[lld] dd8cb3d - [ELF] Support high address DW_EH_sdata4 for ELFCLASS32

via llvm-commits llvm-commits at lists.llvm.org
Mon May 20 00:13:13 PDT 2024


Author: Fangrui Song
Date: 2024-05-20T00:13:09-07:00
New Revision: dd8cb3d4f120edcf5fc3939594ee086c44010274

URL: https://github.com/llvm/llvm-project/commit/dd8cb3d4f120edcf5fc3939594ee086c44010274
DIFF: https://github.com/llvm/llvm-project/commit/dd8cb3d4f120edcf5fc3939594ee086c44010274.diff

LOG: [ELF] Support high address DW_EH_sdata4 for ELFCLASS32

When the address pointer encoding in FDEs uses
DW_EH_PE_absptr|DW_EH_PE_sdata4, the address is sign-extended to 64-bit
by `readFdeAddr`. We should truncate the address to 32-bit for
ELFCLASS32. Otherwise, `isInt<32>(pc - va)` could be false, leading to a
spurious error in `getFdeData`.

In LLVM, this appears a MIPS-specific issue.
Fix #88852

Pull Request: https://github.com/llvm/llvm-project/pull/92438

Added: 
    

Modified: 
    lld/ELF/SyntheticSections.cpp
    lld/test/ELF/mips-eh_frame-pic.s

Removed: 
    


################################################################################
diff  --git a/lld/ELF/SyntheticSections.cpp b/lld/ELF/SyntheticSections.cpp
index 22bfed0852bca..ad280289cebf9 100644
--- a/lld/ELF/SyntheticSections.cpp
+++ b/lld/ELF/SyntheticSections.cpp
@@ -613,7 +613,7 @@ uint64_t EhFrameSection::getFdePc(uint8_t *buf, size_t fdeOff,
   size_t off = fdeOff + 8;
   uint64_t addr = readFdeAddr(buf + off, enc & 0xf);
   if ((enc & 0x70) == DW_EH_PE_absptr)
-    return addr;
+    return config->is64 ? addr : uint32_t(addr);
   if ((enc & 0x70) == DW_EH_PE_pcrel)
     return addr + getParent()->addr + off + outSecOff;
   fatal("unknown FDE size relative encoding");

diff  --git a/lld/test/ELF/mips-eh_frame-pic.s b/lld/test/ELF/mips-eh_frame-pic.s
index 79076e74a7e3f..fd8560bc0163f 100644
--- a/lld/test/ELF/mips-eh_frame-pic.s
+++ b/lld/test/ELF/mips-eh_frame-pic.s
@@ -27,6 +27,11 @@
 ## relative addressing.
 # NOPIC32-ERR: ld.lld: error: relocation R_MIPS_32 cannot be used against local symbol
 
+## https://github.com/llvm/llvm-project/issues/88852: getFdePc should return a
+## 32-bit address.
+# RUN: ld.lld --eh-frame-hdr -Ttext=0x80000000 %t-nopic32.o -o %t-nopic32
+# RUN: llvm-readelf -x .eh_frame_hdr %t-nopic32 | FileCheck %s --check-prefix=NOPIC32-HDR
+
 ## For -fPIC, .eh_frame should contain DW_EH_PE_pcrel | DW_EH_PE_sdata4 values:
 # RUN: llvm-mc -filetype=obj -triple=mips-unknown-linux --position-independent %s -o %t-pic32.o
 # RUN: llvm-readobj -r %t-pic32.o | FileCheck %s --check-prefixes=RELOCS,PIC32-RELOCS
@@ -51,6 +56,10 @@
 ## Note: ld.bfd converts the R_MIPS_64 relocs to DW_EH_PE_pcrel | DW_EH_PE_sdata8
 ## for N64 ABI (and DW_EH_PE_pcrel | DW_EH_PE_sdata4 for MIPS32)
 
+# NOPIC32-HDR: Hex dump of section '.eh_frame_hdr':
+# NOPIC32-HDR: 0x80010038 011b033b 00000010 00000001 fffeffc8 .
+# NOPIC32-HDR: 0x80010048 00000028                            .
+
 .ent func
 .global func
 func:


        


More information about the llvm-commits mailing list