PING: [PATCH v8 0/2] x86: Update _dl_tlsdesc_dynamic to preserve caller-saved registers
H.J. Lu
hjl.tools@gmail.com
Thu Feb 22 20:24:01 GMT 2024
On Fri, Feb 16, 2024 at 7:17 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> Changes in v8:
>
> 1. Remove malloc-for-test.c and move malloc to tst-gnu2-tls2.c.
> 2. Add malloc_counter to verify malloc in tst-gnu2-tls2.c is called for
> TLSDESC call.
> 3. Add BEFORE_TLSDESC_CALL and AFTER_TLSDESC_CALL.
> 4. Use /* ... */ in assembly code comments.
>
> Changes in v7:
>
> 1. Generate malloc-for-test.map at build time to get the correct version
> map for malloc.
>
> Changes in v6:
>
> 1. Drop Tile registers.
>
> Changes in v5:
>
> 1. Also preserve Tile registers.
> 2. Add an error check in i386 dl-tlsdesc-dynamic.h.
>
> Changes in v4:
>
> 1. Add APX registers to STATE_SAVE_MASK so that APX registers are saved
> in ld.so trampoline.
> 2. Also save x87 FPU stack registers for TLSDESC_CALL and TLS_DESC_CALL.
> 3. Change i386 _dl_tlsdesc_dynamic to IFUNC.
> 4. Rename GLRO(dl_x86_64_tlsdesc_dynamic) to GLRO(dl_x86_tlsdesc_dynamic)
> for both i386 and x86-64.
> 5. Update the testcase for i386 with a simple malloc interceptor.
>
> Changes in v3:
>
> 1. Don't add GLRO(dl_x86_64_tlsdesc_dynamic) to libc.a.
>
> Changes in v2:
>
> 1. Add GLRO(dl_x86_64_runtime_resolve) to optimize
> elf_machine_runtime_setup.
> ---
> Add APX registers to STATE_SAVE_MASK so that APX registers are saved in
> ld.so trampoline. This fixes BZ #31371.
>
> Compiler generates the following instruction sequence for GNU2 dynamic
> TLS access:
>
> leaq tls_var@TLSDESC(%rip), %rax
> call *tls_var@TLSCALL(%rax)
>
> or
>
> leal tls_var@TLSDESC(%ebx), %eax
> call *tls_var@TLSCALL(%eax)
>
> CALL instruction is transparent to compiler which assumes all registers,
> except for EFLAGS and RAX/EAX, are unchanged after CALL. When
> _dl_tlsdesc_dynamic is called, it calls __tls_get_addr on the slow
> path. __tls_get_addr is a normal function which doesn't preserve any
> caller-saved registers. _dl_tlsdesc_dynamic saved and restored integer
> caller-saved registers, but didn't preserve any other caller-saved
> registers. Add _dl_tlsdesc_dynamic IFUNC functions for FNSAVE, FXSAVE,
> XSAVE and XSAVEC to save and restore all caller-saved registers. This
> fixes BZ #31372.
>
> Add GLRO(dl_x86_64_runtime_resolve) with GLRO(dl_x86_tlsdesc_dynamic)
> to optimize elf_machine_runtime_setup.
>
> H.J. Lu (2):
> x86-64: Save APX registers in ld.so trampoline
> x86: Update _dl_tlsdesc_dynamic to preserve caller-saved registers
>
> elf/Makefile | 14 ++
> elf/tst-gnu2-tls2.c | 120 ++++++++++++
> elf/tst-gnu2-tls2.h | 36 ++++
> elf/tst-gnu2-tls2mod0.c | 31 +++
> elf/tst-gnu2-tls2mod1.c | 31 +++
> elf/tst-gnu2-tls2mod2.c | 31 +++
> sysdeps/i386/dl-machine.h | 2 +-
> sysdeps/i386/dl-tlsdesc-dynamic.h | 190 +++++++++++++++++++
> sysdeps/i386/dl-tlsdesc.S | 115 +++++------
> sysdeps/x86/Makefile | 7 +-
> sysdeps/x86/cpu-features.c | 56 +++++-
> sysdeps/x86/dl-procinfo.c | 16 ++
> sysdeps/{x86_64 => x86}/features-offsets.sym | 2 +
> sysdeps/x86/sysdep.h | 56 +++++-
> sysdeps/x86/tst-gnu2-tls2.c | 20 ++
> sysdeps/x86_64/Makefile | 2 +-
> sysdeps/x86_64/dl-machine.h | 19 +-
> sysdeps/x86_64/dl-procinfo.c | 16 ++
> sysdeps/x86_64/dl-tlsdesc-dynamic.h | 166 ++++++++++++++++
> sysdeps/x86_64/dl-tlsdesc.S | 108 ++++-------
> sysdeps/x86_64/dl-trampoline-save.h | 34 ++++
> sysdeps/x86_64/dl-trampoline-state.h | 51 +++++
> sysdeps/x86_64/dl-trampoline.S | 20 +-
> sysdeps/x86_64/dl-trampoline.h | 34 +---
> 24 files changed, 959 insertions(+), 218 deletions(-)
> create mode 100644 elf/tst-gnu2-tls2.c
> create mode 100644 elf/tst-gnu2-tls2.h
> create mode 100644 elf/tst-gnu2-tls2mod0.c
> create mode 100644 elf/tst-gnu2-tls2mod1.c
> create mode 100644 elf/tst-gnu2-tls2mod2.c
> create mode 100644 sysdeps/i386/dl-tlsdesc-dynamic.h
> rename sysdeps/{x86_64 => x86}/features-offsets.sym (89%)
> create mode 100644 sysdeps/x86/tst-gnu2-tls2.c
> create mode 100644 sysdeps/x86_64/dl-tlsdesc-dynamic.h
> create mode 100644 sysdeps/x86_64/dl-trampoline-save.h
> create mode 100644 sysdeps/x86_64/dl-trampoline-state.h
>
> --
> 2.43.0
>
PING.
--
H.J.
More information about the Libc-alpha
mailing list