[MC] Fix clang integrated assembler generates incorrect relocations… #83115

yingopq · 2024-02-27T09:56:54Z

… for mips32

When mips asm parse instruction la, check whether .rdata section was accessed and parsed, if it was not, then check the following statement, check if local symbol(eg "hello:") was in .rdata section, if it was in it, IsLocalSym is true.

Fix #65020

github-actions · 2024-02-27T09:59:33Z

✅ With the latest revision this PR passed the C/C++ code formatter.

brad0 · 2024-03-08T23:20:07Z

@wzssyqa

wzssyqa · 2024-03-12T23:39:23Z

llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp

+            CurrentASMContent.take_front(NewlineIndex).take_back(SymbolLength);
+
+        // Get and check if ".rdata" section exist.
+        size_t RdataIndex = CurrentASMContent.find(".rdata");


GCC uses ".rodata.str1.4" for this code.

char *g = "xxxx"; char *x() {return g;}

Maybe the section name won't require to be something like ".rodata".

And I think that we need a more generic solution.
The naive scan string may have lots of bug.
maybe something like, in the next lines, we may have:

do_read_rdata: lw ... lw ...

Now, MipsAsmParser.cpp use following code to parse .rdata，so the name, maybe I think, basically it won’t change.

if (IDVal == ".rdata") { parseRSectionDirective(".rodata"); return false; }

And I have considered the situation you mentioned, the patch contains this situation, it use containsto check if rdata section contains local symbol, instead of checking the next line.

find is not a good way here: you may meet some strange strings from user's asmcode, may be from comment?

# this is a comment about .rdata

wzssyqa · 2024-03-12T23:53:01Z

llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp

    // For O32, "$"-prefixed symbols are recognized as temporary while
    // .L-prefixed symbols are not (PrivateGlobalPrefix is "$"). Recognize ".L"
    // manually.
    if (ABI.IsO32() && Res.getSymA()->getSymbol().getName().starts_with(".L"))
      IsLocalSym = true;
+    else {
+      if (HasParseRdata == false) {


It can be

else if (HasParseRdata == false)

So we won't need 2-level nest.

OK, I would change it.

wzssyqa · 2024-03-12T23:59:46Z

In fact we can update the generic asm scan for all ports, instead mips only.
See MCSymbol *MCContext::createSymbol(StringRef Name, bool AlwaysAddSuffix, bool CanBeUnnamed) in llvm/lib/MC/MCContext.cpp

yingopq · 2024-03-15T03:34:09Z

In fact we can update the generic asm scan for all ports, instead mips only. See MCSymbol *MCContext::createSymbol(StringRef Name, bool AlwaysAddSuffix, bool CanBeUnnamed) in llvm/lib/MC/MCContext.cpp

Can you explain it in more detail?

wzssyqa · 2024-03-15T07:01:41Z

llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp

+        size_t CommaIndex = CurrentASMContent.find_first_of(',');
+        size_t SymbolLength = NewlineIndex - CommaIndex - 2;
+        StringRef LocalSymbol =
+            CurrentASMContent.take_front(NewlineIndex).take_back(SymbolLength);


if the asm code is like

# more spaces and windows style newline LA $a0, localsymbolname \r\n

It won't work.

wzssyqa · 2024-03-15T07:05:17Z

llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp

+            CurrentASMContent.take_front(NewlineIndex).take_back(SymbolLength);
+
+        // Get and check if ".rdata" section exist.
+        size_t RdataIndex = CurrentASMContent.find(".rdata");


find is not a good way here: you may meet some strange strings from user's asmcode, may be from comment?

# this is a comment about .rdata

wzssyqa · 2024-03-15T07:11:20Z

In fact we can update the generic asm scan for all ports, instead mips only. See MCSymbol *MCContext::createSymbol(StringRef Name, bool AlwaysAddSuffix, bool CanBeUnnamed) in llvm/lib/MC/MCContext.cpp

Can you explain it in more detail?

Integrated assembler scan the asm code in MCParser/MCAsmParser.cpp, and if it find a symbol, it will call createSymbol just now, so that it cannot know that this symbol is defined in the same asm code later.

That's the real problem, and this problem effects all architectures.

yingopq · 2024-03-15T07:14:35Z

In fact we can update the generic asm scan for all ports, instead mips only. See MCSymbol *MCContext::createSymbol(StringRef Name, bool AlwaysAddSuffix, bool CanBeUnnamed) in llvm/lib/MC/MCContext.cpp

Can you explain it in more detail?

Integrated assembler scan the asm code in MCParser/MCAsmParser.cpp, and if it find a symbol, it will call createSymbol just now, so that it cannot know that this symbol is defined in the same asm code later.

That's the real problem, and this problem effects all architectures.

OK，I would find another solution.

yingopq · 2024-03-29T08:50:53Z

@wzssyqa Please help review this patch yingopq@dcc86c6

…or mips32 Modify asm scan starting point to ensure scaning .rdata section before .text section. Save begin location, check if have .rdata section, confirm where to begin parse. If have it and .rdata after .text, begin parse from .rdata, when go to Eof, jump to begin location and then continue parse. If did not have .rdata or .rdata is before .text, jump to begin location and then parse. Fix llvm#65020

yingopq · 2024-04-03T08:41:36Z

@topperc @wzssyqa Could you help review this patch and give me some advice?
This patch would generate some fail testcase which I am searching.
I did not konw if we should change asm scan beginning point or scan twice unconditionally.
Thanks!

wzssyqa · 2024-04-04T09:50:57Z

I guess that we can also CC @MaskRay

brad0 · 2024-04-11T04:48:59Z

@MaskRay

yingopq · 2024-04-23T07:08:21Z

@wzssyqa @topperc @MaskRay Could you give me some advice?
I did not konw if we should change asm scan beginning point or scan twice unconditionally.
Thanks!

wzssyqa · 2024-04-23T07:34:57Z

@wzssyqa @topperc @MaskRay Could you give me some advice? I did not konw if we should change asm scan beginning point or scan twice unconditionally. Thanks!

I think that you can try to figure out a patch first. So that we can run some test.
Please go ahead.

yingopq · 2024-04-25T08:47:01Z

@wzssyqa @topperc @MaskRay Could you give me some advice? I did not konw if we should change asm scan beginning point or scan twice unconditionally. Thanks!

I think that you can try to figure out a patch first. So that we can run some test. Please go ahead.

@wzssyqa I tested with this patch about scaning twice, the result of readelf -r would occur double entries than normal, could you have some advice? Is it better to scan it once?

diff --git a/llvm/lib/MC/MCParser/AsmParser.cpp b/llvm/lib/MC/MCParser/AsmParser.cpp
index 76a3e501f459..36c3f550e320 100644
--- a/llvm/lib/MC/MCParser/AsmParser.cpp
+++ b/llvm/lib/MC/MCParser/AsmParser.cpp
@@ -997,6 +997,9 @@ bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) {
 
   getTargetParser().onBeginOfFile();
 
+  int cnt = 0;
+  SMLoc IDLoc_start = getTok().getLoc();
+A:
   // While we have input, parse each statement.
   while (Lexer.isNot(AsmToken::Eof)) {
     ParseStatementInfo Info(&AsmStrRewrites);
@@ -1017,6 +1020,13 @@ bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) {
       eatToEndOfStatement();
   }
 
+  cnt++;
+  if (cnt == 1) {
+    jumpToLoc(IDLoc_start);
+    Lex();
+    goto A;
+  }
+
   getTargetParser().onEndOfFile();
   printPendingErrors();
 
diff --git a/llvm/lib/MC/MCStreamer.cpp b/llvm/lib/MC/MCStreamer.cpp
index 176d55aa890b..e6071abc02ed 100644
--- a/llvm/lib/MC/MCStreamer.cpp
+++ b/llvm/lib/MC/MCStreamer.cpp
@@ -42,7 +42,7 @@
 #include <utility>
 
 using namespace llvm;
-
+#define DEBUG_TYPE "mips-mcstreamer"
 MCTargetStreamer::MCTargetStreamer(MCStreamer &S) : Streamer(S) {
   S.setTargetStreamer(this);
 }
@@ -424,9 +424,11 @@ void MCStreamer::assignFragment(MCSymbol *Symbol, MCFragment *Fragment) {
 void MCStreamer::emitLabel(MCSymbol *Symbol, SMLoc Loc) {
   Symbol->redefineIfPossible();
 
-  if (!Symbol->isUndefined() || Symbol->isVariable())
-    return getContext().reportError(Loc, "symbol '" + Twine(Symbol->getName()) +
-                                             "' is already defined");
+  if (!Symbol->isUndefined() || Symbol->isVariable()) {
+    LLVM_DEBUG(dbgs() << DEBUG_TYPE << "symbol '" + Twine(Symbol->getName()) +
+                                             "' is already defined" << "\n");
+    return;
+  }
 
   assert(!Symbol->isVariable() && "Cannot emit a variable symbol!");
   assert(getCurrentSectionOnly() && "Cannot emit before setting section!");

$ readelf -r 3.o

Relocation section '.rel.text' at offset 0x1b0 contains 11 entries:
 Offset     Info    Type            Sym.Value  Sym. Name
00000000  00000405 R_MIPS_HI16       00000000   _gp_disp
00000004  00000406 R_MIPS_LO16       00000000   _gp_disp
0000001c  0000050b R_MIPS_CALL16     00000000   printf
00000020  00000525 R_MIPS_JALR       00000000   printf
0000003c  00000405 R_MIPS_HI16       00000000   _gp_disp
00000040  00000406 R_MIPS_LO16       00000000   _gp_disp
00000018  00000209 R_MIPS_GOT16      00000000   .rodata
00000054  00000209 R_MIPS_GOT16      00000000   .rodata
00000058  00000206 R_MIPS_LO16       00000000   .rodata
0000005c  0000050b R_MIPS_CALL16     00000000   printf
00000060  00000525 R_MIPS_JALR       00000000   printf

Relocation section '.rel.pdr' at offset 0x208 contains 2 entries:
 Offset     Info    Type            Sym.Value  Sym. Name
00000000  00000302 R_MIPS_32         0000003c   main
00000020  00000302 R_MIPS_32         0000003c   main

MaskRay

I don't have more time spent on investigating the issue, but adding a lot of mips-specific code to the generic AsmParser.cpp is not acceptable.

wzssyqa · 2024-04-27T03:15:37Z

I don't have more time spent on investigating the issue, but adding a lot of mips-specific code to the generic AsmParser.cpp is not acceptable.

Although her code is not correct, while I don't think that it is mips-specific, instead it is a generic problem.
With our current logic, if there are 2 symbols in a single ASM source file, and the previous symbol uses the later one,
the previous symbol don't treat the later symbol as local symbol.

GNU GAS doesn't have this problem.

yingopq · 2024-04-28T02:17:44Z

I don't have more time spent on investigating the issue, but adding a lot of mips-specific code to the generic AsmParser.cpp is not acceptable.

@MaskRay Now we have two modification plans.
The first one is in the files changed in this PR, which changes the starting point of the scan to ensure that the .rdata section is in front of the .text section; the second one is scan twice as recommended by @wzssyqa.

yingopq force-pushed the Fix_bug_65020 branch from b7015a9 to 33a29c9 Compare February 27, 2024 10:10

brad0 requested a review from wzssyqa February 27, 2024 22:20

wzssyqa requested changes Mar 12, 2024

View reviewed changes

wzssyqa reviewed Mar 15, 2024

View reviewed changes

yingopq force-pushed the Fix_bug_65020 branch from 33a29c9 to bfefb62 Compare April 3, 2024 08:29

yingopq force-pushed the Fix_bug_65020 branch from bfefb62 to 07fb14f Compare April 3, 2024 08:34

MaskRay requested changes Apr 26, 2024

View reviewed changes

yingopq changed the title ~~[Mips] Fix clang integrated assembler generates incorrect relocations…~~ [MC] Fix clang integrated assembler generates incorrect relocations… Apr 28, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MC] Fix clang integrated assembler generates incorrect relocations… #83115

[MC] Fix clang integrated assembler generates incorrect relocations… #83115

yingopq commented Feb 27, 2024

github-actions bot commented Feb 27, 2024 •

edited

Loading

brad0 commented Mar 8, 2024

wzssyqa Mar 12, 2024

wzssyqa Mar 12, 2024

yingopq Mar 15, 2024

wzssyqa Mar 15, 2024

wzssyqa Mar 12, 2024

yingopq Mar 15, 2024

wzssyqa commented Mar 12, 2024

yingopq commented Mar 15, 2024

wzssyqa Mar 15, 2024

wzssyqa Mar 15, 2024

wzssyqa commented Mar 15, 2024

yingopq commented Mar 15, 2024

yingopq commented Mar 29, 2024

yingopq commented Apr 3, 2024 •

edited

Loading

wzssyqa commented Apr 4, 2024

brad0 commented Apr 11, 2024

yingopq commented Apr 23, 2024

wzssyqa commented Apr 23, 2024

yingopq commented Apr 25, 2024 •

edited

Loading

MaskRay left a comment

wzssyqa commented Apr 27, 2024

yingopq commented Apr 28, 2024

[MC] Fix clang integrated assembler generates incorrect relocations… #83115

Are you sure you want to change the base?

[MC] Fix clang integrated assembler generates incorrect relocations… #83115

Conversation

yingopq commented Feb 27, 2024

github-actions bot commented Feb 27, 2024 • edited Loading

brad0 commented Mar 8, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wzssyqa commented Mar 12, 2024

yingopq commented Mar 15, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wzssyqa commented Mar 15, 2024

yingopq commented Mar 15, 2024

yingopq commented Mar 29, 2024

yingopq commented Apr 3, 2024 • edited Loading

wzssyqa commented Apr 4, 2024

brad0 commented Apr 11, 2024

yingopq commented Apr 23, 2024

wzssyqa commented Apr 23, 2024

yingopq commented Apr 25, 2024 • edited Loading

MaskRay left a comment

Choose a reason for hiding this comment

wzssyqa commented Apr 27, 2024

yingopq commented Apr 28, 2024

github-actions bot commented Feb 27, 2024 •

edited

Loading

yingopq commented Apr 3, 2024 •

edited

Loading

yingopq commented Apr 25, 2024 •

edited

Loading