[PATCH v2 04/14] riscv: Adjust assembly routines to support landing pad
Deepak Gupta
debug@rivosinc.com
Tue Jul 22 20:27:59 GMT 2025
On Fri, Jul 11, 2025 at 06:52:45AM -0700, Jesse Huang wrote:
>Landing pads, instructions for setting the label value as well as
>alignment directives are inserted at where they should be.
>
>Frame offsets for floating point registers in _dl_runtime_resolve
>are also adjusted because now t2 has to be saved onto the stack.
>
>Co-authored-by: Hau Hsu <hau.hsu@sifive.com>
>Co-authored-by: Kito Cheng <kito.cheng@sifive.com>
>---
> sysdeps/riscv/__longjmp.S | 1 +
> sysdeps/riscv/crti.S | 4 ++
> sysdeps/riscv/crtn.S | 4 ++
> sysdeps/riscv/dl-machine.h | 8 +++
> sysdeps/riscv/dl-trampoline.S | 61 +++++++++++------
> sysdeps/riscv/multiarch/memcpy_noalignment.S | 4 ++
> sysdeps/riscv/setjmp.S | 3 +
> sysdeps/riscv/start.S | 10 +++
> sysdeps/unix/sysv/linux/riscv/clone.S | 2 +
> sysdeps/unix/sysv/linux/riscv/sysdep.S | 1 +
> sysdeps/unix/sysv/linux/riscv/sysdep.h | 71 ++++++++++++++++++++
> sysdeps/unix/sysv/linux/riscv/vfork.S | 1 +
> 12 files changed, 150 insertions(+), 20 deletions(-)
> create mode 100644 sysdeps/riscv/crti.S
> create mode 100644 sysdeps/riscv/crtn.S
>
>diff --git a/sysdeps/riscv/__longjmp.S b/sysdeps/riscv/__longjmp.S
>index 7228b457a6..47ff3aa09e 100644
>--- a/sysdeps/riscv/__longjmp.S
>+++ b/sysdeps/riscv/__longjmp.S
>@@ -20,6 +20,7 @@
> #include <sys/asm.h>
>
> ENTRY (__longjmp)
>+ LPAD
> REG_L ra, 0*SZREG(a0)
> REG_L s0, 1*SZREG(a0)
> REG_L s1, 2*SZREG(a0)
>diff --git a/sysdeps/riscv/crti.S b/sysdeps/riscv/crti.S
>new file mode 100644
>index 0000000000..fb1097c5ca
>--- /dev/null
>+++ b/sysdeps/riscv/crti.S
>@@ -0,0 +1,4 @@
>+/* crti.S is empty because .init_array/.fini_array are used exclusively.
>+ Include sysdep.h to define gnu property if necessary. */
>+
>+#include <sysdep.h>
>diff --git a/sysdeps/riscv/crtn.S b/sysdeps/riscv/crtn.S
>new file mode 100644
>index 0000000000..b2e7cb692e
>--- /dev/null
>+++ b/sysdeps/riscv/crtn.S
>@@ -0,0 +1,4 @@
>+/* crtn.S is empty because .init_array/.fini_array are used exclusively.
>+ Include sysdep.h to define gnu property if necessary. */
>+
>+#include <sysdep.h>
>diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
>index dcc3e0883b..76f43a242b 100644
>--- a/sysdeps/riscv/dl-machine.h
>+++ b/sysdeps/riscv/dl-machine.h
>@@ -28,6 +28,13 @@
> #include <dl-irel.h>
> #include <dl-static-tls.h>
> #include <dl-machine-rel.h>
>+/* This is a marker to remind us to add real expansion to setup the label
>+ for the function signature label scheme in the future */
>+#ifdef __riscv_landing_pad_unlabeled
>+# define SET_LPAD
>+#else
>+# define SET_LPAD
>+#endif
>
> #ifndef _RTLD_PROLOGUE
> # define _RTLD_PROLOGUE(entry) \
>@@ -126,6 +133,7 @@ elf_machine_dynamic (void)
> # Pass our finalizer function to _start.\n\
> lla a0, _dl_fini\n\
> # Jump to the user entry point.\n\
>+ " STRINGXV (SET_LPAD) "\n\
> jr s0\n\
> " _RTLD_EPILOGUE (ENTRY_POINT) \
> _RTLD_EPILOGUE (_dl_start_user) "\
>diff --git a/sysdeps/riscv/dl-trampoline.S b/sysdeps/riscv/dl-trampoline.S
>index 93d04af326..2b25f64a3b 100644
>--- a/sysdeps/riscv/dl-trampoline.S
>+++ b/sysdeps/riscv/dl-trampoline.S
>@@ -25,13 +25,27 @@
> /* Assembler veneer called from the PLT header code for lazy loading.
> The PLT header passes its own args in t0-t2. */
>
>-#ifdef __riscv_float_abi_soft
>-# define FRAME_SIZE (-((-10 * SZREG) & ALMASK))
>-#else
>-# define FRAME_SIZE (-((-10 * SZREG - 8 * SZFREG) & ALMASK))
>+#ifdef __riscv_landing_pad
>+
>+# ifdef __riscv_float_abi_soft
>+# define FRAME_SIZE (-((-11 * SZREG) & ALMASK))
>+# else
>+# define FRAME_SIZE (-((-11 * SZREG - 8 * SZFREG) & ALMASK))
>+# define FREG_BASE_OFFSET (11*SZREG)
>+# endif
>+
>+# else
>+
>+# ifdef __riscv_float_abi_soft
>+# define FRAME_SIZE (-((-10 * SZREG) & ALMASK))
>+# else
>+# define FRAME_SIZE (-((-10 * SZREG - 8 * SZFREG) & ALMASK))
>+# define FREG_BASE_OFFSET (10*SZREG)
>+# endif
> #endif
>
> ENTRY (_dl_runtime_resolve)
>+ LPAD
> # Save arguments to stack.
> addi sp, sp, -FRAME_SIZE
> REG_S ra, 9*SZREG(sp)
>@@ -43,16 +57,19 @@ ENTRY (_dl_runtime_resolve)
> REG_S a5, 6*SZREG(sp)
> REG_S a6, 7*SZREG(sp)
> REG_S a7, 8*SZREG(sp)
>+#ifdef __riscv_landing_pad
>+ REG_S t2, 10*SZREG(sp)
>+#endif
>
> #ifndef __riscv_float_abi_soft
>- FREG_S fa0, (10*SZREG + 0*SZFREG)(sp)
>- FREG_S fa1, (10*SZREG + 1*SZFREG)(sp)
>- FREG_S fa2, (10*SZREG + 2*SZFREG)(sp)
>- FREG_S fa3, (10*SZREG + 3*SZFREG)(sp)
>- FREG_S fa4, (10*SZREG + 4*SZFREG)(sp)
>- FREG_S fa5, (10*SZREG + 5*SZFREG)(sp)
>- FREG_S fa6, (10*SZREG + 6*SZFREG)(sp)
>- FREG_S fa7, (10*SZREG + 7*SZFREG)(sp)
>+ FREG_S fa0, (FREG_BASE_OFFSET + 0*SZFREG)(sp)
>+ FREG_S fa1, (FREG_BASE_OFFSET + 1*SZFREG)(sp)
>+ FREG_S fa2, (FREG_BASE_OFFSET + 2*SZFREG)(sp)
>+ FREG_S fa3, (FREG_BASE_OFFSET + 3*SZFREG)(sp)
>+ FREG_S fa4, (FREG_BASE_OFFSET + 4*SZFREG)(sp)
>+ FREG_S fa5, (FREG_BASE_OFFSET + 5*SZFREG)(sp)
>+ FREG_S fa6, (FREG_BASE_OFFSET + 6*SZFREG)(sp)
>+ FREG_S fa7, (FREG_BASE_OFFSET + 7*SZFREG)(sp)
> #endif
>
> # Update .got.plt and obtain runtime address of callee.
>@@ -60,6 +77,7 @@ ENTRY (_dl_runtime_resolve)
> mv a0, t0 # link map
> add a1, a1, t1 # reloc offset (== thrice the .got.plt offset)
> la a2, _dl_fixup
>+ SET_LPAD
> jalr a2
> mv t1, a0
>
>@@ -73,16 +91,19 @@ ENTRY (_dl_runtime_resolve)
> REG_L a5, 6*SZREG(sp)
> REG_L a6, 7*SZREG(sp)
> REG_L a7, 8*SZREG(sp)
>+#ifdef __riscv_landing_pad
>+ REG_L t2, 10*SZREG(sp)
>+#endif
>
> #ifndef __riscv_float_abi_soft
>- FREG_L fa0, (10*SZREG + 0*SZFREG)(sp)
>- FREG_L fa1, (10*SZREG + 1*SZFREG)(sp)
>- FREG_L fa2, (10*SZREG + 2*SZFREG)(sp)
>- FREG_L fa3, (10*SZREG + 3*SZFREG)(sp)
>- FREG_L fa4, (10*SZREG + 4*SZFREG)(sp)
>- FREG_L fa5, (10*SZREG + 5*SZFREG)(sp)
>- FREG_L fa6, (10*SZREG + 6*SZFREG)(sp)
>- FREG_L fa7, (10*SZREG + 7*SZFREG)(sp)
>+ FREG_L fa0, (FREG_BASE_OFFSET + 0*SZFREG)(sp)
>+ FREG_L fa1, (FREG_BASE_OFFSET + 1*SZFREG)(sp)
>+ FREG_L fa2, (FREG_BASE_OFFSET + 2*SZFREG)(sp)
>+ FREG_L fa3, (FREG_BASE_OFFSET + 3*SZFREG)(sp)
>+ FREG_L fa4, (FREG_BASE_OFFSET + 4*SZFREG)(sp)
>+ FREG_L fa5, (FREG_BASE_OFFSET + 5*SZFREG)(sp)
>+ FREG_L fa6, (FREG_BASE_OFFSET + 6*SZFREG)(sp)
>+ FREG_L fa7, (FREG_BASE_OFFSET + 7*SZFREG)(sp)
> #endif
>
> addi sp, sp, FRAME_SIZE
>diff --git a/sysdeps/riscv/multiarch/memcpy_noalignment.S b/sysdeps/riscv/multiarch/memcpy_noalignment.S
>index dd135f4a4d..e97748e82b 100644
>--- a/sysdeps/riscv/multiarch/memcpy_noalignment.S
>+++ b/sysdeps/riscv/multiarch/memcpy_noalignment.S
>@@ -34,7 +34,11 @@
> #define BLOCK_SIZE (16 * SZREG)
>
> .attribute unaligned_access, 1
>+#ifdef __riscv_landing_pad
>+ .align 2
>+#endif
> ENTRY (__memcpy_noalignment)
>+ LPAD
> beq a2, zero, L(ret)
>
> /* if LEN < SZREG jump to tail handling. */
>diff --git a/sysdeps/riscv/setjmp.S b/sysdeps/riscv/setjmp.S
>index 600ea84db6..df048cb544 100644
>--- a/sysdeps/riscv/setjmp.S
>+++ b/sysdeps/riscv/setjmp.S
>@@ -20,14 +20,17 @@
> #include <sys/asm.h>
>
> ENTRY (_setjmp)
>+ LPAD
> li a1, 0
> j HIDDEN_JUMPTARGET (__sigsetjmp)
> END (_setjmp)
> ENTRY (setjmp)
>+ LPAD
> li a1, 1
> /* Fallthrough */
> END (setjmp)
> ENTRY (__sigsetjmp)
>+ LPAD
> REG_S ra, 0*SZREG(a0)
> REG_S s0, 1*SZREG(a0)
> REG_S s1, 2*SZREG(a0)
>diff --git a/sysdeps/riscv/start.S b/sysdeps/riscv/start.S
>index 2db79c0ae6..ff083b13a4 100644
>--- a/sysdeps/riscv/start.S
>+++ b/sysdeps/riscv/start.S
>@@ -47,12 +47,14 @@ ENTRY (ENTRY_POINT)
> .cfi_label to force starting the FDE. */
> .cfi_label .Ldummy
> cfi_undefined (ra)
>+ LPAD
> call load_gp
> mv a5, a0 /* rtld_fini. */
> /* main may be in a shared library. */
> #if defined PIC && !defined SHARED
> /* Avoid relocation in static PIE since _start is called before it
> is relocated. */
>+ SET_LPAD
> lla a0, __wrap_main
> #else
> la a0, main
>@@ -69,7 +71,11 @@ ENTRY (ENTRY_POINT)
> END (ENTRY_POINT)
>
> #if defined PIC && !defined SHARED
>+#ifdef __riscv_landing_pad
>+ .align 2
>+#endif /* __riscv_landing_pad */
> __wrap_main:
>+ LPAD
> tail main@plt
> #endif
>
>@@ -79,9 +85,13 @@ __wrap_main:
> needs to be initialized before calling __libc_start_main in that case.
> So we redundantly initialize it at the beginning of _start. */
>
>+#ifdef __riscv_landing_pad
>+ .align 2
>+#endif /* __riscv_landing_pad */
> load_gp:
> .option push
> .option norelax
>+ LPAD
> lla gp, __global_pointer$
> .option pop
> ret
>diff --git a/sysdeps/unix/sysv/linux/riscv/clone.S b/sysdeps/unix/sysv/linux/riscv/clone.S
>index 1362dd9c22..d6a0996ec8 100644
>--- a/sysdeps/unix/sysv/linux/riscv/clone.S
>+++ b/sysdeps/unix/sysv/linux/riscv/clone.S
>@@ -31,6 +31,7 @@
>
> .text
> LEAF (__clone)
>+ LPAD
>
> /* Align stack to a 128-bit boundary as per RISC-V ABI. */
> andi a1,a1,ALMASK
>@@ -82,6 +83,7 @@ L (thread_start):
> REG_L a0,SZREG(sp) /* Argument pointer. */
>
> /* Call the user's function. */
>+ SET_LPAD
> jalr a1
>
> /* Call exit with the function's return value. */
>diff --git a/sysdeps/unix/sysv/linux/riscv/sysdep.S b/sysdeps/unix/sysv/linux/riscv/sysdep.S
>index b50671b9b7..e7ea424e45 100644
>--- a/sysdeps/unix/sysv/linux/riscv/sysdep.S
>+++ b/sysdeps/unix/sysv/linux/riscv/sysdep.S
>@@ -23,6 +23,7 @@
> #endif
>
> ENTRY (__syscall_error)
>+ LPAD
> mv t0, ra
> /* Fall through to __syscall_set_errno. */
> END (__syscall_error)
>diff --git a/sysdeps/unix/sysv/linux/riscv/sysdep.h b/sysdeps/unix/sysv/linux/riscv/sysdep.h
>index ee015dfeb6..14a1f3a647 100644
>--- a/sysdeps/unix/sysv/linux/riscv/sysdep.h
>+++ b/sysdeps/unix/sysv/linux/riscv/sysdep.h
>@@ -53,6 +53,75 @@
>
> # include <sys/asm.h>
>
>+/* GNU_PROPERTY_RISCV_* macros from elf.h for use in asm code. */
>+#define FEATURE_1_AND 0xc0000000
>+
>+/* Add a NT_GNU_PROPERTY_TYPE_0 note. */
>+#if __riscv_xlen == 32
>+# define GNU_PROPERTY(type, value) \
>+ .section .note.gnu.property, "a"; \
>+ .p2align 2; \
>+ .word 4; \
>+ .word 12; \
>+ .word 5; \
>+ .asciz "GNU"; \
>+ .word type; \
>+ .word 4; \
>+ .word value; \
>+ .text
>+#else
>+# define GNU_PROPERTY(type, value) \
>+ .section .note.gnu.property, "a"; \
>+ .p2align 3; \
>+ .word 4; \
>+ .word 16; \
>+ .word 5; \
>+ .asciz "GNU"; \
>+ .word type; \
>+ .word 4; \
>+ .word value; \
>+ .word 0; \
>+ .text
>+#endif
+ Heinrich
Do we need a version field for unlabeled landing pad scheme?
I have raised it with Kito, for labeled one's we need version for sure.
Although if we start with unlabaled gnu property, it will look
canonical.
>+
>+/* Add GNU property note with the supported features to all asm code
>+ where sysdep.h is included. */
>+#undef __VALUE_FOR_FEATURE_1_AND
>+#if defined (__riscv_landing_pad) || defined (__riscv_shadow_stack)
>+# if defined (__riscv_landing_pad_unlabeled)
>+# if defined (__riscv_shadow_stack)
>+# define __VALUE_FOR_FEATURE_1_AND 0x3
>+# else
>+# define __VALUE_FOR_FEATURE_1_AND 0x1
>+# endif
>+# elif defined (__riscv_landing_pad_func_sig)
>+# if defined (__riscv_shadow_stack)
>+# define __VALUE_FOR_FEATURE_1_AND 0x6
>+# else
>+# define __VALUE_FOR_FEATURE_1_AND 0x4
>+# endif
>+# else
>+# if defined (__riscv_shadow_stack)
>+# define __VALUE_FOR_FEATURE_1_AND 0x2
>+# else
>+# error "What?"
>+# endif
>+# endif
>+#endif
>+
>+#if defined (__VALUE_FOR_FEATURE_1_AND)
>+GNU_PROPERTY (FEATURE_1_AND, __VALUE_FOR_FEATURE_1_AND)
>+#endif
>+#undef __VALUE_FOR_FEATURE_1_AND
>+
>+#ifdef __riscv_landing_pad_unlabeled
>+# define SET_LPAD
>+# define LPAD lpad 0
>+#else
>+# define SET_LPAD
>+# define LPAD
>+#endif
>+
> # define ENTRY(name) LEAF(name)
>
> # define L(label) .L ## label
>@@ -64,6 +133,7 @@
> .text; \
> .align 2; \
> ENTRY (name); \
>+ LPAD; \
> li a7, SYS_ify (syscall_name); \
> scall; \
> li a7, -4096; \
>@@ -111,6 +181,7 @@
> # define PSEUDO_NOERRNO(name, syscall_name, args) \
> .align 2; \
> ENTRY (name); \
>+ LPAD; \
> li a7, SYS_ify (syscall_name); \
> scall;
>
>diff --git a/sysdeps/unix/sysv/linux/riscv/vfork.S b/sysdeps/unix/sysv/linux/riscv/vfork.S
>index 1e39b7a057..4cf7b5a9d8 100644
>--- a/sysdeps/unix/sysv/linux/riscv/vfork.S
>+++ b/sysdeps/unix/sysv/linux/riscv/vfork.S
>@@ -29,6 +29,7 @@
>
> .text
> LEAF (__libc_vfork)
>+ LPAD
>
> li a0, (CLONE_VFORK | CLONE_VM | SIGCHLD)
> mv a1, sp
>--
>2.39.3
>
More information about the Libc-alpha
mailing list