fix [2068105] implement queueBuffer/lockBuffer/dequeueBuffer properly
Rewrote SurfaceFlinger's buffer management from the ground-up.
The design now support an arbitrary number of buffers per surface, however the current implementation is limited to four. Currently only 2 buffers are used in practice.
The main new feature is to be able to dequeue all buffers at once (very important when there are only two).
A client can dequeue all buffers until there are none available, it can lock all buffers except the last one that is used for composition. The client will block then, until a new buffer is enqueued.
The current implementation requires that buffers are locked in the same order they are dequeued and enqueued in the same order they are locked. Only one buffer can be locked at a time.
eg. Allowed sequence: DQ, DQ, LOCK, Q, LOCK, Q
eg. Forbidden sequence: DQ, DQ, LOCK, LOCK, Q, Q
diff --git a/libs/ui/SurfaceComposerClient.cpp b/libs/ui/SurfaceComposerClient.cpp
index d2cef78..8401cb6 100644
--- a/libs/ui/SurfaceComposerClient.cpp
+++ b/libs/ui/SurfaceComposerClient.cpp
@@ -40,8 +40,8 @@
#include <ui/SurfaceComposerClient.h>
#include <ui/Rect.h>
-#include <private/ui/SharedState.h>
#include <private/ui/LayerState.h>
+#include <private/ui/SharedBufferStack.h>
#include <private/ui/SurfaceFlingerSynchro.h>
#define VERBOSE(...) ((void)0)
@@ -103,169 +103,6 @@
// ---------------------------------------------------------------------------
-// these functions are used by the clients
-status_t per_client_cblk_t::validate(size_t i) const {
- if (uint32_t(i) >= NUM_LAYERS_MAX)
- return BAD_INDEX;
- if (layers[i].swapState & eInvalidSurface)
- return NO_MEMORY;
- return NO_ERROR;
-}
-
-int32_t per_client_cblk_t::lock_layer(size_t i, uint32_t flags)
-{
- int32_t index;
- uint32_t state;
- int timeout = 0;
- status_t result;
- layer_cblk_t * const layer = layers + i;
- const bool blocking = flags & BLOCKING;
- const bool inspect = flags & INSPECT;
-
- do {
- state = layer->swapState;
-
- if (UNLIKELY((state&(eFlipRequested|eNextFlipPending)) == eNextFlipPending)) {
- LOGE("eNextFlipPending set but eFlipRequested not set, "
- "layer=%d (lcblk=%p), state=%08x",
- int(i), layer, int(state));
- return INVALID_OPERATION;
- }
-
- if (UNLIKELY(state&eLocked)) {
- LOGE("eLocked set when entering lock_layer(), "
- "layer=%d (lcblk=%p), state=%08x",
- int(i), layer, int(state));
- return WOULD_BLOCK;
- }
-
-
- if (state & (eFlipRequested | eNextFlipPending | eResizeRequested
- | eInvalidSurface))
- {
- int32_t resizeIndex;
- Mutex::Autolock _l(lock);
- // might block for a very short amount of time
- // will never cause the server to block (trylock())
-
- goto start_loop_here;
-
- // We block the client if:
- // eNextFlipPending: we've used both buffers already, so we need to
- // wait for one to become availlable.
- // eResizeRequested: the buffer we're going to acquire is being
- // resized. Block until it is done.
- // eFlipRequested && eBusy: the buffer we're going to acquire is
- // currently in use by the server.
- // eInvalidSurface: this is a special case, we don't block in this
- // case, we just return an error.
-
- while((state & (eNextFlipPending|eInvalidSurface)) ||
- (state & ((resizeIndex) ? eResizeBuffer1 : eResizeBuffer0)) ||
- ((state & (eFlipRequested|eBusy)) == (eFlipRequested|eBusy)) )
- {
- if (state & eInvalidSurface)
- return NO_MEMORY;
-
- if (!blocking)
- return WOULD_BLOCK;
-
- timeout = 0;
- result = cv.waitRelative(lock, seconds(1));
- if (__builtin_expect(result!=NO_ERROR, false)) {
- const int newState = layer->swapState;
- LOGW( "lock_layer timed out (is the CPU pegged?) "
- "layer=%d, lcblk=%p, state=%08x (was %08x)",
- int(i), layer, newState, int(state));
- timeout = newState != int(state);
- }
-
- start_loop_here:
- state = layer->swapState;
- resizeIndex = (state&eIndex) ^ ((state&eFlipRequested)>>1);
- }
-
- LOGW_IF(timeout,
- "lock_layer() timed out but didn't appear to need "
- "to be locked and we recovered "
- "(layer=%d, lcblk=%p, state=%08x)",
- int(i), layer, int(state));
- }
-
- // eFlipRequested is not set and cannot be set by another thread: it's
- // safe to use the first buffer without synchronization.
-
- // Choose the index depending on eFlipRequested.
- // When it's set, choose the 'other' buffer.
- index = (state&eIndex) ^ ((state&eFlipRequested)>>1);
-
- // make sure this buffer is valid
- status_t err = layer->surface[index].status;
- if (err < 0) {
- return err;
- }
-
- if (inspect) {
- // we just want to inspect this layer. don't lock it.
- goto done;
- }
-
- // last thing before we're done, we need to atomically lock the state
- } while (android_atomic_cmpxchg(state, state|eLocked, &(layer->swapState)));
-
- VERBOSE("locked layer=%d (lcblk=%p), buffer=%d, state=0x%08x",
- int(i), layer, int(index), int(state));
-
- // store the index of the locked buffer (for client use only)
- layer->flags &= ~eBufferIndex;
- layer->flags |= ((index << eBufferIndexShift) & eBufferIndex);
-
-done:
- return index;
-}
-
-uint32_t per_client_cblk_t::unlock_layer_and_post(size_t i)
-{
- // atomically set eFlipRequested and clear eLocked and optionally
- // set eNextFlipPending if eFlipRequested was already set
-
- layer_cblk_t * const layer = layers + i;
- int32_t oldvalue, newvalue;
- do {
- oldvalue = layer->swapState;
- // get current value
-
- newvalue = oldvalue & ~eLocked;
- // clear eLocked
-
- newvalue |= eFlipRequested;
- // set eFlipRequested
-
- if (oldvalue & eFlipRequested)
- newvalue |= eNextFlipPending;
- // if eFlipRequested was already set, set eNextFlipPending
-
- } while (android_atomic_cmpxchg(oldvalue, newvalue, &(layer->swapState)));
-
- VERBOSE("request pageflip for layer=%d, buffer=%d, state=0x%08x",
- int(i), int((layer->flags & eBufferIndex) >> eBufferIndexShift),
- int(newvalue));
-
- // from this point, the server can kick in at any time and use the first
- // buffer, so we cannot use it anymore, and we must use the 'other'
- // buffer instead (or wait if it is not available yet, see lock_layer).
-
- return newvalue;
-}
-
-void per_client_cblk_t::unlock_layer(size_t i)
-{
- layer_cblk_t * const layer = layers + i;
- android_atomic_and(~eLocked, &layer->swapState);
-}
-
-// ---------------------------------------------------------------------------
-
static inline int compare_type( const layer_state_t& lhs,
const layer_state_t& rhs) {
if (lhs.surface < rhs.surface) return -1;
@@ -315,7 +152,7 @@
mControlMemory = mClient->getControlBlock();
mSignalServer = new SurfaceFlingerSynchro(sm);
- mControl = static_cast<per_client_cblk_t *>(mControlMemory->getBase());
+ mControl = static_cast<SharedClient *>(mControlMemory->getBase());
}
SurfaceComposerClient::~SurfaceComposerClient()
@@ -539,18 +376,17 @@
const size_t N = clients.size();
VERBOSE("closeGlobalTransaction (%ld clients)", N);
- if (N == 1) {
- clients[0]->closeTransaction();
- } else {
- const sp<ISurfaceComposer>& sm(_get_surface_manager());
- sm->openGlobalTransaction();
- for (size_t i=0; i<N; i++) {
- clients[i]->closeTransaction();
- }
- sm->closeGlobalTransaction();
+
+ const sp<ISurfaceComposer>& sm(_get_surface_manager());
+ sm->openGlobalTransaction();
+ for (size_t i=0; i<N; i++) {
+ clients[i]->closeTransaction();
}
+ sm->closeGlobalTransaction();
+
}
+
status_t SurfaceComposerClient::freezeDisplay(DisplayID dpy, uint32_t flags)
{
const sp<ISurfaceComposer>& sm(_get_surface_manager());