blob: 34ca64c5547851f0bec6b7435367981b5fe85ed6 [file] [log] [blame]
// Copyright 2019 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef SRC_DEVICES_SYSMEM_DRIVERS_SYSMEM_LOGICAL_BUFFER_COLLECTION_H_
#define SRC_DEVICES_SYSMEM_DRIVERS_SYSMEM_LOGICAL_BUFFER_COLLECTION_H_
#include <fuchsia/sysmem/c/fidl.h>
#include <fuchsia/sysmem/llcpp/fidl.h>
#include <fuchsia/sysmem2/llcpp/fidl.h>
#include <inttypes.h>
#include <lib/async/cpp/task.h>
#include <lib/fidl-async-2/fidl_struct.h>
#include <lib/fidl/llcpp/fidl_allocator.h>
#include <lib/zx/channel.h>
#include <list>
#include <map>
#include <memory>
#include <fbl/ref_counted.h>
#include <fbl/ref_ptr.h>
#include <fbl/string_printf.h>
#include "allocation_result.h"
#include "device.h"
#include "logging.h"
#include "node.h"
#include "node_properties.h"
#include "table_set.h"
namespace sysmem_driver {
class BufferCollectionToken;
class BufferCollection;
class MemoryAllocator;
// This class can be used to hold an inspect snapshot of one set of constraints taken from a client
// at a particular point in time.
struct ConstraintInfoSnapshot {
inspect::Node inspect_node;
inspect::ValueList node_constraints;
};
// TODO(dustingreen): MaybeAllocate() should sweep all related incoming channels for ZX_PEER_CLOSED
// and not attempt allocation until all channel close(es) that were pending at the time have been
// processed. Ignoring new channel closes is fine/good.
class LogicalBufferCollection : public fbl::RefCounted<LogicalBufferCollection> {
public:
using FidlAllocator = fidl::FidlAllocator<>;
using CollectionMap = std::map<BufferCollection*, fbl::RefPtr<BufferCollection>>;
~LogicalBufferCollection();
static void Create(zx::channel buffer_collection_token_request, Device* parent_device);
// |parent_device| the Device* that the calling allocator is part of. The
// tokens_by_koid_ for each Device is separate. If somehow two clients were
// to get connected to two separate sysmem device instances hosted in the
// same devhost, those clients (intentionally) won't be able to share a
// LogicalBufferCollection.
//
// |buffer_collection_token| the client end of the BufferCollectionToken
// being turned in by the client to get a BufferCollection in exchange.
//
// |buffer_collection_request| the server end of a BufferCollection channel
// to be served by the LogicalBufferCollection associated with
// buffer_collection_token.
static void BindSharedCollection(Device* parent_device, zx::channel buffer_collection_token,
zx::channel buffer_collection_request,
const ClientDebugInfo* client_debug_info);
// ZX_OK if the token is known to the server.
// ZX_ERR_NOT_FOUND if the token isn't known to the server.
static zx_status_t ValidateBufferCollectionToken(Device* parent_device,
zx_koid_t token_server_koid);
// This is used to create the initial BufferCollectionToken, and also used
// by BufferCollectionToken::Duplicate().
//
// The |self| parameter exists only because LogicalBufferCollection can't
// hold a std::weak_ptr<> to itself because that requires libc++ (the binary
// not just the headers) which isn't available in Zircon so far.
void CreateBufferCollectionToken(
fbl::RefPtr<LogicalBufferCollection> self, NodeProperties* new_node_properties,
fidl::ServerEnd<fuchsia_sysmem::BufferCollectionToken> token_request);
void AttachLifetimeTracking(zx::eventpair server_end, uint32_t buffers_remaining);
void SweepLifetimeTracking();
void OnSetConstraints();
void SetName(uint32_t priority, std::string name);
void SetDebugTimeoutLogDeadline(int64_t deadline);
uint64_t CreateDispensableOrdinal();
void VLogClient(bool is_error, Location location, const NodeProperties* node_properties,
const char* format, va_list args) const;
void LogClientInfo(Location location, const NodeProperties* node_properties, const char* format,
...) const __PRINTFLIKE(4, 5);
void LogClientError(Location location, const NodeProperties* node_properties, const char* format,
...) const __PRINTFLIKE(4, 5);
void VLogClientInfo(Location location, const NodeProperties* node_properties, const char* format,
va_list args) const;
void VLogClientError(Location location, const NodeProperties* node_properties, const char* format,
va_list args) const;
Device* parent_device() const { return parent_device_; }
// For tests.
std::vector<const BufferCollection*> collection_views() const;
TableSet& table_set() { return table_set_; }
std::optional<std::string> name() const {
return name_ ? std::make_optional(name_->name) : std::optional<std::string>();
}
inspect::Node& inspect_node() { return inspect_node_; }
private:
friend class NodeProperties;
enum class CheckSanitizeStage { kInitial, kNotAggregated, kAggregated };
class Constraints {
public:
Constraints(const Constraints&) = delete;
Constraints(Constraints&&) = default;
Constraints(TableSet& table_set,
fuchsia_sysmem2::wire::BufferCollectionConstraints&& constraints,
const NodeProperties& node_properties)
: buffer_collection_constraints_(table_set, std::move(constraints)),
node_properties_(node_properties) {}
const fuchsia_sysmem2::wire::BufferCollectionConstraints& constraints() const {
return *buffer_collection_constraints_;
}
fuchsia_sysmem2::wire::BufferCollectionConstraints& mutate_constraints() {
return buffer_collection_constraints_.mutate();
}
const ClientDebugInfo& client_debug_info() const {
return node_properties_.client_debug_info();
}
const NodeProperties& node_properties() const { return node_properties_; }
private:
TableHolder<fuchsia_sysmem2::wire::BufferCollectionConstraints> buffer_collection_constraints_;
const NodeProperties& node_properties_;
};
using ConstraintsList = std::list<Constraints>;
struct CollectionName {
uint32_t priority{};
std::string name;
};
LogicalBufferCollection(Device* parent_device);
// Will log an error, and then FailRoot().
void LogAndFailRootNode(Location location, zx_status_t epitaph, const char* format, ...)
__PRINTFLIKE(4, 5);
// This fails the entire LogicalBufferCollection, by failing the root Node, which propagates that
// failure to the entire Node tree, and also results in zero remaining Node(s). Then once all
// outstanding VMOs have been closed, the LogicalBufferCollection is deleted.
//
// This cleans out a lot of state that's unnecessary after a failure.
void FailRootNode(zx_status_t epitaph);
NodeProperties* FindTreeToFail(NodeProperties* failing_node);
// Fails the tree rooted at tree_to_fail. The tree_to_fail is allowed to be the root_, or it can
// be a sub-tree. All the Node(s) from tree_to_fail down are Fail()ed. Any Node(s) that still
// have channels open will send epitaph just before the Node('s) channel closes. All Node(s) from
// tree_to_fail downward are also removed from the tree. If tree_to_fail is the root, then when
// all child VMOs have been closed, the LogicalBufferCollection will be deleted.
//
// Node(s) are Fail()ed and removed from the tree in child-first order.
//
// These two are only appropriate to use if it's already known which node should fail. If it's
// not yet known which node should fail, call FindTreeToFail() first, or use LogAndFailNode() /
// FailNode().
void LogAndFailDownFrom(Location location, NodeProperties* tree_to_fail, zx_status_t epitaph,
const char* format, ...) __PRINTFLIKE(5, 6);
void FailDownFrom(NodeProperties* tree_to_fail, zx_status_t epitaph);
// Find the root-most node of member_node's failure domain, and fail that whole failure domain and
// any of its child failure domains.
void LogAndFailNode(Location location, NodeProperties* member_node, zx_status_t epitaph,
const char* format, ...) __PRINTFLIKE(5, 6);
void FailNode(NodeProperties* member_node, zx_status_t epitaph);
static void LogInfo(Location location, const char* format, ...);
static void LogErrorStatic(Location location, const ClientDebugInfo* client_debug_info,
const char* format, ...) __PRINTFLIKE(3, 4);
void LogError(Location location, const char* format, ...) const __PRINTFLIKE(3, 4);
void VLogError(Location location, const char* format, va_list args) const;
// The caller must keep "this" alive. We require this of the caller since the caller is fairly
// likely to want to keep "this" alive longer than MaybeAllocate() could anyway.
void MaybeAllocate();
void TryAllocate(std::vector<NodeProperties*> nodes);
void TryLateLogicalAllocation(std::vector<NodeProperties*> nodes);
void InitializeConstraintSnapshots(const ConstraintsList& constraints_list);
void SetFailedAllocationResult(zx_status_t status);
void SetAllocationResult(std::vector<NodeProperties*> nodes,
fuchsia_sysmem2::wire::BufferCollectionInfo&& info);
void SendAllocationResult(std::vector<NodeProperties*> nodes);
void SetFailedLateLogicalAllocationResult(NodeProperties* tree, zx_status_t status_param);
void SetSucceededLateLogicalAllocationResult(std::vector<NodeProperties*> nodes);
void BindSharedCollectionInternal(BufferCollectionToken* token,
zx::channel buffer_collection_request);
fit::result<fuchsia_sysmem2::wire::BufferCollectionConstraints, void> CombineConstraints(
ConstraintsList* constraints_list);
bool CheckSanitizeBufferCollectionConstraints(
CheckSanitizeStage stage, fuchsia_sysmem2::wire::BufferCollectionConstraints& constraints);
bool CheckSanitizeBufferUsage(CheckSanitizeStage stage,
fuchsia_sysmem2::wire::BufferUsage& buffer_usage);
bool CheckSanitizeBufferMemoryConstraints(
CheckSanitizeStage stage, const fuchsia_sysmem2::wire::BufferUsage& buffer_usage,
fuchsia_sysmem2::wire::BufferMemoryConstraints& constraints);
bool CheckSanitizeImageFormatConstraints(
CheckSanitizeStage stage, fuchsia_sysmem2::wire::ImageFormatConstraints& constraints);
bool AccumulateConstraintBufferCollection(fuchsia_sysmem2::wire::BufferCollectionConstraints* acc,
fuchsia_sysmem2::wire::BufferCollectionConstraints c);
bool AccumulateConstraintsBufferUsage(fuchsia_sysmem2::wire::BufferUsage* acc,
fuchsia_sysmem2::wire::BufferUsage c);
bool AccumulateConstraintHeapPermitted(fidl::VectorView<fuchsia_sysmem2::wire::HeapType>* acc,
fidl::VectorView<fuchsia_sysmem2::wire::HeapType> c);
bool AccumulateConstraintBufferMemory(fuchsia_sysmem2::wire::BufferMemoryConstraints* acc,
fuchsia_sysmem2::wire::BufferMemoryConstraints c);
bool AccumulateConstraintImageFormats(
fidl::VectorView<fuchsia_sysmem2::wire::ImageFormatConstraints>* acc,
fidl::VectorView<fuchsia_sysmem2::wire::ImageFormatConstraints> c);
bool AccumulateConstraintImageFormat(fuchsia_sysmem2::wire::ImageFormatConstraints* acc,
fuchsia_sysmem2::wire::ImageFormatConstraints c);
bool AccumulateConstraintColorSpaces(fidl::VectorView<fuchsia_sysmem2::wire::ColorSpace>* acc,
fidl::VectorView<fuchsia_sysmem2::wire::ColorSpace> c);
size_t InitialCapacityOrZero(CheckSanitizeStage stage, size_t initial_capacity);
bool IsColorSpaceEqual(const fuchsia_sysmem2::wire::ColorSpace& a,
const fuchsia_sysmem2::wire::ColorSpace& b);
fit::result<fuchsia_sysmem2::wire::BufferCollectionInfo, zx_status_t>
GenerateUnpopulatedBufferCollectionInfo(
const fuchsia_sysmem2::wire::BufferCollectionConstraints& constraints);
fit::result<fuchsia_sysmem2::wire::BufferCollectionInfo, zx_status_t> Allocate(
const fuchsia_sysmem2::wire::BufferCollectionConstraints& constraints,
fuchsia_sysmem2::wire::BufferCollectionInfo* buffer_collection_info);
fit::result<zx::vmo> AllocateVmo(MemoryAllocator* allocator,
const fuchsia_sysmem2::wire::SingleBufferSettings& settings,
uint32_t index);
int32_t CompareImageFormatConstraintsTieBreaker(
const fuchsia_sysmem2::wire::ImageFormatConstraints& a,
const fuchsia_sysmem2::wire::ImageFormatConstraints& b);
int32_t CompareImageFormatConstraintsByIndex(
const fuchsia_sysmem2::wire::BufferCollectionConstraints& constraints, uint32_t index_a,
uint32_t index_b);
void CreationTimedOut(async_dispatcher_t* dispatcher, async::TaskBase* task, zx_status_t status);
AllocationResult allocation_result();
void LogBufferCollectionInfoDiffs(const fuchsia_sysmem2::wire::BufferCollectionInfo& o,
const fuchsia_sysmem2::wire::BufferCollectionInfo& n);
// To be called only by CombineConstraints().
bool IsMinBufferSizeSpecifiedByAnyParticipant(const ConstraintsList& constraints_list);
// This is the root and any sub-trees created via SetDispensable() or AttachToken().
std::vector<NodeProperties*> FailureDomainSubtrees();
// A Node is a pruned sub-tree eligible for logical allocation if it is not yet logically
// allocated, and does not have a non-logically-allocated direct parent.
//
// Such sub-trees will always have a root-most Node that is either the root or an
// ErrorPropagationMode kDoNotPropagate Node.
//
// The returned pruned sub-trees may not be ready for logical allocation (may not have all its
// constraints yet), nor do they necessarily have constraints that are satisfiable.
std::vector<NodeProperties*> PrunedSubtreesEligibleForLogicalAllocation();
// The granularity of logical allocation is a sub-tree pruned of any ErrorPropagationMode
// kDoNotPropagate Node(s) other than the sub-tree's root-most node.
//
// The root-most Node of a pruned subtree can be the root, or it can be a Node that has
// ErrorPropagationMode kDoNotPropagate (implying that AttachToken() was used to create the
// non-root Node).
//
// The pruning is the same for both the root and a non-root Node so that the locally-observable
// (by a participant) logical allocation granularity is the same regardless of whether the logical
// allocation granule has the root as its root-most Node, or has an ErrorPropagationMode
// kDoNotPropagate Node as its root-most Node.
std::vector<NodeProperties*> NodesOfPrunedSubtreeEligibleForLogicalAllocation(
NodeProperties& subtree);
void LogTreeForDebugOnly(NodeProperties* node);
// For NodeProperties:
void AddCountsForNode(const Node& node);
void RemoveCountsForNode(const Node& node);
void AdjustRelevantNodeCounts(const Node& node, fit::function<void(uint32_t& count)> visitor);
void DeleteRoot();
// Diff printing.
#if defined(PRINT_DIFF)
#error "Let's pick a different name for PRINT_DIFF"
#endif
// If LLCPP had generic field objects, we wouldn't need this macro. We #undef the macro name
// after we're done using it so it doesn't escape this header.
#define PRINT_DIFF(child_field_name) \
do { \
const std::string& parent_field_name = field_name; \
if (o.has_##child_field_name() != n.has_##child_field_name()) { \
LogError(FROM_HERE, \
"o%s.has_" #child_field_name "(): %d n%s.has_" #child_field_name "(): %d", \
parent_field_name.c_str(), o.has_##child_field_name(), parent_field_name.c_str(), \
n.has_##child_field_name()); \
} else if (o.has_##child_field_name()) { \
std::string field_name = \
fbl::StringPrintf("%s.%s()", parent_field_name.c_str(), #child_field_name).c_str(); \
DiffPrinter<std::remove_const_t<std::remove_reference_t<decltype(o.child_field_name())>>>:: \
PrintDiff(*this, field_name, o.child_field_name(), n.child_field_name()); \
} \
} while (false)
template <typename FieldType, typename Enable = void>
struct DiffPrinter {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const FieldType& o, const FieldType& n);
};
template <typename ElementType>
struct DiffPrinter<fidl::VectorView<ElementType>,
std::enable_if_t<fidl::IsVectorView<fidl::VectorView<ElementType>>::value>> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const fidl::VectorView<ElementType>& o,
const fidl::VectorView<ElementType>& n) {
if (o.count() != n.count()) {
buffer_collection.LogError(FROM_HERE, "o%s.count(): %" PRIu64 " n%s.count(): %" PRIu64,
field_name.c_str(), o.count(), field_name.c_str(), n.count());
}
for (uint32_t i = 0; i < std::max(o.count(), n.count()); ++i) {
std::string new_field_name = fbl::StringPrintf("%s[%u]", field_name.c_str(), i).c_str();
const ElementType& blank = ElementType();
const ElementType& o_element = i < o.count() ? o[i] : blank;
const ElementType& n_element = i < n.count() ? n[i] : blank;
DiffPrinter<ElementType>::PrintDiff(buffer_collection, new_field_name, o_element,
n_element);
}
}
};
template <typename TableType>
struct DiffPrinter<TableType, std::enable_if_t<fidl::IsTable<TableType>::value>> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const TableType& o, const TableType& n) {
buffer_collection.LogTableDiffs<TableType>(field_name, o, n);
}
};
template <>
struct DiffPrinter<bool, void> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const bool& o, const bool& n) {
if (o != n) {
buffer_collection.LogError(FROM_HERE, "o%s: %d n%s: %d", field_name.c_str(), o,
field_name.c_str(), n);
}
}
};
template <>
struct DiffPrinter<uint32_t, void> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const uint32_t& o, const uint32_t& n) {
if (o != n) {
buffer_collection.LogError(FROM_HERE, "o%s: %u n%s: %u", field_name.c_str(), o,
field_name.c_str(), n);
}
}
};
template <typename EnumType>
struct DiffPrinter<EnumType, std::enable_if_t<std::is_enum_v<EnumType>>> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const EnumType& o, const EnumType& n) {
using UnderlyingType = std::underlying_type_t<EnumType>;
const UnderlyingType o_underlying = static_cast<UnderlyingType>(o);
const UnderlyingType n_underlying = static_cast<UnderlyingType>(n);
DiffPrinter<UnderlyingType>::PrintDiff(buffer_collection, field_name, o_underlying,
n_underlying);
}
};
template <>
struct DiffPrinter<uint64_t, void> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const uint64_t& o, const uint64_t& n) {
if (o != n) {
buffer_collection.LogError(FROM_HERE, "o%s: %" PRIu64 " n%s: %" PRIu64, field_name.c_str(),
o, field_name.c_str(), n);
}
}
};
template <>
struct DiffPrinter<zx::vmo, void> {
static void PrintDiff(const LogicalBufferCollection& buffer_collection,
const std::string& field_name, const zx::vmo& o, const zx::vmo& n) {
// We don't expect to call the zx::vmo variant since !has_vmo() and !has_aux_vmo(), but if we
// do get here, complain + print the values regardless of what the values are or whether they
// differ.
buffer_collection.LogError(FROM_HERE,
"Why did we call zx::vmo PrintDiff? --- o%s: %u n%s: %u",
field_name.c_str(), o.get(), field_name.c_str(), n.get());
}
};
template <typename Table>
void LogTableDiffs(const std::string& field_name, const Table& o, const Table& n) const;
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::BufferMemorySettings>(
const std::string& field_name, const fuchsia_sysmem2::wire::BufferMemorySettings& o,
const fuchsia_sysmem2::wire::BufferMemorySettings& n) const {
PRINT_DIFF(size_bytes);
PRINT_DIFF(is_physically_contiguous);
PRINT_DIFF(is_secure);
PRINT_DIFF(coherency_domain);
PRINT_DIFF(heap);
}
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::PixelFormat>(
const std::string& field_name, const fuchsia_sysmem2::wire::PixelFormat& o,
const fuchsia_sysmem2::wire::PixelFormat& n) const {
PRINT_DIFF(type);
PRINT_DIFF(format_modifier_value);
}
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::ColorSpace>(
const std::string& field_name, const fuchsia_sysmem2::wire::ColorSpace& o,
const fuchsia_sysmem2::wire::ColorSpace& n) const {
PRINT_DIFF(type);
}
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::ImageFormatConstraints>(
const std::string& field_name, const fuchsia_sysmem2::wire::ImageFormatConstraints& o,
const fuchsia_sysmem2::wire::ImageFormatConstraints& n) const {
PRINT_DIFF(pixel_format);
PRINT_DIFF(color_spaces);
PRINT_DIFF(min_coded_width);
PRINT_DIFF(max_coded_width);
PRINT_DIFF(min_coded_height);
PRINT_DIFF(max_coded_height);
PRINT_DIFF(min_bytes_per_row);
PRINT_DIFF(max_bytes_per_row);
PRINT_DIFF(max_coded_width_times_coded_height);
PRINT_DIFF(coded_width_divisor);
PRINT_DIFF(coded_height_divisor);
PRINT_DIFF(bytes_per_row_divisor);
PRINT_DIFF(start_offset_divisor);
PRINT_DIFF(display_width_divisor);
PRINT_DIFF(display_height_divisor);
PRINT_DIFF(required_min_coded_width);
PRINT_DIFF(required_max_coded_width);
PRINT_DIFF(required_min_coded_height);
PRINT_DIFF(required_max_coded_height);
PRINT_DIFF(required_min_bytes_per_row);
PRINT_DIFF(required_max_bytes_per_row);
}
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::SingleBufferSettings>(
const std::string& field_name, const fuchsia_sysmem2::wire::SingleBufferSettings& o,
const fuchsia_sysmem2::wire::SingleBufferSettings& n) const {
PRINT_DIFF(buffer_settings);
PRINT_DIFF(image_format_constraints);
}
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::VmoBuffer>(
const std::string& field_name, const fuchsia_sysmem2::wire::VmoBuffer& o,
const fuchsia_sysmem2::wire::VmoBuffer& n) const {
PRINT_DIFF(vmo);
PRINT_DIFF(vmo_usable_start);
PRINT_DIFF(aux_vmo);
}
template <>
void LogTableDiffs<fuchsia_sysmem2::wire::BufferCollectionInfo>(
const std::string& field_name, const fuchsia_sysmem2::wire::BufferCollectionInfo& o,
const fuchsia_sysmem2::wire::BufferCollectionInfo& n) const {
PRINT_DIFF(settings);
PRINT_DIFF(buffers);
}
#undef PRINT_DIFF
void LogDiffsBufferCollectionInfo(const fuchsia_sysmem2::wire::BufferCollectionInfo& o,
const fuchsia_sysmem2::wire::BufferCollectionInfo& n) const {
LOG(WARNING, "LogDiffsBufferCollectionInfo()");
LogTableDiffs<fuchsia_sysmem2::wire::BufferCollectionInfo>("", o, n);
}
void LogConstraints(Location location, NodeProperties* node_properties, const fuchsia_sysmem2::wire::BufferCollectionConstraints& constraints) const;
Device* parent_device_ = nullptr;
// We occasionally swap out the allocator for a fresh one, to avoid the possibility of churn
// leading to excessive un-used memory allocation in the allocator. This is accomplished via
// TableHolder and TableSet.
TableSet table_set_;
// This owns the current tree of BufferCollectionToken, BufferCollection, OrphanedNode. The
// Location in the tree is determined by creation path. Child Node(s) are children because they
// were created via their parent node. The root node is created by CreateSharedCollection() which
// also creates the LogicalBufferCollection.
//
// We use shared_ptr<> to manage NodeProperties only so that Node can have a std::weak_ptr<> back
// to NodeProperties instead of a raw pointer, since the Node can last longer than NodeProperties,
// despite NodeProperties being the primary non-transient owner of Node. This way we avoid using
// a raw pointer from Node to NodeProperties.
//
// The tree at root_ is the only non-transient ownership of NodeProperties. Transient ownership
// lasts only during a portion of a single dispatch on parent_device_->dispatcher().
std::shared_ptr<NodeProperties> root_;
std::vector<ConstraintInfoSnapshot> constraints_at_allocation_;
bool is_allocate_attempted_ = false;
// Iff true, initial allocation has been attempted and has succeeded or
// failed. Both allocation_result_status_ and allocation_result_info_ are
// not meaningful until has_allocation_result_ is true.
bool has_allocation_result_ = false;
std::optional<TableHolder<fuchsia_sysmem2::wire::BufferCollectionInfo>>
buffer_collection_info_before_population_;
std::optional<fuchsia_sysmem2::wire::BufferCollectionInfo::OwnedEncodedMessage>
linearized_buffer_collection_info_before_population_;
zx_status_t allocation_result_status_ = ZX_OK;
std::optional<TableHolder<fuchsia_sysmem2::wire::BufferCollectionInfo>> allocation_result_info_;
MemoryAllocator* memory_allocator_ = nullptr;
std::optional<CollectionName> name_;
// 0 means not dispensable. Non-zero means dispensable, with each value being a group of
// BufferCollectionToken(s) / BufferCollection(s) that were all created from a single
// BufferCollectionToken that was created with AttachToken() or which had SetDispensable() called
// on it. Each group's constraints are aggregated together and succeed or fail to logically
// allocate as a group, considered in order by when each group's constraints are ready, not in
// order by dispensable_ordinal values.
uint64_t next_dispensable_ordinal_ = 1;
// Information about the current client - only valid while aggregating state for a particular
// client.
const NodeProperties* current_node_properties_ = nullptr;
inspect::Node inspect_node_;
inspect::StringProperty name_property_;
inspect::UintProperty vmo_count_property_;
inspect::ValueList vmo_properties_;
// We keep LogicalBufferCollection alive as long as there are child VMOs
// outstanding (no revoking of child VMOs for now).
//
// This tracking is for the benefit of MemoryAllocator sub-classes that need
// a Delete() call, such as to clean up a slab allocation and/or to inform
// an external allocator of delete.
class TrackedParentVmo {
public:
using DoDelete = fit::callback<void(TrackedParentVmo* parent)>;
// The do_delete callback will be invoked upon the sooner of (A) the client
// code causing ~ParentVmo, or (B) ZX_VMO_ZERO_CHILDREN occurring async
// after StartWait() is called.
TrackedParentVmo(fbl::RefPtr<LogicalBufferCollection> buffer_collection, zx::vmo vmo,
DoDelete do_delete);
~TrackedParentVmo();
// This should only be called after client code has created a child VMO, and
// will begin the wait for ZX_VMO_ZERO_CHILDREN.
zx_status_t StartWait(async_dispatcher_t* dispatcher);
// Cancel the wait. This should only be used by LogicalBufferCollection
zx_status_t CancelWait();
zx::vmo TakeVmo();
[[nodiscard]] const zx::vmo& vmo() const;
void set_child_koid(zx_koid_t koid) { child_koid_ = koid; }
TrackedParentVmo(const TrackedParentVmo&) = delete;
TrackedParentVmo(TrackedParentVmo&&) = delete;
TrackedParentVmo& operator=(const TrackedParentVmo&) = delete;
TrackedParentVmo& operator=(TrackedParentVmo&&) = delete;
private:
void OnZeroChildren(async_dispatcher_t* dispatcher, async::WaitBase* wait, zx_status_t status,
const zx_packet_signal_t* signal);
fbl::RefPtr<LogicalBufferCollection> buffer_collection_;
zx::vmo vmo_;
zx_koid_t child_koid_{};
DoDelete do_delete_;
async::WaitMethod<TrackedParentVmo, &TrackedParentVmo::OnZeroChildren> zero_children_wait_;
// Only for asserts:
bool waiting_ = {};
};
// From buffers_remaining to server_end.
std::multimap<uint32_t, zx::eventpair> lifetime_tracking_;
// It's nice for members containing timers to be last for destruction order purposes, but the
// destructor also explicitly cancels timers to avoid any brittle-ness from members potentially
// added after these.
using ParentVmoMap = std::map<zx_handle_t, std::unique_ptr<TrackedParentVmo>>;
ParentVmoMap parent_vmos_;
async::TaskMethod<LogicalBufferCollection, &LogicalBufferCollection::CreationTimedOut>
creation_timer_{this};
};
} // namespace sysmem_driver
#endif // SRC_DEVICES_SYSMEM_DRIVERS_SYSMEM_LOGICAL_BUFFER_COLLECTION_H_