info(width, height, format, prod_usage, cons_usage);
*tile_enabled = IsUBwcEnabled(format, prod_usage, cons_usage);
GetAlignedWidthAndHeight(info, &alignedw, &alignedh);
*aligned_width = INT(alignedw);
*aligned_height = INT(alignedh);
} break;
case GRALLOC_MODULE_PERFORM_GET_COLOR_SPACE_FROM_HANDLE: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
int *color_space = va_arg(args, int *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!color_space) {
return GRALLOC1_ERROR_BAD_VALUE;
}
*color_space = 0;
#ifdef USE_COLOR_METADATA
ColorMetaData color_metadata;
if (getMetaData(hnd, GET_COLOR_METADATA, &color_metadata) == 0) {
switch (color_metadata.colorPrimaries) {
case ColorPrimaries_BT709_5:
*color_space = HAL_CSC_ITU_R_709;
break;
case ColorPrimaries_BT601_6_525:
case ColorPrimaries_BT601_6_625:
*color_space = ((color_metadata.range) ? HAL_CSC_ITU_R_601_FR : HAL_CSC_ITU_R_601);
break;
case ColorPrimaries_BT2020:
*color_space = (color_metadata.range) ? HAL_CSC_ITU_R_2020_FR : HAL_CSC_ITU_R_2020;
break;
default:
ALOGE("Unknown Color Space = %d", color_metadata.colorPrimaries);
break;
}
break;
}
if (getMetaData(hnd, GET_COLOR_SPACE, &color_metadata) != 0) {
*color_space = 0;
}
#endif
} break;
case GRALLOC_MODULE_PERFORM_GET_YUV_PLANE_INFO: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
android_ycbcr *ycbcr = va_arg(args, struct android_ycbcr *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!ycbcr) {
return GRALLOC1_ERROR_BAD_VALUE;
}
if (GetYUVPlaneInfo(hnd, ycbcr)) {
return GRALLOC1_ERROR_UNDEFINED;
}
} break;
case GRALLOC_MODULE_PERFORM_GET_MAP_SECURE_BUFFER_INFO: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
int *map_secure_buffer = va_arg(args, int *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!map_secure_buffer) {
return GRALLOC1_ERROR_BAD_VALUE;
}
if (getMetaData(hnd, GET_MAP_SECURE_BUFFER, map_secure_buffer) != 0) {
*map_secure_buffer = 0;
}
} break;
case GRALLOC_MODULE_PERFORM_GET_UBWC_FLAG: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
int *flag = va_arg(args, int *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!flag) {
return GRALLOC1_ERROR_BAD_VALUE;
}
*flag = hnd->flags &private_handle_t::PRIV_FLAGS_UBWC_ALIGNED;
int linear_format = 0;
if (getMetaData(hnd, GET_LINEAR_FORMAT, &linear_format) == 0) {
if (linear_format) {
*flag = 0;
}
}
} break;
case GRALLOC_MODULE_PERFORM_GET_RGB_DATA_ADDRESS: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
void **rgb_data = va_arg(args, void **);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!rgb_data) {
return GRALLOC1_ERROR_BAD_VALUE;
}
if (GetRgbDataAddress(hnd, rgb_data)) {
return GRALLOC1_ERROR_UNDEFINED;
}
} break;
case GRALLOC1_MODULE_PERFORM_GET_BUFFER_SIZE_AND_DIMENSIONS: {
int width = va_arg(args, int);
int height = va_arg(args, int);
int format = va_arg(args, int);
uint64_t p_usage = va_arg(args, uint64_t);
uint64_t c_usage = va_arg(args, uint64_t);
gralloc1_producer_usage_t producer_usage = static_cast<gralloc1_producer_usage_t>(p_usage);
gralloc1_consumer_usage_t consumer_usage = static_cast<gralloc1_consumer_usage_t>(c_usage);
uint32_t *aligned_width = va_arg(args, uint32_t *);
uint32_t *aligned_height = va_arg(args, uint32_t *);
uint32_t *size = va_arg(args, uint32_t *);
if (!aligned_width || !aligned_height || !size) {
return GRALLOC1_ERROR_BAD_VALUE;
}
auto info = BufferInfo(width, height, format, producer_usage, consumer_usage);
GetBufferSizeAndDimensions(info, size, aligned_width, aligned_height);
// Align size
auto align = GetDataAlignment(format, producer_usage, consumer_usage);
*size = ALIGN(*size, align);
} break;
case GRALLOC1_MODULE_PERFORM_GET_INTERLACE_FLAG: {
private_handle_t *hnd = va_arg(args, private_handle_t *);
int *flag = va_arg(args, int *);
if (private_handle_t::validate(hnd) != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
if (!flag) {
return GRALLOC1_ERROR_BAD_VALUE;
}
if (getMetaData(hnd, GET_PP_PARAM_INTERLACED, flag) != 0) {
*flag = 0;
}
} break;
default:
break;
}
return GRALLOC1_ERROR_NONE;
}
static bool IsYuvFormat(const private_handle_t *hnd) {
switch (hnd->format) {
case HAL_PIXEL_FORMAT_YCbCr_420_SP:
case HAL_PIXEL_FORMAT_YCbCr_422_SP:
case HAL_PIXEL_FORMAT_YCbCr_420_SP_VENUS:
case HAL_PIXEL_FORMAT_NV12_ENCODEABLE: // Same as YCbCr_420_SP_VENUS
case HAL_PIXEL_FORMAT_YCbCr_420_SP_VENUS_UBWC:
case HAL_PIXEL_FORMAT_YCrCb_420_SP:
case HAL_PIXEL_FORMAT_YCrCb_422_SP:
case HAL_PIXEL_FORMAT_YCrCb_420_SP_ADRENO:
case HAL_PIXEL_FORMAT_NV21_ZSL:
case HAL_PIXEL_FORMAT_RAW16:
case HAL_PIXEL_FORMAT_Y16:
case HAL_PIXEL_FORMAT_RAW12:
case HAL_PIXEL_FORMAT_RAW10:
case HAL_PIXEL_FORMAT_YV12:
case HAL_PIXEL_FORMAT_Y8:
case HAL_PIXEL_FORMAT_YCbCr_420_P010:
case HAL_PIXEL_FORMAT_YCbCr_420_TP10_UBWC:
case HAL_PIXEL_FORMAT_YCbCr_420_P010_UBWC:
return true;
default:
return false;
}
}
gralloc1_error_t BufferManager::GetNumFlexPlanes(const private_handle_t *hnd,
uint32_t *out_num_planes) {
if (!IsYuvFormat(hnd)) {
return GRALLOC1_ERROR_UNSUPPORTED;
} else {
*out_num_planes = 3;
}
return GRALLOC1_ERROR_NONE;
}
gralloc1_error_t BufferManager::GetFlexLayout(const private_handle_t *hnd,
struct android_flex_layout *layout) {
if (!IsYuvFormat(hnd)) {
return GRALLOC1_ERROR_UNSUPPORTED;
}
android_ycbcr yuvPlaneInfo[2];
int err = GetYUVPlaneInfo(hnd, yuvPlaneInfo);
if (err != 0) {
return GRALLOC1_ERROR_BAD_HANDLE;
}
layout->format = FLEX_FORMAT_YCbCr;
layout->num_planes = 3;
for (uint32_t i = 0; i < layout->num_planes; i++) {
layout->planes[i].bits_per_component = 8;
layout->planes[i].bits_used = 8;
layout->planes[i].h_increment = 1;
layout->planes[i].v_increment = 1;
layout->planes[i].h_subsampling = 2;
layout->planes[i].v_subsampling = 2;
}
// We are only returning flex layout for progressive or single field formats.
struct android_ycbcr ycbcr = yuvPlaneInfo[0];
layout->planes[0].top_left = static_cast<uint8_t *>(ycbcr.y);
layout->planes[0].component = FLEX_COMPONENT_Y;
layout->planes[0].v_increment = static_cast<int32_t>(ycbcr.ystride);
layout->planes[1].top_left = static_cast<uint8_t *>(ycbcr.cb);
layout->planes[1].component = FLEX_COMPONENT_Cb;
layout->planes[1].h_increment = static_cast<int32_t>(ycbcr.chroma_step);
layout->planes[1].v_increment = static_cast<int32_t>(ycbcr.cstride);
layout->planes[2].top_left = static_cast<uint8_t *>(ycbcr.cr);
layout->planes[2].component = FLEX_COMPONENT_Cr;
layout->planes[2].h_increment = static_cast<int32_t>(ycbcr.chroma_step);
layout->planes[2].v_increment = static_cast<int32_t>(ycbcr.cstride);
return GRALLOC1_ERROR_NONE;
}
gralloc1_error_t BufferManager::Dump(std::ostringstream *os) {
for (auto it : handles_map_) {
auto buf = it.second;
auto hnd = buf->handle;
*os << "handle id: " << std::setw(4) << hnd->id;
*os << " fd: " << std::setw(3) << hnd->fd;
*os << " fd_meta: " << std::setw(3) << hnd->fd_metadata;
*os << " wxh: " << std::setw(4) << hnd->width <<" x " << std::setw(4) << hnd->height;
*os << " uwxuh: " << std::setw(4) << hnd->unaligned_width << " x ";
*os << std::setw(4) << hnd->unaligned_height;
*os << " size: " << std::setw(9) << hnd->size;
*os << std::hex << std::setfill('0');
*os << " priv_flags: " << "0x" << std::setw(8) << hnd->flags;
*os << " prod_usage: " << "0x" << std::setw(8) << hnd->producer_usage;
*os << " cons_usage: " << "0x" << std::setw(8) << hnd->consumer_usage;
// TODO(user): get format string from qdutils
*os << " format: " << "0x" << std::setw(8) << hnd->format;
*os << std::dec << std::setfill(' ') << std::endl;
}
return GRALLOC1_ERROR_NONE;
}
} // namespace gralloc1