Hi all, during testing libc-test on RV64, I happended to encounter a segfault on math/fenv.exe . When FE_DFL_ENV is passed to fesetenv(), the very first instruction : lw t1, 0(a0) will fail since a0 is -1 . Here's quick hack - - git diff -- src/fenv/riscv64/fenv.S diff --git a/src/fenv/riscv64/fenv.S b/src/fenv/riscv64/fenv.S index 97f74dd6..06215954 100644 --- a/src/fenv/riscv64/fenv.S +++ b/src/fenv/riscv64/fenv.S @@ -45,8 +45,12 @@ fegetenv: .global fesetenv .type fesetenv, %function fesetenv: + li t2, -1 + li t1, 0 + beq a0, t2, setfpcsr lw t1, 0(a0) - fscsr t0, t1 +setfpcsr: + fscsr t1 li a0, 0 ret And the test case will pass. Sincerely, Ruinland Chuan-Tzu Tsai
* Ruinland ChuanTzu Tsai <ruinland@andestech.com> [2019-12-02 19:06:52 +0800]:
> Hi all,
> during testing libc-test on RV64,
> I happended to encounter a segfault on math/fenv.exe .
>
> When FE_DFL_ENV is passed to fesetenv(),
> the very first instruction : lw t1, 0(a0)
> will fail since a0 is -1 .
>
> Here's quick hack - -
>
> git diff -- src/fenv/riscv64/fenv.S
> diff --git a/src/fenv/riscv64/fenv.S b/src/fenv/riscv64/fenv.S
> index 97f74dd6..06215954 100644
> --- a/src/fenv/riscv64/fenv.S
> +++ b/src/fenv/riscv64/fenv.S
> @@ -45,8 +45,12 @@ fegetenv:
> .global fesetenv
> .type fesetenv, %function
> fesetenv:
> + li t2, -1
> + li t1, 0
> + beq a0, t2, setfpcsr
> lw t1, 0(a0)
> - fscsr t0, t1
> +setfpcsr:
> + fscsr t1
> li a0, 0
> ret
>
> And the test case will pass.
thanks, this looks good
On Mon, Dec 02, 2019 at 07:06:52PM +0800, Ruinland ChuanTzu Tsai wrote:
> Hi all,
> during testing libc-test on RV64,
> I happended to encounter a segfault on math/fenv.exe .
>
> When FE_DFL_ENV is passed to fesetenv(),
> the very first instruction : lw t1, 0(a0)
> will fail since a0 is -1 .
>
> Here's quick hack - -
>
> git diff -- src/fenv/riscv64/fenv.S
> diff --git a/src/fenv/riscv64/fenv.S b/src/fenv/riscv64/fenv.S
> index 97f74dd6..06215954 100644
> --- a/src/fenv/riscv64/fenv.S
> +++ b/src/fenv/riscv64/fenv.S
> @@ -45,8 +45,12 @@ fegetenv:
> .global fesetenv
> .type fesetenv, %function
> fesetenv:
> + li t2, -1
> + li t1, 0
> + beq a0, t2, setfpcsr
> lw t1, 0(a0)
> - fscsr t0, t1
> +setfpcsr:
> + fscsr t1
> li a0, 0
> ret
>
> And the test case will pass.
Looks mostly ok. Applying without the label name, 1f/1: instead. I was
a bit confused by the removal of the t0 operand from fscsr but
apparently the one-operand form implicitly uses the zero register as
rd.
Rich