This prevents Malloc from allocating the same chunk of memory twice, as
a byproduct of an alignment adjustment which gave the client access to
unallocated memory.
Prior to this patch, the newly-added test failed with:
$ lldb-test ir-memory-map ... ir-memory-map-overlap1.test
...
Command: malloc(size=64, alignment=32)
Malloc: address = 0x1000cd080
Command: malloc(size=64, alignment=8)
Malloc: address = 0x1000cd0b0
Malloc error: overlapping allocation detected, previous allocation at [0x1000cd080, 0x1000cd0c0)
Differential Revision: https://reviews.llvm.org/D47551
llvm-svn: 333697
Fixes infinite recursion due to change in how FileSpec deals with
removing the last path component.
Fixes timout for TestMiniDumpNew.py
llvm-svn: 333666
Summary:
As discussed in https://bugs.llvm.org/show_bug.cgi?id=37317,
FindGlobalVariables does not properly handle the case where
append=false. As this doesn't seem to be used in the tree, this patch
removes the parameter entirely.
Reviewers: clayborg, jingham, labath
Reviewed By: clayborg
Subscribers: aprantl, lldb-commits, kubamracek, JDevlieghere
Differential Revision: https://reviews.llvm.org/D46885
Patch by Tom Tromey <ttromey@mozilla.com>.
llvm-svn: 333639
rL145086 introduced m_die_array.shrink_to_fit() implemented by
exact_size_die_array.swap, it was before LLVM became written in C++11.
Differential revision: https://reviews.llvm.org/D47492
llvm-svn: 333636
This teaches lldb-test how to launch a process, set up an IRMemoryMap,
and issue memory allocations in the target process through the map. This
makes it possible to test IRMemoryMap in a targeted way.
This has uncovered two bugs so far. The first bug is that Malloc
performs an adjustment on the pointer returned from AllocateMemory (for
alignment purposes) which ultimately allows overlapping memory regions
to be created. The second bug is that after most of the address space on
the host side is exhausted, Malloc may return the same address multiple
times. These bugs (and hopefully more!) can be uncovered and tested for
with targeted lldb-test commands.
At an even higher level, the motivation for addressing these bugs is
that they can lead to strange user-visible failures (e.g, variables
assume the wrong value during expression evaluation, or the debugger
crashes). See my third comment on this swift-lldb PR for an example:
https://github.com/apple/swift-lldb/pull/652
I hope lldb-test is the right place to add this testing harness. Setting
up a gtest-style unit test proved too cumbersome (you need to recreate
or mock way too much debugger state), as did writing end-to-end tests
(it's hard to write a test that actually hits a buggy path).
With lldb-test, it's easy to read/generate the test input and parse the
test output. I'll attach a simple "fuzz" tester which generates failing
test cases to the Phab review. Here's an example:
```
Command: malloc(size=1024, alignment=32)
Malloc: address = 0xca000
Command: malloc(size=64, alignment=16)
Malloc: address = 0xca400
Command: malloc(size=1024, alignment=16)
Malloc: address = 0xca440
Command: malloc(size=16, alignment=8)
Malloc: address = 0xca840
Command: malloc(size=2048, alignment=16)
Malloc: address = 0xcb000
Command: malloc(size=64, alignment=32)
Malloc: address = 0xca860
Command: malloc(size=1024, alignment=16)
Malloc: address = 0xca890
Malloc error: overlapping allocation detected, previous allocation at [0xca860, 0xca8a0)
```
{F6288839}
Differential Revision: https://reviews.llvm.org/D47508
llvm-svn: 333583
When reading DBGSourcePathRemapping from a dSYM, we remove the last two
path components to make the source lookup more general. However, when
dealing with a relative path that has less than 2 components, we ended
up with an invalid (empty) FileSpec.
This patch changes the behavior of removeLastPathComponent to remove the
last path component, if possible. It does this by checking whether a
parent path exists, and if so using that as the new path. We rely
entirely on LLVM's path implementation to do the heavy lifting.
We now also return a boolean which indicates whether the operator was
successful or not.
Differential revision: https://reviews.llvm.org/D47495
rdar://37791687
llvm-svn: 333540
Summary:
Generating LLDB.framework when building with CMake+Ninja will copy the
lldb-private headers because public_headers contains them, even though we try
to make sure they don't get copied by removing root_private_headers from
root_public_headers.
This patch also removes SystemInitializerFull.h from the LLDB.framework headers when building with CMake.
Reviewers: compnerd, sas, labath, beanz, zturner
Reviewed By: labath
Subscribers: clayborg, mgorny, lldb-commits
Differential Revision: https://reviews.llvm.org/D47278
llvm-svn: 333444
GetUnitDIEPtrOnly() needs to return pointer to the first DIE.
But the first element of m_die_array after ExtractDIEsIfNeeded(true)
may move in memory after later ExtractDIEsIfNeeded(false).
DWARFDebugInfoEntry::collection m_die_array is std::vector,
its data may move during its expansion.
Differential revision: https://reviews.llvm.org/D46810
llvm-svn: 333437
Summary:
This improves the process of cross-compiling from macOS to Linux
since these files aren't used / needed at all.
Reviewers: clayborg, labath
Subscribers: lldb-commits, krytarowski
Differential Revision: https://reviews.llvm.org/D47420
llvm-svn: 333400
Summary:
We were treating subprograms and inlined subroutines differently when
building the index. The difference was in which indexes were individual
tags inserted (subprograms went to all indexes, where as inlined
subroutines only into the basename and full name indexes).
This seems like an error, because an inlined subroutine can still
represent an C++ or an ObjC method. I don't see anything in the
subprogram branch which should not apply to an inlined subroutine, so I
propose to just treat them identically. This makes searching for an
inlined method behave the same way as for the apple index.
I write an assembly-based test because I did not want to depend on
particular clang inlining behavior (and because I wanted to see how hard
would it be).
Reviewers: clayborg, JDevlieghere
Subscribers: eraman, lldb-commits
Differential Revision: https://reviews.llvm.org/D47368
llvm-svn: 333398
Summary:
FastDemangle gives us a C-string that we own (which is allocated in SymbolDemangler::GetDemangledCopy).
As we are not deleting the string, we leak memory whenever we call SubsPrimitiveParmItanium.
Reviewers: javed.absar
Subscribers: kristof.beyls, chrib, lldb-commits
Differential Revision: https://reviews.llvm.org/D47418
llvm-svn: 333353
Summary: When compiling with modules, these missing includes cause the build to fail (as the header can't be compiled into a module).
Subscribers: ki.stfu, lldb-commits
Differential Revision: https://reviews.llvm.org/D47412
llvm-svn: 333345
Summary: This resolves unnecessary the header dependency from
Core to DataFormatters. Patch is necessary for the introduction of
C++ modules to the LLDB build system.
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D47409
llvm-svn: 333342
Summary:
It seems to me that files in include/lldb/API/ are headers that should
be exposed to liblldb users. Because SystemInitializerFull.h exposes details of
lldb_private, I think having it there is not the right thing to do. Since it's
only included from files in source/API, we should move it there and treat it as
private.
Reviewers: labath, clayborg
Reviewed By: labath, clayborg
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D47342
llvm-svn: 333304
Several functions were passing a list of 8 NameToDIE arguments around.
This puts those variables in a struct and passes that instead, reducing
code duplication and the possibility of error (swapping two arguments
accidentally).
llvm-svn: 333264
As suggested by Pavel Labath in D46810 DWARFUnit::GetUnitDIEOnly() returning
a pointer to m_first_die should not permit using methods like GetFirstChild().
Differential revision: https://reviews.llvm.org/D47276
llvm-svn: 333224
This new DWARFBaseDIE is going to be used for DWARFUnit::GetUnitDIEOnly() as
other DIEs are unavailable that time so the caller should not have methods
available to access them.
This patch is only a mechanical split without any use of it.
Differential revision: https://reviews.llvm.org/D47275
llvm-svn: 333222
Summary:
For lldb-server, it is sufficient to parse only the native object file
format for its target OS (no other file can be loaded into a running
process). This moves the object file initialization code into specific
initializer classes: lldb-test and liblldb get all object files;
lldb-server gets only one of them. For this to work, I've needed to
create a special SystemInitializer for use in lldb-server, instead of it
calling directly into the common one.
This reduces the size of lldb-server by about 2%, which is not
earth-shattering, but it's an easy win, and it helps.
Reviewers: zturner, clayborg
Subscribers: mgorny, lldb-commits
Differential Revision: https://reviews.llvm.org/D47250
llvm-svn: 333182
Summary:
I think this makes sense for several reasons:
- better separation of concerns: DWARFUnit's job should be to provide a
nice interface to its users to access the unit contents.
ManualDWARFIndex can then use this interface to build an index and
provide it to its users.
- closer alignment with llvm parsers: there is no indexing equivalent in
llvm, and there probably never will be, as the index is very centered
around how lldb wants to access debug info. If we ever switch to
llvm's parser, this will allow us swap out DWARFUnit implementations
and keep indexing as-is.
- closer proximity of the indexing code to AppleDWARFIndex will make it
easier to keep the two in sync (e.g. right now the two use very
different algorithms to determine whether a DW_TAG_subroutine
represents a "method"). This is my primary motivation for making this
change now, but I am leaving this work to a separate patch.
The only interface change to DWARFUnit I needed to make was to add an
efficient way to iterate over the list of all DIEs. Adding this also
aligns us closer to the llvm parser.
Reviewers: JDevlieghere, clayborg, aprantl
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D47253
llvm-svn: 333178
The only reason this was here was so that Module could have a
function called CreateJITModule which created things in a special
order. Instead of making this specific to creating JIT modules,
I converted this into a template function that can create a module
for any type of object file plugin and just forwards arguments
through. Since the template is not instantiated in Core, the linker
(and header file) dependency moves to the point where it is
instantiated, which only happens in Expression. Conceptually, this
location also makes more sense for a dependency on ObjectFileJIT.
After all, we JIT expressions so it's no surprise that Expression
needs to make use of ObjectFileJIT.
Differential Revision: https://reviews.llvm.org/D47228
llvm-svn: 333143
This call was originally being only made when launching for debug (as an
attempt to make sure we don't impart extra privileges on the launched
process), but after the debug and non-debug paths were merged, it made
it's way into generic code. This was causing problems in locked down
android environments which disallowed calling setgid even if it would be
a no-op. This prevented launching llgs from lldb-server platform.
Overall I'm not sure we should be calling setgid in the first place
(it seems random -- e.g. why don't we call setuid then as well).
However, all our other copies of launch code have it, so I choose to
keep it for now.
llvm-svn: 333073
Summary:
Implement FindGlobalVariables and ParseVariableContext methods.
Compile unit information is necessary for resolving variable context, however some PDB symbols do not have this information. For now an empty DWARFExpression is used to construct a lldb::Variable instance with the limitation that using lldb to lookup the value of a global or local variable is not available.
This commit may slow down lit/SymbolFile/PDB/compilands.test since the test includes MS specific modules that spend more time parsing variables.
Reviewers: rnk, zturner, lldb-commits
Subscribers: aprantl, JDevlieghere, llvm-commits
Differential Revision: https://reviews.llvm.org/D45224
llvm-svn: 333049
Most non-local includes of header files living under lldb/sources/
were specified with the full path starting after sources/. However, in
a few instances, other sub-directories were added to include paths, or
Normalize those few instances to follow the style used by the rest of
the codebase, to make it easier to understand.
llvm-svn: 333035
Summary:
The plugin already builds fine on other platforms (linux, at least). All
that was necessary was to revitalize the hack in PlatformDarwinKernel
(not a very pretty hack, but it gets us going at least).
I haven't done a thorough investigation of the state of the plugin on
other platforms, but at least the two core file tests we have seem to
pass, so I enable them.
Reviewers: JDevlieghere, jasonmolenda
Subscribers: lldb-commits, mgorny
Differential Revision: https://reviews.llvm.org/D47133
llvm-svn: 332997
We've seen some cases on macOS where you go to instruction single
step (over a breakpoint), and single step returns but the instruction
hasn't been executed (and the pc hasn't moved.) The ThreadPlanStepOverBreakpoint
used to handle this case by accident, but the patches to handle two adjacent
breakpoints broke that accident.
This patch fixes the logic of ExplainsStop to explicitly handle the case where
the pc didn't move. It also adds a WillPop that re-enables the breakpoint we
were stepping over. We never want an unexpected path through the plan to
fool us into not doing that.
I have no idea how to make this bug happen. It is very inconsistent when it
occurs IRL. We really need a full MockProcess Plugin before we can start to write
tests for this sort of system hiccup.
<rdar://problem/38505726>
llvm-svn: 332922
PathMappingList was broken for relative and empty paths after normalization changes in FileSpec. There were also no tests for PathMappingList so I added those.
Changes include:
Change PathMappingList::ReverseRemapPath() to take FileSpec objects instead of ConstString. The only client of this was doing work to convert to and from ConstString objects for no reason.
Normalize all paths prefix and replacements that are added to the PathMappingList vector so they match the paths that have been already normalized in the debug info
Unify code in the two forms of PathMappingList::RemapPath() so only one contains the actual functionality. Prior to this, there were two versions of this code.
Use FileSpec::AppendPathComponent() and remove a long standing TODO so paths are correctly appended to each other.
Added tests for absolute, relative and empty paths.
Differential Revision: https://reviews.llvm.org/D47021
llvm-svn: 332842
Summary:
- Fix #include path
- Fix warning:
````
error: format specifies type 'unsigned long long' but the argument has type 'uint64_t'
(aka 'unsigned long') [-Werror,-Wformat]
```
Reviewers: labath, javed.absar
Differential Revision: https://reviews.llvm.org/D47072
llvm-svn: 332733
Summary:
This places the `if(m_using_apple_tables)` branches inside the
SymbolFileDWARF class behind an abstract DWARFIndex class. The class
currently has two implementations:
- AppleIndex, which searches using .apple_names and friends
- ManualIndex, which searches using a manually built index
Most of the methods of the class are very simple, and simply extract the
list of DIEs for the given name from the appropriate sub-table. The main
exception are the two GetFunctions overloads, which take a couple of
extra paramenters, including some callbacks. It was not possible to
split these up the same way as other methods, as here we were doing a
lot of post-processing on the results. The post-processing is similar
for the two cases, but not identical. I hope to factor these further in
separate patches.
Other interesting methods are:
- Preload(): do any preprocessing to make lookups faster (noop for
AppleIndex, forces a build of the lookup tables for ManualIndex).
- ReportInvalidDIEOffset(): Used to notify the users of an invalid index
(prints a message for AppleIndex, noop for ManualIndex).
- Dump(): dumps the index state (noop for AppleIndex, prints the lookup
tables for ManualIndex).
Reviewers: clayborg, JDevlieghere
Subscribers: mgorny, aprantl, lldb-commits
Differential Revision: https://reviews.llvm.org/D46889
llvm-svn: 332719
It turns out these class still contained some os-specific functionality,
but I did not notice that originally, as it was #ifdef arm(64). This
adds back the __APPLE__ condition to these particular functions,
unbreaking arm builds on other OSs.
llvm-svn: 332710
Summary:
Before this patch we were unable to write cross-platform MachO tests
because the parsing code did not compile on other platforms. The reason
for that was that ObjectFileMachO depended on
RegisterContextDarwin_arm(64)? (presumably for core file parsing) and
the two Register Context classes uses constants from the system headers
(KERN_SUCCESS, KERN_INVALID_ARGUMENT).
As far as I can tell, these two files don't actually interact with the
darwin kernel -- they are used only in ObjectFileMachO and MacOSX-Kernel
process plugin (even though it has "kernel" in the name, this one
communicates with it via network packets and not syscalls). For the time
being I have created OS-independent definitions of these constants and
made the register context classes use those. Long term, the error
handling in these classes should be probably changed to use more
standard mechanisms such as Status or Error classes.
This is the only change necessary (apart from build system glue) to make
ObjectFileMachO work on other platforms. To demonstrate that, I remove
REQUIRES:darwin from our (only) cross-platform mach-o test.
Reviewers: jasonmolenda, aprantl, clayborg, javed.absar
Subscribers: mgorny, lldb-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D46934
llvm-svn: 332702
