diff options
Diffstat (limited to 'indra/llcommon')
32 files changed, 2718 insertions, 1247 deletions
diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 3d073b009c..516813aae1 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -16,6 +16,7 @@ include(Tracy) set(llcommon_SOURCE_FILES + commoncontrol.cpp indra_constants.cpp llallocator.cpp llallocator_heap_profile.cpp @@ -105,6 +106,7 @@ set(llcommon_SOURCE_FILES lluriparser.cpp lluuid.cpp llworkerthread.cpp + hbxxh.cpp u64.cpp threadpool.cpp workqueue.cpp @@ -116,6 +118,7 @@ set(llcommon_HEADER_FILES chrono.h classic_callback.h + commoncontrol.h ctype_workaround.h fix_macros.h indra_constants.h @@ -173,6 +176,7 @@ set(llcommon_HEADER_FILES llinitdestroyclass.h llinitparam.h llinstancetracker.h + llinstancetrackersubclass.h llkeybind.h llkeythrottle.h llleap.h @@ -241,10 +245,12 @@ set(llcommon_HEADER_FILES llwin32headers.h llwin32headerslean.h llworkerthread.h + hbxxh.h lockstatic.h stdtypes.h stringize.h threadpool.h + threadpool_fwd.h threadsafeschedule.h timer.h tuple.h diff --git a/indra/llcommon/commoncontrol.cpp b/indra/llcommon/commoncontrol.cpp new file mode 100644 index 0000000000..81e66baf8c --- /dev/null +++ b/indra/llcommon/commoncontrol.cpp @@ -0,0 +1,106 @@ +/** + * @file commoncontrol.cpp + * @author Nat Goodspeed + * @date 2022-06-08 + * @brief Implementation for commoncontrol. + * + * $LicenseInfo:firstyear=2022&license=viewerlgpl$ + * Copyright (c) 2022, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "commoncontrol.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "llevents.h" +#include "llsdutil.h" + +LLSD LL::CommonControl::access(const LLSD& params) +{ + // We can't actually introduce a link-time dependency on llxml, or on any + // global LLControlGroup (*koff* gSavedSettings *koff*) but we can issue a + // runtime query. If we're running as part of a viewer with + // LLViewerControlListener, we can use that to interact with any + // instantiated LLControGroup. + LLSD response; + { + LLEventStream reply("reply"); + LLTempBoundListener connection = reply.listen("listener", + [&response] (const LLSD& event) + { + response = event; + return false; + }); + LLSD rparams{ params }; + rparams["reply"] = reply.getName(); + LLEventPumps::instance().obtain("LLViewerControl").post(rparams); + } + // LLViewerControlListener responds immediately. If it's listening at all, + // it will already have set response. + if (! response.isDefined()) + { + LLTHROW(NoListener("No LLViewerControl listener instantiated")); + } + LLSD error{ response["error"] }; + if (error.isDefined()) + { + LLTHROW(ParamError(error)); + } + response.erase("error"); + response.erase("reqid"); + return response; +} + +/// set control group.key to defined default value +LLSD LL::CommonControl::set_default(const std::string& group, const std::string& key) +{ + return access(llsd::map("op", "set", + "group", group, "key", key))["value"]; +} + +/// set control group.key to specified value +LLSD LL::CommonControl::set(const std::string& group, const std::string& key, const LLSD& value) +{ + return access(llsd::map("op", "set", + "group", group, "key", key, "value", value))["value"]; +} + +/// toggle boolean control group.key +LLSD LL::CommonControl::toggle(const std::string& group, const std::string& key) +{ + return access(llsd::map("op", "toggle", + "group", group, "key", key))["value"]; +} + +/// get the definition for control group.key, (! isDefined()) if bad +/// ["name"], ["type"], ["value"], ["comment"] +LLSD LL::CommonControl::get_def(const std::string& group, const std::string& key) +{ + return access(llsd::map("op", "get", + "group", group, "key", key)); +} + +/// get the value of control group.key +LLSD LL::CommonControl::get(const std::string& group, const std::string& key) +{ + return access(llsd::map("op", "get", + "group", group, "key", key))["value"]; +} + +/// get defined groups +std::vector<std::string> LL::CommonControl::get_groups() +{ + auto groups{ access(llsd::map("op", "groups"))["groups"] }; + return { groups.beginArray(), groups.endArray() }; +} + +/// get definitions for all variables in group +LLSD LL::CommonControl::get_vars(const std::string& group) +{ + return access(llsd::map("op", "vars", "group", group))["vars"]; +} diff --git a/indra/llcommon/commoncontrol.h b/indra/llcommon/commoncontrol.h new file mode 100644 index 0000000000..07d4a45ac5 --- /dev/null +++ b/indra/llcommon/commoncontrol.h @@ -0,0 +1,75 @@ +/** + * @file commoncontrol.h + * @author Nat Goodspeed + * @date 2022-06-08 + * @brief Access LLViewerControl LLEventAPI, if process has one. + * + * $LicenseInfo:firstyear=2022&license=viewerlgpl$ + * Copyright (c) 2022, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_COMMONCONTROL_H) +#define LL_COMMONCONTROL_H + +#include <vector> +#include "llexception.h" +#include "llsd.h" + +namespace LL +{ + class CommonControl + { + public: + struct Error: public LLException + { + Error(const std::string& what): LLException(what) {} + }; + + /// Exception thrown if there's no LLViewerControl LLEventAPI + struct NoListener: public Error + { + NoListener(const std::string& what): Error(what) {} + }; + + struct ParamError: public Error + { + ParamError(const std::string& what): Error(what) {} + }; + + /// set control group.key to defined default value + static + LLSD set_default(const std::string& group, const std::string& key); + + /// set control group.key to specified value + static + LLSD set(const std::string& group, const std::string& key, const LLSD& value); + + /// toggle boolean control group.key + static + LLSD toggle(const std::string& group, const std::string& key); + + /// get the definition for control group.key, (! isDefined()) if bad + /// ["name"], ["type"], ["value"], ["comment"] + static + LLSD get_def(const std::string& group, const std::string& key); + + /// get the value of control group.key + static + LLSD get(const std::string& group, const std::string& key); + + /// get defined groups + static + std::vector<std::string> get_groups(); + + /// get definitions for all variables in group + static + LLSD get_vars(const std::string& group); + + private: + static + LLSD access(const LLSD& params); + }; +} // namespace LL + +#endif /* ! defined(LL_COMMONCONTROL_H) */ diff --git a/indra/llcommon/hbxxh.cpp b/indra/llcommon/hbxxh.cpp new file mode 100644 index 0000000000..388269d6c8 --- /dev/null +++ b/indra/llcommon/hbxxh.cpp @@ -0,0 +1,377 @@ +/** + * @file hbxxh.cpp + * @brief High performances vectorized hashing based on xxHash. + * + * $LicenseInfo:firstyear=2023&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (c) 2023, Henri Beauchamp. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "linden_common.h" + +// This define ensures that xxHash will be compiled within this module, with +// vectorized (*) and inlined functions (with no exported API symbol); our +// xxhash "pre-built library" package actually only contains the xxhash.h +// header (no library needed at link time). +// (*) SSE2 is normally used for x86(_64) builds, unless you enabled AVX2 +// in your build, in which case the latter would be used instead. For ARM64 +// builds, this would also automatically enable NEON vectorization. +#define XXH_INLINE_ALL +#include "xxhash/xxhash.h" + +#include "hbxxh.h" + +// How many bytes to grab at a time when hashing files or streams +constexpr size_t BLOCK_LEN = 4096; + +/////////////////////////////////////////////////////////////////////////////// +// HBXXH64 class +/////////////////////////////////////////////////////////////////////////////// + +//static +U64 HBXXH64::digest(const void* buffer, size_t len) +{ + return XXH3_64bits(buffer, len); +} + +//static +U64 HBXXH64::digest(const char* str) +{ + return XXH3_64bits((const void*)str, strlen(str)); +} + +//static +U64 HBXXH64::digest(const std::string& str) +{ + return XXH3_64bits((const void*)str.c_str(), str.size()); +} + +// Must be called by all constructors. +void HBXXH64::init() +{ + mDigest = 0; + mState = (void*)XXH3_createState(); + if (!mState || XXH3_64bits_reset((XXH3_state_t*)mState) != XXH_OK) + { + LL_WARNS() << "Failed to initialize state !" << LL_ENDL; + } +} + +HBXXH64::~HBXXH64() +{ + if (mState) + { + XXH3_freeState((XXH3_state_t*)mState); + } +} + +void HBXXH64::update(const void* buffer, size_t len) +{ + if (mState) + { + XXH3_64bits_update((XXH3_state_t*)mState, buffer, len); + } + else + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + } +} + +void HBXXH64::update(const std::string& str) +{ + if (mState) + { + XXH3_64bits_update((XXH3_state_t*)mState, (const void*)str.c_str(), + str.length()); + } + else + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + } +} + +void HBXXH64::update(std::istream& stream) +{ + if (!mState) + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + return; + } + + char buffer[BLOCK_LEN]; + size_t len; + while (stream.good()) + { + stream.read(buffer, BLOCK_LEN); + len = stream.gcount(); + XXH3_64bits_update((XXH3_state_t*)mState, (const void*)buffer, len); + } +} + +void HBXXH64::update(FILE* file) +{ + if (!mState) + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + return; + } + + char buffer[BLOCK_LEN]; + size_t len; + while ((len = fread((void*)buffer, 1, BLOCK_LEN, file))) + { + XXH3_64bits_update((XXH3_state_t*)mState, (const void*)buffer, len); + } + fclose(file); +} + +void HBXXH64::finalize() +{ + if (!mState) + { + LL_WARNS() << "Already finalized !" << LL_ENDL; + return; + } + mDigest = XXH3_64bits_digest((XXH3_state_t*)mState); + XXH3_freeState((XXH3_state_t*)mState); + mState = NULL; +} + +U64 HBXXH64::digest() const +{ + return mState ? XXH3_64bits_digest((XXH3_state_t*)mState) : mDigest; +} + +std::ostream& operator<<(std::ostream& stream, HBXXH64 context) +{ + stream << context.digest(); + return stream; +} + +/////////////////////////////////////////////////////////////////////////////// +// HBXXH128 class +/////////////////////////////////////////////////////////////////////////////// + +//static +LLUUID HBXXH128::digest(const void* buffer, size_t len) +{ + XXH128_hash_t hash = XXH3_128bits(buffer, len); + LLUUID id; + U64* data = (U64*)id.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; + return id; +} + +//static +LLUUID HBXXH128::digest(const char* str) +{ + XXH128_hash_t hash = XXH3_128bits((const void*)str, strlen(str)); + LLUUID id; + U64* data = (U64*)id.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; + return id; +} + +//static +LLUUID HBXXH128::digest(const std::string& str) +{ + XXH128_hash_t hash = XXH3_128bits((const void*)str.c_str(), str.size()); + LLUUID id; + U64* data = (U64*)id.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; + return id; +} + +//static +void HBXXH128::digest(LLUUID& result, const void* buffer, size_t len) +{ + XXH128_hash_t hash = XXH3_128bits(buffer, len); + U64* data = (U64*)result.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; +} + +//static +void HBXXH128::digest(LLUUID& result, const char* str) +{ + XXH128_hash_t hash = XXH3_128bits((const void*)str, strlen(str)); + U64* data = (U64*)result.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; +} + +//static +void HBXXH128::digest(LLUUID& result, const std::string& str) +{ + XXH128_hash_t hash = XXH3_128bits((const void*)str.c_str(), str.size()); + U64* data = (U64*)result.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; +} + +// Must be called by all constructors. +void HBXXH128::init() +{ + mState = (void*)XXH3_createState(); + if (!mState || XXH3_128bits_reset((XXH3_state_t*)mState) != XXH_OK) + { + LL_WARNS() << "Failed to initialize state !" << LL_ENDL; + } +} + +HBXXH128::~HBXXH128() +{ + if (mState) + { + XXH3_freeState((XXH3_state_t*)mState); + } +} + +void HBXXH128::update(const void* buffer, size_t len) +{ + if (mState) + { + XXH3_128bits_update((XXH3_state_t*)mState, buffer, len); + } + else + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + } +} + +void HBXXH128::update(const std::string& str) +{ + if (mState) + { + XXH3_128bits_update((XXH3_state_t*)mState, (const void*)str.c_str(), + str.length()); + } + else + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + } +} + +void HBXXH128::update(std::istream& stream) +{ + if (!mState) + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + return; + } + + char buffer[BLOCK_LEN]; + size_t len; + while (stream.good()) + { + stream.read(buffer, BLOCK_LEN); + len = stream.gcount(); + XXH3_128bits_update((XXH3_state_t*)mState, (const void*)buffer, len); + } +} + +void HBXXH128::update(FILE* file) +{ + if (!mState) + { + LL_WARNS() << "Cannot update a finalized digest !" << LL_ENDL; + return; + } + + char buffer[BLOCK_LEN]; + size_t len; + while ((len = fread((void*)buffer, 1, BLOCK_LEN, file))) + { + XXH3_128bits_update((XXH3_state_t*)mState, (const void*)buffer, len); + } + fclose(file); +} + +void HBXXH128::finalize() +{ + if (!mState) + { + LL_WARNS() << "Already finalized !" << LL_ENDL; + return; + } + XXH128_hash_t hash = XXH3_128bits_digest((XXH3_state_t*)mState); + U64* data = (U64*)mDigest.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; + XXH3_freeState((XXH3_state_t*)mState); + mState = NULL; +} + +const LLUUID& HBXXH128::digest() const +{ + if (mState) + { + XXH128_hash_t hash = XXH3_128bits_digest((XXH3_state_t*)mState); + // We cheat the const-ness of the method here, but this is OK, since + // mDigest is private and cannot be accessed indirectly by other + // methods than digest() ones, that do check for mState to decide + // wether mDigest's current value may be provided as is or not. This + // cheat saves us a temporary LLLUID copy. + U64* data = (U64*)mDigest.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; + } + return mDigest; +} + +void HBXXH128::digest(LLUUID& result) const +{ + if (!mState) + { + result = mDigest; + return; + } + XXH128_hash_t hash = XXH3_128bits_digest((XXH3_state_t*)mState); + U64* data = (U64*)result.mData; + // Note: we do not check endianness here and we just store in the same + // order as XXH128_hash_t, that is low word "first". + data[0] = hash.low64; + data[1] = hash.high64; +} + +std::ostream& operator<<(std::ostream& stream, HBXXH128 context) +{ + stream << context.digest(); + return stream; +} diff --git a/indra/llcommon/hbxxh.h b/indra/llcommon/hbxxh.h new file mode 100644 index 0000000000..236716722a --- /dev/null +++ b/indra/llcommon/hbxxh.h @@ -0,0 +1,259 @@ +/** + * @file hbxxh.h + * @brief High performances vectorized hashing based on xxHash. + * + * $LicenseInfo:firstyear=2023&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (c) 2023, Henri Beauchamp. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_HBXXH_H +#define LL_HBXXH_H + +#include "lluuid.h" + +// HBXXH* classes are to be used where speed matters and cryptographic quality +// is not required (no "one-way" guarantee, though they are likely not worst in +// this respect than MD5 which got busted and is now considered too weak). The +// xxHash code they are built upon is vectorized and about 50 times faster than +// MD5. A 64 bits hash class is also provided for when 128 bits of entropy are +// not needed. The hashes collision rate is similar to MD5's. +// See https://github.com/Cyan4973/xxHash#readme for details. + +// 64 bits hashing class + +class HBXXH64 +{ + friend std::ostream& operator<<(std::ostream&, HBXXH64); + +protected: + LOG_CLASS(HBXXH64); + +public: + inline HBXXH64() { init(); } + + // Constructors for special circumstances; they all digest the first passed + // parameter. Set 'do_finalize' to false if you do not want to finalize the + // context, which is useful/needed when you want to update() it afterwards. + // Ideally, the compiler should be smart enough to get our clue and + // optimize out the const bool test during inlining... + + inline HBXXH64(const void* buffer, size_t len, + const bool do_finalize = true) + { + init(); + update(buffer, len); + if (do_finalize) + { + finalize(); + } + } + + inline HBXXH64(const std::string& str, const bool do_finalize = true) + { + init(); + update(str); + if (do_finalize) + { + finalize(); + } + } + + inline HBXXH64(std::istream& s, const bool do_finalize = true) + { + init(); + update(s); + if (do_finalize) + { + finalize(); + } + } + + inline HBXXH64(FILE* file, const bool do_finalize = true) + { + init(); + update(file); + if (do_finalize) + { + finalize(); + } + } + + ~HBXXH64(); + + void update(const void* buffer, size_t len); + void update(const std::string& str); + void update(std::istream& s); + void update(FILE* file); + + // Note that unlike what happens with LLMD5, you do not need to finalize() + // HBXXH64 before using digest(), and you may keep updating() it even after + // you got a first digest() (the next digest would of course change after + // any update). It is still useful to use finalize() when you do not want + // to store a final digest() result in a separate U64; after this method + // has been called, digest() simply returns mDigest value. + void finalize(); + + U64 digest() const; + + // Fast static methods. Use them when hashing just one contiguous block of + // data. + static U64 digest(const void* buffer, size_t len); + static U64 digest(const char* str); // str must be NUL-terminated + static U64 digest(const std::string& str); + +private: + void init(); + +private: + // We use a void pointer to avoid including xxhash.h here for XXH3_state_t + // (which cannot either be trivially forward-declared, due to complex API + // related pre-processor macros in xxhash.h). + void* mState; + U64 mDigest; +}; + +inline bool operator==(const HBXXH64& a, const HBXXH64& b) +{ + return a.digest() == b.digest(); +} + +inline bool operator!=(const HBXXH64& a, const HBXXH64& b) +{ + return a.digest() != b.digest(); +} + +// 128 bits hashing class + +class HBXXH128 +{ + friend std::ostream& operator<<(std::ostream&, HBXXH128); + +protected: + LOG_CLASS(HBXXH128); + +public: + inline HBXXH128() { init(); } + + // Constructors for special circumstances; they all digest the first passed + // parameter. Set 'do_finalize' to false if you do not want to finalize the + // context, which is useful/needed when you want to update() it afterwards. + // Ideally, the compiler should be smart enough to get our clue and + // optimize out the const bool test during inlining... + + inline HBXXH128(const void* buffer, size_t len, + const bool do_finalize = true) + { + init(); + update(buffer, len); + if (do_finalize) + { + finalize(); + } + } + + inline HBXXH128(const std::string& str, const bool do_finalize = true) + { + init(); + update(str); + if (do_finalize) + { + finalize(); + } + } + + inline HBXXH128(std::istream& s, const bool do_finalize = true) + { + init(); + update(s); + if (do_finalize) + { + finalize(); + } + } + + inline HBXXH128(FILE* file, const bool do_finalize = true) + { + init(); + update(file); + if (do_finalize) + { + finalize(); + } + } + + ~HBXXH128(); + + void update(const void* buffer, size_t len); + void update(const std::string& str); + void update(std::istream& s); + void update(FILE* file); + + // Note that unlike what happens with LLMD5, you do not need to finalize() + // HBXXH128 before using digest(), and you may keep updating() it even + // after you got a first digest() (the next digest would of course change + // after any update). It is still useful to use finalize() when you do not + // want to store a final digest() result in a separate LLUUID; after this + // method has been called, digest() simply returns a reference on mDigest. + void finalize(); + + // We use an LLUUID for the digest, since this is a 128 bits wide native + // type available in the viewer code, making it easy to manipulate. It also + // allows to use HBXXH128 efficiently in LLUUID generate() and combine() + // methods. + const LLUUID& digest() const; + + // Here, we avoid an LLUUID copy whenever we already got one to store the + // result *and* we did not yet call finalize(). + void digest(LLUUID& result) const; + + // Fast static methods. Use them when hashing just one contiguous block of + // data. + static LLUUID digest(const void* buffer, size_t len); + static LLUUID digest(const char* str); // str must be NUL-terminated + static LLUUID digest(const std::string& str); + // Same as above, but saves you from an LLUUID copy when you already got + // one for storage use. + static void digest(LLUUID& result, const void* buffer, size_t len); + static void digest(LLUUID& result, const char* str); // str NUL-terminated + static void digest(LLUUID& result, const std::string& str); + +private: + void init(); + +private: + // We use a void pointer to avoid including xxhash.h here for XXH3_state_t + // (which cannot either be trivially forward-declared, due to complex API + // related pre-processor macros in xxhash.h). + void* mState; + LLUUID mDigest; +}; + +inline bool operator==(const HBXXH128& a, const HBXXH128& b) +{ + return a.digest() == b.digest(); +} + +inline bool operator!=(const HBXXH128& a, const HBXXH128& b) +{ + return a.digest() != b.digest(); +} + +#endif // LL_HBXXH_H diff --git a/indra/llcommon/llapr.cpp b/indra/llcommon/llapr.cpp index 435531f86f..e49f72722b 100644 --- a/indra/llcommon/llapr.cpp +++ b/indra/llcommon/llapr.cpp @@ -522,6 +522,7 @@ S32 LLAPRFile::seek(apr_file_t* file_handle, apr_seek_where_t where, S32 offset) //static S32 LLAPRFile::readEx(const std::string& filename, void *buf, S32 offset, S32 nbytes, LLVolatileAPRPool* pool) { + LL_PROFILE_ZONE_SCOPED; //***************************************** LLAPRFilePoolScope scope(pool); apr_file_t* file_handle = open(filename, scope.getVolatileAPRPool(), APR_READ|APR_BINARY); @@ -566,6 +567,7 @@ S32 LLAPRFile::readEx(const std::string& filename, void *buf, S32 offset, S32 nb //static S32 LLAPRFile::writeEx(const std::string& filename, void *buf, S32 offset, S32 nbytes, LLVolatileAPRPool* pool) { + LL_PROFILE_ZONE_SCOPED; apr_int32_t flags = APR_CREATE|APR_WRITE|APR_BINARY; if (offset < 0) { diff --git a/indra/llcommon/llassettype.cpp b/indra/llcommon/llassettype.cpp index e6cc06e8d0..f08cc18036 100644 --- a/indra/llcommon/llassettype.cpp +++ b/indra/llcommon/llassettype.cpp @@ -96,6 +96,7 @@ LLAssetDictionary::LLAssetDictionary() addEntry(LLAssetType::AT_WIDGET, new AssetEntry("WIDGET", "widget", "widget", false, false, false)); addEntry(LLAssetType::AT_PERSON, new AssetEntry("PERSON", "person", "person", false, false, false)); addEntry(LLAssetType::AT_SETTINGS, new AssetEntry("SETTINGS", "settings", "settings blob", true, true, true)); + addEntry(LLAssetType::AT_MATERIAL, new AssetEntry("MATERIAL", "material", "render material", true, true, true)); addEntry(LLAssetType::AT_UNKNOWN, new AssetEntry("UNKNOWN", "invalid", NULL, false, false, false)); addEntry(LLAssetType::AT_NONE, new AssetEntry("NONE", "-1", NULL, FALSE, FALSE, FALSE)); diff --git a/indra/llcommon/llassettype.h b/indra/llcommon/llassettype.h index 652c548d59..e8df8574f7 100644 --- a/indra/llcommon/llassettype.h +++ b/indra/llcommon/llassettype.h @@ -127,8 +127,9 @@ public: AT_RESERVED_6 = 55, AT_SETTINGS = 56, // Collection of settings - - AT_COUNT = 57, + AT_MATERIAL = 57, // Render Material + + AT_COUNT = 58, // +*********************************************************+ // | TO ADD AN ELEMENT TO THIS ENUM: | diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index d2c4e66160..6e988260a9 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -37,12 +37,13 @@ thread_local bool gProfilerEnabled = false; #if (TRACY_ENABLE) // Override new/delete for tracy memory profiling -void *operator new(size_t size) + +void* ll_tracy_new(size_t size) { void* ptr; if (gProfilerEnabled) { - LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; + //LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; ptr = (malloc)(size); } else @@ -57,12 +58,22 @@ void *operator new(size_t size) return ptr; } -void operator delete(void *ptr) noexcept +void* operator new(size_t size) +{ + return ll_tracy_new(size); +} + +void* operator new[](std::size_t count) +{ + return ll_tracy_new(count); +} + +void ll_tracy_delete(void* ptr) { TracyFree(ptr); if (gProfilerEnabled) { - LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; + //LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; (free)(ptr); } else @@ -71,6 +82,16 @@ void operator delete(void *ptr) noexcept } } +void operator delete(void *ptr) noexcept +{ + ll_tracy_delete(ptr); +} + +void operator delete[](void* ptr) noexcept +{ + ll_tracy_delete(ptr); +} + // C-style malloc/free can't be so easily overridden, so we define tracy versions and use // a pre-processor #define in linden_common.h to redirect to them. The parens around the native // functions below prevents recursive substitution by the preprocessor. diff --git a/indra/llcommon/llerror.h b/indra/llcommon/llerror.h index d06c0e2132..b7dec3cb7f 100644 --- a/indra/llcommon/llerror.h +++ b/indra/llcommon/llerror.h @@ -82,9 +82,11 @@ const int LL_ERR_NOERR = 0; #ifdef SHOW_ASSERT #define llassert(func) llassert_always_msg(func, #func) +#define llassert_msg(func, msg) llassert_always_msg(func, msg) #define llverify(func) llassert_always_msg(func, #func) #else #define llassert(func) +#define llassert_msg(func, msg) #define llverify(func) do {if (func) {}} while(0) #endif diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp index c54029e8b4..1e9920746b 100644 --- a/indra/llcommon/llframetimer.cpp +++ b/indra/llcommon/llframetimer.cpp @@ -29,11 +29,6 @@ #include "llframetimer.h" -// We don't bother building a stand alone lib; we just need to include the one source file for Tracy support -#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER - #include "TracyClient.cpp" -#endif // LL_PROFILER_CONFIGURATION - // Static members //LLTimer LLFrameTimer::sInternalTimer; U64 LLFrameTimer::sStartTotalTime = totalTime(); diff --git a/indra/llcommon/llinstancetracker.h b/indra/llcommon/llinstancetracker.h index 02535a59e7..97f7817e74 100644 --- a/indra/llcommon/llinstancetracker.h +++ b/indra/llcommon/llinstancetracker.h @@ -104,22 +104,26 @@ public: return LockStatic()->mMap.size(); } - // snapshot of std::pair<const KEY, std::shared_ptr<T>> pairs - class snapshot + // snapshot of std::pair<const KEY, std::shared_ptr<SUBCLASS>> pairs, for + // some SUBCLASS derived from T + template <typename SUBCLASS> + class snapshot_of { // It's very important that what we store in this snapshot are // weak_ptrs, NOT shared_ptrs. That's how we discover whether any // instance has been deleted during the lifespan of a snapshot. typedef std::vector<std::pair<const KEY, weak_t>> VectorType; - // Dereferencing our iterator produces a std::shared_ptr for each - // instance that still exists. Since we store weak_ptrs, that involves - // two chained transformations: + // Dereferencing the iterator we publish produces a + // std::shared_ptr<SUBCLASS> for each instance that still exists. + // Since we store weak_ptr<T>, that involves two chained + // transformations: // - a transform_iterator to lock the weak_ptr and return a shared_ptr - // - a filter_iterator to skip any shared_ptr that has become invalid. + // - a filter_iterator to skip any shared_ptr<T> that has become + // invalid or references any T instance that isn't SUBCLASS. // It is very important that we filter lazily, that is, during // traversal. Any one of our stored weak_ptrs might expire during // traversal. - typedef std::pair<const KEY, ptr_t> strong_pair; + typedef std::pair<const KEY, std::shared_ptr<SUBCLASS>> strong_pair; // Note for future reference: nat has not yet had any luck (up to // Boost 1.67) trying to use boost::transform_iterator with a hand- // coded functor, only with actual functions. In my experience, an @@ -127,7 +131,7 @@ public: // result_type typedef. But this works. static strong_pair strengthen(typename VectorType::value_type& pair) { - return { pair.first, pair.second.lock() }; + return { pair.first, std::dynamic_pointer_cast<SUBCLASS>(pair.second.lock()) }; } static bool dead_skipper(const strong_pair& pair) { @@ -135,7 +139,7 @@ public: } public: - snapshot(): + snapshot_of(): // populate our vector with a snapshot of (locked!) InstanceMap // note, this assigns pair<KEY, shared_ptr> to pair<KEY, weak_ptr> mData(mLock->mMap.begin(), mLock->mMap.end()) @@ -184,44 +188,51 @@ public: #endif // LL_WINDOWS VectorType mData; }; + using snapshot = snapshot_of<T>; - // iterate over this for references to each instance - class instance_snapshot: public snapshot + // iterate over this for references to each SUBCLASS instance + template <typename SUBCLASS> + class instance_snapshot_of: public snapshot_of<SUBCLASS> { private: - static T& instance_getter(typename snapshot::iterator::reference pair) + using super = snapshot_of<SUBCLASS>; + static T& instance_getter(typename super::iterator::reference pair) { return *pair.second; } public: typedef boost::transform_iterator<decltype(instance_getter)*, - typename snapshot::iterator> iterator; - iterator begin() { return iterator(snapshot::begin(), instance_getter); } - iterator end() { return iterator(snapshot::end(), instance_getter); } + typename super::iterator> iterator; + iterator begin() { return iterator(super::begin(), instance_getter); } + iterator end() { return iterator(super::end(), instance_getter); } void deleteAll() { - for (auto it(snapshot::begin()), end(snapshot::end()); it != end; ++it) + for (auto it(super::begin()), end(super::end()); it != end; ++it) { delete it->second.get(); } } - }; + }; + using instance_snapshot = instance_snapshot_of<T>; // iterate over this for each key - class key_snapshot: public snapshot + template <typename SUBCLASS> + class key_snapshot_of: public snapshot_of<SUBCLASS> { private: - static KEY key_getter(typename snapshot::iterator::reference pair) + using super = snapshot_of<SUBCLASS>; + static KEY key_getter(typename super::iterator::reference pair) { return pair.first; } public: typedef boost::transform_iterator<decltype(key_getter)*, - typename snapshot::iterator> iterator; - iterator begin() { return iterator(snapshot::begin(), key_getter); } - iterator end() { return iterator(snapshot::end(), key_getter); } + typename super::iterator> iterator; + iterator begin() { return iterator(super::begin(), key_getter); } + iterator end() { return iterator(super::end(), key_getter); } }; + using key_snapshot = key_snapshot_of<T>; static ptr_t getInstance(const KEY& k) { @@ -368,22 +379,25 @@ public: return LockStatic()->mSet.size(); } - // snapshot of std::shared_ptr<T> pointers - class snapshot + // snapshot of std::shared_ptr<SUBCLASS> pointers + template <typename SUBCLASS> + class snapshot_of { // It's very important that what we store in this snapshot are // weak_ptrs, NOT shared_ptrs. That's how we discover whether any // instance has been deleted during the lifespan of a snapshot. typedef std::vector<weak_t> VectorType; - // Dereferencing our iterator produces a std::shared_ptr for each - // instance that still exists. Since we store weak_ptrs, that involves - // two chained transformations: + // Dereferencing the iterator we publish produces a + // std::shared_ptr<SUBCLASS> for each instance that still exists. + // Since we store weak_ptrs, that involves two chained + // transformations: // - a transform_iterator to lock the weak_ptr and return a shared_ptr - // - a filter_iterator to skip any shared_ptr that has become invalid. - typedef std::shared_ptr<T> strong_ptr; + // - a filter_iterator to skip any shared_ptr that has become invalid + // or references any T instance that isn't SUBCLASS. + typedef std::shared_ptr<SUBCLASS> strong_ptr; static strong_ptr strengthen(typename VectorType::value_type& ptr) { - return ptr.lock(); + return std::dynamic_pointer_cast<SUBCLASS>(ptr.lock()); } static bool dead_skipper(const strong_ptr& ptr) { @@ -391,7 +405,7 @@ public: } public: - snapshot(): + snapshot_of(): // populate our vector with a snapshot of (locked!) InstanceSet // note, this assigns stored shared_ptrs to weak_ptrs for snapshot mData(mLock->mSet.begin(), mLock->mSet.end()) @@ -437,22 +451,33 @@ public: #endif // LL_WINDOWS VectorType mData; }; + using snapshot = snapshot_of<T>; // iterate over this for references to each instance - struct instance_snapshot: public snapshot + template <typename SUBCLASS> + class instance_snapshot_of: public snapshot_of<SUBCLASS> { - typedef boost::indirect_iterator<typename snapshot::iterator> iterator; - iterator begin() { return iterator(snapshot::begin()); } - iterator end() { return iterator(snapshot::end()); } + private: + using super = snapshot_of<SUBCLASS>; + + public: + typedef boost::indirect_iterator<typename super::iterator> iterator; + iterator begin() { return iterator(super::begin()); } + iterator end() { return iterator(super::end()); } void deleteAll() { - for (auto it(snapshot::begin()), end(snapshot::end()); it != end; ++it) + for (auto it(super::begin()), end(super::end()); it != end; ++it) { delete it->get(); } } }; + using instance_snapshot = instance_snapshot_of<T>; + // key_snapshot_of isn't really meaningful, but define it anyway to avoid + // requiring two different LLInstanceTrackerSubclass implementations. + template <typename SUBCLASS> + using key_snapshot_of = instance_snapshot_of<SUBCLASS>; protected: LLInstanceTracker() diff --git a/indra/llcommon/llinstancetrackersubclass.h b/indra/llcommon/llinstancetrackersubclass.h new file mode 100644 index 0000000000..ea9a38200f --- /dev/null +++ b/indra/llcommon/llinstancetrackersubclass.h @@ -0,0 +1,98 @@ +/** + * @file llinstancetrackersubclass.h + * @author Nat Goodspeed + * @date 2022-12-09 + * @brief Intermediate class to get subclass-specific types from + * LLInstanceTracker instance-retrieval methods. + * + * $LicenseInfo:firstyear=2022&license=viewerlgpl$ + * Copyright (c) 2022, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_LLINSTANCETRACKERSUBCLASS_H) +#define LL_LLINSTANCETRACKERSUBCLASS_H + +#include <memory> // std::shared_ptr, std::weak_ptr + +/** + * Derive your subclass S of a subclass T of LLInstanceTracker<T> from + * LLInstanceTrackerSubclass<S, T> to perform appropriate downcasting and + * filtering for LLInstanceTracker access methods. + * + * LLInstanceTracker<T> uses CRTP, so that getWeak(), getInstance(), snapshot + * and instance_snapshot return pointers and references to T. The trouble is + * that subclasses T0 and T1 derived from T also get pointers and references + * to their base class T, requiring explicit downcasting. Moreover, + * T0::getInstance() shouldn't find an instance of any T subclass other than + * T0. Nor should T0::snapshot. + * + * @code + * class Tracked: public LLInstanceTracker<Tracked, std::string> + * { + * private: + * using super = LLInstanceTracker<Tracked, std::string>; + * public: + * Tracked(const std::string& name): super(name) {} + * // All references to Tracked::ptr_t, Tracked::getInstance() etc. + * // appropriately use Tracked. + * // ... + * }; + * + * // But now we derive SubTracked from Tracked. We need SubTracked::ptr_t, + * // SubTracked::getInstance() etc. to use SubTracked, not Tracked. + * // This LLInstanceTrackerSubclass specialization is itself derived from + * // Tracked. + * class SubTracked: public LLInstanceTrackerSubclass<SubTracked, Tracked> + * { + * private: + * using super = LLInstanceTrackerSubclass<SubTracked, Tracked>; + * public: + * // LLInstanceTrackerSubclass's constructor forwards to Tracked's. + * SubTracked(const std::string& name): super(name) {} + * // SubTracked::getInstance() returns std::shared_ptr<SubTracked>, etc. + * // ... + * @endcode + */ +template <typename SUBCLASS, typename T> +class LLInstanceTrackerSubclass: public T +{ +public: + using ptr_t = std::shared_ptr<SUBCLASS>; + using weak_t = std::weak_ptr<SUBCLASS>; + + // forward any constructor call to the corresponding T ctor + template <typename... ARGS> + LLInstanceTrackerSubclass(ARGS&&... args): + T(std::forward<ARGS>(args)...) + {} + + weak_t getWeak() + { + // call base-class getWeak(), try to lock, downcast to SUBCLASS + return std::dynamic_pointer_cast<SUBCLASS>(T::getWeak().lock()); + } + + template <typename KEY> + static ptr_t getInstance(const KEY& k) + { + return std::dynamic_pointer_cast<SUBCLASS>(T::getInstance(k)); + } + + using snapshot = typename T::template snapshot_of<SUBCLASS>; + using instance_snapshot = typename T::template instance_snapshot_of<SUBCLASS>; + using key_snapshot = typename T::template key_snapshot_of<SUBCLASS>; + + static size_t instanceCount() + { + // T::instanceCount() lies because our snapshot, et al., won't + // necessarily return all the T instances -- only those that are also + // SUBCLASS instances. Count those. + size_t count = 0; + for (const auto& pair : snapshot()) + ++count; + return count; + } +}; + +#endif /* ! defined(LL_LLINSTANCETRACKERSUBCLASS_H) */ diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp index d6ae1284d3..7cdf7254ff 100644 --- a/indra/llcommon/llmemory.cpp +++ b/indra/llcommon/llmemory.cpp @@ -35,6 +35,7 @@ # include <sys/types.h> # include <mach/task.h> # include <mach/mach_init.h> +#include <mach/mach_host.h> #elif LL_LINUX # include <unistd.h> #endif @@ -109,6 +110,50 @@ void LLMemory::updateMemoryInfo() { sAvailPhysicalMemInKB = U32Kilobytes(0); } + +#elif defined(LL_DARWIN) + task_vm_info info; + mach_msg_type_number_t infoCount = TASK_VM_INFO_COUNT; + // MACH_TASK_BASIC_INFO reports the same resident_size, but does not tell us the reusable bytes or phys_footprint. + if (task_info(mach_task_self(), TASK_VM_INFO, reinterpret_cast<task_info_t>(&info), &infoCount) == KERN_SUCCESS) + { + // Our Windows definition of PagefileUsage is documented by Microsoft as "the total amount of + // memory that the memory manager has committed for a running process", which is rss. + sAllocatedPageSizeInKB = U32Bytes(info.resident_size); + + // Activity Monitor => Inspect Process => Real Memory Size appears to report resident_size + // Activity monitor => main window memory column appears to report phys_footprint, which spot checks as at least 30% less. + // I think that is because of compression, which isn't going to give us a consistent measurement. We want uncompressed totals. + // + // In between is resident_size - reusable. This is what Chrome source code uses, with source comments saying it is 'the "Real Memory" value + // reported for the app by the Memory Monitor in Instruments.' It is still about 8% bigger than phys_footprint. + // + // (On Windows, we use WorkingSetSize.) + sAllocatedMemInKB = U32Bytes(info.resident_size - info.reusable); + } + else + { + LL_WARNS() << "task_info failed" << LL_ENDL; + } + + // Total installed and available physical memory are properties of the host, not just our process. + vm_statistics64_data_t vmstat; + mach_msg_type_number_t count = HOST_VM_INFO64_COUNT; + mach_port_t host = mach_host_self(); + vm_size_t page_size; + host_page_size(host, &page_size); + kern_return_t result = host_statistics64(host, HOST_VM_INFO64, reinterpret_cast<host_info_t>(&vmstat), &count); + if (result == KERN_SUCCESS) { + // This is what Chrome reports as 'the "Physical Memory Free" value reported by the Memory Monitor in Instruments.' + // Note though that inactive pages are not included here and not yet free, but could become so under memory pressure. + sAvailPhysicalMemInKB = U32Bytes(vmstat.free_count * page_size); + sMaxPhysicalMemInKB = LLMemoryInfo::getHardwareMemSize(); + } + else + { + LL_WARNS() << "task_info failed" << LL_ENDL; + } + #else //not valid for other systems for now. sAllocatedMemInKB = U64Bytes(LLMemory::getCurrentRSS()); diff --git a/indra/llcommon/llpointer.h b/indra/llcommon/llpointer.h index 9a6453ea48..f9de0c7929 100644 --- a/indra/llcommon/llpointer.h +++ b/indra/llcommon/llpointer.h @@ -340,4 +340,28 @@ private: bool mStayUnique; }; + +// boost hash adapter +template <class Type> +struct boost::hash<LLPointer<Type>> +{ + typedef LLPointer<Type> argument_type; + typedef std::size_t result_type; + result_type operator()(argument_type const& s) const + { + return (std::size_t) s.get(); + } +}; + +// Adapt boost hash to std hash +namespace std +{ + template<class Type> struct hash<LLPointer<Type>> + { + std::size_t operator()(LLPointer<Type> const& s) const noexcept + { + return boost::hash<LLPointer<Type>>()(s); + } + }; +} #endif diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index f9d7ae7ce4..736a069f49 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -86,8 +86,12 @@ extern thread_local bool gProfilerEnabled; #define TRACY_ONLY_IPV4 1 #include "Tracy.hpp" - // Mutually exclusive with detailed memory tracing + // Enable OpenGL profiling #define LL_PROFILER_ENABLE_TRACY_OPENGL 0 + + // Enable RenderDoc labeling + #define LL_PROFILER_ENABLE_RENDER_DOC 0 + #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY @@ -104,8 +108,6 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red - #define LL_PROFILE_ALLOC(ptr, size) TracyAlloc(ptr, size) - #define LL_PROFILE_FREE(ptr) TracyFree(ptr) #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER #define LL_PROFILER_FRAME_END @@ -121,8 +123,6 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILE_ZONE_ERR(name) (void)(name); // Not supported #define LL_PROFILE_ZONE_INFO(name) (void)(name); // Not supported #define LL_PROFILE_ZONE_WARN(name) (void)(name); // Not supported - #define LL_PROFILE_ALLOC(ptr, size) (void)(ptr); (void)(size); - #define LL_PROFILE_FREE(ptr) (void)(ptr); #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define LL_PROFILER_FRAME_END FrameMark @@ -138,14 +138,45 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red - #define LL_PROFILE_ALLOC(ptr, size) TracyAlloc(ptr, size) - #define LL_PROFILE_FREE(ptr) TracyFree(ptr) #endif #else #define LL_PROFILER_FRAME_END #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) #endif // LL_PROFILER +#if LL_PROFILER_ENABLE_TRACY_OPENGL +#define LL_PROFILE_GPU_ZONE(name) TracyGpuZone(name) +#define LL_PROFILE_GPU_ZONEC(name,color) TracyGpuZoneC(name,color) +#define LL_PROFILER_GPU_COLLECT TracyGpuCollect +#define LL_PROFILER_GPU_CONTEXT TracyGpuContext + +// disable memory tracking (incompatible with GPU tracing +#define LL_PROFILE_ALLOC(ptr, size) (void)(ptr); (void)(size); +#define LL_PROFILE_FREE(ptr) (void)(ptr); +#else +#define LL_PROFILE_GPU_ZONE(name) (void)name; +#define LL_PROFILE_GPU_ZONEC(name,color) (void)name;(void)color; +#define LL_PROFILER_GPU_COLLECT +#define LL_PROFILER_GPU_CONTEXT + +#define LL_LABEL_OBJECT_GL(type, name, length, label) + +#if LL_PROFILER_CONFIGURATION > 1 +#define LL_PROFILE_ALLOC(ptr, size) TracyAlloc(ptr, size) +#define LL_PROFILE_FREE(ptr) TracyFree(ptr) +#else +#define LL_PROFILE_ALLOC(ptr, size) (void)(ptr); (void)(size); +#define LL_PROFILE_FREE(ptr) (void)(ptr); +#endif + +#endif + +#if LL_PROFILER_ENABLE_RENDER_DOC +#define LL_LABEL_OBJECT_GL(type, name, length, label) glObjectLabel(type, name, length, label) +#else +#define LL_LABEL_OBJECT_GL(type, name, length, label) +#endif + #include "llprofilercategories.h" #endif // LL_PROFILER_H diff --git a/indra/llcommon/llprofilercategories.h b/indra/llcommon/llprofilercategories.h index 8db29468cc..617431f629 100644 --- a/indra/llcommon/llprofilercategories.h +++ b/indra/llcommon/llprofilercategories.h @@ -52,7 +52,7 @@ #define LL_PROFILER_CATEGORY_ENABLE_LOGGING 1 #define LL_PROFILER_CATEGORY_ENABLE_MATERIAL 1 #define LL_PROFILER_CATEGORY_ENABLE_MEDIA 1 -#define LL_PROFILER_CATEGORY_ENABLE_MEMORY 1 +#define LL_PROFILER_CATEGORY_ENABLE_MEMORY 0 #define LL_PROFILER_CATEGORY_ENABLE_NETWORK 1 #define LL_PROFILER_CATEGORY_ENABLE_OCTREE 1 #define LL_PROFILER_CATEGORY_ENABLE_PIPELINE 1 diff --git a/indra/llcommon/llqueuedthread.cpp b/indra/llcommon/llqueuedthread.cpp index 8cef4293cd..9b1de2e9a5 100644 --- a/indra/llcommon/llqueuedthread.cpp +++ b/indra/llcommon/llqueuedthread.cpp @@ -26,20 +26,26 @@ #include "linden_common.h" #include "llqueuedthread.h" +#include <chrono> + #include "llstl.h" #include "lltimer.h" // ms_sleep() -#include "lltracethreadrecorder.h" +#include "llmutex.h" //============================================================================ // MAIN THREAD LLQueuedThread::LLQueuedThread(const std::string& name, bool threaded, bool should_pause) : - LLThread(name), - mThreaded(threaded), - mIdleThread(TRUE), - mNextHandle(0), - mStarted(FALSE) + LLThread(name), + mIdleThread(TRUE), + mNextHandle(0), + mStarted(FALSE), + mThreaded(threaded), + mRequestQueue(name, 1024 * 1024) { + llassert(threaded); // not threaded implementation is deprecated + mMainQueue = LL::WorkQueue::getInstance("mainloop"); + if (mThreaded) { if(should_pause) @@ -69,6 +75,11 @@ void LLQueuedThread::shutdown() unpause(); // MAIN THREAD if (mThreaded) { + if (mRequestQueue.size() == 0) + { + mRequestQueue.close(); + } + S32 timeout = 100; for ( ; timeout>0; timeout--) { @@ -104,6 +115,8 @@ void LLQueuedThread::shutdown() { LL_WARNS() << "~LLQueuedThread() called with active requests: " << active_count << LL_ENDL; } + + mRequestQueue.close(); } //---------------------------------------------------------------------------- @@ -112,6 +125,7 @@ void LLQueuedThread::shutdown() // virtual S32 LLQueuedThread::update(F32 max_time_ms) { + LL_PROFILE_ZONE_SCOPED; if (!mStarted) { if (!mThreaded) @@ -125,29 +139,34 @@ S32 LLQueuedThread::update(F32 max_time_ms) S32 LLQueuedThread::updateQueue(F32 max_time_ms) { - F64 max_time = (F64)max_time_ms * .001; - LLTimer timer; - S32 pending = 1; - + LL_PROFILE_ZONE_SCOPED; // Frame Update if (mThreaded) { - pending = getPending(); - if(pending > 0) + // schedule a call to threadedUpdate for every call to updateQueue + if (!isQuitting()) + { + mRequestQueue.postIfOpen([=]() + { + LL_PROFILE_ZONE_NAMED_CATEGORY_THREAD("qt - update"); + mIdleThread = FALSE; + threadedUpdate(); + mIdleThread = TRUE; + } + ); + } + + if(getPending() > 0) { - unpause(); - } + unpause(); + } } else { - while (pending > 0) - { - pending = processNextRequest(); - if (max_time && timer.getElapsedTimeF64() > max_time) - break; - } + mRequestQueue.runFor(std::chrono::microseconds((int) (max_time_ms*1000.f))); + threadedUpdate(); } - return pending; + return getPending(); } void LLQueuedThread::incQueue() @@ -166,11 +185,7 @@ void LLQueuedThread::incQueue() // May be called from any thread S32 LLQueuedThread::getPending() { - S32 res; - lockData(); - res = mRequestQueue.size(); - unlockData(); - return res; + return mRequestQueue.size(); } // MAIN thread @@ -195,35 +210,28 @@ void LLQueuedThread::waitOnPending() // MAIN thread void LLQueuedThread::printQueueStats() { - lockData(); - if (!mRequestQueue.empty()) + U32 size = mRequestQueue.size(); + if (size > 0) { - QueuedRequest *req = *mRequestQueue.begin(); - LL_INFOS() << llformat("Pending Requests:%d Current status:%d", mRequestQueue.size(), req->getStatus()) << LL_ENDL; + LL_INFOS() << llformat("Pending Requests:%d ", mRequestQueue.size()) << LL_ENDL; } else { LL_INFOS() << "Queued Thread Idle" << LL_ENDL; } - unlockData(); } // MAIN thread LLQueuedThread::handle_t LLQueuedThread::generateHandle() { - lockData(); - while ((mNextHandle == nullHandle()) || (mRequestHash.find(mNextHandle))) - { - mNextHandle++; - } - const LLQueuedThread::handle_t res = mNextHandle++; - unlockData(); + U32 res = ++mNextHandle; return res; } // MAIN thread bool LLQueuedThread::addRequest(QueuedRequest* req) { + LL_PROFILE_ZONE_SCOPED; if (mStatus == QUITTING) { return false; @@ -231,14 +239,14 @@ bool LLQueuedThread::addRequest(QueuedRequest* req) lockData(); req->setStatus(STATUS_QUEUED); - mRequestQueue.insert(req); - mRequestHash.insert(req); + mRequestHash.insert(req); #if _DEBUG // LL_INFOS() << llformat("LLQueuedThread::Added req [%08d]",handle) << LL_ENDL; #endif unlockData(); - incQueue(); + llassert(!mDataLock->isSelfLocked()); + mRequestQueue.post([this, req]() { processRequest(req); }); return true; } @@ -246,6 +254,7 @@ bool LLQueuedThread::addRequest(QueuedRequest* req) // MAIN thread bool LLQueuedThread::waitForResult(LLQueuedThread::handle_t handle, bool auto_complete) { + LL_PROFILE_ZONE_SCOPED; llassert (handle != nullHandle()); bool res = false; bool waspaused = isPaused(); @@ -312,6 +321,7 @@ LLQueuedThread::status_t LLQueuedThread::getRequestStatus(handle_t handle) void LLQueuedThread::abortRequest(handle_t handle, bool autocomplete) { + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lockData(); QueuedRequest* req = (QueuedRequest*)mRequestHash.find(handle); if (req) @@ -333,30 +343,9 @@ void LLQueuedThread::setFlags(handle_t handle, U32 flags) unlockData(); } -void LLQueuedThread::setPriority(handle_t handle, U32 priority) -{ - lockData(); - QueuedRequest* req = (QueuedRequest*)mRequestHash.find(handle); - if (req) - { - if(req->getStatus() == STATUS_INPROGRESS) - { - // not in list - req->setPriority(priority); - } - else if(req->getStatus() == STATUS_QUEUED) - { - // remove from list then re-insert - llverify(mRequestQueue.erase(req) == 1); - req->setPriority(priority); - mRequestQueue.insert(req); - } - } - unlockData(); -} - bool LLQueuedThread::completeRequest(handle_t handle) { + LL_PROFILE_ZONE_SCOPED; bool res = false; lockData(); QueuedRequest* req = (QueuedRequest*)mRequestHash.find(handle); @@ -399,89 +388,120 @@ bool LLQueuedThread::check() //============================================================================ // Runs on its OWN thread -S32 LLQueuedThread::processNextRequest() +void LLQueuedThread::processRequest(LLQueuedThread::QueuedRequest* req) { - QueuedRequest *req; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + + mIdleThread = FALSE; + //threadedUpdate(); + // Get next request from pool lockData(); - while(1) + if ((req->getFlags() & FLAG_ABORT) || (mStatus == QUITTING)) { - req = NULL; - if (mRequestQueue.empty()) + LL_PROFILE_ZONE_NAMED_CATEGORY_THREAD("qtpr - abort"); + req->setStatus(STATUS_ABORTED); + req->finishRequest(false); + if (req->getFlags() & FLAG_AUTO_COMPLETE) { - break; - } - req = *mRequestQueue.begin(); - mRequestQueue.erase(mRequestQueue.begin()); - if ((req->getFlags() & FLAG_ABORT) || (mStatus == QUITTING)) - { - req->setStatus(STATUS_ABORTED); - req->finishRequest(false); - if (req->getFlags() & FLAG_AUTO_COMPLETE) - { - mRequestHash.erase(req); - req->deleteRequest(); -// check(); - } - continue; - } - llassert_always(req->getStatus() == STATUS_QUEUED); - break; - } - U32 start_priority = 0 ; - if (req) - { - req->setStatus(STATUS_INPROGRESS); - start_priority = req->getPriority(); - } - unlockData(); - - // This is the only place we will call req->setStatus() after - // it has initially been seet to STATUS_QUEUED, so it is - // safe to access req. - if (req) - { - // process request - bool complete = req->processRequest(); - - if (complete) - { - lockData(); - req->setStatus(STATUS_COMPLETE); - req->finishRequest(true); - if (req->getFlags() & FLAG_AUTO_COMPLETE) - { - mRequestHash.erase(req); - req->deleteRequest(); + mRequestHash.erase(req); + req->deleteRequest(); // check(); - } - unlockData(); } - else - { - lockData(); - req->setStatus(STATUS_QUEUED); - mRequestQueue.insert(req); - unlockData(); - if (mThreaded && start_priority < PRIORITY_NORMAL) - { - ms_sleep(1); // sleep the thread a little - } - } - - LLTrace::get_thread_recorder()->pushToParent(); + unlockData(); } + else + { + llassert_always(req->getStatus() == STATUS_QUEUED); + + if (req) + { + req->setStatus(STATUS_INPROGRESS); + } + unlockData(); + + // This is the only place we will call req->setStatus() after + // it has initially been seet to STATUS_QUEUED, so it is + // safe to access req. + if (req) + { + // process request + bool complete = req->processRequest(); + + if (complete) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_THREAD("qtpr - complete"); + lockData(); + req->setStatus(STATUS_COMPLETE); + req->finishRequest(true); + if (req->getFlags() & FLAG_AUTO_COMPLETE) + { + mRequestHash.erase(req); + req->deleteRequest(); + // check(); + } + unlockData(); + } + else + { + LL_PROFILE_ZONE_NAMED_CATEGORY_THREAD("qtpr - retry"); + //put back on queue and try again in 0.1ms + lockData(); + req->setStatus(STATUS_QUEUED); + + unlockData(); + + llassert(!mDataLock->isSelfLocked()); + +#if 0 + // try again on next frame + // NOTE: tried using "post" with a time in the future, but this + // would invariably cause this thread to wait for a long time (10+ ms) + // while work is pending + bool ret = LL::WorkQueue::postMaybe( + mMainQueue, + [=]() + { + LL_PROFILE_ZONE_NAMED("processRequest - retry"); + mRequestQueue.post([=]() + { + LL_PROFILE_ZONE_NAMED("processRequest - retry"); // <-- not redundant, track retry on both queues + processRequest(req); + }); + }); + llassert(ret); +#else + using namespace std::chrono_literals; + auto retry_time = LL::WorkQueue::TimePoint::clock::now() + 16ms; + mRequestQueue.post([=] + { + LL_PROFILE_ZONE_NAMED("processRequest - retry"); + if (LL::WorkQueue::TimePoint::clock::now() < retry_time) + { + auto sleep_time = std::chrono::duration_cast<std::chrono::milliseconds>(retry_time - LL::WorkQueue::TimePoint::clock::now()); + + if (sleep_time.count() > 0) + { + ms_sleep(sleep_time.count()); + } + } + processRequest(req); + }); +#endif + + } + } + } - S32 pending = getPending(); - return pending; + mIdleThread = TRUE; } // virtual bool LLQueuedThread::runCondition() { // mRunCondition must be locked here - if (mRequestQueue.empty() && mIdleThread) + if (mRequestQueue.size() == 0 && mIdleThread) return false; else return true; @@ -495,18 +515,13 @@ void LLQueuedThread::run() startThread(); mStarted = TRUE; - while (1) + + /*while (1) { + LL_PROFILE_ZONE_SCOPED; // this will block on the condition until runCondition() returns true, the thread is unpaused, or the thread leaves the RUNNING state. checkPause(); - if (isQuitting()) - { - LLTrace::get_thread_recorder()->pushToParent(); - endThread(); - break; - } - mIdleThread = FALSE; threadedUpdate(); @@ -515,12 +530,18 @@ void LLQueuedThread::run() if (pending_work == 0) { + //LL_PROFILE_ZONE_NAMED("LLQueuedThread - sleep"); mIdleThread = TRUE; - ms_sleep(1); + //ms_sleep(1); } //LLThread::yield(); // thread should yield after each request - } + }*/ + mRequestQueue.runUntilClose(); + + endThread(); LL_INFOS() << "LLQueuedThread " << mName << " EXITING." << LL_ENDL; + + } // virtual @@ -540,10 +561,9 @@ void LLQueuedThread::threadedUpdate() //============================================================================ -LLQueuedThread::QueuedRequest::QueuedRequest(LLQueuedThread::handle_t handle, U32 priority, U32 flags) : +LLQueuedThread::QueuedRequest::QueuedRequest(LLQueuedThread::handle_t handle, U32 flags) : LLSimpleHashEntry<LLQueuedThread::handle_t>(handle), mStatus(STATUS_UNKNOWN), - mPriority(priority), mFlags(flags) { } diff --git a/indra/llcommon/llqueuedthread.h b/indra/llcommon/llqueuedthread.h index 5d3f873646..6b82ccc434 100644 --- a/indra/llcommon/llqueuedthread.h +++ b/indra/llcommon/llqueuedthread.h @@ -36,6 +36,7 @@ #include "llthread.h" #include "llsimplehash.h" +#include "workqueue.h" //============================================================================ // Note: ~LLQueuedThread is O(N) N=# of queued threads, assumed to be small @@ -45,15 +46,6 @@ class LL_COMMON_API LLQueuedThread : public LLThread { //------------------------------------------------------------------------ public: - enum priority_t { - PRIORITY_IMMEDIATE = 0x7FFFFFFF, - PRIORITY_URGENT = 0x40000000, - PRIORITY_HIGH = 0x30000000, - PRIORITY_NORMAL = 0x20000000, - PRIORITY_LOW = 0x10000000, - PRIORITY_LOWBITS = 0x0FFFFFFF, - PRIORITY_HIGHBITS = 0x70000000 - }; enum status_t { STATUS_EXPIRED = -1, STATUS_UNKNOWN = 0, @@ -82,28 +74,17 @@ public: virtual ~QueuedRequest(); // use deleteRequest() public: - QueuedRequest(handle_t handle, U32 priority, U32 flags = 0); + QueuedRequest(handle_t handle, U32 flags = 0); status_t getStatus() { return mStatus; } - U32 getPriority() const - { - return mPriority; - } U32 getFlags() const { return mFlags; } - bool higherPriority(const QueuedRequest& second) const - { - if ( mPriority == second.mPriority) - return mHashKey < second.mHashKey; - else - return mPriority > second.mPriority; - } - + protected: status_t setStatus(status_t newstatus) { @@ -121,28 +102,11 @@ public: virtual void finishRequest(bool completed); // Always called from thread after request has completed or aborted virtual void deleteRequest(); // Only method to delete a request - void setPriority(U32 pri) - { - // Only do this on a request that is not in a queued list! - mPriority = pri; - }; - protected: LLAtomicBase<status_t> mStatus; - U32 mPriority; U32 mFlags; }; -protected: - struct queued_request_less - { - bool operator()(const QueuedRequest* lhs, const QueuedRequest* rhs) const - { - return lhs->higherPriority(*rhs); // higher priority in front of queue (set) - } - }; - - //------------------------------------------------------------------------ public: @@ -167,7 +131,7 @@ private: protected: handle_t generateHandle(); bool addRequest(QueuedRequest* req); - S32 processNextRequest(void); + void processRequest(QueuedRequest* req); void incQueue(); public: @@ -186,7 +150,6 @@ public: status_t getRequestStatus(handle_t handle); void abortRequest(handle_t handle, bool autocomplete); void setFlags(handle_t handle, U32 flags); - void setPriority(handle_t handle, U32 priority); bool completeRequest(handle_t handle); // This is public for support classes like LLWorkerThread, // but generally the methods above should be used. @@ -200,8 +163,10 @@ protected: BOOL mStarted; // required when mThreaded is false to call startThread() from update() LLAtomicBool mIdleThread; // request queue is empty (or we are quitting) and the thread is idle - typedef std::set<QueuedRequest*, queued_request_less> request_queue_t; - request_queue_t mRequestQueue; + //typedef std::set<QueuedRequest*, queued_request_less> request_queue_t; + //request_queue_t mRequestQueue; + LL::WorkQueue mRequestQueue; + LL::WorkQueue::weak_t mMainQueue; enum { REQUEST_HASH_SIZE = 512 }; // must be power of 2 typedef LLSimpleHash<handle_t, REQUEST_HASH_SIZE> request_hash_t; diff --git a/indra/llcommon/llsys.cpp b/indra/llcommon/llsys.cpp index a8b5c7b3a8..aeb8c530f4 100644 --- a/indra/llcommon/llsys.cpp +++ b/indra/llcommon/llsys.cpp @@ -771,20 +771,28 @@ static U32Kilobytes LLMemoryAdjustKBResult(U32Kilobytes inKB) } #endif +#if LL_DARWIN +// static +U32Kilobytes LLMemoryInfo::getHardwareMemSize() +{ + // This might work on Linux as well. Someone check... + uint64_t phys = 0; + int mib[2] = { CTL_HW, HW_MEMSIZE }; + + size_t len = sizeof(phys); + sysctl(mib, 2, &phys, &len, NULL, 0); + + return U64Bytes(phys); +} +#endif + U32Kilobytes LLMemoryInfo::getPhysicalMemoryKB() const { #if LL_WINDOWS return LLMemoryAdjustKBResult(U32Kilobytes(mStatsMap["Total Physical KB"].asInteger())); #elif LL_DARWIN - // This might work on Linux as well. Someone check... - uint64_t phys = 0; - int mib[2] = { CTL_HW, HW_MEMSIZE }; - - size_t len = sizeof(phys); - sysctl(mib, 2, &phys, &len, NULL, 0); - - return U64Bytes(phys); + return getHardwareMemSize(); #elif LL_LINUX U64 phys = 0; diff --git a/indra/llcommon/llsys.h b/indra/llcommon/llsys.h index 5ffbf5a732..0f768938cb 100644 --- a/indra/llcommon/llsys.h +++ b/indra/llcommon/llsys.h @@ -129,7 +129,10 @@ public: LLMemoryInfo(); ///< Default constructor void stream(std::ostream& s) const; ///< output text info to s - U32Kilobytes getPhysicalMemoryKB() const; + U32Kilobytes getPhysicalMemoryKB() const; +#if LL_DARWIN + static U32Kilobytes getHardwareMemSize(); // Because some Mac linkers won't let us reference extern gSysMemory from a different lib. +#endif //get the available memory infomation in KiloBytes. static void getAvailableMemoryKB(U32Kilobytes& avail_physical_mem_kb, U32Kilobytes& avail_virtual_mem_kb); diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp index a807acc56e..4eaa05c335 100644 --- a/indra/llcommon/llthread.cpp +++ b/indra/llcommon/llthread.cpp @@ -42,6 +42,7 @@ #ifdef LL_WINDOWS + const DWORD MS_VC_EXCEPTION=0x406D1388; #pragma pack(push,8) @@ -133,6 +134,15 @@ void LLThread::threadRun() { #ifdef LL_WINDOWS set_thread_name(-1, mName.c_str()); + +#if 0 // probably a bad idea, see usage of SetThreadIdealProcessor in LLWindowWin32) + HANDLE hThread = GetCurrentThread(); + if (hThread) + { + SetThreadAffinityMask(hThread, (DWORD_PTR) 0xFFFFFFFFFFFFFFFE); + } +#endif + #endif LL_PROFILER_SET_THREAD_NAME( mName.c_str() ); diff --git a/indra/llcommon/lltimer.cpp b/indra/llcommon/lltimer.cpp index aaa6df325c..24eaa4c1a9 100644 --- a/indra/llcommon/lltimer.cpp +++ b/indra/llcommon/lltimer.cpp @@ -30,6 +30,9 @@ #include "u64.h" +#include <chrono> +#include <thread> + #if LL_WINDOWS # include "llwin32headerslean.h" #elif LL_LINUX || LL_DARWIN @@ -62,9 +65,18 @@ LLTimer* LLTimer::sTimer = NULL; //--------------------------------------------------------------------------- #if LL_WINDOWS + + +#if 0 void ms_sleep(U32 ms) { - Sleep(ms); + LL_PROFILE_ZONE_SCOPED; + using TimePoint = std::chrono::steady_clock::time_point; + auto resume_time = TimePoint::clock::now() + std::chrono::milliseconds(ms); + while (TimePoint::clock::now() < resume_time) + { + std::this_thread::yield(); //note: don't use LLThread::yield here to avoid yielding for too long + } } U32 micro_sleep(U64 us, U32 max_yields) @@ -74,6 +86,35 @@ U32 micro_sleep(U64 us, U32 max_yields) ms_sleep((U32)(us / 1000)); return 0; } + +#else + +U32 micro_sleep(U64 us, U32 max_yields) +{ + LL_PROFILE_ZONE_SCOPED +#if 0 + LARGE_INTEGER ft; + ft.QuadPart = -static_cast<S64>(us * 10); // '-' using relative time + + HANDLE timer = CreateWaitableTimer(NULL, TRUE, NULL); + SetWaitableTimer(timer, &ft, 0, NULL, NULL, 0); + WaitForSingleObject(timer, INFINITE); + CloseHandle(timer); +#else + Sleep(us / 1000); +#endif + + return 0; +} + +void ms_sleep(U32 ms) +{ + LL_PROFILE_ZONE_SCOPED + micro_sleep(ms * 1000, 0); +} + +#endif + #elif LL_LINUX || LL_DARWIN static void _sleep_loop(struct timespec& thiswait) { diff --git a/indra/llcommon/lluuid.cpp b/indra/llcommon/lluuid.cpp index 6f9e09a587..3619a64ea9 100644 --- a/indra/llcommon/lluuid.cpp +++ b/indra/llcommon/lluuid.cpp @@ -1,31 +1,31 @@ -/** +/** * @file lluuid.cpp * * $LicenseInfo:firstyear=2000&license=viewerlgpl$ * Second Life Viewer Source Code * Copyright (C) 2010, Linden Research, Inc. - * + * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; * version 2.1 of the License only. - * + * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. - * + * * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * + * * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA * $/LicenseInfo$ */ #include "linden_common.h" -// We can't use WIN32_LEAN_AND_MEAN here, needs lots of includes. + // We can't use WIN32_LEAN_AND_MEAN here, needs lots of includes. #if LL_WINDOWS #include "llwin32headers.h" // ugh, this is ugly. We need to straighten out our linking for this library @@ -50,7 +50,7 @@ const LLUUID LLUUID::null; const LLTransactionID LLTransactionID::tnull; // static -LLMutex * LLUUID::mMutex = NULL; +LLMutex* LLUUID::mMutex = NULL; @@ -59,15 +59,15 @@ LLMutex * LLUUID::mMutex = NULL; NOT DONE YET!!! static char BASE85_TABLE[] = { - '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', - 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', - 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', - 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', - 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', - 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', - 'y', 'z', '!', '#', '$', '%', '&', '(', ')', '*', - '+', '-', ';', '[', '=', '>', '?', '@', '^', '_', - '`', '{', '|', '}', '~', '\0' + '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', + 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', + 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', + 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', + 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', + 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', + 'y', 'z', '!', '#', '$', '%', '&', '(', ')', '*', + '+', '-', ';', '[', '=', '>', '?', '@', '^', '_', + '`', '{', '|', '}', '~', '\0' }; @@ -93,28 +93,28 @@ return ret; void LLUUID::toBase85(char* out) { - U32* me = (U32*)&(mData[0]); - for(S32 i = 0; i < 4; ++i) - { - char* o = &out[i*i]; - for(S32 j = 0; j < 5; ++j) - { - o[4-j] = BASE85_TABLE[ me[i] % 85]; - word /= 85; - } - } + U32* me = (U32*)&(mData[0]); + for(S32 i = 0; i < 4; ++i) + { + char* o = &out[i*i]; + for(S32 j = 0; j < 5; ++j) + { + o[4-j] = BASE85_TABLE[ me[i] % 85]; + word /= 85; + } + } } unsigned int decode( char const * fiveChars ) throw( bad_input_data ) { - unsigned int ret = 0; - for( S32 ix = 0; ix < 5; ++ix ) - { - char * s = strchr( encodeTable, fiveChars[ ix ] ); - ret = ret * 85 + (s-encodeTable); - } - return ret; -} + unsigned int ret = 0; + for( S32 ix = 0; ix < 5; ++ix ) + { + char * s = strchr( encodeTable, fiveChars[ ix ] ); + ret = ret * 85 + (s-encodeTable); + } + return ret; +} */ #define LL_USE_JANKY_RANDOM_NUMBER_GENERATOR 0 @@ -129,8 +129,8 @@ static U64 sJankyRandomSeed(LLUUID::getRandomSeed()); */ U32 janky_fast_random_bytes() { - sJankyRandomSeed = U64L(1664525) * sJankyRandomSeed + U64L(1013904223); - return (U32)sJankyRandomSeed; + sJankyRandomSeed = U64L(1664525) * sJankyRandomSeed + U64L(1013904223); + return (U32)sJankyRandomSeed; } /** @@ -138,8 +138,8 @@ U32 janky_fast_random_bytes() */ U32 janky_fast_random_byes_range(U32 val) { - sJankyRandomSeed = U64L(1664525) * sJankyRandomSeed + U64L(1013904223); - return (U32)(sJankyRandomSeed) % val; + sJankyRandomSeed = U64L(1664525) * sJankyRandomSeed + U64L(1013904223); + return (U32)(sJankyRandomSeed) % val; } /** @@ -147,362 +147,362 @@ U32 janky_fast_random_byes_range(U32 val) */ U32 janky_fast_random_seeded_bytes(U32 seed, U32 val) { - seed = U64L(1664525) * (U64)(seed) + U64L(1013904223); - return (U32)(seed) % val; + seed = U64L(1664525) * (U64)(seed)+U64L(1013904223); + return (U32)(seed) % val; } #endif // Common to all UUID implementations void LLUUID::toString(std::string& out) const { - out = llformat( - "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x", - (U8)(mData[0]), - (U8)(mData[1]), - (U8)(mData[2]), - (U8)(mData[3]), - (U8)(mData[4]), - (U8)(mData[5]), - (U8)(mData[6]), - (U8)(mData[7]), - (U8)(mData[8]), - (U8)(mData[9]), - (U8)(mData[10]), - (U8)(mData[11]), - (U8)(mData[12]), - (U8)(mData[13]), - (U8)(mData[14]), - (U8)(mData[15])); + out = llformat( + "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x", + (U8)(mData[0]), + (U8)(mData[1]), + (U8)(mData[2]), + (U8)(mData[3]), + (U8)(mData[4]), + (U8)(mData[5]), + (U8)(mData[6]), + (U8)(mData[7]), + (U8)(mData[8]), + (U8)(mData[9]), + (U8)(mData[10]), + (U8)(mData[11]), + (U8)(mData[12]), + (U8)(mData[13]), + (U8)(mData[14]), + (U8)(mData[15])); } // *TODO: deprecate -void LLUUID::toString(char *out) const +void LLUUID::toString(char* out) const { - std::string buffer; - toString(buffer); - strcpy(out,buffer.c_str()); /* Flawfinder: ignore */ + std::string buffer; + toString(buffer); + strcpy(out, buffer.c_str()); /* Flawfinder: ignore */ } void LLUUID::toCompressedString(std::string& out) const { - char bytes[UUID_BYTES+1]; - memcpy(bytes, mData, UUID_BYTES); /* Flawfinder: ignore */ - bytes[UUID_BYTES] = '\0'; - out.assign(bytes, UUID_BYTES); + char bytes[UUID_BYTES + 1]; + memcpy(bytes, mData, UUID_BYTES); /* Flawfinder: ignore */ + bytes[UUID_BYTES] = '\0'; + out.assign(bytes, UUID_BYTES); } // *TODO: deprecate -void LLUUID::toCompressedString(char *out) const +void LLUUID::toCompressedString(char* out) const { - memcpy(out, mData, UUID_BYTES); /* Flawfinder: ignore */ - out[UUID_BYTES] = '\0'; + memcpy(out, mData, UUID_BYTES); /* Flawfinder: ignore */ + out[UUID_BYTES] = '\0'; } std::string LLUUID::getString() const { - return asString(); + return asString(); } std::string LLUUID::asString() const { - std::string str; - toString(str); - return str; + std::string str; + toString(str); + return str; } BOOL LLUUID::set(const char* in_string, BOOL emit) { - return set(ll_safe_string(in_string),emit); + return set(ll_safe_string(in_string), emit); } BOOL LLUUID::set(const std::string& in_string, BOOL emit) { - BOOL broken_format = FALSE; - - // empty strings should make NULL uuid - if (in_string.empty()) - { - setNull(); - return TRUE; - } - - if (in_string.length() != (UUID_STR_LENGTH - 1)) /* Flawfinder: ignore */ - { - // I'm a moron. First implementation didn't have the right UUID format. - // Shouldn't see any of these any more - if (in_string.length() == (UUID_STR_LENGTH - 2)) /* Flawfinder: ignore */ - { - if(emit) - { - LL_WARNS() << "Warning! Using broken UUID string format" << LL_ENDL; - } - broken_format = TRUE; - } - else - { - // Bad UUID string. Spam as INFO, as most cases we don't care. - if(emit) - { - //don't spam the logs because a resident can't spell. - LL_WARNS() << "Bad UUID string: " << in_string << LL_ENDL; - } - setNull(); - return FALSE; - } - } - - U8 cur_pos = 0; - S32 i; - for (i = 0; i < UUID_BYTES; i++) - { - if ((i == 4) || (i == 6) || (i == 8) || (i == 10)) - { - cur_pos++; - if (broken_format && (i==10)) - { - // Missing - in the broken format - cur_pos--; - } - } - - mData[i] = 0; - - if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) - { - mData[i] += (U8)(in_string[cur_pos] - '0'); - } - else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <='f')) - { - mData[i] += (U8)(10 + in_string[cur_pos] - 'a'); - } - else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <='F')) - { - mData[i] += (U8)(10 + in_string[cur_pos] - 'A'); - } - else - { - if(emit) - { - LL_WARNS() << "Invalid UUID string character" << LL_ENDL; - } - setNull(); - return FALSE; - } - - mData[i] = mData[i] << 4; - cur_pos++; - - if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) - { - mData[i] += (U8)(in_string[cur_pos] - '0'); - } - else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <='f')) - { - mData[i] += (U8)(10 + in_string[cur_pos] - 'a'); - } - else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <='F')) - { - mData[i] += (U8)(10 + in_string[cur_pos] - 'A'); - } - else - { - if(emit) - { - LL_WARNS() << "Invalid UUID string character" << LL_ENDL; - } - setNull(); - return FALSE; - } - cur_pos++; - } - - return TRUE; + BOOL broken_format = FALSE; + + // empty strings should make NULL uuid + if (in_string.empty()) + { + setNull(); + return TRUE; + } + + if (in_string.length() != (UUID_STR_LENGTH - 1)) /* Flawfinder: ignore */ + { + // I'm a moron. First implementation didn't have the right UUID format. + // Shouldn't see any of these any more + if (in_string.length() == (UUID_STR_LENGTH - 2)) /* Flawfinder: ignore */ + { + if (emit) + { + LL_WARNS() << "Warning! Using broken UUID string format" << LL_ENDL; + } + broken_format = TRUE; + } + else + { + // Bad UUID string. Spam as INFO, as most cases we don't care. + if (emit) + { + //don't spam the logs because a resident can't spell. + LL_WARNS() << "Bad UUID string: " << in_string << LL_ENDL; + } + setNull(); + return FALSE; + } + } + + U8 cur_pos = 0; + S32 i; + for (i = 0; i < UUID_BYTES; i++) + { + if ((i == 4) || (i == 6) || (i == 8) || (i == 10)) + { + cur_pos++; + if (broken_format && (i == 10)) + { + // Missing - in the broken format + cur_pos--; + } + } + + mData[i] = 0; + + if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) + { + mData[i] += (U8)(in_string[cur_pos] - '0'); + } + else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <= 'f')) + { + mData[i] += (U8)(10 + in_string[cur_pos] - 'a'); + } + else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <= 'F')) + { + mData[i] += (U8)(10 + in_string[cur_pos] - 'A'); + } + else + { + if (emit) + { + LL_WARNS() << "Invalid UUID string character" << LL_ENDL; + } + setNull(); + return FALSE; + } + + mData[i] = mData[i] << 4; + cur_pos++; + + if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) + { + mData[i] += (U8)(in_string[cur_pos] - '0'); + } + else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <= 'f')) + { + mData[i] += (U8)(10 + in_string[cur_pos] - 'a'); + } + else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <= 'F')) + { + mData[i] += (U8)(10 + in_string[cur_pos] - 'A'); + } + else + { + if (emit) + { + LL_WARNS() << "Invalid UUID string character" << LL_ENDL; + } + setNull(); + return FALSE; + } + cur_pos++; + } + + return TRUE; } BOOL LLUUID::validate(const std::string& in_string) { - BOOL broken_format = FALSE; - if (in_string.length() != (UUID_STR_LENGTH - 1)) /* Flawfinder: ignore */ - { - // I'm a moron. First implementation didn't have the right UUID format. - if (in_string.length() == (UUID_STR_LENGTH - 2)) /* Flawfinder: ignore */ - { - broken_format = TRUE; - } - else - { - return FALSE; - } - } - - U8 cur_pos = 0; - for (U32 i = 0; i < 16; i++) - { - if ((i == 4) || (i == 6) || (i == 8) || (i == 10)) - { - cur_pos++; - if (broken_format && (i==10)) - { - // Missing - in the broken format - cur_pos--; - } - } - - if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) - { - } - else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <='f')) - { - } - else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <='F')) - { - } - else - { - return FALSE; - } - - cur_pos++; - - if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) - { - } - else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <='f')) - { - } - else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <='F')) - { - } - else - { - return FALSE; - } - cur_pos++; - } - return TRUE; + BOOL broken_format = FALSE; + if (in_string.length() != (UUID_STR_LENGTH - 1)) /* Flawfinder: ignore */ + { + // I'm a moron. First implementation didn't have the right UUID format. + if (in_string.length() == (UUID_STR_LENGTH - 2)) /* Flawfinder: ignore */ + { + broken_format = TRUE; + } + else + { + return FALSE; + } + } + + U8 cur_pos = 0; + for (U32 i = 0; i < 16; i++) + { + if ((i == 4) || (i == 6) || (i == 8) || (i == 10)) + { + cur_pos++; + if (broken_format && (i == 10)) + { + // Missing - in the broken format + cur_pos--; + } + } + + if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) + { + } + else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <= 'f')) + { + } + else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <= 'F')) + { + } + else + { + return FALSE; + } + + cur_pos++; + + if ((in_string[cur_pos] >= '0') && (in_string[cur_pos] <= '9')) + { + } + else if ((in_string[cur_pos] >= 'a') && (in_string[cur_pos] <= 'f')) + { + } + else if ((in_string[cur_pos] >= 'A') && (in_string[cur_pos] <= 'F')) + { + } + else + { + return FALSE; + } + cur_pos++; + } + return TRUE; } const LLUUID& LLUUID::operator^=(const LLUUID& rhs) { - U32* me = (U32*)&(mData[0]); - const U32* other = (U32*)&(rhs.mData[0]); - for(S32 i = 0; i < 4; ++i) - { - me[i] = me[i] ^ other[i]; - } - return *this; + U32* me = (U32*)&(mData[0]); + const U32* other = (U32*)&(rhs.mData[0]); + for (S32 i = 0; i < 4; ++i) + { + me[i] = me[i] ^ other[i]; + } + return *this; } LLUUID LLUUID::operator^(const LLUUID& rhs) const { - LLUUID id(*this); - id ^= rhs; - return id; + LLUUID id(*this); + id ^= rhs; + return id; } void LLUUID::combine(const LLUUID& other, LLUUID& result) const { - LLMD5 md5_uuid; - md5_uuid.update((unsigned char*)mData, 16); - md5_uuid.update((unsigned char*)other.mData, 16); - md5_uuid.finalize(); - md5_uuid.raw_digest(result.mData); + LLMD5 md5_uuid; + md5_uuid.update((unsigned char*)mData, 16); + md5_uuid.update((unsigned char*)other.mData, 16); + md5_uuid.finalize(); + md5_uuid.raw_digest(result.mData); } -LLUUID LLUUID::combine(const LLUUID &other) const +LLUUID LLUUID::combine(const LLUUID& other) const { - LLUUID combination; - combine(other, combination); - return combination; + LLUUID combination; + combine(other, combination); + return combination; } -std::ostream& operator<<(std::ostream& s, const LLUUID &uuid) +std::ostream& operator<<(std::ostream& s, const LLUUID& uuid) { - std::string uuid_str; - uuid.toString(uuid_str); - s << uuid_str; - return s; + std::string uuid_str; + uuid.toString(uuid_str); + s << uuid_str; + return s; } -std::istream& operator>>(std::istream &s, LLUUID &uuid) +std::istream& operator>>(std::istream& s, LLUUID& uuid) { - U32 i; - char uuid_str[UUID_STR_LENGTH]; /* Flawfinder: ignore */ - for (i = 0; i < UUID_STR_LENGTH-1; i++) - { - s >> uuid_str[i]; - } - uuid_str[i] = '\0'; - uuid.set(std::string(uuid_str)); - return s; + U32 i; + char uuid_str[UUID_STR_LENGTH]; /* Flawfinder: ignore */ + for (i = 0; i < UUID_STR_LENGTH - 1; i++) + { + s >> uuid_str[i]; + } + uuid_str[i] = '\0'; + uuid.set(std::string(uuid_str)); + return s; } -static void get_random_bytes(void *buf, int nbytes) +static void get_random_bytes(void* buf, int nbytes) { - int i; - char *cp = (char *) buf; - - // *NOTE: If we are not using the janky generator ll_rand() - // generates at least 3 good bytes of data since it is 0 to - // RAND_MAX. This could be made more efficient by copying all the - // bytes. - for (i=0; i < nbytes; i++) + int i; + char* cp = (char*)buf; + + // *NOTE: If we are not using the janky generator ll_rand() + // generates at least 3 good bytes of data since it is 0 to + // RAND_MAX. This could be made more efficient by copying all the + // bytes. + for (i = 0; i < nbytes; i++) #if LL_USE_JANKY_RANDOM_NUMBER_GENERATOR - *cp++ = janky_fast_random_bytes() & 0xFF; + * cp++ = janky_fast_random_bytes() & 0xFF; #else - *cp++ = ll_rand() & 0xFF; + * cp++ = ll_rand() & 0xFF; #endif - return; + return; } #if LL_WINDOWS typedef struct _ASTAT_ { - ADAPTER_STATUS adapt; - NAME_BUFFER NameBuff [30]; + ADAPTER_STATUS adapt; + NAME_BUFFER NameBuff[30]; }ASTAT, * PASTAT; // static -S32 LLUUID::getNodeID(unsigned char *node_id) +S32 LLUUID::getNodeID(unsigned char* node_id) { - ASTAT Adapter; - NCB Ncb; - UCHAR uRetCode; - LANA_ENUM lenum; - int i; - int retval = 0; - - memset( &Ncb, 0, sizeof(Ncb) ); - Ncb.ncb_command = NCBENUM; - Ncb.ncb_buffer = (UCHAR *)&lenum; - Ncb.ncb_length = sizeof(lenum); - uRetCode = Netbios( &Ncb ); - - for(i=0; i < lenum.length ;i++) - { - memset( &Ncb, 0, sizeof(Ncb) ); - Ncb.ncb_command = NCBRESET; - Ncb.ncb_lana_num = lenum.lana[i]; - - uRetCode = Netbios( &Ncb ); - - memset( &Ncb, 0, sizeof (Ncb) ); - Ncb.ncb_command = NCBASTAT; - Ncb.ncb_lana_num = lenum.lana[i]; - - strcpy( (char *)Ncb.ncb_callname, "* " ); /* Flawfinder: ignore */ - Ncb.ncb_buffer = (unsigned char *)&Adapter; - Ncb.ncb_length = sizeof(Adapter); - - uRetCode = Netbios( &Ncb ); - if ( uRetCode == 0 ) - { - memcpy(node_id,Adapter.adapt.adapter_address,6); /* Flawfinder: ignore */ - retval = 1; - } - } - return retval; + ASTAT Adapter; + NCB Ncb; + UCHAR uRetCode; + LANA_ENUM lenum; + int i; + int retval = 0; + + memset(&Ncb, 0, sizeof(Ncb)); + Ncb.ncb_command = NCBENUM; + Ncb.ncb_buffer = (UCHAR*)&lenum; + Ncb.ncb_length = sizeof(lenum); + uRetCode = Netbios(&Ncb); + + for (i = 0; i < lenum.length; i++) + { + memset(&Ncb, 0, sizeof(Ncb)); + Ncb.ncb_command = NCBRESET; + Ncb.ncb_lana_num = lenum.lana[i]; + + uRetCode = Netbios(&Ncb); + + memset(&Ncb, 0, sizeof(Ncb)); + Ncb.ncb_command = NCBASTAT; + Ncb.ncb_lana_num = lenum.lana[i]; + + strcpy((char*)Ncb.ncb_callname, "* "); /* Flawfinder: ignore */ + Ncb.ncb_buffer = (unsigned char*)&Adapter; + Ncb.ncb_length = sizeof(Adapter); + + uRetCode = Netbios(&Ncb); + if (uRetCode == 0) + { + memcpy(node_id, Adapter.adapt.adapter_address, 6); /* Flawfinder: ignore */ + retval = 1; + } + } + return retval; } #elif LL_DARWIN @@ -521,66 +521,66 @@ S32 LLUUID::getNodeID(unsigned char *node_id) #include <net/route.h> #include <ifaddrs.h> -// static -S32 LLUUID::getNodeID(unsigned char *node_id) + // static +S32 LLUUID::getNodeID(unsigned char* node_id) { - int i; - unsigned char *a = NULL; - struct ifaddrs *ifap, *ifa; - int rv; - S32 result = 0; - - if ((rv=getifaddrs(&ifap))==-1) - { - return -1; - } - if (ifap == NULL) - { - return -1; - } - - for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) - { -// printf("Interface %s, address family %d, ", ifa->ifa_name, ifa->ifa_addr->sa_family); - for(i=0; i< ifa->ifa_addr->sa_len; i++) - { -// printf("%02X ", (unsigned char)ifa->ifa_addr->sa_data[i]); - } -// printf("\n"); - - if(ifa->ifa_addr->sa_family == AF_LINK) - { - // This is a link-level address - struct sockaddr_dl *lla = (struct sockaddr_dl *)ifa->ifa_addr; - -// printf("\tLink level address, type %02X\n", lla->sdl_type); - - if(lla->sdl_type == IFT_ETHER) - { - // Use the first ethernet MAC in the list. - // For some reason, the macro LLADDR() defined in net/if_dl.h doesn't expand correctly. This is what it would do. - a = (unsigned char *)&((lla)->sdl_data); - a += (lla)->sdl_nlen; - - if (!a[0] && !a[1] && !a[2] && !a[3] && !a[4] && !a[5]) - { - continue; - } - - if (node_id) - { - memcpy(node_id, a, 6); - result = 1; - } - - // We found one. - break; - } - } - } - freeifaddrs(ifap); - - return result; + int i; + unsigned char* a = NULL; + struct ifaddrs* ifap, * ifa; + int rv; + S32 result = 0; + + if ((rv = getifaddrs(&ifap)) == -1) + { + return -1; + } + if (ifap == NULL) + { + return -1; + } + + for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) + { + // printf("Interface %s, address family %d, ", ifa->ifa_name, ifa->ifa_addr->sa_family); + for (i = 0; i < ifa->ifa_addr->sa_len; i++) + { + // printf("%02X ", (unsigned char)ifa->ifa_addr->sa_data[i]); + } + // printf("\n"); + + if (ifa->ifa_addr->sa_family == AF_LINK) + { + // This is a link-level address + struct sockaddr_dl* lla = (struct sockaddr_dl*)ifa->ifa_addr; + + // printf("\tLink level address, type %02X\n", lla->sdl_type); + + if (lla->sdl_type == IFT_ETHER) + { + // Use the first ethernet MAC in the list. + // For some reason, the macro LLADDR() defined in net/if_dl.h doesn't expand correctly. This is what it would do. + a = (unsigned char*)&((lla)->sdl_data); + a += (lla)->sdl_nlen; + + if (!a[0] && !a[1] && !a[2] && !a[3] && !a[4] && !a[5]) + { + continue; + } + + if (node_id) + { + memcpy(node_id, a, 6); + result = 1; + } + + // We found one. + break; + } + } + } + freeifaddrs(ifap); + + return result; } #else @@ -607,22 +607,22 @@ S32 LLUUID::getNodeID(unsigned char *node_id) #endif #endif -// static -S32 LLUUID::getNodeID(unsigned char *node_id) + // static +S32 LLUUID::getNodeID(unsigned char* node_id) { - int sd; - struct ifreq ifr, *ifrp; - struct ifconf ifc; - char buf[1024]; - int n, i; - unsigned char *a; - -/* - * BSD 4.4 defines the size of an ifreq to be - * max(sizeof(ifreq), sizeof(ifreq.ifr_name)+ifreq.ifr_addr.sa_len - * However, under earlier systems, sa_len isn't present, so the size is - * just sizeof(struct ifreq) - */ + int sd; + struct ifreq ifr, * ifrp; + struct ifconf ifc; + char buf[1024]; + int n, i; + unsigned char* a; + + /* + * BSD 4.4 defines the size of an ifreq to be + * max(sizeof(ifreq), sizeof(ifreq.ifr_name)+ifreq.ifr_addr.sa_len + * However, under earlier systems, sa_len isn't present, so the size is + * just sizeof(struct ifreq) + */ #ifdef HAVE_SA_LEN #ifndef max #define max(a,b) ((a) > (b) ? (a) : (b)) @@ -633,355 +633,356 @@ S32 LLUUID::getNodeID(unsigned char *node_id) #define ifreq_size(i) sizeof(struct ifreq) #endif /* HAVE_SA_LEN*/ - sd = socket(AF_INET, SOCK_DGRAM, IPPROTO_IP); - if (sd < 0) { - return -1; - } - memset(buf, 0, sizeof(buf)); - ifc.ifc_len = sizeof(buf); - ifc.ifc_buf = buf; - if (ioctl (sd, SIOCGIFCONF, (char *)&ifc) < 0) { - close(sd); - return -1; - } - n = ifc.ifc_len; - for (i = 0; i < n; i+= ifreq_size(*ifr) ) { - ifrp = (struct ifreq *)((char *) ifc.ifc_buf+i); - strncpy(ifr.ifr_name, ifrp->ifr_name, IFNAMSIZ); /* Flawfinder: ignore */ + sd = socket(AF_INET, SOCK_DGRAM, IPPROTO_IP); + if (sd < 0) { + return -1; + } + memset(buf, 0, sizeof(buf)); + ifc.ifc_len = sizeof(buf); + ifc.ifc_buf = buf; + if (ioctl(sd, SIOCGIFCONF, (char*)&ifc) < 0) { + close(sd); + return -1; + } + n = ifc.ifc_len; + for (i = 0; i < n; i += ifreq_size(*ifr)) { + ifrp = (struct ifreq*)((char*)ifc.ifc_buf + i); + strncpy(ifr.ifr_name, ifrp->ifr_name, IFNAMSIZ); /* Flawfinder: ignore */ #ifdef SIOCGIFHWADDR - if (ioctl(sd, SIOCGIFHWADDR, &ifr) < 0) - continue; - a = (unsigned char *) &ifr.ifr_hwaddr.sa_data; + if (ioctl(sd, SIOCGIFHWADDR, &ifr) < 0) + continue; + a = (unsigned char*)&ifr.ifr_hwaddr.sa_data; #else #ifdef SIOCGENADDR - if (ioctl(sd, SIOCGENADDR, &ifr) < 0) - continue; - a = (unsigned char *) ifr.ifr_enaddr; + if (ioctl(sd, SIOCGENADDR, &ifr) < 0) + continue; + a = (unsigned char*)ifr.ifr_enaddr; #else - /* - * XXX we don't have a way of getting the hardware - * address - */ - close(sd); - return 0; + /* + * XXX we don't have a way of getting the hardware + * address + */ + close(sd); + return 0; #endif /* SIOCGENADDR */ #endif /* SIOCGIFHWADDR */ - if (!a[0] && !a[1] && !a[2] && !a[3] && !a[4] && !a[5]) - continue; - if (node_id) { - memcpy(node_id, a, 6); /* Flawfinder: ignore */ - close(sd); - return 1; - } - } - close(sd); - return 0; + if (!a[0] && !a[1] && !a[2] && !a[3] && !a[4] && !a[5]) + continue; + if (node_id) { + memcpy(node_id, a, 6); /* Flawfinder: ignore */ + close(sd); + return 1; + } + } + close(sd); + return 0; } #endif -S32 LLUUID::cmpTime(uuid_time_t *t1, uuid_time_t *t2) +S32 LLUUID::cmpTime(uuid_time_t* t1, uuid_time_t* t2) { - // Compare two time values. + // Compare two time values. - if (t1->high < t2->high) return -1; - if (t1->high > t2->high) return 1; - if (t1->low < t2->low) return -1; - if (t1->low > t2->low) return 1; - return 0; + if (t1->high < t2->high) return -1; + if (t1->high > t2->high) return 1; + if (t1->low < t2->low) return -1; + if (t1->low > t2->low) return 1; + return 0; } -void LLUUID::getSystemTime(uuid_time_t *timestamp) +void LLUUID::getSystemTime(uuid_time_t* timestamp) { - // Get system time with 100ns precision. Time is since Oct 15, 1582. + // Get system time with 100ns precision. Time is since Oct 15, 1582. #if LL_WINDOWS - ULARGE_INTEGER time; - GetSystemTimeAsFileTime((FILETIME *)&time); - // NT keeps time in FILETIME format which is 100ns ticks since - // Jan 1, 1601. UUIDs use time in 100ns ticks since Oct 15, 1582. - // The difference is 17 Days in Oct + 30 (Nov) + 31 (Dec) - // + 18 years and 5 leap days. - time.QuadPart += - (unsigned __int64) (1000*1000*10) // seconds - * (unsigned __int64) (60 * 60 * 24) // days - * (unsigned __int64) (17+30+31+365*18+5); // # of days - - timestamp->high = time.HighPart; - timestamp->low = time.LowPart; + ULARGE_INTEGER time; + GetSystemTimeAsFileTime((FILETIME*)&time); + // NT keeps time in FILETIME format which is 100ns ticks since + // Jan 1, 1601. UUIDs use time in 100ns ticks since Oct 15, 1582. + // The difference is 17 Days in Oct + 30 (Nov) + 31 (Dec) + // + 18 years and 5 leap days. + time.QuadPart += + (unsigned __int64)(1000 * 1000 * 10) // seconds + * (unsigned __int64)(60 * 60 * 24) // days + * (unsigned __int64)(17 + 30 + 31 + 365 * 18 + 5); // # of days + + timestamp->high = time.HighPart; + timestamp->low = time.LowPart; #else - struct timeval tp; - gettimeofday(&tp, 0); - - // Offset between UUID formatted times and Unix formatted times. - // UUID UTC base time is October 15, 1582. - // Unix base time is January 1, 1970. - U64 uuid_time = ((U64)tp.tv_sec * 10000000) + (tp.tv_usec * 10) + - U64L(0x01B21DD213814000); - timestamp->high = (U32) (uuid_time >> 32); - timestamp->low = (U32) (uuid_time & 0xFFFFFFFF); + struct timeval tp; + gettimeofday(&tp, 0); + + // Offset between UUID formatted times and Unix formatted times. + // UUID UTC base time is October 15, 1582. + // Unix base time is January 1, 1970. + U64 uuid_time = ((U64)tp.tv_sec * 10000000) + (tp.tv_usec * 10) + + U64L(0x01B21DD213814000); + timestamp->high = (U32)(uuid_time >> 32); + timestamp->low = (U32)(uuid_time & 0xFFFFFFFF); #endif } -void LLUUID::getCurrentTime(uuid_time_t *timestamp) +void LLUUID::getCurrentTime(uuid_time_t* timestamp) { - // Get current time as 60 bit 100ns ticks since whenever. - // Compensate for the fact that real clock resolution is less - // than 100ns. - - const U32 uuids_per_tick = 1024; - - static uuid_time_t time_last; - static U32 uuids_this_tick; - static BOOL init = FALSE; - - if (!init) { - getSystemTime(&time_last); - uuids_this_tick = uuids_per_tick; - init = TRUE; - mMutex = new LLMutex(); - } - - uuid_time_t time_now = {0,0}; - - while (1) { - getSystemTime(&time_now); - - // if clock reading changed since last UUID generated - if (cmpTime(&time_last, &time_now)) { - // reset count of uuid's generated with this clock reading - uuids_this_tick = 0; - break; - } - if (uuids_this_tick < uuids_per_tick) { - uuids_this_tick++; - break; - } - // going too fast for our clock; spin - } - - time_last = time_now; - - if (uuids_this_tick != 0) { - if (time_now.low & 0x80000000) { - time_now.low += uuids_this_tick; - if (!(time_now.low & 0x80000000)) - time_now.high++; - } else - time_now.low += uuids_this_tick; - } - - timestamp->high = time_now.high; - timestamp->low = time_now.low; + // Get current time as 60 bit 100ns ticks since whenever. + // Compensate for the fact that real clock resolution is less + // than 100ns. + + const U32 uuids_per_tick = 1024; + + static uuid_time_t time_last; + static U32 uuids_this_tick; + static BOOL init = FALSE; + + if (!init) { + getSystemTime(&time_last); + uuids_this_tick = uuids_per_tick; + init = TRUE; + mMutex = new LLMutex(); + } + + uuid_time_t time_now = { 0,0 }; + + while (1) { + getSystemTime(&time_now); + + // if clock reading changed since last UUID generated + if (cmpTime(&time_last, &time_now)) { + // reset count of uuid's generated with this clock reading + uuids_this_tick = 0; + break; + } + if (uuids_this_tick < uuids_per_tick) { + uuids_this_tick++; + break; + } + // going too fast for our clock; spin + } + + time_last = time_now; + + if (uuids_this_tick != 0) { + if (time_now.low & 0x80000000) { + time_now.low += uuids_this_tick; + if (!(time_now.low & 0x80000000)) + time_now.high++; + } + else + time_now.low += uuids_this_tick; + } + + timestamp->high = time_now.high; + timestamp->low = time_now.low; } void LLUUID::generate() { - // Create a UUID. - uuid_time_t timestamp; - - static unsigned char node_id[6]; /* Flawfinder: ignore */ - static int has_init = 0; - - // Create a UUID. - static uuid_time_t time_last = {0,0}; - static U16 clock_seq = 0; + // Create a UUID. + uuid_time_t timestamp; + + static unsigned char node_id[6]; /* Flawfinder: ignore */ + static int has_init = 0; + + // Create a UUID. + static uuid_time_t time_last = { 0,0 }; + static U16 clock_seq = 0; #if LL_USE_JANKY_RANDOM_NUMBER_GENERATOR - static U32 seed = 0L; // dummy seed. reset it below + static U32 seed = 0L; // dummy seed. reset it below #endif - if (!has_init) - { - has_init = 1; - if (getNodeID(node_id) <= 0) - { - get_random_bytes(node_id, 6); - /* - * Set multicast bit, to prevent conflicts - * with IEEE 802 addresses obtained from - * network cards - */ - node_id[0] |= 0x80; - } - - getCurrentTime(&time_last); + if (!has_init) + { + has_init = 1; + if (getNodeID(node_id) <= 0) + { + get_random_bytes(node_id, 6); + /* + * Set multicast bit, to prevent conflicts + * with IEEE 802 addresses obtained from + * network cards + */ + node_id[0] |= 0x80; + } + + getCurrentTime(&time_last); #if LL_USE_JANKY_RANDOM_NUMBER_GENERATOR - seed = time_last.low; + seed = time_last.low; #endif #if LL_USE_JANKY_RANDOM_NUMBER_GENERATOR - clock_seq = (U16)janky_fast_random_seeded_bytes(seed, 65536); + clock_seq = (U16)janky_fast_random_seeded_bytes(seed, 65536); #else - clock_seq = (U16)ll_rand(65536); + clock_seq = (U16)ll_rand(65536); #endif - } - - // get current time - getCurrentTime(×tamp); - U16 our_clock_seq = clock_seq; - - // if clock hasn't changed or went backward, change clockseq - if (cmpTime(×tamp, &time_last) != 1) - { - LLMutexLock lock(mMutex); - clock_seq = (clock_seq + 1) & 0x3FFF; - if (clock_seq == 0) - clock_seq++; - our_clock_seq = clock_seq; // Ensure we're using a different clock_seq value from previous time - } + } + + // get current time + getCurrentTime(×tamp); + U16 our_clock_seq = clock_seq; + + // if clock hasn't changed or went backward, change clockseq + if (cmpTime(×tamp, &time_last) != 1) + { + LLMutexLock lock(mMutex); + clock_seq = (clock_seq + 1) & 0x3FFF; + if (clock_seq == 0) + clock_seq++; + our_clock_seq = clock_seq; // Ensure we're using a different clock_seq value from previous time + } time_last = timestamp; - memcpy(mData+10, node_id, 6); /* Flawfinder: ignore */ - U32 tmp; - tmp = timestamp.low; - mData[3] = (unsigned char) tmp; - tmp >>= 8; - mData[2] = (unsigned char) tmp; - tmp >>= 8; - mData[1] = (unsigned char) tmp; - tmp >>= 8; - mData[0] = (unsigned char) tmp; - - tmp = (U16) timestamp.high; - mData[5] = (unsigned char) tmp; - tmp >>= 8; - mData[4] = (unsigned char) tmp; - - tmp = (timestamp.high >> 16) | 0x1000; - mData[7] = (unsigned char) tmp; - tmp >>= 8; - mData[6] = (unsigned char) tmp; - - tmp = our_clock_seq; - - mData[9] = (unsigned char) tmp; - tmp >>= 8; - mData[8] = (unsigned char) tmp; - - LLMD5 md5_uuid; - - md5_uuid.update(mData,16); - md5_uuid.finalize(); - md5_uuid.raw_digest(mData); + memcpy(mData + 10, node_id, 6); /* Flawfinder: ignore */ + U32 tmp; + tmp = timestamp.low; + mData[3] = (unsigned char)tmp; + tmp >>= 8; + mData[2] = (unsigned char)tmp; + tmp >>= 8; + mData[1] = (unsigned char)tmp; + tmp >>= 8; + mData[0] = (unsigned char)tmp; + + tmp = (U16)timestamp.high; + mData[5] = (unsigned char)tmp; + tmp >>= 8; + mData[4] = (unsigned char)tmp; + + tmp = (timestamp.high >> 16) | 0x1000; + mData[7] = (unsigned char)tmp; + tmp >>= 8; + mData[6] = (unsigned char)tmp; + + tmp = our_clock_seq; + + mData[9] = (unsigned char)tmp; + tmp >>= 8; + mData[8] = (unsigned char)tmp; + + LLMD5 md5_uuid; + + md5_uuid.update(mData, 16); + md5_uuid.finalize(); + md5_uuid.raw_digest(mData); } void LLUUID::generate(const std::string& hash_string) { - LLMD5 md5_uuid((U8*)hash_string.c_str()); - md5_uuid.raw_digest(mData); + LLMD5 md5_uuid((U8*)hash_string.c_str()); + md5_uuid.raw_digest(mData); } U32 LLUUID::getRandomSeed() { - static unsigned char seed[16]; /* Flawfinder: ignore */ - - getNodeID(&seed[0]); - - // Incorporate the pid into the seed to prevent - // processes that start on the same host at the same - // time from generating the same seed. - pid_t pid = LLApp::getPid(); - - seed[6]=(unsigned char)(pid >> 8); - seed[7]=(unsigned char)(pid); - getSystemTime((uuid_time_t *)(&seed[8])); - - LLMD5 md5_seed; - - md5_seed.update(seed,16); - md5_seed.finalize(); - md5_seed.raw_digest(seed); - - return(*(U32 *)seed); + static unsigned char seed[16]; /* Flawfinder: ignore */ + + getNodeID(&seed[0]); + + // Incorporate the pid into the seed to prevent + // processes that start on the same host at the same + // time from generating the same seed. + pid_t pid = LLApp::getPid(); + + seed[6] = (unsigned char)(pid >> 8); + seed[7] = (unsigned char)(pid); + getSystemTime((uuid_time_t*)(&seed[8])); + + LLMD5 md5_seed; + + md5_seed.update(seed, 16); + md5_seed.finalize(); + md5_seed.raw_digest(seed); + + return(*(U32*)seed); } BOOL LLUUID::parseUUID(const std::string& buf, LLUUID* value) { - if( buf.empty() || value == NULL) - { - return FALSE; - } - - std::string temp( buf ); - LLStringUtil::trim(temp); - if( LLUUID::validate( temp ) ) - { - value->set( temp ); - return TRUE; - } - return FALSE; + if (buf.empty() || value == NULL) + { + return FALSE; + } + + std::string temp(buf); + LLStringUtil::trim(temp); + if (LLUUID::validate(temp)) + { + value->set(temp); + return TRUE; + } + return FALSE; } //static LLUUID LLUUID::generateNewID(std::string hash_string) { - LLUUID new_id; - if (hash_string.empty()) - { - new_id.generate(); - } - else - { - new_id.generate(hash_string); - } - return new_id; + LLUUID new_id; + if (hash_string.empty()) + { + new_id.generate(); + } + else + { + new_id.generate(hash_string); + } + return new_id; } LLAssetID LLTransactionID::makeAssetID(const LLUUID& session) const { - LLAssetID result; - if (isNull()) - { - result.setNull(); - } - else - { - combine(session, result); - } - return result; + LLAssetID result; + if (isNull()) + { + result.setNull(); + } + else + { + combine(session, result); + } + return result; } // Construct LLUUID::LLUUID() { - setNull(); + setNull(); } // Faster than copying from memory - void LLUUID::setNull() +void LLUUID::setNull() { - U32 *word = (U32 *)mData; - word[0] = 0; - word[1] = 0; - word[2] = 0; - word[3] = 0; + U32* word = (U32*)mData; + word[0] = 0; + word[1] = 0; + word[2] = 0; + word[3] = 0; } // Compare - bool LLUUID::operator==(const LLUUID& rhs) const +bool LLUUID::operator==(const LLUUID& rhs) const { - U32 *tmp = (U32 *)mData; - U32 *rhstmp = (U32 *)rhs.mData; - // Note: binary & to avoid branching - return - (tmp[0] == rhstmp[0]) & - (tmp[1] == rhstmp[1]) & - (tmp[2] == rhstmp[2]) & - (tmp[3] == rhstmp[3]); + U32* tmp = (U32*)mData; + U32* rhstmp = (U32*)rhs.mData; + // Note: binary & to avoid branching + return + (tmp[0] == rhstmp[0]) & + (tmp[1] == rhstmp[1]) & + (tmp[2] == rhstmp[2]) & + (tmp[3] == rhstmp[3]); } - bool LLUUID::operator!=(const LLUUID& rhs) const +bool LLUUID::operator!=(const LLUUID& rhs) const { - U32 *tmp = (U32 *)mData; - U32 *rhstmp = (U32 *)rhs.mData; - // Note: binary | to avoid branching - return - (tmp[0] != rhstmp[0]) | - (tmp[1] != rhstmp[1]) | - (tmp[2] != rhstmp[2]) | - (tmp[3] != rhstmp[3]); + U32* tmp = (U32*)mData; + U32* rhstmp = (U32*)rhs.mData; + // Note: binary | to avoid branching + return + (tmp[0] != rhstmp[0]) | + (tmp[1] != rhstmp[1]) | + (tmp[2] != rhstmp[2]) | + (tmp[3] != rhstmp[3]); } /* @@ -989,94 +990,94 @@ LLUUID::LLUUID() // to integers, among other things. Use isNull() or notNull(). LLUUID::operator bool() const { - U32 *word = (U32 *)mData; - return (word[0] | word[1] | word[2] | word[3]) > 0; + U32 *word = (U32 *)mData; + return (word[0] | word[1] | word[2] | word[3]) > 0; } */ - BOOL LLUUID::notNull() const +BOOL LLUUID::notNull() const { - U32 *word = (U32 *)mData; - return (word[0] | word[1] | word[2] | word[3]) > 0; + U32* word = (U32*)mData; + return (word[0] | word[1] | word[2] | word[3]) > 0; } // Faster than == LLUUID::null because doesn't require // as much memory access. - BOOL LLUUID::isNull() const +BOOL LLUUID::isNull() const { - U32 *word = (U32 *)mData; - // If all bits are zero, return !0 == TRUE - return !(word[0] | word[1] | word[2] | word[3]); + U32* word = (U32*)mData; + // If all bits are zero, return !0 == TRUE + return !(word[0] | word[1] | word[2] | word[3]); } - LLUUID::LLUUID(const char *in_string) +LLUUID::LLUUID(const char* in_string) { - if (!in_string || in_string[0] == 0) - { - setNull(); - return; - } - - set(in_string); + if (!in_string || in_string[0] == 0) + { + setNull(); + return; + } + + set(in_string); } - LLUUID::LLUUID(const std::string& in_string) +LLUUID::LLUUID(const std::string& in_string) { - if (in_string.empty()) - { - setNull(); - return; - } + if (in_string.empty()) + { + setNull(); + return; + } - set(in_string); + set(in_string); } // IW: DON'T "optimize" these w/ U32s or you'll scoogie the sort order // IW: this will make me very sad - bool LLUUID::operator<(const LLUUID &rhs) const +bool LLUUID::operator<(const LLUUID& rhs) const { - U32 i; - for( i = 0; i < (UUID_BYTES - 1); i++ ) - { - if( mData[i] != rhs.mData[i] ) - { - return (mData[i] < rhs.mData[i]); - } - } - return (mData[UUID_BYTES - 1] < rhs.mData[UUID_BYTES - 1]); + U32 i; + for (i = 0; i < (UUID_BYTES - 1); i++) + { + if (mData[i] != rhs.mData[i]) + { + return (mData[i] < rhs.mData[i]); + } + } + return (mData[UUID_BYTES - 1] < rhs.mData[UUID_BYTES - 1]); } - bool LLUUID::operator>(const LLUUID &rhs) const +bool LLUUID::operator>(const LLUUID& rhs) const { - U32 i; - for( i = 0; i < (UUID_BYTES - 1); i++ ) - { - if( mData[i] != rhs.mData[i] ) - { - return (mData[i] > rhs.mData[i]); - } - } - return (mData[UUID_BYTES - 1] > rhs.mData[UUID_BYTES - 1]); + U32 i; + for (i = 0; i < (UUID_BYTES - 1); i++) + { + if (mData[i] != rhs.mData[i]) + { + return (mData[i] > rhs.mData[i]); + } + } + return (mData[UUID_BYTES - 1] > rhs.mData[UUID_BYTES - 1]); } - U16 LLUUID::getCRC16() const +U16 LLUUID::getCRC16() const { - // A UUID is 16 bytes, or 8 shorts. - U16 *short_data = (U16*)mData; - U16 out = 0; - out += short_data[0]; - out += short_data[1]; - out += short_data[2]; - out += short_data[3]; - out += short_data[4]; - out += short_data[5]; - out += short_data[6]; - out += short_data[7]; - return out; + // A UUID is 16 bytes, or 8 shorts. + U16* short_data = (U16*)mData; + U16 out = 0; + out += short_data[0]; + out += short_data[1]; + out += short_data[2]; + out += short_data[3]; + out += short_data[4]; + out += short_data[5]; + out += short_data[6]; + out += short_data[7]; + return out; } - U32 LLUUID::getCRC32() const +U32 LLUUID::getCRC32() const { - U32 *tmp = (U32*)mData; - return tmp[0] + tmp[1] + tmp[2] + tmp[3]; + U32* tmp = (U32*)mData; + return tmp[0] + tmp[1] + tmp[2] + tmp[3]; } diff --git a/indra/llcommon/llworkerthread.cpp b/indra/llcommon/llworkerthread.cpp index 4b91b2caca..cd4d587752 100644 --- a/indra/llcommon/llworkerthread.cpp +++ b/indra/llcommon/llworkerthread.cpp @@ -73,6 +73,7 @@ void LLWorkerThread::clearDeleteList() { (*iter)->mRequestHandle = LLWorkerThread::nullHandle(); (*iter)->clearFlags(LLWorkerClass::WCF_HAVE_WORK); + (*iter)->clearFlags(LLWorkerClass::WCF_WORKING); delete *iter ; } mDeleteList.clear() ; @@ -97,6 +98,7 @@ S32 LLWorkerThread::update(F32 max_time_ms) { if (worker->getFlags(LLWorkerClass::WCF_WORK_FINISHED)) { + worker->setFlags(LLWorkerClass::WCF_DELETE_REQUESTED); delete_list.push_back(worker); mDeleteList.erase(curiter); } @@ -133,11 +135,11 @@ S32 LLWorkerThread::update(F32 max_time_ms) //---------------------------------------------------------------------------- -LLWorkerThread::handle_t LLWorkerThread::addWorkRequest(LLWorkerClass* workerclass, S32 param, U32 priority) +LLWorkerThread::handle_t LLWorkerThread::addWorkRequest(LLWorkerClass* workerclass, S32 param) { handle_t handle = generateHandle(); - WorkRequest* req = new WorkRequest(handle, priority, workerclass, param); + WorkRequest* req = new WorkRequest(handle, workerclass, param); bool res = addRequest(req); if (!res) @@ -160,8 +162,8 @@ void LLWorkerThread::deleteWorker(LLWorkerClass* workerclass) //============================================================================ // Runs on its OWN thread -LLWorkerThread::WorkRequest::WorkRequest(handle_t handle, U32 priority, LLWorkerClass* workerclass, S32 param) : - LLQueuedThread::QueuedRequest(handle, priority), +LLWorkerThread::WorkRequest::WorkRequest(handle_t handle, LLWorkerClass* workerclass, S32 param) : + LLQueuedThread::QueuedRequest(handle), mWorkerClass(workerclass), mParam(param) { @@ -180,6 +182,7 @@ void LLWorkerThread::WorkRequest::deleteRequest() // virtual bool LLWorkerThread::WorkRequest::processRequest() { + LL_PROFILE_ZONE_SCOPED; LLWorkerClass* workerclass = getWorkerClass(); workerclass->setWorking(true); bool complete = workerclass->doWork(getParam()); @@ -190,6 +193,7 @@ bool LLWorkerThread::WorkRequest::processRequest() // virtual void LLWorkerThread::WorkRequest::finishRequest(bool completed) { + LL_PROFILE_ZONE_SCOPED; LLWorkerClass* workerclass = getWorkerClass(); workerclass->finishWork(getParam(), completed); U32 flags = LLWorkerClass::WCF_WORK_FINISHED | (completed ? 0 : LLWorkerClass::WCF_WORK_ABORTED); @@ -203,7 +207,6 @@ LLWorkerClass::LLWorkerClass(LLWorkerThread* workerthread, const std::string& na : mWorkerThread(workerthread), mWorkerClassName(name), mRequestHandle(LLWorkerThread::nullHandle()), - mRequestPriority(LLWorkerThread::PRIORITY_NORMAL), mMutex(), mWorkFlags(0) { @@ -292,7 +295,7 @@ bool LLWorkerClass::yield() //---------------------------------------------------------------------------- // calls startWork, adds doWork() to queue -void LLWorkerClass::addWork(S32 param, U32 priority) +void LLWorkerClass::addWork(S32 param) { mMutex.lock(); llassert_always(!(mWorkFlags & (WCF_WORKING|WCF_HAVE_WORK))); @@ -306,7 +309,7 @@ void LLWorkerClass::addWork(S32 param, U32 priority) startWork(param); clearFlags(WCF_WORK_FINISHED|WCF_WORK_ABORTED); setFlags(WCF_HAVE_WORK); - mRequestHandle = mWorkerThread->addWorkRequest(this, param, priority); + mRequestHandle = mWorkerThread->addWorkRequest(this, param); mMutex.unlock(); } @@ -321,7 +324,6 @@ void LLWorkerClass::abortWork(bool autocomplete) if (mRequestHandle != LLWorkerThread::nullHandle()) { mWorkerThread->abortRequest(mRequestHandle, autocomplete); - mWorkerThread->setPriority(mRequestHandle, LLQueuedThread::PRIORITY_IMMEDIATE); setFlags(WCF_ABORT_REQUESTED); } mMutex.unlock(); @@ -395,16 +397,5 @@ void LLWorkerClass::scheduleDelete() } } -void LLWorkerClass::setPriority(U32 priority) -{ - mMutex.lock(); - if (mRequestHandle != LLWorkerThread::nullHandle() && mRequestPriority != priority) - { - mRequestPriority = priority; - mWorkerThread->setPriority(mRequestHandle, priority); - } - mMutex.unlock(); -} - //============================================================================ diff --git a/indra/llcommon/llworkerthread.h b/indra/llcommon/llworkerthread.h index 0387e75c65..06bbb7369e 100644 --- a/indra/llcommon/llworkerthread.h +++ b/indra/llcommon/llworkerthread.h @@ -56,7 +56,7 @@ public: virtual ~WorkRequest(); // use deleteRequest() public: - WorkRequest(handle_t handle, U32 priority, LLWorkerClass* workerclass, S32 param); + WorkRequest(handle_t handle, LLWorkerClass* workerclass, S32 param); S32 getParam() { @@ -90,7 +90,7 @@ public: /*virtual*/ S32 update(F32 max_time_ms); - handle_t addWorkRequest(LLWorkerClass* workerclass, S32 param, U32 priority = PRIORITY_NORMAL); + handle_t addWorkRequest(LLWorkerClass* workerclass, S32 param); S32 getNumDeletes() { return (S32)mDeleteList.size(); } // debug @@ -151,10 +151,6 @@ public: bool isWorking() { return getFlags(WCF_WORKING); } bool wasAborted() { return getFlags(WCF_ABORT_REQUESTED); } - // setPriority(): changes the priority of a request - void setPriority(U32 priority); - U32 getPriority() { return mRequestPriority; } - const std::string& getName() const { return mWorkerClassName; } protected: @@ -169,7 +165,7 @@ protected: void setWorkerThread(LLWorkerThread* workerthread); // addWork(): calls startWork, adds doWork() to queue - void addWork(S32 param, U32 priority = LLWorkerThread::PRIORITY_NORMAL); + void addWork(S32 param); // abortWork(): requests that work be aborted void abortWork(bool autocomplete); @@ -193,7 +189,6 @@ protected: LLWorkerThread* mWorkerThread; std::string mWorkerClassName; handle_t mRequestHandle; - U32 mRequestPriority; // last priority set private: LLMutex mMutex; diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index 1d73f7aa0d..41aa858084 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -38,7 +38,7 @@ namespace tut { struct workqueue_data { - WorkQueue queue{"queue"}; + WorkSchedule queue{"queue"}; }; typedef test_group<workqueue_data> workqueue_group; typedef workqueue_group::object object; @@ -49,8 +49,8 @@ namespace tut { set_test_name("name"); ensure_equals("didn't capture name", queue.getKey(), "queue"); - ensure("not findable", WorkQueue::getInstance("queue") == queue.getWeak().lock()); - WorkQueue q2; + ensure("not findable", WorkSchedule::getInstance("queue") == queue.getWeak().lock()); + WorkSchedule q2; ensure("has no name", LLStringUtil::startsWith(q2.getKey(), "WorkQueue")); } @@ -73,16 +73,16 @@ namespace tut { set_test_name("postEvery"); // record of runs - using Shared = std::deque<WorkQueue::TimePoint>; + using Shared = std::deque<WorkSchedule::TimePoint>; // This is an example of how to share data between the originator of - // postEvery(work) and the work item itself, since usually a WorkQueue + // postEvery(work) and the work item itself, since usually a WorkSchedule // is used to dispatch work to a different thread. Neither of them // should call any of LLCond's wait methods: you don't want to stall // either the worker thread or the originating thread (conventionally // main). Use LLCond or a subclass even if all you want to do is // signal the work item that it can quit; consider LLOneShotCond. LLCond<Shared> data; - auto start = WorkQueue::TimePoint::clock::now(); + auto start = WorkSchedule::TimePoint::clock::now(); auto interval = 100ms; queue.postEvery( interval, @@ -93,7 +93,7 @@ namespace tut data.update_one( [](Shared& data) { - data.push_back(WorkQueue::TimePoint::clock::now()); + data.push_back(WorkSchedule::TimePoint::clock::now()); }); // by the 3rd call, return false to stop return (++count < 3); @@ -102,7 +102,7 @@ namespace tut // postEvery() running, so run until we have exhausted the iterations // or we time out waiting for (auto finish = start + 10*interval; - WorkQueue::TimePoint::clock::now() < finish && + WorkSchedule::TimePoint::clock::now() < finish && data.get([](const Shared& data){ return data.size(); }) < 3; ) { queue.runPending(); @@ -139,8 +139,8 @@ namespace tut void object::test<4>() { set_test_name("postTo"); - WorkQueue main("main"); - auto qptr = WorkQueue::getInstance("queue"); + WorkSchedule main("main"); + auto qptr = WorkSchedule::getInstance("queue"); int result = 0; main.postTo( qptr, @@ -171,8 +171,8 @@ namespace tut void object::test<5>() { set_test_name("postTo with void return"); - WorkQueue main("main"); - auto qptr = WorkQueue::getInstance("queue"); + WorkSchedule main("main"); + auto qptr = WorkSchedule::getInstance("queue"); std::string observe; main.postTo( qptr, @@ -194,7 +194,7 @@ namespace tut std::string stored; // Try to call waitForResult() on this thread's main coroutine. It // should throw because the main coroutine must service the queue. - auto what{ catch_what<WorkQueue::Error>( + auto what{ catch_what<WorkSchedule::Error>( [this, &stored](){ stored = queue.waitForResult( [](){ return "should throw"; }); }) }; ensure("lambda should not have run", stored.empty()); diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp index d5adf11264..3a9a5a2062 100644 --- a/indra/llcommon/threadpool.cpp +++ b/indra/llcommon/threadpool.cpp @@ -17,18 +17,58 @@ // std headers // external library headers // other Linden headers +#include "commoncontrol.h" #include "llerror.h" #include "llevents.h" +#include "llsd.h" #include "stringize.h" -LL::ThreadPool::ThreadPool(const std::string& name, size_t threads, size_t capacity): +#include <boost/fiber/algo/round_robin.hpp> + +/***************************************************************************** +* Custom fiber scheduler for worker threads +*****************************************************************************/ +// As of 2022-12-06, each of our worker threads only runs a single (default) +// fiber: we don't launch explicit fibers within worker threads, nor do we +// anticipate doing so. So a worker thread that's simply waiting for incoming +// tasks should really sleep a little. Override the default fiber scheduler to +// implement that. +struct sleepy_robin: public boost::fibers::algo::round_robin +{ + virtual void suspend_until( std::chrono::steady_clock::time_point const&) noexcept + { +#if LL_WINDOWS + // round_robin holds a std::condition_variable, and + // round_robin::suspend_until() calls + // std::condition_variable::wait_until(). On Windows, that call seems + // busier than it ought to be. Try just sleeping. + Sleep(1); +#else + // currently unused other than windows, but might as well have something here + // different units than Sleep(), but we actually just want to sleep for any de-minimis duration + usleep(1); +#endif + } + + virtual void notify() noexcept + { + // Since our Sleep() call above will wake up on its own, we need not + // take any special action to wake it. + } +}; + +/***************************************************************************** +* ThreadPoolBase +*****************************************************************************/ +LL::ThreadPoolBase::ThreadPoolBase(const std::string& name, size_t threads, + WorkQueueBase* queue): super(name), - mQueue(name, capacity), mName("ThreadPool:" + name), - mThreadCount(threads) + mThreadCount(getConfiguredWidth(name, threads)), + mQueue(queue) {} -void LL::ThreadPool::start() +void LL::ThreadPoolBase::start() { for (size_t i = 0; i < mThreadCount; ++i) { @@ -56,17 +96,17 @@ void LL::ThreadPool::start() }); } -LL::ThreadPool::~ThreadPool() +LL::ThreadPoolBase::~ThreadPoolBase() { close(); } -void LL::ThreadPool::close() +void LL::ThreadPoolBase::close() { - if (! mQueue.isClosed()) + if (! mQueue->isClosed()) { LL_DEBUGS("ThreadPool") << mName << " closing queue and joining threads" << LL_ENDL; - mQueue.close(); + mQueue->close(); for (auto& pair: mThreads) { LL_DEBUGS("ThreadPool") << mName << " waiting on thread " << pair.first << LL_ENDL; @@ -76,14 +116,74 @@ void LL::ThreadPool::close() } } -void LL::ThreadPool::run(const std::string& name) +void LL::ThreadPoolBase::run(const std::string& name) { +#if LL_WINDOWS + // Try using sleepy_robin fiber scheduler. + boost::fibers::use_scheduling_algorithm<sleepy_robin>(); +#endif // LL_WINDOWS + LL_DEBUGS("ThreadPool") << name << " starting" << LL_ENDL; run(); LL_DEBUGS("ThreadPool") << name << " stopping" << LL_ENDL; } -void LL::ThreadPool::run() +void LL::ThreadPoolBase::run() +{ + mQueue->runUntilClose(); +} + +//static +size_t LL::ThreadPoolBase::getConfiguredWidth(const std::string& name, size_t dft) +{ + LLSD poolSizes; + try + { + poolSizes = LL::CommonControl::get("Global", "ThreadPoolSizes"); + // "ThreadPoolSizes" is actually a map containing the sizes of + // interest -- or should be, if this process has an + // LLViewerControlListener instance and its settings include + // "ThreadPoolSizes". If we failed to retrieve it, perhaps we're in a + // program that doesn't define that, or perhaps there's no such + // setting, or perhaps we're asking too early, before the LLEventAPI + // itself has been instantiated. In any of those cases, it seems worth + // warning. + if (! poolSizes.isDefined()) + { + // Note: we don't warn about absence of an override key for a + // particular ThreadPool name, that's fine. This warning is about + // complete absence of a ThreadPoolSizes setting, which we expect + // in a normal viewer session. + LL_WARNS("ThreadPool") << "No 'ThreadPoolSizes' setting for ThreadPool '" + << name << "'" << LL_ENDL; + } + } + catch (const LL::CommonControl::Error& exc) + { + // We don't want ThreadPool to *require* LLViewerControlListener. + // Just log it and carry on. + LL_WARNS("ThreadPool") << "Can't check 'ThreadPoolSizes': " << exc.what() << LL_ENDL; + } + + LL_DEBUGS("ThreadPool") << "ThreadPoolSizes = " << poolSizes << LL_ENDL; + // LLSD treats an undefined value as an empty map when asked to retrieve a + // key, so we don't need this to be conditional. + LLSD sizeSpec{ poolSizes[name] }; + // We retrieve sizeSpec as LLSD, rather than immediately as LLSD::Integer, + // so we can distinguish the case when it's undefined. + return sizeSpec.isInteger() ? sizeSpec.asInteger() : dft; +} + +//static +size_t LL::ThreadPoolBase::getWidth(const std::string& name, size_t dft) { - mQueue.runUntilClose(); + auto instance{ getInstance(name) }; + if (instance) + { + return instance->getWidth(); + } + else + { + return getConfiguredWidth(name, dft); + } } diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h index f8eec3b457..60f4a0ce1b 100644 --- a/indra/llcommon/threadpool.h +++ b/indra/llcommon/threadpool.h @@ -13,7 +13,9 @@ #if ! defined(LL_THREADPOOL_H) #define LL_THREADPOOL_H +#include "threadpool_fwd.h" #include "workqueue.h" +#include <memory> // std::unique_ptr #include <string> #include <thread> #include <utility> // std::pair @@ -22,17 +24,24 @@ namespace LL { - class ThreadPool: public LLInstanceTracker<ThreadPool, std::string> + class ThreadPoolBase: public LLInstanceTracker<ThreadPoolBase, std::string> { private: - using super = LLInstanceTracker<ThreadPool, std::string>; + using super = LLInstanceTracker<ThreadPoolBase, std::string>; + public: /** - * Pass ThreadPool a string name. This can be used to look up the + * Pass ThreadPoolBase a string name. This can be used to look up the * relevant WorkQueue. + * + * The number of threads you pass sets the compile-time default. But + * if the user has overridden the LLSD map in the "ThreadPoolSizes" + * setting with a key matching this ThreadPool name, that setting + * overrides this parameter. */ - ThreadPool(const std::string& name, size_t threads=1, size_t capacity=1024); - virtual ~ThreadPool(); + ThreadPoolBase(const std::string& name, size_t threads, + WorkQueueBase* queue); + virtual ~ThreadPoolBase(); /** * Launch the ThreadPool. Until this call, a constructed ThreadPool @@ -50,8 +59,6 @@ namespace LL std::string getName() const { return mName; } size_t getWidth() const { return mThreads.size(); } - /// obtain a non-const reference to the WorkQueue to post work to it - WorkQueue& getQueue() { return mQueue; } /** * Override run() if you need special processing. The default run() @@ -59,15 +66,72 @@ namespace LL */ virtual void run(); + /** + * getConfiguredWidth() returns the setting, if any, for the specified + * ThreadPool name. Returns dft if the "ThreadPoolSizes" map does not + * contain the specified name. + */ + static + size_t getConfiguredWidth(const std::string& name, size_t dft=0); + + /** + * This getWidth() returns the width of the instantiated ThreadPool + * with the specified name, if any. If no instance exists, returns its + * getConfiguredWidth() if any. If there's no instance and no relevant + * override, return dft. Presumably dft should match the threads + * parameter passed to the ThreadPool constructor call that will + * eventually instantiate the ThreadPool with that name. + */ + static + size_t getWidth(const std::string& name, size_t dft); + + protected: + std::unique_ptr<WorkQueueBase> mQueue; + private: void run(const std::string& name); - WorkQueue mQueue; std::string mName; size_t mThreadCount; std::vector<std::pair<std::string, std::thread>> mThreads; }; + /** + * Specialize with WorkQueue or, for timestamped tasks, WorkSchedule + */ + template <class QUEUE> + struct ThreadPoolUsing: public ThreadPoolBase + { + using queue_t = QUEUE; + + /** + * Pass ThreadPoolUsing a string name. This can be used to look up the + * relevant WorkQueue. + * + * The number of threads you pass sets the compile-time default. But + * if the user has overridden the LLSD map in the "ThreadPoolSizes" + * setting with a key matching this ThreadPool name, that setting + * overrides this parameter. + * + * Pass an explicit capacity to limit the size of the queue. + * Constraining the queue can cause a submitter to block. Do not + * constrain any ThreadPool accepting work from the main thread. + */ + ThreadPoolUsing(const std::string& name, size_t threads=1, size_t capacity=1024*1024): + ThreadPoolBase(name, threads, new queue_t(name, capacity)) + {} + ~ThreadPoolUsing() override {} + + /** + * obtain a non-const reference to the specific WorkQueue subclass to + * post work to it + */ + queue_t& getQueue() { return static_cast<queue_t&>(*mQueue); } + }; + + /// ThreadPool is shorthand for using the simpler WorkQueue + using ThreadPool = ThreadPoolUsing<WorkQueue>; + } // namespace LL #endif /* ! defined(LL_THREADPOOL_H) */ diff --git a/indra/llcommon/threadpool_fwd.h b/indra/llcommon/threadpool_fwd.h new file mode 100644 index 0000000000..1aa3c4a0e2 --- /dev/null +++ b/indra/llcommon/threadpool_fwd.h @@ -0,0 +1,25 @@ +/** + * @file threadpool_fwd.h + * @author Nat Goodspeed + * @date 2022-12-09 + * @brief Forward declarations for ThreadPool et al. + * + * $LicenseInfo:firstyear=2022&license=viewerlgpl$ + * Copyright (c) 2022, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_THREADPOOL_FWD_H) +#define LL_THREADPOOL_FWD_H + +#include "workqueue.h" + +namespace LL +{ + template <class QUEUE> + struct ThreadPoolUsing; + + using ThreadPool = ThreadPoolUsing<WorkQueue>; +} // namespace LL + +#endif /* ! defined(LL_THREADPOOL_FWD_H) */ diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index eb06890468..83e0216ae7 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -26,83 +26,65 @@ using Mutex = LLCoros::Mutex; using Lock = LLCoros::LockType; -LL::WorkQueue::WorkQueue(const std::string& name, size_t capacity): - super(makeName(name)), - mQueue(capacity) +/***************************************************************************** +* WorkQueueBase +*****************************************************************************/ +LL::WorkQueueBase::WorkQueueBase(const std::string& name): + super(makeName(name)) { // TODO: register for "LLApp" events so we can implicitly close() on // viewer shutdown. } -void LL::WorkQueue::close() -{ - mQueue.close(); -} - -size_t LL::WorkQueue::size() -{ - return mQueue.size(); -} - -bool LL::WorkQueue::isClosed() -{ - return mQueue.isClosed(); -} - -bool LL::WorkQueue::done() -{ - return mQueue.done(); -} - -void LL::WorkQueue::runUntilClose() +void LL::WorkQueueBase::runUntilClose() { try { for (;;) { LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; - callWork(mQueue.pop()); + callWork(pop_()); } } - catch (const Queue::Closed&) + catch (const Closed&) { } } -bool LL::WorkQueue::runPending() +bool LL::WorkQueueBase::runPending() { LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; - for (Work work; mQueue.tryPop(work); ) + for (Work work; tryPop_(work); ) { callWork(work); } - return ! mQueue.done(); + return ! done(); } -bool LL::WorkQueue::runOne() +bool LL::WorkQueueBase::runOne() { Work work; - if (mQueue.tryPop(work)) + if (tryPop_(work)) { callWork(work); } - return ! mQueue.done(); + return ! done(); } -bool LL::WorkQueue::runUntil(const TimePoint& until) +bool LL::WorkQueueBase::runUntil(const TimePoint& until) { LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // Should we subtract some slop to allow for typical Work execution time? // How much slop? // runUntil() is simply a time-bounded runPending(). - for (Work work; TimePoint::clock::now() < until && mQueue.tryPop(work); ) + for (Work work; TimePoint::clock::now() < until && tryPop_(work); ) { callWork(work); } - return ! mQueue.done(); + return ! done(); } -std::string LL::WorkQueue::makeName(const std::string& name) +std::string LL::WorkQueueBase::makeName(const std::string& name) { if (! name.empty()) return name; @@ -120,14 +102,7 @@ std::string LL::WorkQueue::makeName(const std::string& name) return STRINGIZE("WorkQueue" << num); } -void LL::WorkQueue::callWork(const Queue::DataTuple& work) -{ - // ThreadSafeSchedule::pop() always delivers a tuple, even when - // there's only one data field per item, as for us. - callWork(std::get<0>(work)); -} - -void LL::WorkQueue::callWork(const Work& work) +void LL::WorkQueueBase::callWork(const Work& work) { LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; try @@ -142,12 +117,12 @@ void LL::WorkQueue::callWork(const Work& work) } } -void LL::WorkQueue::error(const std::string& msg) +void LL::WorkQueueBase::error(const std::string& msg) { LL_ERRS("WorkQueue") << msg << LL_ENDL; } -void LL::WorkQueue::checkCoroutine(const std::string& method) +void LL::WorkQueueBase::checkCoroutine(const std::string& method) { // By convention, the default coroutine on each thread has an empty name // string. See also LLCoros::logname(). @@ -156,3 +131,130 @@ void LL::WorkQueue::checkCoroutine(const std::string& method) LLTHROW(Error("Do not call " + method + " from a thread's default coroutine")); } } + +/***************************************************************************** +* WorkQueue +*****************************************************************************/ +LL::WorkQueue::WorkQueue(const std::string& name, size_t capacity): + super(name), + mQueue(capacity) +{ +} + +void LL::WorkQueue::close() +{ + mQueue.close(); +} + +size_t LL::WorkQueue::size() +{ + return mQueue.size(); +} + +bool LL::WorkQueue::isClosed() +{ + return mQueue.isClosed(); +} + +bool LL::WorkQueue::done() +{ + return mQueue.done(); +} + +void LL::WorkQueue::post(const Work& callable) +{ + mQueue.push(callable); +} + +bool LL::WorkQueue::postIfOpen(const Work& callable) +{ + return mQueue.pushIfOpen(callable); +} + +bool LL::WorkQueue::tryPost(const Work& callable) +{ + return mQueue.tryPush(callable); +} + +LL::WorkQueue::Work LL::WorkQueue::pop_() +{ + return mQueue.pop(); +} + +bool LL::WorkQueue::tryPop_(Work& work) +{ + return mQueue.tryPop(work); +} + +/***************************************************************************** +* WorkSchedule +*****************************************************************************/ +LL::WorkSchedule::WorkSchedule(const std::string& name, size_t capacity): + super(name), + mQueue(capacity) +{ +} + +void LL::WorkSchedule::close() +{ + mQueue.close(); +} + +size_t LL::WorkSchedule::size() +{ + return mQueue.size(); +} + +bool LL::WorkSchedule::isClosed() +{ + return mQueue.isClosed(); +} + +bool LL::WorkSchedule::done() +{ + return mQueue.done(); +} + +void LL::WorkSchedule::post(const Work& callable) +{ + // Use TimePoint::clock::now() instead of TimePoint's representation of + // the epoch because this WorkSchedule may contain a mix of past-due + // TimedWork items and TimedWork items scheduled for the future. Sift this + // new item into the correct place. + post(callable, TimePoint::clock::now()); +} + +void LL::WorkSchedule::post(const Work& callable, const TimePoint& time) +{ + mQueue.push(TimedWork(time, callable)); +} + +bool LL::WorkSchedule::postIfOpen(const Work& callable) +{ + return postIfOpen(callable, TimePoint::clock::now()); +} + +bool LL::WorkSchedule::postIfOpen(const Work& callable, const TimePoint& time) +{ + return mQueue.pushIfOpen(TimedWork(time, callable)); +} + +bool LL::WorkSchedule::tryPost(const Work& callable) +{ + return tryPost(callable, TimePoint::clock::now()); +} + +bool LL::WorkSchedule::tryPost(const Work& callable, const TimePoint& time) +{ + return mQueue.tryPush(TimedWork(time, callable)); +} + +LL::WorkSchedule::Work LL::WorkSchedule::pop_() +{ + return std::get<0>(mQueue.pop()); +} + +bool LL::WorkSchedule::tryPop_(Work& work) +{ + return mQueue.tryPop(work); +} diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index 70fd65bd0c..5461ce6c23 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -15,6 +15,7 @@ #include "llcoros.h" #include "llexception.h" #include "llinstancetracker.h" +#include "llinstancetrackersubclass.h" #include "threadsafeschedule.h" #include <chrono> #include <exception> // std::current_exception @@ -23,27 +24,23 @@ namespace LL { + +/***************************************************************************** +* WorkQueueBase: API for WorkQueue and WorkSchedule +*****************************************************************************/ /** * A typical WorkQueue has a string name that can be used to find it. */ - class WorkQueue: public LLInstanceTracker<WorkQueue, std::string> + class WorkQueueBase: public LLInstanceTracker<WorkQueueBase, std::string> { private: - using super = LLInstanceTracker<WorkQueue, std::string>; + using super = LLInstanceTracker<WorkQueueBase, std::string>; public: using Work = std::function<void()>; - - private: - using Queue = ThreadSafeSchedule<Work>; - // helper for postEvery() - template <typename Rep, typename Period, typename CALLABLE> - class BackJack; - - public: - using TimePoint = Queue::TimePoint; - using TimedWork = Queue::TimeTuple; - using Closed = Queue::Closed; + using Closed = LLThreadSafeQueueInterrupt; + // for runFor() + using TimePoint = std::chrono::steady_clock::time_point; struct Error: public LLException { @@ -51,18 +48,18 @@ namespace LL }; /** - * You may omit the WorkQueue name, in which case a unique name is + * You may omit the WorkQueueBase name, in which case a unique name is * synthesized; for practical purposes that makes it anonymous. */ - WorkQueue(const std::string& name = std::string(), size_t capacity=1024); + WorkQueueBase(const std::string& name); /** * Since the point of WorkQueue is to pass work to some other worker - * thread(s) asynchronously, it's important that the WorkQueue continue - * to exist until the worker thread(s) have drained it. To communicate - * that it's time for them to quit, close() the queue. + * thread(s) asynchronously, it's important that it continue to exist + * until the worker thread(s) have drained it. To communicate that + * it's time for them to quit, close() the queue. */ - void close(); + virtual void close() = 0; /** * WorkQueue supports multiple producers and multiple consumers. In @@ -78,152 +75,60 @@ namespace LL * * If you're the only consumer, noticing that size() > 0 is * meaningful. */ - size_t size(); + virtual size_t size() = 0; /// producer end: are we prevented from pushing any additional items? - bool isClosed(); + virtual bool isClosed() = 0; /// consumer end: are we done, is the queue entirely drained? - bool done(); + virtual bool done() = 0; /*---------------------- fire and forget API -----------------------*/ - /// fire-and-forget, but at a particular (future?) time - template <typename CALLABLE> - void post(const TimePoint& time, CALLABLE&& callable) - { - // Defer reifying an arbitrary CALLABLE until we hit this or - // postIfOpen(). All other methods should accept CALLABLEs of - // arbitrary type to avoid multiple levels of std::function - // indirection. - mQueue.push(TimedWork(time, std::move(callable))); - } - /// fire-and-forget - template <typename CALLABLE> - void post(CALLABLE&& callable) - { - // We use TimePoint::clock::now() instead of TimePoint's - // representation of the epoch because this WorkQueue may contain - // a mix of past-due TimedWork items and TimedWork items scheduled - // for the future. Sift this new item into the correct place. - post(TimePoint::clock::now(), std::move(callable)); - } - - /** - * post work for a particular time, unless the queue is closed before - * we can post - */ - template <typename CALLABLE> - bool postIfOpen(const TimePoint& time, CALLABLE&& callable) - { - // Defer reifying an arbitrary CALLABLE until we hit this or - // post(). All other methods should accept CALLABLEs of arbitrary - // type to avoid multiple levels of std::function indirection. - return mQueue.pushIfOpen(TimedWork(time, std::move(callable))); - } + virtual void post(const Work&) = 0; /** * post work, unless the queue is closed before we can post */ - template <typename CALLABLE> - bool postIfOpen(CALLABLE&& callable) - { - return postIfOpen(TimePoint::clock::now(), std::move(callable)); - } + virtual bool postIfOpen(const Work&) = 0; /** - * Post work to be run at a specified time to another WorkQueue, which - * may or may not still exist and be open. Return true if we were able - * to post. + * post work, unless the queue is full */ - template <typename CALLABLE> - static bool postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable); + virtual bool tryPost(const Work&) = 0; /** * Post work to another WorkQueue, which may or may not still exist - * and be open. Return true if we were able to post. + * and be open. Support any post() overload. Return true if we were + * able to post. */ - template <typename CALLABLE> - static bool postMaybe(weak_t target, CALLABLE&& callable) - { - return postMaybe(target, TimePoint::clock::now(), - std::forward<CALLABLE>(callable)); - } - - /** - * Launch a callable returning bool that will trigger repeatedly at - * specified interval, until the callable returns false. - * - * If you need to signal that callable from outside, DO NOT bind a - * reference to a simple bool! That's not thread-safe. Instead, bind - * an LLCond variant, e.g. LLOneShotCond or LLBoolCond. - */ - template <typename Rep, typename Period, typename CALLABLE> - void postEvery(const std::chrono::duration<Rep, Period>& interval, - CALLABLE&& callable); - - template <typename CALLABLE> - bool tryPost(CALLABLE&& callable) - { - return mQueue.tryPush(TimedWork(TimePoint::clock::now(), std::move(callable))); - } + template <typename... ARGS> + static bool postMaybe(weak_t target, ARGS&&... args); /*------------------------- handshake API --------------------------*/ /** - * Post work to another WorkQueue to be run at a specified time, - * requesting a specific callback to be run on this WorkQueue on - * completion. - * - * Returns true if able to post, false if the other WorkQueue is - * inaccessible. - */ - // Apparently some Microsoft header file defines a macro CALLBACK? The - // natural template argument name CALLBACK produces very weird Visual - // Studio compile errors that seem utterly unrelated to this source - // code. - template <typename CALLABLE, typename FOLLOWUP> - bool postTo(weak_t target, - const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback); - - /** * Post work to another WorkQueue, requesting a specific callback to - * be run on this WorkQueue on completion. + * be run on this WorkQueue on completion. Optional final argument is + * TimePoint for WorkSchedule. * * Returns true if able to post, false if the other WorkQueue is * inaccessible. */ - template <typename CALLABLE, typename FOLLOWUP> - bool postTo(weak_t target, CALLABLE&& callable, FOLLOWUP&& callback) - { - return postTo(target, TimePoint::clock::now(), - std::move(callable), std::move(callback)); - } - - /** - * Post work to another WorkQueue to be run at a specified time, - * blocking the calling coroutine until then, returning the result to - * caller on completion. - * - * In general, we assume that each thread's default coroutine is busy - * servicing its WorkQueue or whatever. To try to prevent mistakes, we - * forbid calling waitForResult() from a thread's default coroutine. - */ - template <typename CALLABLE> - auto waitForResult(const TimePoint& time, CALLABLE&& callable); + template <typename CALLABLE, typename FOLLOWUP, typename... ARGS> + bool postTo(weak_t target, CALLABLE&& callable, FOLLOWUP&& callback, + ARGS&&... args); /** * Post work to another WorkQueue, blocking the calling coroutine - * until then, returning the result to caller on completion. + * until then, returning the result to caller on completion. Optional + * final argument is TimePoint for WorkSchedule. * * In general, we assume that each thread's default coroutine is busy * servicing its WorkQueue or whatever. To try to prevent mistakes, we * forbid calling waitForResult() from a thread's default coroutine. */ - template <typename CALLABLE> - auto waitForResult(CALLABLE&& callable) - { - return waitForResult(TimePoint::clock::now(), std::move(callable)); - } + template <typename CALLABLE, typename... ARGS> + auto waitForResult(CALLABLE&& callable, ARGS&&... args); /*--------------------------- worker API ---------------------------*/ @@ -270,7 +175,7 @@ namespace LL */ bool runUntil(const TimePoint& until); - private: + protected: template <typename CALLABLE, typename FOLLOWUP> static auto makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback); /// general case: arbitrary C++ return type @@ -290,13 +195,179 @@ namespace LL static void checkCoroutine(const std::string& method); static void error(const std::string& msg); static std::string makeName(const std::string& name); - void callWork(const Queue::DataTuple& work); void callWork(const Work& work); + + private: + virtual Work pop_() = 0; + virtual bool tryPop_(Work&) = 0; + }; + +/***************************************************************************** +* WorkQueue: no timestamped task support +*****************************************************************************/ + class WorkQueue: public LLInstanceTrackerSubclass<WorkQueue, WorkQueueBase> + { + private: + using super = LLInstanceTrackerSubclass<WorkQueue, WorkQueueBase>; + + public: + /** + * You may omit the WorkQueue name, in which case a unique name is + * synthesized; for practical purposes that makes it anonymous. + */ + WorkQueue(const std::string& name = std::string(), size_t capacity=1024); + + /** + * Since the point of WorkQueue is to pass work to some other worker + * thread(s) asynchronously, it's important that it continue to exist + * until the worker thread(s) have drained it. To communicate that + * it's time for them to quit, close() the queue. + */ + void close() override; + + /** + * WorkQueue supports multiple producers and multiple consumers. In + * the general case it's misleading to test size(), since any other + * thread might change it the nanosecond the lock is released. On that + * basis, some might argue against publishing a size() method at all. + * + * But there are two specific cases in which a test based on size() + * might be reasonable: + * + * * If you're the only producer, noticing that size() == 0 is + * meaningful. + * * If you're the only consumer, noticing that size() > 0 is + * meaningful. + */ + size_t size() override; + /// producer end: are we prevented from pushing any additional items? + bool isClosed() override; + /// consumer end: are we done, is the queue entirely drained? + bool done() override; + + /*---------------------- fire and forget API -----------------------*/ + + /// fire-and-forget + void post(const Work&) override; + + /** + * post work, unless the queue is closed before we can post + */ + bool postIfOpen(const Work&) override; + + /** + * post work, unless the queue is full + */ + bool tryPost(const Work&) override; + + private: + using Queue = LLThreadSafeQueue<Work>; Queue mQueue; + + Work pop_() override; + bool tryPop_(Work&) override; + }; + +/***************************************************************************** +* WorkSchedule: add support for timestamped tasks +*****************************************************************************/ + class WorkSchedule: public LLInstanceTrackerSubclass<WorkSchedule, WorkQueueBase> + { + private: + using super = LLInstanceTrackerSubclass<WorkSchedule, WorkQueueBase>; + using Queue = ThreadSafeSchedule<Work>; + // helper for postEvery() + template <typename Rep, typename Period, typename CALLABLE> + class BackJack; + + public: + using TimePoint = Queue::TimePoint; + using TimedWork = Queue::TimeTuple; + + /** + * You may omit the WorkSchedule name, in which case a unique name is + * synthesized; for practical purposes that makes it anonymous. + */ + WorkSchedule(const std::string& name = std::string(), size_t capacity=1024); + + /** + * Since the point of WorkSchedule is to pass work to some other worker + * thread(s) asynchronously, it's important that the WorkSchedule continue + * to exist until the worker thread(s) have drained it. To communicate + * that it's time for them to quit, close() the queue. + */ + void close() override; + + /** + * WorkSchedule supports multiple producers and multiple consumers. In + * the general case it's misleading to test size(), since any other + * thread might change it the nanosecond the lock is released. On that + * basis, some might argue against publishing a size() method at all. + * + * But there are two specific cases in which a test based on size() + * might be reasonable: + * + * * If you're the only producer, noticing that size() == 0 is + * meaningful. + * * If you're the only consumer, noticing that size() > 0 is + * meaningful. + */ + size_t size() override; + /// producer end: are we prevented from pushing any additional items? + bool isClosed() override; + /// consumer end: are we done, is the queue entirely drained? + bool done() override; + + /*---------------------- fire and forget API -----------------------*/ + + /// fire-and-forget + void post(const Work& callable) override; + + /// fire-and-forget, but at a particular (future?) time + void post(const Work& callable, const TimePoint& time); + + /** + * post work, unless the queue is closed before we can post + */ + bool postIfOpen(const Work& callable) override; + + /** + * post work for a particular time, unless the queue is closed before + * we can post + */ + bool postIfOpen(const Work& callable, const TimePoint& time); + + /** + * post work, unless the queue is full + */ + bool tryPost(const Work& callable) override; + + /** + * post work for a particular time, unless the queue is full + */ + bool tryPost(const Work& callable, const TimePoint& time); + + /** + * Launch a callable returning bool that will trigger repeatedly at + * specified interval, until the callable returns false. + * + * If you need to signal that callable from outside, DO NOT bind a + * reference to a simple bool! That's not thread-safe. Instead, bind + * an LLCond variant, e.g. LLOneShotCond or LLBoolCond. + */ + template <typename Rep, typename Period, typename CALLABLE> + void postEvery(const std::chrono::duration<Rep, Period>& interval, + CALLABLE&& callable); + + private: + Queue mQueue; + + Work pop_() override; + bool tryPop_(Work&) override; }; /** - * BackJack is, in effect, a hand-rolled lambda, binding a WorkQueue, a + * BackJack is, in effect, a hand-rolled lambda, binding a WorkSchedule, a * CALLABLE that returns bool, a TimePoint and an interval at which to * relaunch it. As long as the callable continues returning true, BackJack * keeps resubmitting it to the target WorkQueue. @@ -305,7 +376,7 @@ namespace LL // class method gets its own 'this' pointer -- which we need to resubmit // the whole BackJack callable. template <typename Rep, typename Period, typename CALLABLE> - class WorkQueue::BackJack + class WorkSchedule::BackJack { public: // bind the desired data @@ -319,9 +390,10 @@ namespace LL mCallable(std::move(callable)) {} - // Call by target WorkQueue -- note that although WE require a - // callable returning bool, WorkQueue wants a void callable. We - // consume the bool. + // This operator() method, called by target WorkSchedule, is what + // makes this object a Work item. Although WE require a callable + // returning bool, WorkSchedule wants a void callable. We consume the + // bool. void operator()() { // If mCallable() throws an exception, don't catch it here: if it @@ -337,7 +409,7 @@ namespace LL // register our intent to fire at exact mIntervals. mStart += mInterval; - // We're being called at this moment by the target WorkQueue. + // We're being called at this moment by the target WorkSchedule. // Assume it still exists, rather than checking the result of // lock(). // Resubmit the whole *this callable: that's why we're a class @@ -347,7 +419,8 @@ namespace LL // moved-from. try { - mTarget.lock()->post(mStart, std::move(*this)); + auto target{ std::dynamic_pointer_cast<WorkSchedule>(mTarget.lock()) }; + target->post(std::move(*this), mStart); } catch (const Closed&) { @@ -364,8 +437,8 @@ namespace LL }; template <typename Rep, typename Period, typename CALLABLE> - void WorkQueue::postEvery(const std::chrono::duration<Rep, Period>& interval, - CALLABLE&& callable) + void WorkSchedule::postEvery(const std::chrono::duration<Rep, Period>& interval, + CALLABLE&& callable) { if (interval.count() <= 0) { @@ -388,7 +461,7 @@ namespace LL /// general case: arbitrary C++ return type template <typename CALLABLE, typename FOLLOWUP, typename RETURNTYPE> - struct WorkQueue::MakeReplyLambda + struct WorkQueueBase::MakeReplyLambda { auto operator()(CALLABLE&& callable, FOLLOWUP&& callback) { @@ -409,7 +482,7 @@ namespace LL /// specialize for CALLABLE returning void template <typename CALLABLE, typename FOLLOWUP> - struct WorkQueue::MakeReplyLambda<CALLABLE, FOLLOWUP, void> + struct WorkQueueBase::MakeReplyLambda<CALLABLE, FOLLOWUP, void> { auto operator()(CALLABLE&& callable, FOLLOWUP&& callback) { @@ -421,16 +494,16 @@ namespace LL }; template <typename CALLABLE, typename FOLLOWUP> - auto WorkQueue::makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback) + auto WorkQueueBase::makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback) { return MakeReplyLambda<CALLABLE, FOLLOWUP, decltype(std::forward<CALLABLE>(callable)())>() (std::move(callable), std::move(callback)); } - template <typename CALLABLE, typename FOLLOWUP> - bool WorkQueue::postTo(weak_t target, - const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) + template <typename CALLABLE, typename FOLLOWUP, typename... ARGS> + bool WorkQueueBase::postTo(weak_t target, CALLABLE&& callable, FOLLOWUP&& callback, + ARGS&&... args) { LL_PROFILE_ZONE_SCOPED; // We're being asked to post to the WorkQueue at target. @@ -444,12 +517,11 @@ namespace LL // lambda that packages our callable, our callback and a weak_ptr // to this originating WorkQueue. tptr->post( - time, [reply = super::getWeak(), callable = std::move(callable), callback = std::move(callback)] - () - mutable { + () mutable + { // Use postMaybe() below in case this originating WorkQueue // has been closed or destroyed. Remember, the outer lambda is // now running on a thread servicing the target WorkQueue, and @@ -472,14 +544,16 @@ namespace LL // originating WorkQueue. Once there, rethrow it. [exc = std::current_exception()](){ std::rethrow_exception(exc); }); } - }); + }, + // if caller passed a TimePoint, pass it along to post() + std::forward<ARGS>(args)...); // looks like we were able to post() return true; } - template <typename CALLABLE> - bool WorkQueue::postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable) + template <typename... ARGS> + bool WorkQueueBase::postMaybe(weak_t target, ARGS&&... args) { LL_PROFILE_ZONE_SCOPED; // target is a weak_ptr: have to lock it to check it @@ -488,7 +562,7 @@ namespace LL { try { - tptr->post(time, std::forward<CALLABLE>(callable)); + tptr->post(std::forward<ARGS>(args)...); // we were able to post() return true; } @@ -503,13 +577,13 @@ namespace LL /// general case: arbitrary C++ return type template <typename CALLABLE, typename RETURNTYPE> - struct WorkQueue::WaitForResult + struct WorkQueueBase::WaitForResult { - auto operator()(WorkQueue* self, const TimePoint& time, CALLABLE&& callable) + template <typename... ARGS> + auto operator()(WorkQueueBase* self, CALLABLE&& callable, ARGS&&... args) { LLCoros::Promise<RETURNTYPE> promise; self->post( - time, // We dare to bind a reference to Promise because it's // specifically designed for cross-thread communication. [&promise, callable = std::move(callable)]() @@ -523,7 +597,9 @@ namespace LL { promise.set_exception(std::current_exception()); } - }); + }, + // if caller passed a TimePoint, pass it to post() + std::forward<ARGS>(args)...); auto future{ LLCoros::getFuture(promise) }; // now, on the calling thread, wait for that result LLCoros::TempStatus st("waiting for WorkQueue::waitForResult()"); @@ -533,13 +609,13 @@ namespace LL /// specialize for CALLABLE returning void template <typename CALLABLE> - struct WorkQueue::WaitForResult<CALLABLE, void> + struct WorkQueueBase::WaitForResult<CALLABLE, void> { - void operator()(WorkQueue* self, const TimePoint& time, CALLABLE&& callable) + template <typename... ARGS> + void operator()(WorkQueueBase* self, CALLABLE&& callable, ARGS&&... args) { LLCoros::Promise<void> promise; self->post( - time, // &promise is designed for cross-thread access [&promise, callable = std::move(callable)]() mutable { @@ -552,7 +628,9 @@ namespace LL { promise.set_exception(std::current_exception()); } - }); + }, + // if caller passed a TimePoint, pass it to post() + std::forward<ARGS>(args)...); auto future{ LLCoros::getFuture(promise) }; // block until set_value() LLCoros::TempStatus st("waiting for void WorkQueue::waitForResult()"); @@ -560,13 +638,13 @@ namespace LL } }; - template <typename CALLABLE> - auto WorkQueue::waitForResult(const TimePoint& time, CALLABLE&& callable) + template <typename CALLABLE, typename... ARGS> + auto WorkQueueBase::waitForResult(CALLABLE&& callable, ARGS&&... args) { checkCoroutine("waitForResult()"); // derive callable's return type so we can specialize for void return WaitForResult<CALLABLE, decltype(std::forward<CALLABLE>(callable)())>() - (this, time, std::forward<CALLABLE>(callable)); + (this, std::forward<CALLABLE>(callable), std::forward<ARGS>(args)...); } } // namespace LL |