maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/lld/test/MachO/tlv.s

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

133 lines
4.9 KiB
ArmAsm
Raw Normal View History

[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
# REQUIRES: x86
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# RUN: rm -rf %t; split-file %s %t
# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/regular.s -o %t/regular.o
# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/tbss.s -o %t/tbss.o
[lld-macho] Don't emit rebase opcodes for relocs in TLV sections Their addresses are already encoded as section-relative offsets, so there's no need to rebase them at runtime. {D85080} has some context on the weirdness of TLV sections. Fixes llvm.org/PR48491. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D93257
2020-12-15 07:57:01 +08:00
[lld-macho][nfc] Remove unnecessary -pie flags in tests D101513 means that we no longer need to specify `-pie` in most of our test RUN commands. Let's clean up the unused flags so as not to confuse future test writers. Reviewed By: #lld-macho, oontvoo, MaskRay Differential Revision: https://reviews.llvm.org/D113114
2021-11-04 12:01:49 +08:00
# RUN: %lld -lSystem -no_pie -o %t/regular-no-pie %t/regular.o
# RUN: llvm-otool -hv %t/regular-no-pie | FileCheck %s --check-prefix=HEADER
# RUN: llvm-objdump -d --bind --rebase %t/regular-no-pie | FileCheck %s --check-prefixes=REG,LINKEDIT
# RUN: llvm-objdump --macho --section=__DATA,__thread_vars %t/regular-no-pie | \
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# RUN: FileCheck %s --check-prefix=REG-TLVP
[lld-macho] Don't emit rebase opcodes for relocs in TLV sections Their addresses are already encoded as section-relative offsets, so there's no need to rebase them at runtime. {D85080} has some context on the weirdness of TLV sections. Fixes llvm.org/PR48491. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D93257
2020-12-15 07:57:01 +08:00
[lld-macho][nfc] Remove unnecessary -pie flags in tests D101513 means that we no longer need to specify `-pie` in most of our test RUN commands. Let's clean up the unused flags so as not to confuse future test writers. Reviewed By: #lld-macho, oontvoo, MaskRay Differential Revision: https://reviews.llvm.org/D113114
2021-11-04 12:01:49 +08:00
# RUN: %lld -lSystem %t/regular.o -o %t/regular-pie
[lld/mac] Give __DATA,__thread_ptrs type S_THREAD_LOCAL_VARIABLE_POINTERS ...instead of S_NON_LAZY_SYMBOL_POINTERS. This matches ld64. Part of PR50769. While here, also remove an old TODO that was done in D87178. Differential Revision: https://reviews.llvm.org/D104594
2021-06-19 21:54:11 +08:00
# RUN: llvm-otool -hv %t/regular-pie | FileCheck %s --check-prefix=HEADER
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# RUN: llvm-objdump -d --bind --rebase %t/regular-pie | FileCheck %s --check-prefixes=REG,LINKEDIT
# RUN: llvm-objdump --macho --section=__DATA,__thread_vars %t/regular-pie | \
# RUN: FileCheck %s --check-prefix=REG-TLVP
# RUN: %lld -lSystem %t/tbss.o -o %t/tbss -e _f
# RUN: llvm-objdump -d --bind --rebase %t/tbss | FileCheck %s --check-prefixes=TBSS,LINKEDIT
# RUN: llvm-objdump --macho --section=__DATA,__thread_vars %t/tbss | \
# RUN: FileCheck %s --check-prefix=TBSS-TLVP
# RUN: %lld -lSystem %t/regular.o %t/tbss.o -o %t/regular-and-tbss
# RUN: llvm-objdump -d --bind --rebase %t/regular-and-tbss | FileCheck %s --check-prefixes=REG,TBSS,LINKEDIT
# RUN: llvm-objdump --macho --section=__DATA,__thread_vars %t/regular-and-tbss | \
# RUN: FileCheck %s --check-prefix=REG-TBSS-TLVP
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
# RUN: llvm-objdump --section-headers %t/regular-and-tbss | FileCheck %s --check-prefix=SECTIONS
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
## Check that we always put __thread_bss immediately after __thread_data,
## regardless of the order of the input files.
# RUN: %lld -lSystem %t/tbss.o %t/regular.o -o %t/regular-and-tbss
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
# RUN: llvm-objdump --section-headers %t/regular-and-tbss | FileCheck %s --check-prefix=SECTIONS
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
# HEADER: MH_HAS_TLV_DESCRIPTORS
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# REG: <_main>:
[llvm-objdump] Print memory operand addresses as regular comments The patch reuses the common code to print memory operand addresses as instruction comments. This helps to align the comments and enables using target-specific comment markers when `evaluateMemoryOperandAddress()` is implemented for them. Differential Revision: https://reviews.llvm.org/D104861
2021-06-28 15:24:39 +08:00
# REG-NEXT: leaq {{.*}}(%rip), %rax ## {{.*}} <_foo>
# REG-NEXT: leaq {{.*}}(%rip), %rax ## {{.*}} <_bar>
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# REG-NEXT: retq
# TBSS: <_f>:
[llvm-objdump] Print memory operand addresses as regular comments The patch reuses the common code to print memory operand addresses as instruction comments. This helps to align the comments and enables using target-specific comment markers when `evaluateMemoryOperandAddress()` is implemented for them. Differential Revision: https://reviews.llvm.org/D104861
2021-06-28 15:24:39 +08:00
# TBSS-NEXT: leaq {{.*}}(%rip), %rax ## {{.*}} <_baz>
# TBSS-NEXT: leaq {{.*}}(%rip), %rax ## {{.*}} <_qux>
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
# TBSS-NEXT: leaq {{.*}}(%rip), %rax ## {{.*}} <_hoge>
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# TBSS-NEXT: retq
# REG-TLVP: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# REG-TLVP-NEXT: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# REG-TLVP-NEXT: 00 00 00 00 00 00 00 00 08 00 00 00 00 00 00 00
# TBSS-TLVP: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 08 00 00 00 00 00 00 00
# REG-TBSS-TLVP: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# REG-TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
# REG-TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 10 00 00 00 00 00 00 00
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# REG-TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
# REG-TBSS-TLVP-NEXT: 20 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# REG-TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 28 00 00 00 00 00 00 00
# REG-TBSS-TLVP-NEXT: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
# REG-TBSS-TLVP-NEXT: 30 00 00 00 00 00 00 00
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
[lld-macho] Don't emit rebase opcodes for relocs in TLV sections Their addresses are already encoded as section-relative offsets, so there's no need to rebase them at runtime. {D85080} has some context on the weirdness of TLV sections. Fixes llvm.org/PR48491. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D93257
2020-12-15 07:57:01 +08:00
## Make sure we don't emit rebase opcodes for relocations in __thread_vars.
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# LINKEDIT: Rebase table:
# LINKEDIT-NEXT: segment section address type
[llvm-objdump][MachO] Print a newline before lazy bind/bind/weak/exports trie This adds a separator between two pieces of information. Reviewed By: #lld-macho, alexshap Differential Revision: https://reviews.llvm.org/D102114
2021-05-11 00:16:18 +08:00
# LINKEDIT-EMPTY:
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
# LINKEDIT-NEXT: Bind table:
# LINKEDIT: __DATA __thread_vars 0x{{[0-9a-f]*}} pointer 0 libSystem __tlv_bootstrap
# LINKEDIT: __DATA __thread_vars 0x{{[0-9a-f]*}} pointer 0 libSystem __tlv_bootstrap
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
## Make sure we have an odd number of tlv vars, and that the __thread_vars
## section starts 16-bytes aligned. This is the setup required for __thread_data
## not to be automatically 16-bytes aligned, ensuring the linker does its
## expected job of aligning _hoge$tlv$init.
# SECTIONS: __thread_vars {{[0-9]+}}8 {{[0-9]+}}0
# SECTIONS: __thread_data
# SECTIONS: more_thread_data
# SECTIONS-NEXT: __thread_bss
[lld-macho] Don't emit rebase opcodes for relocs in TLV sections Their addresses are already encoded as section-relative offsets, so there's no need to rebase them at runtime. {D85080} has some context on the weirdness of TLV sections. Fixes llvm.org/PR48491. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D93257
2020-12-15 07:57:01 +08:00
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
#--- regular.s
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
.globl _main
_main:
mov _foo@TLVP(%rip), %rax
mov _bar@TLVP(%rip), %rax
ret
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.section __DATA,__thread_data,thread_local_regular
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
_foo$tlv$init:
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
.quad 123
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.section __DATA,more_thread_data,thread_local_regular
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
_bar$tlv$init:
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
.quad 123
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.section __DATA,__thread_vars,thread_local_variables
.globl _foo, _bar
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
_foo:
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.quad __tlv_bootstrap
.quad 0
.quad _foo$tlv$init
[lld-macho] Support static linking of thread-locals Note: What ELF refers to as "TLS", Mach-O seems to refer to as "TLV", i.e. thread-local variables. This diff implements support for TLV relocations that reference defined symbols. On x86_64, TLV relocations are always used with movq opcodes, so for defined TLVs, we don't need to create a synthetic section to store the addresses of the symbols -- we can just convert the `movq` to a `leaq`. One notable quirk of Mach-O's TLVs is that absolute-address relocations inside TLV-defining sections behave differently -- their addresses are no longer absolute, but relative to the start of the target section. (AFAICT, RIP-relative relocations are not allowed in these sections.) Reviewed By: #lld-macho, compnerd, smeenai Differential Revision: https://reviews.llvm.org/D85080
2020-08-08 02:04:52 +08:00
_bar:
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.quad __tlv_bootstrap
.quad 0
.quad _bar$tlv$init
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
#--- tbss.s
.globl _f
_f:
mov _baz@TLVP(%rip), %rax
mov _qux@TLVP(%rip), %rax
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
mov _hoge@TLVP(%rip), %rax
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
ret
.tbss _baz$tlv$init, 8, 3
.tbss _qux$tlv$init, 8, 3
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
.tbss _hoge$tlv$init, 16, 4
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.section __DATA,__thread_vars,thread_local_variables
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
_baz:
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.quad __tlv_bootstrap
.quad 0
.quad _baz$tlv$init
[lld-macho] Fix TLV data initialization We were mishandling the case where both `__tbss` and `__thread_data` sections were present. TLVP relocations should be encoded as offsets from the start of `__thread_data`, even if the symbol is actually located in `__thread_bss`. Previously, we were writing the offset from the start of the containing section, which doesn't really make sense since there's no way `tlv_get_addr()` can know which section a given `tlv$init` symbol is in at runtime. In addition, this patch ensures that we place `__thread_data` immediately before `__thread_bss`. This is what ld64 does, likely for performance reasons. Zerofill sections must also be at the end of their segments; we were already doing this, but now we ensure that `__thread_bss` occurs before `__bss`, so that it's always possible to have it contiguous with `__thread_data`. Fixes llvm.org/PR48657. Reviewed By: #lld-macho, thakis Differential Revision: https://reviews.llvm.org/D94329
2021-01-09 07:47:40 +08:00
_qux:
[lld-macho][nfc] Convert tabs to spaces
2021-04-12 11:23:37 +08:00
.quad __tlv_bootstrap
.quad 0
.quad _qux$tlv$init
[lld-macho] Fix alignment of TLV data sections References from thread-local variable sections are treated as offsets relative to the start of the thread-local data memory area, which is initialized via copying all the TLV data sections (which are all contiguous). If later data sections require a greater alignment than earlier ones, the offsets of data within those sections won't be guaranteed to aligned unless we normalize alignments. We therefore use the largest alignment for all TLV data sections. Reviewed By: #lld-macho, int3 Differential Revision: https://reviews.llvm.org/D116263
2021-12-29 08:01:01 +08:00
_hoge:
.quad __tlv_bootstrap
.quad 0
.quad _hoge$tlv$init