[llvm-commits] [PATCH] Teach llvm-objdump to dump Win64 exception tables
Kai
kai at redstar.de
Tue Nov 27 12:32:31 PST 2012
Hi Sean!
I reworked the function names and the other stuff you mentioned.
But I did not invent an "UnwindInfoIterator". I expect that future
patches extend the capabilities of the tool in other places.
In fact I am already thinking about how to output the language specific
EH data. This would be a future patch in which I also correct the
repeated reinterpret_cast<> expressions.
You are right that the changes to WinEH64.h have different root causes.
In fact they were done by two different people! I noted this in the
commit messages.
Thanks for all your very helpful comments!!!
Regards
Kai
On 25.11.2012 22:45, Sean Silva wrote:
> Just a few tiny nits and the first patch looks ready to be committed
> (although I'd like Michael to check off on it).
>
>> The type casts are necessary, otherwise the output is garbage. I think the
>> ulittle8_t value is interpreted as a char, not a number.
>
> You might want to note that in a comment.
>
> + case UOP_SaveNonVolBig:
> + case UOP_SaveXMM128Big:
> + UsedSlots = 3;
> + break;
> + case UOP_AllocLarge:
> + UsedSlots = (UnwindCode.getOpInfo() == 0) ? 2 : 3;
> + break;
> + }
> + return UsedSlots;
> +}
>
> You can get rid of UsedSlots here and just return the size directly.
>
> +static void printUnwindCode(ArrayRef<UnwindCode> UCs) {
>
> +static void printCOFFUnwindCode(ArrayRef<UnwindCode> UCs) {
>
> I think these names need to be adjusted. The calling sequence
> printCOFFUnwindCode -> printUnwindCode doesn't make sense, since it
> seems like a more specific function is calling a more generic
> function. I'm not super familiar with the terminology here, but
> shouldn't these be plural, since they are printing multiple unwind
> codes `ArrayRef<UnwindCode> UCs`? The naming should also reflect that
> "printCOFFUnwindCode" prints multiple of whatever "printUnwindCode"
> prints. From the comment, it seems like "printUnwindCode" prints
> "unwind code entry"'s, so how about renaming them to
> printCOFFUnwindCodeEntries and printCOFFUnwindCodeEntry?
>
> Also, depending on how useful this would be in other places, (in a
> future patch set) you could write an "unwind code entry iterator"
> which is a forward iterator whose value_type is a class
> UnwindCodeEntry. I'm not sure how useful that would be, but that seems
> like a really clean way to work with these if you need to do something
> more complicated with them besides just printing them.
>
> +// Prints one unwind code entry. Because an entry can occupy up to 3 slots in
> +// the codes array, this function requires that the correct number slots is
> +// provided.
>
> Even though the check has already been done in the caller, it probably
> wouldn't hurt to `assert(UCs.size() >= getNumUsedSlots(UCs.front()))`
> at the start of this function for a bit of bulletproofing.
>
> + uint8_t getVersion() const {
> + return VersionAndFlags & 0x07;
> + }
> + uint8_t getFlags() const {
> + return (VersionAndFlags >> 3) & 0x1f;
> + }
> + uint8_t getFrameRegister() const {
> + return FrameRegisterAndOffset & 0x0f;
> + }
> + uint8_t getFrameOffset() const {
> + return (FrameRegisterAndOffset >> 4) & 0x0f;
> + }
>
> Indent the bodies here.
>
> Regarding 02-win64eh.diff, it looks like there are a number of
> different changes smashed into a single patch. Your original patch
> only contains changes from uint64_t to uint32_t. What was the problem
> that was fixed by that? I think it should be a separate patch so it is
> clear what is being changed and why. The change from bitfields to
> functions should be broken off into its own patch too. Breaking
> changes up like this is fundamental to LLVM's incremental development
> style; the goal of each patch is to communicate a single focused,
> commitable change to the reviewer.
>
> void *getLanguageSpecificData() {
> - return reinterpret_cast<void *>(&unwindCodes[(numCodes+1) & ~1]);
> + return reinterpret_cast<void *>(&UnwindCodes[(NumCodes+1) & ~1]);
> }
>
> You could clean up this function to make this be templated on the
> return type and have it directly return a pointer to the desired type,
> which should greatly reduce the number of reinterpret_cast<>'s in the
> surrounding code. You can do this in a separate patch (now or in the
> future).
>
> So in summary, (as far as I understand what is trying to be
> accomplished in this patch set) I think that you should break these
> changes into three patches:
>
> 1. fix the uint64_t to uint32_t size issues (include a good commit
> message explaining why this is needed)
> 2. cleanup the bitfields to functions
> 3. The changes in 01-llvm-objdump.diff (with the small fixes as requested above)
> [4. (optional) make getLanguageSpecificData be templated on the return
> type and clean up the reinterpret casts (you can put this anywhere in
> the patch set)]
>
> Also, I see that you are using git. I recommend that you use
> git-format-patch to make the patches, since that will preserve your
> commit message (otherwise the committer might make one up that you
> don't like). Don't be afraid to write nice long commit messages
> explaining the change and why it is needed ;)
>
> -- Sean Silva
>
> On Sun, Nov 25, 2012 at 11:01 AM, Kai <kai at redstar.de> wrote:
>> Hi Sean!
>>
>> The hint to use ArrayRef's turned out to be really good. I have the feeling
>> that a lot of the complexity is gone.
>>
>> I also implemented your other suggestions.
>>
>> The type casts are necessary, otherwise the output is garbage. I think the
>> ulittle8_t value is interpreted as a char, not a number. The casts make sure
>> that the value is formatted as a number. However I changed the old-style
>> cast to a static_cast<>.
>>
>> Regards
>> Kai
>>
>>
>> On 24.11.2012 21:47, Sean Silva wrote:
>>>
>>> This is a lot better!
>>>
>>> +unsigned printUnwindCode(const UnwindCode *UnwindCodes, unsigned
>>> SlotsAvail) {
>>>
>>> Use ArrayRef's instead of pointer-and-length. Here and in any callers.
>>>
>>> +void printCOFFUnwindCode(const UnwindCode *UnwindCodes, unsigned
>>> NumCodes) {
>>> + for (unsigned i = 0; i < NumCodes; ) {
>>> + i += printUnwindCode(&UnwindCodes[i], NumCodes - i);
>>> + }
>>> +}
>>>
>>> The `NumCodes-i` here is not immediately obvious; better to use an
>>> ArrayRef. How about
>>>
>>> void printCOFFUnwindCode(ArrayRef<UnwindCode> UCs) {
>>> for (const UnwindCode *I = UCs.begin(), *E = UCs.end(); I < E;)
>>> I += printUnwindCode(ArrayRef<UnwindCode>(I, E));
>>> }
>>>
>>> ?
>>>
>>> + switch (UnwindCodes[0].getUnwindOp()) {
>>> + default: llvm_unreachable("Invalid unwind code");
>>> + case UOP_PushNonVol:
>>> + case UOP_AllocSmall:
>>> + case UOP_SetFPReg:
>>> + case UOP_PushMachFrame:
>>> + UsedSlots = 1;
>>> + break;
>>> + case UOP_SaveNonVol:
>>> + case UOP_SaveXMM128:
>>> + UsedSlots = 2;
>>> + break;
>>> + case UOP_SaveNonVolBig:
>>> + case UOP_SaveXMM128Big:
>>> + UsedSlots = 3;
>>> + break;
>>> + case UOP_AllocLarge:
>>> + UsedSlots = (UnwindCodes[0].getOpInfo() == 0) ? 2 : 3;
>>> + break;
>>> + }
>>>
>>> How about breaking this out into a static function getNumUsedSlots()?
>>> Same for the other switch in this function. You can probably then
>>> inline what remains of this function (which should be very little)
>>> into the loop in printCOFFUnwindCode().
>>>
>>>
>>> +namespace {
>>>
>>> For functions, use static. See
>>> <http://llvm.org/docs/CodingStandards.html#anonymous-namespaces>.
>>>
>>> + if (UsedSlots > SlotsAvail) {
>>> + outs() << "Unwind data corrupted\n";
>>> + return SlotsAvail;
>>> + }
>>>
>>> Can you diagnose the situation better? Maybe say something like
>>> "encountered unwind op <...> of size <N> slots, but only <X> remaining
>>> in buffer".
>>>
>>> + outs() << " Size of prolog: " << (int) UI->PrologSize << "\n";
>>> + outs() << " Number of Codes: " << (int) UI->NumCodes << "\n";
>>>
>>> Why the casts here?
>>>
>>> + if (Name == ".pdata") {
>>> + Pdata = Obj->getCOFFSection(SI);
>>>
>>> I feel like you could use an early exit/continue to reduce the nesting
>>> here.
>>>
>>> + unsigned i = 0;
>>> + while ((Contents.size() - i) >= sizeof(RuntimeFunction)) {
>>> + const RuntimeFunction *RF =
>>> + reinterpret_cast<const RuntimeFunction *>(Contents.data() +
>>> i);
>>>
>>> This loop is wacky. Can you rewrite it as a regular for loop over
>>> RuntimeFunction*'s? It's really hard to tell what is actually being
>>> iterated over.
>>>
>>> + resolveSymbolName(Rels, Offset, Sym);
>>>
>>> not checking the error code?
>>>
>>> -- Sean Silva
>>
>>
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
-------------- next part --------------
>From 5a30baabcb910806ff0d187eae59a6b21bfd0507 Mon Sep 17 00:00:00 2001
From: kai <kai at redstar.de>
Date: Mon, 26 Nov 2012 16:40:05 +0100
Subject: [PATCH 1/3] Change member types of RuntimeFunction and UnwindInfo
from uint64_t to uint32_t.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
These members represent addresses. According to MSDN, they are image relative, that is, they are 32-bit offsets from the starting address of the image that contains the function table entry.
See MSDN for more information:
RUNTIME_FUNCTION: http://msdn.microsoft.com/en-us/library/ft9x1kdx.aspx
UNWIND_INFO: http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
Patch by João Matos.
---
include/llvm/Support/Win64EH.h | 43 +++++++++++++++++++++++++++++++-----------
1 file changed, 32 insertions(+), 11 deletions(-)
diff --git a/include/llvm/Support/Win64EH.h b/include/llvm/Support/Win64EH.h
index 8d74e10..a2c993b 100644
--- a/include/llvm/Support/Win64EH.h
+++ b/include/llvm/Support/Win64EH.h
@@ -60,9 +60,9 @@ enum {
/// RuntimeFunction - An entry in the table of functions with unwind info.
struct RuntimeFunction {
- uint64_t startAddress;
- uint64_t endAddress;
- uint64_t unwindInfoOffset;
+ uint32_t startAddress;
+ uint32_t endAddress;
+ uint32_t unwindInfoOffset;
};
/// UnwindInfo - An entry in the exception table.
@@ -75,22 +75,43 @@ struct UnwindInfo {
frameOffset:4;
UnwindCode unwindCodes[1];
+ /* The data after unwindCodes depends on flags.
+ * If UNW_ExceptionHandler or UNW_TerminateHandler is set then follows
+ * the address of the language-specific exception handler.
+ * If UNW_ChainInfo is set then follows a RuntimeFunction which defines
+ * the chained unwind info.
+ * For more information please see MSDN at:
+ * http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
+ */
+
+ /// getLanguageSpecificData - Return pointer to language specific data part
+ /// of UnwindInfo.
void *getLanguageSpecificData() {
return reinterpret_cast<void *>(&unwindCodes[(numCodes+1) & ~1]);
}
- uint64_t getLanguageSpecificHandlerOffset() {
- return *reinterpret_cast<uint64_t *>(getLanguageSpecificData());
- }
- void setLanguageSpecificHandlerOffset(uint64_t offset) {
- *reinterpret_cast<uint64_t *>(getLanguageSpecificData()) = offset;
+
+ /// getLanguageSpecificHandlerOffset - Return image-relativ offset of
+ /// language-specific exception handler
+ uint32_t getLanguageSpecificHandlerOffset() {
+ return *reinterpret_cast<uint32_t *>(getLanguageSpecificData());
}
- RuntimeFunction *getChainedFunctionEntry() {
- return reinterpret_cast<RuntimeFunction *>(getLanguageSpecificData());
+
+ /// setLanguageSpecificHandlerOffset - Set image-relativ offset of
+ /// language-specific exception handler
+ void setLanguageSpecificHandlerOffset(uint32_t offset) {
+ *reinterpret_cast<uint32_t *>(getLanguageSpecificData()) = offset;
}
+
+ /// getExceptionData - Return pointer to exception-specific data.
void *getExceptionData() {
- return reinterpret_cast<void *>(reinterpret_cast<uint64_t *>(
+ return reinterpret_cast<void *>(reinterpret_cast<uint32_t *>(
getLanguageSpecificData())+1);
}
+
+ /// getChainedFunctionEntry - Return pointer to chained unwind info.
+ RuntimeFunction *getChainedFunctionEntry() {
+ return reinterpret_cast<RuntimeFunction *>(getLanguageSpecificData());
+ }
};
--
1.8.0.msysgit.0
-------------- next part --------------
>From 720546c88ee1c4fc3c259118e9a6e5bbe692da21 Mon Sep 17 00:00:00 2001
From: kai <kai at redstar.de>
Date: Tue, 27 Nov 2012 06:44:30 +0100
Subject: [PATCH 2/3] Make Win64.h platform-neutral.
The standard types unit8_t, uint16_t and uint32_t are replaced with their counterparts from Endian.h. Accessor fucntions are introduced to replace bit fields.
Patch by Kai Nacke.
---
include/llvm/Support/Win64EH.h | 48 +++++++++++++++++++++++++++++-------------
1 file changed, 33 insertions(+), 15 deletions(-)
diff --git a/include/llvm/Support/Win64EH.h b/include/llvm/Support/Win64EH.h
index a2c993b..a50ac3005 100644
--- a/include/llvm/Support/Win64EH.h
+++ b/include/llvm/Support/Win64EH.h
@@ -17,6 +17,7 @@
#define LLVM_SUPPORT_WIN64EH_H
#include "llvm/Support/DataTypes.h"
+#include "llvm/Support/Endian.h"
namespace llvm {
namespace Win64EH {
@@ -39,11 +40,17 @@ enum UnwindOpcodes {
/// or part thereof.
union UnwindCode {
struct {
- uint8_t codeOffset;
- uint8_t unwindOp:4,
- opInfo:4;
+ support::ulittle8_t CodeOffset;
+ support::ulittle8_t UnwindOpAndOpInfo;
} u;
- uint16_t frameOffset;
+ support::ulittle16_t FrameOffset;
+
+ uint8_t getUnwindOp() const {
+ return u.UnwindOpAndOpInfo & 0x0F;
+ }
+ uint8_t getOpInfo() const {
+ return (u.UnwindOpAndOpInfo >> 4) & 0x0F;
+ }
};
enum {
@@ -60,20 +67,31 @@ enum {
/// RuntimeFunction - An entry in the table of functions with unwind info.
struct RuntimeFunction {
- uint32_t startAddress;
- uint32_t endAddress;
- uint32_t unwindInfoOffset;
+ support::ulittle32_t StartAddress;
+ support::ulittle32_t EndAddress;
+ support::ulittle32_t UnwindInfoOffset;
};
/// UnwindInfo - An entry in the exception table.
struct UnwindInfo {
- uint8_t version:3,
- flags:5;
- uint8_t prologSize;
- uint8_t numCodes;
- uint8_t frameRegister:4,
- frameOffset:4;
- UnwindCode unwindCodes[1];
+ support::ulittle8_t VersionAndFlags;
+ support::ulittle8_t PrologSize;
+ support::ulittle8_t NumCodes;
+ support::ulittle8_t FrameRegisterAndOffset;
+ UnwindCode UnwindCodes[1];
+
+ uint8_t getVersion() const {
+ return VersionAndFlags & 0x07;
+ }
+ uint8_t getFlags() const {
+ return (VersionAndFlags >> 3) & 0x1f;
+ }
+ uint8_t getFrameRegister() const {
+ return FrameRegisterAndOffset & 0x0f;
+ }
+ uint8_t getFrameOffset() const {
+ return (FrameRegisterAndOffset >> 4) & 0x0f;
+ }
/* The data after unwindCodes depends on flags.
* If UNW_ExceptionHandler or UNW_TerminateHandler is set then follows
@@ -87,7 +105,7 @@ struct UnwindInfo {
/// getLanguageSpecificData - Return pointer to language specific data part
/// of UnwindInfo.
void *getLanguageSpecificData() {
- return reinterpret_cast<void *>(&unwindCodes[(numCodes+1) & ~1]);
+ return reinterpret_cast<void *>(&UnwindCodes[(NumCodes+1) & ~1]);
}
/// getLanguageSpecificHandlerOffset - Return image-relativ offset of
--
1.8.0.msysgit.0
-------------- next part --------------
>From 9aa629d906b45db3c403b278ca1bdcb28bab26a3 Mon Sep 17 00:00:00 2001
From: kai <kai at redstar.de>
Date: Tue, 27 Nov 2012 20:13:04 +0100
Subject: [PATCH 3/3] Add dump of Win64 EH unwind data.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
The new command line option -unwind-info dumps the Win64 EH unwind data to the console. This is a nice feature if you need to debug generated EH data (e.g. from LLVM).
Includes a test case.
Initial patch by João Matos, extensions and rework by Kai Nacke.
---
test/tools/llvm-objdump/win64-unwind-data.s | 106 +++++++++
tools/llvm-objdump/CMakeLists.txt | 1 +
tools/llvm-objdump/COFFDump.cpp | 355 ++++++++++++++++++++++++++++
tools/llvm-objdump/llvm-objdump.cpp | 29 ++-
tools/llvm-objdump/llvm-objdump.h | 9 +
5 files changed, 497 insertions(+), 3 deletions(-)
create mode 100644 test/tools/llvm-objdump/win64-unwind-data.s
create mode 100644 tools/llvm-objdump/COFFDump.cpp
diff --git a/test/tools/llvm-objdump/win64-unwind-data.s b/test/tools/llvm-objdump/win64-unwind-data.s
new file mode 100644
index 0000000..1e4c742
--- /dev/null
+++ b/test/tools/llvm-objdump/win64-unwind-data.s
@@ -0,0 +1,106 @@
+// This test checks that the unwind data is dumped by llvm-objdump.
+// RUN: llvm-mc -triple x86_64-pc-win32 -filetype=obj %s | llvm-objdump -u - | FileCheck %s
+
+// CHECK: Unwind info:
+// CHECK: Function Table:
+// CHECK-NEXT: Start Address: .text
+// CHECK-NEXT: End Address: .text + 0x001b
+// CHECK-NEXT: Unwind Info Address: .xdata
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 1 UNW_ExceptionHandler
+// CHECK-NEXT: Size of prolog: 18
+// CHECK-NEXT: Number of Codes: 8
+// CHECK-NEXT: Frame register: RBX
+// CHECK-NEXT: Frame offset: 0
+// CHECK-NEXT: Unwind Codes:
+// CHECK-NEXT: 0x00: UOP_SetFPReg
+// CHECK-NEXT: 0x0f: UOP_PushNonVol RBX
+// CHECK-NEXT: 0x0e: UOP_SaveXMM128 XMM8 [0x0000]
+// CHECK-NEXT: 0x09: UOP_SaveNonVol RSI [0x0010]
+// CHECK-NEXT: 0x04: UOP_AllocSmall 24
+// CHECK-NEXT: 0x00: UOP_PushMachFrame w/o error code
+// CHECK: Function Table:
+// CHECK-NEXT: Start Address: .text + 0x0012
+// CHECK-NEXT: End Address: .text + 0x0012
+// CHECK-NEXT: Unwind Info Address: .xdata + 0x001c
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 4 UNW_ChainInfo
+// CHECK-NEXT: Size of prolog: 0
+// CHECK-NEXT: Number of Codes: 0
+// CHECK-NEXT: No frame pointer used
+// CHECK: Function Table:
+// CHECK-NEXT: Start Address: .text + 0x001b
+// CHECK-NEXT: End Address: .text + 0x001c
+// CHECK-NEXT: Unwind Info Address: .xdata + 0x002c
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 0
+// CHECK-NEXT: Size of prolog: 0
+// CHECK-NEXT: Number of Codes: 0
+// CHECK-NEXT: No frame pointer used
+// CHECK: Function Table:
+// CHECK-NEXT: Start Address: .text + 0x001c
+// CHECK-NEXT: End Address: .text + 0x0039
+// CHECK-NEXT: Unwind Info Address: .xdata + 0x0034
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 0
+// CHECK-NEXT: Size of prolog: 14
+// CHECK-NEXT: Number of Codes: 6
+// CHECK-NEXT: No frame pointer used
+// CHECK-NEXT: Unwind Codes:
+// CHECK-NEXT: 0x0e: UOP_AllocLarge 8454128
+// CHECK-NEXT: 0x07: UOP_AllocLarge 8190
+// CHECK-NEXT: 0x00: UOP_PushMachFrame w/o error code
+
+ .text
+ .globl func
+ .def func; .scl 2; .type 32; .endef
+ .seh_proc func
+func:
+ .seh_pushframe @code
+ subq $24, %rsp
+ .seh_stackalloc 24
+ movq %rsi, 16(%rsp)
+ .seh_savereg %rsi, 16
+ movups %xmm8, (%rsp)
+ .seh_savexmm %xmm8, 0
+ pushq %rbx
+ .seh_pushreg 3
+ mov %rsp, %rbx
+ .seh_setframe 3, 0
+ .seh_endprologue
+ .seh_handler __C_specific_handler, @except
+ .seh_handlerdata
+ .long 0
+ .text
+ .seh_startchained
+ .seh_endprologue
+ .seh_endchained
+ lea (%rbx), %rsp
+ pop %rbx
+ addq $24, %rsp
+ ret
+ .seh_endproc
+
+// Test emission of small functions.
+ .globl smallFunc
+ .def smallFunc; .scl 2; .type 32; .endef
+ .seh_proc smallFunc
+smallFunc:
+ ret
+ .seh_endproc
+
+// Function with big stack allocation.
+ .globl smallFunc
+ .def allocFunc; .scl 2; .type 32; .endef
+ .seh_proc smallFunc
+allocFunc:
+ .seh_pushframe @code
+ subq $65520, %rsp
+ .seh_stackalloc 65520
+ sub $8454128, %rsp
+ .seh_stackalloc 8454128
+ .seh_endprologue
+ add $8454128, %rsp
+ addq $65520, %rsp
+ ret
+ .seh_endproc
diff --git a/tools/llvm-objdump/CMakeLists.txt b/tools/llvm-objdump/CMakeLists.txt
index f3b2e1f..5001435 100644
--- a/tools/llvm-objdump/CMakeLists.txt
+++ b/tools/llvm-objdump/CMakeLists.txt
@@ -9,6 +9,7 @@ set(LLVM_LINK_COMPONENTS
add_llvm_tool(llvm-objdump
llvm-objdump.cpp
+ COFFDump.cpp
MachODump.cpp
MCFunction.cpp
)
diff --git a/tools/llvm-objdump/COFFDump.cpp b/tools/llvm-objdump/COFFDump.cpp
new file mode 100644
index 0000000..1218afa
--- /dev/null
+++ b/tools/llvm-objdump/COFFDump.cpp
@@ -0,0 +1,355 @@
+//===-- COFFDump.cpp - COFF-specific dumper ---------------------*- C++ -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// \brief This file implements the COFF-specific dumper for llvm-objdump.
+/// It outputs the Win64 EH data structures as plain text.
+/// The encoding of the unwind codes is decribed in MSDN:
+/// http://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
+///
+//===----------------------------------------------------------------------===//
+
+#include "llvm-objdump.h"
+#include "llvm/Object/COFF.h"
+#include "llvm/Object/ObjectFile.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/system_error.h"
+#include "llvm/Support/Win64EH.h"
+#include <algorithm>
+#include <cstring>
+
+using namespace llvm;
+using namespace object;
+using namespace llvm::Win64EH;
+
+// Returns the name of the unwind code.
+static StringRef getUnwindCodeTypeName(uint8_t Code) {
+ switch(Code) {
+ default: llvm_unreachable("Invalid unwind code");
+ case UOP_PushNonVol: return "UOP_PushNonVol";
+ case UOP_AllocLarge: return "UOP_AllocLarge";
+ case UOP_AllocSmall: return "UOP_AllocSmall";
+ case UOP_SetFPReg: return "UOP_SetFPReg";
+ case UOP_SaveNonVol: return "UOP_SaveNonVol";
+ case UOP_SaveNonVolBig: return "UOP_SaveNonVolBig";
+ case UOP_SaveXMM128: return "UOP_SaveXMM128";
+ case UOP_SaveXMM128Big: return "UOP_SaveXMM128Big";
+ case UOP_PushMachFrame: return "UOP_PushMachFrame";
+ }
+}
+
+// Returns the name of a referenced register.
+static StringRef getUnwindRegisterName(uint8_t Reg) {
+ switch(Reg) {
+ default: llvm_unreachable("Invalid register");
+ case 0: return "RAX";
+ case 1: return "RCX";
+ case 2: return "RDX";
+ case 3: return "RBX";
+ case 4: return "RSP";
+ case 5: return "RBP";
+ case 6: return "RSI";
+ case 7: return "RDI";
+ case 8: return "R8";
+ case 9: return "R9";
+ case 10: return "R10";
+ case 11: return "R11";
+ case 12: return "R12";
+ case 13: return "R13";
+ case 14: return "R14";
+ case 15: return "R15";
+ }
+}
+
+// Calculates the number of array slots required for the unwind code.
+static unsigned getNumUsedSlots(const UnwindCode &UnwindCode) {
+ switch (UnwindCode.getUnwindOp()) {
+ default: llvm_unreachable("Invalid unwind code");
+ case UOP_PushNonVol:
+ case UOP_AllocSmall:
+ case UOP_SetFPReg:
+ case UOP_PushMachFrame:
+ return 1;
+ case UOP_SaveNonVol:
+ case UOP_SaveXMM128:
+ return 2;
+ case UOP_SaveNonVolBig:
+ case UOP_SaveXMM128Big:
+ return 3;
+ case UOP_AllocLarge:
+ return (UnwindCode.getOpInfo() == 0) ? 2 : 3;
+ }
+}
+
+// Prints one unwind code. Because an unwind code can occupy up to 3 slots in
+// the unwind codes array, this function requires that the correct number of
+// slots is provided.
+static void printUnwindCode(ArrayRef<UnwindCode> UCs) {
+ assert(UCs.size() >= getNumUsedSlots(UCs[0]));
+ outs() << format(" 0x%02x: ", unsigned(UCs[0].u.CodeOffset))
+ << getUnwindCodeTypeName(UCs[0].getUnwindOp());
+ switch (UCs[0].getUnwindOp()) {
+ case UOP_PushNonVol:
+ outs() << " " << getUnwindRegisterName(UCs[0].getOpInfo());
+ break;
+ case UOP_AllocLarge:
+ if (UCs[0].getOpInfo() == 0) {
+ outs() << " " << UCs[1].FrameOffset;
+ } else {
+ outs() << " " << UCs[1].FrameOffset
+ + (((uint32_t) UCs[2].FrameOffset) << 16);
+ }
+ break;
+ case UOP_AllocSmall:
+ outs() << " " << ((UCs[0].getOpInfo()+1) * 8);
+ break;
+ case UOP_SetFPReg:
+ outs() << " ";
+ break;
+ case UOP_SaveNonVol:
+ outs() << " " << getUnwindRegisterName(UCs[0].getOpInfo())
+ << format(" [0x%04x]", 8 * UCs[1].FrameOffset);
+ break;
+ case UOP_SaveNonVolBig:
+ outs() << " " << getUnwindRegisterName(UCs[0].getOpInfo())
+ << format(" [0x%08x]", UCs[1].FrameOffset
+ + (((uint32_t) UCs[2].FrameOffset) << 16));
+ break;
+ case UOP_SaveXMM128:
+ outs() << " XMM" << static_cast<uint32_t>(UCs[0].getOpInfo())
+ << format(" [0x%04x]", 16 * UCs[1].FrameOffset);
+ break;
+ case UOP_SaveXMM128Big:
+ outs() << " XMM" << UCs[0].getOpInfo()
+ << format(" [0x%08x]", UCs[1].FrameOffset
+ + (((uint32_t) UCs[2].FrameOffset) << 16));
+ break;
+ case UOP_PushMachFrame:
+ outs() << " " << (UCs[0].getOpInfo() ? "w/o" : "w")
+ << " error code";
+ break;
+ }
+ outs() << "\n";
+}
+
+static void printAllUnwindCodes(ArrayRef<UnwindCode> UCs) {
+ for (const UnwindCode *I = UCs.begin(), *E = UCs.end(); I < E; ) {
+ unsigned UsedSlots = getNumUsedSlots(*I);
+ if (UsedSlots > UCs.size()) {
+ outs() << "Unwind data corrupted: Encountered unwind op "
+ << getUnwindCodeTypeName((*I).getUnwindOp())
+ << " which requires " << UsedSlots
+ << " slots, but only " << UCs.size()
+ << " remaining in buffer";
+ return ;
+ }
+ printUnwindCode(ArrayRef<UnwindCode>(I, E));
+ I += UsedSlots;
+ }
+}
+
+// Given a symbol sym this functions returns the address and section of it.
+static error_code resolveSectionAndAddress(const COFFObjectFile *Obj,
+ const SymbolRef &Sym,
+ const coff_section *&ResolvedSection,
+ uint64_t &ResolvedAddr) {
+ if (error_code ec = Sym.getAddress(ResolvedAddr)) return ec;
+ section_iterator iter(Obj->begin_sections());
+ if (error_code ec = Sym.getSection(iter)) return ec;
+ ResolvedSection = Obj->getCOFFSection(iter);
+ return object_error::success;
+}
+
+// Given a vector of relocations for a section and an offset into this section
+// the function returns the symbol used for the relocation at the offset.
+static error_code resolveSymbol(const std::vector<RelocationRef> &Rels,
+ uint64_t Offset, SymbolRef &Sym) {
+ for (std::vector<RelocationRef>::const_iterator I = Rels.begin(),
+ E = Rels.end();
+ I != E; ++I) {
+ uint64_t Ofs;
+ if (error_code ec = I->getOffset(Ofs)) return ec;
+ if (Ofs == Offset) {
+ if (error_code ec = I->getSymbol(Sym)) return ec;
+ break;
+ }
+ }
+ return object_error::success;
+}
+
+// Given a vector of relocations for a section and an offset into this section
+// the function resolves the symbol used for the relocation at the offset and
+// returns the section content and the address inside the content pointed to
+// by the symbol.
+static error_code getSectionContents(const COFFObjectFile *Obj,
+ const std::vector<RelocationRef> &Rels,
+ uint64_t Offset,
+ ArrayRef<uint8_t> &Contents,
+ uint64_t &Addr) {
+ SymbolRef Sym;
+ if (error_code ec = resolveSymbol(Rels, Offset, Sym)) return ec;
+ const coff_section *Section;
+ if (error_code ec = resolveSectionAndAddress(Obj, Sym, Section, Addr))
+ return ec;
+ if (error_code ec = Obj->getSectionContents(Section, Contents)) return ec;
+ return object_error::success;
+}
+
+// Given a vector of relocations for a section and an offset into this section
+// the function returns the name of the symbol used for the relocation at the
+// offset.
+static error_code resolveSymbolName(const std::vector<RelocationRef> &Rels,
+ uint64_t Offset, StringRef &Name) {
+ SymbolRef Sym;
+ if (error_code ec = resolveSymbol(Rels, Offset, Sym)) return ec;
+ if (error_code ec = Sym.getName(Name)) return ec;
+ return object_error::success;
+}
+
+static void printCOFFSymbolAddress(llvm::raw_ostream &Out,
+ const std::vector<RelocationRef> &Rels,
+ uint64_t Offset, uint32_t Disp) {
+ StringRef Sym;
+ if (error_code ec = resolveSymbolName(Rels, Offset, Sym)) {
+ error(ec);
+ return ;
+ }
+ Out << Sym;
+ if (Disp > 0)
+ Out << format(" + 0x%04x", Disp);
+}
+
+void llvm::printCOFFUnwindInfo(const COFFObjectFile *Obj) {
+ const coff_file_header *Header;
+ if (error(Obj->getHeader(Header))) return;
+
+ if (Header->Machine != COFF::IMAGE_FILE_MACHINE_AMD64) {
+ errs() << "Unsupported image machine type "
+ "(currently only AMD64 is supported).\n";
+ return;
+ }
+
+ const coff_section *Pdata = 0;
+
+ error_code ec;
+ for (section_iterator SI = Obj->begin_sections(),
+ SE = Obj->end_sections();
+ SI != SE; SI.increment(ec)) {
+ if (error(ec)) return;
+
+ StringRef Name;
+ if (error(SI->getName(Name))) continue;
+
+ if (Name != ".pdata") continue;
+
+ Pdata = Obj->getCOFFSection(SI);
+ std::vector<RelocationRef> Rels;
+ for (relocation_iterator RI = SI->begin_relocations(),
+ RE = SI->end_relocations();
+ RI != RE; RI.increment(ec)) {
+ if (error(ec)) break;
+ Rels.push_back(*RI);
+ }
+
+ // Sort relocations by address.
+ std::sort(Rels.begin(), Rels.end(), RelocAddressLess);
+
+ ArrayRef<uint8_t> Contents;
+ if (error(Obj->getSectionContents(Pdata, Contents))) continue;
+ if (Contents.empty()) continue;
+
+ ArrayRef<RuntimeFunction> RFs(
+ reinterpret_cast<const RuntimeFunction *>(Contents.data()),
+ Contents.size() / sizeof(RuntimeFunction));
+ for (const RuntimeFunction *I = RFs.begin(), *E = RFs.end(); I < E; ++I) {
+ const uint64_t SectionOffset = std::distance(RFs.begin(), I)
+ * sizeof(RuntimeFunction);
+
+ outs() << "Function Table:\n";
+
+ outs() << " Start Address: ";
+ printCOFFSymbolAddress(outs(), Rels, SectionOffset +
+ offsetof(RuntimeFunction, StartAddress),
+ I->StartAddress);
+ outs() << "\n";
+
+ outs() << " End Address: ";
+ printCOFFSymbolAddress(outs(), Rels, SectionOffset +
+ offsetof(RuntimeFunction, EndAddress),
+ I->EndAddress);
+ outs() << "\n";
+
+ outs() << " Unwind Info Address: ";
+ printCOFFSymbolAddress(outs(), Rels, SectionOffset +
+ offsetof(RuntimeFunction, UnwindInfoOffset),
+ I->UnwindInfoOffset);
+ outs() << "\n";
+
+ ArrayRef<uint8_t> XContents;
+ uint64_t UnwindInfoOffset = 0;
+ if (error(getSectionContents(Obj, Rels, SectionOffset +
+ offsetof(RuntimeFunction, UnwindInfoOffset),
+ XContents, UnwindInfoOffset))) continue;
+ if (XContents.empty()) continue;
+
+ UnwindInfoOffset += I->UnwindInfoOffset;
+ if (UnwindInfoOffset > XContents.size()) continue;
+
+ const Win64EH::UnwindInfo *UI =
+ reinterpret_cast<const Win64EH::UnwindInfo *>
+ (XContents.data() + UnwindInfoOffset);
+
+ // The casts to int are required in order to output the value as number.
+ // Without the casts the value would be interpreted as char data (which
+ // results in garbage output).
+ outs() << " Version: " << static_cast<int>(UI->getVersion()) << "\n";
+ outs() << " Flags: " << static_cast<int>(UI->getFlags());
+ if (UI->getFlags()) {
+ if (UI->getFlags() & UNW_ExceptionHandler)
+ outs() << " UNW_ExceptionHandler";
+ if (UI->getFlags() & UNW_TerminateHandler)
+ outs() << " UNW_TerminateHandler";
+ if (UI->getFlags() & UNW_ChainInfo)
+ outs() << " UNW_ChainInfo";
+ }
+ outs() << "\n";
+ outs() << " Size of prolog: "
+ << static_cast<int>(UI->PrologSize) << "\n";
+ outs() << " Number of Codes: "
+ << static_cast<int>(UI->NumCodes) << "\n";
+ // Maybe this should move to output of UOP_SetFPReg?
+ if (UI->getFrameRegister()) {
+ outs() << " Frame register: "
+ << getUnwindRegisterName(UI->getFrameRegister())
+ << "\n";
+ outs() << " Frame offset: "
+ << 16 * UI->getFrameOffset()
+ << "\n";
+ } else {
+ outs() << " No frame pointer used\n";
+ }
+ if (UI->getFlags() & (UNW_ExceptionHandler | UNW_TerminateHandler)) {
+ // FIXME: Output exception handler data
+ } else if (UI->getFlags() & UNW_ChainInfo) {
+ // FIXME: Output chained unwind info
+ }
+
+ if (UI->NumCodes)
+ outs() << " Unwind Codes:\n";
+
+ printAllUnwindCodes(ArrayRef<UnwindCode>(&UI->UnwindCodes[0],
+ UI->NumCodes));
+
+ outs() << "\n\n";
+ outs().flush();
+ }
+ }
+}
diff --git a/tools/llvm-objdump/llvm-objdump.cpp b/tools/llvm-objdump/llvm-objdump.cpp
index ddfcca3..282bf01 100644
--- a/tools/llvm-objdump/llvm-objdump.cpp
+++ b/tools/llvm-objdump/llvm-objdump.cpp
@@ -104,9 +104,16 @@ static cl::opt<bool>
NoShowRawInsn("no-show-raw-insn", cl::desc("When disassembling instructions, "
"do not print the instruction bytes."));
+static cl::opt<bool>
+UnwindInfo("unwind-info", cl::desc("Display unwind information"));
+
+static cl::alias
+UnwindInfoShort("u", cl::desc("Alias for --unwind-info"),
+ cl::aliasopt(::UnwindInfo));
+
static StringRef ToolName;
-static bool error(error_code ec) {
+bool llvm::error(error_code ec) {
if (!ec) return false;
outs() << ToolName << ": error reading file: " << ec.message() << ".\n";
@@ -165,7 +172,7 @@ void llvm::DumpBytes(StringRef bytes) {
outs() << output;
}
-static bool RelocAddressLess(RelocationRef a, RelocationRef b) {
+bool llvm::RelocAddressLess(RelocationRef a, RelocationRef b) {
uint64_t a_addr, b_addr;
if (error(a.getAddress(a_addr))) return false;
if (error(b.getAddress(b_addr))) return false;
@@ -573,6 +580,19 @@ static void PrintSymbolTable(const ObjectFile *o) {
}
}
+static void PrintUnwindInfo(const ObjectFile *o) {
+ outs() << "Unwind info:\n\n";
+
+ if (const COFFObjectFile *coff = dyn_cast<COFFObjectFile>(o)) {
+ printCOFFUnwindInfo(coff);
+ } else {
+ // TODO: Extract DWARF dump tool to objdump.
+ errs() << "This operation is only currently supported "
+ "for COFF object files.\n";
+ return;
+ }
+}
+
static void DumpObject(const ObjectFile *o) {
outs() << '\n';
outs() << o->getFileName()
@@ -588,6 +608,8 @@ static void DumpObject(const ObjectFile *o) {
PrintSectionContents(o);
if (SymbolTable)
PrintSymbolTable(o);
+ if (::UnwindInfo)
+ PrintUnwindInfo(o);
}
/// @brief Dump each object file in \a a;
@@ -666,7 +688,8 @@ int main(int argc, char **argv) {
&& !Relocations
&& !SectionHeaders
&& !SectionContents
- && !SymbolTable) {
+ && !SymbolTable
+ && !::UnwindInfo) {
cl::PrintHelpMessage();
return 2;
}
diff --git a/tools/llvm-objdump/llvm-objdump.h b/tools/llvm-objdump/llvm-objdump.h
index aa71b77..9f5a8c3 100644
--- a/tools/llvm-objdump/llvm-objdump.h
+++ b/tools/llvm-objdump/llvm-objdump.h
@@ -17,12 +17,21 @@
namespace llvm {
+namespace object {
+ class COFFObjectFile;
+ class RelocationRef;
+}
+class error_code;
+
extern cl::opt<std::string> TripleName;
extern cl::opt<std::string> ArchName;
// Various helper functions.
+bool error(error_code ec);
+bool RelocAddressLess(object::RelocationRef a, object::RelocationRef b);
void DumpBytes(StringRef bytes);
void DisassembleInputMachO(StringRef Filename);
+void printCOFFUnwindInfo(const object::COFFObjectFile* o);
class StringRefMemoryObject : public MemoryObject {
virtual void anchor();
--
1.8.0.msysgit.0
More information about the llvm-commits
mailing list