xref: /llvm-project/lldb/source/Target/Memory.cpp (revision 0642cd768b80665585c8500bed2933a3b99123dc)
1 //===-- Memory.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 //===----------------------------------------------------------------------===//
8 
9 #include "lldb/Target/Memory.h"
10 #include "lldb/Target/Process.h"
11 #include "lldb/Utility/DataBufferHeap.h"
12 #include "lldb/Utility/LLDBLog.h"
13 #include "lldb/Utility/Log.h"
14 #include "lldb/Utility/RangeMap.h"
15 #include "lldb/Utility/State.h"
16 
17 #include <cinttypes>
18 #include <memory>
19 
20 using namespace lldb;
21 using namespace lldb_private;
22 
23 // MemoryCache constructor
24 MemoryCache::MemoryCache(Process &process)
25     : m_mutex(), m_L1_cache(), m_L2_cache(), m_invalid_ranges(),
26       m_process(process),
27       m_L2_cache_line_byte_size(process.GetMemoryCacheLineSize()) {}
28 
29 // Destructor
30 MemoryCache::~MemoryCache() = default;
31 
32 void MemoryCache::Clear(bool clear_invalid_ranges) {
33   std::lock_guard<std::recursive_mutex> guard(m_mutex);
34   m_L1_cache.clear();
35   m_L2_cache.clear();
36   if (clear_invalid_ranges)
37     m_invalid_ranges.Clear();
38   m_L2_cache_line_byte_size = m_process.GetMemoryCacheLineSize();
39 }
40 
41 void MemoryCache::AddL1CacheData(lldb::addr_t addr, const void *src,
42                                  size_t src_len) {
43   AddL1CacheData(
44       addr, DataBufferSP(new DataBufferHeap(DataBufferHeap(src, src_len))));
45 }
46 
47 void MemoryCache::AddL1CacheData(lldb::addr_t addr,
48                                  const DataBufferSP &data_buffer_sp) {
49   std::lock_guard<std::recursive_mutex> guard(m_mutex);
50   m_L1_cache[addr] = data_buffer_sp;
51 }
52 
53 void MemoryCache::Flush(addr_t addr, size_t size) {
54   if (size == 0)
55     return;
56 
57   std::lock_guard<std::recursive_mutex> guard(m_mutex);
58 
59   // Erase any blocks from the L1 cache that intersect with the flush range
60   if (!m_L1_cache.empty()) {
61     AddrRange flush_range(addr, size);
62     BlockMap::iterator pos = m_L1_cache.upper_bound(addr);
63     if (pos != m_L1_cache.begin()) {
64       --pos;
65     }
66     while (pos != m_L1_cache.end()) {
67       AddrRange chunk_range(pos->first, pos->second->GetByteSize());
68       if (!chunk_range.DoesIntersect(flush_range))
69         break;
70       pos = m_L1_cache.erase(pos);
71     }
72   }
73 
74   if (!m_L2_cache.empty()) {
75     const uint32_t cache_line_byte_size = m_L2_cache_line_byte_size;
76     const addr_t end_addr = (addr + size - 1);
77     const addr_t first_cache_line_addr = addr - (addr % cache_line_byte_size);
78     const addr_t last_cache_line_addr =
79         end_addr - (end_addr % cache_line_byte_size);
80     // Watch for overflow where size will cause us to go off the end of the
81     // 64 bit address space
82     uint32_t num_cache_lines;
83     if (last_cache_line_addr >= first_cache_line_addr)
84       num_cache_lines = ((last_cache_line_addr - first_cache_line_addr) /
85                          cache_line_byte_size) +
86                         1;
87     else
88       num_cache_lines =
89           (UINT64_MAX - first_cache_line_addr + 1) / cache_line_byte_size;
90 
91     uint32_t cache_idx = 0;
92     for (addr_t curr_addr = first_cache_line_addr; cache_idx < num_cache_lines;
93          curr_addr += cache_line_byte_size, ++cache_idx) {
94       BlockMap::iterator pos = m_L2_cache.find(curr_addr);
95       if (pos != m_L2_cache.end())
96         m_L2_cache.erase(pos);
97     }
98   }
99 }
100 
101 void MemoryCache::AddInvalidRange(lldb::addr_t base_addr,
102                                   lldb::addr_t byte_size) {
103   if (byte_size > 0) {
104     std::lock_guard<std::recursive_mutex> guard(m_mutex);
105     InvalidRanges::Entry range(base_addr, byte_size);
106     m_invalid_ranges.Append(range);
107     m_invalid_ranges.Sort();
108   }
109 }
110 
111 bool MemoryCache::RemoveInvalidRange(lldb::addr_t base_addr,
112                                      lldb::addr_t byte_size) {
113   if (byte_size > 0) {
114     std::lock_guard<std::recursive_mutex> guard(m_mutex);
115     const uint32_t idx = m_invalid_ranges.FindEntryIndexThatContains(base_addr);
116     if (idx != UINT32_MAX) {
117       const InvalidRanges::Entry *entry = m_invalid_ranges.GetEntryAtIndex(idx);
118       if (entry->GetRangeBase() == base_addr &&
119           entry->GetByteSize() == byte_size)
120         return m_invalid_ranges.RemoveEntryAtIndex(idx);
121     }
122   }
123   return false;
124 }
125 
126 lldb::DataBufferSP MemoryCache::GetL2CacheLine(lldb::addr_t line_base_addr,
127                                                Status &error) {
128   // This function assumes that the address given is aligned correctly.
129   assert((line_base_addr % m_L2_cache_line_byte_size) == 0);
130 
131   std::lock_guard<std::recursive_mutex> guard(m_mutex);
132   auto pos = m_L2_cache.find(line_base_addr);
133   if (pos != m_L2_cache.end())
134     return pos->second;
135 
136   auto data_buffer_heap_sp =
137       std::make_shared<DataBufferHeap>(m_L2_cache_line_byte_size, 0);
138   size_t process_bytes_read = m_process.ReadMemoryFromInferior(
139       line_base_addr, data_buffer_heap_sp->GetBytes(),
140       data_buffer_heap_sp->GetByteSize(), error);
141 
142   // If we failed a read, not much we can do.
143   if (process_bytes_read == 0)
144     return lldb::DataBufferSP();
145 
146   // If we didn't get a complete read, we can still cache what we did get.
147   if (process_bytes_read < m_L2_cache_line_byte_size)
148     data_buffer_heap_sp->SetByteSize(process_bytes_read);
149 
150   m_L2_cache[line_base_addr] = data_buffer_heap_sp;
151   return data_buffer_heap_sp;
152 }
153 
154 size_t MemoryCache::Read(addr_t addr, void *dst, size_t dst_len,
155                          Status &error) {
156   if (!dst || dst_len == 0)
157     return 0;
158 
159   std::lock_guard<std::recursive_mutex> guard(m_mutex);
160   // FIXME: We should do a more thorough check to make sure that we're not
161   // overlapping with any invalid ranges (e.g. Read 0x100 - 0x200 but there's an
162   // invalid range 0x180 - 0x280). `FindEntryThatContains` has an implementation
163   // that takes a range, but it only checks to see if the argument is contained
164   // by an existing invalid range. It cannot check if the argument contains
165   // invalid ranges and cannot check for overlaps.
166   if (m_invalid_ranges.FindEntryThatContains(addr)) {
167     error = Status::FromErrorStringWithFormat(
168         "memory read failed for 0x%" PRIx64, addr);
169     return 0;
170   }
171 
172   // Check the L1 cache for a range that contains the entire memory read.
173   // L1 cache contains chunks of memory that are not required to be the size of
174   // an L2 cache line. We avoid trying to do partial reads from the L1 cache to
175   // simplify the implementation.
176   if (!m_L1_cache.empty()) {
177     AddrRange read_range(addr, dst_len);
178     BlockMap::iterator pos = m_L1_cache.upper_bound(addr);
179     if (pos != m_L1_cache.begin()) {
180       --pos;
181     }
182     AddrRange chunk_range(pos->first, pos->second->GetByteSize());
183     if (chunk_range.Contains(read_range)) {
184       memcpy(dst, pos->second->GetBytes() + (addr - chunk_range.GetRangeBase()),
185              dst_len);
186       return dst_len;
187     }
188   }
189 
190   // If the size of the read is greater than the size of an L2 cache line, we'll
191   // just read from the inferior. If that read is successful, we'll cache what
192   // we read in the L1 cache for future use.
193   if (dst_len > m_L2_cache_line_byte_size) {
194     size_t bytes_read =
195         m_process.ReadMemoryFromInferior(addr, dst, dst_len, error);
196     if (bytes_read > 0)
197       AddL1CacheData(addr, dst, bytes_read);
198     return bytes_read;
199   }
200 
201   // If the size of the read fits inside one L2 cache line, we'll try reading
202   // from the L2 cache. Note that if the range of memory we're reading sits
203   // between two contiguous cache lines, we'll touch two cache lines instead of
204   // just one.
205 
206   // We're going to have all of our loads and reads be cache line aligned.
207   addr_t cache_line_offset = addr % m_L2_cache_line_byte_size;
208   addr_t cache_line_base_addr = addr - cache_line_offset;
209   DataBufferSP first_cache_line = GetL2CacheLine(cache_line_base_addr, error);
210   // If we get nothing, then the read to the inferior likely failed. Nothing to
211   // do here.
212   if (!first_cache_line)
213     return 0;
214 
215   // If the cache line was not filled out completely and the offset is greater
216   // than what we have available, we can't do anything further here.
217   if (cache_line_offset >= first_cache_line->GetByteSize())
218     return 0;
219 
220   uint8_t *dst_buf = (uint8_t *)dst;
221   size_t bytes_left = dst_len;
222   size_t read_size = first_cache_line->GetByteSize() - cache_line_offset;
223   if (read_size > bytes_left)
224     read_size = bytes_left;
225 
226   memcpy(dst_buf + dst_len - bytes_left,
227          first_cache_line->GetBytes() + cache_line_offset, read_size);
228   bytes_left -= read_size;
229 
230   // If the cache line was not filled out completely and we still have data to
231   // read, we can't do anything further.
232   if (first_cache_line->GetByteSize() < m_L2_cache_line_byte_size &&
233       bytes_left > 0)
234     return dst_len - bytes_left;
235 
236   // We'll hit this scenario if our read straddles two cache lines.
237   if (bytes_left > 0) {
238     cache_line_base_addr += m_L2_cache_line_byte_size;
239 
240     // FIXME: Until we are able to more thoroughly check for invalid ranges, we
241     // will have to check the second line to see if it is in an invalid range as
242     // well. See the check near the beginning of the function for more details.
243     if (m_invalid_ranges.FindEntryThatContains(cache_line_base_addr)) {
244       error = Status::FromErrorStringWithFormat(
245           "memory read failed for 0x%" PRIx64, cache_line_base_addr);
246       return dst_len - bytes_left;
247     }
248 
249     DataBufferSP second_cache_line =
250         GetL2CacheLine(cache_line_base_addr, error);
251     if (!second_cache_line)
252       return dst_len - bytes_left;
253 
254     read_size = bytes_left;
255     if (read_size > second_cache_line->GetByteSize())
256       read_size = second_cache_line->GetByteSize();
257 
258     memcpy(dst_buf + dst_len - bytes_left, second_cache_line->GetBytes(),
259            read_size);
260     bytes_left -= read_size;
261 
262     return dst_len - bytes_left;
263   }
264 
265   return dst_len;
266 }
267 
268 AllocatedBlock::AllocatedBlock(lldb::addr_t addr, uint32_t byte_size,
269                                uint32_t permissions, uint32_t chunk_size)
270     : m_range(addr, byte_size), m_permissions(permissions),
271       m_chunk_size(chunk_size)
272 {
273   // The entire address range is free to start with.
274   m_free_blocks.Append(m_range);
275   assert(byte_size > chunk_size);
276 }
277 
278 AllocatedBlock::~AllocatedBlock() = default;
279 
280 lldb::addr_t AllocatedBlock::ReserveBlock(uint32_t size) {
281   // We must return something valid for zero bytes.
282   if (size == 0)
283     size = 1;
284   Log *log = GetLog(LLDBLog::Process);
285 
286   const size_t free_count = m_free_blocks.GetSize();
287   for (size_t i=0; i<free_count; ++i)
288   {
289     auto &free_block = m_free_blocks.GetEntryRef(i);
290     const lldb::addr_t range_size = free_block.GetByteSize();
291     if (range_size >= size)
292     {
293       // We found a free block that is big enough for our data. Figure out how
294       // many chunks we will need and calculate the resulting block size we
295       // will reserve.
296       addr_t addr = free_block.GetRangeBase();
297       size_t num_chunks = CalculateChunksNeededForSize(size);
298       lldb::addr_t block_size = num_chunks * m_chunk_size;
299       lldb::addr_t bytes_left = range_size - block_size;
300       if (bytes_left == 0)
301       {
302         // The newly allocated block will take all of the bytes in this
303         // available block, so we can just add it to the allocated ranges and
304         // remove the range from the free ranges.
305         m_reserved_blocks.Insert(free_block, false);
306         m_free_blocks.RemoveEntryAtIndex(i);
307       }
308       else
309       {
310         // Make the new allocated range and add it to the allocated ranges.
311         Range<lldb::addr_t, uint32_t> reserved_block(free_block);
312         reserved_block.SetByteSize(block_size);
313         // Insert the reserved range and don't combine it with other blocks in
314         // the reserved blocks list.
315         m_reserved_blocks.Insert(reserved_block, false);
316         // Adjust the free range in place since we won't change the sorted
317         // ordering of the m_free_blocks list.
318         free_block.SetRangeBase(reserved_block.GetRangeEnd());
319         free_block.SetByteSize(bytes_left);
320       }
321       LLDB_LOGV(log, "({0}) (size = {1} ({1:x})) => {2:x}", this, size, addr);
322       return addr;
323     }
324   }
325 
326   LLDB_LOGV(log, "({0}) (size = {1} ({1:x})) => {2:x}", this, size,
327             LLDB_INVALID_ADDRESS);
328   return LLDB_INVALID_ADDRESS;
329 }
330 
331 bool AllocatedBlock::FreeBlock(addr_t addr) {
332   bool success = false;
333   auto entry_idx = m_reserved_blocks.FindEntryIndexThatContains(addr);
334   if (entry_idx != UINT32_MAX)
335   {
336     m_free_blocks.Insert(m_reserved_blocks.GetEntryRef(entry_idx), true);
337     m_reserved_blocks.RemoveEntryAtIndex(entry_idx);
338     success = true;
339   }
340   Log *log = GetLog(LLDBLog::Process);
341   LLDB_LOGV(log, "({0}) (addr = {1:x}) => {2}", this, addr, success);
342   return success;
343 }
344 
345 AllocatedMemoryCache::AllocatedMemoryCache(Process &process)
346     : m_process(process), m_mutex(), m_memory_map() {}
347 
348 AllocatedMemoryCache::~AllocatedMemoryCache() = default;
349 
350 void AllocatedMemoryCache::Clear(bool deallocate_memory) {
351   std::lock_guard<std::recursive_mutex> guard(m_mutex);
352   if (m_process.IsAlive() && deallocate_memory) {
353     PermissionsToBlockMap::iterator pos, end = m_memory_map.end();
354     for (pos = m_memory_map.begin(); pos != end; ++pos)
355       m_process.DoDeallocateMemory(pos->second->GetBaseAddress());
356   }
357   m_memory_map.clear();
358 }
359 
360 AllocatedMemoryCache::AllocatedBlockSP
361 AllocatedMemoryCache::AllocatePage(uint32_t byte_size, uint32_t permissions,
362                                    uint32_t chunk_size, Status &error) {
363   AllocatedBlockSP block_sp;
364   const size_t page_size = 4096;
365   const size_t num_pages = (byte_size + page_size - 1) / page_size;
366   const size_t page_byte_size = num_pages * page_size;
367 
368   addr_t addr = m_process.DoAllocateMemory(page_byte_size, permissions, error);
369 
370   Log *log = GetLog(LLDBLog::Process);
371   if (log) {
372     LLDB_LOGF(log,
373               "Process::DoAllocateMemory (byte_size = 0x%8.8" PRIx32
374               ", permissions = %s) => 0x%16.16" PRIx64,
375               (uint32_t)page_byte_size, GetPermissionsAsCString(permissions),
376               (uint64_t)addr);
377   }
378 
379   if (addr != LLDB_INVALID_ADDRESS) {
380     block_sp = std::make_shared<AllocatedBlock>(addr, page_byte_size,
381                                                 permissions, chunk_size);
382     m_memory_map.insert(std::make_pair(permissions, block_sp));
383   }
384   return block_sp;
385 }
386 
387 lldb::addr_t AllocatedMemoryCache::AllocateMemory(size_t byte_size,
388                                                   uint32_t permissions,
389                                                   Status &error) {
390   std::lock_guard<std::recursive_mutex> guard(m_mutex);
391 
392   addr_t addr = LLDB_INVALID_ADDRESS;
393   std::pair<PermissionsToBlockMap::iterator, PermissionsToBlockMap::iterator>
394       range = m_memory_map.equal_range(permissions);
395 
396   for (PermissionsToBlockMap::iterator pos = range.first; pos != range.second;
397        ++pos) {
398     addr = (*pos).second->ReserveBlock(byte_size);
399     if (addr != LLDB_INVALID_ADDRESS)
400       break;
401   }
402 
403   if (addr == LLDB_INVALID_ADDRESS) {
404     AllocatedBlockSP block_sp(AllocatePage(byte_size, permissions, 16, error));
405 
406     if (block_sp)
407       addr = block_sp->ReserveBlock(byte_size);
408   }
409   Log *log = GetLog(LLDBLog::Process);
410   LLDB_LOGF(log,
411             "AllocatedMemoryCache::AllocateMemory (byte_size = 0x%8.8" PRIx32
412             ", permissions = %s) => 0x%16.16" PRIx64,
413             (uint32_t)byte_size, GetPermissionsAsCString(permissions),
414             (uint64_t)addr);
415   return addr;
416 }
417 
418 bool AllocatedMemoryCache::DeallocateMemory(lldb::addr_t addr) {
419   std::lock_guard<std::recursive_mutex> guard(m_mutex);
420 
421   PermissionsToBlockMap::iterator pos, end = m_memory_map.end();
422   bool success = false;
423   for (pos = m_memory_map.begin(); pos != end; ++pos) {
424     if (pos->second->Contains(addr)) {
425       success = pos->second->FreeBlock(addr);
426       break;
427     }
428   }
429   Log *log = GetLog(LLDBLog::Process);
430   LLDB_LOGF(log,
431             "AllocatedMemoryCache::DeallocateMemory (addr = 0x%16.16" PRIx64
432             ") => %i",
433             (uint64_t)addr, success);
434   return success;
435 }
436