[RISCV] Specify various encodings for DWARF exception handling

[edward-jones]

Summary

This specifies the encodings used for the personality function, LSDA and type info for exception handling.

[edward-jones]

Added tests, rebased against master

[HsiangKai]

I think D63409, D63411, D63415, and D63416 could be merged as one patch. They are all related to exception handling for RISC-V.

[edward-jones]

In D63409#1556886, @HsiangKai wrote:

I think D63409, D63411, D63415, and D63416 could be merged as one patch. They are all related to exception handling for RISC-V.

I kept them as separate changes since they're incremental and I thought it would make things easier to review, but I can squash them into one patch if that is more convenient. At the very least I plan on keeping D63415 separate as it includes generic changes.

[HsiangKai]

In D63409#1557447, @edward-jones wrote:

In D63409#1556886, @HsiangKai wrote:

I think D63409, D63411, D63415, and D63416 could be merged as one patch. They are all related to exception handling for RISC-V.

I kept them as separate changes since they're incremental and I thought it would make things easier to review, but I can squash them into one patch if that is more convenient. At the very least I plan on keeping D63415 separate as it includes generic changes.

I found that the test case in this patch will trigger an assertion failed without D63411 and there is no test case for D63411. So, I suggest these two patches could be combined at least. It will make review and verification easier.

If you want to keep D63415 as a stand alone patch, you should create a test case for it. I think D63415 and D63416 could be combined. You could think D63416 is a show case about how to specify “call site encoding” in architecture dependent part and why you want to do these generic changes. It is only my opinion. For your reference. BTW, your modification looks good.

[asb]

Hi Ed, this is really hard to review without any documentation in the psABI or otherwise. Can you confirm that you've verified this matches gcc/gas? Are you planning to submit document to riscv-elf-psabi-doc to reflect this?

[edward-jones]

In D63409#1587117, @asb wrote:

Hi Ed, this is really hard to review without any documentation in the psABI or otherwise. Can you confirm that you've verified this matches gcc/gas? Are you planning to submit document to riscv-elf-psabi-doc to reflect this?

I've opened a pull request on the psABI docs (https://github.com/riscv/riscv-elf-psabi-doc/pull/110). I'm not entirely sure the details here are relevant to the ABI though. As far as I'm aware provided the individual CIE and FDEs are well-formed it shouldn't matter if the encodings differ from those specified in other CIE and FDEs.

On another note, after taking a look at the GCC code for this I'm not convinced the non-PIC values for the encodings match. GCC appears to just encode everything as pc-relative sdata4, whereas the above encodes things as absolute when PIC is not specified.

[asb]

In D63409#1587351, @edward-jones wrote:

In D63409#1587117, @asb wrote:

Hi Ed, this is really hard to review without any documentation in the psABI or otherwise. Can you confirm that you've verified this matches gcc/gas? Are you planning to submit document to riscv-elf-psabi-doc to reflect this?

I've opened a pull request on the psABI docs (https://github.com/riscv/riscv-elf-psabi-doc/pull/110). I'm not entirely sure the details here are relevant to the ABI though. As far as I'm aware provided the individual CIE and FDEs are well-formed it shouldn't matter if the encodings differ from those specified in other CIE and FDEs.

On another note, after taking a look at the GCC code for this I'm not convinced the non-PIC values for the encodings match. GCC appears to just encode everything as pc-relative sdata4, whereas the above encodes things as absolute when PIC is not specified.

Yes, I was trying to cross-reference to what GCC does and couldn't see a match for everything you specify here.

Many thanks for submitting that PR, as you say it's not clear where it belongs but it's certainly useful to have it written down somewhere!

[asb]

Resigning as a reviewer, as the intent of this change is captured in rL366327.

[edward-jones]

Abandoned since this was superceded by rL366327.

lib/CodeGen/TargetLoweringObjectFileImpl.cpp

void TargetLoweringObjectFileELF::Initialize(MCContext &Ctx,
   case Triple::ppc64:
   case Triple::ppc64le:
     PersonalityEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
       dwarf::DW_EH_PE_udata8;
     LSDAEncoding = dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_udata8;
     TTypeEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
       dwarf::DW_EH_PE_udata8;
     break;
+  case Triple::riscv32:
+  case Triple::riscv64:
+    assert((CM == CodeModel::Small || CM == CodeModel::Medium)
+           && "unhandled code model");
+    if (isPositionIndependent()) {
+      PersonalityEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
+        dwarf::DW_EH_PE_sdata4;
+      LSDAEncoding = dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4;
+      TTypeEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
+        dwarf::DW_EH_PE_sdata4;
+    } else if (CM == CodeModel::Small) {
+      PersonalityEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_udata4;
+      LSDAEncoding = dwarf::DW_EH_PE_udata4;
+      TTypeEncoding = dwarf::DW_EH_PE_udata4;
+    } else {
+      assert(CM == CodeModel::Medium);
+      PersonalityEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_absptr;
+      LSDAEncoding = dwarf::DW_EH_PE_absptr;
+      TTypeEncoding = dwarf::DW_EH_PE_absptr;
+    }
+    break;
   case Triple::sparcel:
   case Triple::sparc:
     if (isPositionIndependent()) {
       LSDAEncoding = dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4;
       PersonalityEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
         dwarf::DW_EH_PE_sdata4;
       TTypeEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
         dwarf::DW_EH_PE_sdata4;

test/CodeGen/RISCV/exception.ll

+; RUN: llc -march=riscv32 --code-model=small  < %s | FileCheck --check-prefixes=CHECK,SMALL %s
+; RUN: llc -march=riscv32 --code-model=medium < %s | FileCheck --check-prefixes=CHECK,MED   %s
+; RUN: llc -march=riscv32 --code-model=small  -relocation-model=pic < %s | FileCheck --check-prefixes=CHECK,PIC %s
+; RUN: llc -march=riscv32 --code-model=medium -relocation-model=pic < %s | FileCheck --check-prefixes=CHECK,PIC %s
+
+; RUN: llc -march=riscv64 --code-model=small  < %s | FileCheck --check-prefixes=CHECK,SMALL %s
+; RUN: llc -march=riscv64 --code-model=medium < %s | FileCheck --check-prefixes=CHECK,MED   %s
+; RUN: llc -march=riscv64 --code-model=small  -relocation-model=pic < %s | FileCheck --check-prefixes=CHECK,PIC %s
+; RUN: llc -march=riscv64 --code-model=medium -relocation-model=pic < %s | FileCheck --check-prefixes=CHECK,PIC %s
+
+declare i8* @__cxa_allocate_exception(i64)
+declare void @__cxa_throw(i8*, i8*, i8*)
+declare i32 @__gxx_personality_v0(...)
+declare i32 @llvm.eh.typeid.for(i8*) nounwind readnone
+declare i8* @__cxa_begin_catch(i8*)
+declare void @__cxa_end_catch()
+
+@_ZTIi = external constant i8*
+
+; CHECK-LABEL: foo:
+; CHECK:      .cfi_startproc
+;
+;   # Personality encoding #
+; SMALL-NEXT: .cfi_personality 131, DW.ref.__gxx_personality_v0
+;                               ^-- Personality encoding: DW_EH_PE_indirect |
+;                                                         DW_EH_PE_udata4
+; MED-NEXT:   .cfi_personality 128, DW.ref.__gxx_personality_v0
+;                               ^-- Personality encoding: DW_EH_PE_indirect |
+;                                                         DW_EH_PE_absptr
+; PIC-NEXT:   .cfi_personality 155, DW.ref.__gxx_personality_v0
+;                               ^-- Personality encoding: DW_EH_PE_indirect |
+;                                                         DW_EH_PE_pcrel |
+;                                                         DW_EH_PE_sdata4
+;   # LSDA encoding #
+; SMALL-NEXT: .cfi_lsda 3, .Lexception0
+;                       ^-- LSDA encoding: DW_EH_PE_udata4
+; MED-NEXT:   .cfi_lsda 0, .Lexception0
+;                       ^-- LSDA encoding: DW_EH_PE_absptr
+; PIC-NEXT:   .cfi_lsda 27, .Lexception0
+;                       ^-- LSDA encoding: DW_EH_PE_pcrel | DW_EH_PE_sdata4
+define signext i32 @foo() personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
+entry:
+  %exception = tail call i8* @__cxa_allocate_exception(i64 4)
+  %0 = bitcast i8* %exception to i32*
+  store i32 5, i32* %0, align 16
+  invoke void @__cxa_throw(i8* %exception, i8* bitcast (i8** @_ZTIi to i8*), i8* null)
+          to label %unreachable unwind label %lpad
+
+lpad:
+  %1 = landingpad { i8*, i32 }
+          catch i8* bitcast (i8** @_ZTIi to i8*)
+  %2 = extractvalue { i8*, i32 } %1, 1
+  %3 = tail call i32 @llvm.eh.typeid.for(i8* bitcast (i8** @_ZTIi to i8*))
+  %matches = icmp eq i32 %2, %3
+  br i1 %matches, label %catch, label %eh.resume
+
+catch:
+  %4 = extractvalue { i8*, i32 } %1, 0
+  %5 = tail call i8* @__cxa_begin_catch(i8* %4)
+  %6 = bitcast i8* %5 to i32*
+  %7 = load i32, i32* %6, align 4
+  tail call void @__cxa_end_catch()
+  ret i32 %7
+
+eh.resume:
+  resume { i8*, i32 } %1
+
+unreachable:
+  unreachable
+}
+
+; CHECK:      .section .gcc_except_table,"a",@progbits
+; CHECK:      GCC_except_table0:
+; CHECK-NEXT: .Lexception0:
+; CHECK-NEXT: .byte  255
+;                     ^-- Landing pad base encoding: DW_EH_PE_omit
+;
+;   # Type table encoding #
+; SMALL-NEXT: .byte  3
+;                    ^-- Type table encoding: DW_EH_PE_udata4
+; MED-NEXT:   .byte  0
+;                    ^-- Type table encoding: DW_EH_PE_absptr
+; PIC-NEXT:   .byte  155
+;                    ^-- Type table encoding: DW_EH_PE_indirect |
+;                                             DW_EH_PE_pcrel |
+;                                             DW_EH_PE_sdata4
+;   # Call site encoding #
+; CHECK:      .Lttbaseref0:
+; CHECK-NEXT: .byte  1
+;                    ^-- Call site encoding: DW_EH_PE_uleb128

test/MC/RISCV/eh-frame.s

+# Test the bits of .eh_frame on riscv that are already implemented correctly.
+
+# RUN: llvm-mc -filetype=obj %s -o %t.o -triple riscv32
+# RUN: llvm-dwarfdump -eh-frame %t.o | FileCheck --check-prefixes=CHECK,RV32 %s
+
+# RUN: llvm-mc -filetype=obj %s -o %t.o -triple riscv64
+# RUN: llvm-dwarfdump -eh-frame %t.o | FileCheck --check-prefixes=CHECK,RV64 %s
+
+func:
+  .cfi_startproc
+  .cfi_endproc
+
+# CHECK: 00000000 00000010 ffffffff CIE
+# CHECK:   Version:               1
+# CHECK:   Augmentation:          "zR"
+# CHECK:   Code alignment factor: 1
+
+# RV32:    Data alignment factor: -4
+# RV64:    Data alignment factor: -8
+#                                 ^-- GAS uses -4. Should be okay provided all
+#                                     offsets are multiples of 8
+
+# CHECK:   Return address column: 1
+# CHECK:   Augmentation data:     1B
+#                                 ^-- FDE encoding: DW_EH_PE_pcrel |
+#                                                   DW_EH_PE_sdata4
+# CHECK:   DW_CFA_def_cfa: reg2 +0
+#
+# CHECK: 00000014 00000010 00000018 FDE cie=00000018 pc=00000000...00000000
+# CHECK:   DW_CFA_nop:
+# CHECK:   DW_CFA_nop:
+# CHECK:   DW_CFA_nop: