[mlir][py] Enable loading only specified dialects during creation. (#121421)
[llvm-project.git] / lldb / source / Core / DataFileCache.cpp
blobef0e07a8b03420d35e291c2f919b061b276d0ff0
1 //===-- DataFileCache.cpp -------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
9 #include "lldb/Core/DataFileCache.h"
10 #include "lldb/Core/Module.h"
11 #include "lldb/Core/ModuleList.h"
12 #include "lldb/Host/FileSystem.h"
13 #include "lldb/Symbol/ObjectFile.h"
14 #include "lldb/Utility/DataEncoder.h"
15 #include "lldb/Utility/LLDBLog.h"
16 #include "lldb/Utility/Log.h"
17 #include "llvm/Support/CachePruning.h"
19 using namespace lldb_private;
22 llvm::CachePruningPolicy DataFileCache::GetLLDBIndexCachePolicy() {
23 static llvm::CachePruningPolicy policy;
24 static llvm::once_flag once_flag;
26 llvm::call_once(once_flag, []() {
27 // Prune the cache based off of the LLDB settings each time we create a
28 // cache object.
29 ModuleListProperties &properties =
30 ModuleList::GetGlobalModuleListProperties();
31 // Only scan once an hour. If we have lots of debug sessions we don't want
32 // to scan this directory too often. A timestamp file is written to the
33 // directory to ensure different processes don't scan the directory too
34 // often. This setting doesn't mean that a thread will continually scan the
35 // cache directory within this process.
36 policy.Interval = std::chrono::hours(1);
37 // Get the user settings for pruning.
38 policy.MaxSizeBytes = properties.GetLLDBIndexCacheMaxByteSize();
39 policy.MaxSizePercentageOfAvailableSpace =
40 properties.GetLLDBIndexCacheMaxPercent();
41 policy.Expiration =
42 std::chrono::hours(properties.GetLLDBIndexCacheExpirationDays() * 24);
43 });
44 return policy;
47 DataFileCache::DataFileCache(llvm::StringRef path, llvm::CachePruningPolicy policy) {
48 m_cache_dir.SetPath(path);
49 pruneCache(path, policy);
51 // This lambda will get called when the data is gotten from the cache and
52 // also after the data was set for a given key. We only need to take
53 // ownership of the data if we are geting the data, so we use the
54 // m_take_ownership member variable to indicate if we need to take
55 // ownership.
57 auto add_buffer = [this](unsigned task, const llvm::Twine &moduleName,
58 std::unique_ptr<llvm::MemoryBuffer> m) {
59 if (m_take_ownership)
60 m_mem_buff_up = std::move(m);
62 llvm::Expected<llvm::FileCache> cache_or_err =
63 llvm::localCache("LLDBModuleCache", "lldb-module", path, add_buffer);
64 if (cache_or_err)
65 m_cache_callback = std::move(*cache_or_err);
66 else {
67 Log *log = GetLog(LLDBLog::Modules);
68 LLDB_LOG_ERROR(log, cache_or_err.takeError(),
69 "failed to create lldb index cache directory: {0}");
73 std::unique_ptr<llvm::MemoryBuffer>
74 DataFileCache::GetCachedData(llvm::StringRef key) {
75 std::lock_guard<std::mutex> guard(m_mutex);
77 const unsigned task = 1;
78 m_take_ownership = true;
79 // If we call the "m_cache_callback" function and the data is cached, it will
80 // call the "add_buffer" lambda function from the constructor which will in
81 // turn take ownership of the member buffer that is passed to the callback and
82 // put it into a member variable.
83 llvm::Expected<llvm::AddStreamFn> add_stream_or_err =
84 m_cache_callback(task, key, "");
85 m_take_ownership = false;
86 // At this point we either already called the "add_buffer" lambda with
87 // the data or we haven't. We can tell if we got the cached data by checking
88 // the add_stream function pointer value below.
89 if (add_stream_or_err) {
90 llvm::AddStreamFn &add_stream = *add_stream_or_err;
91 // If the "add_stream" is nullptr, then the data was cached and we already
92 // called the "add_buffer" lambda. If it is valid, then if we were to call
93 // the add_stream function it would cause a cache file to get generated
94 // and we would be expected to fill in the data. In this function we only
95 // want to check if the data was cached, so we don't want to call
96 // "add_stream" in this function.
97 if (!add_stream)
98 return std::move(m_mem_buff_up);
99 } else {
100 Log *log = GetLog(LLDBLog::Modules);
101 LLDB_LOG_ERROR(log, add_stream_or_err.takeError(),
102 "failed to get the cache add stream callback for key: {0}");
104 // Data was not cached.
105 return std::unique_ptr<llvm::MemoryBuffer>();
108 bool DataFileCache::SetCachedData(llvm::StringRef key,
109 llvm::ArrayRef<uint8_t> data) {
110 std::lock_guard<std::mutex> guard(m_mutex);
111 const unsigned task = 2;
112 // If we call this function and the data is cached, it will call the
113 // add_buffer lambda function from the constructor which will ignore the
114 // data.
115 llvm::Expected<llvm::AddStreamFn> add_stream_or_err =
116 m_cache_callback(task, key, "");
117 // If we reach this code then we either already called the callback with
118 // the data or we haven't. We can tell if we had the cached data by checking
119 // the CacheAddStream function pointer value below.
120 if (add_stream_or_err) {
121 llvm::AddStreamFn &add_stream = *add_stream_or_err;
122 // If the "add_stream" is nullptr, then the data was cached. If it is
123 // valid, then if we call the add_stream function with a task it will
124 // cause the file to get generated, but we only want to check if the data
125 // is cached here, so we don't want to call it here. Note that the
126 // add_buffer will also get called in this case after the data has been
127 // provided, but we won't take ownership of the memory buffer as we just
128 // want to write the data.
129 if (add_stream) {
130 llvm::Expected<std::unique_ptr<llvm::CachedFileStream>> file_or_err =
131 add_stream(task, "");
132 if (file_or_err) {
133 llvm::CachedFileStream *cfs = file_or_err->get();
134 cfs->OS->write((const char *)data.data(), data.size());
135 return true;
136 } else {
137 Log *log = GetLog(LLDBLog::Modules);
138 LLDB_LOG_ERROR(log, file_or_err.takeError(),
139 "failed to get the cache file stream for key: {0}");
142 } else {
143 Log *log = GetLog(LLDBLog::Modules);
144 LLDB_LOG_ERROR(log, add_stream_or_err.takeError(),
145 "failed to get the cache add stream callback for key: {0}");
147 return false;
150 FileSpec DataFileCache::GetCacheFilePath(llvm::StringRef key) {
151 FileSpec cache_file(m_cache_dir);
152 std::string filename("llvmcache-");
153 filename += key.str();
154 cache_file.AppendPathComponent(filename);
155 return cache_file;
158 Status DataFileCache::RemoveCacheFile(llvm::StringRef key) {
159 FileSpec cache_file = GetCacheFilePath(key);
160 FileSystem &fs = FileSystem::Instance();
161 if (!fs.Exists(cache_file))
162 return Status();
163 return fs.RemoveFile(cache_file);
166 CacheSignature::CacheSignature(lldb_private::Module *module) {
167 Clear();
168 UUID uuid = module->GetUUID();
169 if (uuid.IsValid())
170 m_uuid = uuid;
172 std::time_t mod_time = 0;
173 mod_time = llvm::sys::toTimeT(module->GetModificationTime());
174 if (mod_time != 0)
175 m_mod_time = mod_time;
177 mod_time = llvm::sys::toTimeT(module->GetObjectModificationTime());
178 if (mod_time != 0)
179 m_obj_mod_time = mod_time;
182 CacheSignature::CacheSignature(lldb_private::ObjectFile *objfile) {
183 Clear();
184 UUID uuid = objfile->GetUUID();
185 if (uuid.IsValid())
186 m_uuid = uuid;
188 std::time_t mod_time = 0;
189 // Grab the modification time of the object file's file. It isn't always the
190 // same as the module's file when you have a executable file as the main
191 // executable, and you have a object file for a symbol file.
192 FileSystem &fs = FileSystem::Instance();
193 mod_time = llvm::sys::toTimeT(fs.GetModificationTime(objfile->GetFileSpec()));
194 if (mod_time != 0)
195 m_mod_time = mod_time;
197 mod_time =
198 llvm::sys::toTimeT(objfile->GetModule()->GetObjectModificationTime());
199 if (mod_time != 0)
200 m_obj_mod_time = mod_time;
203 enum SignatureEncoding {
204 eSignatureUUID = 1u,
205 eSignatureModTime = 2u,
206 eSignatureObjectModTime = 3u,
207 eSignatureEnd = 255u,
210 bool CacheSignature::Encode(DataEncoder &encoder) const {
211 if (!IsValid())
212 return false; // Invalid signature, return false!
214 if (m_uuid) {
215 llvm::ArrayRef<uint8_t> uuid_bytes = m_uuid->GetBytes();
216 encoder.AppendU8(eSignatureUUID);
217 encoder.AppendU8(uuid_bytes.size());
218 encoder.AppendData(uuid_bytes);
220 if (m_mod_time) {
221 encoder.AppendU8(eSignatureModTime);
222 encoder.AppendU32(*m_mod_time);
224 if (m_obj_mod_time) {
225 encoder.AppendU8(eSignatureObjectModTime);
226 encoder.AppendU32(*m_obj_mod_time);
228 encoder.AppendU8(eSignatureEnd);
229 return true;
232 bool CacheSignature::Decode(const lldb_private::DataExtractor &data,
233 lldb::offset_t *offset_ptr) {
234 Clear();
235 while (uint8_t sig_encoding = data.GetU8(offset_ptr)) {
236 switch (sig_encoding) {
237 case eSignatureUUID: {
238 const uint8_t length = data.GetU8(offset_ptr);
239 const uint8_t *bytes = (const uint8_t *)data.GetData(offset_ptr, length);
240 if (bytes != nullptr && length > 0)
241 m_uuid = UUID(llvm::ArrayRef<uint8_t>(bytes, length));
242 } break;
243 case eSignatureModTime: {
244 uint32_t mod_time = data.GetU32(offset_ptr);
245 if (mod_time > 0)
246 m_mod_time = mod_time;
247 } break;
248 case eSignatureObjectModTime: {
249 uint32_t mod_time = data.GetU32(offset_ptr);
250 if (mod_time > 0)
251 m_obj_mod_time = mod_time;
252 } break;
253 case eSignatureEnd:
254 // The definition of is valid changed to only be valid if the UUID is
255 // valid so make sure that if we attempt to decode an old cache file
256 // that we will fail to decode the cache file if the signature isn't
257 // considered valid.
258 return IsValid();
259 default:
260 break;
263 return false;
266 uint32_t ConstStringTable::Add(ConstString s) {
267 auto [pos, inserted] = m_string_to_offset.try_emplace(s, m_next_offset);
268 if (inserted) {
269 m_strings.push_back(s);
270 m_next_offset += s.GetLength() + 1;
272 return pos->second;
275 static const llvm::StringRef kStringTableIdentifier("STAB");
277 bool ConstStringTable::Encode(DataEncoder &encoder) {
278 // Write an 4 character code into the stream. This will help us when decoding
279 // to make sure we find this identifier when decoding the string table to make
280 // sure we have the rigth data. It also helps to identify the string table
281 // when dumping the hex bytes in a cache file.
282 encoder.AppendData(kStringTableIdentifier);
283 size_t length_offset = encoder.GetByteSize();
284 encoder.AppendU32(0); // Total length of all strings which will be fixed up.
285 size_t strtab_offset = encoder.GetByteSize();
286 encoder.AppendU8(0); // Start the string table with an empty string.
287 for (auto s: m_strings) {
288 // Make sure all of the offsets match up with what we handed out!
289 assert(m_string_to_offset.find(s)->second ==
290 encoder.GetByteSize() - strtab_offset);
291 // Append the C string into the encoder
292 encoder.AppendCString(s.GetStringRef());
294 // Fixup the string table length.
295 encoder.PutU32(length_offset, encoder.GetByteSize() - strtab_offset);
296 return true;
299 bool StringTableReader::Decode(const lldb_private::DataExtractor &data,
300 lldb::offset_t *offset_ptr) {
301 llvm::StringRef identifier((const char *)data.GetData(offset_ptr, 4), 4);
302 if (identifier != kStringTableIdentifier)
303 return false;
304 const uint32_t length = data.GetU32(offset_ptr);
305 // We always have at least one byte for the empty string at offset zero.
306 if (length == 0)
307 return false;
308 const char *bytes = (const char *)data.GetData(offset_ptr, length);
309 if (bytes == nullptr)
310 return false;
311 m_data = llvm::StringRef(bytes, length);
312 return true;
315 llvm::StringRef StringTableReader::Get(uint32_t offset) const {
316 if (offset >= m_data.size())
317 return llvm::StringRef();
318 return llvm::StringRef(m_data.data() + offset);