feature, either build the JIT in debug mode to enable it by default or pass
-jit-emit-debug to lli.
Right now, the only debug information that this communicates to GDB is call
frame information, since it's already being generated to support exceptions in
the JIT. Eventually, when DWARF generation isn't tied so tightly to AsmPrinter,
it will be easy to push that information to GDB through this interface.
Here's a step-by-step breakdown of how the feature works:
- The JIT generates the machine code and DWARF call frame info
(.eh_frame/.debug_frame) for a function into memory.
- The JIT copies that info into an in-memory ELF file with a symbol for the
function.
- The JIT creates a code entry pointing to the ELF buffer and adds it to a
linked list hanging off of a global descriptor at a special symbol that GDB
knows about.
- The JIT calls a function marked noinline that GDB knows about and has put an
internal breakpoint in.
- GDB catches the breakpoint and reads the global descriptor to look for new
code.
- When sees there is new code, it reads the ELF from the inferior's memory and
adds it to itself as an object file.
- The JIT continues, and the next time we stop the program, we are able to
produce a proper backtrace.
Consider running the following program through the JIT:
#include <stdio.h>
void baz(short z) {
long w = z + 1;
printf("%d, %x\n", w, *((int*)NULL)); // SEGFAULT here
}
void bar(short y) {
int z = y + 1;
baz(z);
}
void foo(char x) {
short y = x + 1;
bar(y);
}
int main(int argc, char** argv) {
char x = 1;
foo(x);
}
Here is a backtrace before this patch:
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x2aaaabdfbd10 (LWP 25476)]
0x00002aaaabe7d1a8 in ?? ()
(gdb) bt
#0 0x00002aaaabe7d1a8 in ?? ()
#1 0x0000000000000003 in ?? ()
#2 0x0000000000000004 in ?? ()
#3 0x00032aaaabe7cfd0 in ?? ()
#4 0x00002aaaabe7d12c in ?? ()
#5 0x00022aaa00000003 in ?? ()
#6 0x00002aaaabe7d0aa in ?? ()
#7 0x01000002abe7cff0 in ?? ()
#8 0x00002aaaabe7d02c in ?? ()
#9 0x0100000000000001 in ?? ()
#10 0x00000000014388e0 in ?? ()
#11 0x00007fff00000001 in ?? ()
#12 0x0000000000b870a2 in llvm::JIT::runFunction (this=0x1405b70,
F=0x14024e0, ArgValues=@0x7fffffffe050)
at /home/rnk/llvm-gdb/lib/ExecutionEngine/JIT/JIT.cpp:395
#13 0x0000000000baa4c5 in llvm::ExecutionEngine::runFunctionAsMain
(this=0x1405b70, Fn=0x14024e0, argv=@0x13f06f8, envp=0x7fffffffe3b0)
at /home/rnk/llvm-gdb/lib/ExecutionEngine/ExecutionEngine.cpp:377
#14 0x00000000007ebd52 in main (argc=2, argv=0x7fffffffe398,
envp=0x7fffffffe3b0) at /home/rnk/llvm-gdb/tools/lli/lli.cpp:208
And a backtrace after this patch:
Program received signal SIGSEGV, Segmentation fault.
0x00002aaaabe7d1a8 in baz ()
(gdb) bt
#0 0x00002aaaabe7d1a8 in baz ()
#1 0x00002aaaabe7d12c in bar ()
#2 0x00002aaaabe7d0aa in foo ()
#3 0x00002aaaabe7d02c in main ()
#4 0x0000000000b870a2 in llvm::JIT::runFunction (this=0x1405b70,
F=0x14024e0, ArgValues=...)
at /home/rnk/llvm-gdb/lib/ExecutionEngine/JIT/JIT.cpp:395
#5 0x0000000000baa4c5 in llvm::ExecutionEngine::runFunctionAsMain
(this=0x1405b70, Fn=0x14024e0, argv=..., envp=0x7fffffffe3c0)
at /home/rnk/llvm-gdb/lib/ExecutionEngine/ExecutionEngine.cpp:377
#6 0x00000000007ebd52 in main (argc=2, argv=0x7fffffffe3a8,
envp=0x7fffffffe3c0) at /home/rnk/llvm-gdb/tools/lli/lli.cpp:208
llvm-svn: 82418
and short. Well, it's kinda short. Definitely nasty and brutish.
The front-end generates the register/unregister calls into the SjLj runtime,
call-site indices and landing pad dispatch. The back end fills in the LSDA
with the call-site information provided by the front end. Catch blocks are
not yet implemented.
Built on Darwin and verified no llvm-core "make check" regressions.
llvm-svn: 78625
- Some clients which used DOUT have moved to DEBUG. We are deprecating the
"magic" DOUT behavior which avoided calling printing functions when the
statement was disabled. In addition to being unnecessary magic, it had the
downside of leaving code in -Asserts builds, and of hiding potentially
unnecessary computations.
llvm-svn: 77019
out of memory, and also make the default memory manager allocate more memory
when it runs out.
Also, switch function stubs and global data over to using the BumpPtrAllocator.
This makes it so the JIT no longer mmaps (or the equivalent on Windows) 16 MB
of memory, and instead allocates in 512K slabs. I suspect this size could go
lower, especially on embedded platforms, now that more slabs can be allocated.
llvm-svn: 76828
call to the MachineCodeEmitter interface and made copying the start
line of a function not conditional on whether we're emitting Dwarf
debug information. I'll propagate the processDebugLoc() calls to the
non-X86 targets in a followup patch.
In the long run, it'll probably be better to gather this information
through the DwarfWriter, but the DwarfWriter currently depends on the
AsmPrinter and TargetAsmInfo, and fixing that would be out of the way
for this patch.
There's a bug in OProfile 0.9.4 that makes it ignore line numbers for
addresses above 4G, and a patch fixing it at
http://thread.gmane.org/gmane.linux.oprofile/7634
Sample output:
$ sudo opcontrol --reset; sudo opcontrol --start-daemon; sudo opcontrol --start; `pwd`/Debug/bin/lli fib.bc; sudo opcontrol --stop
Signalling daemon... done
Profiler running.
fib(40) == 165580141
Stopping profiling.
$ opreport -g -d -l `pwd`/Debug/bin/lli|head -60
Overflow stats not available
CPU: Core 2, speed 1998 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask of 0x00 (Unhalted core cycles) count 100000
vma samples % linenr info image name symbol name
00007f67a30370b0 25489 61.2554 fib.c:24 10946.jo fib_left
00007f67a30370b0 1634 6.4106 fib.c:24
00007f67a30370b1 83 0.3256 fib.c:24
00007f67a30370b9 1997 7.8348 fib.c:24
00007f67a30370c6 2080 8.1604 fib.c:27
00007f67a30370c8 988 3.8762 fib.c:27
00007f67a30370cd 1315 5.1591 fib.c:27
00007f67a30370cf 251 0.9847 fib.c:27
00007f67a30370d3 1191 4.6726 fib.c:27
00007f67a30370d6 975 3.8252 fib.c:27
00007f67a30370db 1010 3.9625 fib.c:27
00007f67a30370dd 242 0.9494 fib.c:27
00007f67a30370e1 2782 10.9145 fib.c:28
00007f67a30370e5 3768 14.7828 fib.c:28
00007f67a30370eb 615 2.4128 (no location information)
00007f67a30370f3 6558 25.7287 (no location information)
00007f67a3037100 15603 37.4973 fib.c:29 10946.jo fib_right
00007f67a3037100 1646 10.5493 fib.c:29
00007f67a3037101 45 0.2884 fib.c:29
00007f67a3037109 2372 15.2022 fib.c:29
00007f67a3037116 2234 14.3178 fib.c:32
00007f67a3037118 612 3.9223 fib.c:32
00007f67a303711d 622 3.9864 fib.c:32
00007f67a303711f 385 2.4675 fib.c:32
00007f67a3037123 404 2.5892 fib.c:32
00007f67a3037126 634 4.0633 fib.c:32
00007f67a303712b 870 5.5759 fib.c:32
00007f67a303712d 62 0.3974 fib.c:32
00007f67a3037131 1848 11.8439 fib.c:33
00007f67a3037135 2840 18.2016 fib.c:33
00007f67a303713a 1 0.0064 fib.c:33
00007f67a303713b 1023 6.5564 (no location information)
00007f67a3037143 5 0.0320 (no location information)
000000000080c1e4 15 0.0360 MachineOperand.h:150 lli llvm::MachineOperand::isReg() const
000000000080c1e4 6 40.0000 MachineOperand.h:150
000000000080c1ec 2 13.3333 MachineOperand.h:150
...
llvm-svn: 76102
default, this option is not enabled to support clients who rely on
this behavior.
Fixes http://llvm.org/PR4483
A patch to allocate additional memory for globals after we run out is
forthcoming.
Patch by Reid Kleckner!
llvm-svn: 75059
emitted or the machine code for a function is freed. Chris mentioned that we
may also want a notification when a stub is emitted, but that'll be a future
change. I intend to use this to tell oprofile where functions are emitted and
what lines correspond to what addresses.
llvm-svn: 74157
integer and floating-point opcodes, introducing
FAdd, FSub, and FMul.
For now, the AsmParser, BitcodeReader, and IRBuilder all preserve
backwards compatability, and the Core LLVM APIs preserve backwards
compatibility for IR producers. Most front-ends won't need to change
immediately.
This implements the first step of the plan outlined here:
http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt
llvm-svn: 72897
Introduce a new class (MachineCodeInfo) that the JIT can fill in with details. Right now, just the address and the size of the machine code are reported.
Patch by Evan Phoenix!
llvm-svn: 72040
another stub, but then never calling the jitted function) can cause the JIT to
leave a stub in place. Judging by the comments this is a known deficiency, so
we're just not going to use AssertingVH for the StubToFunctionTy map.
Also shorten some lines longer than 80 columns.
This fixes the "make check" failure with ocaml on x86-64 linux.
llvm-svn: 70185
1. ConstantPoolSDNode alignment field is log2 value of the alignment requirement. This is not consistent with other SDNode variants.
2. MachineConstantPool alignment field is also a log2 value.
3. However, some places are creating ConstantPoolSDNode with alignment value rather than log2 values. This creates entries with artificially large alignments, e.g. 256 for SSE vector values.
4. Constant pool entry offsets are computed when they are created. However, asm printer group them by sections. That means the offsets are no longer valid. However, asm printer uses them to determine size of padding between entries.
5. Asm printer uses expensive data structure multimap to track constant pool entries by sections.
6. Asm printer iterate over SmallPtrSet when it's emitting constant pool entries. This is non-deterministic.
Solutions:
1. ConstantPoolSDNode alignment field is changed to keep non-log2 value.
2. MachineConstantPool alignment field is also changed to keep non-log2 value.
3. Functions that create ConstantPool nodes are passing in non-log2 alignments.
4. MachineConstantPoolEntry no longer keeps an offset field. It's replaced with an alignment field. Offsets are not computed when constant pool entries are created. They are computed on the fly in asm printer and JIT.
5. Asm printer uses cheaper data structure to group constant pool entries.
6. Asm printer compute entry offsets after grouping is done.
7. Change JIT code to compute entry offsets on the fly.
llvm-svn: 66875
1. When the JIT is asked to remove a function, updating it's
mapping to 0, we invalidate any function stubs used only
by that function. Now, also invalidate the JIT's mapping
from the GV the stub pointed to, to the address of the GV.
2. When dlsym stubs for cross-process JIT are enabled, do not
abort just because a named function cannot be found in the
JIT's process.
3. Fix various assumptions about when it is ok to use the lazy
resolver when non-lazy JITing is enabled.
llvm-svn: 66324
This invalidates the stubs in the resolver map when they are no longer referenced,
and should the JIT memory manager ever pick up a deallocateStub interface, the
JIT could reclaim the memory for unused stubs as well.
llvm-svn: 66141
on failure to resolve it.
Do not abort on failure to resolve an external symbol when using dlsym stubs,
since the symbol may not be in the JIT's address space. Just use 0.
Allow dlsym stubs to differentiate between GlobalVars and Functions.
llvm-svn: 66050
that has not been JIT'd yet, the callee is put on a list of pending functions
to JIT. The call is directed through a stub, which is updated with the address
of the function after it has been JIT'd. A new interface for allocating and
updating empty stubs is provided.
Add support for removing the ModuleProvider the JIT was created with, which
would otherwise invalidate the JIT's PassManager, which is initialized with the
ModuleProvider's Module.
Add support under a new ExecutionEngine flag for emitting the infomration
necessary to update Function and GlobalVariable stubs after JITing them, by
recording the address of the stub and the name of the GlobalValue. This allows
code to be copied from one address space to another, where libraries may live
at different virtual addresses, and have the stubs updated with their new
correct target addresses.
llvm-svn: 64906
SingleSource/UnitTests/2007-04-25-weak.c in JIT mode. The test
now passes on systems which are able to produce a correct
reference output to compare with.
llvm-svn: 61674
Since the ARM constant pool handling supercedes the standard LLVM constant
pool entirely, the JIT emitter does not allocate space for the constants,
nor initialize the memory. The constant pool is considered part of the
instruction stream.
Likewise, when resolving relocations into the constant pool, a hook into
the target back end is used to resolve from the constant ID# to the
address where the constant is stored.
For now, the support in the ARM emitter is limited to 32-bit integer. Future
patches will expand this to the full range of constants necessary.
llvm-svn: 58338
model, except for external calls; this makes
addressing modes PC-relative. Incomplete.
The assertion at the top of Emitter::runOnMachineFunction
was obviously bogus (always true) so I removed it.
If someone knows what the correct test should be to cover
all the various targets, please fix.
llvm-svn: 54656
are allocated in the same buffer as the code,
jump tables, etc.
The default JIT memory manager does not handle buffer
overflow well. I didn't introduce this and I'm not
attempting to fix it here, but it is more likely to
be hit now since we're putting more stuff in the
buffer. This affects one test that I know of so far,
MultiSource/Benchmarks/NPB-serial/is.
llvm-svn: 54442
was actually passing a completely incorrect size to sys_icache_invalidate.
Instead of having the JITEmitter do this (which doesn't have the correct
size), just make the target sync its own stubs.
llvm-svn: 46354
The meaning of getTypeSize was not clear - clarifying it is important
now that we have x86 long double and arbitrary precision integers.
The issue with long double is that it requires 80 bits, and this is
not a multiple of its alignment. This gives a primitive type for
which getTypeSize differed from getABITypeSize. For arbitrary precision
integers it is even worse: there is the minimum number of bits needed to
hold the type (eg: 36 for an i36), the maximum number of bits that will
be overwriten when storing the type (40 bits for i36) and the ABI size
(i.e. the storage size rounded up to a multiple of the alignment; 64 bits
for i36).
This patch removes getTypeSize (not really - it is still there but
deprecated to allow for a gradual transition). Instead there is:
(1) getTypeSizeInBits - a number of bits that suffices to hold all
values of the type. For a primitive type, this is the minimum number
of bits. For an i36 this is 36 bits. For x86 long double it is 80.
This corresponds to gcc's TYPE_PRECISION.
(2) getTypeStoreSizeInBits - the maximum number of bits that is
written when storing the type (or read when reading it). For an
i36 this is 40 bits, for an x86 long double it is 80 bits. This
is the size alias analysis is interested in (getTypeStoreSize
returns the number of bytes). There doesn't seem to be anything
corresponding to this in gcc.
(3) getABITypeSizeInBits - this is getTypeStoreSizeInBits rounded
up to a multiple of the alignment. For an i36 this is 64, for an
x86 long double this is 96 or 128 depending on the OS. This is the
spacing between consecutive elements when you form an array out of
this type (getABITypeSize returns the number of bytes). This is
TYPE_SIZE in gcc.
Since successive elements in a SequentialType (arrays, pointers
and vectors) need to be aligned, the spacing between them will be
given by getABITypeSize. This means that the size of an array
is the length times the getABITypeSize. It also means that GEP
computations need to use getABITypeSize when computing offsets.
Furthermore, if an alloca allocates several elements at once then
these too need to be aligned, so the size of the alloca has to be
the number of elements multiplied by getABITypeSize. Logically
speaking this doesn't have to be the case when allocating just
one element, but it is simpler to also use getABITypeSize in this
case. So alloca's and mallocs should use getABITypeSize. Finally,
since gcc's only notion of size is that given by getABITypeSize, if
you want to output assembler etc the same as gcc then getABITypeSize
is the size you want.
Since a store will overwrite no more than getTypeStoreSize bytes,
and a read will read no more than that many bytes, this is the
notion of size appropriate for alias analysis calculations.
In this patch I have corrected all type size uses except some of
those in ScalarReplAggregates, lib/Codegen, lib/Target (the hard
cases). I will get around to auditing these too at some point,
but I could do with some help.
Finally, I made one change which I think wise but others might
consider pointless and suboptimal: in an unpacked struct the
amount of space allocated for a field is now given by the ABI
size rather than getTypeStoreSize. I did this because every
other place that reserves memory for a type (eg: alloca) now
uses getABITypeSize, and I didn't want to make an exception
for unpacked structs, i.e. I did it to make things more uniform.
This only effects structs containing long doubles and arbitrary
precision integers. If someone wants to pack these types more
tightly they can always use a packed struct.
llvm-svn: 43620
Devang Patel