Android6.0 显示系统（四） 图像显示相关

    Linux通常使用Framebuffer来用作显示输出，Framebuffer就是一块内存区域，它通常是显示驱动的内部缓冲区在内存中的映射。一旦用户进程把图像数据复制到Framebuffer中，显示驱动会一个像素一个像素地扫描整个Framebuffer，并根据其中的值更新屏幕上像素点的颜色。驱动中这种更新屏幕的动作是固定的，它的周期就是我们常说的刷新率。

    但是在屏幕更新一半时，用户进程更新了Framebuffer中的数据，将导致屏幕上画面的上半部分是前一帧的画面，下半部分变成了新的画面。当然错误会在下次刷新时纠正过来，但是这样也会有闪烁的感觉。这个可以使用双缓冲机制，双缓冲就是提供两块Framebuffer，一块用于显示，一块用于数据更新。数据准备好后，通过ioctl操作告诉显示设备切换用于显示的FrameBuffer，这样图像就能快速的显示出来。

    但是双缓冲并没有完全解决问题，虽然双缓冲切换的速度很快，但是如果切换的时间点不对，在画面更新一半的时候切换，还是会出现闪烁的问题。当然，我们可以在底层进行控制，收到切换请求的时候，内部并不马上执行，等到刷新完成后再切换，这样完全避免了画面重叠问题。但是这样也有问题，如果用ioctl操作告诉底层可以进行切换了，但是缓冲区没有切换，这样应用层就不能确定何时可以再使用缓冲区，因此只能不断的通过ioctl来查询缓冲区的状态，一直到切换完成了。这种方式效率太低，拖慢了整个系统。解决这个问题就是底层固定发送信号给用户进程，通知进程切换的时机。这个信号就是VSync信号。

VSync信号是一个硬件信号，一般是显示设备刷新的周期到了会发送。

一、VSync信号的产生

Android通过VSync机制来提高显示效果，那么VSync是如何产生的？通常这个信号是由显示驱动产生，这样才能达到最佳效果。但是Android为了能运行在不支持VSync机制的设备上，也提供了软件模拟产生VSync信号的手段。

SurfaceFlinger中用HWComposer类来表示硬件显示设备，

HWComposer::HWComposer(
const sp<SurfaceFlinger>& flinger,
EventHandler& handler)
: mFlinger(flinger),
mFbDev(0), mHwc(0), mNumDisplays(1),
mCBContext(new cb_context),
mEventHandler(handler),
mDebugForceFakeVSync(false)
{
......
bool needVSyncThread = true;

// Note: some devices may insist that the FB HAL be opened before HWC.
int fberr = loadFbHalModule();//装载FrameBuffer的硬件模块
loadHwcModule();//装载HWComposer的硬件模块,这个函数中会将mHwc置为true

......
if (mHwc) {//这个为true代表硬件设备打开了
ALOGI("Using %s version %u.%u", HWC_HARDWARE_COMPOSER,
(hwcApiVersion(mHwc) >> 24) & 0xff,
(hwcApiVersion(mHwc) >> 16) & 0xff);
if (mHwc->registerProcs) {
mCBContext->hwc = this;
mCBContext->procs.invalidate = &hook_invalidate;
mCBContext->procs.vsync = &hook_vsync;//vsync回调函数
if (hwcHasApiVersion(mHwc, HWC_DEVICE_API_VERSION_1_1))
mCBContext->procs.hotplug = &hook_hotplug;
else
mCBContext->procs.hotplug = NULL;
memset(mCBContext->procs.zero, 0, sizeof(mCBContext->procs.zero));
mHwc->registerProcs(mHwc, &mCBContext->procs);
}

// don't need a vsync thread if we have a hardware composer
needVSyncThread = false;//打开硬件设备成功了，将needVSncThread为false
......
}
......

if (needVSyncThread) {
// we don't have VSYNC support, we need to fake it
mVSyncThread = new VSyncThread(*this);
}
}

通过loadHwcModule来装载硬件模块，如果成功，mHwc为true，needVSyncThread为false。如果不成功，needVsyncThread为true，然后就要创建VSyncThread对象了，它就是产生VSync信号的软件手段了。

VSyncThread是一个thread，在onFirstRef中会调用run函数，就是执行threadLoop，这个函数只要返回true就会一直执行。

bool HWComposer::VSyncThread::threadLoop() {
{ // scope for lock
Mutex::Autolock _l(mLock);
while (!mEnabled) {
mCondition.wait(mLock);
}
}

const nsecs_t period = mRefreshPeriod;
const nsecs_t now = systemTime(CLOCK_MONOTONIC);
nsecs_t next_vsync = mNextFakeVSync;
nsecs_t sleep = next_vsync - now;
if (sleep < 0) {
// we missed, find where the next vsync should be
sleep = (period - ((now - next_vsync) % period));
next_vsync = now + sleep;
}
mNextFakeVSync = next_vsync + period;

struct timespec spec;
spec.tv_sec  = next_vsync / 1000000000;
spec.tv_nsec = next_vsync % 1000000000;

int err;
do {
err = clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, &spec, NULL);
} while (err<0 && errno == EINTR);

if (err == 0) {
mHwc.mEventHandler.onVSyncReceived(0, next_vsync);
}

return true;
}

这个函数会间隔模拟产生VSync的信号的原理是在固定时间发送消息给HWCompoer的消息对象mEventHandler，这个其实就到SurfaceFlinger的onVSyncReceived函数了。用软件模拟VSync信号在系统比较忙的时候可能会丢失一些信号。

Android源码再hardware/lib/libhardware/modules下有一个hwcomposer目录，里面是一个Android提供的缺省的硬件HWComposer模块的例子，这个例子只实现了一个open接口，并不能真正工作。在前面HWComposer的构造函数中，有如下代码

mCBContext->procs.vsync = &hook_vsync;

这里指定了vsync的回调函数是hook_vsync，如果硬件中产生了VSync信号，将通过这个函数来通知上层，看看它的代码：

void HWComposer::hook_vsync(const struct hwc_procs* procs, int disp,
int64_t timestamp) {
cb_context* ctx = reinterpret_cast<cb_context*>(
const_cast<hwc_procs_t*>(procs));
ctx->hwc->vsync(disp, timestamp);
}

然后又调用了vsync函数，这个函数最后也是调用了mEventHandler.onVSyncReceived函数，这个函数最后回到SurfaceFlinger中的onVsyncReceived函数中。

void HWComposer::vsync(int disp, int64_t timestamp) {
if (uint32_t(disp) < HWC_NUM_PHYSICAL_DISPLAY_TYPES) {
{
Mutex::Autolock _l(mLock);

// There have been reports of HWCs that signal several vsync events
// with the same timestamp when turning the display off and on. This
// is a bug in the HWC implementation, but filter the extra events
// out here so they don't cause havoc downstream.
if (timestamp == mLastHwVSync[disp]) {
ALOGW("Ignoring duplicate VSYNC event from HWC (t=%" PRId64 ")",
timestamp);
return;
}

mLastHwVSync[disp] = timestamp;
}

char tag[16];
snprintf(tag, sizeof(tag), "HW_VSYNC_%1u", disp);
ATRACE_INT(tag, ++mVSyncCounts[disp] & 1);

mEventHandler.onVSyncReceived(disp, timestamp);
}
}

二、FrameBuffer工作原理

我们先来看下loadFbHalModule函数，hw_get_module是HAl框架中装载HAL模块的函数

int HWComposer::loadFbHalModule()
{
hw_module_t const* module;

int err = hw_get_module(GRALLOC_HARDWARE_MODULE_ID, &module);
if (err != 0) {
ALOGE("%s module not found", GRALLOC_HARDWARE_MODULE_ID);
return err;
}

return framebuffer_open(module, &mFbDev);
}

我们再来看看framebuffer_open函数，

static inline int framebuffer_open(const struct hw_module_t* module,
struct framebuffer_device_t** device) {
return module->methods->open(module,
GRALLOC_HARDWARE_FB0, (struct hw_device_t**)device);
}

GRALLOC_HARDWARE_FB0 就是fb0

#define GRALLOC_HARDWARE_FB0 "fb0"

Gralloc模块在实际设备中有硬件厂商提供。我们来看下这个open函数

static int gralloc_device_open(const hw_module_t* module, const char* name, hw_device_t** device)
{
int status = -EINVAL;

if (!strncmp(name, GRALLOC_HARDWARE_GPU0, MALI_GRALLOC_HARDWARE_MAX_STR_LEN))
{
status = alloc_device_open(module, name, device);//处理gpu的
}
else if (!strncmp(name, GRALLOC_HARDWARE_FB0, MALI_GRALLOC_HARDWARE_MAX_STR_LEN))
{
status = framebuffer_device_open(module, name, device);
}

return status;
}

我们来看framebuffer_device_open函数，如果不支持framebuffer直接退出了（现在很多设备都开始不支持了）。如果支持framebuffer的话先是调用了init_frame_buffer函数来获取设备信息，通过mmap分配一块共享内存，然后设置FrameBuffer的操作函数等。

int framebuffer_device_open(hw_module_t const* module, const char* name, hw_device_t** device)
{
int status = -EINVAL;

log_fbpost = false;
char property[PROPERTY_VALUE_MAX];
if(property_get("debug.gralloc.fbpost", property, "0") > 0) {
if(atoi(property) == 1) {
log_fbpost = true;
ALOGI("enable fbpost log!");
}
}

alloc_device_t* gralloc_device;
#if DISABLE_FRAMEBUFFER_HAL == 1 //不支持FrameBuffer
AERR("Framebuffer HAL not support/disabled %s",
#ifdef MALI_DISPLAY_VERSION
"with MALI display enable");
#else
"");
#endif
return -ENODEV;
#endif
status = gralloc_open(module, &gralloc_device);
if (status < 0)
{
return status;
}

private_module_t* m = (private_module_t*)module;
status = init_frame_buffer(m);

framebuffer_device_t *dev =  reinterpret_cast<framebuffer_device_t*> (malloc(sizeof(framebuffer_device_t)));

/* if either or both of init_frame_buffer() and malloc failed */
if ((status < 0) || (!dev))
{
gralloc_close(gralloc_device);
(!dev) ? (void)(status = -ENOMEM) : free(dev);
return status;
}

memset(dev, 0, sizeof(*dev));

//设置framebuffer的操作函数
dev->common.tag = HARDWARE_DEVICE_TAG;
dev->common.version = 0;
dev->common.module = const_cast<hw_module_t*>(module);
dev->common.close = fb_close;
dev->setSwapInterval = fb_set_swap_interval;
dev->post = fb_post;
dev->enableScreen = fb_enable_screen;
dev->setUpdateRect = 0;
dev->compositionComplete = &compositionComplete;

int stride = m->finfo.line_length / (m->info.bits_per_pixel >> 3);
const_cast<uint32_t&>(dev->flags) = 0;
const_cast<uint32_t&>(dev->width) = m->info.xres;
const_cast<uint32_t&>(dev->height) = m->info.yres;
const_cast<int&>(dev->stride) = stride;
const_cast<int&>(dev->format) = m->fbFormat;
const_cast<float&>(dev->xdpi) = m->xdpi;
const_cast<float&>(dev->ydpi) = m->ydpi;
const_cast<float&>(dev->fps) = m->fps;
const_cast<int&>(dev->minSwapInterval) = 0;
const_cast<int&>(dev->maxSwapInterval) = 1;
const_cast<int&>(dev->numFramebuffers) = m->numBuffers;
*device = &dev->common;

AINF("%s line %d format %d numBuffers %d",__FUNCTION__,__LINE__, dev->format, m->numBuffers);

//init dynamic lcd fps adjustment
dyn_fps_init(m);
#if GRALLOC_VSYNC_NEEDED == 1
gralloc_vsync_enable(dev);//支持vsync
#endif

gralloc_close(gralloc_device);

return status;
}

init_frame_buffer函数主要调用了init_frame_buffer_locked函数

static int init_frame_buffer(struct private_module_t* module)
{
pthread_mutex_lock(&module->lock);
int err = init_frame_buffer_locked(module);
pthread_mutex_unlock(&module->lock);
return err;
}

我们来看看init_frame_buffer_locked函数，先打开设备列表中的一个设备即可，然后通过ioctl获取设备信息，把设备信息放到module中，后面通过mmap分配一块共享内存。

int init_frame_buffer_locked(struct private_module_t* module)
{
if (module->framebuffer)
{
return 0; // Nothing to do, already initialized
}

char const * const device_template[] =//设备列表
{
"/dev/graphics/fb%u",
"/dev/fb%u",
NULL
};

int fd = -1;
int i = 0;
char name[64];

while ((fd == -1) && device_template[i])//只要打开一个设备就好了
{
snprintf(name, 64, device_template[i], 0);
fd = open(name, O_RDWR, 0);
i++;
}

if (fd < 0)
{
return -errno;
}

struct fb_fix_screeninfo finfo;
if (ioctl(fd, FBIOGET_FSCREENINFO, &finfo) == -1)
{
return -errno;
}

struct fb_var_screeninfo info;
if (ioctl(fd, FBIOGET_VSCREENINFO, &info) == -1)
{
return -errno;
}

info.reserved[0] = 0;
info.reserved[1] = 0;
info.reserved[2] = 0;
info.xoffset = 0;
info.yoffset = 0;
info.activate = FB_ACTIVATE_NOW;

if(info.bits_per_pixel == 32)
{
/*
* Explicitly request 8/8/8
*/
info.bits_per_pixel = 32;
info.red.offset     = 16;
info.red.length     = 8;
info.green.offset   = 8;
info.green.length   = 8;
info.blue.offset    = 0;
info.blue.length    = 8;
info.transp.offset  = 24;
info.transp.length  = 8;
}
else
{
/*
* Explicitly request 5/6/5
*/
info.bits_per_pixel = 16;
info.red.offset     = 11;
info.red.length     = 5;
info.green.offset   = 5;
info.green.length   = 6;
info.blue.offset    = 0;
info.blue.length    = 5;
info.transp.offset  = 0;
info.transp.length  = 0;
}

/*
* Request NUM_BUFFERS screens (at lest 2 for page flipping)
*/
info.yres_virtual = info.yres * NUM_BUFFERS;

uint32_t flags = PAGE_FLIP;
if (ioctl(fd, FBIOPUT_VSCREENINFO, &info) == -1)
{
info.yres_virtual = info.yres;
flags &= ~PAGE_FLIP;
AWAR( "FBIOPUT_VSCREENINFO failed, page flipping not supported fd: %d", fd );
}

if (info.yres_virtual < info.yres * 2)
{
// we need at least 2 for page-flipping
info.yres_virtual = info.yres;
flags &= ~PAGE_FLIP;
AWAR( "page flipping not supported (yres_virtual=%d, requested=%d)", info.yres_virtual, info.yres*2 );
}

if (ioctl(fd, FBIOGET_VSCREENINFO, &info) == -1)
{
return -errno;
}

int refreshRate = 0;
if ( info.pixclock > 0 )
{
refreshRate = 1000000000000000LLU /
(
uint64_t( info.upper_margin + info.lower_margin + info.yres + info.hsync_len )
* ( info.left_margin  + info.right_margin + info.xres + info.vsync_len )
* info.pixclock
);
}
else
{
AWAR( "fbdev pixclock is zero for fd: %d", fd );
}

if (refreshRate == 0)
{
refreshRate = 60*1000;  // 60 Hz
}

if (int(info.width) <= 0 || int(info.height) <= 0)
{
// the driver doesn't return that information
// default to 320 dpi
// debugging stuff...
char value[PROPERTY_VALUE_MAX];
int lcd_density;

property_get("ro.sf.lcd_density", value, "320");
lcd_density = atoi(value);

info.width  = ((info.xres * 25.4f) / (float)lcd_density + 0.5f);
info.height = ((info.yres * 25.4f) / (float)lcd_density + 0.5f);
}

float xdpi = (info.xres * 25.4f) / info.width;
float ydpi = (info.yres * 25.4f) / info.height;
float fps  = refreshRate / 1000.0f;

AINF("leadcore fb using (fd=%d)\n"
"id           = %s\n"
"xres         = %d px\n"
"yres         = %d px\n"
"xres_virtual = %d px\n"
"yres_virtual = %d px\n"
"bpp          = %d\n"
"r            = %2u:%u\n"
"g            = %2u:%u\n"
"b            = %2u:%u\n",
fd,
finfo.id,
info.xres,
info.yres,
info.xres_virtual,
info.yres_virtual,
info.bits_per_pixel,
info.red.offset, info.red.length,
info.green.offset, info.green.length,
info.blue.offset, info.blue.length);

AINF("width        = %d mm (%f dpi)\n"
"height       = %d mm (%f dpi)\n"
"refresh rate = %.2f Hz\n",
info.width,  xdpi,
info.height, ydpi,
fps);

if (0 == strncmp(finfo.id, "CLCD FB", 7))
{
module->dpy_type = MALI_DPY_TYPE_CLCD;
}
else if (0 == strncmp(finfo.id, "ARM Mali HDLCD", 14))
{
module->dpy_type = MALI_DPY_TYPE_HDLCD;
}
else
{
module->dpy_type = MALI_DPY_TYPE_UNKNOWN;
}

if (ioctl(fd, FBIOGET_FSCREENINFO, &finfo) == -1)
{
return -errno;
}

if (finfo.smem_len <= 0)
{
return -errno;
}

if(	info.bits_per_pixel == 32 &&
info.red.offset == 16 &&
info.red.length == 8 &&
info.green.offset == 8 &&
info.green.length == 8 &&
info.blue.offset == 0 &&
info.blue.length == 8)
{
module->fbFormat = HAL_PIXEL_FORMAT_BGRA_8888;
}

if(	info.bits_per_pixel == 32 &&
info.red.offset == 0 &&
info.red.length == 8 &&
info.green.offset == 8 &&
info.green.length == 8 &&
info.blue.offset == 16 &&
info.blue.length == 8)
{
module->fbFormat = HAL_PIXEL_FORMAT_RGBA_8888;
}

if(	info.bits_per_pixel == 16 &&
info.red.offset == 0 &&
info.red.length == 5 &&
info.green.offset == 5 &&
info.green.length == 6 &&
info.blue.offset == 11 &&
info.blue.length == 5)
{
module->fbFormat = HAL_PIXEL_FORMAT_RGB_565;
}

module->flags = flags;//设置信息
module->info = info;
module->finfo = finfo;
module->xdpi = xdpi;
module->ydpi = ydpi;
module->fps = fps;
module->swapInterval = 1;

/*
* map the framebuffer
*/
size_t fbSize = round_up_to_page_size(finfo.line_length * info.yres_virtual);
void* vaddr = mmap(0, fbSize, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);//mmap分配一块共享内存
if (vaddr == MAP_FAILED)
{
AERR( "Error mapping the framebuffer (%s)", strerror(errno) );
return -errno;
}

//fix black screen between uboot logo and bootanimation
//memset(vaddr, 0, fbSize);

// Create a "fake" buffer object for the entire frame buffer memory, and store it in the module
module->framebuffer = new private_handle_t(private_handle_t::PRIV_FLAGS_FRAMEBUFFER, GRALLOC_USAGE_HW_FB, fbSize, vaddr,
0, dup(fd), 0, 0);

module->numBuffers = info.yres_virtual / info.yres;
module->bufferMask = 0;

return 0;
}

最后我们再来看看framebuffer的操作函数fb_post，这个函数根据PRIV_FLAGS_FRAMEBUFFER来判断Framebuffer是否支持多缓冲，如果不支持方法很简单，直接把buffer中的数据复制到Framebuffer中就可以了。

Filp是指使用ioctl的FBIOPUT_VSCREENINFO参数设置当前显示的buffer。通过将显示区域指向Framebuffer中的新的数据帧，能非常迅速地完成buffer的切换。单缓冲模式下数据复制到缓冲区还需要一定时间，会加重闪烁感，通过Filp的方式切换缓冲区就不存在这个问题了。

static int fb_post(struct framebuffer_device_t* dev, buffer_handle_t buffer)
{
if (private_handle_t::validate(buffer) < 0)
{
return -EINVAL;
}

private_handle_t const* hnd = reinterpret_cast<private_handle_t const*>(buffer);
private_module_t* m = reinterpret_cast<private_module_t*>(dev->common.module);

if (m->currentBuffer)
{
m->base.unlock(&m->base, m->currentBuffer);
m->currentBuffer = 0;
}

struct timeval tv1, tv2;

if (hnd->flags & private_handle_t::PRIV_FLAGS_FRAMEBUFFER) //framebuffer是否支持多缓冲区（flip）
{
m->base.lock(&m->base, buffer, private_module_t::PRIV_USAGE_LOCKED_FOR_POST,
0, 0, m->info.xres, m->info.yres, NULL);

const size_t offset = (uintptr_t)hnd->base - (uintptr_t)m->framebuffer->base;
int interrupt;
m->info.activate = FB_ACTIVATE_VBL;
m->info.yoffset = offset / m->finfo.line_length;

up_fps(m);

gettimeofday(&tv1, NULL);
if (ioctl(m->framebuffer->fd, FBIOPUT_VSCREENINFO, &m->info) == -1)
{
AERR( "FBIOPUT_VSCREENINFO failed for fd: %d", m->framebuffer->fd );
m->base.unlock(&m->base, buffer);
return -errno;
}
#if GRALLOC_VSYNC_NEEDED
if ( 0 != gralloc_wait_for_vsync(dev) )
{
AERR( "Gralloc wait for vsync failed for fd: %d", m->framebuffer->fd );
m->base.unlock(&m->base, buffer);
return -errno;
}
#endif
gettimeofday(&tv2, NULL);

if((int64_t)tv2.tv_sec * 1000 + tv2.tv_usec/1000 - (int64_t)tv1.tv_sec * 1000 - tv1.tv_usec/1000 > 50)
{
ALOGI("%s line %d FBIOPUT_VSCREENINFO buffer %p blocktime=%lldms now=%lldms lasttime=%lld tid=%d",__FUNCTION__,__LINE__,
buffer,
(int64_t)tv2.tv_sec * 1000 + tv2.tv_usec/1000 - (int64_t)tv1.tv_sec * 1000 - tv1.tv_usec/1000,
systemTime(CLOCK_MONOTONIC)/1000000,
(int64_t)tv2.tv_sec * 1000 + tv2.tv_usec/1000 - lasttime,
gettid());
} else {
ALOGD_IF(log_fbpost, "%s line %d FBIOPUT_VSCREENINFO buffer %p blocktime=%lldms now=%lldms lasttime=%lld tid=%d",__FUNCTION__,__LINE__,
buffer,
(int64_t)tv2.tv_sec * 1000 + tv2.tv_usec/1000 - (int64_t)tv1.tv_sec * 1000 - tv1.tv_usec/1000,
systemTime(CLOCK_MONOTONIC)/1000000,
(int64_t)tv2.tv_sec * 1000 + tv2.tv_usec/1000 - lasttime,
gettid());
}

lasttime = (int64_t)tv2.tv_sec * 1000 + tv2.tv_usec/1000;

postcount++;
if(postcount == 1000)
{
ALOGI("%s line %d avgframerate = %2f",__FUNCTION__,__LINE__,
(float)postcount*1000.0/(lasttime - timecount));
postcount = 0;
timecount = lasttime;
}

m->currentBuffer = buffer;
} else {//不支持多缓冲（flip）
void* fb_vaddr;
void* buffer_vaddr;

m->base.lock(&m->base, m->framebuffer, GRALLOC_USAGE_SW_WRITE_RARELY,
0, 0, m->info.xres, m->info.yres, &fb_vaddr);

m->base.lock(&m->base, buffer, GRALLOC_USAGE_SW_READ_RARELY,
0, 0, m->info.xres, m->info.yres, &buffer_vaddr);

// If buffer's alignment match framebuffer alignment we can do a direct copy.
// If not we must fallback to do an aligned copy of each line.
if ( hnd->byte_stride == (int)m->finfo.line_length )
{
memcpy(fb_vaddr, buffer_vaddr, m->finfo.line_length * m->info.yres);
}
else
{
uintptr_t fb_offset = 0;
uintptr_t buffer_offset = 0;
unsigned int i;

for (i = 0; i < m->info.yres; i++)
{
memcpy((void *)((uintptr_t)fb_vaddr + fb_offset),
(void *)((uintptr_t)buffer_vaddr + buffer_offset),
m->finfo.line_length);

fb_offset += m->finfo.line_length;
buffer_offset += hnd->byte_stride;
}
}
m->base.unlock(&m->base, buffer);
m->base.unlock(&m->base, m->framebuffer);
}

return 0;
}

三、分配图像缓冲区的内存

之前的博客在GraphicBufferAllocator的alloc方法是调用了mAllocDev的alloc，而这个mAllocDev也是Gralloc模块，最后会调用如下方法。

int gralloc_alloc(struct alloc_device_t* dev,
int w, int h, int format, int usage,
buffer_handle_t* pHandle, int* pStride)
{
int rel;

/* TODO: Redirect to specific allocator according to usage. */
if (usage & GRALLOC_USAGE_HW_FB) {
/* Dispatch to framebuffer allocator. */
rel = gralloc_alloc_framebuffer(dev,
w, h, format, usage, pHandle, pStride);
} else {
rel = gc_gralloc_alloc(dev, w, h, format, usage, pHandle, pStride);
}

return rel;
}

gralloc_alloc函数会根据usage中的标志是否有GRALLOC_USAGE_HW_FB来决定是从硬件缓冲中分配缓冲区还是从普通内存分配缓冲区。我们先看看从内存中分配缓冲区的。

如果不支持framebuffer，只能从普通内存中分配缓冲区。gc_gralloc_alloc函数是从普通内存中分配缓冲区，主要使用了匿名共享内存的方法。最后pHandle装了共享内存的fd。

另外从硬件缓冲区分配内存是调用了gralloc_alloc_framebuffer方法，这个函数主要调用了gralloc_alloc_framebuffer_locked方法

static int gralloc_alloc_framebuffer_locked(alloc_device_t* dev, size_t size, int usage, buffer_handle_t* pHandle, int* stride, int* byte_stride)
{
private_module_t* m = reinterpret_cast<private_module_t*>(dev->common.module);

// allocate the framebuffer
if (m->framebuffer == NULL)//如果为空
{
// initialize the framebuffer, the framebuffer is mapped once and forever.
int err = init_frame_buffer_locked(m);//分配一大块内存
if (err < 0)
{
return err;
}
}

const uint32_t bufferMask = m->bufferMask;
const uint32_t numBuffers = m->numBuffers;
/* framebufferSize is used for allocating the handle to the framebuffer and refers
* to the size of the actual framebuffer.
* alignedFramebufferSize is used for allocating a possible internal buffer and
* thus need to consider internal alignment requirements. */
const size_t framebufferSize = m->finfo.line_length * m->info.yres;
const size_t alignedFramebufferSize = GRALLOC_ALIGN(m->finfo.line_length, 64) * m->info.yres;

*stride = m->info.xres;

if (numBuffers == 1)//如果是单缓冲，使用普通内存分配
{
// If we have only one buffer, we never use page-flipping. Instead,
// we return a regular buffer which will be memcpy'ed to the main
// screen when post is called.
int newUsage = (usage & ~GRALLOC_USAGE_HW_FB) | GRALLOC_USAGE_HW_2D;
AWAR( "fallback to single buffering. Virtual Y-res too small %d", m->info.yres );
*byte_stride = GRALLOC_ALIGN(m->finfo.line_length, 64);
return alloc_backend_alloc(dev, alignedFramebufferSize, newUsage, pHandle);
}

if (bufferMask >= ((1LU<<numBuffers)-1))
{
// We ran out of buffers.
return -ENOMEM;
}

uintptr_t framebufferVaddr = (uintptr_t)m->framebuffer->base;
// find a free slot
for (uint32_t i=0 ; i<numBuffers ; i++)//找到一块空闲的快
{
if ((bufferMask & (1LU<<i)) == 0)
{
m->bufferMask |= (1LU<<i);
break;
}
framebufferVaddr += framebufferSize;
}

// The entire framebuffer memory is already mapped, now create a buffer object for parts of this memory
private_handle_t* hnd = new private_handle_t(private_handle_t::PRIV_FLAGS_FRAMEBUFFER, usage, size,
(void*)framebufferVaddr, 0, dup(m->framebuffer->fd),
(framebufferVaddr - (uintptr_t)m->framebuffer->base), 0);

/*
* Perform allocator specific actions. If these fail we fall back to a regular buffer
* which will be memcpy'ed to the main screen when fb_post is called.
*/
if (alloc_backend_alloc_framebuffer(m, hnd) == -1)
{
delete hnd;
int newUsage = (usage & ~GRALLOC_USAGE_HW_FB) | GRALLOC_USAGE_HW_2D;
AERR( "Fallback to single buffering. Unable to map framebuffer memory to handle:%p", hnd );
*byte_stride = GRALLOC_ALIGN(m->finfo.line_length, 64);
return alloc_backend_alloc(dev, alignedFramebufferSize, newUsage, pHandle);
}

*pHandle = hnd;
*byte_stride = m->finfo.line_length;

return 0;
}这个函数如果第一次调用会调用init_frame_buffer_locked来从Framebuffer设备上分配一大块共享内存，内存的大小是屏幕尺寸的整数倍。numBuffers是内存的块数，如果只有一块代表是单缓冲，单缓冲调用分配普通内存的函数（单内存只能分配普通内存也很好理解，因为framebuffer的内存要用于显示。重新分配的话只能就分配普通内存了）。多缓冲的话使用Framebuffer的内存。但是Framebuffer的缓冲区块数也是有限的，因此函数要找一块空闲的缓冲区。如果缓冲区分配完了，返回错误值-ENOMEM。