[lld] [lld] Add target support for SystemZ (s390x) (PR #75643)
Fangrui Song via llvm-commits
llvm-commits at lists.llvm.org
Sat Feb 10 22:01:56 PST 2024
================
@@ -0,0 +1,587 @@
+//===- SystemZ.cpp --------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "OutputSections.h"
+#include "Symbols.h"
+#include "SyntheticSections.h"
+#include "Target.h"
+#include "lld/Common/ErrorHandler.h"
+#include "llvm/BinaryFormat/ELF.h"
+#include "llvm/Support/Endian.h"
+
+using namespace llvm;
+using namespace llvm::object;
+using namespace llvm::support::endian;
+using namespace llvm::ELF;
+using namespace lld;
+using namespace lld::elf;
+
+namespace {
+class SystemZ : public TargetInfo {
+public:
+ SystemZ();
+ RelExpr getRelExpr(RelType type, const Symbol &s,
+ const uint8_t *loc) const override;
+ RelType getDynRel(RelType type) const override;
+ void writeGotHeader(uint8_t *buf) const override;
+ void writeGotPlt(uint8_t *buf, const Symbol &s) const override;
+ void writeIgotPlt(uint8_t *buf, const Symbol &s) const override;
+ void writePltHeader(uint8_t *buf) const override;
+ void addPltHeaderSymbols(InputSection &isd) const override;
+ void writePlt(uint8_t *buf, const Symbol &sym,
+ uint64_t pltEntryAddr) const override;
+ RelExpr adjustTlsExpr(RelType type, RelExpr expr) const override;
+ RelExpr adjustGotPcExpr(RelType type, int64_t addend,
+ const uint8_t *loc) const override;
+ bool relaxOnce(int pass) const override;
+ void relocate(uint8_t *loc, const Relocation &rel,
+ uint64_t val) const override;
+ int64_t getImplicitAddend(const uint8_t *buf, RelType type) const override;
+
+private:
+ void relaxGot(uint8_t *loc, const Relocation &rel, uint64_t val) const;
+ void relaxTlsGdToIe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
+ void relaxTlsGdToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
+ void relaxTlsLdToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
+};
+} // namespace
+
+SystemZ::SystemZ() {
+ copyRel = R_390_COPY;
+ gotRel = R_390_GLOB_DAT;
+ pltRel = R_390_JMP_SLOT;
+ relativeRel = R_390_RELATIVE;
+ iRelativeRel = R_390_IRELATIVE;
+ symbolicRel = R_390_64;
+ tlsGotRel = R_390_TLS_TPOFF;
+ tlsModuleIndexRel = R_390_TLS_DTPMOD;
+ tlsOffsetRel = R_390_TLS_DTPOFF;
+ gotHeaderEntriesNum = 3;
+ gotPltHeaderEntriesNum = 0;
+ gotEntrySize = 8;
+ pltHeaderSize = 32;
+ pltEntrySize = 32;
+ ipltEntrySize = 32;
+
+ // This "trap instruction" is used to fill gaps between sections.
+ // On SystemZ, the behavior of the GNU ld is to fill those gaps
+ // with nop instructions instead - and unfortunately the default
+ // glibc crt object files (used to) rely on that behavior since
+ // they use an alignment on the .init section fragments that causes
+ // gaps which must be filled with nops as they are being executed.
+ // Therefore, we provide a nop instruction as "trapInstr" here.
+ trapInstr = {0x07, 0x07, 0x07, 0x07};
+
+ defaultImageBase = 0x1000000;
+}
+
+RelExpr SystemZ::getRelExpr(RelType type, const Symbol &s,
+ const uint8_t *loc) const {
+ switch (type) {
+ case R_390_NONE:
+ return R_NONE;
+ // Relocations targeting the symbol value.
+ case R_390_8:
+ case R_390_12:
+ case R_390_16:
+ case R_390_20:
+ case R_390_32:
+ case R_390_64:
+ return R_ABS;
+ case R_390_PC16:
+ case R_390_PC32:
+ case R_390_PC64:
+ case R_390_PC12DBL:
+ case R_390_PC16DBL:
+ case R_390_PC24DBL:
+ case R_390_PC32DBL:
+ return R_PC;
+ case R_390_GOTOFF16:
+ case R_390_GOTOFF: // a.k.a. R_390_GOTOFF32
+ case R_390_GOTOFF64:
+ return R_GOTREL;
+ // Relocations targeting the PLT associated with the symbol.
+ case R_390_PLT32:
+ case R_390_PLT64:
+ case R_390_PLT12DBL:
+ case R_390_PLT16DBL:
+ case R_390_PLT24DBL:
+ case R_390_PLT32DBL:
+ return R_PLT_PC;
+ case R_390_PLTOFF16:
+ case R_390_PLTOFF32:
+ case R_390_PLTOFF64:
+ return R_PLT_GOTREL;
+ // Relocations targeting the GOT entry associated with the symbol.
+ case R_390_GOTENT:
+ return R_GOT_PC;
+ case R_390_GOT12:
+ case R_390_GOT16:
+ case R_390_GOT20:
+ case R_390_GOT32:
+ case R_390_GOT64:
+ return R_GOT_OFF;
+ // Relocations targeting the GOTPLT entry associated with the symbol.
+ case R_390_GOTPLTENT:
+ return R_GOTPLT_PC;
+ case R_390_GOTPLT12:
+ case R_390_GOTPLT16:
+ case R_390_GOTPLT20:
+ case R_390_GOTPLT32:
+ case R_390_GOTPLT64:
+ return R_GOTPLT_OFF;
+ // Relocations targeting _GLOBAL_OFFSET_TABLE_.
+ case R_390_GOTPC:
+ case R_390_GOTPCDBL:
+ return R_GOTONLY_PC;
+ // TLS-related relocations.
+ case R_390_TLS_LOAD:
+ return R_NONE;
+ case R_390_TLS_GDCALL:
+ return R_TLSGD_PC;
+ case R_390_TLS_LDCALL:
+ return R_TLSLD_PC;
+ case R_390_TLS_GD32:
+ case R_390_TLS_GD64:
+ return R_TLSGD_GOT;
+ case R_390_TLS_LDM32:
+ case R_390_TLS_LDM64:
+ return R_TLSLD_GOT;
+ case R_390_TLS_LDO32:
+ case R_390_TLS_LDO64:
+ return R_DTPREL;
+ case R_390_TLS_LE32:
+ case R_390_TLS_LE64:
+ return R_TPREL;
+ case R_390_TLS_IE32:
+ case R_390_TLS_IE64:
+ return R_GOT;
+ case R_390_TLS_GOTIE12:
+ case R_390_TLS_GOTIE20:
+ case R_390_TLS_GOTIE32:
+ case R_390_TLS_GOTIE64:
+ return R_GOT_OFF;
+ case R_390_TLS_IEENT:
+ return R_GOT_PC;
+
+ default:
+ error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) +
+ ") against symbol " + toString(s));
+ return R_NONE;
+ }
+}
+
+void SystemZ::writeGotHeader(uint8_t *buf) const {
+ // _GLOBAL_OFFSET_TABLE_[0] holds the value of _DYNAMIC.
+ // _GLOBAL_OFFSET_TABLE_[1] and [2] are reserved.
+ write64be(buf, mainPart->dynamic->getVA());
+}
+
+void SystemZ::writeGotPlt(uint8_t *buf, const Symbol &s) const {
+ write64be(buf, s.getPltVA() + 14);
+}
+
+void SystemZ::writeIgotPlt(uint8_t *buf, const Symbol &s) const {
+ if (config->writeAddends)
+ write64be(buf, s.getVA());
+}
+
+void SystemZ::writePltHeader(uint8_t *buf) const {
+ const uint8_t pltData[] = {
+ 0xe3, 0x10, 0xf0, 0x38, 0x00, 0x24, // stg %r1,56(%r15)
+ 0xc0, 0x10, 0x00, 0x00, 0x00, 0x00, // larl %r1,_GLOBAL_OFFSET_TABLE_
+ 0xd2, 0x07, 0xf0, 0x30, 0x10, 0x08, // mvc 48(8,%r15),8(%r1)
+ 0xe3, 0x10, 0x10, 0x10, 0x00, 0x04, // lg %r1,16(%r1)
+ 0x07, 0xf1, // br %r1
+ 0x07, 0x00, // nopr
+ 0x07, 0x00, // nopr
+ 0x07, 0x00, // nopr
+ };
+ memcpy(buf, pltData, sizeof(pltData));
+ uint64_t got = in.got->getVA();
+ uint64_t plt = in.plt->getVA();
+ write32be(buf + 8, (got - plt - 6) >> 1);
+}
+
+void SystemZ::addPltHeaderSymbols(InputSection &isec) const {
+ // The PLT header needs a reference to _GLOBAL_OFFSET_TABLE_, so we
+ // must ensure the .got section is created even if otherwise unused.
+ in.got->hasGotOffRel.store(true, std::memory_order_relaxed);
+}
+
+void SystemZ::writePlt(uint8_t *buf, const Symbol &sym,
+ uint64_t pltEntryAddr) const {
+ const uint8_t inst[] = {
+ 0xc0, 0x10, 0x00, 0x00, 0x00, 0x00, // larl %r1,<.got.plt slot>
+ 0xe3, 0x10, 0x10, 0x00, 0x00, 0x04, // lg %r1,0(%r1)
+ 0x07, 0xf1, // br %r1
+ 0x0d, 0x10, // basr %r1,%r0
+ 0xe3, 0x10, 0x10, 0x0c, 0x00, 0x14, // lgf %r1,12(%r1)
+ 0xc0, 0xf4, 0x00, 0x00, 0x00, 0x00, // jg <plt header>
+ 0x00, 0x00, 0x00, 0x00, // <relocation offset>
+ };
+ memcpy(buf, inst, sizeof(inst));
+
+ write32be(buf + 2, (sym.getGotPltVA() - pltEntryAddr) >> 1);
+ write32be(buf + 24, (in.plt->getVA() - pltEntryAddr - 22) >> 1);
+ write32be(buf + 28, in.relaPlt->entsize * sym.getPltIdx());
+}
+
+int64_t SystemZ::getImplicitAddend(const uint8_t *buf, RelType type) const {
+ switch (type) {
+ case R_390_8:
+ return SignExtend64<8>(*buf);
+ case R_390_16:
+ case R_390_PC16:
+ return SignExtend64<16>(read16be(buf));
+ case R_390_PC16DBL:
+ return SignExtend64<16>(read16be(buf)) << 1;
+ case R_390_32:
+ case R_390_PC32:
+ return SignExtend64<32>(read32be(buf));
+ case R_390_PC32DBL:
+ return SignExtend64<32>(read32be(buf)) << 1;
+ case R_390_64:
+ case R_390_PC64:
+ case R_390_TLS_DTPMOD:
+ case R_390_TLS_DTPOFF:
+ case R_390_TLS_TPOFF:
+ case R_390_GLOB_DAT:
+ case R_390_RELATIVE:
+ case R_390_IRELATIVE:
+ return read64be(buf);
+ case R_390_COPY:
+ case R_390_JMP_SLOT:
+ case R_390_NONE:
+ // These relocations are defined as not having an implicit addend.
+ return 0;
+ default:
+ internalLinkerError(getErrorLocation(buf),
+ "cannot read addend for relocation " + toString(type));
+ return 0;
+ }
+}
+
+RelType SystemZ::getDynRel(RelType type) const {
+ if (type == R_390_64 || type == R_390_PC64)
+ return type;
+ return R_390_NONE;
+}
+
+RelExpr SystemZ::adjustTlsExpr(RelType type, RelExpr expr) const {
+ if (expr == R_RELAX_TLS_GD_TO_IE)
+ return R_RELAX_TLS_GD_TO_IE_GOT_OFF;
+ return expr;
+}
+
+void SystemZ::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
+ uint64_t val) const {
+ // The general-dynamic code sequence for a global `x`:
+ //
+ // Instruction Relocation Symbol
+ // ear %rX,%a0
+ // sllg %rX,%rX,32
+ // ear %rX,%a1
+ // larl %r12,_GLOBAL_OFFSET_TABLE_ R_390_GOTPCDBL _GLOBAL_OFFSET_TABLE_
+ // lgrl %r2,.LC0 R_390_PC32DBL .LC0
+ // brasl %r14,__tls_get_offset at plt R_390_TLS_GDCALL x
+ // :tls_gdcall:x R_390_PLT32DBL __tls_get_offset
+ // la %r2,0(%r2,%rX)
+ //
+ // .LC0:
+ // .quad x at TLSGD R_390_TLS_GD64 x
+ //
+ // Relaxing to initial-exec entails:
+ // 1) Replacing the call by a load from the GOT.
+ // 2) Replacing the relocation on the constant LC0 by R_390_TLS_GOTIE64.
+
+ switch (rel.type) {
+ case R_390_TLS_GDCALL:
+ /* brasl %r14,__tls_get_offset at plt -> lg %r2,0(%r2,%r12) */
+ write16be(loc, 0xe322);
+ write16be(loc + 2, 0xc000);
+ write16be(loc + 4, 0x0004);
+ break;
+ case R_390_TLS_GD64:
+ relocateNoSym(loc, R_390_TLS_GOTIE64, val);
+ break;
+ default:
+ llvm_unreachable("unsupported relocation for TLS GD to IE relaxation");
+ }
+}
+
+void SystemZ::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
+ uint64_t val) const {
+ // The general-dynamic code sequence for a global `x`:
+ //
+ // Instruction Relocation Symbol
+ // ear %rX,%a0
+ // sllg %rX,%rX,32
+ // ear %rX,%a1
+ // larl %r12,_GLOBAL_OFFSET_TABLE_ R_390_GOTPCDBL _GLOBAL_OFFSET_TABLE_
+ // lgrl %r2,.LC0 R_390_PC32DBL .LC0
+ // brasl %r14,__tls_get_offset at plt R_390_TLS_GDCALL x
+ // :tls_gdcall:x R_390_PLT32DBL __tls_get_offset
+ // la %r2,0(%r2,%rX)
+ //
+ // .LC0:
+ // .quad x at tlsgd R_390_TLS_GD64 x
+ //
+ // Relaxing to local-exec entails:
+ // 1) Replacing the call by a nop.
+ // 2) Replacing the relocation on the constant LC0 by R_390_TLS_LE64.
+
+ switch (rel.type) {
+ case R_390_TLS_GDCALL:
+ /* brasl %r14,__tls_get_offset at plt -> brcl 0,. */
+ write16be(loc, 0xc004);
----------------
MaskRay wrote:
`//`
Use one write32be to replace two write16be?
https://github.com/llvm/llvm-project/pull/75643
More information about the llvm-commits
mailing list